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                                struct lnet_nid *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,
  99                            struct lnet_nid *peernid)
 100 {
 101         struct lu_ucred *ucred = mdt_ucred(info);
 102         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 103         ENTRY;
 104
 105         LASSERT(ucred != NULL);
 106         if (!squash->rsi_uid || ucred->uc_fsuid)
 107                 RETURN(0);
 108
 109         if (match_nosquash_list(&squash->rsi_lock,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nidstr(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nidstr(peernid),
 119                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap.cap[0],
 120                squash->rsi_uid, squash->rsi_gid, 0);
 121
 122         ucred->uc_fsuid = squash->rsi_uid;
 123         ucred->uc_fsgid = squash->rsi_gid;
 124         ucred->uc_cap = CAP_EMPTY_SET;
 125         ucred->uc_suppgids[0] = -1;
 126         ucred->uc_suppgids[1] = -1;
 127
 128         RETURN(0);
 129 }
 130
 131 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 132 {
 133         struct ptlrpc_request   *req = mdt_info_req(info);
 134         const char              *jobid = mdt_req_get_jobid(req);
 135
 136         /* set jobid if specified. */
 137         if (jobid)
 138                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 139         else
 140                 uc->uc_jobid[0] = '\0';
 141 }
 142
 143 static void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
 144 {
 145         if (info && info->mti_exp && info->mti_exp->exp_connection)
 146                 uc->uc_nid = lnet_nid_to_nid4(
 147                         &info->mti_exp->exp_connection->c_peer.nid);
 148         else
 149                 uc->uc_nid = LNET_NID_ANY;
 150 }
 151
 152 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
 153                                     struct lu_ucred *uc)
 154 {
 155         struct lu_nodemap *nodemap = NULL;
 156         bool audit = true;
 157
 158         if (info && info->mti_exp) {
 159                 nodemap = nodemap_get_from_exp(info->mti_exp);
 160                 if (nodemap && !IS_ERR(nodemap)) {
 161                         audit = nodemap->nmf_enable_audit;
 162                         nodemap_putref(nodemap);
 163                 }
 164         }
 165
 166         uc->uc_enable_audit = audit;
 167 }
 168
 169 static void ucred_set_rbac_roles(struct mdt_thread_info *info,
 170                                  struct lu_ucred *uc)
 171 {
 172         struct lu_nodemap *nodemap = NULL;
 173         enum nodemap_rbac_roles rbac = NODEMAP_RBAC_ALL;
 174
 175         if (info && info->mti_exp) {
 176                 nodemap = nodemap_get_from_exp(info->mti_exp);
 177                 if (!IS_ERR_OR_NULL(nodemap)) {
 178                         rbac = nodemap->nmf_rbac;
 179                         nodemap_putref(nodemap);
 180                 }
 181         }
 182
 183         uc->uc_rbac_file_perms = !!(rbac & NODEMAP_RBAC_FILE_PERMS);
 184         uc->uc_rbac_dne_ops = !!(rbac & NODEMAP_RBAC_DNE_OPS);
 185         uc->uc_rbac_quota_ops = !!(rbac & NODEMAP_RBAC_QUOTA_OPS);
 186         uc->uc_rbac_byfid_ops = !!(rbac & NODEMAP_RBAC_BYFID_OPS);
 187         uc->uc_rbac_chlg_ops = !!(rbac & NODEMAP_RBAC_CHLG_OPS);
 188         uc->uc_rbac_fscrypt_admin = !!(rbac & NODEMAP_RBAC_FSCRYPT_ADMIN);
 189 }
 190
 191 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 192                           void *buf)
 193 {
 194         struct ptlrpc_request *req = mdt_info_req(info);
 195         struct mdt_device *mdt = info->mti_mdt;
 196         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 197         struct lu_ucred *ucred = mdt_ucred(info);
 198         struct lu_nodemap *nodemap;
 199         struct lnet_nid peernid = req->rq_peer.nid;
 200         __u32 perm = 0;
 201         int setuid;
 202         int setgid;
 203         bool is_nm_gid_squashed = false;
 204         int rc = 0;
 205
 206         ENTRY;
 207
 208         LASSERT(req->rq_auth_gss);
 209         LASSERT(!req->rq_auth_usr_mdt);
 210         LASSERT(req->rq_user_desc);
 211         LASSERT(ucred != NULL);
 212
 213         ucred->uc_valid = UCRED_INVALID;
 214
 215         nodemap = nodemap_get_from_exp(info->mti_exp);
 216         if (IS_ERR(nodemap))
 217                 RETURN(PTR_ERR(nodemap));
 218
 219         pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 220                                        NODEMAP_CLIENT_TO_FS, pud->pud_uid);
 221         pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 222                                        NODEMAP_CLIENT_TO_FS, pud->pud_gid);
 223         pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 224                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
 225         pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 226                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
 227
 228         ucred->uc_o_uid = pud->pud_uid;
 229         ucred->uc_o_gid = pud->pud_gid;
 230         ucred->uc_o_fsuid = pud->pud_fsuid;
 231         ucred->uc_o_fsgid = pud->pud_fsgid;
 232
 233         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 234                 /* deny access before we get identity ref */
 235                 if (nodemap->nmf_deny_unknown) {
 236                         nodemap_putref(nodemap);
 237                         RETURN(-EACCES);
 238                 }
 239
 240                 ucred->uc_suppgids[0] = -1;
 241                 ucred->uc_suppgids[1] = -1;
 242         }
 243
 244         if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
 245                 is_nm_gid_squashed = true;
 246
 247         nodemap_putref(nodemap);
 248
 249         if (type == BODY_INIT) {
 250                 struct mdt_body *body = (struct mdt_body *)buf;
 251
 252                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 253                 ucred->uc_suppgids[1] = -1;
 254         }
 255
 256         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 257             req->rq_auth_uid != pud->pud_uid) {
 258                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 259                        "while client claims %u:%u/%u:%u\n",
 260                        libcfs_nidstr(&peernid), req->rq_auth_uid,
 261                        pud->pud_uid, pud->pud_gid,
 262                        pud->pud_fsuid, pud->pud_fsgid);
 263                 RETURN(-EACCES);
 264         }
 265
 266         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 267                 ucred->uc_identity = NULL;
 268                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 269         } else {
 270                 struct md_identity *identity;
 271
 272                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 273                                             pud->pud_uid);
 274                 if (IS_ERR(identity)) {
 275                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 276                                 ucred->uc_identity = NULL;
 277                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 278                                        CFS_SETGRP_PERM;
 279                         } else {
 280                                 CDEBUG(D_SEC,
 281                                        "Deny access without identity: uid %u\n",
 282                                        pud->pud_uid);
 283                                 RETURN(-EACCES);
 284                         }
 285                 } else {
 286                         ucred->uc_identity = identity;
 287                         perm = mdt_identity_get_perm(ucred->uc_identity,
 288                                                      &peernid);
 289                 }
 290         }
 291
 292         /* find out the setuid/setgid attempt */
 293         setuid = (pud->pud_uid != pud->pud_fsuid);
 294         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 295                   (ucred->uc_identity &&
 296                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 297
 298         /* check permission of setuid */
 299         if (setuid && !(perm & CFS_SETUID_PERM)) {
 300                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 301                        pud->pud_uid, pud->pud_fsuid, libcfs_nidstr(&peernid));
 302                 GOTO(out, rc = -EACCES);
 303         }
 304
 305         /* check permission of setgid */
 306         if (setgid && !(perm & CFS_SETGID_PERM)) {
 307                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 308                        "from %s\n", pud->pud_uid, pud->pud_gid,
 309                        pud->pud_fsuid, pud->pud_fsgid,
 310                        ucred->uc_identity->mi_gid, libcfs_nidstr(&peernid));
 311                 GOTO(out, rc = -EACCES);
 312         }
 313
 314         if (perm & CFS_SETGRP_PERM) {
 315                 /* only set groups if GID is not squashed */
 316                 if (pud->pud_ngroups && !is_nm_gid_squashed) {
 317                         /* setgroups for local client */
 318                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 319                         if (!ucred->uc_ginfo) {
 320                                 CERROR("failed to alloc %d groups\n",
 321                                        pud->pud_ngroups);
 322                                 GOTO(out, rc = -ENOMEM);
 323                         }
 324
 325                         lustre_groups_from_list(ucred->uc_ginfo,
 326                                                 pud->pud_groups);
 327                         lustre_groups_sort(ucred->uc_ginfo);
 328                 } else {
 329                         ucred->uc_suppgids[0] = -1;
 330                         ucred->uc_suppgids[1] = -1;
 331                         ucred->uc_ginfo = NULL;
 332                 }
 333         } else {
 334                 ucred->uc_suppgids[0] = -1;
 335                 ucred->uc_suppgids[1] = -1;
 336                 ucred->uc_ginfo = NULL;
 337         }
 338
 339         ucred->uc_uid = pud->pud_uid;
 340         ucred->uc_gid = pud->pud_gid;
 341
 342         ucred->uc_cap = CAP_EMPTY_SET;
 343         if (!nodemap || ucred->uc_o_uid != nodemap->nm_squash_uid)
 344                 ucred->uc_cap.cap[0] = pud->pud_cap;
 345
 346         ucred->uc_fsuid = pud->pud_fsuid;
 347         ucred->uc_fsgid = pud->pud_fsgid;
 348
 349         /* process root_squash here. */
 350         mdt_root_squash(info, &peernid);
 351
 352         ucred->uc_valid = UCRED_NEW;
 353         ucred_set_jobid(info, ucred);
 354         ucred_set_nid(info, ucred);
 355         ucred_set_audit_enabled(info, ucred);
 356         ucred_set_rbac_roles(info, ucred);
 357
 358         EXIT;
 359
 360 out:
 361         if (rc) {
 362                 if (ucred->uc_ginfo) {
 363                         put_group_info(ucred->uc_ginfo);
 364                         ucred->uc_ginfo = NULL;
 365                 }
 366                 if (ucred->uc_identity) {
 367                         mdt_identity_put(mdt->mdt_identity_cache,
 368                                          ucred->uc_identity);
 369                         ucred->uc_identity = NULL;
 370                 }
 371         }
 372
 373         return rc;
 374 }
 375
 376 /**
 377  * Check whether allow the client to set supplementary group IDs or not.
 378  *
 379  * \param[in] info      pointer to the thread context
 380  * \param[in] uc        pointer to the RPC user descriptor
 381  *
 382  * \retval              true if allow to set supplementary group IDs
 383  * \retval              false for other cases
 384  */
 385 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 386 {
 387         __u32 perm;
 388
 389         /* 1. If identity_upcall is disabled,
 390          *    permit local client to do anything. */
 391         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 392                 return true;
 393
 394         /* 2. If fail to get related identities, then forbid any client to
 395          *    set supplementary group IDs. */
 396         if (uc->uc_identity == NULL)
 397                 return false;
 398
 399         /* 3. Check the permission in the identities. */
 400         perm = mdt_identity_get_perm(
 401                 uc->uc_identity,
 402                 &mdt_info_req(info)->rq_peer.nid);
 403         if (perm & CFS_SETGRP_PERM)
 404                 return true;
 405
 406         return false;
 407 }
 408
 409 int mdt_check_ucred(struct mdt_thread_info *info)
 410 {
 411         struct ptlrpc_request   *req = mdt_info_req(info);
 412         struct mdt_device       *mdt = info->mti_mdt;
 413         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 414         struct lu_ucred         *ucred = mdt_ucred(info);
 415         struct md_identity      *identity = NULL;
 416         struct lnet_nid          peernid = req->rq_peer.nid;
 417         __u32                    perm = 0;
 418         int                      setuid;
 419         int                      setgid;
 420         int                      rc = 0;
 421
 422         ENTRY;
 423
 424         LASSERT(ucred != NULL);
 425         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 426                 RETURN(0);
 427
 428         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 429                 RETURN(0);
 430
 431         /* sanity check: if we use strong authentication, we expect the
 432          * uid which client claimed is true */
 433         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 434             req->rq_auth_uid != pud->pud_uid) {
 435                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 436                        "while client claims %u:%u/%u:%u\n",
 437                        libcfs_nidstr(&peernid), req->rq_auth_uid,
 438                        pud->pud_uid, pud->pud_gid,
 439                        pud->pud_fsuid, pud->pud_fsgid);
 440                 RETURN(-EACCES);
 441         }
 442
 443         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 444                 RETURN(0);
 445
 446         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 447         if (IS_ERR(identity)) {
 448                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 449                         RETURN(0);
 450                 } else {
 451                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 452                                pud->pud_uid);
 453                         RETURN(-EACCES);
 454                 }
 455         }
 456
 457         perm = mdt_identity_get_perm(identity, &peernid);
 458         /* find out the setuid/setgid attempt */
 459         setuid = (pud->pud_uid != pud->pud_fsuid);
 460         setgid = (pud->pud_gid != pud->pud_fsgid ||
 461                   pud->pud_gid != identity->mi_gid);
 462
 463         /* check permission of setuid */
 464         if (setuid && !(perm & CFS_SETUID_PERM)) {
 465                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 466                        pud->pud_uid, pud->pud_fsuid, libcfs_nidstr(&peernid));
 467                 GOTO(out, rc = -EACCES);
 468         }
 469
 470         /* check permission of setgid */
 471         if (setgid && !(perm & CFS_SETGID_PERM)) {
 472                 CDEBUG(D_SEC,
 473                        "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) from %s\n",
 474                        pud->pud_uid, pud->pud_gid,
 475                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 476                        libcfs_nidstr(&peernid));
 477                 GOTO(out, rc = -EACCES);
 478         }
 479
 480         EXIT;
 481
 482 out:
 483         mdt_identity_put(mdt->mdt_identity_cache, identity);
 484         return rc;
 485 }
 486
 487 static int old_init_ucred_common(struct mdt_thread_info *info,
 488                                  struct lu_nodemap *nodemap)
 489 {
 490         struct lu_ucred         *uc = mdt_ucred(info);
 491         struct mdt_device       *mdt = info->mti_mdt;
 492         struct md_identity      *identity = NULL;
 493
 494         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 495                 /* deny access before we get identity ref */
 496                 if (nodemap->nmf_deny_unknown)
 497                         RETURN(-EACCES);
 498
 499                 uc->uc_cap = CAP_EMPTY_SET;
 500                 uc->uc_suppgids[0] = -1;
 501                 uc->uc_suppgids[1] = -1;
 502         }
 503
 504         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 505                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 506                                             uc->uc_fsuid);
 507                 if (IS_ERR(identity)) {
 508                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 509                                      cap_raised(uc->uc_cap,
 510                                                 CAP_DAC_READ_SEARCH))) {
 511                                 identity = NULL;
 512                         } else {
 513                                 CDEBUG(D_SEC, "Deny access without identity: "
 514                                        "uid %u\n", uc->uc_fsuid);
 515                                 RETURN(-EACCES);
 516                         }
 517                 }
 518         }
 519         uc->uc_identity = identity;
 520
 521         /* process root_squash here. */
 522         mdt_root_squash(info,
 523                         &mdt_info_req(info)->rq_peer.nid);
 524
 525         uc->uc_valid = UCRED_OLD;
 526         ucred_set_jobid(info, uc);
 527         ucred_set_nid(info, uc);
 528         ucred_set_audit_enabled(info, uc);
 529         ucred_set_rbac_roles(info, uc);
 530
 531         EXIT;
 532
 533         return 0;
 534 }
 535
 536 static int old_init_ucred(struct mdt_thread_info *info,
 537                           struct mdt_body *body)
 538 {
 539         struct lu_ucred *uc = mdt_ucred(info);
 540         struct lu_nodemap *nodemap;
 541         int rc;
 542         ENTRY;
 543
 544         nodemap = nodemap_get_from_exp(info->mti_exp);
 545         if (IS_ERR(nodemap))
 546                 RETURN(PTR_ERR(nodemap));
 547
 548         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 549                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 550         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 551                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 552         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 553                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 554         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 555                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 556
 557         LASSERT(uc != NULL);
 558         uc->uc_valid = UCRED_INVALID;
 559         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 560         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 561         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 562         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 563         uc->uc_suppgids[0] = body->mbo_suppgid;
 564         uc->uc_suppgids[1] = -1;
 565         uc->uc_ginfo = NULL;
 566         uc->uc_cap = CAP_EMPTY_SET;
 567         uc->uc_cap.cap[0] = body->mbo_capability;
 568
 569         rc = old_init_ucred_common(info, nodemap);
 570         nodemap_putref(nodemap);
 571
 572         RETURN(rc);
 573 }
 574
 575 static int old_init_ucred_reint(struct mdt_thread_info *info)
 576 {
 577         struct lu_ucred *uc = mdt_ucred(info);
 578         struct lu_nodemap *nodemap;
 579         int rc;
 580         ENTRY;
 581
 582         nodemap = nodemap_get_from_exp(info->mti_exp);
 583         if (IS_ERR(nodemap))
 584                 RETURN(PTR_ERR(nodemap));
 585
 586         LASSERT(uc != NULL);
 587
 588         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 589                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 590         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 591                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 592
 593         uc->uc_valid = UCRED_INVALID;
 594         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 595         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 596         uc->uc_ginfo = NULL;
 597
 598         rc = old_init_ucred_common(info, nodemap);
 599         nodemap_putref(nodemap);
 600
 601         RETURN(rc);
 602 }
 603
 604 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 605                                    struct mdt_body *body)
 606 {
 607         struct ptlrpc_request   *req = mdt_info_req(info);
 608         struct lu_ucred         *uc  = mdt_ucred(info);
 609
 610         LASSERT(uc != NULL);
 611         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 612                 return 0;
 613
 614         mdt_exit_ucred(info);
 615
 616         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 617                 return old_init_ucred(info, body);
 618         else
 619                 return new_init_ucred(info, BODY_INIT, body);
 620 }
 621
 622 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 623 {
 624         return __mdt_init_ucred(info, body);
 625 }
 626
 627 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 628 {
 629         struct ptlrpc_request *req = mdt_info_req(info);
 630         struct lu_ucred       *uc  = mdt_ucred(info);
 631         struct md_attr        *ma  = &info->mti_attr;
 632
 633         LASSERT(uc != NULL);
 634         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 635                 return 0;
 636
 637         /* LU-5564: for normal close request, skip permission check */
 638         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 639             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP))) {
 640                 cap_raise_nfsd_set(uc->uc_cap, CAP_FULL_SET);
 641                 cap_raise_fs_set(uc->uc_cap, CAP_FULL_SET);
 642         }
 643
 644         mdt_exit_ucred(info);
 645
 646         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 647                 return old_init_ucred_reint(info);
 648         else
 649                 return new_init_ucred(info, REC_INIT, NULL);
 650 }
 651
 652 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 653 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 654 {
 655         const struct lov_ost_data_v1 *lod;
 656         __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
 657         __u16 count;
 658         int i;
 659
 660         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 661                 return;
 662
 663         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 664                POSTID(&lmm->lmm_oi), lmm_magic,
 665                le32_to_cpu(lmm->lmm_pattern));
 666
 667         /* No support for compount layouts yet */
 668         if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
 669                 return;
 670
 671         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 672         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 673                le32_to_cpu(lmm->lmm_stripe_size), count);
 674
 675         /* If it's a directory or a released file, then there are
 676          * no actual objects to print, so bail out. */
 677         if (valid & OBD_MD_FLDIREA ||
 678             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 679                 return;
 680
 681         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 682         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 683                 struct ost_id oi;
 684
 685                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 686                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 687                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 688         }
 689 }
 690
 691 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 692 {
 693         const struct lmv_mds_md_v1 *lmm1;
 694         const struct lmv_foreign_md *lfm;
 695         int                        i;
 696
 697         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 698                 return;
 699
 700         /* foreign LMV case */
 701         lfm = &lmv->lmv_foreign_md;
 702         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
 703                 CDEBUG_LIMIT(level,
 704                              "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
 705                              le32_to_cpu(lfm->lfm_magic),
 706                              le32_to_cpu(lfm->lfm_length),
 707                              le32_to_cpu(lfm->lfm_type),
 708                              le32_to_cpu(lfm->lfm_flags),
 709                              le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
 710                 return;
 711         }
 712
 713         lmm1 = &lmv->lmv_md_v1;
 714         CDEBUG(level,
 715                "magic 0x%08X, master %#X stripe_count %d hash_type %#x\n",
 716                le32_to_cpu(lmm1->lmv_magic),
 717                le32_to_cpu(lmm1->lmv_master_mdt_index),
 718                le32_to_cpu(lmm1->lmv_stripe_count),
 719                le32_to_cpu(lmm1->lmv_hash_type));
 720
 721         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 722                 return;
 723
 724         if (le32_to_cpu(lmm1->lmv_stripe_count) > LMV_MAX_STRIPE_COUNT)
 725                 return;
 726
 727         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 728                 struct lu_fid fid;
 729
 730                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 731                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 732         }
 733 }
 734
 735 /* Shrink and/or grow reply buffers */
 736 int mdt_fix_reply(struct mdt_thread_info *info)
 737 {
 738         struct req_capsule *pill = info->mti_pill;
 739         struct mdt_body    *body;
 740         int                md_size, md_packed = 0;
 741         int                acl_size;
 742         int                rc = 0;
 743         ENTRY;
 744
 745         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 746         LASSERT(body != NULL);
 747
 748         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 749                                OBD_MD_LINKNAME))
 750                 md_size = body->mbo_eadatasize;
 751         else
 752                 md_size = 0;
 753
 754         acl_size = body->mbo_aclsize;
 755
 756         /* this replay - not send info to client */
 757         if (info->mti_spec.no_create) {
 758                 md_size = 0;
 759                 acl_size = 0;
 760         }
 761
 762         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 763                md_size, acl_size);
 764 /*
 765             &RMF_MDT_BODY,
 766             &RMF_MDT_MD,
 767             &RMF_ACL, or &RMF_LOGCOOKIES
 768 (optional)  &RMF_CAPA1,
 769 (optional)  &RMF_CAPA2,
 770 (optional)  something else
 771 */
 772
 773         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 774          * buffers before growing it */
 775         if (info->mti_big_lmm_used) {
 776                 /* big_lmm buffer may be used even without packing the result
 777                  * into reply, just for internal server needs */
 778                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 779                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 780                                                          RCL_SERVER);
 781
 782                 /* free big lmm if md_size is not needed */
 783                 if (md_size == 0 || md_packed == 0) {
 784                         info->mti_big_lmm_used = 0;
 785                 } else {
 786                         /* buffer must be allocated separately */
 787                         LASSERT(info->mti_attr.ma_lmm !=
 788                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 789                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 790                 }
 791         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 792                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 793         }
 794
 795         if (info->mti_big_acl_used) {
 796                 if (acl_size == 0)
 797                         info->mti_big_acl_used = 0;
 798                 else
 799                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 800         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 801                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 802         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 803                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 804         }
 805
 806         /* Shrink optional SECCTX buffer if it is not used */
 807         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
 808             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
 809             !(body->mbo_valid & OBD_MD_SECCTX))
 810                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
 811
 812         /* Shrink optional ENCCTX buffer if it is not used */
 813         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
 814             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
 815             !(body->mbo_valid & OBD_MD_ENCCTX))
 816                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
 817
 818         /* Shrink optional default LMV buffer if it is not used */
 819         if (req_capsule_has_field(pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER) &&
 820             req_capsule_get_size(pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER) != 0 &&
 821             !(body->mbo_valid & OBD_MD_DEFAULT_MEA))
 822                 req_capsule_shrink(pill, &RMF_DEFAULT_MDT_MD, 0, RCL_SERVER);
 823
 824         /*
 825          * Some more field should be shrinked if needed.
 826          * This should be done by those who added fields to reply message.
 827          */
 828
 829         /* Grow MD buffer if needed finally */
 830         if (info->mti_big_lmm_used) {
 831                 void *lmm;
 832
 833                 LASSERT(md_size > md_packed);
 834                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 835                        md_size - md_packed);
 836
 837                 /* FIXME: Grow reply buffer for the batch request. */
 838                 if (info->mti_batch_env) {
 839                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 840                         info->mti_big_lmm_used = 0;
 841                         GOTO(check_acl, rc);
 842                 }
 843
 844                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 845                 if (rc) {
 846                         /* we can't answer with proper LOV EA, drop flags,
 847                          * the rc is also returned so this request is
 848                          * considered as failed */
 849                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 850                         /* don't return transno along with error */
 851                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 852                 } else {
 853                         /* now we need to pack right LOV/LMV EA */
 854                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 855                         if (info->mti_attr.ma_valid & MA_LOV) {
 856                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 857                                                              RCL_SERVER) ==
 858                                                 info->mti_attr.ma_lmm_size);
 859                                 memcpy(lmm, info->mti_attr.ma_lmm,
 860                                        info->mti_attr.ma_lmm_size);
 861                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 862                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 863                                                              RCL_SERVER) ==
 864                                                 info->mti_attr.ma_lmv_size);
 865                                 memcpy(lmm, info->mti_attr.ma_lmv,
 866                                        info->mti_attr.ma_lmv_size);
 867                         }
 868                 }
 869
 870                 /* update mdt_max_mdsize so clients will be aware about that */
 871                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 872                         info->mti_mdt->mdt_max_mdsize =
 873                                                 info->mti_attr.ma_lmm_size;
 874                 info->mti_big_lmm_used = 0;
 875         }
 876
 877 check_acl:
 878         if (info->mti_big_acl_used) {
 879                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 880                        acl_size);
 881
 882                 if (info->mti_batch_env) {
 883                         body->mbo_valid &= ~OBD_MD_FLACL;
 884                         info->mti_big_acl_used = 0;
 885                         RETURN(rc);
 886                 }
 887
 888                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 889                 if (rc) {
 890                         body->mbo_valid &= ~OBD_MD_FLACL;
 891                 } else {
 892                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 893
 894                         memcpy(acl, info->mti_big_acl, acl_size);
 895                 }
 896
 897                 info->mti_big_acl_used = 0;
 898         }
 899
 900         RETURN(rc);
 901 }
 902
 903
 904 /* if object is dying, pack the lov/llog data,
 905  * parameter info->mti_attr should be valid at this point!
 906  * Also implements RAoLU policy */
 907 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 908                            struct md_attr *ma)
 909 {
 910         struct mdt_body *repbody = NULL;
 911         const struct lu_attr *la = &ma->ma_attr;
 912         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 913         int rc;
 914         __u64 need = 0;
 915         struct hsm_action_item hai = {
 916                 .hai_len = sizeof(hai),
 917                 .hai_action = HSMA_REMOVE,
 918                 .hai_extent.length = -1,
 919                 .hai_cookie = 0,
 920                 .hai_gid = 0,
 921         };
 922         int archive_id;
 923
 924         ENTRY;
 925
 926         if (mdt_info_req(info) != NULL) {
 927                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 928                 LASSERT(repbody != NULL);
 929         } else {
 930                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 931         }
 932
 933         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 934                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 935
 936         if (ma->ma_valid & MA_LOV) {
 937                 CERROR("No need in LOV EA upon unlink\n");
 938                 dump_stack();
 939         }
 940         if (repbody != NULL)
 941                 repbody->mbo_eadatasize = 0;
 942
 943         /* Only check unlinked and archived if RAoLU and upon last close */
 944         if (!cdt->cdt_remove_archive_on_last_unlink ||
 945             atomic_read(&mo->mot_open_count) != 0)
 946                 RETURN(0);
 947
 948         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 949         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 950                 RETURN(0);
 951
 952         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 953                 RETURN(0);
 954
 955         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 956         if (need != 0) {
 957                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 958                  * try setting them */
 959                 ma->ma_need |= need;
 960                 rc = mdt_attr_get_complex(info, mo, ma);
 961                 if (rc) {
 962                         CERROR("%s: unable to fetch missing attributes of"
 963                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 964                                PFID(mdt_object_fid(mo)), rc);
 965                         RETURN(0);
 966                 }
 967
 968                 if (need & MA_INODE) {
 969                         if (ma->ma_valid & MA_INODE) {
 970                                 if (ma->ma_attr.la_nlink != 0)
 971                                         RETURN(0);
 972                         } else {
 973                                 RETURN(0);
 974                         }
 975                 }
 976
 977                 if (need & MA_HSM) {
 978                         if (ma->ma_valid & MA_HSM) {
 979                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 980                                         RETURN(0);
 981                         } else {
 982                                 RETURN(0);
 983                         }
 984                 }
 985         }
 986
 987         /* RAoLU policy is active, last close on file has occured,
 988          * file is unlinked, file is archived, so create remove request
 989          * for copytool!
 990          * If CDT is not running, requests will be logged for later. */
 991         if (ma->ma_hsm.mh_arch_id != 0)
 992                 archive_id = ma->ma_hsm.mh_arch_id;
 993         else
 994                 archive_id = cdt->cdt_default_archive_id;
 995
 996         hai.hai_fid = *mdt_object_fid(mo);
 997
 998         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
 999                                   &hai);
1000         if (rc)
1001                 CERROR("%s: unable to add HSM remove request for "DFID
1002                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
1003                        PFID(mdt_object_fid(mo)), rc);
1004
1005         RETURN(0);
1006 }
1007
1008 static __u64 mdt_attr_valid_xlate(enum mds_attr_flags in,
1009                                   struct mdt_reint_record *rr,
1010                                   struct md_attr *ma)
1011 {
1012         __u64 out;
1013
1014         out = 0;
1015         if (in & MDS_ATTR_MODE)
1016                 out |= LA_MODE;
1017         if (in & MDS_ATTR_UID)
1018                 out |= LA_UID;
1019         if (in & MDS_ATTR_GID)
1020                 out |= LA_GID;
1021         if (in & MDS_ATTR_SIZE)
1022                 out |= LA_SIZE;
1023         if (in & MDS_ATTR_BLOCKS)
1024                 out |= LA_BLOCKS;
1025         if (in & MDS_ATTR_ATIME_SET)
1026                 out |= LA_ATIME;
1027         if (in & MDS_ATTR_CTIME_SET)
1028                 out |= LA_CTIME;
1029         if (in & MDS_ATTR_MTIME_SET)
1030                 out |= LA_MTIME;
1031         if (in & MDS_ATTR_ATTR_FLAG)
1032                 out |= LA_FLAGS;
1033         if (in & MDS_ATTR_KILL_SUID)
1034                 out |= LA_KILL_SUID;
1035         if (in & MDS_ATTR_KILL_SGID)
1036                 out |= LA_KILL_SGID;
1037         if (in & MDS_ATTR_PROJID)
1038                 out |= LA_PROJID;
1039         if (in & MDS_ATTR_LSIZE)
1040                 out |= LA_LSIZE;
1041         if (in & MDS_ATTR_LBLOCKS)
1042                 out |= LA_LBLOCKS;
1043
1044         if (in & MDS_ATTR_FROM_OPEN)
1045                 rr->rr_flags |= MRF_OPEN_TRUNC;
1046         if (in & MDS_ATTR_OVERRIDE)
1047                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
1048         if (in & MDS_ATTR_FORCE)
1049                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1050
1051         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
1052                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
1053                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
1054                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1055                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1056                 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1057                 MDS_ATTR_OVERRIDE);
1058         if (in != 0)
1059                 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", (u64)in);
1060
1061         return out;
1062 }
1063
1064 /* unpacking */
1065
1066 int mdt_name_unpack(struct req_capsule *pill,
1067                     const struct req_msg_field *field,
1068                     struct lu_name *ln,
1069                     enum mdt_name_flags flags)
1070 {
1071         ln->ln_name = req_capsule_client_get(pill, field);
1072         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1073
1074         if (!lu_name_is_valid(ln)) {
1075                 ln->ln_name = NULL;
1076                 ln->ln_namelen = 0;
1077
1078                 return -EPROTO;
1079         }
1080
1081         if ((flags & MNF_FIX_ANON) &&
1082             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1083                 /* Newer (3.x) kernels use a name of "/" for the
1084                  * "anonymous" disconnected dentries from NFS
1085                  * filehandle conversion. See d_obtain_alias(). */
1086                 ln->ln_name = NULL;
1087                 ln->ln_namelen = 0;
1088         }
1089
1090         return 0;
1091 }
1092
1093 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1094                                   const char **secctx_name,
1095                                   void **secctx, size_t *secctx_size)
1096 {
1097         const char *name;
1098         size_t name_size;
1099
1100         *secctx_name = NULL;
1101         *secctx = NULL;
1102         *secctx_size = 0;
1103
1104         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1105             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1106                 return 0;
1107
1108         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1109                                          RCL_CLIENT);
1110         if (name_size == 0)
1111                 return 0;
1112
1113         if (name_size > XATTR_NAME_MAX + 1)
1114                 return -EPROTO;
1115
1116         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1117         if (strnlen(name, name_size) != name_size - 1)
1118                 return -EPROTO;
1119
1120         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1121             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1122                 return -EPROTO;
1123
1124         *secctx_name = name;
1125         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1126         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1127
1128         return 0;
1129 }
1130
1131 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1132                                   void **encctx, size_t *encctx_size)
1133 {
1134         *encctx = NULL;
1135         *encctx_size = 0;
1136
1137         if (!exp_connect_encrypt(pill->rc_req->rq_export))
1138                 return 0;
1139
1140         if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1141             !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1142                 return -EPROTO;
1143
1144         *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1145         if (*encctx_size == 0)
1146                 return 0;
1147
1148         *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1149
1150         return 0;
1151 }
1152
1153 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1154 {
1155         struct lu_ucred *uc = mdt_ucred(info);
1156         struct md_attr *ma = &info->mti_attr;
1157         struct lu_attr *la = &ma->ma_attr;
1158         struct req_capsule *pill = info->mti_pill;
1159         struct mdt_reint_record *rr = &info->mti_rr;
1160         struct mdt_rec_setattr *rec;
1161         struct lu_nodemap *nodemap;
1162
1163         ENTRY;
1164
1165         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1166         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1167         if (rec == NULL)
1168                 RETURN(-EFAULT);
1169
1170         /* This prior initialization is needed for old_init_ucred_reint() */
1171         uc->uc_fsuid = rec->sa_fsuid;
1172         uc->uc_fsgid = rec->sa_fsgid;
1173         uc->uc_cap = CAP_EMPTY_SET;
1174         uc->uc_cap.cap[0] = rec->sa_cap;
1175         uc->uc_suppgids[0] = rec->sa_suppgid;
1176         uc->uc_suppgids[1] = -1;
1177
1178         rr->rr_fid1 = &rec->sa_fid;
1179         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1180         la->la_mode  = rec->sa_mode;
1181         la->la_flags = rec->sa_attr_flags;
1182
1183         nodemap = nodemap_get_from_exp(info->mti_exp);
1184         if (IS_ERR(nodemap))
1185                 RETURN(PTR_ERR(nodemap));
1186
1187         la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1188                                     NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1189         la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1190                                     NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1191         la->la_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
1192                                        NODEMAP_CLIENT_TO_FS, rec->sa_projid);
1193         nodemap_putref(nodemap);
1194
1195         la->la_size  = rec->sa_size;
1196         la->la_blocks = rec->sa_blocks;
1197         la->la_ctime = rec->sa_ctime;
1198         la->la_atime = rec->sa_atime;
1199         la->la_mtime = rec->sa_mtime;
1200         ma->ma_valid = MA_INODE;
1201
1202         ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1203                                 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1204                                 MDS_PCC_ATTACH);
1205         RETURN(0);
1206 }
1207
1208 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1209 {
1210         struct req_capsule *pill = info->mti_pill;
1211         struct mdt_ioepoch *ioepoch;
1212         ENTRY;
1213
1214         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1215                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1216         else
1217                 ioepoch = NULL;
1218
1219         if (ioepoch == NULL)
1220                 RETURN(-EPROTO);
1221
1222         info->mti_open_handle = ioepoch->mio_open_handle;
1223
1224         RETURN(0);
1225 }
1226
1227 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1228         struct req_capsule      *pill = info->mti_pill;
1229
1230         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1231                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1232                 if (info->mti_dlm_req == NULL)
1233                         RETURN(-EFAULT);
1234         }
1235
1236         RETURN(0);
1237 }
1238
1239 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1240 {
1241         struct mdt_reint_record *rr = &info->mti_rr;
1242         struct md_attr          *ma = &info->mti_attr;
1243         struct req_capsule      *pill = info->mti_pill;
1244         int rc;
1245         ENTRY;
1246
1247         rc = mdt_setattr_unpack_rec(info);
1248         if (rc)
1249                 RETURN(rc);
1250
1251         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1252                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1253                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1254                                                         RCL_CLIENT);
1255
1256                 if (rr->rr_eadatalen > 0) {
1257                         const struct lmv_user_md        *lum;
1258
1259                         lum = rr->rr_eadata;
1260                         /* Sigh ma_valid(from req) does not indicate whether
1261                          * it will set LOV/LMV EA, so we have to check magic */
1262                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1263                                 ma->ma_valid |= MA_LMV;
1264                                 ma->ma_lmv = (void *)rr->rr_eadata;
1265                                 ma->ma_lmv_size = rr->rr_eadatalen;
1266                         } else {
1267                                 ma->ma_valid |= MA_LOV;
1268                                 ma->ma_lmm = (void *)rr->rr_eadata;
1269                                 ma->ma_lmm_size = rr->rr_eadatalen;
1270                         }
1271                 }
1272         }
1273
1274         rc = mdt_dlmreq_unpack(info);
1275         RETURN(rc);
1276 }
1277
1278 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1279 {
1280         struct md_attr          *ma = &info->mti_attr;
1281         struct req_capsule      *pill = info->mti_pill;
1282         ENTRY;
1283
1284         if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1285                 RETURN(0);
1286
1287         req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1288
1289         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1290             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1291                 RETURN(-EFAULT);
1292
1293         RETURN(0);
1294 }
1295
1296 int mdt_close_unpack(struct mdt_thread_info *info)
1297 {
1298         int rc;
1299         ENTRY;
1300
1301         rc = mdt_close_handle_unpack(info);
1302         if (rc)
1303                 RETURN(rc);
1304
1305         rc = mdt_setattr_unpack_rec(info);
1306         if (rc)
1307                 RETURN(rc);
1308
1309         rc = mdt_close_intent_unpack(info);
1310         if (rc)
1311                 RETURN(rc);
1312
1313         RETURN(mdt_init_ucred_reint(info));
1314 }
1315
1316 static int mdt_create_unpack(struct mdt_thread_info *info)
1317 {
1318         struct lu_ucred *uc  = mdt_ucred(info);
1319         struct mdt_rec_create *rec;
1320         struct lu_attr *attr = &info->mti_attr.ma_attr;
1321         struct mdt_reint_record *rr = &info->mti_rr;
1322         struct req_capsule      *pill = info->mti_pill;
1323         struct md_op_spec       *sp = &info->mti_spec;
1324         int rc;
1325
1326         ENTRY;
1327
1328         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1329         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1330         if (rec == NULL)
1331                 RETURN(-EFAULT);
1332
1333         /* This prior initialization is needed for old_init_ucred_reint() */
1334         uc->uc_fsuid = rec->cr_fsuid;
1335         uc->uc_fsgid = rec->cr_fsgid;
1336         uc->uc_cap = CAP_EMPTY_SET;
1337         uc->uc_cap.cap[0] = rec->cr_cap;
1338         uc->uc_suppgids[0] = rec->cr_suppgid1;
1339         uc->uc_suppgids[1] = -1;
1340         uc->uc_umask = rec->cr_umask;
1341
1342         rr->rr_fid1 = &rec->cr_fid1;
1343         rr->rr_fid2 = &rec->cr_fid2;
1344         attr->la_mode = rec->cr_mode;
1345         attr->la_rdev  = rec->cr_rdev;
1346         attr->la_uid   = rec->cr_fsuid;
1347         attr->la_gid   = rec->cr_fsgid;
1348         attr->la_ctime = rec->cr_time;
1349         attr->la_mtime = rec->cr_time;
1350         attr->la_atime = rec->cr_time;
1351         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1352                 LA_CTIME | LA_MTIME | LA_ATIME;
1353         memset(&sp->u, 0, sizeof(sp->u));
1354         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1355
1356         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1357         if (rc < 0)
1358                 RETURN(rc);
1359
1360         if (S_ISLNK(attr->la_mode)) {
1361                 const char *tgt = NULL;
1362                 int sz;
1363
1364                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1365                 sz = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT);
1366                 if (sz) {
1367                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1368                         sp->u.sp_symname.ln_name = tgt;
1369                         sp->u.sp_symname.ln_namelen = sz - 1; /* skip NUL */
1370                 }
1371                 if (tgt == NULL)
1372                         RETURN(-EFAULT);
1373         } else {
1374                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1375                 if (S_ISDIR(attr->la_mode) &&
1376                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1377                         sp->u.sp_ea.eadata =
1378                                 req_capsule_client_get(pill, &RMF_EADATA);
1379                         sp->u.sp_ea.eadatalen =
1380                                 req_capsule_get_size(pill, &RMF_EADATA,
1381                                                      RCL_CLIENT);
1382                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1383                 }
1384         }
1385
1386         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1387                                     &sp->sp_cr_file_secctx,
1388                                     &sp->sp_cr_file_secctx_size);
1389         if (rc < 0)
1390                 RETURN(rc);
1391
1392         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1393                                     &sp->sp_cr_file_encctx_size);
1394         if (rc < 0)
1395                 RETURN(rc);
1396
1397         rc = req_check_sepol(pill);
1398         if (rc)
1399                 RETURN(rc);
1400
1401         rc = mdt_dlmreq_unpack(info);
1402         RETURN(rc);
1403 }
1404
1405 static int mdt_link_unpack(struct mdt_thread_info *info)
1406 {
1407         struct lu_ucred *uc  = mdt_ucred(info);
1408         struct mdt_rec_link *rec;
1409         struct lu_attr *attr = &info->mti_attr.ma_attr;
1410         struct mdt_reint_record *rr = &info->mti_rr;
1411         struct req_capsule *pill = info->mti_pill;
1412         int rc;
1413
1414         ENTRY;
1415
1416         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1417         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1418         if (rec == NULL)
1419                 RETURN(-EFAULT);
1420
1421         /* This prior initialization is needed for old_init_ucred_reint() */
1422         uc->uc_fsuid = rec->lk_fsuid;
1423         uc->uc_fsgid = rec->lk_fsgid;
1424         uc->uc_cap = CAP_EMPTY_SET;
1425         uc->uc_cap.cap[0] = rec->lk_cap;
1426         uc->uc_suppgids[0] = rec->lk_suppgid1;
1427         uc->uc_suppgids[1] = rec->lk_suppgid2;
1428
1429         attr->la_uid = rec->lk_fsuid;
1430         attr->la_gid = rec->lk_fsgid;
1431         rr->rr_fid1 = &rec->lk_fid1;
1432         rr->rr_fid2 = &rec->lk_fid2;
1433         attr->la_ctime = rec->lk_time;
1434         attr->la_mtime = rec->lk_time;
1435         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1436
1437         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1438         if (rc < 0)
1439                 RETURN(rc);
1440
1441         rc = req_check_sepol(pill);
1442         if (rc)
1443                 RETURN(rc);
1444
1445         rc = mdt_dlmreq_unpack(info);
1446
1447         RETURN(rc);
1448 }
1449
1450 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1451 {
1452         struct lu_ucred *uc  = mdt_ucred(info);
1453         struct mdt_rec_unlink *rec;
1454         struct lu_attr *attr = &info->mti_attr.ma_attr;
1455         struct mdt_reint_record *rr = &info->mti_rr;
1456         struct req_capsule      *pill = info->mti_pill;
1457         int rc;
1458
1459         ENTRY;
1460
1461         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1462         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1463         if (rec == NULL)
1464                 RETURN(-EFAULT);
1465
1466         /* This prior initialization is needed for old_init_ucred_reint() */
1467         uc->uc_fsuid = rec->ul_fsuid;
1468         uc->uc_fsgid = rec->ul_fsgid;
1469         uc->uc_cap = CAP_EMPTY_SET;
1470         uc->uc_cap.cap[0] = rec->ul_cap;
1471         uc->uc_suppgids[0] = rec->ul_suppgid1;
1472         uc->uc_suppgids[1] = -1;
1473
1474         attr->la_uid = rec->ul_fsuid;
1475         attr->la_gid = rec->ul_fsgid;
1476         rr->rr_fid1 = &rec->ul_fid1;
1477         rr->rr_fid2 = &rec->ul_fid2;
1478         attr->la_ctime = rec->ul_time;
1479         attr->la_mtime = rec->ul_time;
1480         attr->la_mode  = rec->ul_mode;
1481         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1482         if (rec->ul_bias & MDS_FID_OP)
1483                 info->mti_spec.sp_cr_flags |= MDS_OP_WITH_FID;
1484         else
1485                 info->mti_spec.sp_cr_flags &= ~MDS_OP_WITH_FID;
1486
1487         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1488         if (rc < 0)
1489                 RETURN(rc);
1490
1491         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1492
1493         rc = req_check_sepol(pill);
1494         if (rc)
1495                 RETURN(rc);
1496
1497         rc = mdt_dlmreq_unpack(info);
1498         RETURN(rc);
1499 }
1500
1501 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1502 {
1503         info->mti_spec.sp_rm_entry = 1;
1504         return mdt_unlink_unpack(info);
1505 }
1506
1507 static int mdt_rename_unpack(struct mdt_thread_info *info)
1508 {
1509         struct lu_ucred *uc = mdt_ucred(info);
1510         struct mdt_rec_rename *rec;
1511         struct lu_attr *attr = &info->mti_attr.ma_attr;
1512         struct mdt_reint_record *rr = &info->mti_rr;
1513         struct req_capsule *pill = info->mti_pill;
1514         struct md_op_spec *spec = &info->mti_spec;
1515         int rc;
1516
1517         ENTRY;
1518
1519         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1520         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1521         if (rec == NULL)
1522                 RETURN(-EFAULT);
1523
1524         /* This prior initialization is needed for old_init_ucred_reint() */
1525         uc->uc_fsuid = rec->rn_fsuid;
1526         uc->uc_fsgid = rec->rn_fsgid;
1527         uc->uc_cap = CAP_EMPTY_SET;
1528         uc->uc_cap.cap[0] = rec->rn_cap;
1529         uc->uc_suppgids[0] = rec->rn_suppgid1;
1530         uc->uc_suppgids[1] = rec->rn_suppgid2;
1531
1532         attr->la_uid = rec->rn_fsuid;
1533         attr->la_gid = rec->rn_fsgid;
1534         rr->rr_fid1 = &rec->rn_fid1;
1535         rr->rr_fid2 = &rec->rn_fid2;
1536         attr->la_ctime = rec->rn_time;
1537         attr->la_mtime = rec->rn_time;
1538         /* rename_tgt contains the mode already */
1539         attr->la_mode = rec->rn_mode;
1540         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1541
1542         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1543         if (rc < 0)
1544                 RETURN(rc);
1545
1546         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1547         if (rc < 0)
1548                 RETURN(rc);
1549
1550         spec->no_create = !!req_is_replay(mdt_info_req(info));
1551
1552         rc = req_check_sepol(pill);
1553         if (rc)
1554                 RETURN(rc);
1555
1556         rc = mdt_dlmreq_unpack(info);
1557
1558         RETURN(rc);
1559 }
1560
1561 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1562 {
1563         struct lu_ucred *uc = mdt_ucred(info);
1564         struct mdt_rec_rename *rec;
1565         struct lu_attr *attr = &info->mti_attr.ma_attr;
1566         struct mdt_reint_record *rr = &info->mti_rr;
1567         struct req_capsule *pill = info->mti_pill;
1568         struct md_op_spec *spec = &info->mti_spec;
1569         int rc;
1570
1571         ENTRY;
1572
1573         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1574         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1575         if (rec == NULL)
1576                 RETURN(-EFAULT);
1577
1578         /* This prior initialization is needed for old_init_ucred_reint() */
1579         uc->uc_fsuid = rec->rn_fsuid;
1580         uc->uc_fsgid = rec->rn_fsgid;
1581         uc->uc_cap = CAP_EMPTY_SET;
1582         uc->uc_cap.cap[0] = rec->rn_cap;
1583         uc->uc_suppgids[0] = rec->rn_suppgid1;
1584         uc->uc_suppgids[1] = rec->rn_suppgid2;
1585
1586         attr->la_uid = rec->rn_fsuid;
1587         attr->la_gid = rec->rn_fsgid;
1588         rr->rr_fid1 = &rec->rn_fid1;
1589         rr->rr_fid2 = &rec->rn_fid2;
1590         attr->la_ctime = rec->rn_time;
1591         attr->la_mtime = rec->rn_time;
1592         /* rename_tgt contains the mode already */
1593         attr->la_mode = rec->rn_mode;
1594         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1595         spec->sp_cr_flags = 0;
1596
1597         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1598         if (rc < 0)
1599                 RETURN(rc);
1600
1601         if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1602                 rc = mdt_close_handle_unpack(info);
1603                 if (rc)
1604                         RETURN(rc);
1605
1606                 spec->sp_migrate_close = 1;
1607         } else {
1608                 spec->sp_migrate_close = 0;
1609         }
1610
1611         spec->sp_migrate_nsonly = !!(rec->rn_bias & MDS_MIGRATE_NSONLY);
1612
1613         /* lustre version > 2.11 migration packs lum */
1614         if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1615                 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1616                         rr->rr_eadatalen = req_capsule_get_size(pill,
1617                                                                 &RMF_EADATA,
1618                                                                 RCL_CLIENT);
1619
1620                         if (rr->rr_eadatalen > 0) {
1621                                 struct lmv_user_md_v1 *lmu;
1622
1623                                 lmu = req_capsule_client_get(pill, &RMF_EADATA);
1624                                 lmu->lum_hash_type |=
1625                                         cpu_to_le32(LMV_HASH_FLAG_FIXED);
1626                                 rr->rr_eadata = lmu;
1627                                 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1628                                 spec->u.sp_ea.eadata = rr->rr_eadata;
1629                                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1630                         }
1631                 } else {
1632                         /* old client doesn't provide lum. */
1633                         RETURN(-EOPNOTSUPP);
1634                 }
1635         }
1636
1637         spec->no_create = !!req_is_replay(mdt_info_req(info));
1638
1639         rc = mdt_dlmreq_unpack(info);
1640
1641         RETURN(rc);
1642 }
1643
1644 /*
1645  * please see comment above LOV_MAGIC_V1_DEFINED
1646  */
1647 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1648 {
1649         struct lov_user_md_v1   *v1 = eadata;
1650
1651         LASSERT(v1);
1652
1653         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1654                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1655                         v1->lmm_magic |= LOV_MAGIC_DEFINED;
1656                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1657                          __swab32(LOV_MAGIC_MAGIC))
1658                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1659         }
1660 }
1661
1662 static int mdt_open_unpack(struct mdt_thread_info *info)
1663 {
1664         struct lu_ucred *uc = mdt_ucred(info);
1665         struct mdt_rec_create *rec;
1666         struct lu_attr *attr = &info->mti_attr.ma_attr;
1667         struct req_capsule *pill = info->mti_pill;
1668         struct mdt_reint_record *rr = &info->mti_rr;
1669         struct ptlrpc_request *req = mdt_info_req(info);
1670         struct md_op_spec *sp = &info->mti_spec;
1671         int rc;
1672         ENTRY;
1673
1674         BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1675                      sizeof(struct mdt_rec_reint));
1676         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1677         if (rec == NULL)
1678                 RETURN(-EFAULT);
1679
1680         /* This prior initialization is needed for old_init_ucred_reint() */
1681         uc->uc_fsuid = rec->cr_fsuid;
1682         uc->uc_fsgid = rec->cr_fsgid;
1683         uc->uc_cap = CAP_EMPTY_SET;
1684         uc->uc_cap.cap[0] = rec->cr_cap;
1685         uc->uc_suppgids[0] = rec->cr_suppgid1;
1686         uc->uc_suppgids[1] = rec->cr_suppgid2;
1687         uc->uc_umask = rec->cr_umask;
1688
1689         rr->rr_fid1   = &rec->cr_fid1;
1690         rr->rr_fid2   = &rec->cr_fid2;
1691         rr->rr_open_handle = &rec->cr_open_handle_old;
1692         attr->la_mode = rec->cr_mode;
1693         attr->la_rdev  = rec->cr_rdev;
1694         attr->la_uid   = rec->cr_fsuid;
1695         attr->la_gid   = rec->cr_fsgid;
1696         attr->la_ctime = rec->cr_time;
1697         attr->la_mtime = rec->cr_time;
1698         attr->la_atime = rec->cr_time;
1699         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1700                          LA_CTIME | LA_MTIME | LA_ATIME;
1701         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1702         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1703         /* Do not trigger ASSERTION if client miss to set such flags. */
1704         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1705                 RETURN(-EPROTO);
1706
1707         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1708
1709         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1710
1711         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1712                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1713                                                         RCL_CLIENT);
1714
1715                 if (rr->rr_eadatalen > 0) {
1716                         rr->rr_eadata = req_capsule_client_get(pill,
1717                                                                &RMF_EADATA);
1718                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1719                         sp->u.sp_ea.eadata = rr->rr_eadata;
1720                         sp->sp_archive_id = rec->cr_archive_id;
1721                         sp->no_create = !!req_is_replay(req);
1722                         mdt_fix_lov_magic(info, rr->rr_eadata);
1723                 }
1724
1725                 /*
1726                  * Client default md_size may be 0 right after client start,
1727                  * until all osc are connected, set here just some reasonable
1728                  * value to prevent misbehavior.
1729                  */
1730                 if (rr->rr_eadatalen == 0 &&
1731                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1732                         rr->rr_eadatalen = MIN_MD_SIZE;
1733         }
1734
1735         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1736                                     &sp->sp_cr_file_secctx,
1737                                     &sp->sp_cr_file_secctx_size);
1738         if (rc < 0)
1739                 RETURN(rc);
1740
1741         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1742                                     &sp->sp_cr_file_encctx_size);
1743         if (rc < 0)
1744                 RETURN(rc);
1745
1746         rc = req_check_sepol(pill);
1747         if (rc)
1748                 RETURN(rc);
1749
1750         RETURN(rc);
1751 }
1752
1753 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1754 {
1755         struct mdt_reint_record *rr = &info->mti_rr;
1756         struct lu_ucred *uc = mdt_ucred(info);
1757         struct lu_attr *attr = &info->mti_attr.ma_attr;
1758         struct req_capsule *pill = info->mti_pill;
1759         struct mdt_rec_setxattr *rec;
1760         int rc;
1761         ENTRY;
1762
1763
1764         BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1765                      sizeof(struct mdt_rec_reint));
1766         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1767         if (rec == NULL)
1768                 RETURN(-EFAULT);
1769
1770         /* This prior initialization is needed for old_init_ucred_reint() */
1771         uc->uc_fsuid  = rec->sx_fsuid;
1772         uc->uc_fsgid  = rec->sx_fsgid;
1773         uc->uc_cap = CAP_EMPTY_SET;
1774         uc->uc_cap.cap[0] = rec->sx_cap;
1775         uc->uc_suppgids[0] = rec->sx_suppgid1;
1776         uc->uc_suppgids[1] = -1;
1777
1778         rr->rr_opcode = rec->sx_opcode;
1779         rr->rr_fid1   = &rec->sx_fid;
1780         attr->la_valid = rec->sx_valid;
1781         attr->la_ctime = rec->sx_time;
1782         attr->la_size = rec->sx_size;
1783         attr->la_flags = rec->sx_flags;
1784
1785         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1786         if (rc < 0)
1787                 RETURN(rc);
1788
1789         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1790                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1791                                                         RCL_CLIENT);
1792
1793                 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1794                         RETURN(-E2BIG);
1795
1796                 if (rr->rr_eadatalen > 0) {
1797                         rr->rr_eadata = req_capsule_client_get(pill,
1798                                                                &RMF_EADATA);
1799                         if (rr->rr_eadata == NULL)
1800                                 RETURN(-EFAULT);
1801                 } else {
1802                         rr->rr_eadata = NULL;
1803                 }
1804         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1805                 CDEBUG(D_INFO, "no xattr data supplied\n");
1806                 RETURN(-EFAULT);
1807         }
1808
1809         rc = req_check_sepol(pill);
1810         if (rc)
1811                 RETURN(rc);
1812
1813         if (mdt_dlmreq_unpack(info) < 0)
1814                 RETURN(-EPROTO);
1815
1816         RETURN(0);
1817 }
1818
1819 static int mdt_resync_unpack(struct mdt_thread_info *info)
1820 {
1821         struct req_capsule      *pill = info->mti_pill;
1822         struct mdt_reint_record *rr   = &info->mti_rr;
1823         struct lu_ucred         *uc     = mdt_ucred(info);
1824         struct mdt_rec_resync   *rec;
1825         ENTRY;
1826
1827         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1828         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1829         if (rec == NULL)
1830                 RETURN(-EFAULT);
1831
1832         /* This prior initialization is needed for old_init_ucred_reint() */
1833         uc->uc_fsuid = rec->rs_fsuid;
1834         uc->uc_fsgid = rec->rs_fsgid;
1835         uc->uc_cap = CAP_EMPTY_SET;
1836         uc->uc_cap.cap[0] = rec->rs_cap;
1837
1838         rr->rr_fid1   = &rec->rs_fid;
1839         rr->rr_mirror_id = rec->rs_mirror_id;
1840
1841         /* cookie doesn't need to be swapped but it has been swapped
1842          * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1843          * restoring. */
1844         if (req_capsule_req_need_swab(pill))
1845                 __swab64s(&rec->rs_lease_handle.cookie);
1846         rr->rr_lease_handle = &rec->rs_lease_handle;
1847
1848         RETURN(mdt_dlmreq_unpack(info));
1849 }
1850
1851 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1852
1853 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1854         [REINT_SETATTR]  = mdt_setattr_unpack,
1855         [REINT_CREATE]   = mdt_create_unpack,
1856         [REINT_LINK]     = mdt_link_unpack,
1857         [REINT_UNLINK]   = mdt_unlink_unpack,
1858         [REINT_RENAME]   = mdt_rename_unpack,
1859         [REINT_OPEN]     = mdt_open_unpack,
1860         [REINT_SETXATTR] = mdt_setxattr_unpack,
1861         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1862         [REINT_MIGRATE]  = mdt_migrate_unpack,
1863         [REINT_RESYNC]   = mdt_resync_unpack,
1864 };
1865
1866 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1867 {
1868         int rc;
1869         ENTRY;
1870
1871         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1872         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1873                 info->mti_rr.rr_opcode = op;
1874                 rc = mdt_reint_unpackers[op](info);
1875         } else {
1876                 CERROR("Unexpected opcode %d\n", op);
1877                 rc = -EFAULT;
1878         }
1879         RETURN(rc);
1880 }
1881
1882 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1883                              struct mdt_object *child)
1884 {
1885         char *secctx_name;
1886         struct lu_buf *buffer;
1887         struct mdt_body *repbody;
1888         struct req_capsule *pill = info->mti_pill;
1889         int rc = 0;
1890
1891         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1892             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1893                 secctx_name =
1894                         req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1895                 buffer = &info->mti_buf;
1896
1897                 /* fill reply buffer with security context now */
1898                 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1899                                                       RCL_SERVER);
1900                 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1901                                                         &RMF_FILE_SECCTX);
1902                 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1903                                   buffer, secctx_name);
1904                 if (rc >= 0) {
1905                         CDEBUG(D_SEC,
1906                                "found security context of size %d for "DFID"\n",
1907                                rc, PFID(mdt_object_fid(child)));
1908
1909                         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1910                         repbody->mbo_valid |= OBD_MD_SECCTX;
1911                         if (rc < buffer->lb_len)
1912                                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1913                                                    RCL_SERVER);
1914                         rc = 0;
1915                 } else {
1916                         CDEBUG(D_SEC,
1917                              "security context not found for "DFID": rc = %d\n",
1918                              PFID(mdt_object_fid(child)), rc);
1919                         req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1920                                            RCL_SERVER);
1921                         /* handling -ENOENT is important because it may change
1922                          * object state in DNE env dropping LOHA_EXISTS flag,
1923                          * it is important to return that to the caller.
1924                          * Check LU-13115 for details.
1925                          */
1926                         if (rc != -ENOENT)
1927                                 rc = 0;
1928                 }
1929         }
1930         return rc;
1931 }
1932
1933 /* check whether two FIDs belong to different MDT. */
1934 static int mdt_fids_different_target(struct mdt_thread_info *info,
1935                                      const struct lu_fid *fid1,
1936                                      const struct lu_fid *fid2)
1937 {
1938         const struct lu_env *env = info->mti_env;
1939         struct mdt_device *mdt = info->mti_mdt;
1940         struct lu_seq_range *range = &info->mti_range;
1941         struct seq_server_site *ss;
1942         __u32 index1, index2;
1943         int rc;
1944
1945         if (fid_seq(fid1) == fid_seq(fid2))
1946                 return 0;
1947
1948         ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1949
1950         range->lsr_flags = LU_SEQ_RANGE_MDT;
1951         rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1952         if (rc)
1953                 return rc;
1954
1955         index1 = range->lsr_index;
1956
1957         rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1958         if (rc)
1959                 return rc;
1960
1961         index2 = range->lsr_index;
1962
1963         return index1 != index2;
1964 }
1965
1966 /**
1967  * Check whether \a child is remote object on \a parent.
1968  *
1969  * \param[in]  info     thread environment
1970  * \param[in]  parent   parent object, it's the same as child object in
1971  *                      getattr_by_fid
1972  * \param[in]  child    child object
1973  *
1974  * \retval 1    is remote object.
1975  * \retval 0    isn't remote object.
1976  * \retval < 1  error code
1977  */
1978 int mdt_is_remote_object(struct mdt_thread_info *info,
1979                          struct mdt_object *parent,
1980                          struct mdt_object *child)
1981 {
1982         struct lu_buf *buf = &info->mti_big_buf;
1983         struct linkea_data ldata = { NULL };
1984         struct link_ea_header *leh;
1985         struct link_ea_entry *lee;
1986         struct lu_name name;
1987         struct lu_fid pfid;
1988         int reclen;
1989         int i;
1990         int rc;
1991
1992         ENTRY;
1993
1994         if (fid_is_root(mdt_object_fid(child)))
1995                 RETURN(0);
1996
1997         if (likely(parent != child)) {
1998                 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
1999                         RETURN(1);
2000
2001                 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
2002                         RETURN(0);
2003
2004                 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
2005                                                mdt_object_fid(child));
2006                 RETURN(rc);
2007         }
2008
2009         /* client < 2.13.52 getattr_by_fid parent and child are the same */
2010         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2011         if (!buf->lb_buf)
2012                 RETURN(-ENOMEM);
2013
2014         ldata.ld_buf = buf;
2015         rc = mdt_links_read(info, child, &ldata);
2016         /* can't read linkea, just assume it's remote object */
2017         if (rc == -ENOENT || rc == -ENODATA)
2018                 RETURN(1);
2019         if (rc)
2020                 RETURN(rc);
2021
2022         leh = buf->lb_buf;
2023         lee = (struct link_ea_entry *)(leh + 1);
2024         for (i = 0; i < leh->leh_reccount; i++) {
2025                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
2026                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2027                 if (mdt_fids_different_target(info, &pfid,
2028                                               mdt_object_fid(child)))
2029                         RETURN(1);
2030         }
2031
2032         RETURN(0);
2033 }
2034
2035 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
2036                              struct mdt_object *child)
2037 {
2038         struct lu_buf *buffer;
2039         struct mdt_body *repbody;
2040         struct req_capsule *pill = info->mti_pill;
2041         struct obd_export *exp = mdt_info_req(info)->rq_export;
2042         int rc = 0;
2043
2044         if (!exp_connect_encrypt(exp))
2045                 return rc;
2046
2047         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
2048             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
2049                 struct lu_attr la = { 0 };
2050                 struct dt_object *dt = mdt_obj2dt(child);
2051
2052                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
2053                         dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
2054
2055                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
2056                         buffer = &info->mti_buf;
2057
2058                         /* fill reply buffer with encryption context now */
2059                         buffer->lb_len =
2060                                 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
2061                                                      RCL_SERVER);
2062                         buffer->lb_buf =
2063                                 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
2064                         rc = mo_xattr_get(info->mti_env,
2065                                           mdt_object_child(child),
2066                                           buffer,
2067                                           LL_XATTR_NAME_ENCRYPTION_CONTEXT);
2068                         if (unlikely(rc == -ENODATA))
2069                                 /* For compatibility with 2.14 */
2070                                 rc = mo_xattr_get(info->mti_env,
2071                                           mdt_object_child(child),
2072                                           buffer,
2073                                           LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD);
2074                         if (rc >= 0) {
2075                                 CDEBUG(D_SEC,
2076                                        "found encryption ctx of size %d for "DFID"\n",
2077                                        rc, PFID(mdt_object_fid(child)));
2078
2079                                 repbody = req_capsule_server_get(pill,
2080                                                                  &RMF_MDT_BODY);
2081                                 repbody->mbo_valid |= OBD_MD_ENCCTX;
2082                                 if (rc < buffer->lb_len)
2083                                         req_capsule_shrink(pill,
2084                                                            &RMF_FILE_ENCCTX, rc,
2085                                                            RCL_SERVER);
2086                                 rc = 0;
2087                         } else {
2088                                 CDEBUG(D_SEC,
2089                                        "encryption ctx not found for "DFID": rc = %d\n",
2090                                        PFID(mdt_object_fid(child)), rc);
2091                                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2092                                                    RCL_SERVER);
2093                                 /* handling -ENOENT is important because it may
2094                                  * change object state in DNE env dropping
2095                                  * LOHA_EXISTS flag, it is important to return
2096                                  * that to the caller.
2097                                  * Check LU-13115 for details.
2098                                  */
2099                                 if (rc != -ENOENT)
2100                                         rc = 0;
2101                         }
2102                 } else {
2103                         req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2104                                            RCL_SERVER);
2105                 }
2106         }
2107         return rc;
2108 }