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"
 703                " MDSCAPA = %llx, OSSCAPA = %llx\n",
 704                md_size, acl_size,
 705                (unsigned long long)(body->mbo_valid & OBD_MD_FLMDSCAPA),
 706                (unsigned long long)(body->mbo_valid & OBD_MD_FLOSSCAPA));
 707 /*
 708             &RMF_MDT_BODY,
 709             &RMF_MDT_MD,
 710             &RMF_ACL, or &RMF_LOGCOOKIES
 711 (optional)  &RMF_CAPA1,
 712 (optional)  &RMF_CAPA2,
 713 (optional)  something else
 714 */
 715
 716         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 717          * buffers before growing it */
 718         if (info->mti_big_lmm_used) {
 719                 /* big_lmm buffer may be used even without packing the result
 720                  * into reply, just for internal server needs */
 721                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 722                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 723                                                          RCL_SERVER);
 724
 725                 /* free big lmm if md_size is not needed */
 726                 if (md_size == 0 || md_packed == 0) {
 727                         info->mti_big_lmm_used = 0;
 728                 } else {
 729                         /* buffer must be allocated separately */
 730                         LASSERT(info->mti_attr.ma_lmm !=
 731                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 732                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 733                 }
 734         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 735                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 736         }
 737
 738         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 739                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 740         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 741                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 742                                    acl_size, RCL_SERVER);
 743
 744         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 745             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 746                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 747
 748         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 749             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 750                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 751
 752         /*
 753          * Some more field should be shrinked if needed.
 754          * This should be done by those who added fields to reply message.
 755          */
 756
 757         /* Grow MD buffer if needed finally */
 758         if (info->mti_big_lmm_used) {
 759                 void *lmm;
 760
 761                 LASSERT(md_size > md_packed);
 762                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 763                        md_size - md_packed);
 764                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 765                 if (rc) {
 766                         /* we can't answer with proper LOV EA, drop flags,
 767                          * the rc is also returned so this request is
 768                          * considered as failed */
 769                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 770                         /* don't return transno along with error */
 771                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 772                 } else {
 773                         /* now we need to pack right LOV/LMV EA */
 774                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 775                         if (info->mti_attr.ma_valid & MA_LOV) {
 776                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 777                                                              RCL_SERVER) ==
 778                                                 info->mti_attr.ma_lmm_size);
 779                                 memcpy(lmm, info->mti_attr.ma_lmm,
 780                                        info->mti_attr.ma_lmm_size);
 781                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 782                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 783                                                              RCL_SERVER) ==
 784                                                 info->mti_attr.ma_lmv_size);
 785                                 memcpy(lmm, info->mti_attr.ma_lmv,
 786                                        info->mti_attr.ma_lmv_size);
 787                         }
 788                 }
 789                 /* update mdt_max_mdsize so clients will be aware about that */
 790                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 791                         info->mti_mdt->mdt_max_mdsize =
 792                                                     info->mti_attr.ma_lmm_size;
 793                 info->mti_big_lmm_used = 0;
 794         }
 795         RETURN(rc);
 796 }
 797
 798
 799 /* if object is dying, pack the lov/llog data,
 800  * parameter info->mti_attr should be valid at this point! */
 801 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 802                            const struct md_attr *ma)
 803 {
 804         struct mdt_body       *repbody;
 805         const struct lu_attr *la = &ma->ma_attr;
 806         int rc;
 807         ENTRY;
 808
 809         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 810         LASSERT(repbody != NULL);
 811
 812         if (ma->ma_valid & MA_INODE)
 813                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 814
 815         if (ma->ma_valid & MA_LOV) {
 816                 CERROR("No need in LOV EA upon unlink\n");
 817                 dump_stack();
 818         }
 819         repbody->mbo_eadatasize = 0;
 820
 821         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 822             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 823             repbody->mbo_valid & OBD_MD_FLEASIZE) {
 824                 struct lustre_capa *capa;
 825
 826                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 827                 LASSERT(capa);
 828                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 829                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 830                 if (rc)
 831                         RETURN(rc);
 832
 833                 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
 834         }
 835
 836         RETURN(0);
 837 }
 838
 839 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 840                                   struct md_attr *ma)
 841 {
 842         __u64 out;
 843
 844         out = 0;
 845         if (in & MDS_ATTR_MODE)
 846                 out |= LA_MODE;
 847         if (in & MDS_ATTR_UID)
 848                 out |= LA_UID;
 849         if (in & MDS_ATTR_GID)
 850                 out |= LA_GID;
 851         if (in & MDS_ATTR_SIZE)
 852                 out |= LA_SIZE;
 853         if (in & MDS_ATTR_BLOCKS)
 854                 out |= LA_BLOCKS;
 855         if (in & MDS_ATTR_ATIME_SET)
 856                 out |= LA_ATIME;
 857         if (in & MDS_ATTR_CTIME_SET)
 858                 out |= LA_CTIME;
 859         if (in & MDS_ATTR_MTIME_SET)
 860                 out |= LA_MTIME;
 861         if (in & MDS_ATTR_ATTR_FLAG)
 862                 out |= LA_FLAGS;
 863         if (in & MDS_ATTR_KILL_SUID)
 864                 out |= LA_KILL_SUID;
 865         if (in & MDS_ATTR_KILL_SGID)
 866                 out |= LA_KILL_SGID;
 867
 868         if (in & MDS_ATTR_FROM_OPEN)
 869                 rr->rr_flags |= MRF_OPEN_TRUNC;
 870         if (in & MDS_OPEN_OWNEROVERRIDE)
 871                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 872         if (in & MDS_ATTR_FORCE)
 873                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 874
 875         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 876                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 877                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 878                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 879                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 880                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 881         if (in != 0)
 882                 CERROR("Unknown attr bits: "LPX64"\n", in);
 883         return out;
 884 }
 885
 886 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 887                       const struct lu_fid *fid, struct lustre_capa *capa)
 888 {
 889         struct lu_capainfo *lci;
 890
 891         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 892         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 893             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 894                 return;
 895
 896         lci = lu_capainfo_get(info->mti_env);
 897         LASSERT(lci);
 898         lci->lci_fid[offset]  = *fid;
 899         lci->lci_capa[offset] = capa;
 900 }
 901
 902 #ifdef DEBUG_CAPA
 903 void mdt_dump_capainfo(struct mdt_thread_info *info)
 904 {
 905         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 906         int i;
 907
 908         if (lci == NULL)
 909                 return;
 910
 911         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 912                 if (lci->lci_capa[i] == NULL) {
 913                         CERROR("no capa for index %d "DFID"\n",
 914                                i, PFID(&lci->lci_fid[i]));
 915                         continue;
 916                 }
 917                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 918                         CERROR("bypass for index %d "DFID"\n",
 919                                i, PFID(&lci->lci_fid[i]));
 920                         continue;
 921                 }
 922                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 923         }
 924 }
 925 #endif /* DEBUG_CAPA */
 926
 927 /* unpacking */
 928
 929 int mdt_name_unpack(struct req_capsule *pill,
 930                     const struct req_msg_field *field,
 931                     struct lu_name *ln,
 932                     enum mdt_name_flags flags)
 933 {
 934         ln->ln_name = req_capsule_client_get(pill, field);
 935         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 936
 937         if (!lu_name_is_valid(ln)) {
 938                 ln->ln_name = NULL;
 939                 ln->ln_namelen = 0;
 940
 941                 return -EPROTO;
 942         }
 943
 944         if ((flags & MNF_FIX_ANON) &&
 945             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 946                 /* Newer (3.x) kernels use a name of "/" for the
 947                  * "anonymous" disconnected dentries from NFS
 948                  * filehandle conversion. See d_obtain_alias(). */
 949                 ln->ln_name = NULL;
 950                 ln->ln_namelen = 0;
 951         }
 952
 953         return 0;
 954 }
 955
 956 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 957 {
 958         struct lu_ucred         *uc = mdt_ucred(info);
 959         struct md_attr          *ma = &info->mti_attr;
 960         struct lu_attr          *la = &ma->ma_attr;
 961         struct req_capsule      *pill = info->mti_pill;
 962         struct mdt_reint_record *rr = &info->mti_rr;
 963         struct mdt_rec_setattr  *rec;
 964         struct lu_nodemap       *nodemap =
 965                 info->mti_exp->exp_target_data.ted_nodemap;
 966         ENTRY;
 967
 968         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 969         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 970         if (rec == NULL)
 971                 RETURN(-EFAULT);
 972
 973         /* This prior initialization is needed for old_init_ucred_reint() */
 974         uc->uc_fsuid = rec->sa_fsuid;
 975         uc->uc_fsgid = rec->sa_fsgid;
 976         uc->uc_cap   = rec->sa_cap;
 977         uc->uc_suppgids[0] = rec->sa_suppgid;
 978         uc->uc_suppgids[1] = -1;
 979
 980         rr->rr_fid1 = &rec->sa_fid;
 981         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 982         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 983          *  the client does not have OBD_CONNECT_FULL20, convert it
 984          *  to LA_xTIME. LU-3036 */
 985         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 986                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 987                      (rec->sa_valid & MDS_ATTR_ATIME))
 988                         la->la_valid |= LA_ATIME;
 989                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 990                      (rec->sa_valid & MDS_ATTR_MTIME))
 991                         la->la_valid |= LA_MTIME;
 992                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 993                      (rec->sa_valid & MDS_ATTR_CTIME))
 994                         la->la_valid |= LA_CTIME;
 995         }
 996         la->la_mode  = rec->sa_mode;
 997         la->la_flags = rec->sa_attr_flags;
 998         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 999                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1000         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1001                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1002         la->la_size  = rec->sa_size;
1003         la->la_blocks = rec->sa_blocks;
1004         la->la_ctime = rec->sa_ctime;
1005         la->la_atime = rec->sa_atime;
1006         la->la_mtime = rec->sa_mtime;
1007         ma->ma_valid = MA_INODE;
1008
1009         if (rec->sa_bias & MDS_DATA_MODIFIED)
1010                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
1011         else
1012                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
1013
1014         if (rec->sa_bias & MDS_HSM_RELEASE)
1015                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
1016         else
1017                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
1018
1019         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1020                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1021                                  req_capsule_client_get(pill, &RMF_CAPA1));
1022
1023         RETURN(0);
1024 }
1025
1026 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
1027 {
1028         struct req_capsule *pill = info->mti_pill;
1029         ENTRY;
1030
1031         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1032                 info->mti_ioepoch =
1033                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1034         else
1035                 info->mti_ioepoch = NULL;
1036         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
1037 }
1038
1039 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1040         struct req_capsule      *pill = info->mti_pill;
1041
1042         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1043                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1044                 if (info->mti_dlm_req == NULL)
1045                         RETURN(-EFAULT);
1046         }
1047
1048         RETURN(0);
1049 }
1050
1051 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1052 {
1053         struct mdt_reint_record *rr = &info->mti_rr;
1054         struct md_attr          *ma = &info->mti_attr;
1055         struct req_capsule      *pill = info->mti_pill;
1056         int rc;
1057         ENTRY;
1058
1059         rc = mdt_setattr_unpack_rec(info);
1060         if (rc)
1061                 RETURN(rc);
1062
1063         /* Epoch may be absent */
1064         mdt_ioepoch_unpack(info);
1065
1066         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1067                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1068                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1069                                                         RCL_CLIENT);
1070                 if (rr->rr_eadatalen > 0) {
1071                         const struct lmv_user_md        *lum;
1072
1073                         lum = rr->rr_eadata;
1074                         /* Sigh ma_valid(from req) does not indicate whether
1075                          * it will set LOV/LMV EA, so we have to check magic */
1076                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1077                                 ma->ma_valid |= MA_LMV;
1078                                 ma->ma_lmv = (void *)rr->rr_eadata;
1079                                 ma->ma_lmv_size = rr->rr_eadatalen;
1080                         } else {
1081                                 ma->ma_valid |= MA_LOV;
1082                                 ma->ma_lmm = (void *)rr->rr_eadata;
1083                                 ma->ma_lmm_size = rr->rr_eadatalen;
1084                         }
1085                 }
1086         }
1087
1088         rc = mdt_dlmreq_unpack(info);
1089         RETURN(rc);
1090 }
1091
1092 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1093 {
1094         struct md_attr          *ma = &info->mti_attr;
1095         struct req_capsule      *pill = info->mti_pill;
1096         ENTRY;
1097
1098         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1099                 RETURN(0);
1100
1101         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1102
1103         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1104             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1105                 RETURN(-EFAULT);
1106
1107         RETURN(0);
1108 }
1109
1110 int mdt_close_unpack(struct mdt_thread_info *info)
1111 {
1112         int rc;
1113         ENTRY;
1114
1115         rc = mdt_ioepoch_unpack(info);
1116         if (rc)
1117                 RETURN(rc);
1118
1119         rc = mdt_setattr_unpack_rec(info);
1120         if (rc)
1121                 RETURN(rc);
1122
1123         rc = mdt_hsm_release_unpack(info);
1124         if (rc)
1125                 RETURN(rc);
1126
1127         RETURN(mdt_init_ucred_reint(info));
1128 }
1129
1130 static int mdt_create_unpack(struct mdt_thread_info *info)
1131 {
1132         struct lu_ucred         *uc  = mdt_ucred(info);
1133         struct mdt_rec_create   *rec;
1134         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1135         struct mdt_reint_record *rr = &info->mti_rr;
1136         struct req_capsule      *pill = info->mti_pill;
1137         struct md_op_spec       *sp = &info->mti_spec;
1138         int rc;
1139         ENTRY;
1140
1141         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1142         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1143         if (rec == NULL)
1144                 RETURN(-EFAULT);
1145
1146         /* This prior initialization is needed for old_init_ucred_reint() */
1147         uc->uc_fsuid = rec->cr_fsuid;
1148         uc->uc_fsgid = rec->cr_fsgid;
1149         uc->uc_cap   = rec->cr_cap;
1150         uc->uc_suppgids[0] = rec->cr_suppgid1;
1151         uc->uc_suppgids[1] = -1;
1152         uc->uc_umask = rec->cr_umask;
1153
1154         rr->rr_fid1 = &rec->cr_fid1;
1155         rr->rr_fid2 = &rec->cr_fid2;
1156         attr->la_mode = rec->cr_mode;
1157         attr->la_rdev  = rec->cr_rdev;
1158         attr->la_uid   = rec->cr_fsuid;
1159         attr->la_gid   = rec->cr_fsgid;
1160         attr->la_ctime = rec->cr_time;
1161         attr->la_mtime = rec->cr_time;
1162         attr->la_atime = rec->cr_time;
1163         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1164                          LA_CTIME | LA_MTIME | LA_ATIME;
1165         memset(&sp->u, 0, sizeof(sp->u));
1166         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1167
1168         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1169                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1170                                  req_capsule_client_get(pill, &RMF_CAPA1));
1171         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1172
1173         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1174         if (rc < 0)
1175                 RETURN(rc);
1176
1177         if (S_ISLNK(attr->la_mode)) {
1178                 const char *tgt = NULL;
1179
1180                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1181                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1182                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1183                         sp->u.sp_symname = tgt;
1184                 }
1185                 if (tgt == NULL)
1186                         RETURN(-EFAULT);
1187         } else {
1188                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1189                 if (S_ISDIR(attr->la_mode) &&
1190                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1191                         sp->u.sp_ea.eadata =
1192                                 req_capsule_client_get(pill, &RMF_EADATA);
1193                         sp->u.sp_ea.eadatalen =
1194                                 req_capsule_get_size(pill, &RMF_EADATA,
1195                                                      RCL_CLIENT);
1196                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1197                 }
1198         }
1199
1200         rc = mdt_dlmreq_unpack(info);
1201         RETURN(rc);
1202 }
1203
1204 static int mdt_link_unpack(struct mdt_thread_info *info)
1205 {
1206         struct lu_ucred         *uc  = mdt_ucred(info);
1207         struct mdt_rec_link     *rec;
1208         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1209         struct mdt_reint_record *rr = &info->mti_rr;
1210         struct req_capsule      *pill = info->mti_pill;
1211         int rc;
1212         ENTRY;
1213
1214         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1215         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1216         if (rec == NULL)
1217                 RETURN(-EFAULT);
1218
1219         /* This prior initialization is needed for old_init_ucred_reint() */
1220         uc->uc_fsuid = rec->lk_fsuid;
1221         uc->uc_fsgid = rec->lk_fsgid;
1222         uc->uc_cap   = rec->lk_cap;
1223         uc->uc_suppgids[0] = rec->lk_suppgid1;
1224         uc->uc_suppgids[1] = rec->lk_suppgid2;
1225
1226         attr->la_uid = rec->lk_fsuid;
1227         attr->la_gid = rec->lk_fsgid;
1228         rr->rr_fid1 = &rec->lk_fid1;
1229         rr->rr_fid2 = &rec->lk_fid2;
1230         attr->la_ctime = rec->lk_time;
1231         attr->la_mtime = rec->lk_time;
1232         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1233
1234         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1235                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1236                                  req_capsule_client_get(pill, &RMF_CAPA1));
1237         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1238                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1239                                  req_capsule_client_get(pill, &RMF_CAPA2));
1240
1241         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1242         if (rc < 0)
1243                 RETURN(rc);
1244
1245         rc = mdt_dlmreq_unpack(info);
1246
1247         RETURN(rc);
1248 }
1249
1250 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1251 {
1252         struct lu_ucred         *uc  = mdt_ucred(info);
1253         struct mdt_rec_unlink   *rec;
1254         struct md_attr          *ma = &info->mti_attr;
1255         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1256         struct mdt_reint_record *rr = &info->mti_rr;
1257         struct req_capsule      *pill = info->mti_pill;
1258         int rc;
1259         ENTRY;
1260
1261         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1262         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1263         if (rec == NULL)
1264                 RETURN(-EFAULT);
1265
1266         /* This prior initialization is needed for old_init_ucred_reint() */
1267         uc->uc_fsuid = rec->ul_fsuid;
1268         uc->uc_fsgid = rec->ul_fsgid;
1269         uc->uc_cap   = rec->ul_cap;
1270         uc->uc_suppgids[0] = rec->ul_suppgid1;
1271         uc->uc_suppgids[1] = -1;
1272
1273         attr->la_uid = rec->ul_fsuid;
1274         attr->la_gid = rec->ul_fsgid;
1275         rr->rr_fid1 = &rec->ul_fid1;
1276         rr->rr_fid2 = &rec->ul_fid2;
1277         attr->la_ctime = rec->ul_time;
1278         attr->la_mtime = rec->ul_time;
1279         attr->la_mode  = rec->ul_mode;
1280         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1281
1282         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1283                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1284                                  req_capsule_client_get(pill, &RMF_CAPA1));
1285
1286         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1287         if (rc < 0)
1288                 RETURN(rc);
1289
1290         if (rec->ul_bias & MDS_VTX_BYPASS)
1291                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1292         else
1293                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1294
1295         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1296
1297         rc = mdt_dlmreq_unpack(info);
1298         RETURN(rc);
1299 }
1300
1301 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1302 {
1303         info->mti_spec.sp_rm_entry = 1;
1304         return mdt_unlink_unpack(info);
1305 }
1306
1307 static int mdt_rename_unpack(struct mdt_thread_info *info)
1308 {
1309         struct lu_ucred         *uc = mdt_ucred(info);
1310         struct mdt_rec_rename   *rec;
1311         struct md_attr          *ma = &info->mti_attr;
1312         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1313         struct mdt_reint_record *rr = &info->mti_rr;
1314         struct req_capsule      *pill = info->mti_pill;
1315         int rc;
1316         ENTRY;
1317
1318         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1319         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1320         if (rec == NULL)
1321                 RETURN(-EFAULT);
1322
1323         /* This prior initialization is needed for old_init_ucred_reint() */
1324         uc->uc_fsuid = rec->rn_fsuid;
1325         uc->uc_fsgid = rec->rn_fsgid;
1326         uc->uc_cap   = rec->rn_cap;
1327         uc->uc_suppgids[0] = rec->rn_suppgid1;
1328         uc->uc_suppgids[1] = rec->rn_suppgid2;
1329
1330         attr->la_uid = rec->rn_fsuid;
1331         attr->la_gid = rec->rn_fsgid;
1332         rr->rr_fid1 = &rec->rn_fid1;
1333         rr->rr_fid2 = &rec->rn_fid2;
1334         attr->la_ctime = rec->rn_time;
1335         attr->la_mtime = rec->rn_time;
1336         /* rename_tgt contains the mode already */
1337         attr->la_mode = rec->rn_mode;
1338         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1339
1340         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1341                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1342                                  req_capsule_client_get(pill, &RMF_CAPA1));
1343         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1344                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1345                                  req_capsule_client_get(pill, &RMF_CAPA2));
1346
1347         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1348         if (rc < 0)
1349                 RETURN(rc);
1350
1351         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1352         if (rc < 0)
1353                 RETURN(rc);
1354
1355         if (rec->rn_bias & MDS_VTX_BYPASS)
1356                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1357         else
1358                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1359
1360         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1361
1362
1363         rc = mdt_dlmreq_unpack(info);
1364
1365         RETURN(rc);
1366 }
1367
1368 /*
1369  * please see comment above LOV_MAGIC_V1_DEF
1370  */
1371 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1372 {
1373         struct mdt_reint_record *rr = &info->mti_rr;
1374         struct lov_user_md_v1   *v1;
1375
1376         v1 = (void *)rr->rr_eadata;
1377         LASSERT(v1);
1378
1379         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1380                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1381                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1382                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1383                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1384                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1385                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1386                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1387                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1388                 }
1389         }
1390 }
1391
1392 static int mdt_open_unpack(struct mdt_thread_info *info)
1393 {
1394         struct lu_ucred         *uc = mdt_ucred(info);
1395         struct mdt_rec_create   *rec;
1396         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1397         struct req_capsule      *pill = info->mti_pill;
1398         struct mdt_reint_record *rr   = &info->mti_rr;
1399         struct ptlrpc_request   *req  = mdt_info_req(info);
1400         struct md_op_spec       *sp   = &info->mti_spec;
1401         ENTRY;
1402
1403         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1404         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1405         if (rec == NULL)
1406                 RETURN(-EFAULT);
1407
1408         /* This prior initialization is needed for old_init_ucred_reint() */
1409         uc->uc_fsuid = rec->cr_fsuid;
1410         uc->uc_fsgid = rec->cr_fsgid;
1411         uc->uc_cap   = rec->cr_cap;
1412         uc->uc_suppgids[0] = rec->cr_suppgid1;
1413         uc->uc_suppgids[1] = rec->cr_suppgid2;
1414         uc->uc_umask = rec->cr_umask;
1415
1416         rr->rr_fid1   = &rec->cr_fid1;
1417         rr->rr_fid2   = &rec->cr_fid2;
1418         rr->rr_handle = &rec->cr_old_handle;
1419         attr->la_mode = rec->cr_mode;
1420         attr->la_rdev  = rec->cr_rdev;
1421         attr->la_uid   = rec->cr_fsuid;
1422         attr->la_gid   = rec->cr_fsgid;
1423         attr->la_ctime = rec->cr_time;
1424         attr->la_mtime = rec->cr_time;
1425         attr->la_atime = rec->cr_time;
1426         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1427                          LA_CTIME | LA_MTIME | LA_ATIME;
1428         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1429         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1430         /* Do not trigger ASSERTION if client miss to set such flags. */
1431         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1432                 RETURN(-EPROTO);
1433         info->mti_replayepoch = rec->cr_ioepoch;
1434
1435         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1436
1437         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1438                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1439                                  req_capsule_client_get(pill, &RMF_CAPA1));
1440         if (req_is_replay(req) &&
1441             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1442 #if 0
1443                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1444                                  req_capsule_client_get(pill, &RMF_CAPA2));
1445 #else
1446                 /*
1447                  * FIXME: capa in replay open request might have expired,
1448                  * bypass capa check. Security hole?
1449                  */
1450                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1451                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1452 #endif
1453         }
1454
1455         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1456
1457         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1458                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1459                                                         RCL_CLIENT);
1460                 if (rr->rr_eadatalen > 0) {
1461                         rr->rr_eadata = req_capsule_client_get(pill,
1462                                                                &RMF_EADATA);
1463                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1464                         sp->u.sp_ea.eadata = rr->rr_eadata;
1465                         sp->no_create = !!req_is_replay(req);
1466                         mdt_fix_lov_magic(info);
1467                 }
1468
1469                 /*
1470                  * Client default md_size may be 0 right after client start,
1471                  * until all osc are connected, set here just some reasonable
1472                  * value to prevent misbehavior.
1473                  */
1474                 if (rr->rr_eadatalen == 0 &&
1475                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1476                         rr->rr_eadatalen = MIN_MD_SIZE;
1477         }
1478
1479         RETURN(0);
1480 }
1481
1482 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1483 {
1484         struct mdt_reint_record *rr     = &info->mti_rr;
1485         struct lu_ucred         *uc     = mdt_ucred(info);
1486         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1487         struct req_capsule      *pill   = info->mti_pill;
1488         struct mdt_rec_setxattr *rec;
1489         int                      rc;
1490         ENTRY;
1491
1492
1493         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1494                          sizeof(struct mdt_rec_reint));
1495
1496         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1497         if (rec == NULL)
1498                 RETURN(-EFAULT);
1499
1500         /* This prior initialization is needed for old_init_ucred_reint() */
1501         uc->uc_fsuid  = rec->sx_fsuid;
1502         uc->uc_fsgid  = rec->sx_fsgid;
1503         uc->uc_cap    = rec->sx_cap;
1504         uc->uc_suppgids[0] = rec->sx_suppgid1;
1505         uc->uc_suppgids[1] = -1;
1506
1507         rr->rr_opcode = rec->sx_opcode;
1508         rr->rr_fid1   = &rec->sx_fid;
1509         attr->la_valid = rec->sx_valid;
1510         attr->la_ctime = rec->sx_time;
1511         attr->la_size = rec->sx_size;
1512         attr->la_flags = rec->sx_flags;
1513
1514         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1515                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1516                                  req_capsule_client_get(pill, &RMF_CAPA1));
1517         else
1518                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1519
1520         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1521         if (rc < 0)
1522                 RETURN(rc);
1523
1524         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1525                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1526                                                         RCL_CLIENT);
1527                 if (rr->rr_eadatalen > 0) {
1528                         rr->rr_eadata = req_capsule_client_get(pill,
1529                                                                &RMF_EADATA);
1530                         if (rr->rr_eadata == NULL)
1531                                 RETURN(-EFAULT);
1532                 } else {
1533                         rr->rr_eadata = NULL;
1534                 }
1535         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1536                 CDEBUG(D_INFO, "no xattr data supplied\n");
1537                 RETURN(-EFAULT);
1538         }
1539
1540         if (mdt_dlmreq_unpack(info) < 0)
1541                 RETURN(-EPROTO);
1542
1543         RETURN(0);
1544 }
1545
1546
1547 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1548
1549 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1550         [REINT_SETATTR]  = mdt_setattr_unpack,
1551         [REINT_CREATE]   = mdt_create_unpack,
1552         [REINT_LINK]     = mdt_link_unpack,
1553         [REINT_UNLINK]   = mdt_unlink_unpack,
1554         [REINT_RENAME]   = mdt_rename_unpack,
1555         [REINT_OPEN]     = mdt_open_unpack,
1556         [REINT_SETXATTR] = mdt_setxattr_unpack,
1557         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1558         [REINT_MIGRATE]  = mdt_rename_unpack,
1559 };
1560
1561 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1562 {
1563         int rc;
1564         ENTRY;
1565
1566         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1567         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1568                 info->mti_rr.rr_opcode = op;
1569                 rc = mdt_reint_unpackers[op](info);
1570         } else {
1571                 CERROR("Unexpected opcode %d\n", op);
1572                 rc = -EFAULT;
1573         }
1574         RETURN(rc);
1575 }