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         struct lu_nodemap *nodemap;
 152         lnet_nid_t peernid = req->rq_peer.nid;
 153         __u32 perm = 0;
 154         int setuid;
 155         int setgid;
 156         int rc = 0;
 157
 158         ENTRY;
 159
 160         LASSERT(req->rq_auth_gss);
 161         LASSERT(!req->rq_auth_usr_mdt);
 162         LASSERT(req->rq_user_desc);
 163         LASSERT(ucred != NULL);
 164
 165         ucred->uc_valid = UCRED_INVALID;
 166
 167         nodemap = nodemap_get_from_exp(info->mti_exp);
 168         if (IS_ERR(nodemap))
 169                 RETURN(PTR_ERR(nodemap));
 170
 171         pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 172                                        NODEMAP_CLIENT_TO_FS, pud->pud_uid);
 173         pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 174                                        NODEMAP_CLIENT_TO_FS, pud->pud_gid);
 175         pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 176                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
 177         pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 178                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
 179
 180         ucred->uc_o_uid = pud->pud_uid;
 181         ucred->uc_o_gid = pud->pud_gid;
 182         ucred->uc_o_fsuid = pud->pud_fsuid;
 183         ucred->uc_o_fsgid = pud->pud_fsgid;
 184
 185         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 186                 /* deny access before we get identity ref */
 187                 if (nodemap->nmf_deny_unknown) {
 188                         nodemap_putref(nodemap);
 189                         RETURN(-EACCES);
 190                 }
 191
 192                 ucred->uc_fsuid = nodemap->nm_squash_uid;
 193                 ucred->uc_fsgid = nodemap->nm_squash_gid;
 194                 ucred->uc_cap = 0;
 195                 ucred->uc_suppgids[0] = -1;
 196                 ucred->uc_suppgids[1] = -1;
 197         }
 198         nodemap_putref(nodemap);
 199
 200         if (type == BODY_INIT) {
 201                 struct mdt_body *body = (struct mdt_body *)buf;
 202
 203                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 204                 ucred->uc_suppgids[1] = -1;
 205         }
 206
 207         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 208             req->rq_auth_uid != pud->pud_uid) {
 209                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 210                        "while client claims %u:%u/%u:%u\n",
 211                        libcfs_nid2str(peernid), req->rq_auth_uid,
 212                        pud->pud_uid, pud->pud_gid,
 213                        pud->pud_fsuid, pud->pud_fsgid);
 214                 RETURN(-EACCES);
 215         }
 216
 217         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 218                 ucred->uc_identity = NULL;
 219                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 220         } else {
 221                 struct md_identity *identity;
 222
 223                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 224                                             pud->pud_uid);
 225                 if (IS_ERR(identity)) {
 226                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 227                                 ucred->uc_identity = NULL;
 228                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 229                                        CFS_SETGRP_PERM;
 230                         } else {
 231                                 CDEBUG(D_SEC,
 232                                        "Deny access without identity: uid %u\n",
 233                                        pud->pud_uid);
 234                                 RETURN(-EACCES);
 235                         }
 236                 } else {
 237                         ucred->uc_identity = identity;
 238                         perm = mdt_identity_get_perm(ucred->uc_identity,
 239                                                      peernid);
 240                 }
 241         }
 242
 243         /* find out the setuid/setgid attempt */
 244         setuid = (pud->pud_uid != pud->pud_fsuid);
 245         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 246                   (ucred->uc_identity &&
 247                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 248
 249         /* check permission of setuid */
 250         if (setuid && !(perm & CFS_SETUID_PERM)) {
 251                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 252                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 253                 GOTO(out, rc = -EACCES);
 254         }
 255
 256         /* check permission of setgid */
 257         if (setgid && !(perm & CFS_SETGID_PERM)) {
 258                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 259                        "from %s\n", pud->pud_uid, pud->pud_gid,
 260                        pud->pud_fsuid, pud->pud_fsgid,
 261                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 262                 GOTO(out, rc = -EACCES);
 263         }
 264
 265         if (perm & CFS_SETGRP_PERM) {
 266                 if (pud->pud_ngroups) {
 267                         /* setgroups for local client */
 268                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 269                         if (!ucred->uc_ginfo) {
 270                                 CERROR("failed to alloc %d groups\n",
 271                                        pud->pud_ngroups);
 272                                 GOTO(out, rc = -ENOMEM);
 273                         }
 274
 275                         lustre_groups_from_list(ucred->uc_ginfo,
 276                                                 pud->pud_groups);
 277                         lustre_groups_sort(ucred->uc_ginfo);
 278                 } else {
 279                         ucred->uc_ginfo = NULL;
 280                 }
 281         } else {
 282                 ucred->uc_suppgids[0] = -1;
 283                 ucred->uc_suppgids[1] = -1;
 284                 ucred->uc_ginfo = NULL;
 285         }
 286
 287         ucred->uc_uid = pud->pud_uid;
 288         ucred->uc_gid = pud->pud_gid;
 289         ucred->uc_fsuid = pud->pud_fsuid;
 290         ucred->uc_fsgid = pud->pud_fsgid;
 291
 292         /* process root_squash here. */
 293         mdt_root_squash(info, peernid);
 294
 295         /* remove fs privilege for non-root user. */
 296         if (ucred->uc_fsuid && drop_fs_cap)
 297                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 298         else
 299                 ucred->uc_cap = pud->pud_cap;
 300         ucred->uc_valid = UCRED_NEW;
 301         ucred_set_jobid(info, ucred);
 302
 303         EXIT;
 304
 305 out:
 306         if (rc) {
 307                 if (ucred->uc_ginfo) {
 308                         put_group_info(ucred->uc_ginfo);
 309                         ucred->uc_ginfo = NULL;
 310                 }
 311                 if (ucred->uc_identity) {
 312                         mdt_identity_put(mdt->mdt_identity_cache,
 313                                          ucred->uc_identity);
 314                         ucred->uc_identity = NULL;
 315                 }
 316         }
 317
 318         return rc;
 319 }
 320
 321 /**
 322  * Check whether allow the client to set supplementary group IDs or not.
 323  *
 324  * \param[in] info      pointer to the thread context
 325  * \param[in] uc        pointer to the RPC user descriptor
 326  *
 327  * \retval              true if allow to set supplementary group IDs
 328  * \retval              false for other cases
 329  */
 330 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 331 {
 332         __u32 perm;
 333
 334         /* 1. If identity_upcall is disabled,
 335          *    permit local client to do anything. */
 336         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 337                 return true;
 338
 339         /* 2. If fail to get related identities, then forbid any client to
 340          *    set supplementary group IDs. */
 341         if (uc->uc_identity == NULL)
 342                 return false;
 343
 344         /* 3. Check the permission in the identities. */
 345         perm = mdt_identity_get_perm(uc->uc_identity,
 346                                      mdt_info_req(info)->rq_peer.nid);
 347         if (perm & CFS_SETGRP_PERM)
 348                 return true;
 349
 350         return false;
 351 }
 352
 353 int mdt_check_ucred(struct mdt_thread_info *info)
 354 {
 355         struct ptlrpc_request   *req = mdt_info_req(info);
 356         struct mdt_device       *mdt = info->mti_mdt;
 357         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 358         struct lu_ucred         *ucred = mdt_ucred(info);
 359         struct md_identity      *identity = NULL;
 360         lnet_nid_t               peernid = req->rq_peer.nid;
 361         __u32                    perm = 0;
 362         int                      setuid;
 363         int                      setgid;
 364         int                      rc = 0;
 365
 366         ENTRY;
 367
 368         LASSERT(ucred != NULL);
 369         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 370                 RETURN(0);
 371
 372         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 373                 RETURN(0);
 374
 375         /* sanity check: if we use strong authentication, we expect the
 376          * uid which client claimed is true */
 377         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 378             req->rq_auth_uid != pud->pud_uid) {
 379                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 380                        "while client claims %u:%u/%u:%u\n",
 381                        libcfs_nid2str(peernid), req->rq_auth_uid,
 382                        pud->pud_uid, pud->pud_gid,
 383                        pud->pud_fsuid, pud->pud_fsgid);
 384                 RETURN(-EACCES);
 385         }
 386
 387         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 388                 RETURN(0);
 389
 390         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 391         if (IS_ERR(identity)) {
 392                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 393                         RETURN(0);
 394                 } else {
 395                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 396                                pud->pud_uid);
 397                         RETURN(-EACCES);
 398                 }
 399         }
 400
 401         perm = mdt_identity_get_perm(identity, peernid);
 402         /* find out the setuid/setgid attempt */
 403         setuid = (pud->pud_uid != pud->pud_fsuid);
 404         setgid = (pud->pud_gid != pud->pud_fsgid ||
 405                   pud->pud_gid != identity->mi_gid);
 406
 407         /* check permission of setuid */
 408         if (setuid && !(perm & CFS_SETUID_PERM)) {
 409                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 410                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 411                 GOTO(out, rc = -EACCES);
 412         }
 413
 414         /* check permission of setgid */
 415         if (setgid && !(perm & CFS_SETGID_PERM)) {
 416                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 417                        "from %s\n", pud->pud_uid, pud->pud_gid,
 418                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 419                        libcfs_nid2str(peernid));
 420                 GOTO(out, rc = -EACCES);
 421         }
 422
 423         EXIT;
 424
 425 out:
 426         mdt_identity_put(mdt->mdt_identity_cache, identity);
 427         return rc;
 428 }
 429
 430 static int old_init_ucred_common(struct mdt_thread_info *info,
 431                                  struct lu_nodemap *nodemap,
 432                                  bool drop_fs_cap)
 433 {
 434         struct lu_ucred         *uc = mdt_ucred(info);
 435         struct mdt_device       *mdt = info->mti_mdt;
 436         struct md_identity      *identity = NULL;
 437
 438         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 439                 /* deny access before we get identity ref */
 440                 if (nodemap->nmf_deny_unknown)
 441                         RETURN(-EACCES);
 442
 443                 uc->uc_fsuid = nodemap->nm_squash_uid;
 444                 uc->uc_fsgid = nodemap->nm_squash_gid;
 445                 uc->uc_cap = 0;
 446                 uc->uc_suppgids[0] = -1;
 447                 uc->uc_suppgids[1] = -1;
 448         }
 449
 450         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 451                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 452                                             uc->uc_fsuid);
 453                 if (IS_ERR(identity)) {
 454                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 455                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 456                                 identity = NULL;
 457                         } else {
 458                                 CDEBUG(D_SEC, "Deny access without identity: "
 459                                        "uid %u\n", uc->uc_fsuid);
 460                                 RETURN(-EACCES);
 461                         }
 462                 }
 463         }
 464         uc->uc_identity = identity;
 465
 466         /* process root_squash here. */
 467         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 468
 469         /* remove fs privilege for non-root user. */
 470         if (uc->uc_fsuid && drop_fs_cap)
 471                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 472         uc->uc_valid = UCRED_OLD;
 473         ucred_set_jobid(info, uc);
 474
 475         EXIT;
 476
 477         return 0;
 478 }
 479
 480 static int old_init_ucred(struct mdt_thread_info *info,
 481                           struct mdt_body *body, bool drop_fs_cap)
 482 {
 483         struct lu_ucred *uc = mdt_ucred(info);
 484         struct lu_nodemap *nodemap;
 485         int rc;
 486         ENTRY;
 487
 488         nodemap = nodemap_get_from_exp(info->mti_exp);
 489         if (IS_ERR(nodemap))
 490                 RETURN(PTR_ERR(nodemap));
 491
 492         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 493                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 494         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 495                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 496         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 497                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 498         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 499                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 500
 501         LASSERT(uc != NULL);
 502         uc->uc_valid = UCRED_INVALID;
 503         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 504         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 505         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 506         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 507         uc->uc_suppgids[0] = body->mbo_suppgid;
 508         uc->uc_suppgids[1] = -1;
 509         uc->uc_ginfo = NULL;
 510         uc->uc_cap = body->mbo_capability;
 511
 512         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 513         nodemap_putref(nodemap);
 514
 515         RETURN(rc);
 516 }
 517
 518 static int old_init_ucred_reint(struct mdt_thread_info *info)
 519 {
 520         struct lu_ucred *uc = mdt_ucred(info);
 521         struct lu_nodemap *nodemap;
 522         int rc;
 523         ENTRY;
 524
 525         nodemap = nodemap_get_from_exp(info->mti_exp);
 526         if (IS_ERR(nodemap))
 527                 RETURN(PTR_ERR(nodemap));
 528
 529         LASSERT(uc != NULL);
 530
 531         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 532                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 533         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 534                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 535
 536         uc->uc_valid = UCRED_INVALID;
 537         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 538         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 539         uc->uc_ginfo = NULL;
 540
 541         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 542         nodemap_putref(nodemap);
 543
 544         RETURN(rc);
 545 }
 546
 547 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 548                                    struct mdt_body *body,
 549                                    bool drop_fs_cap)
 550 {
 551         struct ptlrpc_request   *req = mdt_info_req(info);
 552         struct lu_ucred         *uc  = mdt_ucred(info);
 553
 554         LASSERT(uc != NULL);
 555         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 556                 return 0;
 557
 558         mdt_exit_ucred(info);
 559
 560         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 561                 return old_init_ucred(info, body, drop_fs_cap);
 562         else
 563                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 564 }
 565
 566 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 567 {
 568         return __mdt_init_ucred(info, body, true);
 569 }
 570
 571 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 572  * doesn't set correct fsuid explicitely, but raise capability to allow
 573  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 574  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 575  * will fail permission check. */
 576 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 577                                   struct mdt_body *body)
 578 {
 579         return __mdt_init_ucred(info, body, false);
 580 }
 581
 582 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 583 {
 584         struct ptlrpc_request *req = mdt_info_req(info);
 585         struct lu_ucred       *uc  = mdt_ucred(info);
 586         struct md_attr        *ma  = &info->mti_attr;
 587
 588         LASSERT(uc != NULL);
 589         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 590                 return 0;
 591
 592         /* LU-5564: for normal close request, skip permission check */
 593         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 594             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 595                 uc->uc_cap |= CFS_CAP_FS_MASK;
 596
 597         mdt_exit_ucred(info);
 598
 599         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 600                 return old_init_ucred_reint(info);
 601         else
 602                 return new_init_ucred(info, REC_INIT, NULL, true);
 603 }
 604
 605 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 606 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 607 {
 608         const struct lov_ost_data_v1    *lod;
 609         int                              i;
 610         __u16                            count;
 611
 612         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 613                 return;
 614
 615         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 616
 617         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 618                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 619                le32_to_cpu(lmm->lmm_pattern));
 620         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 621                le32_to_cpu(lmm->lmm_stripe_size), count);
 622
 623         /* If it's a directory or a released file, then there are
 624          * no actual objects to print, so bail out. */
 625         if (valid & OBD_MD_FLDIREA ||
 626             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 627                 return;
 628
 629         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 630         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 631                 struct ost_id oi;
 632
 633                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 634                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 635                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 636         }
 637 }
 638
 639 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 640 {
 641         const struct lmv_mds_md_v1 *lmm1;
 642         int                        i;
 643
 644         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 645                 return;
 646
 647         lmm1 = &lmv->lmv_md_v1;
 648         CDEBUG(level,
 649                "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
 650                le32_to_cpu(lmm1->lmv_magic),
 651                le32_to_cpu(lmm1->lmv_master_mdt_index),
 652                le32_to_cpu(lmm1->lmv_stripe_count),
 653                le32_to_cpu(lmm1->lmv_hash_type));
 654
 655         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 656                 return;
 657
 658         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 659                 struct lu_fid fid;
 660
 661                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 662                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 663         }
 664 }
 665
 666 /* Shrink and/or grow reply buffers */
 667 int mdt_fix_reply(struct mdt_thread_info *info)
 668 {
 669         struct req_capsule *pill = info->mti_pill;
 670         struct mdt_body    *body;
 671         int                md_size, md_packed = 0;
 672         int                acl_size;
 673         int                rc = 0;
 674         ENTRY;
 675
 676         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 677         LASSERT(body != NULL);
 678
 679         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 680                                OBD_MD_LINKNAME))
 681                 md_size = body->mbo_eadatasize;
 682         else
 683                 md_size = 0;
 684
 685         acl_size = body->mbo_aclsize;
 686
 687         /* this replay - not send info to client */
 688         if (info->mti_spec.no_create) {
 689                 md_size = 0;
 690                 acl_size = 0;
 691         }
 692
 693         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 694                md_size, acl_size);
 695 /*
 696             &RMF_MDT_BODY,
 697             &RMF_MDT_MD,
 698             &RMF_ACL, or &RMF_LOGCOOKIES
 699 (optional)  &RMF_CAPA1,
 700 (optional)  &RMF_CAPA2,
 701 (optional)  something else
 702 */
 703
 704         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 705          * buffers before growing it */
 706         if (info->mti_big_lmm_used) {
 707                 /* big_lmm buffer may be used even without packing the result
 708                  * into reply, just for internal server needs */
 709                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 710                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 711                                                          RCL_SERVER);
 712
 713                 /* free big lmm if md_size is not needed */
 714                 if (md_size == 0 || md_packed == 0) {
 715                         info->mti_big_lmm_used = 0;
 716                 } else {
 717                         /* buffer must be allocated separately */
 718                         LASSERT(info->mti_attr.ma_lmm !=
 719                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 720                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 721                 }
 722         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 723                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 724         }
 725
 726         if (info->mti_big_acl_used) {
 727                 if (acl_size == 0)
 728                         info->mti_big_acl_used = 0;
 729                 else
 730                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 731         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 732                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 733         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 734                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 735         }
 736
 737         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 738             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 739                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 740
 741         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 742             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 743                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 744
 745         /*
 746          * Some more field should be shrinked if needed.
 747          * This should be done by those who added fields to reply message.
 748          */
 749
 750         /* Grow MD buffer if needed finally */
 751         if (info->mti_big_lmm_used) {
 752                 void *lmm;
 753
 754                 LASSERT(md_size > md_packed);
 755                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 756                        md_size - md_packed);
 757                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 758                 if (rc) {
 759                         /* we can't answer with proper LOV EA, drop flags,
 760                          * the rc is also returned so this request is
 761                          * considered as failed */
 762                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 763                         /* don't return transno along with error */
 764                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 765                 } else {
 766                         /* now we need to pack right LOV/LMV EA */
 767                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 768                         if (info->mti_attr.ma_valid & MA_LOV) {
 769                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 770                                                              RCL_SERVER) ==
 771                                                 info->mti_attr.ma_lmm_size);
 772                                 memcpy(lmm, info->mti_attr.ma_lmm,
 773                                        info->mti_attr.ma_lmm_size);
 774                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 775                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 776                                                              RCL_SERVER) ==
 777                                                 info->mti_attr.ma_lmv_size);
 778                                 memcpy(lmm, info->mti_attr.ma_lmv,
 779                                        info->mti_attr.ma_lmv_size);
 780                         }
 781                 }
 782
 783                 /* update mdt_max_mdsize so clients will be aware about that */
 784                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 785                         info->mti_mdt->mdt_max_mdsize =
 786                                                 info->mti_attr.ma_lmm_size;
 787                 info->mti_big_lmm_used = 0;
 788         }
 789
 790         if (info->mti_big_acl_used) {
 791                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 792                        acl_size);
 793
 794                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 795                 if (rc) {
 796                         body->mbo_valid &= ~OBD_MD_FLACL;
 797                 } else {
 798                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 799
 800                         memcpy(acl, info->mti_big_acl, acl_size);
 801                 }
 802
 803                 info->mti_big_acl_used = 0;
 804         }
 805
 806         RETURN(rc);
 807 }
 808
 809
 810 /* if object is dying, pack the lov/llog data,
 811  * parameter info->mti_attr should be valid at this point!
 812  * Also implements RAoLU policy */
 813 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 814                            struct md_attr *ma)
 815 {
 816         struct mdt_body *repbody = NULL;
 817         const struct lu_attr *la = &ma->ma_attr;
 818         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 819         int rc;
 820         __u64 need = 0;
 821         struct hsm_action_item hai = {
 822                 .hai_len = sizeof(hai),
 823                 .hai_action = HSMA_REMOVE,
 824                 .hai_extent.length = -1,
 825                 .hai_cookie = 0,
 826                 .hai_gid = 0,
 827         };
 828         __u64 compound_id;
 829         int archive_id;
 830
 831         ENTRY;
 832
 833         if (mdt_info_req(info) != NULL) {
 834                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 835                 LASSERT(repbody != NULL);
 836         } else {
 837                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 838         }
 839
 840         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 841                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 842
 843         if (ma->ma_valid & MA_LOV) {
 844                 CERROR("No need in LOV EA upon unlink\n");
 845                 dump_stack();
 846         }
 847         if (repbody != NULL)
 848                 repbody->mbo_eadatasize = 0;
 849
 850         /* Only check unlinked and archived if RAoLU and upon last close */
 851         if (!cdt->cdt_remove_archive_on_last_unlink ||
 852             atomic_read(&mo->mot_open_count) != 0)
 853                 RETURN(0);
 854
 855         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 856         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 857                 RETURN(0);
 858
 859         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 860                 RETURN(0);
 861
 862         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 863         if (need != 0) {
 864                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 865                  * try setting them */
 866                 ma->ma_need |= need;
 867                 rc = mdt_attr_get_complex(info, mo, ma);
 868                 if (rc) {
 869                         CERROR("%s: unable to fetch missing attributes of"
 870                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 871                                PFID(mdt_object_fid(mo)), rc);
 872                         RETURN(0);
 873                 }
 874
 875                 if (need & MA_INODE) {
 876                         if (ma->ma_valid & MA_INODE) {
 877                                 if (ma->ma_attr.la_nlink != 0)
 878                                         RETURN(0);
 879                         } else {
 880                                 RETURN(0);
 881                         }
 882                 }
 883
 884                 if (need & MA_HSM) {
 885                         if (ma->ma_valid & MA_HSM) {
 886                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 887                                         RETURN(0);
 888                         } else {
 889                                 RETURN(0);
 890                         }
 891                 }
 892         }
 893
 894         /* RAoLU policy is active, last close on file has occured,
 895          * file is unlinked, file is archived, so create remove request
 896          * for copytool!
 897          * If CDT is not running, requests will be logged for later. */
 898         compound_id = atomic_inc_return(&cdt->cdt_compound_id);
 899         if (ma->ma_hsm.mh_arch_id != 0)
 900                 archive_id = ma->ma_hsm.mh_arch_id;
 901         else
 902                 archive_id = cdt->cdt_default_archive_id;
 903
 904         hai.hai_fid = *mdt_object_fid(mo);
 905
 906         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt,
 907                                   compound_id, archive_id, 0, &hai);
 908         if (rc)
 909                 CERROR("%s: unable to add HSM remove request for "DFID
 910                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
 911                        PFID(mdt_object_fid(mo)), rc);
 912
 913         RETURN(0);
 914 }
 915
 916 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 917                                   struct md_attr *ma)
 918 {
 919         __u64 out;
 920
 921         out = 0;
 922         if (in & MDS_ATTR_MODE)
 923                 out |= LA_MODE;
 924         if (in & MDS_ATTR_UID)
 925                 out |= LA_UID;
 926         if (in & MDS_ATTR_GID)
 927                 out |= LA_GID;
 928         if (in & MDS_ATTR_SIZE)
 929                 out |= LA_SIZE;
 930         if (in & MDS_ATTR_BLOCKS)
 931                 out |= LA_BLOCKS;
 932         if (in & MDS_ATTR_ATIME_SET)
 933                 out |= LA_ATIME;
 934         if (in & MDS_ATTR_CTIME_SET)
 935                 out |= LA_CTIME;
 936         if (in & MDS_ATTR_MTIME_SET)
 937                 out |= LA_MTIME;
 938         if (in & MDS_ATTR_ATTR_FLAG)
 939                 out |= LA_FLAGS;
 940         if (in & MDS_ATTR_KILL_SUID)
 941                 out |= LA_KILL_SUID;
 942         if (in & MDS_ATTR_KILL_SGID)
 943                 out |= LA_KILL_SGID;
 944         if (in & MDS_ATTR_PROJID)
 945                 out |= LA_PROJID;
 946
 947         if (in & MDS_ATTR_FROM_OPEN)
 948                 rr->rr_flags |= MRF_OPEN_TRUNC;
 949         if (in & MDS_OPEN_OWNEROVERRIDE)
 950                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 951         if (in & MDS_ATTR_FORCE)
 952                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 953
 954         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
 955                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 956                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 957                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 958                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 959                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 960         if (in != 0)
 961                 CERROR("Unknown attr bits: %#llx\n", in);
 962         return out;
 963 }
 964
 965 /* unpacking */
 966
 967 int mdt_name_unpack(struct req_capsule *pill,
 968                     const struct req_msg_field *field,
 969                     struct lu_name *ln,
 970                     enum mdt_name_flags flags)
 971 {
 972         ln->ln_name = req_capsule_client_get(pill, field);
 973         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 974
 975         if (!lu_name_is_valid(ln)) {
 976                 ln->ln_name = NULL;
 977                 ln->ln_namelen = 0;
 978
 979                 return -EPROTO;
 980         }
 981
 982         if ((flags & MNF_FIX_ANON) &&
 983             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 984                 /* Newer (3.x) kernels use a name of "/" for the
 985                  * "anonymous" disconnected dentries from NFS
 986                  * filehandle conversion. See d_obtain_alias(). */
 987                 ln->ln_name = NULL;
 988                 ln->ln_namelen = 0;
 989         }
 990
 991         return 0;
 992 }
 993
 994 static int mdt_file_secctx_unpack(struct req_capsule *pill,
 995                                   const char **secctx_name,
 996                                   void **secctx, size_t *secctx_size)
 997 {
 998         const char *name;
 999         size_t name_size;
1000
1001         *secctx_name = NULL;
1002         *secctx = NULL;
1003         *secctx_size = 0;
1004
1005         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1006             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1007                 return 0;
1008
1009         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1010                                          RCL_CLIENT);
1011         if (name_size == 0)
1012                 return 0;
1013
1014         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1015         if (strnlen(name, name_size) != name_size - 1)
1016                 return -EPROTO;
1017
1018         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1019             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1020                 return -EPROTO;
1021
1022         *secctx_name = name;
1023         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1024         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1025
1026         return 0;
1027 }
1028
1029 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1030 {
1031         struct lu_ucred         *uc = mdt_ucred(info);
1032         struct md_attr          *ma = &info->mti_attr;
1033         struct lu_attr          *la = &ma->ma_attr;
1034         struct req_capsule      *pill = info->mti_pill;
1035         struct mdt_reint_record *rr = &info->mti_rr;
1036         struct mdt_rec_setattr  *rec;
1037         struct lu_nodemap       *nodemap;
1038         ENTRY;
1039
1040         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
1041         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1042         if (rec == NULL)
1043                 RETURN(-EFAULT);
1044
1045         /* This prior initialization is needed for old_init_ucred_reint() */
1046         uc->uc_fsuid = rec->sa_fsuid;
1047         uc->uc_fsgid = rec->sa_fsgid;
1048         uc->uc_cap   = rec->sa_cap;
1049         uc->uc_suppgids[0] = rec->sa_suppgid;
1050         uc->uc_suppgids[1] = -1;
1051
1052         rr->rr_fid1 = &rec->sa_fid;
1053         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1054         la->la_mode  = rec->sa_mode;
1055         la->la_flags = rec->sa_attr_flags;
1056
1057         nodemap = nodemap_get_from_exp(info->mti_exp);
1058         if (IS_ERR(nodemap))
1059                 RETURN(PTR_ERR(nodemap));
1060
1061         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1062                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1063         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1064                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1065         la->la_projid = rec->sa_projid;
1066         nodemap_putref(nodemap);
1067
1068         la->la_size  = rec->sa_size;
1069         la->la_blocks = rec->sa_blocks;
1070         la->la_ctime = rec->sa_ctime;
1071         la->la_atime = rec->sa_atime;
1072         la->la_mtime = rec->sa_mtime;
1073         ma->ma_valid = MA_INODE;
1074
1075         if (rec->sa_bias & MDS_DATA_MODIFIED)
1076                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
1077         else
1078                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
1079
1080         if (rec->sa_bias & MDS_HSM_RELEASE)
1081                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
1082         else
1083                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
1084
1085         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
1086                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
1087         else
1088                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
1089
1090         RETURN(0);
1091 }
1092
1093 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1094 {
1095         struct req_capsule *pill = info->mti_pill;
1096         struct mdt_ioepoch *ioepoch;
1097         ENTRY;
1098
1099         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1100                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1101         else
1102                 ioepoch = NULL;
1103
1104         if (ioepoch == NULL)
1105                 RETURN(-EPROTO);
1106
1107         info->mti_close_handle = ioepoch->mio_handle;
1108
1109         RETURN(0);
1110 }
1111
1112 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1113         struct req_capsule      *pill = info->mti_pill;
1114
1115         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1116                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1117                 if (info->mti_dlm_req == NULL)
1118                         RETURN(-EFAULT);
1119         }
1120
1121         RETURN(0);
1122 }
1123
1124 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1125 {
1126         struct mdt_reint_record *rr = &info->mti_rr;
1127         struct md_attr          *ma = &info->mti_attr;
1128         struct req_capsule      *pill = info->mti_pill;
1129         int rc;
1130         ENTRY;
1131
1132         rc = mdt_setattr_unpack_rec(info);
1133         if (rc)
1134                 RETURN(rc);
1135
1136         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1137                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1138                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1139                                                         RCL_CLIENT);
1140                 if (rr->rr_eadatalen > 0) {
1141                         const struct lmv_user_md        *lum;
1142
1143                         lum = rr->rr_eadata;
1144                         /* Sigh ma_valid(from req) does not indicate whether
1145                          * it will set LOV/LMV EA, so we have to check magic */
1146                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1147                                 ma->ma_valid |= MA_LMV;
1148                                 ma->ma_lmv = (void *)rr->rr_eadata;
1149                                 ma->ma_lmv_size = rr->rr_eadatalen;
1150                         } else {
1151                                 ma->ma_valid |= MA_LOV;
1152                                 ma->ma_lmm = (void *)rr->rr_eadata;
1153                                 ma->ma_lmm_size = rr->rr_eadatalen;
1154                         }
1155                 }
1156         }
1157
1158         rc = mdt_dlmreq_unpack(info);
1159         RETURN(rc);
1160 }
1161
1162 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1163 {
1164         struct md_attr          *ma = &info->mti_attr;
1165         struct req_capsule      *pill = info->mti_pill;
1166         ENTRY;
1167
1168         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1169                 RETURN(0);
1170
1171         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1172
1173         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1174             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1175                 RETURN(-EFAULT);
1176
1177         RETURN(0);
1178 }
1179
1180 int mdt_close_unpack(struct mdt_thread_info *info)
1181 {
1182         int rc;
1183         ENTRY;
1184
1185         rc = mdt_close_handle_unpack(info);
1186         if (rc)
1187                 RETURN(rc);
1188
1189         rc = mdt_setattr_unpack_rec(info);
1190         if (rc)
1191                 RETURN(rc);
1192
1193         rc = mdt_intent_close_unpack(info);
1194         if (rc)
1195                 RETURN(rc);
1196
1197         RETURN(mdt_init_ucred_reint(info));
1198 }
1199
1200 static int mdt_create_unpack(struct mdt_thread_info *info)
1201 {
1202         struct lu_ucred         *uc  = mdt_ucred(info);
1203         struct mdt_rec_create   *rec;
1204         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1205         struct mdt_reint_record *rr = &info->mti_rr;
1206         struct req_capsule      *pill = info->mti_pill;
1207         struct md_op_spec       *sp = &info->mti_spec;
1208         int rc;
1209         ENTRY;
1210
1211         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1212         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1213         if (rec == NULL)
1214                 RETURN(-EFAULT);
1215
1216         /* This prior initialization is needed for old_init_ucred_reint() */
1217         uc->uc_fsuid = rec->cr_fsuid;
1218         uc->uc_fsgid = rec->cr_fsgid;
1219         uc->uc_cap   = rec->cr_cap;
1220         uc->uc_suppgids[0] = rec->cr_suppgid1;
1221         uc->uc_suppgids[1] = -1;
1222         uc->uc_umask = rec->cr_umask;
1223
1224         rr->rr_fid1 = &rec->cr_fid1;
1225         rr->rr_fid2 = &rec->cr_fid2;
1226         attr->la_mode = rec->cr_mode;
1227         attr->la_rdev  = rec->cr_rdev;
1228         attr->la_uid   = rec->cr_fsuid;
1229         attr->la_gid   = rec->cr_fsgid;
1230         attr->la_ctime = rec->cr_time;
1231         attr->la_mtime = rec->cr_time;
1232         attr->la_atime = rec->cr_time;
1233         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1234                          LA_CTIME | LA_MTIME | LA_ATIME;
1235         memset(&sp->u, 0, sizeof(sp->u));
1236         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1237
1238         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1239         if (rc < 0)
1240                 RETURN(rc);
1241
1242         if (S_ISLNK(attr->la_mode)) {
1243                 const char *tgt = NULL;
1244
1245                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1246                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1247                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1248                         sp->u.sp_symname = tgt;
1249                 }
1250                 if (tgt == NULL)
1251                         RETURN(-EFAULT);
1252         } else {
1253                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1254                 if (S_ISDIR(attr->la_mode) &&
1255                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1256                         sp->u.sp_ea.eadata =
1257                                 req_capsule_client_get(pill, &RMF_EADATA);
1258                         sp->u.sp_ea.eadatalen =
1259                                 req_capsule_get_size(pill, &RMF_EADATA,
1260                                                      RCL_CLIENT);
1261                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1262                 }
1263         }
1264
1265         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1266                                     &sp->sp_cr_file_secctx,
1267                                     &sp->sp_cr_file_secctx_size);
1268         if (rc < 0)
1269                 RETURN(rc);
1270
1271         rc = mdt_dlmreq_unpack(info);
1272         RETURN(rc);
1273 }
1274
1275 static int mdt_link_unpack(struct mdt_thread_info *info)
1276 {
1277         struct lu_ucred         *uc  = mdt_ucred(info);
1278         struct mdt_rec_link     *rec;
1279         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1280         struct mdt_reint_record *rr = &info->mti_rr;
1281         struct req_capsule      *pill = info->mti_pill;
1282         int rc;
1283         ENTRY;
1284
1285         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1286         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1287         if (rec == NULL)
1288                 RETURN(-EFAULT);
1289
1290         /* This prior initialization is needed for old_init_ucred_reint() */
1291         uc->uc_fsuid = rec->lk_fsuid;
1292         uc->uc_fsgid = rec->lk_fsgid;
1293         uc->uc_cap   = rec->lk_cap;
1294         uc->uc_suppgids[0] = rec->lk_suppgid1;
1295         uc->uc_suppgids[1] = rec->lk_suppgid2;
1296
1297         attr->la_uid = rec->lk_fsuid;
1298         attr->la_gid = rec->lk_fsgid;
1299         rr->rr_fid1 = &rec->lk_fid1;
1300         rr->rr_fid2 = &rec->lk_fid2;
1301         attr->la_ctime = rec->lk_time;
1302         attr->la_mtime = rec->lk_time;
1303         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1304
1305         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1306         if (rc < 0)
1307                 RETURN(rc);
1308
1309         rc = mdt_dlmreq_unpack(info);
1310
1311         RETURN(rc);
1312 }
1313
1314 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1315 {
1316         struct lu_ucred         *uc  = mdt_ucred(info);
1317         struct mdt_rec_unlink   *rec;
1318         struct md_attr          *ma = &info->mti_attr;
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         ENTRY;
1324
1325         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1326         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1327         if (rec == NULL)
1328                 RETURN(-EFAULT);
1329
1330         /* This prior initialization is needed for old_init_ucred_reint() */
1331         uc->uc_fsuid = rec->ul_fsuid;
1332         uc->uc_fsgid = rec->ul_fsgid;
1333         uc->uc_cap   = rec->ul_cap;
1334         uc->uc_suppgids[0] = rec->ul_suppgid1;
1335         uc->uc_suppgids[1] = -1;
1336
1337         attr->la_uid = rec->ul_fsuid;
1338         attr->la_gid = rec->ul_fsgid;
1339         rr->rr_fid1 = &rec->ul_fid1;
1340         rr->rr_fid2 = &rec->ul_fid2;
1341         attr->la_ctime = rec->ul_time;
1342         attr->la_mtime = rec->ul_time;
1343         attr->la_mode  = rec->ul_mode;
1344         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1345
1346         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1347         if (rc < 0)
1348                 RETURN(rc);
1349
1350         if (rec->ul_bias & MDS_VTX_BYPASS)
1351                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1352         else
1353                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1354
1355         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1356
1357         rc = mdt_dlmreq_unpack(info);
1358         RETURN(rc);
1359 }
1360
1361 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1362 {
1363         info->mti_spec.sp_rm_entry = 1;
1364         return mdt_unlink_unpack(info);
1365 }
1366
1367 static int mdt_rename_unpack(struct mdt_thread_info *info)
1368 {
1369         struct lu_ucred         *uc = mdt_ucred(info);
1370         struct mdt_rec_rename   *rec;
1371         struct md_attr          *ma = &info->mti_attr;
1372         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1373         struct mdt_reint_record *rr = &info->mti_rr;
1374         struct req_capsule      *pill = info->mti_pill;
1375         int rc;
1376         ENTRY;
1377
1378         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1379         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1380         if (rec == NULL)
1381                 RETURN(-EFAULT);
1382
1383         /* This prior initialization is needed for old_init_ucred_reint() */
1384         uc->uc_fsuid = rec->rn_fsuid;
1385         uc->uc_fsgid = rec->rn_fsgid;
1386         uc->uc_cap   = rec->rn_cap;
1387         uc->uc_suppgids[0] = rec->rn_suppgid1;
1388         uc->uc_suppgids[1] = rec->rn_suppgid2;
1389
1390         attr->la_uid = rec->rn_fsuid;
1391         attr->la_gid = rec->rn_fsgid;
1392         rr->rr_fid1 = &rec->rn_fid1;
1393         rr->rr_fid2 = &rec->rn_fid2;
1394         attr->la_ctime = rec->rn_time;
1395         attr->la_mtime = rec->rn_time;
1396         /* rename_tgt contains the mode already */
1397         attr->la_mode = rec->rn_mode;
1398         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1399
1400         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1401         if (rc < 0)
1402                 RETURN(rc);
1403
1404         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1405         if (rc < 0)
1406                 RETURN(rc);
1407
1408         if (rec->rn_bias & MDS_VTX_BYPASS)
1409                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1410         else
1411                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1412
1413         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1414                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1415                 rc = mdt_close_handle_unpack(info);
1416                 if (rc < 0)
1417                         RETURN(rc);
1418                 info->mti_spec.sp_migrate_close = 1;
1419         }
1420
1421         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1422
1423
1424         rc = mdt_dlmreq_unpack(info);
1425
1426         RETURN(rc);
1427 }
1428
1429 /*
1430  * please see comment above LOV_MAGIC_V1_DEF
1431  */
1432 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1433 {
1434         struct lov_user_md_v1   *v1 = eadata;
1435
1436         LASSERT(v1);
1437
1438         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1439                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1440                         v1->lmm_magic |= LOV_MAGIC_DEF;
1441                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1442                          __swab32(LOV_MAGIC_MAGIC))
1443                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEF);
1444         }
1445 }
1446
1447 static int mdt_open_unpack(struct mdt_thread_info *info)
1448 {
1449         struct lu_ucred         *uc = mdt_ucred(info);
1450         struct mdt_rec_create   *rec;
1451         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1452         struct req_capsule      *pill = info->mti_pill;
1453         struct mdt_reint_record *rr   = &info->mti_rr;
1454         struct ptlrpc_request   *req  = mdt_info_req(info);
1455         struct md_op_spec       *sp   = &info->mti_spec;
1456         int rc;
1457         ENTRY;
1458
1459         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1460         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1461         if (rec == NULL)
1462                 RETURN(-EFAULT);
1463
1464         /* This prior initialization is needed for old_init_ucred_reint() */
1465         uc->uc_fsuid = rec->cr_fsuid;
1466         uc->uc_fsgid = rec->cr_fsgid;
1467         uc->uc_cap   = rec->cr_cap;
1468         uc->uc_suppgids[0] = rec->cr_suppgid1;
1469         uc->uc_suppgids[1] = rec->cr_suppgid2;
1470         uc->uc_umask = rec->cr_umask;
1471
1472         rr->rr_fid1   = &rec->cr_fid1;
1473         rr->rr_fid2   = &rec->cr_fid2;
1474         rr->rr_handle = &rec->cr_old_handle;
1475         attr->la_mode = rec->cr_mode;
1476         attr->la_rdev  = rec->cr_rdev;
1477         attr->la_uid   = rec->cr_fsuid;
1478         attr->la_gid   = rec->cr_fsgid;
1479         attr->la_ctime = rec->cr_time;
1480         attr->la_mtime = rec->cr_time;
1481         attr->la_atime = rec->cr_time;
1482         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1483                          LA_CTIME | LA_MTIME | LA_ATIME;
1484         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1485         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1486         /* Do not trigger ASSERTION if client miss to set such flags. */
1487         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1488                 RETURN(-EPROTO);
1489
1490         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1491
1492         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1493
1494         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1495                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1496                                                         RCL_CLIENT);
1497                 if (rr->rr_eadatalen > 0) {
1498                         rr->rr_eadata = req_capsule_client_get(pill,
1499                                                                &RMF_EADATA);
1500                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1501                         sp->u.sp_ea.eadata = rr->rr_eadata;
1502                         sp->no_create = !!req_is_replay(req);
1503                         mdt_fix_lov_magic(info, rr->rr_eadata);
1504                 }
1505
1506                 /*
1507                  * Client default md_size may be 0 right after client start,
1508                  * until all osc are connected, set here just some reasonable
1509                  * value to prevent misbehavior.
1510                  */
1511                 if (rr->rr_eadatalen == 0 &&
1512                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1513                         rr->rr_eadatalen = MIN_MD_SIZE;
1514         }
1515
1516         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1517                                     &sp->sp_cr_file_secctx,
1518                                     &sp->sp_cr_file_secctx_size);
1519
1520         RETURN(rc);
1521 }
1522
1523 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1524 {
1525         struct mdt_reint_record *rr     = &info->mti_rr;
1526         struct lu_ucred         *uc     = mdt_ucred(info);
1527         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1528         struct req_capsule      *pill   = info->mti_pill;
1529         struct mdt_rec_setxattr *rec;
1530         int                      rc;
1531         ENTRY;
1532
1533
1534         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1535                          sizeof(struct mdt_rec_reint));
1536
1537         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1538         if (rec == NULL)
1539                 RETURN(-EFAULT);
1540
1541         /* This prior initialization is needed for old_init_ucred_reint() */
1542         uc->uc_fsuid  = rec->sx_fsuid;
1543         uc->uc_fsgid  = rec->sx_fsgid;
1544         uc->uc_cap    = rec->sx_cap;
1545         uc->uc_suppgids[0] = rec->sx_suppgid1;
1546         uc->uc_suppgids[1] = -1;
1547
1548         rr->rr_opcode = rec->sx_opcode;
1549         rr->rr_fid1   = &rec->sx_fid;
1550         attr->la_valid = rec->sx_valid;
1551         attr->la_ctime = rec->sx_time;
1552         attr->la_size = rec->sx_size;
1553         attr->la_flags = rec->sx_flags;
1554
1555         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1556         if (rc < 0)
1557                 RETURN(rc);
1558
1559         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1560                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1561                                                         RCL_CLIENT);
1562                 if (rr->rr_eadatalen > 0) {
1563                         rr->rr_eadata = req_capsule_client_get(pill,
1564                                                                &RMF_EADATA);
1565                         if (rr->rr_eadata == NULL)
1566                                 RETURN(-EFAULT);
1567                 } else {
1568                         rr->rr_eadata = NULL;
1569                 }
1570         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1571                 CDEBUG(D_INFO, "no xattr data supplied\n");
1572                 RETURN(-EFAULT);
1573         }
1574
1575         if (mdt_dlmreq_unpack(info) < 0)
1576                 RETURN(-EPROTO);
1577
1578         RETURN(0);
1579 }
1580
1581
1582 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1583
1584 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1585         [REINT_SETATTR]  = mdt_setattr_unpack,
1586         [REINT_CREATE]   = mdt_create_unpack,
1587         [REINT_LINK]     = mdt_link_unpack,
1588         [REINT_UNLINK]   = mdt_unlink_unpack,
1589         [REINT_RENAME]   = mdt_rename_unpack,
1590         [REINT_OPEN]     = mdt_open_unpack,
1591         [REINT_SETXATTR] = mdt_setxattr_unpack,
1592         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1593         [REINT_MIGRATE]  = mdt_rename_unpack,
1594 };
1595
1596 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1597 {
1598         int rc;
1599         ENTRY;
1600
1601         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1602         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1603                 info->mti_rr.rr_opcode = op;
1604                 rc = mdt_reint_unpackers[op](info);
1605         } else {
1606                 CERROR("Unexpected opcode %d\n", op);
1607                 rc = -EFAULT;
1608         }
1609         RETURN(rc);
1610 }