lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  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 md_ucred   *uc  = mdt_ucred(info);
  64         struct mdt_device *mdt = info->mti_mdt;
  65
  66         if (uc->mu_valid != UCRED_INIT) {
  67                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
  68                 if (uc->mu_ginfo) {
  69                         cfs_put_group_info(uc->mu_ginfo);
  70                         uc->mu_ginfo = NULL;
  71                 }
  72                 if (uc->mu_identity) {
  73                         mdt_identity_put(mdt->mdt_identity_cache,
  74                                          uc->mu_identity);
  75                         uc->mu_identity = NULL;
  76                 }
  77                 uc->mu_valid = UCRED_INIT;
  78         }
  79 }
  80
  81 static int match_nosquash_list(cfs_rw_semaphore_t *sem,
  82                                cfs_list_t *nidlist,
  83                                lnet_nid_t peernid)
  84 {
  85         int rc;
  86         ENTRY;
  87         cfs_down_read(sem);
  88         rc = cfs_match_nid(peernid, nidlist);
  89         cfs_up_read(sem);
  90         RETURN(rc);
  91 }
  92
  93 /* root_squash for inter-MDS operations */
  94 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  95 {
  96         struct md_ucred *ucred = mdt_ucred(info);
  97         ENTRY;
  98
  99         if (!info->mti_mdt->mdt_squash_uid || ucred->mu_fsuid)
 100                 RETURN(0);
 101
 102         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 103                                 &info->mti_mdt->mdt_nosquash_nids,
 104                                 peernid)) {
 105                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 106                        libcfs_nid2str(peernid));
 107                 RETURN(0);
 108         }
 109
 110         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 111                libcfs_nid2str(peernid),
 112                ucred->mu_fsuid, ucred->mu_fsgid, ucred->mu_cap,
 113                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 114                0);
 115
 116         ucred->mu_fsuid = info->mti_mdt->mdt_squash_uid;
 117         ucred->mu_fsgid = info->mti_mdt->mdt_squash_gid;
 118         ucred->mu_cap = 0;
 119         ucred->mu_suppgids[0] = -1;
 120         ucred->mu_suppgids[1] = -1;
 121
 122         RETURN(0);
 123 }
 124
 125 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 126                           void *buf)
 127 {
 128         struct ptlrpc_request   *req = mdt_info_req(info);
 129         struct mdt_device       *mdt = info->mti_mdt;
 130         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 131         struct md_ucred         *ucred = mdt_ucred(info);
 132         lnet_nid_t               peernid = req->rq_peer.nid;
 133         __u32                    perm = 0;
 134         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 135         int                      setuid;
 136         int                      setgid;
 137         int                      rc = 0;
 138
 139         ENTRY;
 140
 141         LASSERT(req->rq_auth_gss);
 142         LASSERT(!req->rq_auth_usr_mdt);
 143         LASSERT(req->rq_user_desc);
 144
 145         ucred->mu_valid = UCRED_INVALID;
 146
 147         ucred->mu_o_uid   = pud->pud_uid;
 148         ucred->mu_o_gid   = pud->pud_gid;
 149         ucred->mu_o_fsuid = pud->pud_fsuid;
 150         ucred->mu_o_fsgid = pud->pud_fsgid;
 151
 152         if (type == BODY_INIT) {
 153                 struct mdt_body *body = (struct mdt_body *)buf;
 154
 155                 ucred->mu_suppgids[0] = body->suppgid;
 156                 ucred->mu_suppgids[1] = -1;
 157         }
 158
 159         /* sanity check: we expect the uid which client claimed is true */
 160         if (remote) {
 161                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 162                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 163                         RETURN(-EACCES);
 164                 }
 165
 166                 if (ptlrpc_user_desc_do_idmap(req, pud))
 167                         RETURN(-EACCES);
 168
 169                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 170                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 171                                "while client claims %u:%u/%u:%u\n",
 172                                libcfs_nid2str(peernid), req->rq_auth_uid,
 173                                req->rq_auth_mapped_uid,
 174                                pud->pud_uid, pud->pud_gid,
 175                                pud->pud_fsuid, pud->pud_fsgid);
 176                         RETURN(-EACCES);
 177                 }
 178         } else {
 179                 if (req->rq_auth_uid != pud->pud_uid) {
 180                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 181                                "while client claims %u:%u/%u:%u\n",
 182                                libcfs_nid2str(peernid), req->rq_auth_uid,
 183                                pud->pud_uid, pud->pud_gid,
 184                                pud->pud_fsuid, pud->pud_fsgid);
 185                         RETURN(-EACCES);
 186                 }
 187         }
 188
 189         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 190                 if (remote) {
 191                         CDEBUG(D_SEC, "remote client must run with identity_get "
 192                                "enabled!\n");
 193                         RETURN(-EACCES);
 194                 } else {
 195                         ucred->mu_identity = NULL;
 196                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 197                                CFS_SETGRP_PERM;
 198                 }
 199         } else {
 200                 struct md_identity *identity;
 201
 202                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 203                                             pud->pud_uid);
 204                 if (IS_ERR(identity)) {
 205                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 206                                      !remote)) {
 207                                 ucred->mu_identity = NULL;
 208                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 209                                        CFS_SETGRP_PERM;
 210                         } else {
 211                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 212                                        pud->pud_uid);
 213                                 RETURN(-EACCES);
 214                         }
 215                 } else {
 216                         ucred->mu_identity = identity;
 217                         perm = mdt_identity_get_perm(ucred->mu_identity,
 218                                                      remote, peernid);
 219                 }
 220         }
 221
 222         /* find out the setuid/setgid attempt */
 223         setuid = (pud->pud_uid != pud->pud_fsuid);
 224         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 225                   (ucred->mu_identity &&
 226                   (pud->pud_gid != ucred->mu_identity->mi_gid)));
 227
 228         /* check permission of setuid */
 229         if (setuid && !(perm & CFS_SETUID_PERM)) {
 230                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 231                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 232                 GOTO(out, rc = -EACCES);
 233         }
 234
 235         /* check permission of setgid */
 236         if (setgid && !(perm & CFS_SETGID_PERM)) {
 237                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 238                        "from %s\n", pud->pud_uid, pud->pud_gid,
 239                        pud->pud_fsuid, pud->pud_fsgid,
 240                        ucred->mu_identity->mi_gid, libcfs_nid2str(peernid));
 241                 GOTO(out, rc = -EACCES);
 242         }
 243
 244         /*
 245          * NB: remote client not allowed to setgroups anyway.
 246          */
 247         if (!remote && perm & CFS_SETGRP_PERM) {
 248                 if (pud->pud_ngroups) {
 249                         /* setgroups for local client */
 250                         ucred->mu_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 251                         if (!ucred->mu_ginfo) {
 252                                 CERROR("failed to alloc %d groups\n",
 253                                        pud->pud_ngroups);
 254                                 GOTO(out, rc = -ENOMEM);
 255                         }
 256
 257                         lustre_groups_from_list(ucred->mu_ginfo,
 258                                                 pud->pud_groups);
 259                         lustre_groups_sort(ucred->mu_ginfo);
 260                 } else {
 261                         ucred->mu_ginfo = NULL;
 262                 }
 263         } else {
 264                 ucred->mu_suppgids[0] = -1;
 265                 ucred->mu_suppgids[1] = -1;
 266                 ucred->mu_ginfo = NULL;
 267         }
 268
 269         ucred->mu_uid   = pud->pud_uid;
 270         ucred->mu_gid   = pud->pud_gid;
 271         ucred->mu_fsuid = pud->pud_fsuid;
 272         ucred->mu_fsgid = pud->pud_fsgid;
 273
 274         /* process root_squash here. */
 275         mdt_root_squash(info, peernid);
 276
 277         /* remove fs privilege for non-root user. */
 278         if (ucred->mu_fsuid)
 279                 ucred->mu_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 280         else
 281                 ucred->mu_cap = pud->pud_cap;
 282         if (remote && !(perm & CFS_RMTOWN_PERM))
 283                 ucred->mu_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 284                                    CFS_CAP_CHOWN_MASK);
 285         ucred->mu_valid = UCRED_NEW;
 286
 287         EXIT;
 288
 289 out:
 290         if (rc) {
 291                 if (ucred->mu_ginfo) {
 292                         cfs_put_group_info(ucred->mu_ginfo);
 293                         ucred->mu_ginfo = NULL;
 294                 }
 295                 if (ucred->mu_identity) {
 296                         mdt_identity_put(mdt->mdt_identity_cache,
 297                                          ucred->mu_identity);
 298                         ucred->mu_identity = NULL;
 299                 }
 300         }
 301
 302         return rc;
 303 }
 304
 305 int mdt_check_ucred(struct mdt_thread_info *info)
 306 {
 307         struct ptlrpc_request   *req = mdt_info_req(info);
 308         struct mdt_device       *mdt = info->mti_mdt;
 309         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 310         struct md_ucred         *ucred = mdt_ucred(info);
 311         struct md_identity      *identity = NULL;
 312         lnet_nid_t               peernid = req->rq_peer.nid;
 313         __u32                    perm = 0;
 314         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 315         int                      setuid;
 316         int                      setgid;
 317         int                      rc = 0;
 318
 319         ENTRY;
 320
 321         if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
 322                 RETURN(0);
 323
 324         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 325                 RETURN(0);
 326
 327         /* sanity check: if we use strong authentication, we expect the
 328          * uid which client claimed is true */
 329         if (remote) {
 330                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 331                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 332                         RETURN(-EACCES);
 333                 }
 334
 335                 if (ptlrpc_user_desc_do_idmap(req, pud))
 336                         RETURN(-EACCES);
 337
 338                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 339                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 340                                "while client claims %u:%u/%u:%u\n",
 341                                libcfs_nid2str(peernid), req->rq_auth_uid,
 342                                req->rq_auth_mapped_uid,
 343                                pud->pud_uid, pud->pud_gid,
 344                                pud->pud_fsuid, pud->pud_fsgid);
 345                         RETURN(-EACCES);
 346                 }
 347         } else {
 348                 if (req->rq_auth_uid != pud->pud_uid) {
 349                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 350                                "while client claims %u:%u/%u:%u\n",
 351                                libcfs_nid2str(peernid), req->rq_auth_uid,
 352                                pud->pud_uid, pud->pud_gid,
 353                                pud->pud_fsuid, pud->pud_fsgid);
 354                         RETURN(-EACCES);
 355                 }
 356         }
 357
 358         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 359                 if (remote) {
 360                         CDEBUG(D_SEC, "remote client must run with identity_get "
 361                                "enabled!\n");
 362                         RETURN(-EACCES);
 363                 }
 364                 RETURN(0);
 365         }
 366
 367         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 368         if (IS_ERR(identity)) {
 369                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 370                              !remote)) {
 371                         RETURN(0);
 372                 } else {
 373                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 374                                pud->pud_uid);
 375                         RETURN(-EACCES);
 376                }
 377         }
 378
 379         perm = mdt_identity_get_perm(identity, remote, peernid);
 380         /* find out the setuid/setgid attempt */
 381         setuid = (pud->pud_uid != pud->pud_fsuid);
 382         setgid = (pud->pud_gid != pud->pud_fsgid ||
 383                   pud->pud_gid != identity->mi_gid);
 384
 385         /* check permission of setuid */
 386         if (setuid && !(perm & CFS_SETUID_PERM)) {
 387                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 388                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 389                 GOTO(out, rc = -EACCES);
 390         }
 391
 392         /* check permission of setgid */
 393         if (setgid && !(perm & CFS_SETGID_PERM)) {
 394                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 395                        "from %s\n", pud->pud_uid, pud->pud_gid,
 396                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 397                        libcfs_nid2str(peernid));
 398                 GOTO(out, rc = -EACCES);
 399         }
 400
 401         EXIT;
 402
 403 out:
 404         mdt_identity_put(mdt->mdt_identity_cache, identity);
 405         return rc;
 406 }
 407
 408 static int old_init_ucred(struct mdt_thread_info *info,
 409                           struct mdt_body *body)
 410 {
 411         struct md_ucred *uc = mdt_ucred(info);
 412         struct mdt_device  *mdt = info->mti_mdt;
 413         struct md_identity *identity = NULL;
 414
 415         ENTRY;
 416
 417         uc->mu_valid = UCRED_INVALID;
 418         uc->mu_o_uid = uc->mu_uid = body->uid;
 419         uc->mu_o_gid = uc->mu_gid = body->gid;
 420         uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
 421         uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
 422         uc->mu_suppgids[0] = body->suppgid;
 423         uc->mu_suppgids[1] = -1;
 424         uc->mu_ginfo = NULL;
 425         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 426                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 427                                             uc->mu_fsuid);
 428                 if (IS_ERR(identity)) {
 429                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 430                                 identity = NULL;
 431                         } else {
 432                                 CDEBUG(D_SEC, "Deny access without identity: "
 433                                        "uid %u\n", uc->mu_fsuid);
 434                                 RETURN(-EACCES);
 435                         }
 436                 }
 437         }
 438         uc->mu_identity = identity;
 439
 440         /* process root_squash here. */
 441         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 442
 443         /* remove fs privilege for non-root user. */
 444         if (uc->mu_fsuid)
 445                 uc->mu_cap = body->capability & ~CFS_CAP_FS_MASK;
 446         else
 447                 uc->mu_cap = body->capability;
 448         uc->mu_valid = UCRED_OLD;
 449
 450         RETURN(0);
 451 }
 452
 453 static int old_init_ucred_reint(struct mdt_thread_info *info)
 454 {
 455         struct md_ucred *uc = mdt_ucred(info);
 456         struct mdt_device  *mdt = info->mti_mdt;
 457         struct md_identity *identity = NULL;
 458
 459         ENTRY;
 460
 461         uc->mu_valid = UCRED_INVALID;
 462         uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
 463         uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
 464         uc->mu_ginfo = NULL;
 465         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 466                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 467                                             uc->mu_fsuid);
 468                 if (IS_ERR(identity)) {
 469                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 470                                 identity = NULL;
 471                         } else {
 472                                 CDEBUG(D_SEC, "Deny access without identity: "
 473                                        "uid %u\n", uc->mu_fsuid);
 474                                 RETURN(-EACCES);
 475                         }
 476                 }
 477         }
 478         uc->mu_identity = identity;
 479
 480         /* process root_squash here. */
 481         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 482
 483         /* remove fs privilege for non-root user. */
 484         if (uc->mu_fsuid)
 485                 uc->mu_cap &= ~CFS_CAP_FS_MASK;
 486         uc->mu_valid = UCRED_OLD;
 487
 488         RETURN(0);
 489 }
 490
 491 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 492 {
 493         struct ptlrpc_request *req = mdt_info_req(info);
 494         struct md_ucred       *uc  = mdt_ucred(info);
 495
 496         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 497                 return 0;
 498
 499         mdt_exit_ucred(info);
 500
 501         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 502                 return old_init_ucred(info, body);
 503         else
 504                 return new_init_ucred(info, BODY_INIT, body);
 505 }
 506
 507 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 508 {
 509         struct ptlrpc_request *req = mdt_info_req(info);
 510         struct md_ucred       *uc  = mdt_ucred(info);
 511
 512         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 513                 return 0;
 514
 515         mdt_exit_ucred(info);
 516
 517         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 518                 return old_init_ucred_reint(info);
 519         else
 520                 return new_init_ucred(info, REC_INIT, NULL);
 521 }
 522
 523 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 524 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 525 {
 526         const struct lov_ost_data_v1 *lod;
 527         int                           i;
 528         __u16                         count;
 529
 530         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 531
 532         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
 533                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 534                le32_to_cpu(lmm->lmm_pattern));
 535         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 536                le32_to_cpu(lmm->lmm_stripe_size), count);
 537         if (count == LOV_ALL_STRIPES)
 538                 return;
 539         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 540         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++)
 541                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 542                        i, le32_to_cpu(lod->l_ost_idx),
 543                        le64_to_cpu(lod->l_object_seq),
 544                        le64_to_cpu(lod->l_object_id));
 545 }
 546
 547 /* Shrink and/or grow reply buffers */
 548 int mdt_fix_reply(struct mdt_thread_info *info)
 549 {
 550         struct req_capsule *pill = info->mti_pill;
 551         struct mdt_body    *body;
 552         int                md_size, md_packed = 0;
 553         int                acl_size;
 554         int                rc = 0;
 555         ENTRY;
 556
 557         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 558         LASSERT(body != NULL);
 559
 560         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 561                 md_size = body->eadatasize;
 562         else
 563                 md_size = 0;
 564
 565         acl_size = body->aclsize;
 566
 567         /* this replay - not send info to client */
 568         if (info->mti_spec.no_create == 1) {
 569                 md_size = 0;
 570                 acl_size = 0;
 571         }
 572
 573         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 574                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 575                         md_size, acl_size,
 576                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 577                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 578 /*
 579             &RMF_MDT_BODY,
 580             &RMF_MDT_MD,
 581             &RMF_ACL, or &RMF_LOGCOOKIES
 582 (optional)  &RMF_CAPA1,
 583 (optional)  &RMF_CAPA2,
 584 (optional)  something else
 585 */
 586
 587         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 588          * buffers before growing it */
 589         if (info->mti_big_lmm_used) {
 590                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 591                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 592                                                  RCL_SERVER);
 593                 LASSERT(md_packed > 0);
 594                 /* buffer must be allocated separately */
 595                 LASSERT(info->mti_attr.ma_lmm !=
 596                         req_capsule_server_get(pill, &RMF_MDT_MD));
 597                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 598                 /* free big lmm if md_size is not needed */
 599                 if (md_size == 0)
 600                         info->mti_big_lmm_used = 0;
 601         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 602                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 603         }
 604
 605         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 606                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 607         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 608                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 609                                    acl_size, RCL_SERVER);
 610
 611         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 612             !(body->valid & OBD_MD_FLMDSCAPA))
 613                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 614
 615         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 616             !(body->valid & OBD_MD_FLOSSCAPA))
 617                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 618
 619         /*
 620          * Some more field should be shrinked if needed.
 621          * This should be done by those who added fields to reply message.
 622          */
 623
 624         /* Grow MD buffer if needed finally */
 625         if (info->mti_big_lmm_used) {
 626                 void *lmm;
 627
 628                 LASSERT(md_size > md_packed);
 629                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 630                        md_size - md_packed);
 631                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 632                 if (rc) {
 633                         /* we can't answer with proper LOV EA, drop flags,
 634                          * the rc is also returned so this request is
 635                          * considered as failed */
 636                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 637                         /* don't return transno along with error */
 638                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 639                 } else {
 640                         /* now we need to pack right LOV EA */
 641                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 642                         LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 643                                                      RCL_SERVER) ==
 644                                 info->mti_attr.ma_lmm_size);
 645                         memcpy(lmm, info->mti_attr.ma_lmm,
 646                                info->mti_attr.ma_lmm_size);
 647                 }
 648                 /* update mdt_max_mdsize so clients will be aware about that */
 649                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 650                         info->mti_mdt->mdt_max_mdsize =
 651                                                     info->mti_attr.ma_lmm_size;
 652                 info->mti_big_lmm_used = 0;
 653         }
 654         RETURN(rc);
 655 }
 656
 657
 658 /* if object is dying, pack the lov/llog data,
 659  * parameter info->mti_attr should be valid at this point! */
 660 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 661                            const struct md_attr *ma)
 662 {
 663         struct mdt_body       *repbody;
 664         const struct lu_attr *la = &ma->ma_attr;
 665         int rc;
 666         ENTRY;
 667
 668         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 669         LASSERT(repbody != NULL);
 670
 671         if (ma->ma_valid & MA_INODE)
 672                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 673
 674         if (ma->ma_valid & MA_LOV) {
 675                 CERROR("No need in LOV EA upon unlink\n");
 676                 dump_stack();
 677         }
 678         repbody->eadatasize = 0;
 679
 680         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 681                 repbody->aclsize = ma->ma_cookie_size;
 682                 repbody->valid |= OBD_MD_FLCOOKIE;
 683         }
 684
 685         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 686             info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
 687             repbody->valid & OBD_MD_FLEASIZE) {
 688                 struct lustre_capa *capa;
 689
 690                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 691                 LASSERT(capa);
 692                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 693                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 694                 if (rc)
 695                         RETURN(rc);
 696
 697                 repbody->valid |= OBD_MD_FLOSSCAPA;
 698         }
 699
 700         RETURN(0);
 701 }
 702
 703 static inline unsigned int attr_unpack(__u64 sa_valid) {
 704         unsigned int ia_valid = 0;
 705
 706         if (sa_valid & MDS_ATTR_MODE)
 707                 ia_valid |= ATTR_MODE;
 708         if (sa_valid & MDS_ATTR_UID)
 709                 ia_valid |= ATTR_UID;
 710         if (sa_valid & MDS_ATTR_GID)
 711                 ia_valid |= ATTR_GID;
 712         if (sa_valid & MDS_ATTR_SIZE)
 713                 ia_valid |= ATTR_SIZE;
 714         if (sa_valid & MDS_ATTR_ATIME)
 715                 ia_valid |= ATTR_ATIME;
 716         if (sa_valid & MDS_ATTR_MTIME)
 717                 ia_valid |= ATTR_MTIME;
 718         if (sa_valid & MDS_ATTR_CTIME)
 719                 ia_valid |= ATTR_CTIME;
 720         if (sa_valid & MDS_ATTR_ATIME_SET)
 721                 ia_valid |= ATTR_ATIME_SET;
 722         if (sa_valid & MDS_ATTR_MTIME_SET)
 723                 ia_valid |= ATTR_MTIME_SET;
 724         if (sa_valid & MDS_ATTR_FORCE)
 725                 ia_valid |= ATTR_FORCE;
 726         if (sa_valid & MDS_ATTR_ATTR_FLAG)
 727                 ia_valid |= ATTR_ATTR_FLAG;
 728         if (sa_valid & MDS_ATTR_KILL_SUID)
 729                 ia_valid |=  ATTR_KILL_SUID;
 730         if (sa_valid & MDS_ATTR_KILL_SGID)
 731                 ia_valid |= ATTR_KILL_SGID;
 732         if (sa_valid & MDS_ATTR_CTIME_SET)
 733                 ia_valid |= ATTR_CTIME_SET;
 734         if (sa_valid & MDS_ATTR_FROM_OPEN)
 735                 ia_valid |= ATTR_FROM_OPEN;
 736         if (sa_valid & MDS_ATTR_BLOCKS)
 737                 ia_valid |= ATTR_BLOCKS;
 738         if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
 739                 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
 740         return ia_valid;
 741 }
 742
 743 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 744                                   struct md_attr *ma)
 745 {
 746         __u64 out;
 747
 748         out = 0;
 749         if (in & ATTR_MODE)
 750                 out |= LA_MODE;
 751         if (in & ATTR_UID)
 752                 out |= LA_UID;
 753         if (in & ATTR_GID)
 754                 out |= LA_GID;
 755         if (in & ATTR_SIZE)
 756                 out |= LA_SIZE;
 757         if (in & ATTR_BLOCKS)
 758                 out |= LA_BLOCKS;
 759
 760         if (in & ATTR_FROM_OPEN)
 761                 rr->rr_flags |= MRF_OPEN_TRUNC;
 762
 763         if (in & ATTR_ATIME_SET)
 764                 out |= LA_ATIME;
 765
 766         if (in & ATTR_CTIME_SET)
 767                 out |= LA_CTIME;
 768
 769         if (in & ATTR_MTIME_SET)
 770                 out |= LA_MTIME;
 771
 772         if (in & ATTR_ATTR_FLAG)
 773                 out |= LA_FLAGS;
 774
 775         if (in & ATTR_KILL_SUID)
 776                 out |= LA_KILL_SUID;
 777
 778         if (in & ATTR_KILL_SGID)
 779                 out |= LA_KILL_SGID;
 780
 781         if (in & MDS_OPEN_OWNEROVERRIDE)
 782                 ma->ma_attr_flags |= MDS_OPEN_OWNEROVERRIDE;
 783
 784         if (in & ATTR_FORCE)
 785                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 786
 787         /*XXX need ATTR_RAW?*/
 788         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 789                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 790                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 791                 ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
 792                 ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
 793         if (in != 0)
 794                 CERROR("Unknown attr bits: "LPX64"\n", in);
 795         return out;
 796 }
 797 /* unpacking */
 798
 799 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 800 {
 801         struct md_ucred         *uc  = mdt_ucred(info);
 802         struct md_attr          *ma = &info->mti_attr;
 803         struct lu_attr          *la = &ma->ma_attr;
 804         struct req_capsule      *pill = info->mti_pill;
 805         struct mdt_reint_record *rr = &info->mti_rr;
 806         struct mdt_rec_setattr  *rec;
 807         ENTRY;
 808
 809         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 810         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 811         if (rec == NULL)
 812                 RETURN(-EFAULT);
 813
 814         uc->mu_fsuid = rec->sa_fsuid;
 815         uc->mu_fsgid = rec->sa_fsgid;
 816         uc->mu_cap   = rec->sa_cap;
 817         uc->mu_suppgids[0] = rec->sa_suppgid;
 818         uc->mu_suppgids[1] = -1;
 819
 820         rr->rr_fid1 = &rec->sa_fid;
 821         la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
 822         la->la_mode  = rec->sa_mode;
 823         la->la_flags = rec->sa_attr_flags;
 824         la->la_uid   = rec->sa_uid;
 825         la->la_gid   = rec->sa_gid;
 826         la->la_size  = rec->sa_size;
 827         la->la_blocks = rec->sa_blocks;
 828         la->la_ctime = rec->sa_ctime;
 829         la->la_atime = rec->sa_atime;
 830         la->la_mtime = rec->sa_mtime;
 831         ma->ma_valid = MA_INODE;
 832
 833         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 834                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 835                                  req_capsule_client_get(pill, &RMF_CAPA1));
 836
 837         RETURN(0);
 838 }
 839
 840 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 841 {
 842         struct req_capsule *pill = info->mti_pill;
 843         ENTRY;
 844
 845         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 846                 info->mti_ioepoch =
 847                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 848         else
 849                 info->mti_ioepoch = NULL;
 850         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 851 }
 852
 853 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 854         struct req_capsule      *pill = info->mti_pill;
 855
 856         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 857                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 858                 if (info->mti_dlm_req == NULL)
 859                         RETURN(-EFAULT);
 860         }
 861
 862         RETURN(0);
 863 }
 864
 865 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 866 {
 867         struct mdt_reint_record *rr = &info->mti_rr;
 868         struct md_attr          *ma = &info->mti_attr;
 869         struct req_capsule      *pill = info->mti_pill;
 870         int rc;
 871         ENTRY;
 872
 873         rc = mdt_setattr_unpack_rec(info);
 874         if (rc)
 875                 RETURN(rc);
 876
 877         /* Epoch may be absent */
 878         mdt_ioepoch_unpack(info);
 879
 880         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 881                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 882                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 883                                                         RCL_CLIENT);
 884                 ma->ma_lmm_size = rr->rr_eadatalen;
 885                 if (ma->ma_lmm_size > 0) {
 886                         ma->ma_lmm = (void *)rr->rr_eadata;
 887                         ma->ma_valid |= MA_LOV;
 888                 }
 889         }
 890
 891         rc = mdt_dlmreq_unpack(info);
 892         RETURN(rc);
 893 }
 894
 895 int mdt_close_unpack(struct mdt_thread_info *info)
 896 {
 897         int rc;
 898         ENTRY;
 899
 900         rc = mdt_ioepoch_unpack(info);
 901         if (rc)
 902                 RETURN(rc);
 903
 904         RETURN(mdt_setattr_unpack_rec(info));
 905 }
 906
 907 static int mdt_create_unpack(struct mdt_thread_info *info)
 908 {
 909         struct md_ucred         *uc  = mdt_ucred(info);
 910         struct mdt_rec_create   *rec;
 911         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 912         struct mdt_reint_record *rr = &info->mti_rr;
 913         struct req_capsule      *pill = info->mti_pill;
 914         struct md_op_spec       *sp = &info->mti_spec;
 915         int rc;
 916         ENTRY;
 917
 918         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 919         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 920         if (rec == NULL)
 921                 RETURN(-EFAULT);
 922
 923         uc->mu_fsuid = rec->cr_fsuid;
 924         uc->mu_fsgid = rec->cr_fsgid;
 925         uc->mu_cap   = rec->cr_cap;
 926         uc->mu_suppgids[0] = rec->cr_suppgid1;
 927         uc->mu_suppgids[1] = -1;
 928
 929         rr->rr_fid1 = &rec->cr_fid1;
 930         rr->rr_fid2 = &rec->cr_fid2;
 931         attr->la_mode = rec->cr_mode;
 932         attr->la_rdev  = rec->cr_rdev;
 933         attr->la_uid   = rec->cr_fsuid;
 934         attr->la_gid   = rec->cr_fsgid;
 935         attr->la_ctime = rec->cr_time;
 936         attr->la_mtime = rec->cr_time;
 937         attr->la_atime = rec->cr_time;
 938         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
 939                          LA_CTIME | LA_MTIME | LA_ATIME;
 940         memset(&sp->u, 0, sizeof(sp->u));
 941         sp->sp_cr_flags = get_mrc_cr_flags(rec);
 942
 943         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 944                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 945                                  req_capsule_client_get(pill, &RMF_CAPA1));
 946         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 947
 948         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 949         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
 950                                               RCL_CLIENT) - 1;
 951         LASSERT(rr->rr_name && rr->rr_namelen > 0);
 952
 953         if (S_ISLNK(attr->la_mode)) {
 954                 const char *tgt = NULL;
 955
 956                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
 957                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
 958                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
 959                         sp->u.sp_symname = tgt;
 960                 }
 961                 if (tgt == NULL)
 962                         RETURN(-EFAULT);
 963         } else {
 964                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 965         }
 966
 967         rc = mdt_dlmreq_unpack(info);
 968         RETURN(rc);
 969 }
 970
 971 static int mdt_link_unpack(struct mdt_thread_info *info)
 972 {
 973         struct md_ucred         *uc  = mdt_ucred(info);
 974         struct mdt_rec_link     *rec;
 975         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 976         struct mdt_reint_record *rr = &info->mti_rr;
 977         struct req_capsule      *pill = info->mti_pill;
 978         int rc;
 979         ENTRY;
 980
 981         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
 982         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 983         if (rec == NULL)
 984                 RETURN(-EFAULT);
 985
 986         uc->mu_fsuid = rec->lk_fsuid;
 987         uc->mu_fsgid = rec->lk_fsgid;
 988         uc->mu_cap   = rec->lk_cap;
 989         uc->mu_suppgids[0] = rec->lk_suppgid1;
 990         uc->mu_suppgids[1] = rec->lk_suppgid2;
 991
 992         attr->la_uid = rec->lk_fsuid;
 993         attr->la_gid = rec->lk_fsgid;
 994         rr->rr_fid1 = &rec->lk_fid1;
 995         rr->rr_fid2 = &rec->lk_fid2;
 996         attr->la_ctime = rec->lk_time;
 997         attr->la_mtime = rec->lk_time;
 998         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
 999
1000         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1001                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1002                                  req_capsule_client_get(pill, &RMF_CAPA1));
1003         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1004                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1005                                  req_capsule_client_get(pill, &RMF_CAPA2));
1006
1007         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1008         if (rr->rr_name == NULL)
1009                 RETURN(-EFAULT);
1010         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1011
1012         LASSERT(rr->rr_namelen > 0);
1013
1014         rc = mdt_dlmreq_unpack(info);
1015         RETURN(rc);
1016 }
1017
1018 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1019 {
1020         struct md_ucred         *uc  = mdt_ucred(info);
1021         struct mdt_rec_unlink   *rec;
1022         struct md_attr          *ma = &info->mti_attr;
1023         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1024         struct mdt_reint_record *rr = &info->mti_rr;
1025         struct req_capsule      *pill = info->mti_pill;
1026         int rc;
1027         ENTRY;
1028
1029         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1030         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1031         if (rec == NULL)
1032                 RETURN(-EFAULT);
1033
1034         uc->mu_fsuid = rec->ul_fsuid;
1035         uc->mu_fsgid = rec->ul_fsgid;
1036         uc->mu_cap   = rec->ul_cap;
1037         uc->mu_suppgids[0] = rec->ul_suppgid1;
1038         uc->mu_suppgids[1] = -1;
1039
1040         attr->la_uid = rec->ul_fsuid;
1041         attr->la_gid = rec->ul_fsgid;
1042         rr->rr_fid1 = &rec->ul_fid1;
1043         rr->rr_fid2 = &rec->ul_fid2;
1044         attr->la_ctime = rec->ul_time;
1045         attr->la_mtime = rec->ul_time;
1046         attr->la_mode  = rec->ul_mode;
1047         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1048
1049         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1050                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1051                                  req_capsule_client_get(pill, &RMF_CAPA1));
1052
1053         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1054         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1055         if (rr->rr_name == NULL || rr->rr_namelen == 0)
1056                 RETURN(-EFAULT);
1057
1058         if (rec->ul_bias & MDS_VTX_BYPASS)
1059                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1060         else
1061                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1062
1063         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1064
1065         rc = mdt_dlmreq_unpack(info);
1066         RETURN(rc);
1067 }
1068
1069 static int mdt_rename_unpack(struct mdt_thread_info *info)
1070 {
1071         struct md_ucred         *uc = mdt_ucred(info);
1072         struct mdt_rec_rename   *rec;
1073         struct md_attr          *ma = &info->mti_attr;
1074         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1075         struct mdt_reint_record *rr = &info->mti_rr;
1076         struct req_capsule      *pill = info->mti_pill;
1077         int rc;
1078         ENTRY;
1079
1080         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1081         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1082         if (rec == NULL)
1083                 RETURN(-EFAULT);
1084
1085         uc->mu_fsuid = rec->rn_fsuid;
1086         uc->mu_fsgid = rec->rn_fsgid;
1087         uc->mu_cap   = rec->rn_cap;
1088         uc->mu_suppgids[0] = rec->rn_suppgid1;
1089         uc->mu_suppgids[1] = rec->rn_suppgid2;
1090
1091         attr->la_uid = rec->rn_fsuid;
1092         attr->la_gid = rec->rn_fsgid;
1093         rr->rr_fid1 = &rec->rn_fid1;
1094         rr->rr_fid2 = &rec->rn_fid2;
1095         attr->la_ctime = rec->rn_time;
1096         attr->la_mtime = rec->rn_time;
1097         /* rename_tgt contains the mode already */
1098         attr->la_mode = rec->rn_mode;
1099         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1100
1101         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1102                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1103                                  req_capsule_client_get(pill, &RMF_CAPA1));
1104         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1105                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1106                                  req_capsule_client_get(pill, &RMF_CAPA2));
1107
1108         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1109         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1110         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1111                 RETURN(-EFAULT);
1112         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1113         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1114         LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1115
1116         if (rec->rn_bias & MDS_VTX_BYPASS)
1117                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1118         else
1119                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1120
1121         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1122
1123         rc = mdt_dlmreq_unpack(info);
1124         RETURN(rc);
1125 }
1126
1127 /*
1128  * please see comment above LOV_MAGIC_V1_DEF
1129  */
1130 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1131 {
1132         struct mdt_reint_record *rr = &info->mti_rr;
1133         struct lov_user_md_v1   *v1;
1134
1135         v1 = (void *)rr->rr_eadata;
1136         LASSERT(v1);
1137
1138         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1139                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1140                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1141                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1142                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1143                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1144                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1145                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1146                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1147                 }
1148         }
1149 }
1150
1151 static int mdt_open_unpack(struct mdt_thread_info *info)
1152 {
1153         struct md_ucred         *uc = mdt_ucred(info);
1154         struct mdt_rec_create   *rec;
1155         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1156         struct req_capsule      *pill = info->mti_pill;
1157         struct mdt_reint_record *rr   = &info->mti_rr;
1158         struct ptlrpc_request   *req  = mdt_info_req(info);
1159         struct md_op_spec       *sp   = &info->mti_spec;
1160         ENTRY;
1161
1162         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1163         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1164         if (rec == NULL)
1165                 RETURN(-EFAULT);
1166
1167         uc->mu_fsuid = rec->cr_fsuid;
1168         uc->mu_fsgid = rec->cr_fsgid;
1169         uc->mu_cap   = rec->cr_cap;
1170         uc->mu_suppgids[0] = rec->cr_suppgid1;
1171         uc->mu_suppgids[1] = rec->cr_suppgid2;
1172
1173         rr->rr_fid1   = &rec->cr_fid1;
1174         rr->rr_fid2   = &rec->cr_fid2;
1175         rr->rr_handle = &rec->cr_old_handle;
1176         attr->la_mode = rec->cr_mode;
1177         attr->la_rdev  = rec->cr_rdev;
1178         attr->la_uid   = rec->cr_fsuid;
1179         attr->la_gid   = rec->cr_fsgid;
1180         attr->la_ctime = rec->cr_time;
1181         attr->la_mtime = rec->cr_time;
1182         attr->la_atime = rec->cr_time;
1183         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1184                          LA_CTIME | LA_MTIME | LA_ATIME;
1185         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1186         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1187         /* Do not trigger ASSERTION if client miss to set such flags. */
1188         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1189                 RETURN(-EPROTO);
1190         info->mti_replayepoch = rec->cr_ioepoch;
1191
1192         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1193
1194         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1195                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1196                                  req_capsule_client_get(pill, &RMF_CAPA1));
1197         if (req_is_replay(req) &&
1198             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1199 #if 0
1200                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1201                                  req_capsule_client_get(pill, &RMF_CAPA2));
1202 #else
1203                 /*
1204                  * FIXME: capa in replay open request might have expired,
1205                  * bypass capa check. Security hole?
1206                  */
1207                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1208                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1209 #endif
1210         }
1211
1212         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1213         if (rr->rr_name == NULL)
1214                 RETURN(-EFAULT);
1215         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1216
1217         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1218                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1219                                                         RCL_CLIENT);
1220                 if (rr->rr_eadatalen > 0) {
1221                         rr->rr_eadata = req_capsule_client_get(pill,
1222                                                                &RMF_EADATA);
1223                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1224                         sp->u.sp_ea.eadata = rr->rr_eadata;
1225                         sp->no_create = !!req_is_replay(req);
1226                         mdt_fix_lov_magic(info);
1227                 }
1228
1229                 /*
1230                  * Client default md_size may be 0 right after client start,
1231                  * until all osc are connected, set here just some reasonable
1232                  * value to prevent misbehavior.
1233                  */
1234                 if (rr->rr_eadatalen == 0 &&
1235                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1236                         rr->rr_eadatalen = MIN_MD_SIZE;
1237         }
1238
1239         RETURN(0);
1240 }
1241
1242 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1243 {
1244         struct mdt_reint_record   *rr   = &info->mti_rr;
1245         struct md_ucred           *uc   = mdt_ucred(info);
1246         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1247         struct req_capsule        *pill = info->mti_pill;
1248         struct mdt_rec_setxattr   *rec;
1249         ENTRY;
1250
1251
1252         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1253                          sizeof(struct mdt_rec_reint));
1254
1255         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1256         if (rec == NULL)
1257                 RETURN(-EFAULT);
1258
1259         uc->mu_fsuid  = rec->sx_fsuid;
1260         uc->mu_fsgid  = rec->sx_fsgid;
1261         uc->mu_cap    = rec->sx_cap;
1262         uc->mu_suppgids[0] = rec->sx_suppgid1;
1263         uc->mu_suppgids[1] = -1;
1264
1265         rr->rr_opcode = rec->sx_opcode;
1266         rr->rr_fid1   = &rec->sx_fid;
1267         attr->la_valid = rec->sx_valid;
1268         attr->la_ctime = rec->sx_time;
1269         attr->la_size = rec->sx_size;
1270         attr->la_flags = rec->sx_flags;
1271
1272         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1273                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1274                                  req_capsule_client_get(pill, &RMF_CAPA1));
1275         else
1276                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1277
1278         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1279         if (rr->rr_name == NULL)
1280                 RETURN(-EFAULT);
1281         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1282         LASSERT(rr->rr_namelen > 0);
1283
1284         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1285                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1286                                                         RCL_CLIENT);
1287                 if (rr->rr_eadatalen > 0) {
1288                         rr->rr_eadata = req_capsule_client_get(pill,
1289                                                                &RMF_EADATA);
1290                         if (rr->rr_eadata == NULL)
1291                                 RETURN(-EFAULT);
1292                 } else {
1293                         rr->rr_eadata = NULL;
1294                 }
1295         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1296                 CDEBUG(D_INFO, "no xattr data supplied\n");
1297                 RETURN(-EFAULT);
1298         }
1299
1300         RETURN(0);
1301 }
1302
1303
1304 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1305
1306 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1307         [REINT_SETATTR]  = mdt_setattr_unpack,
1308         [REINT_CREATE]   = mdt_create_unpack,
1309         [REINT_LINK]     = mdt_link_unpack,
1310         [REINT_UNLINK]   = mdt_unlink_unpack,
1311         [REINT_RENAME]   = mdt_rename_unpack,
1312         [REINT_OPEN]     = mdt_open_unpack,
1313         [REINT_SETXATTR] = mdt_setxattr_unpack
1314 };
1315
1316 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1317 {
1318         int rc;
1319         ENTRY;
1320
1321         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1322         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1323                 info->mti_rr.rr_opcode = op;
1324                 rc = mdt_reint_unpackers[op](info);
1325         } else {
1326                 CERROR("Unexpected opcode %d\n", op);
1327                 rc = -EFAULT;
1328         }
1329         RETURN(rc);
1330 }