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 [sun.com URL with a
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 struct md_object_operations cml_mo_ops = {
348 .moo_permission = cml_permission,
349 .moo_attr_get = cml_attr_get,
350 .moo_attr_set = cml_attr_set,
351 .moo_xattr_get = cml_xattr_get,
352 .moo_xattr_list = cml_xattr_list,
353 .moo_xattr_set = cml_xattr_set,
354 .moo_xattr_del = cml_xattr_del,
355 .moo_object_create = cml_object_create,
356 .moo_ref_add = cml_ref_add,
357 .moo_ref_del = cml_ref_del,
358 .moo_open = cml_open,
359 .moo_close = cml_close,
360 .moo_readpage = cml_readpage,
361 .moo_readlink = cml_readlink,
362 .moo_capa_get = cml_capa_get
365 /* md_dir operations */
366 static int cml_lookup(const struct lu_env *env, struct md_object *mo_p,
367 const struct lu_name *lname, struct lu_fid *lf,
368 struct md_op_spec *spec)
373 #ifdef HAVE_SPLIT_SUPPORT
374 if (spec != NULL && spec->sp_ck_split) {
375 rc = cmm_split_check(env, mo_p, lname->ln_name);
380 rc = mdo_lookup(env, md_object_next(mo_p), lname, lf, spec);
385 static mdl_mode_t cml_lock_mode(const struct lu_env *env,
386 struct md_object *mo, mdl_mode_t lm)
388 int rc = MDL_MINMODE;
391 #ifdef HAVE_SPLIT_SUPPORT
392 rc = cmm_split_access(env, mo, lm);
398 static int cml_create(const struct lu_env *env, struct md_object *mo_p,
399 const struct lu_name *lname, struct md_object *mo_c,
400 struct md_op_spec *spec, struct md_attr *ma)
405 #ifdef HAVE_SPLIT_SUPPORT
406 /* Lock mode always should be sane. */
407 LASSERT(spec->sp_cr_mode != MDL_MINMODE);
410 * Sigh... This is long story. MDT may have race with detecting if split
411 * is possible in cmm. We know this race and let it live, because
412 * getting it rid (with some sem or spinlock) will also mean that
413 * PDIROPS for create will not work because we kill parallel work, what
414 * is really bad for performance and makes no sense having PDIROPS. So,
415 * we better allow the race to live, but split dir only if some of
416 * concurrent threads takes EX lock, not matter which one. So that, say,
417 * two concurrent threads may have different lock modes on directory (CW
418 * and EX) and not first one which comes here and see that split is
419 * possible should split the dir, but only that one which has EX
420 * lock. And we do not care that in this case, split may happen a bit
421 * later (when dir size will not be necessarily 64K, but may be a bit
422 * larger). So that, we allow concurrent creates and protect split by EX
425 if (spec->sp_cr_mode == MDL_EX) {
427 * Try to split @mo_p. If split is ok, -ERESTART is returned and
428 * current thread will not peoceed with create. Instead it sends
429 * -ERESTART to client to let it know that correct MDT should be
432 rc = cmm_split_dir(env, mo_p);
435 * -ERESTART or some split error is returned, we can't
436 * proceed with create.
441 if (spec != NULL && spec->sp_ck_split) {
443 * Check for possible split directory and let caller know that
444 * it should tell client that directory is split and operation
445 * should repeat to correct MDT.
447 rc = cmm_split_check(env, mo_p, lname->ln_name);
453 rc = mdo_create(env, md_object_next(mo_p), lname, md_object_next(mo_c),
457 #ifdef HAVE_SPLIT_SUPPORT
463 static int cml_create_data(const struct lu_env *env, struct md_object *p,
465 const struct md_op_spec *spec,
470 rc = mdo_create_data(env, md_object_next(p), md_object_next(o),
475 static int cml_link(const struct lu_env *env, struct md_object *mo_p,
476 struct md_object *mo_s, const struct lu_name *lname,
481 rc = mdo_link(env, md_object_next(mo_p), md_object_next(mo_s),
486 static int cml_unlink(const struct lu_env *env, struct md_object *mo_p,
487 struct md_object *mo_c, const struct lu_name *lname,
492 rc = mdo_unlink(env, md_object_next(mo_p), md_object_next(mo_c),
497 /* rename is split to local/remote by location of new parent dir */
498 struct md_object *md_object_find(const struct lu_env *env,
499 struct md_device *md,
500 const struct lu_fid *f)
506 o = lu_object_find(env, md2lu_dev(md)->ld_site, f);
508 m = (struct md_object *)o;
510 o = lu_object_locate(o->lo_header, md2lu_dev(md)->ld_type);
511 m = o ? lu2md(o) : NULL;
516 static int cmm_mode_get(const struct lu_env *env, struct md_device *md,
517 const struct lu_fid *lf, struct md_attr *ma,
520 struct md_object *mo_s = md_object_find(env, md, lf);
521 struct cmm_thread_info *cmi;
522 struct md_attr *tmp_ma;
527 RETURN(PTR_ERR(mo_s));
529 if (remote && (lu_object_exists(&mo_s->mo_lu) < 0))
532 cmi = cmm_env_info(env);
533 tmp_ma = &cmi->cmi_ma;
534 tmp_ma->ma_need = MA_INODE;
535 tmp_ma->ma_valid = 0;
536 /* get type from src, can be remote req */
537 rc = mo_attr_get(env, md_object_next(mo_s), tmp_ma);
539 ma->ma_attr.la_mode = tmp_ma->ma_attr.la_mode;
540 ma->ma_attr.la_uid = tmp_ma->ma_attr.la_uid;
541 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
542 ma->ma_attr.la_flags = tmp_ma->ma_attr.la_flags;
543 ma->ma_attr.la_valid |= LA_MODE | LA_UID | LA_GID | LA_FLAGS;
545 lu_object_put(env, &mo_s->mo_lu);
549 static int cmm_rename_ctime(const struct lu_env *env, struct md_device *md,
550 const struct lu_fid *lf, struct md_attr *ma)
552 struct md_object *mo_s = md_object_find(env, md, lf);
557 RETURN(PTR_ERR(mo_s));
559 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
560 /* set ctime to obj, can be remote req */
561 rc = mo_attr_set(env, md_object_next(mo_s), ma);
562 lu_object_put(env, &mo_s->mo_lu);
566 static inline void cml_rename_warn(const char *fname,
567 struct md_object *mo_po,
568 struct md_object *mo_pn,
569 const struct lu_fid *lf,
571 struct md_object *mo_t,
576 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
577 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t "DFID"] "
578 "[tname %s] [err %d]\n", fname,
579 PFID(lu_object_fid(&mo_po->mo_lu)),
580 PFID(lu_object_fid(&mo_pn->mo_lu)),
582 PFID(lu_object_fid(&mo_t->mo_lu)),
585 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
586 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t NULL] "
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)),
594 static int cml_rename(const struct lu_env *env, struct md_object *mo_po,
595 struct md_object *mo_pn, const struct lu_fid *lf,
596 const struct lu_name *ls_name, struct md_object *mo_t,
597 const struct lu_name *lt_name, struct md_attr *ma)
599 struct cmm_thread_info *cmi;
600 struct md_attr *tmp_ma = NULL;
601 struct md_object *tmp_t = mo_t;
605 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, &remote);
609 if (mo_t && lu_object_exists(&mo_t->mo_lu) < 0) {
610 /* XXX: mo_t is remote object and there is RPC to unlink it.
611 * before that, do local sanity check for rename first. */
613 struct md_object *mo_s = md_object_find(env,
614 md_obj2dev(mo_po), lf);
616 RETURN(PTR_ERR(mo_s));
618 LASSERT(lu_object_exists(&mo_s->mo_lu) > 0);
619 rc = mo_permission(env, md_object_next(mo_po),
620 md_object_next(mo_s),
622 lu_object_put(env, &mo_s->mo_lu);
626 rc = mo_permission(env, NULL, md_object_next(mo_po),
627 ma, MAY_UNLINK | MAY_VTX_FULL);
632 rc = mo_permission(env, NULL, md_object_next(mo_pn), ma,
633 MAY_UNLINK | MAY_VTX_PART);
638 * XXX: @ma will be changed after mo_ref_del, but we will use
639 * it for mdo_rename later, so save it before mo_ref_del.
641 cmi = cmm_env_info(env);
642 tmp_ma = &cmi->cmi_ma;
644 rc = mo_ref_del(env, md_object_next(mo_t), ma);
648 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
652 /* XXX: for src on remote MDS case, change its ctime before local
653 * rename. Firstly, do local sanity check for rename if necessary. */
656 rc = mo_permission(env, NULL, md_object_next(mo_po),
657 ma, MAY_UNLINK | MAY_VTX_FULL);
662 LASSERT(lu_object_exists(&mo_t->mo_lu) > 0);
663 rc = mo_permission(env, md_object_next(mo_pn),
664 md_object_next(mo_t),
672 mask = (S_ISDIR(ma->ma_attr.la_mode) ?
673 MAY_LINK : MAY_CREATE);
676 rc = mo_permission(env, NULL,
677 md_object_next(mo_pn),
683 ma->ma_attr_flags |= MDS_PERM_BYPASS;
685 LASSERT(tmp_ma->ma_attr_flags & MDS_PERM_BYPASS);
688 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf,
689 tmp_ma ? tmp_ma : ma);
691 /* TODO: revoke mo_t if necessary. */
692 cml_rename_warn("cmm_rename_ctime", mo_po,
693 mo_pn, lf, ls_name->ln_name,
694 tmp_t, lt_name->ln_name, rc);
699 /* local rename, mo_t can be NULL */
700 rc = mdo_rename(env, md_object_next(mo_po),
701 md_object_next(mo_pn), lf, ls_name,
702 md_object_next(mo_t), lt_name, tmp_ma ? tmp_ma : ma);
704 /* TODO: revoke all cml_rename */
705 cml_rename_warn("mdo_rename", mo_po, mo_pn, lf,
706 ls_name->ln_name, tmp_t, lt_name->ln_name, rc);
711 static int cml_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
712 struct md_object *mo_t, const struct lu_fid *lf,
713 const struct lu_name *lname, struct md_attr *ma)
718 rc = mdo_rename_tgt(env, md_object_next(mo_p),
719 md_object_next(mo_t), lf, lname, ma);
722 /* used only in case of rename_tgt() when target is not exist */
723 static int cml_name_insert(const struct lu_env *env, struct md_object *p,
724 const struct lu_name *lname, const struct lu_fid *lf,
725 const struct md_attr *ma)
730 rc = mdo_name_insert(env, md_object_next(p), lname, lf, ma);
735 static int cmm_is_subdir(const struct lu_env *env, struct md_object *mo,
736 const struct lu_fid *fid, struct lu_fid *sfid)
738 struct cmm_thread_info *cmi;
742 cmi = cmm_env_info(env);
743 rc = cmm_mode_get(env, md_obj2dev(mo), fid, &cmi->cmi_ma, NULL);
747 if (!S_ISDIR(cmi->cmi_ma.ma_attr.la_mode))
750 rc = mdo_is_subdir(env, md_object_next(mo), fid, sfid);
754 static struct md_dir_operations cml_dir_ops = {
755 .mdo_is_subdir = cmm_is_subdir,
756 .mdo_lookup = cml_lookup,
757 .mdo_lock_mode = cml_lock_mode,
758 .mdo_create = cml_create,
759 .mdo_link = cml_link,
760 .mdo_unlink = cml_unlink,
761 .mdo_name_insert = cml_name_insert,
762 .mdo_rename = cml_rename,
763 .mdo_rename_tgt = cml_rename_tgt,
764 .mdo_create_data = cml_create_data
767 /* -------------------------------------------------------------------
768 * remote CMM object operations. cmr_...
770 static inline struct cmr_object *lu2cmr_obj(struct lu_object *o)
772 return container_of0(o, struct cmr_object, cmm_obj.cmo_obj.mo_lu);
774 static inline struct cmr_object *md2cmr_obj(struct md_object *mo)
776 return container_of0(mo, struct cmr_object, cmm_obj.cmo_obj);
778 static inline struct cmr_object *cmm2cmr_obj(struct cmm_object *co)
780 return container_of0(co, struct cmr_object, cmm_obj);
783 /* get proper child device from MDCs */
784 static struct lu_device *cmr_child_dev(struct cmm_device *d, __u32 num)
786 struct lu_device *next = NULL;
787 struct mdc_device *mdc;
789 spin_lock(&d->cmm_tgt_guard);
790 list_for_each_entry(mdc, &d->cmm_targets, mc_linkage) {
791 if (mdc->mc_num == num) {
792 next = mdc2lu_dev(mdc);
796 spin_unlock(&d->cmm_tgt_guard);
800 /* lu_object operations */
801 static void cmr_object_free(const struct lu_env *env,
802 struct lu_object *lo)
804 struct cmr_object *cro = lu2cmr_obj(lo);
809 static int cmr_object_init(const struct lu_env *env, struct lu_object *lo)
811 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
812 struct lu_device *c_dev;
813 struct lu_object *c_obj;
818 c_dev = cmr_child_dev(cd, lu2cmr_obj(lo)->cmo_num);
822 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
823 lo->lo_header, c_dev);
825 lu_object_add(lo, c_obj);
835 static int cmr_object_print(const struct lu_env *env, void *cookie,
836 lu_printer_t p, const struct lu_object *lo)
838 return (*p)(env, cookie, LUSTRE_CMM_NAME"-remote@%p", lo);
841 static struct lu_object_operations cmr_obj_ops = {
842 .loo_object_init = cmr_object_init,
843 .loo_object_free = cmr_object_free,
844 .loo_object_print = cmr_object_print
847 /* CMM remote md_object operations. All are invalid */
848 static int cmr_object_create(const struct lu_env *env,
849 struct md_object *mo,
850 const struct md_op_spec *spec,
856 static int cmr_permission(const struct lu_env *env,
857 struct md_object *p, struct md_object *c,
858 struct md_attr *attr, int mask)
863 static int cmr_attr_get(const struct lu_env *env, struct md_object *mo,
864 struct md_attr *attr)
869 static int cmr_attr_set(const struct lu_env *env, struct md_object *mo,
870 const struct md_attr *attr)
875 static int cmr_xattr_get(const struct lu_env *env, struct md_object *mo,
876 struct lu_buf *buf, const char *name)
881 static int cmr_readlink(const struct lu_env *env, struct md_object *mo,
887 static int cmr_xattr_list(const struct lu_env *env, struct md_object *mo,
893 static int cmr_xattr_set(const struct lu_env *env, struct md_object *mo,
894 const struct lu_buf *buf, const char *name,
900 static int cmr_xattr_del(const struct lu_env *env, struct md_object *mo,
906 static int cmr_ref_add(const struct lu_env *env, struct md_object *mo,
907 const struct md_attr *ma)
912 static int cmr_ref_del(const struct lu_env *env, struct md_object *mo,
918 static int cmr_open(const struct lu_env *env, struct md_object *mo,
924 static int cmr_close(const struct lu_env *env, struct md_object *mo,
930 static int cmr_readpage(const struct lu_env *env, struct md_object *mo,
931 const struct lu_rdpg *rdpg)
936 static int cmr_capa_get(const struct lu_env *env, struct md_object *mo,
937 struct lustre_capa *capa, int renewal)
942 static struct md_object_operations cmr_mo_ops = {
943 .moo_permission = cmr_permission,
944 .moo_attr_get = cmr_attr_get,
945 .moo_attr_set = cmr_attr_set,
946 .moo_xattr_get = cmr_xattr_get,
947 .moo_xattr_set = cmr_xattr_set,
948 .moo_xattr_list = cmr_xattr_list,
949 .moo_xattr_del = cmr_xattr_del,
950 .moo_object_create = cmr_object_create,
951 .moo_ref_add = cmr_ref_add,
952 .moo_ref_del = cmr_ref_del,
953 .moo_open = cmr_open,
954 .moo_close = cmr_close,
955 .moo_readpage = cmr_readpage,
956 .moo_readlink = cmr_readlink,
957 .moo_capa_get = cmr_capa_get
960 /* remote part of md_dir operations */
961 static int cmr_lookup(const struct lu_env *env, struct md_object *mo_p,
962 const struct lu_name *lname, struct lu_fid *lf,
963 struct md_op_spec *spec)
966 * This can happens while rename() If new parent is remote dir, lookup
973 static mdl_mode_t cmr_lock_mode(const struct lu_env *env,
974 struct md_object *mo, mdl_mode_t lm)
980 * All methods below are cross-ref by nature. They consist of remote call and
981 * local operation. Due to future rollback functionality there are several
982 * limitations for such methods:
983 * 1) remote call should be done at first to do epoch negotiation between all
984 * MDS involved and to avoid the RPC inside transaction.
985 * 2) only one RPC can be sent - also due to epoch negotiation.
986 * For more details see rollback HLD/DLD.
988 static int cmr_create(const struct lu_env *env, struct md_object *mo_p,
989 const struct lu_name *lchild_name, struct md_object *mo_c,
990 struct md_op_spec *spec,
993 struct cmm_thread_info *cmi;
994 struct md_attr *tmp_ma;
998 /* Make sure that name isn't exist before doing remote call. */
999 rc = mdo_lookup(env, md_object_next(mo_p), lchild_name,
1000 &cmm_env_info(env)->cmi_fid, NULL);
1003 else if (rc != -ENOENT)
1006 /* check the SGID attr */
1007 cmi = cmm_env_info(env);
1009 tmp_ma = &cmi->cmi_ma;
1010 tmp_ma->ma_valid = 0;
1011 tmp_ma->ma_need = MA_INODE;
1013 #ifdef CONFIG_FS_POSIX_ACL
1014 if (!S_ISLNK(ma->ma_attr.la_mode)) {
1015 tmp_ma->ma_acl = cmi->cmi_xattr_buf;
1016 tmp_ma->ma_acl_size = sizeof(cmi->cmi_xattr_buf);
1017 tmp_ma->ma_need |= MA_ACL_DEF;
1020 rc = mo_attr_get(env, md_object_next(mo_p), tmp_ma);
1024 if (tmp_ma->ma_attr.la_mode & S_ISGID) {
1025 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
1026 if (S_ISDIR(ma->ma_attr.la_mode)) {
1027 ma->ma_attr.la_mode |= S_ISGID;
1028 ma->ma_attr.la_valid |= LA_MODE;
1032 #ifdef CONFIG_FS_POSIX_ACL
1033 if (tmp_ma->ma_valid & MA_ACL_DEF) {
1034 spec->u.sp_ea.fid = spec->u.sp_pfid;
1035 spec->u.sp_ea.eadata = tmp_ma->ma_acl;
1036 spec->u.sp_ea.eadatalen = tmp_ma->ma_acl_size;
1037 spec->sp_cr_flags |= MDS_CREATE_RMT_ACL;
1041 /* Local permission check for name_insert before remote ops. */
1042 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1043 (S_ISDIR(ma->ma_attr.la_mode) ?
1044 MAY_LINK : MAY_CREATE));
1048 /* Remote object creation and local name insert. */
1050 * XXX: @ma will be changed after mo_object_create, but we will use
1051 * it for mdo_name_insert later, so save it before mo_object_create.
1054 rc = mo_object_create(env, md_object_next(mo_c), spec, ma);
1056 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1057 rc = mdo_name_insert(env, md_object_next(mo_p), lchild_name,
1058 lu_object_fid(&mo_c->mo_lu), tmp_ma);
1060 /* TODO: remove object mo_c on remote MDS */
1061 CWARN("cmr_create failed, should revoke: [mo_p "DFID"]"
1062 " [name %s] [mo_c "DFID"] [err %d]\n",
1063 PFID(lu_object_fid(&mo_p->mo_lu)),
1064 lchild_name->ln_name,
1065 PFID(lu_object_fid(&mo_c->mo_lu)), rc);
1072 static int cmr_link(const struct lu_env *env, struct md_object *mo_p,
1073 struct md_object *mo_s, const struct lu_name *lname,
1079 /* Make sure that name isn't exist before doing remote call. */
1080 rc = mdo_lookup(env, md_object_next(mo_p), lname,
1081 &cmm_env_info(env)->cmi_fid, NULL);
1084 } else if (rc == -ENOENT) {
1085 /* Local permission check for name_insert before remote ops. */
1086 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1091 rc = mo_ref_add(env, md_object_next(mo_s), ma);
1093 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1094 rc = mdo_name_insert(env, md_object_next(mo_p), lname,
1095 lu_object_fid(&mo_s->mo_lu), ma);
1097 /* TODO: ref_del from mo_s on remote MDS */
1098 CWARN("cmr_link failed, should revoke: "
1099 "[mo_p "DFID"] [mo_s "DFID"] "
1100 "[name %s] [err %d]\n",
1101 PFID(lu_object_fid(&mo_p->mo_lu)),
1102 PFID(lu_object_fid(&mo_s->mo_lu)),
1103 lname->ln_name, rc);
1110 static int cmr_unlink(const struct lu_env *env, struct md_object *mo_p,
1111 struct md_object *mo_c, const struct lu_name *lname,
1114 struct cmm_thread_info *cmi;
1115 struct md_attr *tmp_ma;
1119 /* Local permission check for name_remove before remote ops. */
1120 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1121 MAY_UNLINK | MAY_VTX_PART);
1126 * XXX: @ma will be changed after mo_ref_del, but we will use
1127 * it for mdo_name_remove later, so save it before mo_ref_del.
1129 cmi = cmm_env_info(env);
1130 tmp_ma = &cmi->cmi_ma;
1132 rc = mo_ref_del(env, md_object_next(mo_c), ma);
1134 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1135 rc = mdo_name_remove(env, md_object_next(mo_p), lname, tmp_ma);
1137 /* TODO: ref_add to mo_c on remote MDS */
1138 CWARN("cmr_unlink failed, should revoke: [mo_p "DFID"]"
1139 " [mo_c "DFID"] [name %s] [err %d]\n",
1140 PFID(lu_object_fid(&mo_p->mo_lu)),
1141 PFID(lu_object_fid(&mo_c->mo_lu)),
1142 lname->ln_name, rc);
1149 static inline void cmr_rename_warn(const char *fname,
1150 struct md_object *mo_po,
1151 struct md_object *mo_pn,
1152 const struct lu_fid *lf,
1157 CWARN("cmr_rename failed for %s, should revoke: "
1158 "[mo_po "DFID"] [mo_pn "DFID"] [lf "DFID"] "
1159 "[sname %s] [tname %s] [err %d]\n", fname,
1160 PFID(lu_object_fid(&mo_po->mo_lu)),
1161 PFID(lu_object_fid(&mo_pn->mo_lu)),
1162 PFID(lf), s_name, t_name, err);
1165 static int cmr_rename(const struct lu_env *env,
1166 struct md_object *mo_po, struct md_object *mo_pn,
1167 const struct lu_fid *lf, const struct lu_name *ls_name,
1168 struct md_object *mo_t, const struct lu_name *lt_name,
1171 struct cmm_thread_info *cmi;
1172 struct md_attr *tmp_ma;
1176 LASSERT(mo_t == NULL);
1178 /* get real type of src */
1179 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, NULL);
1183 /* Local permission check for name_remove before remote ops. */
1184 rc = mo_permission(env, NULL, md_object_next(mo_po), ma,
1185 MAY_UNLINK | MAY_VTX_FULL);
1190 * XXX: @ma maybe changed after mdo_rename_tgt, but we will use it
1191 * for mdo_name_remove later, so save it before mdo_rename_tgt.
1193 cmi = cmm_env_info(env);
1194 tmp_ma = &cmi->cmi_ma;
1196 /* the mo_pn is remote directory, so we cannot even know if there is
1197 * mo_t or not. Therefore mo_t is NULL here but remote server should do
1198 * lookup and process this further */
1199 rc = mdo_rename_tgt(env, md_object_next(mo_pn),
1200 NULL/* mo_t */, lf, lt_name, ma);
1204 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1206 /* src object maybe on remote MDS, do remote ops first. */
1207 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf, tmp_ma);
1209 /* TODO: revoke mdo_rename_tgt */
1210 cmr_rename_warn("cmm_rename_ctime", mo_po, mo_pn, lf,
1211 ls_name->ln_name, lt_name->ln_name, rc);
1215 /* only old name is removed localy */
1216 rc = mdo_name_remove(env, md_object_next(mo_po), ls_name, tmp_ma);
1218 /* TODO: revoke all cmr_rename */
1219 cmr_rename_warn("mdo_name_remove", mo_po, mo_pn, lf,
1220 ls_name->ln_name, lt_name->ln_name, rc);
1225 /* part of cross-ref rename(). Used to insert new name in new parent
1226 * and unlink target */
1227 static int cmr_rename_tgt(const struct lu_env *env,
1228 struct md_object *mo_p, struct md_object *mo_t,
1229 const struct lu_fid *lf, const struct lu_name *lname,
1232 struct cmm_thread_info *cmi;
1233 struct md_attr *tmp_ma;
1237 /* target object is remote one */
1238 /* Local permission check for rename_tgt before remote ops. */
1239 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1240 MAY_UNLINK | MAY_VTX_PART);
1245 * XXX: @ma maybe changed after mo_ref_del, but we will use
1246 * it for mdo_rename_tgt later, so save it before mo_ref_del.
1248 cmi = cmm_env_info(env);
1249 tmp_ma = &cmi->cmi_ma;
1251 rc = mo_ref_del(env, md_object_next(mo_t), ma);
1252 /* continue locally with name handling only */
1254 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1255 rc = mdo_rename_tgt(env, md_object_next(mo_p),
1256 NULL, lf, lname, tmp_ma);
1258 /* TODO: ref_add to mo_t on remote MDS */
1259 CWARN("cmr_rename_tgt failed, should revoke: "
1260 "[mo_p "DFID"] [mo_t "DFID"] [lf "DFID"] "
1261 "[name %s] [err %d]\n",
1262 PFID(lu_object_fid(&mo_p->mo_lu)),
1263 PFID(lu_object_fid(&mo_t->mo_lu)),
1265 lname->ln_name, rc);
1271 static struct md_dir_operations cmr_dir_ops = {
1272 .mdo_is_subdir = cmm_is_subdir,
1273 .mdo_lookup = cmr_lookup,
1274 .mdo_lock_mode = cmr_lock_mode,
1275 .mdo_create = cmr_create,
1276 .mdo_link = cmr_link,
1277 .mdo_unlink = cmr_unlink,
1278 .mdo_rename = cmr_rename,
1279 .mdo_rename_tgt = cmr_rename_tgt,