4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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>
43 #define DEBUG_SUBSYSTEM S_MDS
45 #include <linux/module.h>
46 #include <ldiskfs/ldiskfs_jbd2.h>
48 #include <obd_class.h>
49 #include <lustre_ver.h>
50 #include <obd_support.h>
51 #include <lprocfs_status.h>
52 #include <ldiskfs/ldiskfs.h>
53 #include <lustre_mds.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_fid.h>
57 #include "mdd_internal.h"
59 static const char dot[] = ".";
60 static const char dotdot[] = "..";
62 static struct lu_name lname_dotdot = {
67 static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
68 const struct lu_name *lname, struct lu_fid* fid,
70 static int mdd_declare_links_add(const struct lu_env *env,
71 struct mdd_object *mdd_obj,
72 struct thandle *handle);
73 static int mdd_links_add(const struct lu_env *env,
74 struct mdd_object *mdd_obj,
75 const struct lu_fid *pfid,
76 const struct lu_name *lname,
77 struct thandle *handle, int first);
78 static int mdd_links_rename(const struct lu_env *env,
79 struct mdd_object *mdd_obj,
80 const struct lu_fid *oldpfid,
81 const struct lu_name *oldlname,
82 const struct lu_fid *newpfid,
83 const struct lu_name *newlname,
84 struct thandle *handle);
87 __mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj,
88 const struct lu_name *lname, struct lu_fid* fid, int mask)
90 const char *name = lname->ln_name;
91 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
92 struct dynlock_handle *dlh;
95 dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT);
96 if (unlikely(dlh == NULL))
98 rc = __mdd_lookup(env, pobj, lname, fid, mask);
99 mdd_pdo_read_unlock(env, mdd_obj, dlh);
104 int mdd_lookup(const struct lu_env *env,
105 struct md_object *pobj, const struct lu_name *lname,
106 struct lu_fid* fid, struct md_op_spec *spec)
110 rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC);
114 static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj,
117 return __mdd_lookup_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0);
121 * For root fid use special function, which does not compare version component
122 * of fid. Version component is different for root fids on all MDTs.
124 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
126 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
127 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
131 * return 1: if lf is the fid of the ancestor of p1;
134 * return -EREMOTE: if remote object is found, in this
135 * case fid of remote object is saved to @pf;
137 * otherwise: values < 0, errors.
139 static int mdd_is_parent(const struct lu_env *env,
140 struct mdd_device *mdd,
141 struct mdd_object *p1,
142 const struct lu_fid *lf,
145 struct mdd_object *parent = NULL;
150 LASSERT(!lu_fid_eq(mdo2fid(p1), lf));
151 pfid = &mdd_env_info(env)->mti_fid;
153 /* Check for root first. */
154 if (mdd_is_root(mdd, mdo2fid(p1)))
158 /* this is done recursively, bypass capa for each obj */
159 mdd_set_capainfo(env, 4, p1, BYPASS_CAPA);
160 rc = mdd_parent_fid(env, p1, pfid);
163 if (mdd_is_root(mdd, pfid))
165 if (lu_fid_eq(pfid, lf))
168 mdd_object_put(env, parent);
169 parent = mdd_object_find(env, mdd, pfid);
171 /* cross-ref parent */
172 if (parent == NULL) {
175 GOTO(out, rc = -EREMOTE);
176 } else if (IS_ERR(parent))
177 GOTO(out, rc = PTR_ERR(parent));
182 if (parent && !IS_ERR(parent))
183 mdd_object_put(env, parent);
188 * No permission check is needed.
190 * returns 1: if fid is ancestor of @mo;
191 * returns 0: if fid is not a ancestor of @mo;
193 * returns EREMOTE if remote object is found, fid of remote object is saved to
196 * returns < 0: if error
198 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
199 const struct lu_fid *fid, struct lu_fid *sfid)
201 struct mdd_device *mdd = mdo2mdd(mo);
205 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
208 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid);
212 } else if (rc == 1) {
213 /* found @fid is parent */
221 * Check that @dir contains no entries except (possibly) dot and dotdot.
226 * -ENOTDIR not a directory object
227 * -ENOTEMPTY not empty
231 static int mdd_dir_is_empty(const struct lu_env *env,
232 struct mdd_object *dir)
235 struct dt_object *obj;
236 const struct dt_it_ops *iops;
240 obj = mdd_object_child(dir);
241 if (!dt_try_as_dir(env, obj))
244 iops = &obj->do_index_ops->dio_it;
245 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
247 result = iops->get(env, it, (const void *)"");
250 for (result = 0, i = 0; result == 0 && i < 3; ++i)
251 result = iops->next(env, it);
254 else if (result == +1)
256 } else if (result == 0)
258 * Huh? Index contains no zero key?
265 result = PTR_ERR(it);
269 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj)
271 struct mdd_device *m = mdd_obj2mdd_dev(obj);
272 struct lu_attr *la = &mdd_env_info(env)->mti_la;
276 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
281 * Subdir count limitation can be broken through.
283 if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink &&
284 !S_ISDIR(la->la_mode))
291 * Check whether it may create the cobj under the pobj.
294 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
295 struct mdd_object *cobj, int check_perm, int check_nlink)
300 if (cobj && mdd_object_exists(cobj))
303 if (mdd_is_dead_obj(pobj))
307 rc = mdd_permission_internal_locked(env, pobj, NULL,
308 MAY_WRITE | MAY_EXEC,
311 if (!rc && check_nlink)
312 rc = __mdd_may_link(env, pobj);
318 * Check whether can unlink from the pobj in the case of "cobj == NULL".
320 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
321 const struct md_attr *ma)
326 if (mdd_is_dead_obj(pobj))
329 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
330 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
333 rc = mdd_permission_internal_locked(env, pobj, NULL,
334 MAY_WRITE | MAY_EXEC,
339 if (mdd_is_append(pobj))
346 * pobj == NULL is remote ops case, under such case, pobj's
347 * VTX feature has been checked already, no need check again.
349 static inline int mdd_is_sticky(const struct lu_env *env,
350 struct mdd_object *pobj,
351 struct mdd_object *cobj)
353 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
354 struct md_ucred *uc = md_ucred(env);
358 rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA);
362 if (!(tmp_la->la_mode & S_ISVTX) ||
363 (tmp_la->la_uid == uc->mu_fsuid))
367 rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA);
371 if (tmp_la->la_uid == uc->mu_fsuid)
374 return !mdd_capable(uc, CFS_CAP_FOWNER);
378 * Check whether it may delete the cobj from the pobj.
381 int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj,
382 struct mdd_object *cobj, struct md_attr *ma,
383 int check_perm, int check_empty)
389 if (!mdd_object_exists(cobj))
392 if (mdd_is_dead_obj(cobj))
396 if (!mdd_object_exists(pobj))
399 if (mdd_is_dead_obj(pobj))
403 rc = mdd_permission_internal_locked(env, pobj, NULL,
404 MAY_WRITE | MAY_EXEC,
410 if (mdd_is_append(pobj))
414 if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) &&
415 mdd_is_sticky(env, pobj, cobj))
418 if (mdd_is_immutable(cobj) || mdd_is_append(cobj))
421 if ((ma->ma_attr.la_valid & LA_FLAGS) &&
422 (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)))
425 if (S_ISDIR(ma->ma_attr.la_mode)) {
426 struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj);
428 if (!S_ISDIR(mdd_object_type(cobj)))
431 if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid))
433 } else if (S_ISDIR(mdd_object_type(cobj)))
436 if (S_ISDIR(ma->ma_attr.la_mode) && check_empty)
437 rc = mdd_dir_is_empty(env, cobj);
444 * has mdd_write_lock on src already, but not on tgt yet
446 int mdd_link_sanity_check(const struct lu_env *env,
447 struct mdd_object *tgt_obj,
448 const struct lu_name *lname,
449 struct mdd_object *src_obj)
451 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
455 if (!mdd_object_exists(src_obj))
458 if (mdd_is_dead_obj(src_obj))
461 /* Local ops, no lookup before link, check filename length here. */
462 if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
463 RETURN(-ENAMETOOLONG);
465 if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj))
468 if (S_ISDIR(mdd_object_type(src_obj)))
471 LASSERT(src_obj != tgt_obj);
473 rc = mdd_may_create(env, tgt_obj, NULL, 1, 0);
478 rc = __mdd_may_link(env, src_obj);
483 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
484 const char *name, struct thandle *handle,
485 struct lustre_capa *capa)
487 struct dt_object *next = mdd_object_child(pobj);
491 if (dt_try_as_dir(env, next)) {
492 rc = next->do_index_ops->dio_delete(env, next,
493 (struct dt_key *)name,
501 static int __mdd_index_insert_only(const struct lu_env *env,
502 struct mdd_object *pobj,
503 const struct lu_fid *lf, const char *name,
504 struct thandle *handle,
505 struct lustre_capa *capa)
507 struct dt_object *next = mdd_object_child(pobj);
511 if (dt_try_as_dir(env, next)) {
512 struct md_ucred *uc = md_ucred(env);
514 rc = next->do_index_ops->dio_insert(env, next,
516 (const struct dt_key *)name,
517 handle, capa, uc->mu_cap &
518 CFS_CAP_SYS_RESOURCE_MASK);
525 /* insert named index, add reference if isdir */
526 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
527 const struct lu_fid *lf, const char *name, int is_dir,
528 struct thandle *handle, struct lustre_capa *capa)
533 rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
534 if (rc == 0 && is_dir) {
535 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
536 mdo_ref_add(env, pobj, handle);
537 mdd_write_unlock(env, pobj);
542 /* delete named index, drop reference if isdir */
543 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
544 const char *name, int is_dir, struct thandle *handle,
545 struct lustre_capa *capa)
550 rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
551 if (rc == 0 && is_dir) {
552 mdd_write_lock(env, pobj, MOR_TGT_PARENT);
553 mdo_ref_del(env, pobj, handle);
554 mdd_write_unlock(env, pobj);
560 int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd,
561 int reclen, struct thandle *handle)
565 /* XXX: this is a temporary solution to declare llog changes
566 * will be fixed in 2.3 with new llog implementation */
568 LASSERT(mdd->mdd_capa);
570 /* XXX: Since we use the 'mdd_capa' as fake llog object here, we
571 * have to set the parameter 'size' as INT_MAX or 0 to inform
572 * OSD that this record write is for a llog write or catalog
573 * header update, and osd declare function will reserve less
574 * credits for optimization purpose.
576 * Reserve 6 blocks for a llog write, since the llog file is
577 * usually small, reserve 2 blocks for catalog header update,
578 * because we know for sure that catalog header is already
581 * This hack should be removed in 2.3.
585 rc = dt_declare_record_write(env, mdd->mdd_capa,
586 DECLARE_LLOG_WRITE, 0, handle);
590 /* header will be updated as well */
591 rc = dt_declare_record_write(env, mdd->mdd_capa,
592 DECLARE_LLOG_WRITE, 0, handle);
596 /* also we should be able to create new plain log */
597 rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle);
601 /* new record referencing new plain llog */
602 rc = dt_declare_record_write(env, mdd->mdd_capa,
603 DECLARE_LLOG_WRITE, 0, handle);
607 /* catalog's header will be updated as well */
608 rc = dt_declare_record_write(env, mdd->mdd_capa,
609 DECLARE_LLOG_REWRITE, 0, handle);
614 int mdd_declare_changelog_store(const struct lu_env *env,
615 struct mdd_device *mdd,
616 const struct lu_name *fname,
617 struct thandle *handle)
622 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
625 /* we'll be writing payload + llog header */
626 reclen = sizeof(struct llog_changelog_rec);
628 reclen += fname->ln_namelen;
629 reclen = llog_data_len(reclen);
631 return mdd_declare_llog_record(env, mdd, reclen, handle);
634 static int mdd_declare_changelog_ext_store(const struct lu_env *env,
635 struct mdd_device *mdd,
636 const struct lu_name *tname,
637 const struct lu_name *sname,
638 struct thandle *handle)
643 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
646 /* we'll be writing payload + llog header */
647 reclen = sizeof(struct llog_changelog_ext_rec);
649 reclen += tname->ln_namelen;
651 reclen += 1 + sname->ln_namelen;
652 reclen = llog_data_len(reclen);
654 return mdd_declare_llog_record(env, mdd, reclen, handle);
657 /** Store a namespace change changelog record
658 * If this fails, we must fail the whole transaction; we don't
659 * want the change to commit without the log entry.
660 * \param target - mdd_object of change
661 * \param parent - parent dir/object
662 * \param tname - target name string
663 * \param handle - transacion handle
665 static int mdd_changelog_ns_store(const struct lu_env *env,
666 struct mdd_device *mdd,
667 enum changelog_rec_type type,
669 struct mdd_object *target,
670 struct mdd_object *parent,
671 const struct lu_name *tname,
672 struct thandle *handle)
674 struct llog_changelog_rec *rec;
681 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
683 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
686 LASSERT(target != NULL);
687 LASSERT(parent != NULL);
688 LASSERT(tname != NULL);
689 LASSERT(handle != NULL);
691 reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen);
692 buf = mdd_buf_alloc(env, reclen);
693 if (buf->lb_buf == NULL)
695 rec = (struct llog_changelog_rec *)buf->lb_buf;
697 rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
698 rec->cr.cr_type = (__u32)type;
699 rec->cr.cr_tfid = *mdo2fid(target);
700 rec->cr.cr_pfid = *mdo2fid(parent);
701 rec->cr.cr_namelen = tname->ln_namelen;
702 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
704 target->mod_cltime = cfs_time_current_64();
706 rc = mdd_changelog_llog_write(mdd, rec, handle);
708 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
709 rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid),
710 PFID(&rec->cr.cr_pfid));
718 /** Store a namespace change changelog record
719 * If this fails, we must fail the whole transaction; we don't
720 * want the change to commit without the log entry.
721 * \param target - mdd_object of change
722 * \param tpfid - target parent dir/object fid
723 * \param sfid - source object fid
724 * \param spfid - source parent fid
725 * \param tname - target name string
726 * \param sname - source name string
727 * \param handle - transacion handle
729 static int mdd_changelog_ext_ns_store(const struct lu_env *env,
730 struct mdd_device *mdd,
731 enum changelog_rec_type type,
733 struct mdd_object *target,
734 const struct lu_fid *tpfid,
735 const struct lu_fid *sfid,
736 const struct lu_fid *spfid,
737 const struct lu_name *tname,
738 const struct lu_name *sname,
739 struct thandle *handle)
741 struct llog_changelog_ext_rec *rec;
748 if (!(mdd->mdd_cl.mc_flags & CLM_ON))
750 if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
753 LASSERT(sfid != NULL);
754 LASSERT(tpfid != NULL);
755 LASSERT(tname != NULL);
756 LASSERT(handle != NULL);
758 reclen = sizeof(*rec) + tname->ln_namelen;
760 reclen += 1 + sname->ln_namelen;
761 reclen = llog_data_len(reclen);
762 buf = mdd_buf_alloc(env, reclen);
763 if (buf->lb_buf == NULL)
765 rec = (struct llog_changelog_ext_rec *)buf->lb_buf;
767 rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags);
768 rec->cr.cr_type = (__u32)type;
769 rec->cr.cr_pfid = *tpfid;
770 rec->cr.cr_sfid = *sfid;
771 rec->cr.cr_spfid = *spfid;
772 rec->cr.cr_namelen = tname->ln_namelen;
773 memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen);
775 LASSERT(sfid != NULL);
776 rec->cr.cr_name[tname->ln_namelen] = '\0';
777 memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name,
779 rec->cr.cr_namelen += 1 + sname->ln_namelen;
782 if (likely(target != NULL)) {
783 rec->cr.cr_tfid = *mdo2fid(target);
784 target->mod_cltime = cfs_time_current_64();
786 fid_zero(&rec->cr.cr_tfid);
789 rc = mdd_changelog_ext_llog_write(mdd, rec, handle);
791 CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n",
792 rc, type, tname->ln_name, PFID(sfid), PFID(tpfid));
799 static int mdd_declare_link(const struct lu_env *env,
800 struct mdd_device *mdd,
801 struct mdd_object *p,
802 struct mdd_object *c,
803 const struct lu_name *name,
804 struct thandle *handle)
808 rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle);
812 rc = mdo_declare_ref_add(env, c, handle);
816 rc = mdo_declare_attr_set(env, p, NULL, handle);
820 rc = mdo_declare_attr_set(env, c, NULL, handle);
824 rc = mdd_declare_links_add(env, c, handle);
828 rc = mdd_declare_changelog_store(env, mdd, name, handle);
833 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
834 struct md_object *src_obj, const struct lu_name *lname,
837 const char *name = lname->ln_name;
838 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
839 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
840 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
841 struct mdd_device *mdd = mdo2mdd(src_obj);
842 struct dynlock_handle *dlh;
843 struct thandle *handle;
844 #ifdef HAVE_QUOTA_SUPPORT
845 struct obd_device *obd = mdd->mdd_obd_dev;
846 struct obd_export *exp = md_quota(env)->mq_exp;
847 struct mds_obd *mds = &obd->u.mds;
848 unsigned int qids[MAXQUOTAS] = { 0, 0 };
849 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
854 #ifdef HAVE_QUOTA_SUPPORT
855 if (mds->mds_quota) {
856 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
858 rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA);
861 mdd_data_get(env, mdd_tobj, &data);
862 quota_opc = FSFILT_OP_LINK;
863 mdd_quota_wrapper(la_tmp, qids);
864 /* get block quota for parent */
865 lquota_chkquota(mds_quota_interface_ref, obd, exp,
866 qids, rec_pending, 1, NULL,
867 LQUOTA_FLAGS_BLK, data, 1);
872 handle = mdd_trans_create(env, mdd);
874 GOTO(out_pending, rc = PTR_ERR(handle));
876 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle);
880 rc = mdd_trans_start(env, mdd, handle);
884 dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD);
886 GOTO(out_trans, rc = -ENOMEM);
887 mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
889 rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj);
891 GOTO(out_unlock, rc);
893 rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
895 mdd_object_capa(env, mdd_tobj));
897 GOTO(out_unlock, rc);
899 mdo_ref_add(env, mdd_sobj, handle);
901 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
902 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
904 la->la_valid = LA_CTIME | LA_MTIME;
905 rc = mdd_attr_check_set_internal_locked(env, mdd_tobj, la, handle, 0);
907 GOTO(out_unlock, rc);
909 la->la_valid = LA_CTIME;
910 rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
912 mdd_links_add(env, mdd_sobj,
913 mdo2fid(mdd_tobj), lname, handle, 0);
918 mdd_write_unlock(env, mdd_sobj);
919 mdd_pdo_write_unlock(env, mdd_tobj, dlh);
922 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
923 mdd_tobj, lname, handle);
925 mdd_trans_stop(env, mdd, rc, handle);
927 #ifdef HAVE_QUOTA_SUPPORT
929 lquota_pending_commit(mds_quota_interface_ref, obd,
930 qids, rec_pending, 1);
931 /* Trigger dqacq for the parent owner. If failed,
932 * the next call for lquota_chkquota will process it. */
933 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
940 int mdd_declare_finish_unlink(const struct lu_env *env,
941 struct mdd_object *obj,
943 struct thandle *handle)
947 rc = orph_declare_index_insert(env, obj, handle);
951 return mdd_declare_object_kill(env, obj, ma, handle);
954 /* caller should take a lock before calling */
955 int mdd_finish_unlink(const struct lu_env *env,
956 struct mdd_object *obj, struct md_attr *ma,
961 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
964 LASSERT(mdd_write_locked(env, obj) != 0);
966 /* read HSM flags, needed to set changelogs flags */
967 ma->ma_need = MA_HSM | MA_INODE;
968 rc = mdd_attr_get_internal(env, obj, ma);
969 if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) {
970 obj->mod_flags |= DEAD_OBJ;
971 /* add new orphan and the object
972 * will be deleted during mdd_close() */
973 if (obj->mod_count) {
974 rc = __mdd_orphan_add(env, obj, th);
976 CDEBUG(D_HA, "Object "DFID" is inserted into "
977 "orphan list, open count = %d\n",
978 PFID(mdd_object_fid(obj)),
981 CERROR("Object "DFID" fail to be an orphan, "
982 "open count = %d, maybe cause failed "
984 PFID(mdd_object_fid(obj)),
987 rc = mdd_object_kill(env, obj, ma, th);
992 /* get the i_nlink */
993 ma->ma_need = MA_INODE;
994 rc = mdd_attr_get_internal(env, obj, ma);
997 ma->ma_valid &= ~(MA_LOV | MA_COOKIE);
1004 * has mdd_write_lock on cobj already, but not on pobj yet
1006 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1007 struct mdd_object *cobj, struct md_attr *ma)
1012 rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1);
1017 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1018 struct mdd_object *p, struct mdd_object *c,
1019 const struct lu_name *name, struct md_attr *ma,
1020 struct thandle *handle)
1024 rc = mdo_declare_index_delete(env, p, name->ln_name, handle);
1028 rc = mdo_declare_ref_del(env, p, handle);
1032 rc = mdo_declare_ref_del(env, c, handle);
1036 rc = mdo_declare_ref_del(env, c, handle);
1040 rc = mdo_declare_attr_set(env, p, NULL, handle);
1044 rc = mdo_declare_attr_set(env, c, NULL, handle);
1048 rc = mdd_declare_finish_unlink(env, c, ma, handle);
1052 rc = mdd_declare_links_add(env, c, handle);
1056 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1061 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1062 struct md_object *cobj, const struct lu_name *lname,
1065 const char *name = lname->ln_name;
1066 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1067 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1068 struct mdd_object *mdd_cobj = md2mdd_obj(cobj);
1069 struct mdd_device *mdd = mdo2mdd(pobj);
1070 struct dynlock_handle *dlh;
1071 struct thandle *handle;
1072 #ifdef HAVE_QUOTA_SUPPORT
1073 struct obd_device *obd = mdd->mdd_obd_dev;
1074 struct mds_obd *mds = &obd->u.mds;
1075 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1076 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1079 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1083 if (mdd_object_exists(mdd_cobj) <= 0)
1086 handle = mdd_trans_create(env, mdd);
1088 RETURN(PTR_ERR(handle));
1090 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1095 rc = mdd_trans_start(env, mdd, handle);
1099 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
1101 GOTO(out_trans, rc = -ENOMEM);
1102 mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD);
1104 rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma);
1108 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle,
1109 mdd_object_capa(env, mdd_pobj));
1113 mdo_ref_del(env, mdd_cobj, handle);
1116 mdo_ref_del(env, mdd_cobj, handle);
1118 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1119 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1121 la->la_valid = LA_CTIME | LA_MTIME;
1122 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
1126 if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
1127 /* update ctime of an unlinked file only if it is still
1128 * opened or a link still exists */
1129 la->la_valid = LA_CTIME;
1130 rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
1135 rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
1136 #ifdef HAVE_QUOTA_SUPPORT
1137 if (mds->mds_quota && ma->ma_valid & MA_INODE &&
1138 ma->ma_attr.la_nlink == 0) {
1139 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1141 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
1143 mdd_quota_wrapper(la_tmp, qpids);
1144 if (mdd_cobj->mod_count == 0) {
1145 quota_opc = FSFILT_OP_UNLINK;
1146 mdd_quota_wrapper(&ma->ma_attr, qcids);
1148 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1154 /* old files may not have link ea; ignore errors */
1155 mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
1156 lname, NULL, NULL, handle);
1160 mdd_write_unlock(env, mdd_cobj);
1161 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
1166 cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
1167 if ((ma->ma_valid & MA_HSM) &&
1168 (ma->ma_hsm.mh_flags & HS_EXISTS))
1169 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1171 rc = mdd_changelog_ns_store(env, mdd,
1172 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1173 mdd_cobj, mdd_pobj, lname, handle);
1177 mdd_trans_stop(env, mdd, rc, handle);
1178 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
1179 if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV &&
1180 ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY)
1181 /* Since echo client is incapable of destorying ost object,
1182 * it will destory the object here. */
1183 rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la, ma, 1);
1185 #warning "please remove this after 2.4 (LOD/OSP)."
1188 #ifdef HAVE_QUOTA_SUPPORT
1190 /* Trigger dqrel on the owner of child and parent. If failed,
1191 * the next call for lquota_chkquota will process it. */
1192 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
1198 /* has not lock on pobj yet */
1199 static int mdd_ni_sanity_check(const struct lu_env *env,
1200 struct md_object *pobj,
1201 const struct md_attr *ma)
1203 struct mdd_object *obj = md2mdd_obj(pobj);
1207 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1210 rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode));
1216 * Partial operation.
1218 static int mdd_name_insert(const struct lu_env *env,
1219 struct md_object *pobj,
1220 const struct lu_name *lname,
1221 const struct lu_fid *fid,
1222 const struct md_attr *ma)
1224 const char *name = lname->ln_name;
1225 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1226 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1227 struct dynlock_handle *dlh;
1228 struct thandle *handle;
1229 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1230 #ifdef HAVE_QUOTA_SUPPORT
1231 struct mdd_device *mdd = mdo2mdd(pobj);
1232 struct md_ucred *uc = md_ucred(env);
1233 struct obd_device *obd = mdd->mdd_obd_dev;
1234 struct obd_export *exp = md_quota(env)->mq_exp;
1235 struct mds_obd *mds = &obd->u.mds;
1236 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1237 int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
1238 cfs_cap_t save = uc->mu_cap;
1243 /* XXX: this code won't be used ever:
1244 * DNE uses slightly different approach */
1247 #ifdef HAVE_QUOTA_SUPPORT
1248 if (mds->mds_quota) {
1249 if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) {
1250 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1252 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1255 mdd_data_get(env, mdd_obj, &data);
1256 quota_opc = FSFILT_OP_LINK;
1257 mdd_quota_wrapper(la_tmp, qids);
1258 /* get block quota for parent */
1259 lquota_chkquota(mds_quota_interface_ref, obd,
1260 exp, qids, rec_pending, 1, NULL,
1261 LQUOTA_FLAGS_BLK, data, 1);
1264 uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK;
1268 handle = mdd_trans_create(env, mdo2mdd(pobj));
1270 GOTO(out_pending, rc = PTR_ERR(handle));
1272 rc = mdd_trans_start(env, mdo2mdd(pobj), handle);
1274 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1276 GOTO(out_trans, rc = -ENOMEM);
1278 rc = mdd_ni_sanity_check(env, pobj, ma);
1280 GOTO(out_unlock, rc);
1282 rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir,
1283 handle, BYPASS_CAPA);
1285 GOTO(out_unlock, rc);
1288 * For some case, no need update obj's ctime (LA_CTIME is not set),
1290 * For other cases, update obj's ctime (LA_CTIME is set),
1293 if (ma->ma_attr.la_valid & LA_CTIME) {
1294 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1295 la->la_valid = LA_CTIME | LA_MTIME;
1296 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1301 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1303 mdd_trans_stop(env, mdo2mdd(pobj), rc, handle);
1305 #ifdef HAVE_QUOTA_SUPPORT
1306 if (mds->mds_quota) {
1308 lquota_pending_commit(mds_quota_interface_ref,
1309 obd, qids, rec_pending, 1);
1310 /* Trigger dqacq for the parent owner. If failed,
1311 * the next call for lquota_chkquota will process it*/
1312 lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
1322 /* has not lock on pobj yet */
1323 static int mdd_nr_sanity_check(const struct lu_env *env,
1324 struct md_object *pobj,
1325 const struct md_attr *ma)
1327 struct mdd_object *obj = md2mdd_obj(pobj);
1331 if (ma->ma_attr_flags & MDS_PERM_BYPASS)
1334 rc = mdd_may_unlink(env, obj, ma);
1340 * Partial operation.
1342 static int mdd_name_remove(const struct lu_env *env,
1343 struct md_object *pobj,
1344 const struct lu_name *lname,
1345 const struct md_attr *ma)
1347 const char *name = lname->ln_name;
1348 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1349 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1350 struct mdd_device *mdd = mdo2mdd(pobj);
1351 struct dynlock_handle *dlh;
1352 struct thandle *handle;
1353 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1354 #ifdef HAVE_QUOTA_SUPPORT
1355 struct obd_device *obd = mdd->mdd_obd_dev;
1356 struct mds_obd *mds = &obd->u.mds;
1357 unsigned int qids[MAXQUOTAS] = { 0, 0 };
1363 /* XXX: this code won't be used ever:
1364 * DNE uses slightly different approach */
1367 #ifdef HAVE_QUOTA_SUPPORT
1368 if (mds->mds_quota) {
1369 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1371 rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA);
1373 quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT;
1374 mdd_quota_wrapper(la_tmp, qids);
1378 handle = mdd_trans_create(env, mdd);
1380 GOTO(out_pending, rc = PTR_ERR(handle));
1382 rc = mdd_trans_start(env, mdd, handle);
1384 dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT);
1386 GOTO(out_trans, rc = -ENOMEM);
1388 rc = mdd_nr_sanity_check(env, pobj, ma);
1390 GOTO(out_unlock, rc);
1392 rc = __mdd_index_delete(env, mdd_obj, name, is_dir,
1393 handle, BYPASS_CAPA);
1395 GOTO(out_unlock, rc);
1398 * For some case, no need update obj's ctime (LA_CTIME is not set),
1400 * For other cases, update obj's ctime (LA_CTIME is set),
1403 if (ma->ma_attr.la_valid & LA_CTIME) {
1404 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1405 la->la_valid = LA_CTIME | LA_MTIME;
1406 rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la,
1411 mdd_pdo_write_unlock(env, mdd_obj, dlh);
1413 mdd_trans_stop(env, mdd, rc, handle);
1415 #ifdef HAVE_QUOTA_SUPPORT
1416 /* Trigger dqrel for the parent owner.
1417 * If failed, the next call for lquota_chkquota will process it. */
1419 lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
1427 * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet
1429 static int mdd_rt_sanity_check(const struct lu_env *env,
1430 struct mdd_object *tgt_pobj,
1431 struct mdd_object *tobj,
1437 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
1440 /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not
1441 * exist. In fact, tobj must exist, otherwise the call trace will be:
1442 * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert.
1443 * When get here, tobj must be NOT NULL, the other case has been
1444 * processed in cmr_rename_tgt before mdd_rename_tgt and enable
1446 * So check may_delete, but not check nlink of tgt_pobj. */
1448 rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
1453 /* Partial rename op on slave MDD */
1454 static int mdd_rename_tgt(const struct lu_env *env,
1455 struct md_object *pobj, struct md_object *tobj,
1456 const struct lu_fid *lf, const struct lu_name *lname,
1459 const char *name = lname->ln_name;
1460 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
1461 struct mdd_object *mdd_tpobj = md2mdd_obj(pobj);
1462 struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
1463 struct mdd_device *mdd = mdo2mdd(pobj);
1464 struct dynlock_handle *dlh;
1465 struct thandle *handle;
1466 #ifdef HAVE_QUOTA_SUPPORT
1467 struct obd_device *obd = mdd->mdd_obd_dev;
1468 struct obd_export *exp = md_quota(env)->mq_exp;
1469 struct mds_obd *mds = &obd->u.mds;
1470 unsigned int qcids[MAXQUOTAS] = { 0, 0 };
1471 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1472 int quota_copc = 0, quota_popc = 0;
1473 int rec_pending[MAXQUOTAS] = { 0, 0 };
1479 /* XXX: this code won't be used ever:
1480 * DNE uses slightly different approach */
1483 #ifdef HAVE_QUOTA_SUPPORT
1484 if (mds->mds_quota && !tobj) {
1485 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
1487 rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA);
1490 mdd_data_get(env, mdd_tpobj, &data);
1491 quota_popc = FSFILT_OP_LINK;
1492 mdd_quota_wrapper(la_tmp, qpids);
1493 /* get block quota for target parent */
1494 lquota_chkquota(mds_quota_interface_ref, obd, exp,
1495 qpids, rec_pending, 1, NULL,
1496 LQUOTA_FLAGS_BLK, data, 1);
1500 handle = mdd_trans_create(env, mdd);
1502 GOTO(out_pending, rc = PTR_ERR(handle));
1504 rc = mdd_trans_start(env, mdd, handle);
1506 dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT);
1508 GOTO(out_trans, rc = -ENOMEM);
1510 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
1512 rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma);
1517 * If rename_tgt is called then we should just re-insert name with
1518 * correct fid, no need to dec/inc parent nlink if obj is dir.
1520 rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA);
1524 rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle,
1529 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1530 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1532 la->la_valid = LA_CTIME | LA_MTIME;
1533 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0);
1538 * For tobj is remote case cmm layer has processed
1539 * and pass NULL tobj to here. So when tobj is NOT NULL,
1540 * it must be local one.
1542 if (tobj && mdd_object_exists(mdd_tobj)) {
1543 mdo_ref_del(env, mdd_tobj, handle);
1545 /* Remove dot reference. */
1546 if (S_ISDIR(ma->ma_attr.la_mode))
1547 mdo_ref_del(env, mdd_tobj, handle);
1549 la->la_valid = LA_CTIME;
1550 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
1554 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
1558 if (ma->ma_valid & MA_INODE && ma->ma_attr.la_nlink == 0) {
1559 cl_flags |= CLF_RENAME_LAST;
1560 #ifdef HAVE_QUOTA_SUPPORT
1561 if (mds->mds_quota && mdd_tobj->mod_count == 0) {
1562 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
1563 mdd_quota_wrapper(&ma->ma_attr, qcids);
1571 mdd_write_unlock(env, mdd_tobj);
1572 mdd_pdo_write_unlock(env, mdd_tpobj, dlh);
1575 /* Bare EXT record with no RENAME in front of it signifies
1576 a partial slave op */
1577 rc = mdd_changelog_ns_store(env, mdd, CL_EXT, cl_flags,
1578 mdd_tobj, mdd_tpobj, lname, handle);
1580 mdd_trans_stop(env, mdd, rc, handle);
1582 #ifdef HAVE_QUOTA_SUPPORT
1583 if (mds->mds_quota) {
1585 lquota_pending_commit(mds_quota_interface_ref, obd,
1586 qpids, rec_pending, 1);
1589 /* Trigger dqrel on the target owner of child.
1590 * If failed, the next call for lquota_chkquota
1591 * will process it. */
1592 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
1600 * The permission has been checked when obj created, no need check again.
1602 static int mdd_cd_sanity_check(const struct lu_env *env,
1603 struct mdd_object *obj)
1608 if (!obj || mdd_is_dead_obj(obj))
1615 static int mdd_declare_create_data(const struct lu_env *env,
1616 struct mdd_device *mdd,
1617 struct mdd_object *obj,
1619 struct thandle *handle)
1621 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1625 buf->lb_len = lmm_size;
1626 rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
1631 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1636 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
1637 struct md_object *cobj, const struct md_op_spec *spec,
1640 struct mdd_device *mdd = mdo2mdd(cobj);
1641 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1642 struct mdd_object *son = md2mdd_obj(cobj);
1643 struct lov_mds_md *lmm = NULL;
1645 struct thandle *handle;
1649 rc = mdd_cd_sanity_check(env, son);
1653 if (!md_should_create(spec->sp_cr_flags))
1655 lmm_size = ma->ma_lmm_size;
1657 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma);
1661 handle = mdd_trans_create(env, mdd);
1663 GOTO(out_free, rc = PTR_ERR(handle));
1665 rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle);
1669 rc = mdd_trans_start(env, mdd, handle);
1674 * XXX: Setting the lov ea is not locked but setting the attr is locked?
1675 * Should this be fixed?
1678 /* Replay creates has objects already */
1680 if (spec->no_create) {
1681 CDEBUG(D_INFO, "we already have lov ea\n");
1682 rc = mdd_lov_set_md(env, mdd_pobj, son,
1683 (struct lov_mds_md *)spec->u.sp_ea.eadata,
1684 spec->u.sp_ea.eadatalen, handle, 0);
1687 /* No need mdd_lsm_sanity_check here */
1688 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm,
1689 lmm_size, handle, 0);
1692 rc = mdd_attr_get_internal_locked(env, son, ma);
1694 /* update lov_objid data, must be before transaction stop! */
1696 mdd_lov_objid_update(mdd, lmm);
1699 mdd_trans_stop(env, mdd, rc, handle);
1701 /* Finish mdd_lov_create() stuff. */
1702 /* if no_create == 0 (not replay), we free lmm allocated by
1703 * mdd_lov_create() */
1704 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
1708 /* Get fid from name and parent */
1710 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
1711 const struct lu_name *lname, struct lu_fid* fid, int mask)
1713 const char *name = lname->ln_name;
1714 const struct dt_key *key = (const struct dt_key *)name;
1715 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
1716 struct mdd_device *m = mdo2mdd(pobj);
1717 struct dt_object *dir = mdd_object_child(mdd_obj);
1721 if (unlikely(mdd_is_dead_obj(mdd_obj)))
1724 rc = mdd_object_exists(mdd_obj);
1725 if (unlikely(rc == 0))
1727 else if (unlikely(rc < 0)) {
1728 CERROR("Object "DFID" locates on remote server\n",
1729 PFID(mdo2fid(mdd_obj)));
1733 /* The common filename length check. */
1734 if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
1735 RETURN(-ENAMETOOLONG);
1737 rc = mdd_permission_internal_locked(env, mdd_obj, NULL, mask,
1742 if (likely(S_ISDIR(mdd_object_type(mdd_obj)) &&
1743 dt_try_as_dir(env, dir))) {
1745 rc = dir->do_index_ops->dio_lookup(env, dir,
1746 (struct dt_rec *)fid, key,
1747 mdd_object_capa(env, mdd_obj));
1758 int mdd_declare_object_initialize(const struct lu_env *env,
1759 struct mdd_object *child,
1761 struct thandle *handle)
1765 rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle);
1766 if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) {
1767 rc = mdo_declare_index_insert(env, child, mdo2fid(child),
1770 rc = mdo_declare_ref_add(env, child, handle);
1773 mdd_declare_links_add(env, child, handle);
1778 int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid,
1779 const struct lu_name *lname, struct mdd_object *child,
1780 struct md_attr *ma, struct thandle *handle,
1781 const struct md_op_spec *spec)
1787 * Update attributes for child.
1790 * (1) the valid bits should be converted between Lustre and Linux;
1791 * (2) maybe, the child attributes should be set in OSD when creation.
1794 rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0);
1798 if (S_ISDIR(ma->ma_attr.la_mode)) {
1799 /* Add "." and ".." for newly created dir */
1800 mdo_ref_add(env, child, handle);
1801 rc = __mdd_index_insert_only(env, child, mdo2fid(child),
1802 dot, handle, BYPASS_CAPA);
1804 rc = __mdd_index_insert_only(env, child, pfid,
1808 mdo_ref_del(env, child, handle);
1811 mdd_links_add(env, child, pfid, lname, handle, 1);
1816 /* has not lock on pobj yet */
1817 static int mdd_create_sanity_check(const struct lu_env *env,
1818 struct md_object *pobj,
1819 const struct lu_name *lname,
1821 struct md_op_spec *spec)
1823 struct mdd_thread_info *info = mdd_env_info(env);
1824 struct lu_attr *la = &info->mti_la;
1825 struct lu_fid *fid = &info->mti_fid;
1826 struct mdd_object *obj = md2mdd_obj(pobj);
1827 struct mdd_device *m = mdo2mdd(pobj);
1828 int lookup = spec->sp_cr_lookup;
1833 if (mdd_is_dead_obj(obj))
1837 * In some cases this lookup is not needed - we know before if name
1838 * exists or not because MDT performs lookup for it.
1839 * name length check is done in lookup.
1843 * Check if the name already exist, though it will be checked in
1844 * _index_insert also, for avoiding rolling back if exists
1847 rc = __mdd_lookup_locked(env, pobj, lname, fid,
1848 MAY_WRITE | MAY_EXEC);
1850 RETURN(rc ? : -EEXIST);
1853 * Check WRITE permission for the parent.
1854 * EXEC permission have been checked
1855 * when lookup before create already.
1857 rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE,
1864 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1868 if (la->la_mode & S_ISGID) {
1869 ma->ma_attr.la_gid = la->la_gid;
1870 if (S_ISDIR(ma->ma_attr.la_mode)) {
1871 ma->ma_attr.la_mode |= S_ISGID;
1872 ma->ma_attr.la_valid |= LA_MODE;
1876 switch (ma->ma_attr.la_mode & S_IFMT) {
1878 unsigned int symlen = strlen(spec->u.sp_symname) + 1;
1880 if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
1881 RETURN(-ENAMETOOLONG);
1900 static int mdd_declare_create(const struct lu_env *env,
1901 struct mdd_device *mdd,
1902 struct mdd_object *p,
1903 struct mdd_object *c,
1904 const struct lu_name *name,
1907 struct thandle *handle,
1908 const struct md_op_spec *spec)
1910 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1913 rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec);
1917 /* if dir, then can inherit default ACl */
1919 buf->lb_len = lmm_size;
1920 if (S_ISDIR(ma->ma_attr.la_mode)) {
1921 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT,
1924 rc = mdo_declare_ref_add(env, p, handle);
1929 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS,
1934 rc = mdd_declare_object_initialize(env, c, ma, handle);
1938 rc = mdo_declare_index_insert(env, p, mdo2fid(c),
1939 name->ln_name, handle);
1943 rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV,
1948 if (S_ISLNK(ma->ma_attr.la_mode)) {
1949 rc = dt_declare_record_write(env, mdd_object_child(c),
1950 strlen(spec->u.sp_symname), 0,
1955 rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle);
1959 rc = mdd_declare_changelog_store(env, mdd, name, handle);
1963 rc = mdd_declare_lov_objid_update(env, mdd, handle);
1971 * Create object and insert it into namespace.
1973 static int mdd_create(const struct lu_env *env,
1974 struct md_object *pobj,
1975 const struct lu_name *lname,
1976 struct md_object *child,
1977 struct md_op_spec *spec,
1980 struct mdd_thread_info *info = mdd_env_info(env);
1981 struct lu_attr *la = &info->mti_la_for_fix;
1982 struct md_attr *ma_acl = &info->mti_ma;
1983 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1984 struct mdd_object *son = md2mdd_obj(child);
1985 struct mdd_device *mdd = mdo2mdd(pobj);
1986 struct lu_attr *attr = &ma->ma_attr;
1987 struct lov_mds_md *lmm = NULL;
1988 struct thandle *handle;
1989 struct dynlock_handle *dlh;
1990 const char *name = lname->ln_name;
1991 int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0;
1992 int got_def_acl = 0;
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 qcids[MAXQUOTAS] = { 0, 0 };
1998 unsigned int qpids[MAXQUOTAS] = { 0, 0 };
1999 int quota_opc = 0, block_count = 0;
2000 int inode_pending[MAXQUOTAS] = { 0, 0 };
2001 int block_pending[MAXQUOTAS] = { 0, 0 };
2002 int parent_pending[MAXQUOTAS] = { 0, 0 };
2007 * Two operations have to be performed:
2009 * - an allocation of a new object (->do_create()), and
2011 * - an insertion into a parent index (->dio_insert()).
2013 * Due to locking, operation order is not important, when both are
2014 * successful, *but* error handling cases are quite different:
2016 * - if insertion is done first, and following object creation fails,
2017 * insertion has to be rolled back, but this operation might fail
2018 * also leaving us with dangling index entry.
2020 * - if creation is done first, is has to be undone if insertion
2021 * fails, leaving us with leaked space, which is neither good, nor
2024 * It seems that creation-first is simplest solution, but it is
2025 * sub-optimal in the frequent
2030 * case, because second mkdir is bound to create object, only to
2031 * destroy it immediately.
2033 * To avoid this follow local file systems that do double lookup:
2035 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2037 * 1. create (mdd_object_create_internal())
2039 * 2. insert (__mdd_index_insert(), lookup again)
2042 /* Sanity checks before big job. */
2043 rc = mdd_create_sanity_check(env, pobj, lname, ma, spec);
2047 #ifdef HAVE_QUOTA_SUPPORT
2048 if (mds->mds_quota) {
2049 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2051 rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA);
2055 quota_opc = FSFILT_OP_CREATE;
2056 mdd_quota_wrapper(&ma->ma_attr, qcids);
2057 mdd_quota_wrapper(la_tmp, qpids);
2058 /* get file quota for child */
2059 lquota_chkquota(mds_quota_interface_ref, obd, exp,
2060 qcids, inode_pending, 1, NULL, 0, NULL,
2062 switch (ma->ma_attr.la_mode & S_IFMT) {
2071 if (qcids[USRQUOTA] == qpids[USRQUOTA] &&
2072 qcids[GRPQUOTA] == qpids[GRPQUOTA]) {
2076 /* get block quota for child and parent */
2078 lquota_chkquota(mds_quota_interface_ref, obd,
2079 exp, qcids, block_pending,
2081 LQUOTA_FLAGS_BLK, NULL, 0);
2083 lquota_chkquota(mds_quota_interface_ref, obd,
2084 exp, qpids, parent_pending, 1,
2085 NULL, LQUOTA_FLAGS_BLK, NULL,
2091 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2092 GOTO(out_pending, rc = -EINPROGRESS);
2095 * No RPC inside the transaction, so OST objects should be created at
2098 if (S_ISREG(attr->la_mode)) {
2099 lmm_size = ma->ma_lmm_size;
2100 rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
2103 GOTO(out_pending, rc);
2106 if (!S_ISLNK(attr->la_mode)) {
2107 ma_acl->ma_acl_size = sizeof info->mti_xattr_buf;
2108 ma_acl->ma_acl = info->mti_xattr_buf;
2109 ma_acl->ma_need = MA_ACL_DEF;
2110 ma_acl->ma_valid = 0;
2112 mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT);
2113 rc = mdd_def_acl_get(env, mdd_pobj, ma_acl);
2114 mdd_read_unlock(env, mdd_pobj);
2117 else if (ma_acl->ma_valid & MA_ACL_DEF)
2121 handle = mdd_trans_create(env, mdd);
2123 GOTO(out_free, rc = PTR_ERR(handle));
2125 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma,
2126 lmm_size, handle, spec);
2130 rc = mdd_trans_start(env, mdd, handle);
2134 dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT);
2136 GOTO(out_trans, rc = -ENOMEM);
2138 mdd_write_lock(env, son, MOR_TGT_CHILD);
2139 rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec);
2141 mdd_write_unlock(env, son);
2147 #ifdef CONFIG_FS_POSIX_ACL
2149 struct lu_buf *acl_buf = &info->mti_buf;
2150 acl_buf->lb_buf = ma_acl->ma_acl;
2151 acl_buf->lb_len = ma_acl->ma_acl_size;
2153 rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle);
2155 mdd_write_unlock(env, son);
2158 ma->ma_attr.la_valid |= LA_MODE;
2163 rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname,
2164 son, ma, handle, spec);
2165 mdd_write_unlock(env, son);
2168 * Object has no links, so it will be destroyed when last
2169 * reference is released. (XXX not now.)
2175 rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son),
2176 name, S_ISDIR(attr->la_mode), handle,
2177 mdd_object_capa(env, mdd_pobj));
2184 /* No need mdd_lsm_sanity_check here */
2185 rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0);
2187 CERROR("error on stripe info copy %d \n", rc);
2190 if (lmm && lmm_size > 0) {
2191 /* Set Lov here, do not get lmm again later */
2192 if (lmm_size > ma->ma_lmm_size) {
2193 /* Reply buffer is smaller, need bigger one */
2194 mdd_max_lmm_buffer(env, lmm_size);
2195 if (unlikely(info->mti_max_lmm == NULL))
2196 GOTO(cleanup, rc = -ENOMEM);
2197 ma->ma_lmm = info->mti_max_lmm;
2198 ma->ma_big_lmm_used = 1;
2200 memcpy(ma->ma_lmm, lmm, lmm_size);
2201 ma->ma_lmm_size = lmm_size;
2202 ma->ma_valid |= MA_LOV;
2205 if (S_ISLNK(attr->la_mode)) {
2206 struct md_ucred *uc = md_ucred(env);
2207 struct dt_object *dt = mdd_object_child(son);
2208 const char *target_name = spec->u.sp_symname;
2209 int sym_len = strlen(target_name);
2210 const struct lu_buf *buf;
2213 buf = mdd_buf_get_const(env, target_name, sym_len);
2214 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle,
2215 mdd_object_capa(env, son),
2217 CFS_CAP_SYS_RESOURCE_MASK);
2222 GOTO(cleanup, rc = -EFAULT);
2226 la->la_valid = LA_CTIME | LA_MTIME;
2227 rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0);
2231 /* Return attr back. */
2232 rc = mdd_attr_get_internal_locked(env, son, ma);
2235 if (rc && created) {
2239 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2240 S_ISDIR(attr->la_mode),
2241 handle, BYPASS_CAPA);
2243 CERROR("error can not cleanup destroy %d\n",
2248 mdd_write_lock(env, son, MOR_TGT_CHILD);
2249 mdo_ref_del(env, son, handle);
2250 if (initialized && S_ISDIR(attr->la_mode))
2251 mdo_ref_del(env, son, handle);
2252 mdd_write_unlock(env, son);
2256 /* update lov_objid data, must be before transaction stop! */
2258 mdd_lov_objid_update(mdd, lmm);
2260 mdd_pdo_write_unlock(env, mdd_pobj, dlh);
2263 rc = mdd_changelog_ns_store(env, mdd,
2264 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2265 S_ISREG(attr->la_mode) ? CL_CREATE :
2266 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2267 0, son, mdd_pobj, lname, handle);
2269 mdd_trans_stop(env, mdd, rc, handle);
2271 /* finish lov_create stuff, free all temporary data */
2272 mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
2274 #ifdef HAVE_QUOTA_SUPPORT
2276 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2278 lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
2280 lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
2282 /* Trigger dqacq on the owner of child and parent. If failed,
2283 * the next call for lquota_chkquota will process it. */
2284 lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
2288 /* The child object shouldn't be cached anymore */
2290 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
2291 &child->mo_lu.lo_header->loh_flags);
2296 * Get locks on parents in proper order
2297 * RETURN: < 0 - error, rename_order if successful
2305 static int mdd_rename_order(const struct lu_env *env,
2306 struct mdd_device *mdd,
2307 struct mdd_object *src_pobj,
2308 struct mdd_object *tgt_pobj)
2310 /* order of locking, 1 - tgt-src, 0 - src-tgt*/
2314 if (src_pobj == tgt_pobj)
2315 RETURN(MDD_RN_SAME);
2317 /* compared the parent child relationship of src_p&tgt_p */
2318 if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(src_pobj))){
2320 } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) {
2323 rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL);
2336 /* has not mdd_write{read}_lock on any obj yet. */
2337 static int mdd_rename_sanity_check(const struct lu_env *env,
2338 struct mdd_object *src_pobj,
2339 struct mdd_object *tgt_pobj,
2340 struct mdd_object *sobj,
2341 struct mdd_object *tobj,
2347 if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS))
2350 /* XXX: when get here, sobj must NOT be NULL,
2351 * the other case has been processed in cml_rename
2352 * before mdd_rename and enable MDS_PERM_BYPASS. */
2355 rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
2359 /* XXX: when get here, "tobj == NULL" means tobj must
2360 * NOT exist (neither on remote MDS, such case has been
2361 * processed in cml_rename before mdd_rename and enable
2363 * So check may_create, but not check may_unlink. */
2365 rc = mdd_may_create(env, tgt_pobj, NULL,
2366 (src_pobj != tgt_pobj), 0);
2368 rc = mdd_may_delete(env, tgt_pobj, tobj, ma,
2369 (src_pobj != tgt_pobj), 1);
2371 if (!rc && !tobj && (src_pobj != tgt_pobj) &&
2372 S_ISDIR(ma->ma_attr.la_mode))
2373 rc = __mdd_may_link(env, tgt_pobj);
2378 static int mdd_declare_rename(const struct lu_env *env,
2379 struct mdd_device *mdd,
2380 struct mdd_object *mdd_spobj,
2381 struct mdd_object *mdd_tpobj,
2382 struct mdd_object *mdd_sobj,
2383 struct mdd_object *mdd_tobj,
2384 const struct lu_name *tname,
2385 const struct lu_name *sname,
2387 struct thandle *handle)
2395 /* name from source dir */
2396 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
2400 /* .. from source child */
2401 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
2402 /* source child can be directory,
2403 * counted by source dir's nlink */
2404 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
2408 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle);
2412 rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj),
2417 /* new target child can be directory,
2418 * counted by target dir's nlink */
2419 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
2425 rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle);
2429 rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle);
2432 mdd_declare_links_add(env, mdd_sobj, handle);
2436 rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle);
2441 rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj),
2442 tname->ln_name, handle);
2446 /* name from target dir (old name), we declare it unconditionally
2447 * as mdd_rename() calls delete unconditionally as well. so just
2448 * to balance declarations vs calls to change ... */
2449 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle);
2453 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
2454 /* delete target child in target parent directory */
2455 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2459 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
2460 /* target child can be directory,
2461 * delete "." reference in target child directory */
2462 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
2466 /* delete ".." reference in target parent directory */
2467 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
2472 rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle);
2476 mdd_declare_links_add(env, mdd_tobj, handle);
2480 rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle);
2485 rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle);
2492 /* src object can be remote that is why we use only fid and type of object */
2493 static int mdd_rename(const struct lu_env *env,
2494 struct md_object *src_pobj, struct md_object *tgt_pobj,
2495 const struct lu_fid *lf, const struct lu_name *lsname,
2496 struct md_object *tobj, const struct lu_name *ltname,
2499 const char *sname = lsname->ln_name;
2500 const char *tname = ltname->ln_name;
2501 struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix;
2502 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
2503 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
2504 struct mdd_device *mdd = mdo2mdd(src_pobj);
2505 struct mdd_object *mdd_sobj = NULL; /* source object */
2506 struct mdd_object *mdd_tobj = NULL;
2507 struct dynlock_handle *sdlh, *tdlh;
2508 struct thandle *handle;
2509 const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
2510 const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
2512 unsigned cl_flags = 0;
2515 #ifdef HAVE_QUOTA_SUPPORT
2516 struct obd_device *obd = mdd->mdd_obd_dev;
2517 struct obd_export *exp = md_quota(env)->mq_exp;
2518 struct mds_obd *mds = &obd->u.mds;
2519 unsigned int qspids[MAXQUOTAS] = { 0, 0 };
2520 unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
2521 unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
2522 int quota_copc = 0, quota_popc = 0;
2523 int rec_pending[MAXQUOTAS] = { 0, 0 };
2527 LASSERT(ma->ma_attr.la_mode & S_IFMT);
2528 is_dir = S_ISDIR(ma->ma_attr.la_mode);
2531 mdd_tobj = md2mdd_obj(tobj);
2533 #ifdef HAVE_QUOTA_SUPPORT
2534 if (mds->mds_quota) {
2535 struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la;
2537 rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA);
2539 mdd_quota_wrapper(la_tmp, qspids);
2541 rc = mdd_la_get(env, mdd_tpobj, la_tmp,
2545 mdd_data_get(env, mdd_tpobj, &data);
2546 quota_popc = FSFILT_OP_LINK;
2547 mdd_quota_wrapper(la_tmp, qtpids);
2548 /* get block quota for target parent */
2549 lquota_chkquota(mds_quota_interface_ref,
2551 rec_pending, 1, NULL,
2559 mdd_sobj = mdd_object_find(env, mdd, lf);
2561 handle = mdd_trans_create(env, mdd);
2563 GOTO(out_pending, rc = PTR_ERR(handle));
2565 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
2566 mdd_tobj, lsname, ltname, ma, handle);
2570 rc = mdd_trans_start(env, mdd, handle);
2574 /* FIXME: Should consider tobj and sobj too in rename_lock. */
2575 rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj);
2577 GOTO(cleanup_unlocked, rc);
2579 /* Get locks in determined order */
2580 if (rc == MDD_RN_SAME) {
2581 sdlh = mdd_pdo_write_lock(env, mdd_spobj,
2582 sname, MOR_SRC_PARENT);
2583 /* check hashes to determine do we need one lock or two */
2584 if (mdd_name2hash(sname) != mdd_name2hash(tname))
2585 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,
2589 } else if (rc == MDD_RN_SRCTGT) {
2590 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT);
2591 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT);
2593 tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT);
2594 sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT);
2596 if (sdlh == NULL || tdlh == NULL)
2597 GOTO(cleanup, rc = -ENOMEM);
2599 rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj,
2600 mdd_sobj, mdd_tobj, ma);
2604 /* Remove source name from source directory */
2605 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle,
2606 mdd_object_capa(env, mdd_spobj));
2610 /* "mv dir1 dir2" needs "dir1/.." link update */
2611 if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
2612 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2613 mdd_object_capa(env, mdd_sobj));
2615 GOTO(fixup_spobj2, rc);
2617 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
2618 handle, mdd_object_capa(env, mdd_sobj));
2620 GOTO(fixup_spobj, rc);
2623 /* Remove target name from target directory
2624 * Here tobj can be remote one, so we do index_delete unconditionally
2625 * and -ENOENT is allowed.
2627 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2628 mdd_object_capa(env, mdd_tpobj));
2631 /* tname might been renamed to something else */
2632 GOTO(fixup_spobj, rc);
2635 GOTO(fixup_spobj, rc);
2638 /* Insert new fid with target name into target dir */
2639 rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
2640 mdd_object_capa(env, mdd_tpobj));
2642 GOTO(fixup_tpobj, rc);
2644 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
2645 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
2647 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
2649 la->la_valid = LA_CTIME;
2650 rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
2653 GOTO(fixup_tpobj, rc);
2656 /* Remove old target object
2657 * For tobj is remote case cmm layer has processed
2658 * and set tobj to NULL then. So when tobj is NOT NULL,
2659 * it must be local one.
2661 if (tobj && mdd_object_exists(mdd_tobj)) {
2662 mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
2663 if (mdd_is_dead_obj(mdd_tobj)) {
2664 mdd_write_unlock(env, mdd_tobj);
2665 /* shld not be dead, something is wrong */
2666 CERROR("tobj is dead, something is wrong\n");
2670 mdo_ref_del(env, mdd_tobj, handle);
2672 /* Remove dot reference. */
2674 mdo_ref_del(env, mdd_tobj, handle);
2676 la->la_valid = LA_CTIME;
2677 rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
2679 GOTO(fixup_tpobj, rc);
2681 rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
2682 mdd_write_unlock(env, mdd_tobj);
2684 GOTO(fixup_tpobj, rc);
2686 if (ma->ma_valid & MA_INODE && ma->ma_attr.la_nlink == 0) {
2687 cl_flags |= CLF_RENAME_LAST;
2688 #ifdef HAVE_QUOTA_SUPPORT
2689 if (mds->mds_quota && mdd_tobj->mod_count == 0) {
2690 quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
2691 mdd_quota_wrapper(&ma->ma_attr, qtcids);
2697 la->la_valid = LA_CTIME | LA_MTIME;
2698 rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
2700 GOTO(fixup_tpobj, rc);
2702 if (mdd_spobj != mdd_tpobj) {
2703 la->la_valid = LA_CTIME | LA_MTIME;
2704 rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la,
2708 if (rc == 0 && mdd_sobj) {
2709 mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD);
2710 rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname,
2711 mdo2fid(mdd_tpobj), ltname, handle);
2713 /* Old files might not have EA entry */
2714 mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
2716 mdd_write_unlock(env, mdd_sobj);
2717 /* We don't fail the transaction if the link ea can't be
2718 updated -- fid2path will use alternate lookup method. */
2726 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
2729 CWARN("tp obj fix error %d\n",rc2);
2731 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
2732 !mdd_is_dead_obj(mdd_tobj)) {
2733 rc2 = __mdd_index_insert(env, mdd_tpobj,
2734 mdo2fid(mdd_tobj), tname,
2739 CWARN("tp obj fix error %d\n",rc2);
2744 if (rc && is_dir && mdd_sobj) {
2745 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
2749 CWARN("sp obj dotdot delete error %d\n",rc2);
2752 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
2753 dotdot, handle, BYPASS_CAPA);
2755 CWARN("sp obj dotdot insert error %d\n",rc2);
2760 rc2 = __mdd_index_insert(env, mdd_spobj,
2761 lf, sname, is_dir, handle, BYPASS_CAPA);
2763 CWARN("sp obj fix error %d\n",rc2);
2766 if (likely(tdlh) && sdlh != tdlh)
2767 mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
2769 mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
2772 rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags,
2773 mdd_tobj, tpobj_fid, lf,
2774 spobj_fid, ltname, lsname,
2778 mdd_trans_stop(env, mdd, rc, handle);
2780 mdd_object_put(env, mdd_sobj);
2782 #ifdef HAVE_QUOTA_SUPPORT
2783 if (mds->mds_quota) {
2785 lquota_pending_commit(mds_quota_interface_ref, obd,
2786 qtpids, rec_pending, 1);
2789 /* Trigger dqrel on the source owner of parent.
2790 * If failed, the next call for lquota_chkquota will
2792 lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
2793 FSFILT_OP_UNLINK_PARTIAL_PARENT);
2795 /* Trigger dqrel on the target owner of child.
2796 * If failed, the next call for lquota_chkquota
2797 * will process it. */
2798 lquota_adjust(mds_quota_interface_ref, obd, qtcids,
2799 qtpids, rc, quota_copc);
2806 /** enable/disable storing of hardlink info */
2807 int mdd_linkea_enable = 1;
2808 CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644,
2809 "record hardlink info in EAs");
2811 /** Read the link EA into a temp buffer.
2812 * Uses the name_buf since it is generally large.
2813 * \retval IS_ERR err
2814 * \retval ptr to \a lu_buf (always \a mti_big_buf)
2816 struct lu_buf *mdd_links_get(const struct lu_env *env,
2817 struct mdd_object *mdd_obj)
2820 struct lustre_capa *capa;
2821 struct link_ea_header *leh;
2824 /* First try a small buf */
2825 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2826 if (buf->lb_buf == NULL)
2827 return ERR_PTR(-ENOMEM);
2829 capa = mdd_object_capa(env, mdd_obj);
2830 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2831 if (rc == -ERANGE) {
2832 /* Buf was too small, figure out what we need. */
2834 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2837 buf = mdd_buf_alloc(env, rc);
2838 if (buf->lb_buf == NULL)
2839 return ERR_PTR(-ENOMEM);
2840 rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
2846 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
2847 leh->leh_magic = LINK_EA_MAGIC;
2848 leh->leh_reccount = __swab32(leh->leh_reccount);
2849 leh->leh_len = __swab64(leh->leh_len);
2850 /* entries are swabbed by mdd_lee_unpack */
2852 if (leh->leh_magic != LINK_EA_MAGIC)
2853 return ERR_PTR(-EINVAL);
2854 if (leh->leh_reccount == 0)
2855 return ERR_PTR(-ENODATA);
2860 /** Pack a link_ea_entry.
2861 * All elements are stored as chars to avoid alignment issues.
2862 * Numbers are always big-endian
2863 * \retval record length
2865 static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
2866 const struct lu_fid *pfid)
2868 struct lu_fid tmpfid;
2871 fid_cpu_to_be(&tmpfid, pfid);
2872 memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
2873 memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
2874 reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
2876 lee->lee_reclen[0] = (reclen >> 8) & 0xff;
2877 lee->lee_reclen[1] = reclen & 0xff;
2881 void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen,
2882 struct lu_name *lname, struct lu_fid *pfid)
2884 *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
2885 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
2886 fid_be_to_cpu(pfid, pfid);
2887 lname->ln_name = lee->lee_name;
2888 lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
2891 /** Add a record to the end of link ea buf */
2892 static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf,
2893 const struct lu_fid *pfid,
2894 const struct lu_name *lname)
2896 struct link_ea_header *leh;
2897 struct link_ea_entry *lee;
2900 if (lname == NULL || pfid == NULL)
2903 /* Make sure our buf is big enough for the new one */
2905 reclen = lname->ln_namelen + sizeof(struct link_ea_entry);
2906 if (leh->leh_len + reclen > buf->lb_len) {
2907 if (mdd_buf_grow(env, leh->leh_len + reclen) < 0)
2912 lee = buf->lb_buf + leh->leh_len;
2913 reclen = mdd_lee_pack(lee, lname, pfid);
2914 leh->leh_len += reclen;
2915 leh->leh_reccount++;
2919 static int mdd_declare_links_add(const struct lu_env *env,
2920 struct mdd_object *mdd_obj,
2921 struct thandle *handle)
2925 /* XXX: max size? */
2926 rc = mdo_declare_xattr_set(env, mdd_obj,
2927 mdd_buf_get_const(env, NULL, 4096),
2928 XATTR_NAME_LINK, 0, handle);
2933 /* For pathologic linkers, we don't want to spend lots of time scanning the
2934 * link ea. Limit ourseleves to something reasonable; links not in the EA
2935 * can be looked up via (slower) parent lookup.
2937 #define LINKEA_MAX_COUNT 128
2939 static int mdd_links_add(const struct lu_env *env,
2940 struct mdd_object *mdd_obj,
2941 const struct lu_fid *pfid,
2942 const struct lu_name *lname,
2943 struct thandle *handle, int first)
2946 struct link_ea_header *leh;
2950 if (!mdd_linkea_enable)
2953 buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
2956 if (rc != -ENODATA) {
2957 CERROR("link_ea read failed %d "DFID"\n", rc,
2958 PFID(mdd_object_fid(mdd_obj)));
2961 /* empty EA; start one */
2962 buf = mdd_buf_alloc(env, CFS_PAGE_SIZE);
2963 if (buf->lb_buf == NULL)
2966 leh->leh_magic = LINK_EA_MAGIC;
2967 leh->leh_len = sizeof(struct link_ea_header);
2968 leh->leh_reccount = 0;
2972 if (leh->leh_reccount > LINKEA_MAX_COUNT)
2975 rc = __mdd_links_add(env, buf, pfid, lname);
2980 rc = __mdd_xattr_set(env, mdd_obj,
2981 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
2982 XATTR_NAME_LINK, 0, handle);
2985 CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
2986 PFID(mdd_object_fid(mdd_obj)));
2988 CERROR("link_ea add failed %d "DFID"\n", rc,
2989 PFID(mdd_object_fid(mdd_obj)));
2992 if (buf->lb_len > OBD_ALLOC_BIG)
2993 /* if we vmalloced a large buffer drop it */
2999 static int mdd_links_rename(const struct lu_env *env,
3000 struct mdd_object *mdd_obj,
3001 const struct lu_fid *oldpfid,
3002 const struct lu_name *oldlname,
3003 const struct lu_fid *newpfid,
3004 const struct lu_name *newlname,
3005 struct thandle *handle)
3008 struct link_ea_header *leh;
3009 struct link_ea_entry *lee;
3010 struct lu_name *tmpname = &mdd_env_info(env)->mti_name;
3011 struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid;
3017 if (!mdd_linkea_enable)
3020 if (mdd_obj->mod_flags & DEAD_OBJ)
3021 /* No more links, don't bother */
3024 buf = mdd_links_get(env, mdd_obj);
3028 CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
3029 rc, PFID(mdd_object_fid(mdd_obj)));
3031 CERROR("link_ea read failed %d "DFID"\n",
3032 rc, PFID(mdd_object_fid(mdd_obj)));
3036 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
3038 /* Find the old record */
3039 for(count = 0; count < leh->leh_reccount; count++) {
3040 mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
3041 if (tmpname->ln_namelen == oldlname->ln_namelen &&
3042 lu_fid_eq(tmpfid, oldpfid) &&
3043 (strncmp(tmpname->ln_name, oldlname->ln_name,
3044 tmpname->ln_namelen) == 0))
3046 lee = (struct link_ea_entry *)((char *)lee + reclen);
3048 if ((count + 1) > leh->leh_reccount) {
3049 CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
3050 oldlname->ln_namelen, oldlname->ln_name);
3051 GOTO(out, rc = -ENOENT);
3054 /* Remove the old record */
3055 leh->leh_reccount--;
3056 leh->leh_len -= reclen;
3057 memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len -
3060 /* If renaming, add the new record */
3061 if (newpfid != NULL) {
3062 /* if the add fails, we still delete the out-of-date old link */
3063 rc2 = __mdd_links_add(env, buf, newpfid, newlname);
3067 rc = __mdd_xattr_set(env, mdd_obj,
3068 mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
3069 XATTR_NAME_LINK, 0, handle);
3075 CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n",
3076 oldlname->ln_namelen, oldlname->ln_name, rc,
3077 PFID(mdd_object_fid(mdd_obj)));
3079 if (buf->lb_len > OBD_ALLOC_BIG)
3080 /* if we vmalloced a large buffer drop it */
3086 const struct md_dir_operations mdd_dir_ops = {
3087 .mdo_is_subdir = mdd_is_subdir,
3088 .mdo_lookup = mdd_lookup,
3089 .mdo_create = mdd_create,
3090 .mdo_rename = mdd_rename,
3091 .mdo_link = mdd_link,
3092 .mdo_unlink = mdd_unlink,
3093 .mdo_lum_lmm_cmp = mdd_lum_lmm_cmp,
3094 .mdo_name_insert = mdd_name_insert,
3095 .mdo_name_remove = mdd_name_remove,
3096 .mdo_rename_tgt = mdd_rename_tgt,
3097 .mdo_create_data = mdd_create_data,