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, LUDA_64BITHASH, 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?
273 result = PTR_ERR(it);
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,
623 struct mdd_object *target,
624 struct mdd_object *parent,
625 const struct lu_fid *tf,
626 const struct lu_name *tname,
627 struct thandle *handle)
629 const struct lu_fid *tfid;
630 const struct lu_fid *tpfid = mdo2fid(parent);
631 struct llog_changelog_rec *rec;
638 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
640 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
643 LASSERT(parent != NULL);
644 LASSERT(tname != NULL);
645 LASSERT(handle != NULL);
648 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
649 buf = mdd_buf_alloc(env, reclen);
650 if (buf->lb_buf == NULL)
652 rec = (struct llog_changelog_rec *)buf->lb_buf;
654 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
655 rec->cr.cr_type = (__u32)type;
656 tfid = tf ? tf : mdo2fid(target);
657 rec->cr.cr_tfid = *tfid;
658 rec->cr.cr_pfid = *tpfid;
659 rec->cr.cr_namelen = tname->ln_namelen;
660 memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
662 target->mod_cltime = cfs_time_current_64();
664 rc = mdd_changelog_llog_write(mdd, rec, handle);
666 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
667 rc, type, tname->ln_name, PFID(tfid), PFID(tpfid));
674 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
675 struct md_object *src_obj, const struct lu_name *lname,
678 const char *name = lname->ln_name;
679 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
680 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
681 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
682 struct mdd_device *mdd = mdo2mdd(src_obj);
683 struct dynlock_handle *dlh;
684 struct thandle *handle;
685 #ifdef HAVE_QUOTA_SUPPORT
686 struct obd_device *obd = mdd->mdd_obd_dev;
687 struct obd_export *exp = md_quota(env)->mq_exp;
688 struct mds_obd *mds = &obd->u.mds;
689 unsigned int qids[MAXQUOTAS] = { 0, 0 };
690 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
695 #ifdef HAVE_QUOTA_SUPPORT
696 if (mds->mds_quota) {
697 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
699 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
702 mdd_data_get(env, mdd_tobj, &data);
703 quota_opc = FSFILT_OP_LINK;
704 mdd_quota_wrapper(la_tmp, qids);
705 /* get block quota for parent */
706 lquota_chkquota(mds_quota_interface_ref, obd, exp,
707 qids, rec_pending, 1, NULL,
708 LQUOTA_FLAGS_BLK, data, 1);
713 mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
714 handle = mdd_trans_start(env, mdd);
716 GOTO(out_pending, rc = PTR_ERR(handle));
718 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
720 GOTO(out_trans, rc = -ENOMEM);
721 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
723 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
725 GOTO(out_unlock, rc);
727 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
729 mdd_object_capa(env, mdd_tobj));
731 GOTO(out_unlock, rc);
733 __mdd_ref_add(env, mdd_sobj, handle);
735 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
736 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
738 la->la_valid = LA_CTIME | LA_MTIME;
739 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
741 GOTO(out_unlock, rc);
743 la->la_valid = LA_CTIME;
744 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
746 mdd_links_add(env, mdd_sobj,
747 mdo2fid(mdd_tobj), lname, handle, 0);
752 mdd_write_unlock(env, mdd_sobj);
753 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
756 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
757 mdd_tobj, NULL, lname, handle);
758 mdd_trans_stop(env, mdd, rc, handle);
760 #ifdef HAVE_QUOTA_SUPPORT
762 lquota_pending_commit(mds_quota_interface_ref, obd,
763 qids, rec_pending, 1);
764 /* Trigger dqacq for the parent owner. If failed,
765 * the next call for lquota_chkquota will process it. */
766 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
773 /* caller should take a lock before calling */
774 int mdd_finish_unlink(const struct lu_env *env,
775 struct mdd_object *obj, struct md_attr *ma,
782 LASSERT(mdd_write_locked(env, obj) != 0);
784 /* read HSM flags, needed to set changelogs flags */
785 ma->ma_need = MA_HSM | MA_INODE;
786 rc = mdd_attr_get_internal(env, obj, ma);
787 if (rc == 0 && ma->ma_attr.la_nlink == 0) {
788 obj->mod_flags |= DEAD_OBJ;
789 /* add new orphan and the object
790 * will be deleted during mdd_close() */
791 if (obj->mod_count) {
792 rc = __mdd_orphan_add(env, obj, th);
794 CDEBUG(D_HA, "Object "DFID" is inserted into "
795 "orphan list, open count = %d\n",
796 PFID(mdd_object_fid(obj)),
799 CERROR("Object "DFID" fail to be an orphan, "
800 "open count = %d, maybe cause failed "
802 PFID(mdd_object_fid(obj)),
805 rc = mdd_object_kill(env, obj, ma);
812 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
819 * has mdd_write_lock on cobj already, but not on pobj yet
821 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
822 struct mdd_object *cobj, struct md_attr *ma)
827 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
832 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
833 struct md_object *cobj, const struct lu_name *lname,
836 const char *name = lname->ln_name;
837 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
838 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
839 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
840 struct mdd_device *mdd = mdo2mdd(pobj);
841 struct dynlock_handle *dlh;
842 struct thandle *handle;
843 #ifdef HAVE_QUOTA_SUPPORT
844 struct obd_device *obd = mdd->mdd_obd_dev;
845 struct mds_obd *mds = &obd->u.mds;
846 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
847 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
850 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
854 LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
855 PFID(mdd_object_fid(mdd_cobj)));
857 rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
861 handle = mdd_trans_start(env, mdd);
863 RETURN(PTR_ERR(handle));
865 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
867 GOTO(out_trans, rc = -ENOMEM);
868 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
870 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
874 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
875 mdd_object_capa(env, mdd_pobj));
879 __mdd_ref_del(env, mdd_cobj, handle, 0);
882 __mdd_ref_del(env, mdd_cobj, handle, 1);
884 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
885 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
887 la->la_valid = LA_CTIME | LA_MTIME;
888 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
892 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
893 /* update ctime of an unlinked file only if it is still
894 * opened or a link still exists */
895 la->la_valid = LA_CTIME;
896 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
901 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
902 #ifdef HAVE_QUOTA_SUPPORT
903 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
904 ma->ma_attr.la_nlink == 0) {
905 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
907 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
909 mdd_quota_wrapper(la_tmp, qpids);
910 if (mdd_cobj->mod_count == 0) {
911 quota_opc = FSFILT_OP_UNLINK;
912 mdd_quota_wrapper(&ma->ma_attr, qcids);
914 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
920 /* old files may not have link ea; ignore errors */
921 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
922 lname, NULL, NULL, handle);
926 mdd_write_unlock(env, mdd_cobj);
927 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
932 cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
933 if ((ma->ma_valid & MA_HSM) &&
934 (ma->ma_hsm.mh_flags & HS_EXISTS))
935 cl_flags |= CLF_UNLINK_HSM_EXISTS;
937 rc = mdd_changelog_ns_store(env, mdd,
938 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
939 mdd_cobj, mdd_pobj, NULL, lname, handle);
942 mdd_trans_stop(env, mdd, rc, handle);
943 #ifdef HAVE_QUOTA_SUPPORT
945 /* Trigger dqrel on the owner of child and parent. If failed,
946 * the next call for lquota_chkquota will process it. */
947 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
953 /* has not lock on pobj yet */
954 static int mdd_ni_sanity_check(const struct lu_env *env,
955 struct md_object *pobj,
956 const struct md_attr *ma)
958 struct mdd_object *obj = md2mdd_obj(pobj);
962 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
965 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
973 static int mdd_name_insert(const struct lu_env *env,
974 struct md_object *pobj,
975 const struct lu_name *lname,
976 const struct lu_fid *fid,
977 const struct md_attr *ma)
979 const char *name = lname->ln_name;
980 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
981 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
982 struct mdd_device *mdd = mdo2mdd(pobj);
983 struct dynlock_handle *dlh;
984 struct thandle *handle;
985 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
986 #ifdef HAVE_QUOTA_SUPPORT
987 struct md_ucred *uc = md_ucred(env);
988 struct obd_device *obd = mdd->mdd_obd_dev;
989 struct obd_export *exp = md_quota(env)->mq_exp;
990 struct mds_obd *mds = &obd->u.mds;
991 unsigned int qids[MAXQUOTAS] = { 0, 0 };
992 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
993 cfs_cap_t save = uc->mu_cap;
998 #ifdef HAVE_QUOTA_SUPPORT
999 if (mds->mds_quota) {
1000 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
1001 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1003 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1006 mdd_data_get(env, mdd_obj, &data);
1007 quota_opc = FSFILT_OP_LINK;
1008 mdd_quota_wrapper(la_tmp, qids);
1009 /* get block quota for parent */
1010 lquota_chkquota(mds_quota_interface_ref, obd,
1011 exp, qids, rec_pending, 1, NULL,
1012 LQUOTA_FLAGS_BLK, data, 1);
1015 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1019 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
1020 handle = mdd_trans_start(env, mdo2mdd(pobj));
1022 GOTO(out_pending, rc = PTR_ERR(handle));
1024 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1026 GOTO(out_trans, rc = -ENOMEM);
1028 rc = mdd_ni_sanity_check(env, pobj, ma);
1030 GOTO(out_unlock, rc);
1032 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1033 handle, BYPASS_CAPA);
1035 GOTO(out_unlock, rc);
1038 * For some case, no need update obj's ctime (LA_CTIME is not set),
1040 * For other cases, update obj's ctime (LA_CTIME is set),
1043 if (ma->ma_attr.la_valid & LA_CTIME) {
1044 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1045 la->la_valid = LA_CTIME | LA_MTIME;
1046 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1051 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1053 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1055 #ifdef HAVE_QUOTA_SUPPORT
1056 if (mds->mds_quota) {
1058 lquota_pending_commit(mds_quota_interface_ref,
1059 obd, qids, rec_pending, 1);
1060 /* Trigger dqacq for the parent owner. If failed,
1061 * the next call for lquota_chkquota will process it*/
1062 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1072 /* has not lock on pobj yet */
1073 static int mdd_nr_sanity_check(const struct lu_env *env,
1074 struct md_object *pobj,
1075 const struct md_attr *ma)
1077 struct mdd_object *obj = md2mdd_obj(pobj);
1081 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1084 rc = mdd_may_unlink(env, obj, ma);
1090 * Partial operation.
1092 static int mdd_name_remove(const struct lu_env *env,
1093 struct md_object *pobj,
1094 const struct lu_name *lname,
1095 const struct md_attr *ma)
1097 const char *name = lname->ln_name;
1098 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1099 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1100 struct mdd_device *mdd = mdo2mdd(pobj);
1101 struct dynlock_handle *dlh;
1102 struct thandle *handle;
1103 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1104 #ifdef HAVE_QUOTA_SUPPORT
1105 struct obd_device *obd = mdd->mdd_obd_dev;
1106 struct mds_obd *mds = &obd->u.mds;
1107 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1113 #ifdef HAVE_QUOTA_SUPPORT
1114 if (mds->mds_quota) {
1115 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1117 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1119 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1120 mdd_quota_wrapper(la_tmp, qids);
1124 mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
1125 handle = mdd_trans_start(env, mdd);
1127 GOTO(out_pending, rc = PTR_ERR(handle));
1129 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1131 GOTO(out_trans, rc = -ENOMEM);
1133 rc = mdd_nr_sanity_check(env, pobj, ma);
1135 GOTO(out_unlock, rc);
1137 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1138 handle, BYPASS_CAPA);
1140 GOTO(out_unlock, rc);
1143 * For some case, no need update obj's ctime (LA_CTIME is not set),
1145 * For other cases, update obj's ctime (LA_CTIME is set),
1148 if (ma->ma_attr.la_valid & LA_CTIME) {
1149 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1150 la->la_valid = LA_CTIME | LA_MTIME;
1151 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1156 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1158 mdd_trans_stop(env, mdd, rc, handle);
1160 #ifdef HAVE_QUOTA_SUPPORT
1161 /* Trigger dqrel for the parent owner.
1162 * If failed, the next call for lquota_chkquota will process it. */
1164 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1172 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1174 static int mdd_rt_sanity_check(const struct lu_env *env,
1175 struct mdd_object *tgt_pobj,
1176 struct mdd_object *tobj,
1182 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1185 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1186 * exist. In fact, tobj must exist, otherwise the call trace will be:
1187 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1188 * When get here, tobj must be NOT NULL, the other case has been
1189 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1191 * So check may_delete, but not check nlink of tgt_pobj. */
1193 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1198 /* Partial rename op on slave MDD */
1199 static int mdd_rename_tgt(const struct lu_env *env,
1200 struct md_object *pobj, struct md_object *tobj,
1201 const struct lu_fid *lf, const struct lu_name *lname,
1204 const char *name = lname->ln_name;
1205 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1206 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1207 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1208 struct mdd_device *mdd = mdo2mdd(pobj);
1209 struct dynlock_handle *dlh;
1210 struct thandle *handle;
1211 #ifdef HAVE_QUOTA_SUPPORT
1212 struct obd_device *obd = mdd->mdd_obd_dev;
1213 struct obd_export *exp = md_quota(env)->mq_exp;
1214 struct mds_obd *mds = &obd->u.mds;
1215 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1216 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1217 int quota_copc = 0, quota_popc = 0;
1218 int rec_pending[MAXQUOTAS] = { 0, 0 };
1223 #ifdef HAVE_QUOTA_SUPPORT
1224 if (mds->mds_quota && !tobj) {
1225 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1227 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1230 mdd_data_get(env, mdd_tpobj, &data);
1231 quota_popc = FSFILT_OP_LINK;
1232 mdd_quota_wrapper(la_tmp, qpids);
1233 /* get block quota for target parent */
1234 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1235 qpids, rec_pending, 1, NULL,
1236 LQUOTA_FLAGS_BLK, data, 1);
1240 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
1241 handle = mdd_trans_start(env, mdd);
1243 GOTO(out_pending, rc = PTR_ERR(handle));
1245 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1247 GOTO(out_trans, rc = -ENOMEM);
1249 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1251 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1256 * If rename_tgt is called then we should just re-insert name with
1257 * correct fid, no need to dec/inc parent nlink if obj is dir.
1259 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1263 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1268 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1269 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1271 la->la_valid = LA_CTIME | LA_MTIME;
1272 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1277 * For tobj is remote case cmm layer has processed
1278 * and pass NULL tobj to here. So when tobj is NOT NULL,
1279 * it must be local one.
1281 if (tobj && mdd_object_exists(mdd_tobj)) {
1282 __mdd_ref_del(env, mdd_tobj, handle, 0);
1284 /* Remove dot reference. */
1285 if (S_ISDIR(ma->ma_attr.la_mode))
1286 __mdd_ref_del(env, mdd_tobj, handle, 1);
1288 la->la_valid = LA_CTIME;
1289 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1293 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1297 #ifdef HAVE_QUOTA_SUPPORT
1298 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1299 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
1300 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1301 mdd_quota_wrapper(&ma->ma_attr, qcids);
1308 mdd_write_unlock(env, mdd_tobj);
1309 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1312 /* Bare EXT record with no RENAME in front of it signifies
1313 a partial slave op */
1314 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
1315 mdd_tpobj, NULL, lname, handle);
1317 mdd_trans_stop(env, mdd, rc, handle);
1319 #ifdef HAVE_QUOTA_SUPPORT
1320 if (mds->mds_quota) {
1322 lquota_pending_commit(mds_quota_interface_ref, obd,
1323 qpids, rec_pending, 1);
1326 /* Trigger dqrel on the target owner of child.
1327 * If failed, the next call for lquota_chkquota
1328 * will process it. */
1329 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1337 * The permission has been checked when obj created, no need check again.
1339 static int mdd_cd_sanity_check(const struct lu_env *env,
1340 struct mdd_object *obj)
1345 if (!obj || mdd_is_dead_obj(obj))
1352 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1353 struct md_object *cobj, const struct md_op_spec *spec,
1356 struct mdd_device *mdd = mdo2mdd(cobj);
1357 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1358 struct mdd_object *son = md2mdd_obj(cobj);
1359 struct lu_attr *attr = &ma->ma_attr;
1360 struct lov_mds_md *lmm = NULL;
1362 struct thandle *handle;
1366 rc = mdd_cd_sanity_check(env, son);
1370 if (!md_should_create(spec->sp_cr_flags))
1372 lmm_size = ma->ma_lmm_size;
1373 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1378 mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
1379 handle = mdd_trans_start(env, mdd);
1381 GOTO(out_free, rc = PTR_ERR(handle));
1384 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1385 * Should this be fixed?
1388 /* Replay creates has objects already */
1390 if (spec->no_create) {
1391 CDEBUG(D_INFO, "we already have lov ea\n");
1392 rc = mdd_lov_set_md(env, mdd_pobj, son,
1393 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1394 spec->u.sp_ea.eadatalen, handle, 0);
1397 /* No need mdd_lsm_sanity_check here */
1398 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1399 lmm_size, handle, 0);
1402 rc = mdd_attr_get_internal_locked(env, son, ma);
1404 /* update lov_objid data, must be before transaction stop! */
1406 mdd_lov_objid_update(mdd, lmm);
1408 mdd_trans_stop(env, mdd, rc, handle);
1410 /* Finish mdd_lov_create() stuff. */
1411 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1415 /* Get fid from name and parent */
1417 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1418 const struct lu_name *lname, struct lu_fid* fid, int mask)
1420 const char *name = lname->ln_name;
1421 const struct dt_key *key = (const struct dt_key *)name;
1422 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1423 struct mdd_device *m = mdo2mdd(pobj);
1424 struct dt_object *dir = mdd_object_child(mdd_obj);
1428 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1431 rc = mdd_object_exists(mdd_obj);
1432 if (unlikely(rc == 0))
1434 else if (unlikely(rc < 0)) {
1435 CERROR("Object "DFID" locates on remote server\n",
1436 PFID(mdo2fid(mdd_obj)));
1440 /* The common filename length check. */
1441 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1442 RETURN(-ENAMETOOLONG);
1444 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1449 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1450 dt_try_as_dir(env, dir))) {
1452 rc = dir->do_index_ops->dio_lookup(env, dir,
1453 (struct dt_rec *)fid, key,
1454 mdd_object_capa(env, mdd_obj));
1465 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1466 const struct lu_name *lname, struct mdd_object *child,
1467 struct md_attr *ma, struct thandle *handle,
1468 const struct md_op_spec *spec)
1474 * Update attributes for child.
1477 * (1) the valid bits should be converted between Lustre and Linux;
1478 * (2) maybe, the child attributes should be set in OSD when creation.
1481 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1485 if (S_ISDIR(ma->ma_attr.la_mode)) {
1486 /* Add "." and ".." for newly created dir */
1487 __mdd_ref_add(env, child, handle);
1488 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1489 dot, handle, BYPASS_CAPA);
1491 rc = __mdd_index_insert_only(env, child, pfid,
1495 __mdd_ref_del(env, child, handle, 1);
1498 mdd_links_add(env, child, pfid, lname, handle, 1);
1503 /* has not lock on pobj yet */
1504 static int mdd_create_sanity_check(const struct lu_env *env,
1505 struct md_object *pobj,
1506 const struct lu_name *lname,
1508 struct md_op_spec *spec)
1510 struct mdd_thread_info *info = mdd_env_info(env);
1511 struct lu_attr *la = &info->mti_la;
1512 struct lu_fid *fid = &info->mti_fid;
1513 struct mdd_object *obj = md2mdd_obj(pobj);
1514 struct mdd_device *m = mdo2mdd(pobj);
1515 int lookup = spec->sp_cr_lookup;
1520 if (mdd_is_dead_obj(obj))
1524 * In some cases this lookup is not needed - we know before if name
1525 * exists or not because MDT performs lookup for it.
1526 * name length check is done in lookup.
1530 * Check if the name already exist, though it will be checked in
1531 * _index_insert also, for avoiding rolling back if exists
1534 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1535 MAY_WRITE | MAY_EXEC);
1537 RETURN(rc ? : -EEXIST);
1540 * Check WRITE permission for the parent.
1541 * EXEC permission have been checked
1542 * when lookup before create already.
1544 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1551 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1555 if (la->la_mode & S_ISGID) {
1556 ma->ma_attr.la_gid = la->la_gid;
1557 if (S_ISDIR(ma->ma_attr.la_mode)) {
1558 ma->ma_attr.la_mode |= S_ISGID;
1559 ma->ma_attr.la_valid |= LA_MODE;
1563 switch (ma->ma_attr.la_mode & S_IFMT) {
1565 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1567 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1568 RETURN(-ENAMETOOLONG);
1588 * Create object and insert it into namespace.
1590 static int mdd_create(const struct lu_env *env,
1591 struct md_object *pobj,
1592 const struct lu_name *lname,
1593 struct md_object *child,
1594 struct md_op_spec *spec,
1597 struct mdd_thread_info *info = mdd_env_info(env);
1598 struct lu_attr *la = &info->mti_la_for_fix;
1599 struct md_attr *ma_acl = &info->mti_ma;
1600 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1601 struct mdd_object *son = md2mdd_obj(child);
1602 struct mdd_device *mdd = mdo2mdd(pobj);
1603 struct lu_attr *attr = &ma->ma_attr;
1604 struct lov_mds_md *lmm = NULL;
1605 struct thandle *handle;
1606 struct dynlock_handle *dlh;
1607 const char *name = lname->ln_name;
1608 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1609 int got_def_acl = 0;
1610 #ifdef HAVE_QUOTA_SUPPORT
1611 struct obd_device *obd = mdd->mdd_obd_dev;
1612 struct obd_export *exp = md_quota(env)->mq_exp;
1613 struct mds_obd *mds = &obd->u.mds;
1614 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1615 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1616 int quota_opc = 0, block_count = 0;
1617 int inode_pending[MAXQUOTAS] = { 0, 0 };
1618 int block_pending[MAXQUOTAS] = { 0, 0 };
1619 int parent_pending[MAXQUOTAS] = { 0, 0 };
1624 * Two operations have to be performed:
1626 * - an allocation of a new object (->do_create()), and
1628 * - an insertion into a parent index (->dio_insert()).
1630 * Due to locking, operation order is not important, when both are
1631 * successful, *but* error handling cases are quite different:
1633 * - if insertion is done first, and following object creation fails,
1634 * insertion has to be rolled back, but this operation might fail
1635 * also leaving us with dangling index entry.
1637 * - if creation is done first, is has to be undone if insertion
1638 * fails, leaving us with leaked space, which is neither good, nor
1641 * It seems that creation-first is simplest solution, but it is
1642 * sub-optimal in the frequent
1647 * case, because second mkdir is bound to create object, only to
1648 * destroy it immediately.
1650 * To avoid this follow local file systems that do double lookup:
1652 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
1654 * 1. create (mdd_object_create_internal())
1656 * 2. insert (__mdd_index_insert(), lookup again)
1659 /* Sanity checks before big job. */
1660 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
1664 #ifdef HAVE_QUOTA_SUPPORT
1665 if (mds->mds_quota) {
1666 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1668 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1672 quota_opc = FSFILT_OP_CREATE;
1673 mdd_quota_wrapper(&ma->ma_attr, qcids);
1674 mdd_quota_wrapper(la_tmp, qpids);
1675 /* get file quota for child */
1676 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1677 qcids, inode_pending, 1, NULL, 0, NULL,
1679 switch (ma->ma_attr.la_mode & S_IFMT) {
1688 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
1689 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
1693 /* get block quota for child and parent */
1695 lquota_chkquota(mds_quota_interface_ref, obd,
1696 exp, qcids, block_pending,
1698 LQUOTA_FLAGS_BLK, NULL, 0);
1700 lquota_chkquota(mds_quota_interface_ref, obd,
1701 exp, qpids, parent_pending, 1,
1702 NULL, LQUOTA_FLAGS_BLK, NULL,
1709 * No RPC inside the transaction, so OST objects should be created at
1712 if (S_ISREG(attr->la_mode)) {
1713 lmm_size = ma->ma_lmm_size;
1714 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
1717 GOTO(out_pending, rc);
1720 if (!S_ISLNK(attr->la_mode)) {
1721 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
1722 ma_acl->ma_acl = info->mti_xattr_buf;
1723 ma_acl->ma_need = MA_ACL_DEF;
1724 ma_acl->ma_valid = 0;
1726 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
1727 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
1728 mdd_read_unlock(env, mdd_pobj);
1731 else if (ma_acl->ma_valid & MA_ACL_DEF)
1735 mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
1736 handle = mdd_trans_start(env, mdd);
1738 GOTO(out_free, rc = PTR_ERR(handle));
1740 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1742 GOTO(out_trans, rc = -ENOMEM);
1744 mdd_write_lock(env, son, MOR_TGT_CHILD);
1745 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
1747 mdd_write_unlock(env, son);
1753 #ifdef CONFIG_FS_POSIX_ACL
1755 struct lu_buf *acl_buf = &info->mti_buf;
1756 acl_buf->lb_buf = ma_acl->ma_acl;
1757 acl_buf->lb_len = ma_acl->ma_acl_size;
1759 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
1761 mdd_write_unlock(env, son);
1764 ma->ma_attr.la_valid |= LA_MODE;
1769 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
1770 son, ma, handle, spec);
1771 mdd_write_unlock(env, son);
1774 * Object has no links, so it will be destroyed when last
1775 * reference is released. (XXX not now.)
1781 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
1782 name, S_ISDIR(attr->la_mode), handle,
1783 mdd_object_capa(env, mdd_pobj));
1790 /* No need mdd_lsm_sanity_check here */
1791 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
1793 CERROR("error on stripe info copy %d \n", rc);
1796 if (lmm && lmm_size > 0) {
1797 /* Set Lov here, do not get lmm again later */
1798 memcpy(ma->ma_lmm, lmm, lmm_size);
1799 ma->ma_lmm_size = lmm_size;
1800 ma->ma_valid |= MA_LOV;
1803 if (S_ISLNK(attr->la_mode)) {
1804 struct md_ucred *uc = md_ucred(env);
1805 struct dt_object *dt = mdd_object_child(son);
1806 const char *target_name = spec->u.sp_symname;
1807 int sym_len = strlen(target_name);
1808 const struct lu_buf *buf;
1811 buf = mdd_buf_get_const(env, target_name, sym_len);
1812 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
1813 mdd_object_capa(env, son),
1815 CFS_CAP_SYS_RESOURCE_MASK);
1820 GOTO(cleanup, rc = -EFAULT);
1824 la->la_valid = LA_CTIME | LA_MTIME;
1825 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1829 /* Return attr back. */
1830 rc = mdd_attr_get_internal_locked(env, son, ma);
1833 if (rc && created) {
1837 rc2 = __mdd_index_delete(env, mdd_pobj, name,
1838 S_ISDIR(attr->la_mode),
1839 handle, BYPASS_CAPA);
1841 CERROR("error can not cleanup destroy %d\n",
1846 mdd_write_lock(env, son, MOR_TGT_CHILD);
1847 __mdd_ref_del(env, son, handle, 0);
1848 if (initialized && S_ISDIR(attr->la_mode))
1849 __mdd_ref_del(env, son, handle, 1);
1850 mdd_write_unlock(env, son);
1854 /* update lov_objid data, must be before transaction stop! */
1856 mdd_lov_objid_update(mdd, lmm);
1858 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1861 rc = mdd_changelog_ns_store(env, mdd,
1862 S_ISDIR(attr->la_mode) ? CL_MKDIR :
1863 S_ISREG(attr->la_mode) ? CL_CREATE :
1864 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
1865 0, son, mdd_pobj, NULL, lname, handle);
1866 mdd_trans_stop(env, mdd, rc, handle);
1868 /* finish lov_create stuff, free all temporary data */
1869 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1871 #ifdef HAVE_QUOTA_SUPPORT
1873 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1875 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
1877 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
1879 /* Trigger dqacq on the owner of child and parent. If failed,
1880 * the next call for lquota_chkquota will process it. */
1881 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1889 * Get locks on parents in proper order
1890 * RETURN: < 0 - error, rename_order if successful
1898 static int mdd_rename_order(const struct lu_env *env,
1899 struct mdd_device *mdd,
1900 struct mdd_object *src_pobj,
1901 struct mdd_object *tgt_pobj)
1903 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
1907 if (src_pobj == tgt_pobj)
1908 RETURN(MDD_RN_SAME);
1910 /* compared the parent child relationship of src_p&tgt_p */
1911 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
1913 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
1916 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
1929 /* has not mdd_write{read}_lock on any obj yet. */
1930 static int mdd_rename_sanity_check(const struct lu_env *env,
1931 struct mdd_object *src_pobj,
1932 struct mdd_object *tgt_pobj,
1933 struct mdd_object *sobj,
1934 struct mdd_object *tobj,
1940 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1943 /* XXX: when get here, sobj must NOT be NULL,
1944 * the other case has been processed in cml_rename
1945 * before mdd_rename and enable MDS_PERM_BYPASS. */
1948 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
1952 /* XXX: when get here, "tobj == NULL" means tobj must
1953 * NOT exist (neither on remote MDS, such case has been
1954 * processed in cml_rename before mdd_rename and enable
1956 * So check may_create, but not check may_unlink. */
1958 rc = mdd_may_create(env, tgt_pobj, NULL,
1959 (src_pobj != tgt_pobj), 0);
1961 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
1962 (src_pobj != tgt_pobj), 1);
1964 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
1965 S_ISDIR(ma->ma_attr.la_mode))
1966 rc = __mdd_may_link(env, tgt_pobj);
1971 /* src object can be remote that is why we use only fid and type of object */
1972 static int mdd_rename(const struct lu_env *env,
1973 struct md_object *src_pobj, struct md_object *tgt_pobj,
1974 const struct lu_fid *lf, const struct lu_name *lsname,
1975 struct md_object *tobj, const struct lu_name *ltname,
1978 const char *sname = lsname->ln_name;
1979 const char *tname = ltname->ln_name;
1980 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1981 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
1982 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
1983 struct mdd_device *mdd = mdo2mdd(src_pobj);
1984 struct mdd_object *mdd_sobj = NULL; /* source object */
1985 struct mdd_object *mdd_tobj = NULL;
1986 struct dynlock_handle *sdlh, *tdlh;
1987 struct thandle *handle;
1988 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
1989 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
1993 #ifdef HAVE_QUOTA_SUPPORT
1994 struct obd_device *obd = mdd->mdd_obd_dev;
1995 struct obd_export *exp = md_quota(env)->mq_exp;
1996 struct mds_obd *mds = &obd->u.mds;
1997 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
1998 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
1999 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2000 int quota_copc = 0, quota_popc = 0;
2001 int rec_pending[MAXQUOTAS] = { 0, 0 };
2005 LASSERT(ma->ma_attr.la_mode & S_IFMT);
2006 is_dir = S_ISDIR(ma->ma_attr.la_mode);
2009 mdd_tobj = md2mdd_obj(tobj);
2011 #ifdef HAVE_QUOTA_SUPPORT
2012 if (mds->mds_quota) {
2013 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2015 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2017 mdd_quota_wrapper(la_tmp, qspids);
2019 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2023 mdd_data_get(env, mdd_tpobj, &data);
2024 quota_popc = FSFILT_OP_LINK;
2025 mdd_quota_wrapper(la_tmp, qtpids);
2026 /* get block quota for target parent */
2027 lquota_chkquota(mds_quota_interface_ref,
2029 rec_pending, 1, NULL,
2037 mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
2038 handle = mdd_trans_start(env, mdd);
2040 GOTO(out_pending, rc = PTR_ERR(handle));
2042 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2043 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2045 GOTO(cleanup_unlocked, rc);
2047 /* Get locks in determined order */
2048 if (rc == MDD_RN_SAME) {
2049 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2050 sname, MOR_SRC_PARENT);
2051 /* check hashes to determine do we need one lock or two */
2052 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2053 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2057 } else if (rc == MDD_RN_SRCTGT) {
2058 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2059 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2061 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2062 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2064 if (sdlh == NULL || tdlh == NULL)
2065 GOTO(cleanup, rc = -ENOMEM);
2067 mdd_sobj = mdd_object_find(env, mdd, lf);
2068 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2069 mdd_sobj, mdd_tobj, ma);
2073 /* Remove source name from source directory */
2074 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2075 mdd_object_capa(env, mdd_spobj));
2079 /* "mv dir1 dir2" needs "dir1/.." link update */
2080 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2081 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2082 mdd_object_capa(env, mdd_sobj));
2084 GOTO(fixup_spobj2, rc);
2086 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2087 handle, mdd_object_capa(env, mdd_sobj));
2089 GOTO(fixup_spobj, rc);
2092 /* Remove target name from target directory
2093 * Here tobj can be remote one, so we do index_delete unconditionally
2094 * and -ENOENT is allowed.
2096 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2097 mdd_object_capa(env, mdd_tpobj));
2100 /* tname might been renamed to something else */
2101 GOTO(fixup_spobj, rc);
2104 GOTO(fixup_spobj, rc);
2107 /* Insert new fid with target name into target dir */
2108 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2109 mdd_object_capa(env, mdd_tpobj));
2111 GOTO(fixup_tpobj, rc);
2113 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2114 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2116 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2118 la->la_valid = LA_CTIME;
2119 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2122 GOTO(fixup_tpobj, rc);
2125 /* Remove old target object
2126 * For tobj is remote case cmm layer has processed
2127 * and set tobj to NULL then. So when tobj is NOT NULL,
2128 * it must be local one.
2130 if (tobj && mdd_object_exists(mdd_tobj)) {
2131 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2132 if (mdd_is_dead_obj(mdd_tobj)) {
2133 mdd_write_unlock(env, mdd_tobj);
2134 /* shld not be dead, something is wrong */
2135 CERROR("tobj is dead, something is wrong\n");
2139 __mdd_ref_del(env, mdd_tobj, handle, 0);
2141 /* Remove dot reference. */
2143 __mdd_ref_del(env, mdd_tobj, handle, 1);
2145 la->la_valid = LA_CTIME;
2146 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2148 GOTO(fixup_tpobj, rc);
2150 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2151 mdd_write_unlock(env, mdd_tobj);
2153 GOTO(fixup_tpobj, rc);
2155 #ifdef HAVE_QUOTA_SUPPORT
2156 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
2157 ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
2158 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2159 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2164 la->la_valid = LA_CTIME | LA_MTIME;
2165 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2167 GOTO(fixup_tpobj, rc);
2169 if (mdd_spobj != mdd_tpobj) {
2170 la->la_valid = LA_CTIME | LA_MTIME;
2171 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2175 if (rc == 0 && mdd_sobj) {
2176 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2177 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2178 mdo2fid(mdd_tpobj), ltname, handle);
2180 /* Old files might not have EA entry */
2181 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2183 mdd_write_unlock(env, mdd_sobj);
2184 /* We don't fail the transaction if the link ea can't be
2185 updated -- fid2path will use alternate lookup method. */
2193 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2196 CWARN("tp obj fix error %d\n",rc2);
2198 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2199 !mdd_is_dead_obj(mdd_tobj)) {
2200 rc2 = __mdd_index_insert(env, mdd_tpobj,
2201 mdo2fid(mdd_tobj), tname,
2206 CWARN("tp obj fix error %d\n",rc2);
2211 if (rc && is_dir && mdd_sobj) {
2212 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2216 CWARN("sp obj dotdot delete error %d\n",rc2);
2219 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2220 dotdot, handle, BYPASS_CAPA);
2222 CWARN("sp obj dotdot insert error %d\n",rc2);
2227 rc2 = __mdd_index_insert(env, mdd_spobj,
2228 lf, sname, is_dir, handle, BYPASS_CAPA);
2230 CWARN("sp obj fix error %d\n",rc2);
2233 if (likely(tdlh) && sdlh != tdlh)
2234 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2236 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2239 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
2240 mdd_spobj, lf, lsname, handle);
2242 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
2243 mdd_tpobj, lf, ltname, handle);
2245 mdd_trans_stop(env, mdd, rc, handle);
2247 mdd_object_put(env, mdd_sobj);
2249 #ifdef HAVE_QUOTA_SUPPORT
2250 if (mds->mds_quota) {
2252 lquota_pending_commit(mds_quota_interface_ref, obd,
2253 qtpids, rec_pending, 1);
2256 /* Trigger dqrel on the source owner of parent.
2257 * If failed, the next call for lquota_chkquota will
2259 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2260 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2262 /* Trigger dqrel on the target owner of child.
2263 * If failed, the next call for lquota_chkquota
2264 * will process it. */
2265 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2266 qtpids, rc, quota_copc);
2273 /** enable/disable storing of hardlink info */
2274 int mdd_linkea_enable = 1;
2275 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2276 "record hardlink info in EAs");
2278 /** Read the link EA into a temp buffer.
2279 * Uses the name_buf since it is generally large.
2280 * \retval IS_ERR err
2281 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2283 struct lu_buf *mdd_links_get(const struct lu_env *env,
2284 struct mdd_object *mdd_obj)
2287 struct lustre_capa *capa;
2288 struct link_ea_header *leh;
2291 /* First try a small buf */
2292 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2293 if (buf->lb_buf == NULL)
2294 return ERR_PTR(-ENOMEM);
2296 capa = mdd_object_capa(env, mdd_obj);
2297 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2298 if (rc == -ERANGE) {
2299 /* Buf was too small, figure out what we need. */
2301 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2304 buf = mdd_buf_alloc(env, rc);
2305 if (buf->lb_buf == NULL)
2306 return ERR_PTR(-ENOMEM);
2307 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2313 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2314 leh->leh_magic = LINK_EA_MAGIC;
2315 leh->leh_reccount = __swab32(leh->leh_reccount);
2316 leh->leh_len = __swab64(leh->leh_len);
2317 /* entries are swabbed by mdd_lee_unpack */
2319 if (leh->leh_magic != LINK_EA_MAGIC)
2320 return ERR_PTR(-EINVAL);
2321 if (leh->leh_reccount == 0)
2322 return ERR_PTR(-ENODATA);
2327 /** Pack a link_ea_entry.
2328 * All elements are stored as chars to avoid alignment issues.
2329 * Numbers are always big-endian
2330 * \retval record length
2332 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2333 const struct lu_fid *pfid)
2335 struct lu_fid tmpfid;
2338 fid_cpu_to_be(&tmpfid, pfid);
2339 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2340 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2341 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2343 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2344 lee->lee_reclen[1] = reclen & 0xff;
2348 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2349 struct lu_name *lname, struct lu_fid *pfid)
2351 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2352 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2353 fid_be_to_cpu(pfid, pfid);
2354 lname->ln_name = lee->lee_name;
2355 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2358 /** Add a record to the end of link ea buf */
2359 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2360 const struct lu_fid *pfid,
2361 const struct lu_name *lname)
2363 struct link_ea_header *leh;
2364 struct link_ea_entry *lee;
2367 if (lname == NULL || pfid == NULL)
2370 /* Make sure our buf is big enough for the new one */
2372 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2373 if (leh->leh_len + reclen > buf->lb_len) {
2374 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2379 lee = buf->lb_buf + leh->leh_len;
2380 reclen = mdd_lee_pack(lee, lname, pfid);
2381 leh->leh_len += reclen;
2382 leh->leh_reccount++;
2386 /* For pathologic linkers, we don't want to spend lots of time scanning the
2387 * link ea. Limit ourseleves to something reasonable; links not in the EA
2388 * can be looked up via (slower) parent lookup.
2390 #define LINKEA_MAX_COUNT 128
2392 static int mdd_links_add(const struct lu_env *env,
2393 struct mdd_object *mdd_obj,
2394 const struct lu_fid *pfid,
2395 const struct lu_name *lname,
2396 struct thandle *handle, int first)
2399 struct link_ea_header *leh;
2403 if (!mdd_linkea_enable)
2406 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2409 if (rc != -ENODATA) {
2410 CERROR("link_ea read failed %d "DFID"\n", rc,
2411 PFID(mdd_object_fid(mdd_obj)));
2414 /* empty EA; start one */
2415 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2416 if (buf->lb_buf == NULL)
2419 leh->leh_magic = LINK_EA_MAGIC;
2420 leh->leh_len = sizeof(struct link_ea_header);
2421 leh->leh_reccount = 0;
2425 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2428 rc = __mdd_links_add(env, buf, pfid, lname);
2433 rc = __mdd_xattr_set(env, mdd_obj,
2434 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2435 XATTR_NAME_LINK, 0, handle);
2438 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2439 PFID(mdd_object_fid(mdd_obj)));
2441 CERROR("link_ea add failed %d "DFID"\n", rc,
2442 PFID(mdd_object_fid(mdd_obj)));
2445 if (buf->lb_len > OBD_ALLOC_BIG)
2446 /* if we vmalloced a large buffer drop it */
2452 static int mdd_links_rename(const struct lu_env *env,
2453 struct mdd_object *mdd_obj,
2454 const struct lu_fid *oldpfid,
2455 const struct lu_name *oldlname,
2456 const struct lu_fid *newpfid,
2457 const struct lu_name *newlname,
2458 struct thandle *handle)
2461 struct link_ea_header *leh;
2462 struct link_ea_entry *lee;
2463 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
2464 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
2470 if (!mdd_linkea_enable)
2473 if (mdd_obj->mod_flags & DEAD_OBJ)
2474 /* No more links, don't bother */
2477 buf = mdd_links_get(env, mdd_obj);
2481 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
2482 rc, PFID(mdd_object_fid(mdd_obj)));
2484 CERROR("link_ea read failed %d "DFID"\n",
2485 rc, PFID(mdd_object_fid(mdd_obj)));
2489 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
2491 /* Find the old record */
2492 for(count = 0; count < leh->leh_reccount; count++) {
2493 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
2494 if (tmpname->ln_namelen == oldlname->ln_namelen &&
2495 lu_fid_eq(tmpfid, oldpfid) &&
2496 (strncmp(tmpname->ln_name, oldlname->ln_name,
2497 tmpname->ln_namelen) == 0))
2499 lee = (struct link_ea_entry *)((char *)lee + reclen);
2501 if ((count + 1) > leh->leh_reccount) {
2502 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
2503 oldlname->ln_namelen, oldlname->ln_name);
2504 GOTO(out, rc = -ENOENT);
2507 /* Remove the old record */
2508 leh->leh_reccount--;
2509 leh->leh_len -= reclen;
2510 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
2513 /* If renaming, add the new record */
2514 if (newpfid != NULL) {
2515 /* if the add fails, we still delete the out-of-date old link */
2516 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
2520 rc = __mdd_xattr_set(env, mdd_obj,
2521 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2522 XATTR_NAME_LINK, 0, handle);
2528 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
2529 oldlname->ln_namelen, oldlname->ln_name, rc,
2530 PFID(mdd_object_fid(mdd_obj)));
2532 if (buf->lb_len > OBD_ALLOC_BIG)
2533 /* if we vmalloced a large buffer drop it */
2539 const struct md_dir_operations mdd_dir_ops = {
2540 .mdo_is_subdir = mdd_is_subdir,
2541 .mdo_lookup = mdd_lookup,
2542 .mdo_create = mdd_create,
2543 .mdo_rename = mdd_rename,
2544 .mdo_link = mdd_link,
2545 .mdo_unlink = mdd_unlink,
2546 .mdo_name_insert = mdd_name_insert,
2547 .mdo_name_remove = mdd_name_remove,
2548 .mdo_rename_tgt = mdd_rename_tgt,
2549 .mdo_create_data = mdd_create_data,