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         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 153         int                      setuid;
 154         int                      setgid;
 155         int                      rc = 0;
 156
 157         ENTRY;
 158
 159         LASSERT(req->rq_auth_gss);
 160         LASSERT(!req->rq_auth_usr_mdt);
 161         LASSERT(req->rq_user_desc);
 162         LASSERT(ucred != NULL);
 163
 164         ucred->uc_valid = UCRED_INVALID;
 165
 166         ucred->uc_o_uid   = pud->pud_uid;
 167         ucred->uc_o_gid   = pud->pud_gid;
 168         ucred->uc_o_fsuid = pud->pud_fsuid;
 169         ucred->uc_o_fsgid = pud->pud_fsgid;
 170
 171         if (type == BODY_INIT) {
 172                 struct mdt_body *body = (struct mdt_body *)buf;
 173
 174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 175                 ucred->uc_suppgids[1] = -1;
 176         }
 177
 178         /* sanity check: we expect the uid which client claimed is true */
 179         if (remote) {
 180                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 181                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 182                         RETURN(-EACCES);
 183                 }
 184
 185                 if (ptlrpc_user_desc_do_idmap(req, pud))
 186                         RETURN(-EACCES);
 187
 188                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 189                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 190                                "while client claims %u:%u/%u:%u\n",
 191                                libcfs_nid2str(peernid), req->rq_auth_uid,
 192                                req->rq_auth_mapped_uid,
 193                                pud->pud_uid, pud->pud_gid,
 194                                pud->pud_fsuid, pud->pud_fsgid);
 195                         RETURN(-EACCES);
 196                 }
 197         } else {
 198                 if (req->rq_auth_uid != pud->pud_uid) {
 199                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 200                                "while client claims %u:%u/%u:%u\n",
 201                                libcfs_nid2str(peernid), req->rq_auth_uid,
 202                                pud->pud_uid, pud->pud_gid,
 203                                pud->pud_fsuid, pud->pud_fsgid);
 204                         RETURN(-EACCES);
 205                 }
 206         }
 207
 208         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 209                 if (remote) {
 210                         CDEBUG(D_SEC, "remote client must run with identity_get "
 211                                "enabled!\n");
 212                         RETURN(-EACCES);
 213                 } else {
 214                         ucred->uc_identity = NULL;
 215                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 216                                CFS_SETGRP_PERM;
 217                 }
 218         } else {
 219                 struct md_identity *identity;
 220
 221                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 222                                             pud->pud_uid);
 223                 if (IS_ERR(identity)) {
 224                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 225                                      !remote)) {
 226                                 ucred->uc_identity = NULL;
 227                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 228                                        CFS_SETGRP_PERM;
 229                         } else {
 230                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 231                                        pud->pud_uid);
 232                                 RETURN(-EACCES);
 233                         }
 234                 } else {
 235                         ucred->uc_identity = identity;
 236                         perm = mdt_identity_get_perm(ucred->uc_identity,
 237                                                      remote, peernid);
 238                 }
 239         }
 240
 241         /* find out the setuid/setgid attempt */
 242         setuid = (pud->pud_uid != pud->pud_fsuid);
 243         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 244                   (ucred->uc_identity &&
 245                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 246
 247         /* check permission of setuid */
 248         if (setuid && !(perm & CFS_SETUID_PERM)) {
 249                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 250                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 251                 GOTO(out, rc = -EACCES);
 252         }
 253
 254         /* check permission of setgid */
 255         if (setgid && !(perm & CFS_SETGID_PERM)) {
 256                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 257                        "from %s\n", pud->pud_uid, pud->pud_gid,
 258                        pud->pud_fsuid, pud->pud_fsgid,
 259                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 260                 GOTO(out, rc = -EACCES);
 261         }
 262
 263         /*
 264          * NB: remote client not allowed to setgroups anyway.
 265          */
 266         if (!remote && perm & CFS_SETGRP_PERM) {
 267                 if (pud->pud_ngroups) {
 268                         /* setgroups for local client */
 269                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 270                         if (!ucred->uc_ginfo) {
 271                                 CERROR("failed to alloc %d groups\n",
 272                                        pud->pud_ngroups);
 273                                 GOTO(out, rc = -ENOMEM);
 274                         }
 275
 276                         lustre_groups_from_list(ucred->uc_ginfo,
 277                                                 pud->pud_groups);
 278                         lustre_groups_sort(ucred->uc_ginfo);
 279                 } else {
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282         } else {
 283                 ucred->uc_suppgids[0] = -1;
 284                 ucred->uc_suppgids[1] = -1;
 285                 ucred->uc_ginfo = NULL;
 286         }
 287
 288         ucred->uc_uid   = pud->pud_uid;
 289         ucred->uc_gid   = pud->pud_gid;
 290         ucred->uc_fsuid = pud->pud_fsuid;
 291         ucred->uc_fsgid = pud->pud_fsgid;
 292
 293         /* process root_squash here. */
 294         mdt_root_squash(info, peernid);
 295
 296         /* remove fs privilege for non-root user. */
 297         if (ucred->uc_fsuid && drop_fs_cap)
 298                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 299         else
 300                 ucred->uc_cap = pud->pud_cap;
 301         if (remote && !(perm & CFS_RMTOWN_PERM))
 302                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 303                                    CFS_CAP_CHOWN_MASK);
 304         ucred->uc_valid = UCRED_NEW;
 305         ucred_set_jobid(info, ucred);
 306
 307         EXIT;
 308
 309 out:
 310         if (rc) {
 311                 if (ucred->uc_ginfo) {
 312                         put_group_info(ucred->uc_ginfo);
 313                         ucred->uc_ginfo = NULL;
 314                 }
 315                 if (ucred->uc_identity) {
 316                         mdt_identity_put(mdt->mdt_identity_cache,
 317                                          ucred->uc_identity);
 318                         ucred->uc_identity = NULL;
 319                 }
 320         }
 321
 322         return rc;
 323 }
 324
 325 /**
 326  * Check whether allow the client to set supplementary group IDs or not.
 327  *
 328  * \param[in] info      pointer to the thread context
 329  * \param[in] uc        pointer to the RPC user descriptor
 330  *
 331  * \retval              true if allow to set supplementary group IDs
 332  * \retval              false for other cases
 333  */
 334 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 335 {
 336         __u32 remote = exp_connect_rmtclient(info->mti_exp);
 337         __u32 perm;
 338
 339         /* 1. If identity_upcall is disabled, then forbid remote client to set
 340          *    supplementary group IDs, but permit local client to do that. */
 341         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) {
 342                 if (remote)
 343                         return false;
 344
 345                 return true;
 346         }
 347
 348         /* 2. If fail to get related identities, then forbid any client to
 349          *    set supplementary group IDs. */
 350         if (uc->uc_identity == NULL)
 351                 return false;
 352
 353         /* 3. Check the permission in the identities. */
 354         perm = mdt_identity_get_perm(uc->uc_identity, remote,
 355                                      mdt_info_req(info)->rq_peer.nid);
 356         if (perm & CFS_SETGRP_PERM)
 357                 return true;
 358
 359         return false;
 360 }
 361
 362 int mdt_check_ucred(struct mdt_thread_info *info)
 363 {
 364         struct ptlrpc_request   *req = mdt_info_req(info);
 365         struct mdt_device       *mdt = info->mti_mdt;
 366         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 367         struct lu_ucred         *ucred = mdt_ucred(info);
 368         struct md_identity      *identity = NULL;
 369         lnet_nid_t               peernid = req->rq_peer.nid;
 370         __u32                    perm = 0;
 371         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 372         int                      setuid;
 373         int                      setgid;
 374         int                      rc = 0;
 375
 376         ENTRY;
 377
 378         LASSERT(ucred != NULL);
 379         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 380                 RETURN(0);
 381
 382         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 383                 RETURN(0);
 384
 385         /* sanity check: if we use strong authentication, we expect the
 386          * uid which client claimed is true */
 387         if (remote) {
 388                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 389                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 390                         RETURN(-EACCES);
 391                 }
 392
 393                 if (ptlrpc_user_desc_do_idmap(req, pud))
 394                         RETURN(-EACCES);
 395
 396                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 397                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 398                                "while client claims %u:%u/%u:%u\n",
 399                                libcfs_nid2str(peernid), req->rq_auth_uid,
 400                                req->rq_auth_mapped_uid,
 401                                pud->pud_uid, pud->pud_gid,
 402                                pud->pud_fsuid, pud->pud_fsgid);
 403                         RETURN(-EACCES);
 404                 }
 405         } else {
 406                 if (req->rq_auth_uid != pud->pud_uid) {
 407                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 408                                "while client claims %u:%u/%u:%u\n",
 409                                libcfs_nid2str(peernid), req->rq_auth_uid,
 410                                pud->pud_uid, pud->pud_gid,
 411                                pud->pud_fsuid, pud->pud_fsgid);
 412                         RETURN(-EACCES);
 413                 }
 414         }
 415
 416         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 417                 if (remote) {
 418                         CDEBUG(D_SEC, "remote client must run with identity_get "
 419                                "enabled!\n");
 420                         RETURN(-EACCES);
 421                 }
 422                 RETURN(0);
 423         }
 424
 425         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 426         if (IS_ERR(identity)) {
 427                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 428                              !remote)) {
 429                         RETURN(0);
 430                 } else {
 431                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 432                                pud->pud_uid);
 433                         RETURN(-EACCES);
 434                }
 435         }
 436
 437         perm = mdt_identity_get_perm(identity, remote, peernid);
 438         /* find out the setuid/setgid attempt */
 439         setuid = (pud->pud_uid != pud->pud_fsuid);
 440         setgid = (pud->pud_gid != pud->pud_fsgid ||
 441                   pud->pud_gid != identity->mi_gid);
 442
 443         /* check permission of setuid */
 444         if (setuid && !(perm & CFS_SETUID_PERM)) {
 445                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 446                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 447                 GOTO(out, rc = -EACCES);
 448         }
 449
 450         /* check permission of setgid */
 451         if (setgid && !(perm & CFS_SETGID_PERM)) {
 452                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 453                        "from %s\n", pud->pud_uid, pud->pud_gid,
 454                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 455                        libcfs_nid2str(peernid));
 456                 GOTO(out, rc = -EACCES);
 457         }
 458
 459         EXIT;
 460
 461 out:
 462         mdt_identity_put(mdt->mdt_identity_cache, identity);
 463         return rc;
 464 }
 465
 466 static int old_init_ucred_common(struct mdt_thread_info *info,
 467                                  struct lu_nodemap *nodemap,
 468                                  bool drop_fs_cap)
 469 {
 470         struct lu_ucred         *uc = mdt_ucred(info);
 471         struct mdt_device       *mdt = info->mti_mdt;
 472         struct md_identity      *identity = NULL;
 473
 474         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 475                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 476                                             uc->uc_fsuid);
 477                 if (IS_ERR(identity)) {
 478                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 479                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 480                                 identity = NULL;
 481                         } else {
 482                                 CDEBUG(D_SEC, "Deny access without identity: "
 483                                        "uid %u\n", uc->uc_fsuid);
 484                                 RETURN(-EACCES);
 485                         }
 486                 }
 487         }
 488         uc->uc_identity = identity;
 489
 490         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 491                 uc->uc_fsuid = nodemap->nm_squash_uid;
 492                 uc->uc_fsgid = nodemap->nm_squash_gid;
 493                 uc->uc_cap = 0;
 494                 uc->uc_suppgids[0] = -1;
 495                 uc->uc_suppgids[1] = -1;
 496         }
 497
 498         /* process root_squash here. */
 499         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 500
 501         /* remove fs privilege for non-root user. */
 502         if (uc->uc_fsuid && drop_fs_cap)
 503                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 504         uc->uc_valid = UCRED_OLD;
 505         ucred_set_jobid(info, uc);
 506
 507         return 0;
 508 }
 509
 510 static int old_init_ucred(struct mdt_thread_info *info,
 511                           struct mdt_body *body, bool drop_fs_cap)
 512 {
 513         struct lu_ucred *uc = mdt_ucred(info);
 514         struct lu_nodemap *nodemap;
 515         int rc;
 516         ENTRY;
 517
 518         nodemap = nodemap_get_from_exp(info->mti_exp);
 519         if (IS_ERR(nodemap))
 520                 RETURN(PTR_ERR(nodemap));
 521
 522         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 523                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 524         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 525                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 526         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 527                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 528         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 529                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 530
 531         LASSERT(uc != NULL);
 532         uc->uc_valid = UCRED_INVALID;
 533         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 534         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 535         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 536         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 537         uc->uc_suppgids[0] = body->mbo_suppgid;
 538         uc->uc_suppgids[1] = -1;
 539         uc->uc_ginfo = NULL;
 540         uc->uc_cap = body->mbo_capability;
 541
 542         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 543         nodemap_putref(nodemap);
 544
 545         RETURN(rc);
 546 }
 547
 548 static int old_init_ucred_reint(struct mdt_thread_info *info)
 549 {
 550         struct lu_ucred *uc = mdt_ucred(info);
 551         struct lu_nodemap *nodemap;
 552         int rc;
 553         ENTRY;
 554
 555         nodemap = nodemap_get_from_exp(info->mti_exp);
 556         if (IS_ERR(nodemap))
 557                 RETURN(PTR_ERR(nodemap));
 558
 559         LASSERT(uc != NULL);
 560
 561         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 562                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 563         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 564                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 565
 566         uc->uc_valid = UCRED_INVALID;
 567         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 568         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 569         uc->uc_ginfo = NULL;
 570
 571         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 572         nodemap_putref(nodemap);
 573
 574         RETURN(rc);
 575 }
 576
 577 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 578                                    struct mdt_body *body,
 579                                    bool drop_fs_cap)
 580 {
 581         struct ptlrpc_request   *req = mdt_info_req(info);
 582         struct lu_ucred         *uc  = mdt_ucred(info);
 583
 584         LASSERT(uc != NULL);
 585         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 586                 return 0;
 587
 588         mdt_exit_ucred(info);
 589
 590         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 591                 return old_init_ucred(info, body, drop_fs_cap);
 592         else
 593                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 594 }
 595
 596 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 597 {
 598         return __mdt_init_ucred(info, body, true);
 599 }
 600
 601 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 602  * doesn't set correct fsuid explicitely, but raise capability to allow
 603  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 604  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 605  * will fail permission check. */
 606 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 607                                   struct mdt_body *body)
 608 {
 609         return __mdt_init_ucred(info, body, false);
 610 }
 611
 612 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 613 {
 614         struct ptlrpc_request *req = mdt_info_req(info);
 615         struct lu_ucred       *uc  = mdt_ucred(info);
 616
 617         LASSERT(uc != NULL);
 618         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 619                 return 0;
 620
 621         mdt_exit_ucred(info);
 622
 623         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 624                 return old_init_ucred_reint(info);
 625         else
 626                 return new_init_ucred(info, REC_INIT, NULL, true);
 627 }
 628
 629 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 630 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 631 {
 632         const struct lov_ost_data_v1    *lod;
 633         int                              i;
 634         __u16                            count;
 635
 636         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 637                 return;
 638
 639         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 640
 641         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 642                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 643                le32_to_cpu(lmm->lmm_pattern));
 644         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 645                le32_to_cpu(lmm->lmm_stripe_size), count);
 646
 647         /* If it's a directory or a released file, then there are
 648          * no actual objects to print, so bail out. */
 649         if (valid & OBD_MD_FLDIREA ||
 650             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 651                 return;
 652
 653         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 654         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 655                 struct ost_id oi;
 656
 657                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 658                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 659                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 660         }
 661 }
 662
 663 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 664 {
 665         const struct lmv_mds_md_v1 *lmm1;
 666         int                        i;
 667
 668         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 669                 return;
 670
 671         lmm1 = &lmv->lmv_md_v1;
 672         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 673                le32_to_cpu(lmm1->lmv_magic),
 674                le32_to_cpu(lmm1->lmv_master_mdt_index),
 675                le32_to_cpu(lmm1->lmv_stripe_count));
 676
 677         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 678                 return;
 679
 680         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 681                 struct lu_fid fid;
 682
 683                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 684                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 685         }
 686 }
 687
 688 /* Shrink and/or grow reply buffers */
 689 int mdt_fix_reply(struct mdt_thread_info *info)
 690 {
 691         struct req_capsule *pill = info->mti_pill;
 692         struct mdt_body    *body;
 693         int                md_size, md_packed = 0;
 694         int                acl_size;
 695         int                rc = 0;
 696         ENTRY;
 697
 698         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 699         LASSERT(body != NULL);
 700
 701         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 702                                OBD_MD_LINKNAME))
 703                 md_size = body->mbo_eadatasize;
 704         else
 705                 md_size = 0;
 706
 707         acl_size = body->mbo_aclsize;
 708
 709         /* this replay - not send info to client */
 710         if (info->mti_spec.no_create) {
 711                 md_size = 0;
 712                 acl_size = 0;
 713         }
 714
 715         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 716                md_size, acl_size);
 717 /*
 718             &RMF_MDT_BODY,
 719             &RMF_MDT_MD,
 720             &RMF_ACL, or &RMF_LOGCOOKIES
 721 (optional)  &RMF_CAPA1,
 722 (optional)  &RMF_CAPA2,
 723 (optional)  something else
 724 */
 725
 726         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 727          * buffers before growing it */
 728         if (info->mti_big_lmm_used) {
 729                 /* big_lmm buffer may be used even without packing the result
 730                  * into reply, just for internal server needs */
 731                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 732                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 733                                                          RCL_SERVER);
 734
 735                 /* free big lmm if md_size is not needed */
 736                 if (md_size == 0 || md_packed == 0) {
 737                         info->mti_big_lmm_used = 0;
 738                 } else {
 739                         /* buffer must be allocated separately */
 740                         LASSERT(info->mti_attr.ma_lmm !=
 741                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 742                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 743                 }
 744         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 745                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 746         }
 747
 748         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 749                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 750         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 751                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 752                                    acl_size, RCL_SERVER);
 753
 754         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 755             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 756                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 757
 758         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 759             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 760                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 761
 762         /*
 763          * Some more field should be shrinked if needed.
 764          * This should be done by those who added fields to reply message.
 765          */
 766
 767         /* Grow MD buffer if needed finally */
 768         if (info->mti_big_lmm_used) {
 769                 void *lmm;
 770
 771                 LASSERT(md_size > md_packed);
 772                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 773                        md_size - md_packed);
 774                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 775                 if (rc) {
 776                         /* we can't answer with proper LOV EA, drop flags,
 777                          * the rc is also returned so this request is
 778                          * considered as failed */
 779                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 780                         /* don't return transno along with error */
 781                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 782                 } else {
 783                         /* now we need to pack right LOV/LMV EA */
 784                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 785                         if (info->mti_attr.ma_valid & MA_LOV) {
 786                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 787                                                              RCL_SERVER) ==
 788                                                 info->mti_attr.ma_lmm_size);
 789                                 memcpy(lmm, info->mti_attr.ma_lmm,
 790                                        info->mti_attr.ma_lmm_size);
 791                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 792                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 793                                                              RCL_SERVER) ==
 794                                                 info->mti_attr.ma_lmv_size);
 795                                 memcpy(lmm, info->mti_attr.ma_lmv,
 796                                        info->mti_attr.ma_lmv_size);
 797                         }
 798                 }
 799                 /* update mdt_max_mdsize so clients will be aware about that */
 800                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 801                         info->mti_mdt->mdt_max_mdsize =
 802                                                     info->mti_attr.ma_lmm_size;
 803                 info->mti_big_lmm_used = 0;
 804         }
 805         RETURN(rc);
 806 }
 807
 808
 809 /* if object is dying, pack the lov/llog data,
 810  * parameter info->mti_attr should be valid at this point! */
 811 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 812                            const struct md_attr *ma)
 813 {
 814         struct mdt_body       *repbody;
 815         const struct lu_attr *la = &ma->ma_attr;
 816         ENTRY;
 817
 818         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 819         LASSERT(repbody != NULL);
 820
 821         if (ma->ma_valid & MA_INODE)
 822                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 823
 824         if (ma->ma_valid & MA_LOV) {
 825                 CERROR("No need in LOV EA upon unlink\n");
 826                 dump_stack();
 827         }
 828         repbody->mbo_eadatasize = 0;
 829
 830         RETURN(0);
 831 }
 832
 833 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 834                                   struct md_attr *ma)
 835 {
 836         __u64 out;
 837
 838         out = 0;
 839         if (in & MDS_ATTR_MODE)
 840                 out |= LA_MODE;
 841         if (in & MDS_ATTR_UID)
 842                 out |= LA_UID;
 843         if (in & MDS_ATTR_GID)
 844                 out |= LA_GID;
 845         if (in & MDS_ATTR_SIZE)
 846                 out |= LA_SIZE;
 847         if (in & MDS_ATTR_BLOCKS)
 848                 out |= LA_BLOCKS;
 849         if (in & MDS_ATTR_ATIME_SET)
 850                 out |= LA_ATIME;
 851         if (in & MDS_ATTR_CTIME_SET)
 852                 out |= LA_CTIME;
 853         if (in & MDS_ATTR_MTIME_SET)
 854                 out |= LA_MTIME;
 855         if (in & MDS_ATTR_ATTR_FLAG)
 856                 out |= LA_FLAGS;
 857         if (in & MDS_ATTR_KILL_SUID)
 858                 out |= LA_KILL_SUID;
 859         if (in & MDS_ATTR_KILL_SGID)
 860                 out |= LA_KILL_SGID;
 861
 862         if (in & MDS_ATTR_FROM_OPEN)
 863                 rr->rr_flags |= MRF_OPEN_TRUNC;
 864         if (in & MDS_OPEN_OWNEROVERRIDE)
 865                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 866         if (in & MDS_ATTR_FORCE)
 867                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 868
 869         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 870                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 871                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 872                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 873                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 874                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 875         if (in != 0)
 876                 CERROR("Unknown attr bits: "LPX64"\n", in);
 877         return out;
 878 }
 879
 880 /* unpacking */
 881
 882 int mdt_name_unpack(struct req_capsule *pill,
 883                     const struct req_msg_field *field,
 884                     struct lu_name *ln,
 885                     enum mdt_name_flags flags)
 886 {
 887         ln->ln_name = req_capsule_client_get(pill, field);
 888         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 889
 890         if (!lu_name_is_valid(ln)) {
 891                 ln->ln_name = NULL;
 892                 ln->ln_namelen = 0;
 893
 894                 return -EPROTO;
 895         }
 896
 897         if ((flags & MNF_FIX_ANON) &&
 898             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 899                 /* Newer (3.x) kernels use a name of "/" for the
 900                  * "anonymous" disconnected dentries from NFS
 901                  * filehandle conversion. See d_obtain_alias(). */
 902                 ln->ln_name = NULL;
 903                 ln->ln_namelen = 0;
 904         }
 905
 906         return 0;
 907 }
 908
 909 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 910 {
 911         struct lu_ucred         *uc = mdt_ucred(info);
 912         struct md_attr          *ma = &info->mti_attr;
 913         struct lu_attr          *la = &ma->ma_attr;
 914         struct req_capsule      *pill = info->mti_pill;
 915         struct mdt_reint_record *rr = &info->mti_rr;
 916         struct mdt_rec_setattr  *rec;
 917         struct lu_nodemap       *nodemap;
 918         ENTRY;
 919
 920         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 921         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 922         if (rec == NULL)
 923                 RETURN(-EFAULT);
 924
 925         /* This prior initialization is needed for old_init_ucred_reint() */
 926         uc->uc_fsuid = rec->sa_fsuid;
 927         uc->uc_fsgid = rec->sa_fsgid;
 928         uc->uc_cap   = rec->sa_cap;
 929         uc->uc_suppgids[0] = rec->sa_suppgid;
 930         uc->uc_suppgids[1] = -1;
 931
 932         rr->rr_fid1 = &rec->sa_fid;
 933         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 934         la->la_mode  = rec->sa_mode;
 935         la->la_flags = rec->sa_attr_flags;
 936
 937         nodemap = nodemap_get_from_exp(info->mti_exp);
 938         if (IS_ERR(nodemap))
 939                 RETURN(PTR_ERR(nodemap));
 940
 941         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 942                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 943         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 944                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 945         nodemap_putref(nodemap);
 946
 947         la->la_size  = rec->sa_size;
 948         la->la_blocks = rec->sa_blocks;
 949         la->la_ctime = rec->sa_ctime;
 950         la->la_atime = rec->sa_atime;
 951         la->la_mtime = rec->sa_mtime;
 952         ma->ma_valid = MA_INODE;
 953
 954         if (rec->sa_bias & MDS_DATA_MODIFIED)
 955                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 956         else
 957                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 958
 959         if (rec->sa_bias & MDS_HSM_RELEASE)
 960                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 961         else
 962                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 963
 964         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 965                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 966         else
 967                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 968
 969         RETURN(0);
 970 }
 971
 972 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 973 {
 974         struct req_capsule *pill = info->mti_pill;
 975         struct mdt_ioepoch *ioepoch;
 976         ENTRY;
 977
 978         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 979                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 980         else
 981                 ioepoch = NULL;
 982
 983         if (ioepoch == NULL)
 984                 RETURN(-EPROTO);
 985
 986         info->mti_close_handle = ioepoch->mio_handle;
 987
 988         RETURN(0);
 989 }
 990
 991 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 992         struct req_capsule      *pill = info->mti_pill;
 993
 994         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 995                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 996                 if (info->mti_dlm_req == NULL)
 997                         RETURN(-EFAULT);
 998         }
 999
1000         RETURN(0);
1001 }
1002
1003 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1004 {
1005         struct mdt_reint_record *rr = &info->mti_rr;
1006         struct md_attr          *ma = &info->mti_attr;
1007         struct req_capsule      *pill = info->mti_pill;
1008         int rc;
1009         ENTRY;
1010
1011         rc = mdt_setattr_unpack_rec(info);
1012         if (rc)
1013                 RETURN(rc);
1014
1015         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1016                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1017                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1018                                                         RCL_CLIENT);
1019                 if (rr->rr_eadatalen > 0) {
1020                         const struct lmv_user_md        *lum;
1021
1022                         lum = rr->rr_eadata;
1023                         /* Sigh ma_valid(from req) does not indicate whether
1024                          * it will set LOV/LMV EA, so we have to check magic */
1025                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1026                                 ma->ma_valid |= MA_LMV;
1027                                 ma->ma_lmv = (void *)rr->rr_eadata;
1028                                 ma->ma_lmv_size = rr->rr_eadatalen;
1029                         } else {
1030                                 ma->ma_valid |= MA_LOV;
1031                                 ma->ma_lmm = (void *)rr->rr_eadata;
1032                                 ma->ma_lmm_size = rr->rr_eadatalen;
1033                         }
1034                 }
1035         }
1036
1037         rc = mdt_dlmreq_unpack(info);
1038         RETURN(rc);
1039 }
1040
1041 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1042 {
1043         struct md_attr          *ma = &info->mti_attr;
1044         struct req_capsule      *pill = info->mti_pill;
1045         ENTRY;
1046
1047         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1048                 RETURN(0);
1049
1050         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1051
1052         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1053             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1054                 RETURN(-EFAULT);
1055
1056         RETURN(0);
1057 }
1058
1059 int mdt_close_unpack(struct mdt_thread_info *info)
1060 {
1061         int rc;
1062         ENTRY;
1063
1064         rc = mdt_close_handle_unpack(info);
1065         if (rc)
1066                 RETURN(rc);
1067
1068         rc = mdt_setattr_unpack_rec(info);
1069         if (rc)
1070                 RETURN(rc);
1071
1072         rc = mdt_intent_close_unpack(info);
1073         if (rc)
1074                 RETURN(rc);
1075
1076         RETURN(mdt_init_ucred_reint(info));
1077 }
1078
1079 static int mdt_create_unpack(struct mdt_thread_info *info)
1080 {
1081         struct lu_ucred         *uc  = mdt_ucred(info);
1082         struct mdt_rec_create   *rec;
1083         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1084         struct mdt_reint_record *rr = &info->mti_rr;
1085         struct req_capsule      *pill = info->mti_pill;
1086         struct md_op_spec       *sp = &info->mti_spec;
1087         int rc;
1088         ENTRY;
1089
1090         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1091         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1092         if (rec == NULL)
1093                 RETURN(-EFAULT);
1094
1095         /* This prior initialization is needed for old_init_ucred_reint() */
1096         uc->uc_fsuid = rec->cr_fsuid;
1097         uc->uc_fsgid = rec->cr_fsgid;
1098         uc->uc_cap   = rec->cr_cap;
1099         uc->uc_suppgids[0] = rec->cr_suppgid1;
1100         uc->uc_suppgids[1] = -1;
1101         uc->uc_umask = rec->cr_umask;
1102
1103         rr->rr_fid1 = &rec->cr_fid1;
1104         rr->rr_fid2 = &rec->cr_fid2;
1105         attr->la_mode = rec->cr_mode;
1106         attr->la_rdev  = rec->cr_rdev;
1107         attr->la_uid   = rec->cr_fsuid;
1108         attr->la_gid   = rec->cr_fsgid;
1109         attr->la_ctime = rec->cr_time;
1110         attr->la_mtime = rec->cr_time;
1111         attr->la_atime = rec->cr_time;
1112         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1113                          LA_CTIME | LA_MTIME | LA_ATIME;
1114         memset(&sp->u, 0, sizeof(sp->u));
1115         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1116
1117         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1118         if (rc < 0)
1119                 RETURN(rc);
1120
1121         if (S_ISLNK(attr->la_mode)) {
1122                 const char *tgt = NULL;
1123
1124                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1125                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1126                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1127                         sp->u.sp_symname = tgt;
1128                 }
1129                 if (tgt == NULL)
1130                         RETURN(-EFAULT);
1131         } else {
1132                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1133                 if (S_ISDIR(attr->la_mode) &&
1134                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1135                         sp->u.sp_ea.eadata =
1136                                 req_capsule_client_get(pill, &RMF_EADATA);
1137                         sp->u.sp_ea.eadatalen =
1138                                 req_capsule_get_size(pill, &RMF_EADATA,
1139                                                      RCL_CLIENT);
1140                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1141                 }
1142         }
1143
1144         rc = mdt_dlmreq_unpack(info);
1145         RETURN(rc);
1146 }
1147
1148 static int mdt_link_unpack(struct mdt_thread_info *info)
1149 {
1150         struct lu_ucred         *uc  = mdt_ucred(info);
1151         struct mdt_rec_link     *rec;
1152         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1153         struct mdt_reint_record *rr = &info->mti_rr;
1154         struct req_capsule      *pill = info->mti_pill;
1155         int rc;
1156         ENTRY;
1157
1158         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1159         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1160         if (rec == NULL)
1161                 RETURN(-EFAULT);
1162
1163         /* This prior initialization is needed for old_init_ucred_reint() */
1164         uc->uc_fsuid = rec->lk_fsuid;
1165         uc->uc_fsgid = rec->lk_fsgid;
1166         uc->uc_cap   = rec->lk_cap;
1167         uc->uc_suppgids[0] = rec->lk_suppgid1;
1168         uc->uc_suppgids[1] = rec->lk_suppgid2;
1169
1170         attr->la_uid = rec->lk_fsuid;
1171         attr->la_gid = rec->lk_fsgid;
1172         rr->rr_fid1 = &rec->lk_fid1;
1173         rr->rr_fid2 = &rec->lk_fid2;
1174         attr->la_ctime = rec->lk_time;
1175         attr->la_mtime = rec->lk_time;
1176         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1177
1178         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1179         if (rc < 0)
1180                 RETURN(rc);
1181
1182         rc = mdt_dlmreq_unpack(info);
1183
1184         RETURN(rc);
1185 }
1186
1187 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1188 {
1189         struct lu_ucred         *uc  = mdt_ucred(info);
1190         struct mdt_rec_unlink   *rec;
1191         struct md_attr          *ma = &info->mti_attr;
1192         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1193         struct mdt_reint_record *rr = &info->mti_rr;
1194         struct req_capsule      *pill = info->mti_pill;
1195         int rc;
1196         ENTRY;
1197
1198         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1199         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1200         if (rec == NULL)
1201                 RETURN(-EFAULT);
1202
1203         /* This prior initialization is needed for old_init_ucred_reint() */
1204         uc->uc_fsuid = rec->ul_fsuid;
1205         uc->uc_fsgid = rec->ul_fsgid;
1206         uc->uc_cap   = rec->ul_cap;
1207         uc->uc_suppgids[0] = rec->ul_suppgid1;
1208         uc->uc_suppgids[1] = -1;
1209
1210         attr->la_uid = rec->ul_fsuid;
1211         attr->la_gid = rec->ul_fsgid;
1212         rr->rr_fid1 = &rec->ul_fid1;
1213         rr->rr_fid2 = &rec->ul_fid2;
1214         attr->la_ctime = rec->ul_time;
1215         attr->la_mtime = rec->ul_time;
1216         attr->la_mode  = rec->ul_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         if (rec->ul_bias & MDS_VTX_BYPASS)
1224                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1225         else
1226                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1227
1228         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1229
1230         rc = mdt_dlmreq_unpack(info);
1231         RETURN(rc);
1232 }
1233
1234 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1235 {
1236         info->mti_spec.sp_rm_entry = 1;
1237         return mdt_unlink_unpack(info);
1238 }
1239
1240 static int mdt_rename_unpack(struct mdt_thread_info *info)
1241 {
1242         struct lu_ucred         *uc = mdt_ucred(info);
1243         struct mdt_rec_rename   *rec;
1244         struct md_attr          *ma = &info->mti_attr;
1245         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1246         struct mdt_reint_record *rr = &info->mti_rr;
1247         struct req_capsule      *pill = info->mti_pill;
1248         int rc;
1249         ENTRY;
1250
1251         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1252         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1253         if (rec == NULL)
1254                 RETURN(-EFAULT);
1255
1256         /* This prior initialization is needed for old_init_ucred_reint() */
1257         uc->uc_fsuid = rec->rn_fsuid;
1258         uc->uc_fsgid = rec->rn_fsgid;
1259         uc->uc_cap   = rec->rn_cap;
1260         uc->uc_suppgids[0] = rec->rn_suppgid1;
1261         uc->uc_suppgids[1] = rec->rn_suppgid2;
1262
1263         attr->la_uid = rec->rn_fsuid;
1264         attr->la_gid = rec->rn_fsgid;
1265         rr->rr_fid1 = &rec->rn_fid1;
1266         rr->rr_fid2 = &rec->rn_fid2;
1267         attr->la_ctime = rec->rn_time;
1268         attr->la_mtime = rec->rn_time;
1269         /* rename_tgt contains the mode already */
1270         attr->la_mode = rec->rn_mode;
1271         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1272
1273         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1274         if (rc < 0)
1275                 RETURN(rc);
1276
1277         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1278         if (rc < 0)
1279                 RETURN(rc);
1280
1281         if (rec->rn_bias & MDS_VTX_BYPASS)
1282                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1283         else
1284                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1285
1286         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1287                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1288                 rc = mdt_close_handle_unpack(info);
1289                 if (rc < 0)
1290                         RETURN(rc);
1291                 info->mti_spec.sp_migrate_close = 1;
1292         }
1293
1294         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1295
1296
1297         rc = mdt_dlmreq_unpack(info);
1298
1299         RETURN(rc);
1300 }
1301
1302 /*
1303  * please see comment above LOV_MAGIC_V1_DEF
1304  */
1305 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1306 {
1307         struct mdt_reint_record *rr = &info->mti_rr;
1308         struct lov_user_md_v1   *v1;
1309
1310         v1 = (void *)rr->rr_eadata;
1311         LASSERT(v1);
1312
1313         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1314                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1315                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1316                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1317                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1318                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1319                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1320                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1321                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1322                 }
1323         }
1324 }
1325
1326 static int mdt_open_unpack(struct mdt_thread_info *info)
1327 {
1328         struct lu_ucred         *uc = mdt_ucred(info);
1329         struct mdt_rec_create   *rec;
1330         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1331         struct req_capsule      *pill = info->mti_pill;
1332         struct mdt_reint_record *rr   = &info->mti_rr;
1333         struct ptlrpc_request   *req  = mdt_info_req(info);
1334         struct md_op_spec       *sp   = &info->mti_spec;
1335         ENTRY;
1336
1337         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1338         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1339         if (rec == NULL)
1340                 RETURN(-EFAULT);
1341
1342         /* This prior initialization is needed for old_init_ucred_reint() */
1343         uc->uc_fsuid = rec->cr_fsuid;
1344         uc->uc_fsgid = rec->cr_fsgid;
1345         uc->uc_cap   = rec->cr_cap;
1346         uc->uc_suppgids[0] = rec->cr_suppgid1;
1347         uc->uc_suppgids[1] = rec->cr_suppgid2;
1348         uc->uc_umask = rec->cr_umask;
1349
1350         rr->rr_fid1   = &rec->cr_fid1;
1351         rr->rr_fid2   = &rec->cr_fid2;
1352         rr->rr_handle = &rec->cr_old_handle;
1353         attr->la_mode = rec->cr_mode;
1354         attr->la_rdev  = rec->cr_rdev;
1355         attr->la_uid   = rec->cr_fsuid;
1356         attr->la_gid   = rec->cr_fsgid;
1357         attr->la_ctime = rec->cr_time;
1358         attr->la_mtime = rec->cr_time;
1359         attr->la_atime = rec->cr_time;
1360         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1361                          LA_CTIME | LA_MTIME | LA_ATIME;
1362         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1363         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1364         /* Do not trigger ASSERTION if client miss to set such flags. */
1365         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1366                 RETURN(-EPROTO);
1367
1368         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1369
1370         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1371
1372         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1373                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1374                                                         RCL_CLIENT);
1375                 if (rr->rr_eadatalen > 0) {
1376                         rr->rr_eadata = req_capsule_client_get(pill,
1377                                                                &RMF_EADATA);
1378                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1379                         sp->u.sp_ea.eadata = rr->rr_eadata;
1380                         sp->no_create = !!req_is_replay(req);
1381                         mdt_fix_lov_magic(info);
1382                 }
1383
1384                 /*
1385                  * Client default md_size may be 0 right after client start,
1386                  * until all osc are connected, set here just some reasonable
1387                  * value to prevent misbehavior.
1388                  */
1389                 if (rr->rr_eadatalen == 0 &&
1390                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1391                         rr->rr_eadatalen = MIN_MD_SIZE;
1392         }
1393
1394         RETURN(0);
1395 }
1396
1397 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1398 {
1399         struct mdt_reint_record *rr     = &info->mti_rr;
1400         struct lu_ucred         *uc     = mdt_ucred(info);
1401         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1402         struct req_capsule      *pill   = info->mti_pill;
1403         struct mdt_rec_setxattr *rec;
1404         int                      rc;
1405         ENTRY;
1406
1407
1408         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1409                          sizeof(struct mdt_rec_reint));
1410
1411         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1412         if (rec == NULL)
1413                 RETURN(-EFAULT);
1414
1415         /* This prior initialization is needed for old_init_ucred_reint() */
1416         uc->uc_fsuid  = rec->sx_fsuid;
1417         uc->uc_fsgid  = rec->sx_fsgid;
1418         uc->uc_cap    = rec->sx_cap;
1419         uc->uc_suppgids[0] = rec->sx_suppgid1;
1420         uc->uc_suppgids[1] = -1;
1421
1422         rr->rr_opcode = rec->sx_opcode;
1423         rr->rr_fid1   = &rec->sx_fid;
1424         attr->la_valid = rec->sx_valid;
1425         attr->la_ctime = rec->sx_time;
1426         attr->la_size = rec->sx_size;
1427         attr->la_flags = rec->sx_flags;
1428
1429         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1430         if (rc < 0)
1431                 RETURN(rc);
1432
1433         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1434                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1435                                                         RCL_CLIENT);
1436                 if (rr->rr_eadatalen > 0) {
1437                         rr->rr_eadata = req_capsule_client_get(pill,
1438                                                                &RMF_EADATA);
1439                         if (rr->rr_eadata == NULL)
1440                                 RETURN(-EFAULT);
1441                 } else {
1442                         rr->rr_eadata = NULL;
1443                 }
1444         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1445                 CDEBUG(D_INFO, "no xattr data supplied\n");
1446                 RETURN(-EFAULT);
1447         }
1448
1449         if (mdt_dlmreq_unpack(info) < 0)
1450                 RETURN(-EPROTO);
1451
1452         RETURN(0);
1453 }
1454
1455
1456 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1457
1458 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1459         [REINT_SETATTR]  = mdt_setattr_unpack,
1460         [REINT_CREATE]   = mdt_create_unpack,
1461         [REINT_LINK]     = mdt_link_unpack,
1462         [REINT_UNLINK]   = mdt_unlink_unpack,
1463         [REINT_RENAME]   = mdt_rename_unpack,
1464         [REINT_OPEN]     = mdt_open_unpack,
1465         [REINT_SETXATTR] = mdt_setxattr_unpack,
1466         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1467         [REINT_MIGRATE]  = mdt_rename_unpack,
1468 };
1469
1470 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1471 {
1472         int rc;
1473         ENTRY;
1474
1475         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1476         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1477                 info->mti_rr.rr_opcode = op;
1478                 rc = mdt_reint_unpackers[op](info);
1479         } else {
1480                 CERROR("Unexpected opcode %d\n", op);
1481                 rc = -EFAULT;
1482         }
1483         RETURN(rc);
1484 }