lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  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
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_lib.c
  37  *
  38  * Lustre Metadata Target (mdt) request unpacking helper.
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Fan Yong <fanyong@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include "mdt_internal.h"
  52 #include <lnet/lib-lnet.h>
  53
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 void mdt_exit_ucred(struct mdt_thread_info *info)
  62 {
  63         struct lu_ucred   *uc  = mdt_ucred(info);
  64         struct mdt_device *mdt = info->mti_mdt;
  65
  66         LASSERT(uc != NULL);
  67         if (uc->uc_valid != UCRED_INIT) {
  68                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  69                 if (uc->uc_ginfo) {
  70                         cfs_put_group_info(uc->uc_ginfo);
  71                         uc->uc_ginfo = NULL;
  72                 }
  73                 if (uc->uc_identity) {
  74                         mdt_identity_put(mdt->mdt_identity_cache,
  75                                          uc->uc_identity);
  76                         uc->uc_identity = NULL;
  77                 }
  78                 uc->uc_valid = UCRED_INIT;
  79         }
  80 }
  81
  82 static int match_nosquash_list(struct rw_semaphore *sem,
  83                                cfs_list_t *nidlist,
  84                                lnet_nid_t peernid)
  85 {
  86         int rc;
  87         ENTRY;
  88         down_read(sem);
  89         rc = cfs_match_nid(peernid, nidlist);
  90         up_read(sem);
  91         RETURN(rc);
  92 }
  93
  94 /* root_squash for inter-MDS operations */
  95 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  96 {
  97         struct lu_ucred *ucred = mdt_ucred(info);
  98         ENTRY;
  99
 100         LASSERT(ucred != NULL);
 101         if (!info->mti_mdt->mdt_squash_uid || ucred->uc_fsuid)
 102                 RETURN(0);
 103
 104         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 105                                 &info->mti_mdt->mdt_nosquash_nids,
 106                                 peernid)) {
 107                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 108                        libcfs_nid2str(peernid));
 109                 RETURN(0);
 110         }
 111
 112         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 113                libcfs_nid2str(peernid),
 114                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 115                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 116                0);
 117
 118         ucred->uc_fsuid = info->mti_mdt->mdt_squash_uid;
 119         ucred->uc_fsgid = info->mti_mdt->mdt_squash_gid;
 120         ucred->uc_cap = 0;
 121         ucred->uc_suppgids[0] = -1;
 122         ucred->uc_suppgids[1] = -1;
 123
 124         RETURN(0);
 125 }
 126
 127 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 128                           void *buf)
 129 {
 130         struct ptlrpc_request   *req = mdt_info_req(info);
 131         struct mdt_device       *mdt = info->mti_mdt;
 132         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 133         struct lu_ucred         *ucred = mdt_ucred(info);
 134         lnet_nid_t               peernid = req->rq_peer.nid;
 135         __u32                    perm = 0;
 136         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 137         int                      setuid;
 138         int                      setgid;
 139         int                      rc = 0;
 140
 141         ENTRY;
 142
 143         LASSERT(req->rq_auth_gss);
 144         LASSERT(!req->rq_auth_usr_mdt);
 145         LASSERT(req->rq_user_desc);
 146         LASSERT(ucred != NULL);
 147
 148         ucred->uc_valid = UCRED_INVALID;
 149
 150         ucred->uc_o_uid   = pud->pud_uid;
 151         ucred->uc_o_gid   = pud->pud_gid;
 152         ucred->uc_o_fsuid = pud->pud_fsuid;
 153         ucred->uc_o_fsgid = pud->pud_fsgid;
 154
 155         if (type == BODY_INIT) {
 156                 struct mdt_body *body = (struct mdt_body *)buf;
 157
 158                 ucred->uc_suppgids[0] = body->suppgid;
 159                 ucred->uc_suppgids[1] = -1;
 160         }
 161
 162         /* sanity check: we expect the uid which client claimed is true */
 163         if (remote) {
 164                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 165                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 166                         RETURN(-EACCES);
 167                 }
 168
 169                 if (ptlrpc_user_desc_do_idmap(req, pud))
 170                         RETURN(-EACCES);
 171
 172                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 173                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 174                                "while client claims %u:%u/%u:%u\n",
 175                                libcfs_nid2str(peernid), req->rq_auth_uid,
 176                                req->rq_auth_mapped_uid,
 177                                pud->pud_uid, pud->pud_gid,
 178                                pud->pud_fsuid, pud->pud_fsgid);
 179                         RETURN(-EACCES);
 180                 }
 181         } else {
 182                 if (req->rq_auth_uid != pud->pud_uid) {
 183                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 184                                "while client claims %u:%u/%u:%u\n",
 185                                libcfs_nid2str(peernid), req->rq_auth_uid,
 186                                pud->pud_uid, pud->pud_gid,
 187                                pud->pud_fsuid, pud->pud_fsgid);
 188                         RETURN(-EACCES);
 189                 }
 190         }
 191
 192         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 193                 if (remote) {
 194                         CDEBUG(D_SEC, "remote client must run with identity_get "
 195                                "enabled!\n");
 196                         RETURN(-EACCES);
 197                 } else {
 198                         ucred->uc_identity = NULL;
 199                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 200                                CFS_SETGRP_PERM;
 201                 }
 202         } else {
 203                 struct md_identity *identity;
 204
 205                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 206                                             pud->pud_uid);
 207                 if (IS_ERR(identity)) {
 208                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 209                                      !remote)) {
 210                                 ucred->uc_identity = NULL;
 211                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 212                                        CFS_SETGRP_PERM;
 213                         } else {
 214                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 215                                        pud->pud_uid);
 216                                 RETURN(-EACCES);
 217                         }
 218                 } else {
 219                         ucred->uc_identity = identity;
 220                         perm = mdt_identity_get_perm(ucred->uc_identity,
 221                                                      remote, peernid);
 222                 }
 223         }
 224
 225         /* find out the setuid/setgid attempt */
 226         setuid = (pud->pud_uid != pud->pud_fsuid);
 227         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 228                   (ucred->uc_identity &&
 229                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 230
 231         /* check permission of setuid */
 232         if (setuid && !(perm & CFS_SETUID_PERM)) {
 233                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 234                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 235                 GOTO(out, rc = -EACCES);
 236         }
 237
 238         /* check permission of setgid */
 239         if (setgid && !(perm & CFS_SETGID_PERM)) {
 240                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 241                        "from %s\n", pud->pud_uid, pud->pud_gid,
 242                        pud->pud_fsuid, pud->pud_fsgid,
 243                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 244                 GOTO(out, rc = -EACCES);
 245         }
 246
 247         /*
 248          * NB: remote client not allowed to setgroups anyway.
 249          */
 250         if (!remote && perm & CFS_SETGRP_PERM) {
 251                 if (pud->pud_ngroups) {
 252                         /* setgroups for local client */
 253                         ucred->uc_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 254                         if (!ucred->uc_ginfo) {
 255                                 CERROR("failed to alloc %d groups\n",
 256                                        pud->pud_ngroups);
 257                                 GOTO(out, rc = -ENOMEM);
 258                         }
 259
 260                         lustre_groups_from_list(ucred->uc_ginfo,
 261                                                 pud->pud_groups);
 262                         lustre_groups_sort(ucred->uc_ginfo);
 263                 } else {
 264                         ucred->uc_ginfo = NULL;
 265                 }
 266         } else {
 267                 ucred->uc_suppgids[0] = -1;
 268                 ucred->uc_suppgids[1] = -1;
 269                 ucred->uc_ginfo = NULL;
 270         }
 271
 272         ucred->uc_uid   = pud->pud_uid;
 273         ucred->uc_gid   = pud->pud_gid;
 274         ucred->uc_fsuid = pud->pud_fsuid;
 275         ucred->uc_fsgid = pud->pud_fsgid;
 276
 277         /* process root_squash here. */
 278         mdt_root_squash(info, peernid);
 279
 280         /* remove fs privilege for non-root user. */
 281         if (ucred->uc_fsuid)
 282                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 283         else
 284                 ucred->uc_cap = pud->pud_cap;
 285         if (remote && !(perm & CFS_RMTOWN_PERM))
 286                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 287                                    CFS_CAP_CHOWN_MASK);
 288         ucred->uc_valid = UCRED_NEW;
 289
 290         EXIT;
 291
 292 out:
 293         if (rc) {
 294                 if (ucred->uc_ginfo) {
 295                         cfs_put_group_info(ucred->uc_ginfo);
 296                         ucred->uc_ginfo = NULL;
 297                 }
 298                 if (ucred->uc_identity) {
 299                         mdt_identity_put(mdt->mdt_identity_cache,
 300                                          ucred->uc_identity);
 301                         ucred->uc_identity = NULL;
 302                 }
 303         }
 304
 305         return rc;
 306 }
 307
 308 int mdt_check_ucred(struct mdt_thread_info *info)
 309 {
 310         struct ptlrpc_request   *req = mdt_info_req(info);
 311         struct mdt_device       *mdt = info->mti_mdt;
 312         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 313         struct lu_ucred         *ucred = mdt_ucred(info);
 314         struct md_identity      *identity = NULL;
 315         lnet_nid_t               peernid = req->rq_peer.nid;
 316         __u32                    perm = 0;
 317         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 318         int                      setuid;
 319         int                      setgid;
 320         int                      rc = 0;
 321
 322         ENTRY;
 323
 324         LASSERT(ucred != NULL);
 325         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 326                 RETURN(0);
 327
 328         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 329                 RETURN(0);
 330
 331         /* sanity check: if we use strong authentication, we expect the
 332          * uid which client claimed is true */
 333         if (remote) {
 334                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 335                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 336                         RETURN(-EACCES);
 337                 }
 338
 339                 if (ptlrpc_user_desc_do_idmap(req, pud))
 340                         RETURN(-EACCES);
 341
 342                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 343                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 344                                "while client claims %u:%u/%u:%u\n",
 345                                libcfs_nid2str(peernid), req->rq_auth_uid,
 346                                req->rq_auth_mapped_uid,
 347                                pud->pud_uid, pud->pud_gid,
 348                                pud->pud_fsuid, pud->pud_fsgid);
 349                         RETURN(-EACCES);
 350                 }
 351         } else {
 352                 if (req->rq_auth_uid != pud->pud_uid) {
 353                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 354                                "while client claims %u:%u/%u:%u\n",
 355                                libcfs_nid2str(peernid), req->rq_auth_uid,
 356                                pud->pud_uid, pud->pud_gid,
 357                                pud->pud_fsuid, pud->pud_fsgid);
 358                         RETURN(-EACCES);
 359                 }
 360         }
 361
 362         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 363                 if (remote) {
 364                         CDEBUG(D_SEC, "remote client must run with identity_get "
 365                                "enabled!\n");
 366                         RETURN(-EACCES);
 367                 }
 368                 RETURN(0);
 369         }
 370
 371         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 372         if (IS_ERR(identity)) {
 373                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 374                              !remote)) {
 375                         RETURN(0);
 376                 } else {
 377                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 378                                pud->pud_uid);
 379                         RETURN(-EACCES);
 380                }
 381         }
 382
 383         perm = mdt_identity_get_perm(identity, remote, peernid);
 384         /* find out the setuid/setgid attempt */
 385         setuid = (pud->pud_uid != pud->pud_fsuid);
 386         setgid = (pud->pud_gid != pud->pud_fsgid ||
 387                   pud->pud_gid != identity->mi_gid);
 388
 389         /* check permission of setuid */
 390         if (setuid && !(perm & CFS_SETUID_PERM)) {
 391                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 392                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 393                 GOTO(out, rc = -EACCES);
 394         }
 395
 396         /* check permission of setgid */
 397         if (setgid && !(perm & CFS_SETGID_PERM)) {
 398                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 399                        "from %s\n", pud->pud_uid, pud->pud_gid,
 400                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 401                        libcfs_nid2str(peernid));
 402                 GOTO(out, rc = -EACCES);
 403         }
 404
 405         EXIT;
 406
 407 out:
 408         mdt_identity_put(mdt->mdt_identity_cache, identity);
 409         return rc;
 410 }
 411
 412 static int old_init_ucred(struct mdt_thread_info *info,
 413                           struct mdt_body *body)
 414 {
 415         struct lu_ucred *uc = mdt_ucred(info);
 416         struct mdt_device  *mdt = info->mti_mdt;
 417         struct md_identity *identity = NULL;
 418
 419         ENTRY;
 420
 421         LASSERT(uc != NULL);
 422         uc->uc_valid = UCRED_INVALID;
 423         uc->uc_o_uid = uc->uc_uid = body->uid;
 424         uc->uc_o_gid = uc->uc_gid = body->gid;
 425         uc->uc_o_fsuid = uc->uc_fsuid = body->fsuid;
 426         uc->uc_o_fsgid = uc->uc_fsgid = body->fsgid;
 427         uc->uc_suppgids[0] = body->suppgid;
 428         uc->uc_suppgids[1] = -1;
 429         uc->uc_ginfo = NULL;
 430         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 431                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 432                                             uc->uc_fsuid);
 433                 if (IS_ERR(identity)) {
 434                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 435                                 identity = NULL;
 436                         } else {
 437                                 CDEBUG(D_SEC, "Deny access without identity: "
 438                                        "uid %u\n", uc->uc_fsuid);
 439                                 RETURN(-EACCES);
 440                         }
 441                 }
 442         }
 443         uc->uc_identity = identity;
 444
 445         /* process root_squash here. */
 446         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 447
 448         /* remove fs privilege for non-root user. */
 449         if (uc->uc_fsuid)
 450                 uc->uc_cap = body->capability & ~CFS_CAP_FS_MASK;
 451         else
 452                 uc->uc_cap = body->capability;
 453         uc->uc_valid = UCRED_OLD;
 454
 455         RETURN(0);
 456 }
 457
 458 static int old_init_ucred_reint(struct mdt_thread_info *info)
 459 {
 460         struct lu_ucred *uc = mdt_ucred(info);
 461         struct mdt_device  *mdt = info->mti_mdt;
 462         struct md_identity *identity = NULL;
 463
 464         ENTRY;
 465
 466         LASSERT(uc != NULL);
 467         uc->uc_valid = UCRED_INVALID;
 468         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 469         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 470         uc->uc_ginfo = NULL;
 471         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 472                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 473                                             uc->uc_fsuid);
 474                 if (IS_ERR(identity)) {
 475                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 476                                 identity = NULL;
 477                         } else {
 478                                 CDEBUG(D_SEC, "Deny access without identity: "
 479                                        "uid %u\n", uc->uc_fsuid);
 480                                 RETURN(-EACCES);
 481                         }
 482                 }
 483         }
 484         uc->uc_identity = identity;
 485
 486         /* process root_squash here. */
 487         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 488
 489         /* remove fs privilege for non-root user. */
 490         if (uc->uc_fsuid)
 491                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 492         uc->uc_valid = UCRED_OLD;
 493
 494         RETURN(0);
 495 }
 496
 497 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 498 {
 499         struct ptlrpc_request *req = mdt_info_req(info);
 500         struct lu_ucred       *uc  = mdt_ucred(info);
 501
 502         LASSERT(uc != NULL);
 503         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 504                 return 0;
 505
 506         mdt_exit_ucred(info);
 507
 508         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 509                 return old_init_ucred(info, body);
 510         else
 511                 return new_init_ucred(info, BODY_INIT, body);
 512 }
 513
 514 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 515 {
 516         struct ptlrpc_request *req = mdt_info_req(info);
 517         struct lu_ucred       *uc  = mdt_ucred(info);
 518
 519         LASSERT(uc != NULL);
 520         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 521                 return 0;
 522
 523         mdt_exit_ucred(info);
 524
 525         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 526                 return old_init_ucred_reint(info);
 527         else
 528                 return new_init_ucred(info, REC_INIT, NULL);
 529 }
 530
 531 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 532 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 533 {
 534         const struct lov_ost_data_v1 *lod;
 535         int                           i;
 536         __u16                         count;
 537
 538         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 539
 540         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 541                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 542                le32_to_cpu(lmm->lmm_pattern));
 543         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 544                le32_to_cpu(lmm->lmm_stripe_size), count);
 545         if (count == LOV_ALL_STRIPES)
 546                 return;
 547
 548         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 549         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 550                 struct ost_id   oi;
 551                 ostid_le_to_cpu((struct ost_id *)&lod->l_ost_oi, &oi);
 552                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 553                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 554         }
 555 }
 556
 557 /* Shrink and/or grow reply buffers */
 558 int mdt_fix_reply(struct mdt_thread_info *info)
 559 {
 560         struct req_capsule *pill = info->mti_pill;
 561         struct mdt_body    *body;
 562         int                md_size, md_packed = 0;
 563         int                acl_size;
 564         int                rc = 0;
 565         ENTRY;
 566
 567         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 568         LASSERT(body != NULL);
 569
 570         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 571                 md_size = body->eadatasize;
 572         else
 573                 md_size = 0;
 574
 575         acl_size = body->aclsize;
 576
 577         /* this replay - not send info to client */
 578         if (info->mti_spec.no_create) {
 579                 md_size = 0;
 580                 acl_size = 0;
 581         }
 582
 583         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 584                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 585                         md_size, acl_size,
 586                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 587                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 588 /*
 589             &RMF_MDT_BODY,
 590             &RMF_MDT_MD,
 591             &RMF_ACL, or &RMF_LOGCOOKIES
 592 (optional)  &RMF_CAPA1,
 593 (optional)  &RMF_CAPA2,
 594 (optional)  something else
 595 */
 596
 597         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 598          * buffers before growing it */
 599         if (info->mti_big_lmm_used) {
 600                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 601                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 602                                                  RCL_SERVER);
 603                 LASSERT(md_packed > 0);
 604                 /* buffer must be allocated separately */
 605                 LASSERT(info->mti_attr.ma_lmm !=
 606                         req_capsule_server_get(pill, &RMF_MDT_MD));
 607                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 608                 /* free big lmm if md_size is not needed */
 609                 if (md_size == 0)
 610                         info->mti_big_lmm_used = 0;
 611         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 612                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 613         }
 614
 615         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 616                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 617         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 618                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 619                                    acl_size, RCL_SERVER);
 620
 621         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 622             !(body->valid & OBD_MD_FLMDSCAPA))
 623                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 624
 625         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 626             !(body->valid & OBD_MD_FLOSSCAPA))
 627                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 628
 629         /*
 630          * Some more field should be shrinked if needed.
 631          * This should be done by those who added fields to reply message.
 632          */
 633
 634         /* Grow MD buffer if needed finally */
 635         if (info->mti_big_lmm_used) {
 636                 void *lmm;
 637
 638                 LASSERT(md_size > md_packed);
 639                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 640                        md_size - md_packed);
 641                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 642                 if (rc) {
 643                         /* we can't answer with proper LOV EA, drop flags,
 644                          * the rc is also returned so this request is
 645                          * considered as failed */
 646                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 647                         /* don't return transno along with error */
 648                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 649                 } else {
 650                         /* now we need to pack right LOV EA */
 651                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 652                         LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 653                                                      RCL_SERVER) ==
 654                                 info->mti_attr.ma_lmm_size);
 655                         memcpy(lmm, info->mti_attr.ma_lmm,
 656                                info->mti_attr.ma_lmm_size);
 657                 }
 658                 /* update mdt_max_mdsize so clients will be aware about that */
 659                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 660                         info->mti_mdt->mdt_max_mdsize =
 661                                                     info->mti_attr.ma_lmm_size;
 662                 info->mti_big_lmm_used = 0;
 663         }
 664         RETURN(rc);
 665 }
 666
 667
 668 /* if object is dying, pack the lov/llog data,
 669  * parameter info->mti_attr should be valid at this point! */
 670 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 671                            const struct md_attr *ma)
 672 {
 673         struct mdt_body       *repbody;
 674         const struct lu_attr *la = &ma->ma_attr;
 675         int rc;
 676         ENTRY;
 677
 678         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 679         LASSERT(repbody != NULL);
 680
 681         if (ma->ma_valid & MA_INODE)
 682                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 683
 684         if (ma->ma_valid & MA_LOV) {
 685                 CERROR("No need in LOV EA upon unlink\n");
 686                 dump_stack();
 687         }
 688         repbody->eadatasize = 0;
 689
 690         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 691                 repbody->aclsize = ma->ma_cookie_size;
 692                 repbody->valid |= OBD_MD_FLCOOKIE;
 693         }
 694
 695         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 696             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 697             repbody->valid & OBD_MD_FLEASIZE) {
 698                 struct lustre_capa *capa;
 699
 700                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 701                 LASSERT(capa);
 702                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 703                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 704                 if (rc)
 705                         RETURN(rc);
 706
 707                 repbody->valid |= OBD_MD_FLOSSCAPA;
 708         }
 709
 710         RETURN(0);
 711 }
 712
 713 static inline unsigned int attr_unpack(__u64 sa_valid) {
 714         unsigned int ia_valid = 0;
 715
 716         if (sa_valid & MDS_ATTR_MODE)
 717                 ia_valid |= ATTR_MODE;
 718         if (sa_valid & MDS_ATTR_UID)
 719                 ia_valid |= ATTR_UID;
 720         if (sa_valid & MDS_ATTR_GID)
 721                 ia_valid |= ATTR_GID;
 722         if (sa_valid & MDS_ATTR_SIZE)
 723                 ia_valid |= ATTR_SIZE;
 724         if (sa_valid & MDS_ATTR_ATIME)
 725                 ia_valid |= ATTR_ATIME;
 726         if (sa_valid & MDS_ATTR_MTIME)
 727                 ia_valid |= ATTR_MTIME;
 728         if (sa_valid & MDS_ATTR_CTIME)
 729                 ia_valid |= ATTR_CTIME;
 730         if (sa_valid & MDS_ATTR_ATIME_SET)
 731                 ia_valid |= ATTR_ATIME_SET;
 732         if (sa_valid & MDS_ATTR_MTIME_SET)
 733                 ia_valid |= ATTR_MTIME_SET;
 734         if (sa_valid & MDS_ATTR_FORCE)
 735                 ia_valid |= ATTR_FORCE;
 736         if (sa_valid & MDS_ATTR_ATTR_FLAG)
 737                 ia_valid |= ATTR_ATTR_FLAG;
 738         if (sa_valid & MDS_ATTR_KILL_SUID)
 739                 ia_valid |=  ATTR_KILL_SUID;
 740         if (sa_valid & MDS_ATTR_KILL_SGID)
 741                 ia_valid |= ATTR_KILL_SGID;
 742         if (sa_valid & MDS_ATTR_CTIME_SET)
 743                 ia_valid |= ATTR_CTIME_SET;
 744         if (sa_valid & MDS_ATTR_FROM_OPEN)
 745                 ia_valid |= ATTR_FROM_OPEN;
 746         if (sa_valid & MDS_ATTR_BLOCKS)
 747                 ia_valid |= ATTR_BLOCKS;
 748         if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
 749                 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
 750         return ia_valid;
 751 }
 752
 753 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 754                                   struct md_attr *ma)
 755 {
 756         __u64 out;
 757
 758         out = 0;
 759         if (in & ATTR_MODE)
 760                 out |= LA_MODE;
 761         if (in & ATTR_UID)
 762                 out |= LA_UID;
 763         if (in & ATTR_GID)
 764                 out |= LA_GID;
 765         if (in & ATTR_SIZE)
 766                 out |= LA_SIZE;
 767         if (in & ATTR_BLOCKS)
 768                 out |= LA_BLOCKS;
 769
 770         if (in & ATTR_FROM_OPEN)
 771                 rr->rr_flags |= MRF_OPEN_TRUNC;
 772
 773         if (in & ATTR_ATIME_SET)
 774                 out |= LA_ATIME;
 775
 776         if (in & ATTR_CTIME_SET)
 777                 out |= LA_CTIME;
 778
 779         if (in & ATTR_MTIME_SET)
 780                 out |= LA_MTIME;
 781
 782         if (in & ATTR_ATTR_FLAG)
 783                 out |= LA_FLAGS;
 784
 785         if (in & ATTR_KILL_SUID)
 786                 out |= LA_KILL_SUID;
 787
 788         if (in & ATTR_KILL_SGID)
 789                 out |= LA_KILL_SGID;
 790
 791         if (in & MDS_OPEN_OWNEROVERRIDE)
 792                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 793
 794         if (in & ATTR_FORCE)
 795                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 796
 797         /*XXX need ATTR_RAW?*/
 798         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 799                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 800                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 801                 ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
 802                 ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
 803         if (in != 0)
 804                 CERROR("Unknown attr bits: "LPX64"\n", in);
 805         return out;
 806 }
 807 /* unpacking */
 808
 809 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 810 {
 811         struct lu_ucred         *uc  = mdt_ucred(info);
 812         struct md_attr          *ma = &info->mti_attr;
 813         struct lu_attr          *la = &ma->ma_attr;
 814         struct req_capsule      *pill = info->mti_pill;
 815         struct mdt_reint_record *rr = &info->mti_rr;
 816         struct mdt_rec_setattr  *rec;
 817         ENTRY;
 818
 819         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 820         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 821         if (rec == NULL)
 822                 RETURN(-EFAULT);
 823
 824         /* This prior initialization is needed for old_init_ucred_reint() */
 825         uc->uc_fsuid = rec->sa_fsuid;
 826         uc->uc_fsgid = rec->sa_fsgid;
 827         uc->uc_cap   = rec->sa_cap;
 828         uc->uc_suppgids[0] = rec->sa_suppgid;
 829         uc->uc_suppgids[1] = -1;
 830
 831         rr->rr_fid1 = &rec->sa_fid;
 832         la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
 833         la->la_mode  = rec->sa_mode;
 834         la->la_flags = rec->sa_attr_flags;
 835         la->la_uid   = rec->sa_uid;
 836         la->la_gid   = rec->sa_gid;
 837         la->la_size  = rec->sa_size;
 838         la->la_blocks = rec->sa_blocks;
 839         la->la_ctime = rec->sa_ctime;
 840         la->la_atime = rec->sa_atime;
 841         la->la_mtime = rec->sa_mtime;
 842         ma->ma_valid = MA_INODE;
 843
 844         if (rec->sa_bias & MDS_DATA_MODIFIED)
 845                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 846         else
 847                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 848
 849         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 850                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 851                                  req_capsule_client_get(pill, &RMF_CAPA1));
 852
 853         RETURN(0);
 854 }
 855
 856 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 857 {
 858         struct req_capsule *pill = info->mti_pill;
 859         ENTRY;
 860
 861         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 862                 info->mti_ioepoch =
 863                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 864         else
 865                 info->mti_ioepoch = NULL;
 866         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 867 }
 868
 869 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 870         struct req_capsule      *pill = info->mti_pill;
 871
 872         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 873                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 874                 if (info->mti_dlm_req == NULL)
 875                         RETURN(-EFAULT);
 876         }
 877
 878         RETURN(0);
 879 }
 880
 881 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 882 {
 883         struct mdt_reint_record *rr = &info->mti_rr;
 884         struct md_attr          *ma = &info->mti_attr;
 885         struct req_capsule      *pill = info->mti_pill;
 886         int rc;
 887         ENTRY;
 888
 889         rc = mdt_setattr_unpack_rec(info);
 890         if (rc)
 891                 RETURN(rc);
 892
 893         /* Epoch may be absent */
 894         mdt_ioepoch_unpack(info);
 895
 896         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 897                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 898                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 899                                                         RCL_CLIENT);
 900                 ma->ma_lmm_size = rr->rr_eadatalen;
 901                 if (ma->ma_lmm_size > 0) {
 902                         ma->ma_lmm = (void *)rr->rr_eadata;
 903                         ma->ma_valid |= MA_LOV;
 904                 }
 905         }
 906
 907         rc = mdt_dlmreq_unpack(info);
 908         RETURN(rc);
 909 }
 910
 911 int mdt_close_unpack(struct mdt_thread_info *info)
 912 {
 913         int rc;
 914         ENTRY;
 915
 916         rc = mdt_ioepoch_unpack(info);
 917         if (rc)
 918                 RETURN(rc);
 919
 920         rc = mdt_setattr_unpack_rec(info);
 921         if (rc)
 922                 RETURN(rc);
 923         RETURN(mdt_init_ucred_reint(info));
 924 }
 925
 926 static int mdt_create_unpack(struct mdt_thread_info *info)
 927 {
 928         struct lu_ucred         *uc  = mdt_ucred(info);
 929         struct mdt_rec_create   *rec;
 930         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 931         struct mdt_reint_record *rr = &info->mti_rr;
 932         struct req_capsule      *pill = info->mti_pill;
 933         struct md_op_spec       *sp = &info->mti_spec;
 934         int rc;
 935         ENTRY;
 936
 937         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 938         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 939         if (rec == NULL)
 940                 RETURN(-EFAULT);
 941
 942         /* This prior initialization is needed for old_init_ucred_reint() */
 943         uc->uc_fsuid = rec->cr_fsuid;
 944         uc->uc_fsgid = rec->cr_fsgid;
 945         uc->uc_cap   = rec->cr_cap;
 946         uc->uc_suppgids[0] = rec->cr_suppgid1;
 947         uc->uc_suppgids[1] = -1;
 948         uc->uc_umask = rec->cr_umask;
 949
 950         rr->rr_fid1 = &rec->cr_fid1;
 951         rr->rr_fid2 = &rec->cr_fid2;
 952         attr->la_mode = rec->cr_mode;
 953         attr->la_rdev  = rec->cr_rdev;
 954         attr->la_uid   = rec->cr_fsuid;
 955         attr->la_gid   = rec->cr_fsgid;
 956         attr->la_ctime = rec->cr_time;
 957         attr->la_mtime = rec->cr_time;
 958         attr->la_atime = rec->cr_time;
 959         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
 960                          LA_CTIME | LA_MTIME | LA_ATIME;
 961         memset(&sp->u, 0, sizeof(sp->u));
 962         sp->sp_cr_flags = get_mrc_cr_flags(rec);
 963
 964         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 965                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 966                                  req_capsule_client_get(pill, &RMF_CAPA1));
 967         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 968
 969         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 970         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
 971                                               RCL_CLIENT) - 1;
 972         LASSERT(rr->rr_name && rr->rr_namelen > 0);
 973
 974         if (S_ISLNK(attr->la_mode)) {
 975                 const char *tgt = NULL;
 976
 977                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
 978                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
 979                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
 980                         sp->u.sp_symname = tgt;
 981                 }
 982                 if (tgt == NULL)
 983                         RETURN(-EFAULT);
 984         } else {
 985                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 986         }
 987
 988         rc = mdt_dlmreq_unpack(info);
 989         RETURN(rc);
 990 }
 991
 992 static int mdt_link_unpack(struct mdt_thread_info *info)
 993 {
 994         struct lu_ucred         *uc  = mdt_ucred(info);
 995         struct mdt_rec_link     *rec;
 996         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 997         struct mdt_reint_record *rr = &info->mti_rr;
 998         struct req_capsule      *pill = info->mti_pill;
 999         int rc;
1000         ENTRY;
1001
1002         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1003         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1004         if (rec == NULL)
1005                 RETURN(-EFAULT);
1006
1007         /* This prior initialization is needed for old_init_ucred_reint() */
1008         uc->uc_fsuid = rec->lk_fsuid;
1009         uc->uc_fsgid = rec->lk_fsgid;
1010         uc->uc_cap   = rec->lk_cap;
1011         uc->uc_suppgids[0] = rec->lk_suppgid1;
1012         uc->uc_suppgids[1] = rec->lk_suppgid2;
1013
1014         attr->la_uid = rec->lk_fsuid;
1015         attr->la_gid = rec->lk_fsgid;
1016         rr->rr_fid1 = &rec->lk_fid1;
1017         rr->rr_fid2 = &rec->lk_fid2;
1018         attr->la_ctime = rec->lk_time;
1019         attr->la_mtime = rec->lk_time;
1020         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1021
1022         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1023                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1024                                  req_capsule_client_get(pill, &RMF_CAPA1));
1025         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1026                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1027                                  req_capsule_client_get(pill, &RMF_CAPA2));
1028
1029         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1030         if (rr->rr_name == NULL)
1031                 RETURN(-EFAULT);
1032         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1033
1034         LASSERT(rr->rr_namelen > 0);
1035
1036         rc = mdt_dlmreq_unpack(info);
1037         RETURN(rc);
1038 }
1039
1040 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1041 {
1042         struct lu_ucred         *uc  = mdt_ucred(info);
1043         struct mdt_rec_unlink   *rec;
1044         struct md_attr          *ma = &info->mti_attr;
1045         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1046         struct mdt_reint_record *rr = &info->mti_rr;
1047         struct req_capsule      *pill = info->mti_pill;
1048         int rc;
1049         ENTRY;
1050
1051         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1052         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1053         if (rec == NULL)
1054                 RETURN(-EFAULT);
1055
1056         /* This prior initialization is needed for old_init_ucred_reint() */
1057         uc->uc_fsuid = rec->ul_fsuid;
1058         uc->uc_fsgid = rec->ul_fsgid;
1059         uc->uc_cap   = rec->ul_cap;
1060         uc->uc_suppgids[0] = rec->ul_suppgid1;
1061         uc->uc_suppgids[1] = -1;
1062
1063         attr->la_uid = rec->ul_fsuid;
1064         attr->la_gid = rec->ul_fsgid;
1065         rr->rr_fid1 = &rec->ul_fid1;
1066         rr->rr_fid2 = &rec->ul_fid2;
1067         attr->la_ctime = rec->ul_time;
1068         attr->la_mtime = rec->ul_time;
1069         attr->la_mode  = rec->ul_mode;
1070         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1071
1072         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1073                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1074                                  req_capsule_client_get(pill, &RMF_CAPA1));
1075
1076         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1077         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1078         if (rr->rr_name == NULL || rr->rr_namelen == 0)
1079                 RETURN(-EFAULT);
1080
1081         if (rec->ul_bias & MDS_VTX_BYPASS)
1082                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1083         else
1084                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1085
1086         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1087
1088         rc = mdt_dlmreq_unpack(info);
1089         RETURN(rc);
1090 }
1091
1092 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1093 {
1094         info->mti_spec.sp_rm_entry = 1;
1095         return mdt_unlink_unpack(info);
1096 }
1097
1098 static int mdt_rename_unpack(struct mdt_thread_info *info)
1099 {
1100         struct lu_ucred         *uc = mdt_ucred(info);
1101         struct mdt_rec_rename   *rec;
1102         struct md_attr          *ma = &info->mti_attr;
1103         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1104         struct mdt_reint_record *rr = &info->mti_rr;
1105         struct req_capsule      *pill = info->mti_pill;
1106         int rc;
1107         ENTRY;
1108
1109         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1110         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1111         if (rec == NULL)
1112                 RETURN(-EFAULT);
1113
1114         /* This prior initialization is needed for old_init_ucred_reint() */
1115         uc->uc_fsuid = rec->rn_fsuid;
1116         uc->uc_fsgid = rec->rn_fsgid;
1117         uc->uc_cap   = rec->rn_cap;
1118         uc->uc_suppgids[0] = rec->rn_suppgid1;
1119         uc->uc_suppgids[1] = rec->rn_suppgid2;
1120
1121         attr->la_uid = rec->rn_fsuid;
1122         attr->la_gid = rec->rn_fsgid;
1123         rr->rr_fid1 = &rec->rn_fid1;
1124         rr->rr_fid2 = &rec->rn_fid2;
1125         attr->la_ctime = rec->rn_time;
1126         attr->la_mtime = rec->rn_time;
1127         /* rename_tgt contains the mode already */
1128         attr->la_mode = rec->rn_mode;
1129         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1130
1131         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1132                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1133                                  req_capsule_client_get(pill, &RMF_CAPA1));
1134         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1135                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1136                                  req_capsule_client_get(pill, &RMF_CAPA2));
1137
1138         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1139         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1140         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1141                 RETURN(-EFAULT);
1142         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1143         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1144         LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1145
1146         if (rec->rn_bias & MDS_VTX_BYPASS)
1147                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1148         else
1149                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1150
1151         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1152
1153         rc = mdt_dlmreq_unpack(info);
1154         RETURN(rc);
1155 }
1156
1157 /*
1158  * please see comment above LOV_MAGIC_V1_DEF
1159  */
1160 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1161 {
1162         struct mdt_reint_record *rr = &info->mti_rr;
1163         struct lov_user_md_v1   *v1;
1164
1165         v1 = (void *)rr->rr_eadata;
1166         LASSERT(v1);
1167
1168         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1169                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1170                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1171                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1172                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1173                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1174                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1175                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1176                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1177                 }
1178         }
1179 }
1180
1181 static int mdt_open_unpack(struct mdt_thread_info *info)
1182 {
1183         struct lu_ucred         *uc = mdt_ucred(info);
1184         struct mdt_rec_create   *rec;
1185         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1186         struct req_capsule      *pill = info->mti_pill;
1187         struct mdt_reint_record *rr   = &info->mti_rr;
1188         struct ptlrpc_request   *req  = mdt_info_req(info);
1189         struct md_op_spec       *sp   = &info->mti_spec;
1190         ENTRY;
1191
1192         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1193         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1194         if (rec == NULL)
1195                 RETURN(-EFAULT);
1196
1197         /* This prior initialization is needed for old_init_ucred_reint() */
1198         uc->uc_fsuid = rec->cr_fsuid;
1199         uc->uc_fsgid = rec->cr_fsgid;
1200         uc->uc_cap   = rec->cr_cap;
1201         uc->uc_suppgids[0] = rec->cr_suppgid1;
1202         uc->uc_suppgids[1] = rec->cr_suppgid2;
1203         uc->uc_umask = rec->cr_umask;
1204
1205         rr->rr_fid1   = &rec->cr_fid1;
1206         rr->rr_fid2   = &rec->cr_fid2;
1207         rr->rr_handle = &rec->cr_old_handle;
1208         attr->la_mode = rec->cr_mode;
1209         attr->la_rdev  = rec->cr_rdev;
1210         attr->la_uid   = rec->cr_fsuid;
1211         attr->la_gid   = rec->cr_fsgid;
1212         attr->la_ctime = rec->cr_time;
1213         attr->la_mtime = rec->cr_time;
1214         attr->la_atime = rec->cr_time;
1215         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1216                          LA_CTIME | LA_MTIME | LA_ATIME;
1217         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1218         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1219         /* Do not trigger ASSERTION if client miss to set such flags. */
1220         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1221                 RETURN(-EPROTO);
1222         info->mti_replayepoch = rec->cr_ioepoch;
1223
1224         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1225
1226         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1227                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1228                                  req_capsule_client_get(pill, &RMF_CAPA1));
1229         if (req_is_replay(req) &&
1230             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1231 #if 0
1232                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1233                                  req_capsule_client_get(pill, &RMF_CAPA2));
1234 #else
1235                 /*
1236                  * FIXME: capa in replay open request might have expired,
1237                  * bypass capa check. Security hole?
1238                  */
1239                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1240                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1241 #endif
1242         }
1243
1244         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1245         if (rr->rr_name == NULL)
1246                 RETURN(-EFAULT);
1247         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1248
1249         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1250                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1251                                                         RCL_CLIENT);
1252                 if (rr->rr_eadatalen > 0) {
1253                         rr->rr_eadata = req_capsule_client_get(pill,
1254                                                                &RMF_EADATA);
1255                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1256                         sp->u.sp_ea.eadata = rr->rr_eadata;
1257                         sp->no_create = !!req_is_replay(req);
1258                         mdt_fix_lov_magic(info);
1259                 }
1260
1261                 /*
1262                  * Client default md_size may be 0 right after client start,
1263                  * until all osc are connected, set here just some reasonable
1264                  * value to prevent misbehavior.
1265                  */
1266                 if (rr->rr_eadatalen == 0 &&
1267                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1268                         rr->rr_eadatalen = MIN_MD_SIZE;
1269         }
1270
1271         RETURN(0);
1272 }
1273
1274 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1275 {
1276         struct mdt_reint_record   *rr   = &info->mti_rr;
1277         struct lu_ucred           *uc   = mdt_ucred(info);
1278         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1279         struct req_capsule        *pill = info->mti_pill;
1280         struct mdt_rec_setxattr   *rec;
1281         ENTRY;
1282
1283
1284         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1285                          sizeof(struct mdt_rec_reint));
1286
1287         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1288         if (rec == NULL)
1289                 RETURN(-EFAULT);
1290
1291         /* This prior initialization is needed for old_init_ucred_reint() */
1292         uc->uc_fsuid  = rec->sx_fsuid;
1293         uc->uc_fsgid  = rec->sx_fsgid;
1294         uc->uc_cap    = rec->sx_cap;
1295         uc->uc_suppgids[0] = rec->sx_suppgid1;
1296         uc->uc_suppgids[1] = -1;
1297
1298         rr->rr_opcode = rec->sx_opcode;
1299         rr->rr_fid1   = &rec->sx_fid;
1300         attr->la_valid = rec->sx_valid;
1301         attr->la_ctime = rec->sx_time;
1302         attr->la_size = rec->sx_size;
1303         attr->la_flags = rec->sx_flags;
1304
1305         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1306                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1307                                  req_capsule_client_get(pill, &RMF_CAPA1));
1308         else
1309                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1310
1311         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1312         if (rr->rr_name == NULL)
1313                 RETURN(-EFAULT);
1314         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1315         LASSERT(rr->rr_namelen > 0);
1316
1317         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1318                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1319                                                         RCL_CLIENT);
1320                 if (rr->rr_eadatalen > 0) {
1321                         rr->rr_eadata = req_capsule_client_get(pill,
1322                                                                &RMF_EADATA);
1323                         if (rr->rr_eadata == NULL)
1324                                 RETURN(-EFAULT);
1325                 } else {
1326                         rr->rr_eadata = NULL;
1327                 }
1328         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1329                 CDEBUG(D_INFO, "no xattr data supplied\n");
1330                 RETURN(-EFAULT);
1331         }
1332
1333         RETURN(0);
1334 }
1335
1336
1337 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1338
1339 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1340         [REINT_SETATTR]  = mdt_setattr_unpack,
1341         [REINT_CREATE]   = mdt_create_unpack,
1342         [REINT_LINK]     = mdt_link_unpack,
1343         [REINT_UNLINK]   = mdt_unlink_unpack,
1344         [REINT_RENAME]   = mdt_rename_unpack,
1345         [REINT_OPEN]     = mdt_open_unpack,
1346         [REINT_SETXATTR] = mdt_setxattr_unpack,
1347         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1348 };
1349
1350 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1351 {
1352         int rc;
1353         ENTRY;
1354
1355         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1356         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1357                 info->mti_rr.rr_opcode = op;
1358                 rc = mdt_reint_unpackers[op](info);
1359         } else {
1360                 CERROR("Unexpected opcode %d\n", op);
1361                 rc = -EFAULT;
1362         }
1363         RETURN(rc);
1364 }