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 #include <linux/jbd.h>
51 #include <obd_class.h>
52 #include <lustre_ver.h>
53 #include <obd_support.h>
54 #include <lprocfs_status.h>
56 #include <linux/ldiskfs_fs.h>
57 #include <lustre_mds.h>
58 #include <lustre/lustre_idl.h>
59 #include <lustre_fid.h>
61 #include "mdd_internal.h"
63 static const char dot[] = ".";
64 static const char dotdot[] = "..";
66 static struct lu_name lname_dotdot = {
71 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
72 const struct lu_name *lname, struct lu_fid* fid,
74 static int mdd_links_add(const struct lu_env *env,
75 struct mdd_object *mdd_obj,
76 const struct lu_fid *pfid,
77 const struct lu_name *lname,
78 struct thandle *handle);
79 static int mdd_links_rename(const struct lu_env *env,
80 struct mdd_object *mdd_obj,
81 const struct lu_fid *oldpfid,
82 const struct lu_name *oldlname,
83 const struct lu_fid *newpfid,
84 const struct lu_name *newlname,
85 struct thandle *handle);
88 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
89 const struct lu_name *lname, struct lu_fid* fid, int mask)
91 const char *name = lname->ln_name;
92 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
93 struct dynlock_handle *dlh;
96 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
97 if (unlikely(dlh == NULL))
99 rc = __mdd_lookup(env, pobj, lname, fid, mask);
100 mdd_pdo_read_unlock(env, mdd_obj, dlh);
105 int mdd_lookup(const struct lu_env *env,
106 struct md_object *pobj, const struct lu_name *lname,
107 struct lu_fid* fid, struct md_op_spec *spec)
111 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
115 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
118 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
122 * For root fid use special function, which does not compare version component
123 * of fid. Version component is different for root fids on all MDTs.
125 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
127 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
128 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
132 * return 1: if lf is the fid of the ancestor of p1;
135 * return -EREMOTE: if remote object is found, in this
136 * case fid of remote object is saved to @pf;
138 * otherwise: values < 0, errors.
140 static int mdd_is_parent(const struct lu_env *env,
141 struct mdd_device *mdd,
142 struct mdd_object *p1,
143 const struct lu_fid *lf,
146 struct mdd_object *parent = NULL;
151 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
152 pfid = &mdd_env_info(env)->mti_fid;
154 /* Check for root first. */
155 if (mdd_is_root(mdd, mdo2fid(p1)))
159 /* this is done recursively, bypass capa for each obj */
160 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
161 rc = mdd_parent_fid(env, p1, pfid);
164 if (mdd_is_root(mdd, pfid))
166 if (lu_fid_eq(pfid, lf))
169 mdd_object_put(env, parent);
170 parent = mdd_object_find(env, mdd, pfid);
172 /* cross-ref parent */
173 if (parent == NULL) {
176 GOTO(out, rc = -EREMOTE);
177 } else if (IS_ERR(parent))
178 GOTO(out, rc = PTR_ERR(parent));
183 if (parent && !IS_ERR(parent))
184 mdd_object_put(env, parent);
189 * No permission check is needed.
191 * returns 1: if fid is ancestor of @mo;
192 * returns 0: if fid is not a ancestor of @mo;
194 * returns EREMOTE if remote object is found, fid of remote object is saved to
197 * returns < 0: if error
199 static int mdd_is_subdir(const struct lu_env *env,
200 struct md_object *mo, const struct lu_fid *fid,
203 struct mdd_device *mdd = mdo2mdd(mo);
207 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
210 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
214 } else if (rc == 1) {
215 /* found @fid is parent */
223 * Check that @dir contains no entries except (possibly) dot and dotdot.
228 * -ENOTDIR not a directory object
229 * -ENOTEMPTY not empty
233 static int mdd_dir_is_empty(const struct lu_env *env,
234 struct mdd_object *dir)
237 struct dt_object *obj;
238 const struct dt_it_ops *iops;
242 obj = mdd_object_child(dir);
243 if (!dt_try_as_dir(env, obj))
246 iops = &obj->do_index_ops->dio_it;
247 it = iops->init(env, obj, BYPASS_CAPA);
249 result = iops->get(env, it, (const void *)"");
252 for (result = 0, i = 0; result == 0 && i < 3; ++i)
253 result = iops->next(env, it);
256 else if (result == +1)
258 } else if (result == 0)
260 * Huh? Index contains no zero key?
271 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
273 struct mdd_device *m = mdd_obj2mdd_dev(obj);
274 struct lu_attr *la = &mdd_env_info(env)->mti_la;
278 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
283 * Subdir count limitation can be broken through.
285 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
286 !S_ISDIR(la->la_mode))
293 * Check whether it may create the cobj under the pobj.
296 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
297 struct mdd_object *cobj, int check_perm, int check_nlink)
302 if (cobj && mdd_object_exists(cobj))
305 if (mdd_is_dead_obj(pobj))
309 rc = mdd_permission_internal_locked(env, pobj, NULL,
310 MAY_WRITE | MAY_EXEC,
313 if (!rc && check_nlink)
314 rc = __mdd_may_link(env, pobj);
320 * Check whether can unlink from the pobj in the case of "cobj == NULL".
322 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
323 const struct md_attr *ma)
328 if (mdd_is_dead_obj(pobj))
331 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
332 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
335 rc = mdd_permission_internal_locked(env, pobj, NULL,
336 MAY_WRITE | MAY_EXEC,
341 if (mdd_is_append(pobj))
348 * pobj == NULL is remote ops case, under such case, pobj's
349 * VTX feature has been checked already, no need check again.
351 static inline int mdd_is_sticky(const struct lu_env *env,
352 struct mdd_object *pobj,
353 struct mdd_object *cobj)
355 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
356 struct md_ucred *uc = md_ucred(env);
360 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
364 if (!(tmp_la->la_mode & S_ISVTX) ||
365 (tmp_la->la_uid == uc->mu_fsuid))
369 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
373 if (tmp_la->la_uid == uc->mu_fsuid)
376 return !mdd_capable(uc, CFS_CAP_FOWNER);
380 * Check whether it may delete the cobj from the pobj.
383 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
384 struct mdd_object *cobj, struct md_attr *ma,
385 int check_perm, int check_empty)
391 if (!mdd_object_exists(cobj))
395 if (mdd_is_dead_obj(pobj))
399 rc = mdd_permission_internal_locked(env, pobj, NULL,
400 MAY_WRITE | MAY_EXEC,
406 if (mdd_is_append(pobj))
410 if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) &&
411 mdd_is_sticky(env, pobj, cobj))
414 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
417 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
418 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
421 if (S_ISDIR(ma->ma_attr.la_mode)) {
422 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
424 if (!S_ISDIR(mdd_object_type(cobj)))
427 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
429 } else if (S_ISDIR(mdd_object_type(cobj)))
432 if (S_ISDIR(ma->ma_attr.la_mode) && check_empty)
433 rc = mdd_dir_is_empty(env, cobj);
440 * has mdd_write_lock on src already, but not on tgt yet
442 int mdd_link_sanity_check(const struct lu_env *env,
443 struct mdd_object *tgt_obj,
444 const struct lu_name *lname,
445 struct mdd_object *src_obj)
447 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
451 /* Local ops, no lookup before link, check filename length here. */
452 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
453 RETURN(-ENAMETOOLONG);
455 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
458 if (S_ISDIR(mdd_object_type(src_obj)))
461 LASSERT(src_obj != tgt_obj);
463 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
468 rc = __mdd_may_link(env, src_obj);
474 * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and
475 * assign i_nlink to 1 which means the i_nlink for subdir count is incredible
476 * (maybe too large to be represented). It is a trick to break through the
477 * "i_nlink" limitation for subdir count.
479 void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj,
480 struct thandle *handle)
482 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
483 struct mdd_device *m = mdd_obj2mdd_dev(obj);
485 if (!mdd_is_mnlink(obj)) {
486 if (S_ISDIR(mdd_object_type(obj))) {
487 if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA))
490 if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) {
491 obj->mod_flags |= MNLINK_OBJ;
492 tmp_la->la_nlink = 1;
493 tmp_la->la_valid = LA_NLINK;
494 mdd_attr_set_internal(env, obj, tmp_la, handle,
499 mdo_ref_add(env, obj, handle);
503 void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj,
504 struct thandle *handle, int is_dot)
506 if (!mdd_is_mnlink(obj) || is_dot)
507 mdo_ref_del(env, obj, handle);
510 /* insert named index, add reference if isdir */
511 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
512 const struct lu_fid *lf, const char *name, int is_dir,
513 struct thandle *handle, struct lustre_capa *capa)
515 struct dt_object *next = mdd_object_child(pobj);
519 if (dt_try_as_dir(env, next)) {
520 struct md_ucred *uc = md_ucred(env);
522 rc = next->do_index_ops->dio_insert(env, next,
523 __mdd_fid_rec(env, lf),
524 (const struct dt_key *)name,
525 handle, capa, uc->mu_cap &
526 CFS_CAP_SYS_RESOURCE_MASK);
533 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
534 __mdd_ref_add(env, pobj, handle);
535 mdd_write_unlock(env, pobj);
541 /* delete named index, drop reference if isdir */
542 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
543 const char *name, int is_dir, struct thandle *handle,
544 struct lustre_capa *capa)
546 struct dt_object *next = mdd_object_child(pobj);
550 if (dt_try_as_dir(env, next)) {
551 rc = next->do_index_ops->dio_delete(env, next,
552 (struct dt_key *)name,
554 if (rc == 0 && is_dir) {
557 if (name != NULL && name[0] == '.' && name[1] == 0)
559 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
560 __mdd_ref_del(env, pobj, handle, is_dot);
561 mdd_write_unlock(env, pobj);
570 __mdd_index_insert_only(const struct lu_env *env, struct mdd_object *pobj,
571 const struct lu_fid *lf, const char *name,
572 struct thandle *handle, struct lustre_capa *capa)
574 struct dt_object *next = mdd_object_child(pobj);
578 if (dt_try_as_dir(env, next)) {
579 struct md_ucred *uc = md_ucred(env);
581 rc = next->do_index_ops->dio_insert(env, next,
582 __mdd_fid_rec(env, lf),
583 (const struct dt_key *)name,
584 handle, capa, uc->mu_cap &
585 CFS_CAP_SYS_RESOURCE_MASK);
592 /** Store a namespace change changelog record
593 * If this fails, we must fail the whole transaction; we don't
594 * want the change to commit without the log entry.
595 * \param target - mdd_object of change
596 * \param parent - parent dir/object
597 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
598 * \param tname - target name string
599 * \param handle - transacion handle
601 static int mdd_changelog_ns_store(const struct lu_env *env,
602 struct mdd_device *mdd,
603 enum changelog_rec_type type,
604 struct mdd_object *target,
605 struct mdd_object *parent,
606 const struct lu_fid *tf,
607 const struct lu_name *tname,
608 struct thandle *handle)
610 const struct lu_fid *tfid;
611 const struct lu_fid *tpfid = mdo2fid(parent);
612 struct llog_changelog_rec *rec;
618 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
621 LASSERT(parent != NULL);
622 LASSERT(tname != NULL);
623 LASSERT(handle != NULL);
626 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
627 buf = mdd_buf_alloc(env, reclen);
628 if (buf->lb_buf == NULL)
630 rec = (struct llog_changelog_rec *)buf->lb_buf;
632 rec->cr_flags = CLF_VERSION;
633 rec->cr_type = (__u32)type;
634 tfid = tf ? tf : mdo2fid(target);
635 rec->cr_tfid = *tfid;
636 rec->cr_pfid = *tpfid;
637 rec->cr_namelen = tname->ln_namelen;
638 memcpy(rec->cr_name, tname->ln_name, rec->cr_namelen);
640 target->mod_cltime = cfs_time_current_64();
642 rc = mdd_changelog_llog_write(mdd, rec, handle);
644 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
645 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
652 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
653 struct md_object *src_obj, const struct lu_name *lname,
656 const char *name = lname->ln_name;
657 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
658 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
659 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
660 struct mdd_device *mdd = mdo2mdd(src_obj);
661 struct dynlock_handle *dlh;
662 struct thandle *handle;
663 #ifdef HAVE_QUOTA_SUPPORT
664 struct obd_device *obd = mdd->mdd_obd_dev;
665 struct mds_obd *mds = &obd->u.mds;
666 unsigned int qids[MAXQUOTAS] = { 0, 0 };
667 int quota_opc = 0, rec_pending = 0;
672 #ifdef HAVE_QUOTA_SUPPORT
673 if (mds->mds_quota) {
674 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
676 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
679 mdd_data_get(env, mdd_tobj, &data);
680 quota_opc = FSFILT_OP_LINK;
681 mdd_quota_wrapper(la_tmp, qids);
682 /* get block quota for parent */
683 lquota_chkquota(mds_quota_interface_ref, obd,
684 qids[USRQUOTA], qids[GRPQUOTA], 1,
685 &rec_pending, NULL, LQUOTA_FLAGS_BLK,
691 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
692 handle = mdd_trans_start(env, mdd);
694 GOTO(out_pending, rc = PTR_ERR(handle));
696 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
698 GOTO(out_trans, rc = -ENOMEM);
699 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
701 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
703 GOTO(out_unlock, rc);
705 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
707 mdd_object_capa(env, mdd_tobj));
709 GOTO(out_unlock, rc);
711 __mdd_ref_add(env, mdd_sobj, handle);
713 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
714 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
716 la->la_valid = LA_CTIME | LA_MTIME;
717 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
719 GOTO(out_unlock, rc);
721 la->la_valid = LA_CTIME;
722 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
724 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
728 mdd_write_unlock(env, mdd_sobj);
729 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
732 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
733 mdd_tobj, NULL, lname, handle);
734 mdd_trans_stop(env, mdd, rc, handle);
736 #ifdef HAVE_QUOTA_SUPPORT
739 lquota_pending_commit(mds_quota_interface_ref, obd,
740 qids[USRQUOTA], qids[GRPQUOTA],
742 /* Trigger dqacq for the parent owner. If failed,
743 * the next call for lquota_chkquota will process it. */
744 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
751 /* caller should take a lock before calling */
752 int mdd_finish_unlink(const struct lu_env *env,
753 struct mdd_object *obj, struct md_attr *ma,
760 rc = mdd_iattr_get(env, obj, ma);
761 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
762 /* add new orphan and the object
763 * will be deleted during mdd_close() */
764 if (obj->mod_count) {
765 rc = __mdd_orphan_add(env, obj, th);
767 obj->mod_flags |= ORPHAN_OBJ;
768 CDEBUG(D_HA, "Object "DFID" is going to be "
769 "an orphan, open count = %d\n",
770 PFID(mdd_object_fid(obj)),
775 obj->mod_flags |= DEAD_OBJ;
776 if (!(obj->mod_flags & ORPHAN_OBJ)) {
777 rc = mdd_object_kill(env, obj, ma);
784 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
791 * has mdd_write_lock on cobj already, but not on pobj yet
793 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
794 struct mdd_object *cobj, struct md_attr *ma)
799 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
804 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
805 struct md_object *cobj, const struct lu_name *lname,
808 const char *name = lname->ln_name;
809 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
810 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
811 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
812 struct mdd_device *mdd = mdo2mdd(pobj);
813 struct dynlock_handle *dlh;
814 struct thandle *handle;
815 #ifdef HAVE_QUOTA_SUPPORT
816 struct obd_device *obd = mdd->mdd_obd_dev;
817 struct mds_obd *mds = &obd->u.mds;
818 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
819 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
822 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
826 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
827 PFID(mdd_object_fid(mdd_cobj)));
829 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
833 handle = mdd_trans_start(env, mdd);
835 RETURN(PTR_ERR(handle));
837 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
839 GOTO(out_trans, rc = -ENOMEM);
840 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
842 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
846 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
847 mdd_object_capa(env, mdd_pobj));
851 __mdd_ref_del(env, mdd_cobj, handle, 0);
854 __mdd_ref_del(env, mdd_cobj, handle, 1);
856 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
857 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
859 la->la_valid = LA_CTIME | LA_MTIME;
860 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
864 la->la_valid = LA_CTIME;
865 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
869 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
870 #ifdef HAVE_QUOTA_SUPPORT
871 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
872 ma->ma_attr.la_nlink == 0) {
873 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
875 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
877 mdd_quota_wrapper(la_tmp, qpids);
878 if (mdd_cobj->mod_count == 0) {
879 quota_opc = FSFILT_OP_UNLINK;
880 mdd_quota_wrapper(&ma->ma_attr, qcids);
882 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
889 obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
890 sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
893 /* old files may not have link ea; ignore errors */
894 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
895 lname, NULL, NULL, handle);
899 mdd_write_unlock(env, mdd_cobj);
900 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
903 rc = mdd_changelog_ns_store(env, mdd,
904 is_dir ? CL_RMDIR : CL_UNLINK,
905 mdd_cobj, mdd_pobj, NULL, lname,
908 mdd_trans_stop(env, mdd, rc, handle);
909 #ifdef HAVE_QUOTA_SUPPORT
911 /* Trigger dqrel on the owner of child and parent. If failed,
912 * the next call for lquota_chkquota will process it. */
913 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
919 /* has not lock on pobj yet */
920 static int mdd_ni_sanity_check(const struct lu_env *env,
921 struct md_object *pobj,
922 const struct md_attr *ma)
924 struct mdd_object *obj = md2mdd_obj(pobj);
928 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
931 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
939 static int mdd_name_insert(const struct lu_env *env,
940 struct md_object *pobj,
941 const struct lu_name *lname,
942 const struct lu_fid *fid,
943 const struct md_attr *ma)
945 const char *name = lname->ln_name;
946 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
947 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
948 struct mdd_device *mdd = mdo2mdd(pobj);
949 struct dynlock_handle *dlh;
950 struct thandle *handle;
951 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
952 #ifdef HAVE_QUOTA_SUPPORT
953 struct md_ucred *uc = md_ucred(env);
954 struct obd_device *obd = mdd->mdd_obd_dev;
955 struct mds_obd *mds = &obd->u.mds;
956 unsigned int qids[MAXQUOTAS] = { 0, 0 };
957 int quota_opc = 0, rec_pending = 0;
958 cfs_cap_t save = uc->mu_cap;
963 #ifdef HAVE_QUOTA_SUPPORT
964 if (mds->mds_quota) {
965 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
966 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
968 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
971 mdd_data_get(env, mdd_obj, &data);
972 quota_opc = FSFILT_OP_LINK;
973 mdd_quota_wrapper(la_tmp, qids);
974 /* get block quota for parent */
975 lquota_chkquota(mds_quota_interface_ref, obd,
976 qids[USRQUOTA], qids[GRPQUOTA],
977 1, &rec_pending, NULL,
978 LQUOTA_FLAGS_BLK, data, 1);
981 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
985 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
986 handle = mdd_trans_start(env, mdo2mdd(pobj));
988 GOTO(out_pending, rc = PTR_ERR(handle));
990 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
992 GOTO(out_trans, rc = -ENOMEM);
994 rc = mdd_ni_sanity_check(env, pobj, ma);
996 GOTO(out_unlock, rc);
998 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
999 handle, BYPASS_CAPA);
1001 GOTO(out_unlock, rc);
1004 * For some case, no need update obj's ctime (LA_CTIME is not set),
1006 * For other cases, update obj's ctime (LA_CTIME is set),
1009 if (ma->ma_attr.la_valid & LA_CTIME) {
1010 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1011 la->la_valid = LA_CTIME | LA_MTIME;
1012 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1017 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1019 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1021 #ifdef HAVE_QUOTA_SUPPORT
1022 if (mds->mds_quota) {
1025 lquota_pending_commit(mds_quota_interface_ref,
1026 obd, qids[USRQUOTA],
1029 /* Trigger dqacq for the parent owner. If failed,
1030 * the next call for lquota_chkquota will process it*/
1031 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1041 /* has not lock on pobj yet */
1042 static int mdd_nr_sanity_check(const struct lu_env *env,
1043 struct md_object *pobj,
1044 const struct md_attr *ma)
1046 struct mdd_object *obj = md2mdd_obj(pobj);
1050 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1053 rc = mdd_may_unlink(env, obj, ma);
1059 * Partial operation.
1061 static int mdd_name_remove(const struct lu_env *env,
1062 struct md_object *pobj,
1063 const struct lu_name *lname,
1064 const struct md_attr *ma)
1066 const char *name = lname->ln_name;
1067 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1068 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1069 struct mdd_device *mdd = mdo2mdd(pobj);
1070 struct dynlock_handle *dlh;
1071 struct thandle *handle;
1072 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1073 #ifdef HAVE_QUOTA_SUPPORT
1074 struct obd_device *obd = mdd->mdd_obd_dev;
1075 struct mds_obd *mds = &obd->u.mds;
1076 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1082 #ifdef HAVE_QUOTA_SUPPORT
1083 if (mds->mds_quota) {
1084 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1086 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1088 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1089 mdd_quota_wrapper(la_tmp, qids);
1093 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1094 handle = mdd_trans_start(env, mdd);
1096 GOTO(out_pending, rc = PTR_ERR(handle));
1098 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1100 GOTO(out_trans, rc = -ENOMEM);
1102 rc = mdd_nr_sanity_check(env, pobj, ma);
1104 GOTO(out_unlock, rc);
1106 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1107 handle, BYPASS_CAPA);
1109 GOTO(out_unlock, rc);
1112 * For some case, no need update obj's ctime (LA_CTIME is not set),
1114 * For other cases, update obj's ctime (LA_CTIME is set),
1117 if (ma->ma_attr.la_valid & LA_CTIME) {
1118 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1119 la->la_valid = LA_CTIME | LA_MTIME;
1120 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1125 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1127 mdd_trans_stop(env, mdd, rc, handle);
1129 #ifdef HAVE_QUOTA_SUPPORT
1130 /* Trigger dqrel for the parent owner.
1131 * If failed, the next call for lquota_chkquota will process it. */
1133 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1141 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1143 static int mdd_rt_sanity_check(const struct lu_env *env,
1144 struct mdd_object *tgt_pobj,
1145 struct mdd_object *tobj,
1151 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1154 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1155 * exist. In fact, tobj must exist, otherwise the call trace will be:
1156 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1157 * When get here, tobj must be NOT NULL, the other case has been
1158 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1160 * So check may_delete, but not check nlink of tgt_pobj. */
1162 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1167 /* Partial rename op on slave MDD */
1168 static int mdd_rename_tgt(const struct lu_env *env,
1169 struct md_object *pobj, struct md_object *tobj,
1170 const struct lu_fid *lf, const struct lu_name *lname,
1173 const char *name = lname->ln_name;
1174 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1175 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1176 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1177 struct mdd_device *mdd = mdo2mdd(pobj);
1178 struct dynlock_handle *dlh;
1179 struct thandle *handle;
1180 #ifdef HAVE_QUOTA_SUPPORT
1181 struct obd_device *obd = mdd->mdd_obd_dev;
1182 struct mds_obd *mds = &obd->u.mds;
1183 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1184 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1185 int quota_opc = 0, rec_pending = 0;
1190 #ifdef HAVE_QUOTA_SUPPORT
1191 if (mds->mds_quota && !tobj) {
1192 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1194 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1197 mdd_data_get(env, mdd_tpobj, &data);
1198 quota_opc = FSFILT_OP_LINK;
1199 mdd_quota_wrapper(la_tmp, qpids);
1200 /* get block quota for target parent */
1201 lquota_chkquota(mds_quota_interface_ref, obd,
1202 qpids[USRQUOTA], qpids[GRPQUOTA], 1,
1203 &rec_pending, NULL, LQUOTA_FLAGS_BLK,
1208 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1209 handle = mdd_trans_start(env, mdd);
1211 GOTO(out_pending, rc = PTR_ERR(handle));
1213 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1215 GOTO(out_trans, rc = -ENOMEM);
1217 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1219 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1224 * If rename_tgt is called then we should just re-insert name with
1225 * correct fid, no need to dec/inc parent nlink if obj is dir.
1227 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1231 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1236 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1237 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1239 la->la_valid = LA_CTIME | LA_MTIME;
1240 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1245 * For tobj is remote case cmm layer has processed
1246 * and pass NULL tobj to here. So when tobj is NOT NULL,
1247 * it must be local one.
1249 if (tobj && mdd_object_exists(mdd_tobj)) {
1250 __mdd_ref_del(env, mdd_tobj, handle, 0);
1252 /* Remove dot reference. */
1253 if (S_ISDIR(ma->ma_attr.la_mode))
1254 __mdd_ref_del(env, mdd_tobj, handle, 1);
1256 la->la_valid = LA_CTIME;
1257 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1261 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1265 #ifdef HAVE_QUOTA_SUPPORT
1266 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1267 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1268 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1269 mdd_quota_wrapper(&ma->ma_attr, qcids);
1276 mdd_write_unlock(env, mdd_tobj);
1277 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1280 /* Bare EXT record with no RENAME in front of it signifies
1281 a partial slave op */
1282 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1283 mdd_tpobj, NULL, lname, handle);
1285 mdd_trans_stop(env, mdd, rc, handle);
1287 #ifdef HAVE_QUOTA_SUPPORT
1288 if (mds->mds_quota) {
1290 lquota_pending_commit(mds_quota_interface_ref, obd,
1295 /* Trigger dqrel/dqacq on the target owner of child and
1296 * parent. If failed, the next call for lquota_chkquota
1297 * will process it. */
1298 lquota_adjust(mds_quota_interface_ref, obd, qcids,
1299 qpids, rc, quota_opc);
1306 * The permission has been checked when obj created, no need check again.
1308 static int mdd_cd_sanity_check(const struct lu_env *env,
1309 struct mdd_object *obj)
1314 if (!obj || mdd_is_dead_obj(obj))
1321 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1322 struct md_object *cobj, const struct md_op_spec *spec,
1325 struct mdd_device *mdd = mdo2mdd(cobj);
1326 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1327 struct mdd_object *son = md2mdd_obj(cobj);
1328 struct lu_attr *attr = &ma->ma_attr;
1329 struct lov_mds_md *lmm = NULL;
1331 struct thandle *handle;
1335 rc = mdd_cd_sanity_check(env, son);
1339 if (!md_should_create(spec->sp_cr_flags))
1342 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1347 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1348 handle = mdd_trans_start(env, mdd);
1350 GOTO(out_free, rc = PTR_ERR(handle));
1353 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1354 * Should this be fixed?
1357 /* Replay creates has objects already */
1359 if (spec->no_create) {
1360 CDEBUG(D_INFO, "we already have lov ea\n");
1361 rc = mdd_lov_set_md(env, mdd_pobj, son,
1362 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1363 spec->u.sp_ea.eadatalen, handle, 0);
1366 /* No need mdd_lsm_sanity_check here */
1367 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1368 lmm_size, handle, 0);
1371 rc = mdd_attr_get_internal_locked(env, son, ma);
1373 /* update lov_objid data, must be before transaction stop! */
1375 mdd_lov_objid_update(mdd, lmm);
1377 mdd_trans_stop(env, mdd, rc, handle);
1379 /* Finish mdd_lov_create() stuff. */
1380 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1384 /* Get fid from name and parent */
1386 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1387 const struct lu_name *lname, struct lu_fid* fid, int mask)
1389 const char *name = lname->ln_name;
1390 const struct dt_key *key = (const struct dt_key *)name;
1391 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1392 struct mdd_device *m = mdo2mdd(pobj);
1393 struct dt_object *dir = mdd_object_child(mdd_obj);
1394 struct lu_fid_pack *pack = &mdd_env_info(env)->mti_pack;
1398 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1401 rc = mdd_object_exists(mdd_obj);
1402 if (unlikely(rc == 0))
1404 else if (unlikely(rc < 0)) {
1405 CERROR("Object "DFID" locates on remote server\n",
1406 PFID(mdo2fid(mdd_obj)));
1410 /* The common filename length check. */
1411 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1412 RETURN(-ENAMETOOLONG);
1414 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1419 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1420 dt_try_as_dir(env, dir))) {
1422 rc = dir->do_index_ops->dio_lookup(env, dir,
1423 (struct dt_rec *)pack, key,
1424 mdd_object_capa(env, mdd_obj));
1426 rc = fid_unpack(pack, fid);
1435 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1436 const struct lu_name *lname, struct mdd_object *child,
1437 struct md_attr *ma, struct thandle *handle,
1438 const struct md_op_spec *spec)
1444 * Update attributes for child.
1447 * (1) the valid bits should be converted between Lustre and Linux;
1448 * (2) maybe, the child attributes should be set in OSD when creation.
1451 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1455 if (S_ISDIR(ma->ma_attr.la_mode)) {
1456 /* Add "." and ".." for newly created dir */
1457 __mdd_ref_add(env, child, handle);
1458 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1459 dot, handle, BYPASS_CAPA);
1461 rc = __mdd_index_insert_only(env, child, pfid,
1467 rc2 = __mdd_index_delete(env, child, dot, 1,
1468 handle, BYPASS_CAPA);
1470 CERROR("Failure to cleanup after dotdot"
1471 " creation: %d (%d)\n", rc2, rc);
1476 mdd_links_add(env, child, pfid, lname, handle);
1481 /* has not lock on pobj yet */
1482 static int mdd_create_sanity_check(const struct lu_env *env,
1483 struct md_object *pobj,
1484 const struct lu_name *lname,
1486 struct md_op_spec *spec)
1488 struct mdd_thread_info *info = mdd_env_info(env);
1489 struct lu_attr *la = &info->mti_la;
1490 struct lu_fid *fid = &info->mti_fid;
1491 struct mdd_object *obj = md2mdd_obj(pobj);
1492 struct mdd_device *m = mdo2mdd(pobj);
1493 int lookup = spec->sp_cr_lookup;
1498 if (mdd_is_dead_obj(obj))
1502 * In some cases this lookup is not needed - we know before if name
1503 * exists or not because MDT performs lookup for it.
1504 * name length check is done in lookup.
1508 * Check if the name already exist, though it will be checked in
1509 * _index_insert also, for avoiding rolling back if exists
1512 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1513 MAY_WRITE | MAY_EXEC);
1515 RETURN(rc ? : -EEXIST);
1518 * Check WRITE permission for the parent.
1519 * EXEC permission have been checked
1520 * when lookup before create already.
1522 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1529 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1533 if (la->la_mode & S_ISGID) {
1534 ma->ma_attr.la_gid = la->la_gid;
1535 if (S_ISDIR(ma->ma_attr.la_mode)) {
1536 ma->ma_attr.la_mode |= S_ISGID;
1537 ma->ma_attr.la_valid |= LA_MODE;
1541 switch (ma->ma_attr.la_mode & S_IFMT) {
1543 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1545 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1546 RETURN(-ENAMETOOLONG);
1566 * Create object and insert it into namespace.
1568 static int mdd_create(const struct lu_env *env,
1569 struct md_object *pobj,
1570 const struct lu_name *lname,
1571 struct md_object *child,
1572 struct md_op_spec *spec,
1575 struct mdd_thread_info *info = mdd_env_info(env);
1576 struct lu_attr *la = &info->mti_la_for_fix;
1577 struct md_attr *ma_acl = &info->mti_ma;
1578 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1579 struct mdd_object *son = md2mdd_obj(child);
1580 struct mdd_device *mdd = mdo2mdd(pobj);
1581 struct lu_attr *attr = &ma->ma_attr;
1582 struct lov_mds_md *lmm = NULL;
1583 struct thandle *handle;
1584 struct dynlock_handle *dlh;
1585 const char *name = lname->ln_name;
1586 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1587 int got_def_acl = 0;
1588 #ifdef HAVE_QUOTA_SUPPORT
1589 struct obd_device *obd = mdd->mdd_obd_dev;
1590 struct mds_obd *mds = &obd->u.mds;
1591 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1592 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1593 int quota_opc = 0, block_count = 0;
1594 int inode_pending = 0, block_pending = 0, parent_pending = 0;
1599 * Two operations have to be performed:
1601 * - an allocation of a new object (->do_create()), and
1603 * - an insertion into a parent index (->dio_insert()).
1605 * Due to locking, operation order is not important, when both are
1606 * successful, *but* error handling cases are quite different:
1608 * - if insertion is done first, and following object creation fails,
1609 * insertion has to be rolled back, but this operation might fail
1610 * also leaving us with dangling index entry.
1612 * - if creation is done first, is has to be undone if insertion
1613 * fails, leaving us with leaked space, which is neither good, nor
1616 * It seems that creation-first is simplest solution, but it is
1617 * sub-optimal in the frequent
1622 * case, because second mkdir is bound to create object, only to
1623 * destroy it immediately.
1625 * To avoid this follow local file systems that do double lookup:
1627 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1629 * 1. create (mdd_object_create_internal())
1631 * 2. insert (__mdd_index_insert(), lookup again)
1634 /* Sanity checks before big job. */
1635 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1639 #ifdef HAVE_QUOTA_SUPPORT
1640 if (mds->mds_quota) {
1641 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1643 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1647 quota_opc = FSFILT_OP_CREATE;
1648 mdd_quota_wrapper(&ma->ma_attr, qcids);
1649 mdd_quota_wrapper(la_tmp, qpids);
1650 /* get file quota for child */
1651 lquota_chkquota(mds_quota_interface_ref, obd,
1652 qcids[USRQUOTA], qcids[GRPQUOTA], 1,
1653 &inode_pending, NULL, 0, NULL, 0);
1654 switch (ma->ma_attr.la_mode & S_IFMT) {
1663 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1664 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1668 /* get block quota for child and parent */
1670 lquota_chkquota(mds_quota_interface_ref, obd,
1671 qcids[USRQUOTA], qcids[GRPQUOTA],
1673 &block_pending, NULL,
1674 LQUOTA_FLAGS_BLK, NULL, 0);
1676 lquota_chkquota(mds_quota_interface_ref, obd,
1677 qpids[USRQUOTA], qpids[GRPQUOTA], 1,
1678 &parent_pending, NULL,
1679 LQUOTA_FLAGS_BLK, NULL, 0);
1685 * No RPC inside the transaction, so OST objects should be created at
1688 if (S_ISREG(attr->la_mode)) {
1689 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1692 GOTO(out_pending, rc);
1695 if (!S_ISLNK(attr->la_mode)) {
1696 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1697 ma_acl->ma_acl = info->mti_xattr_buf;
1698 ma_acl->ma_need = MA_ACL_DEF;
1699 ma_acl->ma_valid = 0;
1701 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1702 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1703 mdd_read_unlock(env, mdd_pobj);
1706 else if (ma_acl->ma_valid & MA_ACL_DEF)
1710 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1711 handle = mdd_trans_start(env, mdd);
1713 GOTO(out_free, rc = PTR_ERR(handle));
1715 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1717 GOTO(out_trans, rc = -ENOMEM);
1719 mdd_write_lock(env, son, MOR_TGT_CHILD);
1720 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1722 mdd_write_unlock(env, son);
1728 #ifdef CONFIG_FS_POSIX_ACL
1730 struct lu_buf *acl_buf = &info->mti_buf;
1731 acl_buf->lb_buf = ma_acl->ma_acl;
1732 acl_buf->lb_len = ma_acl->ma_acl_size;
1734 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1736 mdd_write_unlock(env, son);
1739 ma->ma_attr.la_valid |= LA_MODE;
1744 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1745 son, ma, handle, spec);
1746 mdd_write_unlock(env, son);
1749 * Object has no links, so it will be destroyed when last
1750 * reference is released. (XXX not now.)
1756 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1757 name, S_ISDIR(attr->la_mode), handle,
1758 mdd_object_capa(env, mdd_pobj));
1765 /* No need mdd_lsm_sanity_check here */
1766 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1768 CERROR("error on stripe info copy %d \n", rc);
1771 if (lmm && lmm_size > 0) {
1772 /* Set Lov here, do not get lmm again later */
1773 memcpy(ma->ma_lmm, lmm, lmm_size);
1774 ma->ma_lmm_size = lmm_size;
1775 ma->ma_valid |= MA_LOV;
1778 if (S_ISLNK(attr->la_mode)) {
1779 struct md_ucred *uc = md_ucred(env);
1780 struct dt_object *dt = mdd_object_child(son);
1781 const char *target_name = spec->u.sp_symname;
1782 int sym_len = strlen(target_name);
1783 const struct lu_buf *buf;
1786 buf = mdd_buf_get_const(env, target_name, sym_len);
1787 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1788 mdd_object_capa(env, son),
1790 CFS_CAP_SYS_RESOURCE_MASK);
1795 GOTO(cleanup, rc = -EFAULT);
1799 la->la_valid = LA_CTIME | LA_MTIME;
1800 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1804 /* Return attr back. */
1805 rc = mdd_attr_get_internal_locked(env, son, ma);
1808 if (rc && created) {
1812 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1813 S_ISDIR(attr->la_mode),
1814 handle, BYPASS_CAPA);
1816 CERROR("error can not cleanup destroy %d\n",
1821 mdd_write_lock(env, son, MOR_TGT_CHILD);
1822 __mdd_ref_del(env, son, handle, 0);
1823 if (initialized && S_ISDIR(attr->la_mode))
1824 __mdd_ref_del(env, son, handle, 1);
1825 mdd_write_unlock(env, son);
1829 /* update lov_objid data, must be before transaction stop! */
1831 mdd_lov_objid_update(mdd, lmm);
1833 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1836 rc = mdd_changelog_ns_store(env, mdd,
1837 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1838 S_ISREG(attr->la_mode) ? CL_CREATE :
1839 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1840 son, mdd_pobj, NULL, lname, handle);
1841 mdd_trans_stop(env, mdd, rc, handle);
1843 /* finis lov_create stuff, free all temporary data */
1844 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1846 #ifdef HAVE_QUOTA_SUPPORT
1849 lquota_pending_commit(mds_quota_interface_ref, obd,
1850 qcids[USRQUOTA], qcids[GRPQUOTA],
1853 lquota_pending_commit(mds_quota_interface_ref, obd,
1854 qcids[USRQUOTA], qcids[GRPQUOTA],
1857 lquota_pending_commit(mds_quota_interface_ref, obd,
1858 qpids[USRQUOTA], qpids[GRPQUOTA],
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. */
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);
1972 #ifdef HAVE_QUOTA_SUPPORT
1973 struct obd_device *obd = mdd->mdd_obd_dev;
1974 struct mds_obd *mds = &obd->u.mds;
1975 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1976 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1977 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1978 int quota_opc = 0, rec_pending = 0;
1982 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1983 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1986 mdd_tobj = md2mdd_obj(tobj);
1988 #ifdef HAVE_QUOTA_SUPPORT
1989 if (mds->mds_quota) {
1990 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1992 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
1994 mdd_quota_wrapper(la_tmp, qspids);
1996 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2000 mdd_data_get(env, mdd_tpobj, &data);
2001 quota_opc = FSFILT_OP_LINK;
2002 mdd_quota_wrapper(la_tmp, qtpids);
2003 /* get block quota for target parent */
2004 lquota_chkquota(mds_quota_interface_ref,
2005 obd, qtpids[USRQUOTA],
2006 qtpids[GRPQUOTA], 1,
2015 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2016 handle = mdd_trans_start(env, mdd);
2018 GOTO(out_pending, rc = PTR_ERR(handle));
2020 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2021 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2023 GOTO(cleanup_unlocked, rc);
2025 /* Get locks in determined order */
2026 if (rc == MDD_RN_SAME) {
2027 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2028 sname, MOR_SRC_PARENT);
2029 /* check hashes to determine do we need one lock or two */
2030 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2031 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2035 } else if (rc == MDD_RN_SRCTGT) {
2036 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2037 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2039 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2040 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2042 if (sdlh == NULL || tdlh == NULL)
2043 GOTO(cleanup, rc = -ENOMEM);
2045 mdd_sobj = mdd_object_find(env, mdd, lf);
2046 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2047 mdd_sobj, mdd_tobj, ma);
2051 /* Remove source name from source directory */
2052 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2053 mdd_object_capa(env, mdd_spobj));
2057 /* "mv dir1 dir2" needs "dir1/.." link update */
2058 if (is_dir && mdd_sobj) {
2059 rc = __mdd_index_delete(env, mdd_sobj, dotdot, is_dir, handle,
2060 mdd_object_capa(env, mdd_spobj));
2064 rc = __mdd_index_insert(env, mdd_sobj, tpobj_fid, dotdot,
2066 mdd_object_capa(env, mdd_tpobj));
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));
2077 if (rc != 0 && rc != -ENOENT)
2080 /* Insert new fid with target name into target dir */
2081 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2082 mdd_object_capa(env, mdd_tpobj));
2086 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2087 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2089 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2091 la->la_valid = LA_CTIME;
2092 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2098 /* Remove old target object
2099 * For tobj is remote case cmm layer has processed
2100 * and set tobj to NULL then. So when tobj is NOT NULL,
2101 * it must be local one.
2103 if (tobj && mdd_object_exists(mdd_tobj)) {
2104 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2105 __mdd_ref_del(env, mdd_tobj, handle, 0);
2107 /* Remove dot reference. */
2109 __mdd_ref_del(env, mdd_tobj, handle, 1);
2111 la->la_valid = LA_CTIME;
2112 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2116 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2117 mdd_write_unlock(env, mdd_tobj);
2121 #ifdef HAVE_QUOTA_SUPPORT
2122 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2123 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2124 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2125 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2130 la->la_valid = LA_CTIME | LA_MTIME;
2131 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2135 if (mdd_spobj != mdd_tpobj) {
2136 la->la_valid = LA_CTIME | LA_MTIME;
2137 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2141 if (rc == 0 && mdd_sobj) {
2142 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2143 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2144 mdo2fid(mdd_tpobj), ltname, handle);
2146 /* Old files might not have EA entry */
2147 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2149 mdd_write_unlock(env, mdd_sobj);
2150 /* We don't fail the transaction if the link ea can't be
2151 updated -- fid2path will use alternate lookup method. */
2157 if (likely(tdlh) && sdlh != tdlh)
2158 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2160 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2163 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2164 mdd_spobj, lf, lsname, handle);
2166 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2167 mdd_tpobj, lf, ltname, handle);
2169 mdd_trans_stop(env, mdd, rc, handle);
2171 mdd_object_put(env, mdd_sobj);
2173 #ifdef HAVE_QUOTA_SUPPORT
2174 if (mds->mds_quota) {
2176 lquota_pending_commit(mds_quota_interface_ref, obd,
2180 /* Trigger dqrel on the source owner of parent.
2181 * If failed, the next call for lquota_chkquota will
2183 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2184 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2186 /* Trigger dqrel/dqacq on the target owner of child and
2187 * parent. If failed, the next call for lquota_chkquota
2188 * will process it. */
2189 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2190 qtpids, rc, quota_opc);
2196 /** enable/disable storing of hardlink info */
2197 int mdd_linkea_enable = 1;
2198 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2199 "record hardlink info in EAs");
2201 /** Read the link EA into a temp buffer.
2202 * Uses the name_buf since it is generally large.
2203 * \retval IS_ERR err
2204 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2206 struct lu_buf *mdd_links_get(const struct lu_env *env,
2207 struct mdd_object *mdd_obj)
2210 struct lustre_capa *capa;
2211 struct link_ea_header *leh;
2214 /* First try a small buf */
2215 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2216 if (buf->lb_buf == NULL)
2217 return ERR_PTR(-ENOMEM);
2219 capa = mdd_object_capa(env, mdd_obj);
2220 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2221 if (rc == -ERANGE) {
2222 /* Buf was too small, figure out what we need. */
2225 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2228 buf = mdd_buf_alloc(env, rc);
2229 if (buf->lb_buf == NULL)
2230 return ERR_PTR(-ENOMEM);
2231 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2237 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2238 leh->leh_magic = LINK_EA_MAGIC;
2239 leh->leh_reccount = __swab32(leh->leh_reccount);
2240 leh->leh_len = __swab64(leh->leh_len);
2241 /* entries are swabbed by mdd_lee_unpack */
2243 if (leh->leh_magic != LINK_EA_MAGIC)
2244 return ERR_PTR(-EINVAL);
2245 if (leh->leh_reccount == 0)
2246 return ERR_PTR(-ENODATA);
2251 /** Pack a link_ea_entry.
2252 * All elements are stored as chars to avoid alignment issues.
2253 * Numbers are always big-endian
2254 * \param packbuf is a temp fid buffer
2255 * \retval record length
2257 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2258 const struct lu_fid *pfid, struct lu_fid* packbuf)
2263 fid_pack(&lee->lee_parent_fid, pfid, packbuf);
2264 ptr = (char *)&lee->lee_parent_fid + lee->lee_parent_fid.fp_len;
2265 strncpy(ptr, lname->ln_name, lname->ln_namelen);
2266 reclen = lee->lee_parent_fid.fp_len + lname->ln_namelen +
2267 sizeof(lee->lee_reclen);
2268 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2269 lee->lee_reclen[1] = reclen & 0xff;
2273 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2274 struct lu_name *lname, struct lu_fid *pfid)
2276 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2277 fid_unpack(&lee->lee_parent_fid, pfid);
2278 lname->ln_name = (char *)&lee->lee_parent_fid +
2279 lee->lee_parent_fid.fp_len;
2280 lname->ln_namelen = *reclen - lee->lee_parent_fid.fp_len -
2281 sizeof(lee->lee_reclen);
2284 /** Add a record to the end of link ea buf */
2285 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2286 const struct lu_fid *pfid,
2287 const struct lu_name *lname)
2289 struct link_ea_header *leh;
2290 struct link_ea_entry *lee;
2293 if (lname == NULL || pfid == NULL)
2296 /* Make sure our buf is big enough for the new one */
2298 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2299 if (leh->leh_len + reclen > buf->lb_len) {
2300 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2305 lee = buf->lb_buf + leh->leh_len;
2306 reclen = mdd_lee_pack(lee, lname, pfid, &mdd_env_info(env)->mti_fid2);
2307 leh->leh_len += reclen;
2308 leh->leh_reccount++;
2312 /* For pathologic linkers, we don't want to spend lots of time scanning the
2313 * link ea. Limit ourseleves to something reasonable; links not in the EA
2314 * can be looked up via (slower) parent lookup.
2316 #define LINKEA_MAX_COUNT 128
2318 static int mdd_links_add(const struct lu_env *env,
2319 struct mdd_object *mdd_obj,
2320 const struct lu_fid *pfid,
2321 const struct lu_name *lname,
2322 struct thandle *handle)
2325 struct link_ea_header *leh;
2329 if (!mdd_linkea_enable)
2332 buf = mdd_links_get(env, mdd_obj);
2335 if (rc != -ENODATA) {
2336 CERROR("link_ea read failed %d "DFID"\n", rc,
2337 PFID(mdd_object_fid(mdd_obj)));
2340 /* empty EA; start one */
2341 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2342 if (buf->lb_buf == NULL)
2345 leh->leh_magic = LINK_EA_MAGIC;
2346 leh->leh_len = sizeof(struct link_ea_header);
2347 leh->leh_reccount = 0;
2351 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2354 rc = __mdd_links_add(env, buf, pfid, lname);
2359 rc = __mdd_xattr_set(env, mdd_obj,
2360 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2361 XATTR_NAME_LINK, 0, handle);
2363 CERROR("link_ea add failed %d "DFID"\n", rc,
2364 PFID(mdd_object_fid(mdd_obj)));
2366 if (buf->lb_vmalloc)
2367 /* if we vmalloced a large buffer drop it */
2373 static int mdd_links_rename(const struct lu_env *env,
2374 struct mdd_object *mdd_obj,
2375 const struct lu_fid *oldpfid,
2376 const struct lu_name *oldlname,
2377 const struct lu_fid *newpfid,
2378 const struct lu_name *newlname,
2379 struct thandle *handle)
2382 struct link_ea_header *leh;
2383 struct link_ea_entry *lee;
2384 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2385 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2391 if (!mdd_linkea_enable)
2394 if (mdd_obj->mod_flags & DEAD_OBJ)
2395 /* No more links, don't bother */
2398 buf = mdd_links_get(env, mdd_obj);
2401 CERROR("link_ea read failed %d "DFID"\n",
2402 rc, PFID(mdd_object_fid(mdd_obj)));
2406 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2408 /* Find the old record */
2409 for(count = 0; count <= leh->leh_reccount; count++) {
2410 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2411 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2412 lu_fid_eq(tmpfid, oldpfid) &&
2413 (strncmp(tmpname->ln_name, oldlname->ln_name,
2414 tmpname->ln_namelen) == 0))
2416 lee = (struct link_ea_entry *)((char *)lee + reclen);
2418 if (count > leh->leh_reccount) {
2419 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2420 oldlname->ln_namelen, oldlname->ln_name);
2421 GOTO(out, rc = -ENOENT);
2424 /* Remove the old record */
2425 leh->leh_reccount--;
2426 leh->leh_len -= reclen;
2427 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2430 /* If renaming, add the new record */
2431 if (newpfid != NULL) {
2432 /* if the add fails, we still delete the out-of-date old link */
2433 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2437 rc = __mdd_xattr_set(env, mdd_obj,
2438 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2439 XATTR_NAME_LINK, 0, handle);
2445 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2446 oldlname->ln_namelen, oldlname->ln_name, rc,
2447 PFID(mdd_object_fid(mdd_obj)));
2449 if (buf->lb_vmalloc)
2450 /* if we vmalloced a large buffer drop it */
2456 const struct md_dir_operations mdd_dir_ops = {
2457 .mdo_is_subdir = mdd_is_subdir,
2458 .mdo_lookup = mdd_lookup,
2459 .mdo_create = mdd_create,
2460 .mdo_rename = mdd_rename,
2461 .mdo_link = mdd_link,
2462 .mdo_unlink = mdd_unlink,
2463 .mdo_name_insert = mdd_name_insert,
2464 .mdo_name_remove = mdd_name_remove,
2465 .mdo_rename_tgt = mdd_rename_tgt,
2466 .mdo_create_data = mdd_create_data