1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/cmm/cmm_object.c
38 * Lustre Cluster Metadata Manager (cmm)
40 * Author: Mike Pershin <tappro@clusterfs.com>
44 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_MDS
49 #include <lustre_fid.h>
50 #include "cmm_internal.h"
51 #include "mdc_internal.h"
53 int cmm_fld_lookup(struct cmm_device *cm, const struct lu_fid *fid,
54 mdsno_t *mds, const struct lu_env *env)
59 LASSERT(fid_is_sane(fid));
61 rc = fld_client_lookup(cm->cmm_fld, fid_seq(fid), mds, env);
63 CERROR("Can't find mds by seq "LPX64", rc %d\n",
68 if (*mds > cm->cmm_tgt_count) {
69 CERROR("Got invalid mdsno: "LPU64" (max: %u)\n",
70 *mds, cm->cmm_tgt_count);
73 CDEBUG(D_INFO, "CMM: got MDS "LPU64" for sequence: "
74 LPU64"\n", *mds, fid_seq(fid));
80 static struct md_object_operations cml_mo_ops;
81 static struct md_dir_operations cml_dir_ops;
82 static struct lu_object_operations cml_obj_ops;
84 static struct md_object_operations cmr_mo_ops;
85 static struct md_dir_operations cmr_dir_ops;
86 static struct lu_object_operations cmr_obj_ops;
88 struct lu_object *cmm_object_alloc(const struct lu_env *env,
89 const struct lu_object_header *loh,
92 const struct lu_fid *fid = &loh->loh_fid;
93 struct lu_object *lo = NULL;
94 struct cmm_device *cd;
101 if (cd->cmm_flags & CMM_INITIALIZED) {
102 /* get object location */
103 rc = cmm_fld_lookup(lu2cmm_dev(ld), fid, &mds, env);
108 * Device is not yet initialized, cmm_object is being created
109 * as part of early bootstrap procedure (it is /ROOT, or /fld,
110 * etc.). Such object *has* to be local.
112 mds = cd->cmm_local_num;
114 /* select the proper set of operations based on object location */
115 if (mds == cd->cmm_local_num) {
116 struct cml_object *clo;
120 lo = &clo->cmm_obj.cmo_obj.mo_lu;
121 lu_object_init(lo, NULL, ld);
122 clo->cmm_obj.cmo_obj.mo_ops = &cml_mo_ops;
123 clo->cmm_obj.cmo_obj.mo_dir_ops = &cml_dir_ops;
124 lo->lo_ops = &cml_obj_ops;
127 struct cmr_object *cro;
131 lo = &cro->cmm_obj.cmo_obj.mo_lu;
132 lu_object_init(lo, NULL, ld);
133 cro->cmm_obj.cmo_obj.mo_ops = &cmr_mo_ops;
134 cro->cmm_obj.cmo_obj.mo_dir_ops = &cmr_dir_ops;
135 lo->lo_ops = &cmr_obj_ops;
143 * CMM has two types of objects - local and remote. They have different set
144 * of operations so we are avoiding multiple checks in code.
147 /* get local child device */
148 static struct lu_device *cml_child_dev(struct cmm_device *d)
150 return &d->cmm_child->md_lu_dev;
153 /* lu_object operations */
154 static void cml_object_free(const struct lu_env *env,
155 struct lu_object *lo)
157 struct cml_object *clo = lu2cml_obj(lo);
162 static int cml_object_init(const struct lu_env *env, struct lu_object *lo)
164 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
165 struct lu_device *c_dev;
166 struct lu_object *c_obj;
171 #ifdef HAVE_SPLIT_SUPPORT
172 if (cd->cmm_tgt_count == 0)
173 lu2cml_obj(lo)->clo_split = CMM_SPLIT_DENIED;
175 lu2cml_obj(lo)->clo_split = CMM_SPLIT_UNKNOWN;
177 c_dev = cml_child_dev(cd);
181 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
182 lo->lo_header, c_dev);
184 lu_object_add(lo, c_obj);
194 static int cml_object_print(const struct lu_env *env, void *cookie,
195 lu_printer_t p, const struct lu_object *lo)
197 return (*p)(env, cookie, LUSTRE_CMM_NAME"-local@%p", lo);
200 static struct lu_object_operations cml_obj_ops = {
201 .loo_object_init = cml_object_init,
202 .loo_object_free = cml_object_free,
203 .loo_object_print = cml_object_print
206 /* CMM local md_object operations */
207 static int cml_object_create(const struct lu_env *env,
208 struct md_object *mo,
209 const struct md_op_spec *spec,
210 struct md_attr *attr)
214 rc = mo_object_create(env, md_object_next(mo), spec, attr);
218 static int cml_permission(const struct lu_env *env,
219 struct md_object *p, struct md_object *c,
220 struct md_attr *attr, int mask)
224 rc = mo_permission(env, md_object_next(p), md_object_next(c),
229 static int cml_attr_get(const struct lu_env *env, struct md_object *mo,
230 struct md_attr *attr)
234 rc = mo_attr_get(env, md_object_next(mo), attr);
238 static int cml_attr_set(const struct lu_env *env, struct md_object *mo,
239 const struct md_attr *attr)
243 rc = mo_attr_set(env, md_object_next(mo), attr);
247 static int cml_xattr_get(const struct lu_env *env, struct md_object *mo,
248 struct lu_buf *buf, const char *name)
252 rc = mo_xattr_get(env, md_object_next(mo), buf, name);
256 static int cml_readlink(const struct lu_env *env, struct md_object *mo,
261 rc = mo_readlink(env, md_object_next(mo), buf);
265 static int cml_xattr_list(const struct lu_env *env, struct md_object *mo,
270 rc = mo_xattr_list(env, md_object_next(mo), buf);
274 static int cml_xattr_set(const struct lu_env *env, struct md_object *mo,
275 const struct lu_buf *buf, const char *name,
280 rc = mo_xattr_set(env, md_object_next(mo), buf, name, fl);
284 static int cml_xattr_del(const struct lu_env *env, struct md_object *mo,
289 rc = mo_xattr_del(env, md_object_next(mo), name);
293 static int cml_ref_add(const struct lu_env *env, struct md_object *mo,
294 const struct md_attr *ma)
298 rc = mo_ref_add(env, md_object_next(mo), ma);
302 static int cml_ref_del(const struct lu_env *env, struct md_object *mo,
307 rc = mo_ref_del(env, md_object_next(mo), ma);
311 static int cml_open(const struct lu_env *env, struct md_object *mo,
316 rc = mo_open(env, md_object_next(mo), flags);
320 static int cml_close(const struct lu_env *env, struct md_object *mo,
325 rc = mo_close(env, md_object_next(mo), ma);
329 static int cml_readpage(const struct lu_env *env, struct md_object *mo,
330 const struct lu_rdpg *rdpg)
334 rc = mo_readpage(env, md_object_next(mo), rdpg);
338 static int cml_capa_get(const struct lu_env *env, struct md_object *mo,
339 struct lustre_capa *capa, int renewal)
343 rc = mo_capa_get(env, md_object_next(mo), capa, renewal);
347 static int cml_object_sync(const struct lu_env *env, struct md_object *mo)
351 rc = mo_object_sync(env, md_object_next(mo));
355 static struct md_object_operations cml_mo_ops = {
356 .moo_permission = cml_permission,
357 .moo_attr_get = cml_attr_get,
358 .moo_attr_set = cml_attr_set,
359 .moo_xattr_get = cml_xattr_get,
360 .moo_xattr_list = cml_xattr_list,
361 .moo_xattr_set = cml_xattr_set,
362 .moo_xattr_del = cml_xattr_del,
363 .moo_object_create = cml_object_create,
364 .moo_ref_add = cml_ref_add,
365 .moo_ref_del = cml_ref_del,
366 .moo_open = cml_open,
367 .moo_close = cml_close,
368 .moo_readpage = cml_readpage,
369 .moo_readlink = cml_readlink,
370 .moo_capa_get = cml_capa_get,
371 .moo_object_sync = cml_object_sync,
374 /* md_dir operations */
375 static int cml_lookup(const struct lu_env *env, struct md_object *mo_p,
376 const struct lu_name *lname, struct lu_fid *lf,
377 struct md_op_spec *spec)
382 #ifdef HAVE_SPLIT_SUPPORT
383 if (spec != NULL && spec->sp_ck_split) {
384 rc = cmm_split_check(env, mo_p, lname->ln_name);
389 rc = mdo_lookup(env, md_object_next(mo_p), lname, lf, spec);
394 static mdl_mode_t cml_lock_mode(const struct lu_env *env,
395 struct md_object *mo, mdl_mode_t lm)
397 int rc = MDL_MINMODE;
400 #ifdef HAVE_SPLIT_SUPPORT
401 rc = cmm_split_access(env, mo, lm);
407 static int cml_create(const struct lu_env *env, struct md_object *mo_p,
408 const struct lu_name *lname, struct md_object *mo_c,
409 struct md_op_spec *spec, struct md_attr *ma)
414 #ifdef HAVE_SPLIT_SUPPORT
415 /* Lock mode always should be sane. */
416 LASSERT(spec->sp_cr_mode != MDL_MINMODE);
419 * Sigh... This is long story. MDT may have race with detecting if split
420 * is possible in cmm. We know this race and let it live, because
421 * getting it rid (with some sem or spinlock) will also mean that
422 * PDIROPS for create will not work because we kill parallel work, what
423 * is really bad for performance and makes no sense having PDIROPS. So,
424 * we better allow the race to live, but split dir only if some of
425 * concurrent threads takes EX lock, not matter which one. So that, say,
426 * two concurrent threads may have different lock modes on directory (CW
427 * and EX) and not first one which comes here and see that split is
428 * possible should split the dir, but only that one which has EX
429 * lock. And we do not care that in this case, split may happen a bit
430 * later (when dir size will not be necessarily 64K, but may be a bit
431 * larger). So that, we allow concurrent creates and protect split by EX
434 if (spec->sp_cr_mode == MDL_EX) {
436 * Try to split @mo_p. If split is ok, -ERESTART is returned and
437 * current thread will not peoceed with create. Instead it sends
438 * -ERESTART to client to let it know that correct MDT should be
441 rc = cmm_split_dir(env, mo_p);
444 * -ERESTART or some split error is returned, we can't
445 * proceed with create.
450 if (spec != NULL && spec->sp_ck_split) {
452 * Check for possible split directory and let caller know that
453 * it should tell client that directory is split and operation
454 * should repeat to correct MDT.
456 rc = cmm_split_check(env, mo_p, lname->ln_name);
462 rc = mdo_create(env, md_object_next(mo_p), lname, md_object_next(mo_c),
466 #ifdef HAVE_SPLIT_SUPPORT
472 static int cml_create_data(const struct lu_env *env, struct md_object *p,
474 const struct md_op_spec *spec,
479 rc = mdo_create_data(env, md_object_next(p), md_object_next(o),
484 static int cml_link(const struct lu_env *env, struct md_object *mo_p,
485 struct md_object *mo_s, const struct lu_name *lname,
490 rc = mdo_link(env, md_object_next(mo_p), md_object_next(mo_s),
495 static int cml_unlink(const struct lu_env *env, struct md_object *mo_p,
496 struct md_object *mo_c, const struct lu_name *lname,
501 rc = mdo_unlink(env, md_object_next(mo_p), md_object_next(mo_c),
506 /* rename is split to local/remote by location of new parent dir */
507 struct md_object *md_object_find(const struct lu_env *env,
508 struct md_device *md,
509 const struct lu_fid *f)
515 o = lu_object_find(env, md2lu_dev(md)->ld_site, f);
517 m = (struct md_object *)o;
519 o = lu_object_locate(o->lo_header, md2lu_dev(md)->ld_type);
520 m = o ? lu2md(o) : NULL;
525 static int cmm_mode_get(const struct lu_env *env, struct md_device *md,
526 const struct lu_fid *lf, struct md_attr *ma,
529 struct md_object *mo_s = md_object_find(env, md, lf);
530 struct cmm_thread_info *cmi;
531 struct md_attr *tmp_ma;
536 RETURN(PTR_ERR(mo_s));
538 if (remote && (lu_object_exists(&mo_s->mo_lu) < 0))
541 cmi = cmm_env_info(env);
542 tmp_ma = &cmi->cmi_ma;
543 tmp_ma->ma_need = MA_INODE;
544 tmp_ma->ma_valid = 0;
545 /* get type from src, can be remote req */
546 rc = mo_attr_get(env, md_object_next(mo_s), tmp_ma);
548 ma->ma_attr.la_mode = tmp_ma->ma_attr.la_mode;
549 ma->ma_attr.la_uid = tmp_ma->ma_attr.la_uid;
550 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
551 ma->ma_attr.la_flags = tmp_ma->ma_attr.la_flags;
552 ma->ma_attr.la_valid |= LA_MODE | LA_UID | LA_GID | LA_FLAGS;
554 lu_object_put(env, &mo_s->mo_lu);
558 static int cmm_rename_ctime(const struct lu_env *env, struct md_device *md,
559 const struct lu_fid *lf, struct md_attr *ma)
561 struct md_object *mo_s = md_object_find(env, md, lf);
566 RETURN(PTR_ERR(mo_s));
568 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
569 /* set ctime to obj, can be remote req */
570 rc = mo_attr_set(env, md_object_next(mo_s), ma);
571 lu_object_put(env, &mo_s->mo_lu);
575 static inline void cml_rename_warn(const char *fname,
576 struct md_object *mo_po,
577 struct md_object *mo_pn,
578 const struct lu_fid *lf,
580 struct md_object *mo_t,
585 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
586 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t "DFID"] "
587 "[tname %s] [err %d]\n", fname,
588 PFID(lu_object_fid(&mo_po->mo_lu)),
589 PFID(lu_object_fid(&mo_pn->mo_lu)),
591 PFID(lu_object_fid(&mo_t->mo_lu)),
594 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
595 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t NULL] "
596 "[tname %s] [err %d]\n", fname,
597 PFID(lu_object_fid(&mo_po->mo_lu)),
598 PFID(lu_object_fid(&mo_pn->mo_lu)),
603 static int cml_rename(const struct lu_env *env, struct md_object *mo_po,
604 struct md_object *mo_pn, const struct lu_fid *lf,
605 const struct lu_name *ls_name, struct md_object *mo_t,
606 const struct lu_name *lt_name, struct md_attr *ma)
608 struct cmm_thread_info *cmi;
609 struct md_attr *tmp_ma = NULL;
610 struct md_object *tmp_t = mo_t;
614 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, &remote);
618 if (mo_t && lu_object_exists(&mo_t->mo_lu) < 0) {
619 /* XXX: mo_t is remote object and there is RPC to unlink it.
620 * before that, do local sanity check for rename first. */
622 struct md_object *mo_s = md_object_find(env,
623 md_obj2dev(mo_po), lf);
625 RETURN(PTR_ERR(mo_s));
627 LASSERT(lu_object_exists(&mo_s->mo_lu) > 0);
628 rc = mo_permission(env, md_object_next(mo_po),
629 md_object_next(mo_s),
631 lu_object_put(env, &mo_s->mo_lu);
635 rc = mo_permission(env, NULL, md_object_next(mo_po),
636 ma, MAY_UNLINK | MAY_VTX_FULL);
641 rc = mo_permission(env, NULL, md_object_next(mo_pn), ma,
642 MAY_UNLINK | MAY_VTX_PART);
647 * XXX: @ma will be changed after mo_ref_del, but we will use
648 * it for mdo_rename later, so save it before mo_ref_del.
650 cmi = cmm_env_info(env);
651 tmp_ma = &cmi->cmi_ma;
653 rc = mo_ref_del(env, md_object_next(mo_t), ma);
657 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
661 /* XXX: for src on remote MDS case, change its ctime before local
662 * rename. Firstly, do local sanity check for rename if necessary. */
665 rc = mo_permission(env, NULL, md_object_next(mo_po),
666 ma, MAY_UNLINK | MAY_VTX_FULL);
671 LASSERT(lu_object_exists(&mo_t->mo_lu) > 0);
672 rc = mo_permission(env, md_object_next(mo_pn),
673 md_object_next(mo_t),
681 mask = (S_ISDIR(ma->ma_attr.la_mode) ?
682 MAY_LINK : MAY_CREATE);
685 rc = mo_permission(env, NULL,
686 md_object_next(mo_pn),
692 ma->ma_attr_flags |= MDS_PERM_BYPASS;
694 LASSERT(tmp_ma->ma_attr_flags & MDS_PERM_BYPASS);
697 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf,
698 tmp_ma ? tmp_ma : ma);
700 /* TODO: revoke mo_t if necessary. */
701 cml_rename_warn("cmm_rename_ctime", mo_po,
702 mo_pn, lf, ls_name->ln_name,
703 tmp_t, lt_name->ln_name, rc);
708 /* local rename, mo_t can be NULL */
709 rc = mdo_rename(env, md_object_next(mo_po),
710 md_object_next(mo_pn), lf, ls_name,
711 md_object_next(mo_t), lt_name, tmp_ma ? tmp_ma : ma);
713 /* TODO: revoke all cml_rename */
714 cml_rename_warn("mdo_rename", mo_po, mo_pn, lf,
715 ls_name->ln_name, tmp_t, lt_name->ln_name, rc);
720 static int cml_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
721 struct md_object *mo_t, const struct lu_fid *lf,
722 const struct lu_name *lname, struct md_attr *ma)
727 rc = mdo_rename_tgt(env, md_object_next(mo_p),
728 md_object_next(mo_t), lf, lname, ma);
731 /* used only in case of rename_tgt() when target is not exist */
732 static int cml_name_insert(const struct lu_env *env, struct md_object *p,
733 const struct lu_name *lname, const struct lu_fid *lf,
734 const struct md_attr *ma)
739 rc = mdo_name_insert(env, md_object_next(p), lname, lf, ma);
744 static int cmm_is_subdir(const struct lu_env *env, struct md_object *mo,
745 const struct lu_fid *fid, struct lu_fid *sfid)
747 struct cmm_thread_info *cmi;
751 cmi = cmm_env_info(env);
752 rc = cmm_mode_get(env, md_obj2dev(mo), fid, &cmi->cmi_ma, NULL);
756 if (!S_ISDIR(cmi->cmi_ma.ma_attr.la_mode))
759 rc = mdo_is_subdir(env, md_object_next(mo), fid, sfid);
763 static struct md_dir_operations cml_dir_ops = {
764 .mdo_is_subdir = cmm_is_subdir,
765 .mdo_lookup = cml_lookup,
766 .mdo_lock_mode = cml_lock_mode,
767 .mdo_create = cml_create,
768 .mdo_link = cml_link,
769 .mdo_unlink = cml_unlink,
770 .mdo_name_insert = cml_name_insert,
771 .mdo_rename = cml_rename,
772 .mdo_rename_tgt = cml_rename_tgt,
773 .mdo_create_data = cml_create_data
776 /* -------------------------------------------------------------------
777 * remote CMM object operations. cmr_...
779 static inline struct cmr_object *lu2cmr_obj(struct lu_object *o)
781 return container_of0(o, struct cmr_object, cmm_obj.cmo_obj.mo_lu);
783 static inline struct cmr_object *md2cmr_obj(struct md_object *mo)
785 return container_of0(mo, struct cmr_object, cmm_obj.cmo_obj);
787 static inline struct cmr_object *cmm2cmr_obj(struct cmm_object *co)
789 return container_of0(co, struct cmr_object, cmm_obj);
792 /* get proper child device from MDCs */
793 static struct lu_device *cmr_child_dev(struct cmm_device *d, __u32 num)
795 struct lu_device *next = NULL;
796 struct mdc_device *mdc;
798 spin_lock(&d->cmm_tgt_guard);
799 list_for_each_entry(mdc, &d->cmm_targets, mc_linkage) {
800 if (mdc->mc_num == num) {
801 next = mdc2lu_dev(mdc);
805 spin_unlock(&d->cmm_tgt_guard);
809 /* lu_object operations */
810 static void cmr_object_free(const struct lu_env *env,
811 struct lu_object *lo)
813 struct cmr_object *cro = lu2cmr_obj(lo);
818 static int cmr_object_init(const struct lu_env *env, struct lu_object *lo)
820 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
821 struct lu_device *c_dev;
822 struct lu_object *c_obj;
827 c_dev = cmr_child_dev(cd, lu2cmr_obj(lo)->cmo_num);
831 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
832 lo->lo_header, c_dev);
834 lu_object_add(lo, c_obj);
844 static int cmr_object_print(const struct lu_env *env, void *cookie,
845 lu_printer_t p, const struct lu_object *lo)
847 return (*p)(env, cookie, LUSTRE_CMM_NAME"-remote@%p", lo);
850 static struct lu_object_operations cmr_obj_ops = {
851 .loo_object_init = cmr_object_init,
852 .loo_object_free = cmr_object_free,
853 .loo_object_print = cmr_object_print
856 /* CMM remote md_object operations. All are invalid */
857 static int cmr_object_create(const struct lu_env *env,
858 struct md_object *mo,
859 const struct md_op_spec *spec,
865 static int cmr_permission(const struct lu_env *env,
866 struct md_object *p, struct md_object *c,
867 struct md_attr *attr, int mask)
872 static int cmr_attr_get(const struct lu_env *env, struct md_object *mo,
873 struct md_attr *attr)
878 static int cmr_attr_set(const struct lu_env *env, struct md_object *mo,
879 const struct md_attr *attr)
884 static int cmr_xattr_get(const struct lu_env *env, struct md_object *mo,
885 struct lu_buf *buf, const char *name)
890 static int cmr_readlink(const struct lu_env *env, struct md_object *mo,
896 static int cmr_xattr_list(const struct lu_env *env, struct md_object *mo,
902 static int cmr_xattr_set(const struct lu_env *env, struct md_object *mo,
903 const struct lu_buf *buf, const char *name,
909 static int cmr_xattr_del(const struct lu_env *env, struct md_object *mo,
915 static int cmr_ref_add(const struct lu_env *env, struct md_object *mo,
916 const struct md_attr *ma)
921 static int cmr_ref_del(const struct lu_env *env, struct md_object *mo,
927 static int cmr_open(const struct lu_env *env, struct md_object *mo,
933 static int cmr_close(const struct lu_env *env, struct md_object *mo,
939 static int cmr_readpage(const struct lu_env *env, struct md_object *mo,
940 const struct lu_rdpg *rdpg)
945 static int cmr_capa_get(const struct lu_env *env, struct md_object *mo,
946 struct lustre_capa *capa, int renewal)
951 static int cmr_object_sync(const struct lu_env *env, struct md_object *mo)
956 static struct md_object_operations cmr_mo_ops = {
957 .moo_permission = cmr_permission,
958 .moo_attr_get = cmr_attr_get,
959 .moo_attr_set = cmr_attr_set,
960 .moo_xattr_get = cmr_xattr_get,
961 .moo_xattr_set = cmr_xattr_set,
962 .moo_xattr_list = cmr_xattr_list,
963 .moo_xattr_del = cmr_xattr_del,
964 .moo_object_create = cmr_object_create,
965 .moo_ref_add = cmr_ref_add,
966 .moo_ref_del = cmr_ref_del,
967 .moo_open = cmr_open,
968 .moo_close = cmr_close,
969 .moo_readpage = cmr_readpage,
970 .moo_readlink = cmr_readlink,
971 .moo_capa_get = cmr_capa_get,
972 .moo_object_sync = cmr_object_sync,
975 /* remote part of md_dir operations */
976 static int cmr_lookup(const struct lu_env *env, struct md_object *mo_p,
977 const struct lu_name *lname, struct lu_fid *lf,
978 struct md_op_spec *spec)
981 * This can happens while rename() If new parent is remote dir, lookup
988 static mdl_mode_t cmr_lock_mode(const struct lu_env *env,
989 struct md_object *mo, mdl_mode_t lm)
995 * All methods below are cross-ref by nature. They consist of remote call and
996 * local operation. Due to future rollback functionality there are several
997 * limitations for such methods:
998 * 1) remote call should be done at first to do epoch negotiation between all
999 * MDS involved and to avoid the RPC inside transaction.
1000 * 2) only one RPC can be sent - also due to epoch negotiation.
1001 * For more details see rollback HLD/DLD.
1003 static int cmr_create(const struct lu_env *env, struct md_object *mo_p,
1004 const struct lu_name *lchild_name, struct md_object *mo_c,
1005 struct md_op_spec *spec,
1008 struct cmm_thread_info *cmi;
1009 struct md_attr *tmp_ma;
1013 /* Make sure that name isn't exist before doing remote call. */
1014 rc = mdo_lookup(env, md_object_next(mo_p), lchild_name,
1015 &cmm_env_info(env)->cmi_fid, NULL);
1018 else if (rc != -ENOENT)
1021 /* check the SGID attr */
1022 cmi = cmm_env_info(env);
1024 tmp_ma = &cmi->cmi_ma;
1025 tmp_ma->ma_valid = 0;
1026 tmp_ma->ma_need = MA_INODE;
1028 #ifdef CONFIG_FS_POSIX_ACL
1029 if (!S_ISLNK(ma->ma_attr.la_mode)) {
1030 tmp_ma->ma_acl = cmi->cmi_xattr_buf;
1031 tmp_ma->ma_acl_size = sizeof(cmi->cmi_xattr_buf);
1032 tmp_ma->ma_need |= MA_ACL_DEF;
1035 rc = mo_attr_get(env, md_object_next(mo_p), tmp_ma);
1039 if (tmp_ma->ma_attr.la_mode & S_ISGID) {
1040 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
1041 if (S_ISDIR(ma->ma_attr.la_mode)) {
1042 ma->ma_attr.la_mode |= S_ISGID;
1043 ma->ma_attr.la_valid |= LA_MODE;
1047 #ifdef CONFIG_FS_POSIX_ACL
1048 if (tmp_ma->ma_valid & MA_ACL_DEF) {
1049 spec->u.sp_ea.fid = spec->u.sp_pfid;
1050 spec->u.sp_ea.eadata = tmp_ma->ma_acl;
1051 spec->u.sp_ea.eadatalen = tmp_ma->ma_acl_size;
1052 spec->sp_cr_flags |= MDS_CREATE_RMT_ACL;
1056 /* Local permission check for name_insert before remote ops. */
1057 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1058 (S_ISDIR(ma->ma_attr.la_mode) ?
1059 MAY_LINK : MAY_CREATE));
1063 /* Remote object creation and local name insert. */
1065 * XXX: @ma will be changed after mo_object_create, but we will use
1066 * it for mdo_name_insert later, so save it before mo_object_create.
1069 rc = mo_object_create(env, md_object_next(mo_c), spec, ma);
1071 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1072 rc = mdo_name_insert(env, md_object_next(mo_p), lchild_name,
1073 lu_object_fid(&mo_c->mo_lu), tmp_ma);
1075 /* TODO: remove object mo_c on remote MDS */
1076 CWARN("cmr_create failed, should revoke: [mo_p "DFID"]"
1077 " [name %s] [mo_c "DFID"] [err %d]\n",
1078 PFID(lu_object_fid(&mo_p->mo_lu)),
1079 lchild_name->ln_name,
1080 PFID(lu_object_fid(&mo_c->mo_lu)), rc);
1087 static int cmr_link(const struct lu_env *env, struct md_object *mo_p,
1088 struct md_object *mo_s, const struct lu_name *lname,
1094 /* Make sure that name isn't exist before doing remote call. */
1095 rc = mdo_lookup(env, md_object_next(mo_p), lname,
1096 &cmm_env_info(env)->cmi_fid, NULL);
1099 } else if (rc == -ENOENT) {
1100 /* Local permission check for name_insert before remote ops. */
1101 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1106 rc = mo_ref_add(env, md_object_next(mo_s), ma);
1108 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1109 rc = mdo_name_insert(env, md_object_next(mo_p), lname,
1110 lu_object_fid(&mo_s->mo_lu), ma);
1112 /* TODO: ref_del from mo_s on remote MDS */
1113 CWARN("cmr_link failed, should revoke: "
1114 "[mo_p "DFID"] [mo_s "DFID"] "
1115 "[name %s] [err %d]\n",
1116 PFID(lu_object_fid(&mo_p->mo_lu)),
1117 PFID(lu_object_fid(&mo_s->mo_lu)),
1118 lname->ln_name, rc);
1125 static int cmr_unlink(const struct lu_env *env, struct md_object *mo_p,
1126 struct md_object *mo_c, const struct lu_name *lname,
1129 struct cmm_thread_info *cmi;
1130 struct md_attr *tmp_ma;
1134 /* Local permission check for name_remove before remote ops. */
1135 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1136 MAY_UNLINK | MAY_VTX_PART);
1141 * XXX: @ma will be changed after mo_ref_del, but we will use
1142 * it for mdo_name_remove later, so save it before mo_ref_del.
1144 cmi = cmm_env_info(env);
1145 tmp_ma = &cmi->cmi_ma;
1147 rc = mo_ref_del(env, md_object_next(mo_c), ma);
1149 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1150 rc = mdo_name_remove(env, md_object_next(mo_p), lname, tmp_ma);
1152 /* TODO: ref_add to mo_c on remote MDS */
1153 CWARN("cmr_unlink failed, should revoke: [mo_p "DFID"]"
1154 " [mo_c "DFID"] [name %s] [err %d]\n",
1155 PFID(lu_object_fid(&mo_p->mo_lu)),
1156 PFID(lu_object_fid(&mo_c->mo_lu)),
1157 lname->ln_name, rc);
1164 static inline void cmr_rename_warn(const char *fname,
1165 struct md_object *mo_po,
1166 struct md_object *mo_pn,
1167 const struct lu_fid *lf,
1172 CWARN("cmr_rename failed for %s, should revoke: "
1173 "[mo_po "DFID"] [mo_pn "DFID"] [lf "DFID"] "
1174 "[sname %s] [tname %s] [err %d]\n", fname,
1175 PFID(lu_object_fid(&mo_po->mo_lu)),
1176 PFID(lu_object_fid(&mo_pn->mo_lu)),
1177 PFID(lf), s_name, t_name, err);
1180 static int cmr_rename(const struct lu_env *env,
1181 struct md_object *mo_po, struct md_object *mo_pn,
1182 const struct lu_fid *lf, const struct lu_name *ls_name,
1183 struct md_object *mo_t, const struct lu_name *lt_name,
1186 struct cmm_thread_info *cmi;
1187 struct md_attr *tmp_ma;
1191 LASSERT(mo_t == NULL);
1193 /* get real type of src */
1194 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, NULL);
1198 /* Local permission check for name_remove before remote ops. */
1199 rc = mo_permission(env, NULL, md_object_next(mo_po), ma,
1200 MAY_UNLINK | MAY_VTX_FULL);
1205 * XXX: @ma maybe changed after mdo_rename_tgt, but we will use it
1206 * for mdo_name_remove later, so save it before mdo_rename_tgt.
1208 cmi = cmm_env_info(env);
1209 tmp_ma = &cmi->cmi_ma;
1211 /* the mo_pn is remote directory, so we cannot even know if there is
1212 * mo_t or not. Therefore mo_t is NULL here but remote server should do
1213 * lookup and process this further */
1214 rc = mdo_rename_tgt(env, md_object_next(mo_pn),
1215 NULL/* mo_t */, lf, lt_name, ma);
1219 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1221 /* src object maybe on remote MDS, do remote ops first. */
1222 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf, tmp_ma);
1224 /* TODO: revoke mdo_rename_tgt */
1225 cmr_rename_warn("cmm_rename_ctime", mo_po, mo_pn, lf,
1226 ls_name->ln_name, lt_name->ln_name, rc);
1230 /* only old name is removed localy */
1231 rc = mdo_name_remove(env, md_object_next(mo_po), ls_name, tmp_ma);
1233 /* TODO: revoke all cmr_rename */
1234 cmr_rename_warn("mdo_name_remove", mo_po, mo_pn, lf,
1235 ls_name->ln_name, lt_name->ln_name, rc);
1240 /* part of cross-ref rename(). Used to insert new name in new parent
1241 * and unlink target */
1242 static int cmr_rename_tgt(const struct lu_env *env,
1243 struct md_object *mo_p, struct md_object *mo_t,
1244 const struct lu_fid *lf, const struct lu_name *lname,
1247 struct cmm_thread_info *cmi;
1248 struct md_attr *tmp_ma;
1252 /* target object is remote one */
1253 /* Local permission check for rename_tgt before remote ops. */
1254 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1255 MAY_UNLINK | MAY_VTX_PART);
1260 * XXX: @ma maybe changed after mo_ref_del, but we will use
1261 * it for mdo_rename_tgt later, so save it before mo_ref_del.
1263 cmi = cmm_env_info(env);
1264 tmp_ma = &cmi->cmi_ma;
1266 rc = mo_ref_del(env, md_object_next(mo_t), ma);
1267 /* continue locally with name handling only */
1269 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1270 rc = mdo_rename_tgt(env, md_object_next(mo_p),
1271 NULL, lf, lname, tmp_ma);
1273 /* TODO: ref_add to mo_t on remote MDS */
1274 CWARN("cmr_rename_tgt failed, should revoke: "
1275 "[mo_p "DFID"] [mo_t "DFID"] [lf "DFID"] "
1276 "[name %s] [err %d]\n",
1277 PFID(lu_object_fid(&mo_p->mo_lu)),
1278 PFID(lu_object_fid(&mo_t->mo_lu)),
1280 lname->ln_name, rc);
1286 static struct md_dir_operations cmr_dir_ops = {
1287 .mdo_is_subdir = cmm_is_subdir,
1288 .mdo_lookup = cmr_lookup,
1289 .mdo_lock_mode = cmr_lock_mode,
1290 .mdo_create = cmr_create,
1291 .mdo_link = cmr_link,
1292 .mdo_unlink = cmr_unlink,
1293 .mdo_rename = cmr_rename,
1294 .mdo_rename_tgt = cmr_rename_tgt,