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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_dir.c
38 * Lustre Metadata Server (mdd) routines
40 * Author: Wang Di <wangdi@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
49 #ifdef HAVE_EXT4_LDISKFS
50 #include <ldiskfs/ldiskfs_jbd2.h>
52 #include <linux/jbd.h>
55 #include <obd_class.h>
56 #include <lustre_ver.h>
57 #include <obd_support.h>
58 #include <lprocfs_status.h>
59 #ifdef HAVE_EXT4_LDISKFS
60 #include <ldiskfs/ldiskfs.h>
62 #include <linux/ldiskfs_fs.h>
64 #include <lustre_mds.h>
65 #include <lustre/lustre_idl.h>
66 #include <lustre_fid.h>
68 #include "mdd_internal.h"
70 static const char dot[] = ".";
71 static const char dotdot[] = "..";
73 static struct lu_name lname_dotdot = {
78 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
79 const struct lu_name *lname, struct lu_fid* fid,
81 static int mdd_links_add(const struct lu_env *env,
82 struct mdd_object *mdd_obj,
83 const struct lu_fid *pfid,
84 const struct lu_name *lname,
85 struct thandle *handle, int first);
86 static int mdd_links_rename(const struct lu_env *env,
87 struct mdd_object *mdd_obj,
88 const struct lu_fid *oldpfid,
89 const struct lu_name *oldlname,
90 const struct lu_fid *newpfid,
91 const struct lu_name *newlname,
92 struct thandle *handle);
95 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
96 const struct lu_name *lname, struct lu_fid* fid, int mask)
98 const char *name = lname->ln_name;
99 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
100 struct dynlock_handle *dlh;
103 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
104 if (unlikely(dlh == NULL))
106 rc = __mdd_lookup(env, pobj, lname, fid, mask);
107 mdd_pdo_read_unlock(env, mdd_obj, dlh);
112 int mdd_lookup(const struct lu_env *env,
113 struct md_object *pobj, const struct lu_name *lname,
114 struct lu_fid* fid, struct md_op_spec *spec)
118 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
122 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
125 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
129 * For root fid use special function, which does not compare version component
130 * of fid. Version component is different for root fids on all MDTs.
132 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
134 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
135 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
139 * return 1: if lf is the fid of the ancestor of p1;
142 * return -EREMOTE: if remote object is found, in this
143 * case fid of remote object is saved to @pf;
145 * otherwise: values < 0, errors.
147 static int mdd_is_parent(const struct lu_env *env,
148 struct mdd_device *mdd,
149 struct mdd_object *p1,
150 const struct lu_fid *lf,
153 struct mdd_object *parent = NULL;
158 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
159 pfid = &mdd_env_info(env)->mti_fid;
161 /* Check for root first. */
162 if (mdd_is_root(mdd, mdo2fid(p1)))
166 /* this is done recursively, bypass capa for each obj */
167 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
168 rc = mdd_parent_fid(env, p1, pfid);
171 if (mdd_is_root(mdd, pfid))
173 if (lu_fid_eq(pfid, lf))
176 mdd_object_put(env, parent);
177 parent = mdd_object_find(env, mdd, pfid);
179 /* cross-ref parent */
180 if (parent == NULL) {
183 GOTO(out, rc = -EREMOTE);
184 } else if (IS_ERR(parent))
185 GOTO(out, rc = PTR_ERR(parent));
190 if (parent && !IS_ERR(parent))
191 mdd_object_put(env, parent);
196 * No permission check is needed.
198 * returns 1: if fid is ancestor of @mo;
199 * returns 0: if fid is not a ancestor of @mo;
201 * returns EREMOTE if remote object is found, fid of remote object is saved to
204 * returns < 0: if error
206 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
207 const struct lu_fid *fid, struct lu_fid *sfid)
209 struct mdd_device *mdd = mdo2mdd(mo);
213 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
216 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
220 } else if (rc == 1) {
221 /* found @fid is parent */
229 * Check that @dir contains no entries except (possibly) dot and dotdot.
234 * -ENOTDIR not a directory object
235 * -ENOTEMPTY not empty
239 static int mdd_dir_is_empty(const struct lu_env *env,
240 struct mdd_object *dir)
243 struct dt_object *obj;
244 const struct dt_it_ops *iops;
248 obj = mdd_object_child(dir);
249 if (!dt_try_as_dir(env, obj))
252 iops = &obj->do_index_ops->dio_it;
253 it = iops->init(env, obj, BYPASS_CAPA);
255 result = iops->get(env, it, (const void *)"");
258 for (result = 0, i = 0; result == 0 && i < 3; ++i)
259 result = iops->next(env, it);
262 else if (result == +1)
264 } else if (result == 0)
266 * Huh? Index contains no zero key?
277 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
279 struct mdd_device *m = mdd_obj2mdd_dev(obj);
280 struct lu_attr *la = &mdd_env_info(env)->mti_la;
284 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
289 * Subdir count limitation can be broken through.
291 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
292 !S_ISDIR(la->la_mode))
299 * Check whether it may create the cobj under the pobj.
302 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
303 struct mdd_object *cobj, int check_perm, int check_nlink)
308 if (cobj && mdd_object_exists(cobj))
311 if (mdd_is_dead_obj(pobj))
315 rc = mdd_permission_internal_locked(env, pobj, NULL,
316 MAY_WRITE | MAY_EXEC,
319 if (!rc && check_nlink)
320 rc = __mdd_may_link(env, pobj);
326 * Check whether can unlink from the pobj in the case of "cobj == NULL".
328 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
329 const struct md_attr *ma)
334 if (mdd_is_dead_obj(pobj))
337 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
338 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
341 rc = mdd_permission_internal_locked(env, pobj, NULL,
342 MAY_WRITE | MAY_EXEC,
347 if (mdd_is_append(pobj))
354 * pobj == NULL is remote ops case, under such case, pobj's
355 * VTX feature has been checked already, no need check again.
357 static inline int mdd_is_sticky(const struct lu_env *env,
358 struct mdd_object *pobj,
359 struct mdd_object *cobj)
361 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
362 struct md_ucred *uc = md_ucred(env);
366 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
370 if (!(tmp_la->la_mode & S_ISVTX) ||
371 (tmp_la->la_uid == uc->mu_fsuid))
375 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
379 if (tmp_la->la_uid == uc->mu_fsuid)
382 return !mdd_capable(uc, CFS_CAP_FOWNER);
386 * Check whether it may delete the cobj from the pobj.
389 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
390 struct mdd_object *cobj, struct md_attr *ma,
391 int check_perm, int check_empty)
397 if (!mdd_object_exists(cobj))
400 if (mdd_is_dead_obj(cobj))
404 if (!mdd_object_exists(pobj))
407 if (mdd_is_dead_obj(pobj))
411 rc = mdd_permission_internal_locked(env, pobj, NULL,
412 MAY_WRITE | MAY_EXEC,
418 if (mdd_is_append(pobj))
422 if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) &&
423 mdd_is_sticky(env, pobj, cobj))
426 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
429 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
430 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
433 if (S_ISDIR(ma->ma_attr.la_mode)) {
434 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
436 if (!S_ISDIR(mdd_object_type(cobj)))
439 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
441 } else if (S_ISDIR(mdd_object_type(cobj)))
444 if (S_ISDIR(ma->ma_attr.la_mode) && check_empty)
445 rc = mdd_dir_is_empty(env, cobj);
452 * has mdd_write_lock on src already, but not on tgt yet
454 int mdd_link_sanity_check(const struct lu_env *env,
455 struct mdd_object *tgt_obj,
456 const struct lu_name *lname,
457 struct mdd_object *src_obj)
459 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
463 if (!mdd_object_exists(src_obj))
466 if (mdd_is_dead_obj(src_obj))
469 /* Local ops, no lookup before link, check filename length here. */
470 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
471 RETURN(-ENAMETOOLONG);
473 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
476 if (S_ISDIR(mdd_object_type(src_obj)))
479 LASSERT(src_obj != tgt_obj);
481 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
486 rc = __mdd_may_link(env, src_obj);
492 * If subdir count is up to ddp_max_nlink, then enable MNLINK_OBJ flag and
493 * assign i_nlink to 1 which means the i_nlink for subdir count is incredible
494 * (maybe too large to be represented). It is a trick to break through the
495 * "i_nlink" limitation for subdir count.
497 void __mdd_ref_add(const struct lu_env *env, struct mdd_object *obj,
498 struct thandle *handle)
500 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
501 struct mdd_device *m = mdd_obj2mdd_dev(obj);
503 if (!mdd_is_mnlink(obj)) {
504 if (S_ISDIR(mdd_object_type(obj))) {
505 if (mdd_la_get(env, obj, tmp_la, BYPASS_CAPA))
508 if (tmp_la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) {
509 obj->mod_flags |= MNLINK_OBJ;
510 tmp_la->la_nlink = 1;
511 tmp_la->la_valid = LA_NLINK;
512 mdd_attr_set_internal(env, obj, tmp_la, handle,
517 mdo_ref_add(env, obj, handle);
521 void __mdd_ref_del(const struct lu_env *env, struct mdd_object *obj,
522 struct thandle *handle, int is_dot)
524 if (!mdd_is_mnlink(obj) || is_dot)
525 mdo_ref_del(env, obj, handle);
528 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
529 const char *name, struct thandle *handle,
530 struct lustre_capa *capa)
532 struct dt_object *next = mdd_object_child(pobj);
536 if (dt_try_as_dir(env, next)) {
537 rc = next->do_index_ops->dio_delete(env, next,
538 (struct dt_key *)name,
546 static int __mdd_index_insert_only(const struct lu_env *env,
547 struct mdd_object *pobj,
548 const struct lu_fid *lf, const char *name,
549 struct thandle *handle,
550 struct lustre_capa *capa)
552 struct dt_object *next = mdd_object_child(pobj);
556 if (dt_try_as_dir(env, next)) {
557 struct md_ucred *uc = md_ucred(env);
559 rc = next->do_index_ops->dio_insert(env, next,
561 (const struct dt_key *)name,
562 handle, capa, uc->mu_cap &
563 CFS_CAP_SYS_RESOURCE_MASK);
570 /* insert named index, add reference if isdir */
571 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
572 const struct lu_fid *lf, const char *name, int is_dir,
573 struct thandle *handle, struct lustre_capa *capa)
578 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
579 if (rc == 0 && is_dir) {
580 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
581 __mdd_ref_add(env, pobj, handle);
582 mdd_write_unlock(env, pobj);
587 /* delete named index, drop reference if isdir */
588 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
589 const char *name, int is_dir, struct thandle *handle,
590 struct lustre_capa *capa)
595 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
596 if (rc == 0 && is_dir) {
599 if (name != NULL && name[0] == '.' && name[1] == 0)
601 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
602 __mdd_ref_del(env, pobj, handle, is_dot);
603 mdd_write_unlock(env, pobj);
610 /** Store a namespace change changelog record
611 * If this fails, we must fail the whole transaction; we don't
612 * want the change to commit without the log entry.
613 * \param target - mdd_object of change
614 * \param parent - parent dir/object
615 * \param tf - target lu_fid, overrides fid of \a target if this is non-null
616 * \param tname - target name string
617 * \param handle - transacion handle
619 static int mdd_changelog_ns_store(const struct lu_env *env,
620 struct mdd_device *mdd,
621 enum changelog_rec_type type,
622 struct mdd_object *target,
623 struct mdd_object *parent,
624 const struct lu_fid *tf,
625 const struct lu_name *tname,
626 struct thandle *handle)
628 const struct lu_fid *tfid;
629 const struct lu_fid *tpfid = mdo2fid(parent);
630 struct llog_changelog_rec *rec;
637 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
639 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
642 LASSERT(parent != NULL);
643 LASSERT(tname != NULL);
644 LASSERT(handle != NULL);
647 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
648 buf = mdd_buf_alloc(env, reclen);
649 if (buf->lb_buf == NULL)
651 rec = (struct llog_changelog_rec *)buf->lb_buf;
653 rec->cr.cr_flags = CLF_VERSION;
654 rec->cr.cr_type = (__u32)type;
655 tfid = tf ? tf : mdo2fid(target);
656 rec->cr.cr_tfid = *tfid;
657 rec->cr.cr_pfid = *tpfid;
658 rec->cr.cr_namelen = tname->ln_namelen;
659 memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
661 target->mod_cltime = cfs_time_current_64();
663 rc = mdd_changelog_llog_write(mdd, rec, handle);
665 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
666 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
673 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
674 struct md_object *src_obj, const struct lu_name *lname,
677 const char *name = lname->ln_name;
678 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
679 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
680 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
681 struct mdd_device *mdd = mdo2mdd(src_obj);
682 struct dynlock_handle *dlh;
683 struct thandle *handle;
684 #ifdef HAVE_QUOTA_SUPPORT
685 struct obd_device *obd = mdd->mdd_obd_dev;
686 struct obd_export *exp = md_quota(env)->mq_exp;
687 struct mds_obd *mds = &obd->u.mds;
688 unsigned int qids[MAXQUOTAS] = { 0, 0 };
689 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
694 #ifdef HAVE_QUOTA_SUPPORT
695 if (mds->mds_quota) {
696 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
698 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
701 mdd_data_get(env, mdd_tobj, &data);
702 quota_opc = FSFILT_OP_LINK;
703 mdd_quota_wrapper(la_tmp, qids);
704 /* get block quota for parent */
705 lquota_chkquota(mds_quota_interface_ref, obd, exp,
706 qids, rec_pending, 1, NULL,
707 LQUOTA_FLAGS_BLK, data, 1);
712 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
713 handle = mdd_trans_start(env, mdd);
715 GOTO(out_pending, rc = PTR_ERR(handle));
717 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
719 GOTO(out_trans, rc = -ENOMEM);
720 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
722 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
724 GOTO(out_unlock, rc);
726 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
728 mdd_object_capa(env, mdd_tobj));
730 GOTO(out_unlock, rc);
732 __mdd_ref_add(env, mdd_sobj, handle);
734 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
735 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
737 la->la_valid = LA_CTIME | LA_MTIME;
738 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
740 GOTO(out_unlock, rc);
742 la->la_valid = LA_CTIME;
743 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
745 mdd_links_add(env, mdd_sobj,
746 mdo2fid(mdd_tobj), lname, handle, 0);
751 mdd_write_unlock(env, mdd_sobj);
752 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
755 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
756 mdd_tobj, NULL, lname, handle);
757 mdd_trans_stop(env, mdd, rc, handle);
759 #ifdef HAVE_QUOTA_SUPPORT
761 lquota_pending_commit(mds_quota_interface_ref, obd,
762 qids, rec_pending, 1);
763 /* Trigger dqacq for the parent owner. If failed,
764 * the next call for lquota_chkquota will process it. */
765 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
772 /* caller should take a lock before calling */
773 int mdd_finish_unlink(const struct lu_env *env,
774 struct mdd_object *obj, struct md_attr *ma,
781 LASSERT(mdd_write_locked(env, obj) != 0);
783 rc = mdd_iattr_get(env, obj, ma);
784 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
785 obj->mod_flags |= DEAD_OBJ;
786 /* add new orphan and the object
787 * will be deleted during mdd_close() */
788 if (obj->mod_count) {
789 rc = __mdd_orphan_add(env, obj, th);
791 CDEBUG(D_HA, "Object "DFID" is inserted into "
792 "orphan list, open count = %d\n",
793 PFID(mdd_object_fid(obj)),
796 CERROR("Object "DFID" fail to be an orphan, "
797 "open count = %d, maybe cause failed "
799 PFID(mdd_object_fid(obj)),
802 rc = mdd_object_kill(env, obj, ma);
809 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
816 * has mdd_write_lock on cobj already, but not on pobj yet
818 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
819 struct mdd_object *cobj, struct md_attr *ma)
824 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
829 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
830 struct md_object *cobj, const struct lu_name *lname,
833 const char *name = lname->ln_name;
834 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
835 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
836 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
837 struct mdd_device *mdd = mdo2mdd(pobj);
838 struct dynlock_handle *dlh;
839 struct thandle *handle;
840 #ifdef HAVE_QUOTA_SUPPORT
841 struct obd_device *obd = mdd->mdd_obd_dev;
842 struct mds_obd *mds = &obd->u.mds;
843 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
844 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
847 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
851 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
852 PFID(mdd_object_fid(mdd_cobj)));
854 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
858 handle = mdd_trans_start(env, mdd);
860 RETURN(PTR_ERR(handle));
862 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
864 GOTO(out_trans, rc = -ENOMEM);
865 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
867 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
871 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
872 mdd_object_capa(env, mdd_pobj));
876 __mdd_ref_del(env, mdd_cobj, handle, 0);
879 __mdd_ref_del(env, mdd_cobj, handle, 1);
881 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
882 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
884 la->la_valid = LA_CTIME | LA_MTIME;
885 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
889 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
890 /* update ctime of an unlinked file only if it is still
891 * opened or a link still exists */
892 la->la_valid = LA_CTIME;
893 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
898 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
899 #ifdef HAVE_QUOTA_SUPPORT
900 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
901 ma->ma_attr.la_nlink == 0) {
902 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
904 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
906 mdd_quota_wrapper(la_tmp, qpids);
907 if (mdd_cobj->mod_count == 0) {
908 quota_opc = FSFILT_OP_UNLINK;
909 mdd_quota_wrapper(&ma->ma_attr, qcids);
911 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
917 /* old files may not have link ea; ignore errors */
918 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
919 lname, NULL, NULL, handle);
923 mdd_write_unlock(env, mdd_cobj);
924 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
927 rc = mdd_changelog_ns_store(env, mdd,
928 is_dir ? CL_RMDIR : CL_UNLINK,
929 mdd_cobj, mdd_pobj, NULL, lname,
932 mdd_trans_stop(env, mdd, rc, handle);
933 #ifdef HAVE_QUOTA_SUPPORT
935 /* Trigger dqrel on the owner of child and parent. If failed,
936 * the next call for lquota_chkquota will process it. */
937 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
943 /* has not lock on pobj yet */
944 static int mdd_ni_sanity_check(const struct lu_env *env,
945 struct md_object *pobj,
946 const struct md_attr *ma)
948 struct mdd_object *obj = md2mdd_obj(pobj);
952 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
955 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
963 static int mdd_name_insert(const struct lu_env *env,
964 struct md_object *pobj,
965 const struct lu_name *lname,
966 const struct lu_fid *fid,
967 const struct md_attr *ma)
969 const char *name = lname->ln_name;
970 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
971 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
972 struct mdd_device *mdd = mdo2mdd(pobj);
973 struct dynlock_handle *dlh;
974 struct thandle *handle;
975 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
976 #ifdef HAVE_QUOTA_SUPPORT
977 struct md_ucred *uc = md_ucred(env);
978 struct obd_device *obd = mdd->mdd_obd_dev;
979 struct obd_export *exp = md_quota(env)->mq_exp;
980 struct mds_obd *mds = &obd->u.mds;
981 unsigned int qids[MAXQUOTAS] = { 0, 0 };
982 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
983 cfs_cap_t save = uc->mu_cap;
988 #ifdef HAVE_QUOTA_SUPPORT
989 if (mds->mds_quota) {
990 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
991 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
993 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
996 mdd_data_get(env, mdd_obj, &data);
997 quota_opc = FSFILT_OP_LINK;
998 mdd_quota_wrapper(la_tmp, qids);
999 /* get block quota for parent */
1000 lquota_chkquota(mds_quota_interface_ref, obd,
1001 exp, qids, rec_pending, 1, NULL,
1002 LQUOTA_FLAGS_BLK, data, 1);
1005 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1009 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
1010 handle = mdd_trans_start(env, mdo2mdd(pobj));
1012 GOTO(out_pending, rc = PTR_ERR(handle));
1014 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1016 GOTO(out_trans, rc = -ENOMEM);
1018 rc = mdd_ni_sanity_check(env, pobj, ma);
1020 GOTO(out_unlock, rc);
1022 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1023 handle, BYPASS_CAPA);
1025 GOTO(out_unlock, rc);
1028 * For some case, no need update obj's ctime (LA_CTIME is not set),
1030 * For other cases, update obj's ctime (LA_CTIME is set),
1033 if (ma->ma_attr.la_valid & LA_CTIME) {
1034 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1035 la->la_valid = LA_CTIME | LA_MTIME;
1036 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1041 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1043 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1045 #ifdef HAVE_QUOTA_SUPPORT
1046 if (mds->mds_quota) {
1048 lquota_pending_commit(mds_quota_interface_ref,
1049 obd, qids, rec_pending, 1);
1050 /* Trigger dqacq for the parent owner. If failed,
1051 * the next call for lquota_chkquota will process it*/
1052 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1062 /* has not lock on pobj yet */
1063 static int mdd_nr_sanity_check(const struct lu_env *env,
1064 struct md_object *pobj,
1065 const struct md_attr *ma)
1067 struct mdd_object *obj = md2mdd_obj(pobj);
1071 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1074 rc = mdd_may_unlink(env, obj, ma);
1080 * Partial operation.
1082 static int mdd_name_remove(const struct lu_env *env,
1083 struct md_object *pobj,
1084 const struct lu_name *lname,
1085 const struct md_attr *ma)
1087 const char *name = lname->ln_name;
1088 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1089 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1090 struct mdd_device *mdd = mdo2mdd(pobj);
1091 struct dynlock_handle *dlh;
1092 struct thandle *handle;
1093 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1094 #ifdef HAVE_QUOTA_SUPPORT
1095 struct obd_device *obd = mdd->mdd_obd_dev;
1096 struct mds_obd *mds = &obd->u.mds;
1097 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1103 #ifdef HAVE_QUOTA_SUPPORT
1104 if (mds->mds_quota) {
1105 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1107 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1109 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1110 mdd_quota_wrapper(la_tmp, qids);
1114 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1115 handle = mdd_trans_start(env, mdd);
1117 GOTO(out_pending, rc = PTR_ERR(handle));
1119 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1121 GOTO(out_trans, rc = -ENOMEM);
1123 rc = mdd_nr_sanity_check(env, pobj, ma);
1125 GOTO(out_unlock, rc);
1127 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1128 handle, BYPASS_CAPA);
1130 GOTO(out_unlock, rc);
1133 * For some case, no need update obj's ctime (LA_CTIME is not set),
1135 * For other cases, update obj's ctime (LA_CTIME is set),
1138 if (ma->ma_attr.la_valid & LA_CTIME) {
1139 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1140 la->la_valid = LA_CTIME | LA_MTIME;
1141 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1146 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1148 mdd_trans_stop(env, mdd, rc, handle);
1150 #ifdef HAVE_QUOTA_SUPPORT
1151 /* Trigger dqrel for the parent owner.
1152 * If failed, the next call for lquota_chkquota will process it. */
1154 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1162 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1164 static int mdd_rt_sanity_check(const struct lu_env *env,
1165 struct mdd_object *tgt_pobj,
1166 struct mdd_object *tobj,
1172 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1175 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1176 * exist. In fact, tobj must exist, otherwise the call trace will be:
1177 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1178 * When get here, tobj must be NOT NULL, the other case has been
1179 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1181 * So check may_delete, but not check nlink of tgt_pobj. */
1183 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1188 /* Partial rename op on slave MDD */
1189 static int mdd_rename_tgt(const struct lu_env *env,
1190 struct md_object *pobj, struct md_object *tobj,
1191 const struct lu_fid *lf, const struct lu_name *lname,
1194 const char *name = lname->ln_name;
1195 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1196 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1197 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1198 struct mdd_device *mdd = mdo2mdd(pobj);
1199 struct dynlock_handle *dlh;
1200 struct thandle *handle;
1201 #ifdef HAVE_QUOTA_SUPPORT
1202 struct obd_device *obd = mdd->mdd_obd_dev;
1203 struct obd_export *exp = md_quota(env)->mq_exp;
1204 struct mds_obd *mds = &obd->u.mds;
1205 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1206 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1207 int quota_copc = 0, quota_popc = 0;
1208 int rec_pending[MAXQUOTAS] = { 0, 0 };
1213 #ifdef HAVE_QUOTA_SUPPORT
1214 if (mds->mds_quota && !tobj) {
1215 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1217 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1220 mdd_data_get(env, mdd_tpobj, &data);
1221 quota_popc = FSFILT_OP_LINK;
1222 mdd_quota_wrapper(la_tmp, qpids);
1223 /* get block quota for target parent */
1224 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1225 qpids, rec_pending, 1, NULL,
1226 LQUOTA_FLAGS_BLK, data, 1);
1230 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1231 handle = mdd_trans_start(env, mdd);
1233 GOTO(out_pending, rc = PTR_ERR(handle));
1235 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1237 GOTO(out_trans, rc = -ENOMEM);
1239 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1241 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1246 * If rename_tgt is called then we should just re-insert name with
1247 * correct fid, no need to dec/inc parent nlink if obj is dir.
1249 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1253 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1258 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1259 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1261 la->la_valid = LA_CTIME | LA_MTIME;
1262 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1267 * For tobj is remote case cmm layer has processed
1268 * and pass NULL tobj to here. So when tobj is NOT NULL,
1269 * it must be local one.
1271 if (tobj && mdd_object_exists(mdd_tobj)) {
1272 __mdd_ref_del(env, mdd_tobj, handle, 0);
1274 /* Remove dot reference. */
1275 if (S_ISDIR(ma->ma_attr.la_mode))
1276 __mdd_ref_del(env, mdd_tobj, handle, 1);
1278 la->la_valid = LA_CTIME;
1279 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1283 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1287 #ifdef HAVE_QUOTA_SUPPORT
1288 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1289 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1290 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1291 mdd_quota_wrapper(&ma->ma_attr, qcids);
1298 mdd_write_unlock(env, mdd_tobj);
1299 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1302 /* Bare EXT record with no RENAME in front of it signifies
1303 a partial slave op */
1304 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1305 mdd_tpobj, NULL, lname, handle);
1307 mdd_trans_stop(env, mdd, rc, handle);
1309 #ifdef HAVE_QUOTA_SUPPORT
1310 if (mds->mds_quota) {
1312 lquota_pending_commit(mds_quota_interface_ref, obd,
1313 qpids, rec_pending, 1);
1316 /* Trigger dqrel on the target owner of child.
1317 * If failed, the next call for lquota_chkquota
1318 * will process it. */
1319 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1327 * The permission has been checked when obj created, no need check again.
1329 static int mdd_cd_sanity_check(const struct lu_env *env,
1330 struct mdd_object *obj)
1335 if (!obj || mdd_is_dead_obj(obj))
1342 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1343 struct md_object *cobj, const struct md_op_spec *spec,
1346 struct mdd_device *mdd = mdo2mdd(cobj);
1347 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1348 struct mdd_object *son = md2mdd_obj(cobj);
1349 struct lu_attr *attr = &ma->ma_attr;
1350 struct lov_mds_md *lmm = NULL;
1352 struct thandle *handle;
1356 rc = mdd_cd_sanity_check(env, son);
1360 if (!md_should_create(spec->sp_cr_flags))
1362 lmm_size = ma->ma_lmm_size;
1363 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1368 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1369 handle = mdd_trans_start(env, mdd);
1371 GOTO(out_free, rc = PTR_ERR(handle));
1374 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1375 * Should this be fixed?
1378 /* Replay creates has objects already */
1380 if (spec->no_create) {
1381 CDEBUG(D_INFO, "we already have lov ea\n");
1382 rc = mdd_lov_set_md(env, mdd_pobj, son,
1383 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1384 spec->u.sp_ea.eadatalen, handle, 0);
1387 /* No need mdd_lsm_sanity_check here */
1388 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1389 lmm_size, handle, 0);
1392 rc = mdd_attr_get_internal_locked(env, son, ma);
1394 /* update lov_objid data, must be before transaction stop! */
1396 mdd_lov_objid_update(mdd, lmm);
1398 mdd_trans_stop(env, mdd, rc, handle);
1400 /* Finish mdd_lov_create() stuff. */
1401 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1405 /* Get fid from name and parent */
1407 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1408 const struct lu_name *lname, struct lu_fid* fid, int mask)
1410 const char *name = lname->ln_name;
1411 const struct dt_key *key = (const struct dt_key *)name;
1412 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1413 struct mdd_device *m = mdo2mdd(pobj);
1414 struct dt_object *dir = mdd_object_child(mdd_obj);
1418 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1421 rc = mdd_object_exists(mdd_obj);
1422 if (unlikely(rc == 0))
1424 else if (unlikely(rc < 0)) {
1425 CERROR("Object "DFID" locates on remote server\n",
1426 PFID(mdo2fid(mdd_obj)));
1430 /* The common filename length check. */
1431 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1432 RETURN(-ENAMETOOLONG);
1434 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1439 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1440 dt_try_as_dir(env, dir))) {
1442 rc = dir->do_index_ops->dio_lookup(env, dir,
1443 (struct dt_rec *)fid, key,
1444 mdd_object_capa(env, mdd_obj));
1455 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1456 const struct lu_name *lname, struct mdd_object *child,
1457 struct md_attr *ma, struct thandle *handle,
1458 const struct md_op_spec *spec)
1464 * Update attributes for child.
1467 * (1) the valid bits should be converted between Lustre and Linux;
1468 * (2) maybe, the child attributes should be set in OSD when creation.
1471 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1475 if (S_ISDIR(ma->ma_attr.la_mode)) {
1476 /* Add "." and ".." for newly created dir */
1477 __mdd_ref_add(env, child, handle);
1478 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1479 dot, handle, BYPASS_CAPA);
1481 rc = __mdd_index_insert_only(env, child, pfid,
1485 __mdd_ref_del(env, child, handle, 1);
1488 mdd_links_add(env, child, pfid, lname, handle, 1);
1493 /* has not lock on pobj yet */
1494 static int mdd_create_sanity_check(const struct lu_env *env,
1495 struct md_object *pobj,
1496 const struct lu_name *lname,
1498 struct md_op_spec *spec)
1500 struct mdd_thread_info *info = mdd_env_info(env);
1501 struct lu_attr *la = &info->mti_la;
1502 struct lu_fid *fid = &info->mti_fid;
1503 struct mdd_object *obj = md2mdd_obj(pobj);
1504 struct mdd_device *m = mdo2mdd(pobj);
1505 int lookup = spec->sp_cr_lookup;
1510 if (mdd_is_dead_obj(obj))
1514 * In some cases this lookup is not needed - we know before if name
1515 * exists or not because MDT performs lookup for it.
1516 * name length check is done in lookup.
1520 * Check if the name already exist, though it will be checked in
1521 * _index_insert also, for avoiding rolling back if exists
1524 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1525 MAY_WRITE | MAY_EXEC);
1527 RETURN(rc ? : -EEXIST);
1530 * Check WRITE permission for the parent.
1531 * EXEC permission have been checked
1532 * when lookup before create already.
1534 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1541 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1545 if (la->la_mode & S_ISGID) {
1546 ma->ma_attr.la_gid = la->la_gid;
1547 if (S_ISDIR(ma->ma_attr.la_mode)) {
1548 ma->ma_attr.la_mode |= S_ISGID;
1549 ma->ma_attr.la_valid |= LA_MODE;
1553 switch (ma->ma_attr.la_mode & S_IFMT) {
1555 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1557 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1558 RETURN(-ENAMETOOLONG);
1578 * Create object and insert it into namespace.
1580 static int mdd_create(const struct lu_env *env,
1581 struct md_object *pobj,
1582 const struct lu_name *lname,
1583 struct md_object *child,
1584 struct md_op_spec *spec,
1587 struct mdd_thread_info *info = mdd_env_info(env);
1588 struct lu_attr *la = &info->mti_la_for_fix;
1589 struct md_attr *ma_acl = &info->mti_ma;
1590 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1591 struct mdd_object *son = md2mdd_obj(child);
1592 struct mdd_device *mdd = mdo2mdd(pobj);
1593 struct lu_attr *attr = &ma->ma_attr;
1594 struct lov_mds_md *lmm = NULL;
1595 struct thandle *handle;
1596 struct dynlock_handle *dlh;
1597 const char *name = lname->ln_name;
1598 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1599 int got_def_acl = 0;
1600 #ifdef HAVE_QUOTA_SUPPORT
1601 struct obd_device *obd = mdd->mdd_obd_dev;
1602 struct obd_export *exp = md_quota(env)->mq_exp;
1603 struct mds_obd *mds = &obd->u.mds;
1604 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1605 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1606 int quota_opc = 0, block_count = 0;
1607 int inode_pending[MAXQUOTAS] = { 0, 0 };
1608 int block_pending[MAXQUOTAS] = { 0, 0 };
1609 int parent_pending[MAXQUOTAS] = { 0, 0 };
1614 * Two operations have to be performed:
1616 * - an allocation of a new object (->do_create()), and
1618 * - an insertion into a parent index (->dio_insert()).
1620 * Due to locking, operation order is not important, when both are
1621 * successful, *but* error handling cases are quite different:
1623 * - if insertion is done first, and following object creation fails,
1624 * insertion has to be rolled back, but this operation might fail
1625 * also leaving us with dangling index entry.
1627 * - if creation is done first, is has to be undone if insertion
1628 * fails, leaving us with leaked space, which is neither good, nor
1631 * It seems that creation-first is simplest solution, but it is
1632 * sub-optimal in the frequent
1637 * case, because second mkdir is bound to create object, only to
1638 * destroy it immediately.
1640 * To avoid this follow local file systems that do double lookup:
1642 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1644 * 1. create (mdd_object_create_internal())
1646 * 2. insert (__mdd_index_insert(), lookup again)
1649 /* Sanity checks before big job. */
1650 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1654 #ifdef HAVE_QUOTA_SUPPORT
1655 if (mds->mds_quota) {
1656 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1658 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1662 quota_opc = FSFILT_OP_CREATE;
1663 mdd_quota_wrapper(&ma->ma_attr, qcids);
1664 mdd_quota_wrapper(la_tmp, qpids);
1665 /* get file quota for child */
1666 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1667 qcids, inode_pending, 1, NULL, 0, NULL,
1669 switch (ma->ma_attr.la_mode & S_IFMT) {
1678 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1679 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1683 /* get block quota for child and parent */
1685 lquota_chkquota(mds_quota_interface_ref, obd,
1686 exp, qcids, block_pending,
1688 LQUOTA_FLAGS_BLK, NULL, 0);
1690 lquota_chkquota(mds_quota_interface_ref, obd,
1691 exp, qpids, parent_pending, 1,
1692 NULL, LQUOTA_FLAGS_BLK, NULL,
1699 * No RPC inside the transaction, so OST objects should be created at
1702 if (S_ISREG(attr->la_mode)) {
1703 lmm_size = ma->ma_lmm_size;
1704 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1707 GOTO(out_pending, rc);
1710 if (!S_ISLNK(attr->la_mode)) {
1711 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1712 ma_acl->ma_acl = info->mti_xattr_buf;
1713 ma_acl->ma_need = MA_ACL_DEF;
1714 ma_acl->ma_valid = 0;
1716 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1717 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1718 mdd_read_unlock(env, mdd_pobj);
1721 else if (ma_acl->ma_valid & MA_ACL_DEF)
1725 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1726 handle = mdd_trans_start(env, mdd);
1728 GOTO(out_free, rc = PTR_ERR(handle));
1730 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1732 GOTO(out_trans, rc = -ENOMEM);
1734 mdd_write_lock(env, son, MOR_TGT_CHILD);
1735 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1737 mdd_write_unlock(env, son);
1743 #ifdef CONFIG_FS_POSIX_ACL
1745 struct lu_buf *acl_buf = &info->mti_buf;
1746 acl_buf->lb_buf = ma_acl->ma_acl;
1747 acl_buf->lb_len = ma_acl->ma_acl_size;
1749 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1751 mdd_write_unlock(env, son);
1754 ma->ma_attr.la_valid |= LA_MODE;
1759 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1760 son, ma, handle, spec);
1761 mdd_write_unlock(env, son);
1764 * Object has no links, so it will be destroyed when last
1765 * reference is released. (XXX not now.)
1771 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1772 name, S_ISDIR(attr->la_mode), handle,
1773 mdd_object_capa(env, mdd_pobj));
1780 /* No need mdd_lsm_sanity_check here */
1781 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1783 CERROR("error on stripe info copy %d \n", rc);
1786 if (lmm && lmm_size > 0) {
1787 /* Set Lov here, do not get lmm again later */
1788 memcpy(ma->ma_lmm, lmm, lmm_size);
1789 ma->ma_lmm_size = lmm_size;
1790 ma->ma_valid |= MA_LOV;
1793 if (S_ISLNK(attr->la_mode)) {
1794 struct md_ucred *uc = md_ucred(env);
1795 struct dt_object *dt = mdd_object_child(son);
1796 const char *target_name = spec->u.sp_symname;
1797 int sym_len = strlen(target_name);
1798 const struct lu_buf *buf;
1801 buf = mdd_buf_get_const(env, target_name, sym_len);
1802 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1803 mdd_object_capa(env, son),
1805 CFS_CAP_SYS_RESOURCE_MASK);
1810 GOTO(cleanup, rc = -EFAULT);
1814 la->la_valid = LA_CTIME | LA_MTIME;
1815 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1819 /* Return attr back. */
1820 rc = mdd_attr_get_internal_locked(env, son, ma);
1823 if (rc && created) {
1827 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1828 S_ISDIR(attr->la_mode),
1829 handle, BYPASS_CAPA);
1831 CERROR("error can not cleanup destroy %d\n",
1836 mdd_write_lock(env, son, MOR_TGT_CHILD);
1837 __mdd_ref_del(env, son, handle, 0);
1838 if (initialized && S_ISDIR(attr->la_mode))
1839 __mdd_ref_del(env, son, handle, 1);
1840 mdd_write_unlock(env, son);
1844 /* update lov_objid data, must be before transaction stop! */
1846 mdd_lov_objid_update(mdd, lmm);
1848 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1851 rc = mdd_changelog_ns_store(env, mdd,
1852 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1853 S_ISREG(attr->la_mode) ? CL_CREATE :
1854 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1855 son, mdd_pobj, NULL, lname, handle);
1856 mdd_trans_stop(env, mdd, rc, handle);
1858 /* finis lov_create stuff, free all temporary data */
1859 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1861 #ifdef HAVE_QUOTA_SUPPORT
1863 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1865 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1867 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1869 /* Trigger dqacq on the owner of child and parent. If failed,
1870 * the next call for lquota_chkquota will process it. */
1871 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1879 * Get locks on parents in proper order
1880 * RETURN: < 0 - error, rename_order if successful
1888 static int mdd_rename_order(const struct lu_env *env,
1889 struct mdd_device *mdd,
1890 struct mdd_object *src_pobj,
1891 struct mdd_object *tgt_pobj)
1893 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1897 if (src_pobj == tgt_pobj)
1898 RETURN(MDD_RN_SAME);
1900 /* compared the parent child relationship of src_p&tgt_p */
1901 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1903 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1906 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1919 /* has not mdd_write{read}_lock on any obj yet. */
1920 static int mdd_rename_sanity_check(const struct lu_env *env,
1921 struct mdd_object *src_pobj,
1922 struct mdd_object *tgt_pobj,
1923 struct mdd_object *sobj,
1924 struct mdd_object *tobj,
1930 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1933 /* XXX: when get here, sobj must NOT be NULL,
1934 * the other case has been processed in cml_rename
1935 * before mdd_rename and enable MDS_PERM_BYPASS. */
1938 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1942 /* XXX: when get here, "tobj == NULL" means tobj must
1943 * NOT exist (neither on remote MDS, such case has been
1944 * processed in cml_rename before mdd_rename and enable
1946 * So check may_create, but not check may_unlink. */
1948 rc = mdd_may_create(env, tgt_pobj, NULL,
1949 (src_pobj != tgt_pobj), 0);
1951 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1952 (src_pobj != tgt_pobj), 1);
1954 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1955 S_ISDIR(ma->ma_attr.la_mode))
1956 rc = __mdd_may_link(env, tgt_pobj);
1961 /* src object can be remote that is why we use only fid and type of object */
1962 static int mdd_rename(const struct lu_env *env,
1963 struct md_object *src_pobj, struct md_object *tgt_pobj,
1964 const struct lu_fid *lf, const struct lu_name *lsname,
1965 struct md_object *tobj, const struct lu_name *ltname,
1968 const char *sname = lsname->ln_name;
1969 const char *tname = ltname->ln_name;
1970 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1971 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1972 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1973 struct mdd_device *mdd = mdo2mdd(src_pobj);
1974 struct mdd_object *mdd_sobj = NULL; /* source object */
1975 struct mdd_object *mdd_tobj = NULL;
1976 struct dynlock_handle *sdlh, *tdlh;
1977 struct thandle *handle;
1978 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1979 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1983 #ifdef HAVE_QUOTA_SUPPORT
1984 struct obd_device *obd = mdd->mdd_obd_dev;
1985 struct obd_export *exp = md_quota(env)->mq_exp;
1986 struct mds_obd *mds = &obd->u.mds;
1987 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1988 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1989 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1990 int quota_copc = 0, quota_popc = 0;
1991 int rec_pending[MAXQUOTAS] = { 0, 0 };
1995 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1996 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1999 mdd_tobj = md2mdd_obj(tobj);
2001 #ifdef HAVE_QUOTA_SUPPORT
2002 if (mds->mds_quota) {
2003 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2005 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2007 mdd_quota_wrapper(la_tmp, qspids);
2009 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2013 mdd_data_get(env, mdd_tpobj, &data);
2014 quota_popc = FSFILT_OP_LINK;
2015 mdd_quota_wrapper(la_tmp, qtpids);
2016 /* get block quota for target parent */
2017 lquota_chkquota(mds_quota_interface_ref,
2019 rec_pending, 1, NULL,
2027 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2028 handle = mdd_trans_start(env, mdd);
2030 GOTO(out_pending, rc = PTR_ERR(handle));
2032 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2033 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2035 GOTO(cleanup_unlocked, rc);
2037 /* Get locks in determined order */
2038 if (rc == MDD_RN_SAME) {
2039 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2040 sname, MOR_SRC_PARENT);
2041 /* check hashes to determine do we need one lock or two */
2042 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2043 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2047 } else if (rc == MDD_RN_SRCTGT) {
2048 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2049 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2051 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2052 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2054 if (sdlh == NULL || tdlh == NULL)
2055 GOTO(cleanup, rc = -ENOMEM);
2057 mdd_sobj = mdd_object_find(env, mdd, lf);
2058 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2059 mdd_sobj, mdd_tobj, ma);
2063 /* Remove source name from source directory */
2064 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2065 mdd_object_capa(env, mdd_spobj));
2069 /* "mv dir1 dir2" needs "dir1/.." link update */
2070 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2071 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2072 mdd_object_capa(env, mdd_sobj));
2074 GOTO(fixup_spobj2, rc);
2076 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2077 handle, mdd_object_capa(env, mdd_sobj));
2079 GOTO(fixup_spobj, rc);
2082 /* Remove target name from target directory
2083 * Here tobj can be remote one, so we do index_delete unconditionally
2084 * and -ENOENT is allowed.
2086 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2087 mdd_object_capa(env, mdd_tpobj));
2090 /* tname might been renamed to something else */
2091 GOTO(fixup_spobj, rc);
2094 GOTO(fixup_spobj, rc);
2097 /* Insert new fid with target name into target dir */
2098 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2099 mdd_object_capa(env, mdd_tpobj));
2101 GOTO(fixup_tpobj, rc);
2103 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2104 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2106 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2108 la->la_valid = LA_CTIME;
2109 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2112 GOTO(fixup_tpobj, rc);
2115 /* Remove old target object
2116 * For tobj is remote case cmm layer has processed
2117 * and set tobj to NULL then. So when tobj is NOT NULL,
2118 * it must be local one.
2120 if (tobj && mdd_object_exists(mdd_tobj)) {
2121 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2122 if (mdd_is_dead_obj(mdd_tobj)) {
2123 mdd_write_unlock(env, mdd_tobj);
2124 /* shld not be dead, something is wrong */
2125 CERROR("tobj is dead, something is wrong\n");
2129 __mdd_ref_del(env, mdd_tobj, handle, 0);
2131 /* Remove dot reference. */
2133 __mdd_ref_del(env, mdd_tobj, handle, 1);
2135 la->la_valid = LA_CTIME;
2136 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2138 GOTO(fixup_tpobj, rc);
2140 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2141 mdd_write_unlock(env, mdd_tobj);
2143 GOTO(fixup_tpobj, rc);
2145 #ifdef HAVE_QUOTA_SUPPORT
2146 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2147 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2148 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2149 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2154 la->la_valid = LA_CTIME | LA_MTIME;
2155 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2157 GOTO(fixup_tpobj, rc);
2159 if (mdd_spobj != mdd_tpobj) {
2160 la->la_valid = LA_CTIME | LA_MTIME;
2161 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2165 if (rc == 0 && mdd_sobj) {
2166 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2167 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2168 mdo2fid(mdd_tpobj), ltname, handle);
2170 /* Old files might not have EA entry */
2171 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2173 mdd_write_unlock(env, mdd_sobj);
2174 /* We don't fail the transaction if the link ea can't be
2175 updated -- fid2path will use alternate lookup method. */
2183 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2186 CWARN("tp obj fix error %d\n",rc2);
2188 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2189 !mdd_is_dead_obj(mdd_tobj)) {
2190 rc2 = __mdd_index_insert(env, mdd_tpobj,
2191 mdo2fid(mdd_tobj), tname,
2196 CWARN("tp obj fix error %d\n",rc2);
2201 if (rc && is_dir && mdd_sobj) {
2202 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2206 CWARN("sp obj dotdot delete error %d\n",rc2);
2209 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2210 dotdot, handle, BYPASS_CAPA);
2212 CWARN("sp obj dotdot insert error %d\n",rc2);
2217 rc2 = __mdd_index_insert(env, mdd_spobj,
2218 lf, sname, is_dir, handle, BYPASS_CAPA);
2220 CWARN("sp obj fix error %d\n",rc2);
2223 if (likely(tdlh) && sdlh != tdlh)
2224 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2226 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2229 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2230 mdd_spobj, lf, lsname, handle);
2232 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2233 mdd_tpobj, lf, ltname, handle);
2235 mdd_trans_stop(env, mdd, rc, handle);
2237 mdd_object_put(env, mdd_sobj);
2239 #ifdef HAVE_QUOTA_SUPPORT
2240 if (mds->mds_quota) {
2242 lquota_pending_commit(mds_quota_interface_ref, obd,
2243 qtpids, rec_pending, 1);
2246 /* Trigger dqrel on the source owner of parent.
2247 * If failed, the next call for lquota_chkquota will
2249 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2250 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2252 /* Trigger dqrel on the target owner of child.
2253 * If failed, the next call for lquota_chkquota
2254 * will process it. */
2255 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2256 qtpids, rc, quota_copc);
2263 /** enable/disable storing of hardlink info */
2264 int mdd_linkea_enable = 1;
2265 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2266 "record hardlink info in EAs");
2268 /** Read the link EA into a temp buffer.
2269 * Uses the name_buf since it is generally large.
2270 * \retval IS_ERR err
2271 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2273 struct lu_buf *mdd_links_get(const struct lu_env *env,
2274 struct mdd_object *mdd_obj)
2277 struct lustre_capa *capa;
2278 struct link_ea_header *leh;
2281 /* First try a small buf */
2282 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2283 if (buf->lb_buf == NULL)
2284 return ERR_PTR(-ENOMEM);
2286 capa = mdd_object_capa(env, mdd_obj);
2287 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2288 if (rc == -ERANGE) {
2289 /* Buf was too small, figure out what we need. */
2291 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2294 buf = mdd_buf_alloc(env, rc);
2295 if (buf->lb_buf == NULL)
2296 return ERR_PTR(-ENOMEM);
2297 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2303 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2304 leh->leh_magic = LINK_EA_MAGIC;
2305 leh->leh_reccount = __swab32(leh->leh_reccount);
2306 leh->leh_len = __swab64(leh->leh_len);
2307 /* entries are swabbed by mdd_lee_unpack */
2309 if (leh->leh_magic != LINK_EA_MAGIC)
2310 return ERR_PTR(-EINVAL);
2311 if (leh->leh_reccount == 0)
2312 return ERR_PTR(-ENODATA);
2317 /** Pack a link_ea_entry.
2318 * All elements are stored as chars to avoid alignment issues.
2319 * Numbers are always big-endian
2320 * \retval record length
2322 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2323 const struct lu_fid *pfid)
2325 struct lu_fid tmpfid;
2328 fid_cpu_to_be(&tmpfid, pfid);
2329 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2330 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2331 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2333 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2334 lee->lee_reclen[1] = reclen & 0xff;
2338 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2339 struct lu_name *lname, struct lu_fid *pfid)
2341 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2342 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2343 fid_be_to_cpu(pfid, pfid);
2344 lname->ln_name = lee->lee_name;
2345 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2348 /** Add a record to the end of link ea buf */
2349 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2350 const struct lu_fid *pfid,
2351 const struct lu_name *lname)
2353 struct link_ea_header *leh;
2354 struct link_ea_entry *lee;
2357 if (lname == NULL || pfid == NULL)
2360 /* Make sure our buf is big enough for the new one */
2362 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2363 if (leh->leh_len + reclen > buf->lb_len) {
2364 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2369 lee = buf->lb_buf + leh->leh_len;
2370 reclen = mdd_lee_pack(lee, lname, pfid);
2371 leh->leh_len += reclen;
2372 leh->leh_reccount++;
2376 /* For pathologic linkers, we don't want to spend lots of time scanning the
2377 * link ea. Limit ourseleves to something reasonable; links not in the EA
2378 * can be looked up via (slower) parent lookup.
2380 #define LINKEA_MAX_COUNT 128
2382 static int mdd_links_add(const struct lu_env *env,
2383 struct mdd_object *mdd_obj,
2384 const struct lu_fid *pfid,
2385 const struct lu_name *lname,
2386 struct thandle *handle, int first)
2389 struct link_ea_header *leh;
2393 if (!mdd_linkea_enable)
2396 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2399 if (rc != -ENODATA) {
2400 CERROR("link_ea read failed %d "DFID"\n", rc,
2401 PFID(mdd_object_fid(mdd_obj)));
2404 /* empty EA; start one */
2405 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2406 if (buf->lb_buf == NULL)
2409 leh->leh_magic = LINK_EA_MAGIC;
2410 leh->leh_len = sizeof(struct link_ea_header);
2411 leh->leh_reccount = 0;
2415 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2418 rc = __mdd_links_add(env, buf, pfid, lname);
2423 rc = __mdd_xattr_set(env, mdd_obj,
2424 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2425 XATTR_NAME_LINK, 0, handle);
2428 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2429 PFID(mdd_object_fid(mdd_obj)));
2431 CERROR("link_ea add failed %d "DFID"\n", rc,
2432 PFID(mdd_object_fid(mdd_obj)));
2435 if (buf->lb_vmalloc)
2436 /* if we vmalloced a large buffer drop it */
2442 static int mdd_links_rename(const struct lu_env *env,
2443 struct mdd_object *mdd_obj,
2444 const struct lu_fid *oldpfid,
2445 const struct lu_name *oldlname,
2446 const struct lu_fid *newpfid,
2447 const struct lu_name *newlname,
2448 struct thandle *handle)
2451 struct link_ea_header *leh;
2452 struct link_ea_entry *lee;
2453 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2454 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2460 if (!mdd_linkea_enable)
2463 if (mdd_obj->mod_flags & DEAD_OBJ)
2464 /* No more links, don't bother */
2467 buf = mdd_links_get(env, mdd_obj);
2471 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2472 rc, PFID(mdd_object_fid(mdd_obj)));
2474 CERROR("link_ea read failed %d "DFID"\n",
2475 rc, PFID(mdd_object_fid(mdd_obj)));
2479 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2481 /* Find the old record */
2482 for(count = 0; count < leh->leh_reccount; count++) {
2483 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2484 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2485 lu_fid_eq(tmpfid, oldpfid) &&
2486 (strncmp(tmpname->ln_name, oldlname->ln_name,
2487 tmpname->ln_namelen) == 0))
2489 lee = (struct link_ea_entry *)((char *)lee + reclen);
2491 if ((count + 1) > leh->leh_reccount) {
2492 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2493 oldlname->ln_namelen, oldlname->ln_name);
2494 GOTO(out, rc = -ENOENT);
2497 /* Remove the old record */
2498 leh->leh_reccount--;
2499 leh->leh_len -= reclen;
2500 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2503 /* If renaming, add the new record */
2504 if (newpfid != NULL) {
2505 /* if the add fails, we still delete the out-of-date old link */
2506 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2510 rc = __mdd_xattr_set(env, mdd_obj,
2511 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2512 XATTR_NAME_LINK, 0, handle);
2518 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2519 oldlname->ln_namelen, oldlname->ln_name, rc,
2520 PFID(mdd_object_fid(mdd_obj)));
2522 if (buf->lb_vmalloc)
2523 /* if we vmalloced a large buffer drop it */
2529 const struct md_dir_operations mdd_dir_ops = {
2530 .mdo_is_subdir = mdd_is_subdir,
2531 .mdo_lookup = mdd_lookup,
2532 .mdo_create = mdd_create,
2533 .mdo_rename = mdd_rename,
2534 .mdo_link = mdd_link,
2535 .mdo_unlink = mdd_unlink,
2536 .mdo_name_insert = mdd_name_insert,
2537 .mdo_name_remove = mdd_name_remove,
2538 .mdo_rename_tgt = mdd_rename_tgt,
2539 .mdo_create_data = mdd_create_data