4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lustre/cmm/mdc_object.c
36 * Lustre Cluster Metadata Manager (cmm)
38 * Author: Mike Pershin <tappro@clusterfs.com>
42 # define EXPORT_SYMTAB
45 #define DEBUG_SUBSYSTEM S_MDS
46 #include <obd_support.h>
47 #include <lustre_lib.h>
48 #include <obd_class.h>
49 #include <lustre_mdc.h>
50 #include "cmm_internal.h"
51 #include "mdc_internal.h"
53 static const struct md_object_operations mdc_mo_ops;
54 static const struct md_dir_operations mdc_dir_ops;
55 static const struct lu_object_operations mdc_obj_ops;
57 extern struct lu_context_key mdc_thread_key;
63 * Allocate new mdc object.
65 struct lu_object *mdc_object_alloc(const struct lu_env *env,
66 const struct lu_object_header *hdr,
69 struct mdc_object *mco;
76 lo = &mco->mco_obj.mo_lu;
77 lu_object_init(lo, NULL, ld);
78 mco->mco_obj.mo_ops = &mdc_mo_ops;
79 mco->mco_obj.mo_dir_ops = &mdc_dir_ops;
80 lo->lo_ops = &mdc_obj_ops;
86 /** Free current mdc object */
87 static void mdc_object_free(const struct lu_env *env, struct lu_object *lo)
89 struct mdc_object *mco = lu2mdc_obj(lo);
95 * Initialize mdc object. All of them have loh_attr::LOHA_REMOTE set.
97 static int mdc_object_init(const struct lu_env *env, struct lu_object *lo,
98 const struct lu_object_conf *unused)
101 lo->lo_header->loh_attr |= LOHA_REMOTE;
106 * Instance of lu_object_operations for mdc.
108 static const struct lu_object_operations mdc_obj_ops = {
109 .loo_object_init = mdc_object_init,
110 .loo_object_free = mdc_object_free,
114 * \name The set of md_object_operations.
118 * Get mdc_thread_info from lu_context
121 struct mdc_thread_info *mdc_info_get(const struct lu_env *env)
123 struct mdc_thread_info *mci;
125 mci = lu_context_key_get(&env->le_ctx, &mdc_thread_key);
131 * Initialize mdc_thread_info.
134 struct mdc_thread_info *mdc_info_init(const struct lu_env *env)
136 struct mdc_thread_info *mci = mdc_info_get(env);
137 memset(mci, 0, sizeof(*mci));
142 * Convert attributes from mdt_body to the md_attr.
144 static void mdc_body2attr(struct mdt_body *body, struct md_attr *ma)
146 struct lu_attr *la = &ma->ma_attr;
148 if (body->valid & OBD_MD_FLCTIME && body->ctime >= la->la_ctime) {
149 la->la_ctime = body->ctime;
150 if (body->valid & OBD_MD_FLMTIME)
151 la->la_mtime = body->mtime;
154 if (body->valid & OBD_MD_FLMODE)
155 la->la_mode = body->mode;
156 if (body->valid & OBD_MD_FLSIZE)
157 la->la_size = body->size;
158 if (body->valid & OBD_MD_FLBLOCKS)
159 la->la_blocks = body->blocks;
160 if (body->valid & OBD_MD_FLUID)
161 la->la_uid = body->uid;
162 if (body->valid & OBD_MD_FLGID)
163 la->la_gid = body->gid;
164 if (body->valid & OBD_MD_FLFLAGS)
165 la->la_flags = body->flags;
166 if (body->valid & OBD_MD_FLNLINK)
167 la->la_nlink = body->nlink;
168 if (body->valid & OBD_MD_FLRDEV)
169 la->la_rdev = body->rdev;
171 la->la_valid = body->valid;
172 ma->ma_valid = MA_INODE;
176 * Fill the md_attr \a ma with attributes from request.
178 static int mdc_req2attr_update(const struct lu_env *env,
181 struct mdc_thread_info *mci;
182 struct ptlrpc_request *req;
183 struct mdt_body *body;
184 struct lov_mds_md *md;
185 struct llog_cookie *cookie;
189 mci = mdc_info_get(env);
192 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
194 mdc_body2attr(body, ma);
196 if (body->valid & OBD_MD_FLMDSCAPA) {
197 struct lustre_capa *capa;
199 /* create for cross-ref will fetch mds capa from remote obj */
200 capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA1);
201 LASSERT(capa != NULL);
202 LASSERT(ma->ma_capa != NULL);
203 *ma->ma_capa = *capa;
206 if ((body->valid & OBD_MD_FLEASIZE) || (body->valid & OBD_MD_FLDIREA)) {
207 if (body->eadatasize == 0) {
208 CERROR("No size defined for easize field\n");
212 md = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
217 LASSERT(ma->ma_lmm != NULL);
218 LASSERT(ma->ma_lmm_size >= body->eadatasize);
219 ma->ma_lmm_size = body->eadatasize;
220 memcpy(ma->ma_lmm, md, ma->ma_lmm_size);
221 ma->ma_valid |= MA_LOV;
224 if (body->valid & OBD_MD_FLCOOKIE) {
226 * ACL and cookie share the same body->aclsize, we need
227 * to make sure that they both never come here.
229 LASSERT(!(body->valid & OBD_MD_FLACL));
231 if (body->aclsize == 0) {
232 CERROR("No size defined for cookie field\n");
236 cookie = req_capsule_server_sized_get(&req->rq_pill,
242 LASSERT(ma->ma_cookie != NULL);
243 LASSERT(ma->ma_cookie_size == body->aclsize);
244 memcpy(ma->ma_cookie, cookie, ma->ma_cookie_size);
245 ma->ma_valid |= MA_COOKIE;
248 #ifdef CONFIG_FS_POSIX_ACL
249 if (body->valid & OBD_MD_FLACL) {
250 if (body->aclsize == 0) {
251 CERROR("No size defined for acl field\n");
255 acl = req_capsule_server_sized_get(&req->rq_pill,
261 LASSERT(ma->ma_acl != NULL);
262 LASSERT(ma->ma_acl_size == body->aclsize);
263 memcpy(ma->ma_acl, acl, ma->ma_acl_size);
264 ma->ma_valid |= MA_ACL_DEF;
272 * The md_object_operations::moo_attr_get() in mdc.
274 static int mdc_attr_get(const struct lu_env *env, struct md_object *mo,
277 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
278 struct mdc_thread_info *mci;
282 mci = lu_context_key_get(&env->le_ctx, &mdc_thread_key);
285 memset(&mci->mci_opdata, 0, sizeof(mci->mci_opdata));
287 memcpy(&mci->mci_opdata.op_fid1, lu_object_fid(&mo->mo_lu),
288 sizeof (struct lu_fid));
289 mci->mci_opdata.op_valid = OBD_MD_FLMODE | OBD_MD_FLUID |
290 OBD_MD_FLGID | OBD_MD_FLFLAGS |
293 rc = md_getattr(mc->mc_desc.cl_exp, &mci->mci_opdata, &mci->mci_req);
295 /* get attr from request */
296 rc = mdc_req2attr_update(env, ma);
299 ptlrpc_req_finished(mci->mci_req);
305 * Helper to init timspec \a t.
307 static inline struct timespec *mdc_attr_time(struct timespec *t, obd_time seconds)
315 * The md_object_operations::moo_attr_set() in mdc.
317 * \note It is only used for set ctime when rename's source on remote MDS.
319 static int mdc_attr_set(const struct lu_env *env, struct md_object *mo,
320 const struct md_attr *ma)
322 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
323 const struct lu_attr *la = &ma->ma_attr;
324 struct mdc_thread_info *mci;
325 struct md_ucred *uc = md_ucred(env);
329 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
331 mci = lu_context_key_get(&env->le_ctx, &mdc_thread_key);
334 memset(&mci->mci_opdata, 0, sizeof(mci->mci_opdata));
336 mci->mci_opdata.op_fid1 = *lu_object_fid(&mo->mo_lu);
337 mdc_attr_time(&mci->mci_opdata.op_attr.ia_ctime, la->la_ctime);
338 mci->mci_opdata.op_attr.ia_mode = la->la_mode;
339 mci->mci_opdata.op_attr.ia_valid = ATTR_CTIME_SET;
341 ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))) {
342 mci->mci_opdata.op_fsuid = uc->mu_fsuid;
343 mci->mci_opdata.op_fsgid = uc->mu_fsgid;
344 mci->mci_opdata.op_cap = uc->mu_cap;
345 if (uc->mu_ginfo || (uc->mu_valid == UCRED_OLD)) {
346 mci->mci_opdata.op_suppgids[0] = uc->mu_suppgids[0];
347 mci->mci_opdata.op_suppgids[1] = uc->mu_suppgids[1];
349 mci->mci_opdata.op_suppgids[0] =
350 mci->mci_opdata.op_suppgids[1] = -1;
353 mci->mci_opdata.op_fsuid = la->la_uid;
354 mci->mci_opdata.op_fsgid = la->la_gid;
355 mci->mci_opdata.op_cap = cfs_curproc_cap_pack();
356 mci->mci_opdata.op_suppgids[0] =
357 mci->mci_opdata.op_suppgids[1] = -1;
360 rc = md_setattr(mc->mc_desc.cl_exp, &mci->mci_opdata,
361 NULL, 0, NULL, 0, &mci->mci_req, NULL);
363 ptlrpc_req_finished(mci->mci_req);
369 * The md_object_operations::moo_object_create() in mdc.
371 static int mdc_object_create(const struct lu_env *env,
372 struct md_object *mo,
373 const struct md_op_spec *spec,
376 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
377 struct lu_attr *la = &ma->ma_attr;
378 struct mdc_thread_info *mci;
380 struct md_ucred *uc = md_ucred(env);
387 LASSERT(S_ISDIR(la->la_mode));
388 LASSERT(spec->u.sp_pfid != NULL);
390 mci = mdc_info_init(env);
391 mci->mci_opdata.op_bias = MDS_CROSS_REF;
392 mci->mci_opdata.op_fid2 = *lu_object_fid(&mo->mo_lu);
394 /* Parent fid is needed to create dotdot on the remote node. */
395 mci->mci_opdata.op_fid1 = *(spec->u.sp_pfid);
396 mci->mci_opdata.op_mod_time = la->la_ctime;
398 ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))) {
400 if (la->la_mode & S_ISGID)
405 if (uc->mu_ginfo || (uc->mu_valid == UCRED_OLD))
406 mci->mci_opdata.op_suppgids[0] = uc->mu_suppgids[0];
408 mci->mci_opdata.op_suppgids[0] = -1;
413 mci->mci_opdata.op_suppgids[0] = -1;
416 /* get data from spec */
417 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
418 symname = spec->u.sp_ea.eadata;
419 symlen = spec->u.sp_ea.eadatalen;
420 mci->mci_opdata.op_fid1 = *(spec->u.sp_ea.fid);
421 mci->mci_opdata.op_flags |= MDS_CREATE_SLAVE_OBJ;
422 #ifdef CONFIG_FS_POSIX_ACL
423 } else if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
424 symname = spec->u.sp_ea.eadata;
425 symlen = spec->u.sp_ea.eadatalen;
426 mci->mci_opdata.op_fid1 = *(spec->u.sp_ea.fid);
427 mci->mci_opdata.op_flags |= MDS_CREATE_RMT_ACL;
430 symname = spec->u.sp_symname;
431 symlen = symname ? strlen(symname) + 1 : 0;
434 rc = md_create(mc->mc_desc.cl_exp, &mci->mci_opdata,
435 symname, symlen, la->la_mode, uid, gid,
436 cap, la->la_rdev, &mci->mci_req);
439 /* get attr from request */
440 rc = mdc_req2attr_update(env, ma);
443 ptlrpc_req_finished(mci->mci_req);
449 * The md_object_operations::moo_ref_add() in mdc.
451 static int mdc_ref_add(const struct lu_env *env, struct md_object *mo,
452 const struct md_attr *ma)
454 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
455 const struct lu_attr *la = &ma->ma_attr;
456 struct mdc_thread_info *mci;
457 struct md_ucred *uc = md_ucred(env);
461 mci = lu_context_key_get(&env->le_ctx, &mdc_thread_key);
464 memset(&mci->mci_opdata, 0, sizeof(mci->mci_opdata));
465 mci->mci_opdata.op_bias = MDS_CROSS_REF;
466 mci->mci_opdata.op_fid1 = *lu_object_fid(&mo->mo_lu);
467 mci->mci_opdata.op_mod_time = la->la_ctime;
469 ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))) {
470 mci->mci_opdata.op_fsuid = uc->mu_fsuid;
471 mci->mci_opdata.op_fsgid = uc->mu_fsgid;
472 mci->mci_opdata.op_cap = uc->mu_cap;
473 if (uc->mu_ginfo || (uc->mu_valid == UCRED_OLD)) {
474 mci->mci_opdata.op_suppgids[0] = uc->mu_suppgids[0];
475 mci->mci_opdata.op_suppgids[1] = uc->mu_suppgids[1];
477 mci->mci_opdata.op_suppgids[0] =
478 mci->mci_opdata.op_suppgids[1] = -1;
481 mci->mci_opdata.op_fsuid = la->la_uid;
482 mci->mci_opdata.op_fsgid = la->la_gid;
483 mci->mci_opdata.op_cap = cfs_curproc_cap_pack();
484 mci->mci_opdata.op_suppgids[0] =
485 mci->mci_opdata.op_suppgids[1] = -1;
489 rc = md_link(mc->mc_desc.cl_exp, &mci->mci_opdata, &mci->mci_req);
491 ptlrpc_req_finished(mci->mci_req);
497 * The md_object_operations::moo_ref_del() in mdc.
499 static int mdc_ref_del(const struct lu_env *env, struct md_object *mo,
502 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
503 struct lu_attr *la = &ma->ma_attr;
504 struct mdc_thread_info *mci;
505 struct md_ucred *uc = md_ucred(env);
509 mci = mdc_info_init(env);
510 mci->mci_opdata.op_bias = MDS_CROSS_REF;
511 if (ma->ma_attr_flags & MDS_VTX_BYPASS)
512 mci->mci_opdata.op_bias |= MDS_VTX_BYPASS;
514 mci->mci_opdata.op_bias &= ~MDS_VTX_BYPASS;
515 mci->mci_opdata.op_fid1 = *lu_object_fid(&mo->mo_lu);
516 mci->mci_opdata.op_mode = la->la_mode;
517 mci->mci_opdata.op_mod_time = la->la_ctime;
519 ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))) {
520 mci->mci_opdata.op_fsuid = uc->mu_fsuid;
521 mci->mci_opdata.op_fsgid = uc->mu_fsgid;
522 mci->mci_opdata.op_cap = uc->mu_cap;
523 if (uc->mu_ginfo || (uc->mu_valid == UCRED_OLD))
524 mci->mci_opdata.op_suppgids[0] = uc->mu_suppgids[0];
526 mci->mci_opdata.op_suppgids[0] = -1;
528 mci->mci_opdata.op_fsuid = la->la_uid;
529 mci->mci_opdata.op_fsgid = la->la_gid;
530 mci->mci_opdata.op_cap = cfs_curproc_cap_pack();
531 mci->mci_opdata.op_suppgids[0] = -1;
534 rc = md_unlink(mc->mc_desc.cl_exp, &mci->mci_opdata, &mci->mci_req);
536 /* get attr from request */
537 rc = mdc_req2attr_update(env, ma);
540 ptlrpc_req_finished(mci->mci_req);
545 #ifdef HAVE_SPLIT_SUPPORT
546 /** Send page with directory entries to another MDS. */
547 int mdc_send_page(struct cmm_device *cm, const struct lu_env *env,
548 struct md_object *mo, struct page *page, __u32 offset)
550 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
554 rc = mdc_sendpage(mc->mc_desc.cl_exp, lu_object_fid(&mo->mo_lu),
556 CDEBUG(rc ? D_ERROR : D_INFO, "send page %p offset %d fid "DFID
557 " rc %d \n", page, offset, PFID(lu_object_fid(&mo->mo_lu)), rc);
563 * Instance of md_object_operations for mdc.
565 static const struct md_object_operations mdc_mo_ops = {
566 .moo_attr_get = mdc_attr_get,
567 .moo_attr_set = mdc_attr_set,
568 .moo_object_create = mdc_object_create,
569 .moo_ref_add = mdc_ref_add,
570 .moo_ref_del = mdc_ref_del,
575 * \name The set of md_dir_operations.
579 * The md_dir_operations::mdo_rename_tgt in mdc.
581 static int mdc_rename_tgt(const struct lu_env *env, struct md_object *mo_p,
582 struct md_object *mo_t, const struct lu_fid *lf,
583 const struct lu_name *lname, struct md_attr *ma)
585 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo_p));
586 struct lu_attr *la = &ma->ma_attr;
587 struct mdc_thread_info *mci;
588 struct md_ucred *uc = md_ucred(env);
592 mci = mdc_info_init(env);
593 mci->mci_opdata.op_bias = MDS_CROSS_REF;
594 if (ma->ma_attr_flags & MDS_VTX_BYPASS)
595 mci->mci_opdata.op_bias |= MDS_VTX_BYPASS;
597 mci->mci_opdata.op_bias &= ~MDS_VTX_BYPASS;
598 mci->mci_opdata.op_fid1 = *lu_object_fid(&mo_p->mo_lu);
599 mci->mci_opdata.op_fid2 = *lf;
600 mci->mci_opdata.op_mode = la->la_mode;
601 mci->mci_opdata.op_mod_time = la->la_ctime;
603 ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))) {
604 mci->mci_opdata.op_fsuid = uc->mu_fsuid;
605 mci->mci_opdata.op_fsgid = uc->mu_fsgid;
606 mci->mci_opdata.op_cap = uc->mu_cap;
607 if (uc->mu_ginfo || (uc->mu_valid == UCRED_OLD)) {
608 mci->mci_opdata.op_suppgids[0] = uc->mu_suppgids[0];
609 mci->mci_opdata.op_suppgids[1] = uc->mu_suppgids[1];
611 mci->mci_opdata.op_suppgids[0] =
612 mci->mci_opdata.op_suppgids[1] = -1;
615 mci->mci_opdata.op_fsuid = la->la_uid;
616 mci->mci_opdata.op_fsgid = la->la_gid;
617 mci->mci_opdata.op_cap = cfs_curproc_cap_pack();
618 mci->mci_opdata.op_suppgids[0] =
619 mci->mci_opdata.op_suppgids[1] = -1;
622 rc = md_rename(mc->mc_desc.cl_exp, &mci->mci_opdata, NULL, 0,
623 lname->ln_name, lname->ln_namelen, &mci->mci_req);
625 /* get attr from request */
626 mdc_req2attr_update(env, ma);
629 ptlrpc_req_finished(mci->mci_req);
634 * Check the fids are not relatives.
635 * The md_dir_operations::mdo_is_subdir() in mdc.
637 * Return resulting fid in sfid.
638 * \retval \a sfid = 0 fids are not relatives
639 * \retval \a sfid = FID at which search stopped
641 static int mdc_is_subdir(const struct lu_env *env, struct md_object *mo,
642 const struct lu_fid *fid, struct lu_fid *sfid)
644 struct mdc_device *mc = md2mdc_dev(md_obj2dev(mo));
645 struct mdc_thread_info *mci;
646 struct mdt_body *body;
650 mci = mdc_info_init(env);
652 rc = md_is_subdir(mc->mc_desc.cl_exp, lu_object_fid(&mo->mo_lu),
654 if (rc == 0 || rc == -EREMOTE) {
655 body = req_capsule_server_get(&mci->mci_req->rq_pill,
657 LASSERT(body->valid & OBD_MD_FLID);
659 CDEBUG(D_INFO, "Remote mdo_is_subdir(), new src "DFID"\n",
663 ptlrpc_req_finished(mci->mci_req);
667 /** Instance of md_dir_operations for mdc. */
668 static const struct md_dir_operations mdc_dir_ops = {
669 .mdo_is_subdir = mdc_is_subdir,
670 .mdo_rename_tgt = mdc_rename_tgt