lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_lib.c
  33  *
  34  * Lustre Metadata Target (mdt) request unpacking helper.
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Fan Yong <fanyong@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include "mdt_internal.h"
  52 #include <lnet/nidstr.h>
  53 #include <lustre_nodemap.h>
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
  62 {
  63         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
  64 }
  65
  66 void mdt_exit_ucred(struct mdt_thread_info *info)
  67 {
  68         struct lu_ucred   *uc  = mdt_ucred(info);
  69         struct mdt_device *mdt = info->mti_mdt;
  70
  71         LASSERT(uc != NULL);
  72         if (uc->uc_valid != UCRED_INIT) {
  73                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  74                 if (uc->uc_ginfo) {
  75                         put_group_info(uc->uc_ginfo);
  76                         uc->uc_ginfo = NULL;
  77                 }
  78                 if (uc->uc_identity) {
  79                         mdt_identity_put(mdt->mdt_identity_cache,
  80                                          uc->uc_identity);
  81                         uc->uc_identity = NULL;
  82                 }
  83                 uc->uc_valid = UCRED_INIT;
  84         }
  85 }
  86
  87 static int match_nosquash_list(struct rw_semaphore *sem,
  88                                struct list_head *nidlist,
  89                                lnet_nid_t peernid)
  90 {
  91         int rc;
  92         ENTRY;
  93         down_read(sem);
  94         rc = cfs_match_nid(peernid, nidlist);
  95         up_read(sem);
  96         RETURN(rc);
  97 }
  98
  99 /* root_squash for inter-MDS operations */
 100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 101 {
 102         struct lu_ucred *ucred = mdt_ucred(info);
 103         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 104         ENTRY;
 105
 106         LASSERT(ucred != NULL);
 107         if (!squash->rsi_uid || ucred->uc_fsuid)
 108                 RETURN(0);
 109
 110         if (match_nosquash_list(&squash->rsi_sem,
 111                                 &squash->rsi_nosquash_nids,
 112                                 peernid)) {
 113                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 114                        libcfs_nid2str(peernid));
 115                 RETURN(0);
 116         }
 117
 118         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 119                libcfs_nid2str(peernid),
 120                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 121                squash->rsi_uid, squash->rsi_gid, 0);
 122
 123         ucred->uc_fsuid = squash->rsi_uid;
 124         ucred->uc_fsgid = squash->rsi_gid;
 125         ucred->uc_cap = 0;
 126         ucred->uc_suppgids[0] = -1;
 127         ucred->uc_suppgids[1] = -1;
 128
 129         RETURN(0);
 130 }
 131
 132 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 133 {
 134         struct ptlrpc_request   *req = mdt_info_req(info);
 135         const char              *jobid = mdt_req_get_jobid(req);
 136
 137         /* set jobid if specified. */
 138         if (jobid)
 139                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 140         else
 141                 uc->uc_jobid[0] = '\0';
 142 }
 143
 144 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 145                           void *buf, bool drop_fs_cap)
 146 {
 147         struct ptlrpc_request   *req = mdt_info_req(info);
 148         struct mdt_device       *mdt = info->mti_mdt;
 149         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 150         struct lu_ucred         *ucred = mdt_ucred(info);
 151         lnet_nid_t               peernid = req->rq_peer.nid;
 152         __u32                    perm = 0;
 153         int                      setuid;
 154         int                      setgid;
 155         int                      rc = 0;
 156
 157         ENTRY;
 158
 159         LASSERT(req->rq_auth_gss);
 160         LASSERT(!req->rq_auth_usr_mdt);
 161         LASSERT(req->rq_user_desc);
 162         LASSERT(ucred != NULL);
 163
 164         ucred->uc_valid = UCRED_INVALID;
 165
 166         ucred->uc_o_uid   = pud->pud_uid;
 167         ucred->uc_o_gid   = pud->pud_gid;
 168         ucred->uc_o_fsuid = pud->pud_fsuid;
 169         ucred->uc_o_fsgid = pud->pud_fsgid;
 170
 171         if (type == BODY_INIT) {
 172                 struct mdt_body *body = (struct mdt_body *)buf;
 173
 174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 175                 ucred->uc_suppgids[1] = -1;
 176         }
 177
 178         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 179             req->rq_auth_uid != pud->pud_uid) {
 180                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 181                        "while client claims %u:%u/%u:%u\n",
 182                        libcfs_nid2str(peernid), req->rq_auth_uid,
 183                        pud->pud_uid, pud->pud_gid,
 184                        pud->pud_fsuid, pud->pud_fsgid);
 185                 RETURN(-EACCES);
 186         }
 187
 188         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 189                 ucred->uc_identity = NULL;
 190                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 191         } else {
 192                 struct md_identity *identity;
 193
 194                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 195                                             pud->pud_uid);
 196                 if (IS_ERR(identity)) {
 197                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 198                                 ucred->uc_identity = NULL;
 199                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 200                                        CFS_SETGRP_PERM;
 201                         } else {
 202                                 CDEBUG(D_SEC,
 203                                        "Deny access without identity: uid %u\n",
 204                                        pud->pud_uid);
 205                                 RETURN(-EACCES);
 206                         }
 207                 } else {
 208                         ucred->uc_identity = identity;
 209                         perm = mdt_identity_get_perm(ucred->uc_identity,
 210                                                      peernid);
 211                 }
 212         }
 213
 214         /* find out the setuid/setgid attempt */
 215         setuid = (pud->pud_uid != pud->pud_fsuid);
 216         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 217                   (ucred->uc_identity &&
 218                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 219
 220         /* check permission of setuid */
 221         if (setuid && !(perm & CFS_SETUID_PERM)) {
 222                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 223                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 224                 GOTO(out, rc = -EACCES);
 225         }
 226
 227         /* check permission of setgid */
 228         if (setgid && !(perm & CFS_SETGID_PERM)) {
 229                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 230                        "from %s\n", pud->pud_uid, pud->pud_gid,
 231                        pud->pud_fsuid, pud->pud_fsgid,
 232                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 233                 GOTO(out, rc = -EACCES);
 234         }
 235
 236         if (perm & CFS_SETGRP_PERM) {
 237                 if (pud->pud_ngroups) {
 238                         /* setgroups for local client */
 239                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 240                         if (!ucred->uc_ginfo) {
 241                                 CERROR("failed to alloc %d groups\n",
 242                                        pud->pud_ngroups);
 243                                 GOTO(out, rc = -ENOMEM);
 244                         }
 245
 246                         lustre_groups_from_list(ucred->uc_ginfo,
 247                                                 pud->pud_groups);
 248                         lustre_groups_sort(ucred->uc_ginfo);
 249                 } else {
 250                         ucred->uc_ginfo = NULL;
 251                 }
 252         } else {
 253                 ucred->uc_suppgids[0] = -1;
 254                 ucred->uc_suppgids[1] = -1;
 255                 ucred->uc_ginfo = NULL;
 256         }
 257
 258         ucred->uc_uid   = pud->pud_uid;
 259         ucred->uc_gid   = pud->pud_gid;
 260         ucred->uc_fsuid = pud->pud_fsuid;
 261         ucred->uc_fsgid = pud->pud_fsgid;
 262
 263         /* process root_squash here. */
 264         mdt_root_squash(info, peernid);
 265
 266         /* remove fs privilege for non-root user. */
 267         if (ucred->uc_fsuid && drop_fs_cap)
 268                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 269         else
 270                 ucred->uc_cap = pud->pud_cap;
 271         ucred->uc_valid = UCRED_NEW;
 272         ucred_set_jobid(info, ucred);
 273
 274         EXIT;
 275
 276 out:
 277         if (rc) {
 278                 if (ucred->uc_ginfo) {
 279                         put_group_info(ucred->uc_ginfo);
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282                 if (ucred->uc_identity) {
 283                         mdt_identity_put(mdt->mdt_identity_cache,
 284                                          ucred->uc_identity);
 285                         ucred->uc_identity = NULL;
 286                 }
 287         }
 288
 289         return rc;
 290 }
 291
 292 /**
 293  * Check whether allow the client to set supplementary group IDs or not.
 294  *
 295  * \param[in] info      pointer to the thread context
 296  * \param[in] uc        pointer to the RPC user descriptor
 297  *
 298  * \retval              true if allow to set supplementary group IDs
 299  * \retval              false for other cases
 300  */
 301 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 302 {
 303         __u32 perm;
 304
 305         /* 1. If identity_upcall is disabled,
 306          *    permit local client to do anything. */
 307         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 308                 return true;
 309
 310         /* 2. If fail to get related identities, then forbid any client to
 311          *    set supplementary group IDs. */
 312         if (uc->uc_identity == NULL)
 313                 return false;
 314
 315         /* 3. Check the permission in the identities. */
 316         perm = mdt_identity_get_perm(uc->uc_identity,
 317                                      mdt_info_req(info)->rq_peer.nid);
 318         if (perm & CFS_SETGRP_PERM)
 319                 return true;
 320
 321         return false;
 322 }
 323
 324 int mdt_check_ucred(struct mdt_thread_info *info)
 325 {
 326         struct ptlrpc_request   *req = mdt_info_req(info);
 327         struct mdt_device       *mdt = info->mti_mdt;
 328         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 329         struct lu_ucred         *ucred = mdt_ucred(info);
 330         struct md_identity      *identity = NULL;
 331         lnet_nid_t               peernid = req->rq_peer.nid;
 332         __u32                    perm = 0;
 333         int                      setuid;
 334         int                      setgid;
 335         int                      rc = 0;
 336
 337         ENTRY;
 338
 339         LASSERT(ucred != NULL);
 340         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 341                 RETURN(0);
 342
 343         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 344                 RETURN(0);
 345
 346         /* sanity check: if we use strong authentication, we expect the
 347          * uid which client claimed is true */
 348         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 349             req->rq_auth_uid != pud->pud_uid) {
 350                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 351                        "while client claims %u:%u/%u:%u\n",
 352                        libcfs_nid2str(peernid), req->rq_auth_uid,
 353                        pud->pud_uid, pud->pud_gid,
 354                        pud->pud_fsuid, pud->pud_fsgid);
 355                 RETURN(-EACCES);
 356         }
 357
 358         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 359                 RETURN(0);
 360
 361         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 362         if (IS_ERR(identity)) {
 363                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 364                         RETURN(0);
 365                 } else {
 366                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 367                                pud->pud_uid);
 368                         RETURN(-EACCES);
 369                 }
 370         }
 371
 372         perm = mdt_identity_get_perm(identity, peernid);
 373         /* find out the setuid/setgid attempt */
 374         setuid = (pud->pud_uid != pud->pud_fsuid);
 375         setgid = (pud->pud_gid != pud->pud_fsgid ||
 376                   pud->pud_gid != identity->mi_gid);
 377
 378         /* check permission of setuid */
 379         if (setuid && !(perm & CFS_SETUID_PERM)) {
 380                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 381                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 382                 GOTO(out, rc = -EACCES);
 383         }
 384
 385         /* check permission of setgid */
 386         if (setgid && !(perm & CFS_SETGID_PERM)) {
 387                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 388                        "from %s\n", pud->pud_uid, pud->pud_gid,
 389                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 390                        libcfs_nid2str(peernid));
 391                 GOTO(out, rc = -EACCES);
 392         }
 393
 394         EXIT;
 395
 396 out:
 397         mdt_identity_put(mdt->mdt_identity_cache, identity);
 398         return rc;
 399 }
 400
 401 static int old_init_ucred_common(struct mdt_thread_info *info,
 402                                  struct lu_nodemap *nodemap,
 403                                  bool drop_fs_cap)
 404 {
 405         struct lu_ucred         *uc = mdt_ucred(info);
 406         struct mdt_device       *mdt = info->mti_mdt;
 407         struct md_identity      *identity = NULL;
 408
 409         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 410                 /* deny access before we get identity ref */
 411                 if (nodemap->nmf_deny_unknown)
 412                         RETURN(-EACCES);
 413
 414                 uc->uc_fsuid = nodemap->nm_squash_uid;
 415                 uc->uc_fsgid = nodemap->nm_squash_gid;
 416                 uc->uc_cap = 0;
 417                 uc->uc_suppgids[0] = -1;
 418                 uc->uc_suppgids[1] = -1;
 419         }
 420
 421         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 422                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 423                                             uc->uc_fsuid);
 424                 if (IS_ERR(identity)) {
 425                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 426                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 427                                 identity = NULL;
 428                         } else {
 429                                 CDEBUG(D_SEC, "Deny access without identity: "
 430                                        "uid %u\n", uc->uc_fsuid);
 431                                 RETURN(-EACCES);
 432                         }
 433                 }
 434         }
 435         uc->uc_identity = identity;
 436
 437         /* process root_squash here. */
 438         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 439
 440         /* remove fs privilege for non-root user. */
 441         if (uc->uc_fsuid && drop_fs_cap)
 442                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 443         uc->uc_valid = UCRED_OLD;
 444         ucred_set_jobid(info, uc);
 445
 446         EXIT;
 447
 448         return 0;
 449 }
 450
 451 static int old_init_ucred(struct mdt_thread_info *info,
 452                           struct mdt_body *body, bool drop_fs_cap)
 453 {
 454         struct lu_ucred *uc = mdt_ucred(info);
 455         struct lu_nodemap *nodemap;
 456         int rc;
 457         ENTRY;
 458
 459         nodemap = nodemap_get_from_exp(info->mti_exp);
 460         if (IS_ERR(nodemap))
 461                 RETURN(PTR_ERR(nodemap));
 462
 463         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 464                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 465         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 466                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 467         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 468                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 469         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 470                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 471
 472         LASSERT(uc != NULL);
 473         uc->uc_valid = UCRED_INVALID;
 474         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 475         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 476         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 477         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 478         uc->uc_suppgids[0] = body->mbo_suppgid;
 479         uc->uc_suppgids[1] = -1;
 480         uc->uc_ginfo = NULL;
 481         uc->uc_cap = body->mbo_capability;
 482
 483         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 484         nodemap_putref(nodemap);
 485
 486         RETURN(rc);
 487 }
 488
 489 static int old_init_ucred_reint(struct mdt_thread_info *info)
 490 {
 491         struct lu_ucred *uc = mdt_ucred(info);
 492         struct lu_nodemap *nodemap;
 493         int rc;
 494         ENTRY;
 495
 496         nodemap = nodemap_get_from_exp(info->mti_exp);
 497         if (IS_ERR(nodemap))
 498                 RETURN(PTR_ERR(nodemap));
 499
 500         LASSERT(uc != NULL);
 501
 502         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 503                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 504         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 505                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 506
 507         uc->uc_valid = UCRED_INVALID;
 508         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 509         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 510         uc->uc_ginfo = NULL;
 511
 512         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 513         nodemap_putref(nodemap);
 514
 515         RETURN(rc);
 516 }
 517
 518 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 519                                    struct mdt_body *body,
 520                                    bool drop_fs_cap)
 521 {
 522         struct ptlrpc_request   *req = mdt_info_req(info);
 523         struct lu_ucred         *uc  = mdt_ucred(info);
 524
 525         LASSERT(uc != NULL);
 526         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 527                 return 0;
 528
 529         mdt_exit_ucred(info);
 530
 531         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 532                 return old_init_ucred(info, body, drop_fs_cap);
 533         else
 534                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 535 }
 536
 537 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 538 {
 539         return __mdt_init_ucred(info, body, true);
 540 }
 541
 542 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 543  * doesn't set correct fsuid explicitely, but raise capability to allow
 544  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 545  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 546  * will fail permission check. */
 547 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 548                                   struct mdt_body *body)
 549 {
 550         return __mdt_init_ucred(info, body, false);
 551 }
 552
 553 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 554 {
 555         struct ptlrpc_request *req = mdt_info_req(info);
 556         struct lu_ucred       *uc  = mdt_ucred(info);
 557         struct md_attr        *ma  = &info->mti_attr;
 558
 559         LASSERT(uc != NULL);
 560         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 561                 return 0;
 562
 563         /* LU-5564: for normal close request, skip permission check */
 564         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 565             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 566                 uc->uc_cap |= CFS_CAP_FS_MASK;
 567
 568         mdt_exit_ucred(info);
 569
 570         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 571                 return old_init_ucred_reint(info);
 572         else
 573                 return new_init_ucred(info, REC_INIT, NULL, true);
 574 }
 575
 576 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 577 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 578 {
 579         const struct lov_ost_data_v1    *lod;
 580         int                              i;
 581         __u16                            count;
 582
 583         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 584                 return;
 585
 586         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 587
 588         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 589                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 590                le32_to_cpu(lmm->lmm_pattern));
 591         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 592                le32_to_cpu(lmm->lmm_stripe_size), count);
 593
 594         /* If it's a directory or a released file, then there are
 595          * no actual objects to print, so bail out. */
 596         if (valid & OBD_MD_FLDIREA ||
 597             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 598                 return;
 599
 600         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 601         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 602                 struct ost_id oi;
 603
 604                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 605                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 606                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 607         }
 608 }
 609
 610 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 611 {
 612         const struct lmv_mds_md_v1 *lmm1;
 613         int                        i;
 614
 615         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 616                 return;
 617
 618         lmm1 = &lmv->lmv_md_v1;
 619         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 620                le32_to_cpu(lmm1->lmv_magic),
 621                le32_to_cpu(lmm1->lmv_master_mdt_index),
 622                le32_to_cpu(lmm1->lmv_stripe_count));
 623
 624         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 625                 return;
 626
 627         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 628                 struct lu_fid fid;
 629
 630                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 631                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 632         }
 633 }
 634
 635 /* Shrink and/or grow reply buffers */
 636 int mdt_fix_reply(struct mdt_thread_info *info)
 637 {
 638         struct req_capsule *pill = info->mti_pill;
 639         struct mdt_body    *body;
 640         int                md_size, md_packed = 0;
 641         int                acl_size;
 642         int                rc = 0;
 643         ENTRY;
 644
 645         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 646         LASSERT(body != NULL);
 647
 648         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 649                                OBD_MD_LINKNAME))
 650                 md_size = body->mbo_eadatasize;
 651         else
 652                 md_size = 0;
 653
 654         acl_size = body->mbo_aclsize;
 655
 656         /* this replay - not send info to client */
 657         if (info->mti_spec.no_create) {
 658                 md_size = 0;
 659                 acl_size = 0;
 660         }
 661
 662         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 663                md_size, acl_size);
 664 /*
 665             &RMF_MDT_BODY,
 666             &RMF_MDT_MD,
 667             &RMF_ACL, or &RMF_LOGCOOKIES
 668 (optional)  &RMF_CAPA1,
 669 (optional)  &RMF_CAPA2,
 670 (optional)  something else
 671 */
 672
 673         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 674          * buffers before growing it */
 675         if (info->mti_big_lmm_used) {
 676                 /* big_lmm buffer may be used even without packing the result
 677                  * into reply, just for internal server needs */
 678                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 679                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 680                                                          RCL_SERVER);
 681
 682                 /* free big lmm if md_size is not needed */
 683                 if (md_size == 0 || md_packed == 0) {
 684                         info->mti_big_lmm_used = 0;
 685                 } else {
 686                         /* buffer must be allocated separately */
 687                         LASSERT(info->mti_attr.ma_lmm !=
 688                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 689                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 690                 }
 691         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 692                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 693         }
 694
 695         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 696                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 697         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 698                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 699                                    acl_size, RCL_SERVER);
 700
 701         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 702             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 703                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 704
 705         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 706             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 707                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 708
 709         /*
 710          * Some more field should be shrinked if needed.
 711          * This should be done by those who added fields to reply message.
 712          */
 713
 714         /* Grow MD buffer if needed finally */
 715         if (info->mti_big_lmm_used) {
 716                 void *lmm;
 717
 718                 LASSERT(md_size > md_packed);
 719                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 720                        md_size - md_packed);
 721                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 722                 if (rc) {
 723                         /* we can't answer with proper LOV EA, drop flags,
 724                          * the rc is also returned so this request is
 725                          * considered as failed */
 726                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 727                         /* don't return transno along with error */
 728                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 729                 } else {
 730                         /* now we need to pack right LOV/LMV EA */
 731                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 732                         if (info->mti_attr.ma_valid & MA_LOV) {
 733                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 734                                                              RCL_SERVER) ==
 735                                                 info->mti_attr.ma_lmm_size);
 736                                 memcpy(lmm, info->mti_attr.ma_lmm,
 737                                        info->mti_attr.ma_lmm_size);
 738                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 739                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 740                                                              RCL_SERVER) ==
 741                                                 info->mti_attr.ma_lmv_size);
 742                                 memcpy(lmm, info->mti_attr.ma_lmv,
 743                                        info->mti_attr.ma_lmv_size);
 744                         }
 745                 }
 746                 /* update mdt_max_mdsize so clients will be aware about that */
 747                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 748                         info->mti_mdt->mdt_max_mdsize =
 749                                                     info->mti_attr.ma_lmm_size;
 750                 info->mti_big_lmm_used = 0;
 751         }
 752         RETURN(rc);
 753 }
 754
 755
 756 /* if object is dying, pack the lov/llog data,
 757  * parameter info->mti_attr should be valid at this point! */
 758 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 759                            const struct md_attr *ma)
 760 {
 761         struct mdt_body       *repbody;
 762         const struct lu_attr *la = &ma->ma_attr;
 763         ENTRY;
 764
 765         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 766         LASSERT(repbody != NULL);
 767
 768         if (ma->ma_valid & MA_INODE)
 769                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 770
 771         if (ma->ma_valid & MA_LOV) {
 772                 CERROR("No need in LOV EA upon unlink\n");
 773                 dump_stack();
 774         }
 775         repbody->mbo_eadatasize = 0;
 776
 777         RETURN(0);
 778 }
 779
 780 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 781                                   struct md_attr *ma)
 782 {
 783         __u64 out;
 784
 785         out = 0;
 786         if (in & MDS_ATTR_MODE)
 787                 out |= LA_MODE;
 788         if (in & MDS_ATTR_UID)
 789                 out |= LA_UID;
 790         if (in & MDS_ATTR_GID)
 791                 out |= LA_GID;
 792         if (in & MDS_ATTR_SIZE)
 793                 out |= LA_SIZE;
 794         if (in & MDS_ATTR_BLOCKS)
 795                 out |= LA_BLOCKS;
 796         if (in & MDS_ATTR_ATIME_SET)
 797                 out |= LA_ATIME;
 798         if (in & MDS_ATTR_CTIME_SET)
 799                 out |= LA_CTIME;
 800         if (in & MDS_ATTR_MTIME_SET)
 801                 out |= LA_MTIME;
 802         if (in & MDS_ATTR_ATTR_FLAG)
 803                 out |= LA_FLAGS;
 804         if (in & MDS_ATTR_KILL_SUID)
 805                 out |= LA_KILL_SUID;
 806         if (in & MDS_ATTR_KILL_SGID)
 807                 out |= LA_KILL_SGID;
 808
 809         if (in & MDS_ATTR_FROM_OPEN)
 810                 rr->rr_flags |= MRF_OPEN_TRUNC;
 811         if (in & MDS_OPEN_OWNEROVERRIDE)
 812                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 813         if (in & MDS_ATTR_FORCE)
 814                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 815
 816         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 817                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 818                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 819                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 820                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 821                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 822         if (in != 0)
 823                 CERROR("Unknown attr bits: %#llx\n", in);
 824         return out;
 825 }
 826
 827 /* unpacking */
 828
 829 int mdt_name_unpack(struct req_capsule *pill,
 830                     const struct req_msg_field *field,
 831                     struct lu_name *ln,
 832                     enum mdt_name_flags flags)
 833 {
 834         ln->ln_name = req_capsule_client_get(pill, field);
 835         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 836
 837         if (!lu_name_is_valid(ln)) {
 838                 ln->ln_name = NULL;
 839                 ln->ln_namelen = 0;
 840
 841                 return -EPROTO;
 842         }
 843
 844         if ((flags & MNF_FIX_ANON) &&
 845             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 846                 /* Newer (3.x) kernels use a name of "/" for the
 847                  * "anonymous" disconnected dentries from NFS
 848                  * filehandle conversion. See d_obtain_alias(). */
 849                 ln->ln_name = NULL;
 850                 ln->ln_namelen = 0;
 851         }
 852
 853         return 0;
 854 }
 855
 856 static int mdt_file_secctx_unpack(struct req_capsule *pill,
 857                                   const char **secctx_name,
 858                                   void **secctx, size_t *secctx_size)
 859 {
 860         const char *name;
 861         size_t name_size;
 862
 863         *secctx_name = NULL;
 864         *secctx = NULL;
 865         *secctx_size = 0;
 866
 867         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
 868             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
 869                 return 0;
 870
 871         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
 872                                          RCL_CLIENT);
 873         if (name_size == 0)
 874                 return 0;
 875
 876         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
 877         if (strnlen(name, name_size) != name_size - 1)
 878                 return -EPROTO;
 879
 880         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
 881             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
 882                 return -EPROTO;
 883
 884         *secctx_name = name;
 885         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
 886         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
 887
 888         return 0;
 889 }
 890
 891 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 892 {
 893         struct lu_ucred         *uc = mdt_ucred(info);
 894         struct md_attr          *ma = &info->mti_attr;
 895         struct lu_attr          *la = &ma->ma_attr;
 896         struct req_capsule      *pill = info->mti_pill;
 897         struct mdt_reint_record *rr = &info->mti_rr;
 898         struct mdt_rec_setattr  *rec;
 899         struct lu_nodemap       *nodemap;
 900         ENTRY;
 901
 902         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 903         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 904         if (rec == NULL)
 905                 RETURN(-EFAULT);
 906
 907         /* This prior initialization is needed for old_init_ucred_reint() */
 908         uc->uc_fsuid = rec->sa_fsuid;
 909         uc->uc_fsgid = rec->sa_fsgid;
 910         uc->uc_cap   = rec->sa_cap;
 911         uc->uc_suppgids[0] = rec->sa_suppgid;
 912         uc->uc_suppgids[1] = -1;
 913
 914         rr->rr_fid1 = &rec->sa_fid;
 915         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 916         la->la_mode  = rec->sa_mode;
 917         la->la_flags = rec->sa_attr_flags;
 918
 919         nodemap = nodemap_get_from_exp(info->mti_exp);
 920         if (IS_ERR(nodemap))
 921                 RETURN(PTR_ERR(nodemap));
 922
 923         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 924                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 925         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 926                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 927         nodemap_putref(nodemap);
 928
 929         la->la_size  = rec->sa_size;
 930         la->la_blocks = rec->sa_blocks;
 931         la->la_ctime = rec->sa_ctime;
 932         la->la_atime = rec->sa_atime;
 933         la->la_mtime = rec->sa_mtime;
 934         ma->ma_valid = MA_INODE;
 935
 936         if (rec->sa_bias & MDS_DATA_MODIFIED)
 937                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 938         else
 939                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 940
 941         if (rec->sa_bias & MDS_HSM_RELEASE)
 942                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 943         else
 944                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 945
 946         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 947                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 948         else
 949                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 950
 951         RETURN(0);
 952 }
 953
 954 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 955 {
 956         struct req_capsule *pill = info->mti_pill;
 957         struct mdt_ioepoch *ioepoch;
 958         ENTRY;
 959
 960         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 961                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 962         else
 963                 ioepoch = NULL;
 964
 965         if (ioepoch == NULL)
 966                 RETURN(-EPROTO);
 967
 968         info->mti_close_handle = ioepoch->mio_handle;
 969
 970         RETURN(0);
 971 }
 972
 973 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 974         struct req_capsule      *pill = info->mti_pill;
 975
 976         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 977                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 978                 if (info->mti_dlm_req == NULL)
 979                         RETURN(-EFAULT);
 980         }
 981
 982         RETURN(0);
 983 }
 984
 985 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 986 {
 987         struct mdt_reint_record *rr = &info->mti_rr;
 988         struct md_attr          *ma = &info->mti_attr;
 989         struct req_capsule      *pill = info->mti_pill;
 990         int rc;
 991         ENTRY;
 992
 993         rc = mdt_setattr_unpack_rec(info);
 994         if (rc)
 995                 RETURN(rc);
 996
 997         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 998                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 999                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1000                                                         RCL_CLIENT);
1001                 if (rr->rr_eadatalen > 0) {
1002                         const struct lmv_user_md        *lum;
1003
1004                         lum = rr->rr_eadata;
1005                         /* Sigh ma_valid(from req) does not indicate whether
1006                          * it will set LOV/LMV EA, so we have to check magic */
1007                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1008                                 ma->ma_valid |= MA_LMV;
1009                                 ma->ma_lmv = (void *)rr->rr_eadata;
1010                                 ma->ma_lmv_size = rr->rr_eadatalen;
1011                         } else {
1012                                 ma->ma_valid |= MA_LOV;
1013                                 ma->ma_lmm = (void *)rr->rr_eadata;
1014                                 ma->ma_lmm_size = rr->rr_eadatalen;
1015                         }
1016                 }
1017         }
1018
1019         rc = mdt_dlmreq_unpack(info);
1020         RETURN(rc);
1021 }
1022
1023 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1024 {
1025         struct md_attr          *ma = &info->mti_attr;
1026         struct req_capsule      *pill = info->mti_pill;
1027         ENTRY;
1028
1029         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1030                 RETURN(0);
1031
1032         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1033
1034         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1035             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1036                 RETURN(-EFAULT);
1037
1038         RETURN(0);
1039 }
1040
1041 int mdt_close_unpack(struct mdt_thread_info *info)
1042 {
1043         int rc;
1044         ENTRY;
1045
1046         rc = mdt_close_handle_unpack(info);
1047         if (rc)
1048                 RETURN(rc);
1049
1050         rc = mdt_setattr_unpack_rec(info);
1051         if (rc)
1052                 RETURN(rc);
1053
1054         rc = mdt_intent_close_unpack(info);
1055         if (rc)
1056                 RETURN(rc);
1057
1058         RETURN(mdt_init_ucred_reint(info));
1059 }
1060
1061 static int mdt_create_unpack(struct mdt_thread_info *info)
1062 {
1063         struct lu_ucred         *uc  = mdt_ucred(info);
1064         struct mdt_rec_create   *rec;
1065         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1066         struct mdt_reint_record *rr = &info->mti_rr;
1067         struct req_capsule      *pill = info->mti_pill;
1068         struct md_op_spec       *sp = &info->mti_spec;
1069         int rc;
1070         ENTRY;
1071
1072         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1073         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1074         if (rec == NULL)
1075                 RETURN(-EFAULT);
1076
1077         /* This prior initialization is needed for old_init_ucred_reint() */
1078         uc->uc_fsuid = rec->cr_fsuid;
1079         uc->uc_fsgid = rec->cr_fsgid;
1080         uc->uc_cap   = rec->cr_cap;
1081         uc->uc_suppgids[0] = rec->cr_suppgid1;
1082         uc->uc_suppgids[1] = -1;
1083         uc->uc_umask = rec->cr_umask;
1084
1085         rr->rr_fid1 = &rec->cr_fid1;
1086         rr->rr_fid2 = &rec->cr_fid2;
1087         attr->la_mode = rec->cr_mode;
1088         attr->la_rdev  = rec->cr_rdev;
1089         attr->la_uid   = rec->cr_fsuid;
1090         attr->la_gid   = rec->cr_fsgid;
1091         attr->la_ctime = rec->cr_time;
1092         attr->la_mtime = rec->cr_time;
1093         attr->la_atime = rec->cr_time;
1094         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1095                          LA_CTIME | LA_MTIME | LA_ATIME;
1096         memset(&sp->u, 0, sizeof(sp->u));
1097         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1098
1099         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1100         if (rc < 0)
1101                 RETURN(rc);
1102
1103         if (S_ISLNK(attr->la_mode)) {
1104                 const char *tgt = NULL;
1105
1106                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1107                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1108                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1109                         sp->u.sp_symname = tgt;
1110                 }
1111                 if (tgt == NULL)
1112                         RETURN(-EFAULT);
1113         } else {
1114                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1115                 if (S_ISDIR(attr->la_mode) &&
1116                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1117                         sp->u.sp_ea.eadata =
1118                                 req_capsule_client_get(pill, &RMF_EADATA);
1119                         sp->u.sp_ea.eadatalen =
1120                                 req_capsule_get_size(pill, &RMF_EADATA,
1121                                                      RCL_CLIENT);
1122                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1123                 }
1124         }
1125
1126         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1127                                     &sp->sp_cr_file_secctx,
1128                                     &sp->sp_cr_file_secctx_size);
1129         if (rc < 0)
1130                 RETURN(rc);
1131
1132         rc = mdt_dlmreq_unpack(info);
1133         RETURN(rc);
1134 }
1135
1136 static int mdt_link_unpack(struct mdt_thread_info *info)
1137 {
1138         struct lu_ucred         *uc  = mdt_ucred(info);
1139         struct mdt_rec_link     *rec;
1140         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1141         struct mdt_reint_record *rr = &info->mti_rr;
1142         struct req_capsule      *pill = info->mti_pill;
1143         int rc;
1144         ENTRY;
1145
1146         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1147         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1148         if (rec == NULL)
1149                 RETURN(-EFAULT);
1150
1151         /* This prior initialization is needed for old_init_ucred_reint() */
1152         uc->uc_fsuid = rec->lk_fsuid;
1153         uc->uc_fsgid = rec->lk_fsgid;
1154         uc->uc_cap   = rec->lk_cap;
1155         uc->uc_suppgids[0] = rec->lk_suppgid1;
1156         uc->uc_suppgids[1] = rec->lk_suppgid2;
1157
1158         attr->la_uid = rec->lk_fsuid;
1159         attr->la_gid = rec->lk_fsgid;
1160         rr->rr_fid1 = &rec->lk_fid1;
1161         rr->rr_fid2 = &rec->lk_fid2;
1162         attr->la_ctime = rec->lk_time;
1163         attr->la_mtime = rec->lk_time;
1164         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1165
1166         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1167         if (rc < 0)
1168                 RETURN(rc);
1169
1170         rc = mdt_dlmreq_unpack(info);
1171
1172         RETURN(rc);
1173 }
1174
1175 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1176 {
1177         struct lu_ucred         *uc  = mdt_ucred(info);
1178         struct mdt_rec_unlink   *rec;
1179         struct md_attr          *ma = &info->mti_attr;
1180         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1181         struct mdt_reint_record *rr = &info->mti_rr;
1182         struct req_capsule      *pill = info->mti_pill;
1183         int rc;
1184         ENTRY;
1185
1186         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1187         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1188         if (rec == NULL)
1189                 RETURN(-EFAULT);
1190
1191         /* This prior initialization is needed for old_init_ucred_reint() */
1192         uc->uc_fsuid = rec->ul_fsuid;
1193         uc->uc_fsgid = rec->ul_fsgid;
1194         uc->uc_cap   = rec->ul_cap;
1195         uc->uc_suppgids[0] = rec->ul_suppgid1;
1196         uc->uc_suppgids[1] = -1;
1197
1198         attr->la_uid = rec->ul_fsuid;
1199         attr->la_gid = rec->ul_fsgid;
1200         rr->rr_fid1 = &rec->ul_fid1;
1201         rr->rr_fid2 = &rec->ul_fid2;
1202         attr->la_ctime = rec->ul_time;
1203         attr->la_mtime = rec->ul_time;
1204         attr->la_mode  = rec->ul_mode;
1205         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1206
1207         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1208         if (rc < 0)
1209                 RETURN(rc);
1210
1211         if (rec->ul_bias & MDS_VTX_BYPASS)
1212                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1213         else
1214                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1215
1216         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1217
1218         rc = mdt_dlmreq_unpack(info);
1219         RETURN(rc);
1220 }
1221
1222 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1223 {
1224         info->mti_spec.sp_rm_entry = 1;
1225         return mdt_unlink_unpack(info);
1226 }
1227
1228 static int mdt_rename_unpack(struct mdt_thread_info *info)
1229 {
1230         struct lu_ucred         *uc = mdt_ucred(info);
1231         struct mdt_rec_rename   *rec;
1232         struct md_attr          *ma = &info->mti_attr;
1233         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1234         struct mdt_reint_record *rr = &info->mti_rr;
1235         struct req_capsule      *pill = info->mti_pill;
1236         int rc;
1237         ENTRY;
1238
1239         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1240         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1241         if (rec == NULL)
1242                 RETURN(-EFAULT);
1243
1244         /* This prior initialization is needed for old_init_ucred_reint() */
1245         uc->uc_fsuid = rec->rn_fsuid;
1246         uc->uc_fsgid = rec->rn_fsgid;
1247         uc->uc_cap   = rec->rn_cap;
1248         uc->uc_suppgids[0] = rec->rn_suppgid1;
1249         uc->uc_suppgids[1] = rec->rn_suppgid2;
1250
1251         attr->la_uid = rec->rn_fsuid;
1252         attr->la_gid = rec->rn_fsgid;
1253         rr->rr_fid1 = &rec->rn_fid1;
1254         rr->rr_fid2 = &rec->rn_fid2;
1255         attr->la_ctime = rec->rn_time;
1256         attr->la_mtime = rec->rn_time;
1257         /* rename_tgt contains the mode already */
1258         attr->la_mode = rec->rn_mode;
1259         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1260
1261         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1262         if (rc < 0)
1263                 RETURN(rc);
1264
1265         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1266         if (rc < 0)
1267                 RETURN(rc);
1268
1269         if (rec->rn_bias & MDS_VTX_BYPASS)
1270                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1271         else
1272                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1273
1274         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1275                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1276                 rc = mdt_close_handle_unpack(info);
1277                 if (rc < 0)
1278                         RETURN(rc);
1279                 info->mti_spec.sp_migrate_close = 1;
1280         }
1281
1282         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1283
1284
1285         rc = mdt_dlmreq_unpack(info);
1286
1287         RETURN(rc);
1288 }
1289
1290 /*
1291  * please see comment above LOV_MAGIC_V1_DEF
1292  */
1293 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1294 {
1295         struct mdt_reint_record *rr = &info->mti_rr;
1296         struct lov_user_md_v1   *v1;
1297
1298         v1 = (void *)rr->rr_eadata;
1299         LASSERT(v1);
1300
1301         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1302                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1303                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1304                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1305                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1306                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1307                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1308                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1309                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1310                 }
1311         }
1312 }
1313
1314 static int mdt_open_unpack(struct mdt_thread_info *info)
1315 {
1316         struct lu_ucred         *uc = mdt_ucred(info);
1317         struct mdt_rec_create   *rec;
1318         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1319         struct req_capsule      *pill = info->mti_pill;
1320         struct mdt_reint_record *rr   = &info->mti_rr;
1321         struct ptlrpc_request   *req  = mdt_info_req(info);
1322         struct md_op_spec       *sp   = &info->mti_spec;
1323         int rc;
1324         ENTRY;
1325
1326         CLASSERT(sizeof(struct mdt_rec_create) == 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->cr_fsuid;
1333         uc->uc_fsgid = rec->cr_fsgid;
1334         uc->uc_cap   = rec->cr_cap;
1335         uc->uc_suppgids[0] = rec->cr_suppgid1;
1336         uc->uc_suppgids[1] = rec->cr_suppgid2;
1337         uc->uc_umask = rec->cr_umask;
1338
1339         rr->rr_fid1   = &rec->cr_fid1;
1340         rr->rr_fid2   = &rec->cr_fid2;
1341         rr->rr_handle = &rec->cr_old_handle;
1342         attr->la_mode = rec->cr_mode;
1343         attr->la_rdev  = rec->cr_rdev;
1344         attr->la_uid   = rec->cr_fsuid;
1345         attr->la_gid   = rec->cr_fsgid;
1346         attr->la_ctime = rec->cr_time;
1347         attr->la_mtime = rec->cr_time;
1348         attr->la_atime = rec->cr_time;
1349         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1350                          LA_CTIME | LA_MTIME | LA_ATIME;
1351         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1352         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1353         /* Do not trigger ASSERTION if client miss to set such flags. */
1354         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1355                 RETURN(-EPROTO);
1356
1357         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1358
1359         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1360
1361         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1362                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1363                                                         RCL_CLIENT);
1364                 if (rr->rr_eadatalen > 0) {
1365                         rr->rr_eadata = req_capsule_client_get(pill,
1366                                                                &RMF_EADATA);
1367                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1368                         sp->u.sp_ea.eadata = rr->rr_eadata;
1369                         sp->no_create = !!req_is_replay(req);
1370                         mdt_fix_lov_magic(info);
1371                 }
1372
1373                 /*
1374                  * Client default md_size may be 0 right after client start,
1375                  * until all osc are connected, set here just some reasonable
1376                  * value to prevent misbehavior.
1377                  */
1378                 if (rr->rr_eadatalen == 0 &&
1379                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1380                         rr->rr_eadatalen = MIN_MD_SIZE;
1381         }
1382
1383         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1384                                     &sp->sp_cr_file_secctx,
1385                                     &sp->sp_cr_file_secctx_size);
1386
1387         RETURN(rc);
1388 }
1389
1390 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1391 {
1392         struct mdt_reint_record *rr     = &info->mti_rr;
1393         struct lu_ucred         *uc     = mdt_ucred(info);
1394         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1395         struct req_capsule      *pill   = info->mti_pill;
1396         struct mdt_rec_setxattr *rec;
1397         int                      rc;
1398         ENTRY;
1399
1400
1401         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1402                          sizeof(struct mdt_rec_reint));
1403
1404         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1405         if (rec == NULL)
1406                 RETURN(-EFAULT);
1407
1408         /* This prior initialization is needed for old_init_ucred_reint() */
1409         uc->uc_fsuid  = rec->sx_fsuid;
1410         uc->uc_fsgid  = rec->sx_fsgid;
1411         uc->uc_cap    = rec->sx_cap;
1412         uc->uc_suppgids[0] = rec->sx_suppgid1;
1413         uc->uc_suppgids[1] = -1;
1414
1415         rr->rr_opcode = rec->sx_opcode;
1416         rr->rr_fid1   = &rec->sx_fid;
1417         attr->la_valid = rec->sx_valid;
1418         attr->la_ctime = rec->sx_time;
1419         attr->la_size = rec->sx_size;
1420         attr->la_flags = rec->sx_flags;
1421
1422         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1423         if (rc < 0)
1424                 RETURN(rc);
1425
1426         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1427                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1428                                                         RCL_CLIENT);
1429                 if (rr->rr_eadatalen > 0) {
1430                         rr->rr_eadata = req_capsule_client_get(pill,
1431                                                                &RMF_EADATA);
1432                         if (rr->rr_eadata == NULL)
1433                                 RETURN(-EFAULT);
1434                 } else {
1435                         rr->rr_eadata = NULL;
1436                 }
1437         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1438                 CDEBUG(D_INFO, "no xattr data supplied\n");
1439                 RETURN(-EFAULT);
1440         }
1441
1442         if (mdt_dlmreq_unpack(info) < 0)
1443                 RETURN(-EPROTO);
1444
1445         RETURN(0);
1446 }
1447
1448
1449 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1450
1451 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1452         [REINT_SETATTR]  = mdt_setattr_unpack,
1453         [REINT_CREATE]   = mdt_create_unpack,
1454         [REINT_LINK]     = mdt_link_unpack,
1455         [REINT_UNLINK]   = mdt_unlink_unpack,
1456         [REINT_RENAME]   = mdt_rename_unpack,
1457         [REINT_OPEN]     = mdt_open_unpack,
1458         [REINT_SETXATTR] = mdt_setxattr_unpack,
1459         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1460         [REINT_MIGRATE]  = mdt_rename_unpack,
1461 };
1462
1463 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1464 {
1465         int rc;
1466         ENTRY;
1467
1468         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1469         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1470                 info->mti_rr.rr_opcode = op;
1471                 rc = mdt_reint_unpackers[op](info);
1472         } else {
1473                 CERROR("Unexpected opcode %d\n", op);
1474                 rc = -EFAULT;
1475         }
1476         RETURN(rc);
1477 }