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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #ifdef HAVE_EXT4_LDISKFS
47 #include <ldiskfs/ldiskfs_jbd2.h>
49 #include <linux/jbd.h>
52 #include <obd_class.h>
53 #include <lustre_ver.h>
54 #include <obd_support.h>
55 #include <lprocfs_status.h>
56 #ifdef HAVE_EXT4_LDISKFS
57 #include <ldiskfs/ldiskfs.h>
59 #include <linux/ldiskfs_fs.h>
61 #include <lustre_mds.h>
62 #include <lustre/lustre_idl.h>
63 #include <lustre_fid.h>
65 #include "mdd_internal.h"
67 static const char dot[] = ".";
68 static const char dotdot[] = "..";
70 static struct lu_name lname_dotdot = {
75 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
76 const struct lu_name *lname, struct lu_fid* fid,
78 static int mdd_declare_links_add(const struct lu_env *env,
79 struct mdd_object *mdd_obj,
80 struct thandle *handle);
81 static int mdd_links_add(const struct lu_env *env,
82 struct mdd_object *mdd_obj,
83 const struct lu_fid *pfid,
84 const struct lu_name *lname,
85 struct thandle *handle, int first);
86 static int mdd_links_rename(const struct lu_env *env,
87 struct mdd_object *mdd_obj,
88 const struct lu_fid *oldpfid,
89 const struct lu_name *oldlname,
90 const struct lu_fid *newpfid,
91 const struct lu_name *newlname,
92 struct thandle *handle);
95 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
96 const struct lu_name *lname, struct lu_fid* fid, int mask)
98 const char *name = lname->ln_name;
99 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
100 struct dynlock_handle *dlh;
103 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
104 if (unlikely(dlh == NULL))
106 rc = __mdd_lookup(env, pobj, lname, fid, mask);
107 mdd_pdo_read_unlock(env, mdd_obj, dlh);
112 int mdd_lookup(const struct lu_env *env,
113 struct md_object *pobj, const struct lu_name *lname,
114 struct lu_fid* fid, struct md_op_spec *spec)
118 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
122 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
125 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
129 * For root fid use special function, which does not compare version component
130 * of fid. Version component is different for root fids on all MDTs.
132 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
134 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
135 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
139 * return 1: if lf is the fid of the ancestor of p1;
142 * return -EREMOTE: if remote object is found, in this
143 * case fid of remote object is saved to @pf;
145 * otherwise: values < 0, errors.
147 static int mdd_is_parent(const struct lu_env *env,
148 struct mdd_device *mdd,
149 struct mdd_object *p1,
150 const struct lu_fid *lf,
153 struct mdd_object *parent = NULL;
158 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
159 pfid = &mdd_env_info(env)->mti_fid;
161 /* Check for root first. */
162 if (mdd_is_root(mdd, mdo2fid(p1)))
166 /* this is done recursively, bypass capa for each obj */
167 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
168 rc = mdd_parent_fid(env, p1, pfid);
171 if (mdd_is_root(mdd, pfid))
173 if (lu_fid_eq(pfid, lf))
176 mdd_object_put(env, parent);
177 parent = mdd_object_find(env, mdd, pfid);
179 /* cross-ref parent */
180 if (parent == NULL) {
183 GOTO(out, rc = -EREMOTE);
184 } else if (IS_ERR(parent))
185 GOTO(out, rc = PTR_ERR(parent));
190 if (parent && !IS_ERR(parent))
191 mdd_object_put(env, parent);
196 * No permission check is needed.
198 * returns 1: if fid is ancestor of @mo;
199 * returns 0: if fid is not a ancestor of @mo;
201 * returns EREMOTE if remote object is found, fid of remote object is saved to
204 * returns < 0: if error
206 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
207 const struct lu_fid *fid, struct lu_fid *sfid)
209 struct mdd_device *mdd = mdo2mdd(mo);
213 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
216 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
220 } else if (rc == 1) {
221 /* found @fid is parent */
229 * Check that @dir contains no entries except (possibly) dot and dotdot.
234 * -ENOTDIR not a directory object
235 * -ENOTEMPTY not empty
239 static int mdd_dir_is_empty(const struct lu_env *env,
240 struct mdd_object *dir)
243 struct dt_object *obj;
244 const struct dt_it_ops *iops;
248 obj = mdd_object_child(dir);
249 if (!dt_try_as_dir(env, obj))
252 iops = &obj->do_index_ops->dio_it;
253 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
255 result = iops->get(env, it, (const void *)"");
258 for (result = 0, i = 0; result == 0 && i < 3; ++i)
259 result = iops->next(env, it);
262 else if (result == +1)
264 } else if (result == 0)
266 * Huh? Index contains no zero key?
273 result = PTR_ERR(it);
277 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
279 struct mdd_device *m = mdd_obj2mdd_dev(obj);
280 struct lu_attr *la = &mdd_env_info(env)->mti_la;
284 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
289 * Subdir count limitation can be broken through.
291 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
292 !S_ISDIR(la->la_mode))
299 * Check whether it may create the cobj under the pobj.
302 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
303 struct mdd_object *cobj, int check_perm, int check_nlink)
308 if (cobj && mdd_object_exists(cobj))
311 if (mdd_is_dead_obj(pobj))
315 rc = mdd_permission_internal_locked(env, pobj, NULL,
316 MAY_WRITE | MAY_EXEC,
319 if (!rc && check_nlink)
320 rc = __mdd_may_link(env, pobj);
326 * Check whether can unlink from the pobj in the case of "cobj == NULL".
328 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
329 const struct md_attr *ma)
334 if (mdd_is_dead_obj(pobj))
337 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
338 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
341 rc = mdd_permission_internal_locked(env, pobj, NULL,
342 MAY_WRITE | MAY_EXEC,
347 if (mdd_is_append(pobj))
354 * pobj == NULL is remote ops case, under such case, pobj's
355 * VTX feature has been checked already, no need check again.
357 static inline int mdd_is_sticky(const struct lu_env *env,
358 struct mdd_object *pobj,
359 struct mdd_object *cobj)
361 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
362 struct md_ucred *uc = md_ucred(env);
366 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
370 if (!(tmp_la->la_mode & S_ISVTX) ||
371 (tmp_la->la_uid == uc->mu_fsuid))
375 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
379 if (tmp_la->la_uid == uc->mu_fsuid)
382 return !mdd_capable(uc, CFS_CAP_FOWNER);
386 * Check whether it may delete the cobj from the pobj.
389 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
390 struct mdd_object *cobj, struct md_attr *ma,
391 int check_perm, int check_empty)
397 if (!mdd_object_exists(cobj))
400 if (mdd_is_dead_obj(cobj))
404 if (!mdd_object_exists(pobj))
407 if (mdd_is_dead_obj(pobj))
411 rc = mdd_permission_internal_locked(env, pobj, NULL,
412 MAY_WRITE | MAY_EXEC,
418 if (mdd_is_append(pobj))
422 if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) &&
423 mdd_is_sticky(env, pobj, cobj))
426 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
429 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
430 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
433 if (S_ISDIR(ma->ma_attr.la_mode)) {
434 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
436 if (!S_ISDIR(mdd_object_type(cobj)))
439 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
441 } else if (S_ISDIR(mdd_object_type(cobj)))
444 if (S_ISDIR(ma->ma_attr.la_mode) && check_empty)
445 rc = mdd_dir_is_empty(env, cobj);
452 * has mdd_write_lock on src already, but not on tgt yet
454 int mdd_link_sanity_check(const struct lu_env *env,
455 struct mdd_object *tgt_obj,
456 const struct lu_name *lname,
457 struct mdd_object *src_obj)
459 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
463 if (!mdd_object_exists(src_obj))
466 if (mdd_is_dead_obj(src_obj))
469 /* Local ops, no lookup before link, check filename length here. */
470 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
471 RETURN(-ENAMETOOLONG);
473 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
476 if (S_ISDIR(mdd_object_type(src_obj)))
479 LASSERT(src_obj != tgt_obj);
481 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
486 rc = __mdd_may_link(env, src_obj);
491 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
492 const char *name, struct thandle *handle,
493 struct lustre_capa *capa)
495 struct dt_object *next = mdd_object_child(pobj);
499 if (dt_try_as_dir(env, next)) {
500 rc = next->do_index_ops->dio_delete(env, next,
501 (struct dt_key *)name,
509 static int __mdd_index_insert_only(const struct lu_env *env,
510 struct mdd_object *pobj,
511 const struct lu_fid *lf, const char *name,
512 struct thandle *handle,
513 struct lustre_capa *capa)
515 struct dt_object *next = mdd_object_child(pobj);
519 if (dt_try_as_dir(env, next)) {
520 struct md_ucred *uc = md_ucred(env);
522 rc = next->do_index_ops->dio_insert(env, next,
524 (const struct dt_key *)name,
525 handle, capa, uc->mu_cap &
526 CFS_CAP_SYS_RESOURCE_MASK);
533 /* insert named index, add reference if isdir */
534 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
535 const struct lu_fid *lf, const char *name, int is_dir,
536 struct thandle *handle, struct lustre_capa *capa)
541 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
542 if (rc == 0 && is_dir) {
543 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
544 mdo_ref_add(env, pobj, handle);
545 mdd_write_unlock(env, pobj);
550 /* delete named index, drop reference if isdir */
551 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
552 const char *name, int is_dir, struct thandle *handle,
553 struct lustre_capa *capa)
558 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
559 if (rc == 0 && is_dir) {
560 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
561 mdo_ref_del(env, pobj, handle);
562 mdd_write_unlock(env, pobj);
568 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
569 int reclen, struct thandle *handle)
573 /* XXX: this is a temporary solution to declare llog changes
574 * will be fixed in 2.3 with new llog implementation */
576 LASSERT(mdd->mdd_capa);
578 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
579 * have to set the parameter 'size' as INT_MAX or 0 to inform
580 * OSD that this record write is for a llog write or catalog
581 * header update, and osd declare function will reserve less
582 * credits for optimization purpose.
584 * Reserve 6 blocks for a llog write, since the llog file is
585 * usually small, reserve 2 blocks for catalog header update,
586 * because we know for sure that catalog header is already
589 * This hack should be removed in 2.3.
593 rc = dt_declare_record_write(env, mdd->mdd_capa,
594 DECLARE_LLOG_WRITE, 0, handle);
598 /* header will be updated as well */
599 rc = dt_declare_record_write(env, mdd->mdd_capa,
600 DECLARE_LLOG_WRITE, 0, handle);
604 /* also we should be able to create new plain log */
605 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
609 /* new record referencing new plain llog */
610 rc = dt_declare_record_write(env, mdd->mdd_capa,
611 DECLARE_LLOG_WRITE, 0, handle);
615 /* catalog's header will be updated as well */
616 rc = dt_declare_record_write(env, mdd->mdd_capa,
617 DECLARE_LLOG_REWRITE, 0, handle);
622 int mdd_declare_changelog_store(const struct lu_env *env,
623 struct mdd_device *mdd,
624 const struct lu_name *fname,
625 struct thandle *handle)
630 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
633 /* we'll be writing payload + llog header */
634 reclen = sizeof(struct llog_changelog_rec);
636 reclen += llog_data_len(fname->ln_namelen);
638 return mdd_declare_llog_record(env, mdd, reclen, handle);
641 /** Store a namespace change changelog record
642 * If this fails, we must fail the whole transaction; we don't
643 * want the change to commit without the log entry.
644 * \param target - mdd_object of change
645 * \param parent - parent dir/object
646 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
647 * \param tname - target name string
648 * \param handle - transacion handle
650 static int mdd_changelog_ns_store(const struct lu_env *env,
651 struct mdd_device *mdd,
652 enum changelog_rec_type type,
654 struct mdd_object *target,
655 struct mdd_object *parent,
656 const struct lu_fid *tf,
657 const struct lu_name *tname,
658 struct thandle *handle)
660 const struct lu_fid *tfid;
661 const struct lu_fid *tpfid = mdo2fid(parent);
662 struct llog_changelog_rec *rec;
669 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
671 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
674 LASSERT(parent != NULL);
675 LASSERT(tname != NULL);
676 LASSERT(handle != NULL);
679 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
680 buf = mdd_buf_alloc(env, reclen);
681 if (buf->lb_buf == NULL)
683 rec = (struct llog_changelog_rec *)buf->lb_buf;
685 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
686 rec->cr.cr_type = (__u32)type;
687 tfid = tf ? tf : mdo2fid(target);
688 rec->cr.cr_tfid = *tfid;
689 rec->cr.cr_pfid = *tpfid;
690 rec->cr.cr_namelen = tname->ln_namelen;
691 memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
693 target->mod_cltime = cfs_time_current_64();
695 rc = mdd_changelog_llog_write(mdd, rec, handle);
697 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
698 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
705 static int mdd_declare_link(const struct lu_env *env,
706 struct mdd_device *mdd,
707 struct mdd_object *p,
708 struct mdd_object *c,
709 const struct lu_name *name,
710 struct thandle *handle)
714 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
718 rc = mdo_declare_ref_add(env, c, handle);
722 rc = mdo_declare_attr_set(env, p, NULL, handle);
726 rc = mdo_declare_attr_set(env, c, NULL, handle);
730 rc = mdd_declare_links_add(env, c, handle);
734 rc = mdd_declare_changelog_store(env, mdd, name, handle);
739 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
740 struct md_object *src_obj, const struct lu_name *lname,
743 const char *name = lname->ln_name;
744 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
745 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
746 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
747 struct mdd_device *mdd = mdo2mdd(src_obj);
748 struct dynlock_handle *dlh;
749 struct thandle *handle;
750 #ifdef HAVE_QUOTA_SUPPORT
751 struct obd_device *obd = mdd->mdd_obd_dev;
752 struct obd_export *exp = md_quota(env)->mq_exp;
753 struct mds_obd *mds = &obd->u.mds;
754 unsigned int qids[MAXQUOTAS] = { 0, 0 };
755 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
760 #ifdef HAVE_QUOTA_SUPPORT
761 if (mds->mds_quota) {
762 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
764 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
767 mdd_data_get(env, mdd_tobj, &data);
768 quota_opc = FSFILT_OP_LINK;
769 mdd_quota_wrapper(la_tmp, qids);
770 /* get block quota for parent */
771 lquota_chkquota(mds_quota_interface_ref, obd, exp,
772 qids, rec_pending, 1, NULL,
773 LQUOTA_FLAGS_BLK, data, 1);
778 handle = mdd_trans_create(env, mdd);
780 GOTO(out_pending, rc = PTR_ERR(handle));
782 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
786 rc = mdd_trans_start(env, mdd, handle);
790 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
792 GOTO(out_trans, rc = -ENOMEM);
793 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
795 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
797 GOTO(out_unlock, rc);
799 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
801 mdd_object_capa(env, mdd_tobj));
803 GOTO(out_unlock, rc);
805 mdo_ref_add(env, mdd_sobj, handle);
807 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
808 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
810 la->la_valid = LA_CTIME | LA_MTIME;
811 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
813 GOTO(out_unlock, rc);
815 la->la_valid = LA_CTIME;
816 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
818 mdd_links_add(env, mdd_sobj,
819 mdo2fid(mdd_tobj), lname, handle, 0);
824 mdd_write_unlock(env, mdd_sobj);
825 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
828 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
829 mdd_tobj, NULL, lname, handle);
831 mdd_trans_stop(env, mdd, rc, handle);
833 #ifdef HAVE_QUOTA_SUPPORT
835 lquota_pending_commit(mds_quota_interface_ref, obd,
836 qids, rec_pending, 1);
837 /* Trigger dqacq for the parent owner. If failed,
838 * the next call for lquota_chkquota will process it. */
839 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
846 int mdd_declare_finish_unlink(const struct lu_env *env,
847 struct mdd_object *obj,
849 struct thandle *handle)
853 rc = orph_declare_index_insert(env, obj, handle);
857 return mdd_declare_object_kill(env, obj, ma, handle);
860 /* caller should take a lock before calling */
861 int mdd_finish_unlink(const struct lu_env *env,
862 struct mdd_object *obj, struct md_attr *ma,
867 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
870 LASSERT(mdd_write_locked(env, obj) != 0);
872 /* read HSM flags, needed to set changelogs flags */
873 ma->ma_need = MA_HSM | MA_INODE;
874 rc = mdd_attr_get_internal(env, obj, ma);
875 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
876 obj->mod_flags |= DEAD_OBJ;
877 /* add new orphan and the object
878 * will be deleted during mdd_close() */
879 if (obj->mod_count) {
880 rc = __mdd_orphan_add(env, obj, th);
882 CDEBUG(D_HA, "Object "DFID" is inserted into "
883 "orphan list, open count = %d\n",
884 PFID(mdd_object_fid(obj)),
887 CERROR("Object "DFID" fail to be an orphan, "
888 "open count = %d, maybe cause failed "
890 PFID(mdd_object_fid(obj)),
893 rc = mdd_object_kill(env, obj, ma, th);
898 /* get the i_nlink */
899 ma->ma_need = MA_INODE;
900 rc = mdd_attr_get_internal(env, obj, ma);
903 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
910 * has mdd_write_lock on cobj already, but not on pobj yet
912 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
913 struct mdd_object *cobj, struct md_attr *ma)
918 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
923 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
924 struct mdd_object *p, struct mdd_object *c,
925 const struct lu_name *name, struct md_attr *ma,
926 struct thandle *handle)
930 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
934 rc = mdo_declare_ref_del(env, p, handle);
938 rc = mdo_declare_ref_del(env, c, handle);
942 rc = mdo_declare_ref_del(env, c, handle);
946 rc = mdo_declare_attr_set(env, p, NULL, handle);
950 rc = mdo_declare_attr_set(env, c, NULL, handle);
954 rc = mdd_declare_finish_unlink(env, c, ma, handle);
958 rc = mdd_declare_links_add(env, c, handle);
962 rc = mdd_declare_changelog_store(env, mdd, name, handle);
967 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
968 struct md_object *cobj, const struct lu_name *lname,
971 const char *name = lname->ln_name;
972 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
973 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
974 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
975 struct mdd_device *mdd = mdo2mdd(pobj);
976 struct dynlock_handle *dlh;
977 struct thandle *handle;
978 #ifdef HAVE_QUOTA_SUPPORT
979 struct obd_device *obd = mdd->mdd_obd_dev;
980 struct mds_obd *mds = &obd->u.mds;
981 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
982 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
985 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
989 if (mdd_object_exists(mdd_cobj) <= 0)
992 handle = mdd_trans_create(env, mdd);
994 RETURN(PTR_ERR(handle));
996 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1001 rc = mdd_trans_start(env, mdd, handle);
1005 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1007 GOTO(out_trans, rc = -ENOMEM);
1008 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1010 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
1014 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1015 mdd_object_capa(env, mdd_pobj));
1019 mdo_ref_del(env, mdd_cobj, handle);
1022 mdo_ref_del(env, mdd_cobj, handle);
1024 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1025 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1027 la->la_valid = LA_CTIME | LA_MTIME;
1028 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1032 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
1033 /* update ctime of an unlinked file only if it is still
1034 * opened or a link still exists */
1035 la->la_valid = LA_CTIME;
1036 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1041 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1042 #ifdef HAVE_QUOTA_SUPPORT
1043 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1044 ma->ma_attr.la_nlink == 0) {
1045 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1047 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1049 mdd_quota_wrapper(la_tmp, qpids);
1050 if (mdd_cobj->mod_count == 0) {
1051 quota_opc = FSFILT_OP_UNLINK;
1052 mdd_quota_wrapper(&ma->ma_attr, qcids);
1054 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1060 /* old files may not have link ea; ignore errors */
1061 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1062 lname, NULL, NULL, handle);
1066 mdd_write_unlock(env, mdd_cobj);
1067 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1072 cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1073 if ((ma->ma_valid & MA_HSM) &&
1074 (ma->ma_hsm.mh_flags & HS_EXISTS))
1075 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1077 rc = mdd_changelog_ns_store(env, mdd,
1078 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1079 mdd_cobj, mdd_pobj, NULL, lname, handle);
1083 mdd_trans_stop(env, mdd, rc, handle);
1084 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
1085 if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV &&
1086 ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY)
1087 /* Since echo client is incapable of destorying ost object,
1088 * it will destory the object here. */
1089 rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la,
1090 ma->ma_lmm, ma->ma_lmm_size,
1093 #warning "please remove this after 2.4 (LOD/OSP)."
1096 #ifdef HAVE_QUOTA_SUPPORT
1098 /* Trigger dqrel on the owner of child and parent. If failed,
1099 * the next call for lquota_chkquota will process it. */
1100 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1106 /* has not lock on pobj yet */
1107 static int mdd_ni_sanity_check(const struct lu_env *env,
1108 struct md_object *pobj,
1109 const struct md_attr *ma)
1111 struct mdd_object *obj = md2mdd_obj(pobj);
1115 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1118 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
1124 * Partial operation.
1126 static int mdd_name_insert(const struct lu_env *env,
1127 struct md_object *pobj,
1128 const struct lu_name *lname,
1129 const struct lu_fid *fid,
1130 const struct md_attr *ma)
1132 const char *name = lname->ln_name;
1133 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1134 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1135 struct dynlock_handle *dlh;
1136 struct thandle *handle;
1137 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1138 #ifdef HAVE_QUOTA_SUPPORT
1139 struct mdd_device *mdd = mdo2mdd(pobj);
1140 struct md_ucred *uc = md_ucred(env);
1141 struct obd_device *obd = mdd->mdd_obd_dev;
1142 struct obd_export *exp = md_quota(env)->mq_exp;
1143 struct mds_obd *mds = &obd->u.mds;
1144 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1145 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
1146 cfs_cap_t save = uc->mu_cap;
1151 /* XXX: this code won't be used ever:
1152 * DNE uses slightly different approach */
1155 #ifdef HAVE_QUOTA_SUPPORT
1156 if (mds->mds_quota) {
1157 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
1158 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1160 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1163 mdd_data_get(env, mdd_obj, &data);
1164 quota_opc = FSFILT_OP_LINK;
1165 mdd_quota_wrapper(la_tmp, qids);
1166 /* get block quota for parent */
1167 lquota_chkquota(mds_quota_interface_ref, obd,
1168 exp, qids, rec_pending, 1, NULL,
1169 LQUOTA_FLAGS_BLK, data, 1);
1172 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1176 handle = mdd_trans_create(env, mdo2mdd(pobj));
1178 GOTO(out_pending, rc = PTR_ERR(handle));
1180 rc = mdd_trans_start(env, mdo2mdd(pobj), handle);
1182 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1184 GOTO(out_trans, rc = -ENOMEM);
1186 rc = mdd_ni_sanity_check(env, pobj, ma);
1188 GOTO(out_unlock, rc);
1190 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1191 handle, BYPASS_CAPA);
1193 GOTO(out_unlock, rc);
1196 * For some case, no need update obj's ctime (LA_CTIME is not set),
1198 * For other cases, update obj's ctime (LA_CTIME is set),
1201 if (ma->ma_attr.la_valid & LA_CTIME) {
1202 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1203 la->la_valid = LA_CTIME | LA_MTIME;
1204 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1209 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1211 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1213 #ifdef HAVE_QUOTA_SUPPORT
1214 if (mds->mds_quota) {
1216 lquota_pending_commit(mds_quota_interface_ref,
1217 obd, qids, rec_pending, 1);
1218 /* Trigger dqacq for the parent owner. If failed,
1219 * the next call for lquota_chkquota will process it*/
1220 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1230 /* has not lock on pobj yet */
1231 static int mdd_nr_sanity_check(const struct lu_env *env,
1232 struct md_object *pobj,
1233 const struct md_attr *ma)
1235 struct mdd_object *obj = md2mdd_obj(pobj);
1239 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1242 rc = mdd_may_unlink(env, obj, ma);
1248 * Partial operation.
1250 static int mdd_name_remove(const struct lu_env *env,
1251 struct md_object *pobj,
1252 const struct lu_name *lname,
1253 const struct md_attr *ma)
1255 const char *name = lname->ln_name;
1256 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1257 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1258 struct mdd_device *mdd = mdo2mdd(pobj);
1259 struct dynlock_handle *dlh;
1260 struct thandle *handle;
1261 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1262 #ifdef HAVE_QUOTA_SUPPORT
1263 struct obd_device *obd = mdd->mdd_obd_dev;
1264 struct mds_obd *mds = &obd->u.mds;
1265 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1271 /* XXX: this code won't be used ever:
1272 * DNE uses slightly different approach */
1275 #ifdef HAVE_QUOTA_SUPPORT
1276 if (mds->mds_quota) {
1277 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1279 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1281 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1282 mdd_quota_wrapper(la_tmp, qids);
1286 handle = mdd_trans_create(env, mdd);
1288 GOTO(out_pending, rc = PTR_ERR(handle));
1290 rc = mdd_trans_start(env, mdd, handle);
1292 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1294 GOTO(out_trans, rc = -ENOMEM);
1296 rc = mdd_nr_sanity_check(env, pobj, ma);
1298 GOTO(out_unlock, rc);
1300 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1301 handle, BYPASS_CAPA);
1303 GOTO(out_unlock, rc);
1306 * For some case, no need update obj's ctime (LA_CTIME is not set),
1308 * For other cases, update obj's ctime (LA_CTIME is set),
1311 if (ma->ma_attr.la_valid & LA_CTIME) {
1312 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1313 la->la_valid = LA_CTIME | LA_MTIME;
1314 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1319 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1321 mdd_trans_stop(env, mdd, rc, handle);
1323 #ifdef HAVE_QUOTA_SUPPORT
1324 /* Trigger dqrel for the parent owner.
1325 * If failed, the next call for lquota_chkquota will process it. */
1327 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1335 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1337 static int mdd_rt_sanity_check(const struct lu_env *env,
1338 struct mdd_object *tgt_pobj,
1339 struct mdd_object *tobj,
1345 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1348 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1349 * exist. In fact, tobj must exist, otherwise the call trace will be:
1350 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1351 * When get here, tobj must be NOT NULL, the other case has been
1352 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1354 * So check may_delete, but not check nlink of tgt_pobj. */
1356 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1361 /* Partial rename op on slave MDD */
1362 static int mdd_rename_tgt(const struct lu_env *env,
1363 struct md_object *pobj, struct md_object *tobj,
1364 const struct lu_fid *lf, const struct lu_name *lname,
1367 const char *name = lname->ln_name;
1368 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1369 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1370 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1371 struct mdd_device *mdd = mdo2mdd(pobj);
1372 struct dynlock_handle *dlh;
1373 struct thandle *handle;
1374 #ifdef HAVE_QUOTA_SUPPORT
1375 struct obd_device *obd = mdd->mdd_obd_dev;
1376 struct obd_export *exp = md_quota(env)->mq_exp;
1377 struct mds_obd *mds = &obd->u.mds;
1378 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1379 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1380 int quota_copc = 0, quota_popc = 0;
1381 int rec_pending[MAXQUOTAS] = { 0, 0 };
1386 /* XXX: this code won't be used ever:
1387 * DNE uses slightly different approach */
1390 #ifdef HAVE_QUOTA_SUPPORT
1391 if (mds->mds_quota && !tobj) {
1392 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1394 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1397 mdd_data_get(env, mdd_tpobj, &data);
1398 quota_popc = FSFILT_OP_LINK;
1399 mdd_quota_wrapper(la_tmp, qpids);
1400 /* get block quota for target parent */
1401 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1402 qpids, rec_pending, 1, NULL,
1403 LQUOTA_FLAGS_BLK, data, 1);
1407 handle = mdd_trans_create(env, mdd);
1409 GOTO(out_pending, rc = PTR_ERR(handle));
1411 rc = mdd_trans_start(env, mdd, handle);
1413 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1415 GOTO(out_trans, rc = -ENOMEM);
1417 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1419 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1424 * If rename_tgt is called then we should just re-insert name with
1425 * correct fid, no need to dec/inc parent nlink if obj is dir.
1427 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1431 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1436 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1437 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1439 la->la_valid = LA_CTIME | LA_MTIME;
1440 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1445 * For tobj is remote case cmm layer has processed
1446 * and pass NULL tobj to here. So when tobj is NOT NULL,
1447 * it must be local one.
1449 if (tobj && mdd_object_exists(mdd_tobj)) {
1450 mdo_ref_del(env, mdd_tobj, handle);
1452 /* Remove dot reference. */
1453 if (S_ISDIR(ma->ma_attr.la_mode))
1454 mdo_ref_del(env, mdd_tobj, handle);
1456 la->la_valid = LA_CTIME;
1457 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1461 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1465 #ifdef HAVE_QUOTA_SUPPORT
1466 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1467 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1468 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1469 mdd_quota_wrapper(&ma->ma_attr, qcids);
1476 mdd_write_unlock(env, mdd_tobj);
1477 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1480 /* Bare EXT record with no RENAME in front of it signifies
1481 a partial slave op */
1482 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
1483 mdd_tpobj, NULL, lname, handle);
1485 mdd_trans_stop(env, mdd, rc, handle);
1487 #ifdef HAVE_QUOTA_SUPPORT
1488 if (mds->mds_quota) {
1490 lquota_pending_commit(mds_quota_interface_ref, obd,
1491 qpids, rec_pending, 1);
1494 /* Trigger dqrel on the target owner of child.
1495 * If failed, the next call for lquota_chkquota
1496 * will process it. */
1497 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1505 * The permission has been checked when obj created, no need check again.
1507 static int mdd_cd_sanity_check(const struct lu_env *env,
1508 struct mdd_object *obj)
1513 if (!obj || mdd_is_dead_obj(obj))
1520 static int mdd_declare_create_data(const struct lu_env *env,
1521 struct mdd_device *mdd,
1522 struct mdd_object *obj,
1524 struct thandle *handle)
1526 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1530 buf->lb_len = lmm_size;
1531 rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
1536 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1541 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1542 struct md_object *cobj, const struct md_op_spec *spec,
1545 struct mdd_device *mdd = mdo2mdd(cobj);
1546 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1547 struct mdd_object *son = md2mdd_obj(cobj);
1548 struct lov_mds_md *lmm = NULL;
1550 struct thandle *handle;
1554 rc = mdd_cd_sanity_check(env, son);
1558 if (!md_should_create(spec->sp_cr_flags))
1560 lmm_size = ma->ma_lmm_size;
1562 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma);
1566 handle = mdd_trans_create(env, mdd);
1568 GOTO(out_free, rc = PTR_ERR(handle));
1570 rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle);
1574 rc = mdd_trans_start(env, mdd, handle);
1579 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1580 * Should this be fixed?
1583 /* Replay creates has objects already */
1585 if (spec->no_create) {
1586 CDEBUG(D_INFO, "we already have lov ea\n");
1587 rc = mdd_lov_set_md(env, mdd_pobj, son,
1588 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1589 spec->u.sp_ea.eadatalen, handle, 0);
1592 /* No need mdd_lsm_sanity_check here */
1593 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1594 lmm_size, handle, 0);
1597 rc = mdd_attr_get_internal_locked(env, son, ma);
1599 /* update lov_objid data, must be before transaction stop! */
1601 mdd_lov_objid_update(mdd, lmm);
1604 mdd_trans_stop(env, mdd, rc, handle);
1606 /* Finish mdd_lov_create() stuff. */
1607 /* if no_create == 0 (not replay), we free lmm allocated by
1608 * mdd_lov_create() */
1609 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1613 /* Get fid from name and parent */
1615 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1616 const struct lu_name *lname, struct lu_fid* fid, int mask)
1618 const char *name = lname->ln_name;
1619 const struct dt_key *key = (const struct dt_key *)name;
1620 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1621 struct mdd_device *m = mdo2mdd(pobj);
1622 struct dt_object *dir = mdd_object_child(mdd_obj);
1626 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1629 rc = mdd_object_exists(mdd_obj);
1630 if (unlikely(rc == 0))
1632 else if (unlikely(rc < 0)) {
1633 CERROR("Object "DFID" locates on remote server\n",
1634 PFID(mdo2fid(mdd_obj)));
1638 /* The common filename length check. */
1639 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1640 RETURN(-ENAMETOOLONG);
1642 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1647 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1648 dt_try_as_dir(env, dir))) {
1650 rc = dir->do_index_ops->dio_lookup(env, dir,
1651 (struct dt_rec *)fid, key,
1652 mdd_object_capa(env, mdd_obj));
1663 int mdd_declare_object_initialize(const struct lu_env *env,
1664 struct mdd_object *child,
1666 struct thandle *handle)
1670 rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle);
1671 if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) {
1672 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1675 rc = mdo_declare_ref_add(env, child, handle);
1678 mdd_declare_links_add(env, child, handle);
1683 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1684 const struct lu_name *lname, struct mdd_object *child,
1685 struct md_attr *ma, struct thandle *handle,
1686 const struct md_op_spec *spec)
1692 * Update attributes for child.
1695 * (1) the valid bits should be converted between Lustre and Linux;
1696 * (2) maybe, the child attributes should be set in OSD when creation.
1699 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1703 if (S_ISDIR(ma->ma_attr.la_mode)) {
1704 /* Add "." and ".." for newly created dir */
1705 mdo_ref_add(env, child, handle);
1706 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1707 dot, handle, BYPASS_CAPA);
1709 rc = __mdd_index_insert_only(env, child, pfid,
1713 mdo_ref_del(env, child, handle);
1716 mdd_links_add(env, child, pfid, lname, handle, 1);
1721 /* has not lock on pobj yet */
1722 static int mdd_create_sanity_check(const struct lu_env *env,
1723 struct md_object *pobj,
1724 const struct lu_name *lname,
1726 struct md_op_spec *spec)
1728 struct mdd_thread_info *info = mdd_env_info(env);
1729 struct lu_attr *la = &info->mti_la;
1730 struct lu_fid *fid = &info->mti_fid;
1731 struct mdd_object *obj = md2mdd_obj(pobj);
1732 struct mdd_device *m = mdo2mdd(pobj);
1733 int lookup = spec->sp_cr_lookup;
1738 if (mdd_is_dead_obj(obj))
1742 * In some cases this lookup is not needed - we know before if name
1743 * exists or not because MDT performs lookup for it.
1744 * name length check is done in lookup.
1748 * Check if the name already exist, though it will be checked in
1749 * _index_insert also, for avoiding rolling back if exists
1752 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1753 MAY_WRITE | MAY_EXEC);
1755 RETURN(rc ? : -EEXIST);
1758 * Check WRITE permission for the parent.
1759 * EXEC permission have been checked
1760 * when lookup before create already.
1762 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1769 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1773 if (la->la_mode & S_ISGID) {
1774 ma->ma_attr.la_gid = la->la_gid;
1775 if (S_ISDIR(ma->ma_attr.la_mode)) {
1776 ma->ma_attr.la_mode |= S_ISGID;
1777 ma->ma_attr.la_valid |= LA_MODE;
1781 switch (ma->ma_attr.la_mode & S_IFMT) {
1783 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1785 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1786 RETURN(-ENAMETOOLONG);
1805 static int mdd_declare_create(const struct lu_env *env,
1806 struct mdd_device *mdd,
1807 struct mdd_object *p,
1808 struct mdd_object *c,
1809 const struct lu_name *name,
1812 struct thandle *handle,
1813 const struct md_op_spec *spec)
1815 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1818 rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec);
1822 /* if dir, then can inherit default ACl */
1824 buf->lb_len = lmm_size;
1825 if (S_ISDIR(ma->ma_attr.la_mode)) {
1826 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT,
1829 rc = mdo_declare_ref_add(env, p, handle);
1834 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS,
1839 rc = mdd_declare_object_initialize(env, c, ma, handle);
1843 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1844 name->ln_name, handle);
1848 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1853 if (S_ISLNK(ma->ma_attr.la_mode)) {
1854 rc = dt_declare_record_write(env, mdd_object_child(c),
1855 strlen(spec->u.sp_symname), 0,
1860 rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle);
1864 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1868 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1876 * Create object and insert it into namespace.
1878 static int mdd_create(const struct lu_env *env,
1879 struct md_object *pobj,
1880 const struct lu_name *lname,
1881 struct md_object *child,
1882 struct md_op_spec *spec,
1885 struct mdd_thread_info *info = mdd_env_info(env);
1886 struct lu_attr *la = &info->mti_la_for_fix;
1887 struct md_attr *ma_acl = &info->mti_ma;
1888 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1889 struct mdd_object *son = md2mdd_obj(child);
1890 struct mdd_device *mdd = mdo2mdd(pobj);
1891 struct lu_attr *attr = &ma->ma_attr;
1892 struct lov_mds_md *lmm = NULL;
1893 struct thandle *handle;
1894 struct dynlock_handle *dlh;
1895 const char *name = lname->ln_name;
1896 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1897 int got_def_acl = 0;
1898 #ifdef HAVE_QUOTA_SUPPORT
1899 struct obd_device *obd = mdd->mdd_obd_dev;
1900 struct obd_export *exp = md_quota(env)->mq_exp;
1901 struct mds_obd *mds = &obd->u.mds;
1902 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1903 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1904 int quota_opc = 0, block_count = 0;
1905 int inode_pending[MAXQUOTAS] = { 0, 0 };
1906 int block_pending[MAXQUOTAS] = { 0, 0 };
1907 int parent_pending[MAXQUOTAS] = { 0, 0 };
1912 * Two operations have to be performed:
1914 * - an allocation of a new object (->do_create()), and
1916 * - an insertion into a parent index (->dio_insert()).
1918 * Due to locking, operation order is not important, when both are
1919 * successful, *but* error handling cases are quite different:
1921 * - if insertion is done first, and following object creation fails,
1922 * insertion has to be rolled back, but this operation might fail
1923 * also leaving us with dangling index entry.
1925 * - if creation is done first, is has to be undone if insertion
1926 * fails, leaving us with leaked space, which is neither good, nor
1929 * It seems that creation-first is simplest solution, but it is
1930 * sub-optimal in the frequent
1935 * case, because second mkdir is bound to create object, only to
1936 * destroy it immediately.
1938 * To avoid this follow local file systems that do double lookup:
1940 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1942 * 1. create (mdd_object_create_internal())
1944 * 2. insert (__mdd_index_insert(), lookup again)
1947 /* Sanity checks before big job. */
1948 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1952 #ifdef HAVE_QUOTA_SUPPORT
1953 if (mds->mds_quota) {
1954 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1956 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1960 quota_opc = FSFILT_OP_CREATE;
1961 mdd_quota_wrapper(&ma->ma_attr, qcids);
1962 mdd_quota_wrapper(la_tmp, qpids);
1963 /* get file quota for child */
1964 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1965 qcids, inode_pending, 1, NULL, 0, NULL,
1967 switch (ma->ma_attr.la_mode & S_IFMT) {
1976 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1977 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1981 /* get block quota for child and parent */
1983 lquota_chkquota(mds_quota_interface_ref, obd,
1984 exp, qcids, block_pending,
1986 LQUOTA_FLAGS_BLK, NULL, 0);
1988 lquota_chkquota(mds_quota_interface_ref, obd,
1989 exp, qpids, parent_pending, 1,
1990 NULL, LQUOTA_FLAGS_BLK, NULL,
1996 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
1997 GOTO(out_pending, rc = -EINPROGRESS);
2000 * No RPC inside the transaction, so OST objects should be created at
2003 if (S_ISREG(attr->la_mode)) {
2004 lmm_size = ma->ma_lmm_size;
2005 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
2008 GOTO(out_pending, rc);
2011 if (!S_ISLNK(attr->la_mode)) {
2012 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
2013 ma_acl->ma_acl = info->mti_xattr_buf;
2014 ma_acl->ma_need = MA_ACL_DEF;
2015 ma_acl->ma_valid = 0;
2017 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
2018 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
2019 mdd_read_unlock(env, mdd_pobj);
2022 else if (ma_acl->ma_valid & MA_ACL_DEF)
2026 handle = mdd_trans_create(env, mdd);
2028 GOTO(out_free, rc = PTR_ERR(handle));
2030 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma,
2031 lmm_size, handle, spec);
2035 rc = mdd_trans_start(env, mdd, handle);
2039 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2041 GOTO(out_trans, rc = -ENOMEM);
2043 mdd_write_lock(env, son, MOR_TGT_CHILD);
2044 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
2046 mdd_write_unlock(env, son);
2052 #ifdef CONFIG_FS_POSIX_ACL
2054 struct lu_buf *acl_buf = &info->mti_buf;
2055 acl_buf->lb_buf = ma_acl->ma_acl;
2056 acl_buf->lb_len = ma_acl->ma_acl_size;
2058 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
2060 mdd_write_unlock(env, son);
2063 ma->ma_attr.la_valid |= LA_MODE;
2068 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2069 son, ma, handle, spec);
2070 mdd_write_unlock(env, son);
2073 * Object has no links, so it will be destroyed when last
2074 * reference is released. (XXX not now.)
2080 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2081 name, S_ISDIR(attr->la_mode), handle,
2082 mdd_object_capa(env, mdd_pobj));
2089 /* No need mdd_lsm_sanity_check here */
2090 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
2092 CERROR("error on stripe info copy %d \n", rc);
2095 if (lmm && lmm_size > 0) {
2096 /* Set Lov here, do not get lmm again later */
2097 if (lmm_size > ma->ma_lmm_size) {
2098 /* Reply buffer is smaller, need bigger one */
2099 mdd_max_lmm_buffer(env, lmm_size);
2100 if (unlikely(info->mti_max_lmm == NULL))
2101 GOTO(cleanup, rc = -ENOMEM);
2102 ma->ma_lmm = info->mti_max_lmm;
2103 ma->ma_big_lmm_used = 1;
2105 memcpy(ma->ma_lmm, lmm, lmm_size);
2106 ma->ma_lmm_size = lmm_size;
2107 ma->ma_valid |= MA_LOV;
2110 if (S_ISLNK(attr->la_mode)) {
2111 struct md_ucred *uc = md_ucred(env);
2112 struct dt_object *dt = mdd_object_child(son);
2113 const char *target_name = spec->u.sp_symname;
2114 int sym_len = strlen(target_name);
2115 const struct lu_buf *buf;
2118 buf = mdd_buf_get_const(env, target_name, sym_len);
2119 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2120 mdd_object_capa(env, son),
2122 CFS_CAP_SYS_RESOURCE_MASK);
2127 GOTO(cleanup, rc = -EFAULT);
2131 la->la_valid = LA_CTIME | LA_MTIME;
2132 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
2136 /* Return attr back. */
2137 rc = mdd_attr_get_internal_locked(env, son, ma);
2140 if (rc && created) {
2144 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2145 S_ISDIR(attr->la_mode),
2146 handle, BYPASS_CAPA);
2148 CERROR("error can not cleanup destroy %d\n",
2153 mdd_write_lock(env, son, MOR_TGT_CHILD);
2154 mdo_ref_del(env, son, handle);
2155 if (initialized && S_ISDIR(attr->la_mode))
2156 mdo_ref_del(env, son, handle);
2157 mdd_write_unlock(env, son);
2161 /* update lov_objid data, must be before transaction stop! */
2163 mdd_lov_objid_update(mdd, lmm);
2165 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2168 rc = mdd_changelog_ns_store(env, mdd,
2169 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2170 S_ISREG(attr->la_mode) ? CL_CREATE :
2171 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2172 0, son, mdd_pobj, NULL, lname, handle);
2174 mdd_trans_stop(env, mdd, rc, handle);
2176 /* finish lov_create stuff, free all temporary data */
2177 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
2179 #ifdef HAVE_QUOTA_SUPPORT
2181 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2183 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2185 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
2187 /* Trigger dqacq on the owner of child and parent. If failed,
2188 * the next call for lquota_chkquota will process it. */
2189 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
2193 /* The child object shouldn't be cached anymore */
2195 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
2196 &child->mo_lu.lo_header->loh_flags);
2201 * Get locks on parents in proper order
2202 * RETURN: < 0 - error, rename_order if successful
2210 static int mdd_rename_order(const struct lu_env *env,
2211 struct mdd_device *mdd,
2212 struct mdd_object *src_pobj,
2213 struct mdd_object *tgt_pobj)
2215 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2219 if (src_pobj == tgt_pobj)
2220 RETURN(MDD_RN_SAME);
2222 /* compared the parent child relationship of src_p&tgt_p */
2223 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2225 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2228 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2241 /* has not mdd_write{read}_lock on any obj yet. */
2242 static int mdd_rename_sanity_check(const struct lu_env *env,
2243 struct mdd_object *src_pobj,
2244 struct mdd_object *tgt_pobj,
2245 struct mdd_object *sobj,
2246 struct mdd_object *tobj,
2252 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
2255 /* XXX: when get here, sobj must NOT be NULL,
2256 * the other case has been processed in cml_rename
2257 * before mdd_rename and enable MDS_PERM_BYPASS. */
2260 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
2264 /* XXX: when get here, "tobj == NULL" means tobj must
2265 * NOT exist (neither on remote MDS, such case has been
2266 * processed in cml_rename before mdd_rename and enable
2268 * So check may_create, but not check may_unlink. */
2270 rc = mdd_may_create(env, tgt_pobj, NULL,
2271 (src_pobj != tgt_pobj), 0);
2273 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
2274 (src_pobj != tgt_pobj), 1);
2276 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2277 S_ISDIR(ma->ma_attr.la_mode))
2278 rc = __mdd_may_link(env, tgt_pobj);
2283 static int mdd_declare_rename(const struct lu_env *env,
2284 struct mdd_device *mdd,
2285 struct mdd_object *mdd_spobj,
2286 struct mdd_object *mdd_tpobj,
2287 struct mdd_object *mdd_sobj,
2288 struct mdd_object *mdd_tobj,
2289 const struct lu_name *sname,
2290 const struct lu_name *tname,
2292 struct thandle *handle)
2300 /* name from source dir */
2301 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2305 /* .. from source child */
2306 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2307 /* source child can be directory,
2308 * counted by source dir's nlink */
2309 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2313 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2317 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2322 /* new target child can be directory,
2323 * counted by target dir's nlink */
2324 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2330 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2334 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2337 mdd_declare_links_add(env, mdd_sobj, handle);
2341 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2346 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2347 tname->ln_name, handle);
2351 /* name from target dir (old name), we declare it unconditionally
2352 * as mdd_rename() calls delete unconditionally as well. so just
2353 * to balance declarations vs calls to change ... */
2354 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2358 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2359 /* delete target child in target parent directory */
2360 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2364 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2365 /* target child can be directory,
2366 * delete "." reference in target child directory */
2367 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2371 /* delete ".." reference in target parent directory */
2372 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2377 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2381 mdd_declare_links_add(env, mdd_tobj, handle);
2385 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2390 rc = mdd_declare_changelog_store(env, mdd, tname, handle);
2394 rc = mdd_declare_changelog_store(env, mdd, sname, handle);
2401 /* src object can be remote that is why we use only fid and type of object */
2402 static int mdd_rename(const struct lu_env *env,
2403 struct md_object *src_pobj, struct md_object *tgt_pobj,
2404 const struct lu_fid *lf, const struct lu_name *lsname,
2405 struct md_object *tobj, const struct lu_name *ltname,
2408 const char *sname = lsname->ln_name;
2409 const char *tname = ltname->ln_name;
2410 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2411 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2412 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2413 struct mdd_device *mdd = mdo2mdd(src_pobj);
2414 struct mdd_object *mdd_sobj = NULL; /* source object */
2415 struct mdd_object *mdd_tobj = NULL;
2416 struct dynlock_handle *sdlh, *tdlh;
2417 struct thandle *handle;
2418 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2419 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2423 #ifdef HAVE_QUOTA_SUPPORT
2424 struct obd_device *obd = mdd->mdd_obd_dev;
2425 struct obd_export *exp = md_quota(env)->mq_exp;
2426 struct mds_obd *mds = &obd->u.mds;
2427 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
2428 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
2429 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2430 int quota_copc = 0, quota_popc = 0;
2431 int rec_pending[MAXQUOTAS] = { 0, 0 };
2435 LASSERT(ma->ma_attr.la_mode & S_IFMT);
2436 is_dir = S_ISDIR(ma->ma_attr.la_mode);
2439 mdd_tobj = md2mdd_obj(tobj);
2441 #ifdef HAVE_QUOTA_SUPPORT
2442 if (mds->mds_quota) {
2443 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2445 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2447 mdd_quota_wrapper(la_tmp, qspids);
2449 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2453 mdd_data_get(env, mdd_tpobj, &data);
2454 quota_popc = FSFILT_OP_LINK;
2455 mdd_quota_wrapper(la_tmp, qtpids);
2456 /* get block quota for target parent */
2457 lquota_chkquota(mds_quota_interface_ref,
2459 rec_pending, 1, NULL,
2467 mdd_sobj = mdd_object_find(env, mdd, lf);
2469 handle = mdd_trans_create(env, mdd);
2471 GOTO(out_pending, rc = PTR_ERR(handle));
2473 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2474 mdd_tobj, lsname, ltname, ma, handle);
2478 rc = mdd_trans_start(env, mdd, handle);
2482 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2483 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2485 GOTO(cleanup_unlocked, rc);
2487 /* Get locks in determined order */
2488 if (rc == MDD_RN_SAME) {
2489 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2490 sname, MOR_SRC_PARENT);
2491 /* check hashes to determine do we need one lock or two */
2492 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2493 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2497 } else if (rc == MDD_RN_SRCTGT) {
2498 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2499 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2501 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2502 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2504 if (sdlh == NULL || tdlh == NULL)
2505 GOTO(cleanup, rc = -ENOMEM);
2507 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2508 mdd_sobj, mdd_tobj, ma);
2512 /* Remove source name from source directory */
2513 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2514 mdd_object_capa(env, mdd_spobj));
2518 /* "mv dir1 dir2" needs "dir1/.." link update */
2519 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2520 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2521 mdd_object_capa(env, mdd_sobj));
2523 GOTO(fixup_spobj2, rc);
2525 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2526 handle, mdd_object_capa(env, mdd_sobj));
2528 GOTO(fixup_spobj, rc);
2531 /* Remove target name from target directory
2532 * Here tobj can be remote one, so we do index_delete unconditionally
2533 * and -ENOENT is allowed.
2535 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2536 mdd_object_capa(env, mdd_tpobj));
2539 /* tname might been renamed to something else */
2540 GOTO(fixup_spobj, rc);
2543 GOTO(fixup_spobj, rc);
2546 /* Insert new fid with target name into target dir */
2547 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2548 mdd_object_capa(env, mdd_tpobj));
2550 GOTO(fixup_tpobj, rc);
2552 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2553 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2555 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2557 la->la_valid = LA_CTIME;
2558 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2561 GOTO(fixup_tpobj, rc);
2564 /* Remove old target object
2565 * For tobj is remote case cmm layer has processed
2566 * and set tobj to NULL then. So when tobj is NOT NULL,
2567 * it must be local one.
2569 if (tobj && mdd_object_exists(mdd_tobj)) {
2570 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2571 if (mdd_is_dead_obj(mdd_tobj)) {
2572 mdd_write_unlock(env, mdd_tobj);
2573 /* shld not be dead, something is wrong */
2574 CERROR("tobj is dead, something is wrong\n");
2578 mdo_ref_del(env, mdd_tobj, handle);
2580 /* Remove dot reference. */
2582 mdo_ref_del(env, mdd_tobj, handle);
2584 la->la_valid = LA_CTIME;
2585 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2587 GOTO(fixup_tpobj, rc);
2589 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2590 mdd_write_unlock(env, mdd_tobj);
2592 GOTO(fixup_tpobj, rc);
2594 #ifdef HAVE_QUOTA_SUPPORT
2595 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2596 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2597 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2598 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2603 la->la_valid = LA_CTIME | LA_MTIME;
2604 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2606 GOTO(fixup_tpobj, rc);
2608 if (mdd_spobj != mdd_tpobj) {
2609 la->la_valid = LA_CTIME | LA_MTIME;
2610 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2614 if (rc == 0 && mdd_sobj) {
2615 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2616 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2617 mdo2fid(mdd_tpobj), ltname, handle);
2619 /* Old files might not have EA entry */
2620 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2622 mdd_write_unlock(env, mdd_sobj);
2623 /* We don't fail the transaction if the link ea can't be
2624 updated -- fid2path will use alternate lookup method. */
2632 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2635 CWARN("tp obj fix error %d\n",rc2);
2637 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2638 !mdd_is_dead_obj(mdd_tobj)) {
2639 rc2 = __mdd_index_insert(env, mdd_tpobj,
2640 mdo2fid(mdd_tobj), tname,
2645 CWARN("tp obj fix error %d\n",rc2);
2650 if (rc && is_dir && mdd_sobj) {
2651 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2655 CWARN("sp obj dotdot delete error %d\n",rc2);
2658 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2659 dotdot, handle, BYPASS_CAPA);
2661 CWARN("sp obj dotdot insert error %d\n",rc2);
2666 rc2 = __mdd_index_insert(env, mdd_spobj,
2667 lf, sname, is_dir, handle, BYPASS_CAPA);
2669 CWARN("sp obj fix error %d\n",rc2);
2672 if (likely(tdlh) && sdlh != tdlh)
2673 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2675 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2678 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
2679 mdd_spobj, lf, lsname, handle);
2681 struct lu_fid zero_fid;
2682 fid_zero(&zero_fid);
2683 /* If the rename target exist, The CL_EXT record should save
2684 * the target fid as tfid, otherwise, use zero fid. LU-543 */
2685 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
2687 mdd_tobj ? NULL : &zero_fid,
2692 mdd_trans_stop(env, mdd, rc, handle);
2694 mdd_object_put(env, mdd_sobj);
2696 #ifdef HAVE_QUOTA_SUPPORT
2697 if (mds->mds_quota) {
2699 lquota_pending_commit(mds_quota_interface_ref, obd,
2700 qtpids, rec_pending, 1);
2703 /* Trigger dqrel on the source owner of parent.
2704 * If failed, the next call for lquota_chkquota will
2706 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2707 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2709 /* Trigger dqrel on the target owner of child.
2710 * If failed, the next call for lquota_chkquota
2711 * will process it. */
2712 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2713 qtpids, rc, quota_copc);
2720 /** enable/disable storing of hardlink info */
2721 int mdd_linkea_enable = 1;
2722 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2723 "record hardlink info in EAs");
2725 /** Read the link EA into a temp buffer.
2726 * Uses the name_buf since it is generally large.
2727 * \retval IS_ERR err
2728 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2730 struct lu_buf *mdd_links_get(const struct lu_env *env,
2731 struct mdd_object *mdd_obj)
2734 struct lustre_capa *capa;
2735 struct link_ea_header *leh;
2738 /* First try a small buf */
2739 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2740 if (buf->lb_buf == NULL)
2741 return ERR_PTR(-ENOMEM);
2743 capa = mdd_object_capa(env, mdd_obj);
2744 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2745 if (rc == -ERANGE) {
2746 /* Buf was too small, figure out what we need. */
2748 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2751 buf = mdd_buf_alloc(env, rc);
2752 if (buf->lb_buf == NULL)
2753 return ERR_PTR(-ENOMEM);
2754 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2760 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2761 leh->leh_magic = LINK_EA_MAGIC;
2762 leh->leh_reccount = __swab32(leh->leh_reccount);
2763 leh->leh_len = __swab64(leh->leh_len);
2764 /* entries are swabbed by mdd_lee_unpack */
2766 if (leh->leh_magic != LINK_EA_MAGIC)
2767 return ERR_PTR(-EINVAL);
2768 if (leh->leh_reccount == 0)
2769 return ERR_PTR(-ENODATA);
2774 /** Pack a link_ea_entry.
2775 * All elements are stored as chars to avoid alignment issues.
2776 * Numbers are always big-endian
2777 * \retval record length
2779 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2780 const struct lu_fid *pfid)
2782 struct lu_fid tmpfid;
2785 fid_cpu_to_be(&tmpfid, pfid);
2786 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2787 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2788 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2790 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2791 lee->lee_reclen[1] = reclen & 0xff;
2795 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2796 struct lu_name *lname, struct lu_fid *pfid)
2798 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2799 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2800 fid_be_to_cpu(pfid, pfid);
2801 lname->ln_name = lee->lee_name;
2802 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2805 /** Add a record to the end of link ea buf */
2806 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2807 const struct lu_fid *pfid,
2808 const struct lu_name *lname)
2810 struct link_ea_header *leh;
2811 struct link_ea_entry *lee;
2814 if (lname == NULL || pfid == NULL)
2817 /* Make sure our buf is big enough for the new one */
2819 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2820 if (leh->leh_len + reclen > buf->lb_len) {
2821 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2826 lee = buf->lb_buf + leh->leh_len;
2827 reclen = mdd_lee_pack(lee, lname, pfid);
2828 leh->leh_len += reclen;
2829 leh->leh_reccount++;
2833 static int mdd_declare_links_add(const struct lu_env *env,
2834 struct mdd_object *mdd_obj,
2835 struct thandle *handle)
2839 /* XXX: max size? */
2840 rc = mdo_declare_xattr_set(env, mdd_obj,
2841 mdd_buf_get_const(env, NULL, 4096),
2842 XATTR_NAME_LINK, 0, handle);
2847 /* For pathologic linkers, we don't want to spend lots of time scanning the
2848 * link ea. Limit ourseleves to something reasonable; links not in the EA
2849 * can be looked up via (slower) parent lookup.
2851 #define LINKEA_MAX_COUNT 128
2853 static int mdd_links_add(const struct lu_env *env,
2854 struct mdd_object *mdd_obj,
2855 const struct lu_fid *pfid,
2856 const struct lu_name *lname,
2857 struct thandle *handle, int first)
2860 struct link_ea_header *leh;
2864 if (!mdd_linkea_enable)
2867 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2870 if (rc != -ENODATA) {
2871 CERROR("link_ea read failed %d "DFID"\n", rc,
2872 PFID(mdd_object_fid(mdd_obj)));
2875 /* empty EA; start one */
2876 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2877 if (buf->lb_buf == NULL)
2880 leh->leh_magic = LINK_EA_MAGIC;
2881 leh->leh_len = sizeof(struct link_ea_header);
2882 leh->leh_reccount = 0;
2886 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2889 rc = __mdd_links_add(env, buf, pfid, lname);
2894 rc = __mdd_xattr_set(env, mdd_obj,
2895 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2896 XATTR_NAME_LINK, 0, handle);
2899 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2900 PFID(mdd_object_fid(mdd_obj)));
2902 CERROR("link_ea add failed %d "DFID"\n", rc,
2903 PFID(mdd_object_fid(mdd_obj)));
2906 if (buf->lb_len > OBD_ALLOC_BIG)
2907 /* if we vmalloced a large buffer drop it */
2913 static int mdd_links_rename(const struct lu_env *env,
2914 struct mdd_object *mdd_obj,
2915 const struct lu_fid *oldpfid,
2916 const struct lu_name *oldlname,
2917 const struct lu_fid *newpfid,
2918 const struct lu_name *newlname,
2919 struct thandle *handle)
2922 struct link_ea_header *leh;
2923 struct link_ea_entry *lee;
2924 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2925 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2931 if (!mdd_linkea_enable)
2934 if (mdd_obj->mod_flags & DEAD_OBJ)
2935 /* No more links, don't bother */
2938 buf = mdd_links_get(env, mdd_obj);
2942 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2943 rc, PFID(mdd_object_fid(mdd_obj)));
2945 CERROR("link_ea read failed %d "DFID"\n",
2946 rc, PFID(mdd_object_fid(mdd_obj)));
2950 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2952 /* Find the old record */
2953 for(count = 0; count < leh->leh_reccount; count++) {
2954 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2955 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2956 lu_fid_eq(tmpfid, oldpfid) &&
2957 (strncmp(tmpname->ln_name, oldlname->ln_name,
2958 tmpname->ln_namelen) == 0))
2960 lee = (struct link_ea_entry *)((char *)lee + reclen);
2962 if ((count + 1) > leh->leh_reccount) {
2963 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2964 oldlname->ln_namelen, oldlname->ln_name);
2965 GOTO(out, rc = -ENOENT);
2968 /* Remove the old record */
2969 leh->leh_reccount--;
2970 leh->leh_len -= reclen;
2971 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2974 /* If renaming, add the new record */
2975 if (newpfid != NULL) {
2976 /* if the add fails, we still delete the out-of-date old link */
2977 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2981 rc = __mdd_xattr_set(env, mdd_obj,
2982 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2983 XATTR_NAME_LINK, 0, handle);
2989 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2990 oldlname->ln_namelen, oldlname->ln_name, rc,
2991 PFID(mdd_object_fid(mdd_obj)));
2993 if (buf->lb_len > OBD_ALLOC_BIG)
2994 /* if we vmalloced a large buffer drop it */
3000 const struct md_dir_operations mdd_dir_ops = {
3001 .mdo_is_subdir = mdd_is_subdir,
3002 .mdo_lookup = mdd_lookup,
3003 .mdo_create = mdd_create,
3004 .mdo_rename = mdd_rename,
3005 .mdo_link = mdd_link,
3006 .mdo_unlink = mdd_unlink,
3007 .mdo_lum_lmm_cmp = mdd_lum_lmm_cmp,
3008 .mdo_name_insert = mdd_name_insert,
3009 .mdo_name_remove = mdd_name_remove,
3010 .mdo_rename_tgt = mdd_rename_tgt,
3011 .mdo_create_data = mdd_create_data,