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 mds_obd *mds = &obd->u.mds;
684 unsigned int qids[MAXQUOTAS] = { 0, 0 };
685 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
690 #ifdef HAVE_QUOTA_SUPPORT
691 if (mds->mds_quota) {
692 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
694 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
697 mdd_data_get(env, mdd_tobj, &data);
698 quota_opc = FSFILT_OP_LINK;
699 mdd_quota_wrapper(la_tmp, qids);
700 /* get block quota for parent */
701 lquota_chkquota(mds_quota_interface_ref, obd,
702 qids, rec_pending, 1, NULL,
703 LQUOTA_FLAGS_BLK, data, 1);
708 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
709 handle = mdd_trans_start(env, mdd);
711 GOTO(out_pending, rc = PTR_ERR(handle));
713 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
715 GOTO(out_trans, rc = -ENOMEM);
716 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
718 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
720 GOTO(out_unlock, rc);
722 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
724 mdd_object_capa(env, mdd_tobj));
726 GOTO(out_unlock, rc);
728 __mdd_ref_add(env, mdd_sobj, handle);
730 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
731 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
733 la->la_valid = LA_CTIME | LA_MTIME;
734 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
736 GOTO(out_unlock, rc);
738 la->la_valid = LA_CTIME;
739 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
741 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
745 mdd_write_unlock(env, mdd_sobj);
746 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
749 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
750 mdd_tobj, NULL, lname, handle);
751 mdd_trans_stop(env, mdd, rc, handle);
753 #ifdef HAVE_QUOTA_SUPPORT
755 lquota_pending_commit(mds_quota_interface_ref, obd,
756 qids, rec_pending, 1);
757 /* Trigger dqacq for the parent owner. If failed,
758 * the next call for lquota_chkquota will process it. */
759 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
766 /* caller should take a lock before calling */
767 int mdd_finish_unlink(const struct lu_env *env,
768 struct mdd_object *obj, struct md_attr *ma,
775 LASSERT(mdd_write_locked(env, obj) != 0);
777 rc = mdd_iattr_get(env, obj, ma);
778 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
779 obj->mod_flags |= DEAD_OBJ;
780 /* add new orphan and the object
781 * will be deleted during mdd_close() */
782 if (obj->mod_count) {
783 rc = __mdd_orphan_add(env, obj, th);
785 CDEBUG(D_HA, "Object "DFID" is inserted into "
786 "orphan list, open count = %d\n",
787 PFID(mdd_object_fid(obj)),
790 CERROR("Object "DFID" fail to be an orphan, "
791 "open count = %d, maybe cause failed "
793 PFID(mdd_object_fid(obj)),
796 rc = mdd_object_kill(env, obj, ma);
803 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
810 * has mdd_write_lock on cobj already, but not on pobj yet
812 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
813 struct mdd_object *cobj, struct md_attr *ma)
818 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
823 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
824 struct md_object *cobj, const struct lu_name *lname,
827 const char *name = lname->ln_name;
828 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
829 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
830 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
831 struct mdd_device *mdd = mdo2mdd(pobj);
832 struct dynlock_handle *dlh;
833 struct thandle *handle;
834 #ifdef HAVE_QUOTA_SUPPORT
835 struct obd_device *obd = mdd->mdd_obd_dev;
836 struct mds_obd *mds = &obd->u.mds;
837 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
838 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
841 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
845 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
846 PFID(mdd_object_fid(mdd_cobj)));
848 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
852 handle = mdd_trans_start(env, mdd);
854 RETURN(PTR_ERR(handle));
856 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
858 GOTO(out_trans, rc = -ENOMEM);
859 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
861 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
865 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
866 mdd_object_capa(env, mdd_pobj));
870 __mdd_ref_del(env, mdd_cobj, handle, 0);
873 __mdd_ref_del(env, mdd_cobj, handle, 1);
875 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
876 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
878 la->la_valid = LA_CTIME | LA_MTIME;
879 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
883 la->la_valid = LA_CTIME;
884 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
888 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
889 #ifdef HAVE_QUOTA_SUPPORT
890 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
891 ma->ma_attr.la_nlink == 0) {
892 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
894 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
896 mdd_quota_wrapper(la_tmp, qpids);
897 if (mdd_cobj->mod_count == 0) {
898 quota_opc = FSFILT_OP_UNLINK;
899 mdd_quota_wrapper(&ma->ma_attr, qcids);
901 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
908 obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
909 sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
912 /* old files may not have link ea; ignore errors */
913 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
914 lname, NULL, NULL, handle);
918 mdd_write_unlock(env, mdd_cobj);
919 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
922 rc = mdd_changelog_ns_store(env, mdd,
923 is_dir ? CL_RMDIR : CL_UNLINK,
924 mdd_cobj, mdd_pobj, NULL, lname,
927 mdd_trans_stop(env, mdd, rc, handle);
928 #ifdef HAVE_QUOTA_SUPPORT
930 /* Trigger dqrel on the owner of child and parent. If failed,
931 * the next call for lquota_chkquota will process it. */
932 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
938 /* has not lock on pobj yet */
939 static int mdd_ni_sanity_check(const struct lu_env *env,
940 struct md_object *pobj,
941 const struct md_attr *ma)
943 struct mdd_object *obj = md2mdd_obj(pobj);
947 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
950 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
958 static int mdd_name_insert(const struct lu_env *env,
959 struct md_object *pobj,
960 const struct lu_name *lname,
961 const struct lu_fid *fid,
962 const struct md_attr *ma)
964 const char *name = lname->ln_name;
965 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
966 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
967 struct mdd_device *mdd = mdo2mdd(pobj);
968 struct dynlock_handle *dlh;
969 struct thandle *handle;
970 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
971 #ifdef HAVE_QUOTA_SUPPORT
972 struct md_ucred *uc = md_ucred(env);
973 struct obd_device *obd = mdd->mdd_obd_dev;
974 struct mds_obd *mds = &obd->u.mds;
975 unsigned int qids[MAXQUOTAS] = { 0, 0 };
976 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
977 cfs_cap_t save = uc->mu_cap;
982 #ifdef HAVE_QUOTA_SUPPORT
983 if (mds->mds_quota) {
984 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
985 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
987 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
990 mdd_data_get(env, mdd_obj, &data);
991 quota_opc = FSFILT_OP_LINK;
992 mdd_quota_wrapper(la_tmp, qids);
993 /* get block quota for parent */
994 lquota_chkquota(mds_quota_interface_ref, obd,
995 qids, rec_pending, 1, NULL,
996 LQUOTA_FLAGS_BLK, data, 1);
999 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1003 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
1004 handle = mdd_trans_start(env, mdo2mdd(pobj));
1006 GOTO(out_pending, rc = PTR_ERR(handle));
1008 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1010 GOTO(out_trans, rc = -ENOMEM);
1012 rc = mdd_ni_sanity_check(env, pobj, ma);
1014 GOTO(out_unlock, rc);
1016 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1017 handle, BYPASS_CAPA);
1019 GOTO(out_unlock, rc);
1022 * For some case, no need update obj's ctime (LA_CTIME is not set),
1024 * For other cases, update obj's ctime (LA_CTIME is set),
1027 if (ma->ma_attr.la_valid & LA_CTIME) {
1028 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1029 la->la_valid = LA_CTIME | LA_MTIME;
1030 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1035 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1037 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1039 #ifdef HAVE_QUOTA_SUPPORT
1040 if (mds->mds_quota) {
1042 lquota_pending_commit(mds_quota_interface_ref,
1043 obd, qids, rec_pending, 1);
1044 /* Trigger dqacq for the parent owner. If failed,
1045 * the next call for lquota_chkquota will process it*/
1046 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1056 /* has not lock on pobj yet */
1057 static int mdd_nr_sanity_check(const struct lu_env *env,
1058 struct md_object *pobj,
1059 const struct md_attr *ma)
1061 struct mdd_object *obj = md2mdd_obj(pobj);
1065 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1068 rc = mdd_may_unlink(env, obj, ma);
1074 * Partial operation.
1076 static int mdd_name_remove(const struct lu_env *env,
1077 struct md_object *pobj,
1078 const struct lu_name *lname,
1079 const struct md_attr *ma)
1081 const char *name = lname->ln_name;
1082 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1083 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1084 struct mdd_device *mdd = mdo2mdd(pobj);
1085 struct dynlock_handle *dlh;
1086 struct thandle *handle;
1087 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1088 #ifdef HAVE_QUOTA_SUPPORT
1089 struct obd_device *obd = mdd->mdd_obd_dev;
1090 struct mds_obd *mds = &obd->u.mds;
1091 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1097 #ifdef HAVE_QUOTA_SUPPORT
1098 if (mds->mds_quota) {
1099 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1101 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1103 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1104 mdd_quota_wrapper(la_tmp, qids);
1108 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1109 handle = mdd_trans_start(env, mdd);
1111 GOTO(out_pending, rc = PTR_ERR(handle));
1113 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1115 GOTO(out_trans, rc = -ENOMEM);
1117 rc = mdd_nr_sanity_check(env, pobj, ma);
1119 GOTO(out_unlock, rc);
1121 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1122 handle, BYPASS_CAPA);
1124 GOTO(out_unlock, rc);
1127 * For some case, no need update obj's ctime (LA_CTIME is not set),
1129 * For other cases, update obj's ctime (LA_CTIME is set),
1132 if (ma->ma_attr.la_valid & LA_CTIME) {
1133 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1134 la->la_valid = LA_CTIME | LA_MTIME;
1135 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1140 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1142 mdd_trans_stop(env, mdd, rc, handle);
1144 #ifdef HAVE_QUOTA_SUPPORT
1145 /* Trigger dqrel for the parent owner.
1146 * If failed, the next call for lquota_chkquota will process it. */
1148 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1156 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1158 static int mdd_rt_sanity_check(const struct lu_env *env,
1159 struct mdd_object *tgt_pobj,
1160 struct mdd_object *tobj,
1166 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1169 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1170 * exist. In fact, tobj must exist, otherwise the call trace will be:
1171 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1172 * When get here, tobj must be NOT NULL, the other case has been
1173 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1175 * So check may_delete, but not check nlink of tgt_pobj. */
1177 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1182 /* Partial rename op on slave MDD */
1183 static int mdd_rename_tgt(const struct lu_env *env,
1184 struct md_object *pobj, struct md_object *tobj,
1185 const struct lu_fid *lf, const struct lu_name *lname,
1188 const char *name = lname->ln_name;
1189 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1190 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1191 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1192 struct mdd_device *mdd = mdo2mdd(pobj);
1193 struct dynlock_handle *dlh;
1194 struct thandle *handle;
1195 #ifdef HAVE_QUOTA_SUPPORT
1196 struct obd_device *obd = mdd->mdd_obd_dev;
1197 struct mds_obd *mds = &obd->u.mds;
1198 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1199 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1200 int quota_copc = 0, quota_popc = 0;
1201 int rec_pending[MAXQUOTAS] = { 0, 0 };
1206 #ifdef HAVE_QUOTA_SUPPORT
1207 if (mds->mds_quota && !tobj) {
1208 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1210 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1213 mdd_data_get(env, mdd_tpobj, &data);
1214 quota_popc = FSFILT_OP_LINK;
1215 mdd_quota_wrapper(la_tmp, qpids);
1216 /* get block quota for target parent */
1217 lquota_chkquota(mds_quota_interface_ref, obd,
1218 qpids, rec_pending, 1, NULL,
1219 LQUOTA_FLAGS_BLK, data, 1);
1223 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1224 handle = mdd_trans_start(env, mdd);
1226 GOTO(out_pending, rc = PTR_ERR(handle));
1228 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1230 GOTO(out_trans, rc = -ENOMEM);
1232 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1234 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1239 * If rename_tgt is called then we should just re-insert name with
1240 * correct fid, no need to dec/inc parent nlink if obj is dir.
1242 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1246 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1251 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1252 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1254 la->la_valid = LA_CTIME | LA_MTIME;
1255 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1260 * For tobj is remote case cmm layer has processed
1261 * and pass NULL tobj to here. So when tobj is NOT NULL,
1262 * it must be local one.
1264 if (tobj && mdd_object_exists(mdd_tobj)) {
1265 __mdd_ref_del(env, mdd_tobj, handle, 0);
1267 /* Remove dot reference. */
1268 if (S_ISDIR(ma->ma_attr.la_mode))
1269 __mdd_ref_del(env, mdd_tobj, handle, 1);
1271 la->la_valid = LA_CTIME;
1272 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1276 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1280 #ifdef HAVE_QUOTA_SUPPORT
1281 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1282 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1283 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1284 mdd_quota_wrapper(&ma->ma_attr, qcids);
1291 mdd_write_unlock(env, mdd_tobj);
1292 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1295 /* Bare EXT record with no RENAME in front of it signifies
1296 a partial slave op */
1297 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1298 mdd_tpobj, NULL, lname, handle);
1300 mdd_trans_stop(env, mdd, rc, handle);
1302 #ifdef HAVE_QUOTA_SUPPORT
1303 if (mds->mds_quota) {
1305 lquota_pending_commit(mds_quota_interface_ref, obd,
1306 qpids, rec_pending, 1);
1309 /* Trigger dqrel on the target owner of child.
1310 * If failed, the next call for lquota_chkquota
1311 * will process it. */
1312 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1320 * The permission has been checked when obj created, no need check again.
1322 static int mdd_cd_sanity_check(const struct lu_env *env,
1323 struct mdd_object *obj)
1328 if (!obj || mdd_is_dead_obj(obj))
1335 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1336 struct md_object *cobj, const struct md_op_spec *spec,
1339 struct mdd_device *mdd = mdo2mdd(cobj);
1340 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1341 struct mdd_object *son = md2mdd_obj(cobj);
1342 struct lu_attr *attr = &ma->ma_attr;
1343 struct lov_mds_md *lmm = NULL;
1345 struct thandle *handle;
1349 rc = mdd_cd_sanity_check(env, son);
1353 if (!md_should_create(spec->sp_cr_flags))
1355 lmm_size = ma->ma_lmm_size;
1356 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1361 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1362 handle = mdd_trans_start(env, mdd);
1364 GOTO(out_free, rc = PTR_ERR(handle));
1367 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1368 * Should this be fixed?
1371 /* Replay creates has objects already */
1373 if (spec->no_create) {
1374 CDEBUG(D_INFO, "we already have lov ea\n");
1375 rc = mdd_lov_set_md(env, mdd_pobj, son,
1376 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1377 spec->u.sp_ea.eadatalen, handle, 0);
1380 /* No need mdd_lsm_sanity_check here */
1381 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1382 lmm_size, handle, 0);
1385 rc = mdd_attr_get_internal_locked(env, son, ma);
1387 /* update lov_objid data, must be before transaction stop! */
1389 mdd_lov_objid_update(mdd, lmm);
1391 mdd_trans_stop(env, mdd, rc, handle);
1393 /* Finish mdd_lov_create() stuff. */
1394 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1398 /* Get fid from name and parent */
1400 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1401 const struct lu_name *lname, struct lu_fid* fid, int mask)
1403 const char *name = lname->ln_name;
1404 const struct dt_key *key = (const struct dt_key *)name;
1405 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1406 struct mdd_device *m = mdo2mdd(pobj);
1407 struct dt_object *dir = mdd_object_child(mdd_obj);
1411 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1414 rc = mdd_object_exists(mdd_obj);
1415 if (unlikely(rc == 0))
1417 else if (unlikely(rc < 0)) {
1418 CERROR("Object "DFID" locates on remote server\n",
1419 PFID(mdo2fid(mdd_obj)));
1423 /* The common filename length check. */
1424 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1425 RETURN(-ENAMETOOLONG);
1427 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1432 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1433 dt_try_as_dir(env, dir))) {
1435 rc = dir->do_index_ops->dio_lookup(env, dir,
1436 (struct dt_rec *)fid, key,
1437 mdd_object_capa(env, mdd_obj));
1448 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1449 const struct lu_name *lname, struct mdd_object *child,
1450 struct md_attr *ma, struct thandle *handle,
1451 const struct md_op_spec *spec)
1457 * Update attributes for child.
1460 * (1) the valid bits should be converted between Lustre and Linux;
1461 * (2) maybe, the child attributes should be set in OSD when creation.
1464 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1468 if (S_ISDIR(ma->ma_attr.la_mode)) {
1469 /* Add "." and ".." for newly created dir */
1470 __mdd_ref_add(env, child, handle);
1471 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1472 dot, handle, BYPASS_CAPA);
1474 rc = __mdd_index_insert_only(env, child, pfid,
1478 __mdd_ref_del(env, child, handle, 1);
1481 mdd_links_add(env, child, pfid, lname, handle);
1486 /* has not lock on pobj yet */
1487 static int mdd_create_sanity_check(const struct lu_env *env,
1488 struct md_object *pobj,
1489 const struct lu_name *lname,
1491 struct md_op_spec *spec)
1493 struct mdd_thread_info *info = mdd_env_info(env);
1494 struct lu_attr *la = &info->mti_la;
1495 struct lu_fid *fid = &info->mti_fid;
1496 struct mdd_object *obj = md2mdd_obj(pobj);
1497 struct mdd_device *m = mdo2mdd(pobj);
1498 int lookup = spec->sp_cr_lookup;
1503 if (mdd_is_dead_obj(obj))
1507 * In some cases this lookup is not needed - we know before if name
1508 * exists or not because MDT performs lookup for it.
1509 * name length check is done in lookup.
1513 * Check if the name already exist, though it will be checked in
1514 * _index_insert also, for avoiding rolling back if exists
1517 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1518 MAY_WRITE | MAY_EXEC);
1520 RETURN(rc ? : -EEXIST);
1523 * Check WRITE permission for the parent.
1524 * EXEC permission have been checked
1525 * when lookup before create already.
1527 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1534 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1538 if (la->la_mode & S_ISGID) {
1539 ma->ma_attr.la_gid = la->la_gid;
1540 if (S_ISDIR(ma->ma_attr.la_mode)) {
1541 ma->ma_attr.la_mode |= S_ISGID;
1542 ma->ma_attr.la_valid |= LA_MODE;
1546 switch (ma->ma_attr.la_mode & S_IFMT) {
1548 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1550 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1551 RETURN(-ENAMETOOLONG);
1571 * Create object and insert it into namespace.
1573 static int mdd_create(const struct lu_env *env,
1574 struct md_object *pobj,
1575 const struct lu_name *lname,
1576 struct md_object *child,
1577 struct md_op_spec *spec,
1580 struct mdd_thread_info *info = mdd_env_info(env);
1581 struct lu_attr *la = &info->mti_la_for_fix;
1582 struct md_attr *ma_acl = &info->mti_ma;
1583 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1584 struct mdd_object *son = md2mdd_obj(child);
1585 struct mdd_device *mdd = mdo2mdd(pobj);
1586 struct lu_attr *attr = &ma->ma_attr;
1587 struct lov_mds_md *lmm = NULL;
1588 struct thandle *handle;
1589 struct dynlock_handle *dlh;
1590 const char *name = lname->ln_name;
1591 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1592 int got_def_acl = 0;
1593 #ifdef HAVE_QUOTA_SUPPORT
1594 struct obd_device *obd = mdd->mdd_obd_dev;
1595 struct mds_obd *mds = &obd->u.mds;
1596 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1597 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1598 int quota_opc = 0, block_count = 0;
1599 int inode_pending[MAXQUOTAS] = { 0, 0 };
1600 int block_pending[MAXQUOTAS] = { 0, 0 };
1601 int parent_pending[MAXQUOTAS] = { 0, 0 };
1606 * Two operations have to be performed:
1608 * - an allocation of a new object (->do_create()), and
1610 * - an insertion into a parent index (->dio_insert()).
1612 * Due to locking, operation order is not important, when both are
1613 * successful, *but* error handling cases are quite different:
1615 * - if insertion is done first, and following object creation fails,
1616 * insertion has to be rolled back, but this operation might fail
1617 * also leaving us with dangling index entry.
1619 * - if creation is done first, is has to be undone if insertion
1620 * fails, leaving us with leaked space, which is neither good, nor
1623 * It seems that creation-first is simplest solution, but it is
1624 * sub-optimal in the frequent
1629 * case, because second mkdir is bound to create object, only to
1630 * destroy it immediately.
1632 * To avoid this follow local file systems that do double lookup:
1634 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1636 * 1. create (mdd_object_create_internal())
1638 * 2. insert (__mdd_index_insert(), lookup again)
1641 /* Sanity checks before big job. */
1642 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1646 #ifdef HAVE_QUOTA_SUPPORT
1647 if (mds->mds_quota) {
1648 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1650 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1654 quota_opc = FSFILT_OP_CREATE;
1655 mdd_quota_wrapper(&ma->ma_attr, qcids);
1656 mdd_quota_wrapper(la_tmp, qpids);
1657 /* get file quota for child */
1658 lquota_chkquota(mds_quota_interface_ref, obd, qcids,
1659 inode_pending, 1, NULL, 0, NULL, 0);
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 qcids, block_pending,
1679 LQUOTA_FLAGS_BLK, NULL, 0);
1681 lquota_chkquota(mds_quota_interface_ref, obd,
1682 qpids, parent_pending, 1, NULL,
1683 LQUOTA_FLAGS_BLK, NULL, 0);
1689 * No RPC inside the transaction, so OST objects should be created at
1692 if (S_ISREG(attr->la_mode)) {
1693 lmm_size = ma->ma_lmm_size;
1694 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1697 GOTO(out_pending, rc);
1700 if (!S_ISLNK(attr->la_mode)) {
1701 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1702 ma_acl->ma_acl = info->mti_xattr_buf;
1703 ma_acl->ma_need = MA_ACL_DEF;
1704 ma_acl->ma_valid = 0;
1706 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1707 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1708 mdd_read_unlock(env, mdd_pobj);
1711 else if (ma_acl->ma_valid & MA_ACL_DEF)
1715 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1716 handle = mdd_trans_start(env, mdd);
1718 GOTO(out_free, rc = PTR_ERR(handle));
1720 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1722 GOTO(out_trans, rc = -ENOMEM);
1724 mdd_write_lock(env, son, MOR_TGT_CHILD);
1725 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1727 mdd_write_unlock(env, son);
1733 #ifdef CONFIG_FS_POSIX_ACL
1735 struct lu_buf *acl_buf = &info->mti_buf;
1736 acl_buf->lb_buf = ma_acl->ma_acl;
1737 acl_buf->lb_len = ma_acl->ma_acl_size;
1739 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1741 mdd_write_unlock(env, son);
1744 ma->ma_attr.la_valid |= LA_MODE;
1749 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1750 son, ma, handle, spec);
1751 mdd_write_unlock(env, son);
1754 * Object has no links, so it will be destroyed when last
1755 * reference is released. (XXX not now.)
1761 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1762 name, S_ISDIR(attr->la_mode), handle,
1763 mdd_object_capa(env, mdd_pobj));
1770 /* No need mdd_lsm_sanity_check here */
1771 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1773 CERROR("error on stripe info copy %d \n", rc);
1776 if (lmm && lmm_size > 0) {
1777 /* Set Lov here, do not get lmm again later */
1778 memcpy(ma->ma_lmm, lmm, lmm_size);
1779 ma->ma_lmm_size = lmm_size;
1780 ma->ma_valid |= MA_LOV;
1783 if (S_ISLNK(attr->la_mode)) {
1784 struct md_ucred *uc = md_ucred(env);
1785 struct dt_object *dt = mdd_object_child(son);
1786 const char *target_name = spec->u.sp_symname;
1787 int sym_len = strlen(target_name);
1788 const struct lu_buf *buf;
1791 buf = mdd_buf_get_const(env, target_name, sym_len);
1792 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1793 mdd_object_capa(env, son),
1795 CFS_CAP_SYS_RESOURCE_MASK);
1800 GOTO(cleanup, rc = -EFAULT);
1804 la->la_valid = LA_CTIME | LA_MTIME;
1805 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1809 /* Return attr back. */
1810 rc = mdd_attr_get_internal_locked(env, son, ma);
1813 if (rc && created) {
1817 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1818 S_ISDIR(attr->la_mode),
1819 handle, BYPASS_CAPA);
1821 CERROR("error can not cleanup destroy %d\n",
1826 mdd_write_lock(env, son, MOR_TGT_CHILD);
1827 __mdd_ref_del(env, son, handle, 0);
1828 if (initialized && S_ISDIR(attr->la_mode))
1829 __mdd_ref_del(env, son, handle, 1);
1830 mdd_write_unlock(env, son);
1834 /* update lov_objid data, must be before transaction stop! */
1836 mdd_lov_objid_update(mdd, lmm);
1838 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1841 rc = mdd_changelog_ns_store(env, mdd,
1842 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1843 S_ISREG(attr->la_mode) ? CL_CREATE :
1844 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1845 son, mdd_pobj, NULL, lname, handle);
1846 mdd_trans_stop(env, mdd, rc, handle);
1848 /* finis lov_create stuff, free all temporary data */
1849 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1851 #ifdef HAVE_QUOTA_SUPPORT
1853 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1855 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1857 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1859 /* Trigger dqacq on the owner of child and parent. If failed,
1860 * the next call for lquota_chkquota will process it. */
1861 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1869 * Get locks on parents in proper order
1870 * RETURN: < 0 - error, rename_order if successful
1878 static int mdd_rename_order(const struct lu_env *env,
1879 struct mdd_device *mdd,
1880 struct mdd_object *src_pobj,
1881 struct mdd_object *tgt_pobj)
1883 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1887 if (src_pobj == tgt_pobj)
1888 RETURN(MDD_RN_SAME);
1890 /* compared the parent child relationship of src_p&tgt_p */
1891 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1893 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1896 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1909 /* has not mdd_write{read}_lock on any obj yet. */
1910 static int mdd_rename_sanity_check(const struct lu_env *env,
1911 struct mdd_object *src_pobj,
1912 struct mdd_object *tgt_pobj,
1913 struct mdd_object *sobj,
1914 struct mdd_object *tobj,
1920 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1923 /* XXX: when get here, sobj must NOT be NULL,
1924 * the other case has been processed in cml_rename
1925 * before mdd_rename and enable MDS_PERM_BYPASS. */
1928 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1932 /* XXX: when get here, "tobj == NULL" means tobj must
1933 * NOT exist (neither on remote MDS, such case has been
1934 * processed in cml_rename before mdd_rename and enable
1936 * So check may_create, but not check may_unlink. */
1938 rc = mdd_may_create(env, tgt_pobj, NULL,
1939 (src_pobj != tgt_pobj), 0);
1941 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1942 (src_pobj != tgt_pobj), 1);
1944 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1945 S_ISDIR(ma->ma_attr.la_mode))
1946 rc = __mdd_may_link(env, tgt_pobj);
1951 /* src object can be remote that is why we use only fid and type of object */
1952 static int mdd_rename(const struct lu_env *env,
1953 struct md_object *src_pobj, struct md_object *tgt_pobj,
1954 const struct lu_fid *lf, const struct lu_name *lsname,
1955 struct md_object *tobj, const struct lu_name *ltname,
1958 const char *sname = lsname->ln_name;
1959 const char *tname = ltname->ln_name;
1960 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1961 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1962 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1963 struct mdd_device *mdd = mdo2mdd(src_pobj);
1964 struct mdd_object *mdd_sobj = NULL; /* source object */
1965 struct mdd_object *mdd_tobj = NULL;
1966 struct dynlock_handle *sdlh, *tdlh;
1967 struct thandle *handle;
1968 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1969 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1973 #ifdef HAVE_QUOTA_SUPPORT
1974 struct obd_device *obd = mdd->mdd_obd_dev;
1975 struct mds_obd *mds = &obd->u.mds;
1976 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1977 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1978 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1979 int quota_copc = 0, quota_popc = 0;
1980 int rec_pending[MAXQUOTAS] = { 0, 0 };
1984 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1985 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1988 mdd_tobj = md2mdd_obj(tobj);
1990 #ifdef HAVE_QUOTA_SUPPORT
1991 if (mds->mds_quota) {
1992 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1994 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
1996 mdd_quota_wrapper(la_tmp, qspids);
1998 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2002 mdd_data_get(env, mdd_tpobj, &data);
2003 quota_popc = FSFILT_OP_LINK;
2004 mdd_quota_wrapper(la_tmp, qtpids);
2005 /* get block quota for target parent */
2006 lquota_chkquota(mds_quota_interface_ref,
2008 rec_pending, 1, NULL,
2016 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2017 handle = mdd_trans_start(env, mdd);
2019 GOTO(out_pending, rc = PTR_ERR(handle));
2021 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2022 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2024 GOTO(cleanup_unlocked, rc);
2026 /* Get locks in determined order */
2027 if (rc == MDD_RN_SAME) {
2028 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2029 sname, MOR_SRC_PARENT);
2030 /* check hashes to determine do we need one lock or two */
2031 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2032 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2036 } else if (rc == MDD_RN_SRCTGT) {
2037 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2038 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2040 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2041 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2043 if (sdlh == NULL || tdlh == NULL)
2044 GOTO(cleanup, rc = -ENOMEM);
2046 mdd_sobj = mdd_object_find(env, mdd, lf);
2047 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2048 mdd_sobj, mdd_tobj, ma);
2052 /* Remove source name from source directory */
2053 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2054 mdd_object_capa(env, mdd_spobj));
2058 /* "mv dir1 dir2" needs "dir1/.." link update */
2059 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2060 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2061 mdd_object_capa(env, mdd_sobj));
2063 GOTO(fixup_spobj2, rc);
2065 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2066 handle, mdd_object_capa(env, mdd_sobj));
2068 GOTO(fixup_spobj, rc);
2071 /* Remove target name from target directory
2072 * Here tobj can be remote one, so we do index_delete unconditionally
2073 * and -ENOENT is allowed.
2075 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2076 mdd_object_capa(env, mdd_tpobj));
2079 /* tname might been renamed to something else */
2080 GOTO(fixup_spobj, rc);
2083 GOTO(fixup_spobj, rc);
2086 /* Insert new fid with target name into target dir */
2087 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2088 mdd_object_capa(env, mdd_tpobj));
2090 GOTO(fixup_tpobj, rc);
2092 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2093 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2095 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2097 la->la_valid = LA_CTIME;
2098 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2101 GOTO(fixup_tpobj, rc);
2104 /* Remove old target object
2105 * For tobj is remote case cmm layer has processed
2106 * and set tobj to NULL then. So when tobj is NOT NULL,
2107 * it must be local one.
2109 if (tobj && mdd_object_exists(mdd_tobj)) {
2110 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2111 if (mdd_is_dead_obj(mdd_tobj)) {
2112 mdd_write_unlock(env, mdd_tobj);
2113 /* shld not be dead, something is wrong */
2114 CERROR("tobj is dead, something is wrong\n");
2118 __mdd_ref_del(env, mdd_tobj, handle, 0);
2120 /* Remove dot reference. */
2122 __mdd_ref_del(env, mdd_tobj, handle, 1);
2124 la->la_valid = LA_CTIME;
2125 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2127 GOTO(fixup_tpobj, rc);
2129 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2130 mdd_write_unlock(env, mdd_tobj);
2132 GOTO(fixup_tpobj, rc);
2134 #ifdef HAVE_QUOTA_SUPPORT
2135 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2136 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2137 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2138 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2143 la->la_valid = LA_CTIME | LA_MTIME;
2144 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2146 GOTO(fixup_tpobj, rc);
2148 if (mdd_spobj != mdd_tpobj) {
2149 la->la_valid = LA_CTIME | LA_MTIME;
2150 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2154 if (rc == 0 && mdd_sobj) {
2155 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2156 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2157 mdo2fid(mdd_tpobj), ltname, handle);
2159 /* Old files might not have EA entry */
2160 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2162 mdd_write_unlock(env, mdd_sobj);
2163 /* We don't fail the transaction if the link ea can't be
2164 updated -- fid2path will use alternate lookup method. */
2172 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2175 CWARN("tp obj fix error %d\n",rc2);
2177 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2178 !mdd_is_dead_obj(mdd_tobj)) {
2179 rc2 = __mdd_index_insert(env, mdd_tpobj,
2180 mdo2fid(mdd_tobj), tname,
2185 CWARN("tp obj fix error %d\n",rc2);
2190 if (rc && is_dir && mdd_sobj) {
2191 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2195 CWARN("sp obj dotdot delete error %d\n",rc2);
2198 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2199 dotdot, handle, BYPASS_CAPA);
2201 CWARN("sp obj dotdot insert error %d\n",rc2);
2206 rc2 = __mdd_index_insert(env, mdd_spobj,
2207 lf, sname, is_dir, handle, BYPASS_CAPA);
2209 CWARN("sp obj fix error %d\n",rc2);
2212 if (likely(tdlh) && sdlh != tdlh)
2213 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2215 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2218 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2219 mdd_spobj, lf, lsname, handle);
2221 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2222 mdd_tpobj, lf, ltname, handle);
2224 mdd_trans_stop(env, mdd, rc, handle);
2226 mdd_object_put(env, mdd_sobj);
2228 #ifdef HAVE_QUOTA_SUPPORT
2229 if (mds->mds_quota) {
2231 lquota_pending_commit(mds_quota_interface_ref, obd,
2232 qtpids, rec_pending, 1);
2235 /* Trigger dqrel on the source owner of parent.
2236 * If failed, the next call for lquota_chkquota will
2238 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2239 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2241 /* Trigger dqrel on the target owner of child.
2242 * If failed, the next call for lquota_chkquota
2243 * will process it. */
2244 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2245 qtpids, rc, quota_copc);
2252 /** enable/disable storing of hardlink info */
2253 int mdd_linkea_enable = 1;
2254 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2255 "record hardlink info in EAs");
2257 /** Read the link EA into a temp buffer.
2258 * Uses the name_buf since it is generally large.
2259 * \retval IS_ERR err
2260 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2262 struct lu_buf *mdd_links_get(const struct lu_env *env,
2263 struct mdd_object *mdd_obj)
2266 struct lustre_capa *capa;
2267 struct link_ea_header *leh;
2270 /* First try a small buf */
2271 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2272 if (buf->lb_buf == NULL)
2273 return ERR_PTR(-ENOMEM);
2275 capa = mdd_object_capa(env, mdd_obj);
2276 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2277 if (rc == -ERANGE) {
2278 /* Buf was too small, figure out what we need. */
2281 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2284 buf = mdd_buf_alloc(env, rc);
2285 if (buf->lb_buf == NULL)
2286 return ERR_PTR(-ENOMEM);
2287 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2293 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2294 leh->leh_magic = LINK_EA_MAGIC;
2295 leh->leh_reccount = __swab32(leh->leh_reccount);
2296 leh->leh_len = __swab64(leh->leh_len);
2297 /* entries are swabbed by mdd_lee_unpack */
2299 if (leh->leh_magic != LINK_EA_MAGIC)
2300 return ERR_PTR(-EINVAL);
2301 if (leh->leh_reccount == 0)
2302 return ERR_PTR(-ENODATA);
2307 /** Pack a link_ea_entry.
2308 * All elements are stored as chars to avoid alignment issues.
2309 * Numbers are always big-endian
2310 * \retval record length
2312 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2313 const struct lu_fid *pfid)
2317 fid_cpu_to_be(&lee->lee_parent_fid, pfid);
2318 strncpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2319 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2321 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2322 lee->lee_reclen[1] = reclen & 0xff;
2326 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2327 struct lu_name *lname, struct lu_fid *pfid)
2329 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2330 fid_be_to_cpu(pfid, &lee->lee_parent_fid);
2331 lname->ln_name = lee->lee_name;
2332 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2335 /** Add a record to the end of link ea buf */
2336 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2337 const struct lu_fid *pfid,
2338 const struct lu_name *lname)
2340 struct link_ea_header *leh;
2341 struct link_ea_entry *lee;
2344 if (lname == NULL || pfid == NULL)
2347 /* Make sure our buf is big enough for the new one */
2349 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2350 if (leh->leh_len + reclen > buf->lb_len) {
2351 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2356 lee = buf->lb_buf + leh->leh_len;
2357 reclen = mdd_lee_pack(lee, lname, pfid);
2358 leh->leh_len += reclen;
2359 leh->leh_reccount++;
2363 /* For pathologic linkers, we don't want to spend lots of time scanning the
2364 * link ea. Limit ourseleves to something reasonable; links not in the EA
2365 * can be looked up via (slower) parent lookup.
2367 #define LINKEA_MAX_COUNT 128
2369 static int mdd_links_add(const struct lu_env *env,
2370 struct mdd_object *mdd_obj,
2371 const struct lu_fid *pfid,
2372 const struct lu_name *lname,
2373 struct thandle *handle)
2376 struct link_ea_header *leh;
2380 if (!mdd_linkea_enable)
2383 buf = mdd_links_get(env, mdd_obj);
2386 if (rc != -ENODATA) {
2387 CERROR("link_ea read failed %d "DFID"\n", rc,
2388 PFID(mdd_object_fid(mdd_obj)));
2391 /* empty EA; start one */
2392 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2393 if (buf->lb_buf == NULL)
2396 leh->leh_magic = LINK_EA_MAGIC;
2397 leh->leh_len = sizeof(struct link_ea_header);
2398 leh->leh_reccount = 0;
2402 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2405 rc = __mdd_links_add(env, buf, pfid, lname);
2410 rc = __mdd_xattr_set(env, mdd_obj,
2411 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2412 XATTR_NAME_LINK, 0, handle);
2414 CERROR("link_ea add failed %d "DFID"\n", rc,
2415 PFID(mdd_object_fid(mdd_obj)));
2417 if (buf->lb_vmalloc)
2418 /* if we vmalloced a large buffer drop it */
2424 static int mdd_links_rename(const struct lu_env *env,
2425 struct mdd_object *mdd_obj,
2426 const struct lu_fid *oldpfid,
2427 const struct lu_name *oldlname,
2428 const struct lu_fid *newpfid,
2429 const struct lu_name *newlname,
2430 struct thandle *handle)
2433 struct link_ea_header *leh;
2434 struct link_ea_entry *lee;
2435 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2436 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2442 if (!mdd_linkea_enable)
2445 if (mdd_obj->mod_flags & DEAD_OBJ)
2446 /* No more links, don't bother */
2449 buf = mdd_links_get(env, mdd_obj);
2452 CERROR("link_ea read failed %d "DFID"\n",
2453 rc, PFID(mdd_object_fid(mdd_obj)));
2457 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2459 /* Find the old record */
2460 for(count = 0; count < leh->leh_reccount; count++) {
2461 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2462 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2463 lu_fid_eq(tmpfid, oldpfid) &&
2464 (strncmp(tmpname->ln_name, oldlname->ln_name,
2465 tmpname->ln_namelen) == 0))
2467 lee = (struct link_ea_entry *)((char *)lee + reclen);
2469 if ((count + 1) > leh->leh_reccount) {
2470 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2471 oldlname->ln_namelen, oldlname->ln_name);
2472 GOTO(out, rc = -ENOENT);
2475 /* Remove the old record */
2476 leh->leh_reccount--;
2477 leh->leh_len -= reclen;
2478 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2481 /* If renaming, add the new record */
2482 if (newpfid != NULL) {
2483 /* if the add fails, we still delete the out-of-date old link */
2484 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2488 rc = __mdd_xattr_set(env, mdd_obj,
2489 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2490 XATTR_NAME_LINK, 0, handle);
2496 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2497 oldlname->ln_namelen, oldlname->ln_name, rc,
2498 PFID(mdd_object_fid(mdd_obj)));
2500 if (buf->lb_vmalloc)
2501 /* if we vmalloced a large buffer drop it */
2507 const struct md_dir_operations mdd_dir_ops = {
2508 .mdo_is_subdir = mdd_is_subdir,
2509 .mdo_lookup = mdd_lookup,
2510 .mdo_create = mdd_create,
2511 .mdo_rename = mdd_rename,
2512 .mdo_link = mdd_link,
2513 .mdo_unlink = mdd_unlink,
2514 .mdo_name_insert = mdd_name_insert,
2515 .mdo_name_remove = mdd_name_remove,
2516 .mdo_rename_tgt = mdd_rename_tgt,
2517 .mdo_create_data = mdd_create_data