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;
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 const struct lu_name *tname,
758 const struct lu_name *sname,
759 struct thandle *handle)
761 struct obd_device *obd = mdd2obd_dev(mdd);
762 struct llog_ctxt *ctxt;
763 struct llog_changelog_rec *rec;
765 struct thandle *llog_th;
770 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
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;
944 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
945 changelog_rec_varsize(&rec->cr));
947 /* llog_lvfs_write_rec sets the llog tail len */
948 rec->cr_hdr.lrh_type = CHANGELOG_REC;
949 rec->cr.cr_time = cl_time();
951 spin_lock(&mdd->mdd_cl.mc_lock);
952 /* NB: I suppose it's possible llog_add adds out of order wrt cr_index,
953 * but as long as the MDD transactions are ordered correctly for e.g.
954 * rename conflicts, I don't think this should matter. */
955 rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
956 spin_unlock(&mdd->mdd_cl.mc_lock);
958 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
962 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
964 GOTO(out_put, rc = PTR_ERR(llog_th));
966 /* nested journal transaction */
967 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
969 /* time to recover some space ?? */
970 spin_lock(&mdd->mdd_cl.mc_lock);
971 if (unlikely(mdd->mdd_changelog_gc && (ktime_get_real_seconds() -
972 mdd->mdd_cl.mc_gc_time > mdd->mdd_changelog_min_gc_interval) &&
973 mdd->mdd_cl.mc_gc_task == NULL &&
974 llog_cat_free_space(ctxt->loc_handle) <=
975 mdd->mdd_changelog_min_free_cat_entries)) {
976 CWARN("%s: low on changelog_catalog free entries, starting "
977 "ChangeLog garbage collection thread\n", obd->obd_name);
979 /* indicate further kthread run will occur outside right after
982 mdd->mdd_cl.mc_gc_task = (struct task_struct *)(-1);
985 spin_unlock(&mdd->mdd_cl.mc_lock);
987 struct task_struct *gc_task;
989 gc_task = kthread_run(mdd_chlg_garbage_collect, mdd,
991 if (IS_ERR(gc_task)) {
992 CERROR("%s: cannot start ChangeLog garbage collection "
993 "thread: rc = %ld\n", obd->obd_name,
995 mdd->mdd_cl.mc_gc_task = NULL;
997 CDEBUG(D_HA, "%s: ChangeLog garbage collection thread "
998 "has started with Pid %d\n", obd->obd_name,
1000 mdd->mdd_cl.mc_gc_task = gc_task;
1001 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
1005 llog_ctxt_put(ctxt);
1011 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
1012 const struct lu_fid *sfid,
1013 const struct lu_fid *spfid,
1014 const struct lu_name *sname)
1016 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
1017 size_t extsize = sname->ln_namelen + 1;
1019 LASSERT(sfid != NULL);
1020 LASSERT(spfid != NULL);
1021 LASSERT(sname != NULL);
1023 rnm->cr_sfid = *sfid;
1024 rnm->cr_spfid = *spfid;
1026 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
1027 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
1028 rec->cr_namelen += extsize;
1031 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
1033 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
1035 if (jobid == NULL || jobid[0] == '\0')
1038 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
1041 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
1043 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
1045 ef->cr_extra_flags = eflags;
1048 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
1049 __u64 uid, __u64 gid)
1051 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
1053 uidgid->cr_uid = uid;
1054 uidgid->cr_gid = gid;
1057 /** Store a namespace change changelog record
1058 * If this fails, we must fail the whole transaction; we don't
1059 * want the change to commit without the log entry.
1060 * \param target - mdd_object of change
1061 * \param tpfid - target parent dir/object fid
1062 * \param sfid - source object fid
1063 * \param spfid - source parent fid
1064 * \param tname - target name string
1065 * \param sname - source name string
1066 * \param handle - transaction handle
1068 int mdd_changelog_ns_store(const struct lu_env *env,
1069 struct mdd_device *mdd,
1070 enum changelog_rec_type type,
1071 enum changelog_rec_flags crf,
1072 struct mdd_object *target,
1073 const struct lu_fid *tpfid,
1074 const struct lu_fid *sfid,
1075 const struct lu_fid *spfid,
1076 const struct lu_name *tname,
1077 const struct lu_name *sname,
1078 struct thandle *handle)
1080 const struct lu_ucred *uc = lu_ucred(env);
1081 struct llog_changelog_rec *rec;
1084 __u64 xflags = CLFE_INVALID;
1089 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
1092 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
1095 LASSERT(tpfid != NULL);
1096 LASSERT(tname != NULL);
1097 LASSERT(handle != NULL);
1099 reclen = mdd_llog_record_calc_size(env, tname, sname);
1100 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1101 if (buf->lb_buf == NULL)
1105 crf &= CLF_FLAGMASK;
1106 crf |= CLF_EXTRA_FLAGS;
1108 if (uc != NULL && uc->uc_jobid[0] != '\0')
1116 xflags |= CLFE_UIDGID;
1118 rec->cr.cr_flags = crf;
1120 if (crf & CLF_EXTRA_FLAGS) {
1121 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1122 if (xflags & CLFE_UIDGID)
1123 mdd_changelog_rec_extra_uidgid(&rec->cr,
1124 uc->uc_uid, uc->uc_gid);
1127 rec->cr.cr_type = (__u32)type;
1128 rec->cr.cr_pfid = *tpfid;
1129 rec->cr.cr_namelen = tname->ln_namelen;
1130 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1132 if (crf & CLF_RENAME)
1133 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1135 if (crf & CLF_JOBID)
1136 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1138 if (likely(target != NULL)) {
1139 rec->cr.cr_tfid = *mdo2fid(target);
1140 target->mod_cltime = ktime_get();
1142 fid_zero(&rec->cr.cr_tfid);
1145 rc = mdd_changelog_store(env, mdd, rec, handle);
1147 CERROR("%s: cannot store changelog record: type = %d, "
1148 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1149 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1150 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1157 static int __mdd_links_add(const struct lu_env *env,
1158 struct mdd_object *mdd_obj,
1159 struct linkea_data *ldata,
1160 const struct lu_name *lname,
1161 const struct lu_fid *pfid,
1162 int first, int check)
1166 if (ldata->ld_leh == NULL) {
1167 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1171 rc = linkea_data_new(ldata,
1172 &mdd_env_info(env)->mti_link_buf);
1179 rc = linkea_links_find(ldata, lname, pfid);
1180 if (rc && rc != -ENOENT)
1186 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1187 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1191 linkea_add_buf(ldata, lname, tfid);
1194 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1195 linkea_add_buf(ldata, lname, pfid);
1197 return linkea_add_buf(ldata, lname, pfid);
1200 static int __mdd_links_del(const struct lu_env *env,
1201 struct mdd_object *mdd_obj,
1202 struct linkea_data *ldata,
1203 const struct lu_name *lname,
1204 const struct lu_fid *pfid)
1208 if (ldata->ld_leh == NULL) {
1209 rc = mdd_links_read(env, mdd_obj, ldata);
1214 rc = linkea_links_find(ldata, lname, pfid);
1218 linkea_del_buf(ldata, lname);
1222 static int mdd_linkea_prepare(const struct lu_env *env,
1223 struct mdd_object *mdd_obj,
1224 const struct lu_fid *oldpfid,
1225 const struct lu_name *oldlname,
1226 const struct lu_fid *newpfid,
1227 const struct lu_name *newlname,
1228 int first, int check,
1229 struct linkea_data *ldata)
1234 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1237 LASSERT(oldpfid != NULL || newpfid != NULL);
1239 if (mdd_obj->mod_flags & DEAD_OBJ)
1240 /* Unnecessary to update linkEA for dead object. */
1243 if (oldpfid != NULL) {
1244 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1246 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1249 /* No changes done. */
1254 /* If renaming, add the new record */
1255 if (newpfid != NULL)
1256 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1262 int mdd_links_rename(const struct lu_env *env,
1263 struct mdd_object *mdd_obj,
1264 const struct lu_fid *oldpfid,
1265 const struct lu_name *oldlname,
1266 const struct lu_fid *newpfid,
1267 const struct lu_name *newlname,
1268 struct thandle *handle,
1269 struct linkea_data *ldata,
1270 int first, int check)
1275 if (ldata == NULL) {
1276 ldata = &mdd_env_info(env)->mti_link_data;
1277 memset(ldata, 0, sizeof(*ldata));
1278 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1279 newpfid, newlname, first, check, ldata);
1284 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1285 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1291 if (newlname == NULL)
1292 CERROR("link_ea add failed %d "DFID"\n",
1293 rc, PFID(mdd_object_fid(mdd_obj)));
1294 else if (oldpfid == NULL)
1295 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1296 newlname->ln_namelen, newlname->ln_name, rc,
1297 PFID(mdd_object_fid(mdd_obj)));
1298 else if (newpfid == NULL)
1299 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1300 oldlname->ln_namelen, oldlname->ln_name, rc,
1301 PFID(mdd_object_fid(mdd_obj)));
1303 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1304 "\n", oldlname->ln_namelen, oldlname->ln_name,
1305 newlname->ln_namelen, newlname->ln_name, rc,
1306 PFID(mdd_object_fid(mdd_obj)));
1309 if (is_vmalloc_addr(ldata->ld_buf))
1310 /* if we vmalloced a large buffer drop it */
1311 lu_buf_free(ldata->ld_buf);
1316 static inline int mdd_links_add(const struct lu_env *env,
1317 struct mdd_object *mdd_obj,
1318 const struct lu_fid *pfid,
1319 const struct lu_name *lname,
1320 struct thandle *handle,
1321 struct linkea_data *ldata, int first)
1323 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1324 pfid, lname, handle, ldata, first, 0);
1327 static inline int mdd_links_del(const struct lu_env *env,
1328 struct mdd_object *mdd_obj,
1329 const struct lu_fid *pfid,
1330 const struct lu_name *lname,
1331 struct thandle *handle)
1333 return mdd_links_rename(env, mdd_obj, pfid, lname,
1334 NULL, NULL, handle, NULL, 0, 0);
1337 /** Read the link EA into a temp buffer.
1338 * Uses the name_buf since it is generally large.
1339 * \retval IS_ERR err
1340 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1342 struct lu_buf *mdd_links_get(const struct lu_env *env,
1343 struct mdd_object *mdd_obj)
1345 struct linkea_data ldata = { NULL };
1348 rc = mdd_links_read(env, mdd_obj, &ldata);
1349 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1352 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1353 struct linkea_data *ldata, struct thandle *handle)
1355 const struct lu_buf *buf;
1358 if (ldata == NULL || ldata->ld_buf == NULL ||
1359 ldata->ld_leh == NULL)
1362 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1366 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1367 ldata->ld_leh->leh_len);
1368 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1369 if (unlikely(rc == -ENOSPC)) {
1370 rc = linkea_overflow_shrink(ldata);
1378 static int mdd_declare_links_add(const struct lu_env *env,
1379 struct mdd_object *mdd_obj,
1380 struct thandle *handle,
1381 struct linkea_data *ldata)
1387 if (ldata != NULL && ldata->ld_leh != NULL) {
1388 ea_len = ldata->ld_leh->leh_len;
1389 linkea = ldata->ld_buf->lb_buf;
1391 ea_len = MAX_LINKEA_SIZE;
1395 rc = mdo_declare_xattr_set(env, mdd_obj,
1396 mdd_buf_get_const(env, linkea, ea_len),
1397 XATTR_NAME_LINK, 0, handle);
1402 static inline int mdd_declare_links_del(const struct lu_env *env,
1403 struct mdd_object *c,
1404 struct thandle *handle)
1408 /* For directory, the linkEA will be removed together
1409 * with the object. */
1410 if (!S_ISDIR(mdd_object_type(c)))
1411 rc = mdd_declare_links_add(env, c, handle, NULL);
1416 static int mdd_declare_link(const struct lu_env *env,
1417 struct mdd_device *mdd,
1418 struct mdd_object *p,
1419 struct mdd_object *c,
1420 const struct lu_name *name,
1421 struct thandle *handle,
1423 struct linkea_data *data)
1425 struct lu_fid tfid = *mdo2fid(c);
1428 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1429 tfid.f_oid = cfs_fail_val;
1431 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1432 name->ln_name, handle);
1436 rc = mdo_declare_ref_add(env, c, handle);
1440 la->la_valid = LA_CTIME | LA_MTIME;
1441 rc = mdo_declare_attr_set(env, p, la, handle);
1445 la->la_valid = LA_CTIME;
1446 rc = mdo_declare_attr_set(env, c, la, handle);
1450 rc = mdd_declare_links_add(env, c, handle, data);
1454 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1459 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1460 struct md_object *src_obj, const struct lu_name *lname,
1463 const char *name = lname->ln_name;
1464 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1465 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1466 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1467 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1468 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1469 struct mdd_device *mdd = mdo2mdd(src_obj);
1470 struct thandle *handle;
1471 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1472 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1476 rc = mdd_la_get(env, mdd_sobj, cattr);
1480 rc = mdd_la_get(env, mdd_tobj, tattr);
1485 * If we are using project inheritance, we only allow hard link
1486 * creation in our tree when the project IDs are the same;
1487 * otherwise the tree quota mechanism could be circumvented.
1489 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1490 (tattr->la_projid != cattr->la_projid))
1493 handle = mdd_trans_create(env, mdd);
1495 GOTO(out_pending, rc = PTR_ERR(handle));
1497 memset(ldata, 0, sizeof(*ldata));
1499 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1500 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1502 /* Note: even this function will change ldata, but it comes from
1503 * thread_info, which is completely temporary and only seen in
1504 * this function, so we do not need reset ldata once it fails.*/
1505 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1506 lname, 0, 0, ldata);
1510 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1515 rc = mdd_trans_start(env, mdd, handle);
1519 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1520 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1523 GOTO(out_unlock, rc);
1525 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1526 rc = mdo_ref_add(env, mdd_sobj, handle);
1528 GOTO(out_unlock, rc);
1531 *tfid = *mdo2fid(mdd_sobj);
1532 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1533 tfid->f_oid = cfs_fail_val;
1535 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1536 mdd_object_type(mdd_sobj), name, handle);
1538 mdo_ref_del(env, mdd_sobj, handle);
1539 GOTO(out_unlock, rc);
1542 la->la_valid = LA_CTIME | LA_MTIME;
1543 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1545 GOTO(out_unlock, rc);
1547 la->la_valid = LA_CTIME;
1548 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1550 /* Note: The failure of links_add should not cause the
1551 * link failure, so do not check return value. */
1552 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1553 lname, handle, ldata, 0);
1557 mdd_write_unlock(env, mdd_sobj);
1559 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1560 mdo2fid(mdd_tobj), NULL, NULL,
1561 lname, NULL, handle);
1563 rc = mdd_trans_stop(env, mdd, rc, handle);
1564 if (is_vmalloc_addr(ldata->ld_buf))
1565 /* if we vmalloced a large buffer drop it */
1566 lu_buf_free(ldata->ld_buf);
1571 static int mdd_mark_orphan_object(const struct lu_env *env,
1572 struct mdd_object *obj, struct thandle *handle,
1575 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1578 if (!S_ISDIR(mdd_object_type(obj)))
1581 attr->la_valid = LA_FLAGS;
1582 attr->la_flags = LUSTRE_ORPHAN_FL;
1585 rc = mdo_declare_attr_set(env, obj, attr, handle);
1587 rc = mdo_attr_set(env, obj, attr, handle);
1592 static int mdd_declare_finish_unlink(const struct lu_env *env,
1593 struct mdd_object *obj,
1594 struct thandle *handle)
1598 /* Sigh, we do not know if the unlink object will become orphan in
1599 * declare phase, but fortunately the flags here does not matter
1600 * in current declare implementation */
1601 rc = mdd_mark_orphan_object(env, obj, handle, true);
1605 rc = mdo_declare_destroy(env, obj, handle);
1609 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1613 return mdd_declare_links_del(env, obj, handle);
1616 /* caller should take a lock before calling */
1617 int mdd_finish_unlink(const struct lu_env *env,
1618 struct mdd_object *obj, struct md_attr *ma,
1619 const struct mdd_object *pobj,
1620 const struct lu_name *lname,
1624 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1627 LASSERT(mdd_write_locked(env, obj) != 0);
1629 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1630 /* add new orphan and the object
1631 * will be deleted during mdd_close() */
1632 obj->mod_flags |= DEAD_OBJ;
1633 if (obj->mod_count) {
1634 rc = __mdd_orphan_add(env, obj, th);
1636 CDEBUG(D_HA, "Object "DFID" is inserted into "
1637 "orphan list, open count = %d\n",
1638 PFID(mdd_object_fid(obj)),
1641 CERROR("Object "DFID" fail to be an orphan, "
1642 "open count = %d, maybe cause failed "
1644 PFID(mdd_object_fid(obj)),
1647 /* mark object as an orphan here, not
1648 * before __mdd_orphan_add() as racing
1649 * mdd_la_get() may propagate ORPHAN_OBJ
1650 * causing the asserition */
1651 rc = mdd_mark_orphan_object(env, obj, th, false);
1653 rc = mdo_destroy(env, obj, th);
1655 } else if (!is_dir) {
1656 /* old files may not have link ea; ignore errors */
1657 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1665 * has mdd_write_lock on cobj already, but not on pobj yet
1667 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1668 const struct lu_attr *pattr,
1669 struct mdd_object *cobj,
1670 const struct lu_attr *cattr)
1675 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1680 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1681 struct mdd_object *p, struct mdd_object *c,
1682 const struct lu_name *name, struct md_attr *ma,
1683 struct thandle *handle, int no_name, int is_dir)
1685 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1688 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1689 if (likely(no_name == 0)) {
1690 rc = mdo_declare_index_delete(env, p, name->ln_name,
1697 rc = mdo_declare_ref_del(env, p, handle);
1703 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1704 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1705 la->la_valid = LA_CTIME | LA_MTIME;
1706 rc = mdo_declare_attr_set(env, p, la, handle);
1711 rc = mdo_declare_ref_del(env, c, handle);
1715 rc = mdo_declare_ref_del(env, c, handle);
1719 la->la_valid = LA_CTIME;
1720 rc = mdo_declare_attr_set(env, c, la, handle);
1724 rc = mdd_declare_finish_unlink(env, c, handle);
1728 /* FIXME: need changelog for remove entry */
1729 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1736 * test if a file has an HSM archive
1737 * if HSM attributes are not found in ma update them from
1740 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1741 struct mdd_object *obj,
1746 if (!(ma->ma_valid & MA_HSM)) {
1747 /* no HSM MD provided, read xattr */
1748 struct lu_buf *hsm_buf;
1749 const size_t buflen = sizeof(struct hsm_attrs);
1752 hsm_buf = mdd_buf_get(env, NULL, 0);
1753 lu_buf_alloc(hsm_buf, buflen);
1754 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1755 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1756 lu_buf_free(hsm_buf);
1760 ma->ma_valid |= MA_HSM;
1762 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1768 * Delete name entry and the object.
1769 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1770 * does not exist for this object, and it could only happen during resending
1771 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1772 * is also needed in this case(needed by changelog), so we have to add another
1773 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1774 * the ENOENT failure should be able to be fixed by redo mechanism.
1776 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1777 struct md_object *cobj, const struct lu_name *lname,
1778 struct md_attr *ma, int no_name)
1780 const char *name = lname->ln_name;
1781 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1782 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1783 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1784 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1785 struct mdd_object *mdd_cobj = NULL;
1786 struct mdd_device *mdd = mdo2mdd(pobj);
1787 struct thandle *handle;
1788 int rc, is_dir = 0, cl_flags = 0;
1791 /* cobj == NULL means only delete name entry */
1792 if (likely(cobj != NULL)) {
1793 mdd_cobj = md2mdd_obj(cobj);
1794 if (mdd_object_exists(mdd_cobj) == 0)
1798 rc = mdd_la_get(env, mdd_pobj, pattr);
1802 if (likely(mdd_cobj != NULL)) {
1804 rc = mdd_la_get(env, mdd_cobj, cattr);
1808 is_dir = S_ISDIR(cattr->la_mode);
1809 /* search for an existing archive.
1810 * we should check ahead as the object
1811 * can be destroyed in this transaction */
1812 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1813 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1816 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1820 handle = mdd_trans_create(env, mdd);
1822 RETURN(PTR_ERR(handle));
1824 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1825 lname, ma, handle, no_name, is_dir);
1829 rc = mdd_trans_start(env, mdd, handle);
1833 if (likely(mdd_cobj != NULL))
1834 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1836 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1837 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1842 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1843 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1844 GOTO(cleanup, rc = 0);
1846 if (likely(mdd_cobj != NULL)) {
1847 rc = mdo_ref_del(env, mdd_cobj, handle);
1849 __mdd_index_insert_only(env, mdd_pobj,
1851 mdd_object_type(mdd_cobj),
1858 mdo_ref_del(env, mdd_cobj, handle);
1860 /* fetch updated nlink */
1861 rc = mdd_la_get(env, mdd_cobj, cattr);
1866 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1867 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1869 la->la_valid = LA_CTIME | LA_MTIME;
1870 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1874 /* Enough for only unlink the entry */
1875 if (unlikely(mdd_cobj == NULL))
1878 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1879 /* update ctime of an unlinked file only if it is still
1880 * opened or a link still exists */
1881 la->la_valid = LA_CTIME;
1882 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1887 /* XXX: this transfer to ma will be removed with LOD/OSP */
1888 ma->ma_attr = *cattr;
1889 ma->ma_valid |= MA_INODE;
1890 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1894 /* fetch updated nlink */
1895 rc = mdd_la_get(env, mdd_cobj, cattr);
1896 /* if object is removed then we can't get its attrs,
1898 if (rc == -ENOENT) {
1899 cattr->la_nlink = 0;
1903 if (cattr->la_nlink == 0) {
1904 ma->ma_attr = *cattr;
1905 ma->ma_valid |= MA_INODE;
1910 if (likely(mdd_cobj != NULL))
1911 mdd_write_unlock(env, mdd_cobj);
1914 if (cattr->la_nlink == 0)
1915 cl_flags |= CLF_UNLINK_LAST;
1917 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1919 rc = mdd_changelog_ns_store(env, mdd,
1920 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1921 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1926 rc = mdd_trans_stop(env, mdd, rc, handle);
1932 * The permission has been checked when obj created, no need check again.
1934 static int mdd_cd_sanity_check(const struct lu_env *env,
1935 struct mdd_object *obj)
1940 if (!obj || mdd_is_dead_obj(obj))
1946 static int mdd_create_data(const struct lu_env *env,
1947 struct md_object *pobj,
1948 struct md_object *cobj,
1949 const struct md_op_spec *spec,
1952 struct mdd_device *mdd = mdo2mdd(cobj);
1953 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1954 struct mdd_object *son = md2mdd_obj(cobj);
1955 struct thandle *handle;
1956 const struct lu_buf *buf;
1957 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1958 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1962 rc = mdd_cd_sanity_check(env, son);
1966 if (!md_should_create(spec->sp_cr_flags))
1970 * there are following use cases for this function:
1971 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1972 * striping can be specified or not
1975 rc = mdd_la_get(env, son, attr);
1979 /* calling ->ah_make_hint() is used to transfer information from parent */
1980 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1982 handle = mdd_trans_create(env, mdd);
1984 GOTO(out_free, rc = PTR_ERR(handle));
1987 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1988 * Should this be fixed?
1990 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
1991 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
1992 spec->sp_cr_flags, spec->no_create);
1994 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
1995 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
1996 spec->u.sp_ea.eadatalen);
2001 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2002 XATTR_NAME_LOV, 0, handle);
2006 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
2010 rc = mdd_trans_start(env, mdd, handle);
2014 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2020 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
2023 rc = mdd_trans_stop(env, mdd, rc, handle);
2029 static int mdd_declare_object_initialize(const struct lu_env *env,
2030 struct mdd_object *parent,
2031 struct mdd_object *child,
2032 const struct lu_attr *attr,
2033 struct thandle *handle)
2038 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2039 if (!S_ISDIR(attr->la_mode))
2042 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
2047 rc = mdo_declare_ref_add(env, child, handle);
2051 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
2057 static int mdd_object_initialize(const struct lu_env *env,
2058 const struct lu_fid *pfid,
2059 struct mdd_object *child,
2060 struct lu_attr *attr, struct thandle *handle,
2061 const struct md_op_spec *spec)
2066 if (S_ISDIR(attr->la_mode)) {
2067 /* Add "." and ".." for newly created dir */
2068 mdo_ref_add(env, child, handle);
2069 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
2070 S_IFDIR, dot, handle);
2072 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2075 mdo_ref_del(env, child, handle);
2082 * This function checks whether it can create a file/dir under the
2083 * directory(@pobj). The directory(@pobj) is not being locked by
2086 * \param[in] env execution environment
2087 * \param[in] pobj the directory to create files
2088 * \param[in] pattr the attributes of the directory
2089 * \param[in] lname the name of the created file/dir
2090 * \param[in] cattr the attributes of the file/dir
2091 * \param[in] spec create specification
2093 * \retval = 0 it is allowed to create file/dir under
2095 * \retval negative error not allowed to create file/dir
2096 * under the directory
2098 static int mdd_create_sanity_check(const struct lu_env *env,
2099 struct md_object *pobj,
2100 const struct lu_attr *pattr,
2101 const struct lu_name *lname,
2102 struct lu_attr *cattr,
2103 struct md_op_spec *spec)
2105 struct mdd_thread_info *info = mdd_env_info(env);
2106 struct lu_fid *fid = &info->mti_fid;
2107 struct mdd_object *obj = md2mdd_obj(pobj);
2108 struct mdd_device *m = mdo2mdd(pobj);
2109 bool check_perm = true;
2114 if (mdd_is_dead_obj(obj))
2118 * In some cases this lookup is not needed - we know before if name
2119 * exists or not because MDT performs lookup for it.
2120 * name length check is done in lookup.
2122 if (spec->sp_cr_lookup) {
2124 * Check if the name already exist, though it will be checked in
2125 * _index_insert also, for avoiding rolling back if exists
2128 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2129 MAY_WRITE | MAY_EXEC);
2131 RETURN(rc ? : -EEXIST);
2133 /* Permission is already being checked in mdd_lookup */
2137 if (S_ISDIR(cattr->la_mode) &&
2138 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2139 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2140 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2142 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2143 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2144 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2146 CERROR("%s: invalid lmv_user_md: magic = %x, "
2147 "stripe_offset = %d, stripe_count = %u: "
2148 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2149 le32_to_cpu(lum->lum_magic),
2150 (int)le32_to_cpu(lum->lum_stripe_offset),
2151 le32_to_cpu(lum->lum_stripe_count), rc);
2156 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2161 if (pattr->la_mode & S_ISGID) {
2162 cattr->la_gid = pattr->la_gid;
2163 if (S_ISDIR(cattr->la_mode)) {
2164 cattr->la_mode |= S_ISGID;
2165 cattr->la_valid |= LA_MODE;
2169 /* Inherit project ID from parent directory */
2170 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2171 cattr->la_projid = pattr->la_projid;
2172 if (S_ISDIR(cattr->la_mode)) {
2173 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2174 cattr->la_valid |= LA_FLAGS;
2176 cattr->la_valid |= LA_PROJID;
2179 rc = mdd_name_check(m, lname);
2183 switch (cattr->la_mode & S_IFMT) {
2185 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2187 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2188 RETURN(-ENAMETOOLONG);
2207 static int mdd_declare_create_object(const struct lu_env *env,
2208 struct mdd_device *mdd,
2209 struct mdd_object *p, struct mdd_object *c,
2210 struct lu_attr *attr,
2211 struct thandle *handle,
2212 const struct md_op_spec *spec,
2213 struct lu_buf *def_acl_buf,
2214 struct lu_buf *acl_buf,
2215 struct dt_allocation_hint *hint)
2217 const struct lu_buf *buf;
2220 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2225 #ifdef CONFIG_FS_POSIX_ACL
2226 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2227 /* if dir, then can inherit default ACl */
2228 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2229 XATTR_NAME_ACL_DEFAULT,
2235 if (acl_buf->lb_len > 0) {
2236 rc = mdo_declare_attr_set(env, c, attr, handle);
2240 rc = mdo_declare_xattr_set(env, c, acl_buf,
2241 XATTR_NAME_ACL_ACCESS, 0, handle);
2246 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2250 /* replay case, create LOV EA from client data */
2251 if (spec->no_create ||
2252 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2253 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2254 spec->u.sp_ea.eadatalen);
2255 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2261 if (S_ISLNK(attr->la_mode)) {
2262 const char *target_name = spec->u.sp_symname;
2263 int sym_len = strlen(target_name);
2264 const struct lu_buf *buf;
2266 buf = mdd_buf_get_const(env, target_name, sym_len);
2267 rc = dt_declare_record_write(env, mdd_object_child(c),
2273 if (spec->sp_cr_file_secctx_name != NULL) {
2274 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2275 spec->sp_cr_file_secctx_size);
2276 rc = mdo_declare_xattr_set(env, c, buf,
2277 spec->sp_cr_file_secctx_name, 0,
2286 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2287 struct mdd_object *p, struct mdd_object *c,
2288 const struct lu_name *name,
2289 struct lu_attr *attr,
2290 struct thandle *handle,
2291 const struct md_op_spec *spec,
2292 struct linkea_data *ldata,
2293 struct lu_buf *def_acl_buf,
2294 struct lu_buf *acl_buf,
2295 struct dt_allocation_hint *hint)
2299 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2300 def_acl_buf, acl_buf, hint);
2304 if (S_ISDIR(attr->la_mode)) {
2305 rc = mdo_declare_ref_add(env, p, handle);
2310 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2311 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2315 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2317 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2318 name->ln_name, handle);
2322 rc = mdd_declare_links_add(env, c, handle, ldata);
2327 la->la_valid = LA_CTIME | LA_MTIME;
2328 rc = mdo_declare_attr_set(env, p, la, handle);
2332 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2340 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2341 struct lu_attr *la, struct lu_buf *def_acl_buf,
2342 struct lu_buf *acl_buf)
2347 if (S_ISLNK(la->la_mode)) {
2348 acl_buf->lb_len = 0;
2349 def_acl_buf->lb_len = 0;
2353 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2354 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2355 XATTR_NAME_ACL_DEFAULT);
2356 mdd_read_unlock(env, pobj);
2358 /* If there are default ACL, fix mode/ACL by default ACL */
2359 def_acl_buf->lb_len = rc;
2360 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2361 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2362 acl_buf->lb_len = rc;
2363 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2366 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2367 /* If there are no default ACL, fix mode by mask */
2368 struct lu_ucred *uc = lu_ucred(env);
2370 /* The create triggered by MDT internal events, such as
2371 * LFSCK reset, will not contain valid "uc". */
2372 if (unlikely(uc != NULL))
2373 la->la_mode &= ~uc->uc_umask;
2375 acl_buf->lb_len = 0;
2376 def_acl_buf->lb_len = 0;
2383 * Create a metadata object and initialize it, set acl, xattr.
2385 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2386 struct mdd_object *son, struct lu_attr *attr,
2387 struct md_op_spec *spec, struct lu_buf *acl_buf,
2388 struct lu_buf *def_acl_buf,
2389 struct dt_allocation_hint *hint,
2390 struct thandle *handle)
2392 const struct lu_buf *buf;
2395 mdd_write_lock(env, son, MOR_TGT_CHILD);
2396 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2401 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2402 * created in declare phase, they also needs to be added to master
2403 * object as sub-directory entry. So it has to initialize the master
2404 * object, then set dir striped EA.(in mdo_xattr_set) */
2405 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2408 GOTO(err_destroy, rc);
2411 * in case of replay we just set LOVEA provided by the client
2412 * XXX: I think it would be interesting to try "old" way where
2413 * MDT calls this xattr_set(LOV) in a different transaction.
2414 * probably this way we code can be made better.
2417 /* During creation, there are only a few cases we need do xattr_set to
2419 * 1. regular file: see comments above.
2420 * 2. dir: inherit default striping or pool settings from parent.
2421 * 3. create striped directory with provided stripeEA.
2422 * 4. create striped directory because inherit default layout from the
2425 if (spec->no_create ||
2426 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2427 S_ISDIR(attr->la_mode)) {
2428 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2429 spec->u.sp_ea.eadatalen);
2430 rc = mdo_xattr_set(env, son, buf,
2431 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2435 GOTO(err_destroy, rc);
2438 #ifdef CONFIG_FS_POSIX_ACL
2439 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2440 S_ISDIR(attr->la_mode)) {
2441 /* set default acl */
2442 rc = mdo_xattr_set(env, son, def_acl_buf,
2443 XATTR_NAME_ACL_DEFAULT, 0,
2446 GOTO(err_destroy, rc);
2448 /* set its own acl */
2449 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2450 rc = mdo_xattr_set(env, son, acl_buf,
2451 XATTR_NAME_ACL_ACCESS,
2454 GOTO(err_destroy, rc);
2458 if (S_ISLNK(attr->la_mode)) {
2459 struct lu_ucred *uc = lu_ucred_assert(env);
2460 struct dt_object *dt = mdd_object_child(son);
2461 const char *target_name = spec->u.sp_symname;
2462 int sym_len = strlen(target_name);
2465 buf = mdd_buf_get_const(env, target_name, sym_len);
2466 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2468 CFS_CAP_SYS_RESOURCE_MASK);
2473 GOTO(err_initlized, rc = -EFAULT);
2476 if (spec->sp_cr_file_secctx_name != NULL) {
2477 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2478 spec->sp_cr_file_secctx_size);
2479 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2482 GOTO(err_initlized, rc);
2486 if (unlikely(rc != 0)) {
2488 if (S_ISDIR(attr->la_mode)) {
2489 /* Drop the reference, no need to delete "."/"..",
2490 * because the object to be destroied directly. */
2491 rc2 = mdo_ref_del(env, son, handle);
2495 rc2 = mdo_ref_del(env, son, handle);
2499 mdo_destroy(env, son, handle);
2502 mdd_write_unlock(env, son);
2506 static int mdd_index_delete(const struct lu_env *env,
2507 struct mdd_object *mdd_pobj,
2508 struct lu_attr *cattr,
2509 const struct lu_name *lname)
2511 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2512 struct thandle *handle;
2516 handle = mdd_trans_create(env, mdd);
2518 RETURN(PTR_ERR(handle));
2520 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2525 if (S_ISDIR(cattr->la_mode)) {
2526 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2531 /* Since this will only be used in the error handler path,
2532 * Let's set the thandle to be local and not mess the transno */
2533 handle->th_local = 1;
2534 rc = mdd_trans_start(env, mdd, handle);
2538 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2539 S_ISDIR(cattr->la_mode), handle);
2543 rc = mdd_trans_stop(env, mdd, rc, handle);
2549 * Create object and insert it into namespace.
2551 * Two operations have to be performed:
2553 * - an allocation of a new object (->do_create()), and
2554 * - an insertion into a parent index (->dio_insert()).
2556 * Due to locking, operation order is not important, when both are
2557 * successful, *but* error handling cases are quite different:
2559 * - if insertion is done first, and following object creation fails,
2560 * insertion has to be rolled back, but this operation might fail
2561 * also leaving us with dangling index entry.
2563 * - if creation is done first, is has to be undone if insertion fails,
2564 * leaving us with leaked space, which is not good but not fatal.
2566 * It seems that creation-first is simplest solution, but it is sub-optimal
2572 * case, because second mkdir is bound to create object, only to
2573 * destroy it immediately.
2575 * To avoid this follow local file systems that do double lookup:
2577 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2578 * 1. create (mdd_create_object_internal())
2579 * 2. insert (__mdd_index_insert(), lookup again)
2581 * \param[in] pobj parent object
2582 * \param[in] lname name of child being created
2583 * \param[in,out] child child object being created
2584 * \param[in] spec additional create parameters
2585 * \param[in] ma attributes for new child object
2587 * \retval 0 on success
2588 * \retval negative errno on failure
2590 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2591 const struct lu_name *lname, struct md_object *child,
2592 struct md_op_spec *spec, struct md_attr *ma)
2594 struct mdd_thread_info *info = mdd_env_info(env);
2595 struct lu_attr *la = &info->mti_la_for_fix;
2596 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2597 struct mdd_object *son = md2mdd_obj(child);
2598 struct mdd_device *mdd = mdo2mdd(pobj);
2599 struct lu_attr *attr = &ma->ma_attr;
2600 struct thandle *handle;
2601 struct lu_attr *pattr = &info->mti_pattr;
2602 struct lu_buf acl_buf;
2603 struct lu_buf def_acl_buf;
2604 struct linkea_data *ldata = &info->mti_link_data;
2605 const char *name = lname->ln_name;
2606 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2611 rc = mdd_la_get(env, mdd_pobj, pattr);
2615 /* Sanity checks before big job. */
2616 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2620 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2621 GOTO(out_free, rc = -EINPROGRESS);
2623 handle = mdd_trans_create(env, mdd);
2625 GOTO(out_free, rc = PTR_ERR(handle));
2627 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2628 mdd->mdd_dt_conf.ddp_max_ea_size);
2629 acl_buf = info->mti_xattr_buf;
2630 def_acl_buf.lb_buf = info->mti_key;
2631 def_acl_buf.lb_len = sizeof(info->mti_key);
2632 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2636 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2638 memset(ldata, 0, sizeof(*ldata));
2639 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2640 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2643 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2644 &tfid, lname, 1, 0, ldata);
2646 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2647 mdd_object_fid(mdd_pobj),
2648 lname, 1, 0, ldata);
2651 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2652 handle, spec, ldata, &def_acl_buf, &acl_buf,
2657 rc = mdd_trans_start(env, mdd, handle);
2661 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2662 &def_acl_buf, hint, handle);
2666 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2667 mdd_write_lock(env, son, MOR_TGT_CHILD);
2668 son->mod_flags |= VOLATILE_OBJ;
2669 rc = __mdd_orphan_add(env, son, handle);
2670 GOTO(out_volatile, rc);
2672 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2673 attr->la_mode, name, handle);
2675 GOTO(err_created, rc);
2677 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2680 /* update parent directory mtime/ctime */
2682 la->la_valid = LA_CTIME | LA_MTIME;
2683 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2685 GOTO(err_insert, rc);
2691 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2692 rc2 = __mdd_orphan_del(env, son, handle);
2694 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2695 S_ISDIR(attr->la_mode),
2701 mdd_write_lock(env, son, MOR_TGT_CHILD);
2702 if (S_ISDIR(attr->la_mode)) {
2703 /* Drop the reference, no need to delete "."/"..",
2704 * because the object is to be destroyed directly. */
2705 rc2 = mdo_ref_del(env, son, handle);
2707 mdd_write_unlock(env, son);
2712 /* For volatile files drop one link immediately, since there is
2713 * no filename in the namespace, and save any error returned. */
2714 rc2 = mdo_ref_del(env, son, handle);
2716 mdd_write_unlock(env, son);
2717 if (unlikely(rc == 0))
2722 /* Don't destroy the volatile object on success */
2723 if (likely(rc != 0))
2724 mdo_destroy(env, son, handle);
2725 mdd_write_unlock(env, son);
2728 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2729 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2730 rc = mdd_changelog_ns_store(env, mdd,
2731 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2732 S_ISREG(attr->la_mode) ? CL_CREATE :
2733 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2734 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2737 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2739 /* If creation fails, it is most likely due to the remote update
2740 * failure, because local transaction will mostly succeed at
2741 * this stage. There is no easy way to rollback all of previous
2742 * updates, so let's remove the object from namespace, and
2743 * LFSCK should handle the orphan object. */
2744 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2745 mdd_index_delete(env, mdd_pobj, attr, lname);
2749 if (is_vmalloc_addr(ldata->ld_buf))
2750 /* if we vmalloced a large buffer drop it */
2751 lu_buf_free(ldata->ld_buf);
2753 /* The child object shouldn't be cached anymore */
2755 set_bit(LU_OBJECT_HEARD_BANSHEE,
2756 &child->mo_lu.lo_header->loh_flags);
2761 * Get locks on parents in proper order
2762 * RETURN: < 0 - error, rename_order if successful
2770 static int mdd_rename_order(const struct lu_env *env,
2771 struct mdd_device *mdd,
2772 struct mdd_object *src_pobj,
2773 const struct lu_attr *pattr,
2774 struct mdd_object *tgt_pobj)
2776 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2780 if (src_pobj == tgt_pobj)
2781 RETURN(MDD_RN_SAME);
2783 /* compared the parent child relationship of src_p&tgt_p */
2784 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2786 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2789 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2803 /* has not mdd_write{read}_lock on any obj yet. */
2804 static int mdd_rename_sanity_check(const struct lu_env *env,
2805 struct mdd_object *src_pobj,
2806 const struct lu_attr *pattr,
2807 struct mdd_object *tgt_pobj,
2808 const struct lu_attr *tpattr,
2809 struct mdd_object *sobj,
2810 const struct lu_attr *cattr,
2811 struct mdd_object *tobj,
2812 const struct lu_attr *tattr)
2817 /* XXX: when get here, sobj must NOT be NULL,
2818 * the other case has been processed in cld_rename
2819 * before mdd_rename and enable MDS_PERM_BYPASS. */
2823 * If we are using project inheritance, we only allow renames
2824 * into our tree when the project IDs are the same; otherwise
2825 * tree quota mechanism would be circumvented.
2827 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2828 tpattr->la_projid != cattr->la_projid) ||
2829 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2830 (pattr->la_projid != tpattr->la_projid)))
2833 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2837 /* XXX: when get here, "tobj == NULL" means tobj must
2838 * NOT exist (neither on remote MDS, such case has been
2839 * processed in cld_rename before mdd_rename and enable
2841 * So check may_create, but not check may_unlink. */
2843 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2844 (src_pobj != tgt_pobj));
2846 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2847 (src_pobj != tgt_pobj), 1);
2849 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2850 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2855 static int mdd_declare_rename(const struct lu_env *env,
2856 struct mdd_device *mdd,
2857 struct mdd_object *mdd_spobj,
2858 struct mdd_object *mdd_tpobj,
2859 struct mdd_object *mdd_sobj,
2860 struct mdd_object *mdd_tobj,
2861 const struct lu_name *sname,
2862 const struct lu_name *tname,
2864 struct linkea_data *ldata,
2865 struct thandle *handle)
2867 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2870 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2871 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2877 /* name from source dir */
2878 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2882 /* .. from source child */
2883 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2884 /* source child can be directory,
2885 * counted by source dir's nlink */
2886 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2889 if (mdd_spobj != mdd_tpobj) {
2890 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2895 rc = mdo_declare_index_insert(env, mdd_sobj,
2897 S_IFDIR, dotdot, handle);
2902 /* new target child can be directory,
2903 * counted by target dir's nlink */
2904 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2909 la->la_valid = LA_CTIME | LA_MTIME;
2910 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2914 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2918 la->la_valid = LA_CTIME;
2919 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2923 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2928 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2929 mdd_object_type(mdd_sobj),
2930 tname->ln_name, handle);
2934 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2935 /* delete target child in target parent directory */
2936 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2941 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2945 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2946 /* target child can be directory,
2947 * delete "." reference in target child directory */
2948 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2952 /* delete ".." reference in target parent directory */
2953 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2958 la->la_valid = LA_CTIME;
2959 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2963 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2968 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2975 /* src object can be remote that is why we use only fid and type of object */
2976 static int mdd_rename(const struct lu_env *env,
2977 struct md_object *src_pobj, struct md_object *tgt_pobj,
2978 const struct lu_fid *lf, const struct lu_name *lsname,
2979 struct md_object *tobj, const struct lu_name *ltname,
2982 const char *sname = lsname->ln_name;
2983 const char *tname = ltname->ln_name;
2984 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2985 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2986 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2987 struct mdd_device *mdd = mdo2mdd(src_pobj);
2988 struct mdd_object *mdd_sobj = NULL; /* source object */
2989 struct mdd_object *mdd_tobj = NULL;
2990 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
2991 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
2992 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
2993 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
2994 struct thandle *handle;
2995 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
2996 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2997 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
3000 bool tobj_locked = 0;
3001 unsigned cl_flags = 0;
3006 mdd_tobj = md2mdd_obj(tobj);
3008 mdd_sobj = mdd_object_find(env, mdd, lf);
3009 if (IS_ERR(mdd_sobj))
3010 RETURN(PTR_ERR(mdd_sobj));
3012 rc = mdd_la_get(env, mdd_sobj, cattr);
3014 GOTO(out_pending, rc);
3016 rc = mdd_la_get(env, mdd_spobj, pattr);
3018 GOTO(out_pending, rc);
3021 rc = mdd_la_get(env, mdd_tobj, tattr);
3023 GOTO(out_pending, rc);
3024 /* search for an existing archive.
3025 * we should check ahead as the object
3026 * can be destroyed in this transaction */
3027 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3028 cl_flags |= CLF_RENAME_LAST_EXISTS;
3031 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3033 GOTO(out_pending, rc);
3035 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3036 mdd_sobj, cattr, mdd_tobj, tattr);
3038 GOTO(out_pending, rc);
3040 rc = mdd_name_check(mdd, ltname);
3042 GOTO(out_pending, rc);
3044 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3045 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3047 GOTO(out_pending, rc);
3049 handle = mdd_trans_create(env, mdd);
3051 GOTO(out_pending, rc = PTR_ERR(handle));
3053 memset(ldata, 0, sizeof(*ldata));
3054 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3055 lsname, mdd_object_fid(mdd_tpobj), ltname,
3060 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3061 mdd_tobj, lsname, ltname, ma, ldata, handle);
3065 rc = mdd_trans_start(env, mdd, handle);
3069 is_dir = S_ISDIR(cattr->la_mode);
3071 /* Remove source name from source directory */
3072 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3076 /* "mv dir1 dir2" needs "dir1/.." link update */
3077 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3078 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3080 GOTO(fixup_spobj2, rc);
3082 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3085 GOTO(fixup_spobj, rc);
3088 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3089 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3091 /* tname might been renamed to something else */
3092 GOTO(fixup_spobj, rc);
3095 /* Insert new fid with target name into target dir */
3096 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3099 GOTO(fixup_tpobj, rc);
3101 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3102 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3104 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3105 la->la_valid = LA_CTIME;
3106 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3108 GOTO(fixup_tpobj, rc);
3110 /* Update the linkEA for the source object */
3111 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3112 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3113 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3116 /* Old files might not have EA entry */
3117 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3118 lsname, handle, NULL, 0);
3119 mdd_write_unlock(env, mdd_sobj);
3120 /* We don't fail the transaction if the link ea can't be
3121 updated -- fid2path will use alternate lookup method. */
3124 /* Remove old target object
3125 * For tobj is remote case cmm layer has processed
3126 * and set tobj to NULL then. So when tobj is NOT NULL,
3127 * it must be local one.
3129 if (tobj && mdd_object_exists(mdd_tobj)) {
3130 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3132 if (mdd_is_dead_obj(mdd_tobj)) {
3133 /* shld not be dead, something is wrong */
3134 CERROR("tobj is dead, something is wrong\n");
3138 mdo_ref_del(env, mdd_tobj, handle);
3140 /* Remove dot reference. */
3141 if (S_ISDIR(tattr->la_mode))
3142 mdo_ref_del(env, mdd_tobj, handle);
3145 /* fetch updated nlink */
3146 rc = mdd_la_get(env, mdd_tobj, tattr);
3148 CERROR("%s: Failed to get nlink for tobj "
3150 mdd2obd_dev(mdd)->obd_name,
3151 PFID(tpobj_fid), rc);
3152 GOTO(fixup_tpobj, rc);
3155 la->la_valid = LA_CTIME;
3156 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3158 CERROR("%s: Failed to set ctime for tobj "
3160 mdd2obd_dev(mdd)->obd_name,
3161 PFID(tpobj_fid), rc);
3162 GOTO(fixup_tpobj, rc);
3165 /* XXX: this transfer to ma will be removed with LOD/OSP */
3166 ma->ma_attr = *tattr;
3167 ma->ma_valid |= MA_INODE;
3168 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3171 CERROR("%s: Failed to unlink tobj "
3173 mdd2obd_dev(mdd)->obd_name,
3174 PFID(tpobj_fid), rc);
3175 GOTO(fixup_tpobj, rc);
3178 /* fetch updated nlink */
3179 rc = mdd_la_get(env, mdd_tobj, tattr);
3180 if (rc == -ENOENT) {
3181 /* the object got removed, let's
3182 * return the latest known attributes */
3183 tattr->la_nlink = 0;
3185 } else if (rc != 0) {
3186 CERROR("%s: Failed to get nlink for tobj "
3188 mdd2obd_dev(mdd)->obd_name,
3189 PFID(tpobj_fid), rc);
3190 GOTO(fixup_tpobj, rc);
3192 /* XXX: this transfer to ma will be removed with LOD/OSP */
3193 ma->ma_attr = *tattr;
3194 ma->ma_valid |= MA_INODE;
3196 if (tattr->la_nlink == 0)
3197 cl_flags |= CLF_RENAME_LAST;
3199 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3202 la->la_valid = LA_CTIME | LA_MTIME;
3203 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3205 GOTO(fixup_tpobj, rc);
3207 if (mdd_spobj != mdd_tpobj) {
3208 la->la_valid = LA_CTIME | LA_MTIME;
3209 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3211 GOTO(fixup_tpobj, rc);
3218 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3220 CWARN("tp obj fix error %d\n",rc2);
3222 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3223 !mdd_is_dead_obj(mdd_tobj)) {
3225 mdo_ref_add(env, mdd_tobj, handle);
3227 mdo_ref_add(env, mdd_tobj, handle);
3230 rc2 = __mdd_index_insert(env, mdd_tpobj,
3232 mdd_object_type(mdd_tobj),
3235 CWARN("tp obj fix error: rc = %d\n", rc2);
3240 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3241 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3243 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3244 mdd2obd_dev(mdd)->obd_name, rc2);
3247 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3250 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3251 mdd2obd_dev(mdd)->obd_name, rc2);
3256 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3257 mdd_object_type(mdd_sobj), sname,
3260 CWARN("sp obj fix error: rc = %d\n", rc2);
3265 mdd_write_unlock(env, mdd_tobj);
3268 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3269 mdd_tobj, tpobj_fid, lf, spobj_fid,
3270 ltname, lsname, handle);
3273 rc = mdd_trans_stop(env, mdd, rc, handle);
3276 mdd_object_put(env, mdd_sobj);
3281 * During migration once the parent FID has been changed,
3282 * we need update the parent FID in linkea.
3284 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3285 struct mdd_object *parent,
3286 struct mdd_object *newparent,
3287 struct mdd_object *child,
3288 const char *name, int namelen,
3289 struct thandle *handle,
3292 struct mdd_thread_info *info = mdd_env_info(env);
3293 struct linkea_data ldata = { NULL };
3294 struct lu_buf *buf = &info->mti_link_buf;
3300 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3301 if (buf->lb_buf == NULL)
3305 rc = mdd_links_read(env, child, &ldata);
3307 if (rc == -ENOENT || rc == -ENODATA)
3312 LASSERT(ldata.ld_leh != NULL);
3313 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3314 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3315 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3316 struct lu_name lname;
3319 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3322 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3323 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3324 ldata.ld_lee = (struct link_ea_entry *)
3325 ((char *)ldata.ld_lee +
3330 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3331 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3332 lname.ln_namelen, lname.ln_name,
3333 PFID(mdd_object_fid(newparent)));
3334 /* update to the new parent fid */
3335 linkea_entry_pack(ldata.ld_lee, &lname,
3336 mdd_object_fid(newparent));
3338 rc = mdd_declare_links_add(env, child, handle, &ldata);
3340 rc = mdd_links_write(env, child, &ldata, handle);
3346 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3347 struct mdd_object *parent,
3348 struct mdd_object *newparent,
3349 struct mdd_object *child,
3350 const char *name, int namelen,
3351 struct thandle *handle)
3353 return mdd_linkea_update_child_internal(env, parent, newparent,
3355 namelen, handle, true);
3358 static int mdd_linkea_update_child(const struct lu_env *env,
3359 struct mdd_object *parent,
3360 struct mdd_object *newparent,
3361 struct mdd_object *child,
3362 const char *name, int namelen,
3363 struct thandle *handle)
3365 return mdd_linkea_update_child_internal(env, parent, newparent,
3367 namelen, handle, false);
3370 static int mdd_update_linkea_internal(const struct lu_env *env,
3371 struct mdd_object *mdd_pobj,
3372 struct mdd_object *mdd_sobj,
3373 struct mdd_object *mdd_tobj,
3374 const struct lu_name *child_name,
3375 struct linkea_data *ldata,
3376 struct thandle *handle,
3379 struct mdd_thread_info *info = mdd_env_info(env);
3384 LASSERT(ldata->ld_buf != NULL);
3385 LASSERT(ldata->ld_leh != NULL);
3387 /* If it is mulitple links file, we need update the name entry for
3389 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3390 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3391 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3392 struct mdd_object *pobj;
3393 struct lu_name lname;
3396 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3398 pobj = mdd_object_find(env, mdd, &fid);
3400 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3401 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3406 if (!mdd_object_exists(pobj)) {
3407 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3408 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3412 if (pobj == mdd_pobj &&
3413 lname.ln_namelen == child_name->ln_namelen &&
3414 strncmp(lname.ln_name, child_name->ln_name,
3415 lname.ln_namelen) == 0) {
3416 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3417 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3422 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3423 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3424 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3427 /* Remove source name from source directory */
3428 /* Insert new fid with target name into target dir */
3429 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3434 rc = mdo_declare_index_insert(env, pobj,
3435 mdd_object_fid(mdd_tobj),
3436 mdd_object_type(mdd_tobj),
3437 lname.ln_name, handle);
3441 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3445 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3449 char *tmp_name = info->mti_key;
3451 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3452 /* lnamelen is too big(> NAME_MAX + 16),
3453 * something wrong about this linkea, let's
3455 CWARN("%s: the name %.*s is too long under "
3456 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3457 lname.ln_namelen, lname.ln_name,
3462 /* Note: lname might be without \0 at the end, see
3463 * linkea_entry_unpack(), let's add extra \0 by
3465 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3466 lname.ln_namelen, lname.ln_name);
3467 lname.ln_name = tmp_name;
3469 /* Let's check if this linkEA still valid, before
3470 * it might be packed into the RPC buffer. */
3471 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3472 &info->mti_fid, NULL);
3473 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3474 mdd_object_fid(mdd_sobj)))
3475 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3477 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3481 rc = __mdd_index_insert(env, pobj,
3482 mdd_object_fid(mdd_tobj),
3483 mdd_object_type(mdd_tobj),
3488 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3489 rc = mdo_ref_add(env, mdd_tobj, handle);
3490 mdd_write_unlock(env, mdd_tobj);
3494 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3495 mdo_ref_del(env, mdd_sobj, handle);
3496 mdd_write_unlock(env, mdd_sobj);
3499 mdd_object_put(env, pobj);
3503 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3510 static int mdd_migrate_xattrs(const struct lu_env *env,
3511 struct mdd_object *mdd_sobj,
3512 struct mdd_object *mdd_tobj)
3514 struct mdd_thread_info *info = mdd_env_info(env);
3515 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3517 struct thandle *handle;
3523 struct lu_buf list_xbuf;
3526 /* retrieve xattr list from the old object */
3527 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3528 if (list_xsize == -ENODATA)
3534 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3535 if (info->mti_big_buf.lb_buf == NULL)
3538 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3539 list_xbuf.lb_len = list_xsize;
3540 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3545 xname = list_xbuf.lb_buf;
3547 xlen = strnlen(xname, rem - 1) + 1;
3548 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3549 strcmp(XATTR_NAME_LMV, xname) == 0)
3552 /* For directory, if there are default layout, migrate here */
3553 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3554 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3557 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3558 if (xsize == -ENODATA)
3563 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3564 if (info->mti_link_buf.lb_buf == NULL)
3565 GOTO(out, rc = -ENOMEM);
3567 xbuf.lb_len = xsize;
3568 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3569 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3575 handle = mdd_trans_create(env, mdd);
3577 GOTO(out, rc = PTR_ERR(handle));
3579 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3582 GOTO(stop_trans, rc);
3583 /* Note: this transaction is part of migration, and it is not
3584 * the last step of migration, so we set th_local = 1 to avoid
3585 * update last rcvd for this transaction */
3586 handle->th_local = 1;
3587 rc = mdd_trans_start(env, mdd, handle);
3589 GOTO(stop_trans, rc);
3592 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3594 GOTO(stop_trans, rc = 0);
3596 if (unlikely(rc == -ENOSPC &&
3597 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3598 rc = linkea_overflow_shrink(
3599 (struct linkea_data *)(xbuf.lb_buf));
3600 if (likely(rc > 0)) {
3607 GOTO(stop_trans, rc);
3609 rc = mdd_trans_stop(env, mdd, rc, handle);
3614 memmove(xname, xname + xlen, rem);
3620 static int mdd_declare_migrate_create(const struct lu_env *env,
3621 struct mdd_object *mdd_pobj,
3622 struct mdd_object *mdd_sobj,
3623 struct mdd_object *mdd_tobj,
3624 struct md_op_spec *spec,
3626 union lmv_mds_md *mgr_ea,
3627 struct linkea_data *ldata,
3628 struct thandle *handle)
3630 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3631 const struct lu_buf *buf;
3635 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3636 handle, spec, NULL);
3640 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3645 if (S_ISLNK(la->la_mode)) {
3646 const char *target_name = spec->u.sp_symname;
3647 int sym_len = strlen(target_name);
3648 const struct lu_buf *buf;
3650 buf = mdd_buf_get_const(env, target_name, sym_len);
3651 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3655 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3656 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3661 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3662 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3663 spec->u.sp_ea.eadatalen);
3664 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3670 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3671 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3672 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3677 la_flag->la_valid = LA_FLAGS;
3678 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3679 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3684 static int mdd_migrate_create(const struct lu_env *env,
3685 struct mdd_object *mdd_pobj,
3686 struct mdd_object *mdd_sobj,
3687 struct mdd_object *mdd_tobj,
3688 const struct lu_name *lname,
3691 struct mdd_thread_info *info = mdd_env_info(env);
3692 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3693 struct md_op_spec *spec = &info->mti_spec;
3694 struct lu_buf lmm_buf = { NULL };
3695 struct lu_buf link_buf = { NULL };
3696 struct lu_buf mgr_buf;
3697 struct thandle *handle;
3698 struct lmv_mds_md_v1 *mgr_ea;
3699 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3700 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3702 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3706 /* prepare spec for create */
3707 memset(spec, 0, sizeof(*spec));
3708 spec->sp_cr_lookup = 0;
3709 spec->sp_feat = &dt_directory_features;
3710 if (S_ISLNK(la->la_mode)) {
3711 const struct lu_buf *buf;
3713 buf = lu_buf_check_and_alloc(
3714 &mdd_env_info(env)->mti_big_buf,
3717 link_buf.lb_len = la->la_size + 1;
3718 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3719 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3721 rc = rc != 0 ? rc : -EFAULT;
3722 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3723 mdd2obd_dev(mdd)->obd_name,
3724 PFID(mdd_object_fid(mdd_sobj)), rc);
3727 spec->u.sp_symname = link_buf.lb_buf;
3728 } else if (S_ISREG(la->la_mode)) {
3729 /* retrieve lov of the old object */
3730 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3731 if (rc != 0 && rc != -ENODATA)
3733 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3734 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3735 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3736 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3738 } else if (S_ISDIR(la->la_mode)) {
3739 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3740 if (rc == -ENODATA) {
3741 /* ignore the non-linkEA error */
3749 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3750 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3751 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3752 mgr_buf.lb_len = mgr_easize;
3753 mgr_ea = mgr_buf.lb_buf;
3754 memset(mgr_ea, 0, sizeof(*mgr_ea));
3755 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3756 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3757 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3758 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3759 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3760 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3762 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3764 handle = mdd_trans_create(env, mdd);
3766 GOTO(out_free, rc = PTR_ERR(handle));
3768 /* Note: this transaction is part of migration, and it is not
3769 * the last step of migration, so we set th_local = 1 to avoid
3770 * update last rcvd for this transaction */
3771 handle->th_local = 1;
3772 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3773 la, mgr_buf.lb_buf, ldata, handle);
3775 GOTO(stop_trans, rc);
3777 rc = mdd_trans_start(env, mdd, handle);
3779 GOTO(stop_trans, rc);
3781 /* don't set nlink from the original object */
3782 la->la_valid &= ~LA_NLINK;
3784 /* create the target object */
3785 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3788 GOTO(stop_trans, rc);
3790 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3791 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3793 GOTO(stop_trans, rc);
3796 /* Set MIGRATE EA on the source inode, so once the migration needs
3797 * to be re-done during failover, the re-do process can locate the
3798 * target object which is already being created. */
3799 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3801 GOTO(stop_trans, rc);
3803 /* Set immutable flag, so any modification is disabled until
3804 * the migration is done. Once the migration is interrupted,
3805 * if the resume process find the migrating object has both
3806 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3807 * flag and approve the migration */
3808 la_flag->la_valid = LA_FLAGS;
3809 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3810 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3813 rc = mdd_trans_stop(env, mdd, rc, handle);
3815 if (lmm_buf.lb_buf != NULL)
3816 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3820 static int mdd_migrate_entries(const struct lu_env *env,
3821 struct mdd_object *mdd_sobj,
3822 struct mdd_object *mdd_tobj)
3824 struct dt_object *next = mdd_object_child(mdd_sobj);
3825 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3826 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3827 struct thandle *handle;
3829 const struct dt_it_ops *iops;
3831 struct lu_dirent *ent;
3835 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3839 if (!dt_try_as_dir(env, next))
3840 GOTO(out_ent, rc = -ENOTDIR);
3842 * iterate directories
3844 iops = &next->do_index_ops->dio_it;
3845 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3847 GOTO(out_ent, rc = PTR_ERR(it));
3849 rc = iops->load(env, it, 0);
3851 rc = iops->next(env, it);
3855 * At this point and across for-loop:
3857 * rc == 0 -> ok, proceed.
3858 * rc > 0 -> end of directory.
3862 struct mdd_object *child;
3863 char *name = mdd_env_info(env)->mti_key;
3866 bool target_exist = false;
3868 len = iops->key_size(env, it);
3872 result = iops->rec(env, it, (struct dt_rec *)ent,
3873 LUDA_FID | LUDA_TYPE);
3874 if (result == -ESTALE)
3881 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3883 /* Insert new fid with target name into target dir */
3884 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3885 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3886 ent->lde_name[1] == '.'))
3889 child = mdd_object_find(env, mdd, &ent->lde_fid);
3891 GOTO(out, rc = PTR_ERR(child));
3893 /* child may not exist, but lu_object_attr will assert this,
3894 * get type from loh_attr directly */
3895 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3897 mdd_write_lock(env, child, MOR_SRC_CHILD);
3899 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3901 /* Check whether the name has been inserted to the target */
3902 if (dt_try_as_dir(env, dt_tobj)) {
3903 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3905 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3906 (struct dt_key *)name);
3907 if (unlikely(rc == 0))
3908 target_exist = true;
3911 handle = mdd_trans_create(env, mdd);
3913 GOTO(out_put, rc = PTR_ERR(handle));
3915 /* Note: this transaction is part of migration, and it is not
3916 * the last step of migration, so we set th_local = 1 to avoid
3917 * updating last rcvd for this transaction */
3918 handle->th_local = 1;
3919 if (likely(!target_exist)) {
3920 rc = mdo_declare_index_insert(env, mdd_tobj,
3922 child->mod_obj.mo_lu.lo_header->loh_attr,
3928 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3934 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3939 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3943 /* Update .. for child */
3944 rc = mdo_declare_index_delete(env, child, dotdot,
3949 rc = mdo_declare_index_insert(env, child,
3950 mdd_object_fid(mdd_tobj),
3951 S_IFDIR, dotdot, handle);
3956 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3963 rc = mdd_trans_start(env, mdd, handle);
3965 CERROR("%s: transaction start failed: rc = %d\n",
3966 mdd2obd_dev(mdd)->obd_name, rc);
3970 if (likely(!target_exist)) {
3971 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3972 child->mod_obj.mo_lu.lo_header->loh_attr, name,
3978 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3983 rc = __mdd_index_delete_only(env, child, dotdot,
3988 rc = __mdd_index_insert_only(env, child,
3989 mdd_object_fid(mdd_tobj), S_IFDIR,
3995 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
3997 strlen(name), handle);
4000 mdd_write_unlock(env, child);
4001 mdd_object_put(env, child);
4002 rc = mdd_trans_stop(env, mdd, rc, handle);
4006 result = iops->next(env, it);
4007 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
4008 GOTO(out, rc = -EINTR);
4010 if (result == -ESTALE)
4012 } while (result == 0);
4015 iops->fini(env, it);
4017 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
4021 static int mdd_declare_update_linkea(const struct lu_env *env,
4022 struct mdd_object *mdd_pobj,
4023 struct mdd_object *mdd_sobj,
4024 struct mdd_object *mdd_tobj,
4025 const struct lu_name *child_name,
4026 struct linkea_data *ldata,
4027 struct thandle *handle)
4029 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4030 child_name, ldata, handle, 1);
4033 static int mdd_update_linkea(const struct lu_env *env,
4034 struct mdd_object *mdd_pobj,
4035 struct mdd_object *mdd_sobj,
4036 struct mdd_object *mdd_tobj,
4037 const struct lu_name *child_name,
4038 struct linkea_data *ldata,
4039 struct thandle *handle)
4041 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4042 child_name, ldata, handle, 0);
4045 static int mdd_declare_migrate_update_name(const struct lu_env *env,
4046 struct mdd_object *mdd_pobj,
4047 struct mdd_object *mdd_sobj,
4048 struct mdd_object *mdd_tobj,
4049 const struct lu_name *lname,
4051 struct lu_attr *parent_la,
4052 struct linkea_data *ldata,
4053 struct thandle *handle)
4055 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4056 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4059 /* Revert IMMUTABLE flag */
4060 la_flag->la_valid = LA_FLAGS;
4061 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
4062 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
4066 /* delete entry from source dir */
4067 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
4071 if (ldata->ld_buf != NULL) {
4072 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
4073 mdd_tobj, lname, ldata, handle);
4078 if (S_ISREG(mdd_object_type(mdd_sobj))) {
4079 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4084 handle->th_complex = 1;
4085 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
4087 LU_XATTR_REPLACE, handle);
4092 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4093 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4099 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4100 mdd_object_type(mdd_tobj),
4101 lname->ln_name, handle);
4105 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4109 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4110 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4115 /* delete old object */
4116 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4120 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4121 /* delete old object */
4122 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4125 /* set nlink to 0 */
4126 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4131 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4135 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4139 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
4144 static int mdd_migrate_update_name(const struct lu_env *env,
4145 struct mdd_object *mdd_pobj,
4146 struct mdd_object *mdd_sobj,
4147 struct mdd_object *mdd_tobj,
4148 const struct lu_name *lname,
4151 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4152 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4153 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4154 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4155 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4156 struct thandle *handle;
4157 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4158 const char *name = lname->ln_name;
4162 /* update time for parent */
4163 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4164 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4165 p_la->la_valid = LA_CTIME;
4167 rc = mdd_la_get(env, mdd_sobj, so_attr);
4171 ldata->ld_buf = NULL;
4172 rc = mdd_links_read(env, mdd_sobj, ldata);
4173 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4176 handle = mdd_trans_create(env, mdd);
4178 RETURN(PTR_ERR(handle));
4180 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4181 lname, so_attr, p_la, ldata,
4184 /* If the migration can not be fit in one transaction, just
4185 * leave it in the original MDT */
4187 GOTO(stop_trans, rc = 0);
4189 GOTO(stop_trans, rc);
4192 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4193 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4194 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4195 PFID(mdd_object_fid(mdd_tobj)));
4197 rc = mdd_trans_start(env, mdd, handle);
4199 GOTO(stop_trans, rc);
4201 /* Revert IMMUTABLE flag */
4202 la_flag->la_valid = LA_FLAGS;
4203 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4204 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4206 GOTO(stop_trans, rc);
4208 /* Remove source name from source directory */
4209 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4211 GOTO(stop_trans, rc);
4213 if (ldata->ld_buf != NULL) {
4214 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4215 lname, ldata, handle);
4217 GOTO(stop_trans, rc);
4219 /* linkea update might decrease the source object
4220 * nlink, let's get the attr again after ref_del */
4221 rc = mdd_la_get(env, mdd_sobj, so_attr);
4223 GOTO(stop_trans, rc);
4226 if (S_ISREG(so_attr->la_mode)) {
4227 if (so_attr->la_nlink == 1) {
4228 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4230 if (rc != 0 && rc != -ENODATA)
4231 GOTO(stop_trans, rc);
4233 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4235 LU_XATTR_REPLACE, handle);
4237 GOTO(stop_trans, rc);
4241 /* Insert new fid with target name into target dir */
4242 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4243 mdd_object_type(mdd_tobj), name, handle);
4245 GOTO(stop_trans, rc);
4247 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4249 mdd_sobj->mod_flags |= DEAD_OBJ;
4250 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4252 GOTO(out_unlock, rc);
4254 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4256 GOTO(out_unlock, rc);
4258 mdo_ref_del(env, mdd_sobj, handle);
4260 mdo_ref_del(env, mdd_sobj, handle);
4262 /* Get the attr again after ref_del */
4263 rc = mdd_la_get(env, mdd_sobj, so_attr);
4265 GOTO(out_unlock, rc);
4267 ma->ma_attr = *so_attr;
4268 ma->ma_valid |= MA_INODE;
4270 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4272 GOTO(out_unlock, rc);
4274 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4275 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4276 mdo2fid(mdd_pobj), lname, lname, handle);
4278 CWARN("%s: changelog for migrate %s "DFID
4279 "under "DFID" failed: rc = %d\n",
4280 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4281 PFID(mdd_object_fid(mdd_sobj)),
4282 PFID(mdd_object_fid(mdd_pobj)), rc);
4283 /* Sigh, there are no easy way to migrate back the object, so
4284 * let's reset the result to 0 for now XXX */
4288 mdd_write_unlock(env, mdd_sobj);
4291 rc = mdd_trans_stop(env, mdd, rc, handle);
4296 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4297 const struct lu_fid *fid, __u32 *mdt_index)
4299 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4300 struct seq_server_site *ss;
4303 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4305 range->lsr_flags = LU_SEQ_RANGE_MDT;
4306 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4310 *mdt_index = range->lsr_index;
4315 * Check whether we should migrate the file/dir
4317 * < 0 permission check failed or other error.
4318 * = 0 the file can be migrated.
4319 * > 0 the file does not need to be migrated, mostly
4320 * for multiple link file
4322 static int mdd_migrate_sanity_check(const struct lu_env *env,
4323 struct mdd_object *pobj,
4324 const struct lu_attr *pattr,
4325 struct mdd_object *sobj,
4326 struct lu_attr *sattr)
4328 struct mdd_thread_info *info = mdd_env_info(env);
4329 struct linkea_data *ldata = &info->mti_link_data;
4330 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4332 struct lu_buf *mgr_buf;
4338 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4339 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4340 if (mgr_buf->lb_buf == NULL)
4343 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4345 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4347 /* If the object has migrateEA, it means IMMUTE flag
4348 * is being set by previous migration process, so it
4349 * needs to override the IMMUTE flag, otherwise the
4350 * following sanity check will fail */
4351 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4352 LMV_HASH_FLAG_MIGRATION) {
4353 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4355 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4356 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4357 mdd2obd_dev(mdd)->obd_name,
4358 PFID(mdd_object_fid(sobj)));
4362 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4363 sobj, sattr, NULL, NULL);
4367 /* Then it will check if the file should be migrated. If the file
4368 * has mulitple links, we only need migrate the file if all of its
4369 * entries has been migrated to the remote MDT */
4370 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4373 rc = mdd_links_read(env, sobj, ldata);
4375 /* For multiple links files, if there are no linkEA data at all,
4376 * means the file might be created before linkEA is enabled, and
4377 * all of its links should not be migrated yet, otherwise it
4378 * should have some linkEA there */
4379 if (rc == -ENOENT || rc == -ENODATA)
4384 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4385 /* If there are still links locally, then the file will not be
4387 LASSERT(ldata->ld_leh != NULL);
4389 /* If the linkEA is overflow, then means there are some unknown name
4390 * entries under unknown parents, that will prevent the migration. */
4391 if (unlikely(ldata->ld_leh->leh_overflow_time))
4394 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4395 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4396 struct lu_name lname;
4398 __u32 parent_mdt_index;
4400 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4402 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4405 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4409 /* Migrate the object only if none of its parents are on the
4411 if (parent_mdt_index != mdt_index)
4414 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4415 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4416 lname.ln_name, PFID(&fid));
4424 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4425 struct md_object *sobj, const struct lu_name *lname,
4426 struct md_object *tobj, struct md_attr *ma)
4428 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4429 struct mdd_device *mdd = mdo2mdd(pobj);
4430 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4431 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4432 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4433 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4434 bool created = false;
4438 /* If the file will being migrated, it will check whether
4439 * the file is being opened by someone else right now */
4440 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4441 if (mdd_sobj->mod_count > 0) {
4443 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4444 mdd2obd_dev(mdd)->obd_name,
4445 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4446 mdd_sobj->mod_count, -EBUSY);
4447 mdd_read_unlock(env, mdd_sobj);
4448 GOTO(put, rc = -EBUSY);
4450 mdd_read_unlock(env, mdd_sobj);
4452 rc = mdd_la_get(env, mdd_sobj, so_attr);
4456 rc = mdd_la_get(env, mdd_pobj, pattr);
4460 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4467 /* Sigh, it is impossible to finish all of migration in a single
4468 * transaction, for example migrating big directory entries to the
4469 * new MDT, it needs insert all of name entries of children in the
4472 * So migration will be done in multiple steps and transactions.
4474 * 1. create an orphan object on the remote MDT in one transaction.
4475 * 2. migrate extend attributes to the new target file/directory.
4476 * 3. For directory, migrate the entries to the new MDT and update
4477 * linkEA of each children. Because we can not migrate all entries
4478 * in a single transaction, so the migrating directory will become
4479 * a striped directory during migration, so once the process is
4480 * interrupted, the directory is still accessible. (During lookup,
4481 * client will locate the name by searching both original and target
4483 * 4. Finally, update the name/FID to point to the new file/directory
4484 * in a separate transaction.
4487 /* step 1: Check whether the orphan object has been created, and create
4488 * orphan object on the remote MDT if needed */
4489 if (!mdd_object_exists(mdd_tobj)) {
4490 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4497 LASSERT(mdd_object_exists(mdd_tobj));
4498 /* step 2: migrate xattr */
4499 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4503 /* step 3: migrate name entries to the orphan object */
4504 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4505 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4508 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4509 OBD_FAIL_MDS_REINT_NET_REP)))
4512 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4515 LASSERT(mdd_object_exists(mdd_tobj));
4516 /* step 4: update name entry to the new object */
4517 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4522 /* newly created target was not locked, don't cache its attributes */
4524 mdd_invalidate(env, tobj);
4529 const struct md_dir_operations mdd_dir_ops = {
4530 .mdo_is_subdir = mdd_is_subdir,
4531 .mdo_lookup = mdd_lookup,
4532 .mdo_create = mdd_create,
4533 .mdo_rename = mdd_rename,
4534 .mdo_link = mdd_link,
4535 .mdo_unlink = mdd_unlink,
4536 .mdo_create_data = mdd_create_data,
4537 .mdo_migrate = mdd_migrate,