lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/mdt/mdt_lib.c
  32  *
  33  * Lustre Metadata Target (mdt) request unpacking helper.
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Mike Shaver <shaver@clusterfs.com>
  39  * Author: Nikita Danilov <nikita@clusterfs.com>
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  * Author: Fan Yong <fanyong@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/user_namespace.h>
  47 #include <linux/uidgid.h>
  48
  49 #include "mdt_internal.h"
  50 #include <uapi/linux/lnet/nidstr.h>
  51 #include <lustre_nodemap.h>
  52
  53 typedef enum ucred_init_type {
  54         NONE_INIT       = 0,
  55         BODY_INIT       = 1,
  56         REC_INIT        = 2
  57 } ucred_init_type_t;
  58
  59 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
  60 {
  61         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
  62 }
  63
  64 void mdt_exit_ucred(struct mdt_thread_info *info)
  65 {
  66         struct lu_ucred   *uc  = mdt_ucred(info);
  67         struct mdt_device *mdt = info->mti_mdt;
  68
  69         LASSERT(uc != NULL);
  70         if (uc->uc_valid != UCRED_INIT) {
  71                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  72                 if (uc->uc_ginfo) {
  73                         put_group_info(uc->uc_ginfo);
  74                         uc->uc_ginfo = NULL;
  75                 }
  76                 if (uc->uc_identity) {
  77                         mdt_identity_put(mdt->mdt_identity_cache,
  78                                          uc->uc_identity);
  79                         uc->uc_identity = NULL;
  80                 }
  81                 uc->uc_valid = UCRED_INIT;
  82         }
  83 }
  84
  85 static int match_nosquash_list(struct spinlock *rsi_lock,
  86                                struct list_head *nidlist,
  87                                lnet_nid_t peernid)
  88 {
  89         int rc;
  90         ENTRY;
  91         spin_lock(rsi_lock);
  92         rc = cfs_match_nid(peernid, nidlist);
  93         spin_unlock(rsi_lock);
  94         RETURN(rc);
  95 }
  96
  97 /* root_squash for inter-MDS operations */
  98 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  99 {
 100         struct lu_ucred *ucred = mdt_ucred(info);
 101         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 102         ENTRY;
 103
 104         LASSERT(ucred != NULL);
 105         if (!squash->rsi_uid || ucred->uc_fsuid)
 106                 RETURN(0);
 107
 108         if (match_nosquash_list(&squash->rsi_lock,
 109                                 &squash->rsi_nosquash_nids,
 110                                 peernid)) {
 111                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 112                        libcfs_nid2str(peernid));
 113                 RETURN(0);
 114         }
 115
 116         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 117                libcfs_nid2str(peernid),
 118                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap.cap[0],
 119                squash->rsi_uid, squash->rsi_gid, 0);
 120
 121         ucred->uc_fsuid = squash->rsi_uid;
 122         ucred->uc_fsgid = squash->rsi_gid;
 123         ucred->uc_cap = CAP_EMPTY_SET;
 124         ucred->uc_suppgids[0] = -1;
 125         ucred->uc_suppgids[1] = -1;
 126
 127         RETURN(0);
 128 }
 129
 130 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 131 {
 132         struct ptlrpc_request   *req = mdt_info_req(info);
 133         const char              *jobid = mdt_req_get_jobid(req);
 134
 135         /* set jobid if specified. */
 136         if (jobid)
 137                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 138         else
 139                 uc->uc_jobid[0] = '\0';
 140 }
 141
 142 static void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
 143 {
 144         if (info && info->mti_exp && info->mti_exp->exp_connection)
 145                 uc->uc_nid = info->mti_exp->exp_connection->c_peer.nid;
 146         else
 147                 uc->uc_nid = LNET_NID_ANY;
 148 }
 149
 150 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
 151                                     struct lu_ucred *uc)
 152 {
 153         struct lu_nodemap *nodemap = NULL;
 154         bool audit = true;
 155
 156         if (info && info->mti_exp) {
 157                 nodemap = nodemap_get_from_exp(info->mti_exp);
 158                 if (nodemap && !IS_ERR(nodemap)) {
 159                         audit = nodemap->nmf_enable_audit;
 160                         nodemap_putref(nodemap);
 161                 }
 162         }
 163
 164         uc->uc_enable_audit = audit;
 165 }
 166
 167 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 168                           void *buf)
 169 {
 170         struct ptlrpc_request *req = mdt_info_req(info);
 171         struct mdt_device *mdt = info->mti_mdt;
 172         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 173         struct lu_ucred *ucred = mdt_ucred(info);
 174         struct lu_nodemap *nodemap;
 175         lnet_nid_t peernid = req->rq_peer.nid;
 176         __u32 perm = 0;
 177         int setuid;
 178         int setgid;
 179         bool is_nm_gid_squashed = false;
 180         int rc = 0;
 181
 182         ENTRY;
 183
 184         LASSERT(req->rq_auth_gss);
 185         LASSERT(!req->rq_auth_usr_mdt);
 186         LASSERT(req->rq_user_desc);
 187         LASSERT(ucred != NULL);
 188
 189         ucred->uc_valid = UCRED_INVALID;
 190
 191         nodemap = nodemap_get_from_exp(info->mti_exp);
 192         if (IS_ERR(nodemap))
 193                 RETURN(PTR_ERR(nodemap));
 194
 195         pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 196                                        NODEMAP_CLIENT_TO_FS, pud->pud_uid);
 197         pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 198                                        NODEMAP_CLIENT_TO_FS, pud->pud_gid);
 199         pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 200                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
 201         pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 202                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
 203
 204         ucred->uc_o_uid = pud->pud_uid;
 205         ucred->uc_o_gid = pud->pud_gid;
 206         ucred->uc_o_fsuid = pud->pud_fsuid;
 207         ucred->uc_o_fsgid = pud->pud_fsgid;
 208
 209         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 210                 /* deny access before we get identity ref */
 211                 if (nodemap->nmf_deny_unknown) {
 212                         nodemap_putref(nodemap);
 213                         RETURN(-EACCES);
 214                 }
 215
 216                 ucred->uc_suppgids[0] = -1;
 217                 ucred->uc_suppgids[1] = -1;
 218         }
 219
 220         if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
 221                 is_nm_gid_squashed = true;
 222
 223         nodemap_putref(nodemap);
 224
 225         if (type == BODY_INIT) {
 226                 struct mdt_body *body = (struct mdt_body *)buf;
 227
 228                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 229                 ucred->uc_suppgids[1] = -1;
 230         }
 231
 232         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 233             req->rq_auth_uid != pud->pud_uid) {
 234                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 235                        "while client claims %u:%u/%u:%u\n",
 236                        libcfs_nid2str(peernid), req->rq_auth_uid,
 237                        pud->pud_uid, pud->pud_gid,
 238                        pud->pud_fsuid, pud->pud_fsgid);
 239                 RETURN(-EACCES);
 240         }
 241
 242         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 243                 ucred->uc_identity = NULL;
 244                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 245         } else {
 246                 struct md_identity *identity;
 247
 248                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 249                                             pud->pud_uid);
 250                 if (IS_ERR(identity)) {
 251                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 252                                 ucred->uc_identity = NULL;
 253                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 254                                        CFS_SETGRP_PERM;
 255                         } else {
 256                                 CDEBUG(D_SEC,
 257                                        "Deny access without identity: uid %u\n",
 258                                        pud->pud_uid);
 259                                 RETURN(-EACCES);
 260                         }
 261                 } else {
 262                         ucred->uc_identity = identity;
 263                         perm = mdt_identity_get_perm(ucred->uc_identity,
 264                                                      peernid);
 265                 }
 266         }
 267
 268         /* find out the setuid/setgid attempt */
 269         setuid = (pud->pud_uid != pud->pud_fsuid);
 270         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 271                   (ucred->uc_identity &&
 272                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 273
 274         /* check permission of setuid */
 275         if (setuid && !(perm & CFS_SETUID_PERM)) {
 276                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 277                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 278                 GOTO(out, rc = -EACCES);
 279         }
 280
 281         /* check permission of setgid */
 282         if (setgid && !(perm & CFS_SETGID_PERM)) {
 283                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 284                        "from %s\n", pud->pud_uid, pud->pud_gid,
 285                        pud->pud_fsuid, pud->pud_fsgid,
 286                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 287                 GOTO(out, rc = -EACCES);
 288         }
 289
 290         if (perm & CFS_SETGRP_PERM) {
 291                 /* only set groups if GID is not squashed */
 292                 if (pud->pud_ngroups && !is_nm_gid_squashed) {
 293                         /* setgroups for local client */
 294                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 295                         if (!ucred->uc_ginfo) {
 296                                 CERROR("failed to alloc %d groups\n",
 297                                        pud->pud_ngroups);
 298                                 GOTO(out, rc = -ENOMEM);
 299                         }
 300
 301                         lustre_groups_from_list(ucred->uc_ginfo,
 302                                                 pud->pud_groups);
 303                         lustre_groups_sort(ucred->uc_ginfo);
 304                 } else {
 305                         ucred->uc_suppgids[0] = -1;
 306                         ucred->uc_suppgids[1] = -1;
 307                         ucred->uc_ginfo = NULL;
 308                 }
 309         } else {
 310                 ucred->uc_suppgids[0] = -1;
 311                 ucred->uc_suppgids[1] = -1;
 312                 ucred->uc_ginfo = NULL;
 313         }
 314
 315         ucred->uc_uid = pud->pud_uid;
 316         ucred->uc_gid = pud->pud_gid;
 317
 318         ucred->uc_cap = CAP_EMPTY_SET;
 319         if (!nodemap || ucred->uc_o_uid != nodemap->nm_squash_uid)
 320                 ucred->uc_cap.cap[0] = pud->pud_cap;
 321
 322         ucred->uc_fsuid = pud->pud_fsuid;
 323         ucred->uc_fsgid = pud->pud_fsgid;
 324
 325         /* process root_squash here. */
 326         mdt_root_squash(info, peernid);
 327
 328         ucred->uc_valid = UCRED_NEW;
 329         ucred_set_jobid(info, ucred);
 330         ucred_set_nid(info, ucred);
 331         ucred_set_audit_enabled(info, ucred);
 332
 333         EXIT;
 334
 335 out:
 336         if (rc) {
 337                 if (ucred->uc_ginfo) {
 338                         put_group_info(ucred->uc_ginfo);
 339                         ucred->uc_ginfo = NULL;
 340                 }
 341                 if (ucred->uc_identity) {
 342                         mdt_identity_put(mdt->mdt_identity_cache,
 343                                          ucred->uc_identity);
 344                         ucred->uc_identity = NULL;
 345                 }
 346         }
 347
 348         return rc;
 349 }
 350
 351 /**
 352  * Check whether allow the client to set supplementary group IDs or not.
 353  *
 354  * \param[in] info      pointer to the thread context
 355  * \param[in] uc        pointer to the RPC user descriptor
 356  *
 357  * \retval              true if allow to set supplementary group IDs
 358  * \retval              false for other cases
 359  */
 360 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 361 {
 362         __u32 perm;
 363
 364         /* 1. If identity_upcall is disabled,
 365          *    permit local client to do anything. */
 366         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 367                 return true;
 368
 369         /* 2. If fail to get related identities, then forbid any client to
 370          *    set supplementary group IDs. */
 371         if (uc->uc_identity == NULL)
 372                 return false;
 373
 374         /* 3. Check the permission in the identities. */
 375         perm = mdt_identity_get_perm(uc->uc_identity,
 376                                      mdt_info_req(info)->rq_peer.nid);
 377         if (perm & CFS_SETGRP_PERM)
 378                 return true;
 379
 380         return false;
 381 }
 382
 383 int mdt_check_ucred(struct mdt_thread_info *info)
 384 {
 385         struct ptlrpc_request   *req = mdt_info_req(info);
 386         struct mdt_device       *mdt = info->mti_mdt;
 387         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 388         struct lu_ucred         *ucred = mdt_ucred(info);
 389         struct md_identity      *identity = NULL;
 390         lnet_nid_t               peernid = req->rq_peer.nid;
 391         __u32                    perm = 0;
 392         int                      setuid;
 393         int                      setgid;
 394         int                      rc = 0;
 395
 396         ENTRY;
 397
 398         LASSERT(ucred != NULL);
 399         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 400                 RETURN(0);
 401
 402         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 403                 RETURN(0);
 404
 405         /* sanity check: if we use strong authentication, we expect the
 406          * uid which client claimed is true */
 407         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 408             req->rq_auth_uid != pud->pud_uid) {
 409                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 410                        "while client claims %u:%u/%u:%u\n",
 411                        libcfs_nid2str(peernid), req->rq_auth_uid,
 412                        pud->pud_uid, pud->pud_gid,
 413                        pud->pud_fsuid, pud->pud_fsgid);
 414                 RETURN(-EACCES);
 415         }
 416
 417         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 418                 RETURN(0);
 419
 420         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 421         if (IS_ERR(identity)) {
 422                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 423                         RETURN(0);
 424                 } else {
 425                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 426                                pud->pud_uid);
 427                         RETURN(-EACCES);
 428                 }
 429         }
 430
 431         perm = mdt_identity_get_perm(identity, peernid);
 432         /* find out the setuid/setgid attempt */
 433         setuid = (pud->pud_uid != pud->pud_fsuid);
 434         setgid = (pud->pud_gid != pud->pud_fsgid ||
 435                   pud->pud_gid != identity->mi_gid);
 436
 437         /* check permission of setuid */
 438         if (setuid && !(perm & CFS_SETUID_PERM)) {
 439                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 440                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 441                 GOTO(out, rc = -EACCES);
 442         }
 443
 444         /* check permission of setgid */
 445         if (setgid && !(perm & CFS_SETGID_PERM)) {
 446                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 447                        "from %s\n", pud->pud_uid, pud->pud_gid,
 448                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 449                        libcfs_nid2str(peernid));
 450                 GOTO(out, rc = -EACCES);
 451         }
 452
 453         EXIT;
 454
 455 out:
 456         mdt_identity_put(mdt->mdt_identity_cache, identity);
 457         return rc;
 458 }
 459
 460 static int old_init_ucred_common(struct mdt_thread_info *info,
 461                                  struct lu_nodemap *nodemap)
 462 {
 463         struct lu_ucred         *uc = mdt_ucred(info);
 464         struct mdt_device       *mdt = info->mti_mdt;
 465         struct md_identity      *identity = NULL;
 466
 467         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 468                 /* deny access before we get identity ref */
 469                 if (nodemap->nmf_deny_unknown)
 470                         RETURN(-EACCES);
 471
 472                 uc->uc_cap = CAP_EMPTY_SET;
 473                 uc->uc_suppgids[0] = -1;
 474                 uc->uc_suppgids[1] = -1;
 475         }
 476
 477         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 478                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 479                                             uc->uc_fsuid);
 480                 if (IS_ERR(identity)) {
 481                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 482                                      cap_raised(uc->uc_cap,
 483                                                 CAP_DAC_READ_SEARCH))) {
 484                                 identity = NULL;
 485                         } else {
 486                                 CDEBUG(D_SEC, "Deny access without identity: "
 487                                        "uid %u\n", uc->uc_fsuid);
 488                                 RETURN(-EACCES);
 489                         }
 490                 }
 491         }
 492         uc->uc_identity = identity;
 493
 494         /* process root_squash here. */
 495         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 496
 497         uc->uc_valid = UCRED_OLD;
 498         ucred_set_jobid(info, uc);
 499         ucred_set_nid(info, uc);
 500         ucred_set_audit_enabled(info, uc);
 501
 502         EXIT;
 503
 504         return 0;
 505 }
 506
 507 static int old_init_ucred(struct mdt_thread_info *info,
 508                           struct mdt_body *body)
 509 {
 510         struct lu_ucred *uc = mdt_ucred(info);
 511         struct lu_nodemap *nodemap;
 512         int rc;
 513         ENTRY;
 514
 515         nodemap = nodemap_get_from_exp(info->mti_exp);
 516         if (IS_ERR(nodemap))
 517                 RETURN(PTR_ERR(nodemap));
 518
 519         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 520                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 521         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 522                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 523         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 524                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 525         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 526                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 527
 528         LASSERT(uc != NULL);
 529         uc->uc_valid = UCRED_INVALID;
 530         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 531         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 532         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 533         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 534         uc->uc_suppgids[0] = body->mbo_suppgid;
 535         uc->uc_suppgids[1] = -1;
 536         uc->uc_ginfo = NULL;
 537         uc->uc_cap = CAP_EMPTY_SET;
 538         uc->uc_cap.cap[0] = body->mbo_capability;
 539
 540         rc = old_init_ucred_common(info, nodemap);
 541         nodemap_putref(nodemap);
 542
 543         RETURN(rc);
 544 }
 545
 546 static int old_init_ucred_reint(struct mdt_thread_info *info)
 547 {
 548         struct lu_ucred *uc = mdt_ucred(info);
 549         struct lu_nodemap *nodemap;
 550         int rc;
 551         ENTRY;
 552
 553         nodemap = nodemap_get_from_exp(info->mti_exp);
 554         if (IS_ERR(nodemap))
 555                 RETURN(PTR_ERR(nodemap));
 556
 557         LASSERT(uc != NULL);
 558
 559         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 560                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 561         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 562                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 563
 564         uc->uc_valid = UCRED_INVALID;
 565         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 566         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 567         uc->uc_ginfo = NULL;
 568
 569         rc = old_init_ucred_common(info, nodemap);
 570         nodemap_putref(nodemap);
 571
 572         RETURN(rc);
 573 }
 574
 575 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 576                                    struct mdt_body *body)
 577 {
 578         struct ptlrpc_request   *req = mdt_info_req(info);
 579         struct lu_ucred         *uc  = mdt_ucred(info);
 580
 581         LASSERT(uc != NULL);
 582         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 583                 return 0;
 584
 585         mdt_exit_ucred(info);
 586
 587         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 588                 return old_init_ucred(info, body);
 589         else
 590                 return new_init_ucred(info, BODY_INIT, body);
 591 }
 592
 593 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 594 {
 595         return __mdt_init_ucred(info, body);
 596 }
 597
 598 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 599 {
 600         struct ptlrpc_request *req = mdt_info_req(info);
 601         struct lu_ucred       *uc  = mdt_ucred(info);
 602         struct md_attr        *ma  = &info->mti_attr;
 603
 604         LASSERT(uc != NULL);
 605         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 606                 return 0;
 607
 608         /* LU-5564: for normal close request, skip permission check */
 609         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 610             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP))) {
 611                 cap_raise_nfsd_set(uc->uc_cap, CAP_FULL_SET);
 612                 cap_raise_fs_set(uc->uc_cap, CAP_FULL_SET);
 613         }
 614
 615         mdt_exit_ucred(info);
 616
 617         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 618                 return old_init_ucred_reint(info);
 619         else
 620                 return new_init_ucred(info, REC_INIT, NULL);
 621 }
 622
 623 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 624 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 625 {
 626         const struct lov_ost_data_v1 *lod;
 627         __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
 628         __u16 count;
 629         int i;
 630
 631         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 632                 return;
 633
 634         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 635                POSTID(&lmm->lmm_oi), lmm_magic,
 636                le32_to_cpu(lmm->lmm_pattern));
 637
 638         /* No support for compount layouts yet */
 639         if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
 640                 return;
 641
 642         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 643         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 644                le32_to_cpu(lmm->lmm_stripe_size), count);
 645
 646         /* If it's a directory or a released file, then there are
 647          * no actual objects to print, so bail out. */
 648         if (valid & OBD_MD_FLDIREA ||
 649             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 650                 return;
 651
 652         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 653         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 654                 struct ost_id oi;
 655
 656                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 657                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 658                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 659         }
 660 }
 661
 662 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 663 {
 664         const struct lmv_mds_md_v1 *lmm1;
 665         const struct lmv_foreign_md *lfm;
 666         int                        i;
 667
 668         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 669                 return;
 670
 671         /* foreign LMV case */
 672         lfm = &lmv->lmv_foreign_md;
 673         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
 674                 CDEBUG_LIMIT(level,
 675                              "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
 676                              le32_to_cpu(lfm->lfm_magic),
 677                              le32_to_cpu(lfm->lfm_length),
 678                              le32_to_cpu(lfm->lfm_type),
 679                              le32_to_cpu(lfm->lfm_flags),
 680                              le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
 681                 return;
 682         }
 683
 684         lmm1 = &lmv->lmv_md_v1;
 685         CDEBUG(level,
 686                "magic 0x%08X, master %#X stripe_count %d hash_type %#x\n",
 687                le32_to_cpu(lmm1->lmv_magic),
 688                le32_to_cpu(lmm1->lmv_master_mdt_index),
 689                le32_to_cpu(lmm1->lmv_stripe_count),
 690                le32_to_cpu(lmm1->lmv_hash_type));
 691
 692         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 693                 return;
 694
 695         if (le32_to_cpu(lmm1->lmv_stripe_count) > LMV_MAX_STRIPE_COUNT)
 696                 return;
 697
 698         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 699                 struct lu_fid fid;
 700
 701                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 702                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 703         }
 704 }
 705
 706 /* Shrink and/or grow reply buffers */
 707 int mdt_fix_reply(struct mdt_thread_info *info)
 708 {
 709         struct req_capsule *pill = info->mti_pill;
 710         struct mdt_body    *body;
 711         int                md_size, md_packed = 0;
 712         int                acl_size;
 713         int                rc = 0;
 714         ENTRY;
 715
 716         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 717         LASSERT(body != NULL);
 718
 719         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 720                                OBD_MD_LINKNAME))
 721                 md_size = body->mbo_eadatasize;
 722         else
 723                 md_size = 0;
 724
 725         acl_size = body->mbo_aclsize;
 726
 727         /* this replay - not send info to client */
 728         if (info->mti_spec.no_create) {
 729                 md_size = 0;
 730                 acl_size = 0;
 731         }
 732
 733         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 734                md_size, acl_size);
 735 /*
 736             &RMF_MDT_BODY,
 737             &RMF_MDT_MD,
 738             &RMF_ACL, or &RMF_LOGCOOKIES
 739 (optional)  &RMF_CAPA1,
 740 (optional)  &RMF_CAPA2,
 741 (optional)  something else
 742 */
 743
 744         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 745          * buffers before growing it */
 746         if (info->mti_big_lmm_used) {
 747                 /* big_lmm buffer may be used even without packing the result
 748                  * into reply, just for internal server needs */
 749                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 750                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 751                                                          RCL_SERVER);
 752
 753                 /* free big lmm if md_size is not needed */
 754                 if (md_size == 0 || md_packed == 0) {
 755                         info->mti_big_lmm_used = 0;
 756                 } else {
 757                         /* buffer must be allocated separately */
 758                         LASSERT(info->mti_attr.ma_lmm !=
 759                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 760                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 761                 }
 762         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 763                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 764         }
 765
 766         if (info->mti_big_acl_used) {
 767                 if (acl_size == 0)
 768                         info->mti_big_acl_used = 0;
 769                 else
 770                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 771         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 772                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 773         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 774                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 775         }
 776
 777         /* Shrink optional SECCTX buffer if it is not used */
 778         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
 779             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
 780             !(body->mbo_valid & OBD_MD_SECCTX))
 781                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
 782
 783         /* Shrink optional ENCCTX buffer if it is not used */
 784         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
 785             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
 786             !(body->mbo_valid & OBD_MD_ENCCTX))
 787                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
 788
 789         /*
 790          * Some more field should be shrinked if needed.
 791          * This should be done by those who added fields to reply message.
 792          */
 793
 794         /* Grow MD buffer if needed finally */
 795         if (info->mti_big_lmm_used) {
 796                 void *lmm;
 797
 798                 LASSERT(md_size > md_packed);
 799                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 800                        md_size - md_packed);
 801                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 802                 if (rc) {
 803                         /* we can't answer with proper LOV EA, drop flags,
 804                          * the rc is also returned so this request is
 805                          * considered as failed */
 806                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 807                         /* don't return transno along with error */
 808                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 809                 } else {
 810                         /* now we need to pack right LOV/LMV EA */
 811                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 812                         if (info->mti_attr.ma_valid & MA_LOV) {
 813                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 814                                                              RCL_SERVER) ==
 815                                                 info->mti_attr.ma_lmm_size);
 816                                 memcpy(lmm, info->mti_attr.ma_lmm,
 817                                        info->mti_attr.ma_lmm_size);
 818                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 819                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 820                                                              RCL_SERVER) ==
 821                                                 info->mti_attr.ma_lmv_size);
 822                                 memcpy(lmm, info->mti_attr.ma_lmv,
 823                                        info->mti_attr.ma_lmv_size);
 824                         }
 825                 }
 826
 827                 /* update mdt_max_mdsize so clients will be aware about that */
 828                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 829                         info->mti_mdt->mdt_max_mdsize =
 830                                                 info->mti_attr.ma_lmm_size;
 831                 info->mti_big_lmm_used = 0;
 832         }
 833
 834         if (info->mti_big_acl_used) {
 835                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 836                        acl_size);
 837
 838                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 839                 if (rc) {
 840                         body->mbo_valid &= ~OBD_MD_FLACL;
 841                 } else {
 842                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 843
 844                         memcpy(acl, info->mti_big_acl, acl_size);
 845                 }
 846
 847                 info->mti_big_acl_used = 0;
 848         }
 849
 850         RETURN(rc);
 851 }
 852
 853
 854 /* if object is dying, pack the lov/llog data,
 855  * parameter info->mti_attr should be valid at this point!
 856  * Also implements RAoLU policy */
 857 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 858                            struct md_attr *ma)
 859 {
 860         struct mdt_body *repbody = NULL;
 861         const struct lu_attr *la = &ma->ma_attr;
 862         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 863         int rc;
 864         __u64 need = 0;
 865         struct hsm_action_item hai = {
 866                 .hai_len = sizeof(hai),
 867                 .hai_action = HSMA_REMOVE,
 868                 .hai_extent.length = -1,
 869                 .hai_cookie = 0,
 870                 .hai_gid = 0,
 871         };
 872         int archive_id;
 873
 874         ENTRY;
 875
 876         if (mdt_info_req(info) != NULL) {
 877                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 878                 LASSERT(repbody != NULL);
 879         } else {
 880                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 881         }
 882
 883         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 884                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 885
 886         if (ma->ma_valid & MA_LOV) {
 887                 CERROR("No need in LOV EA upon unlink\n");
 888                 dump_stack();
 889         }
 890         if (repbody != NULL)
 891                 repbody->mbo_eadatasize = 0;
 892
 893         /* Only check unlinked and archived if RAoLU and upon last close */
 894         if (!cdt->cdt_remove_archive_on_last_unlink ||
 895             atomic_read(&mo->mot_open_count) != 0)
 896                 RETURN(0);
 897
 898         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 899         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 900                 RETURN(0);
 901
 902         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 903                 RETURN(0);
 904
 905         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 906         if (need != 0) {
 907                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 908                  * try setting them */
 909                 ma->ma_need |= need;
 910                 rc = mdt_attr_get_complex(info, mo, ma);
 911                 if (rc) {
 912                         CERROR("%s: unable to fetch missing attributes of"
 913                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 914                                PFID(mdt_object_fid(mo)), rc);
 915                         RETURN(0);
 916                 }
 917
 918                 if (need & MA_INODE) {
 919                         if (ma->ma_valid & MA_INODE) {
 920                                 if (ma->ma_attr.la_nlink != 0)
 921                                         RETURN(0);
 922                         } else {
 923                                 RETURN(0);
 924                         }
 925                 }
 926
 927                 if (need & MA_HSM) {
 928                         if (ma->ma_valid & MA_HSM) {
 929                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 930                                         RETURN(0);
 931                         } else {
 932                                 RETURN(0);
 933                         }
 934                 }
 935         }
 936
 937         /* RAoLU policy is active, last close on file has occured,
 938          * file is unlinked, file is archived, so create remove request
 939          * for copytool!
 940          * If CDT is not running, requests will be logged for later. */
 941         if (ma->ma_hsm.mh_arch_id != 0)
 942                 archive_id = ma->ma_hsm.mh_arch_id;
 943         else
 944                 archive_id = cdt->cdt_default_archive_id;
 945
 946         hai.hai_fid = *mdt_object_fid(mo);
 947
 948         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
 949                                   &hai);
 950         if (rc)
 951                 CERROR("%s: unable to add HSM remove request for "DFID
 952                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
 953                        PFID(mdt_object_fid(mo)), rc);
 954
 955         RETURN(0);
 956 }
 957
 958 static __u64 mdt_attr_valid_xlate(enum mds_attr_flags in,
 959                                   struct mdt_reint_record *rr,
 960                                   struct md_attr *ma)
 961 {
 962         __u64 out;
 963
 964         out = 0;
 965         if (in & MDS_ATTR_MODE)
 966                 out |= LA_MODE;
 967         if (in & MDS_ATTR_UID)
 968                 out |= LA_UID;
 969         if (in & MDS_ATTR_GID)
 970                 out |= LA_GID;
 971         if (in & MDS_ATTR_SIZE)
 972                 out |= LA_SIZE;
 973         if (in & MDS_ATTR_BLOCKS)
 974                 out |= LA_BLOCKS;
 975         if (in & MDS_ATTR_ATIME_SET)
 976                 out |= LA_ATIME;
 977         if (in & MDS_ATTR_CTIME_SET)
 978                 out |= LA_CTIME;
 979         if (in & MDS_ATTR_MTIME_SET)
 980                 out |= LA_MTIME;
 981         if (in & MDS_ATTR_ATTR_FLAG)
 982                 out |= LA_FLAGS;
 983         if (in & MDS_ATTR_KILL_SUID)
 984                 out |= LA_KILL_SUID;
 985         if (in & MDS_ATTR_KILL_SGID)
 986                 out |= LA_KILL_SGID;
 987         if (in & MDS_ATTR_PROJID)
 988                 out |= LA_PROJID;
 989         if (in & MDS_ATTR_LSIZE)
 990                 out |= LA_LSIZE;
 991         if (in & MDS_ATTR_LBLOCKS)
 992                 out |= LA_LBLOCKS;
 993
 994         if (in & MDS_ATTR_FROM_OPEN)
 995                 rr->rr_flags |= MRF_OPEN_TRUNC;
 996         if (in & MDS_ATTR_OVERRIDE)
 997                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 998         if (in & MDS_ATTR_FORCE)
 999                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1000
1001         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
1002                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
1003                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
1004                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1005                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1006                 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1007                 MDS_ATTR_OVERRIDE);
1008         if (in != 0)
1009                 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", (u64)in);
1010
1011         return out;
1012 }
1013
1014 /* unpacking */
1015
1016 int mdt_name_unpack(struct req_capsule *pill,
1017                     const struct req_msg_field *field,
1018                     struct lu_name *ln,
1019                     enum mdt_name_flags flags)
1020 {
1021         ln->ln_name = req_capsule_client_get(pill, field);
1022         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1023
1024         if (!lu_name_is_valid(ln)) {
1025                 ln->ln_name = NULL;
1026                 ln->ln_namelen = 0;
1027
1028                 return -EPROTO;
1029         }
1030
1031         if ((flags & MNF_FIX_ANON) &&
1032             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1033                 /* Newer (3.x) kernels use a name of "/" for the
1034                  * "anonymous" disconnected dentries from NFS
1035                  * filehandle conversion. See d_obtain_alias(). */
1036                 ln->ln_name = NULL;
1037                 ln->ln_namelen = 0;
1038         }
1039
1040         return 0;
1041 }
1042
1043 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1044                                   const char **secctx_name,
1045                                   void **secctx, size_t *secctx_size)
1046 {
1047         const char *name;
1048         size_t name_size;
1049
1050         *secctx_name = NULL;
1051         *secctx = NULL;
1052         *secctx_size = 0;
1053
1054         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1055             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1056                 return 0;
1057
1058         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1059                                          RCL_CLIENT);
1060         if (name_size == 0)
1061                 return 0;
1062
1063         if (name_size > XATTR_NAME_MAX + 1)
1064                 return -EPROTO;
1065
1066         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1067         if (strnlen(name, name_size) != name_size - 1)
1068                 return -EPROTO;
1069
1070         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1071             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1072                 return -EPROTO;
1073
1074         *secctx_name = name;
1075         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1076         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1077
1078         return 0;
1079 }
1080
1081 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1082                                   void **encctx, size_t *encctx_size)
1083 {
1084         *encctx = NULL;
1085         *encctx_size = 0;
1086
1087         if (!exp_connect_encrypt(pill->rc_req->rq_export))
1088                 return 0;
1089
1090         if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1091             !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1092                 return -EPROTO;
1093
1094         *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1095         if (*encctx_size == 0)
1096                 return 0;
1097
1098         *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1099
1100         return 0;
1101 }
1102
1103 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1104 {
1105         struct lu_ucred *uc = mdt_ucred(info);
1106         struct md_attr *ma = &info->mti_attr;
1107         struct lu_attr *la = &ma->ma_attr;
1108         struct req_capsule *pill = info->mti_pill;
1109         struct mdt_reint_record *rr = &info->mti_rr;
1110         struct mdt_rec_setattr *rec;
1111         struct lu_nodemap *nodemap;
1112
1113         ENTRY;
1114
1115         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1116         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1117         if (rec == NULL)
1118                 RETURN(-EFAULT);
1119
1120         /* This prior initialization is needed for old_init_ucred_reint() */
1121         uc->uc_fsuid = rec->sa_fsuid;
1122         uc->uc_fsgid = rec->sa_fsgid;
1123         uc->uc_cap = CAP_EMPTY_SET;
1124         uc->uc_cap.cap[0] = rec->sa_cap;
1125         uc->uc_suppgids[0] = rec->sa_suppgid;
1126         uc->uc_suppgids[1] = -1;
1127
1128         rr->rr_fid1 = &rec->sa_fid;
1129         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1130         la->la_mode  = rec->sa_mode;
1131         la->la_flags = rec->sa_attr_flags;
1132
1133         nodemap = nodemap_get_from_exp(info->mti_exp);
1134         if (IS_ERR(nodemap))
1135                 RETURN(PTR_ERR(nodemap));
1136
1137         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1138                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1139         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1140                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1141         la->la_projid = rec->sa_projid;
1142         nodemap_putref(nodemap);
1143
1144         la->la_size  = rec->sa_size;
1145         la->la_blocks = rec->sa_blocks;
1146         la->la_ctime = rec->sa_ctime;
1147         la->la_atime = rec->sa_atime;
1148         la->la_mtime = rec->sa_mtime;
1149         ma->ma_valid = MA_INODE;
1150
1151         ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1152                                 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1153                                 MDS_PCC_ATTACH);
1154         RETURN(0);
1155 }
1156
1157 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1158 {
1159         struct req_capsule *pill = info->mti_pill;
1160         struct mdt_ioepoch *ioepoch;
1161         ENTRY;
1162
1163         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1164                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1165         else
1166                 ioepoch = NULL;
1167
1168         if (ioepoch == NULL)
1169                 RETURN(-EPROTO);
1170
1171         info->mti_open_handle = ioepoch->mio_open_handle;
1172
1173         RETURN(0);
1174 }
1175
1176 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1177         struct req_capsule      *pill = info->mti_pill;
1178
1179         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1180                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1181                 if (info->mti_dlm_req == NULL)
1182                         RETURN(-EFAULT);
1183         }
1184
1185         RETURN(0);
1186 }
1187
1188 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1189 {
1190         struct mdt_reint_record *rr = &info->mti_rr;
1191         struct md_attr          *ma = &info->mti_attr;
1192         struct req_capsule      *pill = info->mti_pill;
1193         int rc;
1194         ENTRY;
1195
1196         rc = mdt_setattr_unpack_rec(info);
1197         if (rc)
1198                 RETURN(rc);
1199
1200         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1201                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1202                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1203                                                         RCL_CLIENT);
1204
1205                 if (rr->rr_eadatalen > 0) {
1206                         const struct lmv_user_md        *lum;
1207
1208                         lum = rr->rr_eadata;
1209                         /* Sigh ma_valid(from req) does not indicate whether
1210                          * it will set LOV/LMV EA, so we have to check magic */
1211                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1212                                 ma->ma_valid |= MA_LMV;
1213                                 ma->ma_lmv = (void *)rr->rr_eadata;
1214                                 ma->ma_lmv_size = rr->rr_eadatalen;
1215                         } else {
1216                                 ma->ma_valid |= MA_LOV;
1217                                 ma->ma_lmm = (void *)rr->rr_eadata;
1218                                 ma->ma_lmm_size = rr->rr_eadatalen;
1219                         }
1220                 }
1221         }
1222
1223         rc = mdt_dlmreq_unpack(info);
1224         RETURN(rc);
1225 }
1226
1227 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1228 {
1229         struct md_attr          *ma = &info->mti_attr;
1230         struct req_capsule      *pill = info->mti_pill;
1231         ENTRY;
1232
1233         if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1234                 RETURN(0);
1235
1236         req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1237
1238         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1239             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1240                 RETURN(-EFAULT);
1241
1242         RETURN(0);
1243 }
1244
1245 int mdt_close_unpack(struct mdt_thread_info *info)
1246 {
1247         int rc;
1248         ENTRY;
1249
1250         rc = mdt_close_handle_unpack(info);
1251         if (rc)
1252                 RETURN(rc);
1253
1254         rc = mdt_setattr_unpack_rec(info);
1255         if (rc)
1256                 RETURN(rc);
1257
1258         rc = mdt_close_intent_unpack(info);
1259         if (rc)
1260                 RETURN(rc);
1261
1262         RETURN(mdt_init_ucred_reint(info));
1263 }
1264
1265 static int mdt_create_unpack(struct mdt_thread_info *info)
1266 {
1267         struct lu_ucred *uc  = mdt_ucred(info);
1268         struct mdt_rec_create *rec;
1269         struct lu_attr *attr = &info->mti_attr.ma_attr;
1270         struct mdt_reint_record *rr = &info->mti_rr;
1271         struct req_capsule      *pill = info->mti_pill;
1272         struct md_op_spec       *sp = &info->mti_spec;
1273         int rc;
1274
1275         ENTRY;
1276
1277         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1278         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1279         if (rec == NULL)
1280                 RETURN(-EFAULT);
1281
1282         /* This prior initialization is needed for old_init_ucred_reint() */
1283         uc->uc_fsuid = rec->cr_fsuid;
1284         uc->uc_fsgid = rec->cr_fsgid;
1285         uc->uc_cap = CAP_EMPTY_SET;
1286         uc->uc_cap.cap[0] = rec->cr_cap;
1287         uc->uc_suppgids[0] = rec->cr_suppgid1;
1288         uc->uc_suppgids[1] = -1;
1289         uc->uc_umask = rec->cr_umask;
1290
1291         rr->rr_fid1 = &rec->cr_fid1;
1292         rr->rr_fid2 = &rec->cr_fid2;
1293         attr->la_mode = rec->cr_mode;
1294         attr->la_rdev  = rec->cr_rdev;
1295         attr->la_uid   = rec->cr_fsuid;
1296         attr->la_gid   = rec->cr_fsgid;
1297         attr->la_ctime = rec->cr_time;
1298         attr->la_mtime = rec->cr_time;
1299         attr->la_atime = rec->cr_time;
1300         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1301                 LA_CTIME | LA_MTIME | LA_ATIME;
1302         memset(&sp->u, 0, sizeof(sp->u));
1303         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1304
1305         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1306         if (rc < 0)
1307                 RETURN(rc);
1308
1309         if (S_ISLNK(attr->la_mode)) {
1310                 const char *tgt = NULL;
1311                 int sz;
1312
1313                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1314                 sz = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT);
1315                 if (sz) {
1316                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1317                         sp->u.sp_symname.ln_name = tgt;
1318                         sp->u.sp_symname.ln_namelen = sz - 1; /* skip NUL */
1319                 }
1320                 if (tgt == NULL)
1321                         RETURN(-EFAULT);
1322         } else {
1323                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1324                 if (S_ISDIR(attr->la_mode) &&
1325                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1326                         sp->u.sp_ea.eadata =
1327                                 req_capsule_client_get(pill, &RMF_EADATA);
1328                         sp->u.sp_ea.eadatalen =
1329                                 req_capsule_get_size(pill, &RMF_EADATA,
1330                                                      RCL_CLIENT);
1331                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1332                 }
1333         }
1334
1335         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1336                                     &sp->sp_cr_file_secctx,
1337                                     &sp->sp_cr_file_secctx_size);
1338         if (rc < 0)
1339                 RETURN(rc);
1340
1341         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1342                                     &sp->sp_cr_file_encctx_size);
1343         if (rc < 0)
1344                 RETURN(rc);
1345
1346         rc = req_check_sepol(pill);
1347         if (rc)
1348                 RETURN(rc);
1349
1350         rc = mdt_dlmreq_unpack(info);
1351         RETURN(rc);
1352 }
1353
1354 static int mdt_link_unpack(struct mdt_thread_info *info)
1355 {
1356         struct lu_ucred *uc  = mdt_ucred(info);
1357         struct mdt_rec_link *rec;
1358         struct lu_attr *attr = &info->mti_attr.ma_attr;
1359         struct mdt_reint_record *rr = &info->mti_rr;
1360         struct req_capsule *pill = info->mti_pill;
1361         int rc;
1362
1363         ENTRY;
1364
1365         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1366         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1367         if (rec == NULL)
1368                 RETURN(-EFAULT);
1369
1370         /* This prior initialization is needed for old_init_ucred_reint() */
1371         uc->uc_fsuid = rec->lk_fsuid;
1372         uc->uc_fsgid = rec->lk_fsgid;
1373         uc->uc_cap = CAP_EMPTY_SET;
1374         uc->uc_cap.cap[0] = rec->lk_cap;
1375         uc->uc_suppgids[0] = rec->lk_suppgid1;
1376         uc->uc_suppgids[1] = rec->lk_suppgid2;
1377
1378         attr->la_uid = rec->lk_fsuid;
1379         attr->la_gid = rec->lk_fsgid;
1380         rr->rr_fid1 = &rec->lk_fid1;
1381         rr->rr_fid2 = &rec->lk_fid2;
1382         attr->la_ctime = rec->lk_time;
1383         attr->la_mtime = rec->lk_time;
1384         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1385
1386         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1387         if (rc < 0)
1388                 RETURN(rc);
1389
1390         rc = req_check_sepol(pill);
1391         if (rc)
1392                 RETURN(rc);
1393
1394         rc = mdt_dlmreq_unpack(info);
1395
1396         RETURN(rc);
1397 }
1398
1399 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1400 {
1401         struct lu_ucred *uc  = mdt_ucred(info);
1402         struct mdt_rec_unlink *rec;
1403         struct lu_attr *attr = &info->mti_attr.ma_attr;
1404         struct mdt_reint_record *rr = &info->mti_rr;
1405         struct req_capsule      *pill = info->mti_pill;
1406         int rc;
1407
1408         ENTRY;
1409
1410         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1411         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1412         if (rec == NULL)
1413                 RETURN(-EFAULT);
1414
1415         /* This prior initialization is needed for old_init_ucred_reint() */
1416         uc->uc_fsuid = rec->ul_fsuid;
1417         uc->uc_fsgid = rec->ul_fsgid;
1418         uc->uc_cap = CAP_EMPTY_SET;
1419         uc->uc_cap.cap[0] = rec->ul_cap;
1420         uc->uc_suppgids[0] = rec->ul_suppgid1;
1421         uc->uc_suppgids[1] = -1;
1422
1423         attr->la_uid = rec->ul_fsuid;
1424         attr->la_gid = rec->ul_fsgid;
1425         rr->rr_fid1 = &rec->ul_fid1;
1426         rr->rr_fid2 = &rec->ul_fid2;
1427         attr->la_ctime = rec->ul_time;
1428         attr->la_mtime = rec->ul_time;
1429         attr->la_mode  = rec->ul_mode;
1430         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1431         if (rec->ul_bias & MDS_FID_OP)
1432                 info->mti_spec.sp_cr_flags |= MDS_OP_WITH_FID;
1433         else
1434                 info->mti_spec.sp_cr_flags &= ~MDS_OP_WITH_FID;
1435
1436         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1437         if (rc < 0)
1438                 RETURN(rc);
1439
1440         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1441
1442         rc = req_check_sepol(pill);
1443         if (rc)
1444                 RETURN(rc);
1445
1446         rc = mdt_dlmreq_unpack(info);
1447         RETURN(rc);
1448 }
1449
1450 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1451 {
1452         info->mti_spec.sp_rm_entry = 1;
1453         return mdt_unlink_unpack(info);
1454 }
1455
1456 static int mdt_rename_unpack(struct mdt_thread_info *info)
1457 {
1458         struct lu_ucred *uc = mdt_ucred(info);
1459         struct mdt_rec_rename *rec;
1460         struct lu_attr *attr = &info->mti_attr.ma_attr;
1461         struct mdt_reint_record *rr = &info->mti_rr;
1462         struct req_capsule *pill = info->mti_pill;
1463         struct md_op_spec *spec = &info->mti_spec;
1464         int rc;
1465
1466         ENTRY;
1467
1468         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1469         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1470         if (rec == NULL)
1471                 RETURN(-EFAULT);
1472
1473         /* This prior initialization is needed for old_init_ucred_reint() */
1474         uc->uc_fsuid = rec->rn_fsuid;
1475         uc->uc_fsgid = rec->rn_fsgid;
1476         uc->uc_cap = CAP_EMPTY_SET;
1477         uc->uc_cap.cap[0] = rec->rn_cap;
1478         uc->uc_suppgids[0] = rec->rn_suppgid1;
1479         uc->uc_suppgids[1] = rec->rn_suppgid2;
1480
1481         attr->la_uid = rec->rn_fsuid;
1482         attr->la_gid = rec->rn_fsgid;
1483         rr->rr_fid1 = &rec->rn_fid1;
1484         rr->rr_fid2 = &rec->rn_fid2;
1485         attr->la_ctime = rec->rn_time;
1486         attr->la_mtime = rec->rn_time;
1487         /* rename_tgt contains the mode already */
1488         attr->la_mode = rec->rn_mode;
1489         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1490
1491         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1492         if (rc < 0)
1493                 RETURN(rc);
1494
1495         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1496         if (rc < 0)
1497                 RETURN(rc);
1498
1499         spec->no_create = !!req_is_replay(mdt_info_req(info));
1500
1501         rc = req_check_sepol(pill);
1502         if (rc)
1503                 RETURN(rc);
1504
1505         rc = mdt_dlmreq_unpack(info);
1506
1507         RETURN(rc);
1508 }
1509
1510 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1511 {
1512         struct lu_ucred *uc = mdt_ucred(info);
1513         struct mdt_rec_rename *rec;
1514         struct lu_attr *attr = &info->mti_attr.ma_attr;
1515         struct mdt_reint_record *rr = &info->mti_rr;
1516         struct req_capsule *pill = info->mti_pill;
1517         struct md_op_spec *spec = &info->mti_spec;
1518         int rc;
1519
1520         ENTRY;
1521
1522         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1523         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1524         if (rec == NULL)
1525                 RETURN(-EFAULT);
1526
1527         /* This prior initialization is needed for old_init_ucred_reint() */
1528         uc->uc_fsuid = rec->rn_fsuid;
1529         uc->uc_fsgid = rec->rn_fsgid;
1530         uc->uc_cap = CAP_EMPTY_SET;
1531         uc->uc_cap.cap[0] = rec->rn_cap;
1532         uc->uc_suppgids[0] = rec->rn_suppgid1;
1533         uc->uc_suppgids[1] = rec->rn_suppgid2;
1534
1535         attr->la_uid = rec->rn_fsuid;
1536         attr->la_gid = rec->rn_fsgid;
1537         rr->rr_fid1 = &rec->rn_fid1;
1538         rr->rr_fid2 = &rec->rn_fid2;
1539         attr->la_ctime = rec->rn_time;
1540         attr->la_mtime = rec->rn_time;
1541         /* rename_tgt contains the mode already */
1542         attr->la_mode = rec->rn_mode;
1543         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1544         spec->sp_cr_flags = 0;
1545
1546         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1547         if (rc < 0)
1548                 RETURN(rc);
1549
1550         if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1551                 rc = mdt_close_handle_unpack(info);
1552                 if (rc)
1553                         RETURN(rc);
1554
1555                 spec->sp_migrate_close = 1;
1556         } else {
1557                 spec->sp_migrate_close = 0;
1558         }
1559         spec->sp_migrate_nsonly = 0;
1560
1561         /* lustre version > 2.11 migration packs lum */
1562         if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1563                 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1564                         rr->rr_eadatalen = req_capsule_get_size(pill,
1565                                                                 &RMF_EADATA,
1566                                                                 RCL_CLIENT);
1567
1568                         if (rr->rr_eadatalen > 0) {
1569                                 struct lmv_user_md_v1 *lmu;
1570
1571                                 lmu = req_capsule_client_get(pill, &RMF_EADATA);
1572                                 lmu->lum_hash_type |=
1573                                         cpu_to_le32(LMV_HASH_FLAG_FIXED);
1574                                 rr->rr_eadata = lmu;
1575                                 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1576                                 spec->u.sp_ea.eadata = rr->rr_eadata;
1577                                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1578                         }
1579                 } else {
1580                         /* old client doesn't provide lum. */
1581                         RETURN(-EOPNOTSUPP);
1582                 }
1583         }
1584
1585         spec->no_create = !!req_is_replay(mdt_info_req(info));
1586
1587         rc = mdt_dlmreq_unpack(info);
1588
1589         RETURN(rc);
1590 }
1591
1592 /*
1593  * please see comment above LOV_MAGIC_V1_DEFINED
1594  */
1595 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1596 {
1597         struct lov_user_md_v1   *v1 = eadata;
1598
1599         LASSERT(v1);
1600
1601         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1602                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1603                         v1->lmm_magic |= LOV_MAGIC_DEFINED;
1604                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1605                          __swab32(LOV_MAGIC_MAGIC))
1606                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1607         }
1608 }
1609
1610 static int mdt_open_unpack(struct mdt_thread_info *info)
1611 {
1612         struct lu_ucred *uc = mdt_ucred(info);
1613         struct mdt_rec_create *rec;
1614         struct lu_attr *attr = &info->mti_attr.ma_attr;
1615         struct req_capsule *pill = info->mti_pill;
1616         struct mdt_reint_record *rr = &info->mti_rr;
1617         struct ptlrpc_request *req = mdt_info_req(info);
1618         struct md_op_spec *sp = &info->mti_spec;
1619         int rc;
1620         ENTRY;
1621
1622         BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1623                      sizeof(struct mdt_rec_reint));
1624         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1625         if (rec == NULL)
1626                 RETURN(-EFAULT);
1627
1628         /* This prior initialization is needed for old_init_ucred_reint() */
1629         uc->uc_fsuid = rec->cr_fsuid;
1630         uc->uc_fsgid = rec->cr_fsgid;
1631         uc->uc_cap = CAP_EMPTY_SET;
1632         uc->uc_cap.cap[0] = rec->cr_cap;
1633         uc->uc_suppgids[0] = rec->cr_suppgid1;
1634         uc->uc_suppgids[1] = rec->cr_suppgid2;
1635         uc->uc_umask = rec->cr_umask;
1636
1637         rr->rr_fid1   = &rec->cr_fid1;
1638         rr->rr_fid2   = &rec->cr_fid2;
1639         rr->rr_open_handle = &rec->cr_open_handle_old;
1640         attr->la_mode = rec->cr_mode;
1641         attr->la_rdev  = rec->cr_rdev;
1642         attr->la_uid   = rec->cr_fsuid;
1643         attr->la_gid   = rec->cr_fsgid;
1644         attr->la_ctime = rec->cr_time;
1645         attr->la_mtime = rec->cr_time;
1646         attr->la_atime = rec->cr_time;
1647         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1648                          LA_CTIME | LA_MTIME | LA_ATIME;
1649         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1650         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1651         /* Do not trigger ASSERTION if client miss to set such flags. */
1652         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1653                 RETURN(-EPROTO);
1654
1655         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1656
1657         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1658
1659         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1660                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1661                                                         RCL_CLIENT);
1662
1663                 if (rr->rr_eadatalen > 0) {
1664                         rr->rr_eadata = req_capsule_client_get(pill,
1665                                                                &RMF_EADATA);
1666                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1667                         sp->u.sp_ea.eadata = rr->rr_eadata;
1668                         sp->sp_archive_id = rec->cr_archive_id;
1669                         sp->no_create = !!req_is_replay(req);
1670                         mdt_fix_lov_magic(info, rr->rr_eadata);
1671                 }
1672
1673                 /*
1674                  * Client default md_size may be 0 right after client start,
1675                  * until all osc are connected, set here just some reasonable
1676                  * value to prevent misbehavior.
1677                  */
1678                 if (rr->rr_eadatalen == 0 &&
1679                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1680                         rr->rr_eadatalen = MIN_MD_SIZE;
1681         }
1682
1683         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1684                                     &sp->sp_cr_file_secctx,
1685                                     &sp->sp_cr_file_secctx_size);
1686         if (rc < 0)
1687                 RETURN(rc);
1688
1689         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1690                                     &sp->sp_cr_file_encctx_size);
1691         if (rc < 0)
1692                 RETURN(rc);
1693
1694         rc = req_check_sepol(pill);
1695         if (rc)
1696                 RETURN(rc);
1697
1698         RETURN(rc);
1699 }
1700
1701 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1702 {
1703         struct mdt_reint_record *rr = &info->mti_rr;
1704         struct lu_ucred *uc = mdt_ucred(info);
1705         struct lu_attr *attr = &info->mti_attr.ma_attr;
1706         struct req_capsule *pill = info->mti_pill;
1707         struct mdt_rec_setxattr *rec;
1708         int rc;
1709         ENTRY;
1710
1711
1712         BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1713                      sizeof(struct mdt_rec_reint));
1714         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1715         if (rec == NULL)
1716                 RETURN(-EFAULT);
1717
1718         /* This prior initialization is needed for old_init_ucred_reint() */
1719         uc->uc_fsuid  = rec->sx_fsuid;
1720         uc->uc_fsgid  = rec->sx_fsgid;
1721         uc->uc_cap = CAP_EMPTY_SET;
1722         uc->uc_cap.cap[0] = rec->sx_cap;
1723         uc->uc_suppgids[0] = rec->sx_suppgid1;
1724         uc->uc_suppgids[1] = -1;
1725
1726         rr->rr_opcode = rec->sx_opcode;
1727         rr->rr_fid1   = &rec->sx_fid;
1728         attr->la_valid = rec->sx_valid;
1729         attr->la_ctime = rec->sx_time;
1730         attr->la_size = rec->sx_size;
1731         attr->la_flags = rec->sx_flags;
1732
1733         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1734         if (rc < 0)
1735                 RETURN(rc);
1736
1737         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1738                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1739                                                         RCL_CLIENT);
1740
1741                 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1742                         RETURN(-E2BIG);
1743
1744                 if (rr->rr_eadatalen > 0) {
1745                         rr->rr_eadata = req_capsule_client_get(pill,
1746                                                                &RMF_EADATA);
1747                         if (rr->rr_eadata == NULL)
1748                                 RETURN(-EFAULT);
1749                 } else {
1750                         rr->rr_eadata = NULL;
1751                 }
1752         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1753                 CDEBUG(D_INFO, "no xattr data supplied\n");
1754                 RETURN(-EFAULT);
1755         }
1756
1757         rc = req_check_sepol(pill);
1758         if (rc)
1759                 RETURN(rc);
1760
1761         if (mdt_dlmreq_unpack(info) < 0)
1762                 RETURN(-EPROTO);
1763
1764         RETURN(0);
1765 }
1766
1767 static int mdt_resync_unpack(struct mdt_thread_info *info)
1768 {
1769         struct req_capsule      *pill = info->mti_pill;
1770         struct mdt_reint_record *rr   = &info->mti_rr;
1771         struct lu_ucred         *uc     = mdt_ucred(info);
1772         struct mdt_rec_resync   *rec;
1773         ENTRY;
1774
1775         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1776         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1777         if (rec == NULL)
1778                 RETURN(-EFAULT);
1779
1780         /* This prior initialization is needed for old_init_ucred_reint() */
1781         uc->uc_fsuid = rec->rs_fsuid;
1782         uc->uc_fsgid = rec->rs_fsgid;
1783         uc->uc_cap = CAP_EMPTY_SET;
1784         uc->uc_cap.cap[0] = rec->rs_cap;
1785
1786         rr->rr_fid1   = &rec->rs_fid;
1787         rr->rr_mirror_id = rec->rs_mirror_id;
1788
1789         /* cookie doesn't need to be swapped but it has been swapped
1790          * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1791          * restoring. */
1792         if (req_capsule_req_need_swab(pill))
1793                 __swab64s(&rec->rs_lease_handle.cookie);
1794         rr->rr_lease_handle = &rec->rs_lease_handle;
1795
1796         RETURN(mdt_dlmreq_unpack(info));
1797 }
1798
1799 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1800
1801 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1802         [REINT_SETATTR]  = mdt_setattr_unpack,
1803         [REINT_CREATE]   = mdt_create_unpack,
1804         [REINT_LINK]     = mdt_link_unpack,
1805         [REINT_UNLINK]   = mdt_unlink_unpack,
1806         [REINT_RENAME]   = mdt_rename_unpack,
1807         [REINT_OPEN]     = mdt_open_unpack,
1808         [REINT_SETXATTR] = mdt_setxattr_unpack,
1809         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1810         [REINT_MIGRATE]  = mdt_migrate_unpack,
1811         [REINT_RESYNC]   = mdt_resync_unpack,
1812 };
1813
1814 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1815 {
1816         int rc;
1817         ENTRY;
1818
1819         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1820         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1821                 info->mti_rr.rr_opcode = op;
1822                 rc = mdt_reint_unpackers[op](info);
1823         } else {
1824                 CERROR("Unexpected opcode %d\n", op);
1825                 rc = -EFAULT;
1826         }
1827         RETURN(rc);
1828 }
1829
1830 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1831                              struct mdt_object *child)
1832 {
1833         char *secctx_name;
1834         struct lu_buf *buffer;
1835         struct mdt_body *repbody;
1836         struct req_capsule *pill = info->mti_pill;
1837         int rc = 0;
1838
1839         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1840             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1841                 secctx_name =
1842                         req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1843                 buffer = &info->mti_buf;
1844
1845                 /* fill reply buffer with security context now */
1846                 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1847                                                       RCL_SERVER);
1848                 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1849                                                         &RMF_FILE_SECCTX);
1850                 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1851                                   buffer, secctx_name);
1852                 if (rc >= 0) {
1853                         CDEBUG(D_SEC,
1854                                "found security context of size %d for "DFID"\n",
1855                                rc, PFID(mdt_object_fid(child)));
1856
1857                         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1858                         repbody->mbo_valid |= OBD_MD_SECCTX;
1859                         if (rc < buffer->lb_len)
1860                                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1861                                                    RCL_SERVER);
1862                         rc = 0;
1863                 } else {
1864                         CDEBUG(D_SEC,
1865                              "security context not found for "DFID": rc = %d\n",
1866                              PFID(mdt_object_fid(child)), rc);
1867                         req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1868                                            RCL_SERVER);
1869                         /* handling -ENOENT is important because it may change
1870                          * object state in DNE env dropping LOHA_EXISTS flag,
1871                          * it is important to return that to the caller.
1872                          * Check LU-13115 for details.
1873                          */
1874                         if (rc != -ENOENT)
1875                                 rc = 0;
1876                 }
1877         }
1878         return rc;
1879 }
1880
1881 /* check whether two FIDs belong to different MDT. */
1882 static int mdt_fids_different_target(struct mdt_thread_info *info,
1883                                      const struct lu_fid *fid1,
1884                                      const struct lu_fid *fid2)
1885 {
1886         const struct lu_env *env = info->mti_env;
1887         struct mdt_device *mdt = info->mti_mdt;
1888         struct lu_seq_range *range = &info->mti_range;
1889         struct seq_server_site *ss;
1890         __u32 index1, index2;
1891         int rc;
1892
1893         if (fid_seq(fid1) == fid_seq(fid2))
1894                 return 0;
1895
1896         ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1897
1898         range->lsr_flags = LU_SEQ_RANGE_MDT;
1899         rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1900         if (rc)
1901                 return rc;
1902
1903         index1 = range->lsr_index;
1904
1905         rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1906         if (rc)
1907                 return rc;
1908
1909         index2 = range->lsr_index;
1910
1911         return index1 != index2;
1912 }
1913
1914 /**
1915  * Check whether \a child is remote object on \a parent.
1916  *
1917  * \param[in]  info     thread environment
1918  * \param[in]  parent   parent object, it's the same as child object in
1919  *                      getattr_by_fid
1920  * \param[in]  child    child object
1921  *
1922  * \retval 1    is remote object.
1923  * \retval 0    isn't remote object.
1924  * \retval < 1  error code
1925  */
1926 int mdt_is_remote_object(struct mdt_thread_info *info,
1927                          struct mdt_object *parent,
1928                          struct mdt_object *child)
1929 {
1930         struct lu_buf *buf = &info->mti_big_buf;
1931         struct linkea_data ldata = { NULL };
1932         struct link_ea_header *leh;
1933         struct link_ea_entry *lee;
1934         struct lu_name name;
1935         struct lu_fid pfid;
1936         int reclen;
1937         int i;
1938         int rc;
1939
1940         ENTRY;
1941
1942         if (fid_is_root(mdt_object_fid(child)))
1943                 RETURN(0);
1944
1945         if (likely(parent != child)) {
1946                 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
1947                         RETURN(1);
1948
1949                 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
1950                         RETURN(0);
1951
1952                 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
1953                                                mdt_object_fid(child));
1954                 RETURN(rc);
1955         }
1956
1957         /* client < 2.13.52 getattr_by_fid parent and child are the same */
1958         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1959         if (!buf->lb_buf)
1960                 RETURN(-ENOMEM);
1961
1962         ldata.ld_buf = buf;
1963         rc = mdt_links_read(info, child, &ldata);
1964         /* can't read linkea, just assume it's remote object */
1965         if (rc == -ENOENT || rc == -ENODATA)
1966                 RETURN(1);
1967         if (rc)
1968                 RETURN(rc);
1969
1970         leh = buf->lb_buf;
1971         lee = (struct link_ea_entry *)(leh + 1);
1972         for (i = 0; i < leh->leh_reccount; i++) {
1973                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1974                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1975                 if (mdt_fids_different_target(info, &pfid,
1976                                               mdt_object_fid(child)))
1977                         RETURN(1);
1978         }
1979
1980         RETURN(0);
1981 }
1982
1983 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1984                              struct mdt_object *child)
1985 {
1986         struct lu_buf *buffer;
1987         struct mdt_body *repbody;
1988         struct req_capsule *pill = info->mti_pill;
1989         struct obd_export *exp = mdt_info_req(info)->rq_export;
1990         int rc = 0;
1991
1992         if (!exp_connect_encrypt(exp))
1993                 return rc;
1994
1995         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
1996             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
1997                 struct lu_attr la = { 0 };
1998                 struct dt_object *dt = mdt_obj2dt(child);
1999
2000                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
2001                         dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
2002
2003                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
2004                         buffer = &info->mti_buf;
2005
2006                         /* fill reply buffer with encryption context now */
2007                         buffer->lb_len =
2008                                 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
2009                                                      RCL_SERVER);
2010                         buffer->lb_buf =
2011                                 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
2012                         rc = mo_xattr_get(info->mti_env,
2013                                           mdt_object_child(child),
2014                                           buffer,
2015                                           LL_XATTR_NAME_ENCRYPTION_CONTEXT);
2016                         if (rc >= 0) {
2017                                 CDEBUG(D_SEC,
2018                                        "found encryption ctx of size %d for "DFID"\n",
2019                                        rc, PFID(mdt_object_fid(child)));
2020
2021                                 repbody = req_capsule_server_get(pill,
2022                                                                  &RMF_MDT_BODY);
2023                                 repbody->mbo_valid |= OBD_MD_ENCCTX;
2024                                 if (rc < buffer->lb_len)
2025                                         req_capsule_shrink(pill,
2026                                                            &RMF_FILE_ENCCTX, rc,
2027                                                            RCL_SERVER);
2028                                 rc = 0;
2029                         } else {
2030                                 CDEBUG(D_SEC,
2031                                        "encryption ctx not found for "DFID": rc = %d\n",
2032                                        PFID(mdt_object_fid(child)), rc);
2033                                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2034                                                    RCL_SERVER);
2035                                 /* handling -ENOENT is important because it may
2036                                  * change object state in DNE env dropping
2037                                  * LOHA_EXISTS flag, it is important to return
2038                                  * that to the caller.
2039                                  * Check LU-13115 for details.
2040                                  */
2041                                 if (rc != -ENOENT)
2042                                         rc = 0;
2043                         }
2044                 } else {
2045                         req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2046                                            RCL_SERVER);
2047                 }
2048         }
2049         return rc;
2050 }