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, 2013, 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                         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 = 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                         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             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 547                 return;
 548
 549         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 550         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 551                 struct ost_id   oi;
 552                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 553                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 554                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 555         }
 556 }
 557
 558 /* Shrink and/or grow reply buffers */
 559 int mdt_fix_reply(struct mdt_thread_info *info)
 560 {
 561         struct req_capsule *pill = info->mti_pill;
 562         struct mdt_body    *body;
 563         int                md_size, md_packed = 0;
 564         int                acl_size;
 565         int                rc = 0;
 566         ENTRY;
 567
 568         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 569         LASSERT(body != NULL);
 570
 571         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 572                 md_size = body->eadatasize;
 573         else
 574                 md_size = 0;
 575
 576         acl_size = body->aclsize;
 577
 578         /* this replay - not send info to client */
 579         if (info->mti_spec.no_create) {
 580                 md_size = 0;
 581                 acl_size = 0;
 582         }
 583
 584         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 585                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 586                         md_size, acl_size,
 587                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 588                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 589 /*
 590             &RMF_MDT_BODY,
 591             &RMF_MDT_MD,
 592             &RMF_ACL, or &RMF_LOGCOOKIES
 593 (optional)  &RMF_CAPA1,
 594 (optional)  &RMF_CAPA2,
 595 (optional)  something else
 596 */
 597
 598         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 599          * buffers before growing it */
 600         if (info->mti_big_lmm_used) {
 601                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 602                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 603                                                  RCL_SERVER);
 604                 LASSERT(md_packed > 0);
 605                 /* buffer must be allocated separately */
 606                 LASSERT(info->mti_attr.ma_lmm !=
 607                         req_capsule_server_get(pill, &RMF_MDT_MD));
 608                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 609                 /* free big lmm if md_size is not needed */
 610                 if (md_size == 0)
 611                         info->mti_big_lmm_used = 0;
 612         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 613                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 614         }
 615
 616         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 617                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 618         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 619                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 620                                    acl_size, RCL_SERVER);
 621
 622         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 623             !(body->valid & OBD_MD_FLMDSCAPA))
 624                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 625
 626         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 627             !(body->valid & OBD_MD_FLOSSCAPA))
 628                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 629
 630         /*
 631          * Some more field should be shrinked if needed.
 632          * This should be done by those who added fields to reply message.
 633          */
 634
 635         /* Grow MD buffer if needed finally */
 636         if (info->mti_big_lmm_used) {
 637                 void *lmm;
 638
 639                 LASSERT(md_size > md_packed);
 640                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 641                        md_size - md_packed);
 642                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 643                 if (rc) {
 644                         /* we can't answer with proper LOV EA, drop flags,
 645                          * the rc is also returned so this request is
 646                          * considered as failed */
 647                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 648                         /* don't return transno along with error */
 649                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 650                 } else {
 651                         /* now we need to pack right LOV/LMV EA */
 652                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 653                         if (info->mti_attr.ma_valid & MA_LOV) {
 654                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 655                                                              RCL_SERVER) ==
 656                                                 info->mti_attr.ma_lmm_size);
 657                                 memcpy(lmm, info->mti_attr.ma_lmm,
 658                                        info->mti_attr.ma_lmm_size);
 659                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 660                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 661                                                              RCL_SERVER) ==
 662                                                 info->mti_attr.ma_lmv_size);
 663                                 memcpy(lmm, info->mti_attr.ma_lmv,
 664                                        info->mti_attr.ma_lmv_size);
 665                         }
 666                 }
 667                 /* update mdt_max_mdsize so clients will be aware about that */
 668                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 669                         info->mti_mdt->mdt_max_mdsize =
 670                                                     info->mti_attr.ma_lmm_size;
 671                 info->mti_big_lmm_used = 0;
 672         }
 673         RETURN(rc);
 674 }
 675
 676
 677 /* if object is dying, pack the lov/llog data,
 678  * parameter info->mti_attr should be valid at this point! */
 679 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 680                            const struct md_attr *ma)
 681 {
 682         struct mdt_body       *repbody;
 683         const struct lu_attr *la = &ma->ma_attr;
 684         int rc;
 685         ENTRY;
 686
 687         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 688         LASSERT(repbody != NULL);
 689
 690         if (ma->ma_valid & MA_INODE)
 691                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 692
 693         if (ma->ma_valid & MA_LOV) {
 694                 CERROR("No need in LOV EA upon unlink\n");
 695                 dump_stack();
 696         }
 697         repbody->eadatasize = 0;
 698
 699         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 700                 repbody->aclsize = ma->ma_cookie_size;
 701                 repbody->valid |= OBD_MD_FLCOOKIE;
 702         }
 703
 704         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 705             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 706             repbody->valid & OBD_MD_FLEASIZE) {
 707                 struct lustre_capa *capa;
 708
 709                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 710                 LASSERT(capa);
 711                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 712                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 713                 if (rc)
 714                         RETURN(rc);
 715
 716                 repbody->valid |= OBD_MD_FLOSSCAPA;
 717         }
 718
 719         RETURN(0);
 720 }
 721
 722 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 723                                   struct md_attr *ma)
 724 {
 725         __u64 out;
 726
 727         out = 0;
 728         if (in & MDS_ATTR_MODE)
 729                 out |= LA_MODE;
 730         if (in & MDS_ATTR_UID)
 731                 out |= LA_UID;
 732         if (in & MDS_ATTR_GID)
 733                 out |= LA_GID;
 734         if (in & MDS_ATTR_SIZE)
 735                 out |= LA_SIZE;
 736         if (in & MDS_ATTR_BLOCKS)
 737                 out |= LA_BLOCKS;
 738         if (in & MDS_ATTR_ATIME_SET)
 739                 out |= LA_ATIME;
 740         if (in & MDS_ATTR_CTIME_SET)
 741                 out |= LA_CTIME;
 742         if (in & MDS_ATTR_MTIME_SET)
 743                 out |= LA_MTIME;
 744         if (in & MDS_ATTR_ATTR_FLAG)
 745                 out |= LA_FLAGS;
 746         if (in & MDS_ATTR_KILL_SUID)
 747                 out |= LA_KILL_SUID;
 748         if (in & MDS_ATTR_KILL_SGID)
 749                 out |= LA_KILL_SGID;
 750
 751         if (in & MDS_ATTR_FROM_OPEN)
 752                 rr->rr_flags |= MRF_OPEN_TRUNC;
 753         if (in & MDS_OPEN_OWNEROVERRIDE)
 754                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 755         if (in & MDS_ATTR_FORCE)
 756                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 757
 758         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 759                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 760                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 761                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 762                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 763                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 764         if (in != 0)
 765                 CERROR("Unknown attr bits: "LPX64"\n", in);
 766         return out;
 767 }
 768
 769 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 770                       const struct lu_fid *fid, struct lustre_capa *capa)
 771 {
 772         struct lu_capainfo *lci;
 773
 774         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 775         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 776             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 777                 return;
 778
 779         lci = lu_capainfo_get(info->mti_env);
 780         LASSERT(lci);
 781         lci->lci_fid[offset]  = *fid;
 782         lci->lci_capa[offset] = capa;
 783 }
 784
 785 #ifdef DEBUG_CAPA
 786 void mdt_dump_capainfo(struct mdt_thread_info *info)
 787 {
 788         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 789         int i;
 790
 791         if (lci == NULL)
 792                 return;
 793
 794         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 795                 if (lci->lci_capa[i] == NULL) {
 796                         CERROR("no capa for index %d "DFID"\n",
 797                                i, PFID(&lci->lci_fid[i]));
 798                         continue;
 799                 }
 800                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 801                         CERROR("bypass for index %d "DFID"\n",
 802                                i, PFID(&lci->lci_fid[i]));
 803                         continue;
 804                 }
 805                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 806         }
 807 }
 808 #endif /* DEBUG_CAPA */
 809
 810 /* unpacking */
 811
 812 int mdt_name_unpack(struct req_capsule *pill,
 813                     const struct req_msg_field *field,
 814                     struct lu_name *ln,
 815                     enum mdt_name_flags flags)
 816 {
 817         ln->ln_name = req_capsule_client_get(pill, field);
 818         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 819
 820         if (!lu_name_is_valid(ln)) {
 821                 ln->ln_name = NULL;
 822                 ln->ln_namelen = 0;
 823
 824                 return -EPROTO;
 825         }
 826
 827         if ((flags & MNF_FIX_ANON) &&
 828             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 829                 /* Newer (3.x) kernels use a name of "/" for the
 830                  * "anonymous" disconnected dentries from NFS
 831                  * filehandle conversion. See d_obtain_alias(). */
 832                 ln->ln_name = NULL;
 833                 ln->ln_namelen = 0;
 834         }
 835
 836         return 0;
 837 }
 838
 839 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 840 {
 841         struct lu_ucred         *uc  = mdt_ucred(info);
 842         struct md_attr          *ma = &info->mti_attr;
 843         struct lu_attr          *la = &ma->ma_attr;
 844         struct req_capsule      *pill = info->mti_pill;
 845         struct mdt_reint_record *rr = &info->mti_rr;
 846         struct mdt_rec_setattr  *rec;
 847         ENTRY;
 848
 849         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 850         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 851         if (rec == NULL)
 852                 RETURN(-EFAULT);
 853
 854         /* This prior initialization is needed for old_init_ucred_reint() */
 855         uc->uc_fsuid = rec->sa_fsuid;
 856         uc->uc_fsgid = rec->sa_fsgid;
 857         uc->uc_cap   = rec->sa_cap;
 858         uc->uc_suppgids[0] = rec->sa_suppgid;
 859         uc->uc_suppgids[1] = -1;
 860
 861         rr->rr_fid1 = &rec->sa_fid;
 862         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 863         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 864          *  the client does not have OBD_CONNECT_FULL20, convert it
 865          *  to LA_xTIME. LU-3036 */
 866         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 867                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 868                      (rec->sa_valid & MDS_ATTR_ATIME))
 869                         la->la_valid |= LA_ATIME;
 870                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 871                      (rec->sa_valid & MDS_ATTR_MTIME))
 872                         la->la_valid |= LA_MTIME;
 873                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 874                      (rec->sa_valid & MDS_ATTR_CTIME))
 875                         la->la_valid |= LA_CTIME;
 876         }
 877         la->la_mode  = rec->sa_mode;
 878         la->la_flags = rec->sa_attr_flags;
 879         la->la_uid   = rec->sa_uid;
 880         la->la_gid   = rec->sa_gid;
 881         la->la_size  = rec->sa_size;
 882         la->la_blocks = rec->sa_blocks;
 883         la->la_ctime = rec->sa_ctime;
 884         la->la_atime = rec->sa_atime;
 885         la->la_mtime = rec->sa_mtime;
 886         ma->ma_valid = MA_INODE;
 887
 888         if (rec->sa_bias & MDS_DATA_MODIFIED)
 889                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 890         else
 891                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 892
 893         if (rec->sa_bias & MDS_HSM_RELEASE)
 894                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 895         else
 896                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 897
 898         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 899                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 900                                  req_capsule_client_get(pill, &RMF_CAPA1));
 901
 902         RETURN(0);
 903 }
 904
 905 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 906 {
 907         struct req_capsule *pill = info->mti_pill;
 908         ENTRY;
 909
 910         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 911                 info->mti_ioepoch =
 912                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 913         else
 914                 info->mti_ioepoch = NULL;
 915         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 916 }
 917
 918 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 919         struct req_capsule      *pill = info->mti_pill;
 920
 921         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 922                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 923                 if (info->mti_dlm_req == NULL)
 924                         RETURN(-EFAULT);
 925         }
 926
 927         RETURN(0);
 928 }
 929
 930 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 931 {
 932         struct mdt_reint_record *rr = &info->mti_rr;
 933         struct md_attr          *ma = &info->mti_attr;
 934         struct req_capsule      *pill = info->mti_pill;
 935         int rc;
 936         ENTRY;
 937
 938         rc = mdt_setattr_unpack_rec(info);
 939         if (rc)
 940                 RETURN(rc);
 941
 942         /* Epoch may be absent */
 943         mdt_ioepoch_unpack(info);
 944
 945         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 946                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 947                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 948                                                         RCL_CLIENT);
 949                 ma->ma_lmm_size = rr->rr_eadatalen;
 950                 if (ma->ma_lmm_size > 0) {
 951                         ma->ma_lmm = (void *)rr->rr_eadata;
 952                         ma->ma_valid |= MA_LOV;
 953                 }
 954         }
 955
 956         rc = mdt_dlmreq_unpack(info);
 957         RETURN(rc);
 958 }
 959
 960 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
 961 {
 962         struct md_attr          *ma = &info->mti_attr;
 963         struct req_capsule      *pill = info->mti_pill;
 964         ENTRY;
 965
 966         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
 967                 RETURN(0);
 968
 969         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
 970
 971         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
 972             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
 973                 RETURN(-EFAULT);
 974
 975         RETURN(0);
 976 }
 977
 978 int mdt_close_unpack(struct mdt_thread_info *info)
 979 {
 980         int rc;
 981         ENTRY;
 982
 983         rc = mdt_ioepoch_unpack(info);
 984         if (rc)
 985                 RETURN(rc);
 986
 987         rc = mdt_setattr_unpack_rec(info);
 988         if (rc)
 989                 RETURN(rc);
 990
 991         rc = mdt_hsm_release_unpack(info);
 992         if (rc)
 993                 RETURN(rc);
 994
 995         RETURN(mdt_init_ucred_reint(info));
 996 }
 997
 998 static int mdt_create_unpack(struct mdt_thread_info *info)
 999 {
1000         struct lu_ucred         *uc  = mdt_ucred(info);
1001         struct mdt_rec_create   *rec;
1002         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1003         struct mdt_reint_record *rr = &info->mti_rr;
1004         struct req_capsule      *pill = info->mti_pill;
1005         struct md_op_spec       *sp = &info->mti_spec;
1006         int rc;
1007         ENTRY;
1008
1009         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1010         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1011         if (rec == NULL)
1012                 RETURN(-EFAULT);
1013
1014         /* This prior initialization is needed for old_init_ucred_reint() */
1015         uc->uc_fsuid = rec->cr_fsuid;
1016         uc->uc_fsgid = rec->cr_fsgid;
1017         uc->uc_cap   = rec->cr_cap;
1018         uc->uc_suppgids[0] = rec->cr_suppgid1;
1019         uc->uc_suppgids[1] = -1;
1020         uc->uc_umask = rec->cr_umask;
1021
1022         rr->rr_fid1 = &rec->cr_fid1;
1023         rr->rr_fid2 = &rec->cr_fid2;
1024         attr->la_mode = rec->cr_mode;
1025         attr->la_rdev  = rec->cr_rdev;
1026         attr->la_uid   = rec->cr_fsuid;
1027         attr->la_gid   = rec->cr_fsgid;
1028         attr->la_ctime = rec->cr_time;
1029         attr->la_mtime = rec->cr_time;
1030         attr->la_atime = rec->cr_time;
1031         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1032                          LA_CTIME | LA_MTIME | LA_ATIME;
1033         memset(&sp->u, 0, sizeof(sp->u));
1034         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1035
1036         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1037                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1038                                  req_capsule_client_get(pill, &RMF_CAPA1));
1039         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1040
1041         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1042         if (rc < 0)
1043                 RETURN(rc);
1044
1045         if (S_ISLNK(attr->la_mode)) {
1046                 const char *tgt = NULL;
1047
1048                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1049                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1050                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1051                         sp->u.sp_symname = tgt;
1052                 }
1053                 if (tgt == NULL)
1054                         RETURN(-EFAULT);
1055         } else {
1056                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1057                 if (S_ISDIR(attr->la_mode) &&
1058                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1059                         sp->u.sp_ea.eadata =
1060                                 req_capsule_client_get(pill, &RMF_EADATA);
1061                         sp->u.sp_ea.eadatalen =
1062                                 req_capsule_get_size(pill, &RMF_EADATA,
1063                                                      RCL_CLIENT);
1064                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1065                 }
1066         }
1067
1068         rc = mdt_dlmreq_unpack(info);
1069         RETURN(rc);
1070 }
1071
1072 static int mdt_link_unpack(struct mdt_thread_info *info)
1073 {
1074         struct lu_ucred         *uc  = mdt_ucred(info);
1075         struct mdt_rec_link     *rec;
1076         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1077         struct mdt_reint_record *rr = &info->mti_rr;
1078         struct req_capsule      *pill = info->mti_pill;
1079         int rc;
1080         ENTRY;
1081
1082         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1083         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1084         if (rec == NULL)
1085                 RETURN(-EFAULT);
1086
1087         /* This prior initialization is needed for old_init_ucred_reint() */
1088         uc->uc_fsuid = rec->lk_fsuid;
1089         uc->uc_fsgid = rec->lk_fsgid;
1090         uc->uc_cap   = rec->lk_cap;
1091         uc->uc_suppgids[0] = rec->lk_suppgid1;
1092         uc->uc_suppgids[1] = rec->lk_suppgid2;
1093
1094         attr->la_uid = rec->lk_fsuid;
1095         attr->la_gid = rec->lk_fsgid;
1096         rr->rr_fid1 = &rec->lk_fid1;
1097         rr->rr_fid2 = &rec->lk_fid2;
1098         attr->la_ctime = rec->lk_time;
1099         attr->la_mtime = rec->lk_time;
1100         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1101
1102         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1103                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1104                                  req_capsule_client_get(pill, &RMF_CAPA1));
1105         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1106                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1107                                  req_capsule_client_get(pill, &RMF_CAPA2));
1108
1109         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1110         if (rc < 0)
1111                 RETURN(rc);
1112
1113         rc = mdt_dlmreq_unpack(info);
1114
1115         RETURN(rc);
1116 }
1117
1118 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1119 {
1120         struct lu_ucred         *uc  = mdt_ucred(info);
1121         struct mdt_rec_unlink   *rec;
1122         struct md_attr          *ma = &info->mti_attr;
1123         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1124         struct mdt_reint_record *rr = &info->mti_rr;
1125         struct req_capsule      *pill = info->mti_pill;
1126         int rc;
1127         ENTRY;
1128
1129         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1130         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1131         if (rec == NULL)
1132                 RETURN(-EFAULT);
1133
1134         /* This prior initialization is needed for old_init_ucred_reint() */
1135         uc->uc_fsuid = rec->ul_fsuid;
1136         uc->uc_fsgid = rec->ul_fsgid;
1137         uc->uc_cap   = rec->ul_cap;
1138         uc->uc_suppgids[0] = rec->ul_suppgid1;
1139         uc->uc_suppgids[1] = -1;
1140
1141         attr->la_uid = rec->ul_fsuid;
1142         attr->la_gid = rec->ul_fsgid;
1143         rr->rr_fid1 = &rec->ul_fid1;
1144         rr->rr_fid2 = &rec->ul_fid2;
1145         attr->la_ctime = rec->ul_time;
1146         attr->la_mtime = rec->ul_time;
1147         attr->la_mode  = rec->ul_mode;
1148         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1149
1150         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1151                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1152                                  req_capsule_client_get(pill, &RMF_CAPA1));
1153
1154         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1155         if (rc < 0)
1156                 RETURN(rc);
1157
1158         if (rec->ul_bias & MDS_VTX_BYPASS)
1159                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1160         else
1161                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1162
1163         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1164
1165         rc = mdt_dlmreq_unpack(info);
1166         RETURN(rc);
1167 }
1168
1169 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1170 {
1171         info->mti_spec.sp_rm_entry = 1;
1172         return mdt_unlink_unpack(info);
1173 }
1174
1175 static int mdt_rename_unpack(struct mdt_thread_info *info)
1176 {
1177         struct lu_ucred         *uc = mdt_ucred(info);
1178         struct mdt_rec_rename   *rec;
1179         struct md_attr          *ma = &info->mti_attr;
1180         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1181         struct mdt_reint_record *rr = &info->mti_rr;
1182         struct req_capsule      *pill = info->mti_pill;
1183         int rc;
1184         ENTRY;
1185
1186         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1187         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1188         if (rec == NULL)
1189                 RETURN(-EFAULT);
1190
1191         /* This prior initialization is needed for old_init_ucred_reint() */
1192         uc->uc_fsuid = rec->rn_fsuid;
1193         uc->uc_fsgid = rec->rn_fsgid;
1194         uc->uc_cap   = rec->rn_cap;
1195         uc->uc_suppgids[0] = rec->rn_suppgid1;
1196         uc->uc_suppgids[1] = rec->rn_suppgid2;
1197
1198         attr->la_uid = rec->rn_fsuid;
1199         attr->la_gid = rec->rn_fsgid;
1200         rr->rr_fid1 = &rec->rn_fid1;
1201         rr->rr_fid2 = &rec->rn_fid2;
1202         attr->la_ctime = rec->rn_time;
1203         attr->la_mtime = rec->rn_time;
1204         /* rename_tgt contains the mode already */
1205         attr->la_mode = rec->rn_mode;
1206         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1207
1208         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1209                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1210                                  req_capsule_client_get(pill, &RMF_CAPA1));
1211         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1212                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1213                                  req_capsule_client_get(pill, &RMF_CAPA2));
1214
1215         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1216         if (rc < 0)
1217                 RETURN(rc);
1218
1219         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1220         if (rc < 0)
1221                 RETURN(rc);
1222
1223         if (rec->rn_bias & MDS_VTX_BYPASS)
1224                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1225         else
1226                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1227
1228         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1229
1230         rc = mdt_dlmreq_unpack(info);
1231         RETURN(rc);
1232 }
1233
1234 /*
1235  * please see comment above LOV_MAGIC_V1_DEF
1236  */
1237 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1238 {
1239         struct mdt_reint_record *rr = &info->mti_rr;
1240         struct lov_user_md_v1   *v1;
1241
1242         v1 = (void *)rr->rr_eadata;
1243         LASSERT(v1);
1244
1245         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1246                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1247                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1248                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1249                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1250                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1251                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1252                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1253                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1254                 }
1255         }
1256 }
1257
1258 static int mdt_open_unpack(struct mdt_thread_info *info)
1259 {
1260         struct lu_ucred         *uc = mdt_ucred(info);
1261         struct mdt_rec_create   *rec;
1262         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1263         struct req_capsule      *pill = info->mti_pill;
1264         struct mdt_reint_record *rr   = &info->mti_rr;
1265         struct ptlrpc_request   *req  = mdt_info_req(info);
1266         struct md_op_spec       *sp   = &info->mti_spec;
1267         ENTRY;
1268
1269         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1270         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1271         if (rec == NULL)
1272                 RETURN(-EFAULT);
1273
1274         /* This prior initialization is needed for old_init_ucred_reint() */
1275         uc->uc_fsuid = rec->cr_fsuid;
1276         uc->uc_fsgid = rec->cr_fsgid;
1277         uc->uc_cap   = rec->cr_cap;
1278         uc->uc_suppgids[0] = rec->cr_suppgid1;
1279         uc->uc_suppgids[1] = rec->cr_suppgid2;
1280         uc->uc_umask = rec->cr_umask;
1281
1282         rr->rr_fid1   = &rec->cr_fid1;
1283         rr->rr_fid2   = &rec->cr_fid2;
1284         rr->rr_handle = &rec->cr_old_handle;
1285         attr->la_mode = rec->cr_mode;
1286         attr->la_rdev  = rec->cr_rdev;
1287         attr->la_uid   = rec->cr_fsuid;
1288         attr->la_gid   = rec->cr_fsgid;
1289         attr->la_ctime = rec->cr_time;
1290         attr->la_mtime = rec->cr_time;
1291         attr->la_atime = rec->cr_time;
1292         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1293                          LA_CTIME | LA_MTIME | LA_ATIME;
1294         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1295         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1296         /* Do not trigger ASSERTION if client miss to set such flags. */
1297         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1298                 RETURN(-EPROTO);
1299         info->mti_replayepoch = rec->cr_ioepoch;
1300
1301         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1302
1303         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1304                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1305                                  req_capsule_client_get(pill, &RMF_CAPA1));
1306         if (req_is_replay(req) &&
1307             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1308 #if 0
1309                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1310                                  req_capsule_client_get(pill, &RMF_CAPA2));
1311 #else
1312                 /*
1313                  * FIXME: capa in replay open request might have expired,
1314                  * bypass capa check. Security hole?
1315                  */
1316                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1317                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1318 #endif
1319         }
1320
1321         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1322
1323         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1324                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1325                                                         RCL_CLIENT);
1326                 if (rr->rr_eadatalen > 0) {
1327                         rr->rr_eadata = req_capsule_client_get(pill,
1328                                                                &RMF_EADATA);
1329                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1330                         sp->u.sp_ea.eadata = rr->rr_eadata;
1331                         sp->no_create = !!req_is_replay(req);
1332                         mdt_fix_lov_magic(info);
1333                 }
1334
1335                 /*
1336                  * Client default md_size may be 0 right after client start,
1337                  * until all osc are connected, set here just some reasonable
1338                  * value to prevent misbehavior.
1339                  */
1340                 if (rr->rr_eadatalen == 0 &&
1341                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1342                         rr->rr_eadatalen = MIN_MD_SIZE;
1343         }
1344
1345         RETURN(0);
1346 }
1347
1348 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1349 {
1350         struct mdt_reint_record *rr     = &info->mti_rr;
1351         struct lu_ucred         *uc     = mdt_ucred(info);
1352         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1353         struct req_capsule      *pill   = info->mti_pill;
1354         struct mdt_rec_setxattr *rec;
1355         int                      rc;
1356         ENTRY;
1357
1358
1359         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1360                          sizeof(struct mdt_rec_reint));
1361
1362         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1363         if (rec == NULL)
1364                 RETURN(-EFAULT);
1365
1366         /* This prior initialization is needed for old_init_ucred_reint() */
1367         uc->uc_fsuid  = rec->sx_fsuid;
1368         uc->uc_fsgid  = rec->sx_fsgid;
1369         uc->uc_cap    = rec->sx_cap;
1370         uc->uc_suppgids[0] = rec->sx_suppgid1;
1371         uc->uc_suppgids[1] = -1;
1372
1373         rr->rr_opcode = rec->sx_opcode;
1374         rr->rr_fid1   = &rec->sx_fid;
1375         attr->la_valid = rec->sx_valid;
1376         attr->la_ctime = rec->sx_time;
1377         attr->la_size = rec->sx_size;
1378         attr->la_flags = rec->sx_flags;
1379
1380         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1381                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1382                                  req_capsule_client_get(pill, &RMF_CAPA1));
1383         else
1384                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1385
1386         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1387         if (rc < 0)
1388                 RETURN(rc);
1389
1390         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1391                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1392                                                         RCL_CLIENT);
1393                 if (rr->rr_eadatalen > 0) {
1394                         rr->rr_eadata = req_capsule_client_get(pill,
1395                                                                &RMF_EADATA);
1396                         if (rr->rr_eadata == NULL)
1397                                 RETURN(-EFAULT);
1398                 } else {
1399                         rr->rr_eadata = NULL;
1400                 }
1401         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1402                 CDEBUG(D_INFO, "no xattr data supplied\n");
1403                 RETURN(-EFAULT);
1404         }
1405
1406         if (mdt_dlmreq_unpack(info) < 0)
1407                 RETURN(-EPROTO);
1408
1409         RETURN(0);
1410 }
1411
1412
1413 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1414
1415 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1416         [REINT_SETATTR]  = mdt_setattr_unpack,
1417         [REINT_CREATE]   = mdt_create_unpack,
1418         [REINT_LINK]     = mdt_link_unpack,
1419         [REINT_UNLINK]   = mdt_unlink_unpack,
1420         [REINT_RENAME]   = mdt_rename_unpack,
1421         [REINT_OPEN]     = mdt_open_unpack,
1422         [REINT_SETXATTR] = mdt_setxattr_unpack,
1423         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1424 };
1425
1426 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1427 {
1428         int rc;
1429         ENTRY;
1430
1431         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1432         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1433                 info->mti_rr.rr_opcode = op;
1434                 rc = mdt_reint_unpackers[op](info);
1435         } else {
1436                 CERROR("Unexpected opcode %d\n", op);
1437                 rc = -EFAULT;
1438         }
1439         RETURN(rc);
1440 }