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