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 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, int flags)
1067 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
1069 omd->cr_openflags = (__u32)flags;
1072 /** Store a namespace change changelog record
1073 * If this fails, we must fail the whole transaction; we don't
1074 * want the change to commit without the log entry.
1075 * \param target - mdd_object of change
1076 * \param tpfid - target parent dir/object fid
1077 * \param sfid - source object fid
1078 * \param spfid - source parent fid
1079 * \param tname - target name string
1080 * \param sname - source name string
1081 * \param handle - transaction handle
1083 int mdd_changelog_ns_store(const struct lu_env *env,
1084 struct mdd_device *mdd,
1085 enum changelog_rec_type type,
1086 enum changelog_rec_flags crf,
1087 struct mdd_object *target,
1088 const struct lu_fid *tpfid,
1089 const struct lu_fid *sfid,
1090 const struct lu_fid *spfid,
1091 const struct lu_name *tname,
1092 const struct lu_name *sname,
1093 struct thandle *handle)
1095 const struct lu_ucred *uc = lu_ucred(env);
1096 struct llog_changelog_rec *rec;
1099 __u64 xflags = CLFE_INVALID;
1104 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
1107 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
1110 LASSERT(tpfid != NULL);
1111 LASSERT(tname != NULL);
1112 LASSERT(handle != NULL);
1114 reclen = mdd_llog_record_calc_size(env, tname, sname);
1115 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
1116 if (buf->lb_buf == NULL)
1120 crf &= CLF_FLAGMASK;
1121 crf |= CLF_EXTRA_FLAGS;
1124 if (uc->uc_jobid[0] != '\0')
1126 xflags |= CLFE_UIDGID;
1135 rec->cr.cr_flags = crf;
1137 if (crf & CLF_EXTRA_FLAGS) {
1138 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1139 if (xflags & CLFE_UIDGID)
1140 mdd_changelog_rec_extra_uidgid(&rec->cr,
1141 uc->uc_uid, uc->uc_gid);
1142 if (xflags & CLFE_NID)
1143 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1146 rec->cr.cr_type = (__u32)type;
1147 rec->cr.cr_pfid = *tpfid;
1148 rec->cr.cr_namelen = tname->ln_namelen;
1149 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1151 if (crf & CLF_RENAME)
1152 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1154 if (crf & CLF_JOBID)
1155 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1157 if (likely(target != NULL)) {
1158 rec->cr.cr_tfid = *mdo2fid(target);
1159 target->mod_cltime = ktime_get();
1161 fid_zero(&rec->cr.cr_tfid);
1164 rc = mdd_changelog_store(env, mdd, rec, handle);
1166 CERROR("%s: cannot store changelog record: type = %d, "
1167 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1168 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1169 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1176 static int __mdd_links_add(const struct lu_env *env,
1177 struct mdd_object *mdd_obj,
1178 struct linkea_data *ldata,
1179 const struct lu_name *lname,
1180 const struct lu_fid *pfid,
1181 int first, int check)
1185 if (ldata->ld_leh == NULL) {
1186 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1190 rc = linkea_data_new(ldata,
1191 &mdd_env_info(env)->mti_link_buf);
1198 rc = linkea_links_find(ldata, lname, pfid);
1199 if (rc && rc != -ENOENT)
1205 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1206 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1210 linkea_add_buf(ldata, lname, tfid);
1213 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1214 linkea_add_buf(ldata, lname, pfid);
1216 return linkea_add_buf(ldata, lname, pfid);
1219 static int __mdd_links_del(const struct lu_env *env,
1220 struct mdd_object *mdd_obj,
1221 struct linkea_data *ldata,
1222 const struct lu_name *lname,
1223 const struct lu_fid *pfid)
1227 if (ldata->ld_leh == NULL) {
1228 rc = mdd_links_read(env, mdd_obj, ldata);
1233 rc = linkea_links_find(ldata, lname, pfid);
1237 linkea_del_buf(ldata, lname);
1241 static int mdd_linkea_prepare(const struct lu_env *env,
1242 struct mdd_object *mdd_obj,
1243 const struct lu_fid *oldpfid,
1244 const struct lu_name *oldlname,
1245 const struct lu_fid *newpfid,
1246 const struct lu_name *newlname,
1247 int first, int check,
1248 struct linkea_data *ldata)
1253 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1256 LASSERT(oldpfid != NULL || newpfid != NULL);
1258 if (mdd_obj->mod_flags & DEAD_OBJ)
1259 /* Unnecessary to update linkEA for dead object. */
1262 if (oldpfid != NULL) {
1263 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1265 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1268 /* No changes done. */
1273 /* If renaming, add the new record */
1274 if (newpfid != NULL)
1275 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1281 int mdd_links_rename(const struct lu_env *env,
1282 struct mdd_object *mdd_obj,
1283 const struct lu_fid *oldpfid,
1284 const struct lu_name *oldlname,
1285 const struct lu_fid *newpfid,
1286 const struct lu_name *newlname,
1287 struct thandle *handle,
1288 struct linkea_data *ldata,
1289 int first, int check)
1294 if (ldata == NULL) {
1295 ldata = &mdd_env_info(env)->mti_link_data;
1296 memset(ldata, 0, sizeof(*ldata));
1297 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1298 newpfid, newlname, first, check, ldata);
1303 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1304 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1310 if (newlname == NULL)
1311 CERROR("link_ea add failed %d "DFID"\n",
1312 rc, PFID(mdd_object_fid(mdd_obj)));
1313 else if (oldpfid == NULL)
1314 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1315 newlname->ln_namelen, newlname->ln_name, rc,
1316 PFID(mdd_object_fid(mdd_obj)));
1317 else if (newpfid == NULL)
1318 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1319 oldlname->ln_namelen, oldlname->ln_name, rc,
1320 PFID(mdd_object_fid(mdd_obj)));
1322 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1323 "\n", oldlname->ln_namelen, oldlname->ln_name,
1324 newlname->ln_namelen, newlname->ln_name, rc,
1325 PFID(mdd_object_fid(mdd_obj)));
1328 if (is_vmalloc_addr(ldata->ld_buf))
1329 /* if we vmalloced a large buffer drop it */
1330 lu_buf_free(ldata->ld_buf);
1335 static inline int mdd_links_add(const struct lu_env *env,
1336 struct mdd_object *mdd_obj,
1337 const struct lu_fid *pfid,
1338 const struct lu_name *lname,
1339 struct thandle *handle,
1340 struct linkea_data *ldata, int first)
1342 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1343 pfid, lname, handle, ldata, first, 0);
1346 static inline int mdd_links_del(const struct lu_env *env,
1347 struct mdd_object *mdd_obj,
1348 const struct lu_fid *pfid,
1349 const struct lu_name *lname,
1350 struct thandle *handle)
1352 return mdd_links_rename(env, mdd_obj, pfid, lname,
1353 NULL, NULL, handle, NULL, 0, 0);
1356 /** Read the link EA into a temp buffer.
1357 * Uses the name_buf since it is generally large.
1358 * \retval IS_ERR err
1359 * \retval ptr to \a lu_buf (always \a mti_big_buf)
1361 struct lu_buf *mdd_links_get(const struct lu_env *env,
1362 struct mdd_object *mdd_obj)
1364 struct linkea_data ldata = { NULL };
1367 rc = mdd_links_read(env, mdd_obj, &ldata);
1368 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1371 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1372 struct linkea_data *ldata, struct thandle *handle)
1374 const struct lu_buf *buf;
1377 if (ldata == NULL || ldata->ld_buf == NULL ||
1378 ldata->ld_leh == NULL)
1381 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1385 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1386 ldata->ld_leh->leh_len);
1387 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1388 if (unlikely(rc == -ENOSPC)) {
1389 rc = linkea_overflow_shrink(ldata);
1397 static int mdd_declare_links_add(const struct lu_env *env,
1398 struct mdd_object *mdd_obj,
1399 struct thandle *handle,
1400 struct linkea_data *ldata)
1406 if (ldata != NULL && ldata->ld_leh != NULL) {
1407 ea_len = ldata->ld_leh->leh_len;
1408 linkea = ldata->ld_buf->lb_buf;
1410 ea_len = MAX_LINKEA_SIZE;
1414 rc = mdo_declare_xattr_set(env, mdd_obj,
1415 mdd_buf_get_const(env, linkea, ea_len),
1416 XATTR_NAME_LINK, 0, handle);
1421 static inline int mdd_declare_links_del(const struct lu_env *env,
1422 struct mdd_object *c,
1423 struct thandle *handle)
1427 /* For directory, the linkEA will be removed together
1428 * with the object. */
1429 if (!S_ISDIR(mdd_object_type(c)))
1430 rc = mdd_declare_links_add(env, c, handle, NULL);
1435 static int mdd_declare_link(const struct lu_env *env,
1436 struct mdd_device *mdd,
1437 struct mdd_object *p,
1438 struct mdd_object *c,
1439 const struct lu_name *name,
1440 struct thandle *handle,
1442 struct linkea_data *data)
1444 struct lu_fid tfid = *mdo2fid(c);
1447 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1448 tfid.f_oid = cfs_fail_val;
1450 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1451 name->ln_name, handle);
1455 rc = mdo_declare_ref_add(env, c, handle);
1459 la->la_valid = LA_CTIME | LA_MTIME;
1460 rc = mdo_declare_attr_set(env, p, la, handle);
1464 la->la_valid = LA_CTIME;
1465 rc = mdo_declare_attr_set(env, c, la, handle);
1469 rc = mdd_declare_links_add(env, c, handle, data);
1473 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1478 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1479 struct md_object *src_obj, const struct lu_name *lname,
1482 const char *name = lname->ln_name;
1483 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1484 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1485 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1486 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1487 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1488 struct mdd_device *mdd = mdo2mdd(src_obj);
1489 struct thandle *handle;
1490 struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
1491 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
1495 rc = mdd_la_get(env, mdd_sobj, cattr);
1499 rc = mdd_la_get(env, mdd_tobj, tattr);
1504 * If we are using project inheritance, we only allow hard link
1505 * creation in our tree when the project IDs are the same;
1506 * otherwise the tree quota mechanism could be circumvented.
1508 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1509 (tattr->la_projid != cattr->la_projid))
1512 handle = mdd_trans_create(env, mdd);
1514 GOTO(out_pending, rc = PTR_ERR(handle));
1516 memset(ldata, 0, sizeof(*ldata));
1518 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1519 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1521 /* Note: even this function will change ldata, but it comes from
1522 * thread_info, which is completely temporary and only seen in
1523 * this function, so we do not need reset ldata once it fails.*/
1524 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
1525 lname, 0, 0, ldata);
1529 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1534 rc = mdd_trans_start(env, mdd, handle);
1538 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
1539 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1542 GOTO(out_unlock, rc);
1544 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1545 rc = mdo_ref_add(env, mdd_sobj, handle);
1547 GOTO(out_unlock, rc);
1550 *tfid = *mdo2fid(mdd_sobj);
1551 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1552 tfid->f_oid = cfs_fail_val;
1554 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1555 mdd_object_type(mdd_sobj), name, handle);
1557 mdo_ref_del(env, mdd_sobj, handle);
1558 GOTO(out_unlock, rc);
1561 la->la_valid = LA_CTIME | LA_MTIME;
1562 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1564 GOTO(out_unlock, rc);
1566 la->la_valid = LA_CTIME;
1567 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1569 /* Note: The failure of links_add should not cause the
1570 * link failure, so do not check return value. */
1571 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
1572 lname, handle, ldata, 0);
1576 mdd_write_unlock(env, mdd_sobj);
1578 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1579 mdo2fid(mdd_tobj), NULL, NULL,
1580 lname, NULL, handle);
1582 rc = mdd_trans_stop(env, mdd, rc, handle);
1583 if (is_vmalloc_addr(ldata->ld_buf))
1584 /* if we vmalloced a large buffer drop it */
1585 lu_buf_free(ldata->ld_buf);
1590 static int mdd_mark_orphan_object(const struct lu_env *env,
1591 struct mdd_object *obj, struct thandle *handle,
1594 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1597 if (!S_ISDIR(mdd_object_type(obj)))
1600 attr->la_valid = LA_FLAGS;
1601 attr->la_flags = LUSTRE_ORPHAN_FL;
1604 rc = mdo_declare_attr_set(env, obj, attr, handle);
1606 rc = mdo_attr_set(env, obj, attr, handle);
1611 static int mdd_declare_finish_unlink(const struct lu_env *env,
1612 struct mdd_object *obj,
1613 struct thandle *handle)
1617 /* Sigh, we do not know if the unlink object will become orphan in
1618 * declare phase, but fortunately the flags here does not matter
1619 * in current declare implementation */
1620 rc = mdd_mark_orphan_object(env, obj, handle, true);
1624 rc = mdo_declare_destroy(env, obj, handle);
1628 rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
1632 return mdd_declare_links_del(env, obj, handle);
1635 /* caller should take a lock before calling */
1636 int mdd_finish_unlink(const struct lu_env *env,
1637 struct mdd_object *obj, struct md_attr *ma,
1638 const struct mdd_object *pobj,
1639 const struct lu_name *lname,
1643 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1646 LASSERT(mdd_write_locked(env, obj) != 0);
1648 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1649 /* add new orphan and the object
1650 * will be deleted during mdd_close() */
1651 obj->mod_flags |= DEAD_OBJ;
1652 if (obj->mod_count) {
1653 rc = __mdd_orphan_add(env, obj, th);
1655 CDEBUG(D_HA, "Object "DFID" is inserted into "
1656 "orphan list, open count = %d\n",
1657 PFID(mdd_object_fid(obj)),
1660 CERROR("Object "DFID" fail to be an orphan, "
1661 "open count = %d, maybe cause failed "
1663 PFID(mdd_object_fid(obj)),
1666 /* mark object as an orphan here, not
1667 * before __mdd_orphan_add() as racing
1668 * mdd_la_get() may propagate ORPHAN_OBJ
1669 * causing the asserition */
1670 rc = mdd_mark_orphan_object(env, obj, th, false);
1672 rc = mdo_destroy(env, obj, th);
1674 } else if (!is_dir) {
1675 /* old files may not have link ea; ignore errors */
1676 mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
1684 * has mdd_write_lock on cobj already, but not on pobj yet
1686 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1687 const struct lu_attr *pattr,
1688 struct mdd_object *cobj,
1689 const struct lu_attr *cattr)
1694 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1699 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1700 struct mdd_object *p, struct mdd_object *c,
1701 const struct lu_name *name, struct md_attr *ma,
1702 struct thandle *handle, int no_name, int is_dir)
1704 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1707 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1708 if (likely(no_name == 0)) {
1709 rc = mdo_declare_index_delete(env, p, name->ln_name,
1716 rc = mdo_declare_ref_del(env, p, handle);
1722 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1723 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1724 la->la_valid = LA_CTIME | LA_MTIME;
1725 rc = mdo_declare_attr_set(env, p, la, handle);
1730 rc = mdo_declare_ref_del(env, c, handle);
1734 rc = mdo_declare_ref_del(env, c, handle);
1738 la->la_valid = LA_CTIME;
1739 rc = mdo_declare_attr_set(env, c, la, handle);
1743 rc = mdd_declare_finish_unlink(env, c, handle);
1747 /* FIXME: need changelog for remove entry */
1748 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
1755 * test if a file has an HSM archive
1756 * if HSM attributes are not found in ma update them from
1759 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1760 struct mdd_object *obj,
1765 if (!(ma->ma_valid & MA_HSM)) {
1766 /* no HSM MD provided, read xattr */
1767 struct lu_buf *hsm_buf;
1768 const size_t buflen = sizeof(struct hsm_attrs);
1771 hsm_buf = mdd_buf_get(env, NULL, 0);
1772 lu_buf_alloc(hsm_buf, buflen);
1773 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1774 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1775 lu_buf_free(hsm_buf);
1779 ma->ma_valid |= MA_HSM;
1781 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1787 * Delete name entry and the object.
1788 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1789 * does not exist for this object, and it could only happen during resending
1790 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1791 * is also needed in this case(needed by changelog), so we have to add another
1792 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1793 * the ENOENT failure should be able to be fixed by redo mechanism.
1795 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1796 struct md_object *cobj, const struct lu_name *lname,
1797 struct md_attr *ma, int no_name)
1799 const char *name = lname->ln_name;
1800 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1801 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1802 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1803 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1804 struct mdd_object *mdd_cobj = NULL;
1805 struct mdd_device *mdd = mdo2mdd(pobj);
1806 struct thandle *handle;
1807 int rc, is_dir = 0, cl_flags = 0;
1810 /* cobj == NULL means only delete name entry */
1811 if (likely(cobj != NULL)) {
1812 mdd_cobj = md2mdd_obj(cobj);
1813 if (mdd_object_exists(mdd_cobj) == 0)
1817 rc = mdd_la_get(env, mdd_pobj, pattr);
1821 if (likely(mdd_cobj != NULL)) {
1823 rc = mdd_la_get(env, mdd_cobj, cattr);
1827 is_dir = S_ISDIR(cattr->la_mode);
1828 /* search for an existing archive.
1829 * we should check ahead as the object
1830 * can be destroyed in this transaction */
1831 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1832 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1835 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1839 handle = mdd_trans_create(env, mdd);
1841 RETURN(PTR_ERR(handle));
1843 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1844 lname, ma, handle, no_name, is_dir);
1848 rc = mdd_trans_start(env, mdd, handle);
1852 if (likely(mdd_cobj != NULL))
1853 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1855 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1856 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1861 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1862 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1863 GOTO(cleanup, rc = 0);
1865 if (likely(mdd_cobj != NULL)) {
1866 rc = mdo_ref_del(env, mdd_cobj, handle);
1868 __mdd_index_insert_only(env, mdd_pobj,
1870 mdd_object_type(mdd_cobj),
1877 mdo_ref_del(env, mdd_cobj, handle);
1879 /* fetch updated nlink */
1880 rc = mdd_la_get(env, mdd_cobj, cattr);
1885 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1886 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1888 la->la_valid = LA_CTIME | LA_MTIME;
1889 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1893 /* Enough for only unlink the entry */
1894 if (unlikely(mdd_cobj == NULL))
1897 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1898 /* update ctime of an unlinked file only if it is still
1899 * opened or a link still exists */
1900 la->la_valid = LA_CTIME;
1901 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1906 /* XXX: this transfer to ma will be removed with LOD/OSP */
1907 ma->ma_attr = *cattr;
1908 ma->ma_valid |= MA_INODE;
1909 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1913 /* fetch updated nlink */
1914 rc = mdd_la_get(env, mdd_cobj, cattr);
1915 /* if object is removed then we can't get its attrs,
1917 if (rc == -ENOENT) {
1918 cattr->la_nlink = 0;
1922 if (cattr->la_nlink == 0) {
1923 ma->ma_attr = *cattr;
1924 ma->ma_valid |= MA_INODE;
1929 if (likely(mdd_cobj != NULL))
1930 mdd_write_unlock(env, mdd_cobj);
1933 if (cattr->la_nlink == 0)
1934 cl_flags |= CLF_UNLINK_LAST;
1936 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1938 rc = mdd_changelog_ns_store(env, mdd,
1939 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1940 mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL,
1945 rc = mdd_trans_stop(env, mdd, rc, handle);
1951 * The permission has been checked when obj created, no need check again.
1953 static int mdd_cd_sanity_check(const struct lu_env *env,
1954 struct mdd_object *obj)
1959 if (!obj || mdd_is_dead_obj(obj))
1965 static int mdd_create_data(const struct lu_env *env,
1966 struct md_object *pobj,
1967 struct md_object *cobj,
1968 const struct md_op_spec *spec,
1971 struct mdd_device *mdd = mdo2mdd(cobj);
1972 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1973 struct mdd_object *son = md2mdd_obj(cobj);
1974 struct thandle *handle;
1975 const struct lu_buf *buf;
1976 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
1977 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
1981 rc = mdd_cd_sanity_check(env, son);
1985 if (!md_should_create(spec->sp_cr_flags))
1989 * there are following use cases for this function:
1990 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
1991 * striping can be specified or not
1994 rc = mdd_la_get(env, son, attr);
1998 /* calling ->ah_make_hint() is used to transfer information from parent */
1999 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2001 handle = mdd_trans_create(env, mdd);
2003 GOTO(out_free, rc = PTR_ERR(handle));
2006 * XXX: Setting the lov ea is not locked but setting the attr is locked?
2007 * Should this be fixed?
2009 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
2010 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
2011 spec->sp_cr_flags, spec->no_create);
2013 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2014 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2015 spec->u.sp_ea.eadatalen);
2020 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2021 XATTR_NAME_LOV, 0, handle);
2025 rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
2029 rc = mdd_trans_start(env, mdd, handle);
2033 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2039 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
2042 rc = mdd_trans_stop(env, mdd, rc, handle);
2048 static int mdd_declare_object_initialize(const struct lu_env *env,
2049 struct mdd_object *parent,
2050 struct mdd_object *child,
2051 const struct lu_attr *attr,
2052 struct thandle *handle)
2057 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2058 if (!S_ISDIR(attr->la_mode))
2061 rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
2066 rc = mdo_declare_ref_add(env, child, handle);
2070 rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR,
2076 static int mdd_object_initialize(const struct lu_env *env,
2077 const struct lu_fid *pfid,
2078 struct mdd_object *child,
2079 struct lu_attr *attr, struct thandle *handle,
2080 const struct md_op_spec *spec)
2085 if (S_ISDIR(attr->la_mode)) {
2086 /* Add "." and ".." for newly created dir */
2087 mdo_ref_add(env, child, handle);
2088 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
2089 S_IFDIR, dot, handle);
2091 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2094 mdo_ref_del(env, child, handle);
2101 * This function checks whether it can create a file/dir under the
2102 * directory(@pobj). The directory(@pobj) is not being locked by
2105 * \param[in] env execution environment
2106 * \param[in] pobj the directory to create files
2107 * \param[in] pattr the attributes of the directory
2108 * \param[in] lname the name of the created file/dir
2109 * \param[in] cattr the attributes of the file/dir
2110 * \param[in] spec create specification
2112 * \retval = 0 it is allowed to create file/dir under
2114 * \retval negative error not allowed to create file/dir
2115 * under the directory
2117 static int mdd_create_sanity_check(const struct lu_env *env,
2118 struct md_object *pobj,
2119 const struct lu_attr *pattr,
2120 const struct lu_name *lname,
2121 struct lu_attr *cattr,
2122 struct md_op_spec *spec)
2124 struct mdd_thread_info *info = mdd_env_info(env);
2125 struct lu_fid *fid = &info->mti_fid;
2126 struct mdd_object *obj = md2mdd_obj(pobj);
2127 struct mdd_device *m = mdo2mdd(pobj);
2128 bool check_perm = true;
2133 if (mdd_is_dead_obj(obj))
2137 * In some cases this lookup is not needed - we know before if name
2138 * exists or not because MDT performs lookup for it.
2139 * name length check is done in lookup.
2141 if (spec->sp_cr_lookup) {
2143 * Check if the name already exist, though it will be checked in
2144 * _index_insert also, for avoiding rolling back if exists
2147 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2148 MAY_WRITE | MAY_EXEC);
2150 RETURN(rc ? : -EEXIST);
2152 /* Permission is already being checked in mdd_lookup */
2156 if (S_ISDIR(cattr->la_mode) &&
2157 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2158 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2159 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2161 if (le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC &&
2162 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_SPECIFIC &&
2163 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2165 CERROR("%s: invalid lmv_user_md: magic = %x, "
2166 "stripe_offset = %d, stripe_count = %u: "
2167 "rc = %d\n", mdd2obd_dev(m)->obd_name,
2168 le32_to_cpu(lum->lum_magic),
2169 (int)le32_to_cpu(lum->lum_stripe_offset),
2170 le32_to_cpu(lum->lum_stripe_count), rc);
2175 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2180 if (pattr->la_mode & S_ISGID) {
2181 cattr->la_gid = pattr->la_gid;
2182 if (S_ISDIR(cattr->la_mode)) {
2183 cattr->la_mode |= S_ISGID;
2184 cattr->la_valid |= LA_MODE;
2188 /* Inherit project ID from parent directory */
2189 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2190 cattr->la_projid = pattr->la_projid;
2191 if (S_ISDIR(cattr->la_mode)) {
2192 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2193 cattr->la_valid |= LA_FLAGS;
2195 cattr->la_valid |= LA_PROJID;
2198 rc = mdd_name_check(m, lname);
2202 switch (cattr->la_mode & S_IFMT) {
2204 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
2206 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2207 RETURN(-ENAMETOOLONG);
2226 static int mdd_declare_create_object(const struct lu_env *env,
2227 struct mdd_device *mdd,
2228 struct mdd_object *p, struct mdd_object *c,
2229 struct lu_attr *attr,
2230 struct thandle *handle,
2231 const struct md_op_spec *spec,
2232 struct lu_buf *def_acl_buf,
2233 struct lu_buf *acl_buf,
2234 struct dt_allocation_hint *hint)
2236 const struct lu_buf *buf;
2239 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2244 #ifdef CONFIG_FS_POSIX_ACL
2245 if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2246 /* if dir, then can inherit default ACl */
2247 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2248 XATTR_NAME_ACL_DEFAULT,
2254 if (acl_buf->lb_len > 0) {
2255 rc = mdo_declare_attr_set(env, c, attr, handle);
2259 rc = mdo_declare_xattr_set(env, c, acl_buf,
2260 XATTR_NAME_ACL_ACCESS, 0, handle);
2265 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2269 /* replay case, create LOV EA from client data */
2270 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2271 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2272 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2273 spec->u.sp_ea.eadatalen);
2274 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
2280 if (S_ISLNK(attr->la_mode)) {
2281 const char *target_name = spec->u.sp_symname;
2282 int sym_len = strlen(target_name);
2283 const struct lu_buf *buf;
2285 buf = mdd_buf_get_const(env, target_name, sym_len);
2286 rc = dt_declare_record_write(env, mdd_object_child(c),
2292 if (spec->sp_cr_file_secctx_name != NULL) {
2293 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2294 spec->sp_cr_file_secctx_size);
2295 rc = mdo_declare_xattr_set(env, c, buf,
2296 spec->sp_cr_file_secctx_name, 0,
2305 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2306 struct mdd_object *p, struct mdd_object *c,
2307 const struct lu_name *name,
2308 struct lu_attr *attr,
2309 struct thandle *handle,
2310 const struct md_op_spec *spec,
2311 struct linkea_data *ldata,
2312 struct lu_buf *def_acl_buf,
2313 struct lu_buf *acl_buf,
2314 struct dt_allocation_hint *hint)
2318 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2319 def_acl_buf, acl_buf, hint);
2323 if (S_ISDIR(attr->la_mode)) {
2324 rc = mdo_declare_ref_add(env, p, handle);
2329 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2330 rc = orph_declare_index_insert(env, c, attr->la_mode, handle);
2334 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2336 rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode,
2337 name->ln_name, handle);
2341 rc = mdd_declare_links_add(env, c, handle, ldata);
2346 la->la_valid = LA_CTIME | LA_MTIME;
2347 rc = mdo_declare_attr_set(env, p, la, handle);
2351 rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle);
2359 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2360 struct lu_attr *la, struct lu_buf *def_acl_buf,
2361 struct lu_buf *acl_buf)
2366 if (S_ISLNK(la->la_mode)) {
2367 acl_buf->lb_len = 0;
2368 def_acl_buf->lb_len = 0;
2372 mdd_read_lock(env, pobj, MOR_TGT_PARENT);
2373 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2374 XATTR_NAME_ACL_DEFAULT);
2375 mdd_read_unlock(env, pobj);
2377 /* If there are default ACL, fix mode/ACL by default ACL */
2378 def_acl_buf->lb_len = rc;
2379 LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len);
2380 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2381 acl_buf->lb_len = rc;
2382 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2385 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2386 /* If there are no default ACL, fix mode by mask */
2387 struct lu_ucred *uc = lu_ucred(env);
2389 /* The create triggered by MDT internal events, such as
2390 * LFSCK reset, will not contain valid "uc". */
2391 if (unlikely(uc != NULL))
2392 la->la_mode &= ~uc->uc_umask;
2394 acl_buf->lb_len = 0;
2395 def_acl_buf->lb_len = 0;
2402 * Create a metadata object and initialize it, set acl, xattr.
2404 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2405 struct mdd_object *son, struct lu_attr *attr,
2406 struct md_op_spec *spec, struct lu_buf *acl_buf,
2407 struct lu_buf *def_acl_buf,
2408 struct dt_allocation_hint *hint,
2409 struct thandle *handle)
2411 const struct lu_buf *buf;
2414 mdd_write_lock(env, son, MOR_TGT_CHILD);
2415 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2420 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2421 * created in declare phase, they also needs to be added to master
2422 * object as sub-directory entry. So it has to initialize the master
2423 * object, then set dir striped EA.(in mdo_xattr_set) */
2424 rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle,
2427 GOTO(err_destroy, rc);
2430 * in case of replay we just set LOVEA provided by the client
2431 * XXX: I think it would be interesting to try "old" way where
2432 * MDT calls this xattr_set(LOV) in a different transaction.
2433 * probably this way we code can be made better.
2436 /* During creation, there are only a few cases we need do xattr_set to
2438 * 1. regular file: see comments above.
2439 * 2. dir: inherit default striping or pool settings from parent.
2440 * 3. create striped directory with provided stripeEA.
2441 * 4. create striped directory because inherit default layout from the
2444 if (spec->no_create ||
2445 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2446 S_ISDIR(attr->la_mode)) {
2447 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2448 spec->u.sp_ea.eadatalen);
2449 rc = mdo_xattr_set(env, son, buf,
2450 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2454 GOTO(err_destroy, rc);
2457 #ifdef CONFIG_FS_POSIX_ACL
2458 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2459 S_ISDIR(attr->la_mode)) {
2460 /* set default acl */
2461 rc = mdo_xattr_set(env, son, def_acl_buf,
2462 XATTR_NAME_ACL_DEFAULT, 0,
2465 GOTO(err_destroy, rc);
2467 /* set its own acl */
2468 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2469 rc = mdo_xattr_set(env, son, acl_buf,
2470 XATTR_NAME_ACL_ACCESS,
2473 GOTO(err_destroy, rc);
2477 if (S_ISLNK(attr->la_mode)) {
2478 struct lu_ucred *uc = lu_ucred_assert(env);
2479 struct dt_object *dt = mdd_object_child(son);
2480 const char *target_name = spec->u.sp_symname;
2481 int sym_len = strlen(target_name);
2484 buf = mdd_buf_get_const(env, target_name, sym_len);
2485 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2487 CFS_CAP_SYS_RESOURCE_MASK);
2492 GOTO(err_initlized, rc = -EFAULT);
2495 if (spec->sp_cr_file_secctx_name != NULL) {
2496 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2497 spec->sp_cr_file_secctx_size);
2498 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2501 GOTO(err_initlized, rc);
2505 if (unlikely(rc != 0)) {
2507 if (S_ISDIR(attr->la_mode)) {
2508 /* Drop the reference, no need to delete "."/"..",
2509 * because the object to be destroied directly. */
2510 rc2 = mdo_ref_del(env, son, handle);
2514 rc2 = mdo_ref_del(env, son, handle);
2518 mdo_destroy(env, son, handle);
2521 mdd_write_unlock(env, son);
2525 static int mdd_index_delete(const struct lu_env *env,
2526 struct mdd_object *mdd_pobj,
2527 struct lu_attr *cattr,
2528 const struct lu_name *lname)
2530 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2531 struct thandle *handle;
2535 handle = mdd_trans_create(env, mdd);
2537 RETURN(PTR_ERR(handle));
2539 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2544 if (S_ISDIR(cattr->la_mode)) {
2545 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2550 /* Since this will only be used in the error handler path,
2551 * Let's set the thandle to be local and not mess the transno */
2552 handle->th_local = 1;
2553 rc = mdd_trans_start(env, mdd, handle);
2557 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2558 S_ISDIR(cattr->la_mode), handle);
2562 rc = mdd_trans_stop(env, mdd, rc, handle);
2568 * Create object and insert it into namespace.
2570 * Two operations have to be performed:
2572 * - an allocation of a new object (->do_create()), and
2573 * - an insertion into a parent index (->dio_insert()).
2575 * Due to locking, operation order is not important, when both are
2576 * successful, *but* error handling cases are quite different:
2578 * - if insertion is done first, and following object creation fails,
2579 * insertion has to be rolled back, but this operation might fail
2580 * also leaving us with dangling index entry.
2582 * - if creation is done first, is has to be undone if insertion fails,
2583 * leaving us with leaked space, which is not good but not fatal.
2585 * It seems that creation-first is simplest solution, but it is sub-optimal
2591 * case, because second mkdir is bound to create object, only to
2592 * destroy it immediately.
2594 * To avoid this follow local file systems that do double lookup:
2596 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2597 * 1. create (mdd_create_object_internal())
2598 * 2. insert (__mdd_index_insert(), lookup again)
2600 * \param[in] pobj parent object
2601 * \param[in] lname name of child being created
2602 * \param[in,out] child child object being created
2603 * \param[in] spec additional create parameters
2604 * \param[in] ma attributes for new child object
2606 * \retval 0 on success
2607 * \retval negative errno on failure
2609 static int mdd_create(const struct lu_env *env, struct md_object *pobj,
2610 const struct lu_name *lname, struct md_object *child,
2611 struct md_op_spec *spec, struct md_attr *ma)
2613 struct mdd_thread_info *info = mdd_env_info(env);
2614 struct lu_attr *la = &info->mti_la_for_fix;
2615 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2616 struct mdd_object *son = md2mdd_obj(child);
2617 struct mdd_device *mdd = mdo2mdd(pobj);
2618 struct lu_attr *attr = &ma->ma_attr;
2619 struct thandle *handle;
2620 struct lu_attr *pattr = &info->mti_pattr;
2621 struct lu_buf acl_buf;
2622 struct lu_buf def_acl_buf;
2623 struct linkea_data *ldata = &info->mti_link_data;
2624 const char *name = lname->ln_name;
2625 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
2630 rc = mdd_la_get(env, mdd_pobj, pattr);
2634 /* Sanity checks before big job. */
2635 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2639 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2640 GOTO(out_free, rc = -EINPROGRESS);
2642 handle = mdd_trans_create(env, mdd);
2644 GOTO(out_free, rc = PTR_ERR(handle));
2646 lu_buf_check_and_alloc(&info->mti_xattr_buf,
2647 mdd->mdd_dt_conf.ddp_max_ea_size);
2648 acl_buf = info->mti_xattr_buf;
2649 def_acl_buf.lb_buf = info->mti_key;
2650 def_acl_buf.lb_len = sizeof(info->mti_key);
2651 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2655 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2657 memset(ldata, 0, sizeof(*ldata));
2658 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2659 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2662 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2663 &tfid, lname, 1, 0, ldata);
2665 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2666 mdd_object_fid(mdd_pobj),
2667 lname, 1, 0, ldata);
2670 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2671 handle, spec, ldata, &def_acl_buf, &acl_buf,
2676 rc = mdd_trans_start(env, mdd, handle);
2680 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2681 &def_acl_buf, hint, handle);
2685 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2686 mdd_write_lock(env, son, MOR_TGT_CHILD);
2687 son->mod_flags |= VOLATILE_OBJ;
2688 rc = __mdd_orphan_add(env, son, handle);
2689 GOTO(out_volatile, rc);
2691 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2692 attr->la_mode, name, handle);
2694 GOTO(err_created, rc);
2696 mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle,
2699 /* update parent directory mtime/ctime */
2701 la->la_valid = LA_CTIME | LA_MTIME;
2702 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2704 GOTO(err_insert, rc);
2710 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2711 rc2 = __mdd_orphan_del(env, son, handle);
2713 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2714 S_ISDIR(attr->la_mode),
2720 mdd_write_lock(env, son, MOR_TGT_CHILD);
2721 if (S_ISDIR(attr->la_mode)) {
2722 /* Drop the reference, no need to delete "."/"..",
2723 * because the object is to be destroyed directly. */
2724 rc2 = mdo_ref_del(env, son, handle);
2726 mdd_write_unlock(env, son);
2731 /* For volatile files drop one link immediately, since there is
2732 * no filename in the namespace, and save any error returned. */
2733 rc2 = mdo_ref_del(env, son, handle);
2735 mdd_write_unlock(env, son);
2736 if (unlikely(rc == 0))
2741 /* Don't destroy the volatile object on success */
2742 if (likely(rc != 0))
2743 mdo_destroy(env, son, handle);
2744 mdd_write_unlock(env, son);
2747 if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) &&
2748 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2749 rc = mdd_changelog_ns_store(env, mdd,
2750 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2751 S_ISREG(attr->la_mode) ? CL_CREATE :
2752 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2753 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname,
2756 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2758 /* If creation fails, it is most likely due to the remote update
2759 * failure, because local transaction will mostly succeed at
2760 * this stage. There is no easy way to rollback all of previous
2761 * updates, so let's remove the object from namespace, and
2762 * LFSCK should handle the orphan object. */
2763 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2764 mdd_index_delete(env, mdd_pobj, attr, lname);
2768 if (is_vmalloc_addr(ldata->ld_buf))
2769 /* if we vmalloced a large buffer drop it */
2770 lu_buf_free(ldata->ld_buf);
2772 /* The child object shouldn't be cached anymore */
2774 set_bit(LU_OBJECT_HEARD_BANSHEE,
2775 &child->mo_lu.lo_header->loh_flags);
2780 * Get locks on parents in proper order
2781 * RETURN: < 0 - error, rename_order if successful
2789 static int mdd_rename_order(const struct lu_env *env,
2790 struct mdd_device *mdd,
2791 struct mdd_object *src_pobj,
2792 const struct lu_attr *pattr,
2793 struct mdd_object *tgt_pobj)
2795 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2799 if (src_pobj == tgt_pobj)
2800 RETURN(MDD_RN_SAME);
2802 /* compared the parent child relationship of src_p&tgt_p */
2803 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2805 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2808 rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj),
2822 /* has not mdd_write{read}_lock on any obj yet. */
2823 static int mdd_rename_sanity_check(const struct lu_env *env,
2824 struct mdd_object *src_pobj,
2825 const struct lu_attr *pattr,
2826 struct mdd_object *tgt_pobj,
2827 const struct lu_attr *tpattr,
2828 struct mdd_object *sobj,
2829 const struct lu_attr *cattr,
2830 struct mdd_object *tobj,
2831 const struct lu_attr *tattr)
2836 /* XXX: when get here, sobj must NOT be NULL,
2837 * the other case has been processed in cld_rename
2838 * before mdd_rename and enable MDS_PERM_BYPASS. */
2842 * If we are using project inheritance, we only allow renames
2843 * into our tree when the project IDs are the same; otherwise
2844 * tree quota mechanism would be circumvented.
2846 if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2847 tpattr->la_projid != cattr->la_projid) ||
2848 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2849 (pattr->la_projid != tpattr->la_projid)))
2852 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2856 /* XXX: when get here, "tobj == NULL" means tobj must
2857 * NOT exist (neither on remote MDS, such case has been
2858 * processed in cld_rename before mdd_rename and enable
2860 * So check may_create, but not check may_unlink. */
2862 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2863 (src_pobj != tgt_pobj));
2865 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2866 (src_pobj != tgt_pobj), 1);
2868 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2869 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2874 static int mdd_declare_rename(const struct lu_env *env,
2875 struct mdd_device *mdd,
2876 struct mdd_object *mdd_spobj,
2877 struct mdd_object *mdd_tpobj,
2878 struct mdd_object *mdd_sobj,
2879 struct mdd_object *mdd_tobj,
2880 const struct lu_name *sname,
2881 const struct lu_name *tname,
2883 struct linkea_data *ldata,
2884 struct thandle *handle)
2886 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2889 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2890 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2896 /* name from source dir */
2897 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2901 /* .. from source child */
2902 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2903 /* source child can be directory,
2904 * counted by source dir's nlink */
2905 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2908 if (mdd_spobj != mdd_tpobj) {
2909 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
2914 rc = mdo_declare_index_insert(env, mdd_sobj,
2916 S_IFDIR, dotdot, handle);
2921 /* new target child can be directory,
2922 * counted by target dir's nlink */
2923 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2928 la->la_valid = LA_CTIME | LA_MTIME;
2929 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
2933 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
2937 la->la_valid = LA_CTIME;
2938 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
2942 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
2947 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2948 mdd_object_type(mdd_sobj),
2949 tname->ln_name, handle);
2953 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2954 /* delete target child in target parent directory */
2955 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
2960 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2964 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2965 /* target child can be directory,
2966 * delete "." reference in target child directory */
2967 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2971 /* delete ".." reference in target parent directory */
2972 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2977 la->la_valid = LA_CTIME;
2978 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
2982 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
2987 rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle);
2994 /* src object can be remote that is why we use only fid and type of object */
2995 static int mdd_rename(const struct lu_env *env,
2996 struct md_object *src_pobj, struct md_object *tgt_pobj,
2997 const struct lu_fid *lf, const struct lu_name *lsname,
2998 struct md_object *tobj, const struct lu_name *ltname,
3001 const char *sname = lsname->ln_name;
3002 const char *tname = ltname->ln_name;
3003 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
3004 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
3005 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
3006 struct mdd_device *mdd = mdo2mdd(src_pobj);
3007 struct mdd_object *mdd_sobj = NULL; /* source object */
3008 struct mdd_object *mdd_tobj = NULL;
3009 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3010 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3011 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3012 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3013 struct thandle *handle;
3014 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3015 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
3016 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
3019 bool tobj_locked = 0;
3020 unsigned cl_flags = 0;
3025 mdd_tobj = md2mdd_obj(tobj);
3027 mdd_sobj = mdd_object_find(env, mdd, lf);
3028 if (IS_ERR(mdd_sobj))
3029 RETURN(PTR_ERR(mdd_sobj));
3031 rc = mdd_la_get(env, mdd_sobj, cattr);
3033 GOTO(out_pending, rc);
3035 rc = mdd_la_get(env, mdd_spobj, pattr);
3037 GOTO(out_pending, rc);
3040 rc = mdd_la_get(env, mdd_tobj, tattr);
3042 GOTO(out_pending, rc);
3043 /* search for an existing archive.
3044 * we should check ahead as the object
3045 * can be destroyed in this transaction */
3046 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3047 cl_flags |= CLF_RENAME_LAST_EXISTS;
3050 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3052 GOTO(out_pending, rc);
3054 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3055 mdd_sobj, cattr, mdd_tobj, tattr);
3057 GOTO(out_pending, rc);
3059 rc = mdd_name_check(mdd, ltname);
3061 GOTO(out_pending, rc);
3063 /* FIXME: Should consider tobj and sobj too in rename_lock. */
3064 rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj);
3066 GOTO(out_pending, rc);
3068 handle = mdd_trans_create(env, mdd);
3070 GOTO(out_pending, rc = PTR_ERR(handle));
3072 memset(ldata, 0, sizeof(*ldata));
3073 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3074 lsname, mdd_object_fid(mdd_tpobj), ltname,
3079 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3080 mdd_tobj, lsname, ltname, ma, ldata, handle);
3084 rc = mdd_trans_start(env, mdd, handle);
3088 is_dir = S_ISDIR(cattr->la_mode);
3090 /* Remove source name from source directory */
3091 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3095 /* "mv dir1 dir2" needs "dir1/.." link update */
3096 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3097 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3099 GOTO(fixup_spobj2, rc);
3101 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3104 GOTO(fixup_spobj, rc);
3107 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3108 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3110 /* tname might been renamed to something else */
3111 GOTO(fixup_spobj, rc);
3114 /* Insert new fid with target name into target dir */
3115 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3118 GOTO(fixup_tpobj, rc);
3120 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3121 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3123 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3124 la->la_valid = LA_CTIME;
3125 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3127 GOTO(fixup_tpobj, rc);
3129 /* Update the linkEA for the source object */
3130 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
3131 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
3132 mdo2fid(mdd_tpobj), ltname, handle, ldata,
3135 /* Old files might not have EA entry */
3136 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
3137 lsname, handle, NULL, 0);
3138 mdd_write_unlock(env, mdd_sobj);
3139 /* We don't fail the transaction if the link ea can't be
3140 updated -- fid2path will use alternate lookup method. */
3143 /* Remove old target object
3144 * For tobj is remote case cmm layer has processed
3145 * and set tobj to NULL then. So when tobj is NOT NULL,
3146 * it must be local one.
3148 if (tobj && mdd_object_exists(mdd_tobj)) {
3149 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
3151 if (mdd_is_dead_obj(mdd_tobj)) {
3152 /* shld not be dead, something is wrong */
3153 CERROR("tobj is dead, something is wrong\n");
3157 mdo_ref_del(env, mdd_tobj, handle);
3159 /* Remove dot reference. */
3160 if (S_ISDIR(tattr->la_mode))
3161 mdo_ref_del(env, mdd_tobj, handle);
3164 /* fetch updated nlink */
3165 rc = mdd_la_get(env, mdd_tobj, tattr);
3167 CERROR("%s: Failed to get nlink for tobj "
3169 mdd2obd_dev(mdd)->obd_name,
3170 PFID(tpobj_fid), rc);
3171 GOTO(fixup_tpobj, rc);
3174 la->la_valid = LA_CTIME;
3175 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3177 CERROR("%s: Failed to set ctime for tobj "
3179 mdd2obd_dev(mdd)->obd_name,
3180 PFID(tpobj_fid), rc);
3181 GOTO(fixup_tpobj, rc);
3184 /* XXX: this transfer to ma will be removed with LOD/OSP */
3185 ma->ma_attr = *tattr;
3186 ma->ma_valid |= MA_INODE;
3187 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3190 CERROR("%s: Failed to unlink tobj "
3192 mdd2obd_dev(mdd)->obd_name,
3193 PFID(tpobj_fid), rc);
3194 GOTO(fixup_tpobj, rc);
3197 /* fetch updated nlink */
3198 rc = mdd_la_get(env, mdd_tobj, tattr);
3199 if (rc == -ENOENT) {
3200 /* the object got removed, let's
3201 * return the latest known attributes */
3202 tattr->la_nlink = 0;
3204 } else if (rc != 0) {
3205 CERROR("%s: Failed to get nlink for tobj "
3207 mdd2obd_dev(mdd)->obd_name,
3208 PFID(tpobj_fid), rc);
3209 GOTO(fixup_tpobj, rc);
3211 /* XXX: this transfer to ma will be removed with LOD/OSP */
3212 ma->ma_attr = *tattr;
3213 ma->ma_valid |= MA_INODE;
3215 if (tattr->la_nlink == 0)
3216 cl_flags |= CLF_RENAME_LAST;
3218 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3221 la->la_valid = LA_CTIME | LA_MTIME;
3222 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3224 GOTO(fixup_tpobj, rc);
3226 if (mdd_spobj != mdd_tpobj) {
3227 la->la_valid = LA_CTIME | LA_MTIME;
3228 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3230 GOTO(fixup_tpobj, rc);
3237 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3239 CWARN("tp obj fix error %d\n",rc2);
3241 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3242 !mdd_is_dead_obj(mdd_tobj)) {
3244 mdo_ref_add(env, mdd_tobj, handle);
3246 mdo_ref_add(env, mdd_tobj, handle);
3249 rc2 = __mdd_index_insert(env, mdd_tpobj,
3251 mdd_object_type(mdd_tobj),
3254 CWARN("tp obj fix error: rc = %d\n", rc2);
3259 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3260 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3262 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3263 mdd2obd_dev(mdd)->obd_name, rc2);
3266 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3269 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3270 mdd2obd_dev(mdd)->obd_name, rc2);
3275 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3276 mdd_object_type(mdd_sobj), sname,
3279 CWARN("sp obj fix error: rc = %d\n", rc2);
3284 mdd_write_unlock(env, mdd_tobj);
3287 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3288 mdd_tobj, tpobj_fid, lf, spobj_fid,
3289 ltname, lsname, handle);
3292 rc = mdd_trans_stop(env, mdd, rc, handle);
3295 mdd_object_put(env, mdd_sobj);
3300 * During migration once the parent FID has been changed,
3301 * we need update the parent FID in linkea.
3303 static int mdd_linkea_update_child_internal(const struct lu_env *env,
3304 struct mdd_object *parent,
3305 struct mdd_object *newparent,
3306 struct mdd_object *child,
3307 const char *name, int namelen,
3308 struct thandle *handle,
3311 struct mdd_thread_info *info = mdd_env_info(env);
3312 struct linkea_data ldata = { NULL };
3313 struct lu_buf *buf = &info->mti_link_buf;
3319 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
3320 if (buf->lb_buf == NULL)
3324 rc = mdd_links_read(env, child, &ldata);
3326 if (rc == -ENOENT || rc == -ENODATA)
3331 LASSERT(ldata.ld_leh != NULL);
3332 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
3333 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
3334 struct mdd_device *mdd = mdo2mdd(&child->mod_obj);
3335 struct lu_name lname;
3338 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3341 if (strncmp(lname.ln_name, name, namelen) != 0 ||
3342 !lu_fid_eq(&fid, mdd_object_fid(parent))) {
3343 ldata.ld_lee = (struct link_ea_entry *)
3344 ((char *)ldata.ld_lee +
3349 CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
3350 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
3351 lname.ln_namelen, lname.ln_name,
3352 PFID(mdd_object_fid(newparent)));
3353 /* update to the new parent fid */
3354 linkea_entry_pack(ldata.ld_lee, &lname,
3355 mdd_object_fid(newparent));
3357 rc = mdd_declare_links_add(env, child, handle, &ldata);
3359 rc = mdd_links_write(env, child, &ldata, handle);
3365 static int mdd_linkea_declare_update_child(const struct lu_env *env,
3366 struct mdd_object *parent,
3367 struct mdd_object *newparent,
3368 struct mdd_object *child,
3369 const char *name, int namelen,
3370 struct thandle *handle)
3372 return mdd_linkea_update_child_internal(env, parent, newparent,
3374 namelen, handle, true);
3377 static int mdd_linkea_update_child(const struct lu_env *env,
3378 struct mdd_object *parent,
3379 struct mdd_object *newparent,
3380 struct mdd_object *child,
3381 const char *name, int namelen,
3382 struct thandle *handle)
3384 return mdd_linkea_update_child_internal(env, parent, newparent,
3386 namelen, handle, false);
3389 static int mdd_update_linkea_internal(const struct lu_env *env,
3390 struct mdd_object *mdd_pobj,
3391 struct mdd_object *mdd_sobj,
3392 struct mdd_object *mdd_tobj,
3393 const struct lu_name *child_name,
3394 struct linkea_data *ldata,
3395 struct thandle *handle,
3398 struct mdd_thread_info *info = mdd_env_info(env);
3403 LASSERT(ldata->ld_buf != NULL);
3404 LASSERT(ldata->ld_leh != NULL);
3406 /* If it is mulitple links file, we need update the name entry for
3408 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
3409 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
3410 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3411 struct mdd_object *pobj;
3412 struct lu_name lname;
3415 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
3417 pobj = mdd_object_find(env, mdd, &fid);
3419 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
3420 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
3425 if (!mdd_object_exists(pobj)) {
3426 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
3427 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
3431 if (pobj == mdd_pobj &&
3432 lname.ln_namelen == child_name->ln_namelen &&
3433 strncmp(lname.ln_name, child_name->ln_name,
3434 lname.ln_namelen) == 0) {
3435 CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
3436 mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
3441 CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
3442 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)),
3443 PNAME(&lname), PFID(mdd_object_fid(mdd_tobj)));
3446 /* Remove source name from source directory */
3447 /* Insert new fid with target name into target dir */
3448 rc = mdo_declare_index_delete(env, pobj, lname.ln_name,
3453 rc = mdo_declare_index_insert(env, pobj,
3454 mdd_object_fid(mdd_tobj),
3455 mdd_object_type(mdd_tobj),
3456 lname.ln_name, handle);
3460 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3464 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3468 char *tmp_name = info->mti_key;
3470 if (lname.ln_namelen >= sizeof(info->mti_key)) {
3471 /* lnamelen is too big(> NAME_MAX + 16),
3472 * something wrong about this linkea, let's
3474 CWARN("%s: the name %.*s is too long under "
3475 DFID"\n", mdd2obd_dev(mdd)->obd_name,
3476 lname.ln_namelen, lname.ln_name,
3481 /* Note: lname might be without \0 at the end, see
3482 * linkea_entry_unpack(), let's add extra \0 by
3484 snprintf(tmp_name, sizeof(info->mti_key), "%.*s",
3485 lname.ln_namelen, lname.ln_name);
3486 lname.ln_name = tmp_name;
3488 /* Let's check if this linkEA still valid, before
3489 * it might be packed into the RPC buffer. */
3490 rc = mdd_lookup(env, &pobj->mod_obj, &lname,
3491 &info->mti_fid, NULL);
3492 if (rc < 0 || !lu_fid_eq(&info->mti_fid,
3493 mdd_object_fid(mdd_sobj)))
3494 GOTO(next_put, rc == -ENOENT ? 0 : rc);
3496 rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
3500 rc = __mdd_index_insert(env, pobj,
3501 mdd_object_fid(mdd_tobj),
3502 mdd_object_type(mdd_tobj),
3507 mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD);
3508 rc = mdo_ref_add(env, mdd_tobj, handle);
3509 mdd_write_unlock(env, mdd_tobj);
3513 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
3514 mdo_ref_del(env, mdd_sobj, handle);
3515 mdd_write_unlock(env, mdd_sobj);
3518 mdd_object_put(env, pobj);
3522 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
3529 static int mdd_migrate_xattrs(const struct lu_env *env,
3530 struct mdd_object *mdd_sobj,
3531 struct mdd_object *mdd_tobj)
3533 struct mdd_thread_info *info = mdd_env_info(env);
3534 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3536 struct thandle *handle;
3542 struct lu_buf list_xbuf;
3545 /* retrieve xattr list from the old object */
3546 list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
3547 if (list_xsize == -ENODATA)
3553 lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize);
3554 if (info->mti_big_buf.lb_buf == NULL)
3557 list_xbuf.lb_buf = info->mti_big_buf.lb_buf;
3558 list_xbuf.lb_len = list_xsize;
3559 rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf);
3564 xname = list_xbuf.lb_buf;
3566 xlen = strnlen(xname, rem - 1) + 1;
3567 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3568 strcmp(XATTR_NAME_LMV, xname) == 0)
3571 /* For directory, if there are default layout, migrate here */
3572 if (strcmp(XATTR_NAME_LOV, xname) == 0 &&
3573 !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu)))
3576 xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname);
3577 if (xsize == -ENODATA)
3582 lu_buf_check_and_alloc(&info->mti_link_buf, xsize);
3583 if (info->mti_link_buf.lb_buf == NULL)
3584 GOTO(out, rc = -ENOMEM);
3586 xbuf.lb_len = xsize;
3587 xbuf.lb_buf = info->mti_link_buf.lb_buf;
3588 rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname);
3594 handle = mdd_trans_create(env, mdd);
3596 GOTO(out, rc = PTR_ERR(handle));
3598 rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0,
3601 GOTO(stop_trans, rc);
3602 /* Note: this transaction is part of migration, and it is not
3603 * the last step of migration, so we set th_local = 1 to avoid
3604 * update last rcvd for this transaction */
3605 handle->th_local = 1;
3606 rc = mdd_trans_start(env, mdd, handle);
3608 GOTO(stop_trans, rc);
3611 rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
3613 GOTO(stop_trans, rc = 0);
3615 if (unlikely(rc == -ENOSPC &&
3616 strcmp(xname, XATTR_NAME_LINK) == 0)) {
3617 rc = linkea_overflow_shrink(
3618 (struct linkea_data *)(xbuf.lb_buf));
3619 if (likely(rc > 0)) {
3626 GOTO(stop_trans, rc);
3628 rc = mdd_trans_stop(env, mdd, rc, handle);
3633 memmove(xname, xname + xlen, rem);
3639 static int mdd_declare_migrate_create(const struct lu_env *env,
3640 struct mdd_object *mdd_pobj,
3641 struct mdd_object *mdd_sobj,
3642 struct mdd_object *mdd_tobj,
3643 struct md_op_spec *spec,
3645 union lmv_mds_md *mgr_ea,
3646 struct linkea_data *ldata,
3647 struct thandle *handle)
3649 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3650 const struct lu_buf *buf;
3654 rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
3655 handle, spec, NULL);
3659 rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la,
3664 if (S_ISLNK(la->la_mode)) {
3665 const char *target_name = spec->u.sp_symname;
3666 int sym_len = strlen(target_name);
3667 const struct lu_buf *buf;
3669 buf = mdd_buf_get_const(env, target_name, sym_len);
3670 rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj),
3674 } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
3675 rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
3680 if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
3681 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
3682 spec->u.sp_ea.eadatalen);
3683 rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV,
3689 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3690 buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
3691 rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV,
3696 la_flag->la_valid = LA_FLAGS;
3697 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3698 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
3703 static int mdd_migrate_create(const struct lu_env *env,
3704 struct mdd_object *mdd_pobj,
3705 struct mdd_object *mdd_sobj,
3706 struct mdd_object *mdd_tobj,
3707 const struct lu_name *lname,
3710 struct mdd_thread_info *info = mdd_env_info(env);
3711 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3712 struct md_op_spec *spec = &info->mti_spec;
3713 struct lu_buf lmm_buf = { NULL };
3714 struct lu_buf link_buf = { NULL };
3715 struct lu_buf mgr_buf;
3716 struct thandle *handle;
3717 struct lmv_mds_md_v1 *mgr_ea;
3718 struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
3719 struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint;
3721 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
3725 /* prepare spec for create */
3726 memset(spec, 0, sizeof(*spec));
3727 spec->sp_cr_lookup = 0;
3728 spec->sp_feat = &dt_directory_features;
3729 if (S_ISLNK(la->la_mode)) {
3730 const struct lu_buf *buf;
3732 buf = lu_buf_check_and_alloc(
3733 &mdd_env_info(env)->mti_big_buf,
3736 link_buf.lb_len = la->la_size + 1;
3737 memset(link_buf.lb_buf, 0, link_buf.lb_len);
3738 rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf);
3740 rc = rc != 0 ? rc : -EFAULT;
3741 CERROR("%s: "DFID" readlink failed: rc = %d\n",
3742 mdd2obd_dev(mdd)->obd_name,
3743 PFID(mdd_object_fid(mdd_sobj)), rc);
3746 spec->u.sp_symname = link_buf.lb_buf;
3747 } else if (S_ISREG(la->la_mode)) {
3748 /* retrieve lov of the old object */
3749 rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf);
3750 if (rc != 0 && rc != -ENODATA)
3752 if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) {
3753 spec->u.sp_ea.eadata = lmm_buf.lb_buf;
3754 spec->u.sp_ea.eadatalen = lmm_buf.lb_len;
3755 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
3757 } else if (S_ISDIR(la->la_mode)) {
3758 rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
3759 if (rc == -ENODATA) {
3760 /* ignore the non-linkEA error */
3768 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
3769 lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
3770 mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
3771 mgr_buf.lb_len = mgr_easize;
3772 mgr_ea = mgr_buf.lb_buf;
3773 memset(mgr_ea, 0, sizeof(*mgr_ea));
3774 mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
3775 mgr_ea->lmv_stripe_count = cpu_to_le32(2);
3776 mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id;
3777 mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
3778 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj));
3779 fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj));
3781 mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint);
3783 handle = mdd_trans_create(env, mdd);
3785 GOTO(out_free, rc = PTR_ERR(handle));
3787 /* Note: this transaction is part of migration, and it is not
3788 * the last step of migration, so we set th_local = 1 to avoid
3789 * update last rcvd for this transaction */
3790 handle->th_local = 1;
3791 rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
3792 la, mgr_buf.lb_buf, ldata, handle);
3794 GOTO(stop_trans, rc);
3796 rc = mdd_trans_start(env, mdd, handle);
3798 GOTO(stop_trans, rc);
3800 /* don't set nlink from the original object */
3801 la->la_valid &= ~LA_NLINK;
3803 /* create the target object */
3804 rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
3807 GOTO(stop_trans, rc);
3809 if (S_ISDIR(la->la_mode) && ldata != NULL) {
3810 rc = mdd_links_write(env, mdd_tobj, ldata, handle);
3812 GOTO(stop_trans, rc);
3815 /* Set MIGRATE EA on the source inode, so once the migration needs
3816 * to be re-done during failover, the re-do process can locate the
3817 * target object which is already being created. */
3818 rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
3820 GOTO(stop_trans, rc);
3822 /* Set immutable flag, so any modification is disabled until
3823 * the migration is done. Once the migration is interrupted,
3824 * if the resume process find the migrating object has both
3825 * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE
3826 * flag and approve the migration */
3827 la_flag->la_valid = LA_FLAGS;
3828 la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
3829 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
3832 rc = mdd_trans_stop(env, mdd, rc, handle);
3834 if (lmm_buf.lb_buf != NULL)
3835 OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
3839 static int mdd_migrate_entries(const struct lu_env *env,
3840 struct mdd_object *mdd_sobj,
3841 struct mdd_object *mdd_tobj)
3843 struct dt_object *next = mdd_object_child(mdd_sobj);
3844 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
3845 struct dt_object *dt_tobj = mdd_object_child(mdd_tobj);
3846 struct thandle *handle;
3848 const struct dt_it_ops *iops;
3850 struct lu_dirent *ent;
3854 OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
3858 if (!dt_try_as_dir(env, next))
3859 GOTO(out_ent, rc = -ENOTDIR);
3861 * iterate directories
3863 iops = &next->do_index_ops->dio_it;
3864 it = iops->init(env, next, LUDA_FID | LUDA_TYPE);
3866 GOTO(out_ent, rc = PTR_ERR(it));
3868 rc = iops->load(env, it, 0);
3870 rc = iops->next(env, it);
3874 * At this point and across for-loop:
3876 * rc == 0 -> ok, proceed.
3877 * rc > 0 -> end of directory.
3881 struct mdd_object *child;
3882 char *name = mdd_env_info(env)->mti_key;
3885 bool target_exist = false;
3887 len = iops->key_size(env, it);
3891 result = iops->rec(env, it, (struct dt_rec *)ent,
3892 LUDA_FID | LUDA_TYPE);
3893 if (result == -ESTALE)
3900 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
3902 /* Insert new fid with target name into target dir */
3903 if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
3904 (ent->lde_namelen == 2 && ent->lde_name[0] == '.' &&
3905 ent->lde_name[1] == '.'))
3908 child = mdd_object_find(env, mdd, &ent->lde_fid);
3910 GOTO(out, rc = PTR_ERR(child));
3912 /* child may not exist, but lu_object_attr will assert this,
3913 * get type from loh_attr directly */
3914 is_dir = S_ISDIR(child->mod_obj.mo_lu.lo_header->loh_attr);
3916 mdd_write_lock(env, child, MOR_SRC_CHILD);
3918 snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name);
3920 /* Check whether the name has been inserted to the target */
3921 if (dt_try_as_dir(env, dt_tobj)) {
3922 struct lu_fid *fid = &mdd_env_info(env)->mti_fid2;
3924 rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid,
3925 (struct dt_key *)name);
3926 if (unlikely(rc == 0))
3927 target_exist = true;
3930 handle = mdd_trans_create(env, mdd);
3932 GOTO(out_put, rc = PTR_ERR(handle));
3934 /* Note: this transaction is part of migration, and it is not
3935 * the last step of migration, so we set th_local = 1 to avoid
3936 * updating last rcvd for this transaction */
3937 handle->th_local = 1;
3938 if (likely(!target_exist)) {
3939 rc = mdo_declare_index_insert(env, mdd_tobj,
3941 child->mod_obj.mo_lu.lo_header->loh_attr,
3947 rc = mdo_declare_ref_add(env, mdd_tobj, handle);
3953 rc = mdo_declare_index_delete(env, mdd_sobj, name, handle);
3958 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
3962 /* Update .. for child */
3963 rc = mdo_declare_index_delete(env, child, dotdot,
3968 rc = mdo_declare_index_insert(env, child,
3969 mdd_object_fid(mdd_tobj),
3970 S_IFDIR, dotdot, handle);
3975 rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
3982 rc = mdd_trans_start(env, mdd, handle);
3984 CERROR("%s: transaction start failed: rc = %d\n",
3985 mdd2obd_dev(mdd)->obd_name, rc);
3989 if (likely(!target_exist)) {
3990 rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid,
3991 child->mod_obj.mo_lu.lo_header->loh_attr, name,
3997 rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle);
4002 rc = __mdd_index_delete_only(env, child, dotdot,
4007 rc = __mdd_index_insert_only(env, child,
4008 mdd_object_fid(mdd_tobj), S_IFDIR,
4014 rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
4016 strlen(name), handle);
4019 mdd_write_unlock(env, child);
4020 mdd_object_put(env, child);
4021 rc = mdd_trans_stop(env, mdd, rc, handle);
4025 result = iops->next(env, it);
4026 if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES))
4027 GOTO(out, rc = -EINTR);
4029 if (result == -ESTALE)
4031 } while (result == 0);
4034 iops->fini(env, it);
4036 OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1);
4040 static int mdd_declare_update_linkea(const struct lu_env *env,
4041 struct mdd_object *mdd_pobj,
4042 struct mdd_object *mdd_sobj,
4043 struct mdd_object *mdd_tobj,
4044 const struct lu_name *child_name,
4045 struct linkea_data *ldata,
4046 struct thandle *handle)
4048 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4049 child_name, ldata, handle, 1);
4052 static int mdd_update_linkea(const struct lu_env *env,
4053 struct mdd_object *mdd_pobj,
4054 struct mdd_object *mdd_sobj,
4055 struct mdd_object *mdd_tobj,
4056 const struct lu_name *child_name,
4057 struct linkea_data *ldata,
4058 struct thandle *handle)
4060 return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj,
4061 child_name, ldata, handle, 0);
4064 static int mdd_declare_migrate_update_name(const struct lu_env *env,
4065 struct mdd_object *mdd_pobj,
4066 struct mdd_object *mdd_sobj,
4067 struct mdd_object *mdd_tobj,
4068 const struct lu_name *lname,
4070 struct lu_attr *parent_la,
4071 struct linkea_data *ldata,
4072 struct thandle *handle)
4074 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4075 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4078 /* Revert IMMUTABLE flag */
4079 la_flag->la_valid = LA_FLAGS;
4080 la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL;
4081 rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle);
4085 /* delete entry from source dir */
4086 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle);
4090 if (ldata->ld_buf != NULL) {
4091 rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj,
4092 mdd_tobj, lname, ldata, handle);
4097 if (S_ISREG(mdd_object_type(mdd_sobj))) {
4098 rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4103 handle->th_complex = 1;
4104 rc = mdo_declare_xattr_set(env, mdd_tobj, NULL,
4106 LU_XATTR_REPLACE, handle);
4111 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4112 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
4118 rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj),
4119 mdd_object_type(mdd_tobj),
4120 lname->ln_name, handle);
4124 rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
4128 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4129 rc = mdo_declare_ref_add(env, mdd_pobj, handle);
4134 /* delete old object */
4135 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4139 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
4140 /* delete old object */
4141 rc = mdo_declare_ref_del(env, mdd_sobj, handle);
4144 /* set nlink to 0 */
4145 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
4150 rc = mdd_declare_finish_unlink(env, mdd_sobj, handle);
4154 rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle);
4158 rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle);
4163 static int mdd_migrate_update_name(const struct lu_env *env,
4164 struct mdd_object *mdd_pobj,
4165 struct mdd_object *mdd_sobj,
4166 struct mdd_object *mdd_tobj,
4167 const struct lu_name *lname,
4170 struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix);
4171 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4172 struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr);
4173 struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
4174 struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
4175 struct thandle *handle;
4176 int is_dir = S_ISDIR(mdd_object_type(mdd_sobj));
4177 const char *name = lname->ln_name;
4181 /* update time for parent */
4182 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
4183 p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime;
4184 p_la->la_valid = LA_CTIME;
4186 rc = mdd_la_get(env, mdd_sobj, so_attr);
4190 ldata->ld_buf = NULL;
4191 rc = mdd_links_read(env, mdd_sobj, ldata);
4192 if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
4195 handle = mdd_trans_create(env, mdd);
4197 RETURN(PTR_ERR(handle));
4199 rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj,
4200 lname, so_attr, p_la, ldata,
4203 /* If the migration can not be fit in one transaction, just
4204 * leave it in the original MDT */
4206 GOTO(stop_trans, rc = 0);
4208 GOTO(stop_trans, rc);
4211 CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n",
4212 mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)),
4213 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4214 PFID(mdd_object_fid(mdd_tobj)));
4216 rc = mdd_trans_start(env, mdd, handle);
4218 GOTO(stop_trans, rc);
4220 /* Revert IMMUTABLE flag */
4221 la_flag->la_valid = LA_FLAGS;
4222 la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL;
4223 rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
4225 GOTO(stop_trans, rc);
4227 /* Remove source name from source directory */
4228 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
4230 GOTO(stop_trans, rc);
4232 if (ldata->ld_buf != NULL) {
4233 rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj,
4234 lname, ldata, handle);
4236 GOTO(stop_trans, rc);
4238 /* linkea update might decrease the source object
4239 * nlink, let's get the attr again after ref_del */
4240 rc = mdd_la_get(env, mdd_sobj, so_attr);
4242 GOTO(stop_trans, rc);
4245 if (S_ISREG(so_attr->la_mode)) {
4246 if (so_attr->la_nlink == 1) {
4247 rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV,
4249 if (rc != 0 && rc != -ENODATA)
4250 GOTO(stop_trans, rc);
4252 rc = mdo_xattr_set(env, mdd_tobj, NULL,
4254 LU_XATTR_REPLACE, handle);
4256 GOTO(stop_trans, rc);
4260 /* Insert new fid with target name into target dir */
4261 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj),
4262 mdd_object_type(mdd_tobj), name, handle);
4264 GOTO(stop_trans, rc);
4266 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
4268 mdd_sobj->mod_flags |= DEAD_OBJ;
4269 rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
4271 GOTO(out_unlock, rc);
4273 rc = __mdd_orphan_add(env, mdd_sobj, handle);
4275 GOTO(out_unlock, rc);
4277 mdo_ref_del(env, mdd_sobj, handle);
4279 mdo_ref_del(env, mdd_sobj, handle);
4281 /* Get the attr again after ref_del */
4282 rc = mdd_la_get(env, mdd_sobj, so_attr);
4284 GOTO(out_unlock, rc);
4286 ma->ma_attr = *so_attr;
4287 ma->ma_valid |= MA_INODE;
4289 rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0);
4291 GOTO(out_unlock, rc);
4293 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj,
4294 mdo2fid(mdd_pobj), mdo2fid(mdd_sobj),
4295 mdo2fid(mdd_pobj), lname, lname, handle);
4297 CWARN("%s: changelog for migrate %s "DFID
4298 "under "DFID" failed: rc = %d\n",
4299 mdd2obd_dev(mdd)->obd_name, lname->ln_name,
4300 PFID(mdd_object_fid(mdd_sobj)),
4301 PFID(mdd_object_fid(mdd_pobj)), rc);
4302 /* Sigh, there are no easy way to migrate back the object, so
4303 * let's reset the result to 0 for now XXX */
4307 mdd_write_unlock(env, mdd_sobj);
4310 rc = mdd_trans_stop(env, mdd, rc, handle);
4315 static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd,
4316 const struct lu_fid *fid, __u32 *mdt_index)
4318 struct lu_seq_range *range = &mdd_env_info(env)->mti_range;
4319 struct seq_server_site *ss;
4322 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
4324 range->lsr_flags = LU_SEQ_RANGE_MDT;
4325 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
4329 *mdt_index = range->lsr_index;
4334 * Check whether we should migrate the file/dir
4336 * < 0 permission check failed or other error.
4337 * = 0 the file can be migrated.
4338 * > 0 the file does not need to be migrated, mostly
4339 * for multiple link file
4341 static int mdd_migrate_sanity_check(const struct lu_env *env,
4342 struct mdd_object *pobj,
4343 const struct lu_attr *pattr,
4344 struct mdd_object *sobj,
4345 struct lu_attr *sattr)
4347 struct mdd_thread_info *info = mdd_env_info(env);
4348 struct linkea_data *ldata = &info->mti_link_data;
4349 struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj);
4351 struct lu_buf *mgr_buf;
4357 mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
4358 mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize);
4359 if (mgr_buf->lb_buf == NULL)
4362 rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV);
4364 union lmv_mds_md *lmm = mgr_buf->lb_buf;
4366 /* If the object has migrateEA, it means IMMUTE flag
4367 * is being set by previous migration process, so it
4368 * needs to override the IMMUTE flag, otherwise the
4369 * following sanity check will fail */
4370 if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) &
4371 LMV_HASH_FLAG_MIGRATION) {
4372 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
4374 sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL;
4375 CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n",
4376 mdd2obd_dev(mdd)->obd_name,
4377 PFID(mdd_object_fid(sobj)));
4381 rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr,
4382 sobj, sattr, NULL, NULL);
4386 /* Then it will check if the file should be migrated. If the file
4387 * has mulitple links, we only need migrate the file if all of its
4388 * entries has been migrated to the remote MDT */
4389 if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2)
4392 rc = mdd_links_read(env, sobj, ldata);
4394 /* For multiple links files, if there are no linkEA data at all,
4395 * means the file might be created before linkEA is enabled, and
4396 * all of its links should not be migrated yet, otherwise it
4397 * should have some linkEA there */
4398 if (rc == -ENOENT || rc == -ENODATA)
4403 mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id;
4404 /* If there are still links locally, then the file will not be
4406 LASSERT(ldata->ld_leh != NULL);
4408 /* If the linkEA is overflow, then means there are some unknown name
4409 * entries under unknown parents, that will prevent the migration. */
4410 if (unlikely(ldata->ld_leh->leh_overflow_time))
4413 ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
4414 for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
4415 struct lu_name lname;
4417 __u32 parent_mdt_index;
4419 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen,
4421 ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee +
4424 rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index);
4428 /* Migrate the object only if none of its parents are on the
4430 if (parent_mdt_index != mdt_index)
4433 CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n",
4434 PFID(mdd_object_fid(sobj)), lname.ln_namelen,
4435 lname.ln_name, PFID(&fid));
4443 static int mdd_migrate(const struct lu_env *env, struct md_object *pobj,
4444 struct md_object *sobj, const struct lu_name *lname,
4445 struct md_object *tobj, struct md_attr *ma)
4447 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
4448 struct mdd_device *mdd = mdo2mdd(pobj);
4449 struct mdd_object *mdd_sobj = md2mdd_obj(sobj);
4450 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
4451 struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
4452 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
4453 bool created = false;
4457 /* If the file will being migrated, it will check whether
4458 * the file is being opened by someone else right now */
4459 mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
4460 if (mdd_sobj->mod_count > 0) {
4462 "%s: "DFID"%s is already opened count %d: rc = %d\n",
4463 mdd2obd_dev(mdd)->obd_name,
4464 PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
4465 mdd_sobj->mod_count, -EBUSY);
4466 mdd_read_unlock(env, mdd_sobj);
4467 GOTO(put, rc = -EBUSY);
4469 mdd_read_unlock(env, mdd_sobj);
4471 rc = mdd_la_get(env, mdd_sobj, so_attr);
4475 rc = mdd_la_get(env, mdd_pobj, pattr);
4479 rc = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr);
4486 /* Sigh, it is impossible to finish all of migration in a single
4487 * transaction, for example migrating big directory entries to the
4488 * new MDT, it needs insert all of name entries of children in the
4491 * So migration will be done in multiple steps and transactions.
4493 * 1. create an orphan object on the remote MDT in one transaction.
4494 * 2. migrate extend attributes to the new target file/directory.
4495 * 3. For directory, migrate the entries to the new MDT and update
4496 * linkEA of each children. Because we can not migrate all entries
4497 * in a single transaction, so the migrating directory will become
4498 * a striped directory during migration, so once the process is
4499 * interrupted, the directory is still accessible. (During lookup,
4500 * client will locate the name by searching both original and target
4502 * 4. Finally, update the name/FID to point to the new file/directory
4503 * in a separate transaction.
4506 /* step 1: Check whether the orphan object has been created, and create
4507 * orphan object on the remote MDT if needed */
4508 if (!mdd_object_exists(mdd_tobj)) {
4509 rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
4516 LASSERT(mdd_object_exists(mdd_tobj));
4517 /* step 2: migrate xattr */
4518 rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj);
4522 /* step 3: migrate name entries to the orphan object */
4523 if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) {
4524 rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj);
4527 if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP,
4528 OBD_FAIL_MDS_REINT_NET_REP)))
4531 OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val);
4534 LASSERT(mdd_object_exists(mdd_tobj));
4535 /* step 4: update name entry to the new object */
4536 rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname,
4541 /* newly created target was not locked, don't cache its attributes */
4543 mdd_invalidate(env, tobj);
4548 const struct md_dir_operations mdd_dir_ops = {
4549 .mdo_is_subdir = mdd_is_subdir,
4550 .mdo_lookup = mdd_lookup,
4551 .mdo_create = mdd_create,
4552 .mdo_rename = mdd_rename,
4553 .mdo_link = mdd_link,
4554 .mdo_unlink = mdd_unlink,
4555 .mdo_create_data = mdd_create_data,
4556 .mdo_migrate = mdd_migrate,