1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/cmm/cmm_object.c
5 * Lustre Cluster Metadata Manager (cmm)
7 * Copyright (c) 2006 Cluster File Systems, Inc.
8 * Author: Mike Pershin <tappro@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
30 # define EXPORT_SYMTAB
33 #define DEBUG_SUBSYSTEM S_MDS
35 #include <lustre_fid.h>
36 #include "cmm_internal.h"
37 #include "mdc_internal.h"
39 int cmm_fld_lookup(struct cmm_device *cm, const struct lu_fid *fid,
40 mdsno_t *mds, const struct lu_env *env)
45 LASSERT(fid_is_sane(fid));
47 rc = fld_client_lookup(cm->cmm_fld, fid_seq(fid), mds, env);
49 CERROR("Can't find mds by seq "LPX64", rc %d\n",
54 if (*mds > cm->cmm_tgt_count) {
55 CERROR("Got invalid mdsno: "LPU64" (max: %u)\n",
56 *mds, cm->cmm_tgt_count);
59 CDEBUG(D_INFO, "CMM: got MDS "LPU64" for sequence: "
60 LPU64"\n", *mds, fid_seq(fid));
66 static struct md_object_operations cml_mo_ops;
67 static struct md_dir_operations cml_dir_ops;
68 static struct lu_object_operations cml_obj_ops;
70 static struct md_object_operations cmr_mo_ops;
71 static struct md_dir_operations cmr_dir_ops;
72 static struct lu_object_operations cmr_obj_ops;
74 struct lu_object *cmm_object_alloc(const struct lu_env *env,
75 const struct lu_object_header *loh,
78 const struct lu_fid *fid = &loh->loh_fid;
79 struct lu_object *lo = NULL;
80 struct cmm_device *cd;
87 if (cd->cmm_flags & CMM_INITIALIZED) {
88 /* get object location */
89 rc = cmm_fld_lookup(lu2cmm_dev(ld), fid, &mds, env);
94 * Device is not yet initialized, cmm_object is being created
95 * as part of early bootstrap procedure (it is /ROOT, or /fld,
96 * etc.). Such object *has* to be local.
98 mds = cd->cmm_local_num;
100 /* select the proper set of operations based on object location */
101 if (mds == cd->cmm_local_num) {
102 struct cml_object *clo;
106 lo = &clo->cmm_obj.cmo_obj.mo_lu;
107 lu_object_init(lo, NULL, ld);
108 clo->cmm_obj.cmo_obj.mo_ops = &cml_mo_ops;
109 clo->cmm_obj.cmo_obj.mo_dir_ops = &cml_dir_ops;
110 lo->lo_ops = &cml_obj_ops;
113 struct cmr_object *cro;
117 lo = &cro->cmm_obj.cmo_obj.mo_lu;
118 lu_object_init(lo, NULL, ld);
119 cro->cmm_obj.cmo_obj.mo_ops = &cmr_mo_ops;
120 cro->cmm_obj.cmo_obj.mo_dir_ops = &cmr_dir_ops;
121 lo->lo_ops = &cmr_obj_ops;
129 * CMM has two types of objects - local and remote. They have different set
130 * of operations so we are avoiding multiple checks in code.
133 /* get local child device */
134 static struct lu_device *cml_child_dev(struct cmm_device *d)
136 return &d->cmm_child->md_lu_dev;
139 /* lu_object operations */
140 static void cml_object_free(const struct lu_env *env,
141 struct lu_object *lo)
143 struct cml_object *clo = lu2cml_obj(lo);
148 static int cml_object_init(const struct lu_env *env, struct lu_object *lo)
150 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
151 struct lu_device *c_dev;
152 struct lu_object *c_obj;
157 #ifdef HAVE_SPLIT_SUPPORT
158 if (cd->cmm_tgt_count == 0)
159 lu2cml_obj(lo)->clo_split = CMM_SPLIT_DENIED;
161 lu2cml_obj(lo)->clo_split = CMM_SPLIT_UNKNOWN;
163 c_dev = cml_child_dev(cd);
167 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
168 lo->lo_header, c_dev);
170 lu_object_add(lo, c_obj);
180 static int cml_object_print(const struct lu_env *env, void *cookie,
181 lu_printer_t p, const struct lu_object *lo)
183 return (*p)(env, cookie, LUSTRE_CMM_NAME"-local@%p", lo);
186 static struct lu_object_operations cml_obj_ops = {
187 .loo_object_init = cml_object_init,
188 .loo_object_free = cml_object_free,
189 .loo_object_print = cml_object_print
192 /* CMM local md_object operations */
193 static int cml_object_create(const struct lu_env *env,
194 struct md_object *mo,
195 const struct md_op_spec *spec,
196 struct md_attr *attr)
200 rc = mo_object_create(env, md_object_next(mo), spec, attr);
204 static int cml_permission(const struct lu_env *env,
205 struct md_object *p, struct md_object *c,
206 struct md_attr *attr, int mask)
210 rc = mo_permission(env, md_object_next(p), md_object_next(c),
215 static int cml_attr_get(const struct lu_env *env, struct md_object *mo,
216 struct md_attr *attr)
220 rc = mo_attr_get(env, md_object_next(mo), attr);
224 static int cml_attr_set(const struct lu_env *env, struct md_object *mo,
225 const struct md_attr *attr)
229 rc = mo_attr_set(env, md_object_next(mo), attr);
233 static int cml_xattr_get(const struct lu_env *env, struct md_object *mo,
234 struct lu_buf *buf, const char *name)
238 rc = mo_xattr_get(env, md_object_next(mo), buf, name);
242 static int cml_readlink(const struct lu_env *env, struct md_object *mo,
247 rc = mo_readlink(env, md_object_next(mo), buf);
251 static int cml_xattr_list(const struct lu_env *env, struct md_object *mo,
256 rc = mo_xattr_list(env, md_object_next(mo), buf);
260 static int cml_xattr_set(const struct lu_env *env, struct md_object *mo,
261 const struct lu_buf *buf,
262 const char *name, int fl)
266 rc = mo_xattr_set(env, md_object_next(mo), buf, name, fl);
270 static int cml_xattr_del(const struct lu_env *env, struct md_object *mo,
275 rc = mo_xattr_del(env, md_object_next(mo), name);
279 static int cml_ref_add(const struct lu_env *env, struct md_object *mo,
280 const struct md_attr *ma)
284 rc = mo_ref_add(env, md_object_next(mo), ma);
288 static int cml_ref_del(const struct lu_env *env, struct md_object *mo,
293 rc = mo_ref_del(env, md_object_next(mo), ma);
297 static int cml_open(const struct lu_env *env, struct md_object *mo,
302 rc = mo_open(env, md_object_next(mo), flags);
306 static int cml_close(const struct lu_env *env, struct md_object *mo,
311 rc = mo_close(env, md_object_next(mo), ma);
315 static int cml_readpage(const struct lu_env *env, struct md_object *mo,
316 const struct lu_rdpg *rdpg)
320 rc = mo_readpage(env, md_object_next(mo), rdpg);
324 static int cml_capa_get(const struct lu_env *env, struct md_object *mo,
325 struct lustre_capa *capa, int renewal)
329 rc = mo_capa_get(env, md_object_next(mo), capa, renewal);
333 static struct md_object_operations cml_mo_ops = {
334 .moo_permission = cml_permission,
335 .moo_attr_get = cml_attr_get,
336 .moo_attr_set = cml_attr_set,
337 .moo_xattr_get = cml_xattr_get,
338 .moo_xattr_list = cml_xattr_list,
339 .moo_xattr_set = cml_xattr_set,
340 .moo_xattr_del = cml_xattr_del,
341 .moo_object_create = cml_object_create,
342 .moo_ref_add = cml_ref_add,
343 .moo_ref_del = cml_ref_del,
344 .moo_open = cml_open,
345 .moo_close = cml_close,
346 .moo_readpage = cml_readpage,
347 .moo_readlink = cml_readlink,
348 .moo_capa_get = cml_capa_get
351 /* md_dir operations */
352 static int cml_lookup(const struct lu_env *env, struct md_object *mo_p,
353 const struct lu_name *lname, struct lu_fid *lf,
354 struct md_op_spec *spec)
359 #ifdef HAVE_SPLIT_SUPPORT
360 if (spec != NULL && spec->sp_ck_split) {
361 rc = cmm_split_check(env, mo_p, lname->ln_name);
366 rc = mdo_lookup(env, md_object_next(mo_p), lname, lf, spec);
371 static mdl_mode_t cml_lock_mode(const struct lu_env *env,
372 struct md_object *mo, mdl_mode_t lm)
374 int rc = MDL_MINMODE;
377 #ifdef HAVE_SPLIT_SUPPORT
378 rc = cmm_split_access(env, mo, lm);
384 static int cml_create(const struct lu_env *env, struct md_object *mo_p,
385 const struct lu_name *lname, struct md_object *mo_c,
386 struct md_op_spec *spec, struct md_attr *ma)
391 #ifdef HAVE_SPLIT_SUPPORT
392 /* Lock mode always should be sane. */
393 LASSERT(spec->sp_cr_mode != MDL_MINMODE);
396 * Sigh... This is long story. MDT may have race with detecting if split
397 * is possible in cmm. We know this race and let it live, because
398 * getting it rid (with some sem or spinlock) will also mean that
399 * PDIROPS for create will not work because we kill parallel work, what
400 * is really bad for performance and makes no sense having PDIROPS. So,
401 * we better allow the race to live, but split dir only if some of
402 * concurrent threads takes EX lock, not matter which one. So that, say,
403 * two concurrent threads may have different lock modes on directory (CW
404 * and EX) and not first one which comes here and see that split is
405 * possible should split the dir, but only that one which has EX
406 * lock. And we do not care that in this case, split may happen a bit
407 * later (when dir size will not be necessarily 64K, but may be a bit
408 * larger). So that, we allow concurrent creates and protect split by EX
411 if (spec->sp_cr_mode == MDL_EX) {
413 * Try to split @mo_p. If split is ok, -ERESTART is returned and
414 * current thread will not peoceed with create. Instead it sends
415 * -ERESTART to client to let it know that correct MDT should be
418 rc = cmm_split_dir(env, mo_p);
421 * -ERESTART or some split error is returned, we can't
422 * proceed with create.
427 if (spec != NULL && spec->sp_ck_split) {
429 * Check for possible split directory and let caller know that
430 * it should tell client that directory is split and operation
431 * should repeat to correct MDT.
433 rc = cmm_split_check(env, mo_p, lname->ln_name);
439 rc = mdo_create(env, md_object_next(mo_p), lname, md_object_next(mo_c),
447 static int cml_create_data(const struct lu_env *env, struct md_object *p,
449 const struct md_op_spec *spec,
454 rc = mdo_create_data(env, md_object_next(p), md_object_next(o),
459 static int cml_link(const struct lu_env *env, struct md_object *mo_p,
460 struct md_object *mo_s, const struct lu_name *lname,
465 rc = mdo_link(env, md_object_next(mo_p), md_object_next(mo_s),
470 static int cml_unlink(const struct lu_env *env, struct md_object *mo_p,
471 struct md_object *mo_c, const struct lu_name *lname,
476 rc = mdo_unlink(env, md_object_next(mo_p), md_object_next(mo_c),
481 /* rename is split to local/remote by location of new parent dir */
482 struct md_object *md_object_find(const struct lu_env *env,
483 struct md_device *md,
484 const struct lu_fid *f)
490 o = lu_object_find(env, md2lu_dev(md)->ld_site, f);
492 m = (struct md_object *)o;
494 o = lu_object_locate(o->lo_header, md2lu_dev(md)->ld_type);
495 m = o ? lu2md(o) : NULL;
500 static int cmm_mode_get(const struct lu_env *env, struct md_device *md,
501 const struct lu_fid *lf, struct md_attr *ma,
504 struct md_object *mo_s = md_object_find(env, md, lf);
505 struct cmm_thread_info *cmi;
506 struct md_attr *tmp_ma;
511 RETURN(PTR_ERR(mo_s));
513 if (remote && (lu_object_exists(&mo_s->mo_lu) < 0))
516 cmi = cmm_env_info(env);
517 tmp_ma = &cmi->cmi_ma;
518 tmp_ma->ma_need = MA_INODE;
519 tmp_ma->ma_valid = 0;
520 /* get type from src, can be remote req */
521 rc = mo_attr_get(env, md_object_next(mo_s), tmp_ma);
523 ma->ma_attr.la_mode = tmp_ma->ma_attr.la_mode;
524 ma->ma_attr.la_uid = tmp_ma->ma_attr.la_uid;
525 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
526 ma->ma_attr.la_flags = tmp_ma->ma_attr.la_flags;
527 ma->ma_attr.la_valid |= LA_MODE | LA_UID | LA_GID | LA_FLAGS;
529 lu_object_put(env, &mo_s->mo_lu);
533 static int cmm_rename_ctime(const struct lu_env *env, struct md_device *md,
534 const struct lu_fid *lf, struct md_attr *ma)
536 struct md_object *mo_s = md_object_find(env, md, lf);
541 RETURN(PTR_ERR(mo_s));
543 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
544 /* set ctime to obj, can be remote req */
545 rc = mo_attr_set(env, md_object_next(mo_s), ma);
546 lu_object_put(env, &mo_s->mo_lu);
550 static inline void cml_rename_warn(const char *fname,
551 struct md_object *mo_po,
552 struct md_object *mo_pn,
553 const struct lu_fid *lf,
555 struct md_object *mo_t,
560 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
561 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t "DFID"] "
562 "[tname %s] [err %d]\n", fname,
563 PFID(lu_object_fid(&mo_po->mo_lu)),
564 PFID(lu_object_fid(&mo_pn->mo_lu)),
566 PFID(lu_object_fid(&mo_t->mo_lu)),
569 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
570 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t NULL] "
571 "[tname %s] [err %d]\n", fname,
572 PFID(lu_object_fid(&mo_po->mo_lu)),
573 PFID(lu_object_fid(&mo_pn->mo_lu)),
578 static int cml_rename(const struct lu_env *env, struct md_object *mo_po,
579 struct md_object *mo_pn, const struct lu_fid *lf,
580 const struct lu_name *ls_name, struct md_object *mo_t,
581 const struct lu_name *lt_name, struct md_attr *ma)
583 struct cmm_thread_info *cmi;
584 struct md_attr *tmp_ma = NULL;
585 struct md_object *tmp_t = mo_t;
589 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, &remote);
593 if (mo_t && lu_object_exists(&mo_t->mo_lu) < 0) {
594 /* XXX: mo_t is remote object and there is RPC to unlink it.
595 * before that, do local sanity check for rename first. */
597 struct md_object *mo_s = md_object_find(env,
598 md_obj2dev(mo_po), lf);
600 RETURN(PTR_ERR(mo_s));
602 LASSERT(lu_object_exists(&mo_s->mo_lu) > 0);
603 rc = mo_permission(env, md_object_next(mo_po),
604 md_object_next(mo_s),
606 lu_object_put(env, &mo_s->mo_lu);
610 rc = mo_permission(env, NULL, md_object_next(mo_po),
611 ma, MAY_UNLINK | MAY_VTX_FULL);
616 rc = mo_permission(env, NULL, md_object_next(mo_pn), ma,
617 MAY_UNLINK | MAY_VTX_PART);
622 * XXX: @ma will be changed after mo_ref_del, but we will use
623 * it for mdo_rename later, so save it before mo_ref_del.
625 cmi = cmm_env_info(env);
626 tmp_ma = &cmi->cmi_ma;
628 rc = mo_ref_del(env, md_object_next(mo_t), ma);
632 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
636 /* XXX: for src on remote MDS case, change its ctime before local
637 * rename. Firstly, do local sanity check for rename if necessary. */
640 rc = mo_permission(env, NULL, md_object_next(mo_po),
641 ma, MAY_UNLINK | MAY_VTX_FULL);
646 LASSERT(lu_object_exists(&mo_t->mo_lu) > 0);
647 rc = mo_permission(env, md_object_next(mo_pn),
648 md_object_next(mo_t),
656 mask = (S_ISDIR(ma->ma_attr.la_mode) ?
657 MAY_LINK : MAY_CREATE);
660 rc = mo_permission(env, NULL,
661 md_object_next(mo_pn),
667 ma->ma_attr_flags |= MDS_PERM_BYPASS;
669 LASSERT(tmp_ma->ma_attr_flags & MDS_PERM_BYPASS);
672 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf,
673 tmp_ma ? tmp_ma : ma);
675 /* TODO: revoke mo_t if necessary. */
676 cml_rename_warn("cmm_rename_ctime", mo_po,
677 mo_pn, lf, ls_name->ln_name,
678 tmp_t, lt_name->ln_name, rc);
683 /* local rename, mo_t can be NULL */
684 rc = mdo_rename(env, md_object_next(mo_po),
685 md_object_next(mo_pn), lf, ls_name,
686 md_object_next(mo_t), lt_name, tmp_ma ? tmp_ma : ma);
688 /* TODO: revoke all cml_rename */
689 cml_rename_warn("mdo_rename", mo_po, mo_pn, lf,
690 ls_name->ln_name, tmp_t, lt_name->ln_name, rc);
695 static int cml_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
696 struct md_object *mo_t, const struct lu_fid *lf,
697 const struct lu_name *lname, struct md_attr *ma)
702 rc = mdo_rename_tgt(env, md_object_next(mo_p),
703 md_object_next(mo_t), lf, lname, ma);
706 /* used only in case of rename_tgt() when target is not exist */
707 static int cml_name_insert(const struct lu_env *env, struct md_object *p,
708 const struct lu_name *lname, const struct lu_fid *lf,
709 const struct md_attr *ma)
714 rc = mdo_name_insert(env, md_object_next(p), lname, lf, ma);
719 static int cmm_is_subdir(const struct lu_env *env, struct md_object *mo,
720 const struct lu_fid *fid, struct lu_fid *sfid)
722 struct cmm_thread_info *cmi;
726 cmi = cmm_env_info(env);
727 rc = cmm_mode_get(env, md_obj2dev(mo), fid, &cmi->cmi_ma, NULL);
731 if (!S_ISDIR(cmi->cmi_ma.ma_attr.la_mode))
734 rc = mdo_is_subdir(env, md_object_next(mo), fid, sfid);
738 static struct md_dir_operations cml_dir_ops = {
739 .mdo_is_subdir = cmm_is_subdir,
740 .mdo_lookup = cml_lookup,
741 .mdo_lock_mode = cml_lock_mode,
742 .mdo_create = cml_create,
743 .mdo_link = cml_link,
744 .mdo_unlink = cml_unlink,
745 .mdo_name_insert = cml_name_insert,
746 .mdo_rename = cml_rename,
747 .mdo_rename_tgt = cml_rename_tgt,
748 .mdo_create_data = cml_create_data
751 /* -------------------------------------------------------------------
752 * remote CMM object operations. cmr_...
754 static inline struct cmr_object *lu2cmr_obj(struct lu_object *o)
756 return container_of0(o, struct cmr_object, cmm_obj.cmo_obj.mo_lu);
758 static inline struct cmr_object *md2cmr_obj(struct md_object *mo)
760 return container_of0(mo, struct cmr_object, cmm_obj.cmo_obj);
762 static inline struct cmr_object *cmm2cmr_obj(struct cmm_object *co)
764 return container_of0(co, struct cmr_object, cmm_obj);
767 /* get proper child device from MDCs */
768 static struct lu_device *cmr_child_dev(struct cmm_device *d, __u32 num)
770 struct lu_device *next = NULL;
771 struct mdc_device *mdc;
773 spin_lock(&d->cmm_tgt_guard);
774 list_for_each_entry(mdc, &d->cmm_targets, mc_linkage) {
775 if (mdc->mc_num == num) {
776 next = mdc2lu_dev(mdc);
780 spin_unlock(&d->cmm_tgt_guard);
784 /* lu_object operations */
785 static void cmr_object_free(const struct lu_env *env,
786 struct lu_object *lo)
788 struct cmr_object *cro = lu2cmr_obj(lo);
793 static int cmr_object_init(const struct lu_env *env, struct lu_object *lo)
795 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
796 struct lu_device *c_dev;
797 struct lu_object *c_obj;
802 c_dev = cmr_child_dev(cd, lu2cmr_obj(lo)->cmo_num);
806 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
807 lo->lo_header, c_dev);
809 lu_object_add(lo, c_obj);
819 static int cmr_object_print(const struct lu_env *env, void *cookie,
820 lu_printer_t p, const struct lu_object *lo)
822 return (*p)(env, cookie, LUSTRE_CMM_NAME"-remote@%p", lo);
825 static struct lu_object_operations cmr_obj_ops = {
826 .loo_object_init = cmr_object_init,
827 .loo_object_free = cmr_object_free,
828 .loo_object_print = cmr_object_print
831 /* CMM remote md_object operations. All are invalid */
832 static int cmr_object_create(const struct lu_env *env,
833 struct md_object *mo,
834 const struct md_op_spec *spec,
840 static int cmr_permission(const struct lu_env *env,
841 struct md_object *p, struct md_object *c,
842 struct md_attr *attr, int mask)
847 static int cmr_attr_get(const struct lu_env *env, struct md_object *mo,
848 struct md_attr *attr)
853 static int cmr_attr_set(const struct lu_env *env, struct md_object *mo,
854 const struct md_attr *attr)
859 static int cmr_xattr_get(const struct lu_env *env, struct md_object *mo,
860 struct lu_buf *buf, const char *name)
865 static int cmr_readlink(const struct lu_env *env, struct md_object *mo,
871 static int cmr_xattr_list(const struct lu_env *env, struct md_object *mo,
877 static int cmr_xattr_set(const struct lu_env *env, struct md_object *mo,
878 const struct lu_buf *buf, const char *name, int fl)
883 static int cmr_xattr_del(const struct lu_env *env, struct md_object *mo,
889 static int cmr_ref_add(const struct lu_env *env, struct md_object *mo,
890 const struct md_attr *ma)
895 static int cmr_ref_del(const struct lu_env *env, struct md_object *mo,
901 static int cmr_open(const struct lu_env *env, struct md_object *mo,
907 static int cmr_close(const struct lu_env *env, struct md_object *mo,
913 static int cmr_readpage(const struct lu_env *env, struct md_object *mo,
914 const struct lu_rdpg *rdpg)
919 static int cmr_capa_get(const struct lu_env *env, struct md_object *mo,
920 struct lustre_capa *capa, int renewal)
925 static struct md_object_operations cmr_mo_ops = {
926 .moo_permission = cmr_permission,
927 .moo_attr_get = cmr_attr_get,
928 .moo_attr_set = cmr_attr_set,
929 .moo_xattr_get = cmr_xattr_get,
930 .moo_xattr_set = cmr_xattr_set,
931 .moo_xattr_list = cmr_xattr_list,
932 .moo_xattr_del = cmr_xattr_del,
933 .moo_object_create = cmr_object_create,
934 .moo_ref_add = cmr_ref_add,
935 .moo_ref_del = cmr_ref_del,
936 .moo_open = cmr_open,
937 .moo_close = cmr_close,
938 .moo_readpage = cmr_readpage,
939 .moo_readlink = cmr_readlink,
940 .moo_capa_get = cmr_capa_get
943 /* remote part of md_dir operations */
944 static int cmr_lookup(const struct lu_env *env, struct md_object *mo_p,
945 const struct lu_name *lname, struct lu_fid *lf,
946 struct md_op_spec *spec)
949 * This can happens while rename() If new parent is remote dir, lookup
956 static mdl_mode_t cmr_lock_mode(const struct lu_env *env,
957 struct md_object *mo, mdl_mode_t lm)
963 * All methods below are cross-ref by nature. They consist of remote call and
964 * local operation. Due to future rollback functionality there are several
965 * limitations for such methods:
966 * 1) remote call should be done at first to do epoch negotiation between all
967 * MDS involved and to avoid the RPC inside transaction.
968 * 2) only one RPC can be sent - also due to epoch negotiation.
969 * For more details see rollback HLD/DLD.
971 static int cmr_create(const struct lu_env *env, struct md_object *mo_p,
972 const struct lu_name *lchild_name, struct md_object *mo_c,
973 struct md_op_spec *spec,
976 struct cmm_thread_info *cmi;
977 struct md_attr *tmp_ma;
981 /* Make sure that name isn't exist before doing remote call. */
982 rc = mdo_lookup(env, md_object_next(mo_p), lchild_name,
983 &cmm_env_info(env)->cmi_fid, NULL);
986 else if (rc != -ENOENT)
989 /* check the SGID attr */
990 cmi = cmm_env_info(env);
992 tmp_ma = &cmi->cmi_ma;
993 tmp_ma->ma_valid = 0;
994 tmp_ma->ma_need = MA_INODE;
996 #ifdef CONFIG_FS_POSIX_ACL
997 if (!S_ISLNK(ma->ma_attr.la_mode)) {
998 tmp_ma->ma_acl = cmi->cmi_xattr_buf;
999 tmp_ma->ma_acl_size = sizeof(cmi->cmi_xattr_buf);
1000 tmp_ma->ma_need |= MA_ACL_DEF;
1003 rc = mo_attr_get(env, md_object_next(mo_p), tmp_ma);
1007 if (tmp_ma->ma_attr.la_mode & S_ISGID) {
1008 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
1009 if (S_ISDIR(ma->ma_attr.la_mode)) {
1010 ma->ma_attr.la_mode |= S_ISGID;
1011 ma->ma_attr.la_valid |= LA_MODE;
1015 #ifdef CONFIG_FS_POSIX_ACL
1016 if (tmp_ma->ma_valid & MA_ACL_DEF) {
1017 spec->u.sp_ea.fid = spec->u.sp_pfid;
1018 spec->u.sp_ea.eadata = tmp_ma->ma_acl;
1019 spec->u.sp_ea.eadatalen = tmp_ma->ma_acl_size;
1020 spec->sp_cr_flags |= MDS_CREATE_RMT_ACL;
1024 /* Local permission check for name_insert before remote ops. */
1025 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1026 (S_ISDIR(ma->ma_attr.la_mode) ?
1027 MAY_LINK : MAY_CREATE));
1031 /* Remote object creation and local name insert. */
1033 * XXX: @ma will be changed after mo_object_create, but we will use
1034 * it for mdo_name_insert later, so save it before mo_object_create.
1037 rc = mo_object_create(env, md_object_next(mo_c), spec, ma);
1039 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1040 rc = mdo_name_insert(env, md_object_next(mo_p), lchild_name,
1041 lu_object_fid(&mo_c->mo_lu), tmp_ma);
1043 /* TODO: remove object mo_c on remote MDS */
1044 CWARN("cmr_create failed, should revoke: [mo_p "DFID"]"
1045 " [name %s] [mo_c "DFID"] [err %d]\n",
1046 PFID(lu_object_fid(&mo_p->mo_lu)),
1047 lchild_name->ln_name,
1048 PFID(lu_object_fid(&mo_c->mo_lu)), rc);
1055 static int cmr_link(const struct lu_env *env, struct md_object *mo_p,
1056 struct md_object *mo_s, const struct lu_name *lname,
1062 /* Make sure that name isn't exist before doing remote call. */
1063 rc = mdo_lookup(env, md_object_next(mo_p), lname,
1064 &cmm_env_info(env)->cmi_fid, NULL);
1067 } else if (rc == -ENOENT) {
1068 /* Local permission check for name_insert before remote ops. */
1069 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1074 rc = mo_ref_add(env, md_object_next(mo_s), ma);
1076 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1077 rc = mdo_name_insert(env, md_object_next(mo_p), lname,
1078 lu_object_fid(&mo_s->mo_lu), ma);
1080 /* TODO: ref_del from mo_s on remote MDS */
1081 CWARN("cmr_link failed, should revoke: "
1082 "[mo_p "DFID"] [mo_s "DFID"] "
1083 "[name %s] [err %d]\n",
1084 PFID(lu_object_fid(&mo_p->mo_lu)),
1085 PFID(lu_object_fid(&mo_s->mo_lu)),
1086 lname->ln_name, rc);
1093 static int cmr_unlink(const struct lu_env *env, struct md_object *mo_p,
1094 struct md_object *mo_c, const struct lu_name *lname,
1097 struct cmm_thread_info *cmi;
1098 struct md_attr *tmp_ma;
1102 /* Local permission check for name_remove before remote ops. */
1103 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1104 MAY_UNLINK | MAY_VTX_PART);
1109 * XXX: @ma will be changed after mo_ref_del, but we will use
1110 * it for mdo_name_remove later, so save it before mo_ref_del.
1112 cmi = cmm_env_info(env);
1113 tmp_ma = &cmi->cmi_ma;
1115 rc = mo_ref_del(env, md_object_next(mo_c), ma);
1117 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1118 rc = mdo_name_remove(env, md_object_next(mo_p), lname, tmp_ma);
1120 /* TODO: ref_add to mo_c on remote MDS */
1121 CWARN("cmr_unlink failed, should revoke: [mo_p "DFID"]"
1122 " [mo_c "DFID"] [name %s] [err %d]\n",
1123 PFID(lu_object_fid(&mo_p->mo_lu)),
1124 PFID(lu_object_fid(&mo_c->mo_lu)),
1125 lname->ln_name, rc);
1132 static inline void cmr_rename_warn(const char *fname,
1133 struct md_object *mo_po,
1134 struct md_object *mo_pn,
1135 const struct lu_fid *lf,
1140 CWARN("cmr_rename failed for %s, should revoke: "
1141 "[mo_po "DFID"] [mo_pn "DFID"] [lf "DFID"] "
1142 "[sname %s] [tname %s] [err %d]\n", fname,
1143 PFID(lu_object_fid(&mo_po->mo_lu)),
1144 PFID(lu_object_fid(&mo_pn->mo_lu)),
1145 PFID(lf), s_name, t_name, err);
1148 static int cmr_rename(const struct lu_env *env,
1149 struct md_object *mo_po, struct md_object *mo_pn,
1150 const struct lu_fid *lf, const struct lu_name *ls_name,
1151 struct md_object *mo_t, const struct lu_name *lt_name,
1154 struct cmm_thread_info *cmi;
1155 struct md_attr *tmp_ma;
1159 LASSERT(mo_t == NULL);
1161 /* get real type of src */
1162 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, NULL);
1166 /* Local permission check for name_remove before remote ops. */
1167 rc = mo_permission(env, NULL, md_object_next(mo_po), ma,
1168 MAY_UNLINK | MAY_VTX_FULL);
1173 * XXX: @ma maybe changed after mdo_rename_tgt, but we will use it
1174 * for mdo_name_remove later, so save it before mdo_rename_tgt.
1176 cmi = cmm_env_info(env);
1177 tmp_ma = &cmi->cmi_ma;
1179 /* the mo_pn is remote directory, so we cannot even know if there is
1180 * mo_t or not. Therefore mo_t is NULL here but remote server should do
1181 * lookup and process this further */
1182 rc = mdo_rename_tgt(env, md_object_next(mo_pn),
1183 NULL/* mo_t */, lf, lt_name, ma);
1187 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1189 /* src object maybe on remote MDS, do remote ops first. */
1190 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf, tmp_ma);
1192 /* TODO: revoke mdo_rename_tgt */
1193 cmr_rename_warn("cmm_rename_ctime", mo_po, mo_pn, lf,
1194 ls_name->ln_name, lt_name->ln_name, rc);
1198 /* only old name is removed localy */
1199 rc = mdo_name_remove(env, md_object_next(mo_po), ls_name, tmp_ma);
1201 /* TODO: revoke all cmr_rename */
1202 cmr_rename_warn("mdo_name_remove", mo_po, mo_pn, lf,
1203 ls_name->ln_name, lt_name->ln_name, rc);
1208 /* part of cross-ref rename(). Used to insert new name in new parent
1209 * and unlink target */
1210 static int cmr_rename_tgt(const struct lu_env *env,
1211 struct md_object *mo_p, struct md_object *mo_t,
1212 const struct lu_fid *lf, const struct lu_name *lname,
1215 struct cmm_thread_info *cmi;
1216 struct md_attr *tmp_ma;
1220 /* target object is remote one */
1221 /* Local permission check for rename_tgt before remote ops. */
1222 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1223 MAY_UNLINK | MAY_VTX_PART);
1228 * XXX: @ma maybe changed after mo_ref_del, but we will use
1229 * it for mdo_rename_tgt later, so save it before mo_ref_del.
1231 cmi = cmm_env_info(env);
1232 tmp_ma = &cmi->cmi_ma;
1234 rc = mo_ref_del(env, md_object_next(mo_t), ma);
1235 /* continue locally with name handling only */
1237 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1238 rc = mdo_rename_tgt(env, md_object_next(mo_p),
1239 NULL, lf, lname, tmp_ma);
1241 /* TODO: ref_add to mo_t on remote MDS */
1242 CWARN("cmr_rename_tgt failed, should revoke: "
1243 "[mo_p "DFID"] [mo_t "DFID"] [lf "DFID"] "
1244 "[name %s] [err %d]\n",
1245 PFID(lu_object_fid(&mo_p->mo_lu)),
1246 PFID(lu_object_fid(&mo_t->mo_lu)),
1248 lname->ln_name, rc);
1254 static struct md_dir_operations cmr_dir_ops = {
1255 .mdo_is_subdir = cmm_is_subdir,
1256 .mdo_lookup = cmr_lookup,
1257 .mdo_lock_mode = cmr_lock_mode,
1258 .mdo_create = cmr_create,
1259 .mdo_link = cmr_link,
1260 .mdo_unlink = cmr_unlink,
1261 .mdo_rename = cmr_rename,
1262 .mdo_rename_tgt = cmr_rename_tgt,