4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_dir.c
34 * Lustre Metadata Server (mdd) routines
36 * Author: Wang Di <wangdi@intel.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <linux/kthread.h>
43 #include <obd_class.h>
44 #include <obd_support.h>
45 #include <lustre_mds.h>
46 #include <lustre_fid.h>
48 #include "mdd_internal.h"
50 static const char dot[] = ".";
51 static const char dotdot[] = "..";
53 static struct lu_name lname_dotdot = {
54 .ln_name = (char *) dotdot,
55 .ln_namelen = sizeof(dotdot) - 1,
59 mdd_name_check(struct mdd_device *m, const struct lu_name *ln)
61 if (!lu_name_is_valid(ln))
63 else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
69 /* Get FID from name and parent */
71 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
72 const struct lu_attr *pattr, const struct lu_name *lname,
73 struct lu_fid* fid, int mask)
75 const char *name = lname->ln_name;
76 const struct dt_key *key = (const struct dt_key *)name;
77 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
78 struct mdd_device *m = mdo2mdd(pobj);
79 struct dt_object *dir = mdd_object_child(mdd_obj);
83 if (unlikely(mdd_is_dead_obj(mdd_obj)))
86 if (!mdd_object_exists(mdd_obj))
89 if (mdd_object_remote(mdd_obj)) {
90 CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
91 mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
94 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
99 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
100 dt_try_as_dir(env, dir)))
101 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
108 int mdd_lookup(const struct lu_env *env,
109 struct md_object *pobj, const struct lu_name *lname,
110 struct lu_fid *fid, struct md_op_spec *spec)
112 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
116 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
120 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
121 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
125 /** Read the link EA into a temp buffer.
126 * Uses the mdd_thread_info::mti_big_buf since it is generally large.
127 * A pointer to the buffer is stored in \a ldata::ld_buf.
131 static int __mdd_links_read(const struct lu_env *env,
132 struct mdd_object *mdd_obj,
133 struct linkea_data *ldata)
137 if (!mdd_object_exists(mdd_obj))
140 /* First try a small buf */
141 LASSERT(env != NULL);
142 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
144 if (ldata->ld_buf->lb_buf == NULL)
147 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
149 /* Buf was too small, figure out what we need. */
150 lu_buf_free(ldata->ld_buf);
151 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
155 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
156 if (ldata->ld_buf->lb_buf == NULL)
158 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
162 lu_buf_free(ldata->ld_buf);
163 ldata->ld_buf = NULL;
167 return linkea_init(ldata);
170 static int mdd_links_read(const struct lu_env *env,
171 struct mdd_object *mdd_obj,
172 struct linkea_data *ldata)
176 rc = __mdd_links_read(env, mdd_obj, ldata);
178 rc = linkea_init(ldata);
183 static int mdd_links_read_with_rec(const struct lu_env *env,
184 struct mdd_object *mdd_obj,
185 struct linkea_data *ldata)
189 rc = __mdd_links_read(env, mdd_obj, ldata);
191 rc = linkea_init_with_rec(ldata);
197 * Get parent FID of the directory
199 * Read parent FID from linkEA, if that fails, then do lookup
200 * dotdot to get the parent FID.
202 * \param[in] env execution environment
203 * \param[in] obj object from which to find the parent FID
204 * \param[in] attr attribute of the object
205 * \param[out] fid fid to get the parent FID
207 * \retval 0 if getting the parent FID succeeds.
208 * \retval negative errno if getting the parent FID fails.
210 static inline int mdd_parent_fid(const struct lu_env *env,
211 struct mdd_object *obj,
212 const struct lu_attr *attr,
215 struct mdd_thread_info *info = mdd_env_info(env);
216 struct linkea_data ldata = { NULL };
217 struct lu_buf *buf = &info->mti_link_buf;
218 struct lu_name lname;
223 LASSERT(S_ISDIR(mdd_object_type(obj)));
225 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
226 if (buf->lb_buf == NULL)
227 GOTO(lookup, rc = 0);
230 rc = mdd_links_read_with_rec(env, obj, &ldata);
234 LASSERT(ldata.ld_leh != NULL);
235 /* Directory should only have 1 parent */
236 if (ldata.ld_leh->leh_reccount > 1)
239 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
241 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
242 if (likely(fid_is_sane(fid)))
245 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
250 * For root fid use special function, which does not compare version component
251 * of fid. Version component is different for root fids on all MDTs.
253 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
255 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
256 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
260 * return 1: if lf is the fid of the ancestor of p1;
263 * return -EREMOTE: if remote object is found, in this
264 * case fid of remote object is saved to @pf;
266 * otherwise: values < 0, errors.
268 static int mdd_is_parent(const struct lu_env *env,
269 struct mdd_device *mdd,
270 struct mdd_object *p1,
271 const struct lu_attr *attr,
272 const struct lu_fid *lf,
275 struct mdd_object *parent = NULL;
280 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
281 pfid = &mdd_env_info(env)->mti_fid;
283 /* Check for root first. */
284 if (mdd_is_root(mdd, mdo2fid(p1)))
288 /* this is done recursively */
289 rc = mdd_parent_fid(env, p1, attr, pfid);
292 if (mdd_is_root(mdd, pfid))
294 if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid))
296 if (lu_fid_eq(pfid, lf))
299 mdd_object_put(env, parent);
301 parent = mdd_object_find(env, mdd, pfid);
303 GOTO(out, rc = PTR_ERR(parent));
305 if (!mdd_object_exists(parent))
306 GOTO(out, rc = -EINVAL);
312 if (parent && !IS_ERR(parent))
313 mdd_object_put(env, parent);
318 * No permission check is needed.
320 * returns 1: if fid is ancestor of @mo;
321 * returns 0: if fid is not an ancestor of @mo;
323 * returns EREMOTE if remote object is found, fid of remote object is saved to
326 * returns < 0: if error
328 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
329 const struct lu_fid *fid, struct lu_fid *sfid)
331 struct mdd_device *mdd = mdo2mdd(mo);
332 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
336 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
339 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
343 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid);
347 } else if (rc == 1) {
348 /* found @fid is parent */
356 * Check that @dir contains no entries except (possibly) dot and dotdot.
361 * -ENOTDIR not a directory object
362 * -ENOTEMPTY not empty
366 static int mdd_dir_is_empty(const struct lu_env *env,
367 struct mdd_object *dir)
370 struct dt_object *obj;
371 const struct dt_it_ops *iops;
375 obj = mdd_object_child(dir);
376 if (!dt_try_as_dir(env, obj))
379 iops = &obj->do_index_ops->dio_it;
380 it = iops->init(env, obj, LUDA_64BITHASH);
382 result = iops->get(env, it, (const struct dt_key *)"");
385 for (result = 0, i = 0; result == 0 && i < 3; ++i)
386 result = iops->next(env, it);
389 else if (result == 1)
391 } else if (result == 0)
393 * Huh? Index contains no zero key?
400 result = PTR_ERR(it);
405 * Determine if the target object can be hard linked, and right now it only
406 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
407 * directory nlink count might exceed the maximum link count(see
408 * osd_object_ref_add), so it only check nlink for non-directories.
410 * \param[in] env thread environment
411 * \param[in] obj object being linked to
412 * \param[in] la attributes of \a obj
414 * \retval 0 if \a obj can be hard linked
415 * \retval negative error if \a obj is a directory or has too
418 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
419 const struct lu_attr *la)
421 struct mdd_device *m = mdd_obj2mdd_dev(obj);
426 /* Subdir count limitation can be broken through
427 * (see osd_object_ref_add), so only check non-directory here. */
428 if (!S_ISDIR(la->la_mode) &&
429 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
436 * Check whether it may create the cobj under the pobj.
438 * \param[in] env execution environment
439 * \param[in] pobj the parent directory
440 * \param[in] pattr the attribute of the parent directory
441 * \param[in] cobj the child to be created
442 * \param[in] check_perm if check WRITE|EXEC permission for parent
444 * \retval = 0 create the child under this dir is allowed
445 * \retval negative errno create the child under this dir is
448 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
449 const struct lu_attr *pattr, struct mdd_object *cobj,
455 if (cobj && mdd_object_exists(cobj))
458 if (mdd_is_dead_obj(pobj))
462 rc = mdd_permission_internal_locked(env, pobj, pattr,
463 MAY_WRITE | MAY_EXEC,
469 * Check whether can unlink from the pobj in the case of "cobj == NULL".
471 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
472 const struct lu_attr *pattr, const struct lu_attr *attr)
477 if (mdd_is_dead_obj(pobj))
480 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
483 rc = mdd_permission_internal_locked(env, pobj, pattr,
484 MAY_WRITE | MAY_EXEC,
489 if (pattr->la_flags & LUSTRE_APPEND_FL)
496 * pobj == NULL is remote ops case, under such case, pobj's
497 * VTX feature has been checked already, no need check again.
499 static inline int mdd_is_sticky(const struct lu_env *env,
500 struct mdd_object *pobj,
501 const struct lu_attr *pattr,
502 struct mdd_object *cobj,
503 const struct lu_attr *cattr)
505 struct lu_ucred *uc = lu_ucred_assert(env);
508 LASSERT(pattr != NULL);
509 if (!(pattr->la_mode & S_ISVTX) ||
510 (pattr->la_uid == uc->uc_fsuid))
514 LASSERT(cattr != NULL);
515 if (cattr->la_uid == uc->uc_fsuid)
518 return !md_capable(uc, CFS_CAP_FOWNER);
521 static int mdd_may_delete_entry(const struct lu_env *env,
522 struct mdd_object *pobj,
523 const struct lu_attr *pattr,
528 LASSERT(pobj != NULL);
529 if (!mdd_object_exists(pobj))
532 if (mdd_is_dead_obj(pobj))
537 rc = mdd_permission_internal_locked(env, pobj, pattr,
538 MAY_WRITE | MAY_EXEC,
544 if (pattr->la_flags & LUSTRE_APPEND_FL)
551 * Check whether it may delete the cobj from the pobj.
554 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
555 const struct lu_attr *tpattr, struct mdd_object *tobj,
556 const struct lu_attr *tattr, const struct lu_attr *cattr,
557 int check_perm, int check_empty)
563 LASSERT(tpattr != NULL);
564 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
572 if (!mdd_object_exists(tobj))
575 if (mdd_is_dead_obj(tobj))
578 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
581 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
584 /* additional check the rename case */
586 if (S_ISDIR(cattr->la_mode)) {
587 struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj);
589 if (!S_ISDIR(tattr->la_mode))
592 if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid))
594 } else if (S_ISDIR(tattr->la_mode))
598 if (S_ISDIR(tattr->la_mode) && check_empty)
599 rc = mdd_dir_is_empty(env, tobj);
605 * Check whether it can create the link file(linked to @src_obj) under
606 * the target directory(@tgt_obj), and src_obj has been locked by
609 * \param[in] env execution environment
610 * \param[in] tgt_obj the target directory
611 * \param[in] tattr attributes of target directory
612 * \param[in] lname the link name
613 * \param[in] src_obj source object for link
614 * \param[in] cattr attributes for source object
616 * \retval = 0 it is allowed to create the link file under tgt_obj
617 * \retval negative error not allowed to create the link file
619 static int mdd_link_sanity_check(const struct lu_env *env,
620 struct mdd_object *tgt_obj,
621 const struct lu_attr *tattr,
622 const struct lu_name *lname,
623 struct mdd_object *src_obj,
624 const struct lu_attr *cattr)
626 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
630 if (!mdd_object_exists(src_obj))
633 if (mdd_is_dead_obj(src_obj))
636 /* Local ops, no lookup before link, check filename length here. */
637 rc = mdd_name_check(m, lname);
641 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
644 if (S_ISDIR(mdd_object_type(src_obj)))
647 LASSERT(src_obj != tgt_obj);
648 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
652 rc = __mdd_may_link(env, src_obj, cattr);
657 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
658 const char *name, struct thandle *handle)
660 struct dt_object *next = mdd_object_child(pobj);
664 if (dt_try_as_dir(env, next))
665 rc = dt_delete(env, next, (struct dt_key *)name, handle);
672 static int __mdd_index_insert_only(const struct lu_env *env,
673 struct mdd_object *pobj,
674 const struct lu_fid *lf, __u32 type,
676 struct thandle *handle)
678 struct dt_object *next = mdd_object_child(pobj);
682 if (dt_try_as_dir(env, next)) {
683 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
684 struct lu_ucred *uc = lu_ucred_check(env);
688 rec->rec_type = type;
689 ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1;
690 rc = dt_insert(env, next, (const struct dt_rec *)rec,
691 (const struct dt_key *)name, handle,
699 /* insert named index, add reference if isdir */
700 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
701 const struct lu_fid *lf, __u32 type,
702 const char *name, struct thandle *handle)
707 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
708 if (rc == 0 && S_ISDIR(type)) {
709 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
710 mdo_ref_add(env, pobj, handle);
711 mdd_write_unlock(env, pobj);
717 /* delete named index, drop reference if isdir */
718 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
719 const char *name, int is_dir,
720 struct thandle *handle)
725 rc = __mdd_index_delete_only(env, pobj, name, handle);
726 if (rc == 0 && is_dir) {
727 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
728 mdo_ref_del(env, pobj, handle);
729 mdd_write_unlock(env, pobj);
735 static int mdd_llog_record_calc_size(const struct lu_env *env,
736 const struct lu_name *tname,
737 const struct lu_name *sname)
739 const struct lu_ucred *uc = lu_ucred(env);
740 enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
741 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
746 if (uc != NULL && uc->uc_jobid[0] != '\0')
749 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
750 changelog_rec_offset(crf, crfe) +
751 (tname != NULL ? tname->ln_namelen : 0) +
752 (sname != NULL ? 1 + sname->ln_namelen : 0));
755 int mdd_declare_changelog_store(const struct lu_env *env,
756 struct mdd_device *mdd,
757 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 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
1060 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
1062 clnid->cr_nid = nid;
1065 /** Store a namespace change changelog record
1066 * If this fails, we must fail the whole transaction; we don't
1067 * want the change to commit without the log entry.
1068 * \param target - mdd_object of change
1069 * \param tpfid - target parent dir/object fid
1070 * \param sfid - source object fid
1071 * \param spfid - source parent fid
1072 * \param tname - target name string
1073 * \param sname - source name string
1074 * \param handle - transaction handle
1076 int mdd_changelog_ns_store(const struct lu_env *env,
1077 struct mdd_device *mdd,
1078 enum changelog_rec_type type,
1079 enum changelog_rec_flags crf,
1080 struct mdd_object *target,
1081 const struct lu_fid *tpfid,
1082 const struct lu_fid *sfid,
1083 const struct lu_fid *spfid,
1084 const struct lu_name *tname,
1085 const struct lu_name *sname,
1086 struct thandle *handle)
1088 const struct lu_ucred *uc = lu_ucred(env);
1089 struct llog_changelog_rec *rec;
1092 __u64 xflags = CLFE_INVALID;
1097 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
1100 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
1103 LASSERT(tpfid != NULL);
1104 LASSERT(tname != NULL);
1105 LASSERT(handle != NULL);
1107 reclen = mdd_llog_record_calc_size(env, tname, sname);
1108 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1109 if (buf->lb_buf == NULL)
1113 crf &= CLF_FLAGMASK;
1114 crf |= CLF_EXTRA_FLAGS;
1117 if (uc->uc_jobid[0] != '\0')
1119 xflags |= CLFE_UIDGID;
1128 rec->cr.cr_flags = crf;
1130 if (crf & CLF_EXTRA_FLAGS) {
1131 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1132 if (xflags & CLFE_UIDGID)
1133 mdd_changelog_rec_extra_uidgid(&rec->cr,
1134 uc->uc_uid, uc->uc_gid);
1135 if (xflags & CLFE_NID)
1136 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1139 rec->cr.cr_type = (__u32)type;
1140 rec->cr.cr_pfid = *tpfid;
1141 rec->cr.cr_namelen = tname->ln_namelen;
1142 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1144 if (crf & CLF_RENAME)
1145 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1147 if (crf & CLF_JOBID)
1148 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1150 if (likely(target != NULL)) {
1151 rec->cr.cr_tfid = *mdo2fid(target);
1152 target->mod_cltime = ktime_get();
1154 fid_zero(&rec->cr.cr_tfid);
1157 rc = mdd_changelog_store(env, mdd, rec, handle);
1159 CERROR("%s: cannot store changelog record: type = %d, "
1160 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1161 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1162 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1169 static int __mdd_links_add(const struct lu_env *env,
1170 struct mdd_object *mdd_obj,
1171 struct linkea_data *ldata,
1172 const struct lu_name *lname,
1173 const struct lu_fid *pfid,
1174 int first, int check)
1178 if (ldata->ld_leh == NULL) {
1179 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1183 rc = linkea_data_new(ldata,
1184 &mdd_env_info(env)->mti_link_buf);
1191 rc = linkea_links_find(ldata, lname, pfid);
1192 if (rc && rc != -ENOENT)
1198 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1199 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1203 linkea_add_buf(ldata, lname, tfid);
1206 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1207 linkea_add_buf(ldata, lname, pfid);
1209 return linkea_add_buf(ldata, lname, pfid);
1212 static int __mdd_links_del(const struct lu_env *env,
1213 struct mdd_object *mdd_obj,
1214 struct linkea_data *ldata,
1215 const struct lu_name *lname,
1216 const struct lu_fid *pfid)
1220 if (ldata->ld_leh == NULL) {
1221 rc = mdd_links_read(env, mdd_obj, ldata);
1226 rc = linkea_links_find(ldata, lname, pfid);
1230 linkea_del_buf(ldata, lname);
1234 static int mdd_linkea_prepare(const struct lu_env *env,
1235 struct mdd_object *mdd_obj,
1236 const struct lu_fid *oldpfid,
1237 const struct lu_name *oldlname,
1238 const struct lu_fid *newpfid,
1239 const struct lu_name *newlname,
1240 int first, int check,
1241 struct linkea_data *ldata)
1246 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1249 LASSERT(oldpfid != NULL || newpfid != NULL);
1251 if (mdd_obj->mod_flags & DEAD_OBJ)
1252 /* Unnecessary to update linkEA for dead object. */
1255 if (oldpfid != NULL) {
1256 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1258 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1261 /* No changes done. */
1266 /* If renaming, add the new record */
1267 if (newpfid != NULL)
1268 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1274 int mdd_links_rename(const struct lu_env *env,
1275 struct mdd_object *mdd_obj,
1276 const struct lu_fid *oldpfid,
1277 const struct lu_name *oldlname,
1278 const struct lu_fid *newpfid,
1279 const struct lu_name *newlname,
1280 struct thandle *handle,
1281 struct linkea_data *ldata,
1282 int first, int check)
1287 if (ldata == NULL) {
1288 ldata = &mdd_env_info(env)->mti_link_data;
1289 memset(ldata, 0, sizeof(*ldata));
1290 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1291 newpfid, newlname, first, check, ldata);
1296 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1297 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1303 if (newlname == NULL)
1304 CERROR("link_ea add failed %d "DFID"\n",
1305 rc, PFID(mdd_object_fid(mdd_obj)));
1306 else if (oldpfid == NULL)
1307 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1308 newlname->ln_namelen, newlname->ln_name, rc,
1309 PFID(mdd_object_fid(mdd_obj)));
1310 else if (newpfid == NULL)
1311 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1312 oldlname->ln_namelen, oldlname->ln_name, rc,
1313 PFID(mdd_object_fid(mdd_obj)));
1315 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1316 "\n", oldlname->ln_namelen, oldlname->ln_name,
1317 newlname->ln_namelen, newlname->ln_name, rc,
1318 PFID(mdd_object_fid(mdd_obj)));
1321 if (is_vmalloc_addr(ldata->ld_buf))
1322 /* if we vmalloced a large buffer drop it */
1323 lu_buf_free(ldata->ld_buf);
1328 static inline int mdd_links_add(const struct lu_env *env,
1329 struct mdd_object *mdd_obj,
1330 const struct lu_fid *pfid,
1331 const struct lu_name *lname,
1332 struct thandle *handle,
1333 struct linkea_data *ldata, int first)
1335 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1336 pfid, lname, handle, ldata, first, 0);
1339 static inline int mdd_links_del(const struct lu_env *env,
1340 struct mdd_object *mdd_obj,
1341 const struct lu_fid *pfid,
1342 const struct lu_name *lname,
1343 struct thandle *handle)
1345 return mdd_links_rename(env, mdd_obj, pfid, lname,
1346 NULL, NULL, handle, NULL, 0, 0);
1349 /** Read the link EA into a temp buffer.
1350 * Uses the name_buf since it is generally large.
1351 * \retval IS_ERR err
1352 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1354 struct lu_buf *mdd_links_get(const struct lu_env *env,
1355 struct mdd_object *mdd_obj)
1357 struct linkea_data ldata = { NULL };
1360 rc = mdd_links_read(env, mdd_obj, &ldata);
1361 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1364 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1365 struct linkea_data *ldata, struct thandle *handle)
1367 const struct lu_buf *buf;
1370 if (ldata == NULL || ldata->ld_buf == NULL ||
1371 ldata->ld_leh == NULL)
1374 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1378 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1379 ldata->ld_leh->leh_len);
1380 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1381 if (unlikely(rc == -ENOSPC)) {
1382 rc = linkea_overflow_shrink(ldata);
1390 static int mdd_declare_links_add(const struct lu_env *env,
1391 struct mdd_object *mdd_obj,
1392 struct thandle *handle,
1393 struct linkea_data *ldata)
1399 if (ldata != NULL && ldata->ld_leh != NULL) {
1400 ea_len = ldata->ld_leh->leh_len;
1401 linkea = ldata->ld_buf->lb_buf;
1403 ea_len = MAX_LINKEA_SIZE;
1407 rc = mdo_declare_xattr_set(env, mdd_obj,
1408 mdd_buf_get_const(env, linkea, ea_len),
1409 XATTR_NAME_LINK, 0, handle);
1414 static inline int mdd_declare_links_del(const struct lu_env *env,
1415 struct mdd_object *c,
1416 struct thandle *handle)
1420 /* For directory, the linkEA will be removed together
1421 * with the object. */
1422 if (!S_ISDIR(mdd_object_type(c)))
1423 rc = mdd_declare_links_add(env, c, handle, NULL);
1428 static int mdd_declare_link(const struct lu_env *env,
1429 struct mdd_device *mdd,
1430 struct mdd_object *p,
1431 struct mdd_object *c,
1432 const struct lu_name *name,
1433 struct thandle *handle,
1435 struct linkea_data *data)
1437 struct lu_fid tfid = *mdo2fid(c);
1440 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1441 tfid.f_oid = cfs_fail_val;
1443 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1444 name->ln_name, handle);
1448 rc = mdo_declare_ref_add(env, c, handle);
1452 la->la_valid = LA_CTIME | LA_MTIME;
1453 rc = mdo_declare_attr_set(env, p, la, handle);
1457 la->la_valid = LA_CTIME;
1458 rc = mdo_declare_attr_set(env, c, la, handle);
1462 rc = mdd_declare_links_add(env, c, handle, data);
1466 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1471 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1472 struct md_object *src_obj, const struct lu_name *lname,
1475 const char *name = lname->ln_name;
1476 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1477 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1478 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1479 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1480 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1481 struct mdd_device *mdd = mdo2mdd(src_obj);
1482 struct thandle *handle;
1483 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1484 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1488 rc = mdd_la_get(env, mdd_sobj, cattr);
1492 rc = mdd_la_get(env, mdd_tobj, tattr);
1497 * If we are using project inheritance, we only allow hard link
1498 * creation in our tree when the project IDs are the same;
1499 * otherwise the tree quota mechanism could be circumvented.
1501 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1502 (tattr->la_projid != cattr->la_projid))
1505 handle = mdd_trans_create(env, mdd);
1507 GOTO(out_pending, rc = PTR_ERR(handle));
1509 memset(ldata, 0, sizeof(*ldata));
1511 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1512 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1514 /* Note: even this function will change ldata, but it comes from
1515 * thread_info, which is completely temporary and only seen in
1516 * this function, so we do not need reset ldata once it fails.*/
1517 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1518 lname, 0, 0, ldata);
1522 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1527 rc = mdd_trans_start(env, mdd, handle);
1531 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1532 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1535 GOTO(out_unlock, rc);
1537 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1538 rc = mdo_ref_add(env, mdd_sobj, handle);
1540 GOTO(out_unlock, rc);
1543 *tfid = *mdo2fid(mdd_sobj);
1544 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1545 tfid->f_oid = cfs_fail_val;
1547 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1548 mdd_object_type(mdd_sobj), name, handle);
1550 mdo_ref_del(env, mdd_sobj, handle);
1551 GOTO(out_unlock, rc);
1554 la->la_valid = LA_CTIME | LA_MTIME;
1555 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1557 GOTO(out_unlock, rc);
1559 la->la_valid = LA_CTIME;
1560 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1562 /* Note: The failure of links_add should not cause the
1563 * link failure, so do not check return value. */
1564 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1565 lname, handle, ldata, 0);
1569 mdd_write_unlock(env, mdd_sobj);
1571 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1572 mdo2fid(mdd_tobj), NULL, NULL,
1573 lname, NULL, handle);
1575 rc = mdd_trans_stop(env, mdd, rc, handle);
1576 if (is_vmalloc_addr(ldata->ld_buf))
1577 /* if we vmalloced a large buffer drop it */
1578 lu_buf_free(ldata->ld_buf);
1583 static int mdd_mark_orphan_object(const struct lu_env *env,
1584 struct mdd_object *obj, struct thandle *handle,
1587 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1590 if (!S_ISDIR(mdd_object_type(obj)))
1593 attr->la_valid = LA_FLAGS;
1594 attr->la_flags = LUSTRE_ORPHAN_FL;
1597 rc = mdo_declare_attr_set(env, obj, attr, handle);
1599 rc = mdo_attr_set(env, obj, attr, handle);
1604 static int mdd_declare_finish_unlink(const struct lu_env *env,
1605 struct mdd_object *obj,
1606 struct thandle *handle)
1610 /* Sigh, we do not know if the unlink object will become orphan in
1611 * declare phase, but fortunately the flags here does not matter
1612 * in current declare implementation */
1613 rc = mdd_mark_orphan_object(env, obj, handle, true);
1617 rc = mdo_declare_destroy(env, obj, handle);
1621 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1625 return mdd_declare_links_del(env, obj, handle);
1628 /* caller should take a lock before calling */
1629 int mdd_finish_unlink(const struct lu_env *env,
1630 struct mdd_object *obj, struct md_attr *ma,
1631 const struct mdd_object *pobj,
1632 const struct lu_name *lname,
1636 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1639 LASSERT(mdd_write_locked(env, obj) != 0);
1641 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1642 /* add new orphan and the object
1643 * will be deleted during mdd_close() */
1644 obj->mod_flags |= DEAD_OBJ;
1645 if (obj->mod_count) {
1646 rc = __mdd_orphan_add(env, obj, th);
1648 CDEBUG(D_HA, "Object "DFID" is inserted into "
1649 "orphan list, open count = %d\n",
1650 PFID(mdd_object_fid(obj)),
1653 CERROR("Object "DFID" fail to be an orphan, "
1654 "open count = %d, maybe cause failed "
1656 PFID(mdd_object_fid(obj)),
1659 /* mark object as an orphan here, not
1660 * before __mdd_orphan_add() as racing
1661 * mdd_la_get() may propagate ORPHAN_OBJ
1662 * causing the asserition */
1663 rc = mdd_mark_orphan_object(env, obj, th, false);
1665 rc = mdo_destroy(env, obj, th);
1667 } else if (!is_dir) {
1668 /* old files may not have link ea; ignore errors */
1669 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1677 * has mdd_write_lock on cobj already, but not on pobj yet
1679 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1680 const struct lu_attr *pattr,
1681 struct mdd_object *cobj,
1682 const struct lu_attr *cattr)
1687 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1692 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1693 struct mdd_object *p, struct mdd_object *c,
1694 const struct lu_name *name, struct md_attr *ma,
1695 struct thandle *handle, int no_name, int is_dir)
1697 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1700 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1701 if (likely(no_name == 0)) {
1702 rc = mdo_declare_index_delete(env, p, name->ln_name,
1709 rc = mdo_declare_ref_del(env, p, handle);
1715 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1716 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1717 la->la_valid = LA_CTIME | LA_MTIME;
1718 rc = mdo_declare_attr_set(env, p, la, handle);
1723 rc = mdo_declare_ref_del(env, c, handle);
1727 rc = mdo_declare_ref_del(env, c, handle);
1731 la->la_valid = LA_CTIME;
1732 rc = mdo_declare_attr_set(env, c, la, handle);
1736 rc = mdd_declare_finish_unlink(env, c, handle);
1740 /* FIXME: need changelog for remove entry */
1741 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1748 * test if a file has an HSM archive
1749 * if HSM attributes are not found in ma update them from
1752 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1753 struct mdd_object *obj,
1758 if (!(ma->ma_valid & MA_HSM)) {
1759 /* no HSM MD provided, read xattr */
1760 struct lu_buf *hsm_buf;
1761 const size_t buflen = sizeof(struct hsm_attrs);
1764 hsm_buf = mdd_buf_get(env, NULL, 0);
1765 lu_buf_alloc(hsm_buf, buflen);
1766 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1767 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1768 lu_buf_free(hsm_buf);
1772 ma->ma_valid |= MA_HSM;
1774 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1780 * Delete name entry and the object.
1781 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1782 * does not exist for this object, and it could only happen during resending
1783 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1784 * is also needed in this case(needed by changelog), so we have to add another
1785 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1786 * the ENOENT failure should be able to be fixed by redo mechanism.
1788 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1789 struct md_object *cobj, const struct lu_name *lname,
1790 struct md_attr *ma, int no_name)
1792 const char *name = lname->ln_name;
1793 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1794 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1795 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1796 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1797 struct mdd_object *mdd_cobj = NULL;
1798 struct mdd_device *mdd = mdo2mdd(pobj);
1799 struct thandle *handle;
1800 int rc, is_dir = 0, cl_flags = 0;
1803 /* cobj == NULL means only delete name entry */
1804 if (likely(cobj != NULL)) {
1805 mdd_cobj = md2mdd_obj(cobj);
1806 if (mdd_object_exists(mdd_cobj) == 0)
1810 rc = mdd_la_get(env, mdd_pobj, pattr);
1814 if (likely(mdd_cobj != NULL)) {
1816 rc = mdd_la_get(env, mdd_cobj, cattr);
1820 is_dir = S_ISDIR(cattr->la_mode);
1821 /* search for an existing archive.
1822 * we should check ahead as the object
1823 * can be destroyed in this transaction */
1824 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1825 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1828 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1832 handle = mdd_trans_create(env, mdd);
1834 RETURN(PTR_ERR(handle));
1836 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1837 lname, ma, handle, no_name, is_dir);
1841 rc = mdd_trans_start(env, mdd, handle);
1845 if (likely(mdd_cobj != NULL))
1846 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1848 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1849 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1854 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1855 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1856 GOTO(cleanup, rc = 0);
1858 if (likely(mdd_cobj != NULL)) {
1859 rc = mdo_ref_del(env, mdd_cobj, handle);
1861 __mdd_index_insert_only(env, mdd_pobj,
1863 mdd_object_type(mdd_cobj),
1870 mdo_ref_del(env, mdd_cobj, handle);
1872 /* fetch updated nlink */
1873 rc = mdd_la_get(env, mdd_cobj, cattr);
1878 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1879 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1881 la->la_valid = LA_CTIME | LA_MTIME;
1882 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1886 /* Enough for only unlink the entry */
1887 if (unlikely(mdd_cobj == NULL))
1890 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1891 /* update ctime of an unlinked file only if it is still
1892 * opened or a link still exists */
1893 la->la_valid = LA_CTIME;
1894 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1899 /* XXX: this transfer to ma will be removed with LOD/OSP */
1900 ma->ma_attr = *cattr;
1901 ma->ma_valid |= MA_INODE;
1902 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1906 /* fetch updated nlink */
1907 rc = mdd_la_get(env, mdd_cobj, cattr);
1908 /* if object is removed then we can't get its attrs,
1910 if (rc == -ENOENT) {
1911 cattr->la_nlink = 0;
1915 if (cattr->la_nlink == 0) {
1916 ma->ma_attr = *cattr;
1917 ma->ma_valid |= MA_INODE;
1922 if (likely(mdd_cobj != NULL))
1923 mdd_write_unlock(env, mdd_cobj);
1926 if (cattr->la_nlink == 0)
1927 cl_flags |= CLF_UNLINK_LAST;
1929 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1931 rc = mdd_changelog_ns_store(env, mdd,
1932 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1933 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1938 rc = mdd_trans_stop(env, mdd, rc, handle);
1944 * The permission has been checked when obj created, no need check again.
1946 static int mdd_cd_sanity_check(const struct lu_env *env,
1947 struct mdd_object *obj)
1952 if (!obj || mdd_is_dead_obj(obj))
1958 static int mdd_create_data(const struct lu_env *env,
1959 struct md_object *pobj,
1960 struct md_object *cobj,
1961 const struct md_op_spec *spec,
1964 struct mdd_device *mdd = mdo2mdd(cobj);
1965 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1966 struct mdd_object *son = md2mdd_obj(cobj);
1967 struct thandle *handle;
1968 const struct lu_buf *buf;
1969 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1970 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1974 rc = mdd_cd_sanity_check(env, son);
1978 if (!md_should_create(spec->sp_cr_flags))
1982 * there are following use cases for this function:
1983 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1984 * striping can be specified or not
1987 rc = mdd_la_get(env, son, attr);
1991 /* calling ->ah_make_hint() is used to transfer information from parent */
1992 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
1994 handle = mdd_trans_create(env, mdd);
1996 GOTO(out_free, rc = PTR_ERR(handle));
1999 * XXX: Setting the lov ea is not locked but setting the attr is locked?
2000 * Should this be fixed?
2002 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
2003 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
2004 spec->sp_cr_flags, spec->no_create);
2006 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2007 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2008 spec->u.sp_ea.eadatalen);
2013 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2014 XATTR_NAME_LOV, 0, handle);
2018 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
2022 rc = mdd_trans_start(env, mdd, handle);
2026 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2032 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
2035 rc = mdd_trans_stop(env, mdd, rc, handle);
2041 static int mdd_declare_object_initialize(const struct lu_env *env,
2042 struct mdd_object *parent,
2043 struct mdd_object *child,
2044 const struct lu_attr *attr,
2045 struct thandle *handle)
2050 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2051 if (!S_ISDIR(attr->la_mode))
2054 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
2059 rc = mdo_declare_ref_add(env, child, handle);
2063 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
2069 static int mdd_object_initialize(const struct lu_env *env,
2070 const struct lu_fid *pfid,
2071 struct mdd_object *child,
2072 struct lu_attr *attr, struct thandle *handle,
2073 const struct md_op_spec *spec)
2078 if (S_ISDIR(attr->la_mode)) {
2079 /* Add "." and ".." for newly created dir */
2080 mdo_ref_add(env, child, handle);
2081 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
2082 S_IFDIR, dot, handle);
2084 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2087 mdo_ref_del(env, child, handle);
2094 * This function checks whether it can create a file/dir under the
2095 * directory(@pobj). The directory(@pobj) is not being locked by
2098 * \param[in] env execution environment
2099 * \param[in] pobj the directory to create files
2100 * \param[in] pattr the attributes of the directory
2101 * \param[in] lname the name of the created file/dir
2102 * \param[in] cattr the attributes of the file/dir
2103 * \param[in] spec create specification
2105 * \retval = 0 it is allowed to create file/dir under
2107 * \retval negative error not allowed to create file/dir
2108 * under the directory
2110 static int mdd_create_sanity_check(const struct lu_env *env,
2111 struct md_object *pobj,
2112 const struct lu_attr *pattr,
2113 const struct lu_name *lname,
2114 struct lu_attr *cattr,
2115 struct md_op_spec *spec)
2117 struct mdd_thread_info *info = mdd_env_info(env);
2118 struct lu_fid *fid = &info->mti_fid;
2119 struct mdd_object *obj = md2mdd_obj(pobj);
2120 struct mdd_device *m = mdo2mdd(pobj);
2121 bool check_perm = true;
2126 if (mdd_is_dead_obj(obj))
2130 * In some cases this lookup is not needed - we know before if name
2131 * exists or not because MDT performs lookup for it.
2132 * name length check is done in lookup.
2134 if (spec->sp_cr_lookup) {
2136 * Check if the name already exist, though it will be checked in
2137 * _index_insert also, for avoiding rolling back if exists
2140 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2141 MAY_WRITE | MAY_EXEC);
2143 RETURN(rc ? : -EEXIST);
2145 /* Permission is already being checked in mdd_lookup */
2149 if (S_ISDIR(cattr->la_mode) &&
2150 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2151 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2152 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2154 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2155 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2156 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2158 CERROR("%s: invalid lmv_user_md: magic = %x, "
2159 "stripe_offset = %d, stripe_count = %u: "
2160 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2161 le32_to_cpu(lum->lum_magic),
2162 (int)le32_to_cpu(lum->lum_stripe_offset),
2163 le32_to_cpu(lum->lum_stripe_count), rc);
2168 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2173 if (pattr->la_mode & S_ISGID) {
2174 cattr->la_gid = pattr->la_gid;
2175 if (S_ISDIR(cattr->la_mode)) {
2176 cattr->la_mode |= S_ISGID;
2177 cattr->la_valid |= LA_MODE;
2181 /* Inherit project ID from parent directory */
2182 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2183 cattr->la_projid = pattr->la_projid;
2184 if (S_ISDIR(cattr->la_mode)) {
2185 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2186 cattr->la_valid |= LA_FLAGS;
2188 cattr->la_valid |= LA_PROJID;
2191 rc = mdd_name_check(m, lname);
2195 switch (cattr->la_mode & S_IFMT) {
2197 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2199 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2200 RETURN(-ENAMETOOLONG);
2219 static int mdd_declare_create_object(const struct lu_env *env,
2220 struct mdd_device *mdd,
2221 struct mdd_object *p, struct mdd_object *c,
2222 struct lu_attr *attr,
2223 struct thandle *handle,
2224 const struct md_op_spec *spec,
2225 struct lu_buf *def_acl_buf,
2226 struct lu_buf *acl_buf,
2227 struct dt_allocation_hint *hint)
2229 const struct lu_buf *buf;
2232 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2237 #ifdef CONFIG_FS_POSIX_ACL
2238 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2239 /* if dir, then can inherit default ACl */
2240 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2241 XATTR_NAME_ACL_DEFAULT,
2247 if (acl_buf->lb_len > 0) {
2248 rc = mdo_declare_attr_set(env, c, attr, handle);
2252 rc = mdo_declare_xattr_set(env, c, acl_buf,
2253 XATTR_NAME_ACL_ACCESS, 0, handle);
2258 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2262 /* replay case, create LOV EA from client data */
2263 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2264 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2265 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2266 spec->u.sp_ea.eadatalen);
2267 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2273 if (S_ISLNK(attr->la_mode)) {
2274 const char *target_name = spec->u.sp_symname;
2275 int sym_len = strlen(target_name);
2276 const struct lu_buf *buf;
2278 buf = mdd_buf_get_const(env, target_name, sym_len);
2279 rc = dt_declare_record_write(env, mdd_object_child(c),
2285 if (spec->sp_cr_file_secctx_name != NULL) {
2286 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2287 spec->sp_cr_file_secctx_size);
2288 rc = mdo_declare_xattr_set(env, c, buf,
2289 spec->sp_cr_file_secctx_name, 0,
2298 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2299 struct mdd_object *p, struct mdd_object *c,
2300 const struct lu_name *name,
2301 struct lu_attr *attr,
2302 struct thandle *handle,
2303 const struct md_op_spec *spec,
2304 struct linkea_data *ldata,
2305 struct lu_buf *def_acl_buf,
2306 struct lu_buf *acl_buf,
2307 struct dt_allocation_hint *hint)
2311 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2312 def_acl_buf, acl_buf, hint);
2316 if (S_ISDIR(attr->la_mode)) {
2317 rc = mdo_declare_ref_add(env, p, handle);
2322 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2323 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2327 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2329 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2330 name->ln_name, handle);
2334 rc = mdd_declare_links_add(env, c, handle, ldata);
2339 la->la_valid = LA_CTIME | LA_MTIME;
2340 rc = mdo_declare_attr_set(env, p, la, handle);
2344 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2352 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2353 struct lu_attr *la, struct lu_buf *def_acl_buf,
2354 struct lu_buf *acl_buf)
2359 if (S_ISLNK(la->la_mode)) {
2360 acl_buf->lb_len = 0;
2361 def_acl_buf->lb_len = 0;
2365 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2366 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2367 XATTR_NAME_ACL_DEFAULT);
2368 mdd_read_unlock(env, pobj);
2370 /* If there are default ACL, fix mode/ACL by default ACL */
2371 def_acl_buf->lb_len = rc;
2372 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2373 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2374 acl_buf->lb_len = rc;
2375 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2378 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2379 /* If there are no default ACL, fix mode by mask */
2380 struct lu_ucred *uc = lu_ucred(env);
2382 /* The create triggered by MDT internal events, such as
2383 * LFSCK reset, will not contain valid "uc". */
2384 if (unlikely(uc != NULL))
2385 la->la_mode &= ~uc->uc_umask;
2387 acl_buf->lb_len = 0;
2388 def_acl_buf->lb_len = 0;
2395 * Create a metadata object and initialize it, set acl, xattr.
2397 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2398 struct mdd_object *son, struct lu_attr *attr,
2399 struct md_op_spec *spec, struct lu_buf *acl_buf,
2400 struct lu_buf *def_acl_buf,
2401 struct dt_allocation_hint *hint,
2402 struct thandle *handle)
2404 const struct lu_buf *buf;
2407 mdd_write_lock(env, son, MOR_TGT_CHILD);
2408 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2413 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2414 * created in declare phase, they also needs to be added to master
2415 * object as sub-directory entry. So it has to initialize the master
2416 * object, then set dir striped EA.(in mdo_xattr_set) */
2417 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2420 GOTO(err_destroy, rc);
2423 * in case of replay we just set LOVEA provided by the client
2424 * XXX: I think it would be interesting to try "old" way where
2425 * MDT calls this xattr_set(LOV) in a different transaction.
2426 * probably this way we code can be made better.
2429 /* During creation, there are only a few cases we need do xattr_set to
2431 * 1. regular file: see comments above.
2432 * 2. dir: inherit default striping or pool settings from parent.
2433 * 3. create striped directory with provided stripeEA.
2434 * 4. create striped directory because inherit default layout from the
2437 if (spec->no_create ||
2438 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2439 S_ISDIR(attr->la_mode)) {
2440 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2441 spec->u.sp_ea.eadatalen);
2442 rc = mdo_xattr_set(env, son, buf,
2443 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2447 GOTO(err_destroy, rc);
2450 #ifdef CONFIG_FS_POSIX_ACL
2451 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2452 S_ISDIR(attr->la_mode)) {
2453 /* set default acl */
2454 rc = mdo_xattr_set(env, son, def_acl_buf,
2455 XATTR_NAME_ACL_DEFAULT, 0,
2458 GOTO(err_destroy, rc);
2460 /* set its own acl */
2461 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2462 rc = mdo_xattr_set(env, son, acl_buf,
2463 XATTR_NAME_ACL_ACCESS,
2466 GOTO(err_destroy, rc);
2470 if (S_ISLNK(attr->la_mode)) {
2471 struct lu_ucred *uc = lu_ucred_assert(env);
2472 struct dt_object *dt = mdd_object_child(son);
2473 const char *target_name = spec->u.sp_symname;
2474 int sym_len = strlen(target_name);
2477 buf = mdd_buf_get_const(env, target_name, sym_len);
2478 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2480 CFS_CAP_SYS_RESOURCE_MASK);
2485 GOTO(err_initlized, rc = -EFAULT);
2488 if (spec->sp_cr_file_secctx_name != NULL) {
2489 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2490 spec->sp_cr_file_secctx_size);
2491 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2494 GOTO(err_initlized, rc);
2498 if (unlikely(rc != 0)) {
2500 if (S_ISDIR(attr->la_mode)) {
2501 /* Drop the reference, no need to delete "."/"..",
2502 * because the object to be destroied directly. */
2503 rc2 = mdo_ref_del(env, son, handle);
2507 rc2 = mdo_ref_del(env, son, handle);
2511 mdo_destroy(env, son, handle);
2514 mdd_write_unlock(env, son);
2518 static int mdd_index_delete(const struct lu_env *env,
2519 struct mdd_object *mdd_pobj,
2520 struct lu_attr *cattr,
2521 const struct lu_name *lname)
2523 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2524 struct thandle *handle;
2528 handle = mdd_trans_create(env, mdd);
2530 RETURN(PTR_ERR(handle));
2532 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2537 if (S_ISDIR(cattr->la_mode)) {
2538 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2543 /* Since this will only be used in the error handler path,
2544 * Let's set the thandle to be local and not mess the transno */
2545 handle->th_local = 1;
2546 rc = mdd_trans_start(env, mdd, handle);
2550 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2551 S_ISDIR(cattr->la_mode), handle);
2555 rc = mdd_trans_stop(env, mdd, rc, handle);
2561 * Create object and insert it into namespace.
2563 * Two operations have to be performed:
2565 * - an allocation of a new object (->do_create()), and
2566 * - an insertion into a parent index (->dio_insert()).
2568 * Due to locking, operation order is not important, when both are
2569 * successful, *but* error handling cases are quite different:
2571 * - if insertion is done first, and following object creation fails,
2572 * insertion has to be rolled back, but this operation might fail
2573 * also leaving us with dangling index entry.
2575 * - if creation is done first, is has to be undone if insertion fails,
2576 * leaving us with leaked space, which is not good but not fatal.
2578 * It seems that creation-first is simplest solution, but it is sub-optimal
2584 * case, because second mkdir is bound to create object, only to
2585 * destroy it immediately.
2587 * To avoid this follow local file systems that do double lookup:
2589 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2590 * 1. create (mdd_create_object_internal())
2591 * 2. insert (__mdd_index_insert(), lookup again)
2593 * \param[in] pobj parent object
2594 * \param[in] lname name of child being created
2595 * \param[in,out] child child object being created
2596 * \param[in] spec additional create parameters
2597 * \param[in] ma attributes for new child object
2599 * \retval 0 on success
2600 * \retval negative errno on failure
2602 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2603 const struct lu_name *lname, struct md_object *child,
2604 struct md_op_spec *spec, struct md_attr *ma)
2606 struct mdd_thread_info *info = mdd_env_info(env);
2607 struct lu_attr *la = &info->mti_la_for_fix;
2608 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2609 struct mdd_object *son = md2mdd_obj(child);
2610 struct mdd_device *mdd = mdo2mdd(pobj);
2611 struct lu_attr *attr = &ma->ma_attr;
2612 struct thandle *handle;
2613 struct lu_attr *pattr = &info->mti_pattr;
2614 struct lu_buf acl_buf;
2615 struct lu_buf def_acl_buf;
2616 struct linkea_data *ldata = &info->mti_link_data;
2617 const char *name = lname->ln_name;
2618 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2623 rc = mdd_la_get(env, mdd_pobj, pattr);
2627 /* Sanity checks before big job. */
2628 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2632 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2633 GOTO(out_free, rc = -EINPROGRESS);
2635 handle = mdd_trans_create(env, mdd);
2637 GOTO(out_free, rc = PTR_ERR(handle));
2639 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2640 mdd->mdd_dt_conf.ddp_max_ea_size);
2641 acl_buf = info->mti_xattr_buf;
2642 def_acl_buf.lb_buf = info->mti_key;
2643 def_acl_buf.lb_len = sizeof(info->mti_key);
2644 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2648 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2650 memset(ldata, 0, sizeof(*ldata));
2651 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2652 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2655 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2656 &tfid, lname, 1, 0, ldata);
2658 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2659 mdd_object_fid(mdd_pobj),
2660 lname, 1, 0, ldata);
2663 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2664 handle, spec, ldata, &def_acl_buf, &acl_buf,
2669 rc = mdd_trans_start(env, mdd, handle);
2673 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2674 &def_acl_buf, hint, handle);
2678 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2679 mdd_write_lock(env, son, MOR_TGT_CHILD);
2680 son->mod_flags |= VOLATILE_OBJ;
2681 rc = __mdd_orphan_add(env, son, handle);
2682 GOTO(out_volatile, rc);
2684 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2685 attr->la_mode, name, handle);
2687 GOTO(err_created, rc);
2689 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2692 /* update parent directory mtime/ctime */
2694 la->la_valid = LA_CTIME | LA_MTIME;
2695 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2697 GOTO(err_insert, rc);
2703 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2704 rc2 = __mdd_orphan_del(env, son, handle);
2706 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2707 S_ISDIR(attr->la_mode),
2713 mdd_write_lock(env, son, MOR_TGT_CHILD);
2714 if (S_ISDIR(attr->la_mode)) {
2715 /* Drop the reference, no need to delete "."/"..",
2716 * because the object is to be destroyed directly. */
2717 rc2 = mdo_ref_del(env, son, handle);
2719 mdd_write_unlock(env, son);
2724 /* For volatile files drop one link immediately, since there is
2725 * no filename in the namespace, and save any error returned. */
2726 rc2 = mdo_ref_del(env, son, handle);
2728 mdd_write_unlock(env, son);
2729 if (unlikely(rc == 0))
2734 /* Don't destroy the volatile object on success */
2735 if (likely(rc != 0))
2736 mdo_destroy(env, son, handle);
2737 mdd_write_unlock(env, son);
2740 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2741 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2742 rc = mdd_changelog_ns_store(env, mdd,
2743 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2744 S_ISREG(attr->la_mode) ? CL_CREATE :
2745 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2746 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2749 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2751 /* If creation fails, it is most likely due to the remote update
2752 * failure, because local transaction will mostly succeed at
2753 * this stage. There is no easy way to rollback all of previous
2754 * updates, so let's remove the object from namespace, and
2755 * LFSCK should handle the orphan object. */
2756 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2757 mdd_index_delete(env, mdd_pobj, attr, lname);
2761 if (is_vmalloc_addr(ldata->ld_buf))
2762 /* if we vmalloced a large buffer drop it */
2763 lu_buf_free(ldata->ld_buf);
2765 /* The child object shouldn't be cached anymore */
2767 set_bit(LU_OBJECT_HEARD_BANSHEE,
2768 &child->mo_lu.lo_header->loh_flags);
2773 * Get locks on parents in proper order
2774 * RETURN: < 0 - error, rename_order if successful
2782 static int mdd_rename_order(const struct lu_env *env,
2783 struct mdd_device *mdd,
2784 struct mdd_object *src_pobj,
2785 const struct lu_attr *pattr,
2786 struct mdd_object *tgt_pobj)
2788 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2792 if (src_pobj == tgt_pobj)
2793 RETURN(MDD_RN_SAME);
2795 /* compared the parent child relationship of src_p&tgt_p */
2796 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2798 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2801 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2815 /* has not mdd_write{read}_lock on any obj yet. */
2816 static int mdd_rename_sanity_check(const struct lu_env *env,
2817 struct mdd_object *src_pobj,
2818 const struct lu_attr *pattr,
2819 struct mdd_object *tgt_pobj,
2820 const struct lu_attr *tpattr,
2821 struct mdd_object *sobj,
2822 const struct lu_attr *cattr,
2823 struct mdd_object *tobj,
2824 const struct lu_attr *tattr)
2829 /* XXX: when get here, sobj must NOT be NULL,
2830 * the other case has been processed in cld_rename
2831 * before mdd_rename and enable MDS_PERM_BYPASS. */
2835 * If we are using project inheritance, we only allow renames
2836 * into our tree when the project IDs are the same; otherwise
2837 * tree quota mechanism would be circumvented.
2839 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2840 tpattr->la_projid != cattr->la_projid) ||
2841 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2842 (pattr->la_projid != tpattr->la_projid)))
2845 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2849 /* XXX: when get here, "tobj == NULL" means tobj must
2850 * NOT exist (neither on remote MDS, such case has been
2851 * processed in cld_rename before mdd_rename and enable
2853 * So check may_create, but not check may_unlink. */
2855 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2856 (src_pobj != tgt_pobj));
2858 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2859 (src_pobj != tgt_pobj), 1);
2861 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2862 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2867 static int mdd_declare_rename(const struct lu_env *env,
2868 struct mdd_device *mdd,
2869 struct mdd_object *mdd_spobj,
2870 struct mdd_object *mdd_tpobj,
2871 struct mdd_object *mdd_sobj,
2872 struct mdd_object *mdd_tobj,
2873 const struct lu_name *sname,
2874 const struct lu_name *tname,
2876 struct linkea_data *ldata,
2877 struct thandle *handle)
2879 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2882 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2883 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2889 /* name from source dir */
2890 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2894 /* .. from source child */
2895 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2896 /* source child can be directory,
2897 * counted by source dir's nlink */
2898 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2901 if (mdd_spobj != mdd_tpobj) {
2902 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2907 rc = mdo_declare_index_insert(env, mdd_sobj,
2909 S_IFDIR, dotdot, handle);
2914 /* new target child can be directory,
2915 * counted by target dir's nlink */
2916 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2921 la->la_valid = LA_CTIME | LA_MTIME;
2922 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2926 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2930 la->la_valid = LA_CTIME;
2931 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2935 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2940 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2941 mdd_object_type(mdd_sobj),
2942 tname->ln_name, handle);
2946 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2947 /* delete target child in target parent directory */
2948 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2953 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2957 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2958 /* target child can be directory,
2959 * delete "." reference in target child directory */
2960 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2964 /* delete ".." reference in target parent directory */
2965 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2970 la->la_valid = LA_CTIME;
2971 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2975 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2980 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2987 /* src object can be remote that is why we use only fid and type of object */
2988 static int mdd_rename(const struct lu_env *env,
2989 struct md_object *src_pobj, struct md_object *tgt_pobj,
2990 const struct lu_fid *lf, const struct lu_name *lsname,
2991 struct md_object *tobj, const struct lu_name *ltname,
2994 const char *sname = lsname->ln_name;
2995 const char *tname = ltname->ln_name;
2996 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2997 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2998 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2999 struct mdd_device *mdd = mdo2mdd(src_pobj);
3000 struct mdd_object *mdd_sobj = NULL; /* source object */
3001 struct mdd_object *mdd_tobj = NULL;
3002 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3003 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3004 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3005 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3006 struct thandle *handle;
3007 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3008 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
3009 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
3012 bool tobj_locked = 0;
3013 unsigned cl_flags = 0;
3018 mdd_tobj = md2mdd_obj(tobj);
3020 mdd_sobj = mdd_object_find(env, mdd, lf);
3021 if (IS_ERR(mdd_sobj))
3022 RETURN(PTR_ERR(mdd_sobj));
3024 rc = mdd_la_get(env, mdd_sobj, cattr);
3026 GOTO(out_pending, rc);
3028 rc = mdd_la_get(env, mdd_spobj, pattr);
3030 GOTO(out_pending, rc);
3033 rc = mdd_la_get(env, mdd_tobj, tattr);
3035 GOTO(out_pending, rc);
3036 /* search for an existing archive.
3037 * we should check ahead as the object
3038 * can be destroyed in this transaction */
3039 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3040 cl_flags |= CLF_RENAME_LAST_EXISTS;
3043 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3045 GOTO(out_pending, rc);
3047 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3048 mdd_sobj, cattr, mdd_tobj, tattr);
3050 GOTO(out_pending, rc);
3052 rc = mdd_name_check(mdd, ltname);
3054 GOTO(out_pending, rc);
3056 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3057 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3059 GOTO(out_pending, rc);
3061 handle = mdd_trans_create(env, mdd);
3063 GOTO(out_pending, rc = PTR_ERR(handle));
3065 memset(ldata, 0, sizeof(*ldata));
3066 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3067 lsname, mdd_object_fid(mdd_tpobj), ltname,
3072 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3073 mdd_tobj, lsname, ltname, ma, ldata, handle);
3077 rc = mdd_trans_start(env, mdd, handle);
3081 is_dir = S_ISDIR(cattr->la_mode);
3083 /* Remove source name from source directory */
3084 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3088 /* "mv dir1 dir2" needs "dir1/.." link update */
3089 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3090 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3092 GOTO(fixup_spobj2, rc);
3094 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3097 GOTO(fixup_spobj, rc);
3100 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3101 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3103 /* tname might been renamed to something else */
3104 GOTO(fixup_spobj, rc);
3107 /* Insert new fid with target name into target dir */
3108 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3111 GOTO(fixup_tpobj, rc);
3113 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3114 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3116 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3117 la->la_valid = LA_CTIME;
3118 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3120 GOTO(fixup_tpobj, rc);
3122 /* Update the linkEA for the source object */
3123 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3124 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3125 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3128 /* Old files might not have EA entry */
3129 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3130 lsname, handle, NULL, 0);
3131 mdd_write_unlock(env, mdd_sobj);
3132 /* We don't fail the transaction if the link ea can't be
3133 updated -- fid2path will use alternate lookup method. */
3136 /* Remove old target object
3137 * For tobj is remote case cmm layer has processed
3138 * and set tobj to NULL then. So when tobj is NOT NULL,
3139 * it must be local one.
3141 if (tobj && mdd_object_exists(mdd_tobj)) {
3142 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3144 if (mdd_is_dead_obj(mdd_tobj)) {
3145 /* shld not be dead, something is wrong */
3146 CERROR("tobj is dead, something is wrong\n");
3150 mdo_ref_del(env, mdd_tobj, handle);
3152 /* Remove dot reference. */
3153 if (S_ISDIR(tattr->la_mode))
3154 mdo_ref_del(env, mdd_tobj, handle);
3157 /* fetch updated nlink */
3158 rc = mdd_la_get(env, mdd_tobj, tattr);
3160 CERROR("%s: Failed to get nlink for tobj "
3162 mdd2obd_dev(mdd)->obd_name,
3163 PFID(tpobj_fid), rc);
3164 GOTO(fixup_tpobj, rc);
3167 la->la_valid = LA_CTIME;
3168 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3170 CERROR("%s: Failed to set ctime for tobj "
3172 mdd2obd_dev(mdd)->obd_name,
3173 PFID(tpobj_fid), rc);
3174 GOTO(fixup_tpobj, rc);
3177 /* XXX: this transfer to ma will be removed with LOD/OSP */
3178 ma->ma_attr = *tattr;
3179 ma->ma_valid |= MA_INODE;
3180 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3183 CERROR("%s: Failed to unlink tobj "
3185 mdd2obd_dev(mdd)->obd_name,
3186 PFID(tpobj_fid), rc);
3187 GOTO(fixup_tpobj, rc);
3190 /* fetch updated nlink */
3191 rc = mdd_la_get(env, mdd_tobj, tattr);
3192 if (rc == -ENOENT) {
3193 /* the object got removed, let's
3194 * return the latest known attributes */
3195 tattr->la_nlink = 0;
3197 } else if (rc != 0) {
3198 CERROR("%s: Failed to get nlink for tobj "
3200 mdd2obd_dev(mdd)->obd_name,
3201 PFID(tpobj_fid), rc);
3202 GOTO(fixup_tpobj, rc);
3204 /* XXX: this transfer to ma will be removed with LOD/OSP */
3205 ma->ma_attr = *tattr;
3206 ma->ma_valid |= MA_INODE;
3208 if (tattr->la_nlink == 0)
3209 cl_flags |= CLF_RENAME_LAST;
3211 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3214 la->la_valid = LA_CTIME | LA_MTIME;
3215 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3217 GOTO(fixup_tpobj, rc);
3219 if (mdd_spobj != mdd_tpobj) {
3220 la->la_valid = LA_CTIME | LA_MTIME;
3221 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3223 GOTO(fixup_tpobj, rc);
3230 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3232 CWARN("tp obj fix error %d\n",rc2);
3234 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3235 !mdd_is_dead_obj(mdd_tobj)) {
3237 mdo_ref_add(env, mdd_tobj, handle);
3239 mdo_ref_add(env, mdd_tobj, handle);
3242 rc2 = __mdd_index_insert(env, mdd_tpobj,
3244 mdd_object_type(mdd_tobj),
3247 CWARN("tp obj fix error: rc = %d\n", rc2);
3252 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3253 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3255 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3256 mdd2obd_dev(mdd)->obd_name, rc2);
3259 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3262 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3263 mdd2obd_dev(mdd)->obd_name, rc2);
3268 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3269 mdd_object_type(mdd_sobj), sname,
3272 CWARN("sp obj fix error: rc = %d\n", rc2);
3277 mdd_write_unlock(env, mdd_tobj);
3280 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3281 mdd_tobj, tpobj_fid, lf, spobj_fid,
3282 ltname, lsname, handle);
3285 rc = mdd_trans_stop(env, mdd, rc, handle);
3288 mdd_object_put(env, mdd_sobj);
3293 * During migration once the parent FID has been changed,
3294 * we need update the parent FID in linkea.
3296 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3297 struct mdd_object *parent,
3298 struct mdd_object *newparent,
3299 struct mdd_object *child,
3300 const char *name, int namelen,
3301 struct thandle *handle,
3304 struct mdd_thread_info *info = mdd_env_info(env);
3305 struct linkea_data ldata = { NULL };
3306 struct lu_buf *buf = &info->mti_link_buf;
3312 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3313 if (buf->lb_buf == NULL)
3317 rc = mdd_links_read(env, child, &ldata);
3319 if (rc == -ENOENT || rc == -ENODATA)
3324 LASSERT(ldata.ld_leh != NULL);
3325 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3326 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3327 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3328 struct lu_name lname;
3331 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3334 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3335 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3336 ldata.ld_lee = (struct link_ea_entry *)
3337 ((char *)ldata.ld_lee +
3342 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3343 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3344 lname.ln_namelen, lname.ln_name,
3345 PFID(mdd_object_fid(newparent)));
3346 /* update to the new parent fid */
3347 linkea_entry_pack(ldata.ld_lee, &lname,
3348 mdd_object_fid(newparent));
3350 rc = mdd_declare_links_add(env, child, handle, &ldata);
3352 rc = mdd_links_write(env, child, &ldata, handle);
3358 static int mdd_linkea_declare_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, true);
3370 static int mdd_linkea_update_child(const struct lu_env *env,
3371 struct mdd_object *parent,
3372 struct mdd_object *newparent,
3373 struct mdd_object *child,
3374 const char *name, int namelen,
3375 struct thandle *handle)
3377 return mdd_linkea_update_child_internal(env, parent, newparent,
3379 namelen, handle, false);
3382 static int mdd_update_linkea_internal(const struct lu_env *env,
3383 struct mdd_object *mdd_pobj,
3384 struct mdd_object *mdd_sobj,
3385 struct mdd_object *mdd_tobj,
3386 const struct lu_name *child_name,
3387 struct linkea_data *ldata,
3388 struct thandle *handle,
3391 struct mdd_thread_info *info = mdd_env_info(env);
3396 LASSERT(ldata->ld_buf != NULL);
3397 LASSERT(ldata->ld_leh != NULL);
3399 /* If it is mulitple links file, we need update the name entry for
3401 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3402 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3403 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3404 struct mdd_object *pobj;
3405 struct lu_name lname;
3408 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3410 pobj = mdd_object_find(env, mdd, &fid);
3412 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3413 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3418 if (!mdd_object_exists(pobj)) {
3419 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3420 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3424 if (pobj == mdd_pobj &&
3425 lname.ln_namelen == child_name->ln_namelen &&
3426 strncmp(lname.ln_name, child_name->ln_name,
3427 lname.ln_namelen) == 0) {
3428 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3429 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3434 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3435 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3436 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3439 /* Remove source name from source directory */
3440 /* Insert new fid with target name into target dir */
3441 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3446 rc = mdo_declare_index_insert(env, pobj,
3447 mdd_object_fid(mdd_tobj),
3448 mdd_object_type(mdd_tobj),
3449 lname.ln_name, handle);
3453 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3457 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3461 char *tmp_name = info->mti_key;
3463 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3464 /* lnamelen is too big(> NAME_MAX + 16),
3465 * something wrong about this linkea, let's
3467 CWARN("%s: the name %.*s is too long under "
3468 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3469 lname.ln_namelen, lname.ln_name,
3474 /* Note: lname might be without \0 at the end, see
3475 * linkea_entry_unpack(), let's add extra \0 by
3477 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3478 lname.ln_namelen, lname.ln_name);
3479 lname.ln_name = tmp_name;
3481 /* Let's check if this linkEA still valid, before
3482 * it might be packed into the RPC buffer. */
3483 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3484 &info->mti_fid, NULL);
3485 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3486 mdd_object_fid(mdd_sobj)))
3487 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3489 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3493 rc = __mdd_index_insert(env, pobj,
3494 mdd_object_fid(mdd_tobj),
3495 mdd_object_type(mdd_tobj),
3500 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3501 rc = mdo_ref_add(env, mdd_tobj, handle);
3502 mdd_write_unlock(env, mdd_tobj);
3506 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3507 mdo_ref_del(env, mdd_sobj, handle);
3508 mdd_write_unlock(env, mdd_sobj);
3511 mdd_object_put(env, pobj);
3515 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3522 static int mdd_migrate_xattrs(const struct lu_env *env,
3523 struct mdd_object *mdd_sobj,
3524 struct mdd_object *mdd_tobj)
3526 struct mdd_thread_info *info = mdd_env_info(env);
3527 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3529 struct thandle *handle;
3535 struct lu_buf list_xbuf;
3538 /* retrieve xattr list from the old object */
3539 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3540 if (list_xsize == -ENODATA)
3546 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3547 if (info->mti_big_buf.lb_buf == NULL)
3550 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3551 list_xbuf.lb_len = list_xsize;
3552 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3557 xname = list_xbuf.lb_buf;
3559 xlen = strnlen(xname, rem - 1) + 1;
3560 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3561 strcmp(XATTR_NAME_LMV, xname) == 0)
3564 /* For directory, if there are default layout, migrate here */
3565 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3566 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3569 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3570 if (xsize == -ENODATA)
3575 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3576 if (info->mti_link_buf.lb_buf == NULL)
3577 GOTO(out, rc = -ENOMEM);
3579 xbuf.lb_len = xsize;
3580 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3581 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3587 handle = mdd_trans_create(env, mdd);
3589 GOTO(out, rc = PTR_ERR(handle));
3591 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3594 GOTO(stop_trans, rc);
3595 /* Note: this transaction is part of migration, and it is not
3596 * the last step of migration, so we set th_local = 1 to avoid
3597 * update last rcvd for this transaction */
3598 handle->th_local = 1;
3599 rc = mdd_trans_start(env, mdd, handle);
3601 GOTO(stop_trans, rc);
3604 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3606 GOTO(stop_trans, rc = 0);
3608 if (unlikely(rc == -ENOSPC &&
3609 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3610 rc = linkea_overflow_shrink(
3611 (struct linkea_data *)(xbuf.lb_buf));
3612 if (likely(rc > 0)) {
3619 GOTO(stop_trans, rc);
3621 rc = mdd_trans_stop(env, mdd, rc, handle);
3626 memmove(xname, xname + xlen, rem);
3632 static int mdd_declare_migrate_create(const struct lu_env *env,
3633 struct mdd_object *mdd_pobj,
3634 struct mdd_object *mdd_sobj,
3635 struct mdd_object *mdd_tobj,
3636 struct md_op_spec *spec,
3638 union lmv_mds_md *mgr_ea,
3639 struct linkea_data *ldata,
3640 struct thandle *handle)
3642 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3643 const struct lu_buf *buf;
3647 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3648 handle, spec, NULL);
3652 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3657 if (S_ISLNK(la->la_mode)) {
3658 const char *target_name = spec->u.sp_symname;
3659 int sym_len = strlen(target_name);
3660 const struct lu_buf *buf;
3662 buf = mdd_buf_get_const(env, target_name, sym_len);
3663 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3667 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3668 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3673 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3674 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3675 spec->u.sp_ea.eadatalen);
3676 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3682 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3683 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3684 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3689 la_flag->la_valid = LA_FLAGS;
3690 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3691 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3696 static int mdd_migrate_create(const struct lu_env *env,
3697 struct mdd_object *mdd_pobj,
3698 struct mdd_object *mdd_sobj,
3699 struct mdd_object *mdd_tobj,
3700 const struct lu_name *lname,
3703 struct mdd_thread_info *info = mdd_env_info(env);
3704 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3705 struct md_op_spec *spec = &info->mti_spec;
3706 struct lu_buf lmm_buf = { NULL };
3707 struct lu_buf link_buf = { NULL };
3708 struct lu_buf mgr_buf;
3709 struct thandle *handle;
3710 struct lmv_mds_md_v1 *mgr_ea;
3711 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3712 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3714 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3718 /* prepare spec for create */
3719 memset(spec, 0, sizeof(*spec));
3720 spec->sp_cr_lookup = 0;
3721 spec->sp_feat = &dt_directory_features;
3722 if (S_ISLNK(la->la_mode)) {
3723 const struct lu_buf *buf;
3725 buf = lu_buf_check_and_alloc(
3726 &mdd_env_info(env)->mti_big_buf,
3729 link_buf.lb_len = la->la_size + 1;
3730 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3731 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3733 rc = rc != 0 ? rc : -EFAULT;
3734 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3735 mdd2obd_dev(mdd)->obd_name,
3736 PFID(mdd_object_fid(mdd_sobj)), rc);
3739 spec->u.sp_symname = link_buf.lb_buf;
3740 } else if (S_ISREG(la->la_mode)) {
3741 /* retrieve lov of the old object */
3742 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3743 if (rc != 0 && rc != -ENODATA)
3745 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3746 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3747 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3748 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3750 } else if (S_ISDIR(la->la_mode)) {
3751 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3752 if (rc == -ENODATA) {
3753 /* ignore the non-linkEA error */
3761 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3762 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3763 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3764 mgr_buf.lb_len = mgr_easize;
3765 mgr_ea = mgr_buf.lb_buf;
3766 memset(mgr_ea, 0, sizeof(*mgr_ea));
3767 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3768 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3769 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3770 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3771 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3772 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3774 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3776 handle = mdd_trans_create(env, mdd);
3778 GOTO(out_free, rc = PTR_ERR(handle));
3780 /* Note: this transaction is part of migration, and it is not
3781 * the last step of migration, so we set th_local = 1 to avoid
3782 * update last rcvd for this transaction */
3783 handle->th_local = 1;
3784 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3785 la, mgr_buf.lb_buf, ldata, handle);
3787 GOTO(stop_trans, rc);
3789 rc = mdd_trans_start(env, mdd, handle);
3791 GOTO(stop_trans, rc);
3793 /* don't set nlink from the original object */
3794 la->la_valid &= ~LA_NLINK;
3796 /* create the target object */
3797 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3800 GOTO(stop_trans, rc);
3802 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3803 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3805 GOTO(stop_trans, rc);
3808 /* Set MIGRATE EA on the source inode, so once the migration needs
3809 * to be re-done during failover, the re-do process can locate the
3810 * target object which is already being created. */
3811 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3813 GOTO(stop_trans, rc);
3815 /* Set immutable flag, so any modification is disabled until
3816 * the migration is done. Once the migration is interrupted,
3817 * if the resume process find the migrating object has both
3818 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3819 * flag and approve the migration */
3820 la_flag->la_valid = LA_FLAGS;
3821 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3822 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3825 rc = mdd_trans_stop(env, mdd, rc, handle);
3827 if (lmm_buf.lb_buf != NULL)
3828 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3832 static int mdd_migrate_entries(const struct lu_env *env,
3833 struct mdd_object *mdd_sobj,
3834 struct mdd_object *mdd_tobj)
3836 struct dt_object *next = mdd_object_child(mdd_sobj);
3837 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3838 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3839 struct thandle *handle;
3841 const struct dt_it_ops *iops;
3843 struct lu_dirent *ent;
3847 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3851 if (!dt_try_as_dir(env, next))
3852 GOTO(out_ent, rc = -ENOTDIR);
3854 * iterate directories
3856 iops = &next->do_index_ops->dio_it;
3857 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3859 GOTO(out_ent, rc = PTR_ERR(it));
3861 rc = iops->load(env, it, 0);
3863 rc = iops->next(env, it);
3867 * At this point and across for-loop:
3869 * rc == 0 -> ok, proceed.
3870 * rc > 0 -> end of directory.
3874 struct mdd_object *child;
3875 char *name = mdd_env_info(env)->mti_key;
3878 bool target_exist = false;
3880 len = iops->key_size(env, it);
3884 result = iops->rec(env, it, (struct dt_rec *)ent,
3885 LUDA_FID | LUDA_TYPE);
3886 if (result == -ESTALE)
3893 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3895 /* Insert new fid with target name into target dir */
3896 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3897 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3898 ent->lde_name[1] == '.'))
3901 child = mdd_object_find(env, mdd, &ent->lde_fid);
3903 GOTO(out, rc = PTR_ERR(child));
3905 /* child may not exist, but lu_object_attr will assert this,
3906 * get type from loh_attr directly */
3907 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3909 mdd_write_lock(env, child, MOR_SRC_CHILD);
3911 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3913 /* Check whether the name has been inserted to the target */
3914 if (dt_try_as_dir(env, dt_tobj)) {
3915 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3917 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3918 (struct dt_key *)name);
3919 if (unlikely(rc == 0))
3920 target_exist = true;
3923 handle = mdd_trans_create(env, mdd);
3925 GOTO(out_put, rc = PTR_ERR(handle));
3927 /* Note: this transaction is part of migration, and it is not
3928 * the last step of migration, so we set th_local = 1 to avoid
3929 * updating last rcvd for this transaction */
3930 handle->th_local = 1;
3931 if (likely(!target_exist)) {
3932 rc = mdo_declare_index_insert(env, mdd_tobj,
3934 child->mod_obj.mo_lu.lo_header->loh_attr,
3940 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3946 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3951 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3955 /* Update .. for child */
3956 rc = mdo_declare_index_delete(env, child, dotdot,
3961 rc = mdo_declare_index_insert(env, child,
3962 mdd_object_fid(mdd_tobj),
3963 S_IFDIR, dotdot, handle);
3968 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3975 rc = mdd_trans_start(env, mdd, handle);
3977 CERROR("%s: transaction start failed: rc = %d\n",
3978 mdd2obd_dev(mdd)->obd_name, rc);
3982 if (likely(!target_exist)) {
3983 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3984 child->mod_obj.mo_lu.lo_header->loh_attr, name,
3990 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
3995 rc = __mdd_index_delete_only(env, child, dotdot,
4000 rc = __mdd_index_insert_only(env, child,
4001 mdd_object_fid(mdd_tobj), S_IFDIR,
4007 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
4009 strlen(name), handle);
4012 mdd_write_unlock(env, child);
4013 mdd_object_put(env, child);
4014 rc = mdd_trans_stop(env, mdd, rc, handle);
4018 result = iops->next(env, it);
4019 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
4020 GOTO(out, rc = -EINTR);
4022 if (result == -ESTALE)
4024 } while (result == 0);
4027 iops->fini(env, it);
4029 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
4033 static int mdd_declare_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, 1);
4045 static int mdd_update_linkea(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 *child_name,
4050 struct linkea_data *ldata,
4051 struct thandle *handle)
4053 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4054 child_name, ldata, handle, 0);
4057 static int mdd_declare_migrate_update_name(const struct lu_env *env,
4058 struct mdd_object *mdd_pobj,
4059 struct mdd_object *mdd_sobj,
4060 struct mdd_object *mdd_tobj,
4061 const struct lu_name *lname,
4063 struct lu_attr *parent_la,
4064 struct linkea_data *ldata,
4065 struct thandle *handle)
4067 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4068 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4071 /* Revert IMMUTABLE flag */
4072 la_flag->la_valid = LA_FLAGS;
4073 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
4074 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
4078 /* delete entry from source dir */
4079 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
4083 if (ldata->ld_buf != NULL) {
4084 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
4085 mdd_tobj, lname, ldata, handle);
4090 if (S_ISREG(mdd_object_type(mdd_sobj))) {
4091 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4096 handle->th_complex = 1;
4097 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
4099 LU_XATTR_REPLACE, handle);
4104 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4105 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4111 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4112 mdd_object_type(mdd_tobj),
4113 lname->ln_name, handle);
4117 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4121 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4122 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4127 /* delete old object */
4128 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4132 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4133 /* delete old object */
4134 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4137 /* set nlink to 0 */
4138 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4143 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4147 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4151 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
4156 static int mdd_migrate_update_name(const struct lu_env *env,
4157 struct mdd_object *mdd_pobj,
4158 struct mdd_object *mdd_sobj,
4159 struct mdd_object *mdd_tobj,
4160 const struct lu_name *lname,
4163 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4164 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4165 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4166 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4167 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4168 struct thandle *handle;
4169 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4170 const char *name = lname->ln_name;
4174 /* update time for parent */
4175 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4176 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4177 p_la->la_valid = LA_CTIME;
4179 rc = mdd_la_get(env, mdd_sobj, so_attr);
4183 ldata->ld_buf = NULL;
4184 rc = mdd_links_read(env, mdd_sobj, ldata);
4185 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4188 handle = mdd_trans_create(env, mdd);
4190 RETURN(PTR_ERR(handle));
4192 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4193 lname, so_attr, p_la, ldata,
4196 /* If the migration can not be fit in one transaction, just
4197 * leave it in the original MDT */
4199 GOTO(stop_trans, rc = 0);
4201 GOTO(stop_trans, rc);
4204 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4205 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4206 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4207 PFID(mdd_object_fid(mdd_tobj)));
4209 rc = mdd_trans_start(env, mdd, handle);
4211 GOTO(stop_trans, rc);
4213 /* Revert IMMUTABLE flag */
4214 la_flag->la_valid = LA_FLAGS;
4215 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4216 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4218 GOTO(stop_trans, rc);
4220 /* Remove source name from source directory */
4221 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4223 GOTO(stop_trans, rc);
4225 if (ldata->ld_buf != NULL) {
4226 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4227 lname, ldata, handle);
4229 GOTO(stop_trans, rc);
4231 /* linkea update might decrease the source object
4232 * nlink, let's get the attr again after ref_del */
4233 rc = mdd_la_get(env, mdd_sobj, so_attr);
4235 GOTO(stop_trans, rc);
4238 if (S_ISREG(so_attr->la_mode)) {
4239 if (so_attr->la_nlink == 1) {
4240 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4242 if (rc != 0 && rc != -ENODATA)
4243 GOTO(stop_trans, rc);
4245 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4247 LU_XATTR_REPLACE, handle);
4249 GOTO(stop_trans, rc);
4253 /* Insert new fid with target name into target dir */
4254 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4255 mdd_object_type(mdd_tobj), name, handle);
4257 GOTO(stop_trans, rc);
4259 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4261 mdd_sobj->mod_flags |= DEAD_OBJ;
4262 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4264 GOTO(out_unlock, rc);
4266 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4268 GOTO(out_unlock, rc);
4270 mdo_ref_del(env, mdd_sobj, handle);
4272 mdo_ref_del(env, mdd_sobj, handle);
4274 /* Get the attr again after ref_del */
4275 rc = mdd_la_get(env, mdd_sobj, so_attr);
4277 GOTO(out_unlock, rc);
4279 ma->ma_attr = *so_attr;
4280 ma->ma_valid |= MA_INODE;
4282 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4284 GOTO(out_unlock, rc);
4286 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4287 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4288 mdo2fid(mdd_pobj), lname, lname, handle);
4290 CWARN("%s: changelog for migrate %s "DFID
4291 "under "DFID" failed: rc = %d\n",
4292 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4293 PFID(mdd_object_fid(mdd_sobj)),
4294 PFID(mdd_object_fid(mdd_pobj)), rc);
4295 /* Sigh, there are no easy way to migrate back the object, so
4296 * let's reset the result to 0 for now XXX */
4300 mdd_write_unlock(env, mdd_sobj);
4303 rc = mdd_trans_stop(env, mdd, rc, handle);
4308 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4309 const struct lu_fid *fid, __u32 *mdt_index)
4311 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4312 struct seq_server_site *ss;
4315 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4317 range->lsr_flags = LU_SEQ_RANGE_MDT;
4318 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4322 *mdt_index = range->lsr_index;
4327 * Check whether we should migrate the file/dir
4329 * < 0 permission check failed or other error.
4330 * = 0 the file can be migrated.
4331 * > 0 the file does not need to be migrated, mostly
4332 * for multiple link file
4334 static int mdd_migrate_sanity_check(const struct lu_env *env,
4335 struct mdd_object *pobj,
4336 const struct lu_attr *pattr,
4337 struct mdd_object *sobj,
4338 struct lu_attr *sattr)
4340 struct mdd_thread_info *info = mdd_env_info(env);
4341 struct linkea_data *ldata = &info->mti_link_data;
4342 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4344 struct lu_buf *mgr_buf;
4350 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4351 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4352 if (mgr_buf->lb_buf == NULL)
4355 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4357 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4359 /* If the object has migrateEA, it means IMMUTE flag
4360 * is being set by previous migration process, so it
4361 * needs to override the IMMUTE flag, otherwise the
4362 * following sanity check will fail */
4363 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4364 LMV_HASH_FLAG_MIGRATION) {
4365 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4367 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4368 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4369 mdd2obd_dev(mdd)->obd_name,
4370 PFID(mdd_object_fid(sobj)));
4374 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4375 sobj, sattr, NULL, NULL);
4379 /* Then it will check if the file should be migrated. If the file
4380 * has mulitple links, we only need migrate the file if all of its
4381 * entries has been migrated to the remote MDT */
4382 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4385 rc = mdd_links_read(env, sobj, ldata);
4387 /* For multiple links files, if there are no linkEA data at all,
4388 * means the file might be created before linkEA is enabled, and
4389 * all of its links should not be migrated yet, otherwise it
4390 * should have some linkEA there */
4391 if (rc == -ENOENT || rc == -ENODATA)
4396 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4397 /* If there are still links locally, then the file will not be
4399 LASSERT(ldata->ld_leh != NULL);
4401 /* If the linkEA is overflow, then means there are some unknown name
4402 * entries under unknown parents, that will prevent the migration. */
4403 if (unlikely(ldata->ld_leh->leh_overflow_time))
4406 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4407 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4408 struct lu_name lname;
4410 __u32 parent_mdt_index;
4412 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4414 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4417 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4421 /* Migrate the object only if none of its parents are on the
4423 if (parent_mdt_index != mdt_index)
4426 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4427 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4428 lname.ln_name, PFID(&fid));
4436 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4437 struct md_object *sobj, const struct lu_name *lname,
4438 struct md_object *tobj, struct md_attr *ma)
4440 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4441 struct mdd_device *mdd = mdo2mdd(pobj);
4442 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4443 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4444 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4445 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4446 bool created = false;
4450 /* If the file will being migrated, it will check whether
4451 * the file is being opened by someone else right now */
4452 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4453 if (mdd_sobj->mod_count > 0) {
4455 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4456 mdd2obd_dev(mdd)->obd_name,
4457 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4458 mdd_sobj->mod_count, -EBUSY);
4459 mdd_read_unlock(env, mdd_sobj);
4460 GOTO(put, rc = -EBUSY);
4462 mdd_read_unlock(env, mdd_sobj);
4464 rc = mdd_la_get(env, mdd_sobj, so_attr);
4468 rc = mdd_la_get(env, mdd_pobj, pattr);
4472 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4479 /* Sigh, it is impossible to finish all of migration in a single
4480 * transaction, for example migrating big directory entries to the
4481 * new MDT, it needs insert all of name entries of children in the
4484 * So migration will be done in multiple steps and transactions.
4486 * 1. create an orphan object on the remote MDT in one transaction.
4487 * 2. migrate extend attributes to the new target file/directory.
4488 * 3. For directory, migrate the entries to the new MDT and update
4489 * linkEA of each children. Because we can not migrate all entries
4490 * in a single transaction, so the migrating directory will become
4491 * a striped directory during migration, so once the process is
4492 * interrupted, the directory is still accessible. (During lookup,
4493 * client will locate the name by searching both original and target
4495 * 4. Finally, update the name/FID to point to the new file/directory
4496 * in a separate transaction.
4499 /* step 1: Check whether the orphan object has been created, and create
4500 * orphan object on the remote MDT if needed */
4501 if (!mdd_object_exists(mdd_tobj)) {
4502 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4509 LASSERT(mdd_object_exists(mdd_tobj));
4510 /* step 2: migrate xattr */
4511 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4515 /* step 3: migrate name entries to the orphan object */
4516 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4517 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4520 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4521 OBD_FAIL_MDS_REINT_NET_REP)))
4524 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4527 LASSERT(mdd_object_exists(mdd_tobj));
4528 /* step 4: update name entry to the new object */
4529 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4534 /* newly created target was not locked, don't cache its attributes */
4536 mdd_invalidate(env, tobj);
4541 const struct md_dir_operations mdd_dir_ops = {
4542 .mdo_is_subdir = mdd_is_subdir,
4543 .mdo_lookup = mdd_lookup,
4544 .mdo_create = mdd_create,
4545 .mdo_rename = mdd_rename,
4546 .mdo_link = mdd_link,
4547 .mdo_unlink = mdd_unlink,
4548 .mdo_create_data = mdd_create_data,
4549 .mdo_migrate = mdd_migrate,