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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_lib.c
  37  *
  38  * Lustre Metadata Target (mdt) request unpacking helper.
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Fan Yong <fanyong@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/user_namespace.h>
  52 #ifdef HAVE_UIDGID_HEADER
  53 # include <linux/uidgid.h>
  54 #endif
  55 #include "mdt_internal.h"
  56 #include <lnet/nidstr.h>
  57 #include <lustre_nodemap.h>
  58
  59 typedef enum ucred_init_type {
  60         NONE_INIT       = 0,
  61         BODY_INIT       = 1,
  62         REC_INIT        = 2
  63 } ucred_init_type_t;
  64
  65 void mdt_exit_ucred(struct mdt_thread_info *info)
  66 {
  67         struct lu_ucred   *uc  = mdt_ucred(info);
  68         struct mdt_device *mdt = info->mti_mdt;
  69
  70         LASSERT(uc != NULL);
  71         if (uc->uc_valid != UCRED_INIT) {
  72                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  73                 if (uc->uc_ginfo) {
  74                         put_group_info(uc->uc_ginfo);
  75                         uc->uc_ginfo = NULL;
  76                 }
  77                 if (uc->uc_identity) {
  78                         mdt_identity_put(mdt->mdt_identity_cache,
  79                                          uc->uc_identity);
  80                         uc->uc_identity = NULL;
  81                 }
  82                 uc->uc_valid = UCRED_INIT;
  83         }
  84 }
  85
  86 static int match_nosquash_list(struct rw_semaphore *sem,
  87                                struct list_head *nidlist,
  88                                lnet_nid_t peernid)
  89 {
  90         int rc;
  91         ENTRY;
  92         down_read(sem);
  93         rc = cfs_match_nid(peernid, nidlist);
  94         up_read(sem);
  95         RETURN(rc);
  96 }
  97
  98 /* root_squash for inter-MDS operations */
  99 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t 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_sem,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nid2str(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nid2str(peernid),
 119                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 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 = 0;
 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 int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 144                           void *buf, bool drop_fs_cap)
 145 {
 146         struct ptlrpc_request   *req = mdt_info_req(info);
 147         struct mdt_device       *mdt = info->mti_mdt;
 148         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 149         struct lu_ucred         *ucred = mdt_ucred(info);
 150         lnet_nid_t               peernid = req->rq_peer.nid;
 151         __u32                    perm = 0;
 152         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 153         int                      setuid;
 154         int                      setgid;
 155         int                      rc = 0;
 156
 157         ENTRY;
 158
 159         LASSERT(req->rq_auth_gss);
 160         LASSERT(!req->rq_auth_usr_mdt);
 161         LASSERT(req->rq_user_desc);
 162         LASSERT(ucred != NULL);
 163
 164         ucred->uc_valid = UCRED_INVALID;
 165
 166         ucred->uc_o_uid   = pud->pud_uid;
 167         ucred->uc_o_gid   = pud->pud_gid;
 168         ucred->uc_o_fsuid = pud->pud_fsuid;
 169         ucred->uc_o_fsgid = pud->pud_fsgid;
 170
 171         if (type == BODY_INIT) {
 172                 struct mdt_body *body = (struct mdt_body *)buf;
 173
 174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 175                 ucred->uc_suppgids[1] = -1;
 176         }
 177
 178         /* sanity check: we expect the uid which client claimed is true */
 179         if (remote) {
 180                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 181                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 182                         RETURN(-EACCES);
 183                 }
 184
 185                 if (ptlrpc_user_desc_do_idmap(req, pud))
 186                         RETURN(-EACCES);
 187
 188                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 189                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 190                                "while client claims %u:%u/%u:%u\n",
 191                                libcfs_nid2str(peernid), req->rq_auth_uid,
 192                                req->rq_auth_mapped_uid,
 193                                pud->pud_uid, pud->pud_gid,
 194                                pud->pud_fsuid, pud->pud_fsgid);
 195                         RETURN(-EACCES);
 196                 }
 197         } else {
 198                 if (req->rq_auth_uid != pud->pud_uid) {
 199                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 200                                "while client claims %u:%u/%u:%u\n",
 201                                libcfs_nid2str(peernid), req->rq_auth_uid,
 202                                pud->pud_uid, pud->pud_gid,
 203                                pud->pud_fsuid, pud->pud_fsgid);
 204                         RETURN(-EACCES);
 205                 }
 206         }
 207
 208         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 209                 if (remote) {
 210                         CDEBUG(D_SEC, "remote client must run with identity_get "
 211                                "enabled!\n");
 212                         RETURN(-EACCES);
 213                 } else {
 214                         ucred->uc_identity = NULL;
 215                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 216                                CFS_SETGRP_PERM;
 217                 }
 218         } else {
 219                 struct md_identity *identity;
 220
 221                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 222                                             pud->pud_uid);
 223                 if (IS_ERR(identity)) {
 224                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 225                                      !remote)) {
 226                                 ucred->uc_identity = NULL;
 227                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 228                                        CFS_SETGRP_PERM;
 229                         } else {
 230                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 231                                        pud->pud_uid);
 232                                 RETURN(-EACCES);
 233                         }
 234                 } else {
 235                         ucred->uc_identity = identity;
 236                         perm = mdt_identity_get_perm(ucred->uc_identity,
 237                                                      remote, peernid);
 238                 }
 239         }
 240
 241         /* find out the setuid/setgid attempt */
 242         setuid = (pud->pud_uid != pud->pud_fsuid);
 243         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 244                   (ucred->uc_identity &&
 245                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 246
 247         /* check permission of setuid */
 248         if (setuid && !(perm & CFS_SETUID_PERM)) {
 249                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 250                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 251                 GOTO(out, rc = -EACCES);
 252         }
 253
 254         /* check permission of setgid */
 255         if (setgid && !(perm & CFS_SETGID_PERM)) {
 256                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 257                        "from %s\n", pud->pud_uid, pud->pud_gid,
 258                        pud->pud_fsuid, pud->pud_fsgid,
 259                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 260                 GOTO(out, rc = -EACCES);
 261         }
 262
 263         /*
 264          * NB: remote client not allowed to setgroups anyway.
 265          */
 266         if (!remote && perm & CFS_SETGRP_PERM) {
 267                 if (pud->pud_ngroups) {
 268                         /* setgroups for local client */
 269                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 270                         if (!ucred->uc_ginfo) {
 271                                 CERROR("failed to alloc %d groups\n",
 272                                        pud->pud_ngroups);
 273                                 GOTO(out, rc = -ENOMEM);
 274                         }
 275
 276                         lustre_groups_from_list(ucred->uc_ginfo,
 277                                                 pud->pud_groups);
 278                         lustre_groups_sort(ucred->uc_ginfo);
 279                 } else {
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282         } else {
 283                 ucred->uc_suppgids[0] = -1;
 284                 ucred->uc_suppgids[1] = -1;
 285                 ucred->uc_ginfo = NULL;
 286         }
 287
 288         ucred->uc_uid   = pud->pud_uid;
 289         ucred->uc_gid   = pud->pud_gid;
 290         ucred->uc_fsuid = pud->pud_fsuid;
 291         ucred->uc_fsgid = pud->pud_fsgid;
 292
 293         /* process root_squash here. */
 294         mdt_root_squash(info, peernid);
 295
 296         /* remove fs privilege for non-root user. */
 297         if (ucred->uc_fsuid && drop_fs_cap)
 298                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 299         else
 300                 ucred->uc_cap = pud->pud_cap;
 301         if (remote && !(perm & CFS_RMTOWN_PERM))
 302                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 303                                    CFS_CAP_CHOWN_MASK);
 304         ucred->uc_valid = UCRED_NEW;
 305         ucred_set_jobid(info, ucred);
 306
 307         EXIT;
 308
 309 out:
 310         if (rc) {
 311                 if (ucred->uc_ginfo) {
 312                         put_group_info(ucred->uc_ginfo);
 313                         ucred->uc_ginfo = NULL;
 314                 }
 315                 if (ucred->uc_identity) {
 316                         mdt_identity_put(mdt->mdt_identity_cache,
 317                                          ucred->uc_identity);
 318                         ucred->uc_identity = NULL;
 319                 }
 320         }
 321
 322         return rc;
 323 }
 324
 325 /**
 326  * Check whether allow the client to set supplementary group IDs or not.
 327  *
 328  * \param[in] info      pointer to the thread context
 329  * \param[in] uc        pointer to the RPC user descriptor
 330  *
 331  * \retval              true if allow to set supplementary group IDs
 332  * \retval              false for other cases
 333  */
 334 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 335 {
 336         __u32 remote = exp_connect_rmtclient(info->mti_exp);
 337         __u32 perm;
 338
 339         /* 1. If identity_upcall is disabled, then forbid remote client to set
 340          *    supplementary group IDs, but permit local client to do that. */
 341         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) {
 342                 if (remote)
 343                         return false;
 344
 345                 return true;
 346         }
 347
 348         /* 2. If fail to get related identities, then forbid any client to
 349          *    set supplementary group IDs. */
 350         if (uc->uc_identity == NULL)
 351                 return false;
 352
 353         /* 3. Check the permission in the identities. */
 354         perm = mdt_identity_get_perm(uc->uc_identity, remote,
 355                                      mdt_info_req(info)->rq_peer.nid);
 356         if (perm & CFS_SETGRP_PERM)
 357                 return true;
 358
 359         return false;
 360 }
 361
 362 int mdt_check_ucred(struct mdt_thread_info *info)
 363 {
 364         struct ptlrpc_request   *req = mdt_info_req(info);
 365         struct mdt_device       *mdt = info->mti_mdt;
 366         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 367         struct lu_ucred         *ucred = mdt_ucred(info);
 368         struct md_identity      *identity = NULL;
 369         lnet_nid_t               peernid = req->rq_peer.nid;
 370         __u32                    perm = 0;
 371         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 372         int                      setuid;
 373         int                      setgid;
 374         int                      rc = 0;
 375
 376         ENTRY;
 377
 378         LASSERT(ucred != NULL);
 379         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 380                 RETURN(0);
 381
 382         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 383                 RETURN(0);
 384
 385         /* sanity check: if we use strong authentication, we expect the
 386          * uid which client claimed is true */
 387         if (remote) {
 388                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 389                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 390                         RETURN(-EACCES);
 391                 }
 392
 393                 if (ptlrpc_user_desc_do_idmap(req, pud))
 394                         RETURN(-EACCES);
 395
 396                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 397                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 398                                "while client claims %u:%u/%u:%u\n",
 399                                libcfs_nid2str(peernid), req->rq_auth_uid,
 400                                req->rq_auth_mapped_uid,
 401                                pud->pud_uid, pud->pud_gid,
 402                                pud->pud_fsuid, pud->pud_fsgid);
 403                         RETURN(-EACCES);
 404                 }
 405         } else {
 406                 if (req->rq_auth_uid != pud->pud_uid) {
 407                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 408                                "while client claims %u:%u/%u:%u\n",
 409                                libcfs_nid2str(peernid), req->rq_auth_uid,
 410                                pud->pud_uid, pud->pud_gid,
 411                                pud->pud_fsuid, pud->pud_fsgid);
 412                         RETURN(-EACCES);
 413                 }
 414         }
 415
 416         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 417                 if (remote) {
 418                         CDEBUG(D_SEC, "remote client must run with identity_get "
 419                                "enabled!\n");
 420                         RETURN(-EACCES);
 421                 }
 422                 RETURN(0);
 423         }
 424
 425         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 426         if (IS_ERR(identity)) {
 427                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 428                              !remote)) {
 429                         RETURN(0);
 430                 } else {
 431                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 432                                pud->pud_uid);
 433                         RETURN(-EACCES);
 434                }
 435         }
 436
 437         perm = mdt_identity_get_perm(identity, remote, peernid);
 438         /* find out the setuid/setgid attempt */
 439         setuid = (pud->pud_uid != pud->pud_fsuid);
 440         setgid = (pud->pud_gid != pud->pud_fsgid ||
 441                   pud->pud_gid != identity->mi_gid);
 442
 443         /* check permission of setuid */
 444         if (setuid && !(perm & CFS_SETUID_PERM)) {
 445                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 446                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 447                 GOTO(out, rc = -EACCES);
 448         }
 449
 450         /* check permission of setgid */
 451         if (setgid && !(perm & CFS_SETGID_PERM)) {
 452                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 453                        "from %s\n", pud->pud_uid, pud->pud_gid,
 454                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 455                        libcfs_nid2str(peernid));
 456                 GOTO(out, rc = -EACCES);
 457         }
 458
 459         EXIT;
 460
 461 out:
 462         mdt_identity_put(mdt->mdt_identity_cache, identity);
 463         return rc;
 464 }
 465
 466 static int old_init_ucred_common(struct mdt_thread_info *info,
 467                                   bool drop_fs_cap)
 468 {
 469         struct lu_ucred         *uc = mdt_ucred(info);
 470         struct mdt_device       *mdt = info->mti_mdt;
 471         struct md_identity      *identity = NULL;
 472         struct lu_nodemap       *nodemap =
 473                 info->mti_exp->exp_target_data.ted_nodemap;
 474
 475         if (nodemap == NULL) {
 476                 CDEBUG(D_SEC, "%s: cli %s/%p nodemap not set.\n",
 477                        mdt2obd_dev(mdt)->obd_name,
 478                        info->mti_exp->exp_client_uuid.uuid, info->mti_exp);
 479                 RETURN(-EACCES);
 480         }
 481
 482         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 483                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 484                                             uc->uc_fsuid);
 485                 if (IS_ERR(identity)) {
 486                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 487                                 identity = NULL;
 488                         } else {
 489                                 CDEBUG(D_SEC, "Deny access without identity: "
 490                                        "uid %u\n", uc->uc_fsuid);
 491                                 RETURN(-EACCES);
 492                         }
 493                 }
 494         }
 495         uc->uc_identity = identity;
 496
 497         if (uc->uc_o_uid == nodemap->nm_squash_uid) {
 498                 uc->uc_fsuid = nodemap->nm_squash_uid;
 499                 uc->uc_fsgid = nodemap->nm_squash_gid;
 500                 uc->uc_cap = 0;
 501                 uc->uc_suppgids[0] = -1;
 502                 uc->uc_suppgids[1] = -1;
 503         }
 504
 505         /* process root_squash here. */
 506         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 507
 508         /* remove fs privilege for non-root user. */
 509         if (uc->uc_fsuid)
 510                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 511         uc->uc_valid = UCRED_OLD;
 512         ucred_set_jobid(info, uc);
 513
 514         return 0;
 515 }
 516
 517 static int old_init_ucred(struct mdt_thread_info *info,
 518                           struct mdt_body *body, bool drop_fs_cap)
 519 {
 520         struct lu_ucred         *uc = mdt_ucred(info);
 521         struct lu_nodemap       *nodemap =
 522                 info->mti_exp->exp_target_data.ted_nodemap;
 523         int                      rc;
 524         ENTRY;
 525
 526         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 527                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 528         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 529                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 530         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 531                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 532         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 533                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 534
 535         LASSERT(uc != NULL);
 536         uc->uc_valid = UCRED_INVALID;
 537         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 538         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 539         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 540         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 541         uc->uc_suppgids[0] = body->mbo_suppgid;
 542         uc->uc_suppgids[1] = -1;
 543         uc->uc_ginfo = NULL;
 544         uc->uc_cap = body->mbo_capability;
 545
 546         rc = old_init_ucred_common(info, drop_fs_cap);
 547
 548         RETURN(rc);
 549 }
 550
 551 static int old_init_ucred_reint(struct mdt_thread_info *info)
 552 {
 553         struct lu_ucred         *uc = mdt_ucred(info);
 554         struct lu_nodemap       *nodemap =
 555                 info->mti_exp->exp_target_data.ted_nodemap;
 556         int                      rc;
 557         ENTRY;
 558
 559         LASSERT(uc != NULL);
 560
 561         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 562                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 563         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 564                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 565
 566         uc->uc_valid = UCRED_INVALID;
 567         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 568         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 569         uc->uc_ginfo = NULL;
 570
 571         rc = old_init_ucred_common(info, true); /* drop_fs_cap = true */
 572
 573         RETURN(rc);
 574 }
 575
 576 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 577                                    struct mdt_body *body,
 578                                    bool drop_fs_cap)
 579 {
 580         struct ptlrpc_request   *req = mdt_info_req(info);
 581         struct lu_ucred         *uc  = mdt_ucred(info);
 582
 583         LASSERT(uc != NULL);
 584         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 585                 return 0;
 586
 587         mdt_exit_ucred(info);
 588
 589         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 590                 return old_init_ucred(info, body, drop_fs_cap);
 591         else
 592                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 593 }
 594
 595 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 596 {
 597         return __mdt_init_ucred(info, body, true);
 598 }
 599
 600 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 601  * doesn't set correct fsuid explicitely, but raise capability to allow
 602  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 603  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 604  * will fail permission check. */
 605 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 606                                   struct mdt_body *body)
 607 {
 608         return __mdt_init_ucred(info, body, false);
 609 }
 610
 611 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 612 {
 613         struct ptlrpc_request *req = mdt_info_req(info);
 614         struct lu_ucred       *uc  = mdt_ucred(info);
 615
 616         LASSERT(uc != NULL);
 617         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 618                 return 0;
 619
 620         mdt_exit_ucred(info);
 621
 622         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 623                 return old_init_ucred_reint(info);
 624         else
 625                 return new_init_ucred(info, REC_INIT, NULL, true);
 626 }
 627
 628 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 629 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 630 {
 631         const struct lov_ost_data_v1    *lod;
 632         int                              i;
 633         __u16                            count;
 634
 635         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 636                 return;
 637
 638         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 639
 640         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 641                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 642                le32_to_cpu(lmm->lmm_pattern));
 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         int                        i;
 666
 667         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 668                 return;
 669
 670         lmm1 = &lmv->lmv_md_v1;
 671         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 672                le32_to_cpu(lmm1->lmv_magic),
 673                le32_to_cpu(lmm1->lmv_master_mdt_index),
 674                le32_to_cpu(lmm1->lmv_stripe_count));
 675
 676         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 677                 return;
 678
 679         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 680                 struct lu_fid fid;
 681
 682                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 683                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 684         }
 685 }
 686
 687 /* Shrink and/or grow reply buffers */
 688 int mdt_fix_reply(struct mdt_thread_info *info)
 689 {
 690         struct req_capsule *pill = info->mti_pill;
 691         struct mdt_body    *body;
 692         int                md_size, md_packed = 0;
 693         int                acl_size;
 694         int                rc = 0;
 695         ENTRY;
 696
 697         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 698         LASSERT(body != NULL);
 699
 700         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 701                                OBD_MD_LINKNAME))
 702                 md_size = body->mbo_eadatasize;
 703         else
 704                 md_size = 0;
 705
 706         acl_size = body->mbo_aclsize;
 707
 708         /* this replay - not send info to client */
 709         if (info->mti_spec.no_create) {
 710                 md_size = 0;
 711                 acl_size = 0;
 712         }
 713
 714         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 715                md_size, acl_size);
 716 /*
 717             &RMF_MDT_BODY,
 718             &RMF_MDT_MD,
 719             &RMF_ACL, or &RMF_LOGCOOKIES
 720 (optional)  &RMF_CAPA1,
 721 (optional)  &RMF_CAPA2,
 722 (optional)  something else
 723 */
 724
 725         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 726          * buffers before growing it */
 727         if (info->mti_big_lmm_used) {
 728                 /* big_lmm buffer may be used even without packing the result
 729                  * into reply, just for internal server needs */
 730                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 731                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 732                                                          RCL_SERVER);
 733
 734                 /* free big lmm if md_size is not needed */
 735                 if (md_size == 0 || md_packed == 0) {
 736                         info->mti_big_lmm_used = 0;
 737                 } else {
 738                         /* buffer must be allocated separately */
 739                         LASSERT(info->mti_attr.ma_lmm !=
 740                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 741                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 742                 }
 743         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 744                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 745         }
 746
 747         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 748                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 749         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 750                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 751                                    acl_size, RCL_SERVER);
 752
 753         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 754             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 755                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 756
 757         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 758             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 759                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 760
 761         /*
 762          * Some more field should be shrinked if needed.
 763          * This should be done by those who added fields to reply message.
 764          */
 765
 766         /* Grow MD buffer if needed finally */
 767         if (info->mti_big_lmm_used) {
 768                 void *lmm;
 769
 770                 LASSERT(md_size > md_packed);
 771                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 772                        md_size - md_packed);
 773                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 774                 if (rc) {
 775                         /* we can't answer with proper LOV EA, drop flags,
 776                          * the rc is also returned so this request is
 777                          * considered as failed */
 778                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 779                         /* don't return transno along with error */
 780                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 781                 } else {
 782                         /* now we need to pack right LOV/LMV EA */
 783                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 784                         if (info->mti_attr.ma_valid & MA_LOV) {
 785                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 786                                                              RCL_SERVER) ==
 787                                                 info->mti_attr.ma_lmm_size);
 788                                 memcpy(lmm, info->mti_attr.ma_lmm,
 789                                        info->mti_attr.ma_lmm_size);
 790                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 791                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 792                                                              RCL_SERVER) ==
 793                                                 info->mti_attr.ma_lmv_size);
 794                                 memcpy(lmm, info->mti_attr.ma_lmv,
 795                                        info->mti_attr.ma_lmv_size);
 796                         }
 797                 }
 798                 /* update mdt_max_mdsize so clients will be aware about that */
 799                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 800                         info->mti_mdt->mdt_max_mdsize =
 801                                                     info->mti_attr.ma_lmm_size;
 802                 info->mti_big_lmm_used = 0;
 803         }
 804         RETURN(rc);
 805 }
 806
 807
 808 /* if object is dying, pack the lov/llog data,
 809  * parameter info->mti_attr should be valid at this point! */
 810 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 811                            const struct md_attr *ma)
 812 {
 813         struct mdt_body       *repbody;
 814         const struct lu_attr *la = &ma->ma_attr;
 815         ENTRY;
 816
 817         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 818         LASSERT(repbody != NULL);
 819
 820         if (ma->ma_valid & MA_INODE)
 821                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 822
 823         if (ma->ma_valid & MA_LOV) {
 824                 CERROR("No need in LOV EA upon unlink\n");
 825                 dump_stack();
 826         }
 827         repbody->mbo_eadatasize = 0;
 828
 829         RETURN(0);
 830 }
 831
 832 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 833                                   struct md_attr *ma)
 834 {
 835         __u64 out;
 836
 837         out = 0;
 838         if (in & MDS_ATTR_MODE)
 839                 out |= LA_MODE;
 840         if (in & MDS_ATTR_UID)
 841                 out |= LA_UID;
 842         if (in & MDS_ATTR_GID)
 843                 out |= LA_GID;
 844         if (in & MDS_ATTR_SIZE)
 845                 out |= LA_SIZE;
 846         if (in & MDS_ATTR_BLOCKS)
 847                 out |= LA_BLOCKS;
 848         if (in & MDS_ATTR_ATIME_SET)
 849                 out |= LA_ATIME;
 850         if (in & MDS_ATTR_CTIME_SET)
 851                 out |= LA_CTIME;
 852         if (in & MDS_ATTR_MTIME_SET)
 853                 out |= LA_MTIME;
 854         if (in & MDS_ATTR_ATTR_FLAG)
 855                 out |= LA_FLAGS;
 856         if (in & MDS_ATTR_KILL_SUID)
 857                 out |= LA_KILL_SUID;
 858         if (in & MDS_ATTR_KILL_SGID)
 859                 out |= LA_KILL_SGID;
 860
 861         if (in & MDS_ATTR_FROM_OPEN)
 862                 rr->rr_flags |= MRF_OPEN_TRUNC;
 863         if (in & MDS_OPEN_OWNEROVERRIDE)
 864                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 865         if (in & MDS_ATTR_FORCE)
 866                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 867
 868         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 869                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 870                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 871                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 872                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 873                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 874         if (in != 0)
 875                 CERROR("Unknown attr bits: "LPX64"\n", in);
 876         return out;
 877 }
 878
 879 /* unpacking */
 880
 881 int mdt_name_unpack(struct req_capsule *pill,
 882                     const struct req_msg_field *field,
 883                     struct lu_name *ln,
 884                     enum mdt_name_flags flags)
 885 {
 886         ln->ln_name = req_capsule_client_get(pill, field);
 887         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 888
 889         if (!lu_name_is_valid(ln)) {
 890                 ln->ln_name = NULL;
 891                 ln->ln_namelen = 0;
 892
 893                 return -EPROTO;
 894         }
 895
 896         if ((flags & MNF_FIX_ANON) &&
 897             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 898                 /* Newer (3.x) kernels use a name of "/" for the
 899                  * "anonymous" disconnected dentries from NFS
 900                  * filehandle conversion. See d_obtain_alias(). */
 901                 ln->ln_name = NULL;
 902                 ln->ln_namelen = 0;
 903         }
 904
 905         return 0;
 906 }
 907
 908 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 909 {
 910         struct lu_ucred         *uc = mdt_ucred(info);
 911         struct md_attr          *ma = &info->mti_attr;
 912         struct lu_attr          *la = &ma->ma_attr;
 913         struct req_capsule      *pill = info->mti_pill;
 914         struct mdt_reint_record *rr = &info->mti_rr;
 915         struct mdt_rec_setattr  *rec;
 916         struct lu_nodemap       *nodemap =
 917                 info->mti_exp->exp_target_data.ted_nodemap;
 918         ENTRY;
 919
 920         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 921         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 922         if (rec == NULL)
 923                 RETURN(-EFAULT);
 924
 925         /* This prior initialization is needed for old_init_ucred_reint() */
 926         uc->uc_fsuid = rec->sa_fsuid;
 927         uc->uc_fsgid = rec->sa_fsgid;
 928         uc->uc_cap   = rec->sa_cap;
 929         uc->uc_suppgids[0] = rec->sa_suppgid;
 930         uc->uc_suppgids[1] = -1;
 931
 932         rr->rr_fid1 = &rec->sa_fid;
 933         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 934         la->la_mode  = rec->sa_mode;
 935         la->la_flags = rec->sa_attr_flags;
 936         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 937                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 938         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 939                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 940         la->la_size  = rec->sa_size;
 941         la->la_blocks = rec->sa_blocks;
 942         la->la_ctime = rec->sa_ctime;
 943         la->la_atime = rec->sa_atime;
 944         la->la_mtime = rec->sa_mtime;
 945         ma->ma_valid = MA_INODE;
 946
 947         if (rec->sa_bias & MDS_DATA_MODIFIED)
 948                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 949         else
 950                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 951
 952         if (rec->sa_bias & MDS_HSM_RELEASE)
 953                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 954         else
 955                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 956
 957         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 958                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 959         else
 960                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 961
 962         RETURN(0);
 963 }
 964
 965 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 966 {
 967         struct req_capsule *pill = info->mti_pill;
 968         struct mdt_ioepoch *ioepoch;
 969         ENTRY;
 970
 971         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 972                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 973         else
 974                 ioepoch = NULL;
 975
 976         if (ioepoch == NULL)
 977                 RETURN(-EPROTO);
 978
 979         info->mti_close_handle = ioepoch->mio_handle;
 980
 981         RETURN(0);
 982 }
 983
 984 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 985         struct req_capsule      *pill = info->mti_pill;
 986
 987         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 988                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 989                 if (info->mti_dlm_req == NULL)
 990                         RETURN(-EFAULT);
 991         }
 992
 993         RETURN(0);
 994 }
 995
 996 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 997 {
 998         struct mdt_reint_record *rr = &info->mti_rr;
 999         struct md_attr          *ma = &info->mti_attr;
1000         struct req_capsule      *pill = info->mti_pill;
1001         int rc;
1002         ENTRY;
1003
1004         rc = mdt_setattr_unpack_rec(info);
1005         if (rc)
1006                 RETURN(rc);
1007
1008         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1009                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1010                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1011                                                         RCL_CLIENT);
1012                 if (rr->rr_eadatalen > 0) {
1013                         const struct lmv_user_md        *lum;
1014
1015                         lum = rr->rr_eadata;
1016                         /* Sigh ma_valid(from req) does not indicate whether
1017                          * it will set LOV/LMV EA, so we have to check magic */
1018                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1019                                 ma->ma_valid |= MA_LMV;
1020                                 ma->ma_lmv = (void *)rr->rr_eadata;
1021                                 ma->ma_lmv_size = rr->rr_eadatalen;
1022                         } else {
1023                                 ma->ma_valid |= MA_LOV;
1024                                 ma->ma_lmm = (void *)rr->rr_eadata;
1025                                 ma->ma_lmm_size = rr->rr_eadatalen;
1026                         }
1027                 }
1028         }
1029
1030         rc = mdt_dlmreq_unpack(info);
1031         RETURN(rc);
1032 }
1033
1034 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1035 {
1036         struct md_attr          *ma = &info->mti_attr;
1037         struct req_capsule      *pill = info->mti_pill;
1038         ENTRY;
1039
1040         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1041                 RETURN(0);
1042
1043         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1044
1045         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1046             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1047                 RETURN(-EFAULT);
1048
1049         RETURN(0);
1050 }
1051
1052 int mdt_close_unpack(struct mdt_thread_info *info)
1053 {
1054         int rc;
1055         ENTRY;
1056
1057         rc = mdt_close_handle_unpack(info);
1058         if (rc)
1059                 RETURN(rc);
1060
1061         rc = mdt_setattr_unpack_rec(info);
1062         if (rc)
1063                 RETURN(rc);
1064
1065         rc = mdt_intent_close_unpack(info);
1066         if (rc)
1067                 RETURN(rc);
1068
1069         RETURN(mdt_init_ucred_reint(info));
1070 }
1071
1072 static int mdt_create_unpack(struct mdt_thread_info *info)
1073 {
1074         struct lu_ucred         *uc  = mdt_ucred(info);
1075         struct mdt_rec_create   *rec;
1076         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1077         struct mdt_reint_record *rr = &info->mti_rr;
1078         struct req_capsule      *pill = info->mti_pill;
1079         struct md_op_spec       *sp = &info->mti_spec;
1080         int rc;
1081         ENTRY;
1082
1083         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1084         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1085         if (rec == NULL)
1086                 RETURN(-EFAULT);
1087
1088         /* This prior initialization is needed for old_init_ucred_reint() */
1089         uc->uc_fsuid = rec->cr_fsuid;
1090         uc->uc_fsgid = rec->cr_fsgid;
1091         uc->uc_cap   = rec->cr_cap;
1092         uc->uc_suppgids[0] = rec->cr_suppgid1;
1093         uc->uc_suppgids[1] = -1;
1094         uc->uc_umask = rec->cr_umask;
1095
1096         rr->rr_fid1 = &rec->cr_fid1;
1097         rr->rr_fid2 = &rec->cr_fid2;
1098         attr->la_mode = rec->cr_mode;
1099         attr->la_rdev  = rec->cr_rdev;
1100         attr->la_uid   = rec->cr_fsuid;
1101         attr->la_gid   = rec->cr_fsgid;
1102         attr->la_ctime = rec->cr_time;
1103         attr->la_mtime = rec->cr_time;
1104         attr->la_atime = rec->cr_time;
1105         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1106                          LA_CTIME | LA_MTIME | LA_ATIME;
1107         memset(&sp->u, 0, sizeof(sp->u));
1108         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1109
1110         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1111         if (rc < 0)
1112                 RETURN(rc);
1113
1114         if (S_ISLNK(attr->la_mode)) {
1115                 const char *tgt = NULL;
1116
1117                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1118                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1119                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1120                         sp->u.sp_symname = tgt;
1121                 }
1122                 if (tgt == NULL)
1123                         RETURN(-EFAULT);
1124         } else {
1125                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1126                 if (S_ISDIR(attr->la_mode) &&
1127                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1128                         sp->u.sp_ea.eadata =
1129                                 req_capsule_client_get(pill, &RMF_EADATA);
1130                         sp->u.sp_ea.eadatalen =
1131                                 req_capsule_get_size(pill, &RMF_EADATA,
1132                                                      RCL_CLIENT);
1133                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1134                 }
1135         }
1136
1137         rc = mdt_dlmreq_unpack(info);
1138         RETURN(rc);
1139 }
1140
1141 static int mdt_link_unpack(struct mdt_thread_info *info)
1142 {
1143         struct lu_ucred         *uc  = mdt_ucred(info);
1144         struct mdt_rec_link     *rec;
1145         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1146         struct mdt_reint_record *rr = &info->mti_rr;
1147         struct req_capsule      *pill = info->mti_pill;
1148         int rc;
1149         ENTRY;
1150
1151         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1152         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1153         if (rec == NULL)
1154                 RETURN(-EFAULT);
1155
1156         /* This prior initialization is needed for old_init_ucred_reint() */
1157         uc->uc_fsuid = rec->lk_fsuid;
1158         uc->uc_fsgid = rec->lk_fsgid;
1159         uc->uc_cap   = rec->lk_cap;
1160         uc->uc_suppgids[0] = rec->lk_suppgid1;
1161         uc->uc_suppgids[1] = rec->lk_suppgid2;
1162
1163         attr->la_uid = rec->lk_fsuid;
1164         attr->la_gid = rec->lk_fsgid;
1165         rr->rr_fid1 = &rec->lk_fid1;
1166         rr->rr_fid2 = &rec->lk_fid2;
1167         attr->la_ctime = rec->lk_time;
1168         attr->la_mtime = rec->lk_time;
1169         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1170
1171         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1172         if (rc < 0)
1173                 RETURN(rc);
1174
1175         rc = mdt_dlmreq_unpack(info);
1176
1177         RETURN(rc);
1178 }
1179
1180 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1181 {
1182         struct lu_ucred         *uc  = mdt_ucred(info);
1183         struct mdt_rec_unlink   *rec;
1184         struct md_attr          *ma = &info->mti_attr;
1185         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1186         struct mdt_reint_record *rr = &info->mti_rr;
1187         struct req_capsule      *pill = info->mti_pill;
1188         int rc;
1189         ENTRY;
1190
1191         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1192         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1193         if (rec == NULL)
1194                 RETURN(-EFAULT);
1195
1196         /* This prior initialization is needed for old_init_ucred_reint() */
1197         uc->uc_fsuid = rec->ul_fsuid;
1198         uc->uc_fsgid = rec->ul_fsgid;
1199         uc->uc_cap   = rec->ul_cap;
1200         uc->uc_suppgids[0] = rec->ul_suppgid1;
1201         uc->uc_suppgids[1] = -1;
1202
1203         attr->la_uid = rec->ul_fsuid;
1204         attr->la_gid = rec->ul_fsgid;
1205         rr->rr_fid1 = &rec->ul_fid1;
1206         rr->rr_fid2 = &rec->ul_fid2;
1207         attr->la_ctime = rec->ul_time;
1208         attr->la_mtime = rec->ul_time;
1209         attr->la_mode  = rec->ul_mode;
1210         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1211
1212         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1213         if (rc < 0)
1214                 RETURN(rc);
1215
1216         if (rec->ul_bias & MDS_VTX_BYPASS)
1217                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1218         else
1219                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1220
1221         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1222
1223         rc = mdt_dlmreq_unpack(info);
1224         RETURN(rc);
1225 }
1226
1227 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1228 {
1229         info->mti_spec.sp_rm_entry = 1;
1230         return mdt_unlink_unpack(info);
1231 }
1232
1233 static int mdt_rename_unpack(struct mdt_thread_info *info)
1234 {
1235         struct lu_ucred         *uc = mdt_ucred(info);
1236         struct mdt_rec_rename   *rec;
1237         struct md_attr          *ma = &info->mti_attr;
1238         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1239         struct mdt_reint_record *rr = &info->mti_rr;
1240         struct req_capsule      *pill = info->mti_pill;
1241         int rc;
1242         ENTRY;
1243
1244         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1245         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1246         if (rec == NULL)
1247                 RETURN(-EFAULT);
1248
1249         /* This prior initialization is needed for old_init_ucred_reint() */
1250         uc->uc_fsuid = rec->rn_fsuid;
1251         uc->uc_fsgid = rec->rn_fsgid;
1252         uc->uc_cap   = rec->rn_cap;
1253         uc->uc_suppgids[0] = rec->rn_suppgid1;
1254         uc->uc_suppgids[1] = rec->rn_suppgid2;
1255
1256         attr->la_uid = rec->rn_fsuid;
1257         attr->la_gid = rec->rn_fsgid;
1258         rr->rr_fid1 = &rec->rn_fid1;
1259         rr->rr_fid2 = &rec->rn_fid2;
1260         attr->la_ctime = rec->rn_time;
1261         attr->la_mtime = rec->rn_time;
1262         /* rename_tgt contains the mode already */
1263         attr->la_mode = rec->rn_mode;
1264         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1265
1266         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1267         if (rc < 0)
1268                 RETURN(rc);
1269
1270         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1271         if (rc < 0)
1272                 RETURN(rc);
1273
1274         if (rec->rn_bias & MDS_VTX_BYPASS)
1275                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1276         else
1277                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1278
1279         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1280                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1281                 rc = mdt_close_handle_unpack(info);
1282                 if (rc < 0)
1283                         RETURN(rc);
1284                 info->mti_spec.sp_migrate_close = 1;
1285         }
1286
1287         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1288
1289
1290         rc = mdt_dlmreq_unpack(info);
1291
1292         RETURN(rc);
1293 }
1294
1295 /*
1296  * please see comment above LOV_MAGIC_V1_DEF
1297  */
1298 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1299 {
1300         struct mdt_reint_record *rr = &info->mti_rr;
1301         struct lov_user_md_v1   *v1;
1302
1303         v1 = (void *)rr->rr_eadata;
1304         LASSERT(v1);
1305
1306         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1307                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1308                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1309                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1310                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1311                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1312                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1313                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1314                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1315                 }
1316         }
1317 }
1318
1319 static int mdt_open_unpack(struct mdt_thread_info *info)
1320 {
1321         struct lu_ucred         *uc = mdt_ucred(info);
1322         struct mdt_rec_create   *rec;
1323         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1324         struct req_capsule      *pill = info->mti_pill;
1325         struct mdt_reint_record *rr   = &info->mti_rr;
1326         struct ptlrpc_request   *req  = mdt_info_req(info);
1327         struct md_op_spec       *sp   = &info->mti_spec;
1328         ENTRY;
1329
1330         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1331         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1332         if (rec == NULL)
1333                 RETURN(-EFAULT);
1334
1335         /* This prior initialization is needed for old_init_ucred_reint() */
1336         uc->uc_fsuid = rec->cr_fsuid;
1337         uc->uc_fsgid = rec->cr_fsgid;
1338         uc->uc_cap   = rec->cr_cap;
1339         uc->uc_suppgids[0] = rec->cr_suppgid1;
1340         uc->uc_suppgids[1] = rec->cr_suppgid2;
1341         uc->uc_umask = rec->cr_umask;
1342
1343         rr->rr_fid1   = &rec->cr_fid1;
1344         rr->rr_fid2   = &rec->cr_fid2;
1345         rr->rr_handle = &rec->cr_old_handle;
1346         attr->la_mode = rec->cr_mode;
1347         attr->la_rdev  = rec->cr_rdev;
1348         attr->la_uid   = rec->cr_fsuid;
1349         attr->la_gid   = rec->cr_fsgid;
1350         attr->la_ctime = rec->cr_time;
1351         attr->la_mtime = rec->cr_time;
1352         attr->la_atime = rec->cr_time;
1353         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1354                          LA_CTIME | LA_MTIME | LA_ATIME;
1355         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1356         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1357         /* Do not trigger ASSERTION if client miss to set such flags. */
1358         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1359                 RETURN(-EPROTO);
1360
1361         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1362
1363         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1364
1365         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1366                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1367                                                         RCL_CLIENT);
1368                 if (rr->rr_eadatalen > 0) {
1369                         rr->rr_eadata = req_capsule_client_get(pill,
1370                                                                &RMF_EADATA);
1371                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1372                         sp->u.sp_ea.eadata = rr->rr_eadata;
1373                         sp->no_create = !!req_is_replay(req);
1374                         mdt_fix_lov_magic(info);
1375                 }
1376
1377                 /*
1378                  * Client default md_size may be 0 right after client start,
1379                  * until all osc are connected, set here just some reasonable
1380                  * value to prevent misbehavior.
1381                  */
1382                 if (rr->rr_eadatalen == 0 &&
1383                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1384                         rr->rr_eadatalen = MIN_MD_SIZE;
1385         }
1386
1387         RETURN(0);
1388 }
1389
1390 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1391 {
1392         struct mdt_reint_record *rr     = &info->mti_rr;
1393         struct lu_ucred         *uc     = mdt_ucred(info);
1394         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1395         struct req_capsule      *pill   = info->mti_pill;
1396         struct mdt_rec_setxattr *rec;
1397         int                      rc;
1398         ENTRY;
1399
1400
1401         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1402                          sizeof(struct mdt_rec_reint));
1403
1404         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1405         if (rec == NULL)
1406                 RETURN(-EFAULT);
1407
1408         /* This prior initialization is needed for old_init_ucred_reint() */
1409         uc->uc_fsuid  = rec->sx_fsuid;
1410         uc->uc_fsgid  = rec->sx_fsgid;
1411         uc->uc_cap    = rec->sx_cap;
1412         uc->uc_suppgids[0] = rec->sx_suppgid1;
1413         uc->uc_suppgids[1] = -1;
1414
1415         rr->rr_opcode = rec->sx_opcode;
1416         rr->rr_fid1   = &rec->sx_fid;
1417         attr->la_valid = rec->sx_valid;
1418         attr->la_ctime = rec->sx_time;
1419         attr->la_size = rec->sx_size;
1420         attr->la_flags = rec->sx_flags;
1421
1422         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1423         if (rc < 0)
1424                 RETURN(rc);
1425
1426         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1427                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1428                                                         RCL_CLIENT);
1429                 if (rr->rr_eadatalen > 0) {
1430                         rr->rr_eadata = req_capsule_client_get(pill,
1431                                                                &RMF_EADATA);
1432                         if (rr->rr_eadata == NULL)
1433                                 RETURN(-EFAULT);
1434                 } else {
1435                         rr->rr_eadata = NULL;
1436                 }
1437         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1438                 CDEBUG(D_INFO, "no xattr data supplied\n");
1439                 RETURN(-EFAULT);
1440         }
1441
1442         if (mdt_dlmreq_unpack(info) < 0)
1443                 RETURN(-EPROTO);
1444
1445         RETURN(0);
1446 }
1447
1448
1449 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1450
1451 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1452         [REINT_SETATTR]  = mdt_setattr_unpack,
1453         [REINT_CREATE]   = mdt_create_unpack,
1454         [REINT_LINK]     = mdt_link_unpack,
1455         [REINT_UNLINK]   = mdt_unlink_unpack,
1456         [REINT_RENAME]   = mdt_rename_unpack,
1457         [REINT_OPEN]     = mdt_open_unpack,
1458         [REINT_SETXATTR] = mdt_setxattr_unpack,
1459         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1460         [REINT_MIGRATE]  = mdt_rename_unpack,
1461 };
1462
1463 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1464 {
1465         int rc;
1466         ENTRY;
1467
1468         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1469         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1470                 info->mti_rr.rr_opcode = op;
1471                 rc = mdt_reint_unpackers[op](info);
1472         } else {
1473                 CERROR("Unexpected opcode %d\n", op);
1474                 rc = -EFAULT;
1475         }
1476         RETURN(rc);
1477 }