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, 2014, 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/nidstr.h>
  53 #include <lustre_nodemap.h>
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 void mdt_exit_ucred(struct mdt_thread_info *info)
  62 {
  63         struct lu_ucred   *uc  = mdt_ucred(info);
  64         struct mdt_device *mdt = info->mti_mdt;
  65
  66         LASSERT(uc != NULL);
  67         if (uc->uc_valid != UCRED_INIT) {
  68                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  69                 if (uc->uc_ginfo) {
  70                         put_group_info(uc->uc_ginfo);
  71                         uc->uc_ginfo = NULL;
  72                 }
  73                 if (uc->uc_identity) {
  74                         mdt_identity_put(mdt->mdt_identity_cache,
  75                                          uc->uc_identity);
  76                         uc->uc_identity = NULL;
  77                 }
  78                 uc->uc_valid = UCRED_INIT;
  79         }
  80 }
  81
  82 static int match_nosquash_list(struct rw_semaphore *sem,
  83                                struct list_head *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         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
  99         ENTRY;
 100
 101         LASSERT(ucred != NULL);
 102         if (!squash->rsi_uid || ucred->uc_fsuid)
 103                 RETURN(0);
 104
 105         if (match_nosquash_list(&squash->rsi_sem,
 106                                 &squash->rsi_nosquash_nids,
 107                                 peernid)) {
 108                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 109                        libcfs_nid2str(peernid));
 110                 RETURN(0);
 111         }
 112
 113         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 114                libcfs_nid2str(peernid),
 115                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 116                squash->rsi_uid, squash->rsi_gid, 0);
 117
 118         ucred->uc_fsuid = squash->rsi_uid;
 119         ucred->uc_fsgid = squash->rsi_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 void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 128 {
 129         struct ptlrpc_request   *req = mdt_info_req(info);
 130         const char              *jobid = mdt_req_get_jobid(req);
 131
 132         /* set jobid if specified. */
 133         if (jobid)
 134                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 135         else
 136                 uc->uc_jobid[0] = '\0';
 137 }
 138
 139 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 140                           void *buf)
 141 {
 142         struct ptlrpc_request   *req = mdt_info_req(info);
 143         struct mdt_device       *mdt = info->mti_mdt;
 144         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 145         struct lu_ucred         *ucred = mdt_ucred(info);
 146         lnet_nid_t               peernid = req->rq_peer.nid;
 147         __u32                    perm = 0;
 148         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 149         int                      setuid;
 150         int                      setgid;
 151         int                      rc = 0;
 152
 153         ENTRY;
 154
 155         LASSERT(req->rq_auth_gss);
 156         LASSERT(!req->rq_auth_usr_mdt);
 157         LASSERT(req->rq_user_desc);
 158         LASSERT(ucred != NULL);
 159
 160         ucred->uc_valid = UCRED_INVALID;
 161
 162         ucred->uc_o_uid   = pud->pud_uid;
 163         ucred->uc_o_gid   = pud->pud_gid;
 164         ucred->uc_o_fsuid = pud->pud_fsuid;
 165         ucred->uc_o_fsgid = pud->pud_fsgid;
 166
 167         if (type == BODY_INIT) {
 168                 struct mdt_body *body = (struct mdt_body *)buf;
 169
 170                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 171                 ucred->uc_suppgids[1] = -1;
 172         }
 173
 174         /* sanity check: we expect the uid which client claimed is true */
 175         if (remote) {
 176                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 177                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 178                         RETURN(-EACCES);
 179                 }
 180
 181                 if (ptlrpc_user_desc_do_idmap(req, pud))
 182                         RETURN(-EACCES);
 183
 184                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 185                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 186                                "while client claims %u:%u/%u:%u\n",
 187                                libcfs_nid2str(peernid), req->rq_auth_uid,
 188                                req->rq_auth_mapped_uid,
 189                                pud->pud_uid, pud->pud_gid,
 190                                pud->pud_fsuid, pud->pud_fsgid);
 191                         RETURN(-EACCES);
 192                 }
 193         } else {
 194                 if (req->rq_auth_uid != pud->pud_uid) {
 195                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 196                                "while client claims %u:%u/%u:%u\n",
 197                                libcfs_nid2str(peernid), req->rq_auth_uid,
 198                                pud->pud_uid, pud->pud_gid,
 199                                pud->pud_fsuid, pud->pud_fsgid);
 200                         RETURN(-EACCES);
 201                 }
 202         }
 203
 204         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 205                 if (remote) {
 206                         CDEBUG(D_SEC, "remote client must run with identity_get "
 207                                "enabled!\n");
 208                         RETURN(-EACCES);
 209                 } else {
 210                         ucred->uc_identity = NULL;
 211                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 212                                CFS_SETGRP_PERM;
 213                 }
 214         } else {
 215                 struct md_identity *identity;
 216
 217                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 218                                             pud->pud_uid);
 219                 if (IS_ERR(identity)) {
 220                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 221                                      !remote)) {
 222                                 ucred->uc_identity = NULL;
 223                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 224                                        CFS_SETGRP_PERM;
 225                         } else {
 226                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 227                                        pud->pud_uid);
 228                                 RETURN(-EACCES);
 229                         }
 230                 } else {
 231                         ucred->uc_identity = identity;
 232                         perm = mdt_identity_get_perm(ucred->uc_identity,
 233                                                      remote, peernid);
 234                 }
 235         }
 236
 237         /* find out the setuid/setgid attempt */
 238         setuid = (pud->pud_uid != pud->pud_fsuid);
 239         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 240                   (ucred->uc_identity &&
 241                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 242
 243         /* check permission of setuid */
 244         if (setuid && !(perm & CFS_SETUID_PERM)) {
 245                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 246                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 247                 GOTO(out, rc = -EACCES);
 248         }
 249
 250         /* check permission of setgid */
 251         if (setgid && !(perm & CFS_SETGID_PERM)) {
 252                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 253                        "from %s\n", pud->pud_uid, pud->pud_gid,
 254                        pud->pud_fsuid, pud->pud_fsgid,
 255                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 256                 GOTO(out, rc = -EACCES);
 257         }
 258
 259         /*
 260          * NB: remote client not allowed to setgroups anyway.
 261          */
 262         if (!remote && perm & CFS_SETGRP_PERM) {
 263                 if (pud->pud_ngroups) {
 264                         /* setgroups for local client */
 265                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 266                         if (!ucred->uc_ginfo) {
 267                                 CERROR("failed to alloc %d groups\n",
 268                                        pud->pud_ngroups);
 269                                 GOTO(out, rc = -ENOMEM);
 270                         }
 271
 272                         lustre_groups_from_list(ucred->uc_ginfo,
 273                                                 pud->pud_groups);
 274                         lustre_groups_sort(ucred->uc_ginfo);
 275                 } else {
 276                         ucred->uc_ginfo = NULL;
 277                 }
 278         } else {
 279                 ucred->uc_suppgids[0] = -1;
 280                 ucred->uc_suppgids[1] = -1;
 281                 ucred->uc_ginfo = NULL;
 282         }
 283
 284         ucred->uc_uid   = pud->pud_uid;
 285         ucred->uc_gid   = pud->pud_gid;
 286         ucred->uc_fsuid = pud->pud_fsuid;
 287         ucred->uc_fsgid = pud->pud_fsgid;
 288
 289         /* process root_squash here. */
 290         mdt_root_squash(info, peernid);
 291
 292         /* remove fs privilege for non-root user. */
 293         if (ucred->uc_fsuid)
 294                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 295         else
 296                 ucred->uc_cap = pud->pud_cap;
 297         if (remote && !(perm & CFS_RMTOWN_PERM))
 298                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 299                                    CFS_CAP_CHOWN_MASK);
 300         ucred->uc_valid = UCRED_NEW;
 301         ucred_set_jobid(info, ucred);
 302
 303         EXIT;
 304
 305 out:
 306         if (rc) {
 307                 if (ucred->uc_ginfo) {
 308                         put_group_info(ucred->uc_ginfo);
 309                         ucred->uc_ginfo = NULL;
 310                 }
 311                 if (ucred->uc_identity) {
 312                         mdt_identity_put(mdt->mdt_identity_cache,
 313                                          ucred->uc_identity);
 314                         ucred->uc_identity = NULL;
 315                 }
 316         }
 317
 318         return rc;
 319 }
 320
 321 /**
 322  * Check whether allow the client to set supplementary group IDs or not.
 323  *
 324  * \param[in] info      pointer to the thread context
 325  * \param[in] uc        pointer to the RPC user descriptor
 326  *
 327  * \retval              true if allow to set supplementary group IDs
 328  * \retval              false for other cases
 329  */
 330 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 331 {
 332         __u32 remote = exp_connect_rmtclient(info->mti_exp);
 333         __u32 perm;
 334
 335         /* 1. If identity_upcall is disabled, then forbid remote client to set
 336          *    supplementary group IDs, but permit local client to do that. */
 337         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) {
 338                 if (remote)
 339                         return false;
 340
 341                 return true;
 342         }
 343
 344         /* 2. If fail to get related identities, then forbid any client to
 345          *    set supplementary group IDs. */
 346         if (uc->uc_identity == NULL)
 347                 return false;
 348
 349         /* 3. Check the permission in the identities. */
 350         perm = mdt_identity_get_perm(uc->uc_identity, remote,
 351                                      mdt_info_req(info)->rq_peer.nid);
 352         if (perm & CFS_SETGRP_PERM)
 353                 return true;
 354
 355         return false;
 356 }
 357
 358 int mdt_check_ucred(struct mdt_thread_info *info)
 359 {
 360         struct ptlrpc_request   *req = mdt_info_req(info);
 361         struct mdt_device       *mdt = info->mti_mdt;
 362         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 363         struct lu_ucred         *ucred = mdt_ucred(info);
 364         struct md_identity      *identity = NULL;
 365         lnet_nid_t               peernid = req->rq_peer.nid;
 366         __u32                    perm = 0;
 367         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 368         int                      setuid;
 369         int                      setgid;
 370         int                      rc = 0;
 371
 372         ENTRY;
 373
 374         LASSERT(ucred != NULL);
 375         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 376                 RETURN(0);
 377
 378         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 379                 RETURN(0);
 380
 381         /* sanity check: if we use strong authentication, we expect the
 382          * uid which client claimed is true */
 383         if (remote) {
 384                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 385                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 386                         RETURN(-EACCES);
 387                 }
 388
 389                 if (ptlrpc_user_desc_do_idmap(req, pud))
 390                         RETURN(-EACCES);
 391
 392                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 393                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 394                                "while client claims %u:%u/%u:%u\n",
 395                                libcfs_nid2str(peernid), req->rq_auth_uid,
 396                                req->rq_auth_mapped_uid,
 397                                pud->pud_uid, pud->pud_gid,
 398                                pud->pud_fsuid, pud->pud_fsgid);
 399                         RETURN(-EACCES);
 400                 }
 401         } else {
 402                 if (req->rq_auth_uid != pud->pud_uid) {
 403                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 404                                "while client claims %u:%u/%u:%u\n",
 405                                libcfs_nid2str(peernid), req->rq_auth_uid,
 406                                pud->pud_uid, pud->pud_gid,
 407                                pud->pud_fsuid, pud->pud_fsgid);
 408                         RETURN(-EACCES);
 409                 }
 410         }
 411
 412         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 413                 if (remote) {
 414                         CDEBUG(D_SEC, "remote client must run with identity_get "
 415                                "enabled!\n");
 416                         RETURN(-EACCES);
 417                 }
 418                 RETURN(0);
 419         }
 420
 421         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 422         if (IS_ERR(identity)) {
 423                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 424                              !remote)) {
 425                         RETURN(0);
 426                 } else {
 427                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 428                                pud->pud_uid);
 429                         RETURN(-EACCES);
 430                }
 431         }
 432
 433         perm = mdt_identity_get_perm(identity, remote, peernid);
 434         /* find out the setuid/setgid attempt */
 435         setuid = (pud->pud_uid != pud->pud_fsuid);
 436         setgid = (pud->pud_gid != pud->pud_fsgid ||
 437                   pud->pud_gid != identity->mi_gid);
 438
 439         /* check permission of setuid */
 440         if (setuid && !(perm & CFS_SETUID_PERM)) {
 441                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 442                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 443                 GOTO(out, rc = -EACCES);
 444         }
 445
 446         /* check permission of setgid */
 447         if (setgid && !(perm & CFS_SETGID_PERM)) {
 448                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 449                        "from %s\n", pud->pud_uid, pud->pud_gid,
 450                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 451                        libcfs_nid2str(peernid));
 452                 GOTO(out, rc = -EACCES);
 453         }
 454
 455         EXIT;
 456
 457 out:
 458         mdt_identity_put(mdt->mdt_identity_cache, identity);
 459         return rc;
 460 }
 461
 462 static void mdt_squash_nodemap_id(struct lu_ucred *ucred,
 463                                   struct lu_nodemap *nodemap)
 464 {
 465         if (ucred->uc_o_uid == nodemap->nm_squash_uid) {
 466                 ucred->uc_fsuid = nodemap->nm_squash_uid;
 467                 ucred->uc_fsgid = nodemap->nm_squash_gid;
 468                 ucred->uc_cap = 0;
 469                 ucred->uc_suppgids[0] = -1;
 470                 ucred->uc_suppgids[1] = -1;
 471         }
 472 }
 473
 474
 475 static int old_init_ucred(struct mdt_thread_info *info,
 476                           struct mdt_body *body)
 477 {
 478         struct lu_ucred         *uc = mdt_ucred(info);
 479         struct mdt_device       *mdt = info->mti_mdt;
 480         struct md_identity      *identity = NULL;
 481         struct lu_nodemap       *nodemap =
 482                 info->mti_exp->exp_target_data.ted_nodemap;
 483         ENTRY;
 484
 485         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 486                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 487         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 488                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 489         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 490                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 491         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 492                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 493
 494         LASSERT(uc != NULL);
 495         uc->uc_valid = UCRED_INVALID;
 496         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 497         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 498         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 499         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 500         uc->uc_suppgids[0] = body->mbo_suppgid;
 501         uc->uc_suppgids[1] = -1;
 502         uc->uc_ginfo = NULL;
 503         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 504                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 505                                             uc->uc_fsuid);
 506                 if (IS_ERR(identity)) {
 507                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 508                                 identity = NULL;
 509                         } else {
 510                                 CDEBUG(D_SEC, "Deny access without identity: "
 511                                        "uid %u\n", uc->uc_fsuid);
 512                                 RETURN(-EACCES);
 513                         }
 514                 }
 515         }
 516         uc->uc_identity = identity;
 517
 518         mdt_squash_nodemap_id(uc, nodemap);
 519
 520         /* process root_squash here. */
 521         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 522
 523         /* remove fs privilege for non-root user. */
 524         if (uc->uc_fsuid)
 525                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 526         else
 527                 uc->uc_cap = body->mbo_capability;
 528         uc->uc_valid = UCRED_OLD;
 529         ucred_set_jobid(info, uc);
 530
 531         RETURN(0);
 532 }
 533
 534 static int old_init_ucred_reint(struct mdt_thread_info *info)
 535 {
 536         struct lu_ucred         *uc = mdt_ucred(info);
 537         struct mdt_device       *mdt = info->mti_mdt;
 538         struct md_identity      *identity = NULL;
 539         struct lu_nodemap       *nodemap =
 540                 info->mti_exp->exp_target_data.ted_nodemap;
 541         ENTRY;
 542
 543         LASSERT(uc != NULL);
 544
 545         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 546                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 547         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 548                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 549
 550         uc->uc_valid = UCRED_INVALID;
 551         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 552         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 553         uc->uc_ginfo = NULL;
 554
 555         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 556                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 557                                             uc->uc_fsuid);
 558                 if (IS_ERR(identity)) {
 559                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 560                                 identity = NULL;
 561                         } else {
 562                                 CDEBUG(D_SEC, "Deny access without identity: "
 563                                        "uid %u\n", uc->uc_fsuid);
 564                                 RETURN(-EACCES);
 565                         }
 566                 }
 567         }
 568         uc->uc_identity = identity;
 569
 570         /* process root_squash here. */
 571         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 572
 573         /* remove fs privilege for non-root user. */
 574         if (uc->uc_fsuid)
 575                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 576         uc->uc_valid = UCRED_OLD;
 577         ucred_set_jobid(info, uc);
 578
 579         RETURN(0);
 580 }
 581
 582 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 583 {
 584         struct ptlrpc_request *req = mdt_info_req(info);
 585         struct lu_ucred       *uc  = mdt_ucred(info);
 586
 587         LASSERT(uc != NULL);
 588         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 589                 return 0;
 590
 591         mdt_exit_ucred(info);
 592
 593         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 594                 return old_init_ucred(info, body);
 595         else
 596                 return new_init_ucred(info, BODY_INIT, body);
 597 }
 598
 599 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 600 {
 601         struct ptlrpc_request *req = mdt_info_req(info);
 602         struct lu_ucred       *uc  = mdt_ucred(info);
 603
 604         LASSERT(uc != NULL);
 605         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 606                 return 0;
 607
 608         mdt_exit_ucred(info);
 609
 610         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 611                 return old_init_ucred_reint(info);
 612         else
 613                 return new_init_ucred(info, REC_INIT, NULL);
 614 }
 615
 616 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 617 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 618 {
 619         const struct lov_ost_data_v1    *lod;
 620         int                              i;
 621         __u16                            count;
 622
 623         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 624                 return;
 625
 626         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 627
 628         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 629                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 630                le32_to_cpu(lmm->lmm_pattern));
 631         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 632                le32_to_cpu(lmm->lmm_stripe_size), count);
 633
 634         /* If it's a directory or a released file, then there are
 635          * no actual objects to print, so bail out. */
 636         if (valid & OBD_MD_FLDIREA ||
 637             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 638                 return;
 639
 640         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 641         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 642                 struct ost_id oi;
 643
 644                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 645                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 646                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 647         }
 648 }
 649
 650 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 651 {
 652         const struct lmv_mds_md_v1 *lmm1;
 653         int                        i;
 654
 655         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 656                 return;
 657
 658         lmm1 = &lmv->lmv_md_v1;
 659         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 660                le32_to_cpu(lmm1->lmv_magic),
 661                le32_to_cpu(lmm1->lmv_master_mdt_index),
 662                le32_to_cpu(lmm1->lmv_stripe_count));
 663
 664         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 665                 return;
 666
 667         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 668                 struct lu_fid fid;
 669
 670                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 671                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 672         }
 673 }
 674
 675 /* Shrink and/or grow reply buffers */
 676 int mdt_fix_reply(struct mdt_thread_info *info)
 677 {
 678         struct req_capsule *pill = info->mti_pill;
 679         struct mdt_body    *body;
 680         int                md_size, md_packed = 0;
 681         int                acl_size;
 682         int                rc = 0;
 683         ENTRY;
 684
 685         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 686         LASSERT(body != NULL);
 687
 688         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 689                                OBD_MD_LINKNAME))
 690                 md_size = body->mbo_eadatasize;
 691         else
 692                 md_size = 0;
 693
 694         acl_size = body->mbo_aclsize;
 695
 696         /* this replay - not send info to client */
 697         if (info->mti_spec.no_create) {
 698                 md_size = 0;
 699                 acl_size = 0;
 700         }
 701
 702         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 703                md_size, acl_size);
 704 /*
 705             &RMF_MDT_BODY,
 706             &RMF_MDT_MD,
 707             &RMF_ACL, or &RMF_LOGCOOKIES
 708 (optional)  &RMF_CAPA1,
 709 (optional)  &RMF_CAPA2,
 710 (optional)  something else
 711 */
 712
 713         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 714          * buffers before growing it */
 715         if (info->mti_big_lmm_used) {
 716                 /* big_lmm buffer may be used even without packing the result
 717                  * into reply, just for internal server needs */
 718                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 719                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 720                                                          RCL_SERVER);
 721
 722                 /* free big lmm if md_size is not needed */
 723                 if (md_size == 0 || md_packed == 0) {
 724                         info->mti_big_lmm_used = 0;
 725                 } else {
 726                         /* buffer must be allocated separately */
 727                         LASSERT(info->mti_attr.ma_lmm !=
 728                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 729                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 730                 }
 731         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 732                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 733         }
 734
 735         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 736                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 737         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 738                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 739                                    acl_size, RCL_SERVER);
 740
 741         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 742             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 743                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 744
 745         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 746             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 747                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 748
 749         /*
 750          * Some more field should be shrinked if needed.
 751          * This should be done by those who added fields to reply message.
 752          */
 753
 754         /* Grow MD buffer if needed finally */
 755         if (info->mti_big_lmm_used) {
 756                 void *lmm;
 757
 758                 LASSERT(md_size > md_packed);
 759                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 760                        md_size - md_packed);
 761                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 762                 if (rc) {
 763                         /* we can't answer with proper LOV EA, drop flags,
 764                          * the rc is also returned so this request is
 765                          * considered as failed */
 766                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 767                         /* don't return transno along with error */
 768                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 769                 } else {
 770                         /* now we need to pack right LOV/LMV EA */
 771                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 772                         if (info->mti_attr.ma_valid & MA_LOV) {
 773                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 774                                                              RCL_SERVER) ==
 775                                                 info->mti_attr.ma_lmm_size);
 776                                 memcpy(lmm, info->mti_attr.ma_lmm,
 777                                        info->mti_attr.ma_lmm_size);
 778                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 779                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 780                                                              RCL_SERVER) ==
 781                                                 info->mti_attr.ma_lmv_size);
 782                                 memcpy(lmm, info->mti_attr.ma_lmv,
 783                                        info->mti_attr.ma_lmv_size);
 784                         }
 785                 }
 786                 /* update mdt_max_mdsize so clients will be aware about that */
 787                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 788                         info->mti_mdt->mdt_max_mdsize =
 789                                                     info->mti_attr.ma_lmm_size;
 790                 info->mti_big_lmm_used = 0;
 791         }
 792         RETURN(rc);
 793 }
 794
 795
 796 /* if object is dying, pack the lov/llog data,
 797  * parameter info->mti_attr should be valid at this point! */
 798 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 799                            const struct md_attr *ma)
 800 {
 801         struct mdt_body       *repbody;
 802         const struct lu_attr *la = &ma->ma_attr;
 803         ENTRY;
 804
 805         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 806         LASSERT(repbody != NULL);
 807
 808         if (ma->ma_valid & MA_INODE)
 809                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 810
 811         if (ma->ma_valid & MA_LOV) {
 812                 CERROR("No need in LOV EA upon unlink\n");
 813                 dump_stack();
 814         }
 815         repbody->mbo_eadatasize = 0;
 816
 817         RETURN(0);
 818 }
 819
 820 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 821                                   struct md_attr *ma)
 822 {
 823         __u64 out;
 824
 825         out = 0;
 826         if (in & MDS_ATTR_MODE)
 827                 out |= LA_MODE;
 828         if (in & MDS_ATTR_UID)
 829                 out |= LA_UID;
 830         if (in & MDS_ATTR_GID)
 831                 out |= LA_GID;
 832         if (in & MDS_ATTR_SIZE)
 833                 out |= LA_SIZE;
 834         if (in & MDS_ATTR_BLOCKS)
 835                 out |= LA_BLOCKS;
 836         if (in & MDS_ATTR_ATIME_SET)
 837                 out |= LA_ATIME;
 838         if (in & MDS_ATTR_CTIME_SET)
 839                 out |= LA_CTIME;
 840         if (in & MDS_ATTR_MTIME_SET)
 841                 out |= LA_MTIME;
 842         if (in & MDS_ATTR_ATTR_FLAG)
 843                 out |= LA_FLAGS;
 844         if (in & MDS_ATTR_KILL_SUID)
 845                 out |= LA_KILL_SUID;
 846         if (in & MDS_ATTR_KILL_SGID)
 847                 out |= LA_KILL_SGID;
 848
 849         if (in & MDS_ATTR_FROM_OPEN)
 850                 rr->rr_flags |= MRF_OPEN_TRUNC;
 851         if (in & MDS_OPEN_OWNEROVERRIDE)
 852                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 853         if (in & MDS_ATTR_FORCE)
 854                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 855
 856         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 857                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 858                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 859                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 860                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 861                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 862         if (in != 0)
 863                 CERROR("Unknown attr bits: "LPX64"\n", in);
 864         return out;
 865 }
 866
 867 /* unpacking */
 868
 869 int mdt_name_unpack(struct req_capsule *pill,
 870                     const struct req_msg_field *field,
 871                     struct lu_name *ln,
 872                     enum mdt_name_flags flags)
 873 {
 874         ln->ln_name = req_capsule_client_get(pill, field);
 875         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 876
 877         if (!lu_name_is_valid(ln)) {
 878                 ln->ln_name = NULL;
 879                 ln->ln_namelen = 0;
 880
 881                 return -EPROTO;
 882         }
 883
 884         if ((flags & MNF_FIX_ANON) &&
 885             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 886                 /* Newer (3.x) kernels use a name of "/" for the
 887                  * "anonymous" disconnected dentries from NFS
 888                  * filehandle conversion. See d_obtain_alias(). */
 889                 ln->ln_name = NULL;
 890                 ln->ln_namelen = 0;
 891         }
 892
 893         return 0;
 894 }
 895
 896 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 897 {
 898         struct lu_ucred         *uc = mdt_ucred(info);
 899         struct md_attr          *ma = &info->mti_attr;
 900         struct lu_attr          *la = &ma->ma_attr;
 901         struct req_capsule      *pill = info->mti_pill;
 902         struct mdt_reint_record *rr = &info->mti_rr;
 903         struct mdt_rec_setattr  *rec;
 904         struct lu_nodemap       *nodemap =
 905                 info->mti_exp->exp_target_data.ted_nodemap;
 906         ENTRY;
 907
 908         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 909         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 910         if (rec == NULL)
 911                 RETURN(-EFAULT);
 912
 913         /* This prior initialization is needed for old_init_ucred_reint() */
 914         uc->uc_fsuid = rec->sa_fsuid;
 915         uc->uc_fsgid = rec->sa_fsgid;
 916         uc->uc_cap   = rec->sa_cap;
 917         uc->uc_suppgids[0] = rec->sa_suppgid;
 918         uc->uc_suppgids[1] = -1;
 919
 920         rr->rr_fid1 = &rec->sa_fid;
 921         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 922         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 923          *  the client does not have OBD_CONNECT_FULL20, convert it
 924          *  to LA_xTIME. LU-3036 */
 925         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 926                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 927                      (rec->sa_valid & MDS_ATTR_ATIME))
 928                         la->la_valid |= LA_ATIME;
 929                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 930                      (rec->sa_valid & MDS_ATTR_MTIME))
 931                         la->la_valid |= LA_MTIME;
 932                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 933                      (rec->sa_valid & MDS_ATTR_CTIME))
 934                         la->la_valid |= LA_CTIME;
 935         }
 936         la->la_mode  = rec->sa_mode;
 937         la->la_flags = rec->sa_attr_flags;
 938         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 939                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 940         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 941                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 942         la->la_size  = rec->sa_size;
 943         la->la_blocks = rec->sa_blocks;
 944         la->la_ctime = rec->sa_ctime;
 945         la->la_atime = rec->sa_atime;
 946         la->la_mtime = rec->sa_mtime;
 947         ma->ma_valid = MA_INODE;
 948
 949         if (rec->sa_bias & MDS_DATA_MODIFIED)
 950                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 951         else
 952                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 953
 954         if (rec->sa_bias & MDS_HSM_RELEASE)
 955                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 956         else
 957                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 958
 959         RETURN(0);
 960 }
 961
 962 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 963 {
 964         struct req_capsule *pill = info->mti_pill;
 965         struct mdt_ioepoch *ioepoch;
 966         ENTRY;
 967
 968         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 969                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 970         else
 971                 ioepoch = NULL;
 972
 973         if (ioepoch == NULL)
 974                 RETURN(-EPROTO);
 975
 976         info->mti_close_handle = ioepoch->handle;
 977
 978         RETURN(0);
 979 }
 980
 981 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 982         struct req_capsule      *pill = info->mti_pill;
 983
 984         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 985                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 986                 if (info->mti_dlm_req == NULL)
 987                         RETURN(-EFAULT);
 988         }
 989
 990         RETURN(0);
 991 }
 992
 993 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 994 {
 995         struct mdt_reint_record *rr = &info->mti_rr;
 996         struct md_attr          *ma = &info->mti_attr;
 997         struct req_capsule      *pill = info->mti_pill;
 998         int rc;
 999         ENTRY;
1000
1001         rc = mdt_setattr_unpack_rec(info);
1002         if (rc)
1003                 RETURN(rc);
1004
1005         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1006                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1007                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1008                                                         RCL_CLIENT);
1009                 if (rr->rr_eadatalen > 0) {
1010                         const struct lmv_user_md        *lum;
1011
1012                         lum = rr->rr_eadata;
1013                         /* Sigh ma_valid(from req) does not indicate whether
1014                          * it will set LOV/LMV EA, so we have to check magic */
1015                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1016                                 ma->ma_valid |= MA_LMV;
1017                                 ma->ma_lmv = (void *)rr->rr_eadata;
1018                                 ma->ma_lmv_size = rr->rr_eadatalen;
1019                         } else {
1020                                 ma->ma_valid |= MA_LOV;
1021                                 ma->ma_lmm = (void *)rr->rr_eadata;
1022                                 ma->ma_lmm_size = rr->rr_eadatalen;
1023                         }
1024                 }
1025         }
1026
1027         rc = mdt_dlmreq_unpack(info);
1028         RETURN(rc);
1029 }
1030
1031 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1032 {
1033         struct md_attr          *ma = &info->mti_attr;
1034         struct req_capsule      *pill = info->mti_pill;
1035         ENTRY;
1036
1037         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1038                 RETURN(0);
1039
1040         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1041
1042         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1043             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1044                 RETURN(-EFAULT);
1045
1046         RETURN(0);
1047 }
1048
1049 int mdt_close_unpack(struct mdt_thread_info *info)
1050 {
1051         int rc;
1052         ENTRY;
1053
1054         rc = mdt_close_handle_unpack(info);
1055         if (rc)
1056                 RETURN(rc);
1057
1058         rc = mdt_setattr_unpack_rec(info);
1059         if (rc)
1060                 RETURN(rc);
1061
1062         rc = mdt_hsm_release_unpack(info);
1063         if (rc)
1064                 RETURN(rc);
1065
1066         RETURN(mdt_init_ucred_reint(info));
1067 }
1068
1069 static int mdt_create_unpack(struct mdt_thread_info *info)
1070 {
1071         struct lu_ucred         *uc  = mdt_ucred(info);
1072         struct mdt_rec_create   *rec;
1073         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1074         struct mdt_reint_record *rr = &info->mti_rr;
1075         struct req_capsule      *pill = info->mti_pill;
1076         struct md_op_spec       *sp = &info->mti_spec;
1077         int rc;
1078         ENTRY;
1079
1080         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1081         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1082         if (rec == NULL)
1083                 RETURN(-EFAULT);
1084
1085         /* This prior initialization is needed for old_init_ucred_reint() */
1086         uc->uc_fsuid = rec->cr_fsuid;
1087         uc->uc_fsgid = rec->cr_fsgid;
1088         uc->uc_cap   = rec->cr_cap;
1089         uc->uc_suppgids[0] = rec->cr_suppgid1;
1090         uc->uc_suppgids[1] = -1;
1091         uc->uc_umask = rec->cr_umask;
1092
1093         rr->rr_fid1 = &rec->cr_fid1;
1094         rr->rr_fid2 = &rec->cr_fid2;
1095         attr->la_mode = rec->cr_mode;
1096         attr->la_rdev  = rec->cr_rdev;
1097         attr->la_uid   = rec->cr_fsuid;
1098         attr->la_gid   = rec->cr_fsgid;
1099         attr->la_ctime = rec->cr_time;
1100         attr->la_mtime = rec->cr_time;
1101         attr->la_atime = rec->cr_time;
1102         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1103                          LA_CTIME | LA_MTIME | LA_ATIME;
1104         memset(&sp->u, 0, sizeof(sp->u));
1105         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1106
1107         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1108         if (rc < 0)
1109                 RETURN(rc);
1110
1111         if (S_ISLNK(attr->la_mode)) {
1112                 const char *tgt = NULL;
1113
1114                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1115                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1116                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1117                         sp->u.sp_symname = tgt;
1118                 }
1119                 if (tgt == NULL)
1120                         RETURN(-EFAULT);
1121         } else {
1122                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1123                 if (S_ISDIR(attr->la_mode) &&
1124                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1125                         sp->u.sp_ea.eadata =
1126                                 req_capsule_client_get(pill, &RMF_EADATA);
1127                         sp->u.sp_ea.eadatalen =
1128                                 req_capsule_get_size(pill, &RMF_EADATA,
1129                                                      RCL_CLIENT);
1130                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1131                 }
1132         }
1133
1134         rc = mdt_dlmreq_unpack(info);
1135         RETURN(rc);
1136 }
1137
1138 static int mdt_link_unpack(struct mdt_thread_info *info)
1139 {
1140         struct lu_ucred         *uc  = mdt_ucred(info);
1141         struct mdt_rec_link     *rec;
1142         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1143         struct mdt_reint_record *rr = &info->mti_rr;
1144         struct req_capsule      *pill = info->mti_pill;
1145         int rc;
1146         ENTRY;
1147
1148         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1149         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1150         if (rec == NULL)
1151                 RETURN(-EFAULT);
1152
1153         /* This prior initialization is needed for old_init_ucred_reint() */
1154         uc->uc_fsuid = rec->lk_fsuid;
1155         uc->uc_fsgid = rec->lk_fsgid;
1156         uc->uc_cap   = rec->lk_cap;
1157         uc->uc_suppgids[0] = rec->lk_suppgid1;
1158         uc->uc_suppgids[1] = rec->lk_suppgid2;
1159
1160         attr->la_uid = rec->lk_fsuid;
1161         attr->la_gid = rec->lk_fsgid;
1162         rr->rr_fid1 = &rec->lk_fid1;
1163         rr->rr_fid2 = &rec->lk_fid2;
1164         attr->la_ctime = rec->lk_time;
1165         attr->la_mtime = rec->lk_time;
1166         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1167
1168         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1169         if (rc < 0)
1170                 RETURN(rc);
1171
1172         rc = mdt_dlmreq_unpack(info);
1173
1174         RETURN(rc);
1175 }
1176
1177 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1178 {
1179         struct lu_ucred         *uc  = mdt_ucred(info);
1180         struct mdt_rec_unlink   *rec;
1181         struct md_attr          *ma = &info->mti_attr;
1182         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1183         struct mdt_reint_record *rr = &info->mti_rr;
1184         struct req_capsule      *pill = info->mti_pill;
1185         int rc;
1186         ENTRY;
1187
1188         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1189         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1190         if (rec == NULL)
1191                 RETURN(-EFAULT);
1192
1193         /* This prior initialization is needed for old_init_ucred_reint() */
1194         uc->uc_fsuid = rec->ul_fsuid;
1195         uc->uc_fsgid = rec->ul_fsgid;
1196         uc->uc_cap   = rec->ul_cap;
1197         uc->uc_suppgids[0] = rec->ul_suppgid1;
1198         uc->uc_suppgids[1] = -1;
1199
1200         attr->la_uid = rec->ul_fsuid;
1201         attr->la_gid = rec->ul_fsgid;
1202         rr->rr_fid1 = &rec->ul_fid1;
1203         rr->rr_fid2 = &rec->ul_fid2;
1204         attr->la_ctime = rec->ul_time;
1205         attr->la_mtime = rec->ul_time;
1206         attr->la_mode  = rec->ul_mode;
1207         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1208
1209         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1210         if (rc < 0)
1211                 RETURN(rc);
1212
1213         if (rec->ul_bias & MDS_VTX_BYPASS)
1214                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1215         else
1216                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1217
1218         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1219
1220         rc = mdt_dlmreq_unpack(info);
1221         RETURN(rc);
1222 }
1223
1224 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1225 {
1226         info->mti_spec.sp_rm_entry = 1;
1227         return mdt_unlink_unpack(info);
1228 }
1229
1230 static int mdt_rename_unpack(struct mdt_thread_info *info)
1231 {
1232         struct lu_ucred         *uc = mdt_ucred(info);
1233         struct mdt_rec_rename   *rec;
1234         struct md_attr          *ma = &info->mti_attr;
1235         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1236         struct mdt_reint_record *rr = &info->mti_rr;
1237         struct req_capsule      *pill = info->mti_pill;
1238         int rc;
1239         ENTRY;
1240
1241         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1242         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1243         if (rec == NULL)
1244                 RETURN(-EFAULT);
1245
1246         /* This prior initialization is needed for old_init_ucred_reint() */
1247         uc->uc_fsuid = rec->rn_fsuid;
1248         uc->uc_fsgid = rec->rn_fsgid;
1249         uc->uc_cap   = rec->rn_cap;
1250         uc->uc_suppgids[0] = rec->rn_suppgid1;
1251         uc->uc_suppgids[1] = rec->rn_suppgid2;
1252
1253         attr->la_uid = rec->rn_fsuid;
1254         attr->la_gid = rec->rn_fsgid;
1255         rr->rr_fid1 = &rec->rn_fid1;
1256         rr->rr_fid2 = &rec->rn_fid2;
1257         attr->la_ctime = rec->rn_time;
1258         attr->la_mtime = rec->rn_time;
1259         /* rename_tgt contains the mode already */
1260         attr->la_mode = rec->rn_mode;
1261         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1262
1263         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1264         if (rc < 0)
1265                 RETURN(rc);
1266
1267         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1268         if (rc < 0)
1269                 RETURN(rc);
1270
1271         if (rec->rn_bias & MDS_VTX_BYPASS)
1272                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1273         else
1274                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1275
1276         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1277
1278
1279         rc = mdt_dlmreq_unpack(info);
1280
1281         RETURN(rc);
1282 }
1283
1284 /*
1285  * please see comment above LOV_MAGIC_V1_DEF
1286  */
1287 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1288 {
1289         struct mdt_reint_record *rr = &info->mti_rr;
1290         struct lov_user_md_v1   *v1;
1291
1292         v1 = (void *)rr->rr_eadata;
1293         LASSERT(v1);
1294
1295         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1296                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1297                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1298                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1299                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1300                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1301                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1302                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1303                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1304                 }
1305         }
1306 }
1307
1308 static int mdt_open_unpack(struct mdt_thread_info *info)
1309 {
1310         struct lu_ucred         *uc = mdt_ucred(info);
1311         struct mdt_rec_create   *rec;
1312         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1313         struct req_capsule      *pill = info->mti_pill;
1314         struct mdt_reint_record *rr   = &info->mti_rr;
1315         struct ptlrpc_request   *req  = mdt_info_req(info);
1316         struct md_op_spec       *sp   = &info->mti_spec;
1317         ENTRY;
1318
1319         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1320         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1321         if (rec == NULL)
1322                 RETURN(-EFAULT);
1323
1324         /* This prior initialization is needed for old_init_ucred_reint() */
1325         uc->uc_fsuid = rec->cr_fsuid;
1326         uc->uc_fsgid = rec->cr_fsgid;
1327         uc->uc_cap   = rec->cr_cap;
1328         uc->uc_suppgids[0] = rec->cr_suppgid1;
1329         uc->uc_suppgids[1] = rec->cr_suppgid2;
1330         uc->uc_umask = rec->cr_umask;
1331
1332         rr->rr_fid1   = &rec->cr_fid1;
1333         rr->rr_fid2   = &rec->cr_fid2;
1334         rr->rr_handle = &rec->cr_old_handle;
1335         attr->la_mode = rec->cr_mode;
1336         attr->la_rdev  = rec->cr_rdev;
1337         attr->la_uid   = rec->cr_fsuid;
1338         attr->la_gid   = rec->cr_fsgid;
1339         attr->la_ctime = rec->cr_time;
1340         attr->la_mtime = rec->cr_time;
1341         attr->la_atime = rec->cr_time;
1342         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1343                          LA_CTIME | LA_MTIME | LA_ATIME;
1344         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1345         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1346         /* Do not trigger ASSERTION if client miss to set such flags. */
1347         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1348                 RETURN(-EPROTO);
1349
1350         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1351
1352         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
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                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1361                         sp->u.sp_ea.eadata = rr->rr_eadata;
1362                         sp->no_create = !!req_is_replay(req);
1363                         mdt_fix_lov_magic(info);
1364                 }
1365
1366                 /*
1367                  * Client default md_size may be 0 right after client start,
1368                  * until all osc are connected, set here just some reasonable
1369                  * value to prevent misbehavior.
1370                  */
1371                 if (rr->rr_eadatalen == 0 &&
1372                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1373                         rr->rr_eadatalen = MIN_MD_SIZE;
1374         }
1375
1376         RETURN(0);
1377 }
1378
1379 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1380 {
1381         struct mdt_reint_record *rr     = &info->mti_rr;
1382         struct lu_ucred         *uc     = mdt_ucred(info);
1383         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1384         struct req_capsule      *pill   = info->mti_pill;
1385         struct mdt_rec_setxattr *rec;
1386         int                      rc;
1387         ENTRY;
1388
1389
1390         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1391                          sizeof(struct mdt_rec_reint));
1392
1393         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1394         if (rec == NULL)
1395                 RETURN(-EFAULT);
1396
1397         /* This prior initialization is needed for old_init_ucred_reint() */
1398         uc->uc_fsuid  = rec->sx_fsuid;
1399         uc->uc_fsgid  = rec->sx_fsgid;
1400         uc->uc_cap    = rec->sx_cap;
1401         uc->uc_suppgids[0] = rec->sx_suppgid1;
1402         uc->uc_suppgids[1] = -1;
1403
1404         rr->rr_opcode = rec->sx_opcode;
1405         rr->rr_fid1   = &rec->sx_fid;
1406         attr->la_valid = rec->sx_valid;
1407         attr->la_ctime = rec->sx_time;
1408         attr->la_size = rec->sx_size;
1409         attr->la_flags = rec->sx_flags;
1410
1411         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1412         if (rc < 0)
1413                 RETURN(rc);
1414
1415         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1416                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1417                                                         RCL_CLIENT);
1418                 if (rr->rr_eadatalen > 0) {
1419                         rr->rr_eadata = req_capsule_client_get(pill,
1420                                                                &RMF_EADATA);
1421                         if (rr->rr_eadata == NULL)
1422                                 RETURN(-EFAULT);
1423                 } else {
1424                         rr->rr_eadata = NULL;
1425                 }
1426         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1427                 CDEBUG(D_INFO, "no xattr data supplied\n");
1428                 RETURN(-EFAULT);
1429         }
1430
1431         if (mdt_dlmreq_unpack(info) < 0)
1432                 RETURN(-EPROTO);
1433
1434         RETURN(0);
1435 }
1436
1437
1438 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1439
1440 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1441         [REINT_SETATTR]  = mdt_setattr_unpack,
1442         [REINT_CREATE]   = mdt_create_unpack,
1443         [REINT_LINK]     = mdt_link_unpack,
1444         [REINT_UNLINK]   = mdt_unlink_unpack,
1445         [REINT_RENAME]   = mdt_rename_unpack,
1446         [REINT_OPEN]     = mdt_open_unpack,
1447         [REINT_SETXATTR] = mdt_setxattr_unpack,
1448         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1449         [REINT_MIGRATE]  = mdt_rename_unpack,
1450 };
1451
1452 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1453 {
1454         int rc;
1455         ENTRY;
1456
1457         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1458         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1459                 info->mti_rr.rr_opcode = op;
1460                 rc = mdt_reint_unpackers[op](info);
1461         } else {
1462                 CERROR("Unexpected opcode %d\n", op);
1463                 rc = -EFAULT;
1464         }
1465         RETURN(rc);
1466 }