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 const struct md_object_operations cml_mo_ops;
81 static const struct md_dir_operations cml_dir_ops;
82 static const struct lu_object_operations cml_obj_ops;
84 static const struct md_object_operations cmr_mo_ops;
85 static const struct md_dir_operations cmr_dir_ops;
86 static const 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,
163 const struct lu_object_conf *_)
165 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
166 struct lu_device *c_dev;
167 struct lu_object *c_obj;
172 #ifdef HAVE_SPLIT_SUPPORT
173 if (cd->cmm_tgt_count == 0)
174 lu2cml_obj(lo)->clo_split = CMM_SPLIT_DENIED;
176 lu2cml_obj(lo)->clo_split = CMM_SPLIT_UNKNOWN;
178 c_dev = cml_child_dev(cd);
182 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
183 lo->lo_header, c_dev);
185 lu_object_add(lo, c_obj);
195 static int cml_object_print(const struct lu_env *env, void *cookie,
196 lu_printer_t p, const struct lu_object *lo)
198 return (*p)(env, cookie, "[local]");
201 static const struct lu_object_operations cml_obj_ops = {
202 .loo_object_init = cml_object_init,
203 .loo_object_free = cml_object_free,
204 .loo_object_print = cml_object_print
207 /* CMM local md_object operations */
208 static int cml_object_create(const struct lu_env *env,
209 struct md_object *mo,
210 const struct md_op_spec *spec,
211 struct md_attr *attr)
215 rc = mo_object_create(env, md_object_next(mo), spec, attr);
219 static int cml_permission(const struct lu_env *env,
220 struct md_object *p, struct md_object *c,
221 struct md_attr *attr, int mask)
225 rc = mo_permission(env, md_object_next(p), md_object_next(c),
230 static int cml_attr_get(const struct lu_env *env, struct md_object *mo,
231 struct md_attr *attr)
235 rc = mo_attr_get(env, md_object_next(mo), attr);
239 static int cml_attr_set(const struct lu_env *env, struct md_object *mo,
240 const struct md_attr *attr)
244 rc = mo_attr_set(env, md_object_next(mo), attr);
248 static int cml_xattr_get(const struct lu_env *env, struct md_object *mo,
249 struct lu_buf *buf, const char *name)
253 rc = mo_xattr_get(env, md_object_next(mo), buf, name);
257 static int cml_readlink(const struct lu_env *env, struct md_object *mo,
262 rc = mo_readlink(env, md_object_next(mo), buf);
266 static int cml_xattr_list(const struct lu_env *env, struct md_object *mo,
271 rc = mo_xattr_list(env, md_object_next(mo), buf);
275 static int cml_xattr_set(const struct lu_env *env, struct md_object *mo,
276 const struct lu_buf *buf, const char *name,
281 rc = mo_xattr_set(env, md_object_next(mo), buf, name, fl);
285 static int cml_xattr_del(const struct lu_env *env, struct md_object *mo,
290 rc = mo_xattr_del(env, md_object_next(mo), name);
294 static int cml_ref_add(const struct lu_env *env, struct md_object *mo,
295 const struct md_attr *ma)
299 rc = mo_ref_add(env, md_object_next(mo), ma);
303 static int cml_ref_del(const struct lu_env *env, struct md_object *mo,
308 rc = mo_ref_del(env, md_object_next(mo), ma);
312 static int cml_open(const struct lu_env *env, struct md_object *mo,
317 rc = mo_open(env, md_object_next(mo), flags);
321 static int cml_close(const struct lu_env *env, struct md_object *mo,
326 rc = mo_close(env, md_object_next(mo), ma);
330 static int cml_readpage(const struct lu_env *env, struct md_object *mo,
331 const struct lu_rdpg *rdpg)
335 rc = mo_readpage(env, md_object_next(mo), rdpg);
339 static int cml_capa_get(const struct lu_env *env, struct md_object *mo,
340 struct lustre_capa *capa, int renewal)
344 rc = mo_capa_get(env, md_object_next(mo), capa, renewal);
348 static int cml_object_sync(const struct lu_env *env, struct md_object *mo)
352 rc = mo_object_sync(env, md_object_next(mo));
356 static const struct md_object_operations cml_mo_ops = {
357 .moo_permission = cml_permission,
358 .moo_attr_get = cml_attr_get,
359 .moo_attr_set = cml_attr_set,
360 .moo_xattr_get = cml_xattr_get,
361 .moo_xattr_list = cml_xattr_list,
362 .moo_xattr_set = cml_xattr_set,
363 .moo_xattr_del = cml_xattr_del,
364 .moo_object_create = cml_object_create,
365 .moo_ref_add = cml_ref_add,
366 .moo_ref_del = cml_ref_del,
367 .moo_open = cml_open,
368 .moo_close = cml_close,
369 .moo_readpage = cml_readpage,
370 .moo_readlink = cml_readlink,
371 .moo_capa_get = cml_capa_get,
372 .moo_object_sync = cml_object_sync,
375 /* md_dir operations */
376 static int cml_lookup(const struct lu_env *env, struct md_object *mo_p,
377 const struct lu_name *lname, struct lu_fid *lf,
378 struct md_op_spec *spec)
383 #ifdef HAVE_SPLIT_SUPPORT
384 if (spec != NULL && spec->sp_ck_split) {
385 rc = cmm_split_check(env, mo_p, lname->ln_name);
390 rc = mdo_lookup(env, md_object_next(mo_p), lname, lf, spec);
395 static mdl_mode_t cml_lock_mode(const struct lu_env *env,
396 struct md_object *mo, mdl_mode_t lm)
398 int rc = MDL_MINMODE;
401 #ifdef HAVE_SPLIT_SUPPORT
402 rc = cmm_split_access(env, mo, lm);
408 static int cml_create(const struct lu_env *env, struct md_object *mo_p,
409 const struct lu_name *lname, struct md_object *mo_c,
410 struct md_op_spec *spec, struct md_attr *ma)
415 #ifdef HAVE_SPLIT_SUPPORT
416 /* Lock mode always should be sane. */
417 LASSERT(spec->sp_cr_mode != MDL_MINMODE);
420 * Sigh... This is long story. MDT may have race with detecting if split
421 * is possible in cmm. We know this race and let it live, because
422 * getting it rid (with some sem or spinlock) will also mean that
423 * PDIROPS for create will not work because we kill parallel work, what
424 * is really bad for performance and makes no sense having PDIROPS. So,
425 * we better allow the race to live, but split dir only if some of
426 * concurrent threads takes EX lock, not matter which one. So that, say,
427 * two concurrent threads may have different lock modes on directory (CW
428 * and EX) and not first one which comes here and see that split is
429 * possible should split the dir, but only that one which has EX
430 * lock. And we do not care that in this case, split may happen a bit
431 * later (when dir size will not be necessarily 64K, but may be a bit
432 * larger). So that, we allow concurrent creates and protect split by EX
435 if (spec->sp_cr_mode == MDL_EX) {
437 * Try to split @mo_p. If split is ok, -ERESTART is returned and
438 * current thread will not peoceed with create. Instead it sends
439 * -ERESTART to client to let it know that correct MDT should be
442 rc = cmm_split_dir(env, mo_p);
445 * -ERESTART or some split error is returned, we can't
446 * proceed with create.
451 if (spec != NULL && spec->sp_ck_split) {
453 * Check for possible split directory and let caller know that
454 * it should tell client that directory is split and operation
455 * should repeat to correct MDT.
457 rc = cmm_split_check(env, mo_p, lname->ln_name);
463 rc = mdo_create(env, md_object_next(mo_p), lname, md_object_next(mo_c),
467 #ifdef HAVE_SPLIT_SUPPORT
473 static int cml_create_data(const struct lu_env *env, struct md_object *p,
475 const struct md_op_spec *spec,
480 rc = mdo_create_data(env, md_object_next(p), md_object_next(o),
485 static int cml_link(const struct lu_env *env, struct md_object *mo_p,
486 struct md_object *mo_s, const struct lu_name *lname,
491 rc = mdo_link(env, md_object_next(mo_p), md_object_next(mo_s),
496 static int cml_unlink(const struct lu_env *env, struct md_object *mo_p,
497 struct md_object *mo_c, const struct lu_name *lname,
502 rc = mdo_unlink(env, md_object_next(mo_p), md_object_next(mo_c),
507 static int cmm_mode_get(const struct lu_env *env, struct md_device *md,
508 const struct lu_fid *lf, struct md_attr *ma,
511 struct md_object *mo_s = md_object_find_slice(env, md, lf);
512 struct cmm_thread_info *cmi;
513 struct md_attr *tmp_ma;
518 RETURN(PTR_ERR(mo_s));
520 if (remote && (lu_object_exists(&mo_s->mo_lu) < 0))
523 cmi = cmm_env_info(env);
524 tmp_ma = &cmi->cmi_ma;
525 tmp_ma->ma_need = MA_INODE;
526 tmp_ma->ma_valid = 0;
527 /* get type from src, can be remote req */
528 rc = mo_attr_get(env, md_object_next(mo_s), tmp_ma);
530 ma->ma_attr.la_mode = tmp_ma->ma_attr.la_mode;
531 ma->ma_attr.la_uid = tmp_ma->ma_attr.la_uid;
532 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
533 ma->ma_attr.la_flags = tmp_ma->ma_attr.la_flags;
534 ma->ma_attr.la_valid |= LA_MODE | LA_UID | LA_GID | LA_FLAGS;
536 lu_object_put(env, &mo_s->mo_lu);
540 static int cmm_rename_ctime(const struct lu_env *env, struct md_device *md,
541 const struct lu_fid *lf, struct md_attr *ma)
543 struct md_object *mo_s = md_object_find_slice(env, md, lf);
548 RETURN(PTR_ERR(mo_s));
550 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
551 /* set ctime to obj, can be remote req */
552 rc = mo_attr_set(env, md_object_next(mo_s), ma);
553 lu_object_put(env, &mo_s->mo_lu);
557 static inline void cml_rename_warn(const char *fname,
558 struct md_object *mo_po,
559 struct md_object *mo_pn,
560 const struct lu_fid *lf,
562 struct md_object *mo_t,
567 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
568 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t "DFID"] "
569 "[tname %s] [err %d]\n", fname,
570 PFID(lu_object_fid(&mo_po->mo_lu)),
571 PFID(lu_object_fid(&mo_pn->mo_lu)),
573 PFID(lu_object_fid(&mo_t->mo_lu)),
576 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
577 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t NULL] "
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)),
585 static int cml_rename(const struct lu_env *env, struct md_object *mo_po,
586 struct md_object *mo_pn, const struct lu_fid *lf,
587 const struct lu_name *ls_name, struct md_object *mo_t,
588 const struct lu_name *lt_name, struct md_attr *ma)
590 struct cmm_thread_info *cmi;
591 struct md_attr *tmp_ma = NULL;
592 struct md_object *tmp_t = mo_t;
596 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, &remote);
600 if (mo_t && lu_object_exists(&mo_t->mo_lu) < 0) {
601 /* XXX: mo_t is remote object and there is RPC to unlink it.
602 * before that, do local sanity check for rename first. */
604 struct md_object *mo_s = md_object_find_slice(env,
605 md_obj2dev(mo_po), lf);
607 RETURN(PTR_ERR(mo_s));
609 LASSERT(lu_object_exists(&mo_s->mo_lu) > 0);
610 rc = mo_permission(env, md_object_next(mo_po),
611 md_object_next(mo_s),
613 lu_object_put(env, &mo_s->mo_lu);
617 rc = mo_permission(env, NULL, md_object_next(mo_po),
618 ma, MAY_UNLINK | MAY_VTX_FULL);
623 rc = mo_permission(env, NULL, md_object_next(mo_pn), ma,
624 MAY_UNLINK | MAY_VTX_PART);
629 * XXX: @ma will be changed after mo_ref_del, but we will use
630 * it for mdo_rename later, so save it before mo_ref_del.
632 cmi = cmm_env_info(env);
633 tmp_ma = &cmi->cmi_ma;
635 rc = mo_ref_del(env, md_object_next(mo_t), ma);
639 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
643 /* XXX: for src on remote MDS case, change its ctime before local
644 * rename. Firstly, do local sanity check for rename if necessary. */
647 rc = mo_permission(env, NULL, md_object_next(mo_po),
648 ma, MAY_UNLINK | MAY_VTX_FULL);
653 LASSERT(lu_object_exists(&mo_t->mo_lu) > 0);
654 rc = mo_permission(env, md_object_next(mo_pn),
655 md_object_next(mo_t),
663 mask = (S_ISDIR(ma->ma_attr.la_mode) ?
664 MAY_LINK : MAY_CREATE);
667 rc = mo_permission(env, NULL,
668 md_object_next(mo_pn),
674 ma->ma_attr_flags |= MDS_PERM_BYPASS;
676 LASSERT(tmp_ma->ma_attr_flags & MDS_PERM_BYPASS);
679 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf,
680 tmp_ma ? tmp_ma : ma);
682 /* TODO: revoke mo_t if necessary. */
683 cml_rename_warn("cmm_rename_ctime", mo_po,
684 mo_pn, lf, ls_name->ln_name,
685 tmp_t, lt_name->ln_name, rc);
690 /* local rename, mo_t can be NULL */
691 rc = mdo_rename(env, md_object_next(mo_po),
692 md_object_next(mo_pn), lf, ls_name,
693 md_object_next(mo_t), lt_name, tmp_ma ? tmp_ma : ma);
695 /* TODO: revoke all cml_rename */
696 cml_rename_warn("mdo_rename", mo_po, mo_pn, lf,
697 ls_name->ln_name, tmp_t, lt_name->ln_name, rc);
702 static int cml_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
703 struct md_object *mo_t, const struct lu_fid *lf,
704 const struct lu_name *lname, struct md_attr *ma)
709 rc = mdo_rename_tgt(env, md_object_next(mo_p),
710 md_object_next(mo_t), lf, lname, ma);
713 /* used only in case of rename_tgt() when target is not exist */
714 static int cml_name_insert(const struct lu_env *env, struct md_object *p,
715 const struct lu_name *lname, const struct lu_fid *lf,
716 const struct md_attr *ma)
721 rc = mdo_name_insert(env, md_object_next(p), lname, lf, ma);
726 static int cmm_is_subdir(const struct lu_env *env, struct md_object *mo,
727 const struct lu_fid *fid, struct lu_fid *sfid)
729 struct cmm_thread_info *cmi;
733 cmi = cmm_env_info(env);
734 rc = cmm_mode_get(env, md_obj2dev(mo), fid, &cmi->cmi_ma, NULL);
738 if (!S_ISDIR(cmi->cmi_ma.ma_attr.la_mode))
741 rc = mdo_is_subdir(env, md_object_next(mo), fid, sfid);
745 static const struct md_dir_operations cml_dir_ops = {
746 .mdo_is_subdir = cmm_is_subdir,
747 .mdo_lookup = cml_lookup,
748 .mdo_lock_mode = cml_lock_mode,
749 .mdo_create = cml_create,
750 .mdo_link = cml_link,
751 .mdo_unlink = cml_unlink,
752 .mdo_name_insert = cml_name_insert,
753 .mdo_rename = cml_rename,
754 .mdo_rename_tgt = cml_rename_tgt,
755 .mdo_create_data = cml_create_data
758 /* -------------------------------------------------------------------
759 * remote CMM object operations. cmr_...
761 static inline struct cmr_object *lu2cmr_obj(struct lu_object *o)
763 return container_of0(o, struct cmr_object, cmm_obj.cmo_obj.mo_lu);
765 static inline struct cmr_object *md2cmr_obj(struct md_object *mo)
767 return container_of0(mo, struct cmr_object, cmm_obj.cmo_obj);
769 static inline struct cmr_object *cmm2cmr_obj(struct cmm_object *co)
771 return container_of0(co, struct cmr_object, cmm_obj);
774 /* get proper child device from MDCs */
775 static struct lu_device *cmr_child_dev(struct cmm_device *d, __u32 num)
777 struct lu_device *next = NULL;
778 struct mdc_device *mdc;
780 spin_lock(&d->cmm_tgt_guard);
781 list_for_each_entry(mdc, &d->cmm_targets, mc_linkage) {
782 if (mdc->mc_num == num) {
783 next = mdc2lu_dev(mdc);
787 spin_unlock(&d->cmm_tgt_guard);
791 /* lu_object operations */
792 static void cmr_object_free(const struct lu_env *env,
793 struct lu_object *lo)
795 struct cmr_object *cro = lu2cmr_obj(lo);
800 static int cmr_object_init(const struct lu_env *env, struct lu_object *lo,
801 const struct lu_object_conf *_)
803 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
804 struct lu_device *c_dev;
805 struct lu_object *c_obj;
810 c_dev = cmr_child_dev(cd, lu2cmr_obj(lo)->cmo_num);
814 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
815 lo->lo_header, c_dev);
817 lu_object_add(lo, c_obj);
827 static int cmr_object_print(const struct lu_env *env, void *cookie,
828 lu_printer_t p, const struct lu_object *lo)
830 return (*p)(env, cookie, "[remote]");
833 static const struct lu_object_operations cmr_obj_ops = {
834 .loo_object_init = cmr_object_init,
835 .loo_object_free = cmr_object_free,
836 .loo_object_print = cmr_object_print
839 /* CMM remote md_object operations. All are invalid */
840 static int cmr_object_create(const struct lu_env *env,
841 struct md_object *mo,
842 const struct md_op_spec *spec,
848 static int cmr_permission(const struct lu_env *env,
849 struct md_object *p, struct md_object *c,
850 struct md_attr *attr, int mask)
855 static int cmr_attr_get(const struct lu_env *env, struct md_object *mo,
856 struct md_attr *attr)
861 static int cmr_attr_set(const struct lu_env *env, struct md_object *mo,
862 const struct md_attr *attr)
867 static int cmr_xattr_get(const struct lu_env *env, struct md_object *mo,
868 struct lu_buf *buf, const char *name)
873 static int cmr_readlink(const struct lu_env *env, struct md_object *mo,
879 static int cmr_xattr_list(const struct lu_env *env, struct md_object *mo,
885 static int cmr_xattr_set(const struct lu_env *env, struct md_object *mo,
886 const struct lu_buf *buf, const char *name,
892 static int cmr_xattr_del(const struct lu_env *env, struct md_object *mo,
898 static int cmr_ref_add(const struct lu_env *env, struct md_object *mo,
899 const struct md_attr *ma)
904 static int cmr_ref_del(const struct lu_env *env, struct md_object *mo,
910 static int cmr_open(const struct lu_env *env, struct md_object *mo,
916 static int cmr_close(const struct lu_env *env, struct md_object *mo,
922 static int cmr_readpage(const struct lu_env *env, struct md_object *mo,
923 const struct lu_rdpg *rdpg)
928 static int cmr_capa_get(const struct lu_env *env, struct md_object *mo,
929 struct lustre_capa *capa, int renewal)
934 static int cmr_object_sync(const struct lu_env *env, struct md_object *mo)
939 static const struct md_object_operations cmr_mo_ops = {
940 .moo_permission = cmr_permission,
941 .moo_attr_get = cmr_attr_get,
942 .moo_attr_set = cmr_attr_set,
943 .moo_xattr_get = cmr_xattr_get,
944 .moo_xattr_set = cmr_xattr_set,
945 .moo_xattr_list = cmr_xattr_list,
946 .moo_xattr_del = cmr_xattr_del,
947 .moo_object_create = cmr_object_create,
948 .moo_ref_add = cmr_ref_add,
949 .moo_ref_del = cmr_ref_del,
950 .moo_open = cmr_open,
951 .moo_close = cmr_close,
952 .moo_readpage = cmr_readpage,
953 .moo_readlink = cmr_readlink,
954 .moo_capa_get = cmr_capa_get,
955 .moo_object_sync = cmr_object_sync,
958 /* remote part of md_dir operations */
959 static int cmr_lookup(const struct lu_env *env, struct md_object *mo_p,
960 const struct lu_name *lname, struct lu_fid *lf,
961 struct md_op_spec *spec)
964 * This can happens while rename() If new parent is remote dir, lookup
971 static mdl_mode_t cmr_lock_mode(const struct lu_env *env,
972 struct md_object *mo, mdl_mode_t lm)
978 * All methods below are cross-ref by nature. They consist of remote call and
979 * local operation. Due to future rollback functionality there are several
980 * limitations for such methods:
981 * 1) remote call should be done at first to do epoch negotiation between all
982 * MDS involved and to avoid the RPC inside transaction.
983 * 2) only one RPC can be sent - also due to epoch negotiation.
984 * For more details see rollback HLD/DLD.
986 static int cmr_create(const struct lu_env *env, struct md_object *mo_p,
987 const struct lu_name *lchild_name, struct md_object *mo_c,
988 struct md_op_spec *spec,
991 struct cmm_thread_info *cmi;
992 struct md_attr *tmp_ma;
996 /* Make sure that name isn't exist before doing remote call. */
997 rc = mdo_lookup(env, md_object_next(mo_p), lchild_name,
998 &cmm_env_info(env)->cmi_fid, NULL);
1001 else if (rc != -ENOENT)
1004 /* check the SGID attr */
1005 cmi = cmm_env_info(env);
1007 tmp_ma = &cmi->cmi_ma;
1008 tmp_ma->ma_valid = 0;
1009 tmp_ma->ma_need = MA_INODE;
1011 #ifdef CONFIG_FS_POSIX_ACL
1012 if (!S_ISLNK(ma->ma_attr.la_mode)) {
1013 tmp_ma->ma_acl = cmi->cmi_xattr_buf;
1014 tmp_ma->ma_acl_size = sizeof(cmi->cmi_xattr_buf);
1015 tmp_ma->ma_need |= MA_ACL_DEF;
1018 rc = mo_attr_get(env, md_object_next(mo_p), tmp_ma);
1022 if (tmp_ma->ma_attr.la_mode & S_ISGID) {
1023 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
1024 if (S_ISDIR(ma->ma_attr.la_mode)) {
1025 ma->ma_attr.la_mode |= S_ISGID;
1026 ma->ma_attr.la_valid |= LA_MODE;
1030 #ifdef CONFIG_FS_POSIX_ACL
1031 if (tmp_ma->ma_valid & MA_ACL_DEF) {
1032 spec->u.sp_ea.fid = spec->u.sp_pfid;
1033 spec->u.sp_ea.eadata = tmp_ma->ma_acl;
1034 spec->u.sp_ea.eadatalen = tmp_ma->ma_acl_size;
1035 spec->sp_cr_flags |= MDS_CREATE_RMT_ACL;
1039 /* Local permission check for name_insert before remote ops. */
1040 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1041 (S_ISDIR(ma->ma_attr.la_mode) ?
1042 MAY_LINK : MAY_CREATE));
1046 /* Remote object creation and local name insert. */
1048 * XXX: @ma will be changed after mo_object_create, but we will use
1049 * it for mdo_name_insert later, so save it before mo_object_create.
1052 rc = mo_object_create(env, md_object_next(mo_c), spec, ma);
1054 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1055 rc = mdo_name_insert(env, md_object_next(mo_p), lchild_name,
1056 lu_object_fid(&mo_c->mo_lu), tmp_ma);
1058 /* TODO: remove object mo_c on remote MDS */
1059 CWARN("cmr_create failed, should revoke: [mo_p "DFID"]"
1060 " [name %s] [mo_c "DFID"] [err %d]\n",
1061 PFID(lu_object_fid(&mo_p->mo_lu)),
1062 lchild_name->ln_name,
1063 PFID(lu_object_fid(&mo_c->mo_lu)), rc);
1070 static int cmr_link(const struct lu_env *env, struct md_object *mo_p,
1071 struct md_object *mo_s, const struct lu_name *lname,
1077 /* Make sure that name isn't exist before doing remote call. */
1078 rc = mdo_lookup(env, md_object_next(mo_p), lname,
1079 &cmm_env_info(env)->cmi_fid, NULL);
1082 } else if (rc == -ENOENT) {
1083 /* Local permission check for name_insert before remote ops. */
1084 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1089 rc = mo_ref_add(env, md_object_next(mo_s), ma);
1091 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1092 rc = mdo_name_insert(env, md_object_next(mo_p), lname,
1093 lu_object_fid(&mo_s->mo_lu), ma);
1095 /* TODO: ref_del from mo_s on remote MDS */
1096 CWARN("cmr_link failed, should revoke: "
1097 "[mo_p "DFID"] [mo_s "DFID"] "
1098 "[name %s] [err %d]\n",
1099 PFID(lu_object_fid(&mo_p->mo_lu)),
1100 PFID(lu_object_fid(&mo_s->mo_lu)),
1101 lname->ln_name, rc);
1108 static int cmr_unlink(const struct lu_env *env, struct md_object *mo_p,
1109 struct md_object *mo_c, const struct lu_name *lname,
1112 struct cmm_thread_info *cmi;
1113 struct md_attr *tmp_ma;
1117 /* Local permission check for name_remove before remote ops. */
1118 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1119 MAY_UNLINK | MAY_VTX_PART);
1124 * XXX: @ma will be changed after mo_ref_del, but we will use
1125 * it for mdo_name_remove later, so save it before mo_ref_del.
1127 cmi = cmm_env_info(env);
1128 tmp_ma = &cmi->cmi_ma;
1130 rc = mo_ref_del(env, md_object_next(mo_c), ma);
1132 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1133 rc = mdo_name_remove(env, md_object_next(mo_p), lname, tmp_ma);
1135 /* TODO: ref_add to mo_c on remote MDS */
1136 CWARN("cmr_unlink failed, should revoke: [mo_p "DFID"]"
1137 " [mo_c "DFID"] [name %s] [err %d]\n",
1138 PFID(lu_object_fid(&mo_p->mo_lu)),
1139 PFID(lu_object_fid(&mo_c->mo_lu)),
1140 lname->ln_name, rc);
1147 static inline void cmr_rename_warn(const char *fname,
1148 struct md_object *mo_po,
1149 struct md_object *mo_pn,
1150 const struct lu_fid *lf,
1155 CWARN("cmr_rename failed for %s, should revoke: "
1156 "[mo_po "DFID"] [mo_pn "DFID"] [lf "DFID"] "
1157 "[sname %s] [tname %s] [err %d]\n", fname,
1158 PFID(lu_object_fid(&mo_po->mo_lu)),
1159 PFID(lu_object_fid(&mo_pn->mo_lu)),
1160 PFID(lf), s_name, t_name, err);
1163 static int cmr_rename(const struct lu_env *env,
1164 struct md_object *mo_po, struct md_object *mo_pn,
1165 const struct lu_fid *lf, const struct lu_name *ls_name,
1166 struct md_object *mo_t, const struct lu_name *lt_name,
1169 struct cmm_thread_info *cmi;
1170 struct md_attr *tmp_ma;
1174 LASSERT(mo_t == NULL);
1176 /* get real type of src */
1177 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, NULL);
1181 /* Local permission check for name_remove before remote ops. */
1182 rc = mo_permission(env, NULL, md_object_next(mo_po), ma,
1183 MAY_UNLINK | MAY_VTX_FULL);
1188 * XXX: @ma maybe changed after mdo_rename_tgt, but we will use it
1189 * for mdo_name_remove later, so save it before mdo_rename_tgt.
1191 cmi = cmm_env_info(env);
1192 tmp_ma = &cmi->cmi_ma;
1194 /* the mo_pn is remote directory, so we cannot even know if there is
1195 * mo_t or not. Therefore mo_t is NULL here but remote server should do
1196 * lookup and process this further */
1197 rc = mdo_rename_tgt(env, md_object_next(mo_pn),
1198 NULL/* mo_t */, lf, lt_name, ma);
1202 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1204 /* src object maybe on remote MDS, do remote ops first. */
1205 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf, tmp_ma);
1207 /* TODO: revoke mdo_rename_tgt */
1208 cmr_rename_warn("cmm_rename_ctime", mo_po, mo_pn, lf,
1209 ls_name->ln_name, lt_name->ln_name, rc);
1213 /* only old name is removed localy */
1214 rc = mdo_name_remove(env, md_object_next(mo_po), ls_name, tmp_ma);
1216 /* TODO: revoke all cmr_rename */
1217 cmr_rename_warn("mdo_name_remove", mo_po, mo_pn, lf,
1218 ls_name->ln_name, lt_name->ln_name, rc);
1223 /* part of cross-ref rename(). Used to insert new name in new parent
1224 * and unlink target */
1225 static int cmr_rename_tgt(const struct lu_env *env,
1226 struct md_object *mo_p, struct md_object *mo_t,
1227 const struct lu_fid *lf, const struct lu_name *lname,
1230 struct cmm_thread_info *cmi;
1231 struct md_attr *tmp_ma;
1235 /* target object is remote one */
1236 /* Local permission check for rename_tgt before remote ops. */
1237 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1238 MAY_UNLINK | MAY_VTX_PART);
1243 * XXX: @ma maybe changed after mo_ref_del, but we will use
1244 * it for mdo_rename_tgt later, so save it before mo_ref_del.
1246 cmi = cmm_env_info(env);
1247 tmp_ma = &cmi->cmi_ma;
1249 rc = mo_ref_del(env, md_object_next(mo_t), ma);
1250 /* continue locally with name handling only */
1252 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1253 rc = mdo_rename_tgt(env, md_object_next(mo_p),
1254 NULL, lf, lname, tmp_ma);
1256 /* TODO: ref_add to mo_t on remote MDS */
1257 CWARN("cmr_rename_tgt failed, should revoke: "
1258 "[mo_p "DFID"] [mo_t "DFID"] [lf "DFID"] "
1259 "[name %s] [err %d]\n",
1260 PFID(lu_object_fid(&mo_p->mo_lu)),
1261 PFID(lu_object_fid(&mo_t->mo_lu)),
1263 lname->ln_name, rc);
1269 static const struct md_dir_operations cmr_dir_ops = {
1270 .mdo_is_subdir = cmm_is_subdir,
1271 .mdo_lookup = cmr_lookup,
1272 .mdo_lock_mode = cmr_lock_mode,
1273 .mdo_create = cmr_create,
1274 .mdo_link = cmr_link,
1275 .mdo_unlink = cmr_unlink,
1276 .mdo_rename = cmr_rename,
1277 .mdo_rename_tgt = cmr_rename_tgt,