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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #ifdef HAVE_EXT4_LDISKFS
50 #include <ldiskfs/ldiskfs_jbd2.h>
52 #include <linux/jbd.h>
55 #include <obd_class.h>
56 #include <lustre_ver.h>
57 #include <obd_support.h>
58 #include <lprocfs_status.h>
59 #ifdef HAVE_EXT4_LDISKFS
60 #include <ldiskfs/ldiskfs.h>
62 #include <linux/ldiskfs_fs.h>
64 #include <lustre_mds.h>
65 #include <lustre/lustre_idl.h>
66 #include <lustre_fid.h>
68 #include "mdd_internal.h"
70 static const char dot[] = ".";
71 static const char dotdot[] = "..";
73 static struct lu_name lname_dotdot = {
78 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
79 const struct lu_name *lname, struct lu_fid* fid,
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);
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, 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?
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);
492 * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and
493 * assign i_nlink to 1 which means the i_nlink for subdir count is incredible
494 * (maybe too large to be represented). It is a trick to break through the
495 * "i_nlink" limitation for subdir count.
497 void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj,
498 struct thandle *handle)
500 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
501 struct mdd_device *m = mdd_obj2mdd_dev(obj);
503 if (!mdd_is_mnlink(obj)) {
504 if (S_ISDIR(mdd_object_type(obj))) {
505 if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA))
508 if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) {
509 obj->mod_flags |= MNLINK_OBJ;
510 tmp_la->la_nlink = 1;
511 tmp_la->la_valid = LA_NLINK;
512 mdd_attr_set_internal(env, obj, tmp_la, handle,
517 mdo_ref_add(env, obj, handle);
521 void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj,
522 struct thandle *handle, int is_dot)
524 if (!mdd_is_mnlink(obj) || is_dot)
525 mdo_ref_del(env, obj, handle);
528 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
529 const char *name, struct thandle *handle,
530 struct lustre_capa *capa)
532 struct dt_object *next = mdd_object_child(pobj);
536 if (dt_try_as_dir(env, next)) {
537 rc = next->do_index_ops->dio_delete(env, next,
538 (struct dt_key *)name,
546 static int __mdd_index_insert_only(const struct lu_env *env,
547 struct mdd_object *pobj,
548 const struct lu_fid *lf, const char *name,
549 struct thandle *handle,
550 struct lustre_capa *capa)
552 struct dt_object *next = mdd_object_child(pobj);
556 if (dt_try_as_dir(env, next)) {
557 struct md_ucred *uc = md_ucred(env);
559 rc = next->do_index_ops->dio_insert(env, next,
561 (const struct dt_key *)name,
562 handle, capa, uc->mu_cap &
563 CFS_CAP_SYS_RESOURCE_MASK);
570 /* insert named index, add reference if isdir */
571 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
572 const struct lu_fid *lf, const char *name, int is_dir,
573 struct thandle *handle, struct lustre_capa *capa)
578 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
579 if (rc == 0 && is_dir) {
580 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
581 __mdd_ref_add(env, pobj, handle);
582 mdd_write_unlock(env, pobj);
587 /* delete named index, drop reference if isdir */
588 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
589 const char *name, int is_dir, struct thandle *handle,
590 struct lustre_capa *capa)
595 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
596 if (rc == 0 && is_dir) {
599 if (name != NULL && name[0] == '.' && name[1] == 0)
601 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
602 __mdd_ref_del(env, pobj, handle, is_dot);
603 mdd_write_unlock(env, pobj);
610 /** Store a namespace change changelog record
611 * If this fails, we must fail the whole transaction; we don't
612 * want the change to commit without the log entry.
613 * \param target - mdd_object of change
614 * \param parent - parent dir/object
615 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
616 * \param tname - target name string
617 * \param handle - transacion handle
619 static int mdd_changelog_ns_store(const struct lu_env *env,
620 struct mdd_device *mdd,
621 enum changelog_rec_type type,
622 struct mdd_object *target,
623 struct mdd_object *parent,
624 const struct lu_fid *tf,
625 const struct lu_name *tname,
626 struct thandle *handle)
628 const struct lu_fid *tfid;
629 const struct lu_fid *tpfid = mdo2fid(parent);
630 struct llog_changelog_rec *rec;
636 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
639 LASSERT(parent != NULL);
640 LASSERT(tname != NULL);
641 LASSERT(handle != NULL);
644 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
645 buf = mdd_buf_alloc(env, reclen);
646 if (buf->lb_buf == NULL)
648 rec = (struct llog_changelog_rec *)buf->lb_buf;
650 rec->cr.cr_flags = CLF_VERSION;
651 rec->cr.cr_type = (__u32)type;
652 tfid = tf ? tf : mdo2fid(target);
653 rec->cr.cr_tfid = *tfid;
654 rec->cr.cr_pfid = *tpfid;
655 rec->cr.cr_namelen = tname->ln_namelen;
656 memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
658 target->mod_cltime = cfs_time_current_64();
660 rc = mdd_changelog_llog_write(mdd, rec, handle);
662 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
663 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
670 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
671 struct md_object *src_obj, const struct lu_name *lname,
674 const char *name = lname->ln_name;
675 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
676 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
677 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
678 struct mdd_device *mdd = mdo2mdd(src_obj);
679 struct dynlock_handle *dlh;
680 struct thandle *handle;
681 #ifdef HAVE_QUOTA_SUPPORT
682 struct obd_device *obd = mdd->mdd_obd_dev;
683 struct obd_export *exp = md_quota(env)->mq_exp;
684 struct mds_obd *mds = &obd->u.mds;
685 unsigned int qids[MAXQUOTAS] = { 0, 0 };
686 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
691 #ifdef HAVE_QUOTA_SUPPORT
692 if (mds->mds_quota) {
693 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
695 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
698 mdd_data_get(env, mdd_tobj, &data);
699 quota_opc = FSFILT_OP_LINK;
700 mdd_quota_wrapper(la_tmp, qids);
701 /* get block quota for parent */
702 lquota_chkquota(mds_quota_interface_ref, obd, exp,
703 qids, rec_pending, 1, NULL,
704 LQUOTA_FLAGS_BLK, data, 1);
709 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
710 handle = mdd_trans_start(env, mdd);
712 GOTO(out_pending, rc = PTR_ERR(handle));
714 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
716 GOTO(out_trans, rc = -ENOMEM);
717 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
719 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
721 GOTO(out_unlock, rc);
723 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
725 mdd_object_capa(env, mdd_tobj));
727 GOTO(out_unlock, rc);
729 __mdd_ref_add(env, mdd_sobj, handle);
731 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
732 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
734 la->la_valid = LA_CTIME | LA_MTIME;
735 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
737 GOTO(out_unlock, rc);
739 la->la_valid = LA_CTIME;
740 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
742 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
746 mdd_write_unlock(env, mdd_sobj);
747 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
750 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
751 mdd_tobj, NULL, lname, handle);
752 mdd_trans_stop(env, mdd, rc, handle);
754 #ifdef HAVE_QUOTA_SUPPORT
756 lquota_pending_commit(mds_quota_interface_ref, obd,
757 qids, rec_pending, 1);
758 /* Trigger dqacq for the parent owner. If failed,
759 * the next call for lquota_chkquota will process it. */
760 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
767 /* caller should take a lock before calling */
768 int mdd_finish_unlink(const struct lu_env *env,
769 struct mdd_object *obj, struct md_attr *ma,
776 LASSERT(mdd_write_locked(env, obj) != 0);
778 rc = mdd_iattr_get(env, obj, ma);
779 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
780 obj->mod_flags |= DEAD_OBJ;
781 /* add new orphan and the object
782 * will be deleted during mdd_close() */
783 if (obj->mod_count) {
784 rc = __mdd_orphan_add(env, obj, th);
786 CDEBUG(D_HA, "Object "DFID" is inserted into "
787 "orphan list, open count = %d\n",
788 PFID(mdd_object_fid(obj)),
791 CERROR("Object "DFID" fail to be an orphan, "
792 "open count = %d, maybe cause failed "
794 PFID(mdd_object_fid(obj)),
797 rc = mdd_object_kill(env, obj, ma);
804 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
811 * has mdd_write_lock on cobj already, but not on pobj yet
813 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
814 struct mdd_object *cobj, struct md_attr *ma)
819 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
824 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
825 struct md_object *cobj, const struct lu_name *lname,
828 const char *name = lname->ln_name;
829 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
830 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
831 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
832 struct mdd_device *mdd = mdo2mdd(pobj);
833 struct dynlock_handle *dlh;
834 struct thandle *handle;
835 #ifdef HAVE_QUOTA_SUPPORT
836 struct obd_device *obd = mdd->mdd_obd_dev;
837 struct mds_obd *mds = &obd->u.mds;
838 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
839 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
842 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
846 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
847 PFID(mdd_object_fid(mdd_cobj)));
849 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
853 handle = mdd_trans_start(env, mdd);
855 RETURN(PTR_ERR(handle));
857 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
859 GOTO(out_trans, rc = -ENOMEM);
860 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
862 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
866 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
867 mdd_object_capa(env, mdd_pobj));
871 __mdd_ref_del(env, mdd_cobj, handle, 0);
874 __mdd_ref_del(env, mdd_cobj, handle, 1);
876 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
877 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
879 la->la_valid = LA_CTIME | LA_MTIME;
880 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
884 la->la_valid = LA_CTIME;
885 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
889 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
890 #ifdef HAVE_QUOTA_SUPPORT
891 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
892 ma->ma_attr.la_nlink == 0) {
893 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
895 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
897 mdd_quota_wrapper(la_tmp, qpids);
898 if (mdd_cobj->mod_count == 0) {
899 quota_opc = FSFILT_OP_UNLINK;
900 mdd_quota_wrapper(&ma->ma_attr, qcids);
902 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
908 /* old files may not have link ea; ignore errors */
909 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
910 lname, NULL, NULL, handle);
914 mdd_write_unlock(env, mdd_cobj);
915 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
918 rc = mdd_changelog_ns_store(env, mdd,
919 is_dir ? CL_RMDIR : CL_UNLINK,
920 mdd_cobj, mdd_pobj, NULL, lname,
923 mdd_trans_stop(env, mdd, rc, handle);
924 #ifdef HAVE_QUOTA_SUPPORT
926 /* Trigger dqrel on the owner of child and parent. If failed,
927 * the next call for lquota_chkquota will process it. */
928 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
934 /* has not lock on pobj yet */
935 static int mdd_ni_sanity_check(const struct lu_env *env,
936 struct md_object *pobj,
937 const struct md_attr *ma)
939 struct mdd_object *obj = md2mdd_obj(pobj);
943 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
946 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
954 static int mdd_name_insert(const struct lu_env *env,
955 struct md_object *pobj,
956 const struct lu_name *lname,
957 const struct lu_fid *fid,
958 const struct md_attr *ma)
960 const char *name = lname->ln_name;
961 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
962 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
963 struct mdd_device *mdd = mdo2mdd(pobj);
964 struct dynlock_handle *dlh;
965 struct thandle *handle;
966 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
967 #ifdef HAVE_QUOTA_SUPPORT
968 struct md_ucred *uc = md_ucred(env);
969 struct obd_device *obd = mdd->mdd_obd_dev;
970 struct obd_export *exp = md_quota(env)->mq_exp;
971 struct mds_obd *mds = &obd->u.mds;
972 unsigned int qids[MAXQUOTAS] = { 0, 0 };
973 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
974 cfs_cap_t save = uc->mu_cap;
979 #ifdef HAVE_QUOTA_SUPPORT
980 if (mds->mds_quota) {
981 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
982 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
984 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
987 mdd_data_get(env, mdd_obj, &data);
988 quota_opc = FSFILT_OP_LINK;
989 mdd_quota_wrapper(la_tmp, qids);
990 /* get block quota for parent */
991 lquota_chkquota(mds_quota_interface_ref, obd,
992 exp, qids, rec_pending, 1, NULL,
993 LQUOTA_FLAGS_BLK, data, 1);
996 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1000 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
1001 handle = mdd_trans_start(env, mdo2mdd(pobj));
1003 GOTO(out_pending, rc = PTR_ERR(handle));
1005 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1007 GOTO(out_trans, rc = -ENOMEM);
1009 rc = mdd_ni_sanity_check(env, pobj, ma);
1011 GOTO(out_unlock, rc);
1013 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1014 handle, BYPASS_CAPA);
1016 GOTO(out_unlock, rc);
1019 * For some case, no need update obj's ctime (LA_CTIME is not set),
1021 * For other cases, update obj's ctime (LA_CTIME is set),
1024 if (ma->ma_attr.la_valid & LA_CTIME) {
1025 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1026 la->la_valid = LA_CTIME | LA_MTIME;
1027 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1032 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1034 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1036 #ifdef HAVE_QUOTA_SUPPORT
1037 if (mds->mds_quota) {
1039 lquota_pending_commit(mds_quota_interface_ref,
1040 obd, qids, rec_pending, 1);
1041 /* Trigger dqacq for the parent owner. If failed,
1042 * the next call for lquota_chkquota will process it*/
1043 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1053 /* has not lock on pobj yet */
1054 static int mdd_nr_sanity_check(const struct lu_env *env,
1055 struct md_object *pobj,
1056 const struct md_attr *ma)
1058 struct mdd_object *obj = md2mdd_obj(pobj);
1062 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1065 rc = mdd_may_unlink(env, obj, ma);
1071 * Partial operation.
1073 static int mdd_name_remove(const struct lu_env *env,
1074 struct md_object *pobj,
1075 const struct lu_name *lname,
1076 const struct md_attr *ma)
1078 const char *name = lname->ln_name;
1079 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1080 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1081 struct mdd_device *mdd = mdo2mdd(pobj);
1082 struct dynlock_handle *dlh;
1083 struct thandle *handle;
1084 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1085 #ifdef HAVE_QUOTA_SUPPORT
1086 struct obd_device *obd = mdd->mdd_obd_dev;
1087 struct mds_obd *mds = &obd->u.mds;
1088 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1094 #ifdef HAVE_QUOTA_SUPPORT
1095 if (mds->mds_quota) {
1096 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1098 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1100 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1101 mdd_quota_wrapper(la_tmp, qids);
1105 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1106 handle = mdd_trans_start(env, mdd);
1108 GOTO(out_pending, rc = PTR_ERR(handle));
1110 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1112 GOTO(out_trans, rc = -ENOMEM);
1114 rc = mdd_nr_sanity_check(env, pobj, ma);
1116 GOTO(out_unlock, rc);
1118 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1119 handle, BYPASS_CAPA);
1121 GOTO(out_unlock, rc);
1124 * For some case, no need update obj's ctime (LA_CTIME is not set),
1126 * For other cases, update obj's ctime (LA_CTIME is set),
1129 if (ma->ma_attr.la_valid & LA_CTIME) {
1130 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1131 la->la_valid = LA_CTIME | LA_MTIME;
1132 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1137 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1139 mdd_trans_stop(env, mdd, rc, handle);
1141 #ifdef HAVE_QUOTA_SUPPORT
1142 /* Trigger dqrel for the parent owner.
1143 * If failed, the next call for lquota_chkquota will process it. */
1145 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1153 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1155 static int mdd_rt_sanity_check(const struct lu_env *env,
1156 struct mdd_object *tgt_pobj,
1157 struct mdd_object *tobj,
1163 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1166 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1167 * exist. In fact, tobj must exist, otherwise the call trace will be:
1168 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1169 * When get here, tobj must be NOT NULL, the other case has been
1170 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1172 * So check may_delete, but not check nlink of tgt_pobj. */
1174 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1179 /* Partial rename op on slave MDD */
1180 static int mdd_rename_tgt(const struct lu_env *env,
1181 struct md_object *pobj, struct md_object *tobj,
1182 const struct lu_fid *lf, const struct lu_name *lname,
1185 const char *name = lname->ln_name;
1186 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1187 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1188 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1189 struct mdd_device *mdd = mdo2mdd(pobj);
1190 struct dynlock_handle *dlh;
1191 struct thandle *handle;
1192 #ifdef HAVE_QUOTA_SUPPORT
1193 struct obd_device *obd = mdd->mdd_obd_dev;
1194 struct obd_export *exp = md_quota(env)->mq_exp;
1195 struct mds_obd *mds = &obd->u.mds;
1196 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1197 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1198 int quota_copc = 0, quota_popc = 0;
1199 int rec_pending[MAXQUOTAS] = { 0, 0 };
1204 #ifdef HAVE_QUOTA_SUPPORT
1205 if (mds->mds_quota && !tobj) {
1206 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1208 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1211 mdd_data_get(env, mdd_tpobj, &data);
1212 quota_popc = FSFILT_OP_LINK;
1213 mdd_quota_wrapper(la_tmp, qpids);
1214 /* get block quota for target parent */
1215 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1216 qpids, rec_pending, 1, NULL,
1217 LQUOTA_FLAGS_BLK, data, 1);
1221 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1222 handle = mdd_trans_start(env, mdd);
1224 GOTO(out_pending, rc = PTR_ERR(handle));
1226 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1228 GOTO(out_trans, rc = -ENOMEM);
1230 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1232 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1237 * If rename_tgt is called then we should just re-insert name with
1238 * correct fid, no need to dec/inc parent nlink if obj is dir.
1240 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1244 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1249 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1250 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1252 la->la_valid = LA_CTIME | LA_MTIME;
1253 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1258 * For tobj is remote case cmm layer has processed
1259 * and pass NULL tobj to here. So when tobj is NOT NULL,
1260 * it must be local one.
1262 if (tobj && mdd_object_exists(mdd_tobj)) {
1263 __mdd_ref_del(env, mdd_tobj, handle, 0);
1265 /* Remove dot reference. */
1266 if (S_ISDIR(ma->ma_attr.la_mode))
1267 __mdd_ref_del(env, mdd_tobj, handle, 1);
1269 la->la_valid = LA_CTIME;
1270 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1274 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1278 #ifdef HAVE_QUOTA_SUPPORT
1279 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1280 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1281 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1282 mdd_quota_wrapper(&ma->ma_attr, qcids);
1289 mdd_write_unlock(env, mdd_tobj);
1290 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1293 /* Bare EXT record with no RENAME in front of it signifies
1294 a partial slave op */
1295 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1296 mdd_tpobj, NULL, lname, handle);
1298 mdd_trans_stop(env, mdd, rc, handle);
1300 #ifdef HAVE_QUOTA_SUPPORT
1301 if (mds->mds_quota) {
1303 lquota_pending_commit(mds_quota_interface_ref, obd,
1304 qpids, rec_pending, 1);
1307 /* Trigger dqrel on the target owner of child.
1308 * If failed, the next call for lquota_chkquota
1309 * will process it. */
1310 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1318 * The permission has been checked when obj created, no need check again.
1320 static int mdd_cd_sanity_check(const struct lu_env *env,
1321 struct mdd_object *obj)
1326 if (!obj || mdd_is_dead_obj(obj))
1333 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1334 struct md_object *cobj, const struct md_op_spec *spec,
1337 struct mdd_device *mdd = mdo2mdd(cobj);
1338 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1339 struct mdd_object *son = md2mdd_obj(cobj);
1340 struct lu_attr *attr = &ma->ma_attr;
1341 struct lov_mds_md *lmm = NULL;
1343 struct thandle *handle;
1347 rc = mdd_cd_sanity_check(env, son);
1351 if (!md_should_create(spec->sp_cr_flags))
1353 lmm_size = ma->ma_lmm_size;
1354 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1359 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1360 handle = mdd_trans_start(env, mdd);
1362 GOTO(out_free, rc = PTR_ERR(handle));
1365 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1366 * Should this be fixed?
1369 /* Replay creates has objects already */
1371 if (spec->no_create) {
1372 CDEBUG(D_INFO, "we already have lov ea\n");
1373 rc = mdd_lov_set_md(env, mdd_pobj, son,
1374 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1375 spec->u.sp_ea.eadatalen, handle, 0);
1378 /* No need mdd_lsm_sanity_check here */
1379 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1380 lmm_size, handle, 0);
1383 rc = mdd_attr_get_internal_locked(env, son, ma);
1385 /* update lov_objid data, must be before transaction stop! */
1387 mdd_lov_objid_update(mdd, lmm);
1389 mdd_trans_stop(env, mdd, rc, handle);
1391 /* Finish mdd_lov_create() stuff. */
1392 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1396 /* Get fid from name and parent */
1398 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1399 const struct lu_name *lname, struct lu_fid* fid, int mask)
1401 const char *name = lname->ln_name;
1402 const struct dt_key *key = (const struct dt_key *)name;
1403 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1404 struct mdd_device *m = mdo2mdd(pobj);
1405 struct dt_object *dir = mdd_object_child(mdd_obj);
1409 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1412 rc = mdd_object_exists(mdd_obj);
1413 if (unlikely(rc == 0))
1415 else if (unlikely(rc < 0)) {
1416 CERROR("Object "DFID" locates on remote server\n",
1417 PFID(mdo2fid(mdd_obj)));
1421 /* The common filename length check. */
1422 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1423 RETURN(-ENAMETOOLONG);
1425 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1430 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1431 dt_try_as_dir(env, dir))) {
1433 rc = dir->do_index_ops->dio_lookup(env, dir,
1434 (struct dt_rec *)fid, key,
1435 mdd_object_capa(env, mdd_obj));
1446 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1447 const struct lu_name *lname, struct mdd_object *child,
1448 struct md_attr *ma, struct thandle *handle,
1449 const struct md_op_spec *spec)
1455 * Update attributes for child.
1458 * (1) the valid bits should be converted between Lustre and Linux;
1459 * (2) maybe, the child attributes should be set in OSD when creation.
1462 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1466 if (S_ISDIR(ma->ma_attr.la_mode)) {
1467 /* Add "." and ".." for newly created dir */
1468 __mdd_ref_add(env, child, handle);
1469 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1470 dot, handle, BYPASS_CAPA);
1472 rc = __mdd_index_insert_only(env, child, pfid,
1476 __mdd_ref_del(env, child, handle, 1);
1479 mdd_links_add(env, child, pfid, lname, handle);
1484 /* has not lock on pobj yet */
1485 static int mdd_create_sanity_check(const struct lu_env *env,
1486 struct md_object *pobj,
1487 const struct lu_name *lname,
1489 struct md_op_spec *spec)
1491 struct mdd_thread_info *info = mdd_env_info(env);
1492 struct lu_attr *la = &info->mti_la;
1493 struct lu_fid *fid = &info->mti_fid;
1494 struct mdd_object *obj = md2mdd_obj(pobj);
1495 struct mdd_device *m = mdo2mdd(pobj);
1496 int lookup = spec->sp_cr_lookup;
1501 if (mdd_is_dead_obj(obj))
1505 * In some cases this lookup is not needed - we know before if name
1506 * exists or not because MDT performs lookup for it.
1507 * name length check is done in lookup.
1511 * Check if the name already exist, though it will be checked in
1512 * _index_insert also, for avoiding rolling back if exists
1515 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1516 MAY_WRITE | MAY_EXEC);
1518 RETURN(rc ? : -EEXIST);
1521 * Check WRITE permission for the parent.
1522 * EXEC permission have been checked
1523 * when lookup before create already.
1525 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1532 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1536 if (la->la_mode & S_ISGID) {
1537 ma->ma_attr.la_gid = la->la_gid;
1538 if (S_ISDIR(ma->ma_attr.la_mode)) {
1539 ma->ma_attr.la_mode |= S_ISGID;
1540 ma->ma_attr.la_valid |= LA_MODE;
1544 switch (ma->ma_attr.la_mode & S_IFMT) {
1546 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1548 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1549 RETURN(-ENAMETOOLONG);
1569 * Create object and insert it into namespace.
1571 static int mdd_create(const struct lu_env *env,
1572 struct md_object *pobj,
1573 const struct lu_name *lname,
1574 struct md_object *child,
1575 struct md_op_spec *spec,
1578 struct mdd_thread_info *info = mdd_env_info(env);
1579 struct lu_attr *la = &info->mti_la_for_fix;
1580 struct md_attr *ma_acl = &info->mti_ma;
1581 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1582 struct mdd_object *son = md2mdd_obj(child);
1583 struct mdd_device *mdd = mdo2mdd(pobj);
1584 struct lu_attr *attr = &ma->ma_attr;
1585 struct lov_mds_md *lmm = NULL;
1586 struct thandle *handle;
1587 struct dynlock_handle *dlh;
1588 const char *name = lname->ln_name;
1589 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1590 int got_def_acl = 0;
1591 #ifdef HAVE_QUOTA_SUPPORT
1592 struct obd_device *obd = mdd->mdd_obd_dev;
1593 struct obd_export *exp = md_quota(env)->mq_exp;
1594 struct mds_obd *mds = &obd->u.mds;
1595 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1596 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1597 int quota_opc = 0, block_count = 0;
1598 int inode_pending[MAXQUOTAS] = { 0, 0 };
1599 int block_pending[MAXQUOTAS] = { 0, 0 };
1600 int parent_pending[MAXQUOTAS] = { 0, 0 };
1605 * Two operations have to be performed:
1607 * - an allocation of a new object (->do_create()), and
1609 * - an insertion into a parent index (->dio_insert()).
1611 * Due to locking, operation order is not important, when both are
1612 * successful, *but* error handling cases are quite different:
1614 * - if insertion is done first, and following object creation fails,
1615 * insertion has to be rolled back, but this operation might fail
1616 * also leaving us with dangling index entry.
1618 * - if creation is done first, is has to be undone if insertion
1619 * fails, leaving us with leaked space, which is neither good, nor
1622 * It seems that creation-first is simplest solution, but it is
1623 * sub-optimal in the frequent
1628 * case, because second mkdir is bound to create object, only to
1629 * destroy it immediately.
1631 * To avoid this follow local file systems that do double lookup:
1633 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1635 * 1. create (mdd_object_create_internal())
1637 * 2. insert (__mdd_index_insert(), lookup again)
1640 /* Sanity checks before big job. */
1641 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1645 #ifdef HAVE_QUOTA_SUPPORT
1646 if (mds->mds_quota) {
1647 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1649 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1653 quota_opc = FSFILT_OP_CREATE;
1654 mdd_quota_wrapper(&ma->ma_attr, qcids);
1655 mdd_quota_wrapper(la_tmp, qpids);
1656 /* get file quota for child */
1657 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1658 qcids, inode_pending, 1, NULL, 0, NULL,
1660 switch (ma->ma_attr.la_mode & S_IFMT) {
1669 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1670 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1674 /* get block quota for child and parent */
1676 lquota_chkquota(mds_quota_interface_ref, obd,
1677 exp, qcids, block_pending,
1679 LQUOTA_FLAGS_BLK, NULL, 0);
1681 lquota_chkquota(mds_quota_interface_ref, obd,
1682 exp, qpids, parent_pending, 1,
1683 NULL, LQUOTA_FLAGS_BLK, NULL,
1690 * No RPC inside the transaction, so OST objects should be created at
1693 if (S_ISREG(attr->la_mode)) {
1694 lmm_size = ma->ma_lmm_size;
1695 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1698 GOTO(out_pending, rc);
1701 if (!S_ISLNK(attr->la_mode)) {
1702 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1703 ma_acl->ma_acl = info->mti_xattr_buf;
1704 ma_acl->ma_need = MA_ACL_DEF;
1705 ma_acl->ma_valid = 0;
1707 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1708 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1709 mdd_read_unlock(env, mdd_pobj);
1712 else if (ma_acl->ma_valid & MA_ACL_DEF)
1716 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1717 handle = mdd_trans_start(env, mdd);
1719 GOTO(out_free, rc = PTR_ERR(handle));
1721 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1723 GOTO(out_trans, rc = -ENOMEM);
1725 mdd_write_lock(env, son, MOR_TGT_CHILD);
1726 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1728 mdd_write_unlock(env, son);
1734 #ifdef CONFIG_FS_POSIX_ACL
1736 struct lu_buf *acl_buf = &info->mti_buf;
1737 acl_buf->lb_buf = ma_acl->ma_acl;
1738 acl_buf->lb_len = ma_acl->ma_acl_size;
1740 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1742 mdd_write_unlock(env, son);
1745 ma->ma_attr.la_valid |= LA_MODE;
1750 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1751 son, ma, handle, spec);
1752 mdd_write_unlock(env, son);
1755 * Object has no links, so it will be destroyed when last
1756 * reference is released. (XXX not now.)
1762 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1763 name, S_ISDIR(attr->la_mode), handle,
1764 mdd_object_capa(env, mdd_pobj));
1771 /* No need mdd_lsm_sanity_check here */
1772 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1774 CERROR("error on stripe info copy %d \n", rc);
1777 if (lmm && lmm_size > 0) {
1778 /* Set Lov here, do not get lmm again later */
1779 memcpy(ma->ma_lmm, lmm, lmm_size);
1780 ma->ma_lmm_size = lmm_size;
1781 ma->ma_valid |= MA_LOV;
1784 if (S_ISLNK(attr->la_mode)) {
1785 struct md_ucred *uc = md_ucred(env);
1786 struct dt_object *dt = mdd_object_child(son);
1787 const char *target_name = spec->u.sp_symname;
1788 int sym_len = strlen(target_name);
1789 const struct lu_buf *buf;
1792 buf = mdd_buf_get_const(env, target_name, sym_len);
1793 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1794 mdd_object_capa(env, son),
1796 CFS_CAP_SYS_RESOURCE_MASK);
1801 GOTO(cleanup, rc = -EFAULT);
1805 la->la_valid = LA_CTIME | LA_MTIME;
1806 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1810 /* Return attr back. */
1811 rc = mdd_attr_get_internal_locked(env, son, ma);
1814 if (rc && created) {
1818 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1819 S_ISDIR(attr->la_mode),
1820 handle, BYPASS_CAPA);
1822 CERROR("error can not cleanup destroy %d\n",
1827 mdd_write_lock(env, son, MOR_TGT_CHILD);
1828 __mdd_ref_del(env, son, handle, 0);
1829 if (initialized && S_ISDIR(attr->la_mode))
1830 __mdd_ref_del(env, son, handle, 1);
1831 mdd_write_unlock(env, son);
1835 /* update lov_objid data, must be before transaction stop! */
1837 mdd_lov_objid_update(mdd, lmm);
1839 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1842 rc = mdd_changelog_ns_store(env, mdd,
1843 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1844 S_ISREG(attr->la_mode) ? CL_CREATE :
1845 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1846 son, mdd_pobj, NULL, lname, handle);
1847 mdd_trans_stop(env, mdd, rc, handle);
1849 /* finis lov_create stuff, free all temporary data */
1850 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1852 #ifdef HAVE_QUOTA_SUPPORT
1854 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1856 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1858 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1860 /* Trigger dqacq on the owner of child and parent. If failed,
1861 * the next call for lquota_chkquota will process it. */
1862 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1870 * Get locks on parents in proper order
1871 * RETURN: < 0 - error, rename_order if successful
1879 static int mdd_rename_order(const struct lu_env *env,
1880 struct mdd_device *mdd,
1881 struct mdd_object *src_pobj,
1882 struct mdd_object *tgt_pobj)
1884 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1888 if (src_pobj == tgt_pobj)
1889 RETURN(MDD_RN_SAME);
1891 /* compared the parent child relationship of src_p&tgt_p */
1892 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1894 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1897 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1910 /* has not mdd_write{read}_lock on any obj yet. */
1911 static int mdd_rename_sanity_check(const struct lu_env *env,
1912 struct mdd_object *src_pobj,
1913 struct mdd_object *tgt_pobj,
1914 struct mdd_object *sobj,
1915 struct mdd_object *tobj,
1921 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1924 /* XXX: when get here, sobj must NOT be NULL,
1925 * the other case has been processed in cml_rename
1926 * before mdd_rename and enable MDS_PERM_BYPASS. */
1929 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1933 /* XXX: when get here, "tobj == NULL" means tobj must
1934 * NOT exist (neither on remote MDS, such case has been
1935 * processed in cml_rename before mdd_rename and enable
1937 * So check may_create, but not check may_unlink. */
1939 rc = mdd_may_create(env, tgt_pobj, NULL,
1940 (src_pobj != tgt_pobj), 0);
1942 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1943 (src_pobj != tgt_pobj), 1);
1945 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1946 S_ISDIR(ma->ma_attr.la_mode))
1947 rc = __mdd_may_link(env, tgt_pobj);
1952 /* src object can be remote that is why we use only fid and type of object */
1953 static int mdd_rename(const struct lu_env *env,
1954 struct md_object *src_pobj, struct md_object *tgt_pobj,
1955 const struct lu_fid *lf, const struct lu_name *lsname,
1956 struct md_object *tobj, const struct lu_name *ltname,
1959 const char *sname = lsname->ln_name;
1960 const char *tname = ltname->ln_name;
1961 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1962 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1963 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1964 struct mdd_device *mdd = mdo2mdd(src_pobj);
1965 struct mdd_object *mdd_sobj = NULL; /* source object */
1966 struct mdd_object *mdd_tobj = NULL;
1967 struct dynlock_handle *sdlh, *tdlh;
1968 struct thandle *handle;
1969 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1970 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1974 #ifdef HAVE_QUOTA_SUPPORT
1975 struct obd_device *obd = mdd->mdd_obd_dev;
1976 struct obd_export *exp = md_quota(env)->mq_exp;
1977 struct mds_obd *mds = &obd->u.mds;
1978 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1979 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1980 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1981 int quota_copc = 0, quota_popc = 0;
1982 int rec_pending[MAXQUOTAS] = { 0, 0 };
1986 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1987 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1990 mdd_tobj = md2mdd_obj(tobj);
1992 #ifdef HAVE_QUOTA_SUPPORT
1993 if (mds->mds_quota) {
1994 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1996 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
1998 mdd_quota_wrapper(la_tmp, qspids);
2000 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2004 mdd_data_get(env, mdd_tpobj, &data);
2005 quota_popc = FSFILT_OP_LINK;
2006 mdd_quota_wrapper(la_tmp, qtpids);
2007 /* get block quota for target parent */
2008 lquota_chkquota(mds_quota_interface_ref,
2010 rec_pending, 1, NULL,
2018 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2019 handle = mdd_trans_start(env, mdd);
2021 GOTO(out_pending, rc = PTR_ERR(handle));
2023 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2024 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2026 GOTO(cleanup_unlocked, rc);
2028 /* Get locks in determined order */
2029 if (rc == MDD_RN_SAME) {
2030 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2031 sname, MOR_SRC_PARENT);
2032 /* check hashes to determine do we need one lock or two */
2033 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2034 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2038 } else if (rc == MDD_RN_SRCTGT) {
2039 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2040 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2042 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2043 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2045 if (sdlh == NULL || tdlh == NULL)
2046 GOTO(cleanup, rc = -ENOMEM);
2048 mdd_sobj = mdd_object_find(env, mdd, lf);
2049 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2050 mdd_sobj, mdd_tobj, ma);
2054 /* Remove source name from source directory */
2055 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2056 mdd_object_capa(env, mdd_spobj));
2060 /* "mv dir1 dir2" needs "dir1/.." link update */
2061 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2062 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2063 mdd_object_capa(env, mdd_sobj));
2065 GOTO(fixup_spobj2, rc);
2067 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2068 handle, mdd_object_capa(env, mdd_sobj));
2070 GOTO(fixup_spobj, rc);
2073 /* Remove target name from target directory
2074 * Here tobj can be remote one, so we do index_delete unconditionally
2075 * and -ENOENT is allowed.
2077 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2078 mdd_object_capa(env, mdd_tpobj));
2081 /* tname might been renamed to something else */
2082 GOTO(fixup_spobj, rc);
2085 GOTO(fixup_spobj, rc);
2088 /* Insert new fid with target name into target dir */
2089 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2090 mdd_object_capa(env, mdd_tpobj));
2092 GOTO(fixup_tpobj, rc);
2094 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2095 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2097 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2099 la->la_valid = LA_CTIME;
2100 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2103 GOTO(fixup_tpobj, rc);
2106 /* Remove old target object
2107 * For tobj is remote case cmm layer has processed
2108 * and set tobj to NULL then. So when tobj is NOT NULL,
2109 * it must be local one.
2111 if (tobj && mdd_object_exists(mdd_tobj)) {
2112 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2113 if (mdd_is_dead_obj(mdd_tobj)) {
2114 mdd_write_unlock(env, mdd_tobj);
2115 /* shld not be dead, something is wrong */
2116 CERROR("tobj is dead, something is wrong\n");
2120 __mdd_ref_del(env, mdd_tobj, handle, 0);
2122 /* Remove dot reference. */
2124 __mdd_ref_del(env, mdd_tobj, handle, 1);
2126 la->la_valid = LA_CTIME;
2127 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2129 GOTO(fixup_tpobj, rc);
2131 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2132 mdd_write_unlock(env, mdd_tobj);
2134 GOTO(fixup_tpobj, rc);
2136 #ifdef HAVE_QUOTA_SUPPORT
2137 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2138 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2139 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2140 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2145 la->la_valid = LA_CTIME | LA_MTIME;
2146 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2148 GOTO(fixup_tpobj, rc);
2150 if (mdd_spobj != mdd_tpobj) {
2151 la->la_valid = LA_CTIME | LA_MTIME;
2152 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2156 if (rc == 0 && mdd_sobj) {
2157 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2158 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2159 mdo2fid(mdd_tpobj), ltname, handle);
2161 /* Old files might not have EA entry */
2162 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2164 mdd_write_unlock(env, mdd_sobj);
2165 /* We don't fail the transaction if the link ea can't be
2166 updated -- fid2path will use alternate lookup method. */
2174 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2177 CWARN("tp obj fix error %d\n",rc2);
2179 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2180 !mdd_is_dead_obj(mdd_tobj)) {
2181 rc2 = __mdd_index_insert(env, mdd_tpobj,
2182 mdo2fid(mdd_tobj), tname,
2187 CWARN("tp obj fix error %d\n",rc2);
2192 if (rc && is_dir && mdd_sobj) {
2193 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2197 CWARN("sp obj dotdot delete error %d\n",rc2);
2200 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2201 dotdot, handle, BYPASS_CAPA);
2203 CWARN("sp obj dotdot insert error %d\n",rc2);
2208 rc2 = __mdd_index_insert(env, mdd_spobj,
2209 lf, sname, is_dir, handle, BYPASS_CAPA);
2211 CWARN("sp obj fix error %d\n",rc2);
2214 if (likely(tdlh) && sdlh != tdlh)
2215 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2217 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2220 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2221 mdd_spobj, lf, lsname, handle);
2223 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2224 mdd_tpobj, lf, ltname, handle);
2226 mdd_trans_stop(env, mdd, rc, handle);
2228 mdd_object_put(env, mdd_sobj);
2230 #ifdef HAVE_QUOTA_SUPPORT
2231 if (mds->mds_quota) {
2233 lquota_pending_commit(mds_quota_interface_ref, obd,
2234 qtpids, rec_pending, 1);
2237 /* Trigger dqrel on the source owner of parent.
2238 * If failed, the next call for lquota_chkquota will
2240 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2241 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2243 /* Trigger dqrel on the target owner of child.
2244 * If failed, the next call for lquota_chkquota
2245 * will process it. */
2246 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2247 qtpids, rc, quota_copc);
2254 /** enable/disable storing of hardlink info */
2255 int mdd_linkea_enable = 1;
2256 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2257 "record hardlink info in EAs");
2259 /** Read the link EA into a temp buffer.
2260 * Uses the name_buf since it is generally large.
2261 * \retval IS_ERR err
2262 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2264 struct lu_buf *mdd_links_get(const struct lu_env *env,
2265 struct mdd_object *mdd_obj)
2268 struct lustre_capa *capa;
2269 struct link_ea_header *leh;
2272 /* First try a small buf */
2273 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2274 if (buf->lb_buf == NULL)
2275 return ERR_PTR(-ENOMEM);
2277 capa = mdd_object_capa(env, mdd_obj);
2278 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2279 if (rc == -ERANGE) {
2280 /* Buf was too small, figure out what we need. */
2282 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2285 buf = mdd_buf_alloc(env, rc);
2286 if (buf->lb_buf == NULL)
2287 return ERR_PTR(-ENOMEM);
2288 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2294 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2295 leh->leh_magic = LINK_EA_MAGIC;
2296 leh->leh_reccount = __swab32(leh->leh_reccount);
2297 leh->leh_len = __swab64(leh->leh_len);
2298 /* entries are swabbed by mdd_lee_unpack */
2300 if (leh->leh_magic != LINK_EA_MAGIC)
2301 return ERR_PTR(-EINVAL);
2302 if (leh->leh_reccount == 0)
2303 return ERR_PTR(-ENODATA);
2308 /** Pack a link_ea_entry.
2309 * All elements are stored as chars to avoid alignment issues.
2310 * Numbers are always big-endian
2311 * \retval record length
2313 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2314 const struct lu_fid *pfid)
2316 struct lu_fid tmpfid;
2319 fid_cpu_to_be(&tmpfid, pfid);
2320 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2321 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2322 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2324 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2325 lee->lee_reclen[1] = reclen & 0xff;
2329 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2330 struct lu_name *lname, struct lu_fid *pfid)
2332 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2333 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2334 fid_be_to_cpu(pfid, pfid);
2335 lname->ln_name = lee->lee_name;
2336 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2339 /** Add a record to the end of link ea buf */
2340 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2341 const struct lu_fid *pfid,
2342 const struct lu_name *lname)
2344 struct link_ea_header *leh;
2345 struct link_ea_entry *lee;
2348 if (lname == NULL || pfid == NULL)
2351 /* Make sure our buf is big enough for the new one */
2353 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2354 if (leh->leh_len + reclen > buf->lb_len) {
2355 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2360 lee = buf->lb_buf + leh->leh_len;
2361 reclen = mdd_lee_pack(lee, lname, pfid);
2362 leh->leh_len += reclen;
2363 leh->leh_reccount++;
2367 /* For pathologic linkers, we don't want to spend lots of time scanning the
2368 * link ea. Limit ourseleves to something reasonable; links not in the EA
2369 * can be looked up via (slower) parent lookup.
2371 #define LINKEA_MAX_COUNT 128
2373 static int mdd_links_add(const struct lu_env *env,
2374 struct mdd_object *mdd_obj,
2375 const struct lu_fid *pfid,
2376 const struct lu_name *lname,
2377 struct thandle *handle)
2380 struct link_ea_header *leh;
2384 if (!mdd_linkea_enable)
2387 buf = mdd_links_get(env, mdd_obj);
2390 if (rc != -ENODATA) {
2391 CERROR("link_ea read failed %d "DFID"\n", rc,
2392 PFID(mdd_object_fid(mdd_obj)));
2395 /* empty EA; start one */
2396 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2397 if (buf->lb_buf == NULL)
2400 leh->leh_magic = LINK_EA_MAGIC;
2401 leh->leh_len = sizeof(struct link_ea_header);
2402 leh->leh_reccount = 0;
2406 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2409 rc = __mdd_links_add(env, buf, pfid, lname);
2414 rc = __mdd_xattr_set(env, mdd_obj,
2415 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2416 XATTR_NAME_LINK, 0, handle);
2419 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2420 PFID(mdd_object_fid(mdd_obj)));
2422 CERROR("link_ea add failed %d "DFID"\n", rc,
2423 PFID(mdd_object_fid(mdd_obj)));
2426 if (buf->lb_vmalloc)
2427 /* if we vmalloced a large buffer drop it */
2433 static int mdd_links_rename(const struct lu_env *env,
2434 struct mdd_object *mdd_obj,
2435 const struct lu_fid *oldpfid,
2436 const struct lu_name *oldlname,
2437 const struct lu_fid *newpfid,
2438 const struct lu_name *newlname,
2439 struct thandle *handle)
2442 struct link_ea_header *leh;
2443 struct link_ea_entry *lee;
2444 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2445 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2451 if (!mdd_linkea_enable)
2454 if (mdd_obj->mod_flags & DEAD_OBJ)
2455 /* No more links, don't bother */
2458 buf = mdd_links_get(env, mdd_obj);
2462 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2463 rc, PFID(mdd_object_fid(mdd_obj)));
2465 CERROR("link_ea read failed %d "DFID"\n",
2466 rc, PFID(mdd_object_fid(mdd_obj)));
2470 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2472 /* Find the old record */
2473 for(count = 0; count < leh->leh_reccount; count++) {
2474 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2475 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2476 lu_fid_eq(tmpfid, oldpfid) &&
2477 (strncmp(tmpname->ln_name, oldlname->ln_name,
2478 tmpname->ln_namelen) == 0))
2480 lee = (struct link_ea_entry *)((char *)lee + reclen);
2482 if ((count + 1) > leh->leh_reccount) {
2483 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2484 oldlname->ln_namelen, oldlname->ln_name);
2485 GOTO(out, rc = -ENOENT);
2488 /* Remove the old record */
2489 leh->leh_reccount--;
2490 leh->leh_len -= reclen;
2491 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2494 /* If renaming, add the new record */
2495 if (newpfid != NULL) {
2496 /* if the add fails, we still delete the out-of-date old link */
2497 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2501 rc = __mdd_xattr_set(env, mdd_obj,
2502 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2503 XATTR_NAME_LINK, 0, handle);
2509 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2510 oldlname->ln_namelen, oldlname->ln_name, rc,
2511 PFID(mdd_object_fid(mdd_obj)));
2513 if (buf->lb_vmalloc)
2514 /* if we vmalloced a large buffer drop it */
2520 const struct md_dir_operations mdd_dir_ops = {
2521 .mdo_is_subdir = mdd_is_subdir,
2522 .mdo_lookup = mdd_lookup,
2523 .mdo_create = mdd_create,
2524 .mdo_rename = mdd_rename,
2525 .mdo_link = mdd_link,
2526 .mdo_unlink = mdd_unlink,
2527 .mdo_name_insert = mdd_name_insert,
2528 .mdo_name_remove = mdd_name_remove,
2529 .mdo_rename_tgt = mdd_rename_tgt,
2530 .mdo_create_data = mdd_create_data