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 (c) 2007, 2010, Oracle and/or its affiliates. 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"
54 * Lookup MDS number \a mds by FID \a fid.
56 * \param fid FID of object to find MDS
57 * \param mds mds number to return.
59 int cmm_fld_lookup(struct cmm_device *cm, const struct lu_fid *fid,
60 mdsno_t *mds, const struct lu_env *env)
65 LASSERT(fid_is_sane(fid));
67 rc = fld_client_lookup(cm->cmm_fld, fid_seq(fid), mds,
68 LU_SEQ_RANGE_MDT, env);
70 CERROR("Can't find mds by seq "LPX64", rc %d\n",
75 if (*mds > cm->cmm_tgt_count) {
76 CERROR("Got invalid mdsno: %x (max: %x)\n",
77 *mds, cm->cmm_tgt_count);
80 CDEBUG(D_INFO, "CMM: got MDS %x for sequence: "
81 LPX64"\n", *mds, fid_seq(fid));
91 static const struct md_object_operations cml_mo_ops;
92 static const struct md_dir_operations cml_dir_ops;
93 static const struct lu_object_operations cml_obj_ops;
95 static const struct md_object_operations cmr_mo_ops;
96 static const struct md_dir_operations cmr_dir_ops;
97 static const struct lu_object_operations cmr_obj_ops;
101 * Allocate CMM object.
103 struct lu_object *cmm_object_alloc(const struct lu_env *env,
104 const struct lu_object_header *loh,
105 struct lu_device *ld)
107 const struct lu_fid *fid = &loh->loh_fid;
108 struct lu_object *lo = NULL;
109 struct cmm_device *cd;
116 if (cd->cmm_flags & CMM_INITIALIZED) {
117 /* get object location */
118 rc = cmm_fld_lookup(lu2cmm_dev(ld), fid, &mds, env);
123 * Device is not yet initialized, cmm_object is being created
124 * as part of early bootstrap procedure (it is /ROOT, or /fld,
125 * etc.). Such object *has* to be local.
127 mds = cd->cmm_local_num;
129 /* select the proper set of operations based on object location */
130 if (mds == cd->cmm_local_num) {
131 struct cml_object *clo;
135 lo = &clo->cmm_obj.cmo_obj.mo_lu;
136 lu_object_init(lo, NULL, ld);
137 clo->cmm_obj.cmo_obj.mo_ops = &cml_mo_ops;
138 clo->cmm_obj.cmo_obj.mo_dir_ops = &cml_dir_ops;
139 lo->lo_ops = &cml_obj_ops;
142 struct cmr_object *cro;
146 lo = &cro->cmm_obj.cmo_obj.mo_lu;
147 lu_object_init(lo, NULL, ld);
148 cro->cmm_obj.cmo_obj.mo_ops = &cmr_mo_ops;
149 cro->cmm_obj.cmo_obj.mo_dir_ops = &cmr_dir_ops;
150 lo->lo_ops = &cmr_obj_ops;
158 * Get local child device.
160 static struct lu_device *cml_child_dev(struct cmm_device *d)
162 return &d->cmm_child->md_lu_dev;
168 static void cml_object_free(const struct lu_env *env,
169 struct lu_object *lo)
171 struct cml_object *clo = lu2cml_obj(lo);
177 * Initialize cml_object.
179 static int cml_object_init(const struct lu_env *env, struct lu_object *lo,
180 const struct lu_object_conf *unused)
182 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
183 struct lu_device *c_dev;
184 struct lu_object *c_obj;
189 #ifdef HAVE_SPLIT_SUPPORT
190 if (cd->cmm_tgt_count == 0)
191 lu2cml_obj(lo)->clo_split = CMM_SPLIT_DENIED;
193 lu2cml_obj(lo)->clo_split = CMM_SPLIT_UNKNOWN;
195 c_dev = cml_child_dev(cd);
199 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
200 lo->lo_header, c_dev);
202 lu_object_add(lo, c_obj);
212 static int cml_object_print(const struct lu_env *env, void *cookie,
213 lu_printer_t p, const struct lu_object *lo)
215 return (*p)(env, cookie, "[local]");
218 static const struct lu_object_operations cml_obj_ops = {
219 .loo_object_init = cml_object_init,
220 .loo_object_free = cml_object_free,
221 .loo_object_print = cml_object_print
225 * \name CMM local md_object operations.
226 * All of them call just corresponding operations on next layer.
229 static int cml_object_create(const struct lu_env *env,
230 struct md_object *mo,
231 const struct md_op_spec *spec,
232 struct md_attr *attr)
236 rc = mo_object_create(env, md_object_next(mo), spec, attr);
240 static int cml_permission(const struct lu_env *env,
241 struct md_object *p, struct md_object *c,
242 struct md_attr *attr, int mask)
246 rc = mo_permission(env, md_object_next(p), md_object_next(c),
251 static int cml_attr_get(const struct lu_env *env, struct md_object *mo,
252 struct md_attr *attr)
256 rc = mo_attr_get(env, md_object_next(mo), attr);
260 static int cml_attr_set(const struct lu_env *env, struct md_object *mo,
261 const struct md_attr *attr)
265 rc = mo_attr_set(env, md_object_next(mo), attr);
269 static int cml_xattr_get(const struct lu_env *env, struct md_object *mo,
270 struct lu_buf *buf, const char *name)
274 rc = mo_xattr_get(env, md_object_next(mo), buf, name);
278 static int cml_readlink(const struct lu_env *env, struct md_object *mo,
283 rc = mo_readlink(env, md_object_next(mo), buf);
287 static int cml_xattr_list(const struct lu_env *env, struct md_object *mo,
292 rc = mo_xattr_list(env, md_object_next(mo), buf);
296 static int cml_xattr_set(const struct lu_env *env, struct md_object *mo,
297 const struct lu_buf *buf, const char *name,
302 rc = mo_xattr_set(env, md_object_next(mo), buf, name, fl);
306 static int cml_xattr_del(const struct lu_env *env, struct md_object *mo,
311 rc = mo_xattr_del(env, md_object_next(mo), name);
315 static int cml_ref_add(const struct lu_env *env, struct md_object *mo,
316 const struct md_attr *ma)
320 rc = mo_ref_add(env, md_object_next(mo), ma);
324 static int cml_ref_del(const struct lu_env *env, struct md_object *mo,
329 rc = mo_ref_del(env, md_object_next(mo), ma);
333 static int cml_open(const struct lu_env *env, struct md_object *mo,
338 rc = mo_open(env, md_object_next(mo), flags);
342 static int cml_close(const struct lu_env *env, struct md_object *mo,
347 rc = mo_close(env, md_object_next(mo), ma);
351 static int cml_readpage(const struct lu_env *env, struct md_object *mo,
352 const struct lu_rdpg *rdpg)
356 rc = mo_readpage(env, md_object_next(mo), rdpg);
360 static int cml_capa_get(const struct lu_env *env, struct md_object *mo,
361 struct lustre_capa *capa, int renewal)
365 rc = mo_capa_get(env, md_object_next(mo), capa, renewal);
369 static int cml_path(const struct lu_env *env, struct md_object *mo,
370 char *path, int pathlen, __u64 *recno, int *linkno)
374 rc = mo_path(env, md_object_next(mo), path, pathlen, recno, linkno);
378 static int cml_object_sync(const struct lu_env *env, struct md_object *mo)
382 rc = mo_object_sync(env, md_object_next(mo));
386 static dt_obj_version_t cml_version_get(const struct lu_env *env,
387 struct md_object *mo)
389 return mo_version_get(env, md_object_next(mo));
392 static void cml_version_set(const struct lu_env *env, struct md_object *mo,
393 dt_obj_version_t version)
395 return mo_version_set(env, md_object_next(mo), version);
398 static const struct md_object_operations cml_mo_ops = {
399 .moo_permission = cml_permission,
400 .moo_attr_get = cml_attr_get,
401 .moo_attr_set = cml_attr_set,
402 .moo_xattr_get = cml_xattr_get,
403 .moo_xattr_list = cml_xattr_list,
404 .moo_xattr_set = cml_xattr_set,
405 .moo_xattr_del = cml_xattr_del,
406 .moo_object_create = cml_object_create,
407 .moo_ref_add = cml_ref_add,
408 .moo_ref_del = cml_ref_del,
409 .moo_open = cml_open,
410 .moo_close = cml_close,
411 .moo_readpage = cml_readpage,
412 .moo_readlink = cml_readlink,
413 .moo_capa_get = cml_capa_get,
414 .moo_object_sync = cml_object_sync,
415 .moo_version_get = cml_version_get,
416 .moo_version_set = cml_version_set,
417 .moo_path = cml_path,
422 * \name CMM local md_dir_operations.
426 * cml lookup object fid by name.
427 * This returns only FID by name.
429 static int cml_lookup(const struct lu_env *env, struct md_object *mo_p,
430 const struct lu_name *lname, struct lu_fid *lf,
431 struct md_op_spec *spec)
436 #ifdef HAVE_SPLIT_SUPPORT
437 if (spec != NULL && spec->sp_ck_split) {
438 rc = cmm_split_check(env, mo_p, lname->ln_name);
443 rc = mdo_lookup(env, md_object_next(mo_p), lname, lf, spec);
449 * Helper to return lock mode. Used in split cases only.
451 static mdl_mode_t cml_lock_mode(const struct lu_env *env,
452 struct md_object *mo, mdl_mode_t lm)
454 int rc = MDL_MINMODE;
457 #ifdef HAVE_SPLIT_SUPPORT
458 rc = cmm_split_access(env, mo, lm);
465 * Create operation for cml.
466 * Objects are local, but split can happen.
467 * If split is not needed this will call next layer mdo_create().
469 * \param mo_p Parent directory. Local object.
470 * \param lname name of file to create.
471 * \param mo_c Child object. It has no real inode yet.
472 * \param spec creation specification.
473 * \param ma child object attributes.
475 static int cml_create(const struct lu_env *env, struct md_object *mo_p,
476 const struct lu_name *lname, struct md_object *mo_c,
477 struct md_op_spec *spec, struct md_attr *ma)
482 #ifdef HAVE_SPLIT_SUPPORT
483 /* Lock mode always should be sane. */
484 LASSERT(spec->sp_cr_mode != MDL_MINMODE);
487 * Sigh... This is long story. MDT may have race with detecting if split
488 * is possible in cmm. We know this race and let it live, because
489 * getting it rid (with some sem or spinlock) will also mean that
490 * PDIROPS for create will not work because we kill parallel work, what
491 * is really bad for performance and makes no sense having PDIROPS. So,
492 * we better allow the race to live, but split dir only if some of
493 * concurrent threads takes EX lock, not matter which one. So that, say,
494 * two concurrent threads may have different lock modes on directory (CW
495 * and EX) and not first one which comes here and see that split is
496 * possible should split the dir, but only that one which has EX
497 * lock. And we do not care that in this case, split may happen a bit
498 * later (when dir size will not be necessarily 64K, but may be a bit
499 * larger). So that, we allow concurrent creates and protect split by EX
502 if (spec->sp_cr_mode == MDL_EX) {
505 * - Try to split \a mo_p upon each create operation.
506 * If split is ok, -ERESTART is returned and current thread
507 * will not peoceed with create. Instead it sends -ERESTART
508 * to client to let it know that correct MDT must be chosen.
509 * \see cmm_split_dir()
511 rc = cmm_split_dir(env, mo_p);
514 * -ERESTART or some split error is returned, we can't
515 * proceed with create.
520 if (spec != NULL && spec->sp_ck_split) {
522 * - Directory is split already. Let the caller know that
523 * it should tell client that directory is split and operation
524 * should repeat to correct MDT.
525 * \see cmm_split_check()
527 rc = cmm_split_check(env, mo_p, lname->ln_name);
533 rc = mdo_create(env, md_object_next(mo_p), lname, md_object_next(mo_c),
537 #ifdef HAVE_SPLIT_SUPPORT
543 /** Call mdo_create_data() on next layer. All objects are local. */
544 static int cml_create_data(const struct lu_env *env, struct md_object *p,
546 const struct md_op_spec *spec,
551 rc = mdo_create_data(env, md_object_next(p), md_object_next(o),
556 /** Call mdo_link() on next layer. All objects are local. */
557 static int cml_link(const struct lu_env *env, struct md_object *mo_p,
558 struct md_object *mo_s, const struct lu_name *lname,
563 rc = mdo_link(env, md_object_next(mo_p), md_object_next(mo_s),
568 /** Call mdo_unlink() on next layer. All objects are local. */
569 static int cml_unlink(const struct lu_env *env, struct md_object *mo_p,
570 struct md_object *mo_c, const struct lu_name *lname,
575 rc = mdo_unlink(env, md_object_next(mo_p), md_object_next(mo_c),
582 * Get mode of object.
583 * Used in both cml and cmr hence can produce RPC to another server.
585 static int cmm_mode_get(const struct lu_env *env, struct md_device *md,
586 const struct lu_fid *lf, struct md_attr *ma,
589 struct md_object *mo_s = md_object_find_slice(env, md, lf);
590 struct cmm_thread_info *cmi;
591 struct md_attr *tmp_ma;
596 RETURN(PTR_ERR(mo_s));
598 if (remote && (lu_object_exists(&mo_s->mo_lu) < 0))
601 cmi = cmm_env_info(env);
602 tmp_ma = &cmi->cmi_ma;
603 tmp_ma->ma_need = MA_INODE;
604 tmp_ma->ma_valid = 0;
605 /* get type from src, can be remote req */
606 rc = mo_attr_get(env, md_object_next(mo_s), tmp_ma);
608 ma->ma_attr.la_mode = tmp_ma->ma_attr.la_mode;
609 ma->ma_attr.la_uid = tmp_ma->ma_attr.la_uid;
610 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
611 ma->ma_attr.la_flags = tmp_ma->ma_attr.la_flags;
612 ma->ma_attr.la_valid |= LA_MODE | LA_UID | LA_GID | LA_FLAGS;
614 lu_object_put(env, &mo_s->mo_lu);
620 * Set ctime for object.
621 * Used in both cml and cmr hence can produce RPC to another server.
623 static int cmm_rename_ctime(const struct lu_env *env, struct md_device *md,
624 const struct lu_fid *lf, struct md_attr *ma)
626 struct md_object *mo_s = md_object_find_slice(env, md, lf);
631 RETURN(PTR_ERR(mo_s));
633 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
634 /* set ctime to obj, can be remote req */
635 rc = mo_attr_set(env, md_object_next(mo_s), ma);
636 lu_object_put(env, &mo_s->mo_lu);
640 /** Helper to output debug information about rename operation. */
641 static inline void cml_rename_warn(const char *fname,
642 struct md_object *mo_po,
643 struct md_object *mo_pn,
644 const struct lu_fid *lf,
646 struct md_object *mo_t,
651 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
652 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t "DFID"] "
653 "[tname %s] [err %d]\n", fname,
654 PFID(lu_object_fid(&mo_po->mo_lu)),
655 PFID(lu_object_fid(&mo_pn->mo_lu)),
657 PFID(lu_object_fid(&mo_t->mo_lu)),
660 CWARN("cml_rename failed for %s, should revoke: [mo_po "DFID"] "
661 "[mo_pn "DFID"] [lf "DFID"] [sname %s] [mo_t NULL] "
662 "[tname %s] [err %d]\n", fname,
663 PFID(lu_object_fid(&mo_po->mo_lu)),
664 PFID(lu_object_fid(&mo_pn->mo_lu)),
670 * Rename operation for cml.
672 * This is the most complex cross-reference operation. It may consist of up to 4
673 * MDS server and require several RPCs to be sent.
675 * \param mo_po Old parent object.
676 * \param mo_pn New parent object.
677 * \param lf FID of object to rename.
678 * \param ls_name Source file name.
679 * \param mo_t target object. Should be NULL here.
680 * \param lt_name Name of target file.
681 * \param ma object attributes.
683 static int cml_rename(const struct lu_env *env, struct md_object *mo_po,
684 struct md_object *mo_pn, const struct lu_fid *lf,
685 const struct lu_name *ls_name, struct md_object *mo_t,
686 const struct lu_name *lt_name, struct md_attr *ma)
688 struct cmm_thread_info *cmi;
689 struct md_attr *tmp_ma = NULL;
690 struct md_object *tmp_t = mo_t;
694 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, &remote);
698 if (mo_t && lu_object_exists(&mo_t->mo_lu) < 0) {
700 * \note \a mo_t is remote object and there is RPC to unlink it.
701 * Before that, do local sanity check for rename first.
704 struct md_object *mo_s = md_object_find_slice(env,
705 md_obj2dev(mo_po), lf);
707 RETURN(PTR_ERR(mo_s));
709 LASSERT(lu_object_exists(&mo_s->mo_lu) > 0);
710 rc = mo_permission(env, md_object_next(mo_po),
711 md_object_next(mo_s),
713 lu_object_put(env, &mo_s->mo_lu);
717 rc = mo_permission(env, NULL, md_object_next(mo_po),
718 ma, MAY_UNLINK | MAY_VTX_FULL);
723 rc = mo_permission(env, NULL, md_object_next(mo_pn), ma,
724 MAY_UNLINK | MAY_VTX_PART);
729 * /note \a ma will be changed after mo_ref_del(), but we will use
730 * it for mdo_rename() later, so save it before mo_ref_del().
732 cmi = cmm_env_info(env);
733 tmp_ma = &cmi->cmi_ma;
735 rc = mo_ref_del(env, md_object_next(mo_t), ma);
739 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
744 * \note for src on remote MDS case, change its ctime before local
745 * rename. Firstly, do local sanity check for rename if necessary.
749 rc = mo_permission(env, NULL, md_object_next(mo_po),
750 ma, MAY_UNLINK | MAY_VTX_FULL);
755 LASSERT(lu_object_exists(&mo_t->mo_lu) > 0);
756 rc = mo_permission(env, md_object_next(mo_pn),
757 md_object_next(mo_t),
765 mask = (S_ISDIR(ma->ma_attr.la_mode) ?
766 MAY_LINK : MAY_CREATE);
769 rc = mo_permission(env, NULL,
770 md_object_next(mo_pn),
776 ma->ma_attr_flags |= MDS_PERM_BYPASS;
778 LASSERT(tmp_ma->ma_attr_flags & MDS_PERM_BYPASS);
781 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf,
782 tmp_ma ? tmp_ma : ma);
784 /* TODO: revoke mo_t if necessary. */
785 cml_rename_warn("cmm_rename_ctime", mo_po,
786 mo_pn, lf, ls_name->ln_name,
787 tmp_t, lt_name->ln_name, rc);
792 /* local rename, mo_t can be NULL */
793 rc = mdo_rename(env, md_object_next(mo_po),
794 md_object_next(mo_pn), lf, ls_name,
795 md_object_next(mo_t), lt_name, tmp_ma ? tmp_ma : ma);
797 /* TODO: revoke all cml_rename */
798 cml_rename_warn("mdo_rename", mo_po, mo_pn, lf,
799 ls_name->ln_name, tmp_t, lt_name->ln_name, rc);
805 * Rename target partial operation.
806 * Used for cross-ref rename.
808 static int cml_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
809 struct md_object *mo_t, const struct lu_fid *lf,
810 const struct lu_name *lname, struct md_attr *ma)
815 rc = mdo_rename_tgt(env, md_object_next(mo_p),
816 md_object_next(mo_t), lf, lname, ma);
821 * Name insert only operation.
822 * used only in case of rename_tgt() when target doesn't exist.
824 static int cml_name_insert(const struct lu_env *env, struct md_object *p,
825 const struct lu_name *lname, const struct lu_fid *lf,
826 const struct md_attr *ma)
831 rc = mdo_name_insert(env, md_object_next(p), lname, lf, ma);
838 * Check two fids are not subdirectories.
840 static int cmm_is_subdir(const struct lu_env *env, struct md_object *mo,
841 const struct lu_fid *fid, struct lu_fid *sfid)
843 struct cmm_thread_info *cmi;
847 cmi = cmm_env_info(env);
848 rc = cmm_mode_get(env, md_obj2dev(mo), fid, &cmi->cmi_ma, NULL);
852 if (!S_ISDIR(cmi->cmi_ma.ma_attr.la_mode))
855 rc = mdo_is_subdir(env, md_object_next(mo), fid, sfid);
859 static const struct md_dir_operations cml_dir_ops = {
860 .mdo_is_subdir = cmm_is_subdir,
861 .mdo_lookup = cml_lookup,
862 .mdo_lock_mode = cml_lock_mode,
863 .mdo_create = cml_create,
864 .mdo_link = cml_link,
865 .mdo_unlink = cml_unlink,
866 .mdo_name_insert = cml_name_insert,
867 .mdo_rename = cml_rename,
868 .mdo_rename_tgt = cml_rename_tgt,
869 .mdo_create_data = cml_create_data
882 /** Get cmr_object from lu_object. */
883 static inline struct cmr_object *lu2cmr_obj(struct lu_object *o)
885 return container_of0(o, struct cmr_object, cmm_obj.cmo_obj.mo_lu);
887 /** Get cmr_object from md_object. */
888 static inline struct cmr_object *md2cmr_obj(struct md_object *mo)
890 return container_of0(mo, struct cmr_object, cmm_obj.cmo_obj);
892 /** Get cmr_object from cmm_object. */
893 static inline struct cmr_object *cmm2cmr_obj(struct cmm_object *co)
895 return container_of0(co, struct cmr_object, cmm_obj);
900 * Get proper child device from MDCs.
902 static struct lu_device *cmr_child_dev(struct cmm_device *d, __u32 num)
904 struct lu_device *next = NULL;
905 struct mdc_device *mdc;
907 cfs_spin_lock(&d->cmm_tgt_guard);
908 cfs_list_for_each_entry(mdc, &d->cmm_targets, mc_linkage) {
909 if (mdc->mc_num == num) {
910 next = mdc2lu_dev(mdc);
914 cfs_spin_unlock(&d->cmm_tgt_guard);
921 static void cmr_object_free(const struct lu_env *env,
922 struct lu_object *lo)
924 struct cmr_object *cro = lu2cmr_obj(lo);
930 * Initialize cmr object.
932 static int cmr_object_init(const struct lu_env *env, struct lu_object *lo,
933 const struct lu_object_conf *unused)
935 struct cmm_device *cd = lu2cmm_dev(lo->lo_dev);
936 struct lu_device *c_dev;
937 struct lu_object *c_obj;
942 c_dev = cmr_child_dev(cd, lu2cmr_obj(lo)->cmo_num);
946 c_obj = c_dev->ld_ops->ldo_object_alloc(env,
947 lo->lo_header, c_dev);
949 lu_object_add(lo, c_obj);
960 * Output lu_object data.
962 static int cmr_object_print(const struct lu_env *env, void *cookie,
963 lu_printer_t p, const struct lu_object *lo)
965 const struct cmr_object *cro = lu2cmr_obj((struct lu_object *)lo);
966 return (*p)(env, cookie, "[remote](mds_num=%d)", cro->cmo_num);
970 * Cmr instance of lu_object_operations.
972 static const struct lu_object_operations cmr_obj_ops = {
973 .loo_object_init = cmr_object_init,
974 .loo_object_free = cmr_object_free,
975 .loo_object_print = cmr_object_print
979 * \name cmr remote md_object operations.
980 * All operations here are invalid and return errors. There is no local object
981 * so these operations return two kinds of error:
982 * -# -EFAULT if operation is prohibited.
983 * -# -EREMOTE if operation can be done just to notify upper level about remote
988 static int cmr_object_create(const struct lu_env *env,
989 struct md_object *mo,
990 const struct md_op_spec *spec,
996 static int cmr_permission(const struct lu_env *env,
997 struct md_object *p, struct md_object *c,
998 struct md_attr *attr, int mask)
1003 static int cmr_attr_get(const struct lu_env *env, struct md_object *mo,
1004 struct md_attr *attr)
1009 static int cmr_attr_set(const struct lu_env *env, struct md_object *mo,
1010 const struct md_attr *attr)
1015 static int cmr_xattr_get(const struct lu_env *env, struct md_object *mo,
1016 struct lu_buf *buf, const char *name)
1021 static int cmr_readlink(const struct lu_env *env, struct md_object *mo,
1027 static int cmr_xattr_list(const struct lu_env *env, struct md_object *mo,
1033 static int cmr_xattr_set(const struct lu_env *env, struct md_object *mo,
1034 const struct lu_buf *buf, const char *name,
1040 static int cmr_xattr_del(const struct lu_env *env, struct md_object *mo,
1046 static int cmr_ref_add(const struct lu_env *env, struct md_object *mo,
1047 const struct md_attr *ma)
1052 static int cmr_ref_del(const struct lu_env *env, struct md_object *mo,
1058 static int cmr_open(const struct lu_env *env, struct md_object *mo,
1064 static int cmr_close(const struct lu_env *env, struct md_object *mo,
1070 static int cmr_readpage(const struct lu_env *env, struct md_object *mo,
1071 const struct lu_rdpg *rdpg)
1076 static int cmr_capa_get(const struct lu_env *env, struct md_object *mo,
1077 struct lustre_capa *capa, int renewal)
1082 static int cmr_path(const struct lu_env *env, struct md_object *obj,
1083 char *path, int pathlen, __u64 *recno, int *linkno)
1088 static int cmr_object_sync(const struct lu_env *env, struct md_object *mo)
1094 * cmr moo_version_get().
1096 static dt_obj_version_t cmr_version_get(const struct lu_env *env,
1097 struct md_object *mo)
1099 /** Don't check remote object version */
1105 * cmr moo_version_set().
1106 * No need to update remote object version here, it is done as a part
1107 * of reintegration of partial operation on the remote server.
1109 static void cmr_version_set(const struct lu_env *env, struct md_object *mo,
1110 dt_obj_version_t version)
1115 /** Set of md_object_operations for cmr. */
1116 static const struct md_object_operations cmr_mo_ops = {
1117 .moo_permission = cmr_permission,
1118 .moo_attr_get = cmr_attr_get,
1119 .moo_attr_set = cmr_attr_set,
1120 .moo_xattr_get = cmr_xattr_get,
1121 .moo_xattr_set = cmr_xattr_set,
1122 .moo_xattr_list = cmr_xattr_list,
1123 .moo_xattr_del = cmr_xattr_del,
1124 .moo_object_create = cmr_object_create,
1125 .moo_ref_add = cmr_ref_add,
1126 .moo_ref_del = cmr_ref_del,
1127 .moo_open = cmr_open,
1128 .moo_close = cmr_close,
1129 .moo_readpage = cmr_readpage,
1130 .moo_readlink = cmr_readlink,
1131 .moo_capa_get = cmr_capa_get,
1132 .moo_object_sync = cmr_object_sync,
1133 .moo_version_get = cmr_version_get,
1134 .moo_version_set = cmr_version_set,
1135 .moo_path = cmr_path,
1140 * \name cmr md_dir operations.
1142 * All methods below are cross-ref by nature. They consist of remote call and
1143 * local operation. Due to future rollback functionality there are several
1144 * limitations for such methods:
1145 * -# remote call should be done at first to do epoch negotiation between all
1146 * MDS involved and to avoid the RPC inside transaction.
1147 * -# only one RPC can be sent - also due to epoch negotiation.
1148 * For more details see rollback HLD/DLD.
1151 static int cmr_lookup(const struct lu_env *env, struct md_object *mo_p,
1152 const struct lu_name *lname, struct lu_fid *lf,
1153 struct md_op_spec *spec)
1156 * This can happens while rename() If new parent is remote dir, lookup
1163 /** Return lock mode. */
1164 static mdl_mode_t cmr_lock_mode(const struct lu_env *env,
1165 struct md_object *mo, mdl_mode_t lm)
1171 * Create operation for cmr.
1172 * Remote object creation and local name insert.
1174 * \param mo_p Parent directory. Local object.
1175 * \param lchild_name name of file to create.
1176 * \param mo_c Child object. It has no real inode yet.
1177 * \param spec creation specification.
1178 * \param ma child object attributes.
1180 static int cmr_create(const struct lu_env *env, struct md_object *mo_p,
1181 const struct lu_name *lchild_name, struct md_object *mo_c,
1182 struct md_op_spec *spec,
1185 struct cmm_thread_info *cmi;
1186 struct md_attr *tmp_ma;
1190 /* Make sure that name isn't exist before doing remote call. */
1191 rc = mdo_lookup(env, md_object_next(mo_p), lchild_name,
1192 &cmm_env_info(env)->cmi_fid, NULL);
1195 else if (rc != -ENOENT)
1198 /* check the SGID attr */
1199 cmi = cmm_env_info(env);
1201 tmp_ma = &cmi->cmi_ma;
1202 tmp_ma->ma_valid = 0;
1203 tmp_ma->ma_need = MA_INODE;
1205 #ifdef CONFIG_FS_POSIX_ACL
1206 if (!S_ISLNK(ma->ma_attr.la_mode)) {
1207 tmp_ma->ma_acl = cmi->cmi_xattr_buf;
1208 tmp_ma->ma_acl_size = sizeof(cmi->cmi_xattr_buf);
1209 tmp_ma->ma_need |= MA_ACL_DEF;
1212 rc = mo_attr_get(env, md_object_next(mo_p), tmp_ma);
1216 if (tmp_ma->ma_attr.la_mode & S_ISGID) {
1217 ma->ma_attr.la_gid = tmp_ma->ma_attr.la_gid;
1218 if (S_ISDIR(ma->ma_attr.la_mode)) {
1219 ma->ma_attr.la_mode |= S_ISGID;
1220 ma->ma_attr.la_valid |= LA_MODE;
1224 #ifdef CONFIG_FS_POSIX_ACL
1225 if (tmp_ma->ma_valid & MA_ACL_DEF) {
1226 spec->u.sp_ea.fid = spec->u.sp_pfid;
1227 spec->u.sp_ea.eadata = tmp_ma->ma_acl;
1228 spec->u.sp_ea.eadatalen = tmp_ma->ma_acl_size;
1229 spec->sp_cr_flags |= MDS_CREATE_RMT_ACL;
1233 /* Local permission check for name_insert before remote ops. */
1234 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1235 (S_ISDIR(ma->ma_attr.la_mode) ?
1236 MAY_LINK : MAY_CREATE));
1241 * \note \a ma will be changed after mo_object_create(), but we will use
1242 * it for mdo_name_insert() later, so save it before mo_object_create().
1245 rc = mo_object_create(env, md_object_next(mo_c), spec, ma);
1247 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1248 rc = mdo_name_insert(env, md_object_next(mo_p), lchild_name,
1249 lu_object_fid(&mo_c->mo_lu), tmp_ma);
1251 /* TODO: remove object mo_c on remote MDS */
1252 CWARN("cmr_create failed, should revoke: [mo_p "DFID"]"
1253 " [name %s] [mo_c "DFID"] [err %d]\n",
1254 PFID(lu_object_fid(&mo_p->mo_lu)),
1255 lchild_name->ln_name,
1256 PFID(lu_object_fid(&mo_c->mo_lu)), rc);
1264 * Link operations for cmr.
1266 * The link RPC is always issued to the server where source parent is living.
1267 * The first operation to do is object nlink increment on remote server.
1268 * Second one is local mdo_name_insert().
1270 * \param mo_p parent directory. It is local.
1271 * \param mo_s source object to link. It is remote.
1272 * \param lname Name of link file.
1273 * \param ma object attributes.
1275 static int cmr_link(const struct lu_env *env, struct md_object *mo_p,
1276 struct md_object *mo_s, const struct lu_name *lname,
1282 /* Make sure that name isn't exist before doing remote call. */
1283 rc = mdo_lookup(env, md_object_next(mo_p), lname,
1284 &cmm_env_info(env)->cmi_fid, NULL);
1287 } else if (rc == -ENOENT) {
1288 /* Local permission check for name_insert before remote ops. */
1289 rc = mo_permission(env, NULL, md_object_next(mo_p), NULL,
1294 rc = mo_ref_add(env, md_object_next(mo_s), ma);
1296 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1297 rc = mdo_name_insert(env, md_object_next(mo_p), lname,
1298 lu_object_fid(&mo_s->mo_lu), ma);
1300 /* TODO: ref_del from mo_s on remote MDS */
1301 CWARN("cmr_link failed, should revoke: "
1302 "[mo_p "DFID"] [mo_s "DFID"] "
1303 "[name %s] [err %d]\n",
1304 PFID(lu_object_fid(&mo_p->mo_lu)),
1305 PFID(lu_object_fid(&mo_s->mo_lu)),
1306 lname->ln_name, rc);
1314 * Unlink operations for cmr.
1316 * The unlink RPC is always issued to the server where parent is living. Hence
1317 * the first operation to do is object unlink on remote server. Second one is
1318 * local mdo_name_remove().
1320 * \param mo_p parent md_object. It is local.
1321 * \param mo_c child object to be unlinked. It is remote.
1322 * \param lname Name of file to unlink.
1323 * \param ma object attributes.
1325 static int cmr_unlink(const struct lu_env *env, struct md_object *mo_p,
1326 struct md_object *mo_c, const struct lu_name *lname,
1329 struct cmm_thread_info *cmi;
1330 struct md_attr *tmp_ma;
1334 /* Local permission check for name_remove before remote ops. */
1335 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1336 MAY_UNLINK | MAY_VTX_PART);
1341 * \note \a ma will be changed after mo_ref_del, but we will use
1342 * it for mdo_name_remove() later, so save it before mo_ref_del().
1344 cmi = cmm_env_info(env);
1345 tmp_ma = &cmi->cmi_ma;
1347 rc = mo_ref_del(env, md_object_next(mo_c), ma);
1349 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1350 rc = mdo_name_remove(env, md_object_next(mo_p), lname, tmp_ma);
1352 /* TODO: ref_add to mo_c on remote MDS */
1353 CWARN("cmr_unlink failed, should revoke: [mo_p "DFID"]"
1354 " [mo_c "DFID"] [name %s] [err %d]\n",
1355 PFID(lu_object_fid(&mo_p->mo_lu)),
1356 PFID(lu_object_fid(&mo_c->mo_lu)),
1357 lname->ln_name, rc);
1364 /** Helper which outputs error message during cmr_rename() */
1365 static inline void cmr_rename_warn(const char *fname,
1366 struct md_object *mo_po,
1367 struct md_object *mo_pn,
1368 const struct lu_fid *lf,
1373 CWARN("cmr_rename failed for %s, should revoke: "
1374 "[mo_po "DFID"] [mo_pn "DFID"] [lf "DFID"] "
1375 "[sname %s] [tname %s] [err %d]\n", fname,
1376 PFID(lu_object_fid(&mo_po->mo_lu)),
1377 PFID(lu_object_fid(&mo_pn->mo_lu)),
1378 PFID(lf), s_name, t_name, err);
1382 * Rename operation for cmr.
1384 * This is the most complex cross-reference operation. It may consist of up to 4
1385 * MDS server and require several RPCs to be sent.
1387 * \param mo_po Old parent object.
1388 * \param mo_pn New parent object.
1389 * \param lf FID of object to rename.
1390 * \param ls_name Source file name.
1391 * \param mo_t target object. Should be NULL here.
1392 * \param lt_name Name of target file.
1393 * \param ma object attributes.
1395 static int cmr_rename(const struct lu_env *env,
1396 struct md_object *mo_po, struct md_object *mo_pn,
1397 const struct lu_fid *lf, const struct lu_name *ls_name,
1398 struct md_object *mo_t, const struct lu_name *lt_name,
1401 struct cmm_thread_info *cmi;
1402 struct md_attr *tmp_ma;
1406 LASSERT(mo_t == NULL);
1408 /* get real type of src */
1409 rc = cmm_mode_get(env, md_obj2dev(mo_po), lf, ma, NULL);
1413 /* Local permission check for name_remove before remote ops. */
1414 rc = mo_permission(env, NULL, md_object_next(mo_po), ma,
1415 MAY_UNLINK | MAY_VTX_FULL);
1420 * \todo \a ma maybe changed after mdo_rename_tgt(), but we will use it
1421 * for mdo_name_remove() later, so save it before mdo_rename_tgt.
1423 cmi = cmm_env_info(env);
1424 tmp_ma = &cmi->cmi_ma;
1427 * \note The \a mo_pn is remote directory, so we cannot even know if there is
1428 * \a mo_t or not. Therefore \a mo_t is NULL here but remote server should do
1429 * lookup and process this further.
1431 rc = mdo_rename_tgt(env, md_object_next(mo_pn),
1432 NULL/* mo_t */, lf, lt_name, ma);
1436 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1438 /* src object maybe on remote MDS, do remote ops first. */
1439 rc = cmm_rename_ctime(env, md_obj2dev(mo_po), lf, tmp_ma);
1441 /* TODO: revoke mdo_rename_tgt */
1442 cmr_rename_warn("cmm_rename_ctime", mo_po, mo_pn, lf,
1443 ls_name->ln_name, lt_name->ln_name, rc);
1447 /* only old name is removed localy */
1448 rc = mdo_name_remove(env, md_object_next(mo_po), ls_name, tmp_ma);
1450 /* TODO: revoke all cmr_rename */
1451 cmr_rename_warn("mdo_name_remove", mo_po, mo_pn, lf,
1452 ls_name->ln_name, lt_name->ln_name, rc);
1458 * Part of cross-ref rename().
1459 * Used to insert new name in new parent and unlink target.
1461 static int cmr_rename_tgt(const struct lu_env *env,
1462 struct md_object *mo_p, struct md_object *mo_t,
1463 const struct lu_fid *lf, const struct lu_name *lname,
1466 struct cmm_thread_info *cmi;
1467 struct md_attr *tmp_ma;
1471 /* target object is remote one */
1472 /* Local permission check for rename_tgt before remote ops. */
1473 rc = mo_permission(env, NULL, md_object_next(mo_p), ma,
1474 MAY_UNLINK | MAY_VTX_PART);
1479 * XXX: @ma maybe changed after mo_ref_del, but we will use
1480 * it for mdo_rename_tgt later, so save it before mo_ref_del.
1482 cmi = cmm_env_info(env);
1483 tmp_ma = &cmi->cmi_ma;
1485 rc = mo_ref_del(env, md_object_next(mo_t), ma);
1486 /* continue locally with name handling only */
1488 tmp_ma->ma_attr_flags |= MDS_PERM_BYPASS;
1489 rc = mdo_rename_tgt(env, md_object_next(mo_p),
1490 NULL, lf, lname, tmp_ma);
1492 /* TODO: ref_add to mo_t on remote MDS */
1493 CWARN("cmr_rename_tgt failed, should revoke: "
1494 "[mo_p "DFID"] [mo_t "DFID"] [lf "DFID"] "
1495 "[name %s] [err %d]\n",
1496 PFID(lu_object_fid(&mo_p->mo_lu)),
1497 PFID(lu_object_fid(&mo_t->mo_lu)),
1499 lname->ln_name, rc);
1506 * The md_dir_operations for cmr.
1508 static const struct md_dir_operations cmr_dir_ops = {
1509 .mdo_is_subdir = cmm_is_subdir,
1510 .mdo_lookup = cmr_lookup,
1511 .mdo_lock_mode = cmr_lock_mode,
1512 .mdo_create = cmr_create,
1513 .mdo_link = cmr_link,
1514 .mdo_unlink = cmr_unlink,
1515 .mdo_rename = cmr_rename,
1516 .mdo_rename_tgt = cmr_rename_tgt