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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_lib.c
  33  *
  34  * Lustre Metadata Target (mdt) request unpacking helper.
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Fan Yong <fanyong@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include "mdt_internal.h"
  52 #include <lnet/nidstr.h>
  53 #include <lustre_nodemap.h>
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
  62 {
  63         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
  64 }
  65
  66 void mdt_exit_ucred(struct mdt_thread_info *info)
  67 {
  68         struct lu_ucred   *uc  = mdt_ucred(info);
  69         struct mdt_device *mdt = info->mti_mdt;
  70
  71         LASSERT(uc != NULL);
  72         if (uc->uc_valid != UCRED_INIT) {
  73                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  74                 if (uc->uc_ginfo) {
  75                         put_group_info(uc->uc_ginfo);
  76                         uc->uc_ginfo = NULL;
  77                 }
  78                 if (uc->uc_identity) {
  79                         mdt_identity_put(mdt->mdt_identity_cache,
  80                                          uc->uc_identity);
  81                         uc->uc_identity = NULL;
  82                 }
  83                 uc->uc_valid = UCRED_INIT;
  84         }
  85 }
  86
  87 static int match_nosquash_list(struct rw_semaphore *sem,
  88                                struct list_head *nidlist,
  89                                lnet_nid_t peernid)
  90 {
  91         int rc;
  92         ENTRY;
  93         down_read(sem);
  94         rc = cfs_match_nid(peernid, nidlist);
  95         up_read(sem);
  96         RETURN(rc);
  97 }
  98
  99 /* root_squash for inter-MDS operations */
 100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 101 {
 102         struct lu_ucred *ucred = mdt_ucred(info);
 103         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 104         ENTRY;
 105
 106         LASSERT(ucred != NULL);
 107         if (!squash->rsi_uid || ucred->uc_fsuid)
 108                 RETURN(0);
 109
 110         if (match_nosquash_list(&squash->rsi_sem,
 111                                 &squash->rsi_nosquash_nids,
 112                                 peernid)) {
 113                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 114                        libcfs_nid2str(peernid));
 115                 RETURN(0);
 116         }
 117
 118         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 119                libcfs_nid2str(peernid),
 120                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 121                squash->rsi_uid, squash->rsi_gid, 0);
 122
 123         ucred->uc_fsuid = squash->rsi_uid;
 124         ucred->uc_fsgid = squash->rsi_gid;
 125         ucred->uc_cap = 0;
 126         ucred->uc_suppgids[0] = -1;
 127         ucred->uc_suppgids[1] = -1;
 128
 129         RETURN(0);
 130 }
 131
 132 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 133 {
 134         struct ptlrpc_request   *req = mdt_info_req(info);
 135         const char              *jobid = mdt_req_get_jobid(req);
 136
 137         /* set jobid if specified. */
 138         if (jobid)
 139                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 140         else
 141                 uc->uc_jobid[0] = '\0';
 142 }
 143
 144 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 145                           void *buf, bool drop_fs_cap)
 146 {
 147         struct ptlrpc_request   *req = mdt_info_req(info);
 148         struct mdt_device       *mdt = info->mti_mdt;
 149         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 150         struct lu_ucred         *ucred = mdt_ucred(info);
 151         lnet_nid_t               peernid = req->rq_peer.nid;
 152         __u32                    perm = 0;
 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         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 179             req->rq_auth_uid != pud->pud_uid) {
 180                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 181                        "while client claims %u:%u/%u:%u\n",
 182                        libcfs_nid2str(peernid), req->rq_auth_uid,
 183                        pud->pud_uid, pud->pud_gid,
 184                        pud->pud_fsuid, pud->pud_fsgid);
 185                 RETURN(-EACCES);
 186         }
 187
 188         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 189                 ucred->uc_identity = NULL;
 190                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 191         } else {
 192                 struct md_identity *identity;
 193
 194                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 195                                             pud->pud_uid);
 196                 if (IS_ERR(identity)) {
 197                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 198                                 ucred->uc_identity = NULL;
 199                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 200                                        CFS_SETGRP_PERM;
 201                         } else {
 202                                 CDEBUG(D_SEC,
 203                                        "Deny access without identity: uid %u\n",
 204                                        pud->pud_uid);
 205                                 RETURN(-EACCES);
 206                         }
 207                 } else {
 208                         ucred->uc_identity = identity;
 209                         perm = mdt_identity_get_perm(ucred->uc_identity,
 210                                                      peernid);
 211                 }
 212         }
 213
 214         /* find out the setuid/setgid attempt */
 215         setuid = (pud->pud_uid != pud->pud_fsuid);
 216         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 217                   (ucred->uc_identity &&
 218                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 219
 220         /* check permission of setuid */
 221         if (setuid && !(perm & CFS_SETUID_PERM)) {
 222                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 223                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 224                 GOTO(out, rc = -EACCES);
 225         }
 226
 227         /* check permission of setgid */
 228         if (setgid && !(perm & CFS_SETGID_PERM)) {
 229                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 230                        "from %s\n", pud->pud_uid, pud->pud_gid,
 231                        pud->pud_fsuid, pud->pud_fsgid,
 232                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 233                 GOTO(out, rc = -EACCES);
 234         }
 235
 236         if (perm & CFS_SETGRP_PERM) {
 237                 if (pud->pud_ngroups) {
 238                         /* setgroups for local client */
 239                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 240                         if (!ucred->uc_ginfo) {
 241                                 CERROR("failed to alloc %d groups\n",
 242                                        pud->pud_ngroups);
 243                                 GOTO(out, rc = -ENOMEM);
 244                         }
 245
 246                         lustre_groups_from_list(ucred->uc_ginfo,
 247                                                 pud->pud_groups);
 248                         lustre_groups_sort(ucred->uc_ginfo);
 249                 } else {
 250                         ucred->uc_ginfo = NULL;
 251                 }
 252         } else {
 253                 ucred->uc_suppgids[0] = -1;
 254                 ucred->uc_suppgids[1] = -1;
 255                 ucred->uc_ginfo = NULL;
 256         }
 257
 258         ucred->uc_uid   = pud->pud_uid;
 259         ucred->uc_gid   = pud->pud_gid;
 260         ucred->uc_fsuid = pud->pud_fsuid;
 261         ucred->uc_fsgid = pud->pud_fsgid;
 262
 263         /* process root_squash here. */
 264         mdt_root_squash(info, peernid);
 265
 266         /* remove fs privilege for non-root user. */
 267         if (ucred->uc_fsuid && drop_fs_cap)
 268                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 269         else
 270                 ucred->uc_cap = pud->pud_cap;
 271         ucred->uc_valid = UCRED_NEW;
 272         ucred_set_jobid(info, ucred);
 273
 274         EXIT;
 275
 276 out:
 277         if (rc) {
 278                 if (ucred->uc_ginfo) {
 279                         put_group_info(ucred->uc_ginfo);
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282                 if (ucred->uc_identity) {
 283                         mdt_identity_put(mdt->mdt_identity_cache,
 284                                          ucred->uc_identity);
 285                         ucred->uc_identity = NULL;
 286                 }
 287         }
 288
 289         return rc;
 290 }
 291
 292 /**
 293  * Check whether allow the client to set supplementary group IDs or not.
 294  *
 295  * \param[in] info      pointer to the thread context
 296  * \param[in] uc        pointer to the RPC user descriptor
 297  *
 298  * \retval              true if allow to set supplementary group IDs
 299  * \retval              false for other cases
 300  */
 301 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 302 {
 303         __u32 perm;
 304
 305         /* 1. If identity_upcall is disabled,
 306          *    permit local client to do anything. */
 307         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 308                 return true;
 309
 310         /* 2. If fail to get related identities, then forbid any client to
 311          *    set supplementary group IDs. */
 312         if (uc->uc_identity == NULL)
 313                 return false;
 314
 315         /* 3. Check the permission in the identities. */
 316         perm = mdt_identity_get_perm(uc->uc_identity,
 317                                      mdt_info_req(info)->rq_peer.nid);
 318         if (perm & CFS_SETGRP_PERM)
 319                 return true;
 320
 321         return false;
 322 }
 323
 324 int mdt_check_ucred(struct mdt_thread_info *info)
 325 {
 326         struct ptlrpc_request   *req = mdt_info_req(info);
 327         struct mdt_device       *mdt = info->mti_mdt;
 328         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 329         struct lu_ucred         *ucred = mdt_ucred(info);
 330         struct md_identity      *identity = NULL;
 331         lnet_nid_t               peernid = req->rq_peer.nid;
 332         __u32                    perm = 0;
 333         int                      setuid;
 334         int                      setgid;
 335         int                      rc = 0;
 336
 337         ENTRY;
 338
 339         LASSERT(ucred != NULL);
 340         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 341                 RETURN(0);
 342
 343         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 344                 RETURN(0);
 345
 346         /* sanity check: if we use strong authentication, we expect the
 347          * uid which client claimed is true */
 348         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 349             req->rq_auth_uid != pud->pud_uid) {
 350                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 351                        "while client claims %u:%u/%u:%u\n",
 352                        libcfs_nid2str(peernid), req->rq_auth_uid,
 353                        pud->pud_uid, pud->pud_gid,
 354                        pud->pud_fsuid, pud->pud_fsgid);
 355                 RETURN(-EACCES);
 356         }
 357
 358         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 359                 RETURN(0);
 360
 361         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 362         if (IS_ERR(identity)) {
 363                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 364                         RETURN(0);
 365                 } else {
 366                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 367                                pud->pud_uid);
 368                         RETURN(-EACCES);
 369                 }
 370         }
 371
 372         perm = mdt_identity_get_perm(identity, peernid);
 373         /* find out the setuid/setgid attempt */
 374         setuid = (pud->pud_uid != pud->pud_fsuid);
 375         setgid = (pud->pud_gid != pud->pud_fsgid ||
 376                   pud->pud_gid != identity->mi_gid);
 377
 378         /* check permission of setuid */
 379         if (setuid && !(perm & CFS_SETUID_PERM)) {
 380                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 381                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 382                 GOTO(out, rc = -EACCES);
 383         }
 384
 385         /* check permission of setgid */
 386         if (setgid && !(perm & CFS_SETGID_PERM)) {
 387                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 388                        "from %s\n", pud->pud_uid, pud->pud_gid,
 389                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 390                        libcfs_nid2str(peernid));
 391                 GOTO(out, rc = -EACCES);
 392         }
 393
 394         EXIT;
 395
 396 out:
 397         mdt_identity_put(mdt->mdt_identity_cache, identity);
 398         return rc;
 399 }
 400
 401 static int old_init_ucred_common(struct mdt_thread_info *info,
 402                                  struct lu_nodemap *nodemap,
 403                                  bool drop_fs_cap)
 404 {
 405         struct lu_ucred         *uc = mdt_ucred(info);
 406         struct mdt_device       *mdt = info->mti_mdt;
 407         struct md_identity      *identity = NULL;
 408
 409         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 410                 /* deny access before we get identity ref */
 411                 if (nodemap->nmf_deny_unknown)
 412                         RETURN(-EACCES);
 413
 414                 uc->uc_fsuid = nodemap->nm_squash_uid;
 415                 uc->uc_fsgid = nodemap->nm_squash_gid;
 416                 uc->uc_cap = 0;
 417                 uc->uc_suppgids[0] = -1;
 418                 uc->uc_suppgids[1] = -1;
 419         }
 420
 421         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 422                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 423                                             uc->uc_fsuid);
 424                 if (IS_ERR(identity)) {
 425                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 426                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 427                                 identity = NULL;
 428                         } else {
 429                                 CDEBUG(D_SEC, "Deny access without identity: "
 430                                        "uid %u\n", uc->uc_fsuid);
 431                                 RETURN(-EACCES);
 432                         }
 433                 }
 434         }
 435         uc->uc_identity = identity;
 436
 437         /* process root_squash here. */
 438         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 439
 440         /* remove fs privilege for non-root user. */
 441         if (uc->uc_fsuid && drop_fs_cap)
 442                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 443         uc->uc_valid = UCRED_OLD;
 444         ucred_set_jobid(info, uc);
 445
 446         EXIT;
 447
 448         return 0;
 449 }
 450
 451 static int old_init_ucred(struct mdt_thread_info *info,
 452                           struct mdt_body *body, bool drop_fs_cap)
 453 {
 454         struct lu_ucred *uc = mdt_ucred(info);
 455         struct lu_nodemap *nodemap;
 456         int rc;
 457         ENTRY;
 458
 459         nodemap = nodemap_get_from_exp(info->mti_exp);
 460         if (IS_ERR(nodemap))
 461                 RETURN(PTR_ERR(nodemap));
 462
 463         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 464                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 465         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 466                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 467         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 468                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 469         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 470                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 471
 472         LASSERT(uc != NULL);
 473         uc->uc_valid = UCRED_INVALID;
 474         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 475         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 476         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 477         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 478         uc->uc_suppgids[0] = body->mbo_suppgid;
 479         uc->uc_suppgids[1] = -1;
 480         uc->uc_ginfo = NULL;
 481         uc->uc_cap = body->mbo_capability;
 482
 483         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 484         nodemap_putref(nodemap);
 485
 486         RETURN(rc);
 487 }
 488
 489 static int old_init_ucred_reint(struct mdt_thread_info *info)
 490 {
 491         struct lu_ucred *uc = mdt_ucred(info);
 492         struct lu_nodemap *nodemap;
 493         int rc;
 494         ENTRY;
 495
 496         nodemap = nodemap_get_from_exp(info->mti_exp);
 497         if (IS_ERR(nodemap))
 498                 RETURN(PTR_ERR(nodemap));
 499
 500         LASSERT(uc != NULL);
 501
 502         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 503                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 504         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 505                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 506
 507         uc->uc_valid = UCRED_INVALID;
 508         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 509         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 510         uc->uc_ginfo = NULL;
 511
 512         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 513         nodemap_putref(nodemap);
 514
 515         RETURN(rc);
 516 }
 517
 518 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 519                                    struct mdt_body *body,
 520                                    bool drop_fs_cap)
 521 {
 522         struct ptlrpc_request   *req = mdt_info_req(info);
 523         struct lu_ucred         *uc  = mdt_ucred(info);
 524
 525         LASSERT(uc != NULL);
 526         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 527                 return 0;
 528
 529         mdt_exit_ucred(info);
 530
 531         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 532                 return old_init_ucred(info, body, drop_fs_cap);
 533         else
 534                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 535 }
 536
 537 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 538 {
 539         return __mdt_init_ucred(info, body, true);
 540 }
 541
 542 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 543  * doesn't set correct fsuid explicitely, but raise capability to allow
 544  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 545  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 546  * will fail permission check. */
 547 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 548                                   struct mdt_body *body)
 549 {
 550         return __mdt_init_ucred(info, body, false);
 551 }
 552
 553 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 554 {
 555         struct ptlrpc_request *req = mdt_info_req(info);
 556         struct lu_ucred       *uc  = mdt_ucred(info);
 557         struct md_attr        *ma  = &info->mti_attr;
 558
 559         LASSERT(uc != NULL);
 560         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 561                 return 0;
 562
 563         /* LU-5564: for normal close request, skip permission check */
 564         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 565             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 566                 uc->uc_cap |= CFS_CAP_FS_MASK;
 567
 568         mdt_exit_ucred(info);
 569
 570         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 571                 return old_init_ucred_reint(info);
 572         else
 573                 return new_init_ucred(info, REC_INIT, NULL, true);
 574 }
 575
 576 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 577 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 578 {
 579         const struct lov_ost_data_v1    *lod;
 580         int                              i;
 581         __u16                            count;
 582
 583         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 584                 return;
 585
 586         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 587
 588         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 589                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 590                le32_to_cpu(lmm->lmm_pattern));
 591         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 592                le32_to_cpu(lmm->lmm_stripe_size), count);
 593
 594         /* If it's a directory or a released file, then there are
 595          * no actual objects to print, so bail out. */
 596         if (valid & OBD_MD_FLDIREA ||
 597             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 598                 return;
 599
 600         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 601         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 602                 struct ost_id oi;
 603
 604                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 605                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 606                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 607         }
 608 }
 609
 610 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 611 {
 612         const struct lmv_mds_md_v1 *lmm1;
 613         int                        i;
 614
 615         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 616                 return;
 617
 618         lmm1 = &lmv->lmv_md_v1;
 619         CDEBUG(level,
 620                "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
 621                le32_to_cpu(lmm1->lmv_magic),
 622                le32_to_cpu(lmm1->lmv_master_mdt_index),
 623                le32_to_cpu(lmm1->lmv_stripe_count),
 624                le32_to_cpu(lmm1->lmv_hash_type));
 625
 626         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 627                 return;
 628
 629         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 630                 struct lu_fid fid;
 631
 632                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 633                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 634         }
 635 }
 636
 637 /* Shrink and/or grow reply buffers */
 638 int mdt_fix_reply(struct mdt_thread_info *info)
 639 {
 640         struct req_capsule *pill = info->mti_pill;
 641         struct mdt_body    *body;
 642         int                md_size, md_packed = 0;
 643         int                acl_size;
 644         int                rc = 0;
 645         ENTRY;
 646
 647         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 648         LASSERT(body != NULL);
 649
 650         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 651                                OBD_MD_LINKNAME))
 652                 md_size = body->mbo_eadatasize;
 653         else
 654                 md_size = 0;
 655
 656         acl_size = body->mbo_aclsize;
 657
 658         /* this replay - not send info to client */
 659         if (info->mti_spec.no_create) {
 660                 md_size = 0;
 661                 acl_size = 0;
 662         }
 663
 664         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 665                md_size, acl_size);
 666 /*
 667             &RMF_MDT_BODY,
 668             &RMF_MDT_MD,
 669             &RMF_ACL, or &RMF_LOGCOOKIES
 670 (optional)  &RMF_CAPA1,
 671 (optional)  &RMF_CAPA2,
 672 (optional)  something else
 673 */
 674
 675         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 676          * buffers before growing it */
 677         if (info->mti_big_lmm_used) {
 678                 /* big_lmm buffer may be used even without packing the result
 679                  * into reply, just for internal server needs */
 680                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 681                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 682                                                          RCL_SERVER);
 683
 684                 /* free big lmm if md_size is not needed */
 685                 if (md_size == 0 || md_packed == 0) {
 686                         info->mti_big_lmm_used = 0;
 687                 } else {
 688                         /* buffer must be allocated separately */
 689                         LASSERT(info->mti_attr.ma_lmm !=
 690                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 691                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 692                 }
 693         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 694                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 695         }
 696
 697         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 698                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 699         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 700                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 701                                    acl_size, RCL_SERVER);
 702
 703         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 704             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 705                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 706
 707         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 708             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 709                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 710
 711         /*
 712          * Some more field should be shrinked if needed.
 713          * This should be done by those who added fields to reply message.
 714          */
 715
 716         /* Grow MD buffer if needed finally */
 717         if (info->mti_big_lmm_used) {
 718                 void *lmm;
 719
 720                 LASSERT(md_size > md_packed);
 721                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 722                        md_size - md_packed);
 723                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 724                 if (rc) {
 725                         /* we can't answer with proper LOV EA, drop flags,
 726                          * the rc is also returned so this request is
 727                          * considered as failed */
 728                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 729                         /* don't return transno along with error */
 730                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 731                 } else {
 732                         /* now we need to pack right LOV/LMV EA */
 733                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 734                         if (info->mti_attr.ma_valid & MA_LOV) {
 735                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 736                                                              RCL_SERVER) ==
 737                                                 info->mti_attr.ma_lmm_size);
 738                                 memcpy(lmm, info->mti_attr.ma_lmm,
 739                                        info->mti_attr.ma_lmm_size);
 740                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 741                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 742                                                              RCL_SERVER) ==
 743                                                 info->mti_attr.ma_lmv_size);
 744                                 memcpy(lmm, info->mti_attr.ma_lmv,
 745                                        info->mti_attr.ma_lmv_size);
 746                         }
 747                 }
 748                 /* update mdt_max_mdsize so clients will be aware about that */
 749                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 750                         info->mti_mdt->mdt_max_mdsize =
 751                                                     info->mti_attr.ma_lmm_size;
 752                 info->mti_big_lmm_used = 0;
 753         }
 754         RETURN(rc);
 755 }
 756
 757
 758 /* if object is dying, pack the lov/llog data,
 759  * parameter info->mti_attr should be valid at this point! */
 760 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 761                            const struct md_attr *ma)
 762 {
 763         struct mdt_body       *repbody;
 764         const struct lu_attr *la = &ma->ma_attr;
 765         ENTRY;
 766
 767         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 768         LASSERT(repbody != NULL);
 769
 770         if (ma->ma_valid & MA_INODE)
 771                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 772
 773         if (ma->ma_valid & MA_LOV) {
 774                 CERROR("No need in LOV EA upon unlink\n");
 775                 dump_stack();
 776         }
 777         repbody->mbo_eadatasize = 0;
 778
 779         RETURN(0);
 780 }
 781
 782 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 783                                   struct md_attr *ma)
 784 {
 785         __u64 out;
 786
 787         out = 0;
 788         if (in & MDS_ATTR_MODE)
 789                 out |= LA_MODE;
 790         if (in & MDS_ATTR_UID)
 791                 out |= LA_UID;
 792         if (in & MDS_ATTR_GID)
 793                 out |= LA_GID;
 794         if (in & MDS_ATTR_SIZE)
 795                 out |= LA_SIZE;
 796         if (in & MDS_ATTR_BLOCKS)
 797                 out |= LA_BLOCKS;
 798         if (in & MDS_ATTR_ATIME_SET)
 799                 out |= LA_ATIME;
 800         if (in & MDS_ATTR_CTIME_SET)
 801                 out |= LA_CTIME;
 802         if (in & MDS_ATTR_MTIME_SET)
 803                 out |= LA_MTIME;
 804         if (in & MDS_ATTR_ATTR_FLAG)
 805                 out |= LA_FLAGS;
 806         if (in & MDS_ATTR_KILL_SUID)
 807                 out |= LA_KILL_SUID;
 808         if (in & MDS_ATTR_KILL_SGID)
 809                 out |= LA_KILL_SGID;
 810         if (in & MDS_ATTR_PROJID)
 811                 out |= LA_PROJID;
 812
 813         if (in & MDS_ATTR_FROM_OPEN)
 814                 rr->rr_flags |= MRF_OPEN_TRUNC;
 815         if (in & MDS_OPEN_OWNEROVERRIDE)
 816                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 817         if (in & MDS_ATTR_FORCE)
 818                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 819
 820         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
 821                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 822                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 823                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 824                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 825                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 826         if (in != 0)
 827                 CERROR("Unknown attr bits: %#llx\n", in);
 828         return out;
 829 }
 830
 831 /* unpacking */
 832
 833 int mdt_name_unpack(struct req_capsule *pill,
 834                     const struct req_msg_field *field,
 835                     struct lu_name *ln,
 836                     enum mdt_name_flags flags)
 837 {
 838         ln->ln_name = req_capsule_client_get(pill, field);
 839         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 840
 841         if (!lu_name_is_valid(ln)) {
 842                 ln->ln_name = NULL;
 843                 ln->ln_namelen = 0;
 844
 845                 return -EPROTO;
 846         }
 847
 848         if ((flags & MNF_FIX_ANON) &&
 849             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 850                 /* Newer (3.x) kernels use a name of "/" for the
 851                  * "anonymous" disconnected dentries from NFS
 852                  * filehandle conversion. See d_obtain_alias(). */
 853                 ln->ln_name = NULL;
 854                 ln->ln_namelen = 0;
 855         }
 856
 857         return 0;
 858 }
 859
 860 static int mdt_file_secctx_unpack(struct req_capsule *pill,
 861                                   const char **secctx_name,
 862                                   void **secctx, size_t *secctx_size)
 863 {
 864         const char *name;
 865         size_t name_size;
 866
 867         *secctx_name = NULL;
 868         *secctx = NULL;
 869         *secctx_size = 0;
 870
 871         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
 872             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
 873                 return 0;
 874
 875         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
 876                                          RCL_CLIENT);
 877         if (name_size == 0)
 878                 return 0;
 879
 880         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
 881         if (strnlen(name, name_size) != name_size - 1)
 882                 return -EPROTO;
 883
 884         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
 885             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
 886                 return -EPROTO;
 887
 888         *secctx_name = name;
 889         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
 890         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
 891
 892         return 0;
 893 }
 894
 895 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 896 {
 897         struct lu_ucred         *uc = mdt_ucred(info);
 898         struct md_attr          *ma = &info->mti_attr;
 899         struct lu_attr          *la = &ma->ma_attr;
 900         struct req_capsule      *pill = info->mti_pill;
 901         struct mdt_reint_record *rr = &info->mti_rr;
 902         struct mdt_rec_setattr  *rec;
 903         struct lu_nodemap       *nodemap;
 904         ENTRY;
 905
 906         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 907         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 908         if (rec == NULL)
 909                 RETURN(-EFAULT);
 910
 911         /* This prior initialization is needed for old_init_ucred_reint() */
 912         uc->uc_fsuid = rec->sa_fsuid;
 913         uc->uc_fsgid = rec->sa_fsgid;
 914         uc->uc_cap   = rec->sa_cap;
 915         uc->uc_suppgids[0] = rec->sa_suppgid;
 916         uc->uc_suppgids[1] = -1;
 917
 918         rr->rr_fid1 = &rec->sa_fid;
 919         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 920         la->la_mode  = rec->sa_mode;
 921         la->la_flags = rec->sa_attr_flags;
 922
 923         nodemap = nodemap_get_from_exp(info->mti_exp);
 924         if (IS_ERR(nodemap))
 925                 RETURN(PTR_ERR(nodemap));
 926
 927         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 928                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 929         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 930                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 931         nodemap_putref(nodemap);
 932
 933         la->la_size  = rec->sa_size;
 934         la->la_blocks = rec->sa_blocks;
 935         la->la_ctime = rec->sa_ctime;
 936         la->la_atime = rec->sa_atime;
 937         la->la_mtime = rec->sa_mtime;
 938         ma->ma_valid = MA_INODE;
 939
 940         if (rec->sa_bias & MDS_DATA_MODIFIED)
 941                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 942         else
 943                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 944
 945         if (rec->sa_bias & MDS_HSM_RELEASE)
 946                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 947         else
 948                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 949
 950         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 951                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 952         else
 953                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 954
 955         RETURN(0);
 956 }
 957
 958 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 959 {
 960         struct req_capsule *pill = info->mti_pill;
 961         struct mdt_ioepoch *ioepoch;
 962         ENTRY;
 963
 964         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 965                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 966         else
 967                 ioepoch = NULL;
 968
 969         if (ioepoch == NULL)
 970                 RETURN(-EPROTO);
 971
 972         info->mti_close_handle = ioepoch->mio_handle;
 973
 974         RETURN(0);
 975 }
 976
 977 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 978         struct req_capsule      *pill = info->mti_pill;
 979
 980         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 981                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 982                 if (info->mti_dlm_req == NULL)
 983                         RETURN(-EFAULT);
 984         }
 985
 986         RETURN(0);
 987 }
 988
 989 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 990 {
 991         struct mdt_reint_record *rr = &info->mti_rr;
 992         struct md_attr          *ma = &info->mti_attr;
 993         struct req_capsule      *pill = info->mti_pill;
 994         int rc;
 995         ENTRY;
 996
 997         rc = mdt_setattr_unpack_rec(info);
 998         if (rc)
 999                 RETURN(rc);
1000
1001         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1002                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1003                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1004                                                         RCL_CLIENT);
1005                 if (rr->rr_eadatalen > 0) {
1006                         const struct lmv_user_md        *lum;
1007
1008                         lum = rr->rr_eadata;
1009                         /* Sigh ma_valid(from req) does not indicate whether
1010                          * it will set LOV/LMV EA, so we have to check magic */
1011                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1012                                 ma->ma_valid |= MA_LMV;
1013                                 ma->ma_lmv = (void *)rr->rr_eadata;
1014                                 ma->ma_lmv_size = rr->rr_eadatalen;
1015                         } else {
1016                                 ma->ma_valid |= MA_LOV;
1017                                 ma->ma_lmm = (void *)rr->rr_eadata;
1018                                 ma->ma_lmm_size = rr->rr_eadatalen;
1019                         }
1020                 }
1021         }
1022
1023         rc = mdt_dlmreq_unpack(info);
1024         RETURN(rc);
1025 }
1026
1027 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1028 {
1029         struct md_attr          *ma = &info->mti_attr;
1030         struct req_capsule      *pill = info->mti_pill;
1031         ENTRY;
1032
1033         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1034                 RETURN(0);
1035
1036         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1037
1038         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1039             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1040                 RETURN(-EFAULT);
1041
1042         RETURN(0);
1043 }
1044
1045 int mdt_close_unpack(struct mdt_thread_info *info)
1046 {
1047         int rc;
1048         ENTRY;
1049
1050         rc = mdt_close_handle_unpack(info);
1051         if (rc)
1052                 RETURN(rc);
1053
1054         rc = mdt_setattr_unpack_rec(info);
1055         if (rc)
1056                 RETURN(rc);
1057
1058         rc = mdt_intent_close_unpack(info);
1059         if (rc)
1060                 RETURN(rc);
1061
1062         RETURN(mdt_init_ucred_reint(info));
1063 }
1064
1065 static int mdt_create_unpack(struct mdt_thread_info *info)
1066 {
1067         struct lu_ucred         *uc  = mdt_ucred(info);
1068         struct mdt_rec_create   *rec;
1069         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1070         struct mdt_reint_record *rr = &info->mti_rr;
1071         struct req_capsule      *pill = info->mti_pill;
1072         struct md_op_spec       *sp = &info->mti_spec;
1073         int rc;
1074         ENTRY;
1075
1076         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1077         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1078         if (rec == NULL)
1079                 RETURN(-EFAULT);
1080
1081         /* This prior initialization is needed for old_init_ucred_reint() */
1082         uc->uc_fsuid = rec->cr_fsuid;
1083         uc->uc_fsgid = rec->cr_fsgid;
1084         uc->uc_cap   = rec->cr_cap;
1085         uc->uc_suppgids[0] = rec->cr_suppgid1;
1086         uc->uc_suppgids[1] = -1;
1087         uc->uc_umask = rec->cr_umask;
1088
1089         rr->rr_fid1 = &rec->cr_fid1;
1090         rr->rr_fid2 = &rec->cr_fid2;
1091         attr->la_mode = rec->cr_mode;
1092         attr->la_rdev  = rec->cr_rdev;
1093         attr->la_uid   = rec->cr_fsuid;
1094         attr->la_gid   = rec->cr_fsgid;
1095         attr->la_ctime = rec->cr_time;
1096         attr->la_mtime = rec->cr_time;
1097         attr->la_atime = rec->cr_time;
1098         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1099                          LA_CTIME | LA_MTIME | LA_ATIME;
1100         memset(&sp->u, 0, sizeof(sp->u));
1101         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1102
1103         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1104         if (rc < 0)
1105                 RETURN(rc);
1106
1107         if (S_ISLNK(attr->la_mode)) {
1108                 const char *tgt = NULL;
1109
1110                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1111                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1112                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1113                         sp->u.sp_symname = tgt;
1114                 }
1115                 if (tgt == NULL)
1116                         RETURN(-EFAULT);
1117         } else {
1118                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1119                 if (S_ISDIR(attr->la_mode) &&
1120                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1121                         sp->u.sp_ea.eadata =
1122                                 req_capsule_client_get(pill, &RMF_EADATA);
1123                         sp->u.sp_ea.eadatalen =
1124                                 req_capsule_get_size(pill, &RMF_EADATA,
1125                                                      RCL_CLIENT);
1126                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1127                 }
1128         }
1129
1130         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1131                                     &sp->sp_cr_file_secctx,
1132                                     &sp->sp_cr_file_secctx_size);
1133         if (rc < 0)
1134                 RETURN(rc);
1135
1136         rc = mdt_dlmreq_unpack(info);
1137         RETURN(rc);
1138 }
1139
1140 static int mdt_link_unpack(struct mdt_thread_info *info)
1141 {
1142         struct lu_ucred         *uc  = mdt_ucred(info);
1143         struct mdt_rec_link     *rec;
1144         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1145         struct mdt_reint_record *rr = &info->mti_rr;
1146         struct req_capsule      *pill = info->mti_pill;
1147         int rc;
1148         ENTRY;
1149
1150         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1151         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1152         if (rec == NULL)
1153                 RETURN(-EFAULT);
1154
1155         /* This prior initialization is needed for old_init_ucred_reint() */
1156         uc->uc_fsuid = rec->lk_fsuid;
1157         uc->uc_fsgid = rec->lk_fsgid;
1158         uc->uc_cap   = rec->lk_cap;
1159         uc->uc_suppgids[0] = rec->lk_suppgid1;
1160         uc->uc_suppgids[1] = rec->lk_suppgid2;
1161
1162         attr->la_uid = rec->lk_fsuid;
1163         attr->la_gid = rec->lk_fsgid;
1164         rr->rr_fid1 = &rec->lk_fid1;
1165         rr->rr_fid2 = &rec->lk_fid2;
1166         attr->la_ctime = rec->lk_time;
1167         attr->la_mtime = rec->lk_time;
1168         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1169
1170         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1171         if (rc < 0)
1172                 RETURN(rc);
1173
1174         rc = mdt_dlmreq_unpack(info);
1175
1176         RETURN(rc);
1177 }
1178
1179 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1180 {
1181         struct lu_ucred         *uc  = mdt_ucred(info);
1182         struct mdt_rec_unlink   *rec;
1183         struct md_attr          *ma = &info->mti_attr;
1184         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1185         struct mdt_reint_record *rr = &info->mti_rr;
1186         struct req_capsule      *pill = info->mti_pill;
1187         int rc;
1188         ENTRY;
1189
1190         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1191         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1192         if (rec == NULL)
1193                 RETURN(-EFAULT);
1194
1195         /* This prior initialization is needed for old_init_ucred_reint() */
1196         uc->uc_fsuid = rec->ul_fsuid;
1197         uc->uc_fsgid = rec->ul_fsgid;
1198         uc->uc_cap   = rec->ul_cap;
1199         uc->uc_suppgids[0] = rec->ul_suppgid1;
1200         uc->uc_suppgids[1] = -1;
1201
1202         attr->la_uid = rec->ul_fsuid;
1203         attr->la_gid = rec->ul_fsgid;
1204         rr->rr_fid1 = &rec->ul_fid1;
1205         rr->rr_fid2 = &rec->ul_fid2;
1206         attr->la_ctime = rec->ul_time;
1207         attr->la_mtime = rec->ul_time;
1208         attr->la_mode  = rec->ul_mode;
1209         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1210
1211         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1212         if (rc < 0)
1213                 RETURN(rc);
1214
1215         if (rec->ul_bias & MDS_VTX_BYPASS)
1216                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1217         else
1218                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1219
1220         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1221
1222         rc = mdt_dlmreq_unpack(info);
1223         RETURN(rc);
1224 }
1225
1226 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1227 {
1228         info->mti_spec.sp_rm_entry = 1;
1229         return mdt_unlink_unpack(info);
1230 }
1231
1232 static int mdt_rename_unpack(struct mdt_thread_info *info)
1233 {
1234         struct lu_ucred         *uc = mdt_ucred(info);
1235         struct mdt_rec_rename   *rec;
1236         struct md_attr          *ma = &info->mti_attr;
1237         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1238         struct mdt_reint_record *rr = &info->mti_rr;
1239         struct req_capsule      *pill = info->mti_pill;
1240         int rc;
1241         ENTRY;
1242
1243         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1244         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1245         if (rec == NULL)
1246                 RETURN(-EFAULT);
1247
1248         /* This prior initialization is needed for old_init_ucred_reint() */
1249         uc->uc_fsuid = rec->rn_fsuid;
1250         uc->uc_fsgid = rec->rn_fsgid;
1251         uc->uc_cap   = rec->rn_cap;
1252         uc->uc_suppgids[0] = rec->rn_suppgid1;
1253         uc->uc_suppgids[1] = rec->rn_suppgid2;
1254
1255         attr->la_uid = rec->rn_fsuid;
1256         attr->la_gid = rec->rn_fsgid;
1257         rr->rr_fid1 = &rec->rn_fid1;
1258         rr->rr_fid2 = &rec->rn_fid2;
1259         attr->la_ctime = rec->rn_time;
1260         attr->la_mtime = rec->rn_time;
1261         /* rename_tgt contains the mode already */
1262         attr->la_mode = rec->rn_mode;
1263         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1264
1265         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1266         if (rc < 0)
1267                 RETURN(rc);
1268
1269         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1270         if (rc < 0)
1271                 RETURN(rc);
1272
1273         if (rec->rn_bias & MDS_VTX_BYPASS)
1274                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1275         else
1276                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1277
1278         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1279                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1280                 rc = mdt_close_handle_unpack(info);
1281                 if (rc < 0)
1282                         RETURN(rc);
1283                 info->mti_spec.sp_migrate_close = 1;
1284         }
1285
1286         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1287
1288
1289         rc = mdt_dlmreq_unpack(info);
1290
1291         RETURN(rc);
1292 }
1293
1294 /*
1295  * please see comment above LOV_MAGIC_V1_DEF
1296  */
1297 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1298 {
1299         struct lov_user_md_v1   *v1 = eadata;
1300
1301         LASSERT(v1);
1302
1303         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1304                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1305                         v1->lmm_magic |= LOV_MAGIC_DEF;
1306                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1307                          __swab32(LOV_MAGIC_MAGIC))
1308                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEF);
1309         }
1310 }
1311
1312 static int mdt_open_unpack(struct mdt_thread_info *info)
1313 {
1314         struct lu_ucred         *uc = mdt_ucred(info);
1315         struct mdt_rec_create   *rec;
1316         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1317         struct req_capsule      *pill = info->mti_pill;
1318         struct mdt_reint_record *rr   = &info->mti_rr;
1319         struct ptlrpc_request   *req  = mdt_info_req(info);
1320         struct md_op_spec       *sp   = &info->mti_spec;
1321         int rc;
1322         ENTRY;
1323
1324         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1325         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1326         if (rec == NULL)
1327                 RETURN(-EFAULT);
1328
1329         /* This prior initialization is needed for old_init_ucred_reint() */
1330         uc->uc_fsuid = rec->cr_fsuid;
1331         uc->uc_fsgid = rec->cr_fsgid;
1332         uc->uc_cap   = rec->cr_cap;
1333         uc->uc_suppgids[0] = rec->cr_suppgid1;
1334         uc->uc_suppgids[1] = rec->cr_suppgid2;
1335         uc->uc_umask = rec->cr_umask;
1336
1337         rr->rr_fid1   = &rec->cr_fid1;
1338         rr->rr_fid2   = &rec->cr_fid2;
1339         rr->rr_handle = &rec->cr_old_handle;
1340         attr->la_mode = rec->cr_mode;
1341         attr->la_rdev  = rec->cr_rdev;
1342         attr->la_uid   = rec->cr_fsuid;
1343         attr->la_gid   = rec->cr_fsgid;
1344         attr->la_ctime = rec->cr_time;
1345         attr->la_mtime = rec->cr_time;
1346         attr->la_atime = rec->cr_time;
1347         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1348                          LA_CTIME | LA_MTIME | LA_ATIME;
1349         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1350         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1351         /* Do not trigger ASSERTION if client miss to set such flags. */
1352         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1353                 RETURN(-EPROTO);
1354
1355         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1356
1357         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1358
1359         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1360                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1361                                                         RCL_CLIENT);
1362                 if (rr->rr_eadatalen > 0) {
1363                         rr->rr_eadata = req_capsule_client_get(pill,
1364                                                                &RMF_EADATA);
1365                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1366                         sp->u.sp_ea.eadata = rr->rr_eadata;
1367                         sp->no_create = !!req_is_replay(req);
1368                         mdt_fix_lov_magic(info, rr->rr_eadata);
1369                 }
1370
1371                 /*
1372                  * Client default md_size may be 0 right after client start,
1373                  * until all osc are connected, set here just some reasonable
1374                  * value to prevent misbehavior.
1375                  */
1376                 if (rr->rr_eadatalen == 0 &&
1377                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1378                         rr->rr_eadatalen = MIN_MD_SIZE;
1379         }
1380
1381         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1382                                     &sp->sp_cr_file_secctx,
1383                                     &sp->sp_cr_file_secctx_size);
1384
1385         RETURN(rc);
1386 }
1387
1388 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1389 {
1390         struct mdt_reint_record *rr     = &info->mti_rr;
1391         struct lu_ucred         *uc     = mdt_ucred(info);
1392         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1393         struct req_capsule      *pill   = info->mti_pill;
1394         struct mdt_rec_setxattr *rec;
1395         int                      rc;
1396         ENTRY;
1397
1398
1399         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1400                          sizeof(struct mdt_rec_reint));
1401
1402         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1403         if (rec == NULL)
1404                 RETURN(-EFAULT);
1405
1406         /* This prior initialization is needed for old_init_ucred_reint() */
1407         uc->uc_fsuid  = rec->sx_fsuid;
1408         uc->uc_fsgid  = rec->sx_fsgid;
1409         uc->uc_cap    = rec->sx_cap;
1410         uc->uc_suppgids[0] = rec->sx_suppgid1;
1411         uc->uc_suppgids[1] = -1;
1412
1413         rr->rr_opcode = rec->sx_opcode;
1414         rr->rr_fid1   = &rec->sx_fid;
1415         attr->la_valid = rec->sx_valid;
1416         attr->la_ctime = rec->sx_time;
1417         attr->la_size = rec->sx_size;
1418         attr->la_flags = rec->sx_flags;
1419
1420         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1421         if (rc < 0)
1422                 RETURN(rc);
1423
1424         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1425                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1426                                                         RCL_CLIENT);
1427                 if (rr->rr_eadatalen > 0) {
1428                         rr->rr_eadata = req_capsule_client_get(pill,
1429                                                                &RMF_EADATA);
1430                         if (rr->rr_eadata == NULL)
1431                                 RETURN(-EFAULT);
1432                 } else {
1433                         rr->rr_eadata = NULL;
1434                 }
1435         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1436                 CDEBUG(D_INFO, "no xattr data supplied\n");
1437                 RETURN(-EFAULT);
1438         }
1439
1440         if (mdt_dlmreq_unpack(info) < 0)
1441                 RETURN(-EPROTO);
1442
1443         RETURN(0);
1444 }
1445
1446
1447 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1448
1449 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1450         [REINT_SETATTR]  = mdt_setattr_unpack,
1451         [REINT_CREATE]   = mdt_create_unpack,
1452         [REINT_LINK]     = mdt_link_unpack,
1453         [REINT_UNLINK]   = mdt_unlink_unpack,
1454         [REINT_RENAME]   = mdt_rename_unpack,
1455         [REINT_OPEN]     = mdt_open_unpack,
1456         [REINT_SETXATTR] = mdt_setxattr_unpack,
1457         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1458         [REINT_MIGRATE]  = mdt_rename_unpack,
1459 };
1460
1461 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1462 {
1463         int rc;
1464         ENTRY;
1465
1466         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1467         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1468                 info->mti_rr.rr_opcode = op;
1469                 rc = mdt_reint_unpackers[op](info);
1470         } else {
1471                 CERROR("Unexpected opcode %d\n", op);
1472                 rc = -EFAULT;
1473         }
1474         RETURN(rc);
1475 }