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