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);
678 quota_opc = FSFILT_OP_LINK;
679 mdd_quota_wrapper(la_tmp, qids);
680 /* get block quota for parent */
681 lquota_chkquota(mds_quota_interface_ref, obd,
682 qids[USRQUOTA], qids[GRPQUOTA], 1,
683 &rec_pending, NULL, LQUOTA_FLAGS_BLK);
688 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
689 handle = mdd_trans_start(env, mdd);
691 GOTO(out_pending, rc = PTR_ERR(handle));
693 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
695 GOTO(out_trans, rc = -ENOMEM);
696 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
698 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
700 GOTO(out_unlock, rc);
702 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
704 mdd_object_capa(env, mdd_tobj));
706 GOTO(out_unlock, rc);
708 __mdd_ref_add(env, mdd_sobj, handle);
710 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
711 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
713 la->la_valid = LA_CTIME | LA_MTIME;
714 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
716 GOTO(out_unlock, rc);
718 la->la_valid = LA_CTIME;
719 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
721 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
725 mdd_write_unlock(env, mdd_sobj);
726 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
729 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
730 mdd_tobj, NULL, lname, handle);
731 mdd_trans_stop(env, mdd, rc, handle);
733 #ifdef HAVE_QUOTA_SUPPORT
736 lquota_pending_commit(mds_quota_interface_ref, obd,
737 qids[USRQUOTA], qids[GRPQUOTA],
739 /* Trigger dqacq for the parent owner. If failed,
740 * the next call for lquota_chkquota will process it. */
741 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
748 /* caller should take a lock before calling */
749 int mdd_finish_unlink(const struct lu_env *env,
750 struct mdd_object *obj, struct md_attr *ma,
757 rc = mdd_iattr_get(env, obj, ma);
758 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
759 /* add new orphan and the object
760 * will be deleted during mdd_close() */
761 if (obj->mod_count) {
762 rc = __mdd_orphan_add(env, obj, th);
764 obj->mod_flags |= ORPHAN_OBJ;
767 obj->mod_flags |= DEAD_OBJ;
768 if (!(obj->mod_flags & ORPHAN_OBJ)) {
769 rc = mdd_object_kill(env, obj, ma);
776 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
783 * has mdd_write_lock on cobj already, but not on pobj yet
785 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
786 struct mdd_object *cobj, struct md_attr *ma)
791 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
796 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
797 struct md_object *cobj, const struct lu_name *lname,
800 const char *name = lname->ln_name;
801 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
802 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
803 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
804 struct mdd_device *mdd = mdo2mdd(pobj);
805 struct dynlock_handle *dlh;
806 struct thandle *handle;
807 #ifdef HAVE_QUOTA_SUPPORT
808 struct obd_device *obd = mdd->mdd_obd_dev;
809 struct mds_obd *mds = &obd->u.mds;
810 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
811 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
814 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
818 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
819 PFID(mdd_object_fid(mdd_cobj)));
821 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
825 handle = mdd_trans_start(env, mdd);
827 RETURN(PTR_ERR(handle));
829 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
831 GOTO(out_trans, rc = -ENOMEM);
832 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
834 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
838 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
839 mdd_object_capa(env, mdd_pobj));
843 __mdd_ref_del(env, mdd_cobj, handle, 0);
846 __mdd_ref_del(env, mdd_cobj, handle, 1);
848 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
849 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
851 la->la_valid = LA_CTIME | LA_MTIME;
852 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
856 la->la_valid = LA_CTIME;
857 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
861 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
862 #ifdef HAVE_QUOTA_SUPPORT
863 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
864 ma->ma_attr.la_nlink == 0) {
865 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
867 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
869 mdd_quota_wrapper(la_tmp, qpids);
870 if (mdd_cobj->mod_count == 0) {
871 quota_opc = FSFILT_OP_UNLINK;
872 mdd_quota_wrapper(&ma->ma_attr, qcids);
874 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
881 obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
882 sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
885 /* old files may not have link ea; ignore errors */
886 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
887 lname, NULL, NULL, handle);
891 mdd_write_unlock(env, mdd_cobj);
892 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
895 rc = mdd_changelog_ns_store(env, mdd,
896 is_dir ? CL_RMDIR : CL_UNLINK,
897 mdd_cobj, mdd_pobj, NULL, lname,
900 mdd_trans_stop(env, mdd, rc, handle);
901 #ifdef HAVE_QUOTA_SUPPORT
903 /* Trigger dqrel on the owner of child and parent. If failed,
904 * the next call for lquota_chkquota will process it. */
905 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
911 /* has not lock on pobj yet */
912 static int mdd_ni_sanity_check(const struct lu_env *env,
913 struct md_object *pobj,
914 const struct md_attr *ma)
916 struct mdd_object *obj = md2mdd_obj(pobj);
920 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
923 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
931 static int mdd_name_insert(const struct lu_env *env,
932 struct md_object *pobj,
933 const struct lu_name *lname,
934 const struct lu_fid *fid,
935 const struct md_attr *ma)
937 const char *name = lname->ln_name;
938 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
939 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
940 struct mdd_device *mdd = mdo2mdd(pobj);
941 struct dynlock_handle *dlh;
942 struct thandle *handle;
943 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
944 #ifdef HAVE_QUOTA_SUPPORT
945 struct md_ucred *uc = md_ucred(env);
946 struct obd_device *obd = mdd->mdd_obd_dev;
947 struct mds_obd *mds = &obd->u.mds;
948 unsigned int qids[MAXQUOTAS] = { 0, 0 };
949 int quota_opc = 0, rec_pending = 0;
950 cfs_cap_t save = uc->mu_cap;
955 #ifdef HAVE_QUOTA_SUPPORT
956 if (mds->mds_quota) {
957 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
958 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
960 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
962 quota_opc = FSFILT_OP_LINK;
963 mdd_quota_wrapper(la_tmp, qids);
964 /* get block quota for parent */
965 lquota_chkquota(mds_quota_interface_ref, obd,
966 qids[USRQUOTA], qids[GRPQUOTA],
967 1, &rec_pending, NULL,
971 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
975 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
976 handle = mdd_trans_start(env, mdo2mdd(pobj));
978 GOTO(out_pending, rc = PTR_ERR(handle));
980 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
982 GOTO(out_trans, rc = -ENOMEM);
984 rc = mdd_ni_sanity_check(env, pobj, ma);
986 GOTO(out_unlock, rc);
988 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
989 handle, BYPASS_CAPA);
991 GOTO(out_unlock, rc);
994 * For some case, no need update obj's ctime (LA_CTIME is not set),
996 * For other cases, update obj's ctime (LA_CTIME is set),
999 if (ma->ma_attr.la_valid & LA_CTIME) {
1000 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1001 la->la_valid = LA_CTIME | LA_MTIME;
1002 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1007 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1009 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1011 #ifdef HAVE_QUOTA_SUPPORT
1012 if (mds->mds_quota) {
1015 lquota_pending_commit(mds_quota_interface_ref,
1016 obd, qids[USRQUOTA],
1017 qids[GRPQUOTA], 1, 1);
1018 /* Trigger dqacq for the parent owner. If failed,
1019 * the next call for lquota_chkquota will process it*/
1020 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1030 /* has not lock on pobj yet */
1031 static int mdd_nr_sanity_check(const struct lu_env *env,
1032 struct md_object *pobj,
1033 const struct md_attr *ma)
1035 struct mdd_object *obj = md2mdd_obj(pobj);
1039 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1042 rc = mdd_may_unlink(env, obj, ma);
1048 * Partial operation.
1050 static int mdd_name_remove(const struct lu_env *env,
1051 struct md_object *pobj,
1052 const struct lu_name *lname,
1053 const struct md_attr *ma)
1055 const char *name = lname->ln_name;
1056 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1057 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1058 struct mdd_device *mdd = mdo2mdd(pobj);
1059 struct dynlock_handle *dlh;
1060 struct thandle *handle;
1061 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1062 #ifdef HAVE_QUOTA_SUPPORT
1063 struct obd_device *obd = mdd->mdd_obd_dev;
1064 struct mds_obd *mds = &obd->u.mds;
1065 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1071 #ifdef HAVE_QUOTA_SUPPORT
1072 if (mds->mds_quota) {
1073 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1075 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1077 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1078 mdd_quota_wrapper(la_tmp, qids);
1082 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1083 handle = mdd_trans_start(env, mdd);
1085 GOTO(out_pending, rc = PTR_ERR(handle));
1087 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1089 GOTO(out_trans, rc = -ENOMEM);
1091 rc = mdd_nr_sanity_check(env, pobj, ma);
1093 GOTO(out_unlock, rc);
1095 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1096 handle, BYPASS_CAPA);
1098 GOTO(out_unlock, rc);
1101 * For some case, no need update obj's ctime (LA_CTIME is not set),
1103 * For other cases, update obj's ctime (LA_CTIME is set),
1106 if (ma->ma_attr.la_valid & LA_CTIME) {
1107 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1108 la->la_valid = LA_CTIME | LA_MTIME;
1109 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1114 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1116 mdd_trans_stop(env, mdd, rc, handle);
1118 #ifdef HAVE_QUOTA_SUPPORT
1119 /* Trigger dqrel for the parent owner.
1120 * If failed, the next call for lquota_chkquota will process it. */
1122 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1130 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1132 static int mdd_rt_sanity_check(const struct lu_env *env,
1133 struct mdd_object *tgt_pobj,
1134 struct mdd_object *tobj,
1140 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1143 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1144 * exist. In fact, tobj must exist, otherwise the call trace will be:
1145 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1146 * When get here, tobj must be NOT NULL, the other case has been
1147 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1149 * So check may_delete, but not check nlink of tgt_pobj. */
1151 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1156 /* Partial rename op on slave MDD */
1157 static int mdd_rename_tgt(const struct lu_env *env,
1158 struct md_object *pobj, struct md_object *tobj,
1159 const struct lu_fid *lf, const struct lu_name *lname,
1162 const char *name = lname->ln_name;
1163 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1164 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1165 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1166 struct mdd_device *mdd = mdo2mdd(pobj);
1167 struct dynlock_handle *dlh;
1168 struct thandle *handle;
1169 #ifdef HAVE_QUOTA_SUPPORT
1170 struct obd_device *obd = mdd->mdd_obd_dev;
1171 struct mds_obd *mds = &obd->u.mds;
1172 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1173 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1174 int quota_opc = 0, rec_pending = 0;
1179 #ifdef HAVE_QUOTA_SUPPORT
1180 if (mds->mds_quota && !tobj) {
1181 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1183 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1185 quota_opc = FSFILT_OP_LINK;
1186 mdd_quota_wrapper(la_tmp, qpids);
1187 /* get block quota for target parent */
1188 lquota_chkquota(mds_quota_interface_ref, obd,
1189 qpids[USRQUOTA], qpids[GRPQUOTA], 1,
1190 &rec_pending, NULL, LQUOTA_FLAGS_BLK);
1194 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1195 handle = mdd_trans_start(env, mdd);
1197 GOTO(out_pending, rc = PTR_ERR(handle));
1199 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1201 GOTO(out_trans, rc = -ENOMEM);
1203 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1205 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1210 * If rename_tgt is called then we should just re-insert name with
1211 * correct fid, no need to dec/inc parent nlink if obj is dir.
1213 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1217 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1222 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1223 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1225 la->la_valid = LA_CTIME | LA_MTIME;
1226 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1231 * For tobj is remote case cmm layer has processed
1232 * and pass NULL tobj to here. So when tobj is NOT NULL,
1233 * it must be local one.
1235 if (tobj && mdd_object_exists(mdd_tobj)) {
1236 __mdd_ref_del(env, mdd_tobj, handle, 0);
1238 /* Remove dot reference. */
1239 if (S_ISDIR(ma->ma_attr.la_mode))
1240 __mdd_ref_del(env, mdd_tobj, handle, 1);
1242 la->la_valid = LA_CTIME;
1243 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1247 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1251 #ifdef HAVE_QUOTA_SUPPORT
1252 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1253 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1254 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1255 mdd_quota_wrapper(&ma->ma_attr, qcids);
1262 mdd_write_unlock(env, mdd_tobj);
1263 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1266 /* Bare EXT record with no RENAME in front of it signifies
1267 a partial slave op */
1268 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1269 mdd_tpobj, NULL, lname, handle);
1271 mdd_trans_stop(env, mdd, rc, handle);
1273 #ifdef HAVE_QUOTA_SUPPORT
1274 if (mds->mds_quota) {
1276 lquota_pending_commit(mds_quota_interface_ref, obd,
1281 /* Trigger dqrel/dqacq on the target owner of child and
1282 * parent. If failed, the next call for lquota_chkquota
1283 * will process it. */
1284 lquota_adjust(mds_quota_interface_ref, obd, qcids,
1285 qpids, rc, quota_opc);
1292 * The permission has been checked when obj created, no need check again.
1294 static int mdd_cd_sanity_check(const struct lu_env *env,
1295 struct mdd_object *obj)
1300 if (!obj || mdd_is_dead_obj(obj))
1307 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1308 struct md_object *cobj, const struct md_op_spec *spec,
1311 struct mdd_device *mdd = mdo2mdd(cobj);
1312 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1313 struct mdd_object *son = md2mdd_obj(cobj);
1314 struct lu_attr *attr = &ma->ma_attr;
1315 struct lov_mds_md *lmm = NULL;
1317 struct thandle *handle;
1321 rc = mdd_cd_sanity_check(env, son);
1325 if (!md_should_create(spec->sp_cr_flags))
1328 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1333 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1334 handle = mdd_trans_start(env, mdd);
1336 GOTO(out_free, rc = PTR_ERR(handle));
1339 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1340 * Should this be fixed?
1343 /* Replay creates has objects already */
1345 if (spec->no_create) {
1346 CDEBUG(D_INFO, "we already have lov ea\n");
1347 rc = mdd_lov_set_md(env, mdd_pobj, son,
1348 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1349 spec->u.sp_ea.eadatalen, handle, 0);
1352 /* No need mdd_lsm_sanity_check here */
1353 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1354 lmm_size, handle, 0);
1357 rc = mdd_attr_get_internal_locked(env, son, ma);
1359 /* update lov_objid data, must be before transaction stop! */
1361 mdd_lov_objid_update(mdd, lmm);
1363 mdd_trans_stop(env, mdd, rc, handle);
1365 /* Finish mdd_lov_create() stuff. */
1366 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1370 /* Get fid from name and parent */
1372 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1373 const struct lu_name *lname, struct lu_fid* fid, int mask)
1375 const char *name = lname->ln_name;
1376 const struct dt_key *key = (const struct dt_key *)name;
1377 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1378 struct mdd_device *m = mdo2mdd(pobj);
1379 struct dt_object *dir = mdd_object_child(mdd_obj);
1380 struct lu_fid_pack *pack = &mdd_env_info(env)->mti_pack;
1384 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1387 rc = mdd_object_exists(mdd_obj);
1388 if (unlikely(rc == 0))
1390 else if (unlikely(rc < 0)) {
1391 CERROR("Object "DFID" locates on remote server\n",
1392 PFID(mdo2fid(mdd_obj)));
1396 /* The common filename length check. */
1397 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1398 RETURN(-ENAMETOOLONG);
1400 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1405 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1406 dt_try_as_dir(env, dir))) {
1408 rc = dir->do_index_ops->dio_lookup(env, dir,
1409 (struct dt_rec *)pack, key,
1410 mdd_object_capa(env, mdd_obj));
1412 rc = fid_unpack(pack, fid);
1421 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1422 const struct lu_name *lname, struct mdd_object *child,
1423 struct md_attr *ma, struct thandle *handle,
1424 const struct md_op_spec *spec)
1430 * Update attributes for child.
1433 * (1) the valid bits should be converted between Lustre and Linux;
1434 * (2) maybe, the child attributes should be set in OSD when creation.
1437 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1441 if (S_ISDIR(ma->ma_attr.la_mode)) {
1442 /* Add "." and ".." for newly created dir */
1443 __mdd_ref_add(env, child, handle);
1444 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1445 dot, handle, BYPASS_CAPA);
1447 rc = __mdd_index_insert_only(env, child, pfid,
1453 rc2 = __mdd_index_delete(env, child, dot, 1,
1454 handle, BYPASS_CAPA);
1456 CERROR("Failure to cleanup after dotdot"
1457 " creation: %d (%d)\n", rc2, rc);
1462 mdd_links_add(env, child, pfid, lname, handle);
1467 /* has not lock on pobj yet */
1468 static int mdd_create_sanity_check(const struct lu_env *env,
1469 struct md_object *pobj,
1470 const struct lu_name *lname,
1472 struct md_op_spec *spec)
1474 struct mdd_thread_info *info = mdd_env_info(env);
1475 struct lu_attr *la = &info->mti_la;
1476 struct lu_fid *fid = &info->mti_fid;
1477 struct mdd_object *obj = md2mdd_obj(pobj);
1478 struct mdd_device *m = mdo2mdd(pobj);
1479 int lookup = spec->sp_cr_lookup;
1484 if (mdd_is_dead_obj(obj))
1488 * In some cases this lookup is not needed - we know before if name
1489 * exists or not because MDT performs lookup for it.
1490 * name length check is done in lookup.
1494 * Check if the name already exist, though it will be checked in
1495 * _index_insert also, for avoiding rolling back if exists
1498 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1499 MAY_WRITE | MAY_EXEC);
1501 RETURN(rc ? : -EEXIST);
1504 * Check WRITE permission for the parent.
1505 * EXEC permission have been checked
1506 * when lookup before create already.
1508 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1515 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1519 if (la->la_mode & S_ISGID) {
1520 ma->ma_attr.la_gid = la->la_gid;
1521 if (S_ISDIR(ma->ma_attr.la_mode)) {
1522 ma->ma_attr.la_mode |= S_ISGID;
1523 ma->ma_attr.la_valid |= LA_MODE;
1527 switch (ma->ma_attr.la_mode & S_IFMT) {
1529 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1531 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1532 RETURN(-ENAMETOOLONG);
1552 * Create object and insert it into namespace.
1554 static int mdd_create(const struct lu_env *env,
1555 struct md_object *pobj,
1556 const struct lu_name *lname,
1557 struct md_object *child,
1558 struct md_op_spec *spec,
1561 struct mdd_thread_info *info = mdd_env_info(env);
1562 struct lu_attr *la = &info->mti_la_for_fix;
1563 struct md_attr *ma_acl = &info->mti_ma;
1564 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1565 struct mdd_object *son = md2mdd_obj(child);
1566 struct mdd_device *mdd = mdo2mdd(pobj);
1567 struct lu_attr *attr = &ma->ma_attr;
1568 struct lov_mds_md *lmm = NULL;
1569 struct thandle *handle;
1570 struct dynlock_handle *dlh;
1571 const char *name = lname->ln_name;
1572 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1573 int got_def_acl = 0;
1574 #ifdef HAVE_QUOTA_SUPPORT
1575 struct obd_device *obd = mdd->mdd_obd_dev;
1576 struct mds_obd *mds = &obd->u.mds;
1577 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1578 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1579 int quota_opc = 0, block_count = 0;
1580 int inode_pending = 0, block_pending = 0, parent_pending = 0;
1585 * Two operations have to be performed:
1587 * - an allocation of a new object (->do_create()), and
1589 * - an insertion into a parent index (->dio_insert()).
1591 * Due to locking, operation order is not important, when both are
1592 * successful, *but* error handling cases are quite different:
1594 * - if insertion is done first, and following object creation fails,
1595 * insertion has to be rolled back, but this operation might fail
1596 * also leaving us with dangling index entry.
1598 * - if creation is done first, is has to be undone if insertion
1599 * fails, leaving us with leaked space, which is neither good, nor
1602 * It seems that creation-first is simplest solution, but it is
1603 * sub-optimal in the frequent
1608 * case, because second mkdir is bound to create object, only to
1609 * destroy it immediately.
1611 * To avoid this follow local file systems that do double lookup:
1613 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1615 * 1. create (mdd_object_create_internal())
1617 * 2. insert (__mdd_index_insert(), lookup again)
1620 /* Sanity checks before big job. */
1621 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1625 #ifdef HAVE_QUOTA_SUPPORT
1626 if (mds->mds_quota) {
1627 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1629 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1633 quota_opc = FSFILT_OP_CREATE;
1634 mdd_quota_wrapper(&ma->ma_attr, qcids);
1635 mdd_quota_wrapper(la_tmp, qpids);
1636 /* get file quota for child */
1637 lquota_chkquota(mds_quota_interface_ref, obd,
1638 qcids[USRQUOTA], qcids[GRPQUOTA], 1,
1639 &inode_pending, NULL, 0);
1640 switch (ma->ma_attr.la_mode & S_IFMT) {
1649 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1650 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1654 /* get block quota for child and parent */
1656 lquota_chkquota(mds_quota_interface_ref, obd,
1657 qcids[USRQUOTA], qcids[GRPQUOTA],
1659 &block_pending, NULL,
1662 lquota_chkquota(mds_quota_interface_ref, obd,
1663 qpids[USRQUOTA], qpids[GRPQUOTA], 1,
1664 &parent_pending, NULL,
1671 * No RPC inside the transaction, so OST objects should be created at
1674 if (S_ISREG(attr->la_mode)) {
1675 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1678 GOTO(out_pending, rc);
1681 if (!S_ISLNK(attr->la_mode)) {
1682 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1683 ma_acl->ma_acl = info->mti_xattr_buf;
1684 ma_acl->ma_need = MA_ACL_DEF;
1685 ma_acl->ma_valid = 0;
1687 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1688 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1689 mdd_read_unlock(env, mdd_pobj);
1692 else if (ma_acl->ma_valid & MA_ACL_DEF)
1696 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1697 handle = mdd_trans_start(env, mdd);
1699 GOTO(out_free, rc = PTR_ERR(handle));
1701 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1703 GOTO(out_trans, rc = -ENOMEM);
1705 mdd_write_lock(env, son, MOR_TGT_CHILD);
1706 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1708 mdd_write_unlock(env, son);
1714 #ifdef CONFIG_FS_POSIX_ACL
1716 struct lu_buf *acl_buf = &info->mti_buf;
1717 acl_buf->lb_buf = ma_acl->ma_acl;
1718 acl_buf->lb_len = ma_acl->ma_acl_size;
1720 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1722 mdd_write_unlock(env, son);
1725 ma->ma_attr.la_valid |= LA_MODE;
1730 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1731 son, ma, handle, spec);
1732 mdd_write_unlock(env, son);
1735 * Object has no links, so it will be destroyed when last
1736 * reference is released. (XXX not now.)
1742 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1743 name, S_ISDIR(attr->la_mode), handle,
1744 mdd_object_capa(env, mdd_pobj));
1751 /* No need mdd_lsm_sanity_check here */
1752 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1754 CERROR("error on stripe info copy %d \n", rc);
1757 if (lmm && lmm_size > 0) {
1758 /* Set Lov here, do not get lmm again later */
1759 memcpy(ma->ma_lmm, lmm, lmm_size);
1760 ma->ma_lmm_size = lmm_size;
1761 ma->ma_valid |= MA_LOV;
1764 if (S_ISLNK(attr->la_mode)) {
1765 struct md_ucred *uc = md_ucred(env);
1766 struct dt_object *dt = mdd_object_child(son);
1767 const char *target_name = spec->u.sp_symname;
1768 int sym_len = strlen(target_name);
1769 const struct lu_buf *buf;
1772 buf = mdd_buf_get_const(env, target_name, sym_len);
1773 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1774 mdd_object_capa(env, son),
1776 CFS_CAP_SYS_RESOURCE_MASK);
1781 GOTO(cleanup, rc = -EFAULT);
1785 la->la_valid = LA_CTIME | LA_MTIME;
1786 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1790 /* Return attr back. */
1791 rc = mdd_attr_get_internal_locked(env, son, ma);
1794 if (rc && created) {
1798 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1799 S_ISDIR(attr->la_mode),
1800 handle, BYPASS_CAPA);
1802 CERROR("error can not cleanup destroy %d\n",
1807 mdd_write_lock(env, son, MOR_TGT_CHILD);
1808 __mdd_ref_del(env, son, handle, 0);
1809 if (initialized && S_ISDIR(attr->la_mode))
1810 __mdd_ref_del(env, son, handle, 1);
1811 mdd_write_unlock(env, son);
1815 /* update lov_objid data, must be before transaction stop! */
1817 mdd_lov_objid_update(mdd, lmm);
1819 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1822 rc = mdd_changelog_ns_store(env, mdd,
1823 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1824 S_ISREG(attr->la_mode) ? CL_CREATE :
1825 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1826 son, mdd_pobj, NULL, lname, handle);
1827 mdd_trans_stop(env, mdd, rc, handle);
1829 /* finis lov_create stuff, free all temporary data */
1830 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1832 #ifdef HAVE_QUOTA_SUPPORT
1835 lquota_pending_commit(mds_quota_interface_ref, obd,
1836 qcids[USRQUOTA], qcids[GRPQUOTA],
1839 lquota_pending_commit(mds_quota_interface_ref, obd,
1840 qcids[USRQUOTA], qcids[GRPQUOTA],
1843 lquota_pending_commit(mds_quota_interface_ref, obd,
1844 qpids[USRQUOTA], qpids[GRPQUOTA],
1846 /* Trigger dqacq on the owner of child and parent. If failed,
1847 * the next call for lquota_chkquota will process it. */
1848 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1856 * Get locks on parents in proper order
1857 * RETURN: < 0 - error, rename_order if successful
1865 static int mdd_rename_order(const struct lu_env *env,
1866 struct mdd_device *mdd,
1867 struct mdd_object *src_pobj,
1868 struct mdd_object *tgt_pobj)
1870 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1874 if (src_pobj == tgt_pobj)
1875 RETURN(MDD_RN_SAME);
1877 /* compared the parent child relationship of src_p&tgt_p */
1878 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1880 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1883 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1896 /* has not mdd_write{read}_lock on any obj yet. */
1897 static int mdd_rename_sanity_check(const struct lu_env *env,
1898 struct mdd_object *src_pobj,
1899 struct mdd_object *tgt_pobj,
1900 struct mdd_object *sobj,
1901 struct mdd_object *tobj,
1907 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1910 /* XXX: when get here, sobj must NOT be NULL,
1911 * the other case has been processed in cml_rename
1912 * before mdd_rename and enable MDS_PERM_BYPASS. */
1914 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1918 /* XXX: when get here, "tobj == NULL" means tobj must
1919 * NOT exist (neither on remote MDS, such case has been
1920 * processed in cml_rename before mdd_rename and enable
1922 * So check may_create, but not check may_unlink. */
1924 rc = mdd_may_create(env, tgt_pobj, NULL,
1925 (src_pobj != tgt_pobj), 0);
1927 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1928 (src_pobj != tgt_pobj), 1);
1930 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1931 S_ISDIR(ma->ma_attr.la_mode))
1932 rc = __mdd_may_link(env, tgt_pobj);
1937 /* src object can be remote that is why we use only fid and type of object */
1938 static int mdd_rename(const struct lu_env *env,
1939 struct md_object *src_pobj, struct md_object *tgt_pobj,
1940 const struct lu_fid *lf, const struct lu_name *lsname,
1941 struct md_object *tobj, const struct lu_name *ltname,
1944 const char *sname = lsname->ln_name;
1945 const char *tname = ltname->ln_name;
1946 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1947 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1948 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1949 struct mdd_device *mdd = mdo2mdd(src_pobj);
1950 struct mdd_object *mdd_sobj = NULL; /* source object */
1951 struct mdd_object *mdd_tobj = NULL;
1952 struct dynlock_handle *sdlh, *tdlh;
1953 struct thandle *handle;
1954 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1958 #ifdef HAVE_QUOTA_SUPPORT
1959 struct obd_device *obd = mdd->mdd_obd_dev;
1960 struct mds_obd *mds = &obd->u.mds;
1961 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1962 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1963 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1964 int quota_opc = 0, rec_pending = 0;
1968 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1969 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1972 mdd_tobj = md2mdd_obj(tobj);
1974 #ifdef HAVE_QUOTA_SUPPORT
1975 if (mds->mds_quota) {
1976 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1978 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
1980 mdd_quota_wrapper(la_tmp, qspids);
1982 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
1985 quota_opc = FSFILT_OP_LINK;
1986 mdd_quota_wrapper(la_tmp, qtpids);
1987 /* get block quota for target parent */
1988 lquota_chkquota(mds_quota_interface_ref,
1989 obd, qtpids[USRQUOTA],
1990 qtpids[GRPQUOTA], 1,
1998 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
1999 handle = mdd_trans_start(env, mdd);
2001 GOTO(out_pending, rc = PTR_ERR(handle));
2003 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2004 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2006 GOTO(cleanup_unlocked, rc);
2008 /* Get locks in determined order */
2009 if (rc == MDD_RN_SAME) {
2010 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2011 sname, MOR_SRC_PARENT);
2012 /* check hashes to determine do we need one lock or two */
2013 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2014 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2018 } else if (rc == MDD_RN_SRCTGT) {
2019 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2020 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2022 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2023 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2025 if (sdlh == NULL || tdlh == NULL)
2026 GOTO(cleanup, rc = -ENOMEM);
2028 mdd_sobj = mdd_object_find(env, mdd, lf);
2029 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2030 mdd_sobj, mdd_tobj, ma);
2034 /* Remove source name from source directory */
2035 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2036 mdd_object_capa(env, mdd_spobj));
2040 /* "mv dir1 dir2" needs "dir1/.." link update */
2041 if (is_dir && mdd_sobj) {
2042 rc = __mdd_index_delete(env, mdd_sobj, dotdot, is_dir, handle,
2043 mdd_object_capa(env, mdd_spobj));
2047 rc = __mdd_index_insert(env, mdd_sobj, tpobj_fid, dotdot,
2049 mdd_object_capa(env, mdd_tpobj));
2054 /* Remove target name from target directory
2055 * Here tobj can be remote one, so we do index_delete unconditionally
2056 * and -ENOENT is allowed.
2058 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2059 mdd_object_capa(env, mdd_tpobj));
2060 if (rc != 0 && rc != -ENOENT)
2063 /* Insert new fid with target name into target dir */
2064 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2065 mdd_object_capa(env, mdd_tpobj));
2069 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2070 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2072 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2074 la->la_valid = LA_CTIME;
2075 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2081 /* Remove old target object
2082 * For tobj is remote case cmm layer has processed
2083 * and set tobj to NULL then. So when tobj is NOT NULL,
2084 * it must be local one.
2086 if (tobj && mdd_object_exists(mdd_tobj)) {
2087 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2088 __mdd_ref_del(env, mdd_tobj, handle, 0);
2090 /* Remove dot reference. */
2092 __mdd_ref_del(env, mdd_tobj, handle, 1);
2094 la->la_valid = LA_CTIME;
2095 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2099 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2100 mdd_write_unlock(env, mdd_tobj);
2104 #ifdef HAVE_QUOTA_SUPPORT
2105 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2106 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2107 quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2108 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2113 la->la_valid = LA_CTIME | LA_MTIME;
2114 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2118 if (mdd_spobj != mdd_tpobj) {
2119 la->la_valid = LA_CTIME | LA_MTIME;
2120 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2124 if (rc == 0 && mdd_sobj) {
2125 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2126 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2127 mdo2fid(mdd_tpobj), ltname, handle);
2129 /* Old files might not have EA entry */
2130 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2132 mdd_write_unlock(env, mdd_sobj);
2133 /* We don't fail the transaction if the link ea can't be
2134 updated -- fid2path will use alternate lookup method. */
2140 if (likely(tdlh) && sdlh != tdlh)
2141 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2143 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2146 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2147 mdd_spobj, lf, lsname, handle);
2149 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2150 mdd_tpobj, lf, ltname, handle);
2152 mdd_trans_stop(env, mdd, rc, handle);
2154 mdd_object_put(env, mdd_sobj);
2156 #ifdef HAVE_QUOTA_SUPPORT
2157 if (mds->mds_quota) {
2159 lquota_pending_commit(mds_quota_interface_ref, obd,
2163 /* Trigger dqrel on the source owner of parent.
2164 * If failed, the next call for lquota_chkquota will
2166 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2167 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2169 /* Trigger dqrel/dqacq on the target owner of child and
2170 * parent. If failed, the next call for lquota_chkquota
2171 * will process it. */
2172 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2173 qtpids, rc, quota_opc);
2179 /** enable/disable storing of hardlink info */
2180 int mdd_linkea_enable = 1;
2181 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2182 "record hardlink info in EAs");
2184 /** Read the link EA into a temp buffer.
2185 * Uses the name_buf since it is generally large.
2186 * \retval IS_ERR err
2187 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2189 struct lu_buf *mdd_links_get(const struct lu_env *env,
2190 struct mdd_object *mdd_obj)
2193 struct lustre_capa *capa;
2194 struct link_ea_header *leh;
2197 /* First try a small buf */
2198 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2199 if (buf->lb_buf == NULL)
2200 return ERR_PTR(-ENOMEM);
2202 capa = mdd_object_capa(env, mdd_obj);
2203 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2204 if (rc == -ERANGE) {
2205 /* Buf was too small, figure out what we need. */
2208 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2211 buf = mdd_buf_alloc(env, rc);
2212 if (buf->lb_buf == NULL)
2213 return ERR_PTR(-ENOMEM);
2214 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2220 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2221 leh->leh_magic = LINK_EA_MAGIC;
2222 leh->leh_reccount = __swab32(leh->leh_reccount);
2223 leh->leh_len = __swab64(leh->leh_len);
2224 /* entries are swabbed by mdd_lee_unpack */
2226 if (leh->leh_magic != LINK_EA_MAGIC)
2227 return ERR_PTR(-EINVAL);
2232 /** Pack a link_ea_entry.
2233 * All elements are stored as chars to avoid alignment issues.
2234 * Numbers are always big-endian
2235 * \param packbuf is a temp fid buffer
2236 * \retval record length
2238 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2239 const struct lu_fid *pfid, struct lu_fid* packbuf)
2244 fid_pack(&lee->lee_parent_fid, pfid, packbuf);
2245 ptr = (char *)&lee->lee_parent_fid + lee->lee_parent_fid.fp_len;
2246 strncpy(ptr, lname->ln_name, lname->ln_namelen);
2247 reclen = lee->lee_parent_fid.fp_len + lname->ln_namelen +
2248 sizeof(lee->lee_reclen);
2249 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2250 lee->lee_reclen[1] = reclen & 0xff;
2254 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2255 struct lu_name *lname, struct lu_fid *pfid)
2257 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2258 fid_unpack(&lee->lee_parent_fid, pfid);
2259 lname->ln_name = (char *)&lee->lee_parent_fid +
2260 lee->lee_parent_fid.fp_len;
2261 lname->ln_namelen = *reclen - lee->lee_parent_fid.fp_len -
2262 sizeof(lee->lee_reclen);
2265 /** Add a record to the end of link ea buf */
2266 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2267 const struct lu_fid *pfid,
2268 const struct lu_name *lname)
2270 struct link_ea_header *leh;
2271 struct link_ea_entry *lee;
2274 if (lname == NULL || pfid == NULL)
2277 /* Make sure our buf is big enough for the new one */
2279 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2280 if (leh->leh_len + reclen > buf->lb_len) {
2281 mdd_buf_grow(env, leh->leh_len + reclen);
2282 if (buf->lb_buf == NULL)
2287 lee = buf->lb_buf + leh->leh_len;
2288 reclen = mdd_lee_pack(lee, lname, pfid, &mdd_env_info(env)->mti_fid2);
2289 leh->leh_len += reclen;
2290 leh->leh_reccount++;
2294 /* For pathologic linkers, we don't want to spend lots of time scanning the
2295 * link ea. Limit ourseleves to something reasonable; links not in the EA
2296 * can be looked up via (slower) parent lookup.
2298 #define LINKEA_MAX_COUNT 128
2300 static int mdd_links_add(const struct lu_env *env,
2301 struct mdd_object *mdd_obj,
2302 const struct lu_fid *pfid,
2303 const struct lu_name *lname,
2304 struct thandle *handle)
2307 struct link_ea_header *leh;
2311 if (!mdd_linkea_enable)
2314 buf = mdd_links_get(env, mdd_obj);
2317 if (rc != -ENODATA) {
2318 CERROR("link_ea read failed %d "DFID"\n", rc,
2319 PFID(mdd_object_fid(mdd_obj)));
2322 /* empty EA; start one */
2323 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2324 if (buf->lb_buf == NULL)
2327 leh->leh_magic = LINK_EA_MAGIC;
2328 leh->leh_len = sizeof(struct link_ea_header);
2329 leh->leh_reccount = 0;
2333 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2336 rc = __mdd_links_add(env, buf, pfid, lname);
2341 rc = __mdd_xattr_set(env, mdd_obj,
2342 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2343 XATTR_NAME_LINK, 0, handle);
2345 CERROR("link_ea add failed %d "DFID"\n", rc,
2346 PFID(mdd_object_fid(mdd_obj)));
2348 if (buf->lb_vmalloc)
2349 /* if we vmalloced a large buffer drop it */
2355 static int mdd_links_rename(const struct lu_env *env,
2356 struct mdd_object *mdd_obj,
2357 const struct lu_fid *oldpfid,
2358 const struct lu_name *oldlname,
2359 const struct lu_fid *newpfid,
2360 const struct lu_name *newlname,
2361 struct thandle *handle)
2364 struct link_ea_header *leh;
2365 struct link_ea_entry *lee;
2366 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2367 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2372 if (!mdd_linkea_enable)
2375 if (mdd_obj->mod_flags & DEAD_OBJ)
2376 /* No more links, don't bother */
2379 buf = mdd_links_get(env, mdd_obj);
2382 CERROR("link_ea read failed %d "DFID"\n",
2383 rc, PFID(mdd_object_fid(mdd_obj)));
2387 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2389 /* Find the old record */
2390 for(count = 0; count <= leh->leh_reccount; count++) {
2391 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2392 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2393 lu_fid_eq(tmpfid, oldpfid) &&
2394 (strncmp(tmpname->ln_name, oldlname->ln_name,
2395 tmpname->ln_namelen) == 0))
2397 lee = (struct link_ea_entry *)((char *)lee + reclen);
2399 if (count > leh->leh_reccount) {
2400 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2401 oldlname->ln_namelen, oldlname->ln_name);
2402 GOTO(out, rc = -ENOENT);
2405 /* Remove the old record */
2406 leh->leh_reccount--;
2407 leh->leh_len -= reclen;
2408 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2411 /* If renaming, add the new record */
2412 if (newpfid != NULL) {
2413 rc = __mdd_links_add(env, buf, newpfid, newlname);
2419 rc = __mdd_xattr_set(env, mdd_obj,
2420 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2421 XATTR_NAME_LINK, 0, handle);
2425 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2426 oldlname->ln_namelen, oldlname->ln_name, rc,
2427 PFID(mdd_object_fid(mdd_obj)));
2429 if (buf->lb_vmalloc)
2430 /* if we vmalloced a large buffer drop it */
2436 const struct md_dir_operations mdd_dir_ops = {
2437 .mdo_is_subdir = mdd_is_subdir,
2438 .mdo_lookup = mdd_lookup,
2439 .mdo_create = mdd_create,
2440 .mdo_rename = mdd_rename,
2441 .mdo_link = mdd_link,
2442 .mdo_unlink = mdd_unlink,
2443 .mdo_name_insert = mdd_name_insert,
2444 .mdo_name_remove = mdd_name_remove,
2445 .mdo_rename_tgt = mdd_rename_tgt,
2446 .mdo_create_data = mdd_create_data