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, 2013, 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 "mdt_internal.h"
  52 #include <lnet/nidstr.h>
  53
  54 typedef enum ucred_init_type {
  55         NONE_INIT       = 0,
  56         BODY_INIT       = 1,
  57         REC_INIT        = 2
  58 } ucred_init_type_t;
  59
  60 void mdt_exit_ucred(struct mdt_thread_info *info)
  61 {
  62         struct lu_ucred   *uc  = mdt_ucred(info);
  63         struct mdt_device *mdt = info->mti_mdt;
  64
  65         LASSERT(uc != NULL);
  66         if (uc->uc_valid != UCRED_INIT) {
  67                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  68                 if (uc->uc_ginfo) {
  69                         put_group_info(uc->uc_ginfo);
  70                         uc->uc_ginfo = NULL;
  71                 }
  72                 if (uc->uc_identity) {
  73                         mdt_identity_put(mdt->mdt_identity_cache,
  74                                          uc->uc_identity);
  75                         uc->uc_identity = NULL;
  76                 }
  77                 uc->uc_valid = UCRED_INIT;
  78         }
  79 }
  80
  81 static int match_nosquash_list(struct rw_semaphore *sem,
  82                                struct list_head *nidlist,
  83                                lnet_nid_t peernid)
  84 {
  85         int rc;
  86         ENTRY;
  87         down_read(sem);
  88         rc = cfs_match_nid(peernid, nidlist);
  89         up_read(sem);
  90         RETURN(rc);
  91 }
  92
  93 /* root_squash for inter-MDS operations */
  94 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  95 {
  96         struct lu_ucred *ucred = mdt_ucred(info);
  97         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
  98         ENTRY;
  99
 100         LASSERT(ucred != NULL);
 101         if (!squash->rsi_uid || ucred->uc_fsuid)
 102                 RETURN(0);
 103
 104         if (match_nosquash_list(&squash->rsi_sem,
 105                                 &squash->rsi_nosquash_nids,
 106                                 peernid)) {
 107                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 108                        libcfs_nid2str(peernid));
 109                 RETURN(0);
 110         }
 111
 112         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 113                libcfs_nid2str(peernid),
 114                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 115                squash->rsi_uid, squash->rsi_gid, 0);
 116
 117         ucred->uc_fsuid = squash->rsi_uid;
 118         ucred->uc_fsgid = squash->rsi_gid;
 119         ucred->uc_cap = 0;
 120         ucred->uc_suppgids[0] = -1;
 121         ucred->uc_suppgids[1] = -1;
 122
 123         RETURN(0);
 124 }
 125
 126 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 127                           void *buf)
 128 {
 129         struct ptlrpc_request   *req = mdt_info_req(info);
 130         struct mdt_device       *mdt = info->mti_mdt;
 131         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 132         struct lu_ucred         *ucred = mdt_ucred(info);
 133         lnet_nid_t               peernid = req->rq_peer.nid;
 134         __u32                    perm = 0;
 135         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 136         int                      setuid;
 137         int                      setgid;
 138         int                      rc = 0;
 139
 140         ENTRY;
 141
 142         LASSERT(req->rq_auth_gss);
 143         LASSERT(!req->rq_auth_usr_mdt);
 144         LASSERT(req->rq_user_desc);
 145         LASSERT(ucred != NULL);
 146
 147         ucred->uc_valid = UCRED_INVALID;
 148
 149         ucred->uc_o_uid   = pud->pud_uid;
 150         ucred->uc_o_gid   = pud->pud_gid;
 151         ucred->uc_o_fsuid = pud->pud_fsuid;
 152         ucred->uc_o_fsgid = pud->pud_fsgid;
 153
 154         if (type == BODY_INIT) {
 155                 struct mdt_body *body = (struct mdt_body *)buf;
 156
 157                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 158                 ucred->uc_suppgids[1] = -1;
 159         }
 160
 161         /* sanity check: we expect the uid which client claimed is true */
 162         if (remote) {
 163                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 164                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 165                         RETURN(-EACCES);
 166                 }
 167
 168                 if (ptlrpc_user_desc_do_idmap(req, pud))
 169                         RETURN(-EACCES);
 170
 171                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 172                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 173                                "while client claims %u:%u/%u:%u\n",
 174                                libcfs_nid2str(peernid), req->rq_auth_uid,
 175                                req->rq_auth_mapped_uid,
 176                                pud->pud_uid, pud->pud_gid,
 177                                pud->pud_fsuid, pud->pud_fsgid);
 178                         RETURN(-EACCES);
 179                 }
 180         } else {
 181                 if (req->rq_auth_uid != pud->pud_uid) {
 182                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 183                                "while client claims %u:%u/%u:%u\n",
 184                                libcfs_nid2str(peernid), req->rq_auth_uid,
 185                                pud->pud_uid, pud->pud_gid,
 186                                pud->pud_fsuid, pud->pud_fsgid);
 187                         RETURN(-EACCES);
 188                 }
 189         }
 190
 191         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 192                 if (remote) {
 193                         CDEBUG(D_SEC, "remote client must run with identity_get "
 194                                "enabled!\n");
 195                         RETURN(-EACCES);
 196                 } else {
 197                         ucred->uc_identity = NULL;
 198                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 199                                CFS_SETGRP_PERM;
 200                 }
 201         } else {
 202                 struct md_identity *identity;
 203
 204                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 205                                             pud->pud_uid);
 206                 if (IS_ERR(identity)) {
 207                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 208                                      !remote)) {
 209                                 ucred->uc_identity = NULL;
 210                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 211                                        CFS_SETGRP_PERM;
 212                         } else {
 213                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 214                                        pud->pud_uid);
 215                                 RETURN(-EACCES);
 216                         }
 217                 } else {
 218                         ucred->uc_identity = identity;
 219                         perm = mdt_identity_get_perm(ucred->uc_identity,
 220                                                      remote, peernid);
 221                 }
 222         }
 223
 224         /* find out the setuid/setgid attempt */
 225         setuid = (pud->pud_uid != pud->pud_fsuid);
 226         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 227                   (ucred->uc_identity &&
 228                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 229
 230         /* check permission of setuid */
 231         if (setuid && !(perm & CFS_SETUID_PERM)) {
 232                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 233                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 234                 GOTO(out, rc = -EACCES);
 235         }
 236
 237         /* check permission of setgid */
 238         if (setgid && !(perm & CFS_SETGID_PERM)) {
 239                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 240                        "from %s\n", pud->pud_uid, pud->pud_gid,
 241                        pud->pud_fsuid, pud->pud_fsgid,
 242                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 243                 GOTO(out, rc = -EACCES);
 244         }
 245
 246         /*
 247          * NB: remote client not allowed to setgroups anyway.
 248          */
 249         if (!remote && perm & CFS_SETGRP_PERM) {
 250                 if (pud->pud_ngroups) {
 251                         /* setgroups for local client */
 252                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 253                         if (!ucred->uc_ginfo) {
 254                                 CERROR("failed to alloc %d groups\n",
 255                                        pud->pud_ngroups);
 256                                 GOTO(out, rc = -ENOMEM);
 257                         }
 258
 259                         lustre_groups_from_list(ucred->uc_ginfo,
 260                                                 pud->pud_groups);
 261                         lustre_groups_sort(ucred->uc_ginfo);
 262                 } else {
 263                         ucred->uc_ginfo = NULL;
 264                 }
 265         } else {
 266                 ucred->uc_suppgids[0] = -1;
 267                 ucred->uc_suppgids[1] = -1;
 268                 ucred->uc_ginfo = NULL;
 269         }
 270
 271         ucred->uc_uid   = pud->pud_uid;
 272         ucred->uc_gid   = pud->pud_gid;
 273         ucred->uc_fsuid = pud->pud_fsuid;
 274         ucred->uc_fsgid = pud->pud_fsgid;
 275
 276         /* process root_squash here. */
 277         mdt_root_squash(info, peernid);
 278
 279         /* remove fs privilege for non-root user. */
 280         if (ucred->uc_fsuid)
 281                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 282         else
 283                 ucred->uc_cap = pud->pud_cap;
 284         if (remote && !(perm & CFS_RMTOWN_PERM))
 285                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 286                                    CFS_CAP_CHOWN_MASK);
 287         ucred->uc_valid = UCRED_NEW;
 288
 289         EXIT;
 290
 291 out:
 292         if (rc) {
 293                 if (ucred->uc_ginfo) {
 294                         put_group_info(ucred->uc_ginfo);
 295                         ucred->uc_ginfo = NULL;
 296                 }
 297                 if (ucred->uc_identity) {
 298                         mdt_identity_put(mdt->mdt_identity_cache,
 299                                          ucred->uc_identity);
 300                         ucred->uc_identity = NULL;
 301                 }
 302         }
 303
 304         return rc;
 305 }
 306
 307 int mdt_check_ucred(struct mdt_thread_info *info)
 308 {
 309         struct ptlrpc_request   *req = mdt_info_req(info);
 310         struct mdt_device       *mdt = info->mti_mdt;
 311         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 312         struct lu_ucred         *ucred = mdt_ucred(info);
 313         struct md_identity      *identity = NULL;
 314         lnet_nid_t               peernid = req->rq_peer.nid;
 315         __u32                    perm = 0;
 316         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 317         int                      setuid;
 318         int                      setgid;
 319         int                      rc = 0;
 320
 321         ENTRY;
 322
 323         LASSERT(ucred != NULL);
 324         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 325                 RETURN(0);
 326
 327         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 328                 RETURN(0);
 329
 330         /* sanity check: if we use strong authentication, we expect the
 331          * uid which client claimed is true */
 332         if (remote) {
 333                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 334                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 335                         RETURN(-EACCES);
 336                 }
 337
 338                 if (ptlrpc_user_desc_do_idmap(req, pud))
 339                         RETURN(-EACCES);
 340
 341                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 342                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 343                                "while client claims %u:%u/%u:%u\n",
 344                                libcfs_nid2str(peernid), req->rq_auth_uid,
 345                                req->rq_auth_mapped_uid,
 346                                pud->pud_uid, pud->pud_gid,
 347                                pud->pud_fsuid, pud->pud_fsgid);
 348                         RETURN(-EACCES);
 349                 }
 350         } else {
 351                 if (req->rq_auth_uid != pud->pud_uid) {
 352                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 353                                "while client claims %u:%u/%u:%u\n",
 354                                libcfs_nid2str(peernid), req->rq_auth_uid,
 355                                pud->pud_uid, pud->pud_gid,
 356                                pud->pud_fsuid, pud->pud_fsgid);
 357                         RETURN(-EACCES);
 358                 }
 359         }
 360
 361         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 362                 if (remote) {
 363                         CDEBUG(D_SEC, "remote client must run with identity_get "
 364                                "enabled!\n");
 365                         RETURN(-EACCES);
 366                 }
 367                 RETURN(0);
 368         }
 369
 370         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 371         if (IS_ERR(identity)) {
 372                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 373                              !remote)) {
 374                         RETURN(0);
 375                 } else {
 376                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 377                                pud->pud_uid);
 378                         RETURN(-EACCES);
 379                }
 380         }
 381
 382         perm = mdt_identity_get_perm(identity, remote, peernid);
 383         /* find out the setuid/setgid attempt */
 384         setuid = (pud->pud_uid != pud->pud_fsuid);
 385         setgid = (pud->pud_gid != pud->pud_fsgid ||
 386                   pud->pud_gid != identity->mi_gid);
 387
 388         /* check permission of setuid */
 389         if (setuid && !(perm & CFS_SETUID_PERM)) {
 390                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 391                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 392                 GOTO(out, rc = -EACCES);
 393         }
 394
 395         /* check permission of setgid */
 396         if (setgid && !(perm & CFS_SETGID_PERM)) {
 397                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 398                        "from %s\n", pud->pud_uid, pud->pud_gid,
 399                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 400                        libcfs_nid2str(peernid));
 401                 GOTO(out, rc = -EACCES);
 402         }
 403
 404         EXIT;
 405
 406 out:
 407         mdt_identity_put(mdt->mdt_identity_cache, identity);
 408         return rc;
 409 }
 410
 411 static int old_init_ucred(struct mdt_thread_info *info,
 412                           struct mdt_body *body)
 413 {
 414         struct lu_ucred *uc = mdt_ucred(info);
 415         struct mdt_device  *mdt = info->mti_mdt;
 416         struct md_identity *identity = NULL;
 417
 418         ENTRY;
 419
 420         LASSERT(uc != NULL);
 421         uc->uc_valid = UCRED_INVALID;
 422         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 423         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 424         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 425         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 426         uc->uc_suppgids[0] = body->mbo_suppgid;
 427         uc->uc_suppgids[1] = -1;
 428         uc->uc_ginfo = NULL;
 429         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 430                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 431                                             uc->uc_fsuid);
 432                 if (IS_ERR(identity)) {
 433                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 434                                 identity = NULL;
 435                         } else {
 436                                 CDEBUG(D_SEC, "Deny access without identity: "
 437                                        "uid %u\n", uc->uc_fsuid);
 438                                 RETURN(-EACCES);
 439                         }
 440                 }
 441         }
 442         uc->uc_identity = identity;
 443
 444         /* process root_squash here. */
 445         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 446
 447         /* remove fs privilege for non-root user. */
 448         if (uc->uc_fsuid)
 449                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 450         else
 451                 uc->uc_cap = body->mbo_capability;
 452         uc->uc_valid = UCRED_OLD;
 453
 454         RETURN(0);
 455 }
 456
 457 static int old_init_ucred_reint(struct mdt_thread_info *info)
 458 {
 459         struct lu_ucred *uc = mdt_ucred(info);
 460         struct mdt_device  *mdt = info->mti_mdt;
 461         struct md_identity *identity = NULL;
 462
 463         ENTRY;
 464
 465         LASSERT(uc != NULL);
 466         uc->uc_valid = UCRED_INVALID;
 467         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 468         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 469         uc->uc_ginfo = NULL;
 470         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 471                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 472                                             uc->uc_fsuid);
 473                 if (IS_ERR(identity)) {
 474                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 475                                 identity = NULL;
 476                         } else {
 477                                 CDEBUG(D_SEC, "Deny access without identity: "
 478                                        "uid %u\n", uc->uc_fsuid);
 479                                 RETURN(-EACCES);
 480                         }
 481                 }
 482         }
 483         uc->uc_identity = identity;
 484
 485         /* process root_squash here. */
 486         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 487
 488         /* remove fs privilege for non-root user. */
 489         if (uc->uc_fsuid)
 490                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 491         uc->uc_valid = UCRED_OLD;
 492
 493         RETURN(0);
 494 }
 495
 496 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 497 {
 498         struct ptlrpc_request *req = mdt_info_req(info);
 499         struct lu_ucred       *uc  = mdt_ucred(info);
 500
 501         LASSERT(uc != NULL);
 502         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 503                 return 0;
 504
 505         mdt_exit_ucred(info);
 506
 507         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 508                 return old_init_ucred(info, body);
 509         else
 510                 return new_init_ucred(info, BODY_INIT, body);
 511 }
 512
 513 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 514 {
 515         struct ptlrpc_request *req = mdt_info_req(info);
 516         struct lu_ucred       *uc  = mdt_ucred(info);
 517
 518         LASSERT(uc != NULL);
 519         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 520                 return 0;
 521
 522         mdt_exit_ucred(info);
 523
 524         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 525                 return old_init_ucred_reint(info);
 526         else
 527                 return new_init_ucred(info, REC_INIT, NULL);
 528 }
 529
 530 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 531 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 532 {
 533         const struct lov_ost_data_v1    *lod;
 534         int                              i;
 535         __u16                            count;
 536
 537         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 538                 return;
 539
 540         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 541
 542         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 543                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 544                le32_to_cpu(lmm->lmm_pattern));
 545         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 546                le32_to_cpu(lmm->lmm_stripe_size), count);
 547
 548         /* If it's a directory or a released file, then there are
 549          * no actual objects to print, so bail out. */
 550         if (valid & OBD_MD_FLDIREA ||
 551             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 552                 return;
 553
 554         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 555         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 556                 struct ost_id oi;
 557
 558                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 559                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 560                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 561         }
 562 }
 563
 564 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 565 {
 566         const struct lmv_mds_md_v1 *lmm1;
 567         int                        i;
 568
 569         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 570                 return;
 571
 572         lmm1 = &lmv->lmv_md_v1;
 573         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 574                le32_to_cpu(lmm1->lmv_magic),
 575                le32_to_cpu(lmm1->lmv_master_mdt_index),
 576                le32_to_cpu(lmm1->lmv_stripe_count));
 577
 578         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 579                 return;
 580
 581         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 582                 struct lu_fid fid;
 583
 584                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 585                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 586         }
 587 }
 588
 589 /* Shrink and/or grow reply buffers */
 590 int mdt_fix_reply(struct mdt_thread_info *info)
 591 {
 592         struct req_capsule *pill = info->mti_pill;
 593         struct mdt_body    *body;
 594         int                md_size, md_packed = 0;
 595         int                acl_size;
 596         int                rc = 0;
 597         ENTRY;
 598
 599         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 600         LASSERT(body != NULL);
 601
 602         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 603                                OBD_MD_LINKNAME))
 604                 md_size = body->mbo_eadatasize;
 605         else
 606                 md_size = 0;
 607
 608         acl_size = body->mbo_aclsize;
 609
 610         /* this replay - not send info to client */
 611         if (info->mti_spec.no_create) {
 612                 md_size = 0;
 613                 acl_size = 0;
 614         }
 615
 616         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 617                " MDSCAPA = %llx, OSSCAPA = %llx\n",
 618                md_size, acl_size,
 619                (unsigned long long)(body->mbo_valid & OBD_MD_FLMDSCAPA),
 620                (unsigned long long)(body->mbo_valid & OBD_MD_FLOSSCAPA));
 621 /*
 622             &RMF_MDT_BODY,
 623             &RMF_MDT_MD,
 624             &RMF_ACL, or &RMF_LOGCOOKIES
 625 (optional)  &RMF_CAPA1,
 626 (optional)  &RMF_CAPA2,
 627 (optional)  something else
 628 */
 629
 630         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 631          * buffers before growing it */
 632         if (info->mti_big_lmm_used) {
 633                 /* big_lmm buffer may be used even without packing the result
 634                  * into reply, just for internal server needs */
 635                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 636                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 637                                                          RCL_SERVER);
 638
 639                 /* free big lmm if md_size is not needed */
 640                 if (md_size == 0 || md_packed == 0) {
 641                         info->mti_big_lmm_used = 0;
 642                 } else {
 643                         /* buffer must be allocated separately */
 644                         LASSERT(info->mti_attr.ma_lmm !=
 645                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 646                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 647                 }
 648         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 649                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 650         }
 651
 652         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 653                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 654         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 655                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 656                                    acl_size, RCL_SERVER);
 657
 658         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 659             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 660                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 661
 662         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 663             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 664                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 665
 666         /*
 667          * Some more field should be shrinked if needed.
 668          * This should be done by those who added fields to reply message.
 669          */
 670
 671         /* Grow MD buffer if needed finally */
 672         if (info->mti_big_lmm_used) {
 673                 void *lmm;
 674
 675                 LASSERT(md_size > md_packed);
 676                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 677                        md_size - md_packed);
 678                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 679                 if (rc) {
 680                         /* we can't answer with proper LOV EA, drop flags,
 681                          * the rc is also returned so this request is
 682                          * considered as failed */
 683                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 684                         /* don't return transno along with error */
 685                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 686                 } else {
 687                         /* now we need to pack right LOV/LMV EA */
 688                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 689                         if (info->mti_attr.ma_valid & MA_LOV) {
 690                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 691                                                              RCL_SERVER) ==
 692                                                 info->mti_attr.ma_lmm_size);
 693                                 memcpy(lmm, info->mti_attr.ma_lmm,
 694                                        info->mti_attr.ma_lmm_size);
 695                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 696                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 697                                                              RCL_SERVER) ==
 698                                                 info->mti_attr.ma_lmv_size);
 699                                 memcpy(lmm, info->mti_attr.ma_lmv,
 700                                        info->mti_attr.ma_lmv_size);
 701                         }
 702                 }
 703                 /* update mdt_max_mdsize so clients will be aware about that */
 704                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 705                         info->mti_mdt->mdt_max_mdsize =
 706                                                     info->mti_attr.ma_lmm_size;
 707                 info->mti_big_lmm_used = 0;
 708         }
 709         RETURN(rc);
 710 }
 711
 712
 713 /* if object is dying, pack the lov/llog data,
 714  * parameter info->mti_attr should be valid at this point! */
 715 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 716                            const struct md_attr *ma)
 717 {
 718         struct mdt_body       *repbody;
 719         const struct lu_attr *la = &ma->ma_attr;
 720         int rc;
 721         ENTRY;
 722
 723         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 724         LASSERT(repbody != NULL);
 725
 726         if (ma->ma_valid & MA_INODE)
 727                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 728
 729         if (ma->ma_valid & MA_LOV) {
 730                 CERROR("No need in LOV EA upon unlink\n");
 731                 dump_stack();
 732         }
 733         repbody->mbo_eadatasize = 0;
 734
 735         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 736             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 737             repbody->mbo_valid & OBD_MD_FLEASIZE) {
 738                 struct lustre_capa *capa;
 739
 740                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 741                 LASSERT(capa);
 742                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 743                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 744                 if (rc)
 745                         RETURN(rc);
 746
 747                 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
 748         }
 749
 750         RETURN(0);
 751 }
 752
 753 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 754                                   struct md_attr *ma)
 755 {
 756         __u64 out;
 757
 758         out = 0;
 759         if (in & MDS_ATTR_MODE)
 760                 out |= LA_MODE;
 761         if (in & MDS_ATTR_UID)
 762                 out |= LA_UID;
 763         if (in & MDS_ATTR_GID)
 764                 out |= LA_GID;
 765         if (in & MDS_ATTR_SIZE)
 766                 out |= LA_SIZE;
 767         if (in & MDS_ATTR_BLOCKS)
 768                 out |= LA_BLOCKS;
 769         if (in & MDS_ATTR_ATIME_SET)
 770                 out |= LA_ATIME;
 771         if (in & MDS_ATTR_CTIME_SET)
 772                 out |= LA_CTIME;
 773         if (in & MDS_ATTR_MTIME_SET)
 774                 out |= LA_MTIME;
 775         if (in & MDS_ATTR_ATTR_FLAG)
 776                 out |= LA_FLAGS;
 777         if (in & MDS_ATTR_KILL_SUID)
 778                 out |= LA_KILL_SUID;
 779         if (in & MDS_ATTR_KILL_SGID)
 780                 out |= LA_KILL_SGID;
 781
 782         if (in & MDS_ATTR_FROM_OPEN)
 783                 rr->rr_flags |= MRF_OPEN_TRUNC;
 784         if (in & MDS_OPEN_OWNEROVERRIDE)
 785                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 786         if (in & MDS_ATTR_FORCE)
 787                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 788
 789         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 790                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 791                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 792                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 793                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 794                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 795         if (in != 0)
 796                 CERROR("Unknown attr bits: "LPX64"\n", in);
 797         return out;
 798 }
 799
 800 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 801                       const struct lu_fid *fid, struct lustre_capa *capa)
 802 {
 803         struct lu_capainfo *lci;
 804
 805         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 806         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 807             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 808                 return;
 809
 810         lci = lu_capainfo_get(info->mti_env);
 811         LASSERT(lci);
 812         lci->lci_fid[offset]  = *fid;
 813         lci->lci_capa[offset] = capa;
 814 }
 815
 816 #ifdef DEBUG_CAPA
 817 void mdt_dump_capainfo(struct mdt_thread_info *info)
 818 {
 819         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 820         int i;
 821
 822         if (lci == NULL)
 823                 return;
 824
 825         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 826                 if (lci->lci_capa[i] == NULL) {
 827                         CERROR("no capa for index %d "DFID"\n",
 828                                i, PFID(&lci->lci_fid[i]));
 829                         continue;
 830                 }
 831                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 832                         CERROR("bypass for index %d "DFID"\n",
 833                                i, PFID(&lci->lci_fid[i]));
 834                         continue;
 835                 }
 836                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 837         }
 838 }
 839 #endif /* DEBUG_CAPA */
 840
 841 /* unpacking */
 842
 843 int mdt_name_unpack(struct req_capsule *pill,
 844                     const struct req_msg_field *field,
 845                     struct lu_name *ln,
 846                     enum mdt_name_flags flags)
 847 {
 848         ln->ln_name = req_capsule_client_get(pill, field);
 849         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 850
 851         if (!lu_name_is_valid(ln)) {
 852                 ln->ln_name = NULL;
 853                 ln->ln_namelen = 0;
 854
 855                 return -EPROTO;
 856         }
 857
 858         if ((flags & MNF_FIX_ANON) &&
 859             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 860                 /* Newer (3.x) kernels use a name of "/" for the
 861                  * "anonymous" disconnected dentries from NFS
 862                  * filehandle conversion. See d_obtain_alias(). */
 863                 ln->ln_name = NULL;
 864                 ln->ln_namelen = 0;
 865         }
 866
 867         return 0;
 868 }
 869
 870 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 871 {
 872         struct lu_ucred         *uc  = mdt_ucred(info);
 873         struct md_attr          *ma = &info->mti_attr;
 874         struct lu_attr          *la = &ma->ma_attr;
 875         struct req_capsule      *pill = info->mti_pill;
 876         struct mdt_reint_record *rr = &info->mti_rr;
 877         struct mdt_rec_setattr  *rec;
 878         ENTRY;
 879
 880         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 881         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 882         if (rec == NULL)
 883                 RETURN(-EFAULT);
 884
 885         /* This prior initialization is needed for old_init_ucred_reint() */
 886         uc->uc_fsuid = rec->sa_fsuid;
 887         uc->uc_fsgid = rec->sa_fsgid;
 888         uc->uc_cap   = rec->sa_cap;
 889         uc->uc_suppgids[0] = rec->sa_suppgid;
 890         uc->uc_suppgids[1] = -1;
 891
 892         rr->rr_fid1 = &rec->sa_fid;
 893         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 894         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 895          *  the client does not have OBD_CONNECT_FULL20, convert it
 896          *  to LA_xTIME. LU-3036 */
 897         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 898                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 899                      (rec->sa_valid & MDS_ATTR_ATIME))
 900                         la->la_valid |= LA_ATIME;
 901                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 902                      (rec->sa_valid & MDS_ATTR_MTIME))
 903                         la->la_valid |= LA_MTIME;
 904                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 905                      (rec->sa_valid & MDS_ATTR_CTIME))
 906                         la->la_valid |= LA_CTIME;
 907         }
 908         la->la_mode  = rec->sa_mode;
 909         la->la_flags = rec->sa_attr_flags;
 910         la->la_uid   = rec->sa_uid;
 911         la->la_gid   = rec->sa_gid;
 912         la->la_size  = rec->sa_size;
 913         la->la_blocks = rec->sa_blocks;
 914         la->la_ctime = rec->sa_ctime;
 915         la->la_atime = rec->sa_atime;
 916         la->la_mtime = rec->sa_mtime;
 917         ma->ma_valid = MA_INODE;
 918
 919         if (rec->sa_bias & MDS_DATA_MODIFIED)
 920                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 921         else
 922                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 923
 924         if (rec->sa_bias & MDS_HSM_RELEASE)
 925                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 926         else
 927                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 928
 929         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 930                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 931                                  req_capsule_client_get(pill, &RMF_CAPA1));
 932
 933         RETURN(0);
 934 }
 935
 936 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 937 {
 938         struct req_capsule *pill = info->mti_pill;
 939         ENTRY;
 940
 941         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 942                 info->mti_ioepoch =
 943                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 944         else
 945                 info->mti_ioepoch = NULL;
 946         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 947 }
 948
 949 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 950         struct req_capsule      *pill = info->mti_pill;
 951
 952         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 953                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 954                 if (info->mti_dlm_req == NULL)
 955                         RETURN(-EFAULT);
 956         }
 957
 958         RETURN(0);
 959 }
 960
 961 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 962 {
 963         struct mdt_reint_record *rr = &info->mti_rr;
 964         struct md_attr          *ma = &info->mti_attr;
 965         struct req_capsule      *pill = info->mti_pill;
 966         int rc;
 967         ENTRY;
 968
 969         rc = mdt_setattr_unpack_rec(info);
 970         if (rc)
 971                 RETURN(rc);
 972
 973         /* Epoch may be absent */
 974         mdt_ioepoch_unpack(info);
 975
 976         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 977                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 978                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 979                                                         RCL_CLIENT);
 980                 if (rr->rr_eadatalen > 0) {
 981                         const struct lmv_user_md        *lum;
 982
 983                         lum = rr->rr_eadata;
 984                         /* Sigh ma_valid(from req) does not indicate whether
 985                          * it will set LOV/LMV EA, so we have to check magic */
 986                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 987                                 ma->ma_valid |= MA_LMV;
 988                                 ma->ma_lmv = (void *)rr->rr_eadata;
 989                                 ma->ma_lmv_size = rr->rr_eadatalen;
 990                         } else {
 991                                 ma->ma_valid |= MA_LOV;
 992                                 ma->ma_lmm = (void *)rr->rr_eadata;
 993                                 ma->ma_lmm_size = rr->rr_eadatalen;
 994                         }
 995                 }
 996         }
 997
 998         rc = mdt_dlmreq_unpack(info);
 999         RETURN(rc);
1000 }
1001
1002 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1003 {
1004         struct md_attr          *ma = &info->mti_attr;
1005         struct req_capsule      *pill = info->mti_pill;
1006         ENTRY;
1007
1008         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1009                 RETURN(0);
1010
1011         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1012
1013         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1014             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1015                 RETURN(-EFAULT);
1016
1017         RETURN(0);
1018 }
1019
1020 int mdt_close_unpack(struct mdt_thread_info *info)
1021 {
1022         int rc;
1023         ENTRY;
1024
1025         rc = mdt_ioepoch_unpack(info);
1026         if (rc)
1027                 RETURN(rc);
1028
1029         rc = mdt_setattr_unpack_rec(info);
1030         if (rc)
1031                 RETURN(rc);
1032
1033         rc = mdt_hsm_release_unpack(info);
1034         if (rc)
1035                 RETURN(rc);
1036
1037         RETURN(mdt_init_ucred_reint(info));
1038 }
1039
1040 static int mdt_create_unpack(struct mdt_thread_info *info)
1041 {
1042         struct lu_ucred         *uc  = mdt_ucred(info);
1043         struct mdt_rec_create   *rec;
1044         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1045         struct mdt_reint_record *rr = &info->mti_rr;
1046         struct req_capsule      *pill = info->mti_pill;
1047         struct md_op_spec       *sp = &info->mti_spec;
1048         int rc;
1049         ENTRY;
1050
1051         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1052         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1053         if (rec == NULL)
1054                 RETURN(-EFAULT);
1055
1056         /* This prior initialization is needed for old_init_ucred_reint() */
1057         uc->uc_fsuid = rec->cr_fsuid;
1058         uc->uc_fsgid = rec->cr_fsgid;
1059         uc->uc_cap   = rec->cr_cap;
1060         uc->uc_suppgids[0] = rec->cr_suppgid1;
1061         uc->uc_suppgids[1] = -1;
1062         uc->uc_umask = rec->cr_umask;
1063
1064         rr->rr_fid1 = &rec->cr_fid1;
1065         rr->rr_fid2 = &rec->cr_fid2;
1066         attr->la_mode = rec->cr_mode;
1067         attr->la_rdev  = rec->cr_rdev;
1068         attr->la_uid   = rec->cr_fsuid;
1069         attr->la_gid   = rec->cr_fsgid;
1070         attr->la_ctime = rec->cr_time;
1071         attr->la_mtime = rec->cr_time;
1072         attr->la_atime = rec->cr_time;
1073         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1074                          LA_CTIME | LA_MTIME | LA_ATIME;
1075         memset(&sp->u, 0, sizeof(sp->u));
1076         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1077
1078         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1079                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1080                                  req_capsule_client_get(pill, &RMF_CAPA1));
1081         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1082
1083         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1084         if (rc < 0)
1085                 RETURN(rc);
1086
1087         if (S_ISLNK(attr->la_mode)) {
1088                 const char *tgt = NULL;
1089
1090                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1091                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1092                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1093                         sp->u.sp_symname = tgt;
1094                 }
1095                 if (tgt == NULL)
1096                         RETURN(-EFAULT);
1097         } else {
1098                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1099                 if (S_ISDIR(attr->la_mode) &&
1100                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1101                         sp->u.sp_ea.eadata =
1102                                 req_capsule_client_get(pill, &RMF_EADATA);
1103                         sp->u.sp_ea.eadatalen =
1104                                 req_capsule_get_size(pill, &RMF_EADATA,
1105                                                      RCL_CLIENT);
1106                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1107                 }
1108         }
1109
1110         rc = mdt_dlmreq_unpack(info);
1111         RETURN(rc);
1112 }
1113
1114 static int mdt_link_unpack(struct mdt_thread_info *info)
1115 {
1116         struct lu_ucred         *uc  = mdt_ucred(info);
1117         struct mdt_rec_link     *rec;
1118         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1119         struct mdt_reint_record *rr = &info->mti_rr;
1120         struct req_capsule      *pill = info->mti_pill;
1121         int rc;
1122         ENTRY;
1123
1124         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1125         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1126         if (rec == NULL)
1127                 RETURN(-EFAULT);
1128
1129         /* This prior initialization is needed for old_init_ucred_reint() */
1130         uc->uc_fsuid = rec->lk_fsuid;
1131         uc->uc_fsgid = rec->lk_fsgid;
1132         uc->uc_cap   = rec->lk_cap;
1133         uc->uc_suppgids[0] = rec->lk_suppgid1;
1134         uc->uc_suppgids[1] = rec->lk_suppgid2;
1135
1136         attr->la_uid = rec->lk_fsuid;
1137         attr->la_gid = rec->lk_fsgid;
1138         rr->rr_fid1 = &rec->lk_fid1;
1139         rr->rr_fid2 = &rec->lk_fid2;
1140         attr->la_ctime = rec->lk_time;
1141         attr->la_mtime = rec->lk_time;
1142         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1143
1144         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1145                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1146                                  req_capsule_client_get(pill, &RMF_CAPA1));
1147         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1148                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1149                                  req_capsule_client_get(pill, &RMF_CAPA2));
1150
1151         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1152         if (rc < 0)
1153                 RETURN(rc);
1154
1155         rc = mdt_dlmreq_unpack(info);
1156
1157         RETURN(rc);
1158 }
1159
1160 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1161 {
1162         struct lu_ucred         *uc  = mdt_ucred(info);
1163         struct mdt_rec_unlink   *rec;
1164         struct md_attr          *ma = &info->mti_attr;
1165         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1166         struct mdt_reint_record *rr = &info->mti_rr;
1167         struct req_capsule      *pill = info->mti_pill;
1168         int rc;
1169         ENTRY;
1170
1171         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1172         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1173         if (rec == NULL)
1174                 RETURN(-EFAULT);
1175
1176         /* This prior initialization is needed for old_init_ucred_reint() */
1177         uc->uc_fsuid = rec->ul_fsuid;
1178         uc->uc_fsgid = rec->ul_fsgid;
1179         uc->uc_cap   = rec->ul_cap;
1180         uc->uc_suppgids[0] = rec->ul_suppgid1;
1181         uc->uc_suppgids[1] = -1;
1182
1183         attr->la_uid = rec->ul_fsuid;
1184         attr->la_gid = rec->ul_fsgid;
1185         rr->rr_fid1 = &rec->ul_fid1;
1186         rr->rr_fid2 = &rec->ul_fid2;
1187         attr->la_ctime = rec->ul_time;
1188         attr->la_mtime = rec->ul_time;
1189         attr->la_mode  = rec->ul_mode;
1190         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1191
1192         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1193                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1194                                  req_capsule_client_get(pill, &RMF_CAPA1));
1195
1196         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1197         if (rc < 0)
1198                 RETURN(rc);
1199
1200         if (rec->ul_bias & MDS_VTX_BYPASS)
1201                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1202         else
1203                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1204
1205         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1206
1207         rc = mdt_dlmreq_unpack(info);
1208         RETURN(rc);
1209 }
1210
1211 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1212 {
1213         info->mti_spec.sp_rm_entry = 1;
1214         return mdt_unlink_unpack(info);
1215 }
1216
1217 static int mdt_rename_unpack(struct mdt_thread_info *info)
1218 {
1219         struct lu_ucred         *uc = mdt_ucred(info);
1220         struct mdt_rec_rename   *rec;
1221         struct md_attr          *ma = &info->mti_attr;
1222         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1223         struct mdt_reint_record *rr = &info->mti_rr;
1224         struct req_capsule      *pill = info->mti_pill;
1225         int rc;
1226         ENTRY;
1227
1228         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1229         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1230         if (rec == NULL)
1231                 RETURN(-EFAULT);
1232
1233         /* This prior initialization is needed for old_init_ucred_reint() */
1234         uc->uc_fsuid = rec->rn_fsuid;
1235         uc->uc_fsgid = rec->rn_fsgid;
1236         uc->uc_cap   = rec->rn_cap;
1237         uc->uc_suppgids[0] = rec->rn_suppgid1;
1238         uc->uc_suppgids[1] = rec->rn_suppgid2;
1239
1240         attr->la_uid = rec->rn_fsuid;
1241         attr->la_gid = rec->rn_fsgid;
1242         rr->rr_fid1 = &rec->rn_fid1;
1243         rr->rr_fid2 = &rec->rn_fid2;
1244         attr->la_ctime = rec->rn_time;
1245         attr->la_mtime = rec->rn_time;
1246         /* rename_tgt contains the mode already */
1247         attr->la_mode = rec->rn_mode;
1248         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1249
1250         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1251                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1252                                  req_capsule_client_get(pill, &RMF_CAPA1));
1253         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1254                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1255                                  req_capsule_client_get(pill, &RMF_CAPA2));
1256
1257         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1258         if (rc < 0)
1259                 RETURN(rc);
1260
1261         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1262         if (rc < 0)
1263                 RETURN(rc);
1264
1265         if (rec->rn_bias & MDS_VTX_BYPASS)
1266                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1267         else
1268                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1269
1270         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1271
1272
1273         rc = mdt_dlmreq_unpack(info);
1274
1275         RETURN(rc);
1276 }
1277
1278 /*
1279  * please see comment above LOV_MAGIC_V1_DEF
1280  */
1281 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1282 {
1283         struct mdt_reint_record *rr = &info->mti_rr;
1284         struct lov_user_md_v1   *v1;
1285
1286         v1 = (void *)rr->rr_eadata;
1287         LASSERT(v1);
1288
1289         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1290                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1291                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1292                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1293                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1294                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1295                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1296                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1297                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1298                 }
1299         }
1300 }
1301
1302 static int mdt_open_unpack(struct mdt_thread_info *info)
1303 {
1304         struct lu_ucred         *uc = mdt_ucred(info);
1305         struct mdt_rec_create   *rec;
1306         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1307         struct req_capsule      *pill = info->mti_pill;
1308         struct mdt_reint_record *rr   = &info->mti_rr;
1309         struct ptlrpc_request   *req  = mdt_info_req(info);
1310         struct md_op_spec       *sp   = &info->mti_spec;
1311         ENTRY;
1312
1313         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1314         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1315         if (rec == NULL)
1316                 RETURN(-EFAULT);
1317
1318         /* This prior initialization is needed for old_init_ucred_reint() */
1319         uc->uc_fsuid = rec->cr_fsuid;
1320         uc->uc_fsgid = rec->cr_fsgid;
1321         uc->uc_cap   = rec->cr_cap;
1322         uc->uc_suppgids[0] = rec->cr_suppgid1;
1323         uc->uc_suppgids[1] = rec->cr_suppgid2;
1324         uc->uc_umask = rec->cr_umask;
1325
1326         rr->rr_fid1   = &rec->cr_fid1;
1327         rr->rr_fid2   = &rec->cr_fid2;
1328         rr->rr_handle = &rec->cr_old_handle;
1329         attr->la_mode = rec->cr_mode;
1330         attr->la_rdev  = rec->cr_rdev;
1331         attr->la_uid   = rec->cr_fsuid;
1332         attr->la_gid   = rec->cr_fsgid;
1333         attr->la_ctime = rec->cr_time;
1334         attr->la_mtime = rec->cr_time;
1335         attr->la_atime = rec->cr_time;
1336         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1337                          LA_CTIME | LA_MTIME | LA_ATIME;
1338         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1339         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1340         /* Do not trigger ASSERTION if client miss to set such flags. */
1341         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1342                 RETURN(-EPROTO);
1343         info->mti_replayepoch = rec->cr_ioepoch;
1344
1345         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1346
1347         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1348                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1349                                  req_capsule_client_get(pill, &RMF_CAPA1));
1350         if (req_is_replay(req) &&
1351             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1352 #if 0
1353                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1354                                  req_capsule_client_get(pill, &RMF_CAPA2));
1355 #else
1356                 /*
1357                  * FIXME: capa in replay open request might have expired,
1358                  * bypass capa check. Security hole?
1359                  */
1360                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1361                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1362 #endif
1363         }
1364
1365         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1366
1367         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1368                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1369                                                         RCL_CLIENT);
1370                 if (rr->rr_eadatalen > 0) {
1371                         rr->rr_eadata = req_capsule_client_get(pill,
1372                                                                &RMF_EADATA);
1373                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1374                         sp->u.sp_ea.eadata = rr->rr_eadata;
1375                         sp->no_create = !!req_is_replay(req);
1376                         mdt_fix_lov_magic(info);
1377                 }
1378
1379                 /*
1380                  * Client default md_size may be 0 right after client start,
1381                  * until all osc are connected, set here just some reasonable
1382                  * value to prevent misbehavior.
1383                  */
1384                 if (rr->rr_eadatalen == 0 &&
1385                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1386                         rr->rr_eadatalen = MIN_MD_SIZE;
1387         }
1388
1389         RETURN(0);
1390 }
1391
1392 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1393 {
1394         struct mdt_reint_record *rr     = &info->mti_rr;
1395         struct lu_ucred         *uc     = mdt_ucred(info);
1396         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1397         struct req_capsule      *pill   = info->mti_pill;
1398         struct mdt_rec_setxattr *rec;
1399         int                      rc;
1400         ENTRY;
1401
1402
1403         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1404                          sizeof(struct mdt_rec_reint));
1405
1406         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1407         if (rec == NULL)
1408                 RETURN(-EFAULT);
1409
1410         /* This prior initialization is needed for old_init_ucred_reint() */
1411         uc->uc_fsuid  = rec->sx_fsuid;
1412         uc->uc_fsgid  = rec->sx_fsgid;
1413         uc->uc_cap    = rec->sx_cap;
1414         uc->uc_suppgids[0] = rec->sx_suppgid1;
1415         uc->uc_suppgids[1] = -1;
1416
1417         rr->rr_opcode = rec->sx_opcode;
1418         rr->rr_fid1   = &rec->sx_fid;
1419         attr->la_valid = rec->sx_valid;
1420         attr->la_ctime = rec->sx_time;
1421         attr->la_size = rec->sx_size;
1422         attr->la_flags = rec->sx_flags;
1423
1424         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1425                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1426                                  req_capsule_client_get(pill, &RMF_CAPA1));
1427         else
1428                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1429
1430         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1431         if (rc < 0)
1432                 RETURN(rc);
1433
1434         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1435                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1436                                                         RCL_CLIENT);
1437                 if (rr->rr_eadatalen > 0) {
1438                         rr->rr_eadata = req_capsule_client_get(pill,
1439                                                                &RMF_EADATA);
1440                         if (rr->rr_eadata == NULL)
1441                                 RETURN(-EFAULT);
1442                 } else {
1443                         rr->rr_eadata = NULL;
1444                 }
1445         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1446                 CDEBUG(D_INFO, "no xattr data supplied\n");
1447                 RETURN(-EFAULT);
1448         }
1449
1450         if (mdt_dlmreq_unpack(info) < 0)
1451                 RETURN(-EPROTO);
1452
1453         RETURN(0);
1454 }
1455
1456
1457 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1458
1459 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1460         [REINT_SETATTR]  = mdt_setattr_unpack,
1461         [REINT_CREATE]   = mdt_create_unpack,
1462         [REINT_LINK]     = mdt_link_unpack,
1463         [REINT_UNLINK]   = mdt_unlink_unpack,
1464         [REINT_RENAME]   = mdt_rename_unpack,
1465         [REINT_OPEN]     = mdt_open_unpack,
1466         [REINT_SETXATTR] = mdt_setxattr_unpack,
1467         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1468         [REINT_MIGRATE]  = mdt_rename_unpack,
1469 };
1470
1471 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1472 {
1473         int rc;
1474         ENTRY;
1475
1476         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1477         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1478                 info->mti_rr.rr_opcode = op;
1479                 rc = mdt_reint_unpackers[op](info);
1480         } else {
1481                 CERROR("Unexpected opcode %d\n", op);
1482                 rc = -EFAULT;
1483         }
1484         RETURN(rc);
1485 }