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, 2017, 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 <uapi/linux/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 void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
 145 {
 146         if (info && info->mti_exp && info->mti_exp->exp_connection)
 147                 uc->uc_nid = info->mti_exp->exp_connection->c_peer.nid;
 148         else
 149                 uc->uc_nid = LNET_NID_ANY;
 150 }
 151
 152 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
 153                                     struct lu_ucred *uc)
 154 {
 155         struct lu_nodemap *nodemap = NULL;
 156         bool audit = true;
 157
 158         if (info && info->mti_exp) {
 159                 nodemap = nodemap_get_from_exp(info->mti_exp);
 160                 if (nodemap && !IS_ERR(nodemap)) {
 161                         audit = nodemap->nmf_enable_audit;
 162                         nodemap_putref(nodemap);
 163                 }
 164         }
 165
 166         uc->uc_enable_audit = audit;
 167 }
 168
 169 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 170                           void *buf, bool drop_fs_cap)
 171 {
 172         struct ptlrpc_request *req = mdt_info_req(info);
 173         struct mdt_device *mdt = info->mti_mdt;
 174         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 175         struct lu_ucred *ucred = mdt_ucred(info);
 176         struct lu_nodemap *nodemap;
 177         lnet_nid_t peernid = req->rq_peer.nid;
 178         __u32 perm = 0;
 179         int setuid;
 180         int setgid;
 181         bool is_nm_gid_squashed = false;
 182         int rc = 0;
 183
 184         ENTRY;
 185
 186         LASSERT(req->rq_auth_gss);
 187         LASSERT(!req->rq_auth_usr_mdt);
 188         LASSERT(req->rq_user_desc);
 189         LASSERT(ucred != NULL);
 190
 191         ucred->uc_valid = UCRED_INVALID;
 192
 193         nodemap = nodemap_get_from_exp(info->mti_exp);
 194         if (IS_ERR(nodemap))
 195                 RETURN(PTR_ERR(nodemap));
 196
 197         pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 198                                        NODEMAP_CLIENT_TO_FS, pud->pud_uid);
 199         pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 200                                        NODEMAP_CLIENT_TO_FS, pud->pud_gid);
 201         pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 202                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
 203         pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 204                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
 205
 206         ucred->uc_o_uid = pud->pud_uid;
 207         ucred->uc_o_gid = pud->pud_gid;
 208         ucred->uc_o_fsuid = pud->pud_fsuid;
 209         ucred->uc_o_fsgid = pud->pud_fsgid;
 210
 211         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 212                 /* deny access before we get identity ref */
 213                 if (nodemap->nmf_deny_unknown) {
 214                         nodemap_putref(nodemap);
 215                         RETURN(-EACCES);
 216                 }
 217
 218                 ucred->uc_suppgids[0] = -1;
 219                 ucred->uc_suppgids[1] = -1;
 220         }
 221
 222         if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
 223                 is_nm_gid_squashed = true;
 224
 225         nodemap_putref(nodemap);
 226
 227         if (type == BODY_INIT) {
 228                 struct mdt_body *body = (struct mdt_body *)buf;
 229
 230                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 231                 ucred->uc_suppgids[1] = -1;
 232         }
 233
 234         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 235             req->rq_auth_uid != pud->pud_uid) {
 236                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 237                        "while client claims %u:%u/%u:%u\n",
 238                        libcfs_nid2str(peernid), req->rq_auth_uid,
 239                        pud->pud_uid, pud->pud_gid,
 240                        pud->pud_fsuid, pud->pud_fsgid);
 241                 RETURN(-EACCES);
 242         }
 243
 244         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 245                 ucred->uc_identity = NULL;
 246                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 247         } else {
 248                 struct md_identity *identity;
 249
 250                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 251                                             pud->pud_uid);
 252                 if (IS_ERR(identity)) {
 253                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 254                                 ucred->uc_identity = NULL;
 255                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 256                                        CFS_SETGRP_PERM;
 257                         } else {
 258                                 CDEBUG(D_SEC,
 259                                        "Deny access without identity: uid %u\n",
 260                                        pud->pud_uid);
 261                                 RETURN(-EACCES);
 262                         }
 263                 } else {
 264                         ucred->uc_identity = identity;
 265                         perm = mdt_identity_get_perm(ucred->uc_identity,
 266                                                      peernid);
 267                 }
 268         }
 269
 270         /* find out the setuid/setgid attempt */
 271         setuid = (pud->pud_uid != pud->pud_fsuid);
 272         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 273                   (ucred->uc_identity &&
 274                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 275
 276         /* check permission of setuid */
 277         if (setuid && !(perm & CFS_SETUID_PERM)) {
 278                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 279                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 280                 GOTO(out, rc = -EACCES);
 281         }
 282
 283         /* check permission of setgid */
 284         if (setgid && !(perm & CFS_SETGID_PERM)) {
 285                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 286                        "from %s\n", pud->pud_uid, pud->pud_gid,
 287                        pud->pud_fsuid, pud->pud_fsgid,
 288                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 289                 GOTO(out, rc = -EACCES);
 290         }
 291
 292         if (perm & CFS_SETGRP_PERM) {
 293                 /* only set groups if GID is not squashed */
 294                 if (pud->pud_ngroups && !is_nm_gid_squashed) {
 295                         /* setgroups for local client */
 296                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 297                         if (!ucred->uc_ginfo) {
 298                                 CERROR("failed to alloc %d groups\n",
 299                                        pud->pud_ngroups);
 300                                 GOTO(out, rc = -ENOMEM);
 301                         }
 302
 303                         lustre_groups_from_list(ucred->uc_ginfo,
 304                                                 pud->pud_groups);
 305                         lustre_groups_sort(ucred->uc_ginfo);
 306                 } else {
 307                         ucred->uc_suppgids[0] = -1;
 308                         ucred->uc_suppgids[1] = -1;
 309                         ucred->uc_ginfo = NULL;
 310                 }
 311         } else {
 312                 ucred->uc_suppgids[0] = -1;
 313                 ucred->uc_suppgids[1] = -1;
 314                 ucred->uc_ginfo = NULL;
 315         }
 316
 317         ucred->uc_uid = pud->pud_uid;
 318         ucred->uc_gid = pud->pud_gid;
 319
 320         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 321                 ucred->uc_cap = 0;
 322         } else {
 323                 ucred->uc_cap = pud->pud_cap;
 324         }
 325         ucred->uc_fsuid = pud->pud_fsuid;
 326         ucred->uc_fsgid = pud->pud_fsgid;
 327
 328         /* process root_squash here. */
 329         mdt_root_squash(info, peernid);
 330
 331         /* remove fs privilege for non-root user. */
 332         if (ucred->uc_fsuid && drop_fs_cap)
 333                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 334         else
 335                 ucred->uc_cap = pud->pud_cap;
 336         ucred->uc_valid = UCRED_NEW;
 337         ucred_set_jobid(info, ucred);
 338         ucred_set_nid(info, ucred);
 339         ucred_set_audit_enabled(info, ucred);
 340
 341         EXIT;
 342
 343 out:
 344         if (rc) {
 345                 if (ucred->uc_ginfo) {
 346                         put_group_info(ucred->uc_ginfo);
 347                         ucred->uc_ginfo = NULL;
 348                 }
 349                 if (ucred->uc_identity) {
 350                         mdt_identity_put(mdt->mdt_identity_cache,
 351                                          ucred->uc_identity);
 352                         ucred->uc_identity = NULL;
 353                 }
 354         }
 355
 356         return rc;
 357 }
 358
 359 /**
 360  * Check whether allow the client to set supplementary group IDs or not.
 361  *
 362  * \param[in] info      pointer to the thread context
 363  * \param[in] uc        pointer to the RPC user descriptor
 364  *
 365  * \retval              true if allow to set supplementary group IDs
 366  * \retval              false for other cases
 367  */
 368 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 369 {
 370         __u32 perm;
 371
 372         /* 1. If identity_upcall is disabled,
 373          *    permit local client to do anything. */
 374         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 375                 return true;
 376
 377         /* 2. If fail to get related identities, then forbid any client to
 378          *    set supplementary group IDs. */
 379         if (uc->uc_identity == NULL)
 380                 return false;
 381
 382         /* 3. Check the permission in the identities. */
 383         perm = mdt_identity_get_perm(uc->uc_identity,
 384                                      mdt_info_req(info)->rq_peer.nid);
 385         if (perm & CFS_SETGRP_PERM)
 386                 return true;
 387
 388         return false;
 389 }
 390
 391 int mdt_check_ucred(struct mdt_thread_info *info)
 392 {
 393         struct ptlrpc_request   *req = mdt_info_req(info);
 394         struct mdt_device       *mdt = info->mti_mdt;
 395         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 396         struct lu_ucred         *ucred = mdt_ucred(info);
 397         struct md_identity      *identity = NULL;
 398         lnet_nid_t               peernid = req->rq_peer.nid;
 399         __u32                    perm = 0;
 400         int                      setuid;
 401         int                      setgid;
 402         int                      rc = 0;
 403
 404         ENTRY;
 405
 406         LASSERT(ucred != NULL);
 407         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 408                 RETURN(0);
 409
 410         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 411                 RETURN(0);
 412
 413         /* sanity check: if we use strong authentication, we expect the
 414          * uid which client claimed is true */
 415         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 416             req->rq_auth_uid != pud->pud_uid) {
 417                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 418                        "while client claims %u:%u/%u:%u\n",
 419                        libcfs_nid2str(peernid), req->rq_auth_uid,
 420                        pud->pud_uid, pud->pud_gid,
 421                        pud->pud_fsuid, pud->pud_fsgid);
 422                 RETURN(-EACCES);
 423         }
 424
 425         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 426                 RETURN(0);
 427
 428         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 429         if (IS_ERR(identity)) {
 430                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 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, 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 (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 477                 /* deny access before we get identity ref */
 478                 if (nodemap->nmf_deny_unknown)
 479                         RETURN(-EACCES);
 480
 481                 uc->uc_cap = 0;
 482                 uc->uc_suppgids[0] = -1;
 483                 uc->uc_suppgids[1] = -1;
 484         }
 485
 486         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 487                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 488                                             uc->uc_fsuid);
 489                 if (IS_ERR(identity)) {
 490                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 491                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 492                                 identity = NULL;
 493                         } else {
 494                                 CDEBUG(D_SEC, "Deny access without identity: "
 495                                        "uid %u\n", uc->uc_fsuid);
 496                                 RETURN(-EACCES);
 497                         }
 498                 }
 499         }
 500         uc->uc_identity = identity;
 501
 502         /* process root_squash here. */
 503         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 504
 505         /* remove fs privilege for non-root user. */
 506         if (uc->uc_fsuid && drop_fs_cap)
 507                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 508         uc->uc_valid = UCRED_OLD;
 509         ucred_set_jobid(info, uc);
 510         ucred_set_nid(info, uc);
 511         ucred_set_audit_enabled(info, uc);
 512
 513         EXIT;
 514
 515         return 0;
 516 }
 517
 518 static int old_init_ucred(struct mdt_thread_info *info,
 519                           struct mdt_body *body, bool drop_fs_cap)
 520 {
 521         struct lu_ucred *uc = mdt_ucred(info);
 522         struct lu_nodemap *nodemap;
 523         int rc;
 524         ENTRY;
 525
 526         nodemap = nodemap_get_from_exp(info->mti_exp);
 527         if (IS_ERR(nodemap))
 528                 RETURN(PTR_ERR(nodemap));
 529
 530         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 531                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 532         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 533                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 534         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 535                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 536         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 537                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 538
 539         LASSERT(uc != NULL);
 540         uc->uc_valid = UCRED_INVALID;
 541         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 542         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 543         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 544         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 545         uc->uc_suppgids[0] = body->mbo_suppgid;
 546         uc->uc_suppgids[1] = -1;
 547         uc->uc_ginfo = NULL;
 548         uc->uc_cap = body->mbo_capability;
 549
 550         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 551         nodemap_putref(nodemap);
 552
 553         RETURN(rc);
 554 }
 555
 556 static int old_init_ucred_reint(struct mdt_thread_info *info)
 557 {
 558         struct lu_ucred *uc = mdt_ucred(info);
 559         struct lu_nodemap *nodemap;
 560         int rc;
 561         ENTRY;
 562
 563         nodemap = nodemap_get_from_exp(info->mti_exp);
 564         if (IS_ERR(nodemap))
 565                 RETURN(PTR_ERR(nodemap));
 566
 567         LASSERT(uc != NULL);
 568
 569         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 570                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 571         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 572                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 573
 574         uc->uc_valid = UCRED_INVALID;
 575         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 576         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 577         uc->uc_ginfo = NULL;
 578
 579         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 580         nodemap_putref(nodemap);
 581
 582         RETURN(rc);
 583 }
 584
 585 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 586                                    struct mdt_body *body,
 587                                    bool drop_fs_cap)
 588 {
 589         struct ptlrpc_request   *req = mdt_info_req(info);
 590         struct lu_ucred         *uc  = mdt_ucred(info);
 591
 592         LASSERT(uc != NULL);
 593         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 594                 return 0;
 595
 596         mdt_exit_ucred(info);
 597
 598         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 599                 return old_init_ucred(info, body, drop_fs_cap);
 600         else
 601                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 602 }
 603
 604 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 605 {
 606         return __mdt_init_ucred(info, body, true);
 607 }
 608
 609 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 610  * doesn't set correct fsuid explicitely, but raise capability to allow
 611  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 612  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 613  * will fail permission check. */
 614 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 615                                   struct mdt_body *body)
 616 {
 617         return __mdt_init_ucred(info, body, false);
 618 }
 619
 620 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 621 {
 622         struct ptlrpc_request *req = mdt_info_req(info);
 623         struct lu_ucred       *uc  = mdt_ucred(info);
 624         struct md_attr        *ma  = &info->mti_attr;
 625
 626         LASSERT(uc != NULL);
 627         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 628                 return 0;
 629
 630         /* LU-5564: for normal close request, skip permission check */
 631         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 632             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 633                 uc->uc_cap |= CFS_CAP_FS_MASK;
 634
 635         mdt_exit_ucred(info);
 636
 637         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 638                 return old_init_ucred_reint(info);
 639         else
 640                 return new_init_ucred(info, REC_INIT, NULL, true);
 641 }
 642
 643 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 644 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 645 {
 646         const struct lov_ost_data_v1 *lod;
 647         __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
 648         __u16 count;
 649         int i;
 650
 651         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 652                 return;
 653
 654         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 655                POSTID(&lmm->lmm_oi), lmm_magic,
 656                le32_to_cpu(lmm->lmm_pattern));
 657
 658         /* No support for compount layouts yet */
 659         if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
 660                 return;
 661
 662         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 663         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 664                le32_to_cpu(lmm->lmm_stripe_size), count);
 665
 666         /* If it's a directory or a released file, then there are
 667          * no actual objects to print, so bail out. */
 668         if (valid & OBD_MD_FLDIREA ||
 669             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 670                 return;
 671
 672         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 673         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 674                 struct ost_id oi;
 675
 676                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 677                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 678                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 679         }
 680 }
 681
 682 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 683 {
 684         const struct lmv_mds_md_v1 *lmm1;
 685         int                        i;
 686
 687         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 688                 return;
 689
 690         lmm1 = &lmv->lmv_md_v1;
 691         CDEBUG(level,
 692                "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
 693                le32_to_cpu(lmm1->lmv_magic),
 694                le32_to_cpu(lmm1->lmv_master_mdt_index),
 695                le32_to_cpu(lmm1->lmv_stripe_count),
 696                le32_to_cpu(lmm1->lmv_hash_type));
 697
 698         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 699                 return;
 700
 701         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 702                 struct lu_fid fid;
 703
 704                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 705                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 706         }
 707 }
 708
 709 /* Shrink and/or grow reply buffers */
 710 int mdt_fix_reply(struct mdt_thread_info *info)
 711 {
 712         struct req_capsule *pill = info->mti_pill;
 713         struct mdt_body    *body;
 714         int                md_size, md_packed = 0;
 715         int                acl_size;
 716         int                rc = 0;
 717         ENTRY;
 718
 719         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 720         LASSERT(body != NULL);
 721
 722         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 723                                OBD_MD_LINKNAME))
 724                 md_size = body->mbo_eadatasize;
 725         else
 726                 md_size = 0;
 727
 728         acl_size = body->mbo_aclsize;
 729
 730         /* this replay - not send info to client */
 731         if (info->mti_spec.no_create) {
 732                 md_size = 0;
 733                 acl_size = 0;
 734         }
 735
 736         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 737                md_size, acl_size);
 738 /*
 739             &RMF_MDT_BODY,
 740             &RMF_MDT_MD,
 741             &RMF_ACL, or &RMF_LOGCOOKIES
 742 (optional)  &RMF_CAPA1,
 743 (optional)  &RMF_CAPA2,
 744 (optional)  something else
 745 */
 746
 747         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 748          * buffers before growing it */
 749         if (info->mti_big_lmm_used) {
 750                 /* big_lmm buffer may be used even without packing the result
 751                  * into reply, just for internal server needs */
 752                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 753                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 754                                                          RCL_SERVER);
 755
 756                 /* free big lmm if md_size is not needed */
 757                 if (md_size == 0 || md_packed == 0) {
 758                         info->mti_big_lmm_used = 0;
 759                 } else {
 760                         /* buffer must be allocated separately */
 761                         LASSERT(info->mti_attr.ma_lmm !=
 762                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 763                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 764                 }
 765         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 766                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 767         }
 768
 769         if (info->mti_big_acl_used) {
 770                 if (acl_size == 0)
 771                         info->mti_big_acl_used = 0;
 772                 else
 773                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 774         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 775                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 776         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 777                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 778         }
 779
 780         /* Shrink optional SECCTX buffer if it is not used */
 781         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
 782             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
 783             !(body->mbo_valid & OBD_MD_SECCTX))
 784                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
 785
 786         /*
 787          * Some more field should be shrinked if needed.
 788          * This should be done by those who added fields to reply message.
 789          */
 790
 791         /* Grow MD buffer if needed finally */
 792         if (info->mti_big_lmm_used) {
 793                 void *lmm;
 794
 795                 LASSERT(md_size > md_packed);
 796                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 797                        md_size - md_packed);
 798                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 799                 if (rc) {
 800                         /* we can't answer with proper LOV EA, drop flags,
 801                          * the rc is also returned so this request is
 802                          * considered as failed */
 803                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 804                         /* don't return transno along with error */
 805                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 806                 } else {
 807                         /* now we need to pack right LOV/LMV EA */
 808                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 809                         if (info->mti_attr.ma_valid & MA_LOV) {
 810                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 811                                                              RCL_SERVER) ==
 812                                                 info->mti_attr.ma_lmm_size);
 813                                 memcpy(lmm, info->mti_attr.ma_lmm,
 814                                        info->mti_attr.ma_lmm_size);
 815                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 816                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 817                                                              RCL_SERVER) ==
 818                                                 info->mti_attr.ma_lmv_size);
 819                                 memcpy(lmm, info->mti_attr.ma_lmv,
 820                                        info->mti_attr.ma_lmv_size);
 821                         }
 822                 }
 823
 824                 /* update mdt_max_mdsize so clients will be aware about that */
 825                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 826                         info->mti_mdt->mdt_max_mdsize =
 827                                                 info->mti_attr.ma_lmm_size;
 828                 info->mti_big_lmm_used = 0;
 829         }
 830
 831         if (info->mti_big_acl_used) {
 832                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 833                        acl_size);
 834
 835                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 836                 if (rc) {
 837                         body->mbo_valid &= ~OBD_MD_FLACL;
 838                 } else {
 839                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 840
 841                         memcpy(acl, info->mti_big_acl, acl_size);
 842                 }
 843
 844                 info->mti_big_acl_used = 0;
 845         }
 846
 847         RETURN(rc);
 848 }
 849
 850
 851 /* if object is dying, pack the lov/llog data,
 852  * parameter info->mti_attr should be valid at this point!
 853  * Also implements RAoLU policy */
 854 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 855                            struct md_attr *ma)
 856 {
 857         struct mdt_body *repbody = NULL;
 858         const struct lu_attr *la = &ma->ma_attr;
 859         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 860         int rc;
 861         __u64 need = 0;
 862         struct hsm_action_item hai = {
 863                 .hai_len = sizeof(hai),
 864                 .hai_action = HSMA_REMOVE,
 865                 .hai_extent.length = -1,
 866                 .hai_cookie = 0,
 867                 .hai_gid = 0,
 868         };
 869         int archive_id;
 870
 871         ENTRY;
 872
 873         if (mdt_info_req(info) != NULL) {
 874                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 875                 LASSERT(repbody != NULL);
 876         } else {
 877                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 878         }
 879
 880         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 881                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 882
 883         if (ma->ma_valid & MA_LOV) {
 884                 CERROR("No need in LOV EA upon unlink\n");
 885                 dump_stack();
 886         }
 887         if (repbody != NULL)
 888                 repbody->mbo_eadatasize = 0;
 889
 890         /* Only check unlinked and archived if RAoLU and upon last close */
 891         if (!cdt->cdt_remove_archive_on_last_unlink ||
 892             atomic_read(&mo->mot_open_count) != 0)
 893                 RETURN(0);
 894
 895         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 896         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 897                 RETURN(0);
 898
 899         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 900                 RETURN(0);
 901
 902         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 903         if (need != 0) {
 904                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 905                  * try setting them */
 906                 ma->ma_need |= need;
 907                 rc = mdt_attr_get_complex(info, mo, ma);
 908                 if (rc) {
 909                         CERROR("%s: unable to fetch missing attributes of"
 910                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 911                                PFID(mdt_object_fid(mo)), rc);
 912                         RETURN(0);
 913                 }
 914
 915                 if (need & MA_INODE) {
 916                         if (ma->ma_valid & MA_INODE) {
 917                                 if (ma->ma_attr.la_nlink != 0)
 918                                         RETURN(0);
 919                         } else {
 920                                 RETURN(0);
 921                         }
 922                 }
 923
 924                 if (need & MA_HSM) {
 925                         if (ma->ma_valid & MA_HSM) {
 926                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 927                                         RETURN(0);
 928                         } else {
 929                                 RETURN(0);
 930                         }
 931                 }
 932         }
 933
 934         /* RAoLU policy is active, last close on file has occured,
 935          * file is unlinked, file is archived, so create remove request
 936          * for copytool!
 937          * If CDT is not running, requests will be logged for later. */
 938         if (ma->ma_hsm.mh_arch_id != 0)
 939                 archive_id = ma->ma_hsm.mh_arch_id;
 940         else
 941                 archive_id = cdt->cdt_default_archive_id;
 942
 943         hai.hai_fid = *mdt_object_fid(mo);
 944
 945         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
 946                                   &hai);
 947         if (rc)
 948                 CERROR("%s: unable to add HSM remove request for "DFID
 949                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
 950                        PFID(mdt_object_fid(mo)), rc);
 951
 952         RETURN(0);
 953 }
 954
 955 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 956                                   struct md_attr *ma)
 957 {
 958         __u64 out;
 959
 960         out = 0;
 961         if (in & MDS_ATTR_MODE)
 962                 out |= LA_MODE;
 963         if (in & MDS_ATTR_UID)
 964                 out |= LA_UID;
 965         if (in & MDS_ATTR_GID)
 966                 out |= LA_GID;
 967         if (in & MDS_ATTR_SIZE)
 968                 out |= LA_SIZE;
 969         if (in & MDS_ATTR_BLOCKS)
 970                 out |= LA_BLOCKS;
 971         if (in & MDS_ATTR_ATIME_SET)
 972                 out |= LA_ATIME;
 973         if (in & MDS_ATTR_CTIME_SET)
 974                 out |= LA_CTIME;
 975         if (in & MDS_ATTR_MTIME_SET)
 976                 out |= LA_MTIME;
 977         if (in & MDS_ATTR_ATTR_FLAG)
 978                 out |= LA_FLAGS;
 979         if (in & MDS_ATTR_KILL_SUID)
 980                 out |= LA_KILL_SUID;
 981         if (in & MDS_ATTR_KILL_SGID)
 982                 out |= LA_KILL_SGID;
 983         if (in & MDS_ATTR_PROJID)
 984                 out |= LA_PROJID;
 985         if (in & MDS_ATTR_LSIZE)
 986                 out |= LA_LSIZE;
 987         if (in & MDS_ATTR_LBLOCKS)
 988                 out |= LA_LBLOCKS;
 989
 990         if (in & MDS_ATTR_FROM_OPEN)
 991                 rr->rr_flags |= MRF_OPEN_TRUNC;
 992         if (in & MDS_ATTR_OVERRIDE)
 993                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 994         if (in & MDS_ATTR_FORCE)
 995                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 996
 997         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
 998                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 999                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
1000                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1001                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1002                 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1003                 MDS_ATTR_OVERRIDE);
1004         if (in != 0)
1005                 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", in);
1006         return out;
1007 }
1008
1009 /* unpacking */
1010
1011 int mdt_name_unpack(struct req_capsule *pill,
1012                     const struct req_msg_field *field,
1013                     struct lu_name *ln,
1014                     enum mdt_name_flags flags)
1015 {
1016         ln->ln_name = req_capsule_client_get(pill, field);
1017         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1018
1019         if (!lu_name_is_valid(ln)) {
1020                 ln->ln_name = NULL;
1021                 ln->ln_namelen = 0;
1022
1023                 return -EPROTO;
1024         }
1025
1026         if ((flags & MNF_FIX_ANON) &&
1027             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1028                 /* Newer (3.x) kernels use a name of "/" for the
1029                  * "anonymous" disconnected dentries from NFS
1030                  * filehandle conversion. See d_obtain_alias(). */
1031                 ln->ln_name = NULL;
1032                 ln->ln_namelen = 0;
1033         }
1034
1035         return 0;
1036 }
1037
1038 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1039                                   const char **secctx_name,
1040                                   void **secctx, size_t *secctx_size)
1041 {
1042         const char *name;
1043         size_t name_size;
1044
1045         *secctx_name = NULL;
1046         *secctx = NULL;
1047         *secctx_size = 0;
1048
1049         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1050             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1051                 return 0;
1052
1053         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1054                                          RCL_CLIENT);
1055         if (name_size == 0)
1056                 return 0;
1057
1058         if (name_size > XATTR_NAME_MAX + 1)
1059                 return -EPROTO;
1060
1061         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1062         if (strnlen(name, name_size) != name_size - 1)
1063                 return -EPROTO;
1064
1065         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1066             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1067                 return -EPROTO;
1068
1069         *secctx_name = name;
1070         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1071         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1072
1073         return 0;
1074 }
1075
1076 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1077 {
1078         struct lu_ucred *uc = mdt_ucred(info);
1079         struct md_attr *ma = &info->mti_attr;
1080         struct lu_attr *la = &ma->ma_attr;
1081         struct req_capsule *pill = info->mti_pill;
1082         struct mdt_reint_record *rr = &info->mti_rr;
1083         struct mdt_rec_setattr *rec;
1084         struct lu_nodemap *nodemap;
1085
1086         ENTRY;
1087
1088         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1089         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1090         if (rec == NULL)
1091                 RETURN(-EFAULT);
1092
1093         /* This prior initialization is needed for old_init_ucred_reint() */
1094         uc->uc_fsuid = rec->sa_fsuid;
1095         uc->uc_fsgid = rec->sa_fsgid;
1096         uc->uc_cap   = rec->sa_cap;
1097         uc->uc_suppgids[0] = rec->sa_suppgid;
1098         uc->uc_suppgids[1] = -1;
1099
1100         rr->rr_fid1 = &rec->sa_fid;
1101         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1102         la->la_mode  = rec->sa_mode;
1103         la->la_flags = rec->sa_attr_flags;
1104
1105         nodemap = nodemap_get_from_exp(info->mti_exp);
1106         if (IS_ERR(nodemap))
1107                 RETURN(PTR_ERR(nodemap));
1108
1109         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1110                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1111         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1112                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1113         la->la_projid = rec->sa_projid;
1114         nodemap_putref(nodemap);
1115
1116         la->la_size  = rec->sa_size;
1117         la->la_blocks = rec->sa_blocks;
1118         la->la_ctime = rec->sa_ctime;
1119         la->la_atime = rec->sa_atime;
1120         la->la_mtime = rec->sa_mtime;
1121         ma->ma_valid = MA_INODE;
1122
1123         ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1124                                 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE);
1125         RETURN(0);
1126 }
1127
1128 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1129 {
1130         struct req_capsule *pill = info->mti_pill;
1131         struct mdt_ioepoch *ioepoch;
1132         ENTRY;
1133
1134         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1135                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1136         else
1137                 ioepoch = NULL;
1138
1139         if (ioepoch == NULL)
1140                 RETURN(-EPROTO);
1141
1142         info->mti_open_handle = ioepoch->mio_open_handle;
1143
1144         RETURN(0);
1145 }
1146
1147 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1148         struct req_capsule      *pill = info->mti_pill;
1149
1150         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1151                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1152                 if (info->mti_dlm_req == NULL)
1153                         RETURN(-EFAULT);
1154         }
1155
1156         RETURN(0);
1157 }
1158
1159 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1160 {
1161         struct mdt_reint_record *rr = &info->mti_rr;
1162         struct md_attr          *ma = &info->mti_attr;
1163         struct req_capsule      *pill = info->mti_pill;
1164         int rc;
1165         ENTRY;
1166
1167         rc = mdt_setattr_unpack_rec(info);
1168         if (rc)
1169                 RETURN(rc);
1170
1171         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1172                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1173                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1174                                                         RCL_CLIENT);
1175                 if (rr->rr_eadatalen > 0) {
1176                         const struct lmv_user_md        *lum;
1177
1178                         lum = rr->rr_eadata;
1179                         /* Sigh ma_valid(from req) does not indicate whether
1180                          * it will set LOV/LMV EA, so we have to check magic */
1181                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1182                                 ma->ma_valid |= MA_LMV;
1183                                 ma->ma_lmv = (void *)rr->rr_eadata;
1184                                 ma->ma_lmv_size = rr->rr_eadatalen;
1185                         } else {
1186                                 ma->ma_valid |= MA_LOV;
1187                                 ma->ma_lmm = (void *)rr->rr_eadata;
1188                                 ma->ma_lmm_size = rr->rr_eadatalen;
1189                         }
1190                 }
1191         }
1192
1193         rc = mdt_dlmreq_unpack(info);
1194         RETURN(rc);
1195 }
1196
1197 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1198 {
1199         struct md_attr          *ma = &info->mti_attr;
1200         struct req_capsule      *pill = info->mti_pill;
1201         ENTRY;
1202
1203         if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1204                 RETURN(0);
1205
1206         req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1207
1208         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1209             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1210                 RETURN(-EFAULT);
1211
1212         RETURN(0);
1213 }
1214
1215 int mdt_close_unpack(struct mdt_thread_info *info)
1216 {
1217         int rc;
1218         ENTRY;
1219
1220         rc = mdt_close_handle_unpack(info);
1221         if (rc)
1222                 RETURN(rc);
1223
1224         rc = mdt_setattr_unpack_rec(info);
1225         if (rc)
1226                 RETURN(rc);
1227
1228         rc = mdt_close_intent_unpack(info);
1229         if (rc)
1230                 RETURN(rc);
1231
1232         RETURN(mdt_init_ucred_reint(info));
1233 }
1234
1235 static int mdt_create_unpack(struct mdt_thread_info *info)
1236 {
1237         struct lu_ucred *uc  = mdt_ucred(info);
1238         struct mdt_rec_create *rec;
1239         struct lu_attr *attr = &info->mti_attr.ma_attr;
1240         struct mdt_reint_record *rr = &info->mti_rr;
1241         struct req_capsule      *pill = info->mti_pill;
1242         struct md_op_spec       *sp = &info->mti_spec;
1243         int rc;
1244
1245         ENTRY;
1246
1247         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1248         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1249         if (rec == NULL)
1250                 RETURN(-EFAULT);
1251
1252         /* This prior initialization is needed for old_init_ucred_reint() */
1253         uc->uc_fsuid = rec->cr_fsuid;
1254         uc->uc_fsgid = rec->cr_fsgid;
1255         uc->uc_cap   = rec->cr_cap;
1256         uc->uc_suppgids[0] = rec->cr_suppgid1;
1257         uc->uc_suppgids[1] = -1;
1258         uc->uc_umask = rec->cr_umask;
1259
1260         rr->rr_fid1 = &rec->cr_fid1;
1261         rr->rr_fid2 = &rec->cr_fid2;
1262         attr->la_mode = rec->cr_mode;
1263         attr->la_rdev  = rec->cr_rdev;
1264         attr->la_uid   = rec->cr_fsuid;
1265         attr->la_gid   = rec->cr_fsgid;
1266         attr->la_ctime = rec->cr_time;
1267         attr->la_mtime = rec->cr_time;
1268         attr->la_atime = rec->cr_time;
1269         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1270                          LA_CTIME | LA_MTIME | LA_ATIME;
1271         memset(&sp->u, 0, sizeof(sp->u));
1272         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1273
1274         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1275         if (rc < 0)
1276                 RETURN(rc);
1277
1278         if (S_ISLNK(attr->la_mode)) {
1279                 const char *tgt = NULL;
1280
1281                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1282                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1283                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1284                         sp->u.sp_symname = tgt;
1285                 }
1286                 if (tgt == NULL)
1287                         RETURN(-EFAULT);
1288         } else {
1289                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1290                 if (S_ISDIR(attr->la_mode) &&
1291                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1292                         sp->u.sp_ea.eadata =
1293                                 req_capsule_client_get(pill, &RMF_EADATA);
1294                         sp->u.sp_ea.eadatalen =
1295                                 req_capsule_get_size(pill, &RMF_EADATA,
1296                                                      RCL_CLIENT);
1297                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1298                 }
1299         }
1300
1301         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1302                                     &sp->sp_cr_file_secctx,
1303                                     &sp->sp_cr_file_secctx_size);
1304         if (rc < 0)
1305                 RETURN(rc);
1306
1307         rc = req_check_sepol(pill);
1308         if (rc)
1309                 RETURN(rc);
1310
1311         rc = mdt_dlmreq_unpack(info);
1312         RETURN(rc);
1313 }
1314
1315 static int mdt_link_unpack(struct mdt_thread_info *info)
1316 {
1317         struct lu_ucred *uc  = mdt_ucred(info);
1318         struct mdt_rec_link *rec;
1319         struct lu_attr *attr = &info->mti_attr.ma_attr;
1320         struct mdt_reint_record *rr = &info->mti_rr;
1321         struct req_capsule *pill = info->mti_pill;
1322         int rc;
1323
1324         ENTRY;
1325
1326         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1327         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1328         if (rec == NULL)
1329                 RETURN(-EFAULT);
1330
1331         /* This prior initialization is needed for old_init_ucred_reint() */
1332         uc->uc_fsuid = rec->lk_fsuid;
1333         uc->uc_fsgid = rec->lk_fsgid;
1334         uc->uc_cap   = rec->lk_cap;
1335         uc->uc_suppgids[0] = rec->lk_suppgid1;
1336         uc->uc_suppgids[1] = rec->lk_suppgid2;
1337
1338         attr->la_uid = rec->lk_fsuid;
1339         attr->la_gid = rec->lk_fsgid;
1340         rr->rr_fid1 = &rec->lk_fid1;
1341         rr->rr_fid2 = &rec->lk_fid2;
1342         attr->la_ctime = rec->lk_time;
1343         attr->la_mtime = rec->lk_time;
1344         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1345
1346         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1347         if (rc < 0)
1348                 RETURN(rc);
1349
1350         rc = req_check_sepol(pill);
1351         if (rc)
1352                 RETURN(rc);
1353
1354         rc = mdt_dlmreq_unpack(info);
1355
1356         RETURN(rc);
1357 }
1358
1359 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1360 {
1361         struct lu_ucred *uc  = mdt_ucred(info);
1362         struct mdt_rec_unlink *rec;
1363         struct lu_attr *attr = &info->mti_attr.ma_attr;
1364         struct mdt_reint_record *rr = &info->mti_rr;
1365         struct req_capsule      *pill = info->mti_pill;
1366         int rc;
1367
1368         ENTRY;
1369
1370         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1371         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1372         if (rec == NULL)
1373                 RETURN(-EFAULT);
1374
1375         /* This prior initialization is needed for old_init_ucred_reint() */
1376         uc->uc_fsuid = rec->ul_fsuid;
1377         uc->uc_fsgid = rec->ul_fsgid;
1378         uc->uc_cap   = rec->ul_cap;
1379         uc->uc_suppgids[0] = rec->ul_suppgid1;
1380         uc->uc_suppgids[1] = -1;
1381
1382         attr->la_uid = rec->ul_fsuid;
1383         attr->la_gid = rec->ul_fsgid;
1384         rr->rr_fid1 = &rec->ul_fid1;
1385         rr->rr_fid2 = &rec->ul_fid2;
1386         attr->la_ctime = rec->ul_time;
1387         attr->la_mtime = rec->ul_time;
1388         attr->la_mode  = rec->ul_mode;
1389         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1390
1391         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1392         if (rc < 0)
1393                 RETURN(rc);
1394
1395         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1396
1397         rc = req_check_sepol(pill);
1398         if (rc)
1399                 RETURN(rc);
1400
1401         rc = mdt_dlmreq_unpack(info);
1402         RETURN(rc);
1403 }
1404
1405 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1406 {
1407         info->mti_spec.sp_rm_entry = 1;
1408         return mdt_unlink_unpack(info);
1409 }
1410
1411 static int mdt_rename_unpack(struct mdt_thread_info *info)
1412 {
1413         struct lu_ucred *uc = mdt_ucred(info);
1414         struct mdt_rec_rename *rec;
1415         struct lu_attr *attr = &info->mti_attr.ma_attr;
1416         struct mdt_reint_record *rr = &info->mti_rr;
1417         struct req_capsule *pill = info->mti_pill;
1418         struct md_op_spec *spec = &info->mti_spec;
1419         int rc;
1420
1421         ENTRY;
1422
1423         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1424         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1425         if (rec == NULL)
1426                 RETURN(-EFAULT);
1427
1428         /* This prior initialization is needed for old_init_ucred_reint() */
1429         uc->uc_fsuid = rec->rn_fsuid;
1430         uc->uc_fsgid = rec->rn_fsgid;
1431         uc->uc_cap   = rec->rn_cap;
1432         uc->uc_suppgids[0] = rec->rn_suppgid1;
1433         uc->uc_suppgids[1] = rec->rn_suppgid2;
1434
1435         attr->la_uid = rec->rn_fsuid;
1436         attr->la_gid = rec->rn_fsgid;
1437         rr->rr_fid1 = &rec->rn_fid1;
1438         rr->rr_fid2 = &rec->rn_fid2;
1439         attr->la_ctime = rec->rn_time;
1440         attr->la_mtime = rec->rn_time;
1441         /* rename_tgt contains the mode already */
1442         attr->la_mode = rec->rn_mode;
1443         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1444
1445         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1446         if (rc < 0)
1447                 RETURN(rc);
1448
1449         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1450         if (rc < 0)
1451                 RETURN(rc);
1452
1453         spec->no_create = !!req_is_replay(mdt_info_req(info));
1454
1455         rc = req_check_sepol(pill);
1456         if (rc)
1457                 RETURN(rc);
1458
1459         rc = mdt_dlmreq_unpack(info);
1460
1461         RETURN(rc);
1462 }
1463
1464 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1465 {
1466         struct lu_ucred *uc = mdt_ucred(info);
1467         struct mdt_rec_rename *rec;
1468         struct lu_attr *attr = &info->mti_attr.ma_attr;
1469         struct mdt_reint_record *rr = &info->mti_rr;
1470         struct req_capsule *pill = info->mti_pill;
1471         struct md_op_spec *spec = &info->mti_spec;
1472         int rc;
1473
1474         ENTRY;
1475
1476         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1477         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1478         if (rec == NULL)
1479                 RETURN(-EFAULT);
1480
1481         /* This prior initialization is needed for old_init_ucred_reint() */
1482         uc->uc_fsuid = rec->rn_fsuid;
1483         uc->uc_fsgid = rec->rn_fsgid;
1484         uc->uc_cap   = rec->rn_cap;
1485         uc->uc_suppgids[0] = rec->rn_suppgid1;
1486         uc->uc_suppgids[1] = rec->rn_suppgid2;
1487
1488         attr->la_uid = rec->rn_fsuid;
1489         attr->la_gid = rec->rn_fsgid;
1490         rr->rr_fid1 = &rec->rn_fid1;
1491         rr->rr_fid2 = &rec->rn_fid2;
1492         attr->la_ctime = rec->rn_time;
1493         attr->la_mtime = rec->rn_time;
1494         /* rename_tgt contains the mode already */
1495         attr->la_mode = rec->rn_mode;
1496         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1497         spec->sp_cr_flags = 0;
1498
1499         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1500         if (rc < 0)
1501                 RETURN(rc);
1502
1503         if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1504                 rc = mdt_close_handle_unpack(info);
1505                 if (rc)
1506                         RETURN(rc);
1507
1508                 spec->sp_migrate_close = 1;
1509         }
1510
1511         /* lustre version > 2.11 migration packs lum */
1512         if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1513                 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1514                         rr->rr_eadatalen = req_capsule_get_size(pill,
1515                                                                 &RMF_EADATA,
1516                                                                 RCL_CLIENT);
1517                         if (rr->rr_eadatalen > 0) {
1518                                 rr->rr_eadata = req_capsule_client_get(pill,
1519                                                                 &RMF_EADATA);
1520                                 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1521                                 spec->u.sp_ea.eadata = rr->rr_eadata;
1522                                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1523                         }
1524                 } else {
1525                         /* old client doesn't provide lum. */
1526                         RETURN(-EOPNOTSUPP);
1527                 }
1528         }
1529
1530         spec->no_create = !!req_is_replay(mdt_info_req(info));
1531
1532         rc = mdt_dlmreq_unpack(info);
1533
1534         RETURN(rc);
1535 }
1536
1537 /*
1538  * please see comment above LOV_MAGIC_V1_DEFINED
1539  */
1540 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1541 {
1542         struct lov_user_md_v1   *v1 = eadata;
1543
1544         LASSERT(v1);
1545
1546         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1547                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1548                         v1->lmm_magic |= LOV_MAGIC_DEFINED;
1549                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1550                          __swab32(LOV_MAGIC_MAGIC))
1551                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1552         }
1553 }
1554
1555 static int mdt_open_unpack(struct mdt_thread_info *info)
1556 {
1557         struct lu_ucred *uc = mdt_ucred(info);
1558         struct mdt_rec_create *rec;
1559         struct lu_attr *attr = &info->mti_attr.ma_attr;
1560         struct req_capsule *pill = info->mti_pill;
1561         struct mdt_reint_record *rr = &info->mti_rr;
1562         struct ptlrpc_request *req = mdt_info_req(info);
1563         struct md_op_spec *sp = &info->mti_spec;
1564         int rc;
1565         ENTRY;
1566
1567         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1568         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1569         if (rec == NULL)
1570                 RETURN(-EFAULT);
1571
1572         /* This prior initialization is needed for old_init_ucred_reint() */
1573         uc->uc_fsuid = rec->cr_fsuid;
1574         uc->uc_fsgid = rec->cr_fsgid;
1575         uc->uc_cap   = rec->cr_cap;
1576         uc->uc_suppgids[0] = rec->cr_suppgid1;
1577         uc->uc_suppgids[1] = rec->cr_suppgid2;
1578         uc->uc_umask = rec->cr_umask;
1579
1580         rr->rr_fid1   = &rec->cr_fid1;
1581         rr->rr_fid2   = &rec->cr_fid2;
1582         rr->rr_open_handle = &rec->cr_open_handle_old;
1583         attr->la_mode = rec->cr_mode;
1584         attr->la_rdev  = rec->cr_rdev;
1585         attr->la_uid   = rec->cr_fsuid;
1586         attr->la_gid   = rec->cr_fsgid;
1587         attr->la_ctime = rec->cr_time;
1588         attr->la_mtime = rec->cr_time;
1589         attr->la_atime = rec->cr_time;
1590         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1591                          LA_CTIME | LA_MTIME | LA_ATIME;
1592         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1593         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1594         /* Do not trigger ASSERTION if client miss to set such flags. */
1595         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1596                 RETURN(-EPROTO);
1597
1598         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1599
1600         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1601
1602         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1603                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1604                                                         RCL_CLIENT);
1605                 if (rr->rr_eadatalen > 0) {
1606                         rr->rr_eadata = req_capsule_client_get(pill,
1607                                                                &RMF_EADATA);
1608                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1609                         sp->u.sp_ea.eadata = rr->rr_eadata;
1610                         sp->no_create = !!req_is_replay(req);
1611                         mdt_fix_lov_magic(info, rr->rr_eadata);
1612                 }
1613
1614                 /*
1615                  * Client default md_size may be 0 right after client start,
1616                  * until all osc are connected, set here just some reasonable
1617                  * value to prevent misbehavior.
1618                  */
1619                 if (rr->rr_eadatalen == 0 &&
1620                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1621                         rr->rr_eadatalen = MIN_MD_SIZE;
1622         }
1623
1624         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1625                                     &sp->sp_cr_file_secctx,
1626                                     &sp->sp_cr_file_secctx_size);
1627         if (rc < 0)
1628                 RETURN(rc);
1629
1630         rc = req_check_sepol(pill);
1631         if (rc)
1632                 RETURN(rc);
1633
1634         RETURN(rc);
1635 }
1636
1637 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1638 {
1639         struct mdt_reint_record *rr = &info->mti_rr;
1640         struct lu_ucred *uc = mdt_ucred(info);
1641         struct lu_attr *attr = &info->mti_attr.ma_attr;
1642         struct req_capsule *pill = info->mti_pill;
1643         struct mdt_rec_setxattr *rec;
1644         int rc;
1645         ENTRY;
1646
1647
1648         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1649                  sizeof(struct mdt_rec_reint));
1650
1651         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1652         if (rec == NULL)
1653                 RETURN(-EFAULT);
1654
1655         /* This prior initialization is needed for old_init_ucred_reint() */
1656         uc->uc_fsuid  = rec->sx_fsuid;
1657         uc->uc_fsgid  = rec->sx_fsgid;
1658         uc->uc_cap    = rec->sx_cap;
1659         uc->uc_suppgids[0] = rec->sx_suppgid1;
1660         uc->uc_suppgids[1] = -1;
1661
1662         rr->rr_opcode = rec->sx_opcode;
1663         rr->rr_fid1   = &rec->sx_fid;
1664         attr->la_valid = rec->sx_valid;
1665         attr->la_ctime = rec->sx_time;
1666         attr->la_size = rec->sx_size;
1667         attr->la_flags = rec->sx_flags;
1668
1669         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1670         if (rc < 0)
1671                 RETURN(rc);
1672
1673         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1674                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1675                                                         RCL_CLIENT);
1676                 if (rr->rr_eadatalen > 0) {
1677                         rr->rr_eadata = req_capsule_client_get(pill,
1678                                                                &RMF_EADATA);
1679                         if (rr->rr_eadata == NULL)
1680                                 RETURN(-EFAULT);
1681                 } else {
1682                         rr->rr_eadata = NULL;
1683                 }
1684         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1685                 CDEBUG(D_INFO, "no xattr data supplied\n");
1686                 RETURN(-EFAULT);
1687         }
1688
1689         rc = req_check_sepol(pill);
1690         if (rc)
1691                 RETURN(rc);
1692
1693         if (mdt_dlmreq_unpack(info) < 0)
1694                 RETURN(-EPROTO);
1695
1696         RETURN(0);
1697 }
1698
1699 static int mdt_resync_unpack(struct mdt_thread_info *info)
1700 {
1701         struct req_capsule      *pill = info->mti_pill;
1702         struct mdt_reint_record *rr   = &info->mti_rr;
1703         struct lu_ucred         *uc     = mdt_ucred(info);
1704         struct mdt_rec_resync   *rec;
1705         ENTRY;
1706
1707         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1708         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1709         if (rec == NULL)
1710                 RETURN(-EFAULT);
1711
1712         /* This prior initialization is needed for old_init_ucred_reint() */
1713         uc->uc_fsuid = rec->rs_fsuid;
1714         uc->uc_fsgid = rec->rs_fsgid;
1715         uc->uc_cap   = rec->rs_cap;
1716
1717         rr->rr_fid1   = &rec->rs_fid;
1718         rr->rr_mirror_id = rec->rs_mirror_id;
1719
1720         /* cookie doesn't need to be swapped but it has been swapped
1721          * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1722          * restoring. */
1723         if (ptlrpc_req_need_swab(mdt_info_req(info)))
1724                 __swab64s(&rec->rs_lease_handle.cookie);
1725         rr->rr_lease_handle = &rec->rs_lease_handle;
1726
1727         RETURN(mdt_dlmreq_unpack(info));
1728 }
1729
1730 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1731
1732 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1733         [REINT_SETATTR]  = mdt_setattr_unpack,
1734         [REINT_CREATE]   = mdt_create_unpack,
1735         [REINT_LINK]     = mdt_link_unpack,
1736         [REINT_UNLINK]   = mdt_unlink_unpack,
1737         [REINT_RENAME]   = mdt_rename_unpack,
1738         [REINT_OPEN]     = mdt_open_unpack,
1739         [REINT_SETXATTR] = mdt_setxattr_unpack,
1740         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1741         [REINT_MIGRATE]  = mdt_migrate_unpack,
1742         [REINT_RESYNC]   = mdt_resync_unpack,
1743 };
1744
1745 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1746 {
1747         int rc;
1748         ENTRY;
1749
1750         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1751         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1752                 info->mti_rr.rr_opcode = op;
1753                 rc = mdt_reint_unpackers[op](info);
1754         } else {
1755                 CERROR("Unexpected opcode %d\n", op);
1756                 rc = -EFAULT;
1757         }
1758         RETURN(rc);
1759 }
1760
1761 /* check whether two FIDs belong to different MDT. */
1762 static int mdt_fids_different_target(struct mdt_thread_info *info,
1763                                      const struct lu_fid *fid1,
1764                                      const struct lu_fid *fid2)
1765 {
1766         const struct lu_env *env = info->mti_env;
1767         struct mdt_device *mdt = info->mti_mdt;
1768         struct lu_seq_range *range = &info->mti_range;
1769         struct seq_server_site *ss;
1770         __u32 index1, index2;
1771         int rc;
1772
1773         if (fid_seq(fid1) == fid_seq(fid2))
1774                 return 0;
1775
1776         ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1777
1778         range->lsr_flags = LU_SEQ_RANGE_MDT;
1779         rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1780         if (rc)
1781                 return rc;
1782
1783         index1 = range->lsr_index;
1784
1785         rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1786         if (rc)
1787                 return rc;
1788
1789         index2 = range->lsr_index;
1790
1791         return index1 != index2;
1792 }
1793
1794 /**
1795  * Check whether \a child is remote object on \a parent.
1796  *
1797  * \param[in]  info     thread environment
1798  * \param[in]  parent   parent object, it's the same as child object in
1799  *                      getattr_by_fid
1800  * \param[in]  child    child object
1801  *
1802  * \retval 1    is remote object.
1803  * \retval 0    isn't remote object.
1804  * \retval < 1  error code
1805  */
1806 int mdt_is_remote_object(struct mdt_thread_info *info,
1807                          struct mdt_object *parent,
1808                          struct mdt_object *child)
1809 {
1810         struct lu_buf *buf = &info->mti_big_buf;
1811         struct linkea_data ldata = { NULL };
1812         struct link_ea_header *leh;
1813         struct link_ea_entry *lee;
1814         struct lu_name name;
1815         struct lu_fid pfid;
1816         int reclen;
1817         int i;
1818         int rc;
1819
1820         ENTRY;
1821
1822         if (fid_is_root(mdt_object_fid(child)))
1823                 RETURN(0);
1824
1825         if (likely(parent != child)) {
1826                 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
1827                         RETURN(1);
1828
1829                 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
1830                         RETURN(0);
1831
1832                 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
1833                                                mdt_object_fid(child));
1834                 RETURN(rc);
1835         }
1836
1837         /* client < 2.13.52 getattr_by_fid parent and child are the same */
1838         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1839         if (!buf->lb_buf)
1840                 RETURN(-ENOMEM);
1841
1842         ldata.ld_buf = buf;
1843         rc = mdt_links_read(info, child, &ldata);
1844         /* can't read linkea, just assume it's remote object */
1845         if (rc == -ENOENT || rc == -ENODATA)
1846                 RETURN(1);
1847         if (rc)
1848                 RETURN(rc);
1849
1850         leh = buf->lb_buf;
1851         lee = (struct link_ea_entry *)(leh + 1);
1852         for (i = 0; i < leh->leh_reccount; i++) {
1853                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1854                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1855                 if (mdt_fids_different_target(info, &pfid,
1856                                               mdt_object_fid(child)))
1857                         RETURN(1);
1858         }
1859
1860         RETURN(0);
1861 }
1862
1863 void mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1864                               struct mdt_object *child)
1865 {
1866         char *secctx_name;
1867         struct lu_buf *buffer;
1868         struct mdt_body *repbody;
1869         struct req_capsule *pill = info->mti_pill;
1870         int rc;
1871
1872         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1873             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1874                 secctx_name =
1875                         req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1876                 buffer = &info->mti_buf;
1877
1878                 /* fill reply buffer with security context now */
1879                 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1880                                                       RCL_SERVER);
1881                 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1882                                                         &RMF_FILE_SECCTX);
1883                 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1884                                   buffer, secctx_name);
1885                 if (rc >= 0) {
1886                         CDEBUG(D_SEC,
1887                                "found security context of size %d for "DFID"\n",
1888                                rc, PFID(mdt_object_fid(child)));
1889
1890                         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1891                         repbody->mbo_valid |= OBD_MD_SECCTX;
1892                         if (rc < buffer->lb_len)
1893                                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1894                                                    RCL_SERVER);
1895                 } else {
1896                         CDEBUG(D_SEC,
1897                              "security context not found for "DFID": rc = %d\n",
1898                              PFID(mdt_object_fid(child)), rc);
1899                         req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1900                                            RCL_SERVER);
1901                 }
1902         }
1903 }