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                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 488                                 identity = NULL;
 489                         } else {
 490                                 CDEBUG(D_SEC, "Deny access without identity: "
 491                                        "uid %u\n", uc->uc_fsuid);
 492                                 RETURN(-EACCES);
 493                         }
 494                 }
 495         }
 496         uc->uc_identity = identity;
 497
 498         if (uc->uc_o_uid == nodemap->nm_squash_uid) {
 499                 uc->uc_fsuid = nodemap->nm_squash_uid;
 500                 uc->uc_fsgid = nodemap->nm_squash_gid;
 501                 uc->uc_cap = 0;
 502                 uc->uc_suppgids[0] = -1;
 503                 uc->uc_suppgids[1] = -1;
 504         }
 505
 506         /* process root_squash here. */
 507         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 508
 509         /* remove fs privilege for non-root user. */
 510         if (uc->uc_fsuid && drop_fs_cap)
 511                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 512         uc->uc_valid = UCRED_OLD;
 513         ucred_set_jobid(info, uc);
 514
 515         return 0;
 516 }
 517
 518 static int old_init_ucred(struct mdt_thread_info *info,
 519                           struct mdt_body *body, bool drop_fs_cap)
 520 {
 521         struct lu_ucred         *uc = mdt_ucred(info);
 522         struct lu_nodemap       *nodemap =
 523                 info->mti_exp->exp_target_data.ted_nodemap;
 524         int                      rc;
 525         ENTRY;
 526
 527         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 528                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 529         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 530                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 531         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 532                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 533         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 534                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 535
 536         LASSERT(uc != NULL);
 537         uc->uc_valid = UCRED_INVALID;
 538         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 539         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 540         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 541         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 542         uc->uc_suppgids[0] = body->mbo_suppgid;
 543         uc->uc_suppgids[1] = -1;
 544         uc->uc_ginfo = NULL;
 545         uc->uc_cap = body->mbo_capability;
 546
 547         rc = old_init_ucred_common(info, drop_fs_cap);
 548
 549         RETURN(rc);
 550 }
 551
 552 static int old_init_ucred_reint(struct mdt_thread_info *info)
 553 {
 554         struct lu_ucred         *uc = mdt_ucred(info);
 555         struct lu_nodemap       *nodemap =
 556                 info->mti_exp->exp_target_data.ted_nodemap;
 557         int                      rc;
 558         ENTRY;
 559
 560         LASSERT(uc != NULL);
 561
 562         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 563                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 564         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 565                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 566
 567         uc->uc_valid = UCRED_INVALID;
 568         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 569         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 570         uc->uc_ginfo = NULL;
 571
 572         rc = old_init_ucred_common(info, true); /* drop_fs_cap = true */
 573
 574         RETURN(rc);
 575 }
 576
 577 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 578                                    struct mdt_body *body,
 579                                    bool drop_fs_cap)
 580 {
 581         struct ptlrpc_request   *req = mdt_info_req(info);
 582         struct lu_ucred         *uc  = mdt_ucred(info);
 583
 584         LASSERT(uc != NULL);
 585         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 586                 return 0;
 587
 588         mdt_exit_ucred(info);
 589
 590         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 591                 return old_init_ucred(info, body, drop_fs_cap);
 592         else
 593                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 594 }
 595
 596 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 597 {
 598         return __mdt_init_ucred(info, body, true);
 599 }
 600
 601 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 602  * doesn't set correct fsuid explicitely, but raise capability to allow
 603  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 604  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 605  * will fail permission check. */
 606 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 607                                   struct mdt_body *body)
 608 {
 609         return __mdt_init_ucred(info, body, false);
 610 }
 611
 612 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 613 {
 614         struct ptlrpc_request *req = mdt_info_req(info);
 615         struct lu_ucred       *uc  = mdt_ucred(info);
 616
 617         LASSERT(uc != NULL);
 618         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 619                 return 0;
 620
 621         mdt_exit_ucred(info);
 622
 623         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 624                 return old_init_ucred_reint(info);
 625         else
 626                 return new_init_ucred(info, REC_INIT, NULL, true);
 627 }
 628
 629 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 630 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 631 {
 632         const struct lov_ost_data_v1    *lod;
 633         int                              i;
 634         __u16                            count;
 635
 636         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 637                 return;
 638
 639         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 640
 641         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 642                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 643                le32_to_cpu(lmm->lmm_pattern));
 644         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 645                le32_to_cpu(lmm->lmm_stripe_size), count);
 646
 647         /* If it's a directory or a released file, then there are
 648          * no actual objects to print, so bail out. */
 649         if (valid & OBD_MD_FLDIREA ||
 650             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 651                 return;
 652
 653         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 654         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 655                 struct ost_id oi;
 656
 657                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 658                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 659                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 660         }
 661 }
 662
 663 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 664 {
 665         const struct lmv_mds_md_v1 *lmm1;
 666         int                        i;
 667
 668         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 669                 return;
 670
 671         lmm1 = &lmv->lmv_md_v1;
 672         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 673                le32_to_cpu(lmm1->lmv_magic),
 674                le32_to_cpu(lmm1->lmv_master_mdt_index),
 675                le32_to_cpu(lmm1->lmv_stripe_count));
 676
 677         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 678                 return;
 679
 680         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 681                 struct lu_fid fid;
 682
 683                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 684                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 685         }
 686 }
 687
 688 /* Shrink and/or grow reply buffers */
 689 int mdt_fix_reply(struct mdt_thread_info *info)
 690 {
 691         struct req_capsule *pill = info->mti_pill;
 692         struct mdt_body    *body;
 693         int                md_size, md_packed = 0;
 694         int                acl_size;
 695         int                rc = 0;
 696         ENTRY;
 697
 698         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 699         LASSERT(body != NULL);
 700
 701         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 702                                OBD_MD_LINKNAME))
 703                 md_size = body->mbo_eadatasize;
 704         else
 705                 md_size = 0;
 706
 707         acl_size = body->mbo_aclsize;
 708
 709         /* this replay - not send info to client */
 710         if (info->mti_spec.no_create) {
 711                 md_size = 0;
 712                 acl_size = 0;
 713         }
 714
 715         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 716                md_size, acl_size);
 717 /*
 718             &RMF_MDT_BODY,
 719             &RMF_MDT_MD,
 720             &RMF_ACL, or &RMF_LOGCOOKIES
 721 (optional)  &RMF_CAPA1,
 722 (optional)  &RMF_CAPA2,
 723 (optional)  something else
 724 */
 725
 726         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 727          * buffers before growing it */
 728         if (info->mti_big_lmm_used) {
 729                 /* big_lmm buffer may be used even without packing the result
 730                  * into reply, just for internal server needs */
 731                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 732                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 733                                                          RCL_SERVER);
 734
 735                 /* free big lmm if md_size is not needed */
 736                 if (md_size == 0 || md_packed == 0) {
 737                         info->mti_big_lmm_used = 0;
 738                 } else {
 739                         /* buffer must be allocated separately */
 740                         LASSERT(info->mti_attr.ma_lmm !=
 741                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 742                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 743                 }
 744         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 745                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 746         }
 747
 748         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 749                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 750         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 751                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 752                                    acl_size, RCL_SERVER);
 753
 754         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 755             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 756                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 757
 758         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 759             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 760                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 761
 762         /*
 763          * Some more field should be shrinked if needed.
 764          * This should be done by those who added fields to reply message.
 765          */
 766
 767         /* Grow MD buffer if needed finally */
 768         if (info->mti_big_lmm_used) {
 769                 void *lmm;
 770
 771                 LASSERT(md_size > md_packed);
 772                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 773                        md_size - md_packed);
 774                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 775                 if (rc) {
 776                         /* we can't answer with proper LOV EA, drop flags,
 777                          * the rc is also returned so this request is
 778                          * considered as failed */
 779                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 780                         /* don't return transno along with error */
 781                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 782                 } else {
 783                         /* now we need to pack right LOV/LMV EA */
 784                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 785                         if (info->mti_attr.ma_valid & MA_LOV) {
 786                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 787                                                              RCL_SERVER) ==
 788                                                 info->mti_attr.ma_lmm_size);
 789                                 memcpy(lmm, info->mti_attr.ma_lmm,
 790                                        info->mti_attr.ma_lmm_size);
 791                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 792                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 793                                                              RCL_SERVER) ==
 794                                                 info->mti_attr.ma_lmv_size);
 795                                 memcpy(lmm, info->mti_attr.ma_lmv,
 796                                        info->mti_attr.ma_lmv_size);
 797                         }
 798                 }
 799                 /* update mdt_max_mdsize so clients will be aware about that */
 800                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 801                         info->mti_mdt->mdt_max_mdsize =
 802                                                     info->mti_attr.ma_lmm_size;
 803                 info->mti_big_lmm_used = 0;
 804         }
 805         RETURN(rc);
 806 }
 807
 808
 809 /* if object is dying, pack the lov/llog data,
 810  * parameter info->mti_attr should be valid at this point! */
 811 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 812                            struct md_attr *ma)
 813 {
 814         struct mdt_body      *repbody = NULL;
 815         const struct lu_attr *la = &ma->ma_attr;
 816         struct coordinator   *cdt = &info->mti_mdt->mdt_coordinator;
 817         int                  rc;
 818         __u64                need = 0;
 819         struct hsm_action_item hai = {
 820                 .hai_len = sizeof(hai),
 821                 .hai_action = HSMA_REMOVE,
 822                 .hai_extent.length = -1,
 823                 .hai_cookie = 0,
 824                 .hai_gid = 0,
 825         };
 826         __u64 compound_id;
 827         int archive_id;
 828
 829         ENTRY;
 830
 831         if (mdt_info_req(info) != NULL) {
 832                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 833                 LASSERT(repbody != NULL);
 834         } else {
 835                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 836         }
 837
 838         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 839                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 840
 841         if (ma->ma_valid & MA_LOV) {
 842                 CERROR("No need in LOV EA upon unlink\n");
 843                 dump_stack();
 844         }
 845         if (repbody != NULL)
 846                 repbody->mbo_eadatasize = 0;
 847
 848         /* Only check unlinked and archived if RAoLU and upon last close */
 849         if (!cdt->cdt_remove_archive_on_last_unlink ||
 850             atomic_read(&mo->mot_open_count) != 0)
 851                 RETURN(0);
 852
 853         if (ma->ma_valid & MA_INODE) {
 854                 if (ma->ma_attr.la_nlink != 0)
 855                         RETURN(0);
 856         } else {
 857                 need |= MA_INODE;
 858         }
 859
 860         if (ma->ma_valid & MA_HSM) {
 861                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 862                         RETURN(0);
 863         } else {
 864                 need |= MA_HSM;
 865         }
 866
 867         if (need != 0) {
 868                 ma->ma_need |= need;
 869                 rc = mdt_attr_get_complex(info, mo, ma);
 870                 if (rc) {
 871                         CERROR("%s: unable to fetch missing attributes of"
 872                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 873                                PFID(mdt_object_fid(mo)), rc);
 874                         RETURN(0);
 875                 }
 876
 877                 if (need & MA_INODE) {
 878                         if (ma->ma_valid & MA_INODE) {
 879                                 if (ma->ma_attr.la_nlink != 0)
 880                                         RETURN(0);
 881                         } else {
 882                                 RETURN(0);
 883                         }
 884                 }
 885
 886                 if (need & MA_HSM) {
 887                         if (ma->ma_valid & MA_HSM) {
 888                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 889                                         RETURN(0);
 890                         } else {
 891                                 RETURN(0);
 892                         }
 893                 }
 894         }
 895
 896         /* RAoLU policy is active, last close on file has occured,
 897          * file is unlinked, file is archived, so create remove request
 898          * for copytool!
 899          * If CDT is not running, requests will be logged for later. */
 900         compound_id = atomic_inc_return(&cdt->cdt_compound_id);
 901         if (ma->ma_hsm.mh_arch_id != 0)
 902                 archive_id = ma->ma_hsm.mh_arch_id;
 903         else
 904                 archive_id = cdt->cdt_default_archive_id;
 905
 906         hai.hai_fid = *mdt_object_fid(mo);
 907
 908         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt,
 909                                   compound_id, archive_id, 0, &hai);
 910         if (rc)
 911                 CERROR("%s: unable to add HSM remove request for "DFID
 912                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
 913                        PFID(mdt_object_fid(mo)), rc);
 914
 915         RETURN(0);
 916 }
 917
 918 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 919                                   struct md_attr *ma)
 920 {
 921         __u64 out;
 922
 923         out = 0;
 924         if (in & MDS_ATTR_MODE)
 925                 out |= LA_MODE;
 926         if (in & MDS_ATTR_UID)
 927                 out |= LA_UID;
 928         if (in & MDS_ATTR_GID)
 929                 out |= LA_GID;
 930         if (in & MDS_ATTR_SIZE)
 931                 out |= LA_SIZE;
 932         if (in & MDS_ATTR_BLOCKS)
 933                 out |= LA_BLOCKS;
 934         if (in & MDS_ATTR_ATIME_SET)
 935                 out |= LA_ATIME;
 936         if (in & MDS_ATTR_CTIME_SET)
 937                 out |= LA_CTIME;
 938         if (in & MDS_ATTR_MTIME_SET)
 939                 out |= LA_MTIME;
 940         if (in & MDS_ATTR_ATTR_FLAG)
 941                 out |= LA_FLAGS;
 942         if (in & MDS_ATTR_KILL_SUID)
 943                 out |= LA_KILL_SUID;
 944         if (in & MDS_ATTR_KILL_SGID)
 945                 out |= LA_KILL_SGID;
 946
 947         if (in & MDS_ATTR_FROM_OPEN)
 948                 rr->rr_flags |= MRF_OPEN_TRUNC;
 949         if (in & MDS_OPEN_OWNEROVERRIDE)
 950                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 951         if (in & MDS_ATTR_FORCE)
 952                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 953
 954         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 955                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 956                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 957                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 958                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 959                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 960         if (in != 0)
 961                 CERROR("Unknown attr bits: "LPX64"\n", in);
 962         return out;
 963 }
 964
 965 /* unpacking */
 966
 967 int mdt_name_unpack(struct req_capsule *pill,
 968                     const struct req_msg_field *field,
 969                     struct lu_name *ln,
 970                     enum mdt_name_flags flags)
 971 {
 972         ln->ln_name = req_capsule_client_get(pill, field);
 973         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 974
 975         if (!lu_name_is_valid(ln)) {
 976                 ln->ln_name = NULL;
 977                 ln->ln_namelen = 0;
 978
 979                 return -EPROTO;
 980         }
 981
 982         if ((flags & MNF_FIX_ANON) &&
 983             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 984                 /* Newer (3.x) kernels use a name of "/" for the
 985                  * "anonymous" disconnected dentries from NFS
 986                  * filehandle conversion. See d_obtain_alias(). */
 987                 ln->ln_name = NULL;
 988                 ln->ln_namelen = 0;
 989         }
 990
 991         return 0;
 992 }
 993
 994 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 995 {
 996         struct lu_ucred         *uc = mdt_ucred(info);
 997         struct md_attr          *ma = &info->mti_attr;
 998         struct lu_attr          *la = &ma->ma_attr;
 999         struct req_capsule      *pill = info->mti_pill;
1000         struct mdt_reint_record *rr = &info->mti_rr;
1001         struct mdt_rec_setattr  *rec;
1002         struct lu_nodemap       *nodemap =
1003                 info->mti_exp->exp_target_data.ted_nodemap;
1004         ENTRY;
1005
1006         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
1007         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1008         if (rec == NULL)
1009                 RETURN(-EFAULT);
1010
1011         /* This prior initialization is needed for old_init_ucred_reint() */
1012         uc->uc_fsuid = rec->sa_fsuid;
1013         uc->uc_fsgid = rec->sa_fsgid;
1014         uc->uc_cap   = rec->sa_cap;
1015         uc->uc_suppgids[0] = rec->sa_suppgid;
1016         uc->uc_suppgids[1] = -1;
1017
1018         rr->rr_fid1 = &rec->sa_fid;
1019         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1020         la->la_mode  = rec->sa_mode;
1021         la->la_flags = rec->sa_attr_flags;
1022         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1023                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1024         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1025                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1026         la->la_size  = rec->sa_size;
1027         la->la_blocks = rec->sa_blocks;
1028         la->la_ctime = rec->sa_ctime;
1029         la->la_atime = rec->sa_atime;
1030         la->la_mtime = rec->sa_mtime;
1031         ma->ma_valid = MA_INODE;
1032
1033         if (rec->sa_bias & MDS_DATA_MODIFIED)
1034                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
1035         else
1036                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
1037
1038         if (rec->sa_bias & MDS_HSM_RELEASE)
1039                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
1040         else
1041                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
1042
1043         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
1044                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
1045         else
1046                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
1047
1048         RETURN(0);
1049 }
1050
1051 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1052 {
1053         struct req_capsule *pill = info->mti_pill;
1054         struct mdt_ioepoch *ioepoch;
1055         ENTRY;
1056
1057         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1058                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1059         else
1060                 ioepoch = NULL;
1061
1062         if (ioepoch == NULL)
1063                 RETURN(-EPROTO);
1064
1065         info->mti_close_handle = ioepoch->mio_handle;
1066
1067         RETURN(0);
1068 }
1069
1070 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1071         struct req_capsule      *pill = info->mti_pill;
1072
1073         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1074                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1075                 if (info->mti_dlm_req == NULL)
1076                         RETURN(-EFAULT);
1077         }
1078
1079         RETURN(0);
1080 }
1081
1082 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1083 {
1084         struct mdt_reint_record *rr = &info->mti_rr;
1085         struct md_attr          *ma = &info->mti_attr;
1086         struct req_capsule      *pill = info->mti_pill;
1087         int rc;
1088         ENTRY;
1089
1090         rc = mdt_setattr_unpack_rec(info);
1091         if (rc)
1092                 RETURN(rc);
1093
1094         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1095                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1096                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1097                                                         RCL_CLIENT);
1098                 if (rr->rr_eadatalen > 0) {
1099                         const struct lmv_user_md        *lum;
1100
1101                         lum = rr->rr_eadata;
1102                         /* Sigh ma_valid(from req) does not indicate whether
1103                          * it will set LOV/LMV EA, so we have to check magic */
1104                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1105                                 ma->ma_valid |= MA_LMV;
1106                                 ma->ma_lmv = (void *)rr->rr_eadata;
1107                                 ma->ma_lmv_size = rr->rr_eadatalen;
1108                         } else {
1109                                 ma->ma_valid |= MA_LOV;
1110                                 ma->ma_lmm = (void *)rr->rr_eadata;
1111                                 ma->ma_lmm_size = rr->rr_eadatalen;
1112                         }
1113                 }
1114         }
1115
1116         rc = mdt_dlmreq_unpack(info);
1117         RETURN(rc);
1118 }
1119
1120 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
1121 {
1122         struct md_attr          *ma = &info->mti_attr;
1123         struct req_capsule      *pill = info->mti_pill;
1124         ENTRY;
1125
1126         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
1127                 RETURN(0);
1128
1129         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
1130
1131         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1132             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1133                 RETURN(-EFAULT);
1134
1135         RETURN(0);
1136 }
1137
1138 int mdt_close_unpack(struct mdt_thread_info *info)
1139 {
1140         int rc;
1141         ENTRY;
1142
1143         rc = mdt_close_handle_unpack(info);
1144         if (rc)
1145                 RETURN(rc);
1146
1147         rc = mdt_setattr_unpack_rec(info);
1148         if (rc)
1149                 RETURN(rc);
1150
1151         rc = mdt_intent_close_unpack(info);
1152         if (rc)
1153                 RETURN(rc);
1154
1155         RETURN(mdt_init_ucred_reint(info));
1156 }
1157
1158 static int mdt_create_unpack(struct mdt_thread_info *info)
1159 {
1160         struct lu_ucred         *uc  = mdt_ucred(info);
1161         struct mdt_rec_create   *rec;
1162         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1163         struct mdt_reint_record *rr = &info->mti_rr;
1164         struct req_capsule      *pill = info->mti_pill;
1165         struct md_op_spec       *sp = &info->mti_spec;
1166         int rc;
1167         ENTRY;
1168
1169         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1170         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1171         if (rec == NULL)
1172                 RETURN(-EFAULT);
1173
1174         /* This prior initialization is needed for old_init_ucred_reint() */
1175         uc->uc_fsuid = rec->cr_fsuid;
1176         uc->uc_fsgid = rec->cr_fsgid;
1177         uc->uc_cap   = rec->cr_cap;
1178         uc->uc_suppgids[0] = rec->cr_suppgid1;
1179         uc->uc_suppgids[1] = -1;
1180         uc->uc_umask = rec->cr_umask;
1181
1182         rr->rr_fid1 = &rec->cr_fid1;
1183         rr->rr_fid2 = &rec->cr_fid2;
1184         attr->la_mode = rec->cr_mode;
1185         attr->la_rdev  = rec->cr_rdev;
1186         attr->la_uid   = rec->cr_fsuid;
1187         attr->la_gid   = rec->cr_fsgid;
1188         attr->la_ctime = rec->cr_time;
1189         attr->la_mtime = rec->cr_time;
1190         attr->la_atime = rec->cr_time;
1191         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1192                          LA_CTIME | LA_MTIME | LA_ATIME;
1193         memset(&sp->u, 0, sizeof(sp->u));
1194         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1195
1196         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1197         if (rc < 0)
1198                 RETURN(rc);
1199
1200         if (S_ISLNK(attr->la_mode)) {
1201                 const char *tgt = NULL;
1202
1203                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1204                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1205                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1206                         sp->u.sp_symname = tgt;
1207                 }
1208                 if (tgt == NULL)
1209                         RETURN(-EFAULT);
1210         } else {
1211                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1212                 if (S_ISDIR(attr->la_mode) &&
1213                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1214                         sp->u.sp_ea.eadata =
1215                                 req_capsule_client_get(pill, &RMF_EADATA);
1216                         sp->u.sp_ea.eadatalen =
1217                                 req_capsule_get_size(pill, &RMF_EADATA,
1218                                                      RCL_CLIENT);
1219                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1220                 }
1221         }
1222
1223         rc = mdt_dlmreq_unpack(info);
1224         RETURN(rc);
1225 }
1226
1227 static int mdt_link_unpack(struct mdt_thread_info *info)
1228 {
1229         struct lu_ucred         *uc  = mdt_ucred(info);
1230         struct mdt_rec_link     *rec;
1231         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1232         struct mdt_reint_record *rr = &info->mti_rr;
1233         struct req_capsule      *pill = info->mti_pill;
1234         int rc;
1235         ENTRY;
1236
1237         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1238         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1239         if (rec == NULL)
1240                 RETURN(-EFAULT);
1241
1242         /* This prior initialization is needed for old_init_ucred_reint() */
1243         uc->uc_fsuid = rec->lk_fsuid;
1244         uc->uc_fsgid = rec->lk_fsgid;
1245         uc->uc_cap   = rec->lk_cap;
1246         uc->uc_suppgids[0] = rec->lk_suppgid1;
1247         uc->uc_suppgids[1] = rec->lk_suppgid2;
1248
1249         attr->la_uid = rec->lk_fsuid;
1250         attr->la_gid = rec->lk_fsgid;
1251         rr->rr_fid1 = &rec->lk_fid1;
1252         rr->rr_fid2 = &rec->lk_fid2;
1253         attr->la_ctime = rec->lk_time;
1254         attr->la_mtime = rec->lk_time;
1255         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1256
1257         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1258         if (rc < 0)
1259                 RETURN(rc);
1260
1261         rc = mdt_dlmreq_unpack(info);
1262
1263         RETURN(rc);
1264 }
1265
1266 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1267 {
1268         struct lu_ucred         *uc  = mdt_ucred(info);
1269         struct mdt_rec_unlink   *rec;
1270         struct md_attr          *ma = &info->mti_attr;
1271         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1272         struct mdt_reint_record *rr = &info->mti_rr;
1273         struct req_capsule      *pill = info->mti_pill;
1274         int rc;
1275         ENTRY;
1276
1277         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1278         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1279         if (rec == NULL)
1280                 RETURN(-EFAULT);
1281
1282         /* This prior initialization is needed for old_init_ucred_reint() */
1283         uc->uc_fsuid = rec->ul_fsuid;
1284         uc->uc_fsgid = rec->ul_fsgid;
1285         uc->uc_cap   = rec->ul_cap;
1286         uc->uc_suppgids[0] = rec->ul_suppgid1;
1287         uc->uc_suppgids[1] = -1;
1288
1289         attr->la_uid = rec->ul_fsuid;
1290         attr->la_gid = rec->ul_fsgid;
1291         rr->rr_fid1 = &rec->ul_fid1;
1292         rr->rr_fid2 = &rec->ul_fid2;
1293         attr->la_ctime = rec->ul_time;
1294         attr->la_mtime = rec->ul_time;
1295         attr->la_mode  = rec->ul_mode;
1296         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1297
1298         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1299         if (rc < 0)
1300                 RETURN(rc);
1301
1302         if (rec->ul_bias & MDS_VTX_BYPASS)
1303                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1304         else
1305                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1306
1307         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1308
1309         rc = mdt_dlmreq_unpack(info);
1310         RETURN(rc);
1311 }
1312
1313 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1314 {
1315         info->mti_spec.sp_rm_entry = 1;
1316         return mdt_unlink_unpack(info);
1317 }
1318
1319 static int mdt_rename_unpack(struct mdt_thread_info *info)
1320 {
1321         struct lu_ucred         *uc = mdt_ucred(info);
1322         struct mdt_rec_rename   *rec;
1323         struct md_attr          *ma = &info->mti_attr;
1324         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1325         struct mdt_reint_record *rr = &info->mti_rr;
1326         struct req_capsule      *pill = info->mti_pill;
1327         int rc;
1328         ENTRY;
1329
1330         CLASSERT(sizeof(struct mdt_rec_rename) == 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->rn_fsuid;
1337         uc->uc_fsgid = rec->rn_fsgid;
1338         uc->uc_cap   = rec->rn_cap;
1339         uc->uc_suppgids[0] = rec->rn_suppgid1;
1340         uc->uc_suppgids[1] = rec->rn_suppgid2;
1341
1342         attr->la_uid = rec->rn_fsuid;
1343         attr->la_gid = rec->rn_fsgid;
1344         rr->rr_fid1 = &rec->rn_fid1;
1345         rr->rr_fid2 = &rec->rn_fid2;
1346         attr->la_ctime = rec->rn_time;
1347         attr->la_mtime = rec->rn_time;
1348         /* rename_tgt contains the mode already */
1349         attr->la_mode = rec->rn_mode;
1350         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1351
1352         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1353         if (rc < 0)
1354                 RETURN(rc);
1355
1356         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1357         if (rc < 0)
1358                 RETURN(rc);
1359
1360         if (rec->rn_bias & MDS_VTX_BYPASS)
1361                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1362         else
1363                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1364
1365         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1366                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1367                 rc = mdt_close_handle_unpack(info);
1368                 if (rc < 0)
1369                         RETURN(rc);
1370                 info->mti_spec.sp_migrate_close = 1;
1371         }
1372
1373         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1374
1375
1376         rc = mdt_dlmreq_unpack(info);
1377
1378         RETURN(rc);
1379 }
1380
1381 /*
1382  * please see comment above LOV_MAGIC_V1_DEF
1383  */
1384 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1385 {
1386         struct mdt_reint_record *rr = &info->mti_rr;
1387         struct lov_user_md_v1   *v1;
1388
1389         v1 = (void *)rr->rr_eadata;
1390         LASSERT(v1);
1391
1392         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1393                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1394                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1395                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1396                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1397                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1398                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1399                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1400                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1401                 }
1402         }
1403 }
1404
1405 static int mdt_open_unpack(struct mdt_thread_info *info)
1406 {
1407         struct lu_ucred         *uc = mdt_ucred(info);
1408         struct mdt_rec_create   *rec;
1409         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1410         struct req_capsule      *pill = info->mti_pill;
1411         struct mdt_reint_record *rr   = &info->mti_rr;
1412         struct ptlrpc_request   *req  = mdt_info_req(info);
1413         struct md_op_spec       *sp   = &info->mti_spec;
1414         ENTRY;
1415
1416         CLASSERT(sizeof(struct mdt_rec_create) == 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->cr_fsuid;
1423         uc->uc_fsgid = rec->cr_fsgid;
1424         uc->uc_cap   = rec->cr_cap;
1425         uc->uc_suppgids[0] = rec->cr_suppgid1;
1426         uc->uc_suppgids[1] = rec->cr_suppgid2;
1427         uc->uc_umask = rec->cr_umask;
1428
1429         rr->rr_fid1   = &rec->cr_fid1;
1430         rr->rr_fid2   = &rec->cr_fid2;
1431         rr->rr_handle = &rec->cr_old_handle;
1432         attr->la_mode = rec->cr_mode;
1433         attr->la_rdev  = rec->cr_rdev;
1434         attr->la_uid   = rec->cr_fsuid;
1435         attr->la_gid   = rec->cr_fsgid;
1436         attr->la_ctime = rec->cr_time;
1437         attr->la_mtime = rec->cr_time;
1438         attr->la_atime = rec->cr_time;
1439         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1440                          LA_CTIME | LA_MTIME | LA_ATIME;
1441         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1442         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1443         /* Do not trigger ASSERTION if client miss to set such flags. */
1444         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1445                 RETURN(-EPROTO);
1446
1447         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1448
1449         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1450
1451         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1452                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1453                                                         RCL_CLIENT);
1454                 if (rr->rr_eadatalen > 0) {
1455                         rr->rr_eadata = req_capsule_client_get(pill,
1456                                                                &RMF_EADATA);
1457                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1458                         sp->u.sp_ea.eadata = rr->rr_eadata;
1459                         sp->no_create = !!req_is_replay(req);
1460                         mdt_fix_lov_magic(info);
1461                 }
1462
1463                 /*
1464                  * Client default md_size may be 0 right after client start,
1465                  * until all osc are connected, set here just some reasonable
1466                  * value to prevent misbehavior.
1467                  */
1468                 if (rr->rr_eadatalen == 0 &&
1469                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1470                         rr->rr_eadatalen = MIN_MD_SIZE;
1471         }
1472
1473         RETURN(0);
1474 }
1475
1476 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1477 {
1478         struct mdt_reint_record *rr     = &info->mti_rr;
1479         struct lu_ucred         *uc     = mdt_ucred(info);
1480         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1481         struct req_capsule      *pill   = info->mti_pill;
1482         struct mdt_rec_setxattr *rec;
1483         int                      rc;
1484         ENTRY;
1485
1486
1487         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1488                          sizeof(struct mdt_rec_reint));
1489
1490         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1491         if (rec == NULL)
1492                 RETURN(-EFAULT);
1493
1494         /* This prior initialization is needed for old_init_ucred_reint() */
1495         uc->uc_fsuid  = rec->sx_fsuid;
1496         uc->uc_fsgid  = rec->sx_fsgid;
1497         uc->uc_cap    = rec->sx_cap;
1498         uc->uc_suppgids[0] = rec->sx_suppgid1;
1499         uc->uc_suppgids[1] = -1;
1500
1501         rr->rr_opcode = rec->sx_opcode;
1502         rr->rr_fid1   = &rec->sx_fid;
1503         attr->la_valid = rec->sx_valid;
1504         attr->la_ctime = rec->sx_time;
1505         attr->la_size = rec->sx_size;
1506         attr->la_flags = rec->sx_flags;
1507
1508         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1509         if (rc < 0)
1510                 RETURN(rc);
1511
1512         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1513                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1514                                                         RCL_CLIENT);
1515                 if (rr->rr_eadatalen > 0) {
1516                         rr->rr_eadata = req_capsule_client_get(pill,
1517                                                                &RMF_EADATA);
1518                         if (rr->rr_eadata == NULL)
1519                                 RETURN(-EFAULT);
1520                 } else {
1521                         rr->rr_eadata = NULL;
1522                 }
1523         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1524                 CDEBUG(D_INFO, "no xattr data supplied\n");
1525                 RETURN(-EFAULT);
1526         }
1527
1528         if (mdt_dlmreq_unpack(info) < 0)
1529                 RETURN(-EPROTO);
1530
1531         RETURN(0);
1532 }
1533
1534
1535 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1536
1537 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1538         [REINT_SETATTR]  = mdt_setattr_unpack,
1539         [REINT_CREATE]   = mdt_create_unpack,
1540         [REINT_LINK]     = mdt_link_unpack,
1541         [REINT_UNLINK]   = mdt_unlink_unpack,
1542         [REINT_RENAME]   = mdt_rename_unpack,
1543         [REINT_OPEN]     = mdt_open_unpack,
1544         [REINT_SETXATTR] = mdt_setxattr_unpack,
1545         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1546         [REINT_MIGRATE]  = mdt_rename_unpack,
1547 };
1548
1549 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1550 {
1551         int rc;
1552         ENTRY;
1553
1554         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1555         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1556                 info->mti_rr.rr_opcode = op;
1557                 rc = mdt_reint_unpackers[op](info);
1558         } else {
1559                 CERROR("Unexpected opcode %d\n", op);
1560                 rc = -EFAULT;
1561         }
1562         RETURN(rc);
1563 }