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                         cfs_put_group_info(uc->uc_ginfo);
  71                         uc->uc_ginfo = NULL;
  72                 }
  73                 if (uc->uc_identity) {
  74                         mdt_identity_put(mdt->mdt_identity_cache,
  75                                          uc->uc_identity);
  76                         uc->uc_identity = NULL;
  77                 }
  78                 uc->uc_valid = UCRED_INIT;
  79         }
  80 }
  81
  82 static int match_nosquash_list(struct rw_semaphore *sem,
  83                                cfs_list_t *nidlist,
  84                                lnet_nid_t peernid)
  85 {
  86         int rc;
  87         ENTRY;
  88         down_read(sem);
  89         rc = cfs_match_nid(peernid, nidlist);
  90         up_read(sem);
  91         RETURN(rc);
  92 }
  93
  94 /* root_squash for inter-MDS operations */
  95 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  96 {
  97         struct lu_ucred *ucred = mdt_ucred(info);
  98         ENTRY;
  99
 100         LASSERT(ucred != NULL);
 101         if (!info->mti_mdt->mdt_squash_uid || ucred->uc_fsuid)
 102                 RETURN(0);
 103
 104         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 105                                 &info->mti_mdt->mdt_nosquash_nids,
 106                                 peernid)) {
 107                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 108                        libcfs_nid2str(peernid));
 109                 RETURN(0);
 110         }
 111
 112         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 113                libcfs_nid2str(peernid),
 114                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 115                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 116                0);
 117
 118         ucred->uc_fsuid = info->mti_mdt->mdt_squash_uid;
 119         ucred->uc_fsgid = info->mti_mdt->mdt_squash_gid;
 120         ucred->uc_cap = 0;
 121         ucred->uc_suppgids[0] = -1;
 122         ucred->uc_suppgids[1] = -1;
 123
 124         RETURN(0);
 125 }
 126
 127 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 128                           void *buf)
 129 {
 130         struct ptlrpc_request   *req = mdt_info_req(info);
 131         struct mdt_device       *mdt = info->mti_mdt;
 132         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 133         struct lu_ucred         *ucred = mdt_ucred(info);
 134         lnet_nid_t               peernid = req->rq_peer.nid;
 135         __u32                    perm = 0;
 136         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 137         int                      setuid;
 138         int                      setgid;
 139         int                      rc = 0;
 140
 141         ENTRY;
 142
 143         LASSERT(req->rq_auth_gss);
 144         LASSERT(!req->rq_auth_usr_mdt);
 145         LASSERT(req->rq_user_desc);
 146         LASSERT(ucred != NULL);
 147
 148         ucred->uc_valid = UCRED_INVALID;
 149
 150         ucred->uc_o_uid   = pud->pud_uid;
 151         ucred->uc_o_gid   = pud->pud_gid;
 152         ucred->uc_o_fsuid = pud->pud_fsuid;
 153         ucred->uc_o_fsgid = pud->pud_fsgid;
 154
 155         if (type == BODY_INIT) {
 156                 struct mdt_body *body = (struct mdt_body *)buf;
 157
 158                 ucred->uc_suppgids[0] = body->suppgid;
 159                 ucred->uc_suppgids[1] = -1;
 160         }
 161
 162         /* sanity check: we expect the uid which client claimed is true */
 163         if (remote) {
 164                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 165                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 166                         RETURN(-EACCES);
 167                 }
 168
 169                 if (ptlrpc_user_desc_do_idmap(req, pud))
 170                         RETURN(-EACCES);
 171
 172                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 173                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 174                                "while client claims %u:%u/%u:%u\n",
 175                                libcfs_nid2str(peernid), req->rq_auth_uid,
 176                                req->rq_auth_mapped_uid,
 177                                pud->pud_uid, pud->pud_gid,
 178                                pud->pud_fsuid, pud->pud_fsgid);
 179                         RETURN(-EACCES);
 180                 }
 181         } else {
 182                 if (req->rq_auth_uid != pud->pud_uid) {
 183                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 184                                "while client claims %u:%u/%u:%u\n",
 185                                libcfs_nid2str(peernid), req->rq_auth_uid,
 186                                pud->pud_uid, pud->pud_gid,
 187                                pud->pud_fsuid, pud->pud_fsgid);
 188                         RETURN(-EACCES);
 189                 }
 190         }
 191
 192         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 193                 if (remote) {
 194                         CDEBUG(D_SEC, "remote client must run with identity_get "
 195                                "enabled!\n");
 196                         RETURN(-EACCES);
 197                 } else {
 198                         ucred->uc_identity = NULL;
 199                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 200                                CFS_SETGRP_PERM;
 201                 }
 202         } else {
 203                 struct md_identity *identity;
 204
 205                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 206                                             pud->pud_uid);
 207                 if (IS_ERR(identity)) {
 208                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 209                                      !remote)) {
 210                                 ucred->uc_identity = NULL;
 211                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 212                                        CFS_SETGRP_PERM;
 213                         } else {
 214                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 215                                        pud->pud_uid);
 216                                 RETURN(-EACCES);
 217                         }
 218                 } else {
 219                         ucred->uc_identity = identity;
 220                         perm = mdt_identity_get_perm(ucred->uc_identity,
 221                                                      remote, peernid);
 222                 }
 223         }
 224
 225         /* find out the setuid/setgid attempt */
 226         setuid = (pud->pud_uid != pud->pud_fsuid);
 227         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 228                   (ucred->uc_identity &&
 229                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 230
 231         /* check permission of setuid */
 232         if (setuid && !(perm & CFS_SETUID_PERM)) {
 233                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 234                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 235                 GOTO(out, rc = -EACCES);
 236         }
 237
 238         /* check permission of setgid */
 239         if (setgid && !(perm & CFS_SETGID_PERM)) {
 240                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 241                        "from %s\n", pud->pud_uid, pud->pud_gid,
 242                        pud->pud_fsuid, pud->pud_fsgid,
 243                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 244                 GOTO(out, rc = -EACCES);
 245         }
 246
 247         /*
 248          * NB: remote client not allowed to setgroups anyway.
 249          */
 250         if (!remote && perm & CFS_SETGRP_PERM) {
 251                 if (pud->pud_ngroups) {
 252                         /* setgroups for local client */
 253                         ucred->uc_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 254                         if (!ucred->uc_ginfo) {
 255                                 CERROR("failed to alloc %d groups\n",
 256                                        pud->pud_ngroups);
 257                                 GOTO(out, rc = -ENOMEM);
 258                         }
 259
 260                         lustre_groups_from_list(ucred->uc_ginfo,
 261                                                 pud->pud_groups);
 262                         lustre_groups_sort(ucred->uc_ginfo);
 263                 } else {
 264                         ucred->uc_ginfo = NULL;
 265                 }
 266         } else {
 267                 ucred->uc_suppgids[0] = -1;
 268                 ucred->uc_suppgids[1] = -1;
 269                 ucred->uc_ginfo = NULL;
 270         }
 271
 272         ucred->uc_uid   = pud->pud_uid;
 273         ucred->uc_gid   = pud->pud_gid;
 274         ucred->uc_fsuid = pud->pud_fsuid;
 275         ucred->uc_fsgid = pud->pud_fsgid;
 276
 277         /* process root_squash here. */
 278         mdt_root_squash(info, peernid);
 279
 280         /* remove fs privilege for non-root user. */
 281         if (ucred->uc_fsuid)
 282                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 283         else
 284                 ucred->uc_cap = pud->pud_cap;
 285         if (remote && !(perm & CFS_RMTOWN_PERM))
 286                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 287                                    CFS_CAP_CHOWN_MASK);
 288         ucred->uc_valid = UCRED_NEW;
 289
 290         EXIT;
 291
 292 out:
 293         if (rc) {
 294                 if (ucred->uc_ginfo) {
 295                         cfs_put_group_info(ucred->uc_ginfo);
 296                         ucred->uc_ginfo = NULL;
 297                 }
 298                 if (ucred->uc_identity) {
 299                         mdt_identity_put(mdt->mdt_identity_cache,
 300                                          ucred->uc_identity);
 301                         ucred->uc_identity = NULL;
 302                 }
 303         }
 304
 305         return rc;
 306 }
 307
 308 int mdt_check_ucred(struct mdt_thread_info *info)
 309 {
 310         struct ptlrpc_request   *req = mdt_info_req(info);
 311         struct mdt_device       *mdt = info->mti_mdt;
 312         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 313         struct lu_ucred         *ucred = mdt_ucred(info);
 314         struct md_identity      *identity = NULL;
 315         lnet_nid_t               peernid = req->rq_peer.nid;
 316         __u32                    perm = 0;
 317         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 318         int                      setuid;
 319         int                      setgid;
 320         int                      rc = 0;
 321
 322         ENTRY;
 323
 324         LASSERT(ucred != NULL);
 325         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 326                 RETURN(0);
 327
 328         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 329                 RETURN(0);
 330
 331         /* sanity check: if we use strong authentication, we expect the
 332          * uid which client claimed is true */
 333         if (remote) {
 334                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 335                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 336                         RETURN(-EACCES);
 337                 }
 338
 339                 if (ptlrpc_user_desc_do_idmap(req, pud))
 340                         RETURN(-EACCES);
 341
 342                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 343                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 344                                "while client claims %u:%u/%u:%u\n",
 345                                libcfs_nid2str(peernid), req->rq_auth_uid,
 346                                req->rq_auth_mapped_uid,
 347                                pud->pud_uid, pud->pud_gid,
 348                                pud->pud_fsuid, pud->pud_fsgid);
 349                         RETURN(-EACCES);
 350                 }
 351         } else {
 352                 if (req->rq_auth_uid != pud->pud_uid) {
 353                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 354                                "while client claims %u:%u/%u:%u\n",
 355                                libcfs_nid2str(peernid), req->rq_auth_uid,
 356                                pud->pud_uid, pud->pud_gid,
 357                                pud->pud_fsuid, pud->pud_fsgid);
 358                         RETURN(-EACCES);
 359                 }
 360         }
 361
 362         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 363                 if (remote) {
 364                         CDEBUG(D_SEC, "remote client must run with identity_get "
 365                                "enabled!\n");
 366                         RETURN(-EACCES);
 367                 }
 368                 RETURN(0);
 369         }
 370
 371         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 372         if (IS_ERR(identity)) {
 373                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 374                              !remote)) {
 375                         RETURN(0);
 376                 } else {
 377                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 378                                pud->pud_uid);
 379                         RETURN(-EACCES);
 380                }
 381         }
 382
 383         perm = mdt_identity_get_perm(identity, remote, peernid);
 384         /* find out the setuid/setgid attempt */
 385         setuid = (pud->pud_uid != pud->pud_fsuid);
 386         setgid = (pud->pud_gid != pud->pud_fsgid ||
 387                   pud->pud_gid != identity->mi_gid);
 388
 389         /* check permission of setuid */
 390         if (setuid && !(perm & CFS_SETUID_PERM)) {
 391                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 392                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 393                 GOTO(out, rc = -EACCES);
 394         }
 395
 396         /* check permission of setgid */
 397         if (setgid && !(perm & CFS_SETGID_PERM)) {
 398                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 399                        "from %s\n", pud->pud_uid, pud->pud_gid,
 400                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 401                        libcfs_nid2str(peernid));
 402                 GOTO(out, rc = -EACCES);
 403         }
 404
 405         EXIT;
 406
 407 out:
 408         mdt_identity_put(mdt->mdt_identity_cache, identity);
 409         return rc;
 410 }
 411
 412 static int old_init_ucred(struct mdt_thread_info *info,
 413                           struct mdt_body *body)
 414 {
 415         struct lu_ucred *uc = mdt_ucred(info);
 416         struct mdt_device  *mdt = info->mti_mdt;
 417         struct md_identity *identity = NULL;
 418
 419         ENTRY;
 420
 421         LASSERT(uc != NULL);
 422         uc->uc_valid = UCRED_INVALID;
 423         uc->uc_o_uid = uc->uc_uid = body->uid;
 424         uc->uc_o_gid = uc->uc_gid = body->gid;
 425         uc->uc_o_fsuid = uc->uc_fsuid = body->fsuid;
 426         uc->uc_o_fsgid = uc->uc_fsgid = body->fsgid;
 427         uc->uc_suppgids[0] = body->suppgid;
 428         uc->uc_suppgids[1] = -1;
 429         uc->uc_ginfo = NULL;
 430         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 431                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 432                                             uc->uc_fsuid);
 433                 if (IS_ERR(identity)) {
 434                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 435                                 identity = NULL;
 436                         } else {
 437                                 CDEBUG(D_SEC, "Deny access without identity: "
 438                                        "uid %u\n", uc->uc_fsuid);
 439                                 RETURN(-EACCES);
 440                         }
 441                 }
 442         }
 443         uc->uc_identity = identity;
 444
 445         /* process root_squash here. */
 446         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 447
 448         /* remove fs privilege for non-root user. */
 449         if (uc->uc_fsuid)
 450                 uc->uc_cap = body->capability & ~CFS_CAP_FS_MASK;
 451         else
 452                 uc->uc_cap = body->capability;
 453         uc->uc_valid = UCRED_OLD;
 454
 455         RETURN(0);
 456 }
 457
 458 static int old_init_ucred_reint(struct mdt_thread_info *info)
 459 {
 460         struct lu_ucred *uc = mdt_ucred(info);
 461         struct mdt_device  *mdt = info->mti_mdt;
 462         struct md_identity *identity = NULL;
 463
 464         ENTRY;
 465
 466         LASSERT(uc != NULL);
 467         uc->uc_valid = UCRED_INVALID;
 468         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 469         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 470         uc->uc_ginfo = NULL;
 471         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 472                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 473                                             uc->uc_fsuid);
 474                 if (IS_ERR(identity)) {
 475                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 476                                 identity = NULL;
 477                         } else {
 478                                 CDEBUG(D_SEC, "Deny access without identity: "
 479                                        "uid %u\n", uc->uc_fsuid);
 480                                 RETURN(-EACCES);
 481                         }
 482                 }
 483         }
 484         uc->uc_identity = identity;
 485
 486         /* process root_squash here. */
 487         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 488
 489         /* remove fs privilege for non-root user. */
 490         if (uc->uc_fsuid)
 491                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 492         uc->uc_valid = UCRED_OLD;
 493
 494         RETURN(0);
 495 }
 496
 497 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 498 {
 499         struct ptlrpc_request *req = mdt_info_req(info);
 500         struct lu_ucred       *uc  = mdt_ucred(info);
 501
 502         LASSERT(uc != NULL);
 503         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 504                 return 0;
 505
 506         mdt_exit_ucred(info);
 507
 508         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 509                 return old_init_ucred(info, body);
 510         else
 511                 return new_init_ucred(info, BODY_INIT, body);
 512 }
 513
 514 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 515 {
 516         struct ptlrpc_request *req = mdt_info_req(info);
 517         struct lu_ucred       *uc  = mdt_ucred(info);
 518
 519         LASSERT(uc != NULL);
 520         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 521                 return 0;
 522
 523         mdt_exit_ucred(info);
 524
 525         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 526                 return old_init_ucred_reint(info);
 527         else
 528                 return new_init_ucred(info, REC_INIT, NULL);
 529 }
 530
 531 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 532 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 533 {
 534         const struct lov_ost_data_v1 *lod;
 535         int                           i;
 536         __u16                         count;
 537
 538         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 539
 540         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 541                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 542                le32_to_cpu(lmm->lmm_pattern));
 543         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 544                le32_to_cpu(lmm->lmm_stripe_size), count);
 545         if (count == LOV_ALL_STRIPES ||
 546             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_opts.mo_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_opts.mo_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 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 805 {
 806         struct lu_ucred         *uc  = mdt_ucred(info);
 807         struct md_attr          *ma = &info->mti_attr;
 808         struct lu_attr          *la = &ma->ma_attr;
 809         struct req_capsule      *pill = info->mti_pill;
 810         struct mdt_reint_record *rr = &info->mti_rr;
 811         struct mdt_rec_setattr  *rec;
 812         ENTRY;
 813
 814         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 815         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 816         if (rec == NULL)
 817                 RETURN(-EFAULT);
 818
 819         /* This prior initialization is needed for old_init_ucred_reint() */
 820         uc->uc_fsuid = rec->sa_fsuid;
 821         uc->uc_fsgid = rec->sa_fsgid;
 822         uc->uc_cap   = rec->sa_cap;
 823         uc->uc_suppgids[0] = rec->sa_suppgid;
 824         uc->uc_suppgids[1] = -1;
 825
 826         rr->rr_fid1 = &rec->sa_fid;
 827         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 828         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 829          *  the client does not have OBD_CONNECT_FULL20, convert it
 830          *  to LA_xTIME. LU-3036 */
 831         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 832                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 833                      (rec->sa_valid & MDS_ATTR_ATIME))
 834                         la->la_valid |= LA_ATIME;
 835                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 836                      (rec->sa_valid & MDS_ATTR_MTIME))
 837                         la->la_valid |= LA_MTIME;
 838                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 839                      (rec->sa_valid & MDS_ATTR_CTIME))
 840                         la->la_valid |= LA_CTIME;
 841         }
 842         la->la_mode  = rec->sa_mode;
 843         la->la_flags = rec->sa_attr_flags;
 844         la->la_uid   = rec->sa_uid;
 845         la->la_gid   = rec->sa_gid;
 846         la->la_size  = rec->sa_size;
 847         la->la_blocks = rec->sa_blocks;
 848         la->la_ctime = rec->sa_ctime;
 849         la->la_atime = rec->sa_atime;
 850         la->la_mtime = rec->sa_mtime;
 851         ma->ma_valid = MA_INODE;
 852
 853         if (rec->sa_bias & MDS_DATA_MODIFIED)
 854                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 855         else
 856                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 857
 858         if (rec->sa_bias & MDS_HSM_RELEASE)
 859                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 860         else
 861                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 862
 863         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 864                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 865                                  req_capsule_client_get(pill, &RMF_CAPA1));
 866
 867         RETURN(0);
 868 }
 869
 870 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 871 {
 872         struct req_capsule *pill = info->mti_pill;
 873         ENTRY;
 874
 875         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 876                 info->mti_ioepoch =
 877                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 878         else
 879                 info->mti_ioepoch = NULL;
 880         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 881 }
 882
 883 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 884         struct req_capsule      *pill = info->mti_pill;
 885
 886         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 887                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 888                 if (info->mti_dlm_req == NULL)
 889                         RETURN(-EFAULT);
 890         }
 891
 892         RETURN(0);
 893 }
 894
 895 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 896 {
 897         struct mdt_reint_record *rr = &info->mti_rr;
 898         struct md_attr          *ma = &info->mti_attr;
 899         struct req_capsule      *pill = info->mti_pill;
 900         int rc;
 901         ENTRY;
 902
 903         rc = mdt_setattr_unpack_rec(info);
 904         if (rc)
 905                 RETURN(rc);
 906
 907         /* Epoch may be absent */
 908         mdt_ioepoch_unpack(info);
 909
 910         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 911                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 912                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 913                                                         RCL_CLIENT);
 914                 ma->ma_lmm_size = rr->rr_eadatalen;
 915                 if (ma->ma_lmm_size > 0) {
 916                         ma->ma_lmm = (void *)rr->rr_eadata;
 917                         ma->ma_valid |= MA_LOV;
 918                 }
 919         }
 920
 921         rc = mdt_dlmreq_unpack(info);
 922         RETURN(rc);
 923 }
 924
 925 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
 926 {
 927         struct md_attr          *ma = &info->mti_attr;
 928         struct req_capsule      *pill = info->mti_pill;
 929         ENTRY;
 930
 931         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
 932                 RETURN(0);
 933
 934         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
 935
 936         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
 937             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
 938                 RETURN(-EFAULT);
 939
 940         RETURN(0);
 941 }
 942
 943 int mdt_close_unpack(struct mdt_thread_info *info)
 944 {
 945         int rc;
 946         ENTRY;
 947
 948         rc = mdt_ioepoch_unpack(info);
 949         if (rc)
 950                 RETURN(rc);
 951
 952         rc = mdt_setattr_unpack_rec(info);
 953         if (rc)
 954                 RETURN(rc);
 955
 956         rc = mdt_hsm_release_unpack(info);
 957         if (rc)
 958                 RETURN(rc);
 959
 960         RETURN(mdt_init_ucred_reint(info));
 961 }
 962
 963 static int mdt_create_unpack(struct mdt_thread_info *info)
 964 {
 965         struct lu_ucred         *uc  = mdt_ucred(info);
 966         struct mdt_rec_create   *rec;
 967         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 968         struct mdt_reint_record *rr = &info->mti_rr;
 969         struct req_capsule      *pill = info->mti_pill;
 970         struct md_op_spec       *sp = &info->mti_spec;
 971         int rc;
 972         ENTRY;
 973
 974         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 975         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 976         if (rec == NULL)
 977                 RETURN(-EFAULT);
 978
 979         /* This prior initialization is needed for old_init_ucred_reint() */
 980         uc->uc_fsuid = rec->cr_fsuid;
 981         uc->uc_fsgid = rec->cr_fsgid;
 982         uc->uc_cap   = rec->cr_cap;
 983         uc->uc_suppgids[0] = rec->cr_suppgid1;
 984         uc->uc_suppgids[1] = -1;
 985         uc->uc_umask = rec->cr_umask;
 986
 987         rr->rr_fid1 = &rec->cr_fid1;
 988         rr->rr_fid2 = &rec->cr_fid2;
 989         attr->la_mode = rec->cr_mode;
 990         attr->la_rdev  = rec->cr_rdev;
 991         attr->la_uid   = rec->cr_fsuid;
 992         attr->la_gid   = rec->cr_fsgid;
 993         attr->la_ctime = rec->cr_time;
 994         attr->la_mtime = rec->cr_time;
 995         attr->la_atime = rec->cr_time;
 996         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
 997                          LA_CTIME | LA_MTIME | LA_ATIME;
 998         memset(&sp->u, 0, sizeof(sp->u));
 999         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1000
1001         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1002                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1003                                  req_capsule_client_get(pill, &RMF_CAPA1));
1004         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1005
1006         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1007         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
1008                                               RCL_CLIENT) - 1;
1009         LASSERT(rr->rr_name && rr->rr_namelen > 0);
1010
1011         if (S_ISLNK(attr->la_mode)) {
1012                 const char *tgt = NULL;
1013
1014                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1015                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1016                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1017                         sp->u.sp_symname = tgt;
1018                 }
1019                 if (tgt == NULL)
1020                         RETURN(-EFAULT);
1021         } else {
1022                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1023         }
1024
1025         rc = mdt_dlmreq_unpack(info);
1026         RETURN(rc);
1027 }
1028
1029 static int mdt_link_unpack(struct mdt_thread_info *info)
1030 {
1031         struct lu_ucred         *uc  = mdt_ucred(info);
1032         struct mdt_rec_link     *rec;
1033         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1034         struct mdt_reint_record *rr = &info->mti_rr;
1035         struct req_capsule      *pill = info->mti_pill;
1036         int rc;
1037         ENTRY;
1038
1039         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1040         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1041         if (rec == NULL)
1042                 RETURN(-EFAULT);
1043
1044         /* This prior initialization is needed for old_init_ucred_reint() */
1045         uc->uc_fsuid = rec->lk_fsuid;
1046         uc->uc_fsgid = rec->lk_fsgid;
1047         uc->uc_cap   = rec->lk_cap;
1048         uc->uc_suppgids[0] = rec->lk_suppgid1;
1049         uc->uc_suppgids[1] = rec->lk_suppgid2;
1050
1051         attr->la_uid = rec->lk_fsuid;
1052         attr->la_gid = rec->lk_fsgid;
1053         rr->rr_fid1 = &rec->lk_fid1;
1054         rr->rr_fid2 = &rec->lk_fid2;
1055         attr->la_ctime = rec->lk_time;
1056         attr->la_mtime = rec->lk_time;
1057         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1058
1059         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1060                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1061                                  req_capsule_client_get(pill, &RMF_CAPA1));
1062         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1063                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1064                                  req_capsule_client_get(pill, &RMF_CAPA2));
1065
1066         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1067         if (rr->rr_name == NULL)
1068                 RETURN(-EFAULT);
1069         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1070
1071         LASSERT(rr->rr_namelen > 0);
1072
1073         rc = mdt_dlmreq_unpack(info);
1074         RETURN(rc);
1075 }
1076
1077 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1078 {
1079         struct lu_ucred         *uc  = mdt_ucred(info);
1080         struct mdt_rec_unlink   *rec;
1081         struct md_attr          *ma = &info->mti_attr;
1082         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1083         struct mdt_reint_record *rr = &info->mti_rr;
1084         struct req_capsule      *pill = info->mti_pill;
1085         int rc;
1086         ENTRY;
1087
1088         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1089         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1090         if (rec == NULL)
1091                 RETURN(-EFAULT);
1092
1093         /* This prior initialization is needed for old_init_ucred_reint() */
1094         uc->uc_fsuid = rec->ul_fsuid;
1095         uc->uc_fsgid = rec->ul_fsgid;
1096         uc->uc_cap   = rec->ul_cap;
1097         uc->uc_suppgids[0] = rec->ul_suppgid1;
1098         uc->uc_suppgids[1] = -1;
1099
1100         attr->la_uid = rec->ul_fsuid;
1101         attr->la_gid = rec->ul_fsgid;
1102         rr->rr_fid1 = &rec->ul_fid1;
1103         rr->rr_fid2 = &rec->ul_fid2;
1104         attr->la_ctime = rec->ul_time;
1105         attr->la_mtime = rec->ul_time;
1106         attr->la_mode  = rec->ul_mode;
1107         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1108
1109         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1110                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1111                                  req_capsule_client_get(pill, &RMF_CAPA1));
1112
1113         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1114         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1115         if (rr->rr_name == NULL || rr->rr_namelen == 0)
1116                 RETURN(-EFAULT);
1117
1118         if (rec->ul_bias & MDS_VTX_BYPASS)
1119                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1120         else
1121                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1122
1123         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1124
1125         rc = mdt_dlmreq_unpack(info);
1126         RETURN(rc);
1127 }
1128
1129 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1130 {
1131         info->mti_spec.sp_rm_entry = 1;
1132         return mdt_unlink_unpack(info);
1133 }
1134
1135 static int mdt_rename_unpack(struct mdt_thread_info *info)
1136 {
1137         struct lu_ucred         *uc = mdt_ucred(info);
1138         struct mdt_rec_rename   *rec;
1139         struct md_attr          *ma = &info->mti_attr;
1140         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1141         struct mdt_reint_record *rr = &info->mti_rr;
1142         struct req_capsule      *pill = info->mti_pill;
1143         int rc;
1144         ENTRY;
1145
1146         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1147         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1148         if (rec == NULL)
1149                 RETURN(-EFAULT);
1150
1151         /* This prior initialization is needed for old_init_ucred_reint() */
1152         uc->uc_fsuid = rec->rn_fsuid;
1153         uc->uc_fsgid = rec->rn_fsgid;
1154         uc->uc_cap   = rec->rn_cap;
1155         uc->uc_suppgids[0] = rec->rn_suppgid1;
1156         uc->uc_suppgids[1] = rec->rn_suppgid2;
1157
1158         attr->la_uid = rec->rn_fsuid;
1159         attr->la_gid = rec->rn_fsgid;
1160         rr->rr_fid1 = &rec->rn_fid1;
1161         rr->rr_fid2 = &rec->rn_fid2;
1162         attr->la_ctime = rec->rn_time;
1163         attr->la_mtime = rec->rn_time;
1164         /* rename_tgt contains the mode already */
1165         attr->la_mode = rec->rn_mode;
1166         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1167
1168         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1169                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1170                                  req_capsule_client_get(pill, &RMF_CAPA1));
1171         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1172                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1173                                  req_capsule_client_get(pill, &RMF_CAPA2));
1174
1175         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1176         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1177         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1178                 RETURN(-EFAULT);
1179         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1180         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1181         LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1182
1183         if (rec->rn_bias & MDS_VTX_BYPASS)
1184                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1185         else
1186                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1187
1188         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1189
1190         rc = mdt_dlmreq_unpack(info);
1191         RETURN(rc);
1192 }
1193
1194 /*
1195  * please see comment above LOV_MAGIC_V1_DEF
1196  */
1197 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1198 {
1199         struct mdt_reint_record *rr = &info->mti_rr;
1200         struct lov_user_md_v1   *v1;
1201
1202         v1 = (void *)rr->rr_eadata;
1203         LASSERT(v1);
1204
1205         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1206                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1207                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1208                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1209                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1210                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1211                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1212                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1213                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1214                 }
1215         }
1216 }
1217
1218 static int mdt_open_unpack(struct mdt_thread_info *info)
1219 {
1220         struct lu_ucred         *uc = mdt_ucred(info);
1221         struct mdt_rec_create   *rec;
1222         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1223         struct req_capsule      *pill = info->mti_pill;
1224         struct mdt_reint_record *rr   = &info->mti_rr;
1225         struct ptlrpc_request   *req  = mdt_info_req(info);
1226         struct md_op_spec       *sp   = &info->mti_spec;
1227         ENTRY;
1228
1229         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1230         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1231         if (rec == NULL)
1232                 RETURN(-EFAULT);
1233
1234         /* This prior initialization is needed for old_init_ucred_reint() */
1235         uc->uc_fsuid = rec->cr_fsuid;
1236         uc->uc_fsgid = rec->cr_fsgid;
1237         uc->uc_cap   = rec->cr_cap;
1238         uc->uc_suppgids[0] = rec->cr_suppgid1;
1239         uc->uc_suppgids[1] = rec->cr_suppgid2;
1240         uc->uc_umask = rec->cr_umask;
1241
1242         rr->rr_fid1   = &rec->cr_fid1;
1243         rr->rr_fid2   = &rec->cr_fid2;
1244         rr->rr_handle = &rec->cr_old_handle;
1245         attr->la_mode = rec->cr_mode;
1246         attr->la_rdev  = rec->cr_rdev;
1247         attr->la_uid   = rec->cr_fsuid;
1248         attr->la_gid   = rec->cr_fsgid;
1249         attr->la_ctime = rec->cr_time;
1250         attr->la_mtime = rec->cr_time;
1251         attr->la_atime = rec->cr_time;
1252         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1253                          LA_CTIME | LA_MTIME | LA_ATIME;
1254         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1255         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1256         /* Do not trigger ASSERTION if client miss to set such flags. */
1257         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1258                 RETURN(-EPROTO);
1259         info->mti_replayepoch = rec->cr_ioepoch;
1260
1261         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1262
1263         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1264                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1265                                  req_capsule_client_get(pill, &RMF_CAPA1));
1266         if (req_is_replay(req) &&
1267             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1268 #if 0
1269                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1270                                  req_capsule_client_get(pill, &RMF_CAPA2));
1271 #else
1272                 /*
1273                  * FIXME: capa in replay open request might have expired,
1274                  * bypass capa check. Security hole?
1275                  */
1276                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1277                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1278 #endif
1279         }
1280
1281         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1282         if (rr->rr_name == NULL)
1283                 RETURN(-EFAULT);
1284         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1285
1286         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1287                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1288                                                         RCL_CLIENT);
1289                 if (rr->rr_eadatalen > 0) {
1290                         rr->rr_eadata = req_capsule_client_get(pill,
1291                                                                &RMF_EADATA);
1292                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1293                         sp->u.sp_ea.eadata = rr->rr_eadata;
1294                         sp->no_create = !!req_is_replay(req);
1295                         mdt_fix_lov_magic(info);
1296                 }
1297
1298                 /*
1299                  * Client default md_size may be 0 right after client start,
1300                  * until all osc are connected, set here just some reasonable
1301                  * value to prevent misbehavior.
1302                  */
1303                 if (rr->rr_eadatalen == 0 &&
1304                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1305                         rr->rr_eadatalen = MIN_MD_SIZE;
1306         }
1307
1308         RETURN(0);
1309 }
1310
1311 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1312 {
1313         struct mdt_reint_record   *rr   = &info->mti_rr;
1314         struct lu_ucred           *uc   = mdt_ucred(info);
1315         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1316         struct req_capsule        *pill = info->mti_pill;
1317         struct mdt_rec_setxattr   *rec;
1318         ENTRY;
1319
1320
1321         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1322                          sizeof(struct mdt_rec_reint));
1323
1324         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1325         if (rec == NULL)
1326                 RETURN(-EFAULT);
1327
1328         /* This prior initialization is needed for old_init_ucred_reint() */
1329         uc->uc_fsuid  = rec->sx_fsuid;
1330         uc->uc_fsgid  = rec->sx_fsgid;
1331         uc->uc_cap    = rec->sx_cap;
1332         uc->uc_suppgids[0] = rec->sx_suppgid1;
1333         uc->uc_suppgids[1] = -1;
1334
1335         rr->rr_opcode = rec->sx_opcode;
1336         rr->rr_fid1   = &rec->sx_fid;
1337         attr->la_valid = rec->sx_valid;
1338         attr->la_ctime = rec->sx_time;
1339         attr->la_size = rec->sx_size;
1340         attr->la_flags = rec->sx_flags;
1341
1342         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1343                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1344                                  req_capsule_client_get(pill, &RMF_CAPA1));
1345         else
1346                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1347
1348         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1349         if (rr->rr_name == NULL)
1350                 RETURN(-EFAULT);
1351         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1352         LASSERT(rr->rr_namelen > 0);
1353
1354         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1355                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1356                                                         RCL_CLIENT);
1357                 if (rr->rr_eadatalen > 0) {
1358                         rr->rr_eadata = req_capsule_client_get(pill,
1359                                                                &RMF_EADATA);
1360                         if (rr->rr_eadata == NULL)
1361                                 RETURN(-EFAULT);
1362                 } else {
1363                         rr->rr_eadata = NULL;
1364                 }
1365         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1366                 CDEBUG(D_INFO, "no xattr data supplied\n");
1367                 RETURN(-EFAULT);
1368         }
1369
1370         RETURN(0);
1371 }
1372
1373
1374 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1375
1376 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1377         [REINT_SETATTR]  = mdt_setattr_unpack,
1378         [REINT_CREATE]   = mdt_create_unpack,
1379         [REINT_LINK]     = mdt_link_unpack,
1380         [REINT_UNLINK]   = mdt_unlink_unpack,
1381         [REINT_RENAME]   = mdt_rename_unpack,
1382         [REINT_OPEN]     = mdt_open_unpack,
1383         [REINT_SETXATTR] = mdt_setxattr_unpack,
1384         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1385 };
1386
1387 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1388 {
1389         int rc;
1390         ENTRY;
1391
1392         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1393         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1394                 info->mti_rr.rr_opcode = op;
1395                 rc = mdt_reint_unpackers[op](info);
1396         } else {
1397                 CERROR("Unexpected opcode %d\n", op);
1398                 rc = -EFAULT;
1399         }
1400         RETURN(rc);
1401 }