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, 2015, 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 <linux/user_namespace.h>
  52 #ifdef HAVE_UIDGID_HEADER
  53 # include <linux/uidgid.h>
  54 #endif
  55 #include "mdt_internal.h"
  56 #include <lnet/nidstr.h>
  57 #include <lustre_nodemap.h>
  58
  59 typedef enum ucred_init_type {
  60         NONE_INIT       = 0,
  61         BODY_INIT       = 1,
  62         REC_INIT        = 2
  63 } ucred_init_type_t;
  64
  65 void mdt_exit_ucred(struct mdt_thread_info *info)
  66 {
  67         struct lu_ucred   *uc  = mdt_ucred(info);
  68         struct mdt_device *mdt = info->mti_mdt;
  69
  70         LASSERT(uc != NULL);
  71         if (uc->uc_valid != UCRED_INIT) {
  72                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  73                 if (uc->uc_ginfo) {
  74                         put_group_info(uc->uc_ginfo);
  75                         uc->uc_ginfo = NULL;
  76                 }
  77                 if (uc->uc_identity) {
  78                         mdt_identity_put(mdt->mdt_identity_cache,
  79                                          uc->uc_identity);
  80                         uc->uc_identity = NULL;
  81                 }
  82                 uc->uc_valid = UCRED_INIT;
  83         }
  84 }
  85
  86 static int match_nosquash_list(struct rw_semaphore *sem,
  87                                struct list_head *nidlist,
  88                                lnet_nid_t peernid)
  89 {
  90         int rc;
  91         ENTRY;
  92         down_read(sem);
  93         rc = cfs_match_nid(peernid, nidlist);
  94         up_read(sem);
  95         RETURN(rc);
  96 }
  97
  98 /* root_squash for inter-MDS operations */
  99 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 100 {
 101         struct lu_ucred *ucred = mdt_ucred(info);
 102         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 103         ENTRY;
 104
 105         LASSERT(ucred != NULL);
 106         if (!squash->rsi_uid || ucred->uc_fsuid)
 107                 RETURN(0);
 108
 109         if (match_nosquash_list(&squash->rsi_sem,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nid2str(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nid2str(peernid),
 119                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 120                squash->rsi_uid, squash->rsi_gid, 0);
 121
 122         ucred->uc_fsuid = squash->rsi_uid;
 123         ucred->uc_fsgid = squash->rsi_gid;
 124         ucred->uc_cap = 0;
 125         ucred->uc_suppgids[0] = -1;
 126         ucred->uc_suppgids[1] = -1;
 127
 128         RETURN(0);
 129 }
 130
 131 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 132 {
 133         struct ptlrpc_request   *req = mdt_info_req(info);
 134         const char              *jobid = mdt_req_get_jobid(req);
 135
 136         /* set jobid if specified. */
 137         if (jobid)
 138                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 139         else
 140                 uc->uc_jobid[0] = '\0';
 141 }
 142
 143 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 144                           void *buf, bool drop_fs_cap)
 145 {
 146         struct ptlrpc_request   *req = mdt_info_req(info);
 147         struct mdt_device       *mdt = info->mti_mdt;
 148         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 149         struct lu_ucred         *ucred = mdt_ucred(info);
 150         lnet_nid_t               peernid = req->rq_peer.nid;
 151         __u32                    perm = 0;
 152         int                      setuid;
 153         int                      setgid;
 154         int                      rc = 0;
 155
 156         ENTRY;
 157
 158         LASSERT(req->rq_auth_gss);
 159         LASSERT(!req->rq_auth_usr_mdt);
 160         LASSERT(req->rq_user_desc);
 161         LASSERT(ucred != NULL);
 162
 163         ucred->uc_valid = UCRED_INVALID;
 164
 165         ucred->uc_o_uid   = pud->pud_uid;
 166         ucred->uc_o_gid   = pud->pud_gid;
 167         ucred->uc_o_fsuid = pud->pud_fsuid;
 168         ucred->uc_o_fsgid = pud->pud_fsgid;
 169
 170         if (type == BODY_INIT) {
 171                 struct mdt_body *body = (struct mdt_body *)buf;
 172
 173                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 174                 ucred->uc_suppgids[1] = -1;
 175         }
 176
 177         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 178             req->rq_auth_uid != pud->pud_uid) {
 179                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 180                        "while client claims %u:%u/%u:%u\n",
 181                        libcfs_nid2str(peernid), req->rq_auth_uid,
 182                        pud->pud_uid, pud->pud_gid,
 183                        pud->pud_fsuid, pud->pud_fsgid);
 184                 RETURN(-EACCES);
 185         }
 186
 187         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 188                 ucred->uc_identity = NULL;
 189                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 190         } else {
 191                 struct md_identity *identity;
 192
 193                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 194                                             pud->pud_uid);
 195                 if (IS_ERR(identity)) {
 196                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 197                                 ucred->uc_identity = NULL;
 198                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 199                                        CFS_SETGRP_PERM;
 200                         } else {
 201                                 CDEBUG(D_SEC,
 202                                        "Deny access without identity: uid %u\n",
 203                                        pud->pud_uid);
 204                                 RETURN(-EACCES);
 205                         }
 206                 } else {
 207                         ucred->uc_identity = identity;
 208                         perm = mdt_identity_get_perm(ucred->uc_identity,
 209                                                      peernid);
 210                 }
 211         }
 212
 213         /* find out the setuid/setgid attempt */
 214         setuid = (pud->pud_uid != pud->pud_fsuid);
 215         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 216                   (ucred->uc_identity &&
 217                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 218
 219         /* check permission of setuid */
 220         if (setuid && !(perm & CFS_SETUID_PERM)) {
 221                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 222                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 223                 GOTO(out, rc = -EACCES);
 224         }
 225
 226         /* check permission of setgid */
 227         if (setgid && !(perm & CFS_SETGID_PERM)) {
 228                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 229                        "from %s\n", pud->pud_uid, pud->pud_gid,
 230                        pud->pud_fsuid, pud->pud_fsgid,
 231                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 232                 GOTO(out, rc = -EACCES);
 233         }
 234
 235         if (perm & CFS_SETGRP_PERM) {
 236                 if (pud->pud_ngroups) {
 237                         /* setgroups for local client */
 238                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 239                         if (!ucred->uc_ginfo) {
 240                                 CERROR("failed to alloc %d groups\n",
 241                                        pud->pud_ngroups);
 242                                 GOTO(out, rc = -ENOMEM);
 243                         }
 244
 245                         lustre_groups_from_list(ucred->uc_ginfo,
 246                                                 pud->pud_groups);
 247                         lustre_groups_sort(ucred->uc_ginfo);
 248                 } else {
 249                         ucred->uc_ginfo = NULL;
 250                 }
 251         } else {
 252                 ucred->uc_suppgids[0] = -1;
 253                 ucred->uc_suppgids[1] = -1;
 254                 ucred->uc_ginfo = NULL;
 255         }
 256
 257         ucred->uc_uid   = pud->pud_uid;
 258         ucred->uc_gid   = pud->pud_gid;
 259         ucred->uc_fsuid = pud->pud_fsuid;
 260         ucred->uc_fsgid = pud->pud_fsgid;
 261
 262         /* process root_squash here. */
 263         mdt_root_squash(info, peernid);
 264
 265         /* remove fs privilege for non-root user. */
 266         if (ucred->uc_fsuid && drop_fs_cap)
 267                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 268         else
 269                 ucred->uc_cap = pud->pud_cap;
 270         ucred->uc_valid = UCRED_NEW;
 271         ucred_set_jobid(info, ucred);
 272
 273         EXIT;
 274
 275 out:
 276         if (rc) {
 277                 if (ucred->uc_ginfo) {
 278                         put_group_info(ucred->uc_ginfo);
 279                         ucred->uc_ginfo = NULL;
 280                 }
 281                 if (ucred->uc_identity) {
 282                         mdt_identity_put(mdt->mdt_identity_cache,
 283                                          ucred->uc_identity);
 284                         ucred->uc_identity = NULL;
 285                 }
 286         }
 287
 288         return rc;
 289 }
 290
 291 /**
 292  * Check whether allow the client to set supplementary group IDs or not.
 293  *
 294  * \param[in] info      pointer to the thread context
 295  * \param[in] uc        pointer to the RPC user descriptor
 296  *
 297  * \retval              true if allow to set supplementary group IDs
 298  * \retval              false for other cases
 299  */
 300 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 301 {
 302         __u32 perm;
 303
 304         /* 1. If identity_upcall is disabled,
 305          *    permit local client to do anything. */
 306         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 307                 return true;
 308
 309         /* 2. If fail to get related identities, then forbid any client to
 310          *    set supplementary group IDs. */
 311         if (uc->uc_identity == NULL)
 312                 return false;
 313
 314         /* 3. Check the permission in the identities. */
 315         perm = mdt_identity_get_perm(uc->uc_identity,
 316                                      mdt_info_req(info)->rq_peer.nid);
 317         if (perm & CFS_SETGRP_PERM)
 318                 return true;
 319
 320         return false;
 321 }
 322
 323 int mdt_check_ucred(struct mdt_thread_info *info)
 324 {
 325         struct ptlrpc_request   *req = mdt_info_req(info);
 326         struct mdt_device       *mdt = info->mti_mdt;
 327         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 328         struct lu_ucred         *ucred = mdt_ucred(info);
 329         struct md_identity      *identity = NULL;
 330         lnet_nid_t               peernid = req->rq_peer.nid;
 331         __u32                    perm = 0;
 332         int                      setuid;
 333         int                      setgid;
 334         int                      rc = 0;
 335
 336         ENTRY;
 337
 338         LASSERT(ucred != NULL);
 339         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 340                 RETURN(0);
 341
 342         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 343                 RETURN(0);
 344
 345         /* sanity check: if we use strong authentication, we expect the
 346          * uid which client claimed is true */
 347         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 348             req->rq_auth_uid != pud->pud_uid) {
 349                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 350                        "while client claims %u:%u/%u:%u\n",
 351                        libcfs_nid2str(peernid), req->rq_auth_uid,
 352                        pud->pud_uid, pud->pud_gid,
 353                        pud->pud_fsuid, pud->pud_fsgid);
 354                 RETURN(-EACCES);
 355         }
 356
 357         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 358                 RETURN(0);
 359
 360         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 361         if (IS_ERR(identity)) {
 362                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 363                         RETURN(0);
 364                 } else {
 365                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 366                                pud->pud_uid);
 367                         RETURN(-EACCES);
 368                 }
 369         }
 370
 371         perm = mdt_identity_get_perm(identity, peernid);
 372         /* find out the setuid/setgid attempt */
 373         setuid = (pud->pud_uid != pud->pud_fsuid);
 374         setgid = (pud->pud_gid != pud->pud_fsgid ||
 375                   pud->pud_gid != identity->mi_gid);
 376
 377         /* check permission of setuid */
 378         if (setuid && !(perm & CFS_SETUID_PERM)) {
 379                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 380                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 381                 GOTO(out, rc = -EACCES);
 382         }
 383
 384         /* check permission of setgid */
 385         if (setgid && !(perm & CFS_SETGID_PERM)) {
 386                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 387                        "from %s\n", pud->pud_uid, pud->pud_gid,
 388                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 389                        libcfs_nid2str(peernid));
 390                 GOTO(out, rc = -EACCES);
 391         }
 392
 393         EXIT;
 394
 395 out:
 396         mdt_identity_put(mdt->mdt_identity_cache, identity);
 397         return rc;
 398 }
 399
 400 static int old_init_ucred_common(struct mdt_thread_info *info,
 401                                  struct lu_nodemap *nodemap,
 402                                  bool drop_fs_cap)
 403 {
 404         struct lu_ucred         *uc = mdt_ucred(info);
 405         struct mdt_device       *mdt = info->mti_mdt;
 406         struct md_identity      *identity = NULL;
 407
 408         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 409                 /* deny access before we get identity ref */
 410                 if (nodemap->nmf_deny_unknown)
 411                         RETURN(-EACCES);
 412
 413                 uc->uc_fsuid = nodemap->nm_squash_uid;
 414                 uc->uc_fsgid = nodemap->nm_squash_gid;
 415                 uc->uc_cap = 0;
 416                 uc->uc_suppgids[0] = -1;
 417                 uc->uc_suppgids[1] = -1;
 418         }
 419
 420         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 421                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 422                                             uc->uc_fsuid);
 423                 if (IS_ERR(identity)) {
 424                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 425                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 426                                 identity = NULL;
 427                         } else {
 428                                 CDEBUG(D_SEC, "Deny access without identity: "
 429                                        "uid %u\n", uc->uc_fsuid);
 430                                 RETURN(-EACCES);
 431                         }
 432                 }
 433         }
 434         uc->uc_identity = identity;
 435
 436         /* process root_squash here. */
 437         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 438
 439         /* remove fs privilege for non-root user. */
 440         if (uc->uc_fsuid && drop_fs_cap)
 441                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 442         uc->uc_valid = UCRED_OLD;
 443         ucred_set_jobid(info, uc);
 444
 445         EXIT;
 446
 447         return 0;
 448 }
 449
 450 static int old_init_ucred(struct mdt_thread_info *info,
 451                           struct mdt_body *body, bool drop_fs_cap)
 452 {
 453         struct lu_ucred *uc = mdt_ucred(info);
 454         struct lu_nodemap *nodemap;
 455         int rc;
 456         ENTRY;
 457
 458         nodemap = nodemap_get_from_exp(info->mti_exp);
 459         if (IS_ERR(nodemap))
 460                 RETURN(PTR_ERR(nodemap));
 461
 462         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 463                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 464         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 465                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 466         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 467                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 468         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 469                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 470
 471         LASSERT(uc != NULL);
 472         uc->uc_valid = UCRED_INVALID;
 473         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 474         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 475         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 476         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 477         uc->uc_suppgids[0] = body->mbo_suppgid;
 478         uc->uc_suppgids[1] = -1;
 479         uc->uc_ginfo = NULL;
 480         uc->uc_cap = body->mbo_capability;
 481
 482         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 483         nodemap_putref(nodemap);
 484
 485         RETURN(rc);
 486 }
 487
 488 static int old_init_ucred_reint(struct mdt_thread_info *info)
 489 {
 490         struct lu_ucred *uc = mdt_ucred(info);
 491         struct lu_nodemap *nodemap;
 492         int rc;
 493         ENTRY;
 494
 495         nodemap = nodemap_get_from_exp(info->mti_exp);
 496         if (IS_ERR(nodemap))
 497                 RETURN(PTR_ERR(nodemap));
 498
 499         LASSERT(uc != NULL);
 500
 501         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 502                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 503         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 504                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 505
 506         uc->uc_valid = UCRED_INVALID;
 507         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 508         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 509         uc->uc_ginfo = NULL;
 510
 511         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 512         nodemap_putref(nodemap);
 513
 514         RETURN(rc);
 515 }
 516
 517 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 518                                    struct mdt_body *body,
 519                                    bool drop_fs_cap)
 520 {
 521         struct ptlrpc_request   *req = mdt_info_req(info);
 522         struct lu_ucred         *uc  = mdt_ucred(info);
 523
 524         LASSERT(uc != NULL);
 525         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 526                 return 0;
 527
 528         mdt_exit_ucred(info);
 529
 530         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 531                 return old_init_ucred(info, body, drop_fs_cap);
 532         else
 533                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 534 }
 535
 536 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 537 {
 538         return __mdt_init_ucred(info, body, true);
 539 }
 540
 541 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 542  * doesn't set correct fsuid explicitely, but raise capability to allow
 543  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 544  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 545  * will fail permission check. */
 546 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 547                                   struct mdt_body *body)
 548 {
 549         return __mdt_init_ucred(info, body, false);
 550 }
 551
 552 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 553 {
 554         struct ptlrpc_request *req = mdt_info_req(info);
 555         struct lu_ucred       *uc  = mdt_ucred(info);
 556         struct md_attr        *ma  = &info->mti_attr;
 557
 558         LASSERT(uc != NULL);
 559         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 560                 return 0;
 561
 562         /* LU-5564: for normal close request, skip permission check */
 563         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 564             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 565                 uc->uc_cap |= CFS_CAP_FS_MASK;
 566
 567         mdt_exit_ucred(info);
 568
 569         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 570                 return old_init_ucred_reint(info);
 571         else
 572                 return new_init_ucred(info, REC_INIT, NULL, true);
 573 }
 574
 575 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 576 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 577 {
 578         const struct lov_ost_data_v1    *lod;
 579         int                              i;
 580         __u16                            count;
 581
 582         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 583                 return;
 584
 585         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 586
 587         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 588                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 589                le32_to_cpu(lmm->lmm_pattern));
 590         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 591                le32_to_cpu(lmm->lmm_stripe_size), count);
 592
 593         /* If it's a directory or a released file, then there are
 594          * no actual objects to print, so bail out. */
 595         if (valid & OBD_MD_FLDIREA ||
 596             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 597                 return;
 598
 599         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 600         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 601                 struct ost_id oi;
 602
 603                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 604                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 605                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 606         }
 607 }
 608
 609 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 610 {
 611         const struct lmv_mds_md_v1 *lmm1;
 612         int                        i;
 613
 614         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 615                 return;
 616
 617         lmm1 = &lmv->lmv_md_v1;
 618         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 619                le32_to_cpu(lmm1->lmv_magic),
 620                le32_to_cpu(lmm1->lmv_master_mdt_index),
 621                le32_to_cpu(lmm1->lmv_stripe_count));
 622
 623         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 624                 return;
 625
 626         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 627                 struct lu_fid fid;
 628
 629                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 630                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 631         }
 632 }
 633
 634 /* Shrink and/or grow reply buffers */
 635 int mdt_fix_reply(struct mdt_thread_info *info)
 636 {
 637         struct req_capsule *pill = info->mti_pill;
 638         struct mdt_body    *body;
 639         int                md_size, md_packed = 0;
 640         int                acl_size;
 641         int                rc = 0;
 642         ENTRY;
 643
 644         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 645         LASSERT(body != NULL);
 646
 647         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 648                                OBD_MD_LINKNAME))
 649                 md_size = body->mbo_eadatasize;
 650         else
 651                 md_size = 0;
 652
 653         acl_size = body->mbo_aclsize;
 654
 655         /* this replay - not send info to client */
 656         if (info->mti_spec.no_create) {
 657                 md_size = 0;
 658                 acl_size = 0;
 659         }
 660
 661         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 662                md_size, acl_size);
 663 /*
 664             &RMF_MDT_BODY,
 665             &RMF_MDT_MD,
 666             &RMF_ACL, or &RMF_LOGCOOKIES
 667 (optional)  &RMF_CAPA1,
 668 (optional)  &RMF_CAPA2,
 669 (optional)  something else
 670 */
 671
 672         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 673          * buffers before growing it */
 674         if (info->mti_big_lmm_used) {
 675                 /* big_lmm buffer may be used even without packing the result
 676                  * into reply, just for internal server needs */
 677                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 678                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 679                                                          RCL_SERVER);
 680
 681                 /* free big lmm if md_size is not needed */
 682                 if (md_size == 0 || md_packed == 0) {
 683                         info->mti_big_lmm_used = 0;
 684                 } else {
 685                         /* buffer must be allocated separately */
 686                         LASSERT(info->mti_attr.ma_lmm !=
 687                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 688                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 689                 }
 690         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 691                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 692         }
 693
 694         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 695                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 696         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 697                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 698                                    acl_size, RCL_SERVER);
 699
 700         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 701             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 702                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 703
 704         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 705             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 706                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 707
 708         /*
 709          * Some more field should be shrinked if needed.
 710          * This should be done by those who added fields to reply message.
 711          */
 712
 713         /* Grow MD buffer if needed finally */
 714         if (info->mti_big_lmm_used) {
 715                 void *lmm;
 716
 717                 LASSERT(md_size > md_packed);
 718                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 719                        md_size - md_packed);
 720                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 721                 if (rc) {
 722                         /* we can't answer with proper LOV EA, drop flags,
 723                          * the rc is also returned so this request is
 724                          * considered as failed */
 725                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 726                         /* don't return transno along with error */
 727                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 728                 } else {
 729                         /* now we need to pack right LOV/LMV EA */
 730                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 731                         if (info->mti_attr.ma_valid & MA_LOV) {
 732                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 733                                                              RCL_SERVER) ==
 734                                                 info->mti_attr.ma_lmm_size);
 735                                 memcpy(lmm, info->mti_attr.ma_lmm,
 736                                        info->mti_attr.ma_lmm_size);
 737                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 738                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 739                                                              RCL_SERVER) ==
 740                                                 info->mti_attr.ma_lmv_size);
 741                                 memcpy(lmm, info->mti_attr.ma_lmv,
 742                                        info->mti_attr.ma_lmv_size);
 743                         }
 744                 }
 745                 /* update mdt_max_mdsize so clients will be aware about that */
 746                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 747                         info->mti_mdt->mdt_max_mdsize =
 748                                                     info->mti_attr.ma_lmm_size;
 749                 info->mti_big_lmm_used = 0;
 750         }
 751         RETURN(rc);
 752 }
 753
 754
 755 /* if object is dying, pack the lov/llog data,
 756  * parameter info->mti_attr should be valid at this point! */
 757 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 758                            const struct md_attr *ma)
 759 {
 760         struct mdt_body       *repbody;
 761         const struct lu_attr *la = &ma->ma_attr;
 762         ENTRY;
 763
 764         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 765         LASSERT(repbody != NULL);
 766
 767         if (ma->ma_valid & MA_INODE)
 768                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 769
 770         if (ma->ma_valid & MA_LOV) {
 771                 CERROR("No need in LOV EA upon unlink\n");
 772                 dump_stack();
 773         }
 774         repbody->mbo_eadatasize = 0;
 775
 776         RETURN(0);
 777 }
 778
 779 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 780                                   struct md_attr *ma)
 781 {
 782         __u64 out;
 783
 784         out = 0;
 785         if (in & MDS_ATTR_MODE)
 786                 out |= LA_MODE;
 787         if (in & MDS_ATTR_UID)
 788                 out |= LA_UID;
 789         if (in & MDS_ATTR_GID)
 790                 out |= LA_GID;
 791         if (in & MDS_ATTR_SIZE)
 792                 out |= LA_SIZE;
 793         if (in & MDS_ATTR_BLOCKS)
 794                 out |= LA_BLOCKS;
 795         if (in & MDS_ATTR_ATIME_SET)
 796                 out |= LA_ATIME;
 797         if (in & MDS_ATTR_CTIME_SET)
 798                 out |= LA_CTIME;
 799         if (in & MDS_ATTR_MTIME_SET)
 800                 out |= LA_MTIME;
 801         if (in & MDS_ATTR_ATTR_FLAG)
 802                 out |= LA_FLAGS;
 803         if (in & MDS_ATTR_KILL_SUID)
 804                 out |= LA_KILL_SUID;
 805         if (in & MDS_ATTR_KILL_SGID)
 806                 out |= LA_KILL_SGID;
 807
 808         if (in & MDS_ATTR_FROM_OPEN)
 809                 rr->rr_flags |= MRF_OPEN_TRUNC;
 810         if (in & MDS_OPEN_OWNEROVERRIDE)
 811                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 812         if (in & MDS_ATTR_FORCE)
 813                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 814
 815         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 816                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 817                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 818                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 819                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 820                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 821         if (in != 0)
 822                 CERROR("Unknown attr bits: "LPX64"\n", in);
 823         return out;
 824 }
 825
 826 /* unpacking */
 827
 828 int mdt_name_unpack(struct req_capsule *pill,
 829                     const struct req_msg_field *field,
 830                     struct lu_name *ln,
 831                     enum mdt_name_flags flags)
 832 {
 833         ln->ln_name = req_capsule_client_get(pill, field);
 834         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 835
 836         if (!lu_name_is_valid(ln)) {
 837                 ln->ln_name = NULL;
 838                 ln->ln_namelen = 0;
 839
 840                 return -EPROTO;
 841         }
 842
 843         if ((flags & MNF_FIX_ANON) &&
 844             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 845                 /* Newer (3.x) kernels use a name of "/" for the
 846                  * "anonymous" disconnected dentries from NFS
 847                  * filehandle conversion. See d_obtain_alias(). */
 848                 ln->ln_name = NULL;
 849                 ln->ln_namelen = 0;
 850         }
 851
 852         return 0;
 853 }
 854
 855 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 856 {
 857         struct lu_ucred         *uc = mdt_ucred(info);
 858         struct md_attr          *ma = &info->mti_attr;
 859         struct lu_attr          *la = &ma->ma_attr;
 860         struct req_capsule      *pill = info->mti_pill;
 861         struct mdt_reint_record *rr = &info->mti_rr;
 862         struct mdt_rec_setattr  *rec;
 863         struct lu_nodemap       *nodemap;
 864         ENTRY;
 865
 866         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 867         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 868         if (rec == NULL)
 869                 RETURN(-EFAULT);
 870
 871         /* This prior initialization is needed for old_init_ucred_reint() */
 872         uc->uc_fsuid = rec->sa_fsuid;
 873         uc->uc_fsgid = rec->sa_fsgid;
 874         uc->uc_cap   = rec->sa_cap;
 875         uc->uc_suppgids[0] = rec->sa_suppgid;
 876         uc->uc_suppgids[1] = -1;
 877
 878         rr->rr_fid1 = &rec->sa_fid;
 879         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 880         la->la_mode  = rec->sa_mode;
 881         la->la_flags = rec->sa_attr_flags;
 882
 883         nodemap = nodemap_get_from_exp(info->mti_exp);
 884         if (IS_ERR(nodemap))
 885                 RETURN(PTR_ERR(nodemap));
 886
 887         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 888                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 889         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 890                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 891         nodemap_putref(nodemap);
 892
 893         la->la_size  = rec->sa_size;
 894         la->la_blocks = rec->sa_blocks;
 895         la->la_ctime = rec->sa_ctime;
 896         la->la_atime = rec->sa_atime;
 897         la->la_mtime = rec->sa_mtime;
 898         ma->ma_valid = MA_INODE;
 899
 900         if (rec->sa_bias & MDS_DATA_MODIFIED)
 901                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 902         else
 903                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 904
 905         if (rec->sa_bias & MDS_HSM_RELEASE)
 906                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 907         else
 908                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 909
 910         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 911                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 912         else
 913                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 914
 915         RETURN(0);
 916 }
 917
 918 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 919 {
 920         struct req_capsule *pill = info->mti_pill;
 921         struct mdt_ioepoch *ioepoch;
 922         ENTRY;
 923
 924         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 925                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 926         else
 927                 ioepoch = NULL;
 928
 929         if (ioepoch == NULL)
 930                 RETURN(-EPROTO);
 931
 932         info->mti_close_handle = ioepoch->mio_handle;
 933
 934         RETURN(0);
 935 }
 936
 937 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 938         struct req_capsule      *pill = info->mti_pill;
 939
 940         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 941                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 942                 if (info->mti_dlm_req == NULL)
 943                         RETURN(-EFAULT);
 944         }
 945
 946         RETURN(0);
 947 }
 948
 949 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 950 {
 951         struct mdt_reint_record *rr = &info->mti_rr;
 952         struct md_attr          *ma = &info->mti_attr;
 953         struct req_capsule      *pill = info->mti_pill;
 954         int rc;
 955         ENTRY;
 956
 957         rc = mdt_setattr_unpack_rec(info);
 958         if (rc)
 959                 RETURN(rc);
 960
 961         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 962                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 963                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 964                                                         RCL_CLIENT);
 965                 if (rr->rr_eadatalen > 0) {
 966                         const struct lmv_user_md        *lum;
 967
 968                         lum = rr->rr_eadata;
 969                         /* Sigh ma_valid(from req) does not indicate whether
 970                          * it will set LOV/LMV EA, so we have to check magic */
 971                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 972                                 ma->ma_valid |= MA_LMV;
 973                                 ma->ma_lmv = (void *)rr->rr_eadata;
 974                                 ma->ma_lmv_size = rr->rr_eadatalen;
 975                         } else {
 976                                 ma->ma_valid |= MA_LOV;
 977                                 ma->ma_lmm = (void *)rr->rr_eadata;
 978                                 ma->ma_lmm_size = rr->rr_eadatalen;
 979                         }
 980                 }
 981         }
 982
 983         rc = mdt_dlmreq_unpack(info);
 984         RETURN(rc);
 985 }
 986
 987 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
 988 {
 989         struct md_attr          *ma = &info->mti_attr;
 990         struct req_capsule      *pill = info->mti_pill;
 991         ENTRY;
 992
 993         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 994                 RETURN(0);
 995
 996         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
 997
 998         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
 999             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1000                 RETURN(-EFAULT);
1001
1002         RETURN(0);
1003 }
1004
1005 int mdt_close_unpack(struct mdt_thread_info *info)
1006 {
1007         int rc;
1008         ENTRY;
1009
1010         rc = mdt_close_handle_unpack(info);
1011         if (rc)
1012                 RETURN(rc);
1013
1014         rc = mdt_setattr_unpack_rec(info);
1015         if (rc)
1016                 RETURN(rc);
1017
1018         rc = mdt_intent_close_unpack(info);
1019         if (rc)
1020                 RETURN(rc);
1021
1022         RETURN(mdt_init_ucred_reint(info));
1023 }
1024
1025 static int mdt_create_unpack(struct mdt_thread_info *info)
1026 {
1027         struct lu_ucred         *uc  = mdt_ucred(info);
1028         struct mdt_rec_create   *rec;
1029         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1030         struct mdt_reint_record *rr = &info->mti_rr;
1031         struct req_capsule      *pill = info->mti_pill;
1032         struct md_op_spec       *sp = &info->mti_spec;
1033         int rc;
1034         ENTRY;
1035
1036         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1037         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1038         if (rec == NULL)
1039                 RETURN(-EFAULT);
1040
1041         /* This prior initialization is needed for old_init_ucred_reint() */
1042         uc->uc_fsuid = rec->cr_fsuid;
1043         uc->uc_fsgid = rec->cr_fsgid;
1044         uc->uc_cap   = rec->cr_cap;
1045         uc->uc_suppgids[0] = rec->cr_suppgid1;
1046         uc->uc_suppgids[1] = -1;
1047         uc->uc_umask = rec->cr_umask;
1048
1049         rr->rr_fid1 = &rec->cr_fid1;
1050         rr->rr_fid2 = &rec->cr_fid2;
1051         attr->la_mode = rec->cr_mode;
1052         attr->la_rdev  = rec->cr_rdev;
1053         attr->la_uid   = rec->cr_fsuid;
1054         attr->la_gid   = rec->cr_fsgid;
1055         attr->la_ctime = rec->cr_time;
1056         attr->la_mtime = rec->cr_time;
1057         attr->la_atime = rec->cr_time;
1058         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1059                          LA_CTIME | LA_MTIME | LA_ATIME;
1060         memset(&sp->u, 0, sizeof(sp->u));
1061         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1062
1063         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1064         if (rc < 0)
1065                 RETURN(rc);
1066
1067         if (S_ISLNK(attr->la_mode)) {
1068                 const char *tgt = NULL;
1069
1070                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1071                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1072                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1073                         sp->u.sp_symname = tgt;
1074                 }
1075                 if (tgt == NULL)
1076                         RETURN(-EFAULT);
1077         } else {
1078                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1079                 if (S_ISDIR(attr->la_mode) &&
1080                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1081                         sp->u.sp_ea.eadata =
1082                                 req_capsule_client_get(pill, &RMF_EADATA);
1083                         sp->u.sp_ea.eadatalen =
1084                                 req_capsule_get_size(pill, &RMF_EADATA,
1085                                                      RCL_CLIENT);
1086                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1087                 }
1088         }
1089
1090         rc = mdt_dlmreq_unpack(info);
1091         RETURN(rc);
1092 }
1093
1094 static int mdt_link_unpack(struct mdt_thread_info *info)
1095 {
1096         struct lu_ucred         *uc  = mdt_ucred(info);
1097         struct mdt_rec_link     *rec;
1098         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1099         struct mdt_reint_record *rr = &info->mti_rr;
1100         struct req_capsule      *pill = info->mti_pill;
1101         int rc;
1102         ENTRY;
1103
1104         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1105         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1106         if (rec == NULL)
1107                 RETURN(-EFAULT);
1108
1109         /* This prior initialization is needed for old_init_ucred_reint() */
1110         uc->uc_fsuid = rec->lk_fsuid;
1111         uc->uc_fsgid = rec->lk_fsgid;
1112         uc->uc_cap   = rec->lk_cap;
1113         uc->uc_suppgids[0] = rec->lk_suppgid1;
1114         uc->uc_suppgids[1] = rec->lk_suppgid2;
1115
1116         attr->la_uid = rec->lk_fsuid;
1117         attr->la_gid = rec->lk_fsgid;
1118         rr->rr_fid1 = &rec->lk_fid1;
1119         rr->rr_fid2 = &rec->lk_fid2;
1120         attr->la_ctime = rec->lk_time;
1121         attr->la_mtime = rec->lk_time;
1122         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1123
1124         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1125         if (rc < 0)
1126                 RETURN(rc);
1127
1128         rc = mdt_dlmreq_unpack(info);
1129
1130         RETURN(rc);
1131 }
1132
1133 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1134 {
1135         struct lu_ucred         *uc  = mdt_ucred(info);
1136         struct mdt_rec_unlink   *rec;
1137         struct md_attr          *ma = &info->mti_attr;
1138         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1139         struct mdt_reint_record *rr = &info->mti_rr;
1140         struct req_capsule      *pill = info->mti_pill;
1141         int rc;
1142         ENTRY;
1143
1144         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1145         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1146         if (rec == NULL)
1147                 RETURN(-EFAULT);
1148
1149         /* This prior initialization is needed for old_init_ucred_reint() */
1150         uc->uc_fsuid = rec->ul_fsuid;
1151         uc->uc_fsgid = rec->ul_fsgid;
1152         uc->uc_cap   = rec->ul_cap;
1153         uc->uc_suppgids[0] = rec->ul_suppgid1;
1154         uc->uc_suppgids[1] = -1;
1155
1156         attr->la_uid = rec->ul_fsuid;
1157         attr->la_gid = rec->ul_fsgid;
1158         rr->rr_fid1 = &rec->ul_fid1;
1159         rr->rr_fid2 = &rec->ul_fid2;
1160         attr->la_ctime = rec->ul_time;
1161         attr->la_mtime = rec->ul_time;
1162         attr->la_mode  = rec->ul_mode;
1163         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1164
1165         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1166         if (rc < 0)
1167                 RETURN(rc);
1168
1169         if (rec->ul_bias & MDS_VTX_BYPASS)
1170                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1171         else
1172                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1173
1174         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1175
1176         rc = mdt_dlmreq_unpack(info);
1177         RETURN(rc);
1178 }
1179
1180 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1181 {
1182         info->mti_spec.sp_rm_entry = 1;
1183         return mdt_unlink_unpack(info);
1184 }
1185
1186 static int mdt_rename_unpack(struct mdt_thread_info *info)
1187 {
1188         struct lu_ucred         *uc = mdt_ucred(info);
1189         struct mdt_rec_rename   *rec;
1190         struct md_attr          *ma = &info->mti_attr;
1191         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1192         struct mdt_reint_record *rr = &info->mti_rr;
1193         struct req_capsule      *pill = info->mti_pill;
1194         int rc;
1195         ENTRY;
1196
1197         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1198         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1199         if (rec == NULL)
1200                 RETURN(-EFAULT);
1201
1202         /* This prior initialization is needed for old_init_ucred_reint() */
1203         uc->uc_fsuid = rec->rn_fsuid;
1204         uc->uc_fsgid = rec->rn_fsgid;
1205         uc->uc_cap   = rec->rn_cap;
1206         uc->uc_suppgids[0] = rec->rn_suppgid1;
1207         uc->uc_suppgids[1] = rec->rn_suppgid2;
1208
1209         attr->la_uid = rec->rn_fsuid;
1210         attr->la_gid = rec->rn_fsgid;
1211         rr->rr_fid1 = &rec->rn_fid1;
1212         rr->rr_fid2 = &rec->rn_fid2;
1213         attr->la_ctime = rec->rn_time;
1214         attr->la_mtime = rec->rn_time;
1215         /* rename_tgt contains the mode already */
1216         attr->la_mode = rec->rn_mode;
1217         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1218
1219         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1220         if (rc < 0)
1221                 RETURN(rc);
1222
1223         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1224         if (rc < 0)
1225                 RETURN(rc);
1226
1227         if (rec->rn_bias & MDS_VTX_BYPASS)
1228                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1229         else
1230                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1231
1232         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1233                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1234                 rc = mdt_close_handle_unpack(info);
1235                 if (rc < 0)
1236                         RETURN(rc);
1237                 info->mti_spec.sp_migrate_close = 1;
1238         }
1239
1240         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1241
1242
1243         rc = mdt_dlmreq_unpack(info);
1244
1245         RETURN(rc);
1246 }
1247
1248 /*
1249  * please see comment above LOV_MAGIC_V1_DEF
1250  */
1251 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1252 {
1253         struct mdt_reint_record *rr = &info->mti_rr;
1254         struct lov_user_md_v1   *v1;
1255
1256         v1 = (void *)rr->rr_eadata;
1257         LASSERT(v1);
1258
1259         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1260                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1261                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1262                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1263                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1264                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1265                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1266                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1267                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1268                 }
1269         }
1270 }
1271
1272 static int mdt_open_unpack(struct mdt_thread_info *info)
1273 {
1274         struct lu_ucred         *uc = mdt_ucred(info);
1275         struct mdt_rec_create   *rec;
1276         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1277         struct req_capsule      *pill = info->mti_pill;
1278         struct mdt_reint_record *rr   = &info->mti_rr;
1279         struct ptlrpc_request   *req  = mdt_info_req(info);
1280         struct md_op_spec       *sp   = &info->mti_spec;
1281         ENTRY;
1282
1283         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1284         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1285         if (rec == NULL)
1286                 RETURN(-EFAULT);
1287
1288         /* This prior initialization is needed for old_init_ucred_reint() */
1289         uc->uc_fsuid = rec->cr_fsuid;
1290         uc->uc_fsgid = rec->cr_fsgid;
1291         uc->uc_cap   = rec->cr_cap;
1292         uc->uc_suppgids[0] = rec->cr_suppgid1;
1293         uc->uc_suppgids[1] = rec->cr_suppgid2;
1294         uc->uc_umask = rec->cr_umask;
1295
1296         rr->rr_fid1   = &rec->cr_fid1;
1297         rr->rr_fid2   = &rec->cr_fid2;
1298         rr->rr_handle = &rec->cr_old_handle;
1299         attr->la_mode = rec->cr_mode;
1300         attr->la_rdev  = rec->cr_rdev;
1301         attr->la_uid   = rec->cr_fsuid;
1302         attr->la_gid   = rec->cr_fsgid;
1303         attr->la_ctime = rec->cr_time;
1304         attr->la_mtime = rec->cr_time;
1305         attr->la_atime = rec->cr_time;
1306         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1307                          LA_CTIME | LA_MTIME | LA_ATIME;
1308         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1309         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1310         /* Do not trigger ASSERTION if client miss to set such flags. */
1311         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1312                 RETURN(-EPROTO);
1313
1314         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1315
1316         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1317
1318         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1319                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1320                                                         RCL_CLIENT);
1321                 if (rr->rr_eadatalen > 0) {
1322                         rr->rr_eadata = req_capsule_client_get(pill,
1323                                                                &RMF_EADATA);
1324                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1325                         sp->u.sp_ea.eadata = rr->rr_eadata;
1326                         sp->no_create = !!req_is_replay(req);
1327                         mdt_fix_lov_magic(info);
1328                 }
1329
1330                 /*
1331                  * Client default md_size may be 0 right after client start,
1332                  * until all osc are connected, set here just some reasonable
1333                  * value to prevent misbehavior.
1334                  */
1335                 if (rr->rr_eadatalen == 0 &&
1336                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1337                         rr->rr_eadatalen = MIN_MD_SIZE;
1338         }
1339
1340         RETURN(0);
1341 }
1342
1343 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1344 {
1345         struct mdt_reint_record *rr     = &info->mti_rr;
1346         struct lu_ucred         *uc     = mdt_ucred(info);
1347         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1348         struct req_capsule      *pill   = info->mti_pill;
1349         struct mdt_rec_setxattr *rec;
1350         int                      rc;
1351         ENTRY;
1352
1353
1354         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1355                          sizeof(struct mdt_rec_reint));
1356
1357         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1358         if (rec == NULL)
1359                 RETURN(-EFAULT);
1360
1361         /* This prior initialization is needed for old_init_ucred_reint() */
1362         uc->uc_fsuid  = rec->sx_fsuid;
1363         uc->uc_fsgid  = rec->sx_fsgid;
1364         uc->uc_cap    = rec->sx_cap;
1365         uc->uc_suppgids[0] = rec->sx_suppgid1;
1366         uc->uc_suppgids[1] = -1;
1367
1368         rr->rr_opcode = rec->sx_opcode;
1369         rr->rr_fid1   = &rec->sx_fid;
1370         attr->la_valid = rec->sx_valid;
1371         attr->la_ctime = rec->sx_time;
1372         attr->la_size = rec->sx_size;
1373         attr->la_flags = rec->sx_flags;
1374
1375         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1376         if (rc < 0)
1377                 RETURN(rc);
1378
1379         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1380                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1381                                                         RCL_CLIENT);
1382                 if (rr->rr_eadatalen > 0) {
1383                         rr->rr_eadata = req_capsule_client_get(pill,
1384                                                                &RMF_EADATA);
1385                         if (rr->rr_eadata == NULL)
1386                                 RETURN(-EFAULT);
1387                 } else {
1388                         rr->rr_eadata = NULL;
1389                 }
1390         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1391                 CDEBUG(D_INFO, "no xattr data supplied\n");
1392                 RETURN(-EFAULT);
1393         }
1394
1395         if (mdt_dlmreq_unpack(info) < 0)
1396                 RETURN(-EPROTO);
1397
1398         RETURN(0);
1399 }
1400
1401
1402 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1403
1404 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1405         [REINT_SETATTR]  = mdt_setattr_unpack,
1406         [REINT_CREATE]   = mdt_create_unpack,
1407         [REINT_LINK]     = mdt_link_unpack,
1408         [REINT_UNLINK]   = mdt_unlink_unpack,
1409         [REINT_RENAME]   = mdt_rename_unpack,
1410         [REINT_OPEN]     = mdt_open_unpack,
1411         [REINT_SETXATTR] = mdt_setxattr_unpack,
1412         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1413         [REINT_MIGRATE]  = mdt_rename_unpack,
1414 };
1415
1416 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1417 {
1418         int rc;
1419         ENTRY;
1420
1421         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1422         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1423                 info->mti_rr.rr_opcode = op;
1424                 rc = mdt_reint_unpackers[op](info);
1425         } else {
1426                 CERROR("Unexpected opcode %d\n", op);
1427                 rc = -EFAULT;
1428         }
1429         RETURN(rc);
1430 }