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         ENTRY;
  99
 100         LASSERT(ucred != NULL);
 101         if (!info->mti_mdt->mdt_squash_uid || ucred->uc_fsuid)
 102                 RETURN(0);
 103
 104         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 105                                 &info->mti_mdt->mdt_nosquash_nids,
 106                                 peernid)) {
 107                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 108                        libcfs_nid2str(peernid));
 109                 RETURN(0);
 110         }
 111
 112         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 113                libcfs_nid2str(peernid),
 114                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 115                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 116                0);
 117
 118         ucred->uc_fsuid = info->mti_mdt->mdt_squash_uid;
 119         ucred->uc_fsgid = info->mti_mdt->mdt_squash_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->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 (req->rq_auth_mapped_uid == INVALID_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 (req->rq_auth_mapped_uid == INVALID_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->uid;
 424         uc->uc_o_gid = uc->uc_gid = body->gid;
 425         uc->uc_o_fsuid = uc->uc_fsuid = body->fsuid;
 426         uc->uc_o_fsgid = uc->uc_fsgid = body->fsgid;
 427         uc->uc_suppgids[0] = body->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->capability & ~CFS_CAP_FS_MASK;
 451         else
 452                 uc->uc_cap = body->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)
 533 {
 534         const struct lov_ost_data_v1 *lod;
 535         int                           i;
 536         __u16                         count;
 537
 538         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 539
 540         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 541                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 542                le32_to_cpu(lmm->lmm_pattern));
 543         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 544                le32_to_cpu(lmm->lmm_stripe_size), count);
 545         if (count == LOV_ALL_STRIPES ||
 546             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 547                 return;
 548
 549         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 550         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 551                 struct ost_id   oi;
 552                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 553                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 554                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 555         }
 556 }
 557
 558 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 559 {
 560         const struct lmv_mds_md_v1 *lmm1;
 561         int                        i;
 562
 563         lmm1 = &lmv->lmv_md_v1;
 564         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 565                le32_to_cpu(lmm1->lmv_magic),
 566                le32_to_cpu(lmm1->lmv_master_mdt_index),
 567                le32_to_cpu(lmm1->lmv_stripe_count));
 568         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 569                 struct lu_fid fid;
 570
 571                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 572                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 573         }
 574 }
 575
 576 /* Shrink and/or grow reply buffers */
 577 int mdt_fix_reply(struct mdt_thread_info *info)
 578 {
 579         struct req_capsule *pill = info->mti_pill;
 580         struct mdt_body    *body;
 581         int                md_size, md_packed = 0;
 582         int                acl_size;
 583         int                rc = 0;
 584         ENTRY;
 585
 586         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 587         LASSERT(body != NULL);
 588
 589         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 590                 md_size = body->eadatasize;
 591         else
 592                 md_size = 0;
 593
 594         acl_size = body->aclsize;
 595
 596         /* this replay - not send info to client */
 597         if (info->mti_spec.no_create) {
 598                 md_size = 0;
 599                 acl_size = 0;
 600         }
 601
 602         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 603                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 604                         md_size, acl_size,
 605                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 606                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 607 /*
 608             &RMF_MDT_BODY,
 609             &RMF_MDT_MD,
 610             &RMF_ACL, or &RMF_LOGCOOKIES
 611 (optional)  &RMF_CAPA1,
 612 (optional)  &RMF_CAPA2,
 613 (optional)  something else
 614 */
 615
 616         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 617          * buffers before growing it */
 618         if (info->mti_big_lmm_used) {
 619                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 620                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 621                                                  RCL_SERVER);
 622                 LASSERT(md_packed > 0);
 623                 /* buffer must be allocated separately */
 624                 LASSERT(info->mti_attr.ma_lmm !=
 625                         req_capsule_server_get(pill, &RMF_MDT_MD));
 626                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 627                 /* free big lmm if md_size is not needed */
 628                 if (md_size == 0)
 629                         info->mti_big_lmm_used = 0;
 630         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 631                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 632         }
 633
 634         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 635                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 636         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 637                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 638                                    acl_size, RCL_SERVER);
 639
 640         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 641             !(body->valid & OBD_MD_FLMDSCAPA))
 642                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 643
 644         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 645             !(body->valid & OBD_MD_FLOSSCAPA))
 646                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 647
 648         /*
 649          * Some more field should be shrinked if needed.
 650          * This should be done by those who added fields to reply message.
 651          */
 652
 653         /* Grow MD buffer if needed finally */
 654         if (info->mti_big_lmm_used) {
 655                 void *lmm;
 656
 657                 LASSERT(md_size > md_packed);
 658                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 659                        md_size - md_packed);
 660                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 661                 if (rc) {
 662                         /* we can't answer with proper LOV EA, drop flags,
 663                          * the rc is also returned so this request is
 664                          * considered as failed */
 665                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 666                         /* don't return transno along with error */
 667                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 668                 } else {
 669                         /* now we need to pack right LOV/LMV EA */
 670                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 671                         if (info->mti_attr.ma_valid & MA_LOV) {
 672                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 673                                                              RCL_SERVER) ==
 674                                                 info->mti_attr.ma_lmm_size);
 675                                 memcpy(lmm, info->mti_attr.ma_lmm,
 676                                        info->mti_attr.ma_lmm_size);
 677                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 678                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 679                                                              RCL_SERVER) ==
 680                                                 info->mti_attr.ma_lmv_size);
 681                                 memcpy(lmm, info->mti_attr.ma_lmv,
 682                                        info->mti_attr.ma_lmv_size);
 683                         }
 684                 }
 685                 /* update mdt_max_mdsize so clients will be aware about that */
 686                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 687                         info->mti_mdt->mdt_max_mdsize =
 688                                                     info->mti_attr.ma_lmm_size;
 689                 info->mti_big_lmm_used = 0;
 690         }
 691         RETURN(rc);
 692 }
 693
 694
 695 /* if object is dying, pack the lov/llog data,
 696  * parameter info->mti_attr should be valid at this point! */
 697 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 698                            const struct md_attr *ma)
 699 {
 700         struct mdt_body       *repbody;
 701         const struct lu_attr *la = &ma->ma_attr;
 702         int rc;
 703         ENTRY;
 704
 705         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 706         LASSERT(repbody != NULL);
 707
 708         if (ma->ma_valid & MA_INODE)
 709                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 710
 711         if (ma->ma_valid & MA_LOV) {
 712                 CERROR("No need in LOV EA upon unlink\n");
 713                 dump_stack();
 714         }
 715         repbody->eadatasize = 0;
 716
 717         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 718                 repbody->aclsize = ma->ma_cookie_size;
 719                 repbody->valid |= OBD_MD_FLCOOKIE;
 720         }
 721
 722         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 723             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 724             repbody->valid & OBD_MD_FLEASIZE) {
 725                 struct lustre_capa *capa;
 726
 727                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 728                 LASSERT(capa);
 729                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 730                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 731                 if (rc)
 732                         RETURN(rc);
 733
 734                 repbody->valid |= OBD_MD_FLOSSCAPA;
 735         }
 736
 737         RETURN(0);
 738 }
 739
 740 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 741                                   struct md_attr *ma)
 742 {
 743         __u64 out;
 744
 745         out = 0;
 746         if (in & MDS_ATTR_MODE)
 747                 out |= LA_MODE;
 748         if (in & MDS_ATTR_UID)
 749                 out |= LA_UID;
 750         if (in & MDS_ATTR_GID)
 751                 out |= LA_GID;
 752         if (in & MDS_ATTR_SIZE)
 753                 out |= LA_SIZE;
 754         if (in & MDS_ATTR_BLOCKS)
 755                 out |= LA_BLOCKS;
 756         if (in & MDS_ATTR_ATIME_SET)
 757                 out |= LA_ATIME;
 758         if (in & MDS_ATTR_CTIME_SET)
 759                 out |= LA_CTIME;
 760         if (in & MDS_ATTR_MTIME_SET)
 761                 out |= LA_MTIME;
 762         if (in & MDS_ATTR_ATTR_FLAG)
 763                 out |= LA_FLAGS;
 764         if (in & MDS_ATTR_KILL_SUID)
 765                 out |= LA_KILL_SUID;
 766         if (in & MDS_ATTR_KILL_SGID)
 767                 out |= LA_KILL_SGID;
 768
 769         if (in & MDS_ATTR_FROM_OPEN)
 770                 rr->rr_flags |= MRF_OPEN_TRUNC;
 771         if (in & MDS_OPEN_OWNEROVERRIDE)
 772                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 773         if (in & MDS_ATTR_FORCE)
 774                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 775
 776         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 777                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 778                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 779                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 780                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 781                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 782         if (in != 0)
 783                 CERROR("Unknown attr bits: "LPX64"\n", in);
 784         return out;
 785 }
 786
 787 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 788                       const struct lu_fid *fid, struct lustre_capa *capa)
 789 {
 790         struct lu_capainfo *lci;
 791
 792         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 793         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 794             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 795                 return;
 796
 797         lci = lu_capainfo_get(info->mti_env);
 798         LASSERT(lci);
 799         lci->lci_fid[offset]  = *fid;
 800         lci->lci_capa[offset] = capa;
 801 }
 802
 803 #ifdef DEBUG_CAPA
 804 void mdt_dump_capainfo(struct mdt_thread_info *info)
 805 {
 806         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 807         int i;
 808
 809         if (lci == NULL)
 810                 return;
 811
 812         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 813                 if (lci->lci_capa[i] == NULL) {
 814                         CERROR("no capa for index %d "DFID"\n",
 815                                i, PFID(&lci->lci_fid[i]));
 816                         continue;
 817                 }
 818                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 819                         CERROR("bypass for index %d "DFID"\n",
 820                                i, PFID(&lci->lci_fid[i]));
 821                         continue;
 822                 }
 823                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 824         }
 825 }
 826 #endif /* DEBUG_CAPA */
 827
 828 /* unpacking */
 829
 830 int mdt_name_unpack(struct req_capsule *pill,
 831                     const struct req_msg_field *field,
 832                     struct lu_name *ln,
 833                     enum mdt_name_flags flags)
 834 {
 835         ln->ln_name = req_capsule_client_get(pill, field);
 836         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 837
 838         if (!lu_name_is_valid(ln)) {
 839                 ln->ln_name = NULL;
 840                 ln->ln_namelen = 0;
 841
 842                 return -EPROTO;
 843         }
 844
 845         if ((flags & MNF_FIX_ANON) &&
 846             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 847                 /* Newer (3.x) kernels use a name of "/" for the
 848                  * "anonymous" disconnected dentries from NFS
 849                  * filehandle conversion. See d_obtain_alias(). */
 850                 ln->ln_name = NULL;
 851                 ln->ln_namelen = 0;
 852         }
 853
 854         return 0;
 855 }
 856
 857 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 858 {
 859         struct lu_ucred         *uc  = mdt_ucred(info);
 860         struct md_attr          *ma = &info->mti_attr;
 861         struct lu_attr          *la = &ma->ma_attr;
 862         struct req_capsule      *pill = info->mti_pill;
 863         struct mdt_reint_record *rr = &info->mti_rr;
 864         struct mdt_rec_setattr  *rec;
 865         ENTRY;
 866
 867         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 868         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 869         if (rec == NULL)
 870                 RETURN(-EFAULT);
 871
 872         /* This prior initialization is needed for old_init_ucred_reint() */
 873         uc->uc_fsuid = rec->sa_fsuid;
 874         uc->uc_fsgid = rec->sa_fsgid;
 875         uc->uc_cap   = rec->sa_cap;
 876         uc->uc_suppgids[0] = rec->sa_suppgid;
 877         uc->uc_suppgids[1] = -1;
 878
 879         rr->rr_fid1 = &rec->sa_fid;
 880         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 881         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 882          *  the client does not have OBD_CONNECT_FULL20, convert it
 883          *  to LA_xTIME. LU-3036 */
 884         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 885                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 886                      (rec->sa_valid & MDS_ATTR_ATIME))
 887                         la->la_valid |= LA_ATIME;
 888                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 889                      (rec->sa_valid & MDS_ATTR_MTIME))
 890                         la->la_valid |= LA_MTIME;
 891                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 892                      (rec->sa_valid & MDS_ATTR_CTIME))
 893                         la->la_valid |= LA_CTIME;
 894         }
 895         la->la_mode  = rec->sa_mode;
 896         la->la_flags = rec->sa_attr_flags;
 897         la->la_uid   = rec->sa_uid;
 898         la->la_gid   = rec->sa_gid;
 899         la->la_size  = rec->sa_size;
 900         la->la_blocks = rec->sa_blocks;
 901         la->la_ctime = rec->sa_ctime;
 902         la->la_atime = rec->sa_atime;
 903         la->la_mtime = rec->sa_mtime;
 904         ma->ma_valid = MA_INODE;
 905
 906         if (rec->sa_bias & MDS_DATA_MODIFIED)
 907                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 908         else
 909                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 910
 911         if (rec->sa_bias & MDS_HSM_RELEASE)
 912                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 913         else
 914                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 915
 916         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 917                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 918                                  req_capsule_client_get(pill, &RMF_CAPA1));
 919
 920         RETURN(0);
 921 }
 922
 923 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 924 {
 925         struct req_capsule *pill = info->mti_pill;
 926         ENTRY;
 927
 928         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 929                 info->mti_ioepoch =
 930                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 931         else
 932                 info->mti_ioepoch = NULL;
 933         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 934 }
 935
 936 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 937         struct req_capsule      *pill = info->mti_pill;
 938
 939         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 940                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 941                 if (info->mti_dlm_req == NULL)
 942                         RETURN(-EFAULT);
 943         }
 944
 945         RETURN(0);
 946 }
 947
 948 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 949 {
 950         struct mdt_reint_record *rr = &info->mti_rr;
 951         struct md_attr          *ma = &info->mti_attr;
 952         struct req_capsule      *pill = info->mti_pill;
 953         int rc;
 954         ENTRY;
 955
 956         rc = mdt_setattr_unpack_rec(info);
 957         if (rc)
 958                 RETURN(rc);
 959
 960         /* Epoch may be absent */
 961         mdt_ioepoch_unpack(info);
 962
 963         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 964                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 965                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 966                                                         RCL_CLIENT);
 967                 if (rr->rr_eadatalen > 0) {
 968                         const struct lmv_user_md        *lum;
 969
 970                         lum = rr->rr_eadata;
 971                         /* Sigh ma_valid(from req) does not indicate whether
 972                          * it will set LOV/LMV EA, so we have to check magic */
 973                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 974                                 ma->ma_valid |= MA_LMV;
 975                                 ma->ma_lmv = (void *)rr->rr_eadata;
 976                                 ma->ma_lmv_size = rr->rr_eadatalen;
 977                         } else {
 978                                 ma->ma_valid |= MA_LOV;
 979                                 ma->ma_lmm = (void *)rr->rr_eadata;
 980                                 ma->ma_lmm_size = rr->rr_eadatalen;
 981                         }
 982                 }
 983         }
 984
 985         rc = mdt_dlmreq_unpack(info);
 986         RETURN(rc);
 987 }
 988
 989 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
 990 {
 991         struct md_attr          *ma = &info->mti_attr;
 992         struct req_capsule      *pill = info->mti_pill;
 993         ENTRY;
 994
 995         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
 996                 RETURN(0);
 997
 998         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
 999
1000         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1001             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1002                 RETURN(-EFAULT);
1003
1004         RETURN(0);
1005 }
1006
1007 int mdt_close_unpack(struct mdt_thread_info *info)
1008 {
1009         int rc;
1010         ENTRY;
1011
1012         rc = mdt_ioepoch_unpack(info);
1013         if (rc)
1014                 RETURN(rc);
1015
1016         rc = mdt_setattr_unpack_rec(info);
1017         if (rc)
1018                 RETURN(rc);
1019
1020         rc = mdt_hsm_release_unpack(info);
1021         if (rc)
1022                 RETURN(rc);
1023
1024         RETURN(mdt_init_ucred_reint(info));
1025 }
1026
1027 static int mdt_create_unpack(struct mdt_thread_info *info)
1028 {
1029         struct lu_ucred         *uc  = mdt_ucred(info);
1030         struct mdt_rec_create   *rec;
1031         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1032         struct mdt_reint_record *rr = &info->mti_rr;
1033         struct req_capsule      *pill = info->mti_pill;
1034         struct md_op_spec       *sp = &info->mti_spec;
1035         int rc;
1036         ENTRY;
1037
1038         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1039         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1040         if (rec == NULL)
1041                 RETURN(-EFAULT);
1042
1043         /* This prior initialization is needed for old_init_ucred_reint() */
1044         uc->uc_fsuid = rec->cr_fsuid;
1045         uc->uc_fsgid = rec->cr_fsgid;
1046         uc->uc_cap   = rec->cr_cap;
1047         uc->uc_suppgids[0] = rec->cr_suppgid1;
1048         uc->uc_suppgids[1] = -1;
1049         uc->uc_umask = rec->cr_umask;
1050
1051         rr->rr_fid1 = &rec->cr_fid1;
1052         rr->rr_fid2 = &rec->cr_fid2;
1053         attr->la_mode = rec->cr_mode;
1054         attr->la_rdev  = rec->cr_rdev;
1055         attr->la_uid   = rec->cr_fsuid;
1056         attr->la_gid   = rec->cr_fsgid;
1057         attr->la_ctime = rec->cr_time;
1058         attr->la_mtime = rec->cr_time;
1059         attr->la_atime = rec->cr_time;
1060         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1061                          LA_CTIME | LA_MTIME | LA_ATIME;
1062         memset(&sp->u, 0, sizeof(sp->u));
1063         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1064
1065         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1066                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1067                                  req_capsule_client_get(pill, &RMF_CAPA1));
1068         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1069
1070         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1071         if (rc < 0)
1072                 RETURN(rc);
1073
1074         if (S_ISLNK(attr->la_mode)) {
1075                 const char *tgt = NULL;
1076
1077                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1078                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1079                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1080                         sp->u.sp_symname = tgt;
1081                 }
1082                 if (tgt == NULL)
1083                         RETURN(-EFAULT);
1084         } else {
1085                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1086                 if (S_ISDIR(attr->la_mode) &&
1087                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1088                         sp->u.sp_ea.eadata =
1089                                 req_capsule_client_get(pill, &RMF_EADATA);
1090                         sp->u.sp_ea.eadatalen =
1091                                 req_capsule_get_size(pill, &RMF_EADATA,
1092                                                      RCL_CLIENT);
1093                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1094                 }
1095         }
1096
1097         rc = mdt_dlmreq_unpack(info);
1098         RETURN(rc);
1099 }
1100
1101 static int mdt_link_unpack(struct mdt_thread_info *info)
1102 {
1103         struct lu_ucred         *uc  = mdt_ucred(info);
1104         struct mdt_rec_link     *rec;
1105         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1106         struct mdt_reint_record *rr = &info->mti_rr;
1107         struct req_capsule      *pill = info->mti_pill;
1108         int rc;
1109         ENTRY;
1110
1111         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1112         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1113         if (rec == NULL)
1114                 RETURN(-EFAULT);
1115
1116         /* This prior initialization is needed for old_init_ucred_reint() */
1117         uc->uc_fsuid = rec->lk_fsuid;
1118         uc->uc_fsgid = rec->lk_fsgid;
1119         uc->uc_cap   = rec->lk_cap;
1120         uc->uc_suppgids[0] = rec->lk_suppgid1;
1121         uc->uc_suppgids[1] = rec->lk_suppgid2;
1122
1123         attr->la_uid = rec->lk_fsuid;
1124         attr->la_gid = rec->lk_fsgid;
1125         rr->rr_fid1 = &rec->lk_fid1;
1126         rr->rr_fid2 = &rec->lk_fid2;
1127         attr->la_ctime = rec->lk_time;
1128         attr->la_mtime = rec->lk_time;
1129         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1130
1131         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1132                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1133                                  req_capsule_client_get(pill, &RMF_CAPA1));
1134         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1135                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1136                                  req_capsule_client_get(pill, &RMF_CAPA2));
1137
1138         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1139         if (rc < 0)
1140                 RETURN(rc);
1141
1142         rc = mdt_dlmreq_unpack(info);
1143
1144         RETURN(rc);
1145 }
1146
1147 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1148 {
1149         struct lu_ucred         *uc  = mdt_ucred(info);
1150         struct mdt_rec_unlink   *rec;
1151         struct md_attr          *ma = &info->mti_attr;
1152         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1153         struct mdt_reint_record *rr = &info->mti_rr;
1154         struct req_capsule      *pill = info->mti_pill;
1155         int rc;
1156         ENTRY;
1157
1158         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1159         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1160         if (rec == NULL)
1161                 RETURN(-EFAULT);
1162
1163         /* This prior initialization is needed for old_init_ucred_reint() */
1164         uc->uc_fsuid = rec->ul_fsuid;
1165         uc->uc_fsgid = rec->ul_fsgid;
1166         uc->uc_cap   = rec->ul_cap;
1167         uc->uc_suppgids[0] = rec->ul_suppgid1;
1168         uc->uc_suppgids[1] = -1;
1169
1170         attr->la_uid = rec->ul_fsuid;
1171         attr->la_gid = rec->ul_fsgid;
1172         rr->rr_fid1 = &rec->ul_fid1;
1173         rr->rr_fid2 = &rec->ul_fid2;
1174         attr->la_ctime = rec->ul_time;
1175         attr->la_mtime = rec->ul_time;
1176         attr->la_mode  = rec->ul_mode;
1177         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1178
1179         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1180                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1181                                  req_capsule_client_get(pill, &RMF_CAPA1));
1182
1183         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1184         if (rc < 0)
1185                 RETURN(rc);
1186
1187         if (rec->ul_bias & MDS_VTX_BYPASS)
1188                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1189         else
1190                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1191
1192         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1193
1194         rc = mdt_dlmreq_unpack(info);
1195         RETURN(rc);
1196 }
1197
1198 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1199 {
1200         info->mti_spec.sp_rm_entry = 1;
1201         return mdt_unlink_unpack(info);
1202 }
1203
1204 static int mdt_rename_unpack(struct mdt_thread_info *info)
1205 {
1206         struct lu_ucred         *uc = mdt_ucred(info);
1207         struct mdt_rec_rename   *rec;
1208         struct md_attr          *ma = &info->mti_attr;
1209         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1210         struct mdt_reint_record *rr = &info->mti_rr;
1211         struct req_capsule      *pill = info->mti_pill;
1212         int rc;
1213         ENTRY;
1214
1215         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1216         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1217         if (rec == NULL)
1218                 RETURN(-EFAULT);
1219
1220         /* This prior initialization is needed for old_init_ucred_reint() */
1221         uc->uc_fsuid = rec->rn_fsuid;
1222         uc->uc_fsgid = rec->rn_fsgid;
1223         uc->uc_cap   = rec->rn_cap;
1224         uc->uc_suppgids[0] = rec->rn_suppgid1;
1225         uc->uc_suppgids[1] = rec->rn_suppgid2;
1226
1227         attr->la_uid = rec->rn_fsuid;
1228         attr->la_gid = rec->rn_fsgid;
1229         rr->rr_fid1 = &rec->rn_fid1;
1230         rr->rr_fid2 = &rec->rn_fid2;
1231         attr->la_ctime = rec->rn_time;
1232         attr->la_mtime = rec->rn_time;
1233         /* rename_tgt contains the mode already */
1234         attr->la_mode = rec->rn_mode;
1235         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1236
1237         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1238                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1239                                  req_capsule_client_get(pill, &RMF_CAPA1));
1240         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1241                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1242                                  req_capsule_client_get(pill, &RMF_CAPA2));
1243
1244         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1245         if (rc < 0)
1246                 RETURN(rc);
1247
1248         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1249         if (rc < 0)
1250                 RETURN(rc);
1251
1252         if (rec->rn_bias & MDS_VTX_BYPASS)
1253                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1254         else
1255                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1256
1257         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1258
1259         rc = mdt_dlmreq_unpack(info);
1260         RETURN(rc);
1261 }
1262
1263 /*
1264  * please see comment above LOV_MAGIC_V1_DEF
1265  */
1266 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1267 {
1268         struct mdt_reint_record *rr = &info->mti_rr;
1269         struct lov_user_md_v1   *v1;
1270
1271         v1 = (void *)rr->rr_eadata;
1272         LASSERT(v1);
1273
1274         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1275                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1276                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1277                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1278                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1279                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1280                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1281                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1282                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1283                 }
1284         }
1285 }
1286
1287 static int mdt_open_unpack(struct mdt_thread_info *info)
1288 {
1289         struct lu_ucred         *uc = mdt_ucred(info);
1290         struct mdt_rec_create   *rec;
1291         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1292         struct req_capsule      *pill = info->mti_pill;
1293         struct mdt_reint_record *rr   = &info->mti_rr;
1294         struct ptlrpc_request   *req  = mdt_info_req(info);
1295         struct md_op_spec       *sp   = &info->mti_spec;
1296         ENTRY;
1297
1298         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1299         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1300         if (rec == NULL)
1301                 RETURN(-EFAULT);
1302
1303         /* This prior initialization is needed for old_init_ucred_reint() */
1304         uc->uc_fsuid = rec->cr_fsuid;
1305         uc->uc_fsgid = rec->cr_fsgid;
1306         uc->uc_cap   = rec->cr_cap;
1307         uc->uc_suppgids[0] = rec->cr_suppgid1;
1308         uc->uc_suppgids[1] = rec->cr_suppgid2;
1309         uc->uc_umask = rec->cr_umask;
1310
1311         rr->rr_fid1   = &rec->cr_fid1;
1312         rr->rr_fid2   = &rec->cr_fid2;
1313         rr->rr_handle = &rec->cr_old_handle;
1314         attr->la_mode = rec->cr_mode;
1315         attr->la_rdev  = rec->cr_rdev;
1316         attr->la_uid   = rec->cr_fsuid;
1317         attr->la_gid   = rec->cr_fsgid;
1318         attr->la_ctime = rec->cr_time;
1319         attr->la_mtime = rec->cr_time;
1320         attr->la_atime = rec->cr_time;
1321         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1322                          LA_CTIME | LA_MTIME | LA_ATIME;
1323         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1324         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1325         /* Do not trigger ASSERTION if client miss to set such flags. */
1326         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1327                 RETURN(-EPROTO);
1328         info->mti_replayepoch = rec->cr_ioepoch;
1329
1330         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1331
1332         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1333                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1334                                  req_capsule_client_get(pill, &RMF_CAPA1));
1335         if (req_is_replay(req) &&
1336             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1337 #if 0
1338                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1339                                  req_capsule_client_get(pill, &RMF_CAPA2));
1340 #else
1341                 /*
1342                  * FIXME: capa in replay open request might have expired,
1343                  * bypass capa check. Security hole?
1344                  */
1345                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1346                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1347 #endif
1348         }
1349
1350         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1351
1352         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1353                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1354                                                         RCL_CLIENT);
1355                 if (rr->rr_eadatalen > 0) {
1356                         rr->rr_eadata = req_capsule_client_get(pill,
1357                                                                &RMF_EADATA);
1358                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1359                         sp->u.sp_ea.eadata = rr->rr_eadata;
1360                         sp->no_create = !!req_is_replay(req);
1361                         mdt_fix_lov_magic(info);
1362                 }
1363
1364                 /*
1365                  * Client default md_size may be 0 right after client start,
1366                  * until all osc are connected, set here just some reasonable
1367                  * value to prevent misbehavior.
1368                  */
1369                 if (rr->rr_eadatalen == 0 &&
1370                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1371                         rr->rr_eadatalen = MIN_MD_SIZE;
1372         }
1373
1374         RETURN(0);
1375 }
1376
1377 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1378 {
1379         struct mdt_reint_record *rr     = &info->mti_rr;
1380         struct lu_ucred         *uc     = mdt_ucred(info);
1381         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1382         struct req_capsule      *pill   = info->mti_pill;
1383         struct mdt_rec_setxattr *rec;
1384         int                      rc;
1385         ENTRY;
1386
1387
1388         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1389                          sizeof(struct mdt_rec_reint));
1390
1391         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1392         if (rec == NULL)
1393                 RETURN(-EFAULT);
1394
1395         /* This prior initialization is needed for old_init_ucred_reint() */
1396         uc->uc_fsuid  = rec->sx_fsuid;
1397         uc->uc_fsgid  = rec->sx_fsgid;
1398         uc->uc_cap    = rec->sx_cap;
1399         uc->uc_suppgids[0] = rec->sx_suppgid1;
1400         uc->uc_suppgids[1] = -1;
1401
1402         rr->rr_opcode = rec->sx_opcode;
1403         rr->rr_fid1   = &rec->sx_fid;
1404         attr->la_valid = rec->sx_valid;
1405         attr->la_ctime = rec->sx_time;
1406         attr->la_size = rec->sx_size;
1407         attr->la_flags = rec->sx_flags;
1408
1409         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1410                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1411                                  req_capsule_client_get(pill, &RMF_CAPA1));
1412         else
1413                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1414
1415         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1416         if (rc < 0)
1417                 RETURN(rc);
1418
1419         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1420                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1421                                                         RCL_CLIENT);
1422                 if (rr->rr_eadatalen > 0) {
1423                         rr->rr_eadata = req_capsule_client_get(pill,
1424                                                                &RMF_EADATA);
1425                         if (rr->rr_eadata == NULL)
1426                                 RETURN(-EFAULT);
1427                 } else {
1428                         rr->rr_eadata = NULL;
1429                 }
1430         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1431                 CDEBUG(D_INFO, "no xattr data supplied\n");
1432                 RETURN(-EFAULT);
1433         }
1434
1435         if (mdt_dlmreq_unpack(info) < 0)
1436                 RETURN(-EPROTO);
1437
1438         RETURN(0);
1439 }
1440
1441
1442 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1443
1444 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1445         [REINT_SETATTR]  = mdt_setattr_unpack,
1446         [REINT_CREATE]   = mdt_create_unpack,
1447         [REINT_LINK]     = mdt_link_unpack,
1448         [REINT_UNLINK]   = mdt_unlink_unpack,
1449         [REINT_RENAME]   = mdt_rename_unpack,
1450         [REINT_OPEN]     = mdt_open_unpack,
1451         [REINT_SETXATTR] = mdt_setxattr_unpack,
1452         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1453 };
1454
1455 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1456 {
1457         int rc;
1458         ENTRY;
1459
1460         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1461         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1462                 info->mti_rr.rr_opcode = op;
1463                 rc = mdt_reint_unpackers[op](info);
1464         } else {
1465                 CERROR("Unexpected opcode %d\n", op);
1466                 rc = -EFAULT;
1467         }
1468         RETURN(rc);
1469 }