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, 2012, Whamcloud, Inc.
  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 md_ucred   *uc  = mdt_ucred(info);
  64         struct mdt_device *mdt = info->mti_mdt;
  65
  66         if (uc->mu_valid != UCRED_INIT) {
  67                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
  68                 if (uc->mu_ginfo) {
  69                         cfs_put_group_info(uc->mu_ginfo);
  70                         uc->mu_ginfo = NULL;
  71                 }
  72                 if (uc->mu_identity) {
  73                         mdt_identity_put(mdt->mdt_identity_cache,
  74                                          uc->mu_identity);
  75                         uc->mu_identity = NULL;
  76                 }
  77                 uc->mu_valid = UCRED_INIT;
  78         }
  79 }
  80
  81 static int match_nosquash_list(cfs_rw_semaphore_t *sem,
  82                                cfs_list_t *nidlist,
  83                                lnet_nid_t peernid)
  84 {
  85         int rc;
  86         ENTRY;
  87         cfs_down_read(sem);
  88         rc = cfs_match_nid(peernid, nidlist);
  89         cfs_up_read(sem);
  90         RETURN(rc);
  91 }
  92
  93 /* root_squash for inter-MDS operations */
  94 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  95 {
  96         struct md_ucred *ucred = mdt_ucred(info);
  97         ENTRY;
  98
  99         if (!info->mti_mdt->mdt_squash_uid || ucred->mu_fsuid)
 100                 RETURN(0);
 101
 102         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 103                                 &info->mti_mdt->mdt_nosquash_nids,
 104                                 peernid)) {
 105                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 106                        libcfs_nid2str(peernid));
 107                 RETURN(0);
 108         }
 109
 110         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 111                libcfs_nid2str(peernid),
 112                ucred->mu_fsuid, ucred->mu_fsgid, ucred->mu_cap,
 113                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 114                0);
 115
 116         ucred->mu_fsuid = info->mti_mdt->mdt_squash_uid;
 117         ucred->mu_fsgid = info->mti_mdt->mdt_squash_gid;
 118         ucred->mu_cap = 0;
 119         ucred->mu_suppgids[0] = -1;
 120         ucred->mu_suppgids[1] = -1;
 121
 122         RETURN(0);
 123 }
 124
 125 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 126                           void *buf)
 127 {
 128         struct ptlrpc_request   *req = mdt_info_req(info);
 129         struct mdt_device       *mdt = info->mti_mdt;
 130         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 131         struct md_ucred         *ucred = mdt_ucred(info);
 132         lnet_nid_t               peernid = req->rq_peer.nid;
 133         __u32                    perm = 0;
 134         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 135         int                      setuid;
 136         int                      setgid;
 137         int                      rc = 0;
 138
 139         ENTRY;
 140
 141         LASSERT(req->rq_auth_gss);
 142         LASSERT(!req->rq_auth_usr_mdt);
 143         LASSERT(req->rq_user_desc);
 144
 145         ucred->mu_valid = UCRED_INVALID;
 146
 147         ucred->mu_o_uid   = pud->pud_uid;
 148         ucred->mu_o_gid   = pud->pud_gid;
 149         ucred->mu_o_fsuid = pud->pud_fsuid;
 150         ucred->mu_o_fsgid = pud->pud_fsgid;
 151
 152         if (type == BODY_INIT) {
 153                 struct mdt_body *body = (struct mdt_body *)buf;
 154
 155                 ucred->mu_suppgids[0] = body->suppgid;
 156                 ucred->mu_suppgids[1] = -1;
 157         }
 158
 159         /* sanity check: we expect the uid which client claimed is true */
 160         if (remote) {
 161                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 162                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 163                         RETURN(-EACCES);
 164                 }
 165
 166                 if (ptlrpc_user_desc_do_idmap(req, pud))
 167                         RETURN(-EACCES);
 168
 169                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 170                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 171                                "while client claims %u:%u/%u:%u\n",
 172                                libcfs_nid2str(peernid), req->rq_auth_uid,
 173                                req->rq_auth_mapped_uid,
 174                                pud->pud_uid, pud->pud_gid,
 175                                pud->pud_fsuid, pud->pud_fsgid);
 176                         RETURN(-EACCES);
 177                 }
 178         } else {
 179                 if (req->rq_auth_uid != pud->pud_uid) {
 180                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 181                                "while client claims %u:%u/%u:%u\n",
 182                                libcfs_nid2str(peernid), req->rq_auth_uid,
 183                                pud->pud_uid, pud->pud_gid,
 184                                pud->pud_fsuid, pud->pud_fsgid);
 185                         RETURN(-EACCES);
 186                 }
 187         }
 188
 189         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 190                 if (remote) {
 191                         CDEBUG(D_SEC, "remote client must run with identity_get "
 192                                "enabled!\n");
 193                         RETURN(-EACCES);
 194                 } else {
 195                         ucred->mu_identity = NULL;
 196                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 197                                CFS_SETGRP_PERM;
 198                 }
 199         } else {
 200                 struct md_identity *identity;
 201
 202                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 203                                             pud->pud_uid);
 204                 if (IS_ERR(identity)) {
 205                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 206                                      !remote)) {
 207                                 ucred->mu_identity = NULL;
 208                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 209                                        CFS_SETGRP_PERM;
 210                         } else {
 211                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 212                                        pud->pud_uid);
 213                                 RETURN(-EACCES);
 214                         }
 215                 } else {
 216                         ucred->mu_identity = identity;
 217                         perm = mdt_identity_get_perm(ucred->mu_identity,
 218                                                      remote, peernid);
 219                 }
 220         }
 221
 222         /* find out the setuid/setgid attempt */
 223         setuid = (pud->pud_uid != pud->pud_fsuid);
 224         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 225                   (ucred->mu_identity &&
 226                   (pud->pud_gid != ucred->mu_identity->mi_gid)));
 227
 228         /* check permission of setuid */
 229         if (setuid && !(perm & CFS_SETUID_PERM)) {
 230                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 231                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 232                 GOTO(out, rc = -EACCES);
 233         }
 234
 235         /* check permission of setgid */
 236         if (setgid && !(perm & CFS_SETGID_PERM)) {
 237                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 238                        "from %s\n", pud->pud_uid, pud->pud_gid,
 239                        pud->pud_fsuid, pud->pud_fsgid,
 240                        ucred->mu_identity->mi_gid, libcfs_nid2str(peernid));
 241                 GOTO(out, rc = -EACCES);
 242         }
 243
 244         /*
 245          * NB: remote client not allowed to setgroups anyway.
 246          */
 247         if (!remote && perm & CFS_SETGRP_PERM) {
 248                 if (pud->pud_ngroups) {
 249                         /* setgroups for local client */
 250                         ucred->mu_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 251                         if (!ucred->mu_ginfo) {
 252                                 CERROR("failed to alloc %d groups\n",
 253                                        pud->pud_ngroups);
 254                                 GOTO(out, rc = -ENOMEM);
 255                         }
 256
 257                         lustre_groups_from_list(ucred->mu_ginfo,
 258                                                 pud->pud_groups);
 259                         lustre_groups_sort(ucred->mu_ginfo);
 260                 } else {
 261                         ucred->mu_ginfo = NULL;
 262                 }
 263         } else {
 264                 ucred->mu_suppgids[0] = -1;
 265                 ucred->mu_suppgids[1] = -1;
 266                 ucred->mu_ginfo = NULL;
 267         }
 268
 269         ucred->mu_uid   = pud->pud_uid;
 270         ucred->mu_gid   = pud->pud_gid;
 271         ucred->mu_fsuid = pud->pud_fsuid;
 272         ucred->mu_fsgid = pud->pud_fsgid;
 273
 274         /* process root_squash here. */
 275         mdt_root_squash(info, peernid);
 276
 277         /* remove fs privilege for non-root user. */
 278         if (ucred->mu_fsuid)
 279                 ucred->mu_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 280         else
 281                 ucred->mu_cap = pud->pud_cap;
 282         if (remote && !(perm & CFS_RMTOWN_PERM))
 283                 ucred->mu_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 284                                    CFS_CAP_CHOWN_MASK);
 285         ucred->mu_valid = UCRED_NEW;
 286
 287         EXIT;
 288
 289 out:
 290         if (rc) {
 291                 if (ucred->mu_ginfo) {
 292                         cfs_put_group_info(ucred->mu_ginfo);
 293                         ucred->mu_ginfo = NULL;
 294                 }
 295                 if (ucred->mu_identity) {
 296                         mdt_identity_put(mdt->mdt_identity_cache,
 297                                          ucred->mu_identity);
 298                         ucred->mu_identity = NULL;
 299                 }
 300         }
 301
 302         return rc;
 303 }
 304
 305 int mdt_check_ucred(struct mdt_thread_info *info)
 306 {
 307         struct ptlrpc_request   *req = mdt_info_req(info);
 308         struct mdt_device       *mdt = info->mti_mdt;
 309         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 310         struct md_ucred         *ucred = mdt_ucred(info);
 311         struct md_identity      *identity = NULL;
 312         lnet_nid_t               peernid = req->rq_peer.nid;
 313         __u32                    perm = 0;
 314         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 315         int                      setuid;
 316         int                      setgid;
 317         int                      rc = 0;
 318
 319         ENTRY;
 320
 321         if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
 322                 RETURN(0);
 323
 324         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 325                 RETURN(0);
 326
 327         /* sanity check: if we use strong authentication, we expect the
 328          * uid which client claimed is true */
 329         if (remote) {
 330                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 331                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 332                         RETURN(-EACCES);
 333                 }
 334
 335                 if (ptlrpc_user_desc_do_idmap(req, pud))
 336                         RETURN(-EACCES);
 337
 338                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 339                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 340                                "while client claims %u:%u/%u:%u\n",
 341                                libcfs_nid2str(peernid), req->rq_auth_uid,
 342                                req->rq_auth_mapped_uid,
 343                                pud->pud_uid, pud->pud_gid,
 344                                pud->pud_fsuid, pud->pud_fsgid);
 345                         RETURN(-EACCES);
 346                 }
 347         } else {
 348                 if (req->rq_auth_uid != pud->pud_uid) {
 349                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 350                                "while client claims %u:%u/%u:%u\n",
 351                                libcfs_nid2str(peernid), req->rq_auth_uid,
 352                                pud->pud_uid, pud->pud_gid,
 353                                pud->pud_fsuid, pud->pud_fsgid);
 354                         RETURN(-EACCES);
 355                 }
 356         }
 357
 358         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 359                 if (remote) {
 360                         CDEBUG(D_SEC, "remote client must run with identity_get "
 361                                "enabled!\n");
 362                         RETURN(-EACCES);
 363                 }
 364                 RETURN(0);
 365         }
 366
 367         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 368         if (IS_ERR(identity)) {
 369                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 370                              !remote)) {
 371                         RETURN(0);
 372                 } else {
 373                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 374                                pud->pud_uid);
 375                         RETURN(-EACCES);
 376                }
 377         }
 378
 379         perm = mdt_identity_get_perm(identity, remote, peernid);
 380         /* find out the setuid/setgid attempt */
 381         setuid = (pud->pud_uid != pud->pud_fsuid);
 382         setgid = (pud->pud_gid != pud->pud_fsgid ||
 383                   pud->pud_gid != identity->mi_gid);
 384
 385         /* check permission of setuid */
 386         if (setuid && !(perm & CFS_SETUID_PERM)) {
 387                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 388                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 389                 GOTO(out, rc = -EACCES);
 390         }
 391
 392         /* check permission of setgid */
 393         if (setgid && !(perm & CFS_SETGID_PERM)) {
 394                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 395                        "from %s\n", pud->pud_uid, pud->pud_gid,
 396                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 397                        libcfs_nid2str(peernid));
 398                 GOTO(out, rc = -EACCES);
 399         }
 400
 401         EXIT;
 402
 403 out:
 404         mdt_identity_put(mdt->mdt_identity_cache, identity);
 405         return rc;
 406 }
 407
 408 static int old_init_ucred(struct mdt_thread_info *info,
 409                           struct mdt_body *body)
 410 {
 411         struct md_ucred *uc = mdt_ucred(info);
 412         struct mdt_device  *mdt = info->mti_mdt;
 413         struct md_identity *identity = NULL;
 414
 415         ENTRY;
 416
 417         uc->mu_valid = UCRED_INVALID;
 418         uc->mu_o_uid = uc->mu_uid = body->uid;
 419         uc->mu_o_gid = uc->mu_gid = body->gid;
 420         uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
 421         uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
 422         uc->mu_suppgids[0] = body->suppgid;
 423         uc->mu_suppgids[1] = -1;
 424         uc->mu_ginfo = NULL;
 425         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 426                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 427                                             uc->mu_fsuid);
 428                 if (IS_ERR(identity)) {
 429                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 430                                 identity = NULL;
 431                         } else {
 432                                 CDEBUG(D_SEC, "Deny access without identity: "
 433                                        "uid %u\n", uc->mu_fsuid);
 434                                 RETURN(-EACCES);
 435                         }
 436                 }
 437         }
 438         uc->mu_identity = identity;
 439
 440         /* process root_squash here. */
 441         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 442
 443         /* remove fs privilege for non-root user. */
 444         if (uc->mu_fsuid)
 445                 uc->mu_cap = body->capability & ~CFS_CAP_FS_MASK;
 446         else
 447                 uc->mu_cap = body->capability;
 448         uc->mu_valid = UCRED_OLD;
 449
 450         RETURN(0);
 451 }
 452
 453 static int old_init_ucred_reint(struct mdt_thread_info *info)
 454 {
 455         struct md_ucred *uc = mdt_ucred(info);
 456         struct mdt_device  *mdt = info->mti_mdt;
 457         struct md_identity *identity = NULL;
 458
 459         ENTRY;
 460
 461         uc->mu_valid = UCRED_INVALID;
 462         uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
 463         uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
 464         uc->mu_ginfo = NULL;
 465         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 466                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 467                                             uc->mu_fsuid);
 468                 if (IS_ERR(identity)) {
 469                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 470                                 identity = NULL;
 471                         } else {
 472                                 CDEBUG(D_SEC, "Deny access without identity: "
 473                                        "uid %u\n", uc->mu_fsuid);
 474                                 RETURN(-EACCES);
 475                         }
 476                 }
 477         }
 478         uc->mu_identity = identity;
 479
 480         /* process root_squash here. */
 481         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 482
 483         /* remove fs privilege for non-root user. */
 484         if (uc->mu_fsuid)
 485                 uc->mu_cap &= ~CFS_CAP_FS_MASK;
 486         uc->mu_valid = UCRED_OLD;
 487
 488         RETURN(0);
 489 }
 490
 491 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 492 {
 493         struct ptlrpc_request *req = mdt_info_req(info);
 494         struct md_ucred       *uc  = mdt_ucred(info);
 495
 496         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 497                 return 0;
 498
 499         mdt_exit_ucred(info);
 500
 501         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 502                 return old_init_ucred(info, body);
 503         else
 504                 return new_init_ucred(info, BODY_INIT, body);
 505 }
 506
 507 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 508 {
 509         struct ptlrpc_request *req = mdt_info_req(info);
 510         struct md_ucred       *uc  = mdt_ucred(info);
 511
 512         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 513                 return 0;
 514
 515         mdt_exit_ucred(info);
 516
 517         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 518                 return old_init_ucred_reint(info);
 519         else
 520                 return new_init_ucred(info, REC_INIT, NULL);
 521 }
 522
 523 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 524 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 525 {
 526         const struct lov_ost_data_v1 *lod;
 527         int                           i;
 528         __u16                         count;
 529
 530         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 531
 532         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
 533                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 534                le32_to_cpu(lmm->lmm_pattern));
 535         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 536                le32_to_cpu(lmm->lmm_stripe_size), count);
 537         if (count == LOV_ALL_STRIPES)
 538                 return;
 539         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 540         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++)
 541                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 542                        i, le32_to_cpu(lod->l_ost_idx),
 543                        le64_to_cpu(lod->l_object_seq),
 544                        le64_to_cpu(lod->l_object_id));
 545 }
 546
 547 /* Shrink and/or grow reply buffers */
 548 int mdt_fix_reply(struct mdt_thread_info *info)
 549 {
 550         struct req_capsule *pill = info->mti_pill;
 551         struct mdt_body    *body;
 552         int                md_size, md_packed = 0;
 553         int                acl_size;
 554         int                rc = 0;
 555         ENTRY;
 556
 557         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 558         LASSERT(body != NULL);
 559
 560         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 561                 md_size = body->eadatasize;
 562         else
 563                 md_size = 0;
 564
 565         acl_size = body->aclsize;
 566
 567         /* this replay - not send info to client */
 568         if (info->mti_spec.no_create == 1) {
 569                 md_size = 0;
 570                 acl_size = 0;
 571         }
 572
 573         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 574                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 575                         md_size, acl_size,
 576                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 577                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 578 /*
 579             &RMF_MDT_BODY,
 580             &RMF_MDT_MD,
 581             &RMF_ACL, or &RMF_LOGCOOKIES
 582 (optional)  &RMF_CAPA1,
 583 (optional)  &RMF_CAPA2,
 584 (optional)  something else
 585 */
 586
 587         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 588          * buffers before growing it */
 589         if (info->mti_big_lmm_used) {
 590                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 591                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 592                                                  RCL_SERVER);
 593                 LASSERT(md_packed > 0);
 594                 /* buffer must be allocated separately */
 595                 LASSERT(info->mti_attr.ma_lmm !=
 596                         req_capsule_server_get(pill, &RMF_MDT_MD));
 597                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 598                 /* free big lmm if md_size is not needed */
 599                 if (md_size == 0)
 600                         info->mti_big_lmm_used = 0;
 601         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 602                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 603         }
 604
 605         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 606                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 607         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 608                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 609                                    acl_size, RCL_SERVER);
 610
 611         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 612             !(body->valid & OBD_MD_FLMDSCAPA))
 613                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 614
 615         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 616             !(body->valid & OBD_MD_FLOSSCAPA))
 617                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 618
 619         /*
 620          * Some more field should be shrinked if needed.
 621          * This should be done by those who added fields to reply message.
 622          */
 623
 624         /* Grow MD buffer if needed finally */
 625         if (info->mti_big_lmm_used) {
 626                 void *lmm;
 627
 628                 LASSERT(md_size > md_packed);
 629                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 630                        md_size - md_packed);
 631                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 632                 if (rc) {
 633                         /* we can't answer with proper LOV EA, drop flags,
 634                          * the rc is also returned so this request is
 635                          * considered as failed */
 636                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 637                         /* don't return transno along with error */
 638                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 639                 } else {
 640                         /* now we need to pack right LOV EA */
 641                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 642                         LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 643                                                      RCL_SERVER) ==
 644                                 info->mti_attr.ma_lmm_size);
 645                         memcpy(lmm, info->mti_attr.ma_lmm,
 646                                info->mti_attr.ma_lmm_size);
 647                 }
 648                 /* update mdt_max_mdsize so clients will be aware about that */
 649                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 650                         info->mti_mdt->mdt_max_mdsize =
 651                                                     info->mti_attr.ma_lmm_size;
 652                 info->mti_big_lmm_used = 0;
 653         }
 654         RETURN(rc);
 655 }
 656
 657
 658 /* if object is dying, pack the lov/llog data,
 659  * parameter info->mti_attr should be valid at this point! */
 660 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 661                            const struct md_attr *ma)
 662 {
 663         struct mdt_body       *repbody;
 664         const struct lu_attr *la = &ma->ma_attr;
 665         int rc;
 666         ENTRY;
 667
 668         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 669         LASSERT(repbody != NULL);
 670
 671         if (ma->ma_valid & MA_INODE)
 672                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 673
 674         if (ma->ma_valid & MA_LOV) {
 675                 __u32 mode;
 676
 677                 if (mdt_object_exists(mo) < 0)
 678                         /* If it is a remote object, and we do not retrieve
 679                          * EA back unlink reg file*/
 680                         mode = S_IFREG;
 681                 else
 682                         mode = lu_object_attr(&mo->mot_obj.mo_lu);
 683
 684                 LASSERT(ma->ma_lmm_size);
 685                 mdt_dump_lmm(D_INFO, ma->ma_lmm);
 686                 repbody->eadatasize = ma->ma_lmm_size;
 687                 if (S_ISREG(mode))
 688                         repbody->valid |= OBD_MD_FLEASIZE;
 689                 else if (S_ISDIR(mode))
 690                         repbody->valid |= OBD_MD_FLDIREA;
 691                 else
 692                         LBUG();
 693         }
 694
 695         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 696                 repbody->aclsize = ma->ma_cookie_size;
 697                 repbody->valid |= OBD_MD_FLCOOKIE;
 698         }
 699
 700         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 701             info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
 702             repbody->valid & OBD_MD_FLEASIZE) {
 703                 struct lustre_capa *capa;
 704
 705                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 706                 LASSERT(capa);
 707                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 708                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 709                 if (rc)
 710                         RETURN(rc);
 711
 712                 repbody->valid |= OBD_MD_FLOSSCAPA;
 713         }
 714
 715         RETURN(0);
 716 }
 717
 718 static inline unsigned int attr_unpack(__u64 sa_valid) {
 719         unsigned int ia_valid = 0;
 720
 721         if (sa_valid & MDS_ATTR_MODE)
 722                 ia_valid |= ATTR_MODE;
 723         if (sa_valid & MDS_ATTR_UID)
 724                 ia_valid |= ATTR_UID;
 725         if (sa_valid & MDS_ATTR_GID)
 726                 ia_valid |= ATTR_GID;
 727         if (sa_valid & MDS_ATTR_SIZE)
 728                 ia_valid |= ATTR_SIZE;
 729         if (sa_valid & MDS_ATTR_ATIME)
 730                 ia_valid |= ATTR_ATIME;
 731         if (sa_valid & MDS_ATTR_MTIME)
 732                 ia_valid |= ATTR_MTIME;
 733         if (sa_valid & MDS_ATTR_CTIME)
 734                 ia_valid |= ATTR_CTIME;
 735         if (sa_valid & MDS_ATTR_ATIME_SET)
 736                 ia_valid |= ATTR_ATIME_SET;
 737         if (sa_valid & MDS_ATTR_MTIME_SET)
 738                 ia_valid |= ATTR_MTIME_SET;
 739         if (sa_valid & MDS_ATTR_FORCE)
 740                 ia_valid |= ATTR_FORCE;
 741         if (sa_valid & MDS_ATTR_ATTR_FLAG)
 742                 ia_valid |= ATTR_ATTR_FLAG;
 743         if (sa_valid & MDS_ATTR_KILL_SUID)
 744                 ia_valid |=  ATTR_KILL_SUID;
 745         if (sa_valid & MDS_ATTR_KILL_SGID)
 746                 ia_valid |= ATTR_KILL_SGID;
 747         if (sa_valid & MDS_ATTR_CTIME_SET)
 748                 ia_valid |= ATTR_CTIME_SET;
 749         if (sa_valid & MDS_ATTR_FROM_OPEN)
 750                 ia_valid |= ATTR_FROM_OPEN;
 751         if (sa_valid & MDS_ATTR_BLOCKS)
 752                 ia_valid |= ATTR_BLOCKS;
 753         if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
 754                 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
 755         return ia_valid;
 756 }
 757
 758 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 759                                   struct md_attr *ma)
 760 {
 761         __u64 out;
 762
 763         out = 0;
 764         if (in & ATTR_MODE)
 765                 out |= LA_MODE;
 766         if (in & ATTR_UID)
 767                 out |= LA_UID;
 768         if (in & ATTR_GID)
 769                 out |= LA_GID;
 770         if (in & ATTR_SIZE)
 771                 out |= LA_SIZE;
 772         if (in & ATTR_BLOCKS)
 773                 out |= LA_BLOCKS;
 774
 775         if (in & ATTR_FROM_OPEN)
 776                 rr->rr_flags |= MRF_OPEN_TRUNC;
 777
 778         if (in & ATTR_ATIME_SET)
 779                 out |= LA_ATIME;
 780
 781         if (in & ATTR_CTIME_SET)
 782                 out |= LA_CTIME;
 783
 784         if (in & ATTR_MTIME_SET)
 785                 out |= LA_MTIME;
 786
 787         if (in & ATTR_ATTR_FLAG)
 788                 out |= LA_FLAGS;
 789
 790         if (in & ATTR_KILL_SUID)
 791                 out |= LA_KILL_SUID;
 792
 793         if (in & ATTR_KILL_SGID)
 794                 out |= LA_KILL_SGID;
 795
 796         if (in & MDS_OPEN_OWNEROVERRIDE)
 797                 ma->ma_attr_flags |= MDS_OPEN_OWNEROVERRIDE;
 798
 799         if (in & ATTR_FORCE)
 800                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 801
 802         /*XXX need ATTR_RAW?*/
 803         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 804                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 805                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 806                 ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
 807                 ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
 808         if (in != 0)
 809                 CERROR("Unknown attr bits: "LPX64"\n", in);
 810         return out;
 811 }
 812 /* unpacking */
 813
 814 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 815 {
 816         struct md_ucred         *uc  = mdt_ucred(info);
 817         struct md_attr          *ma = &info->mti_attr;
 818         struct lu_attr          *la = &ma->ma_attr;
 819         struct req_capsule      *pill = info->mti_pill;
 820         struct mdt_reint_record *rr = &info->mti_rr;
 821         struct mdt_rec_setattr  *rec;
 822         ENTRY;
 823
 824         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 825         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 826         if (rec == NULL)
 827                 RETURN(-EFAULT);
 828
 829         uc->mu_fsuid = rec->sa_fsuid;
 830         uc->mu_fsgid = rec->sa_fsgid;
 831         uc->mu_cap   = rec->sa_cap;
 832         uc->mu_suppgids[0] = rec->sa_suppgid;
 833         uc->mu_suppgids[1] = -1;
 834
 835         rr->rr_fid1 = &rec->sa_fid;
 836         la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
 837         la->la_mode  = rec->sa_mode;
 838         la->la_flags = rec->sa_attr_flags;
 839         la->la_uid   = rec->sa_uid;
 840         la->la_gid   = rec->sa_gid;
 841         la->la_size  = rec->sa_size;
 842         la->la_blocks = rec->sa_blocks;
 843         la->la_ctime = rec->sa_ctime;
 844         la->la_atime = rec->sa_atime;
 845         la->la_mtime = rec->sa_mtime;
 846         ma->ma_valid = MA_INODE;
 847
 848         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 849                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 850                                  req_capsule_client_get(pill, &RMF_CAPA1));
 851
 852         RETURN(0);
 853 }
 854
 855 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 856 {
 857         struct req_capsule *pill = info->mti_pill;
 858         ENTRY;
 859
 860         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 861                 info->mti_ioepoch =
 862                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 863         else
 864                 info->mti_ioepoch = NULL;
 865         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 866 }
 867
 868 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 869         struct req_capsule      *pill = info->mti_pill;
 870
 871         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 872                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 873                 if (info->mti_dlm_req == NULL)
 874                         RETURN(-EFAULT);
 875         }
 876
 877         RETURN(0);
 878 }
 879
 880 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 881 {
 882         struct mdt_reint_record *rr = &info->mti_rr;
 883         struct md_attr          *ma = &info->mti_attr;
 884         struct req_capsule      *pill = info->mti_pill;
 885         int rc;
 886         ENTRY;
 887
 888         rc = mdt_setattr_unpack_rec(info);
 889         if (rc)
 890                 RETURN(rc);
 891
 892         /* Epoch may be absent */
 893         mdt_ioepoch_unpack(info);
 894
 895         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 896                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 897                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 898                                                         RCL_CLIENT);
 899                 ma->ma_lmm_size = rr->rr_eadatalen;
 900                 if (ma->ma_lmm_size > 0) {
 901                         ma->ma_lmm = (void *)rr->rr_eadata;
 902                         ma->ma_valid |= MA_LOV;
 903                 }
 904         }
 905
 906         ma->ma_cookie_size = req_capsule_get_size(pill, &RMF_LOGCOOKIES,
 907                                                   RCL_CLIENT);
 908         if (ma->ma_cookie_size) {
 909                 ma->ma_cookie = req_capsule_client_get(pill, &RMF_LOGCOOKIES);
 910                 ma->ma_valid |= MA_COOKIE;
 911         }
 912
 913         rc = mdt_dlmreq_unpack(info);
 914         RETURN(rc);
 915 }
 916
 917 int mdt_close_unpack(struct mdt_thread_info *info)
 918 {
 919         int rc;
 920         ENTRY;
 921
 922         rc = mdt_ioepoch_unpack(info);
 923         if (rc)
 924                 RETURN(rc);
 925
 926         RETURN(mdt_setattr_unpack_rec(info));
 927 }
 928
 929 static int mdt_create_unpack(struct mdt_thread_info *info)
 930 {
 931         struct md_ucred         *uc  = mdt_ucred(info);
 932         struct mdt_rec_create   *rec;
 933         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 934         struct mdt_reint_record *rr = &info->mti_rr;
 935         struct req_capsule      *pill = info->mti_pill;
 936         struct md_op_spec       *sp = &info->mti_spec;
 937         int rc;
 938         ENTRY;
 939
 940         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 941         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 942         if (rec == NULL)
 943                 RETURN(-EFAULT);
 944
 945         uc->mu_fsuid = rec->cr_fsuid;
 946         uc->mu_fsgid = rec->cr_fsgid;
 947         uc->mu_cap   = rec->cr_cap;
 948         uc->mu_suppgids[0] = rec->cr_suppgid1;
 949         uc->mu_suppgids[1] = -1;
 950
 951         rr->rr_fid1 = &rec->cr_fid1;
 952         rr->rr_fid2 = &rec->cr_fid2;
 953         attr->la_mode = rec->cr_mode;
 954         attr->la_rdev  = rec->cr_rdev;
 955         attr->la_uid   = rec->cr_fsuid;
 956         attr->la_gid   = rec->cr_fsgid;
 957         attr->la_ctime = rec->cr_time;
 958         attr->la_mtime = rec->cr_time;
 959         attr->la_atime = rec->cr_time;
 960         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
 961                          LA_CTIME | LA_MTIME | LA_ATIME;
 962         memset(&sp->u, 0, sizeof(sp->u));
 963         sp->sp_cr_flags = get_mrc_cr_flags(rec);
 964         sp->sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
 965         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
 966
 967         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 968                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 969                                  req_capsule_client_get(pill, &RMF_CAPA1));
 970         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 971
 972         if (!info->mti_cross_ref) {
 973                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 974                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
 975                                                       RCL_CLIENT) - 1;
 976                 LASSERT(rr->rr_name && rr->rr_namelen > 0);
 977         } else {
 978                 rr->rr_name = NULL;
 979                 rr->rr_namelen = 0;
 980         }
 981
 982 #ifdef CONFIG_FS_POSIX_ACL
 983         if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
 984                 if (S_ISDIR(attr->la_mode))
 985                         sp->u.sp_pfid = rr->rr_fid1;
 986                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 987                 LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
 988                                                   RCL_CLIENT));
 989                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 990                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 991                                                         RCL_CLIENT);
 992                 sp->u.sp_ea.eadata = rr->rr_eadata;
 993                 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
 994                 sp->u.sp_ea.fid = rr->rr_fid1;
 995                 RETURN(0);
 996         }
 997 #endif
 998         if (S_ISDIR(attr->la_mode)) {
 999                 /* pass parent fid for cross-ref cases */
1000                 sp->u.sp_pfid = rr->rr_fid1;
1001                 if (sp->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1002                         /* create salve object req, need
1003                          * unpack split ea here
1004                          */
1005                        req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SLAVE);
1006                        LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
1007                                                          RCL_CLIENT));
1008                        rr->rr_eadata = req_capsule_client_get(pill,
1009                                                               &RMF_EADATA);
1010                        rr->rr_eadatalen = req_capsule_get_size(pill,
1011                                                                &RMF_EADATA,
1012                                                                RCL_CLIENT);
1013                        sp->u.sp_ea.eadata = rr->rr_eadata;
1014                        sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1015                        sp->u.sp_ea.fid = rr->rr_fid1;
1016                        RETURN(0);
1017                 }
1018                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1019         } else if (S_ISLNK(attr->la_mode)) {
1020                 const char *tgt = NULL;
1021
1022                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1023                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1024                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1025                         sp->u.sp_symname = tgt;
1026                 }
1027                 if (tgt == NULL)
1028                         RETURN(-EFAULT);
1029         } else {
1030                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1031         }
1032         rc = mdt_dlmreq_unpack(info);
1033         RETURN(rc);
1034 }
1035
1036 static int mdt_link_unpack(struct mdt_thread_info *info)
1037 {
1038         struct md_ucred         *uc  = mdt_ucred(info);
1039         struct mdt_rec_link     *rec;
1040         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1041         struct mdt_reint_record *rr = &info->mti_rr;
1042         struct req_capsule      *pill = info->mti_pill;
1043         int rc;
1044         ENTRY;
1045
1046         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1047         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1048         if (rec == NULL)
1049                 RETURN(-EFAULT);
1050
1051         uc->mu_fsuid = rec->lk_fsuid;
1052         uc->mu_fsgid = rec->lk_fsgid;
1053         uc->mu_cap   = rec->lk_cap;
1054         uc->mu_suppgids[0] = rec->lk_suppgid1;
1055         uc->mu_suppgids[1] = rec->lk_suppgid2;
1056
1057         attr->la_uid = rec->lk_fsuid;
1058         attr->la_gid = rec->lk_fsgid;
1059         rr->rr_fid1 = &rec->lk_fid1;
1060         rr->rr_fid2 = &rec->lk_fid2;
1061         attr->la_ctime = rec->lk_time;
1062         attr->la_mtime = rec->lk_time;
1063         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
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         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1069                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1070                                  req_capsule_client_get(pill, &RMF_CAPA2));
1071
1072         info->mti_spec.sp_ck_split = !!(rec->lk_bias & MDS_CHECK_SPLIT);
1073         info->mti_cross_ref = !!(rec->lk_bias & MDS_CROSS_REF);
1074         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1075         if (rr->rr_name == NULL)
1076                 RETURN(-EFAULT);
1077         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1078         if (!info->mti_cross_ref)
1079                 LASSERT(rr->rr_namelen > 0);
1080
1081         rc = mdt_dlmreq_unpack(info);
1082         RETURN(rc);
1083 }
1084
1085 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1086 {
1087         struct md_ucred         *uc  = mdt_ucred(info);
1088         struct mdt_rec_unlink   *rec;
1089         struct md_attr          *ma = &info->mti_attr;
1090         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1091         struct mdt_reint_record *rr = &info->mti_rr;
1092         struct req_capsule      *pill = info->mti_pill;
1093         int rc;
1094         ENTRY;
1095
1096         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1097         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1098         if (rec == NULL)
1099                 RETURN(-EFAULT);
1100
1101         uc->mu_fsuid = rec->ul_fsuid;
1102         uc->mu_fsgid = rec->ul_fsgid;
1103         uc->mu_cap   = rec->ul_cap;
1104         uc->mu_suppgids[0] = rec->ul_suppgid1;
1105         uc->mu_suppgids[1] = -1;
1106
1107         attr->la_uid = rec->ul_fsuid;
1108         attr->la_gid = rec->ul_fsgid;
1109         rr->rr_fid1 = &rec->ul_fid1;
1110         rr->rr_fid2 = &rec->ul_fid2;
1111         attr->la_ctime = rec->ul_time;
1112         attr->la_mtime = rec->ul_time;
1113         attr->la_mode  = rec->ul_mode;
1114         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1115
1116         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1117                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1118                                  req_capsule_client_get(pill, &RMF_CAPA1));
1119
1120         info->mti_cross_ref = !!(rec->ul_bias & MDS_CROSS_REF);
1121         if (!info->mti_cross_ref) {
1122                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1123                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1124                 if (rr->rr_name == NULL || rr->rr_namelen == 0)
1125                         RETURN(-EFAULT);
1126         } else {
1127                 rr->rr_name = NULL;
1128                 rr->rr_namelen = 0;
1129         }
1130         info->mti_spec.sp_ck_split = !!(rec->ul_bias & MDS_CHECK_SPLIT);
1131         if (rec->ul_bias & MDS_VTX_BYPASS)
1132                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1133         else
1134                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1135
1136         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1137         /* last unlink need LOV EA sent back */
1138         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1139
1140         rc = mdt_dlmreq_unpack(info);
1141         RETURN(rc);
1142 }
1143
1144 static int mdt_rename_unpack(struct mdt_thread_info *info)
1145 {
1146         struct md_ucred         *uc = mdt_ucred(info);
1147         struct mdt_rec_rename   *rec;
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1150         struct mdt_reint_record *rr = &info->mti_rr;
1151         struct req_capsule      *pill = info->mti_pill;
1152         int rc;
1153         ENTRY;
1154
1155         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1156         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1157         if (rec == NULL)
1158                 RETURN(-EFAULT);
1159
1160         uc->mu_fsuid = rec->rn_fsuid;
1161         uc->mu_fsgid = rec->rn_fsgid;
1162         uc->mu_cap   = rec->rn_cap;
1163         uc->mu_suppgids[0] = rec->rn_suppgid1;
1164         uc->mu_suppgids[1] = rec->rn_suppgid2;
1165
1166         attr->la_uid = rec->rn_fsuid;
1167         attr->la_gid = rec->rn_fsgid;
1168         rr->rr_fid1 = &rec->rn_fid1;
1169         rr->rr_fid2 = &rec->rn_fid2;
1170         attr->la_ctime = rec->rn_time;
1171         attr->la_mtime = rec->rn_time;
1172         /* rename_tgt contains the mode already */
1173         attr->la_mode = rec->rn_mode;
1174         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1175
1176         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1177                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1178                                  req_capsule_client_get(pill, &RMF_CAPA1));
1179         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1180                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1181                                  req_capsule_client_get(pill, &RMF_CAPA2));
1182
1183         info->mti_spec.sp_ck_split = !!(rec->rn_bias & MDS_CHECK_SPLIT);
1184         info->mti_cross_ref = !!(rec->rn_bias & MDS_CROSS_REF);
1185         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1186         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1187         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1188                 RETURN(-EFAULT);
1189         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1190         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1191         if (!info->mti_cross_ref)
1192                 LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1193         if (rec->rn_bias & MDS_VTX_BYPASS)
1194                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1195         else
1196                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1197
1198         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1199         /* rename may contain unlink so we might need LOV EA sent back */
1200         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1201
1202         rc = mdt_dlmreq_unpack(info);
1203         RETURN(rc);
1204 }
1205
1206 static int mdt_open_unpack(struct mdt_thread_info *info)
1207 {
1208         struct md_ucred         *uc = mdt_ucred(info);
1209         struct mdt_rec_create   *rec;
1210         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1211         struct req_capsule      *pill = info->mti_pill;
1212         struct mdt_reint_record *rr   = &info->mti_rr;
1213         struct ptlrpc_request   *req  = mdt_info_req(info);
1214         struct md_op_spec       *sp   = &info->mti_spec;
1215         ENTRY;
1216
1217         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1218         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1219         if (rec == NULL)
1220                 RETURN(-EFAULT);
1221
1222         uc->mu_fsuid = rec->cr_fsuid;
1223         uc->mu_fsgid = rec->cr_fsgid;
1224         uc->mu_cap   = rec->cr_cap;
1225         uc->mu_suppgids[0] = rec->cr_suppgid1;
1226         uc->mu_suppgids[1] = rec->cr_suppgid2;
1227
1228         rr->rr_fid1   = &rec->cr_fid1;
1229         rr->rr_fid2   = &rec->cr_fid2;
1230         rr->rr_handle = &rec->cr_old_handle;
1231         attr->la_mode = rec->cr_mode;
1232         attr->la_rdev  = rec->cr_rdev;
1233         attr->la_uid   = rec->cr_fsuid;
1234         attr->la_gid   = rec->cr_fsgid;
1235         attr->la_ctime = rec->cr_time;
1236         attr->la_mtime = rec->cr_time;
1237         attr->la_atime = rec->cr_time;
1238         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1239                          LA_CTIME | LA_MTIME | LA_ATIME;
1240         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1241         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1242         /* Do not trigger ASSERTION if client miss to set such flags. */
1243         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1244                 RETURN(-EPROTO);
1245         info->mti_replayepoch = rec->cr_ioepoch;
1246
1247         info->mti_spec.sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
1248         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1249
1250         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1251                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1252                                  req_capsule_client_get(pill, &RMF_CAPA1));
1253         if (req_is_replay(req) &&
1254             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1255 #if 0
1256                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1257                                  req_capsule_client_get(pill, &RMF_CAPA2));
1258 #else
1259                 /*
1260                  * FIXME: capa in replay open request might have expired,
1261                  * bypass capa check. Security hole?
1262                  */
1263                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1264                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1265 #endif
1266         }
1267
1268         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1269         if (rr->rr_name == NULL)
1270                 RETURN(-EFAULT);
1271         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1272
1273         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1274                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1275                                                         RCL_CLIENT);
1276                 if (rr->rr_eadatalen > 0) {
1277                         rr->rr_eadata = req_capsule_client_get(pill,
1278                                                                &RMF_EADATA);
1279                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1280                         sp->u.sp_ea.eadata = rr->rr_eadata;
1281                         sp->no_create = !!req_is_replay(req);
1282                 }
1283
1284                 /*
1285                  * Client default md_size may be 0 right after client start,
1286                  * until all osc are connected, set here just some reasonable
1287                  * value to prevent misbehavior.
1288                  */
1289                 if (rr->rr_eadatalen == 0 &&
1290                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1291                         rr->rr_eadatalen = MIN_MD_SIZE;
1292         }
1293
1294         RETURN(0);
1295 }
1296
1297 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1298 {
1299         struct mdt_reint_record   *rr   = &info->mti_rr;
1300         struct md_ucred           *uc   = mdt_ucred(info);
1301         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1302         struct req_capsule        *pill = info->mti_pill;
1303         struct mdt_rec_setxattr   *rec;
1304         ENTRY;
1305
1306
1307         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1308                          sizeof(struct mdt_rec_reint));
1309
1310         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1311         if (rec == NULL)
1312                 RETURN(-EFAULT);
1313
1314         uc->mu_fsuid  = rec->sx_fsuid;
1315         uc->mu_fsgid  = rec->sx_fsgid;
1316         uc->mu_cap    = rec->sx_cap;
1317         uc->mu_suppgids[0] = rec->sx_suppgid1;
1318         uc->mu_suppgids[1] = -1;
1319
1320         rr->rr_opcode = rec->sx_opcode;
1321         rr->rr_fid1   = &rec->sx_fid;
1322         attr->la_valid = rec->sx_valid;
1323         attr->la_ctime = rec->sx_time;
1324         attr->la_size = rec->sx_size;
1325         attr->la_flags = rec->sx_flags;
1326
1327         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1328                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1329                                  req_capsule_client_get(pill, &RMF_CAPA1));
1330         else
1331                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1332
1333         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1334         if (rr->rr_name == NULL)
1335                 RETURN(-EFAULT);
1336         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1337         LASSERT(rr->rr_namelen > 0);
1338
1339         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1340                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1341                                                         RCL_CLIENT);
1342                 if (rr->rr_eadatalen > 0) {
1343                         rr->rr_eadata = req_capsule_client_get(pill,
1344                                                                &RMF_EADATA);
1345                         if (rr->rr_eadata == NULL)
1346                                 RETURN(-EFAULT);
1347                 } else {
1348                         rr->rr_eadata = NULL;
1349                 }
1350         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1351                 CDEBUG(D_INFO, "no xattr data supplied\n");
1352                 RETURN(-EFAULT);
1353         }
1354
1355         RETURN(0);
1356 }
1357
1358
1359 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1360
1361 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1362         [REINT_SETATTR]  = mdt_setattr_unpack,
1363         [REINT_CREATE]   = mdt_create_unpack,
1364         [REINT_LINK]     = mdt_link_unpack,
1365         [REINT_UNLINK]   = mdt_unlink_unpack,
1366         [REINT_RENAME]   = mdt_rename_unpack,
1367         [REINT_OPEN]     = mdt_open_unpack,
1368         [REINT_SETXATTR] = mdt_setxattr_unpack
1369 };
1370
1371 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1372 {
1373         int rc;
1374         ENTRY;
1375
1376         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1377         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1378                 info->mti_rr.rr_opcode = op;
1379                 rc = mdt_reint_unpackers[op](info);
1380         } else {
1381                 CERROR("Unexpected opcode %d\n", op);
1382                 rc = -EFAULT;
1383         }
1384         RETURN(rc);
1385 }