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/lib-lnet.h>
  53
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 void mdt_exit_ucred(struct mdt_thread_info *info)
  62 {
  63         struct lu_ucred   *uc  = mdt_ucred(info);
  64         struct mdt_device *mdt = info->mti_mdt;
  65
  66         LASSERT(uc != NULL);
  67         if (uc->uc_valid != UCRED_INIT) {
  68                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  69                 if (uc->uc_ginfo) {
  70                         put_group_info(uc->uc_ginfo);
  71                         uc->uc_ginfo = NULL;
  72                 }
  73                 if (uc->uc_identity) {
  74                         mdt_identity_put(mdt->mdt_identity_cache,
  75                                          uc->uc_identity);
  76                         uc->uc_identity = NULL;
  77                 }
  78                 uc->uc_valid = UCRED_INIT;
  79         }
  80 }
  81
  82 static int match_nosquash_list(struct rw_semaphore *sem,
  83                                cfs_list_t *nidlist,
  84                                lnet_nid_t peernid)
  85 {
  86         int rc;
  87         ENTRY;
  88         down_read(sem);
  89         rc = cfs_match_nid(peernid, nidlist);
  90         up_read(sem);
  91         RETURN(rc);
  92 }
  93
  94 /* root_squash for inter-MDS operations */
  95 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  96 {
  97         struct lu_ucred *ucred = mdt_ucred(info);
  98         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
  99         ENTRY;
 100
 101         LASSERT(ucred != NULL);
 102         if (!squash->rsi_uid || ucred->uc_fsuid)
 103                 RETURN(0);
 104
 105         if (match_nosquash_list(&squash->rsi_sem,
 106                                 &squash->rsi_nosquash_nids,
 107                                 peernid)) {
 108                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 109                        libcfs_nid2str(peernid));
 110                 RETURN(0);
 111         }
 112
 113         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 114                libcfs_nid2str(peernid),
 115                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 116                squash->rsi_uid, squash->rsi_gid, 0);
 117
 118         ucred->uc_fsuid = squash->rsi_uid;
 119         ucred->uc_fsgid = squash->rsi_gid;
 120         ucred->uc_cap = 0;
 121         ucred->uc_suppgids[0] = -1;
 122         ucred->uc_suppgids[1] = -1;
 123
 124         RETURN(0);
 125 }
 126
 127 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 128                           void *buf)
 129 {
 130         struct ptlrpc_request   *req = mdt_info_req(info);
 131         struct mdt_device       *mdt = info->mti_mdt;
 132         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 133         struct lu_ucred         *ucred = mdt_ucred(info);
 134         lnet_nid_t               peernid = req->rq_peer.nid;
 135         __u32                    perm = 0;
 136         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 137         int                      setuid;
 138         int                      setgid;
 139         int                      rc = 0;
 140
 141         ENTRY;
 142
 143         LASSERT(req->rq_auth_gss);
 144         LASSERT(!req->rq_auth_usr_mdt);
 145         LASSERT(req->rq_user_desc);
 146         LASSERT(ucred != NULL);
 147
 148         ucred->uc_valid = UCRED_INVALID;
 149
 150         ucred->uc_o_uid   = pud->pud_uid;
 151         ucred->uc_o_gid   = pud->pud_gid;
 152         ucred->uc_o_fsuid = pud->pud_fsuid;
 153         ucred->uc_o_fsgid = pud->pud_fsgid;
 154
 155         if (type == BODY_INIT) {
 156                 struct mdt_body *body = (struct mdt_body *)buf;
 157
 158                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 159                 ucred->uc_suppgids[1] = -1;
 160         }
 161
 162         /* sanity check: we expect the uid which client claimed is true */
 163         if (remote) {
 164                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 165                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 166                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 167                         RETURN(-EACCES);
 168                 }
 169
 170                 if (ptlrpc_user_desc_do_idmap(req, pud))
 171                         RETURN(-EACCES);
 172
 173                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 174                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 175                                "while client claims %u:%u/%u:%u\n",
 176                                libcfs_nid2str(peernid), req->rq_auth_uid,
 177                                req->rq_auth_mapped_uid,
 178                                pud->pud_uid, pud->pud_gid,
 179                                pud->pud_fsuid, pud->pud_fsgid);
 180                         RETURN(-EACCES);
 181                 }
 182         } else {
 183                 if (req->rq_auth_uid != pud->pud_uid) {
 184                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 185                                "while client claims %u:%u/%u:%u\n",
 186                                libcfs_nid2str(peernid), req->rq_auth_uid,
 187                                pud->pud_uid, pud->pud_gid,
 188                                pud->pud_fsuid, pud->pud_fsgid);
 189                         RETURN(-EACCES);
 190                 }
 191         }
 192
 193         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 194                 if (remote) {
 195                         CDEBUG(D_SEC, "remote client must run with identity_get "
 196                                "enabled!\n");
 197                         RETURN(-EACCES);
 198                 } else {
 199                         ucred->uc_identity = NULL;
 200                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 201                                CFS_SETGRP_PERM;
 202                 }
 203         } else {
 204                 struct md_identity *identity;
 205
 206                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 207                                             pud->pud_uid);
 208                 if (IS_ERR(identity)) {
 209                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 210                                      !remote)) {
 211                                 ucred->uc_identity = NULL;
 212                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 213                                        CFS_SETGRP_PERM;
 214                         } else {
 215                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 216                                        pud->pud_uid);
 217                                 RETURN(-EACCES);
 218                         }
 219                 } else {
 220                         ucred->uc_identity = identity;
 221                         perm = mdt_identity_get_perm(ucred->uc_identity,
 222                                                      remote, peernid);
 223                 }
 224         }
 225
 226         /* find out the setuid/setgid attempt */
 227         setuid = (pud->pud_uid != pud->pud_fsuid);
 228         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 229                   (ucred->uc_identity &&
 230                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 231
 232         /* check permission of setuid */
 233         if (setuid && !(perm & CFS_SETUID_PERM)) {
 234                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 235                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 236                 GOTO(out, rc = -EACCES);
 237         }
 238
 239         /* check permission of setgid */
 240         if (setgid && !(perm & CFS_SETGID_PERM)) {
 241                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 242                        "from %s\n", pud->pud_uid, pud->pud_gid,
 243                        pud->pud_fsuid, pud->pud_fsgid,
 244                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 245                 GOTO(out, rc = -EACCES);
 246         }
 247
 248         /*
 249          * NB: remote client not allowed to setgroups anyway.
 250          */
 251         if (!remote && perm & CFS_SETGRP_PERM) {
 252                 if (pud->pud_ngroups) {
 253                         /* setgroups for local client */
 254                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 255                         if (!ucred->uc_ginfo) {
 256                                 CERROR("failed to alloc %d groups\n",
 257                                        pud->pud_ngroups);
 258                                 GOTO(out, rc = -ENOMEM);
 259                         }
 260
 261                         lustre_groups_from_list(ucred->uc_ginfo,
 262                                                 pud->pud_groups);
 263                         lustre_groups_sort(ucred->uc_ginfo);
 264                 } else {
 265                         ucred->uc_ginfo = NULL;
 266                 }
 267         } else {
 268                 ucred->uc_suppgids[0] = -1;
 269                 ucred->uc_suppgids[1] = -1;
 270                 ucred->uc_ginfo = NULL;
 271         }
 272
 273         ucred->uc_uid   = pud->pud_uid;
 274         ucred->uc_gid   = pud->pud_gid;
 275         ucred->uc_fsuid = pud->pud_fsuid;
 276         ucred->uc_fsgid = pud->pud_fsgid;
 277
 278         /* process root_squash here. */
 279         mdt_root_squash(info, peernid);
 280
 281         /* remove fs privilege for non-root user. */
 282         if (ucred->uc_fsuid)
 283                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 284         else
 285                 ucred->uc_cap = pud->pud_cap;
 286         if (remote && !(perm & CFS_RMTOWN_PERM))
 287                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 288                                    CFS_CAP_CHOWN_MASK);
 289         ucred->uc_valid = UCRED_NEW;
 290
 291         EXIT;
 292
 293 out:
 294         if (rc) {
 295                 if (ucred->uc_ginfo) {
 296                         put_group_info(ucred->uc_ginfo);
 297                         ucred->uc_ginfo = NULL;
 298                 }
 299                 if (ucred->uc_identity) {
 300                         mdt_identity_put(mdt->mdt_identity_cache,
 301                                          ucred->uc_identity);
 302                         ucred->uc_identity = NULL;
 303                 }
 304         }
 305
 306         return rc;
 307 }
 308
 309 int mdt_check_ucred(struct mdt_thread_info *info)
 310 {
 311         struct ptlrpc_request   *req = mdt_info_req(info);
 312         struct mdt_device       *mdt = info->mti_mdt;
 313         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 314         struct lu_ucred         *ucred = mdt_ucred(info);
 315         struct md_identity      *identity = NULL;
 316         lnet_nid_t               peernid = req->rq_peer.nid;
 317         __u32                    perm = 0;
 318         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 319         int                      setuid;
 320         int                      setgid;
 321         int                      rc = 0;
 322
 323         ENTRY;
 324
 325         LASSERT(ucred != NULL);
 326         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 327                 RETURN(0);
 328
 329         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 330                 RETURN(0);
 331
 332         /* sanity check: if we use strong authentication, we expect the
 333          * uid which client claimed is true */
 334         if (remote) {
 335                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 336                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 337                         RETURN(-EACCES);
 338                 }
 339
 340                 if (ptlrpc_user_desc_do_idmap(req, pud))
 341                         RETURN(-EACCES);
 342
 343                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 344                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 345                                "while client claims %u:%u/%u:%u\n",
 346                                libcfs_nid2str(peernid), req->rq_auth_uid,
 347                                req->rq_auth_mapped_uid,
 348                                pud->pud_uid, pud->pud_gid,
 349                                pud->pud_fsuid, pud->pud_fsgid);
 350                         RETURN(-EACCES);
 351                 }
 352         } else {
 353                 if (req->rq_auth_uid != pud->pud_uid) {
 354                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 355                                "while client claims %u:%u/%u:%u\n",
 356                                libcfs_nid2str(peernid), req->rq_auth_uid,
 357                                pud->pud_uid, pud->pud_gid,
 358                                pud->pud_fsuid, pud->pud_fsgid);
 359                         RETURN(-EACCES);
 360                 }
 361         }
 362
 363         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 364                 if (remote) {
 365                         CDEBUG(D_SEC, "remote client must run with identity_get "
 366                                "enabled!\n");
 367                         RETURN(-EACCES);
 368                 }
 369                 RETURN(0);
 370         }
 371
 372         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 373         if (IS_ERR(identity)) {
 374                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 375                              !remote)) {
 376                         RETURN(0);
 377                 } else {
 378                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 379                                pud->pud_uid);
 380                         RETURN(-EACCES);
 381                }
 382         }
 383
 384         perm = mdt_identity_get_perm(identity, remote, peernid);
 385         /* find out the setuid/setgid attempt */
 386         setuid = (pud->pud_uid != pud->pud_fsuid);
 387         setgid = (pud->pud_gid != pud->pud_fsgid ||
 388                   pud->pud_gid != identity->mi_gid);
 389
 390         /* check permission of setuid */
 391         if (setuid && !(perm & CFS_SETUID_PERM)) {
 392                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 393                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 394                 GOTO(out, rc = -EACCES);
 395         }
 396
 397         /* check permission of setgid */
 398         if (setgid && !(perm & CFS_SETGID_PERM)) {
 399                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 400                        "from %s\n", pud->pud_uid, pud->pud_gid,
 401                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 402                        libcfs_nid2str(peernid));
 403                 GOTO(out, rc = -EACCES);
 404         }
 405
 406         EXIT;
 407
 408 out:
 409         mdt_identity_put(mdt->mdt_identity_cache, identity);
 410         return rc;
 411 }
 412
 413 static int old_init_ucred(struct mdt_thread_info *info,
 414                           struct mdt_body *body)
 415 {
 416         struct lu_ucred *uc = mdt_ucred(info);
 417         struct mdt_device  *mdt = info->mti_mdt;
 418         struct md_identity *identity = NULL;
 419
 420         ENTRY;
 421
 422         LASSERT(uc != NULL);
 423         uc->uc_valid = UCRED_INVALID;
 424         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 425         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 426         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 427         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 428         uc->uc_suppgids[0] = body->mbo_suppgid;
 429         uc->uc_suppgids[1] = -1;
 430         uc->uc_ginfo = NULL;
 431         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 432                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 433                                             uc->uc_fsuid);
 434                 if (IS_ERR(identity)) {
 435                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 436                                 identity = NULL;
 437                         } else {
 438                                 CDEBUG(D_SEC, "Deny access without identity: "
 439                                        "uid %u\n", uc->uc_fsuid);
 440                                 RETURN(-EACCES);
 441                         }
 442                 }
 443         }
 444         uc->uc_identity = identity;
 445
 446         /* process root_squash here. */
 447         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 448
 449         /* remove fs privilege for non-root user. */
 450         if (uc->uc_fsuid)
 451                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 452         else
 453                 uc->uc_cap = body->mbo_capability;
 454         uc->uc_valid = UCRED_OLD;
 455
 456         RETURN(0);
 457 }
 458
 459 static int old_init_ucred_reint(struct mdt_thread_info *info)
 460 {
 461         struct lu_ucred *uc = mdt_ucred(info);
 462         struct mdt_device  *mdt = info->mti_mdt;
 463         struct md_identity *identity = NULL;
 464
 465         ENTRY;
 466
 467         LASSERT(uc != NULL);
 468         uc->uc_valid = UCRED_INVALID;
 469         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 470         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 471         uc->uc_ginfo = NULL;
 472         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 473                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 474                                             uc->uc_fsuid);
 475                 if (IS_ERR(identity)) {
 476                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 477                                 identity = NULL;
 478                         } else {
 479                                 CDEBUG(D_SEC, "Deny access without identity: "
 480                                        "uid %u\n", uc->uc_fsuid);
 481                                 RETURN(-EACCES);
 482                         }
 483                 }
 484         }
 485         uc->uc_identity = identity;
 486
 487         /* process root_squash here. */
 488         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 489
 490         /* remove fs privilege for non-root user. */
 491         if (uc->uc_fsuid)
 492                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 493         uc->uc_valid = UCRED_OLD;
 494
 495         RETURN(0);
 496 }
 497
 498 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 499 {
 500         struct ptlrpc_request *req = mdt_info_req(info);
 501         struct lu_ucred       *uc  = mdt_ucred(info);
 502
 503         LASSERT(uc != NULL);
 504         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 505                 return 0;
 506
 507         mdt_exit_ucred(info);
 508
 509         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 510                 return old_init_ucred(info, body);
 511         else
 512                 return new_init_ucred(info, BODY_INIT, body);
 513 }
 514
 515 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 516 {
 517         struct ptlrpc_request *req = mdt_info_req(info);
 518         struct lu_ucred       *uc  = mdt_ucred(info);
 519
 520         LASSERT(uc != NULL);
 521         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 522                 return 0;
 523
 524         mdt_exit_ucred(info);
 525
 526         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 527                 return old_init_ucred_reint(info);
 528         else
 529                 return new_init_ucred(info, REC_INIT, NULL);
 530 }
 531
 532 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 533 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 534 {
 535         const struct lov_ost_data_v1    *lod;
 536         int                              i;
 537         __u16                            count;
 538
 539         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 540                 return;
 541
 542         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 543
 544         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 545                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 546                le32_to_cpu(lmm->lmm_pattern));
 547         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 548                le32_to_cpu(lmm->lmm_stripe_size), count);
 549
 550         /* If it's a directory or a released file, then there are
 551          * no actual objects to print, so bail out. */
 552         if (valid & OBD_MD_FLDIREA ||
 553             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 554                 return;
 555
 556         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 557         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 558                 struct ost_id oi;
 559
 560                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 561                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 562                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 563         }
 564 }
 565
 566 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 567 {
 568         const struct lmv_mds_md_v1 *lmm1;
 569         int                        i;
 570
 571         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 572                 return;
 573
 574         lmm1 = &lmv->lmv_md_v1;
 575         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 576                le32_to_cpu(lmm1->lmv_magic),
 577                le32_to_cpu(lmm1->lmv_master_mdt_index),
 578                le32_to_cpu(lmm1->lmv_stripe_count));
 579
 580         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 581                 return;
 582
 583         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 584                 struct lu_fid fid;
 585
 586                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 587                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 588         }
 589 }
 590
 591 /* Shrink and/or grow reply buffers */
 592 int mdt_fix_reply(struct mdt_thread_info *info)
 593 {
 594         struct req_capsule *pill = info->mti_pill;
 595         struct mdt_body    *body;
 596         int                md_size, md_packed = 0;
 597         int                acl_size;
 598         int                rc = 0;
 599         ENTRY;
 600
 601         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 602         LASSERT(body != NULL);
 603
 604         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 605                                OBD_MD_LINKNAME))
 606                 md_size = body->mbo_eadatasize;
 607         else
 608                 md_size = 0;
 609
 610         acl_size = body->mbo_aclsize;
 611
 612         /* this replay - not send info to client */
 613         if (info->mti_spec.no_create) {
 614                 md_size = 0;
 615                 acl_size = 0;
 616         }
 617
 618         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 619                " MDSCAPA = %llx, OSSCAPA = %llx\n",
 620                md_size, acl_size,
 621                (unsigned long long)(body->mbo_valid & OBD_MD_FLMDSCAPA),
 622                (unsigned long long)(body->mbo_valid & OBD_MD_FLOSSCAPA));
 623 /*
 624             &RMF_MDT_BODY,
 625             &RMF_MDT_MD,
 626             &RMF_ACL, or &RMF_LOGCOOKIES
 627 (optional)  &RMF_CAPA1,
 628 (optional)  &RMF_CAPA2,
 629 (optional)  something else
 630 */
 631
 632         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 633          * buffers before growing it */
 634         if (info->mti_big_lmm_used) {
 635                 /* big_lmm buffer may be used even without packing the result
 636                  * into reply, just for internal server needs */
 637                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 638                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 639                                                          RCL_SERVER);
 640
 641                 /* free big lmm if md_size is not needed */
 642                 if (md_size == 0 || md_packed == 0) {
 643                         info->mti_big_lmm_used = 0;
 644                 } else {
 645                         /* buffer must be allocated separately */
 646                         LASSERT(info->mti_attr.ma_lmm !=
 647                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 648                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 649                 }
 650         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 651                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 652         }
 653
 654         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 655                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 656         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 657                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 658                                    acl_size, RCL_SERVER);
 659
 660         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 661             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 662                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 663
 664         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 665             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 666                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 667
 668         /*
 669          * Some more field should be shrinked if needed.
 670          * This should be done by those who added fields to reply message.
 671          */
 672
 673         /* Grow MD buffer if needed finally */
 674         if (info->mti_big_lmm_used) {
 675                 void *lmm;
 676
 677                 LASSERT(md_size > md_packed);
 678                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 679                        md_size - md_packed);
 680                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 681                 if (rc) {
 682                         /* we can't answer with proper LOV EA, drop flags,
 683                          * the rc is also returned so this request is
 684                          * considered as failed */
 685                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 686                         /* don't return transno along with error */
 687                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 688                 } else {
 689                         /* now we need to pack right LOV/LMV EA */
 690                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 691                         if (info->mti_attr.ma_valid & MA_LOV) {
 692                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 693                                                              RCL_SERVER) ==
 694                                                 info->mti_attr.ma_lmm_size);
 695                                 memcpy(lmm, info->mti_attr.ma_lmm,
 696                                        info->mti_attr.ma_lmm_size);
 697                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 698                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 699                                                              RCL_SERVER) ==
 700                                                 info->mti_attr.ma_lmv_size);
 701                                 memcpy(lmm, info->mti_attr.ma_lmv,
 702                                        info->mti_attr.ma_lmv_size);
 703                         }
 704                 }
 705                 /* update mdt_max_mdsize so clients will be aware about that */
 706                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 707                         info->mti_mdt->mdt_max_mdsize =
 708                                                     info->mti_attr.ma_lmm_size;
 709                 info->mti_big_lmm_used = 0;
 710         }
 711         RETURN(rc);
 712 }
 713
 714
 715 /* if object is dying, pack the lov/llog data,
 716  * parameter info->mti_attr should be valid at this point! */
 717 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 718                            const struct md_attr *ma)
 719 {
 720         struct mdt_body       *repbody;
 721         const struct lu_attr *la = &ma->ma_attr;
 722         int rc;
 723         ENTRY;
 724
 725         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 726         LASSERT(repbody != NULL);
 727
 728         if (ma->ma_valid & MA_INODE)
 729                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 730
 731         if (ma->ma_valid & MA_LOV) {
 732                 CERROR("No need in LOV EA upon unlink\n");
 733                 dump_stack();
 734         }
 735         repbody->mbo_eadatasize = 0;
 736
 737         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 738             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 739             repbody->mbo_valid & OBD_MD_FLEASIZE) {
 740                 struct lustre_capa *capa;
 741
 742                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 743                 LASSERT(capa);
 744                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 745                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 746                 if (rc)
 747                         RETURN(rc);
 748
 749                 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
 750         }
 751
 752         RETURN(0);
 753 }
 754
 755 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 756                                   struct md_attr *ma)
 757 {
 758         __u64 out;
 759
 760         out = 0;
 761         if (in & MDS_ATTR_MODE)
 762                 out |= LA_MODE;
 763         if (in & MDS_ATTR_UID)
 764                 out |= LA_UID;
 765         if (in & MDS_ATTR_GID)
 766                 out |= LA_GID;
 767         if (in & MDS_ATTR_SIZE)
 768                 out |= LA_SIZE;
 769         if (in & MDS_ATTR_BLOCKS)
 770                 out |= LA_BLOCKS;
 771         if (in & MDS_ATTR_ATIME_SET)
 772                 out |= LA_ATIME;
 773         if (in & MDS_ATTR_CTIME_SET)
 774                 out |= LA_CTIME;
 775         if (in & MDS_ATTR_MTIME_SET)
 776                 out |= LA_MTIME;
 777         if (in & MDS_ATTR_ATTR_FLAG)
 778                 out |= LA_FLAGS;
 779         if (in & MDS_ATTR_KILL_SUID)
 780                 out |= LA_KILL_SUID;
 781         if (in & MDS_ATTR_KILL_SGID)
 782                 out |= LA_KILL_SGID;
 783
 784         if (in & MDS_ATTR_FROM_OPEN)
 785                 rr->rr_flags |= MRF_OPEN_TRUNC;
 786         if (in & MDS_OPEN_OWNEROVERRIDE)
 787                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 788         if (in & MDS_ATTR_FORCE)
 789                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 790
 791         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 792                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 793                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 794                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 795                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 796                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 797         if (in != 0)
 798                 CERROR("Unknown attr bits: "LPX64"\n", in);
 799         return out;
 800 }
 801
 802 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 803                       const struct lu_fid *fid, struct lustre_capa *capa)
 804 {
 805         struct lu_capainfo *lci;
 806
 807         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 808         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 809             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 810                 return;
 811
 812         lci = lu_capainfo_get(info->mti_env);
 813         LASSERT(lci);
 814         lci->lci_fid[offset]  = *fid;
 815         lci->lci_capa[offset] = capa;
 816 }
 817
 818 #ifdef DEBUG_CAPA
 819 void mdt_dump_capainfo(struct mdt_thread_info *info)
 820 {
 821         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 822         int i;
 823
 824         if (lci == NULL)
 825                 return;
 826
 827         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 828                 if (lci->lci_capa[i] == NULL) {
 829                         CERROR("no capa for index %d "DFID"\n",
 830                                i, PFID(&lci->lci_fid[i]));
 831                         continue;
 832                 }
 833                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 834                         CERROR("bypass for index %d "DFID"\n",
 835                                i, PFID(&lci->lci_fid[i]));
 836                         continue;
 837                 }
 838                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 839         }
 840 }
 841 #endif /* DEBUG_CAPA */
 842
 843 /* unpacking */
 844
 845 int mdt_name_unpack(struct req_capsule *pill,
 846                     const struct req_msg_field *field,
 847                     struct lu_name *ln,
 848                     enum mdt_name_flags flags)
 849 {
 850         ln->ln_name = req_capsule_client_get(pill, field);
 851         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 852
 853         if (!lu_name_is_valid(ln)) {
 854                 ln->ln_name = NULL;
 855                 ln->ln_namelen = 0;
 856
 857                 return -EPROTO;
 858         }
 859
 860         if ((flags & MNF_FIX_ANON) &&
 861             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 862                 /* Newer (3.x) kernels use a name of "/" for the
 863                  * "anonymous" disconnected dentries from NFS
 864                  * filehandle conversion. See d_obtain_alias(). */
 865                 ln->ln_name = NULL;
 866                 ln->ln_namelen = 0;
 867         }
 868
 869         return 0;
 870 }
 871
 872 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 873 {
 874         struct lu_ucred         *uc  = mdt_ucred(info);
 875         struct md_attr          *ma = &info->mti_attr;
 876         struct lu_attr          *la = &ma->ma_attr;
 877         struct req_capsule      *pill = info->mti_pill;
 878         struct mdt_reint_record *rr = &info->mti_rr;
 879         struct mdt_rec_setattr  *rec;
 880         ENTRY;
 881
 882         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 883         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 884         if (rec == NULL)
 885                 RETURN(-EFAULT);
 886
 887         /* This prior initialization is needed for old_init_ucred_reint() */
 888         uc->uc_fsuid = rec->sa_fsuid;
 889         uc->uc_fsgid = rec->sa_fsgid;
 890         uc->uc_cap   = rec->sa_cap;
 891         uc->uc_suppgids[0] = rec->sa_suppgid;
 892         uc->uc_suppgids[1] = -1;
 893
 894         rr->rr_fid1 = &rec->sa_fid;
 895         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 896         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 897          *  the client does not have OBD_CONNECT_FULL20, convert it
 898          *  to LA_xTIME. LU-3036 */
 899         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 900                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 901                      (rec->sa_valid & MDS_ATTR_ATIME))
 902                         la->la_valid |= LA_ATIME;
 903                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 904                      (rec->sa_valid & MDS_ATTR_MTIME))
 905                         la->la_valid |= LA_MTIME;
 906                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 907                      (rec->sa_valid & MDS_ATTR_CTIME))
 908                         la->la_valid |= LA_CTIME;
 909         }
 910         la->la_mode  = rec->sa_mode;
 911         la->la_flags = rec->sa_attr_flags;
 912         la->la_uid   = rec->sa_uid;
 913         la->la_gid   = rec->sa_gid;
 914         la->la_size  = rec->sa_size;
 915         la->la_blocks = rec->sa_blocks;
 916         la->la_ctime = rec->sa_ctime;
 917         la->la_atime = rec->sa_atime;
 918         la->la_mtime = rec->sa_mtime;
 919         ma->ma_valid = MA_INODE;
 920
 921         if (rec->sa_bias & MDS_DATA_MODIFIED)
 922                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 923         else
 924                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 925
 926         if (rec->sa_bias & MDS_HSM_RELEASE)
 927                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 928         else
 929                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 930
 931         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 932                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 933                                  req_capsule_client_get(pill, &RMF_CAPA1));
 934
 935         RETURN(0);
 936 }
 937
 938 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 939 {
 940         struct req_capsule *pill = info->mti_pill;
 941         ENTRY;
 942
 943         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 944                 info->mti_ioepoch =
 945                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 946         else
 947                 info->mti_ioepoch = NULL;
 948         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 949 }
 950
 951 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 952         struct req_capsule      *pill = info->mti_pill;
 953
 954         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 955                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 956                 if (info->mti_dlm_req == NULL)
 957                         RETURN(-EFAULT);
 958         }
 959
 960         RETURN(0);
 961 }
 962
 963 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 964 {
 965         struct mdt_reint_record *rr = &info->mti_rr;
 966         struct md_attr          *ma = &info->mti_attr;
 967         struct req_capsule      *pill = info->mti_pill;
 968         int rc;
 969         ENTRY;
 970
 971         rc = mdt_setattr_unpack_rec(info);
 972         if (rc)
 973                 RETURN(rc);
 974
 975         /* Epoch may be absent */
 976         mdt_ioepoch_unpack(info);
 977
 978         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 979                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 980                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 981                                                         RCL_CLIENT);
 982                 if (rr->rr_eadatalen > 0) {
 983                         const struct lmv_user_md        *lum;
 984
 985                         lum = rr->rr_eadata;
 986                         /* Sigh ma_valid(from req) does not indicate whether
 987                          * it will set LOV/LMV EA, so we have to check magic */
 988                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 989                                 ma->ma_valid |= MA_LMV;
 990                                 ma->ma_lmv = (void *)rr->rr_eadata;
 991                                 ma->ma_lmv_size = rr->rr_eadatalen;
 992                         } else {
 993                                 ma->ma_valid |= MA_LOV;
 994                                 ma->ma_lmm = (void *)rr->rr_eadata;
 995                                 ma->ma_lmm_size = rr->rr_eadatalen;
 996                         }
 997                 }
 998         }
 999
1000         rc = mdt_dlmreq_unpack(info);
1001         RETURN(rc);
1002 }
1003
1004 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1005 {
1006         struct md_attr          *ma = &info->mti_attr;
1007         struct req_capsule      *pill = info->mti_pill;
1008         ENTRY;
1009
1010         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1011                 RETURN(0);
1012
1013         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1014
1015         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1016             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1017                 RETURN(-EFAULT);
1018
1019         RETURN(0);
1020 }
1021
1022 int mdt_close_unpack(struct mdt_thread_info *info)
1023 {
1024         int rc;
1025         ENTRY;
1026
1027         rc = mdt_ioepoch_unpack(info);
1028         if (rc)
1029                 RETURN(rc);
1030
1031         rc = mdt_setattr_unpack_rec(info);
1032         if (rc)
1033                 RETURN(rc);
1034
1035         rc = mdt_hsm_release_unpack(info);
1036         if (rc)
1037                 RETURN(rc);
1038
1039         RETURN(mdt_init_ucred_reint(info));
1040 }
1041
1042 static int mdt_create_unpack(struct mdt_thread_info *info)
1043 {
1044         struct lu_ucred         *uc  = mdt_ucred(info);
1045         struct mdt_rec_create   *rec;
1046         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1047         struct mdt_reint_record *rr = &info->mti_rr;
1048         struct req_capsule      *pill = info->mti_pill;
1049         struct md_op_spec       *sp = &info->mti_spec;
1050         int rc;
1051         ENTRY;
1052
1053         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1054         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1055         if (rec == NULL)
1056                 RETURN(-EFAULT);
1057
1058         /* This prior initialization is needed for old_init_ucred_reint() */
1059         uc->uc_fsuid = rec->cr_fsuid;
1060         uc->uc_fsgid = rec->cr_fsgid;
1061         uc->uc_cap   = rec->cr_cap;
1062         uc->uc_suppgids[0] = rec->cr_suppgid1;
1063         uc->uc_suppgids[1] = -1;
1064         uc->uc_umask = rec->cr_umask;
1065
1066         rr->rr_fid1 = &rec->cr_fid1;
1067         rr->rr_fid2 = &rec->cr_fid2;
1068         attr->la_mode = rec->cr_mode;
1069         attr->la_rdev  = rec->cr_rdev;
1070         attr->la_uid   = rec->cr_fsuid;
1071         attr->la_gid   = rec->cr_fsgid;
1072         attr->la_ctime = rec->cr_time;
1073         attr->la_mtime = rec->cr_time;
1074         attr->la_atime = rec->cr_time;
1075         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1076                          LA_CTIME | LA_MTIME | LA_ATIME;
1077         memset(&sp->u, 0, sizeof(sp->u));
1078         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1079
1080         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1081                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1082                                  req_capsule_client_get(pill, &RMF_CAPA1));
1083         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1084
1085         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1086         if (rc < 0)
1087                 RETURN(rc);
1088
1089         if (S_ISLNK(attr->la_mode)) {
1090                 const char *tgt = NULL;
1091
1092                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1093                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1094                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1095                         sp->u.sp_symname = tgt;
1096                 }
1097                 if (tgt == NULL)
1098                         RETURN(-EFAULT);
1099         } else {
1100                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1101                 if (S_ISDIR(attr->la_mode) &&
1102                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1103                         sp->u.sp_ea.eadata =
1104                                 req_capsule_client_get(pill, &RMF_EADATA);
1105                         sp->u.sp_ea.eadatalen =
1106                                 req_capsule_get_size(pill, &RMF_EADATA,
1107                                                      RCL_CLIENT);
1108                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1109                 }
1110         }
1111
1112         rc = mdt_dlmreq_unpack(info);
1113         RETURN(rc);
1114 }
1115
1116 static int mdt_link_unpack(struct mdt_thread_info *info)
1117 {
1118         struct lu_ucred         *uc  = mdt_ucred(info);
1119         struct mdt_rec_link     *rec;
1120         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1121         struct mdt_reint_record *rr = &info->mti_rr;
1122         struct req_capsule      *pill = info->mti_pill;
1123         int rc;
1124         ENTRY;
1125
1126         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1127         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1128         if (rec == NULL)
1129                 RETURN(-EFAULT);
1130
1131         /* This prior initialization is needed for old_init_ucred_reint() */
1132         uc->uc_fsuid = rec->lk_fsuid;
1133         uc->uc_fsgid = rec->lk_fsgid;
1134         uc->uc_cap   = rec->lk_cap;
1135         uc->uc_suppgids[0] = rec->lk_suppgid1;
1136         uc->uc_suppgids[1] = rec->lk_suppgid2;
1137
1138         attr->la_uid = rec->lk_fsuid;
1139         attr->la_gid = rec->lk_fsgid;
1140         rr->rr_fid1 = &rec->lk_fid1;
1141         rr->rr_fid2 = &rec->lk_fid2;
1142         attr->la_ctime = rec->lk_time;
1143         attr->la_mtime = rec->lk_time;
1144         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1145
1146         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1147                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1148                                  req_capsule_client_get(pill, &RMF_CAPA1));
1149         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1150                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1151                                  req_capsule_client_get(pill, &RMF_CAPA2));
1152
1153         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1154         if (rc < 0)
1155                 RETURN(rc);
1156
1157         rc = mdt_dlmreq_unpack(info);
1158
1159         RETURN(rc);
1160 }
1161
1162 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1163 {
1164         struct lu_ucred         *uc  = mdt_ucred(info);
1165         struct mdt_rec_unlink   *rec;
1166         struct md_attr          *ma = &info->mti_attr;
1167         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1168         struct mdt_reint_record *rr = &info->mti_rr;
1169         struct req_capsule      *pill = info->mti_pill;
1170         int rc;
1171         ENTRY;
1172
1173         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1174         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1175         if (rec == NULL)
1176                 RETURN(-EFAULT);
1177
1178         /* This prior initialization is needed for old_init_ucred_reint() */
1179         uc->uc_fsuid = rec->ul_fsuid;
1180         uc->uc_fsgid = rec->ul_fsgid;
1181         uc->uc_cap   = rec->ul_cap;
1182         uc->uc_suppgids[0] = rec->ul_suppgid1;
1183         uc->uc_suppgids[1] = -1;
1184
1185         attr->la_uid = rec->ul_fsuid;
1186         attr->la_gid = rec->ul_fsgid;
1187         rr->rr_fid1 = &rec->ul_fid1;
1188         rr->rr_fid2 = &rec->ul_fid2;
1189         attr->la_ctime = rec->ul_time;
1190         attr->la_mtime = rec->ul_time;
1191         attr->la_mode  = rec->ul_mode;
1192         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1193
1194         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1195                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1196                                  req_capsule_client_get(pill, &RMF_CAPA1));
1197
1198         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1199         if (rc < 0)
1200                 RETURN(rc);
1201
1202         if (rec->ul_bias & MDS_VTX_BYPASS)
1203                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1204         else
1205                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1206
1207         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1208
1209         rc = mdt_dlmreq_unpack(info);
1210         RETURN(rc);
1211 }
1212
1213 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1214 {
1215         info->mti_spec.sp_rm_entry = 1;
1216         return mdt_unlink_unpack(info);
1217 }
1218
1219 static int mdt_rename_unpack(struct mdt_thread_info *info)
1220 {
1221         struct lu_ucred         *uc = mdt_ucred(info);
1222         struct mdt_rec_rename   *rec;
1223         struct md_attr          *ma = &info->mti_attr;
1224         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1225         struct mdt_reint_record *rr = &info->mti_rr;
1226         struct req_capsule      *pill = info->mti_pill;
1227         int rc;
1228         ENTRY;
1229
1230         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1231         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1232         if (rec == NULL)
1233                 RETURN(-EFAULT);
1234
1235         /* This prior initialization is needed for old_init_ucred_reint() */
1236         uc->uc_fsuid = rec->rn_fsuid;
1237         uc->uc_fsgid = rec->rn_fsgid;
1238         uc->uc_cap   = rec->rn_cap;
1239         uc->uc_suppgids[0] = rec->rn_suppgid1;
1240         uc->uc_suppgids[1] = rec->rn_suppgid2;
1241
1242         attr->la_uid = rec->rn_fsuid;
1243         attr->la_gid = rec->rn_fsgid;
1244         rr->rr_fid1 = &rec->rn_fid1;
1245         rr->rr_fid2 = &rec->rn_fid2;
1246         attr->la_ctime = rec->rn_time;
1247         attr->la_mtime = rec->rn_time;
1248         /* rename_tgt contains the mode already */
1249         attr->la_mode = rec->rn_mode;
1250         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1251
1252         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1253                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1254                                  req_capsule_client_get(pill, &RMF_CAPA1));
1255         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1256                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1257                                  req_capsule_client_get(pill, &RMF_CAPA2));
1258
1259         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1260         if (rc < 0)
1261                 RETURN(rc);
1262
1263         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1264         if (rc < 0)
1265                 RETURN(rc);
1266
1267         if (rec->rn_bias & MDS_VTX_BYPASS)
1268                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1269         else
1270                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1271
1272         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1273
1274
1275         rc = mdt_dlmreq_unpack(info);
1276
1277         RETURN(rc);
1278 }
1279
1280 /*
1281  * please see comment above LOV_MAGIC_V1_DEF
1282  */
1283 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1284 {
1285         struct mdt_reint_record *rr = &info->mti_rr;
1286         struct lov_user_md_v1   *v1;
1287
1288         v1 = (void *)rr->rr_eadata;
1289         LASSERT(v1);
1290
1291         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1292                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1293                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1294                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1295                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1296                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1297                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1298                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1299                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1300                 }
1301         }
1302 }
1303
1304 static int mdt_open_unpack(struct mdt_thread_info *info)
1305 {
1306         struct lu_ucred         *uc = mdt_ucred(info);
1307         struct mdt_rec_create   *rec;
1308         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1309         struct req_capsule      *pill = info->mti_pill;
1310         struct mdt_reint_record *rr   = &info->mti_rr;
1311         struct ptlrpc_request   *req  = mdt_info_req(info);
1312         struct md_op_spec       *sp   = &info->mti_spec;
1313         ENTRY;
1314
1315         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1316         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1317         if (rec == NULL)
1318                 RETURN(-EFAULT);
1319
1320         /* This prior initialization is needed for old_init_ucred_reint() */
1321         uc->uc_fsuid = rec->cr_fsuid;
1322         uc->uc_fsgid = rec->cr_fsgid;
1323         uc->uc_cap   = rec->cr_cap;
1324         uc->uc_suppgids[0] = rec->cr_suppgid1;
1325         uc->uc_suppgids[1] = rec->cr_suppgid2;
1326         uc->uc_umask = rec->cr_umask;
1327
1328         rr->rr_fid1   = &rec->cr_fid1;
1329         rr->rr_fid2   = &rec->cr_fid2;
1330         rr->rr_handle = &rec->cr_old_handle;
1331         attr->la_mode = rec->cr_mode;
1332         attr->la_rdev  = rec->cr_rdev;
1333         attr->la_uid   = rec->cr_fsuid;
1334         attr->la_gid   = rec->cr_fsgid;
1335         attr->la_ctime = rec->cr_time;
1336         attr->la_mtime = rec->cr_time;
1337         attr->la_atime = rec->cr_time;
1338         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1339                          LA_CTIME | LA_MTIME | LA_ATIME;
1340         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1341         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1342         /* Do not trigger ASSERTION if client miss to set such flags. */
1343         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1344                 RETURN(-EPROTO);
1345         info->mti_replayepoch = rec->cr_ioepoch;
1346
1347         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1348
1349         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1350                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1351                                  req_capsule_client_get(pill, &RMF_CAPA1));
1352         if (req_is_replay(req) &&
1353             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1354 #if 0
1355                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1356                                  req_capsule_client_get(pill, &RMF_CAPA2));
1357 #else
1358                 /*
1359                  * FIXME: capa in replay open request might have expired,
1360                  * bypass capa check. Security hole?
1361                  */
1362                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1363                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1364 #endif
1365         }
1366
1367         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1368
1369         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1370                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1371                                                         RCL_CLIENT);
1372                 if (rr->rr_eadatalen > 0) {
1373                         rr->rr_eadata = req_capsule_client_get(pill,
1374                                                                &RMF_EADATA);
1375                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1376                         sp->u.sp_ea.eadata = rr->rr_eadata;
1377                         sp->no_create = !!req_is_replay(req);
1378                         mdt_fix_lov_magic(info);
1379                 }
1380
1381                 /*
1382                  * Client default md_size may be 0 right after client start,
1383                  * until all osc are connected, set here just some reasonable
1384                  * value to prevent misbehavior.
1385                  */
1386                 if (rr->rr_eadatalen == 0 &&
1387                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1388                         rr->rr_eadatalen = MIN_MD_SIZE;
1389         }
1390
1391         RETURN(0);
1392 }
1393
1394 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1395 {
1396         struct mdt_reint_record *rr     = &info->mti_rr;
1397         struct lu_ucred         *uc     = mdt_ucred(info);
1398         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1399         struct req_capsule      *pill   = info->mti_pill;
1400         struct mdt_rec_setxattr *rec;
1401         int                      rc;
1402         ENTRY;
1403
1404
1405         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1406                          sizeof(struct mdt_rec_reint));
1407
1408         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1409         if (rec == NULL)
1410                 RETURN(-EFAULT);
1411
1412         /* This prior initialization is needed for old_init_ucred_reint() */
1413         uc->uc_fsuid  = rec->sx_fsuid;
1414         uc->uc_fsgid  = rec->sx_fsgid;
1415         uc->uc_cap    = rec->sx_cap;
1416         uc->uc_suppgids[0] = rec->sx_suppgid1;
1417         uc->uc_suppgids[1] = -1;
1418
1419         rr->rr_opcode = rec->sx_opcode;
1420         rr->rr_fid1   = &rec->sx_fid;
1421         attr->la_valid = rec->sx_valid;
1422         attr->la_ctime = rec->sx_time;
1423         attr->la_size = rec->sx_size;
1424         attr->la_flags = rec->sx_flags;
1425
1426         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1427                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1428                                  req_capsule_client_get(pill, &RMF_CAPA1));
1429         else
1430                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1431
1432         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1433         if (rc < 0)
1434                 RETURN(rc);
1435
1436         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1437                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1438                                                         RCL_CLIENT);
1439                 if (rr->rr_eadatalen > 0) {
1440                         rr->rr_eadata = req_capsule_client_get(pill,
1441                                                                &RMF_EADATA);
1442                         if (rr->rr_eadata == NULL)
1443                                 RETURN(-EFAULT);
1444                 } else {
1445                         rr->rr_eadata = NULL;
1446                 }
1447         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1448                 CDEBUG(D_INFO, "no xattr data supplied\n");
1449                 RETURN(-EFAULT);
1450         }
1451
1452         if (mdt_dlmreq_unpack(info) < 0)
1453                 RETURN(-EPROTO);
1454
1455         RETURN(0);
1456 }
1457
1458
1459 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1460
1461 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1462         [REINT_SETATTR]  = mdt_setattr_unpack,
1463         [REINT_CREATE]   = mdt_create_unpack,
1464         [REINT_LINK]     = mdt_link_unpack,
1465         [REINT_UNLINK]   = mdt_unlink_unpack,
1466         [REINT_RENAME]   = mdt_rename_unpack,
1467         [REINT_OPEN]     = mdt_open_unpack,
1468         [REINT_SETXATTR] = mdt_setxattr_unpack,
1469         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1470         [REINT_MIGRATE]  = mdt_rename_unpack,
1471 };
1472
1473 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1474 {
1475         int rc;
1476         ENTRY;
1477
1478         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1479         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1480                 info->mti_rr.rr_opcode = op;
1481                 rc = mdt_reint_unpackers[op](info);
1482         } else {
1483                 CERROR("Unexpected opcode %d\n", op);
1484                 rc = -EFAULT;
1485         }
1486         RETURN(rc);
1487 }