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 (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 199                     req->rq_auth_uid != pud->pud_uid) {
 200                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 201                                "while client claims %u:%u/%u:%u\n",
 202                                libcfs_nid2str(peernid), req->rq_auth_uid,
 203                                pud->pud_uid, pud->pud_gid,
 204                                pud->pud_fsuid, pud->pud_fsgid);
 205                         RETURN(-EACCES);
 206                 }
 207         }
 208
 209         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 210                 if (remote) {
 211                         CDEBUG(D_SEC, "remote client must run with identity_get "
 212                                "enabled!\n");
 213                         RETURN(-EACCES);
 214                 } else {
 215                         ucred->uc_identity = NULL;
 216                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 217                                CFS_SETGRP_PERM;
 218                 }
 219         } else {
 220                 struct md_identity *identity;
 221
 222                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 223                                             pud->pud_uid);
 224                 if (IS_ERR(identity)) {
 225                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 226                                      !remote)) {
 227                                 ucred->uc_identity = NULL;
 228                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 229                                        CFS_SETGRP_PERM;
 230                         } else {
 231                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 232                                        pud->pud_uid);
 233                                 RETURN(-EACCES);
 234                         }
 235                 } else {
 236                         ucred->uc_identity = identity;
 237                         perm = mdt_identity_get_perm(ucred->uc_identity,
 238                                                      remote, peernid);
 239                 }
 240         }
 241
 242         /* find out the setuid/setgid attempt */
 243         setuid = (pud->pud_uid != pud->pud_fsuid);
 244         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 245                   (ucred->uc_identity &&
 246                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 247
 248         /* check permission of setuid */
 249         if (setuid && !(perm & CFS_SETUID_PERM)) {
 250                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 251                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 252                 GOTO(out, rc = -EACCES);
 253         }
 254
 255         /* check permission of setgid */
 256         if (setgid && !(perm & CFS_SETGID_PERM)) {
 257                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 258                        "from %s\n", pud->pud_uid, pud->pud_gid,
 259                        pud->pud_fsuid, pud->pud_fsgid,
 260                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 261                 GOTO(out, rc = -EACCES);
 262         }
 263
 264         /*
 265          * NB: remote client not allowed to setgroups anyway.
 266          */
 267         if (!remote && perm & CFS_SETGRP_PERM) {
 268                 if (pud->pud_ngroups) {
 269                         /* setgroups for local client */
 270                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 271                         if (!ucred->uc_ginfo) {
 272                                 CERROR("failed to alloc %d groups\n",
 273                                        pud->pud_ngroups);
 274                                 GOTO(out, rc = -ENOMEM);
 275                         }
 276
 277                         lustre_groups_from_list(ucred->uc_ginfo,
 278                                                 pud->pud_groups);
 279                         lustre_groups_sort(ucred->uc_ginfo);
 280                 } else {
 281                         ucred->uc_ginfo = NULL;
 282                 }
 283         } else {
 284                 ucred->uc_suppgids[0] = -1;
 285                 ucred->uc_suppgids[1] = -1;
 286                 ucred->uc_ginfo = NULL;
 287         }
 288
 289         ucred->uc_uid   = pud->pud_uid;
 290         ucred->uc_gid   = pud->pud_gid;
 291         ucred->uc_fsuid = pud->pud_fsuid;
 292         ucred->uc_fsgid = pud->pud_fsgid;
 293
 294         /* process root_squash here. */
 295         mdt_root_squash(info, peernid);
 296
 297         /* remove fs privilege for non-root user. */
 298         if (ucred->uc_fsuid && drop_fs_cap)
 299                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 300         else
 301                 ucred->uc_cap = pud->pud_cap;
 302         if (remote && !(perm & CFS_RMTOWN_PERM))
 303                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 304                                    CFS_CAP_CHOWN_MASK);
 305         ucred->uc_valid = UCRED_NEW;
 306         ucred_set_jobid(info, ucred);
 307
 308         EXIT;
 309
 310 out:
 311         if (rc) {
 312                 if (ucred->uc_ginfo) {
 313                         put_group_info(ucred->uc_ginfo);
 314                         ucred->uc_ginfo = NULL;
 315                 }
 316                 if (ucred->uc_identity) {
 317                         mdt_identity_put(mdt->mdt_identity_cache,
 318                                          ucred->uc_identity);
 319                         ucred->uc_identity = NULL;
 320                 }
 321         }
 322
 323         return rc;
 324 }
 325
 326 /**
 327  * Check whether allow the client to set supplementary group IDs or not.
 328  *
 329  * \param[in] info      pointer to the thread context
 330  * \param[in] uc        pointer to the RPC user descriptor
 331  *
 332  * \retval              true if allow to set supplementary group IDs
 333  * \retval              false for other cases
 334  */
 335 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 336 {
 337         __u32 remote = exp_connect_rmtclient(info->mti_exp);
 338         __u32 perm;
 339
 340         /* 1. If identity_upcall is disabled, then forbid remote client to set
 341          *    supplementary group IDs, but permit local client to do that. */
 342         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) {
 343                 if (remote)
 344                         return false;
 345
 346                 return true;
 347         }
 348
 349         /* 2. If fail to get related identities, then forbid any client to
 350          *    set supplementary group IDs. */
 351         if (uc->uc_identity == NULL)
 352                 return false;
 353
 354         /* 3. Check the permission in the identities. */
 355         perm = mdt_identity_get_perm(uc->uc_identity, remote,
 356                                      mdt_info_req(info)->rq_peer.nid);
 357         if (perm & CFS_SETGRP_PERM)
 358                 return true;
 359
 360         return false;
 361 }
 362
 363 int mdt_check_ucred(struct mdt_thread_info *info)
 364 {
 365         struct ptlrpc_request   *req = mdt_info_req(info);
 366         struct mdt_device       *mdt = info->mti_mdt;
 367         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 368         struct lu_ucred         *ucred = mdt_ucred(info);
 369         struct md_identity      *identity = NULL;
 370         lnet_nid_t               peernid = req->rq_peer.nid;
 371         __u32                    perm = 0;
 372         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 373         int                      setuid;
 374         int                      setgid;
 375         int                      rc = 0;
 376
 377         ENTRY;
 378
 379         LASSERT(ucred != NULL);
 380         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 381                 RETURN(0);
 382
 383         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 384                 RETURN(0);
 385
 386         /* sanity check: if we use strong authentication, we expect the
 387          * uid which client claimed is true */
 388         if (remote) {
 389                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 390                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 391                         RETURN(-EACCES);
 392                 }
 393
 394                 if (ptlrpc_user_desc_do_idmap(req, pud))
 395                         RETURN(-EACCES);
 396
 397                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 398                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 399                                "while client claims %u:%u/%u:%u\n",
 400                                libcfs_nid2str(peernid), req->rq_auth_uid,
 401                                req->rq_auth_mapped_uid,
 402                                pud->pud_uid, pud->pud_gid,
 403                                pud->pud_fsuid, pud->pud_fsgid);
 404                         RETURN(-EACCES);
 405                 }
 406         } else {
 407                 if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 408                     req->rq_auth_uid != pud->pud_uid) {
 409                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 410                                "while client claims %u:%u/%u:%u\n",
 411                                libcfs_nid2str(peernid), req->rq_auth_uid,
 412                                pud->pud_uid, pud->pud_gid,
 413                                pud->pud_fsuid, pud->pud_fsgid);
 414                         RETURN(-EACCES);
 415                 }
 416         }
 417
 418         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 419                 if (remote) {
 420                         CDEBUG(D_SEC, "remote client must run with identity_get "
 421                                "enabled!\n");
 422                         RETURN(-EACCES);
 423                 }
 424                 RETURN(0);
 425         }
 426
 427         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 428         if (IS_ERR(identity)) {
 429                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 430                              !remote)) {
 431                         RETURN(0);
 432                 } else {
 433                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 434                                pud->pud_uid);
 435                         RETURN(-EACCES);
 436                }
 437         }
 438
 439         perm = mdt_identity_get_perm(identity, remote, peernid);
 440         /* find out the setuid/setgid attempt */
 441         setuid = (pud->pud_uid != pud->pud_fsuid);
 442         setgid = (pud->pud_gid != pud->pud_fsgid ||
 443                   pud->pud_gid != identity->mi_gid);
 444
 445         /* check permission of setuid */
 446         if (setuid && !(perm & CFS_SETUID_PERM)) {
 447                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 448                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 449                 GOTO(out, rc = -EACCES);
 450         }
 451
 452         /* check permission of setgid */
 453         if (setgid && !(perm & CFS_SETGID_PERM)) {
 454                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 455                        "from %s\n", pud->pud_uid, pud->pud_gid,
 456                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 457                        libcfs_nid2str(peernid));
 458                 GOTO(out, rc = -EACCES);
 459         }
 460
 461         EXIT;
 462
 463 out:
 464         mdt_identity_put(mdt->mdt_identity_cache, identity);
 465         return rc;
 466 }
 467
 468 static int old_init_ucred_common(struct mdt_thread_info *info,
 469                                  struct lu_nodemap *nodemap,
 470                                  bool drop_fs_cap)
 471 {
 472         struct lu_ucred         *uc = mdt_ucred(info);
 473         struct mdt_device       *mdt = info->mti_mdt;
 474         struct md_identity      *identity = NULL;
 475
 476         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 477                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 478                                             uc->uc_fsuid);
 479                 if (IS_ERR(identity)) {
 480                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 481                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 482                                 identity = NULL;
 483                         } else {
 484                                 CDEBUG(D_SEC, "Deny access without identity: "
 485                                        "uid %u\n", uc->uc_fsuid);
 486                                 RETURN(-EACCES);
 487                         }
 488                 }
 489         }
 490         uc->uc_identity = identity;
 491
 492         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 493                 uc->uc_fsuid = nodemap->nm_squash_uid;
 494                 uc->uc_fsgid = nodemap->nm_squash_gid;
 495                 uc->uc_cap = 0;
 496                 uc->uc_suppgids[0] = -1;
 497                 uc->uc_suppgids[1] = -1;
 498         }
 499
 500         /* process root_squash here. */
 501         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 502
 503         /* remove fs privilege for non-root user. */
 504         if (uc->uc_fsuid && drop_fs_cap)
 505                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 506         uc->uc_valid = UCRED_OLD;
 507         ucred_set_jobid(info, uc);
 508
 509         return 0;
 510 }
 511
 512 static int old_init_ucred(struct mdt_thread_info *info,
 513                           struct mdt_body *body, bool drop_fs_cap)
 514 {
 515         struct lu_ucred *uc = mdt_ucred(info);
 516         struct lu_nodemap *nodemap;
 517         int rc;
 518         ENTRY;
 519
 520         nodemap = nodemap_get_from_exp(info->mti_exp);
 521         if (IS_ERR(nodemap))
 522                 RETURN(PTR_ERR(nodemap));
 523
 524         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 525                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 526         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 527                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 528         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 529                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 530         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 531                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 532
 533         LASSERT(uc != NULL);
 534         uc->uc_valid = UCRED_INVALID;
 535         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 536         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 537         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 538         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 539         uc->uc_suppgids[0] = body->mbo_suppgid;
 540         uc->uc_suppgids[1] = -1;
 541         uc->uc_ginfo = NULL;
 542         uc->uc_cap = body->mbo_capability;
 543
 544         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 545         nodemap_putref(nodemap);
 546
 547         RETURN(rc);
 548 }
 549
 550 static int old_init_ucred_reint(struct mdt_thread_info *info)
 551 {
 552         struct lu_ucred *uc = mdt_ucred(info);
 553         struct lu_nodemap *nodemap;
 554         int rc;
 555         ENTRY;
 556
 557         nodemap = nodemap_get_from_exp(info->mti_exp);
 558         if (IS_ERR(nodemap))
 559                 RETURN(PTR_ERR(nodemap));
 560
 561         LASSERT(uc != NULL);
 562
 563         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 564                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 565         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 566                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 567
 568         uc->uc_valid = UCRED_INVALID;
 569         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 570         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 571         uc->uc_ginfo = NULL;
 572
 573         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 574         nodemap_putref(nodemap);
 575
 576         RETURN(rc);
 577 }
 578
 579 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 580                                    struct mdt_body *body,
 581                                    bool drop_fs_cap)
 582 {
 583         struct ptlrpc_request   *req = mdt_info_req(info);
 584         struct lu_ucred         *uc  = mdt_ucred(info);
 585
 586         LASSERT(uc != NULL);
 587         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 588                 return 0;
 589
 590         mdt_exit_ucred(info);
 591
 592         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 593                 return old_init_ucred(info, body, drop_fs_cap);
 594         else
 595                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 596 }
 597
 598 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 599 {
 600         return __mdt_init_ucred(info, body, true);
 601 }
 602
 603 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 604  * doesn't set correct fsuid explicitely, but raise capability to allow
 605  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 606  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 607  * will fail permission check. */
 608 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 609                                   struct mdt_body *body)
 610 {
 611         return __mdt_init_ucred(info, body, false);
 612 }
 613
 614 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 615 {
 616         struct ptlrpc_request *req = mdt_info_req(info);
 617         struct lu_ucred       *uc  = mdt_ucred(info);
 618
 619         LASSERT(uc != NULL);
 620         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 621                 return 0;
 622
 623         mdt_exit_ucred(info);
 624
 625         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 626                 return old_init_ucred_reint(info);
 627         else
 628                 return new_init_ucred(info, REC_INIT, NULL, true);
 629 }
 630
 631 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 632 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 633 {
 634         const struct lov_ost_data_v1    *lod;
 635         int                              i;
 636         __u16                            count;
 637
 638         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 639                 return;
 640
 641         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 642
 643         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 644                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 645                le32_to_cpu(lmm->lmm_pattern));
 646         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 647                le32_to_cpu(lmm->lmm_stripe_size), count);
 648
 649         /* If it's a directory or a released file, then there are
 650          * no actual objects to print, so bail out. */
 651         if (valid & OBD_MD_FLDIREA ||
 652             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 653                 return;
 654
 655         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 656         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 657                 struct ost_id oi;
 658
 659                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 660                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 661                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 662         }
 663 }
 664
 665 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 666 {
 667         const struct lmv_mds_md_v1 *lmm1;
 668         int                        i;
 669
 670         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 671                 return;
 672
 673         lmm1 = &lmv->lmv_md_v1;
 674         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 675                le32_to_cpu(lmm1->lmv_magic),
 676                le32_to_cpu(lmm1->lmv_master_mdt_index),
 677                le32_to_cpu(lmm1->lmv_stripe_count));
 678
 679         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 680                 return;
 681
 682         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 683                 struct lu_fid fid;
 684
 685                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 686                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 687         }
 688 }
 689
 690 /* Shrink and/or grow reply buffers */
 691 int mdt_fix_reply(struct mdt_thread_info *info)
 692 {
 693         struct req_capsule *pill = info->mti_pill;
 694         struct mdt_body    *body;
 695         int                md_size, md_packed = 0;
 696         int                acl_size;
 697         int                rc = 0;
 698         ENTRY;
 699
 700         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 701         LASSERT(body != NULL);
 702
 703         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 704                                OBD_MD_LINKNAME))
 705                 md_size = body->mbo_eadatasize;
 706         else
 707                 md_size = 0;
 708
 709         acl_size = body->mbo_aclsize;
 710
 711         /* this replay - not send info to client */
 712         if (info->mti_spec.no_create) {
 713                 md_size = 0;
 714                 acl_size = 0;
 715         }
 716
 717         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 718                md_size, acl_size);
 719 /*
 720             &RMF_MDT_BODY,
 721             &RMF_MDT_MD,
 722             &RMF_ACL, or &RMF_LOGCOOKIES
 723 (optional)  &RMF_CAPA1,
 724 (optional)  &RMF_CAPA2,
 725 (optional)  something else
 726 */
 727
 728         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 729          * buffers before growing it */
 730         if (info->mti_big_lmm_used) {
 731                 /* big_lmm buffer may be used even without packing the result
 732                  * into reply, just for internal server needs */
 733                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 734                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 735                                                          RCL_SERVER);
 736
 737                 /* free big lmm if md_size is not needed */
 738                 if (md_size == 0 || md_packed == 0) {
 739                         info->mti_big_lmm_used = 0;
 740                 } else {
 741                         /* buffer must be allocated separately */
 742                         LASSERT(info->mti_attr.ma_lmm !=
 743                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 744                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 745                 }
 746         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 747                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 748         }
 749
 750         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 751                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 752         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 753                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 754                                    acl_size, RCL_SERVER);
 755
 756         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 757             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 758                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 759
 760         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 761             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 762                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 763
 764         /*
 765          * Some more field should be shrinked if needed.
 766          * This should be done by those who added fields to reply message.
 767          */
 768
 769         /* Grow MD buffer if needed finally */
 770         if (info->mti_big_lmm_used) {
 771                 void *lmm;
 772
 773                 LASSERT(md_size > md_packed);
 774                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 775                        md_size - md_packed);
 776                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 777                 if (rc) {
 778                         /* we can't answer with proper LOV EA, drop flags,
 779                          * the rc is also returned so this request is
 780                          * considered as failed */
 781                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 782                         /* don't return transno along with error */
 783                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 784                 } else {
 785                         /* now we need to pack right LOV/LMV EA */
 786                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 787                         if (info->mti_attr.ma_valid & MA_LOV) {
 788                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 789                                                              RCL_SERVER) ==
 790                                                 info->mti_attr.ma_lmm_size);
 791                                 memcpy(lmm, info->mti_attr.ma_lmm,
 792                                        info->mti_attr.ma_lmm_size);
 793                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 794                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 795                                                              RCL_SERVER) ==
 796                                                 info->mti_attr.ma_lmv_size);
 797                                 memcpy(lmm, info->mti_attr.ma_lmv,
 798                                        info->mti_attr.ma_lmv_size);
 799                         }
 800                 }
 801                 /* update mdt_max_mdsize so clients will be aware about that */
 802                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 803                         info->mti_mdt->mdt_max_mdsize =
 804                                                     info->mti_attr.ma_lmm_size;
 805                 info->mti_big_lmm_used = 0;
 806         }
 807         RETURN(rc);
 808 }
 809
 810
 811 /* if object is dying, pack the lov/llog data,
 812  * parameter info->mti_attr should be valid at this point! */
 813 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 814                            const struct md_attr *ma)
 815 {
 816         struct mdt_body       *repbody;
 817         const struct lu_attr *la = &ma->ma_attr;
 818         ENTRY;
 819
 820         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 821         LASSERT(repbody != NULL);
 822
 823         if (ma->ma_valid & MA_INODE)
 824                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 825
 826         if (ma->ma_valid & MA_LOV) {
 827                 CERROR("No need in LOV EA upon unlink\n");
 828                 dump_stack();
 829         }
 830         repbody->mbo_eadatasize = 0;
 831
 832         RETURN(0);
 833 }
 834
 835 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 836                                   struct md_attr *ma)
 837 {
 838         __u64 out;
 839
 840         out = 0;
 841         if (in & MDS_ATTR_MODE)
 842                 out |= LA_MODE;
 843         if (in & MDS_ATTR_UID)
 844                 out |= LA_UID;
 845         if (in & MDS_ATTR_GID)
 846                 out |= LA_GID;
 847         if (in & MDS_ATTR_SIZE)
 848                 out |= LA_SIZE;
 849         if (in & MDS_ATTR_BLOCKS)
 850                 out |= LA_BLOCKS;
 851         if (in & MDS_ATTR_ATIME_SET)
 852                 out |= LA_ATIME;
 853         if (in & MDS_ATTR_CTIME_SET)
 854                 out |= LA_CTIME;
 855         if (in & MDS_ATTR_MTIME_SET)
 856                 out |= LA_MTIME;
 857         if (in & MDS_ATTR_ATTR_FLAG)
 858                 out |= LA_FLAGS;
 859         if (in & MDS_ATTR_KILL_SUID)
 860                 out |= LA_KILL_SUID;
 861         if (in & MDS_ATTR_KILL_SGID)
 862                 out |= LA_KILL_SGID;
 863
 864         if (in & MDS_ATTR_FROM_OPEN)
 865                 rr->rr_flags |= MRF_OPEN_TRUNC;
 866         if (in & MDS_OPEN_OWNEROVERRIDE)
 867                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 868         if (in & MDS_ATTR_FORCE)
 869                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 870
 871         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 872                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 873                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 874                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 875                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 876                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 877         if (in != 0)
 878                 CERROR("Unknown attr bits: "LPX64"\n", in);
 879         return out;
 880 }
 881
 882 /* unpacking */
 883
 884 int mdt_name_unpack(struct req_capsule *pill,
 885                     const struct req_msg_field *field,
 886                     struct lu_name *ln,
 887                     enum mdt_name_flags flags)
 888 {
 889         ln->ln_name = req_capsule_client_get(pill, field);
 890         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 891
 892         if (!lu_name_is_valid(ln)) {
 893                 ln->ln_name = NULL;
 894                 ln->ln_namelen = 0;
 895
 896                 return -EPROTO;
 897         }
 898
 899         if ((flags & MNF_FIX_ANON) &&
 900             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 901                 /* Newer (3.x) kernels use a name of "/" for the
 902                  * "anonymous" disconnected dentries from NFS
 903                  * filehandle conversion. See d_obtain_alias(). */
 904                 ln->ln_name = NULL;
 905                 ln->ln_namelen = 0;
 906         }
 907
 908         return 0;
 909 }
 910
 911 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 912 {
 913         struct lu_ucred         *uc = mdt_ucred(info);
 914         struct md_attr          *ma = &info->mti_attr;
 915         struct lu_attr          *la = &ma->ma_attr;
 916         struct req_capsule      *pill = info->mti_pill;
 917         struct mdt_reint_record *rr = &info->mti_rr;
 918         struct mdt_rec_setattr  *rec;
 919         struct lu_nodemap       *nodemap;
 920         ENTRY;
 921
 922         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 923         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 924         if (rec == NULL)
 925                 RETURN(-EFAULT);
 926
 927         /* This prior initialization is needed for old_init_ucred_reint() */
 928         uc->uc_fsuid = rec->sa_fsuid;
 929         uc->uc_fsgid = rec->sa_fsgid;
 930         uc->uc_cap   = rec->sa_cap;
 931         uc->uc_suppgids[0] = rec->sa_suppgid;
 932         uc->uc_suppgids[1] = -1;
 933
 934         rr->rr_fid1 = &rec->sa_fid;
 935         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 936         la->la_mode  = rec->sa_mode;
 937         la->la_flags = rec->sa_attr_flags;
 938
 939         nodemap = nodemap_get_from_exp(info->mti_exp);
 940         if (IS_ERR(nodemap))
 941                 RETURN(PTR_ERR(nodemap));
 942
 943         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 944                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 945         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 946                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 947         nodemap_putref(nodemap);
 948
 949         la->la_size  = rec->sa_size;
 950         la->la_blocks = rec->sa_blocks;
 951         la->la_ctime = rec->sa_ctime;
 952         la->la_atime = rec->sa_atime;
 953         la->la_mtime = rec->sa_mtime;
 954         ma->ma_valid = MA_INODE;
 955
 956         if (rec->sa_bias & MDS_DATA_MODIFIED)
 957                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 958         else
 959                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 960
 961         if (rec->sa_bias & MDS_HSM_RELEASE)
 962                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 963         else
 964                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 965
 966         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 967                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 968         else
 969                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 970
 971         RETURN(0);
 972 }
 973
 974 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 975 {
 976         struct req_capsule *pill = info->mti_pill;
 977         struct mdt_ioepoch *ioepoch;
 978         ENTRY;
 979
 980         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 981                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 982         else
 983                 ioepoch = NULL;
 984
 985         if (ioepoch == NULL)
 986                 RETURN(-EPROTO);
 987
 988         info->mti_close_handle = ioepoch->mio_handle;
 989
 990         RETURN(0);
 991 }
 992
 993 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 994         struct req_capsule      *pill = info->mti_pill;
 995
 996         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 997                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 998                 if (info->mti_dlm_req == NULL)
 999                         RETURN(-EFAULT);
1000         }
1001
1002         RETURN(0);
1003 }
1004
1005 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1006 {
1007         struct mdt_reint_record *rr = &info->mti_rr;
1008         struct md_attr          *ma = &info->mti_attr;
1009         struct req_capsule      *pill = info->mti_pill;
1010         int rc;
1011         ENTRY;
1012
1013         rc = mdt_setattr_unpack_rec(info);
1014         if (rc)
1015                 RETURN(rc);
1016
1017         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1018                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1019                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1020                                                         RCL_CLIENT);
1021                 if (rr->rr_eadatalen > 0) {
1022                         const struct lmv_user_md        *lum;
1023
1024                         lum = rr->rr_eadata;
1025                         /* Sigh ma_valid(from req) does not indicate whether
1026                          * it will set LOV/LMV EA, so we have to check magic */
1027                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1028                                 ma->ma_valid |= MA_LMV;
1029                                 ma->ma_lmv = (void *)rr->rr_eadata;
1030                                 ma->ma_lmv_size = rr->rr_eadatalen;
1031                         } else {
1032                                 ma->ma_valid |= MA_LOV;
1033                                 ma->ma_lmm = (void *)rr->rr_eadata;
1034                                 ma->ma_lmm_size = rr->rr_eadatalen;
1035                         }
1036                 }
1037         }
1038
1039         rc = mdt_dlmreq_unpack(info);
1040         RETURN(rc);
1041 }
1042
1043 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1044 {
1045         struct md_attr          *ma = &info->mti_attr;
1046         struct req_capsule      *pill = info->mti_pill;
1047         ENTRY;
1048
1049         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1050                 RETURN(0);
1051
1052         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1053
1054         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1055             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1056                 RETURN(-EFAULT);
1057
1058         RETURN(0);
1059 }
1060
1061 int mdt_close_unpack(struct mdt_thread_info *info)
1062 {
1063         int rc;
1064         ENTRY;
1065
1066         rc = mdt_close_handle_unpack(info);
1067         if (rc)
1068                 RETURN(rc);
1069
1070         rc = mdt_setattr_unpack_rec(info);
1071         if (rc)
1072                 RETURN(rc);
1073
1074         rc = mdt_intent_close_unpack(info);
1075         if (rc)
1076                 RETURN(rc);
1077
1078         RETURN(mdt_init_ucred_reint(info));
1079 }
1080
1081 static int mdt_create_unpack(struct mdt_thread_info *info)
1082 {
1083         struct lu_ucred         *uc  = mdt_ucred(info);
1084         struct mdt_rec_create   *rec;
1085         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1086         struct mdt_reint_record *rr = &info->mti_rr;
1087         struct req_capsule      *pill = info->mti_pill;
1088         struct md_op_spec       *sp = &info->mti_spec;
1089         int rc;
1090         ENTRY;
1091
1092         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1093         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1094         if (rec == NULL)
1095                 RETURN(-EFAULT);
1096
1097         /* This prior initialization is needed for old_init_ucred_reint() */
1098         uc->uc_fsuid = rec->cr_fsuid;
1099         uc->uc_fsgid = rec->cr_fsgid;
1100         uc->uc_cap   = rec->cr_cap;
1101         uc->uc_suppgids[0] = rec->cr_suppgid1;
1102         uc->uc_suppgids[1] = -1;
1103         uc->uc_umask = rec->cr_umask;
1104
1105         rr->rr_fid1 = &rec->cr_fid1;
1106         rr->rr_fid2 = &rec->cr_fid2;
1107         attr->la_mode = rec->cr_mode;
1108         attr->la_rdev  = rec->cr_rdev;
1109         attr->la_uid   = rec->cr_fsuid;
1110         attr->la_gid   = rec->cr_fsgid;
1111         attr->la_ctime = rec->cr_time;
1112         attr->la_mtime = rec->cr_time;
1113         attr->la_atime = rec->cr_time;
1114         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1115                          LA_CTIME | LA_MTIME | LA_ATIME;
1116         memset(&sp->u, 0, sizeof(sp->u));
1117         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1118
1119         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1120         if (rc < 0)
1121                 RETURN(rc);
1122
1123         if (S_ISLNK(attr->la_mode)) {
1124                 const char *tgt = NULL;
1125
1126                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1127                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1128                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1129                         sp->u.sp_symname = tgt;
1130                 }
1131                 if (tgt == NULL)
1132                         RETURN(-EFAULT);
1133         } else {
1134                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1135                 if (S_ISDIR(attr->la_mode) &&
1136                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1137                         sp->u.sp_ea.eadata =
1138                                 req_capsule_client_get(pill, &RMF_EADATA);
1139                         sp->u.sp_ea.eadatalen =
1140                                 req_capsule_get_size(pill, &RMF_EADATA,
1141                                                      RCL_CLIENT);
1142                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1143                 }
1144         }
1145
1146         rc = mdt_dlmreq_unpack(info);
1147         RETURN(rc);
1148 }
1149
1150 static int mdt_link_unpack(struct mdt_thread_info *info)
1151 {
1152         struct lu_ucred         *uc  = mdt_ucred(info);
1153         struct mdt_rec_link     *rec;
1154         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1155         struct mdt_reint_record *rr = &info->mti_rr;
1156         struct req_capsule      *pill = info->mti_pill;
1157         int rc;
1158         ENTRY;
1159
1160         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1161         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1162         if (rec == NULL)
1163                 RETURN(-EFAULT);
1164
1165         /* This prior initialization is needed for old_init_ucred_reint() */
1166         uc->uc_fsuid = rec->lk_fsuid;
1167         uc->uc_fsgid = rec->lk_fsgid;
1168         uc->uc_cap   = rec->lk_cap;
1169         uc->uc_suppgids[0] = rec->lk_suppgid1;
1170         uc->uc_suppgids[1] = rec->lk_suppgid2;
1171
1172         attr->la_uid = rec->lk_fsuid;
1173         attr->la_gid = rec->lk_fsgid;
1174         rr->rr_fid1 = &rec->lk_fid1;
1175         rr->rr_fid2 = &rec->lk_fid2;
1176         attr->la_ctime = rec->lk_time;
1177         attr->la_mtime = rec->lk_time;
1178         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1179
1180         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1181         if (rc < 0)
1182                 RETURN(rc);
1183
1184         rc = mdt_dlmreq_unpack(info);
1185
1186         RETURN(rc);
1187 }
1188
1189 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1190 {
1191         struct lu_ucred         *uc  = mdt_ucred(info);
1192         struct mdt_rec_unlink   *rec;
1193         struct md_attr          *ma = &info->mti_attr;
1194         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1195         struct mdt_reint_record *rr = &info->mti_rr;
1196         struct req_capsule      *pill = info->mti_pill;
1197         int rc;
1198         ENTRY;
1199
1200         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1201         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1202         if (rec == NULL)
1203                 RETURN(-EFAULT);
1204
1205         /* This prior initialization is needed for old_init_ucred_reint() */
1206         uc->uc_fsuid = rec->ul_fsuid;
1207         uc->uc_fsgid = rec->ul_fsgid;
1208         uc->uc_cap   = rec->ul_cap;
1209         uc->uc_suppgids[0] = rec->ul_suppgid1;
1210         uc->uc_suppgids[1] = -1;
1211
1212         attr->la_uid = rec->ul_fsuid;
1213         attr->la_gid = rec->ul_fsgid;
1214         rr->rr_fid1 = &rec->ul_fid1;
1215         rr->rr_fid2 = &rec->ul_fid2;
1216         attr->la_ctime = rec->ul_time;
1217         attr->la_mtime = rec->ul_time;
1218         attr->la_mode  = rec->ul_mode;
1219         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1220
1221         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1222         if (rc < 0)
1223                 RETURN(rc);
1224
1225         if (rec->ul_bias & MDS_VTX_BYPASS)
1226                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1227         else
1228                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1229
1230         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1231
1232         rc = mdt_dlmreq_unpack(info);
1233         RETURN(rc);
1234 }
1235
1236 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1237 {
1238         info->mti_spec.sp_rm_entry = 1;
1239         return mdt_unlink_unpack(info);
1240 }
1241
1242 static int mdt_rename_unpack(struct mdt_thread_info *info)
1243 {
1244         struct lu_ucred         *uc = mdt_ucred(info);
1245         struct mdt_rec_rename   *rec;
1246         struct md_attr          *ma = &info->mti_attr;
1247         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1248         struct mdt_reint_record *rr = &info->mti_rr;
1249         struct req_capsule      *pill = info->mti_pill;
1250         int rc;
1251         ENTRY;
1252
1253         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1254         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1255         if (rec == NULL)
1256                 RETURN(-EFAULT);
1257
1258         /* This prior initialization is needed for old_init_ucred_reint() */
1259         uc->uc_fsuid = rec->rn_fsuid;
1260         uc->uc_fsgid = rec->rn_fsgid;
1261         uc->uc_cap   = rec->rn_cap;
1262         uc->uc_suppgids[0] = rec->rn_suppgid1;
1263         uc->uc_suppgids[1] = rec->rn_suppgid2;
1264
1265         attr->la_uid = rec->rn_fsuid;
1266         attr->la_gid = rec->rn_fsgid;
1267         rr->rr_fid1 = &rec->rn_fid1;
1268         rr->rr_fid2 = &rec->rn_fid2;
1269         attr->la_ctime = rec->rn_time;
1270         attr->la_mtime = rec->rn_time;
1271         /* rename_tgt contains the mode already */
1272         attr->la_mode = rec->rn_mode;
1273         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1274
1275         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1276         if (rc < 0)
1277                 RETURN(rc);
1278
1279         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1280         if (rc < 0)
1281                 RETURN(rc);
1282
1283         if (rec->rn_bias & MDS_VTX_BYPASS)
1284                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1285         else
1286                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1287
1288         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1289                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1290                 rc = mdt_close_handle_unpack(info);
1291                 if (rc < 0)
1292                         RETURN(rc);
1293                 info->mti_spec.sp_migrate_close = 1;
1294         }
1295
1296         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1297
1298
1299         rc = mdt_dlmreq_unpack(info);
1300
1301         RETURN(rc);
1302 }
1303
1304 /*
1305  * please see comment above LOV_MAGIC_V1_DEF
1306  */
1307 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1308 {
1309         struct mdt_reint_record *rr = &info->mti_rr;
1310         struct lov_user_md_v1   *v1;
1311
1312         v1 = (void *)rr->rr_eadata;
1313         LASSERT(v1);
1314
1315         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1316                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1317                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1318                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1319                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1320                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1321                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1322                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1323                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1324                 }
1325         }
1326 }
1327
1328 static int mdt_open_unpack(struct mdt_thread_info *info)
1329 {
1330         struct lu_ucred         *uc = mdt_ucred(info);
1331         struct mdt_rec_create   *rec;
1332         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1333         struct req_capsule      *pill = info->mti_pill;
1334         struct mdt_reint_record *rr   = &info->mti_rr;
1335         struct ptlrpc_request   *req  = mdt_info_req(info);
1336         struct md_op_spec       *sp   = &info->mti_spec;
1337         ENTRY;
1338
1339         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1340         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1341         if (rec == NULL)
1342                 RETURN(-EFAULT);
1343
1344         /* This prior initialization is needed for old_init_ucred_reint() */
1345         uc->uc_fsuid = rec->cr_fsuid;
1346         uc->uc_fsgid = rec->cr_fsgid;
1347         uc->uc_cap   = rec->cr_cap;
1348         uc->uc_suppgids[0] = rec->cr_suppgid1;
1349         uc->uc_suppgids[1] = rec->cr_suppgid2;
1350         uc->uc_umask = rec->cr_umask;
1351
1352         rr->rr_fid1   = &rec->cr_fid1;
1353         rr->rr_fid2   = &rec->cr_fid2;
1354         rr->rr_handle = &rec->cr_old_handle;
1355         attr->la_mode = rec->cr_mode;
1356         attr->la_rdev  = rec->cr_rdev;
1357         attr->la_uid   = rec->cr_fsuid;
1358         attr->la_gid   = rec->cr_fsgid;
1359         attr->la_ctime = rec->cr_time;
1360         attr->la_mtime = rec->cr_time;
1361         attr->la_atime = rec->cr_time;
1362         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1363                          LA_CTIME | LA_MTIME | LA_ATIME;
1364         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1365         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1366         /* Do not trigger ASSERTION if client miss to set such flags. */
1367         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1368                 RETURN(-EPROTO);
1369
1370         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1371
1372         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1373
1374         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1375                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1376                                                         RCL_CLIENT);
1377                 if (rr->rr_eadatalen > 0) {
1378                         rr->rr_eadata = req_capsule_client_get(pill,
1379                                                                &RMF_EADATA);
1380                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1381                         sp->u.sp_ea.eadata = rr->rr_eadata;
1382                         sp->no_create = !!req_is_replay(req);
1383                         mdt_fix_lov_magic(info);
1384                 }
1385
1386                 /*
1387                  * Client default md_size may be 0 right after client start,
1388                  * until all osc are connected, set here just some reasonable
1389                  * value to prevent misbehavior.
1390                  */
1391                 if (rr->rr_eadatalen == 0 &&
1392                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1393                         rr->rr_eadatalen = MIN_MD_SIZE;
1394         }
1395
1396         RETURN(0);
1397 }
1398
1399 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1400 {
1401         struct mdt_reint_record *rr     = &info->mti_rr;
1402         struct lu_ucred         *uc     = mdt_ucred(info);
1403         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1404         struct req_capsule      *pill   = info->mti_pill;
1405         struct mdt_rec_setxattr *rec;
1406         int                      rc;
1407         ENTRY;
1408
1409
1410         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1411                          sizeof(struct mdt_rec_reint));
1412
1413         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1414         if (rec == NULL)
1415                 RETURN(-EFAULT);
1416
1417         /* This prior initialization is needed for old_init_ucred_reint() */
1418         uc->uc_fsuid  = rec->sx_fsuid;
1419         uc->uc_fsgid  = rec->sx_fsgid;
1420         uc->uc_cap    = rec->sx_cap;
1421         uc->uc_suppgids[0] = rec->sx_suppgid1;
1422         uc->uc_suppgids[1] = -1;
1423
1424         rr->rr_opcode = rec->sx_opcode;
1425         rr->rr_fid1   = &rec->sx_fid;
1426         attr->la_valid = rec->sx_valid;
1427         attr->la_ctime = rec->sx_time;
1428         attr->la_size = rec->sx_size;
1429         attr->la_flags = rec->sx_flags;
1430
1431         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1432         if (rc < 0)
1433                 RETURN(rc);
1434
1435         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1436                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1437                                                         RCL_CLIENT);
1438                 if (rr->rr_eadatalen > 0) {
1439                         rr->rr_eadata = req_capsule_client_get(pill,
1440                                                                &RMF_EADATA);
1441                         if (rr->rr_eadata == NULL)
1442                                 RETURN(-EFAULT);
1443                 } else {
1444                         rr->rr_eadata = NULL;
1445                 }
1446         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1447                 CDEBUG(D_INFO, "no xattr data supplied\n");
1448                 RETURN(-EFAULT);
1449         }
1450
1451         if (mdt_dlmreq_unpack(info) < 0)
1452                 RETURN(-EPROTO);
1453
1454         RETURN(0);
1455 }
1456
1457
1458 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1459
1460 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1461         [REINT_SETATTR]  = mdt_setattr_unpack,
1462         [REINT_CREATE]   = mdt_create_unpack,
1463         [REINT_LINK]     = mdt_link_unpack,
1464         [REINT_UNLINK]   = mdt_unlink_unpack,
1465         [REINT_RENAME]   = mdt_rename_unpack,
1466         [REINT_OPEN]     = mdt_open_unpack,
1467         [REINT_SETXATTR] = mdt_setxattr_unpack,
1468         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1469         [REINT_MIGRATE]  = mdt_rename_unpack,
1470 };
1471
1472 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1473 {
1474         int rc;
1475         ENTRY;
1476
1477         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1478         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1479                 info->mti_rr.rr_opcode = op;
1480                 rc = mdt_reint_unpackers[op](info);
1481         } else {
1482                 CERROR("Unexpected opcode %d\n", op);
1483                 rc = -EFAULT;
1484         }
1485         RETURN(rc);
1486 }