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);
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, mdo2fid(mdd_tobj), lname, handle);
749 mdd_write_unlock(env, mdd_sobj);
750 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
753 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
754 mdd_tobj, NULL, lname, handle);
755 mdd_trans_stop(env, mdd, rc, handle);
757 #ifdef HAVE_QUOTA_SUPPORT
759 lquota_pending_commit(mds_quota_interface_ref, obd,
760 qids, rec_pending, 1);
761 /* Trigger dqacq for the parent owner. If failed,
762 * the next call for lquota_chkquota will process it. */
763 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
770 /* caller should take a lock before calling */
771 int mdd_finish_unlink(const struct lu_env *env,
772 struct mdd_object *obj, struct md_attr *ma,
779 LASSERT(mdd_write_locked(env, obj) != 0);
781 rc = mdd_iattr_get(env, obj, ma);
782 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
783 obj->mod_flags |= DEAD_OBJ;
784 /* add new orphan and the object
785 * will be deleted during mdd_close() */
786 if (obj->mod_count) {
787 rc = __mdd_orphan_add(env, obj, th);
789 CDEBUG(D_HA, "Object "DFID" is inserted into "
790 "orphan list, open count = %d\n",
791 PFID(mdd_object_fid(obj)),
794 CERROR("Object "DFID" fail to be an orphan, "
795 "open count = %d, maybe cause failed "
797 PFID(mdd_object_fid(obj)),
800 rc = mdd_object_kill(env, obj, ma);
807 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
814 * has mdd_write_lock on cobj already, but not on pobj yet
816 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
817 struct mdd_object *cobj, struct md_attr *ma)
822 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
827 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
828 struct md_object *cobj, const struct lu_name *lname,
831 const char *name = lname->ln_name;
832 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
833 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
834 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
835 struct mdd_device *mdd = mdo2mdd(pobj);
836 struct dynlock_handle *dlh;
837 struct thandle *handle;
838 #ifdef HAVE_QUOTA_SUPPORT
839 struct obd_device *obd = mdd->mdd_obd_dev;
840 struct mds_obd *mds = &obd->u.mds;
841 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
842 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
845 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
849 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
850 PFID(mdd_object_fid(mdd_cobj)));
852 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
856 handle = mdd_trans_start(env, mdd);
858 RETURN(PTR_ERR(handle));
860 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
862 GOTO(out_trans, rc = -ENOMEM);
863 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
865 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
869 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
870 mdd_object_capa(env, mdd_pobj));
874 __mdd_ref_del(env, mdd_cobj, handle, 0);
877 __mdd_ref_del(env, mdd_cobj, handle, 1);
879 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
880 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
882 la->la_valid = LA_CTIME | LA_MTIME;
883 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
887 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
888 /* update ctime of an unlinked file only if it is still
889 * opened or a link still exists */
890 la->la_valid = LA_CTIME;
891 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
896 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
897 #ifdef HAVE_QUOTA_SUPPORT
898 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
899 ma->ma_attr.la_nlink == 0) {
900 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
902 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
904 mdd_quota_wrapper(la_tmp, qpids);
905 if (mdd_cobj->mod_count == 0) {
906 quota_opc = FSFILT_OP_UNLINK;
907 mdd_quota_wrapper(&ma->ma_attr, qcids);
909 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
915 /* old files may not have link ea; ignore errors */
916 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
917 lname, NULL, NULL, handle);
921 mdd_write_unlock(env, mdd_cobj);
922 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
925 rc = mdd_changelog_ns_store(env, mdd,
926 is_dir ? CL_RMDIR : CL_UNLINK,
927 mdd_cobj, mdd_pobj, NULL, lname,
930 mdd_trans_stop(env, mdd, rc, handle);
931 #ifdef HAVE_QUOTA_SUPPORT
933 /* Trigger dqrel on the owner of child and parent. If failed,
934 * the next call for lquota_chkquota will process it. */
935 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
941 /* has not lock on pobj yet */
942 static int mdd_ni_sanity_check(const struct lu_env *env,
943 struct md_object *pobj,
944 const struct md_attr *ma)
946 struct mdd_object *obj = md2mdd_obj(pobj);
950 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
953 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
961 static int mdd_name_insert(const struct lu_env *env,
962 struct md_object *pobj,
963 const struct lu_name *lname,
964 const struct lu_fid *fid,
965 const struct md_attr *ma)
967 const char *name = lname->ln_name;
968 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
969 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
970 struct mdd_device *mdd = mdo2mdd(pobj);
971 struct dynlock_handle *dlh;
972 struct thandle *handle;
973 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
974 #ifdef HAVE_QUOTA_SUPPORT
975 struct md_ucred *uc = md_ucred(env);
976 struct obd_device *obd = mdd->mdd_obd_dev;
977 struct obd_export *exp = md_quota(env)->mq_exp;
978 struct mds_obd *mds = &obd->u.mds;
979 unsigned int qids[MAXQUOTAS] = { 0, 0 };
980 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
981 cfs_cap_t save = uc->mu_cap;
986 #ifdef HAVE_QUOTA_SUPPORT
987 if (mds->mds_quota) {
988 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
989 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
991 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
994 mdd_data_get(env, mdd_obj, &data);
995 quota_opc = FSFILT_OP_LINK;
996 mdd_quota_wrapper(la_tmp, qids);
997 /* get block quota for parent */
998 lquota_chkquota(mds_quota_interface_ref, obd,
999 exp, qids, rec_pending, 1, NULL,
1000 LQUOTA_FLAGS_BLK, data, 1);
1003 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1007 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
1008 handle = mdd_trans_start(env, mdo2mdd(pobj));
1010 GOTO(out_pending, rc = PTR_ERR(handle));
1012 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1014 GOTO(out_trans, rc = -ENOMEM);
1016 rc = mdd_ni_sanity_check(env, pobj, ma);
1018 GOTO(out_unlock, rc);
1020 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1021 handle, BYPASS_CAPA);
1023 GOTO(out_unlock, rc);
1026 * For some case, no need update obj's ctime (LA_CTIME is not set),
1028 * For other cases, update obj's ctime (LA_CTIME is set),
1031 if (ma->ma_attr.la_valid & LA_CTIME) {
1032 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1033 la->la_valid = LA_CTIME | LA_MTIME;
1034 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1039 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1041 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1043 #ifdef HAVE_QUOTA_SUPPORT
1044 if (mds->mds_quota) {
1046 lquota_pending_commit(mds_quota_interface_ref,
1047 obd, qids, rec_pending, 1);
1048 /* Trigger dqacq for the parent owner. If failed,
1049 * the next call for lquota_chkquota will process it*/
1050 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1060 /* has not lock on pobj yet */
1061 static int mdd_nr_sanity_check(const struct lu_env *env,
1062 struct md_object *pobj,
1063 const struct md_attr *ma)
1065 struct mdd_object *obj = md2mdd_obj(pobj);
1069 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1072 rc = mdd_may_unlink(env, obj, ma);
1078 * Partial operation.
1080 static int mdd_name_remove(const struct lu_env *env,
1081 struct md_object *pobj,
1082 const struct lu_name *lname,
1083 const struct md_attr *ma)
1085 const char *name = lname->ln_name;
1086 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1087 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1088 struct mdd_device *mdd = mdo2mdd(pobj);
1089 struct dynlock_handle *dlh;
1090 struct thandle *handle;
1091 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1092 #ifdef HAVE_QUOTA_SUPPORT
1093 struct obd_device *obd = mdd->mdd_obd_dev;
1094 struct mds_obd *mds = &obd->u.mds;
1095 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1101 #ifdef HAVE_QUOTA_SUPPORT
1102 if (mds->mds_quota) {
1103 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1105 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1107 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1108 mdd_quota_wrapper(la_tmp, qids);
1112 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1113 handle = mdd_trans_start(env, mdd);
1115 GOTO(out_pending, rc = PTR_ERR(handle));
1117 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1119 GOTO(out_trans, rc = -ENOMEM);
1121 rc = mdd_nr_sanity_check(env, pobj, ma);
1123 GOTO(out_unlock, rc);
1125 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1126 handle, BYPASS_CAPA);
1128 GOTO(out_unlock, rc);
1131 * For some case, no need update obj's ctime (LA_CTIME is not set),
1133 * For other cases, update obj's ctime (LA_CTIME is set),
1136 if (ma->ma_attr.la_valid & LA_CTIME) {
1137 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1138 la->la_valid = LA_CTIME | LA_MTIME;
1139 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1144 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1146 mdd_trans_stop(env, mdd, rc, handle);
1148 #ifdef HAVE_QUOTA_SUPPORT
1149 /* Trigger dqrel for the parent owner.
1150 * If failed, the next call for lquota_chkquota will process it. */
1152 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1160 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1162 static int mdd_rt_sanity_check(const struct lu_env *env,
1163 struct mdd_object *tgt_pobj,
1164 struct mdd_object *tobj,
1170 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1173 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1174 * exist. In fact, tobj must exist, otherwise the call trace will be:
1175 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1176 * When get here, tobj must be NOT NULL, the other case has been
1177 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1179 * So check may_delete, but not check nlink of tgt_pobj. */
1181 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1186 /* Partial rename op on slave MDD */
1187 static int mdd_rename_tgt(const struct lu_env *env,
1188 struct md_object *pobj, struct md_object *tobj,
1189 const struct lu_fid *lf, const struct lu_name *lname,
1192 const char *name = lname->ln_name;
1193 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1194 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1195 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1196 struct mdd_device *mdd = mdo2mdd(pobj);
1197 struct dynlock_handle *dlh;
1198 struct thandle *handle;
1199 #ifdef HAVE_QUOTA_SUPPORT
1200 struct obd_device *obd = mdd->mdd_obd_dev;
1201 struct obd_export *exp = md_quota(env)->mq_exp;
1202 struct mds_obd *mds = &obd->u.mds;
1203 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1204 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1205 int quota_copc = 0, quota_popc = 0;
1206 int rec_pending[MAXQUOTAS] = { 0, 0 };
1211 #ifdef HAVE_QUOTA_SUPPORT
1212 if (mds->mds_quota && !tobj) {
1213 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1215 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1218 mdd_data_get(env, mdd_tpobj, &data);
1219 quota_popc = FSFILT_OP_LINK;
1220 mdd_quota_wrapper(la_tmp, qpids);
1221 /* get block quota for target parent */
1222 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1223 qpids, rec_pending, 1, NULL,
1224 LQUOTA_FLAGS_BLK, data, 1);
1228 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1229 handle = mdd_trans_start(env, mdd);
1231 GOTO(out_pending, rc = PTR_ERR(handle));
1233 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1235 GOTO(out_trans, rc = -ENOMEM);
1237 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1239 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1244 * If rename_tgt is called then we should just re-insert name with
1245 * correct fid, no need to dec/inc parent nlink if obj is dir.
1247 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1251 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1256 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1257 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1259 la->la_valid = LA_CTIME | LA_MTIME;
1260 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1265 * For tobj is remote case cmm layer has processed
1266 * and pass NULL tobj to here. So when tobj is NOT NULL,
1267 * it must be local one.
1269 if (tobj && mdd_object_exists(mdd_tobj)) {
1270 __mdd_ref_del(env, mdd_tobj, handle, 0);
1272 /* Remove dot reference. */
1273 if (S_ISDIR(ma->ma_attr.la_mode))
1274 __mdd_ref_del(env, mdd_tobj, handle, 1);
1276 la->la_valid = LA_CTIME;
1277 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1281 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1285 #ifdef HAVE_QUOTA_SUPPORT
1286 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1287 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1288 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1289 mdd_quota_wrapper(&ma->ma_attr, qcids);
1296 mdd_write_unlock(env, mdd_tobj);
1297 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1300 /* Bare EXT record with no RENAME in front of it signifies
1301 a partial slave op */
1302 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
1303 mdd_tpobj, NULL, lname, handle);
1305 mdd_trans_stop(env, mdd, rc, handle);
1307 #ifdef HAVE_QUOTA_SUPPORT
1308 if (mds->mds_quota) {
1310 lquota_pending_commit(mds_quota_interface_ref, obd,
1311 qpids, rec_pending, 1);
1314 /* Trigger dqrel on the target owner of child.
1315 * If failed, the next call for lquota_chkquota
1316 * will process it. */
1317 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1325 * The permission has been checked when obj created, no need check again.
1327 static int mdd_cd_sanity_check(const struct lu_env *env,
1328 struct mdd_object *obj)
1333 if (!obj || mdd_is_dead_obj(obj))
1340 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1341 struct md_object *cobj, const struct md_op_spec *spec,
1344 struct mdd_device *mdd = mdo2mdd(cobj);
1345 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1346 struct mdd_object *son = md2mdd_obj(cobj);
1347 struct lu_attr *attr = &ma->ma_attr;
1348 struct lov_mds_md *lmm = NULL;
1350 struct thandle *handle;
1354 rc = mdd_cd_sanity_check(env, son);
1358 if (!md_should_create(spec->sp_cr_flags))
1360 lmm_size = ma->ma_lmm_size;
1361 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1366 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1367 handle = mdd_trans_start(env, mdd);
1369 GOTO(out_free, rc = PTR_ERR(handle));
1372 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1373 * Should this be fixed?
1376 /* Replay creates has objects already */
1378 if (spec->no_create) {
1379 CDEBUG(D_INFO, "we already have lov ea\n");
1380 rc = mdd_lov_set_md(env, mdd_pobj, son,
1381 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1382 spec->u.sp_ea.eadatalen, handle, 0);
1385 /* No need mdd_lsm_sanity_check here */
1386 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1387 lmm_size, handle, 0);
1390 rc = mdd_attr_get_internal_locked(env, son, ma);
1392 /* update lov_objid data, must be before transaction stop! */
1394 mdd_lov_objid_update(mdd, lmm);
1396 mdd_trans_stop(env, mdd, rc, handle);
1398 /* Finish mdd_lov_create() stuff. */
1399 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1403 /* Get fid from name and parent */
1405 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1406 const struct lu_name *lname, struct lu_fid* fid, int mask)
1408 const char *name = lname->ln_name;
1409 const struct dt_key *key = (const struct dt_key *)name;
1410 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1411 struct mdd_device *m = mdo2mdd(pobj);
1412 struct dt_object *dir = mdd_object_child(mdd_obj);
1416 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1419 rc = mdd_object_exists(mdd_obj);
1420 if (unlikely(rc == 0))
1422 else if (unlikely(rc < 0)) {
1423 CERROR("Object "DFID" locates on remote server\n",
1424 PFID(mdo2fid(mdd_obj)));
1428 /* The common filename length check. */
1429 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1430 RETURN(-ENAMETOOLONG);
1432 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1437 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1438 dt_try_as_dir(env, dir))) {
1440 rc = dir->do_index_ops->dio_lookup(env, dir,
1441 (struct dt_rec *)fid, key,
1442 mdd_object_capa(env, mdd_obj));
1453 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1454 const struct lu_name *lname, struct mdd_object *child,
1455 struct md_attr *ma, struct thandle *handle,
1456 const struct md_op_spec *spec)
1462 * Update attributes for child.
1465 * (1) the valid bits should be converted between Lustre and Linux;
1466 * (2) maybe, the child attributes should be set in OSD when creation.
1469 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1473 if (S_ISDIR(ma->ma_attr.la_mode)) {
1474 /* Add "." and ".." for newly created dir */
1475 __mdd_ref_add(env, child, handle);
1476 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1477 dot, handle, BYPASS_CAPA);
1479 rc = __mdd_index_insert_only(env, child, pfid,
1483 __mdd_ref_del(env, child, handle, 1);
1486 mdd_links_add(env, child, pfid, lname, handle);
1491 /* has not lock on pobj yet */
1492 static int mdd_create_sanity_check(const struct lu_env *env,
1493 struct md_object *pobj,
1494 const struct lu_name *lname,
1496 struct md_op_spec *spec)
1498 struct mdd_thread_info *info = mdd_env_info(env);
1499 struct lu_attr *la = &info->mti_la;
1500 struct lu_fid *fid = &info->mti_fid;
1501 struct mdd_object *obj = md2mdd_obj(pobj);
1502 struct mdd_device *m = mdo2mdd(pobj);
1503 int lookup = spec->sp_cr_lookup;
1508 if (mdd_is_dead_obj(obj))
1512 * In some cases this lookup is not needed - we know before if name
1513 * exists or not because MDT performs lookup for it.
1514 * name length check is done in lookup.
1518 * Check if the name already exist, though it will be checked in
1519 * _index_insert also, for avoiding rolling back if exists
1522 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1523 MAY_WRITE | MAY_EXEC);
1525 RETURN(rc ? : -EEXIST);
1528 * Check WRITE permission for the parent.
1529 * EXEC permission have been checked
1530 * when lookup before create already.
1532 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1539 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1543 if (la->la_mode & S_ISGID) {
1544 ma->ma_attr.la_gid = la->la_gid;
1545 if (S_ISDIR(ma->ma_attr.la_mode)) {
1546 ma->ma_attr.la_mode |= S_ISGID;
1547 ma->ma_attr.la_valid |= LA_MODE;
1551 switch (ma->ma_attr.la_mode & S_IFMT) {
1553 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1555 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1556 RETURN(-ENAMETOOLONG);
1576 * Create object and insert it into namespace.
1578 static int mdd_create(const struct lu_env *env,
1579 struct md_object *pobj,
1580 const struct lu_name *lname,
1581 struct md_object *child,
1582 struct md_op_spec *spec,
1585 struct mdd_thread_info *info = mdd_env_info(env);
1586 struct lu_attr *la = &info->mti_la_for_fix;
1587 struct md_attr *ma_acl = &info->mti_ma;
1588 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1589 struct mdd_object *son = md2mdd_obj(child);
1590 struct mdd_device *mdd = mdo2mdd(pobj);
1591 struct lu_attr *attr = &ma->ma_attr;
1592 struct lov_mds_md *lmm = NULL;
1593 struct thandle *handle;
1594 struct dynlock_handle *dlh;
1595 const char *name = lname->ln_name;
1596 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1597 int got_def_acl = 0;
1598 #ifdef HAVE_QUOTA_SUPPORT
1599 struct obd_device *obd = mdd->mdd_obd_dev;
1600 struct obd_export *exp = md_quota(env)->mq_exp;
1601 struct mds_obd *mds = &obd->u.mds;
1602 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1603 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1604 int quota_opc = 0, block_count = 0;
1605 int inode_pending[MAXQUOTAS] = { 0, 0 };
1606 int block_pending[MAXQUOTAS] = { 0, 0 };
1607 int parent_pending[MAXQUOTAS] = { 0, 0 };
1612 * Two operations have to be performed:
1614 * - an allocation of a new object (->do_create()), and
1616 * - an insertion into a parent index (->dio_insert()).
1618 * Due to locking, operation order is not important, when both are
1619 * successful, *but* error handling cases are quite different:
1621 * - if insertion is done first, and following object creation fails,
1622 * insertion has to be rolled back, but this operation might fail
1623 * also leaving us with dangling index entry.
1625 * - if creation is done first, is has to be undone if insertion
1626 * fails, leaving us with leaked space, which is neither good, nor
1629 * It seems that creation-first is simplest solution, but it is
1630 * sub-optimal in the frequent
1635 * case, because second mkdir is bound to create object, only to
1636 * destroy it immediately.
1638 * To avoid this follow local file systems that do double lookup:
1640 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1642 * 1. create (mdd_object_create_internal())
1644 * 2. insert (__mdd_index_insert(), lookup again)
1647 /* Sanity checks before big job. */
1648 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1652 #ifdef HAVE_QUOTA_SUPPORT
1653 if (mds->mds_quota) {
1654 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1656 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1660 quota_opc = FSFILT_OP_CREATE;
1661 mdd_quota_wrapper(&ma->ma_attr, qcids);
1662 mdd_quota_wrapper(la_tmp, qpids);
1663 /* get file quota for child */
1664 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1665 qcids, inode_pending, 1, NULL, 0, NULL,
1667 switch (ma->ma_attr.la_mode & S_IFMT) {
1676 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1677 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1681 /* get block quota for child and parent */
1683 lquota_chkquota(mds_quota_interface_ref, obd,
1684 exp, qcids, block_pending,
1686 LQUOTA_FLAGS_BLK, NULL, 0);
1688 lquota_chkquota(mds_quota_interface_ref, obd,
1689 exp, qpids, parent_pending, 1,
1690 NULL, LQUOTA_FLAGS_BLK, NULL,
1697 * No RPC inside the transaction, so OST objects should be created at
1700 if (S_ISREG(attr->la_mode)) {
1701 lmm_size = ma->ma_lmm_size;
1702 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1705 GOTO(out_pending, rc);
1708 if (!S_ISLNK(attr->la_mode)) {
1709 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1710 ma_acl->ma_acl = info->mti_xattr_buf;
1711 ma_acl->ma_need = MA_ACL_DEF;
1712 ma_acl->ma_valid = 0;
1714 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1715 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1716 mdd_read_unlock(env, mdd_pobj);
1719 else if (ma_acl->ma_valid & MA_ACL_DEF)
1723 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1724 handle = mdd_trans_start(env, mdd);
1726 GOTO(out_free, rc = PTR_ERR(handle));
1728 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1730 GOTO(out_trans, rc = -ENOMEM);
1732 mdd_write_lock(env, son, MOR_TGT_CHILD);
1733 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1735 mdd_write_unlock(env, son);
1741 #ifdef CONFIG_FS_POSIX_ACL
1743 struct lu_buf *acl_buf = &info->mti_buf;
1744 acl_buf->lb_buf = ma_acl->ma_acl;
1745 acl_buf->lb_len = ma_acl->ma_acl_size;
1747 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1749 mdd_write_unlock(env, son);
1752 ma->ma_attr.la_valid |= LA_MODE;
1757 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1758 son, ma, handle, spec);
1759 mdd_write_unlock(env, son);
1762 * Object has no links, so it will be destroyed when last
1763 * reference is released. (XXX not now.)
1769 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1770 name, S_ISDIR(attr->la_mode), handle,
1771 mdd_object_capa(env, mdd_pobj));
1778 /* No need mdd_lsm_sanity_check here */
1779 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1781 CERROR("error on stripe info copy %d \n", rc);
1784 if (lmm && lmm_size > 0) {
1785 /* Set Lov here, do not get lmm again later */
1786 memcpy(ma->ma_lmm, lmm, lmm_size);
1787 ma->ma_lmm_size = lmm_size;
1788 ma->ma_valid |= MA_LOV;
1791 if (S_ISLNK(attr->la_mode)) {
1792 struct md_ucred *uc = md_ucred(env);
1793 struct dt_object *dt = mdd_object_child(son);
1794 const char *target_name = spec->u.sp_symname;
1795 int sym_len = strlen(target_name);
1796 const struct lu_buf *buf;
1799 buf = mdd_buf_get_const(env, target_name, sym_len);
1800 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1801 mdd_object_capa(env, son),
1803 CFS_CAP_SYS_RESOURCE_MASK);
1808 GOTO(cleanup, rc = -EFAULT);
1812 la->la_valid = LA_CTIME | LA_MTIME;
1813 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1817 /* Return attr back. */
1818 rc = mdd_attr_get_internal_locked(env, son, ma);
1821 if (rc && created) {
1825 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1826 S_ISDIR(attr->la_mode),
1827 handle, BYPASS_CAPA);
1829 CERROR("error can not cleanup destroy %d\n",
1834 mdd_write_lock(env, son, MOR_TGT_CHILD);
1835 __mdd_ref_del(env, son, handle, 0);
1836 if (initialized && S_ISDIR(attr->la_mode))
1837 __mdd_ref_del(env, son, handle, 1);
1838 mdd_write_unlock(env, son);
1842 /* update lov_objid data, must be before transaction stop! */
1844 mdd_lov_objid_update(mdd, lmm);
1846 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1849 rc = mdd_changelog_ns_store(env, mdd,
1850 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1851 S_ISREG(attr->la_mode) ? CL_CREATE :
1852 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1853 son, mdd_pobj, NULL, lname, handle);
1854 mdd_trans_stop(env, mdd, rc, handle);
1856 /* finis lov_create stuff, free all temporary data */
1857 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1859 #ifdef HAVE_QUOTA_SUPPORT
1861 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1863 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1865 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1867 /* Trigger dqacq on the owner of child and parent. If failed,
1868 * the next call for lquota_chkquota will process it. */
1869 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1877 * Get locks on parents in proper order
1878 * RETURN: < 0 - error, rename_order if successful
1886 static int mdd_rename_order(const struct lu_env *env,
1887 struct mdd_device *mdd,
1888 struct mdd_object *src_pobj,
1889 struct mdd_object *tgt_pobj)
1891 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1895 if (src_pobj == tgt_pobj)
1896 RETURN(MDD_RN_SAME);
1898 /* compared the parent child relationship of src_p&tgt_p */
1899 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1901 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1904 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1917 /* has not mdd_write{read}_lock on any obj yet. */
1918 static int mdd_rename_sanity_check(const struct lu_env *env,
1919 struct mdd_object *src_pobj,
1920 struct mdd_object *tgt_pobj,
1921 struct mdd_object *sobj,
1922 struct mdd_object *tobj,
1928 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1931 /* XXX: when get here, sobj must NOT be NULL,
1932 * the other case has been processed in cml_rename
1933 * before mdd_rename and enable MDS_PERM_BYPASS. */
1936 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1940 /* XXX: when get here, "tobj == NULL" means tobj must
1941 * NOT exist (neither on remote MDS, such case has been
1942 * processed in cml_rename before mdd_rename and enable
1944 * So check may_create, but not check may_unlink. */
1946 rc = mdd_may_create(env, tgt_pobj, NULL,
1947 (src_pobj != tgt_pobj), 0);
1949 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1950 (src_pobj != tgt_pobj), 1);
1952 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1953 S_ISDIR(ma->ma_attr.la_mode))
1954 rc = __mdd_may_link(env, tgt_pobj);
1959 /* src object can be remote that is why we use only fid and type of object */
1960 static int mdd_rename(const struct lu_env *env,
1961 struct md_object *src_pobj, struct md_object *tgt_pobj,
1962 const struct lu_fid *lf, const struct lu_name *lsname,
1963 struct md_object *tobj, const struct lu_name *ltname,
1966 const char *sname = lsname->ln_name;
1967 const char *tname = ltname->ln_name;
1968 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1969 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1970 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1971 struct mdd_device *mdd = mdo2mdd(src_pobj);
1972 struct mdd_object *mdd_sobj = NULL; /* source object */
1973 struct mdd_object *mdd_tobj = NULL;
1974 struct dynlock_handle *sdlh, *tdlh;
1975 struct thandle *handle;
1976 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1977 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1981 #ifdef HAVE_QUOTA_SUPPORT
1982 struct obd_device *obd = mdd->mdd_obd_dev;
1983 struct obd_export *exp = md_quota(env)->mq_exp;
1984 struct mds_obd *mds = &obd->u.mds;
1985 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1986 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1987 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
1988 int quota_copc = 0, quota_popc = 0;
1989 int rec_pending[MAXQUOTAS] = { 0, 0 };
1993 LASSERT(ma->ma_attr.la_mode & S_IFMT);
1994 is_dir = S_ISDIR(ma->ma_attr.la_mode);
1997 mdd_tobj = md2mdd_obj(tobj);
1999 #ifdef HAVE_QUOTA_SUPPORT
2000 if (mds->mds_quota) {
2001 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2003 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2005 mdd_quota_wrapper(la_tmp, qspids);
2007 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2011 mdd_data_get(env, mdd_tpobj, &data);
2012 quota_popc = FSFILT_OP_LINK;
2013 mdd_quota_wrapper(la_tmp, qtpids);
2014 /* get block quota for target parent */
2015 lquota_chkquota(mds_quota_interface_ref,
2017 rec_pending, 1, NULL,
2025 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2026 handle = mdd_trans_start(env, mdd);
2028 GOTO(out_pending, rc = PTR_ERR(handle));
2030 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2031 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2033 GOTO(cleanup_unlocked, rc);
2035 /* Get locks in determined order */
2036 if (rc == MDD_RN_SAME) {
2037 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2038 sname, MOR_SRC_PARENT);
2039 /* check hashes to determine do we need one lock or two */
2040 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2041 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2045 } else if (rc == MDD_RN_SRCTGT) {
2046 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2047 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2049 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2050 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2052 if (sdlh == NULL || tdlh == NULL)
2053 GOTO(cleanup, rc = -ENOMEM);
2055 mdd_sobj = mdd_object_find(env, mdd, lf);
2056 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2057 mdd_sobj, mdd_tobj, ma);
2061 /* Remove source name from source directory */
2062 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2063 mdd_object_capa(env, mdd_spobj));
2067 /* "mv dir1 dir2" needs "dir1/.." link update */
2068 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2069 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2070 mdd_object_capa(env, mdd_sobj));
2072 GOTO(fixup_spobj2, rc);
2074 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2075 handle, mdd_object_capa(env, mdd_sobj));
2077 GOTO(fixup_spobj, rc);
2080 /* Remove target name from target directory
2081 * Here tobj can be remote one, so we do index_delete unconditionally
2082 * and -ENOENT is allowed.
2084 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2085 mdd_object_capa(env, mdd_tpobj));
2088 /* tname might been renamed to something else */
2089 GOTO(fixup_spobj, rc);
2092 GOTO(fixup_spobj, rc);
2095 /* Insert new fid with target name into target dir */
2096 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2097 mdd_object_capa(env, mdd_tpobj));
2099 GOTO(fixup_tpobj, rc);
2101 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2102 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2104 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2106 la->la_valid = LA_CTIME;
2107 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2110 GOTO(fixup_tpobj, rc);
2113 /* Remove old target object
2114 * For tobj is remote case cmm layer has processed
2115 * and set tobj to NULL then. So when tobj is NOT NULL,
2116 * it must be local one.
2118 if (tobj && mdd_object_exists(mdd_tobj)) {
2119 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2120 if (mdd_is_dead_obj(mdd_tobj)) {
2121 mdd_write_unlock(env, mdd_tobj);
2122 /* shld not be dead, something is wrong */
2123 CERROR("tobj is dead, something is wrong\n");
2127 __mdd_ref_del(env, mdd_tobj, handle, 0);
2129 /* Remove dot reference. */
2131 __mdd_ref_del(env, mdd_tobj, handle, 1);
2133 la->la_valid = LA_CTIME;
2134 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2136 GOTO(fixup_tpobj, rc);
2138 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2139 mdd_write_unlock(env, mdd_tobj);
2141 GOTO(fixup_tpobj, rc);
2143 #ifdef HAVE_QUOTA_SUPPORT
2144 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2145 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2146 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2147 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2152 la->la_valid = LA_CTIME | LA_MTIME;
2153 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2155 GOTO(fixup_tpobj, rc);
2157 if (mdd_spobj != mdd_tpobj) {
2158 la->la_valid = LA_CTIME | LA_MTIME;
2159 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2163 if (rc == 0 && mdd_sobj) {
2164 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2165 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2166 mdo2fid(mdd_tpobj), ltname, handle);
2168 /* Old files might not have EA entry */
2169 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2171 mdd_write_unlock(env, mdd_sobj);
2172 /* We don't fail the transaction if the link ea can't be
2173 updated -- fid2path will use alternate lookup method. */
2181 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2184 CWARN("tp obj fix error %d\n",rc2);
2186 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2187 !mdd_is_dead_obj(mdd_tobj)) {
2188 rc2 = __mdd_index_insert(env, mdd_tpobj,
2189 mdo2fid(mdd_tobj), tname,
2194 CWARN("tp obj fix error %d\n",rc2);
2199 if (rc && is_dir && mdd_sobj) {
2200 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2204 CWARN("sp obj dotdot delete error %d\n",rc2);
2207 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2208 dotdot, handle, BYPASS_CAPA);
2210 CWARN("sp obj dotdot insert error %d\n",rc2);
2215 rc2 = __mdd_index_insert(env, mdd_spobj,
2216 lf, sname, is_dir, handle, BYPASS_CAPA);
2218 CWARN("sp obj fix error %d\n",rc2);
2221 if (likely(tdlh) && sdlh != tdlh)
2222 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2224 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2227 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
2228 mdd_spobj, lf, lsname, handle);
2230 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
2231 mdd_tpobj, lf, ltname, handle);
2233 mdd_trans_stop(env, mdd, rc, handle);
2235 mdd_object_put(env, mdd_sobj);
2237 #ifdef HAVE_QUOTA_SUPPORT
2238 if (mds->mds_quota) {
2240 lquota_pending_commit(mds_quota_interface_ref, obd,
2241 qtpids, rec_pending, 1);
2244 /* Trigger dqrel on the source owner of parent.
2245 * If failed, the next call for lquota_chkquota will
2247 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2248 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2250 /* Trigger dqrel on the target owner of child.
2251 * If failed, the next call for lquota_chkquota
2252 * will process it. */
2253 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2254 qtpids, rc, quota_copc);
2261 /** enable/disable storing of hardlink info */
2262 int mdd_linkea_enable = 1;
2263 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2264 "record hardlink info in EAs");
2266 /** Read the link EA into a temp buffer.
2267 * Uses the name_buf since it is generally large.
2268 * \retval IS_ERR err
2269 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2271 struct lu_buf *mdd_links_get(const struct lu_env *env,
2272 struct mdd_object *mdd_obj)
2275 struct lustre_capa *capa;
2276 struct link_ea_header *leh;
2279 /* First try a small buf */
2280 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2281 if (buf->lb_buf == NULL)
2282 return ERR_PTR(-ENOMEM);
2284 capa = mdd_object_capa(env, mdd_obj);
2285 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2286 if (rc == -ERANGE) {
2287 /* Buf was too small, figure out what we need. */
2289 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2292 buf = mdd_buf_alloc(env, rc);
2293 if (buf->lb_buf == NULL)
2294 return ERR_PTR(-ENOMEM);
2295 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2301 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2302 leh->leh_magic = LINK_EA_MAGIC;
2303 leh->leh_reccount = __swab32(leh->leh_reccount);
2304 leh->leh_len = __swab64(leh->leh_len);
2305 /* entries are swabbed by mdd_lee_unpack */
2307 if (leh->leh_magic != LINK_EA_MAGIC)
2308 return ERR_PTR(-EINVAL);
2309 if (leh->leh_reccount == 0)
2310 return ERR_PTR(-ENODATA);
2315 /** Pack a link_ea_entry.
2316 * All elements are stored as chars to avoid alignment issues.
2317 * Numbers are always big-endian
2318 * \retval record length
2320 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2321 const struct lu_fid *pfid)
2323 struct lu_fid tmpfid;
2326 fid_cpu_to_be(&tmpfid, pfid);
2327 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2328 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2329 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2331 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2332 lee->lee_reclen[1] = reclen & 0xff;
2336 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2337 struct lu_name *lname, struct lu_fid *pfid)
2339 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2340 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2341 fid_be_to_cpu(pfid, pfid);
2342 lname->ln_name = lee->lee_name;
2343 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2346 /** Add a record to the end of link ea buf */
2347 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2348 const struct lu_fid *pfid,
2349 const struct lu_name *lname)
2351 struct link_ea_header *leh;
2352 struct link_ea_entry *lee;
2355 if (lname == NULL || pfid == NULL)
2358 /* Make sure our buf is big enough for the new one */
2360 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2361 if (leh->leh_len + reclen > buf->lb_len) {
2362 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2367 lee = buf->lb_buf + leh->leh_len;
2368 reclen = mdd_lee_pack(lee, lname, pfid);
2369 leh->leh_len += reclen;
2370 leh->leh_reccount++;
2374 /* For pathologic linkers, we don't want to spend lots of time scanning the
2375 * link ea. Limit ourseleves to something reasonable; links not in the EA
2376 * can be looked up via (slower) parent lookup.
2378 #define LINKEA_MAX_COUNT 128
2380 static int mdd_links_add(const struct lu_env *env,
2381 struct mdd_object *mdd_obj,
2382 const struct lu_fid *pfid,
2383 const struct lu_name *lname,
2384 struct thandle *handle)
2387 struct link_ea_header *leh;
2391 if (!mdd_linkea_enable)
2394 buf = mdd_links_get(env, mdd_obj);
2397 if (rc != -ENODATA) {
2398 CERROR("link_ea read failed %d "DFID"\n", rc,
2399 PFID(mdd_object_fid(mdd_obj)));
2402 /* empty EA; start one */
2403 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2404 if (buf->lb_buf == NULL)
2407 leh->leh_magic = LINK_EA_MAGIC;
2408 leh->leh_len = sizeof(struct link_ea_header);
2409 leh->leh_reccount = 0;
2413 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2416 rc = __mdd_links_add(env, buf, pfid, lname);
2421 rc = __mdd_xattr_set(env, mdd_obj,
2422 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2423 XATTR_NAME_LINK, 0, handle);
2426 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2427 PFID(mdd_object_fid(mdd_obj)));
2429 CERROR("link_ea add failed %d "DFID"\n", rc,
2430 PFID(mdd_object_fid(mdd_obj)));
2433 if (buf->lb_vmalloc)
2434 /* if we vmalloced a large buffer drop it */
2440 static int mdd_links_rename(const struct lu_env *env,
2441 struct mdd_object *mdd_obj,
2442 const struct lu_fid *oldpfid,
2443 const struct lu_name *oldlname,
2444 const struct lu_fid *newpfid,
2445 const struct lu_name *newlname,
2446 struct thandle *handle)
2449 struct link_ea_header *leh;
2450 struct link_ea_entry *lee;
2451 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2452 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2458 if (!mdd_linkea_enable)
2461 if (mdd_obj->mod_flags & DEAD_OBJ)
2462 /* No more links, don't bother */
2465 buf = mdd_links_get(env, mdd_obj);
2469 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2470 rc, PFID(mdd_object_fid(mdd_obj)));
2472 CERROR("link_ea read failed %d "DFID"\n",
2473 rc, PFID(mdd_object_fid(mdd_obj)));
2477 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2479 /* Find the old record */
2480 for(count = 0; count < leh->leh_reccount; count++) {
2481 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2482 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2483 lu_fid_eq(tmpfid, oldpfid) &&
2484 (strncmp(tmpname->ln_name, oldlname->ln_name,
2485 tmpname->ln_namelen) == 0))
2487 lee = (struct link_ea_entry *)((char *)lee + reclen);
2489 if ((count + 1) > leh->leh_reccount) {
2490 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2491 oldlname->ln_namelen, oldlname->ln_name);
2492 GOTO(out, rc = -ENOENT);
2495 /* Remove the old record */
2496 leh->leh_reccount--;
2497 leh->leh_len -= reclen;
2498 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2501 /* If renaming, add the new record */
2502 if (newpfid != NULL) {
2503 /* if the add fails, we still delete the out-of-date old link */
2504 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2508 rc = __mdd_xattr_set(env, mdd_obj,
2509 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2510 XATTR_NAME_LINK, 0, handle);
2516 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2517 oldlname->ln_namelen, oldlname->ln_name, rc,
2518 PFID(mdd_object_fid(mdd_obj)));
2520 if (buf->lb_vmalloc)
2521 /* if we vmalloced a large buffer drop it */
2527 const struct md_dir_operations mdd_dir_ops = {
2528 .mdo_is_subdir = mdd_is_subdir,
2529 .mdo_lookup = mdd_lookup,
2530 .mdo_create = mdd_create,
2531 .mdo_rename = mdd_rename,
2532 .mdo_link = mdd_link,
2533 .mdo_unlink = mdd_unlink,
2534 .mdo_name_insert = mdd_name_insert,
2535 .mdo_name_remove = mdd_name_remove,
2536 .mdo_rename_tgt = mdd_rename_tgt,
2537 .mdo_create_data = mdd_create_data