1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/mdd/mdd_dir.c
40 * Lustre Metadata Server (mdd) routines
42 * Author: Wang Di <wangdi@clusterfs.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_MDS
50 #include <linux/module.h>
51 #ifdef HAVE_EXT4_LDISKFS
52 #include <ldiskfs/ldiskfs_jbd2.h>
54 #include <linux/jbd.h>
57 #include <obd_class.h>
58 #include <lustre_ver.h>
59 #include <obd_support.h>
60 #include <lprocfs_status.h>
61 #ifdef HAVE_EXT4_LDISKFS
62 #include <ldiskfs/ldiskfs.h>
64 #include <linux/ldiskfs_fs.h>
66 #include <lustre_mds.h>
67 #include <lustre/lustre_idl.h>
68 #include <lustre_fid.h>
70 #include "mdd_internal.h"
72 static const char dot[] = ".";
73 static const char dotdot[] = "..";
75 static struct lu_name lname_dotdot = {
80 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
81 const struct lu_name *lname, struct lu_fid* fid,
83 static int mdd_declare_links_add(const struct lu_env *env,
84 struct mdd_object *mdd_obj,
85 struct thandle *handle);
86 static int mdd_links_add(const struct lu_env *env,
87 struct mdd_object *mdd_obj,
88 const struct lu_fid *pfid,
89 const struct lu_name *lname,
90 struct thandle *handle, int first);
91 static int mdd_links_rename(const struct lu_env *env,
92 struct mdd_object *mdd_obj,
93 const struct lu_fid *oldpfid,
94 const struct lu_name *oldlname,
95 const struct lu_fid *newpfid,
96 const struct lu_name *newlname,
97 struct thandle *handle);
100 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
101 const struct lu_name *lname, struct lu_fid* fid, int mask)
103 const char *name = lname->ln_name;
104 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
105 struct dynlock_handle *dlh;
108 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
109 if (unlikely(dlh == NULL))
111 rc = __mdd_lookup(env, pobj, lname, fid, mask);
112 mdd_pdo_read_unlock(env, mdd_obj, dlh);
117 int mdd_lookup(const struct lu_env *env,
118 struct md_object *pobj, const struct lu_name *lname,
119 struct lu_fid* fid, struct md_op_spec *spec)
123 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
127 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
130 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
134 * For root fid use special function, which does not compare version component
135 * of fid. Version component is different for root fids on all MDTs.
137 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
139 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
140 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
144 * return 1: if lf is the fid of the ancestor of p1;
147 * return -EREMOTE: if remote object is found, in this
148 * case fid of remote object is saved to @pf;
150 * otherwise: values < 0, errors.
152 static int mdd_is_parent(const struct lu_env *env,
153 struct mdd_device *mdd,
154 struct mdd_object *p1,
155 const struct lu_fid *lf,
158 struct mdd_object *parent = NULL;
163 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
164 pfid = &mdd_env_info(env)->mti_fid;
166 /* Check for root first. */
167 if (mdd_is_root(mdd, mdo2fid(p1)))
171 /* this is done recursively, bypass capa for each obj */
172 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
173 rc = mdd_parent_fid(env, p1, pfid);
176 if (mdd_is_root(mdd, pfid))
178 if (lu_fid_eq(pfid, lf))
181 mdd_object_put(env, parent);
182 parent = mdd_object_find(env, mdd, pfid);
184 /* cross-ref parent */
185 if (parent == NULL) {
188 GOTO(out, rc = -EREMOTE);
189 } else if (IS_ERR(parent))
190 GOTO(out, rc = PTR_ERR(parent));
195 if (parent && !IS_ERR(parent))
196 mdd_object_put(env, parent);
201 * No permission check is needed.
203 * returns 1: if fid is ancestor of @mo;
204 * returns 0: if fid is not a ancestor of @mo;
206 * returns EREMOTE if remote object is found, fid of remote object is saved to
209 * returns < 0: if error
211 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
212 const struct lu_fid *fid, struct lu_fid *sfid)
214 struct mdd_device *mdd = mdo2mdd(mo);
218 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
221 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
225 } else if (rc == 1) {
226 /* found @fid is parent */
234 * Check that @dir contains no entries except (possibly) dot and dotdot.
239 * -ENOTDIR not a directory object
240 * -ENOTEMPTY not empty
244 static int mdd_dir_is_empty(const struct lu_env *env,
245 struct mdd_object *dir)
248 struct dt_object *obj;
249 const struct dt_it_ops *iops;
253 obj = mdd_object_child(dir);
254 if (!dt_try_as_dir(env, obj))
257 iops = &obj->do_index_ops->dio_it;
258 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
260 result = iops->get(env, it, (const void *)"");
263 for (result = 0, i = 0; result == 0 && i < 3; ++i)
264 result = iops->next(env, it);
267 else if (result == +1)
269 } else if (result == 0)
271 * Huh? Index contains no zero key?
278 result = PTR_ERR(it);
282 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
284 struct mdd_device *m = mdd_obj2mdd_dev(obj);
285 struct lu_attr *la = &mdd_env_info(env)->mti_la;
289 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
294 * Subdir count limitation can be broken through.
296 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
297 !S_ISDIR(la->la_mode))
304 * Check whether it may create the cobj under the pobj.
307 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
308 struct mdd_object *cobj, int check_perm, int check_nlink)
313 if (cobj && mdd_object_exists(cobj))
316 if (mdd_is_dead_obj(pobj))
320 rc = mdd_permission_internal_locked(env, pobj, NULL,
321 MAY_WRITE | MAY_EXEC,
324 if (!rc && check_nlink)
325 rc = __mdd_may_link(env, pobj);
331 * Check whether can unlink from the pobj in the case of "cobj == NULL".
333 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
334 const struct md_attr *ma)
339 if (mdd_is_dead_obj(pobj))
342 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
343 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
346 rc = mdd_permission_internal_locked(env, pobj, NULL,
347 MAY_WRITE | MAY_EXEC,
352 if (mdd_is_append(pobj))
359 * pobj == NULL is remote ops case, under such case, pobj's
360 * VTX feature has been checked already, no need check again.
362 static inline int mdd_is_sticky(const struct lu_env *env,
363 struct mdd_object *pobj,
364 struct mdd_object *cobj)
366 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
367 struct md_ucred *uc = md_ucred(env);
371 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
375 if (!(tmp_la->la_mode & S_ISVTX) ||
376 (tmp_la->la_uid == uc->mu_fsuid))
380 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
384 if (tmp_la->la_uid == uc->mu_fsuid)
387 return !mdd_capable(uc, CFS_CAP_FOWNER);
391 * Check whether it may delete the cobj from the pobj.
394 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
395 struct mdd_object *cobj, struct md_attr *ma,
396 int check_perm, int check_empty)
402 if (!mdd_object_exists(cobj))
405 if (mdd_is_dead_obj(cobj))
409 if (!mdd_object_exists(pobj))
412 if (mdd_is_dead_obj(pobj))
416 rc = mdd_permission_internal_locked(env, pobj, NULL,
417 MAY_WRITE | MAY_EXEC,
423 if (mdd_is_append(pobj))
427 if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) &&
428 mdd_is_sticky(env, pobj, cobj))
431 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
434 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
435 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
438 if (S_ISDIR(ma->ma_attr.la_mode)) {
439 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
441 if (!S_ISDIR(mdd_object_type(cobj)))
444 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
446 } else if (S_ISDIR(mdd_object_type(cobj)))
449 if (S_ISDIR(ma->ma_attr.la_mode) && check_empty)
450 rc = mdd_dir_is_empty(env, cobj);
457 * has mdd_write_lock on src already, but not on tgt yet
459 int mdd_link_sanity_check(const struct lu_env *env,
460 struct mdd_object *tgt_obj,
461 const struct lu_name *lname,
462 struct mdd_object *src_obj)
464 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
468 if (!mdd_object_exists(src_obj))
471 if (mdd_is_dead_obj(src_obj))
474 /* Local ops, no lookup before link, check filename length here. */
475 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
476 RETURN(-ENAMETOOLONG);
478 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
481 if (S_ISDIR(mdd_object_type(src_obj)))
484 LASSERT(src_obj != tgt_obj);
486 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
491 rc = __mdd_may_link(env, src_obj);
496 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
497 const char *name, struct thandle *handle,
498 struct lustre_capa *capa)
500 struct dt_object *next = mdd_object_child(pobj);
504 if (dt_try_as_dir(env, next)) {
505 rc = next->do_index_ops->dio_delete(env, next,
506 (struct dt_key *)name,
514 static int __mdd_index_insert_only(const struct lu_env *env,
515 struct mdd_object *pobj,
516 const struct lu_fid *lf, const char *name,
517 struct thandle *handle,
518 struct lustre_capa *capa)
520 struct dt_object *next = mdd_object_child(pobj);
524 if (dt_try_as_dir(env, next)) {
525 struct md_ucred *uc = md_ucred(env);
527 rc = next->do_index_ops->dio_insert(env, next,
529 (const struct dt_key *)name,
530 handle, capa, uc->mu_cap &
531 CFS_CAP_SYS_RESOURCE_MASK);
538 /* insert named index, add reference if isdir */
539 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
540 const struct lu_fid *lf, const char *name, int is_dir,
541 struct thandle *handle, struct lustre_capa *capa)
546 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
547 if (rc == 0 && is_dir) {
548 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
549 mdo_ref_add(env, pobj, handle);
550 mdd_write_unlock(env, pobj);
555 /* delete named index, drop reference if isdir */
556 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
557 const char *name, int is_dir, struct thandle *handle,
558 struct lustre_capa *capa)
563 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
564 if (rc == 0 && is_dir) {
565 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
566 mdo_ref_del(env, pobj, handle);
567 mdd_write_unlock(env, pobj);
573 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
574 int reclen, struct thandle *handle)
578 /* XXX: this is a temporary solution to declare llog changes
579 * will be fixed in 2.3 with new llog implementation */
581 LASSERT(mdd->mdd_capa);
583 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
584 * have to set the parameter 'size' as INT_MAX or 0 to inform
585 * OSD that this record write is for a llog write or catalog
586 * header update, and osd declare function will reserve less
587 * credits for optimization purpose.
589 * Reserve 6 blocks for a llog write, since the llog file is
590 * usually small, reserve 2 blocks for catalog header update,
591 * because we know for sure that catalog header is already
594 * This hack should be removed in 2.3.
598 rc = dt_declare_record_write(env, mdd->mdd_capa,
599 DECLARE_LLOG_WRITE, 0, handle);
603 /* header will be updated as well */
604 rc = dt_declare_record_write(env, mdd->mdd_capa,
605 DECLARE_LLOG_WRITE, 0, handle);
609 /* also we should be able to create new plain log */
610 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
614 /* new record referencing new plain llog */
615 rc = dt_declare_record_write(env, mdd->mdd_capa,
616 DECLARE_LLOG_WRITE, 0, handle);
620 /* catalog's header will be updated as well */
621 rc = dt_declare_record_write(env, mdd->mdd_capa,
622 DECLARE_LLOG_REWRITE, 0, handle);
627 int mdd_declare_changelog_store(const struct lu_env *env,
628 struct mdd_device *mdd,
629 const struct lu_name *fname,
630 struct thandle *handle)
635 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
638 /* we'll be writing payload + llog header */
639 reclen = sizeof(struct llog_changelog_rec);
641 reclen += llog_data_len(fname->ln_namelen);
643 return mdd_declare_llog_record(env, mdd, reclen, handle);
646 /** Store a namespace change changelog record
647 * If this fails, we must fail the whole transaction; we don't
648 * want the change to commit without the log entry.
649 * \param target - mdd_object of change
650 * \param parent - parent dir/object
651 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
652 * \param tname - target name string
653 * \param handle - transacion handle
655 static int mdd_changelog_ns_store(const struct lu_env *env,
656 struct mdd_device *mdd,
657 enum changelog_rec_type type,
659 struct mdd_object *target,
660 struct mdd_object *parent,
661 const struct lu_fid *tf,
662 const struct lu_name *tname,
663 struct thandle *handle)
665 const struct lu_fid *tfid;
666 const struct lu_fid *tpfid = mdo2fid(parent);
667 struct llog_changelog_rec *rec;
674 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
676 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
679 LASSERT(parent != NULL);
680 LASSERT(tname != NULL);
681 LASSERT(handle != NULL);
684 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
685 buf = mdd_buf_alloc(env, reclen);
686 if (buf->lb_buf == NULL)
688 rec = (struct llog_changelog_rec *)buf->lb_buf;
690 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
691 rec->cr.cr_type = (__u32)type;
692 tfid = tf ? tf : mdo2fid(target);
693 rec->cr.cr_tfid = *tfid;
694 rec->cr.cr_pfid = *tpfid;
695 rec->cr.cr_namelen = tname->ln_namelen;
696 memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
698 target->mod_cltime = cfs_time_current_64();
700 rc = mdd_changelog_llog_write(mdd, rec, handle);
702 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
703 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
710 static int mdd_declare_link(const struct lu_env *env,
711 struct mdd_device *mdd,
712 struct mdd_object *p,
713 struct mdd_object *c,
714 const struct lu_name *name,
715 struct thandle *handle)
719 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
723 rc = mdo_declare_ref_add(env, c, handle);
727 rc = mdo_declare_attr_set(env, p, NULL, handle);
731 rc = mdo_declare_attr_set(env, c, NULL, handle);
735 rc = mdd_declare_links_add(env, c, handle);
739 rc = mdd_declare_changelog_store(env, mdd, name, handle);
744 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
745 struct md_object *src_obj, const struct lu_name *lname,
748 const char *name = lname->ln_name;
749 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
750 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
751 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
752 struct mdd_device *mdd = mdo2mdd(src_obj);
753 struct dynlock_handle *dlh;
754 struct thandle *handle;
755 #ifdef HAVE_QUOTA_SUPPORT
756 struct obd_device *obd = mdd->mdd_obd_dev;
757 struct obd_export *exp = md_quota(env)->mq_exp;
758 struct mds_obd *mds = &obd->u.mds;
759 unsigned int qids[MAXQUOTAS] = { 0, 0 };
760 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
765 #ifdef HAVE_QUOTA_SUPPORT
766 if (mds->mds_quota) {
767 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
769 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
772 mdd_data_get(env, mdd_tobj, &data);
773 quota_opc = FSFILT_OP_LINK;
774 mdd_quota_wrapper(la_tmp, qids);
775 /* get block quota for parent */
776 lquota_chkquota(mds_quota_interface_ref, obd, exp,
777 qids, rec_pending, 1, NULL,
778 LQUOTA_FLAGS_BLK, data, 1);
783 handle = mdd_trans_create(env, mdd);
785 GOTO(out_pending, rc = PTR_ERR(handle));
787 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
791 rc = mdd_trans_start(env, mdd, handle);
795 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
797 GOTO(out_trans, rc = -ENOMEM);
798 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
800 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
802 GOTO(out_unlock, rc);
804 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
806 mdd_object_capa(env, mdd_tobj));
808 GOTO(out_unlock, rc);
810 mdo_ref_add(env, mdd_sobj, handle);
812 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
813 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
815 la->la_valid = LA_CTIME | LA_MTIME;
816 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
818 GOTO(out_unlock, rc);
820 la->la_valid = LA_CTIME;
821 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
823 mdd_links_add(env, mdd_sobj,
824 mdo2fid(mdd_tobj), lname, handle, 0);
829 mdd_write_unlock(env, mdd_sobj);
830 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
833 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
834 mdd_tobj, NULL, lname, handle);
836 mdd_trans_stop(env, mdd, rc, handle);
838 #ifdef HAVE_QUOTA_SUPPORT
840 lquota_pending_commit(mds_quota_interface_ref, obd,
841 qids, rec_pending, 1);
842 /* Trigger dqacq for the parent owner. If failed,
843 * the next call for lquota_chkquota will process it. */
844 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
851 int mdd_declare_finish_unlink(const struct lu_env *env,
852 struct mdd_object *obj,
854 struct thandle *handle)
858 rc = orph_declare_index_insert(env, obj, handle);
862 return mdd_declare_object_kill(env, obj, ma, handle);
865 /* caller should take a lock before calling */
866 int mdd_finish_unlink(const struct lu_env *env,
867 struct mdd_object *obj, struct md_attr *ma,
872 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
875 LASSERT(mdd_write_locked(env, obj) != 0);
877 /* read HSM flags, needed to set changelogs flags */
878 ma->ma_need = MA_HSM | MA_INODE;
879 rc = mdd_attr_get_internal(env, obj, ma);
880 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
881 obj->mod_flags |= DEAD_OBJ;
882 /* add new orphan and the object
883 * will be deleted during mdd_close() */
884 if (obj->mod_count) {
885 rc = __mdd_orphan_add(env, obj, th);
887 CDEBUG(D_HA, "Object "DFID" is inserted into "
888 "orphan list, open count = %d\n",
889 PFID(mdd_object_fid(obj)),
892 CERROR("Object "DFID" fail to be an orphan, "
893 "open count = %d, maybe cause failed "
895 PFID(mdd_object_fid(obj)),
898 rc = mdd_object_kill(env, obj, ma, th);
903 /* get the i_nlink */
904 ma->ma_need = MA_INODE;
905 rc = mdd_attr_get_internal(env, obj, ma);
908 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
915 * has mdd_write_lock on cobj already, but not on pobj yet
917 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
918 struct mdd_object *cobj, struct md_attr *ma)
923 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
928 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
929 struct mdd_object *p, struct mdd_object *c,
930 const struct lu_name *name, struct md_attr *ma,
931 struct thandle *handle)
935 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
939 rc = mdo_declare_ref_del(env, p, handle);
943 rc = mdo_declare_ref_del(env, c, handle);
947 rc = mdo_declare_ref_del(env, c, handle);
951 rc = mdo_declare_attr_set(env, p, NULL, handle);
955 rc = mdo_declare_attr_set(env, c, NULL, handle);
959 rc = mdd_declare_finish_unlink(env, c, ma, handle);
963 rc = mdd_declare_links_add(env, c, handle);
967 rc = mdd_declare_changelog_store(env, mdd, name, handle);
972 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
973 struct md_object *cobj, const struct lu_name *lname,
976 const char *name = lname->ln_name;
977 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
978 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
979 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
980 struct mdd_device *mdd = mdo2mdd(pobj);
981 struct dynlock_handle *dlh;
982 struct thandle *handle;
983 #ifdef HAVE_QUOTA_SUPPORT
984 struct obd_device *obd = mdd->mdd_obd_dev;
985 struct mds_obd *mds = &obd->u.mds;
986 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
987 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
990 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
994 if (mdd_object_exists(mdd_cobj) <= 0)
997 handle = mdd_trans_create(env, mdd);
999 RETURN(PTR_ERR(handle));
1001 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1006 rc = mdd_trans_start(env, mdd, handle);
1010 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1012 GOTO(out_trans, rc = -ENOMEM);
1013 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1015 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
1019 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1020 mdd_object_capa(env, mdd_pobj));
1024 mdo_ref_del(env, mdd_cobj, handle);
1027 mdo_ref_del(env, mdd_cobj, handle);
1029 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1030 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1032 la->la_valid = LA_CTIME | LA_MTIME;
1033 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1037 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
1038 /* update ctime of an unlinked file only if it is still
1039 * opened or a link still exists */
1040 la->la_valid = LA_CTIME;
1041 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1046 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1047 #ifdef HAVE_QUOTA_SUPPORT
1048 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1049 ma->ma_attr.la_nlink == 0) {
1050 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1052 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1054 mdd_quota_wrapper(la_tmp, qpids);
1055 if (mdd_cobj->mod_count == 0) {
1056 quota_opc = FSFILT_OP_UNLINK;
1057 mdd_quota_wrapper(&ma->ma_attr, qcids);
1059 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1065 /* old files may not have link ea; ignore errors */
1066 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1067 lname, NULL, NULL, handle);
1071 mdd_write_unlock(env, mdd_cobj);
1072 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1077 cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1078 if ((ma->ma_valid & MA_HSM) &&
1079 (ma->ma_hsm.mh_flags & HS_EXISTS))
1080 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1082 rc = mdd_changelog_ns_store(env, mdd,
1083 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1084 mdd_cobj, mdd_pobj, NULL, lname, handle);
1088 mdd_trans_stop(env, mdd, rc, handle);
1089 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
1090 if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV &&
1091 ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY)
1092 /* Since echo client is incapable of destorying ost object,
1093 * it will destory the object here. */
1094 rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la,
1095 ma->ma_lmm, ma->ma_lmm_size,
1098 #warning "please remove this after 2.4 (LOD/OSP)."
1101 #ifdef HAVE_QUOTA_SUPPORT
1103 /* Trigger dqrel on the owner of child and parent. If failed,
1104 * the next call for lquota_chkquota will process it. */
1105 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1111 /* has not lock on pobj yet */
1112 static int mdd_ni_sanity_check(const struct lu_env *env,
1113 struct md_object *pobj,
1114 const struct md_attr *ma)
1116 struct mdd_object *obj = md2mdd_obj(pobj);
1120 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1123 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
1129 * Partial operation.
1131 static int mdd_name_insert(const struct lu_env *env,
1132 struct md_object *pobj,
1133 const struct lu_name *lname,
1134 const struct lu_fid *fid,
1135 const struct md_attr *ma)
1137 const char *name = lname->ln_name;
1138 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1139 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1140 struct mdd_device *mdd = mdo2mdd(pobj);
1141 struct dynlock_handle *dlh;
1142 struct thandle *handle;
1143 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1144 #ifdef HAVE_QUOTA_SUPPORT
1145 struct md_ucred *uc = md_ucred(env);
1146 struct obd_device *obd = mdd->mdd_obd_dev;
1147 struct obd_export *exp = md_quota(env)->mq_exp;
1148 struct mds_obd *mds = &obd->u.mds;
1149 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1150 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
1151 cfs_cap_t save = uc->mu_cap;
1156 /* XXX: this code won't be used ever:
1157 * DNE uses slightly different approach */
1160 #ifdef HAVE_QUOTA_SUPPORT
1161 if (mds->mds_quota) {
1162 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
1163 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1165 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1168 mdd_data_get(env, mdd_obj, &data);
1169 quota_opc = FSFILT_OP_LINK;
1170 mdd_quota_wrapper(la_tmp, qids);
1171 /* get block quota for parent */
1172 lquota_chkquota(mds_quota_interface_ref, obd,
1173 exp, qids, rec_pending, 1, NULL,
1174 LQUOTA_FLAGS_BLK, data, 1);
1177 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1181 handle = mdd_trans_create(env, mdo2mdd(pobj));
1183 GOTO(out_pending, rc = PTR_ERR(handle));
1185 rc = mdd_trans_start(env, mdo2mdd(pobj), handle);
1187 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1189 GOTO(out_trans, rc = -ENOMEM);
1191 rc = mdd_ni_sanity_check(env, pobj, ma);
1193 GOTO(out_unlock, rc);
1195 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1196 handle, BYPASS_CAPA);
1198 GOTO(out_unlock, rc);
1201 * For some case, no need update obj's ctime (LA_CTIME is not set),
1203 * For other cases, update obj's ctime (LA_CTIME is set),
1206 if (ma->ma_attr.la_valid & LA_CTIME) {
1207 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1208 la->la_valid = LA_CTIME | LA_MTIME;
1209 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1214 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1216 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1218 #ifdef HAVE_QUOTA_SUPPORT
1219 if (mds->mds_quota) {
1221 lquota_pending_commit(mds_quota_interface_ref,
1222 obd, qids, rec_pending, 1);
1223 /* Trigger dqacq for the parent owner. If failed,
1224 * the next call for lquota_chkquota will process it*/
1225 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1235 /* has not lock on pobj yet */
1236 static int mdd_nr_sanity_check(const struct lu_env *env,
1237 struct md_object *pobj,
1238 const struct md_attr *ma)
1240 struct mdd_object *obj = md2mdd_obj(pobj);
1244 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1247 rc = mdd_may_unlink(env, obj, ma);
1253 * Partial operation.
1255 static int mdd_name_remove(const struct lu_env *env,
1256 struct md_object *pobj,
1257 const struct lu_name *lname,
1258 const struct md_attr *ma)
1260 const char *name = lname->ln_name;
1261 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1262 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1263 struct mdd_device *mdd = mdo2mdd(pobj);
1264 struct dynlock_handle *dlh;
1265 struct thandle *handle;
1266 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1267 #ifdef HAVE_QUOTA_SUPPORT
1268 struct obd_device *obd = mdd->mdd_obd_dev;
1269 struct mds_obd *mds = &obd->u.mds;
1270 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1276 /* XXX: this code won't be used ever:
1277 * DNE uses slightly different approach */
1280 #ifdef HAVE_QUOTA_SUPPORT
1281 if (mds->mds_quota) {
1282 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1284 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1286 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1287 mdd_quota_wrapper(la_tmp, qids);
1291 handle = mdd_trans_create(env, mdd);
1293 GOTO(out_pending, rc = PTR_ERR(handle));
1295 rc = mdd_trans_start(env, mdd, handle);
1297 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1299 GOTO(out_trans, rc = -ENOMEM);
1301 rc = mdd_nr_sanity_check(env, pobj, ma);
1303 GOTO(out_unlock, rc);
1305 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1306 handle, BYPASS_CAPA);
1308 GOTO(out_unlock, rc);
1311 * For some case, no need update obj's ctime (LA_CTIME is not set),
1313 * For other cases, update obj's ctime (LA_CTIME is set),
1316 if (ma->ma_attr.la_valid & LA_CTIME) {
1317 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1318 la->la_valid = LA_CTIME | LA_MTIME;
1319 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1324 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1326 mdd_trans_stop(env, mdd, rc, handle);
1328 #ifdef HAVE_QUOTA_SUPPORT
1329 /* Trigger dqrel for the parent owner.
1330 * If failed, the next call for lquota_chkquota will process it. */
1332 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1340 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1342 static int mdd_rt_sanity_check(const struct lu_env *env,
1343 struct mdd_object *tgt_pobj,
1344 struct mdd_object *tobj,
1350 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1353 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1354 * exist. In fact, tobj must exist, otherwise the call trace will be:
1355 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1356 * When get here, tobj must be NOT NULL, the other case has been
1357 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1359 * So check may_delete, but not check nlink of tgt_pobj. */
1361 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1366 /* Partial rename op on slave MDD */
1367 static int mdd_rename_tgt(const struct lu_env *env,
1368 struct md_object *pobj, struct md_object *tobj,
1369 const struct lu_fid *lf, const struct lu_name *lname,
1372 const char *name = lname->ln_name;
1373 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1374 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1375 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1376 struct mdd_device *mdd = mdo2mdd(pobj);
1377 struct dynlock_handle *dlh;
1378 struct thandle *handle;
1379 #ifdef HAVE_QUOTA_SUPPORT
1380 struct obd_device *obd = mdd->mdd_obd_dev;
1381 struct obd_export *exp = md_quota(env)->mq_exp;
1382 struct mds_obd *mds = &obd->u.mds;
1383 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1384 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1385 int quota_copc = 0, quota_popc = 0;
1386 int rec_pending[MAXQUOTAS] = { 0, 0 };
1391 /* XXX: this code won't be used ever:
1392 * DNE uses slightly different approach */
1395 #ifdef HAVE_QUOTA_SUPPORT
1396 if (mds->mds_quota && !tobj) {
1397 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1399 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1402 mdd_data_get(env, mdd_tpobj, &data);
1403 quota_popc = FSFILT_OP_LINK;
1404 mdd_quota_wrapper(la_tmp, qpids);
1405 /* get block quota for target parent */
1406 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1407 qpids, rec_pending, 1, NULL,
1408 LQUOTA_FLAGS_BLK, data, 1);
1412 handle = mdd_trans_create(env, mdd);
1414 GOTO(out_pending, rc = PTR_ERR(handle));
1416 rc = mdd_trans_start(env, mdd, handle);
1418 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1420 GOTO(out_trans, rc = -ENOMEM);
1422 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1424 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1429 * If rename_tgt is called then we should just re-insert name with
1430 * correct fid, no need to dec/inc parent nlink if obj is dir.
1432 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1436 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1441 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1442 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1444 la->la_valid = LA_CTIME | LA_MTIME;
1445 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1450 * For tobj is remote case cmm layer has processed
1451 * and pass NULL tobj to here. So when tobj is NOT NULL,
1452 * it must be local one.
1454 if (tobj && mdd_object_exists(mdd_tobj)) {
1455 mdo_ref_del(env, mdd_tobj, handle);
1457 /* Remove dot reference. */
1458 if (S_ISDIR(ma->ma_attr.la_mode))
1459 mdo_ref_del(env, mdd_tobj, handle);
1461 la->la_valid = LA_CTIME;
1462 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1466 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1470 #ifdef HAVE_QUOTA_SUPPORT
1471 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1472 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1473 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1474 mdd_quota_wrapper(&ma->ma_attr, qcids);
1481 mdd_write_unlock(env, mdd_tobj);
1482 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1485 /* Bare EXT record with no RENAME in front of it signifies
1486 a partial slave op */
1487 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
1488 mdd_tpobj, NULL, lname, handle);
1490 mdd_trans_stop(env, mdd, rc, handle);
1492 #ifdef HAVE_QUOTA_SUPPORT
1493 if (mds->mds_quota) {
1495 lquota_pending_commit(mds_quota_interface_ref, obd,
1496 qpids, rec_pending, 1);
1499 /* Trigger dqrel on the target owner of child.
1500 * If failed, the next call for lquota_chkquota
1501 * will process it. */
1502 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1510 * The permission has been checked when obj created, no need check again.
1512 static int mdd_cd_sanity_check(const struct lu_env *env,
1513 struct mdd_object *obj)
1518 if (!obj || mdd_is_dead_obj(obj))
1525 static int mdd_declare_create_data(const struct lu_env *env,
1526 struct mdd_device *mdd,
1527 struct mdd_object *obj,
1529 struct thandle *handle)
1531 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1535 buf->lb_len = lmm_size;
1536 rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
1541 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1546 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1547 struct md_object *cobj, const struct md_op_spec *spec,
1550 struct mdd_device *mdd = mdo2mdd(cobj);
1551 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1552 struct mdd_object *son = md2mdd_obj(cobj);
1553 struct lov_mds_md *lmm = NULL;
1555 struct thandle *handle;
1559 rc = mdd_cd_sanity_check(env, son);
1563 if (!md_should_create(spec->sp_cr_flags))
1565 lmm_size = ma->ma_lmm_size;
1567 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma);
1571 handle = mdd_trans_create(env, mdd);
1573 GOTO(out_free, rc = PTR_ERR(handle));
1575 rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle);
1579 rc = mdd_trans_start(env, mdd, handle);
1584 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1585 * Should this be fixed?
1588 /* Replay creates has objects already */
1590 if (spec->no_create) {
1591 CDEBUG(D_INFO, "we already have lov ea\n");
1592 rc = mdd_lov_set_md(env, mdd_pobj, son,
1593 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1594 spec->u.sp_ea.eadatalen, handle, 0);
1597 /* No need mdd_lsm_sanity_check here */
1598 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1599 lmm_size, handle, 0);
1602 rc = mdd_attr_get_internal_locked(env, son, ma);
1604 /* update lov_objid data, must be before transaction stop! */
1606 mdd_lov_objid_update(mdd, lmm);
1609 mdd_trans_stop(env, mdd, rc, handle);
1611 /* Finish mdd_lov_create() stuff. */
1612 /* if no_create == 0 (not replay), we free lmm allocated by
1613 * mdd_lov_create() */
1614 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1618 /* Get fid from name and parent */
1620 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1621 const struct lu_name *lname, struct lu_fid* fid, int mask)
1623 const char *name = lname->ln_name;
1624 const struct dt_key *key = (const struct dt_key *)name;
1625 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1626 struct mdd_device *m = mdo2mdd(pobj);
1627 struct dt_object *dir = mdd_object_child(mdd_obj);
1631 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1634 rc = mdd_object_exists(mdd_obj);
1635 if (unlikely(rc == 0))
1637 else if (unlikely(rc < 0)) {
1638 CERROR("Object "DFID" locates on remote server\n",
1639 PFID(mdo2fid(mdd_obj)));
1643 /* The common filename length check. */
1644 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1645 RETURN(-ENAMETOOLONG);
1647 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1652 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1653 dt_try_as_dir(env, dir))) {
1655 rc = dir->do_index_ops->dio_lookup(env, dir,
1656 (struct dt_rec *)fid, key,
1657 mdd_object_capa(env, mdd_obj));
1668 int mdd_declare_object_initialize(const struct lu_env *env,
1669 struct mdd_object *child,
1671 struct thandle *handle)
1675 rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle);
1676 if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) {
1677 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1680 rc = mdo_declare_ref_add(env, child, handle);
1683 mdd_declare_links_add(env, child, handle);
1688 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1689 const struct lu_name *lname, struct mdd_object *child,
1690 struct md_attr *ma, struct thandle *handle,
1691 const struct md_op_spec *spec)
1697 * Update attributes for child.
1700 * (1) the valid bits should be converted between Lustre and Linux;
1701 * (2) maybe, the child attributes should be set in OSD when creation.
1704 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1708 if (S_ISDIR(ma->ma_attr.la_mode)) {
1709 /* Add "." and ".." for newly created dir */
1710 mdo_ref_add(env, child, handle);
1711 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1712 dot, handle, BYPASS_CAPA);
1714 rc = __mdd_index_insert_only(env, child, pfid,
1718 mdo_ref_del(env, child, handle);
1721 mdd_links_add(env, child, pfid, lname, handle, 1);
1726 /* has not lock on pobj yet */
1727 static int mdd_create_sanity_check(const struct lu_env *env,
1728 struct md_object *pobj,
1729 const struct lu_name *lname,
1731 struct md_op_spec *spec)
1733 struct mdd_thread_info *info = mdd_env_info(env);
1734 struct lu_attr *la = &info->mti_la;
1735 struct lu_fid *fid = &info->mti_fid;
1736 struct mdd_object *obj = md2mdd_obj(pobj);
1737 struct mdd_device *m = mdo2mdd(pobj);
1738 int lookup = spec->sp_cr_lookup;
1743 if (mdd_is_dead_obj(obj))
1747 * In some cases this lookup is not needed - we know before if name
1748 * exists or not because MDT performs lookup for it.
1749 * name length check is done in lookup.
1753 * Check if the name already exist, though it will be checked in
1754 * _index_insert also, for avoiding rolling back if exists
1757 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1758 MAY_WRITE | MAY_EXEC);
1760 RETURN(rc ? : -EEXIST);
1763 * Check WRITE permission for the parent.
1764 * EXEC permission have been checked
1765 * when lookup before create already.
1767 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1774 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1778 if (la->la_mode & S_ISGID) {
1779 ma->ma_attr.la_gid = la->la_gid;
1780 if (S_ISDIR(ma->ma_attr.la_mode)) {
1781 ma->ma_attr.la_mode |= S_ISGID;
1782 ma->ma_attr.la_valid |= LA_MODE;
1786 switch (ma->ma_attr.la_mode & S_IFMT) {
1788 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1790 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1791 RETURN(-ENAMETOOLONG);
1810 static int mdd_declare_create(const struct lu_env *env,
1811 struct mdd_device *mdd,
1812 struct mdd_object *p,
1813 struct mdd_object *c,
1814 const struct lu_name *name,
1817 struct thandle *handle,
1818 const struct md_op_spec *spec)
1820 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1823 rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec);
1827 /* if dir, then can inherit default ACl */
1829 buf->lb_len = lmm_size;
1830 if (S_ISDIR(ma->ma_attr.la_mode)) {
1831 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT,
1834 rc = mdo_declare_ref_add(env, p, handle);
1839 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS,
1844 rc = mdd_declare_object_initialize(env, c, ma, handle);
1848 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1849 name->ln_name, handle);
1853 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1858 if (S_ISLNK(ma->ma_attr.la_mode)) {
1859 rc = dt_declare_record_write(env, mdd_object_child(c),
1860 strlen(spec->u.sp_symname), 0,
1865 rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle);
1869 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1873 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1881 * Create object and insert it into namespace.
1883 static int mdd_create(const struct lu_env *env,
1884 struct md_object *pobj,
1885 const struct lu_name *lname,
1886 struct md_object *child,
1887 struct md_op_spec *spec,
1890 struct mdd_thread_info *info = mdd_env_info(env);
1891 struct lu_attr *la = &info->mti_la_for_fix;
1892 struct md_attr *ma_acl = &info->mti_ma;
1893 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1894 struct mdd_object *son = md2mdd_obj(child);
1895 struct mdd_device *mdd = mdo2mdd(pobj);
1896 struct lu_attr *attr = &ma->ma_attr;
1897 struct lov_mds_md *lmm = NULL;
1898 struct thandle *handle;
1899 struct dynlock_handle *dlh;
1900 const char *name = lname->ln_name;
1901 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1902 int got_def_acl = 0;
1903 #ifdef HAVE_QUOTA_SUPPORT
1904 struct obd_device *obd = mdd->mdd_obd_dev;
1905 struct obd_export *exp = md_quota(env)->mq_exp;
1906 struct mds_obd *mds = &obd->u.mds;
1907 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1908 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1909 int quota_opc = 0, block_count = 0;
1910 int inode_pending[MAXQUOTAS] = { 0, 0 };
1911 int block_pending[MAXQUOTAS] = { 0, 0 };
1912 int parent_pending[MAXQUOTAS] = { 0, 0 };
1917 * Two operations have to be performed:
1919 * - an allocation of a new object (->do_create()), and
1921 * - an insertion into a parent index (->dio_insert()).
1923 * Due to locking, operation order is not important, when both are
1924 * successful, *but* error handling cases are quite different:
1926 * - if insertion is done first, and following object creation fails,
1927 * insertion has to be rolled back, but this operation might fail
1928 * also leaving us with dangling index entry.
1930 * - if creation is done first, is has to be undone if insertion
1931 * fails, leaving us with leaked space, which is neither good, nor
1934 * It seems that creation-first is simplest solution, but it is
1935 * sub-optimal in the frequent
1940 * case, because second mkdir is bound to create object, only to
1941 * destroy it immediately.
1943 * To avoid this follow local file systems that do double lookup:
1945 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1947 * 1. create (mdd_object_create_internal())
1949 * 2. insert (__mdd_index_insert(), lookup again)
1952 /* Sanity checks before big job. */
1953 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1957 #ifdef HAVE_QUOTA_SUPPORT
1958 if (mds->mds_quota) {
1959 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1961 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1965 quota_opc = FSFILT_OP_CREATE;
1966 mdd_quota_wrapper(&ma->ma_attr, qcids);
1967 mdd_quota_wrapper(la_tmp, qpids);
1968 /* get file quota for child */
1969 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1970 qcids, inode_pending, 1, NULL, 0, NULL,
1972 switch (ma->ma_attr.la_mode & S_IFMT) {
1981 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1982 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1986 /* get block quota for child and parent */
1988 lquota_chkquota(mds_quota_interface_ref, obd,
1989 exp, qcids, block_pending,
1991 LQUOTA_FLAGS_BLK, NULL, 0);
1993 lquota_chkquota(mds_quota_interface_ref, obd,
1994 exp, qpids, parent_pending, 1,
1995 NULL, LQUOTA_FLAGS_BLK, NULL,
2002 * No RPC inside the transaction, so OST objects should be created at
2005 if (S_ISREG(attr->la_mode)) {
2006 lmm_size = ma->ma_lmm_size;
2007 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
2010 GOTO(out_pending, rc);
2013 if (!S_ISLNK(attr->la_mode)) {
2014 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
2015 ma_acl->ma_acl = info->mti_xattr_buf;
2016 ma_acl->ma_need = MA_ACL_DEF;
2017 ma_acl->ma_valid = 0;
2019 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
2020 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
2021 mdd_read_unlock(env, mdd_pobj);
2024 else if (ma_acl->ma_valid & MA_ACL_DEF)
2028 handle = mdd_trans_create(env, mdd);
2030 GOTO(out_free, rc = PTR_ERR(handle));
2032 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma,
2033 lmm_size, handle, spec);
2037 rc = mdd_trans_start(env, mdd, handle);
2041 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2043 GOTO(out_trans, rc = -ENOMEM);
2045 mdd_write_lock(env, son, MOR_TGT_CHILD);
2046 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
2048 mdd_write_unlock(env, son);
2054 #ifdef CONFIG_FS_POSIX_ACL
2056 struct lu_buf *acl_buf = &info->mti_buf;
2057 acl_buf->lb_buf = ma_acl->ma_acl;
2058 acl_buf->lb_len = ma_acl->ma_acl_size;
2060 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
2062 mdd_write_unlock(env, son);
2065 ma->ma_attr.la_valid |= LA_MODE;
2070 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2071 son, ma, handle, spec);
2072 mdd_write_unlock(env, son);
2075 * Object has no links, so it will be destroyed when last
2076 * reference is released. (XXX not now.)
2082 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2083 name, S_ISDIR(attr->la_mode), handle,
2084 mdd_object_capa(env, mdd_pobj));
2091 /* No need mdd_lsm_sanity_check here */
2092 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
2094 CERROR("error on stripe info copy %d \n", rc);
2097 if (lmm && lmm_size > 0) {
2098 /* Set Lov here, do not get lmm again later */
2099 if (lmm_size > ma->ma_lmm_size) {
2100 /* Reply buffer is smaller, need bigger one */
2101 mdd_max_lmm_buffer(env, lmm_size);
2102 if (unlikely(info->mti_max_lmm == NULL))
2103 GOTO(cleanup, rc = -ENOMEM);
2104 ma->ma_lmm = info->mti_max_lmm;
2105 ma->ma_big_lmm_used = 1;
2107 memcpy(ma->ma_lmm, lmm, lmm_size);
2108 ma->ma_lmm_size = lmm_size;
2109 ma->ma_valid |= MA_LOV;
2112 if (S_ISLNK(attr->la_mode)) {
2113 struct md_ucred *uc = md_ucred(env);
2114 struct dt_object *dt = mdd_object_child(son);
2115 const char *target_name = spec->u.sp_symname;
2116 int sym_len = strlen(target_name);
2117 const struct lu_buf *buf;
2120 buf = mdd_buf_get_const(env, target_name, sym_len);
2121 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2122 mdd_object_capa(env, son),
2124 CFS_CAP_SYS_RESOURCE_MASK);
2129 GOTO(cleanup, rc = -EFAULT);
2133 la->la_valid = LA_CTIME | LA_MTIME;
2134 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
2138 /* Return attr back. */
2139 rc = mdd_attr_get_internal_locked(env, son, ma);
2142 if (rc && created) {
2146 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2147 S_ISDIR(attr->la_mode),
2148 handle, BYPASS_CAPA);
2150 CERROR("error can not cleanup destroy %d\n",
2155 mdd_write_lock(env, son, MOR_TGT_CHILD);
2156 mdo_ref_del(env, son, handle);
2157 if (initialized && S_ISDIR(attr->la_mode))
2158 mdo_ref_del(env, son, handle);
2159 mdd_write_unlock(env, son);
2163 /* update lov_objid data, must be before transaction stop! */
2165 mdd_lov_objid_update(mdd, lmm);
2167 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2170 rc = mdd_changelog_ns_store(env, mdd,
2171 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2172 S_ISREG(attr->la_mode) ? CL_CREATE :
2173 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2174 0, son, mdd_pobj, NULL, lname, handle);
2176 mdd_trans_stop(env, mdd, rc, handle);
2178 /* finish lov_create stuff, free all temporary data */
2179 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
2181 #ifdef HAVE_QUOTA_SUPPORT
2183 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2185 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2187 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
2189 /* Trigger dqacq on the owner of child and parent. If failed,
2190 * the next call for lquota_chkquota will process it. */
2191 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
2199 * Get locks on parents in proper order
2200 * RETURN: < 0 - error, rename_order if successful
2208 static int mdd_rename_order(const struct lu_env *env,
2209 struct mdd_device *mdd,
2210 struct mdd_object *src_pobj,
2211 struct mdd_object *tgt_pobj)
2213 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2217 if (src_pobj == tgt_pobj)
2218 RETURN(MDD_RN_SAME);
2220 /* compared the parent child relationship of src_p&tgt_p */
2221 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2223 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2226 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2239 /* has not mdd_write{read}_lock on any obj yet. */
2240 static int mdd_rename_sanity_check(const struct lu_env *env,
2241 struct mdd_object *src_pobj,
2242 struct mdd_object *tgt_pobj,
2243 struct mdd_object *sobj,
2244 struct mdd_object *tobj,
2250 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
2253 /* XXX: when get here, sobj must NOT be NULL,
2254 * the other case has been processed in cml_rename
2255 * before mdd_rename and enable MDS_PERM_BYPASS. */
2258 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
2262 /* XXX: when get here, "tobj == NULL" means tobj must
2263 * NOT exist (neither on remote MDS, such case has been
2264 * processed in cml_rename before mdd_rename and enable
2266 * So check may_create, but not check may_unlink. */
2268 rc = mdd_may_create(env, tgt_pobj, NULL,
2269 (src_pobj != tgt_pobj), 0);
2271 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
2272 (src_pobj != tgt_pobj), 1);
2274 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2275 S_ISDIR(ma->ma_attr.la_mode))
2276 rc = __mdd_may_link(env, tgt_pobj);
2281 static int mdd_declare_rename(const struct lu_env *env,
2282 struct mdd_device *mdd,
2283 struct mdd_object *mdd_spobj,
2284 struct mdd_object *mdd_tpobj,
2285 struct mdd_object *mdd_sobj,
2286 struct mdd_object *mdd_tobj,
2287 const struct lu_name *sname,
2288 const struct lu_name *tname,
2290 struct thandle *handle)
2298 /* name from source dir */
2299 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2303 /* .. from source child */
2304 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2305 /* source child can be directory,
2306 * counted by source dir's nlink */
2307 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2311 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2315 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2320 /* new target child can be directory,
2321 * counted by target dir's nlink */
2322 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2328 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2332 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2335 mdd_declare_links_add(env, mdd_sobj, handle);
2339 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2344 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2345 tname->ln_name, handle);
2349 /* name from target dir (old name), we declare it unconditionally
2350 * as mdd_rename() calls delete unconditionally as well. so just
2351 * to balance declarations vs calls to change ... */
2352 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2356 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2357 /* delete target child in target parent directory */
2358 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2362 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2363 /* target child can be directory,
2364 * delete "." reference in target child directory */
2365 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2369 /* delete ".." reference in target parent directory */
2370 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2375 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2379 mdd_declare_links_add(env, mdd_tobj, handle);
2383 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2388 rc = mdd_declare_changelog_store(env, mdd, tname, handle);
2392 rc = mdd_declare_changelog_store(env, mdd, sname, handle);
2399 /* src object can be remote that is why we use only fid and type of object */
2400 static int mdd_rename(const struct lu_env *env,
2401 struct md_object *src_pobj, struct md_object *tgt_pobj,
2402 const struct lu_fid *lf, const struct lu_name *lsname,
2403 struct md_object *tobj, const struct lu_name *ltname,
2406 const char *sname = lsname->ln_name;
2407 const char *tname = ltname->ln_name;
2408 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2409 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2410 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2411 struct mdd_device *mdd = mdo2mdd(src_pobj);
2412 struct mdd_object *mdd_sobj = NULL; /* source object */
2413 struct mdd_object *mdd_tobj = NULL;
2414 struct dynlock_handle *sdlh, *tdlh;
2415 struct thandle *handle;
2416 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2417 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2421 #ifdef HAVE_QUOTA_SUPPORT
2422 struct obd_device *obd = mdd->mdd_obd_dev;
2423 struct obd_export *exp = md_quota(env)->mq_exp;
2424 struct mds_obd *mds = &obd->u.mds;
2425 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
2426 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
2427 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2428 int quota_copc = 0, quota_popc = 0;
2429 int rec_pending[MAXQUOTAS] = { 0, 0 };
2433 LASSERT(ma->ma_attr.la_mode & S_IFMT);
2434 is_dir = S_ISDIR(ma->ma_attr.la_mode);
2437 mdd_tobj = md2mdd_obj(tobj);
2439 #ifdef HAVE_QUOTA_SUPPORT
2440 if (mds->mds_quota) {
2441 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2443 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2445 mdd_quota_wrapper(la_tmp, qspids);
2447 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2451 mdd_data_get(env, mdd_tpobj, &data);
2452 quota_popc = FSFILT_OP_LINK;
2453 mdd_quota_wrapper(la_tmp, qtpids);
2454 /* get block quota for target parent */
2455 lquota_chkquota(mds_quota_interface_ref,
2457 rec_pending, 1, NULL,
2465 mdd_sobj = mdd_object_find(env, mdd, lf);
2467 handle = mdd_trans_create(env, mdd);
2469 GOTO(out_pending, rc = PTR_ERR(handle));
2471 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2472 mdd_tobj, lsname, ltname, ma, handle);
2476 rc = mdd_trans_start(env, mdd, handle);
2480 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2481 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2483 GOTO(cleanup_unlocked, rc);
2485 /* Get locks in determined order */
2486 if (rc == MDD_RN_SAME) {
2487 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2488 sname, MOR_SRC_PARENT);
2489 /* check hashes to determine do we need one lock or two */
2490 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2491 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2495 } else if (rc == MDD_RN_SRCTGT) {
2496 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2497 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2499 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2500 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2502 if (sdlh == NULL || tdlh == NULL)
2503 GOTO(cleanup, rc = -ENOMEM);
2505 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2506 mdd_sobj, mdd_tobj, ma);
2510 /* Remove source name from source directory */
2511 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2512 mdd_object_capa(env, mdd_spobj));
2516 /* "mv dir1 dir2" needs "dir1/.." link update */
2517 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2518 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2519 mdd_object_capa(env, mdd_sobj));
2521 GOTO(fixup_spobj2, rc);
2523 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2524 handle, mdd_object_capa(env, mdd_sobj));
2526 GOTO(fixup_spobj, rc);
2529 /* Remove target name from target directory
2530 * Here tobj can be remote one, so we do index_delete unconditionally
2531 * and -ENOENT is allowed.
2533 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2534 mdd_object_capa(env, mdd_tpobj));
2537 /* tname might been renamed to something else */
2538 GOTO(fixup_spobj, rc);
2541 GOTO(fixup_spobj, rc);
2544 /* Insert new fid with target name into target dir */
2545 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2546 mdd_object_capa(env, mdd_tpobj));
2548 GOTO(fixup_tpobj, rc);
2550 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2551 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2553 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2555 la->la_valid = LA_CTIME;
2556 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2559 GOTO(fixup_tpobj, rc);
2562 /* Remove old target object
2563 * For tobj is remote case cmm layer has processed
2564 * and set tobj to NULL then. So when tobj is NOT NULL,
2565 * it must be local one.
2567 if (tobj && mdd_object_exists(mdd_tobj)) {
2568 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2569 if (mdd_is_dead_obj(mdd_tobj)) {
2570 mdd_write_unlock(env, mdd_tobj);
2571 /* shld not be dead, something is wrong */
2572 CERROR("tobj is dead, something is wrong\n");
2576 mdo_ref_del(env, mdd_tobj, handle);
2578 /* Remove dot reference. */
2580 mdo_ref_del(env, mdd_tobj, handle);
2582 la->la_valid = LA_CTIME;
2583 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2585 GOTO(fixup_tpobj, rc);
2587 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2588 mdd_write_unlock(env, mdd_tobj);
2590 GOTO(fixup_tpobj, rc);
2592 #ifdef HAVE_QUOTA_SUPPORT
2593 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2594 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2595 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2596 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2601 la->la_valid = LA_CTIME | LA_MTIME;
2602 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2604 GOTO(fixup_tpobj, rc);
2606 if (mdd_spobj != mdd_tpobj) {
2607 la->la_valid = LA_CTIME | LA_MTIME;
2608 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2612 if (rc == 0 && mdd_sobj) {
2613 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2614 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2615 mdo2fid(mdd_tpobj), ltname, handle);
2617 /* Old files might not have EA entry */
2618 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2620 mdd_write_unlock(env, mdd_sobj);
2621 /* We don't fail the transaction if the link ea can't be
2622 updated -- fid2path will use alternate lookup method. */
2630 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2633 CWARN("tp obj fix error %d\n",rc2);
2635 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2636 !mdd_is_dead_obj(mdd_tobj)) {
2637 rc2 = __mdd_index_insert(env, mdd_tpobj,
2638 mdo2fid(mdd_tobj), tname,
2643 CWARN("tp obj fix error %d\n",rc2);
2648 if (rc && is_dir && mdd_sobj) {
2649 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2653 CWARN("sp obj dotdot delete error %d\n",rc2);
2656 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2657 dotdot, handle, BYPASS_CAPA);
2659 CWARN("sp obj dotdot insert error %d\n",rc2);
2664 rc2 = __mdd_index_insert(env, mdd_spobj,
2665 lf, sname, is_dir, handle, BYPASS_CAPA);
2667 CWARN("sp obj fix error %d\n",rc2);
2670 if (likely(tdlh) && sdlh != tdlh)
2671 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2673 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2676 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
2677 mdd_spobj, lf, lsname, handle);
2679 struct lu_fid zero_fid;
2680 fid_zero(&zero_fid);
2681 /* If the rename target exist, The CL_EXT record should save
2682 * the target fid as tfid, otherwise, use zero fid. LU-543 */
2683 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
2685 mdd_tobj ? NULL : &zero_fid,
2690 mdd_trans_stop(env, mdd, rc, handle);
2692 mdd_object_put(env, mdd_sobj);
2694 #ifdef HAVE_QUOTA_SUPPORT
2695 if (mds->mds_quota) {
2697 lquota_pending_commit(mds_quota_interface_ref, obd,
2698 qtpids, rec_pending, 1);
2701 /* Trigger dqrel on the source owner of parent.
2702 * If failed, the next call for lquota_chkquota will
2704 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2705 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2707 /* Trigger dqrel on the target owner of child.
2708 * If failed, the next call for lquota_chkquota
2709 * will process it. */
2710 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2711 qtpids, rc, quota_copc);
2718 /** enable/disable storing of hardlink info */
2719 int mdd_linkea_enable = 1;
2720 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2721 "record hardlink info in EAs");
2723 /** Read the link EA into a temp buffer.
2724 * Uses the name_buf since it is generally large.
2725 * \retval IS_ERR err
2726 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2728 struct lu_buf *mdd_links_get(const struct lu_env *env,
2729 struct mdd_object *mdd_obj)
2732 struct lustre_capa *capa;
2733 struct link_ea_header *leh;
2736 /* First try a small buf */
2737 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2738 if (buf->lb_buf == NULL)
2739 return ERR_PTR(-ENOMEM);
2741 capa = mdd_object_capa(env, mdd_obj);
2742 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2743 if (rc == -ERANGE) {
2744 /* Buf was too small, figure out what we need. */
2746 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2749 buf = mdd_buf_alloc(env, rc);
2750 if (buf->lb_buf == NULL)
2751 return ERR_PTR(-ENOMEM);
2752 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2758 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2759 leh->leh_magic = LINK_EA_MAGIC;
2760 leh->leh_reccount = __swab32(leh->leh_reccount);
2761 leh->leh_len = __swab64(leh->leh_len);
2762 /* entries are swabbed by mdd_lee_unpack */
2764 if (leh->leh_magic != LINK_EA_MAGIC)
2765 return ERR_PTR(-EINVAL);
2766 if (leh->leh_reccount == 0)
2767 return ERR_PTR(-ENODATA);
2772 /** Pack a link_ea_entry.
2773 * All elements are stored as chars to avoid alignment issues.
2774 * Numbers are always big-endian
2775 * \retval record length
2777 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2778 const struct lu_fid *pfid)
2780 struct lu_fid tmpfid;
2783 fid_cpu_to_be(&tmpfid, pfid);
2784 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2785 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2786 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2788 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2789 lee->lee_reclen[1] = reclen & 0xff;
2793 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2794 struct lu_name *lname, struct lu_fid *pfid)
2796 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2797 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2798 fid_be_to_cpu(pfid, pfid);
2799 lname->ln_name = lee->lee_name;
2800 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2803 /** Add a record to the end of link ea buf */
2804 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2805 const struct lu_fid *pfid,
2806 const struct lu_name *lname)
2808 struct link_ea_header *leh;
2809 struct link_ea_entry *lee;
2812 if (lname == NULL || pfid == NULL)
2815 /* Make sure our buf is big enough for the new one */
2817 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2818 if (leh->leh_len + reclen > buf->lb_len) {
2819 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2824 lee = buf->lb_buf + leh->leh_len;
2825 reclen = mdd_lee_pack(lee, lname, pfid);
2826 leh->leh_len += reclen;
2827 leh->leh_reccount++;
2831 static int mdd_declare_links_add(const struct lu_env *env,
2832 struct mdd_object *mdd_obj,
2833 struct thandle *handle)
2837 /* XXX: max size? */
2838 rc = mdo_declare_xattr_set(env, mdd_obj,
2839 mdd_buf_get_const(env, NULL, 4096),
2840 XATTR_NAME_LINK, 0, handle);
2845 /* For pathologic linkers, we don't want to spend lots of time scanning the
2846 * link ea. Limit ourseleves to something reasonable; links not in the EA
2847 * can be looked up via (slower) parent lookup.
2849 #define LINKEA_MAX_COUNT 128
2851 static int mdd_links_add(const struct lu_env *env,
2852 struct mdd_object *mdd_obj,
2853 const struct lu_fid *pfid,
2854 const struct lu_name *lname,
2855 struct thandle *handle, int first)
2858 struct link_ea_header *leh;
2862 if (!mdd_linkea_enable)
2865 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2868 if (rc != -ENODATA) {
2869 CERROR("link_ea read failed %d "DFID"\n", rc,
2870 PFID(mdd_object_fid(mdd_obj)));
2873 /* empty EA; start one */
2874 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2875 if (buf->lb_buf == NULL)
2878 leh->leh_magic = LINK_EA_MAGIC;
2879 leh->leh_len = sizeof(struct link_ea_header);
2880 leh->leh_reccount = 0;
2884 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2887 rc = __mdd_links_add(env, buf, pfid, lname);
2892 rc = __mdd_xattr_set(env, mdd_obj,
2893 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2894 XATTR_NAME_LINK, 0, handle);
2897 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2898 PFID(mdd_object_fid(mdd_obj)));
2900 CERROR("link_ea add failed %d "DFID"\n", rc,
2901 PFID(mdd_object_fid(mdd_obj)));
2904 if (buf->lb_len > OBD_ALLOC_BIG)
2905 /* if we vmalloced a large buffer drop it */
2911 static int mdd_links_rename(const struct lu_env *env,
2912 struct mdd_object *mdd_obj,
2913 const struct lu_fid *oldpfid,
2914 const struct lu_name *oldlname,
2915 const struct lu_fid *newpfid,
2916 const struct lu_name *newlname,
2917 struct thandle *handle)
2920 struct link_ea_header *leh;
2921 struct link_ea_entry *lee;
2922 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2923 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2929 if (!mdd_linkea_enable)
2932 if (mdd_obj->mod_flags & DEAD_OBJ)
2933 /* No more links, don't bother */
2936 buf = mdd_links_get(env, mdd_obj);
2940 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2941 rc, PFID(mdd_object_fid(mdd_obj)));
2943 CERROR("link_ea read failed %d "DFID"\n",
2944 rc, PFID(mdd_object_fid(mdd_obj)));
2948 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2950 /* Find the old record */
2951 for(count = 0; count < leh->leh_reccount; count++) {
2952 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2953 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2954 lu_fid_eq(tmpfid, oldpfid) &&
2955 (strncmp(tmpname->ln_name, oldlname->ln_name,
2956 tmpname->ln_namelen) == 0))
2958 lee = (struct link_ea_entry *)((char *)lee + reclen);
2960 if ((count + 1) > leh->leh_reccount) {
2961 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2962 oldlname->ln_namelen, oldlname->ln_name);
2963 GOTO(out, rc = -ENOENT);
2966 /* Remove the old record */
2967 leh->leh_reccount--;
2968 leh->leh_len -= reclen;
2969 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2972 /* If renaming, add the new record */
2973 if (newpfid != NULL) {
2974 /* if the add fails, we still delete the out-of-date old link */
2975 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2979 rc = __mdd_xattr_set(env, mdd_obj,
2980 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2981 XATTR_NAME_LINK, 0, handle);
2987 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2988 oldlname->ln_namelen, oldlname->ln_name, rc,
2989 PFID(mdd_object_fid(mdd_obj)));
2991 if (buf->lb_len > OBD_ALLOC_BIG)
2992 /* if we vmalloced a large buffer drop it */
2998 const struct md_dir_operations mdd_dir_ops = {
2999 .mdo_is_subdir = mdd_is_subdir,
3000 .mdo_lookup = mdd_lookup,
3001 .mdo_create = mdd_create,
3002 .mdo_rename = mdd_rename,
3003 .mdo_link = mdd_link,
3004 .mdo_unlink = mdd_unlink,
3005 .mdo_lum_lmm_cmp = mdd_lum_lmm_cmp,
3006 .mdo_name_insert = mdd_name_insert,
3007 .mdo_name_remove = mdd_name_remove,
3008 .mdo_rename_tgt = mdd_rename_tgt,
3009 .mdo_create_data = mdd_create_data,