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 if (mdd_is_dead_obj(src_obj))
454 /* Local ops, no lookup before link, check filename length here. */
455 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
456 RETURN(-ENAMETOOLONG);
458 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
461 if (S_ISDIR(mdd_object_type(src_obj)))
464 LASSERT(src_obj != tgt_obj);
466 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
471 rc = __mdd_may_link(env, src_obj);
477 * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and
478 * assign i_nlink to 1 which means the i_nlink for subdir count is incredible
479 * (maybe too large to be represented). It is a trick to break through the
480 * "i_nlink" limitation for subdir count.
482 void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj,
483 struct thandle *handle)
485 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
486 struct mdd_device *m = mdd_obj2mdd_dev(obj);
488 if (!mdd_is_mnlink(obj)) {
489 if (S_ISDIR(mdd_object_type(obj))) {
490 if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA))
493 if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) {
494 obj->mod_flags |= MNLINK_OBJ;
495 tmp_la->la_nlink = 1;
496 tmp_la->la_valid = LA_NLINK;
497 mdd_attr_set_internal(env, obj, tmp_la, handle,
502 mdo_ref_add(env, obj, handle);
506 void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj,
507 struct thandle *handle, int is_dot)
509 if (!mdd_is_mnlink(obj) || is_dot)
510 mdo_ref_del(env, obj, handle);
513 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
514 const char *name, struct thandle *handle,
515 struct lustre_capa *capa)
517 struct dt_object *next = mdd_object_child(pobj);
521 if (dt_try_as_dir(env, next)) {
522 rc = next->do_index_ops->dio_delete(env, next,
523 (struct dt_key *)name,
531 static int __mdd_index_insert_only(const struct lu_env *env,
532 struct mdd_object *pobj,
533 const struct lu_fid *lf, const char *name,
534 struct thandle *handle,
535 struct lustre_capa *capa)
537 struct dt_object *next = mdd_object_child(pobj);
541 if (dt_try_as_dir(env, next)) {
542 struct md_ucred *uc = md_ucred(env);
544 rc = next->do_index_ops->dio_insert(env, next,
545 __mdd_fid_rec(env, lf),
546 (const struct dt_key *)name,
547 handle, capa, uc->mu_cap &
548 CFS_CAP_SYS_RESOURCE_MASK);
555 /* insert named index, add reference if isdir */
556 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
557 const struct lu_fid *lf, const char *name, int is_dir,
558 struct thandle *handle, struct lustre_capa *capa)
563 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
564 if (rc == 0 && is_dir) {
565 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
566 __mdd_ref_add(env, pobj, handle);
567 mdd_write_unlock(env, pobj);
572 /* delete named index, drop reference if isdir */
573 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
574 const char *name, int is_dir, struct thandle *handle,
575 struct lustre_capa *capa)
580 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
581 if (rc == 0 && is_dir) {
584 if (name != NULL && name[0] == '.' && name[1] == 0)
586 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
587 __mdd_ref_del(env, pobj, handle, is_dot);
588 mdd_write_unlock(env, pobj);
595 /** Store a namespace change changelog record
596 * If this fails, we must fail the whole transaction; we don't
597 * want the change to commit without the log entry.
598 * \param target - mdd_object of change
599 * \param parent - parent dir/object
600 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
601 * \param tname - target name string
602 * \param handle - transacion handle
604 static int mdd_changelog_ns_store(const struct lu_env *env,
605 struct mdd_device *mdd,
606 enum changelog_rec_type type,
607 struct mdd_object *target,
608 struct mdd_object *parent,
609 const struct lu_fid *tf,
610 const struct lu_name *tname,
611 struct thandle *handle)
613 const struct lu_fid *tfid;
614 const struct lu_fid *tpfid = mdo2fid(parent);
615 struct llog_changelog_rec *rec;
621 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
624 LASSERT(parent != NULL);
625 LASSERT(tname != NULL);
626 LASSERT(handle != NULL);
629 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
630 buf = mdd_buf_alloc(env, reclen);
631 if (buf->lb_buf == NULL)
633 rec = (struct llog_changelog_rec *)buf->lb_buf;
635 rec->cr_flags = CLF_VERSION;
636 rec->cr_type = (__u32)type;
637 tfid = tf ? tf : mdo2fid(target);
638 rec->cr_tfid = *tfid;
639 rec->cr_pfid = *tpfid;
640 rec->cr_namelen = tname->ln_namelen;
641 memcpy(rec->cr_name, tname->ln_name, rec->cr_namelen);
643 target->mod_cltime = cfs_time_current_64();
645 rc = mdd_changelog_llog_write(mdd, rec, handle);
647 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
648 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
655 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
656 struct md_object *src_obj, const struct lu_name *lname,
659 const char *name = lname->ln_name;
660 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
661 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
662 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
663 struct mdd_device *mdd = mdo2mdd(src_obj);
664 struct dynlock_handle *dlh;
665 struct thandle *handle;
666 #ifdef HAVE_QUOTA_SUPPORT
667 struct obd_device *obd = mdd->mdd_obd_dev;
668 struct mds_obd *mds = &obd->u.mds;
669 unsigned int qids[MAXQUOTAS] = { 0, 0 };
670 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
675 #ifdef HAVE_QUOTA_SUPPORT
676 if (mds->mds_quota) {
677 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
679 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
682 mdd_data_get(env, mdd_tobj, &data);
683 quota_opc = FSFILT_OP_LINK;
684 mdd_quota_wrapper(la_tmp, qids);
685 /* get block quota for parent */
686 lquota_chkquota(mds_quota_interface_ref, obd,
687 qids, rec_pending, 1, NULL,
688 LQUOTA_FLAGS_BLK, data, 1);
693 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
694 handle = mdd_trans_start(env, mdd);
696 GOTO(out_pending, rc = PTR_ERR(handle));
698 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
700 GOTO(out_trans, rc = -ENOMEM);
701 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
703 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
705 GOTO(out_unlock, rc);
707 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
709 mdd_object_capa(env, mdd_tobj));
711 GOTO(out_unlock, rc);
713 __mdd_ref_add(env, mdd_sobj, handle);
715 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
716 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
718 la->la_valid = LA_CTIME | LA_MTIME;
719 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
721 GOTO(out_unlock, rc);
723 la->la_valid = LA_CTIME;
724 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
726 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
730 mdd_write_unlock(env, mdd_sobj);
731 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
734 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
735 mdd_tobj, NULL, lname, handle);
736 mdd_trans_stop(env, mdd, rc, handle);
738 #ifdef HAVE_QUOTA_SUPPORT
740 lquota_pending_commit(mds_quota_interface_ref, obd,
741 qids, rec_pending, 1);
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 if (mdd_is_dead_obj(cobj))
802 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
807 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
808 struct md_object *cobj, const struct lu_name *lname,
811 const char *name = lname->ln_name;
812 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
813 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
814 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
815 struct mdd_device *mdd = mdo2mdd(pobj);
816 struct dynlock_handle *dlh;
817 struct thandle *handle;
818 #ifdef HAVE_QUOTA_SUPPORT
819 struct obd_device *obd = mdd->mdd_obd_dev;
820 struct mds_obd *mds = &obd->u.mds;
821 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
822 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
825 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
829 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
830 PFID(mdd_object_fid(mdd_cobj)));
832 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
836 handle = mdd_trans_start(env, mdd);
838 RETURN(PTR_ERR(handle));
840 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
842 GOTO(out_trans, rc = -ENOMEM);
843 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
845 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
849 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
850 mdd_object_capa(env, mdd_pobj));
854 __mdd_ref_del(env, mdd_cobj, handle, 0);
857 __mdd_ref_del(env, mdd_cobj, handle, 1);
859 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
860 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
862 la->la_valid = LA_CTIME | LA_MTIME;
863 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
867 la->la_valid = LA_CTIME;
868 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
872 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
873 #ifdef HAVE_QUOTA_SUPPORT
874 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
875 ma->ma_attr.la_nlink == 0) {
876 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
878 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
880 mdd_quota_wrapper(la_tmp, qpids);
881 if (mdd_cobj->mod_count == 0) {
882 quota_opc = FSFILT_OP_UNLINK;
883 mdd_quota_wrapper(&ma->ma_attr, qcids);
885 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
892 obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
893 sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
896 /* old files may not have link ea; ignore errors */
897 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
898 lname, NULL, NULL, handle);
902 mdd_write_unlock(env, mdd_cobj);
903 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
906 rc = mdd_changelog_ns_store(env, mdd,
907 is_dir ? CL_RMDIR : CL_UNLINK,
908 mdd_cobj, mdd_pobj, NULL, lname,
911 mdd_trans_stop(env, mdd, rc, handle);
912 #ifdef HAVE_QUOTA_SUPPORT
914 /* Trigger dqrel on the owner of child and parent. If failed,
915 * the next call for lquota_chkquota will process it. */
916 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
922 /* has not lock on pobj yet */
923 static int mdd_ni_sanity_check(const struct lu_env *env,
924 struct md_object *pobj,
925 const struct md_attr *ma)
927 struct mdd_object *obj = md2mdd_obj(pobj);
931 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
934 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
942 static int mdd_name_insert(const struct lu_env *env,
943 struct md_object *pobj,
944 const struct lu_name *lname,
945 const struct lu_fid *fid,
946 const struct md_attr *ma)
948 const char *name = lname->ln_name;
949 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
950 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
951 struct mdd_device *mdd = mdo2mdd(pobj);
952 struct dynlock_handle *dlh;
953 struct thandle *handle;
954 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
955 #ifdef HAVE_QUOTA_SUPPORT
956 struct md_ucred *uc = md_ucred(env);
957 struct obd_device *obd = mdd->mdd_obd_dev;
958 struct mds_obd *mds = &obd->u.mds;
959 unsigned int qids[MAXQUOTAS] = { 0, 0 };
960 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
961 cfs_cap_t save = uc->mu_cap;
966 #ifdef HAVE_QUOTA_SUPPORT
967 if (mds->mds_quota) {
968 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
969 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
971 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
974 mdd_data_get(env, mdd_obj, &data);
975 quota_opc = FSFILT_OP_LINK;
976 mdd_quota_wrapper(la_tmp, qids);
977 /* get block quota for parent */
978 lquota_chkquota(mds_quota_interface_ref, obd,
979 qids, rec_pending, 1, NULL,
980 LQUOTA_FLAGS_BLK, data, 1);
983 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
987 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
988 handle = mdd_trans_start(env, mdo2mdd(pobj));
990 GOTO(out_pending, rc = PTR_ERR(handle));
992 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
994 GOTO(out_trans, rc = -ENOMEM);
996 rc = mdd_ni_sanity_check(env, pobj, ma);
998 GOTO(out_unlock, rc);
1000 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1001 handle, BYPASS_CAPA);
1003 GOTO(out_unlock, rc);
1006 * For some case, no need update obj's ctime (LA_CTIME is not set),
1008 * For other cases, update obj's ctime (LA_CTIME is set),
1011 if (ma->ma_attr.la_valid & LA_CTIME) {
1012 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1013 la->la_valid = LA_CTIME | LA_MTIME;
1014 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1019 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1021 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1023 #ifdef HAVE_QUOTA_SUPPORT
1024 if (mds->mds_quota) {
1026 lquota_pending_commit(mds_quota_interface_ref,
1027 obd, qids, rec_pending, 1);
1028 /* Trigger dqacq for the parent owner. If failed,
1029 * the next call for lquota_chkquota will process it*/
1030 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1040 /* has not lock on pobj yet */
1041 static int mdd_nr_sanity_check(const struct lu_env *env,
1042 struct md_object *pobj,
1043 const struct md_attr *ma)
1045 struct mdd_object *obj = md2mdd_obj(pobj);
1049 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1052 rc = mdd_may_unlink(env, obj, ma);
1058 * Partial operation.
1060 static int mdd_name_remove(const struct lu_env *env,
1061 struct md_object *pobj,
1062 const struct lu_name *lname,
1063 const struct md_attr *ma)
1065 const char *name = lname->ln_name;
1066 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1067 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1068 struct mdd_device *mdd = mdo2mdd(pobj);
1069 struct dynlock_handle *dlh;
1070 struct thandle *handle;
1071 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1072 #ifdef HAVE_QUOTA_SUPPORT
1073 struct obd_device *obd = mdd->mdd_obd_dev;
1074 struct mds_obd *mds = &obd->u.mds;
1075 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1081 #ifdef HAVE_QUOTA_SUPPORT
1082 if (mds->mds_quota) {
1083 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1085 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1087 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1088 mdd_quota_wrapper(la_tmp, qids);
1092 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1093 handle = mdd_trans_start(env, mdd);
1095 GOTO(out_pending, rc = PTR_ERR(handle));
1097 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1099 GOTO(out_trans, rc = -ENOMEM);
1101 rc = mdd_nr_sanity_check(env, pobj, ma);
1103 GOTO(out_unlock, rc);
1105 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1106 handle, BYPASS_CAPA);
1108 GOTO(out_unlock, rc);
1111 * For some case, no need update obj's ctime (LA_CTIME is not set),
1113 * For other cases, update obj's ctime (LA_CTIME is set),
1116 if (ma->ma_attr.la_valid & LA_CTIME) {
1117 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1118 la->la_valid = LA_CTIME | LA_MTIME;
1119 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1124 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1126 mdd_trans_stop(env, mdd, rc, handle);
1128 #ifdef HAVE_QUOTA_SUPPORT
1129 /* Trigger dqrel for the parent owner.
1130 * If failed, the next call for lquota_chkquota will process it. */
1132 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1140 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1142 static int mdd_rt_sanity_check(const struct lu_env *env,
1143 struct mdd_object *tgt_pobj,
1144 struct mdd_object *tobj,
1150 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1153 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1154 * exist. In fact, tobj must exist, otherwise the call trace will be:
1155 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1156 * When get here, tobj must be NOT NULL, the other case has been
1157 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1159 * So check may_delete, but not check nlink of tgt_pobj. */
1161 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1166 /* Partial rename op on slave MDD */
1167 static int mdd_rename_tgt(const struct lu_env *env,
1168 struct md_object *pobj, struct md_object *tobj,
1169 const struct lu_fid *lf, const struct lu_name *lname,
1172 const char *name = lname->ln_name;
1173 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1174 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1175 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1176 struct mdd_device *mdd = mdo2mdd(pobj);
1177 struct dynlock_handle *dlh;
1178 struct thandle *handle;
1179 #ifdef HAVE_QUOTA_SUPPORT
1180 struct obd_device *obd = mdd->mdd_obd_dev;
1181 struct mds_obd *mds = &obd->u.mds;
1182 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1183 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1184 int quota_copc = 0, quota_popc = 0;
1185 int rec_pending[MAXQUOTAS] = { 0, 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_popc = 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, rec_pending, 1, NULL,
1203 LQUOTA_FLAGS_BLK, data, 1);
1207 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1208 handle = mdd_trans_start(env, mdd);
1210 GOTO(out_pending, rc = PTR_ERR(handle));
1212 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1214 GOTO(out_trans, rc = -ENOMEM);
1216 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1218 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1223 * If rename_tgt is called then we should just re-insert name with
1224 * correct fid, no need to dec/inc parent nlink if obj is dir.
1226 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1230 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1235 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1236 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1238 la->la_valid = LA_CTIME | LA_MTIME;
1239 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1244 * For tobj is remote case cmm layer has processed
1245 * and pass NULL tobj to here. So when tobj is NOT NULL,
1246 * it must be local one.
1248 if (tobj && mdd_object_exists(mdd_tobj)) {
1249 __mdd_ref_del(env, mdd_tobj, handle, 0);
1251 /* Remove dot reference. */
1252 if (S_ISDIR(ma->ma_attr.la_mode))
1253 __mdd_ref_del(env, mdd_tobj, handle, 1);
1255 la->la_valid = LA_CTIME;
1256 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1260 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1264 #ifdef HAVE_QUOTA_SUPPORT
1265 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1266 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1267 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1268 mdd_quota_wrapper(&ma->ma_attr, qcids);
1275 mdd_write_unlock(env, mdd_tobj);
1276 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1279 /* Bare EXT record with no RENAME in front of it signifies
1280 a partial slave op */
1281 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1282 mdd_tpobj, NULL, lname, handle);
1284 mdd_trans_stop(env, mdd, rc, handle);
1286 #ifdef HAVE_QUOTA_SUPPORT
1287 if (mds->mds_quota) {
1289 lquota_pending_commit(mds_quota_interface_ref, obd,
1290 qpids, rec_pending, 1);
1293 /* Trigger dqrel on the target owner of child.
1294 * If failed, the next call for lquota_chkquota
1295 * will process it. */
1296 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1304 * The permission has been checked when obj created, no need check again.
1306 static int mdd_cd_sanity_check(const struct lu_env *env,
1307 struct mdd_object *obj)
1312 if (!obj || mdd_is_dead_obj(obj))
1319 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1320 struct md_object *cobj, const struct md_op_spec *spec,
1323 struct mdd_device *mdd = mdo2mdd(cobj);
1324 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1325 struct mdd_object *son = md2mdd_obj(cobj);
1326 struct lu_attr *attr = &ma->ma_attr;
1327 struct lov_mds_md *lmm = NULL;
1329 struct thandle *handle;
1333 rc = mdd_cd_sanity_check(env, son);
1337 if (!md_should_create(spec->sp_cr_flags))
1339 lmm_size = ma->ma_lmm_size;
1340 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1345 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1346 handle = mdd_trans_start(env, mdd);
1348 GOTO(out_free, rc = PTR_ERR(handle));
1351 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1352 * Should this be fixed?
1355 /* Replay creates has objects already */
1357 if (spec->no_create) {
1358 CDEBUG(D_INFO, "we already have lov ea\n");
1359 rc = mdd_lov_set_md(env, mdd_pobj, son,
1360 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1361 spec->u.sp_ea.eadatalen, handle, 0);
1364 /* No need mdd_lsm_sanity_check here */
1365 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1366 lmm_size, handle, 0);
1369 rc = mdd_attr_get_internal_locked(env, son, ma);
1371 /* update lov_objid data, must be before transaction stop! */
1373 mdd_lov_objid_update(mdd, lmm);
1375 mdd_trans_stop(env, mdd, rc, handle);
1377 /* Finish mdd_lov_create() stuff. */
1378 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1382 /* Get fid from name and parent */
1384 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1385 const struct lu_name *lname, struct lu_fid* fid, int mask)
1387 const char *name = lname->ln_name;
1388 const struct dt_key *key = (const struct dt_key *)name;
1389 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1390 struct mdd_device *m = mdo2mdd(pobj);
1391 struct dt_object *dir = mdd_object_child(mdd_obj);
1392 struct lu_fid_pack *pack = &mdd_env_info(env)->mti_pack;
1396 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1399 rc = mdd_object_exists(mdd_obj);
1400 if (unlikely(rc == 0))
1402 else if (unlikely(rc < 0)) {
1403 CERROR("Object "DFID" locates on remote server\n",
1404 PFID(mdo2fid(mdd_obj)));
1408 /* The common filename length check. */
1409 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1410 RETURN(-ENAMETOOLONG);
1412 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1417 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1418 dt_try_as_dir(env, dir))) {
1420 rc = dir->do_index_ops->dio_lookup(env, dir,
1421 (struct dt_rec *)pack, key,
1422 mdd_object_capa(env, mdd_obj));
1424 rc = fid_unpack(pack, fid);
1433 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1434 const struct lu_name *lname, struct mdd_object *child,
1435 struct md_attr *ma, struct thandle *handle,
1436 const struct md_op_spec *spec)
1442 * Update attributes for child.
1445 * (1) the valid bits should be converted between Lustre and Linux;
1446 * (2) maybe, the child attributes should be set in OSD when creation.
1449 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1453 if (S_ISDIR(ma->ma_attr.la_mode)) {
1454 /* Add "." and ".." for newly created dir */
1455 __mdd_ref_add(env, child, handle);
1456 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1457 dot, handle, BYPASS_CAPA);
1459 rc = __mdd_index_insert_only(env, child, pfid,
1465 rc2 = __mdd_index_delete(env, child, dot, 1,
1466 handle, BYPASS_CAPA);
1468 CERROR("Failure to cleanup after dotdot"
1469 " creation: %d (%d)\n", rc2, rc);
1474 mdd_links_add(env, child, pfid, lname, handle);
1479 /* has not lock on pobj yet */
1480 static int mdd_create_sanity_check(const struct lu_env *env,
1481 struct md_object *pobj,
1482 const struct lu_name *lname,
1484 struct md_op_spec *spec)
1486 struct mdd_thread_info *info = mdd_env_info(env);
1487 struct lu_attr *la = &info->mti_la;
1488 struct lu_fid *fid = &info->mti_fid;
1489 struct mdd_object *obj = md2mdd_obj(pobj);
1490 struct mdd_device *m = mdo2mdd(pobj);
1491 int lookup = spec->sp_cr_lookup;
1496 if (mdd_is_dead_obj(obj))
1500 * In some cases this lookup is not needed - we know before if name
1501 * exists or not because MDT performs lookup for it.
1502 * name length check is done in lookup.
1506 * Check if the name already exist, though it will be checked in
1507 * _index_insert also, for avoiding rolling back if exists
1510 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1511 MAY_WRITE | MAY_EXEC);
1513 RETURN(rc ? : -EEXIST);
1516 * Check WRITE permission for the parent.
1517 * EXEC permission have been checked
1518 * when lookup before create already.
1520 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1527 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1531 if (la->la_mode & S_ISGID) {
1532 ma->ma_attr.la_gid = la->la_gid;
1533 if (S_ISDIR(ma->ma_attr.la_mode)) {
1534 ma->ma_attr.la_mode |= S_ISGID;
1535 ma->ma_attr.la_valid |= LA_MODE;
1539 switch (ma->ma_attr.la_mode & S_IFMT) {
1541 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1543 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1544 RETURN(-ENAMETOOLONG);
1564 * Create object and insert it into namespace.
1566 static int mdd_create(const struct lu_env *env,
1567 struct md_object *pobj,
1568 const struct lu_name *lname,
1569 struct md_object *child,
1570 struct md_op_spec *spec,
1573 struct mdd_thread_info *info = mdd_env_info(env);
1574 struct lu_attr *la = &info->mti_la_for_fix;
1575 struct md_attr *ma_acl = &info->mti_ma;
1576 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1577 struct mdd_object *son = md2mdd_obj(child);
1578 struct mdd_device *mdd = mdo2mdd(pobj);
1579 struct lu_attr *attr = &ma->ma_attr;
1580 struct lov_mds_md *lmm = NULL;
1581 struct thandle *handle;
1582 struct dynlock_handle *dlh;
1583 const char *name = lname->ln_name;
1584 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1585 int got_def_acl = 0;
1586 #ifdef HAVE_QUOTA_SUPPORT
1587 struct obd_device *obd = mdd->mdd_obd_dev;
1588 struct mds_obd *mds = &obd->u.mds;
1589 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1590 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1591 int quota_opc = 0, block_count = 0;
1592 int inode_pending[MAXQUOTAS] = { 0, 0 };
1593 int block_pending[MAXQUOTAS] = { 0, 0 };
1594 int parent_pending[MAXQUOTAS] = { 0, 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, qcids,
1652 inode_pending, 1, NULL, 0, NULL, 0);
1653 switch (ma->ma_attr.la_mode & S_IFMT) {
1662 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1663 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1667 /* get block quota for child and parent */
1669 lquota_chkquota(mds_quota_interface_ref, obd,
1670 qcids, block_pending,
1672 LQUOTA_FLAGS_BLK, NULL, 0);
1674 lquota_chkquota(mds_quota_interface_ref, obd,
1675 qpids, parent_pending, 1, NULL,
1676 LQUOTA_FLAGS_BLK, NULL, 0);
1682 * No RPC inside the transaction, so OST objects should be created at
1685 if (S_ISREG(attr->la_mode)) {
1686 lmm_size = ma->ma_lmm_size;
1687 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1690 GOTO(out_pending, rc);
1693 if (!S_ISLNK(attr->la_mode)) {
1694 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1695 ma_acl->ma_acl = info->mti_xattr_buf;
1696 ma_acl->ma_need = MA_ACL_DEF;
1697 ma_acl->ma_valid = 0;
1699 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1700 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1701 mdd_read_unlock(env, mdd_pobj);
1704 else if (ma_acl->ma_valid & MA_ACL_DEF)
1708 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1709 handle = mdd_trans_start(env, mdd);
1711 GOTO(out_free, rc = PTR_ERR(handle));
1713 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1715 GOTO(out_trans, rc = -ENOMEM);
1717 mdd_write_lock(env, son, MOR_TGT_CHILD);
1718 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1720 mdd_write_unlock(env, son);
1726 #ifdef CONFIG_FS_POSIX_ACL
1728 struct lu_buf *acl_buf = &info->mti_buf;
1729 acl_buf->lb_buf = ma_acl->ma_acl;
1730 acl_buf->lb_len = ma_acl->ma_acl_size;
1732 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1734 mdd_write_unlock(env, son);
1737 ma->ma_attr.la_valid |= LA_MODE;
1742 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1743 son, ma, handle, spec);
1744 mdd_write_unlock(env, son);
1747 * Object has no links, so it will be destroyed when last
1748 * reference is released. (XXX not now.)
1754 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1755 name, S_ISDIR(attr->la_mode), handle,
1756 mdd_object_capa(env, mdd_pobj));
1763 /* No need mdd_lsm_sanity_check here */
1764 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1766 CERROR("error on stripe info copy %d \n", rc);
1769 if (lmm && lmm_size > 0) {
1770 /* Set Lov here, do not get lmm again later */
1771 memcpy(ma->ma_lmm, lmm, lmm_size);
1772 ma->ma_lmm_size = lmm_size;
1773 ma->ma_valid |= MA_LOV;
1776 if (S_ISLNK(attr->la_mode)) {
1777 struct md_ucred *uc = md_ucred(env);
1778 struct dt_object *dt = mdd_object_child(son);
1779 const char *target_name = spec->u.sp_symname;
1780 int sym_len = strlen(target_name);
1781 const struct lu_buf *buf;
1784 buf = mdd_buf_get_const(env, target_name, sym_len);
1785 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1786 mdd_object_capa(env, son),
1788 CFS_CAP_SYS_RESOURCE_MASK);
1793 GOTO(cleanup, rc = -EFAULT);
1797 la->la_valid = LA_CTIME | LA_MTIME;
1798 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1802 /* Return attr back. */
1803 rc = mdd_attr_get_internal_locked(env, son, ma);
1806 if (rc && created) {
1810 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1811 S_ISDIR(attr->la_mode),
1812 handle, BYPASS_CAPA);
1814 CERROR("error can not cleanup destroy %d\n",
1819 mdd_write_lock(env, son, MOR_TGT_CHILD);
1820 __mdd_ref_del(env, son, handle, 0);
1821 if (initialized && S_ISDIR(attr->la_mode))
1822 __mdd_ref_del(env, son, handle, 1);
1823 mdd_write_unlock(env, son);
1827 /* update lov_objid data, must be before transaction stop! */
1829 mdd_lov_objid_update(mdd, lmm);
1831 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1834 rc = mdd_changelog_ns_store(env, mdd,
1835 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1836 S_ISREG(attr->la_mode) ? CL_CREATE :
1837 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1838 son, mdd_pobj, NULL, lname, handle);
1839 mdd_trans_stop(env, mdd, rc, handle);
1841 /* finis lov_create stuff, free all temporary data */
1842 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1844 #ifdef HAVE_QUOTA_SUPPORT
1846 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1848 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1850 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1852 /* Trigger dqacq on the owner of child and parent. If failed,
1853 * the next call for lquota_chkquota will process it. */
1854 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1862 * Get locks on parents in proper order
1863 * RETURN: < 0 - error, rename_order if successful
1871 static int mdd_rename_order(const struct lu_env *env,
1872 struct mdd_device *mdd,
1873 struct mdd_object *src_pobj,
1874 struct mdd_object *tgt_pobj)
1876 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1880 if (src_pobj == tgt_pobj)
1881 RETURN(MDD_RN_SAME);
1883 /* compared the parent child relationship of src_p&tgt_p */
1884 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1886 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1889 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1902 /* has not mdd_write{read}_lock on any obj yet. */
1903 static int mdd_rename_sanity_check(const struct lu_env *env,
1904 struct mdd_object *src_pobj,
1905 struct mdd_object *tgt_pobj,
1906 struct mdd_object *sobj,
1907 struct mdd_object *tobj,
1913 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1916 /* XXX: when get here, sobj must NOT be NULL,
1917 * the other case has been processed in cml_rename
1918 * before mdd_rename and enable MDS_PERM_BYPASS. */
1921 if (mdd_is_dead_obj(sobj))
1924 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1928 /* XXX: when get here, "tobj == NULL" means tobj must
1929 * NOT exist (neither on remote MDS, such case has been
1930 * processed in cml_rename before mdd_rename and enable
1932 * So check may_create, but not check may_unlink. */
1934 rc = mdd_may_create(env, tgt_pobj, NULL,
1935 (src_pobj != tgt_pobj), 0);
1937 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1938 (src_pobj != tgt_pobj), 1);
1940 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1941 S_ISDIR(ma->ma_attr.la_mode))
1942 rc = __mdd_may_link(env, tgt_pobj);
1947 /* src object can be remote that is why we use only fid and type of object */
1948 static int mdd_rename(const struct lu_env *env,
1949 struct md_object *src_pobj, struct md_object *tgt_pobj,
1950 const struct lu_fid *lf, const struct lu_name *lsname,
1951 struct md_object *tobj, const struct lu_name *ltname,
1954 const char *sname = lsname->ln_name;
1955 const char *tname = ltname->ln_name;
1956 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1957 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1958 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1959 struct mdd_device *mdd = mdo2mdd(src_pobj);
1960 struct mdd_object *mdd_sobj = NULL; /* source object */
1961 struct mdd_object *mdd_tobj = NULL;
1962 struct dynlock_handle *sdlh, *tdlh;
1963 struct thandle *handle;
1964 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1965 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1969 #ifdef HAVE_QUOTA_SUPPORT
1970 struct obd_device *obd = mdd->mdd_obd_dev;
1971 struct mds_obd *mds = &obd->u.mds;
1972 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1973 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1974 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1975 int quota_copc = 0, quota_popc = 0;
1976 int rec_pending[MAXQUOTAS] = { 0, 0 };
1980 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1981 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1984 mdd_tobj = md2mdd_obj(tobj);
1986 #ifdef HAVE_QUOTA_SUPPORT
1987 if (mds->mds_quota) {
1988 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1990 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
1992 mdd_quota_wrapper(la_tmp, qspids);
1994 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
1998 mdd_data_get(env, mdd_tpobj, &data);
1999 quota_popc = FSFILT_OP_LINK;
2000 mdd_quota_wrapper(la_tmp, qtpids);
2001 /* get block quota for target parent */
2002 lquota_chkquota(mds_quota_interface_ref,
2004 rec_pending, 1, NULL,
2012 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2013 handle = mdd_trans_start(env, mdd);
2015 GOTO(out_pending, rc = PTR_ERR(handle));
2017 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2018 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2020 GOTO(cleanup_unlocked, rc);
2022 /* Get locks in determined order */
2023 if (rc == MDD_RN_SAME) {
2024 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2025 sname, MOR_SRC_PARENT);
2026 /* check hashes to determine do we need one lock or two */
2027 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2028 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2032 } else if (rc == MDD_RN_SRCTGT) {
2033 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2034 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2036 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2037 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2039 if (sdlh == NULL || tdlh == NULL)
2040 GOTO(cleanup, rc = -ENOMEM);
2042 mdd_sobj = mdd_object_find(env, mdd, lf);
2043 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2044 mdd_sobj, mdd_tobj, ma);
2048 /* Remove source name from source directory */
2049 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2050 mdd_object_capa(env, mdd_spobj));
2054 /* "mv dir1 dir2" needs "dir1/.." link update */
2055 if (is_dir && mdd_sobj) {
2056 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2057 mdd_object_capa(env, mdd_spobj));
2059 GOTO(fixup_spobj2, rc);
2061 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2062 handle, mdd_object_capa(env, mdd_tpobj));
2064 GOTO(fixup_spobj, rc);
2068 /* Remove target name from target directory
2069 * Here tobj can be remote one, so we do index_delete unconditionally
2070 * and -ENOENT is allowed.
2072 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2073 mdd_object_capa(env, mdd_tpobj));
2075 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2076 /* tname might been renamed to something else */
2077 GOTO(fixup_spobj, rc);
2080 GOTO(fixup_spobj, rc);
2083 /* Insert new fid with target name into target dir */
2084 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2085 mdd_object_capa(env, mdd_tpobj));
2087 GOTO(fixup_spobj, rc);
2089 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2090 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2092 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2094 la->la_valid = LA_CTIME;
2095 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2098 GOTO(fixup_tpobj, rc);
2101 /* Remove old target object
2102 * For tobj is remote case cmm layer has processed
2103 * and set tobj to NULL then. So when tobj is NOT NULL,
2104 * it must be local one.
2106 if (tobj && mdd_object_exists(mdd_tobj)) {
2107 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2108 if (mdd_is_dead_obj(mdd_tobj)) {
2109 mdd_write_unlock(env, mdd_tobj);
2110 /* shld not be dead, something is wrong */
2111 CERROR("tobj is dead, something is wrong\n");
2115 __mdd_ref_del(env, mdd_tobj, handle, 0);
2117 /* Remove dot reference. */
2119 __mdd_ref_del(env, mdd_tobj, handle, 1);
2121 la->la_valid = LA_CTIME;
2122 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2124 GOTO(fixup_tpobj, rc);
2126 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2127 mdd_write_unlock(env, mdd_tobj);
2129 GOTO(fixup_tpobj, rc);
2131 #ifdef HAVE_QUOTA_SUPPORT
2132 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2133 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2134 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2135 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2140 la->la_valid = LA_CTIME | LA_MTIME;
2141 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2143 GOTO(fixup_tpobj, rc);
2145 if (mdd_spobj != mdd_tpobj) {
2146 la->la_valid = LA_CTIME | LA_MTIME;
2147 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2151 if (rc == 0 && mdd_sobj) {
2152 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2153 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2154 mdo2fid(mdd_tpobj), ltname, handle);
2156 /* Old files might not have EA entry */
2157 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2159 mdd_write_unlock(env, mdd_sobj);
2160 /* We don't fail the transaction if the link ea can't be
2161 updated -- fid2path will use alternate lookup method. */
2169 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2170 mdd_object_capa(env, mdd_tpobj));
2172 CWARN("tp obj fix error %d\n",rc2);
2176 if (rc && is_dir && mdd_sobj) {
2177 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2178 mdd_object_capa(env, mdd_spobj));
2181 CWARN("sp obj dotdot delete error %d\n",rc2);
2184 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2186 mdd_object_capa(env, mdd_spobj));
2188 CWARN("sp obj dotdot insert error %d\n",rc2);
2193 rc2 = __mdd_index_insert(env, mdd_spobj,
2194 lf, sname, is_dir, handle,
2195 mdd_object_capa(env, mdd_tpobj));
2197 CWARN("sp obj fix error %d\n",rc2);
2200 if (likely(tdlh) && sdlh != tdlh)
2201 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2203 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2206 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2207 mdd_spobj, lf, lsname, handle);
2209 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2210 mdd_tpobj, lf, ltname, handle);
2212 mdd_trans_stop(env, mdd, rc, handle);
2214 mdd_object_put(env, mdd_sobj);
2216 #ifdef HAVE_QUOTA_SUPPORT
2217 if (mds->mds_quota) {
2219 lquota_pending_commit(mds_quota_interface_ref, obd,
2220 qtpids, rec_pending, 1);
2223 /* Trigger dqrel on the source owner of parent.
2224 * If failed, the next call for lquota_chkquota will
2226 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2227 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2229 /* Trigger dqrel on the target owner of child.
2230 * If failed, the next call for lquota_chkquota
2231 * will process it. */
2232 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2233 qtpids, rc, quota_copc);
2240 /** enable/disable storing of hardlink info */
2241 int mdd_linkea_enable = 1;
2242 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2243 "record hardlink info in EAs");
2245 /** Read the link EA into a temp buffer.
2246 * Uses the name_buf since it is generally large.
2247 * \retval IS_ERR err
2248 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2250 struct lu_buf *mdd_links_get(const struct lu_env *env,
2251 struct mdd_object *mdd_obj)
2254 struct lustre_capa *capa;
2255 struct link_ea_header *leh;
2258 /* First try a small buf */
2259 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2260 if (buf->lb_buf == NULL)
2261 return ERR_PTR(-ENOMEM);
2263 capa = mdd_object_capa(env, mdd_obj);
2264 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2265 if (rc == -ERANGE) {
2266 /* Buf was too small, figure out what we need. */
2269 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2272 buf = mdd_buf_alloc(env, rc);
2273 if (buf->lb_buf == NULL)
2274 return ERR_PTR(-ENOMEM);
2275 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2281 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2282 leh->leh_magic = LINK_EA_MAGIC;
2283 leh->leh_reccount = __swab32(leh->leh_reccount);
2284 leh->leh_len = __swab64(leh->leh_len);
2285 /* entries are swabbed by mdd_lee_unpack */
2287 if (leh->leh_magic != LINK_EA_MAGIC)
2288 return ERR_PTR(-EINVAL);
2289 if (leh->leh_reccount == 0)
2290 return ERR_PTR(-ENODATA);
2295 /** Pack a link_ea_entry.
2296 * All elements are stored as chars to avoid alignment issues.
2297 * Numbers are always big-endian
2298 * \param packbuf is a temp fid buffer
2299 * \retval record length
2301 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2302 const struct lu_fid *pfid, struct lu_fid* packbuf)
2307 fid_pack(&lee->lee_parent_fid, pfid, packbuf);
2308 ptr = (char *)&lee->lee_parent_fid + lee->lee_parent_fid.fp_len;
2309 strncpy(ptr, lname->ln_name, lname->ln_namelen);
2310 reclen = lee->lee_parent_fid.fp_len + lname->ln_namelen +
2311 sizeof(lee->lee_reclen);
2312 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2313 lee->lee_reclen[1] = reclen & 0xff;
2317 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2318 struct lu_name *lname, struct lu_fid *pfid)
2320 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2321 fid_unpack(&lee->lee_parent_fid, pfid);
2322 lname->ln_name = (char *)&lee->lee_parent_fid +
2323 lee->lee_parent_fid.fp_len;
2324 lname->ln_namelen = *reclen - lee->lee_parent_fid.fp_len -
2325 sizeof(lee->lee_reclen);
2328 /** Add a record to the end of link ea buf */
2329 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2330 const struct lu_fid *pfid,
2331 const struct lu_name *lname)
2333 struct link_ea_header *leh;
2334 struct link_ea_entry *lee;
2337 if (lname == NULL || pfid == NULL)
2340 /* Make sure our buf is big enough for the new one */
2342 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2343 if (leh->leh_len + reclen > buf->lb_len) {
2344 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2349 lee = buf->lb_buf + leh->leh_len;
2350 reclen = mdd_lee_pack(lee, lname, pfid, &mdd_env_info(env)->mti_fid2);
2351 leh->leh_len += reclen;
2352 leh->leh_reccount++;
2356 /* For pathologic linkers, we don't want to spend lots of time scanning the
2357 * link ea. Limit ourseleves to something reasonable; links not in the EA
2358 * can be looked up via (slower) parent lookup.
2360 #define LINKEA_MAX_COUNT 128
2362 static int mdd_links_add(const struct lu_env *env,
2363 struct mdd_object *mdd_obj,
2364 const struct lu_fid *pfid,
2365 const struct lu_name *lname,
2366 struct thandle *handle)
2369 struct link_ea_header *leh;
2373 if (!mdd_linkea_enable)
2376 buf = mdd_links_get(env, mdd_obj);
2379 if (rc != -ENODATA) {
2380 CERROR("link_ea read failed %d "DFID"\n", rc,
2381 PFID(mdd_object_fid(mdd_obj)));
2384 /* empty EA; start one */
2385 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2386 if (buf->lb_buf == NULL)
2389 leh->leh_magic = LINK_EA_MAGIC;
2390 leh->leh_len = sizeof(struct link_ea_header);
2391 leh->leh_reccount = 0;
2395 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2398 rc = __mdd_links_add(env, buf, pfid, lname);
2403 rc = __mdd_xattr_set(env, mdd_obj,
2404 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2405 XATTR_NAME_LINK, 0, handle);
2407 CERROR("link_ea add failed %d "DFID"\n", rc,
2408 PFID(mdd_object_fid(mdd_obj)));
2410 if (buf->lb_vmalloc)
2411 /* if we vmalloced a large buffer drop it */
2417 static int mdd_links_rename(const struct lu_env *env,
2418 struct mdd_object *mdd_obj,
2419 const struct lu_fid *oldpfid,
2420 const struct lu_name *oldlname,
2421 const struct lu_fid *newpfid,
2422 const struct lu_name *newlname,
2423 struct thandle *handle)
2426 struct link_ea_header *leh;
2427 struct link_ea_entry *lee;
2428 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2429 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2435 if (!mdd_linkea_enable)
2438 if (mdd_obj->mod_flags & DEAD_OBJ)
2439 /* No more links, don't bother */
2442 buf = mdd_links_get(env, mdd_obj);
2445 CERROR("link_ea read failed %d "DFID"\n",
2446 rc, PFID(mdd_object_fid(mdd_obj)));
2450 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2452 /* Find the old record */
2453 for(count = 0; count <= leh->leh_reccount; count++) {
2454 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2455 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2456 lu_fid_eq(tmpfid, oldpfid) &&
2457 (strncmp(tmpname->ln_name, oldlname->ln_name,
2458 tmpname->ln_namelen) == 0))
2460 lee = (struct link_ea_entry *)((char *)lee + reclen);
2462 if (count > leh->leh_reccount) {
2463 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2464 oldlname->ln_namelen, oldlname->ln_name);
2465 GOTO(out, rc = -ENOENT);
2468 /* Remove the old record */
2469 leh->leh_reccount--;
2470 leh->leh_len -= reclen;
2471 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2474 /* If renaming, add the new record */
2475 if (newpfid != NULL) {
2476 /* if the add fails, we still delete the out-of-date old link */
2477 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2481 rc = __mdd_xattr_set(env, mdd_obj,
2482 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2483 XATTR_NAME_LINK, 0, handle);
2489 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2490 oldlname->ln_namelen, oldlname->ln_name, rc,
2491 PFID(mdd_object_fid(mdd_obj)));
2493 if (buf->lb_vmalloc)
2494 /* if we vmalloced a large buffer drop it */
2500 const struct md_dir_operations mdd_dir_ops = {
2501 .mdo_is_subdir = mdd_is_subdir,
2502 .mdo_lookup = mdd_lookup,
2503 .mdo_create = mdd_create,
2504 .mdo_rename = mdd_rename,
2505 .mdo_link = mdd_link,
2506 .mdo_unlink = mdd_unlink,
2507 .mdo_name_insert = mdd_name_insert,
2508 .mdo_name_remove = mdd_name_remove,
2509 .mdo_rename_tgt = mdd_rename_tgt,
2510 .mdo_create_data = mdd_create_data