lustre/mdt/mdt_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/mdt/mdt_lib.c
  39  *
  40  * Lustre Metadata Target (mdt) request unpacking helper.
  41  *
  42  * Author: Peter Braam <braam@clusterfs.com>
  43  * Author: Andreas Dilger <adilger@clusterfs.com>
  44  * Author: Phil Schwan <phil@clusterfs.com>
  45  * Author: Mike Shaver <shaver@clusterfs.com>
  46  * Author: Nikita Danilov <nikita@clusterfs.com>
  47  * Author: Huang Hua <huanghua@clusterfs.com>
  48  * Author: Fan Yong <fanyong@clusterfs.com>
  49  */
  50
  51
  52 #ifndef EXPORT_SYMTAB
  53 # define EXPORT_SYMTAB
  54 #endif
  55 #define DEBUG_SUBSYSTEM S_MDS
  56
  57 #include "mdt_internal.h"
  58 #include <lnet/lib-lnet.h>
  59
  60
  61 typedef enum ucred_init_type {
  62         NONE_INIT       = 0,
  63         BODY_INIT       = 1,
  64         REC_INIT        = 2
  65 } ucred_init_type_t;
  66
  67 void mdt_exit_ucred(struct mdt_thread_info *info)
  68 {
  69         struct md_ucred   *uc  = mdt_ucred(info);
  70         struct mdt_device *mdt = info->mti_mdt;
  71
  72         if (uc->mu_valid != UCRED_INIT) {
  73                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
  74                 if (uc->mu_ginfo) {
  75                         cfs_put_group_info(uc->mu_ginfo);
  76                         uc->mu_ginfo = NULL;
  77                 }
  78                 if (uc->mu_identity) {
  79                         mdt_identity_put(mdt->mdt_identity_cache,
  80                                          uc->mu_identity);
  81                         uc->mu_identity = NULL;
  82                 }
  83                 uc->mu_valid = UCRED_INIT;
  84         }
  85 }
  86
  87 static int match_nosquash_list(cfs_rw_semaphore_t *sem,
  88                                cfs_list_t *nidlist,
  89                                lnet_nid_t peernid)
  90 {
  91         int rc;
  92         ENTRY;
  93         cfs_down_read(sem);
  94         rc = cfs_match_nid(peernid, nidlist);
  95         cfs_up_read(sem);
  96         RETURN(rc);
  97 }
  98
  99 /* root_squash for inter-MDS operations */
 100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 101 {
 102         struct md_ucred *ucred = mdt_ucred(info);
 103         ENTRY;
 104
 105         if (!info->mti_mdt->mdt_squash_uid || ucred->mu_fsuid)
 106                 RETURN(0);
 107
 108         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 109                                 &info->mti_mdt->mdt_nosquash_nids,
 110                                 peernid)) {
 111                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 112                        libcfs_nid2str(peernid));
 113                 RETURN(0);
 114         }
 115
 116         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 117                libcfs_nid2str(peernid),
 118                ucred->mu_fsuid, ucred->mu_fsgid, ucred->mu_cap,
 119                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 120                0);
 121
 122         ucred->mu_fsuid = info->mti_mdt->mdt_squash_uid;
 123         ucred->mu_fsgid = info->mti_mdt->mdt_squash_gid;
 124         ucred->mu_cap = 0;
 125         ucred->mu_suppgids[0] = -1;
 126         ucred->mu_suppgids[1] = -1;
 127
 128         RETURN(0);
 129 }
 130
 131 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 132                           void *buf)
 133 {
 134         struct ptlrpc_request   *req = mdt_info_req(info);
 135         struct mdt_device       *mdt = info->mti_mdt;
 136         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 137         struct md_ucred         *ucred = mdt_ucred(info);
 138         lnet_nid_t               peernid = req->rq_peer.nid;
 139         __u32                    perm = 0;
 140         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 141         int                      setuid;
 142         int                      setgid;
 143         int                      rc = 0;
 144
 145         ENTRY;
 146
 147         LASSERT(req->rq_auth_gss);
 148         LASSERT(!req->rq_auth_usr_mdt);
 149         LASSERT(req->rq_user_desc);
 150
 151         ucred->mu_valid = UCRED_INVALID;
 152
 153         ucred->mu_o_uid   = pud->pud_uid;
 154         ucred->mu_o_gid   = pud->pud_gid;
 155         ucred->mu_o_fsuid = pud->pud_fsuid;
 156         ucred->mu_o_fsgid = pud->pud_fsgid;
 157
 158         if (type == BODY_INIT) {
 159                 struct mdt_body *body = (struct mdt_body *)buf;
 160
 161                 ucred->mu_suppgids[0] = body->suppgid;
 162                 ucred->mu_suppgids[1] = -1;
 163         }
 164
 165         /* sanity check: we expect the uid which client claimed is true */
 166         if (remote) {
 167                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 168                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 169                         RETURN(-EACCES);
 170                 }
 171
 172                 if (ptlrpc_user_desc_do_idmap(req, pud))
 173                         RETURN(-EACCES);
 174
 175                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 176                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 177                                "while client claims %u:%u/%u:%u\n",
 178                                libcfs_nid2str(peernid), req->rq_auth_uid,
 179                                req->rq_auth_mapped_uid,
 180                                pud->pud_uid, pud->pud_gid,
 181                                pud->pud_fsuid, pud->pud_fsgid);
 182                         RETURN(-EACCES);
 183                 }
 184         } else {
 185                 if (req->rq_auth_uid != pud->pud_uid) {
 186                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 187                                "while client claims %u:%u/%u:%u\n",
 188                                libcfs_nid2str(peernid), req->rq_auth_uid,
 189                                pud->pud_uid, pud->pud_gid,
 190                                pud->pud_fsuid, pud->pud_fsgid);
 191                         RETURN(-EACCES);
 192                 }
 193         }
 194
 195         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 196                 if (remote) {
 197                         CDEBUG(D_SEC, "remote client must run with identity_get "
 198                                "enabled!\n");
 199                         RETURN(-EACCES);
 200                 } else {
 201                         ucred->mu_identity = NULL;
 202                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 203                                CFS_SETGRP_PERM;
 204                 }
 205         } else {
 206                 struct md_identity *identity;
 207
 208                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 209                                             pud->pud_uid);
 210                 if (IS_ERR(identity)) {
 211                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 212                                      !remote)) {
 213                                 ucred->mu_identity = NULL;
 214                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 215                                        CFS_SETGRP_PERM;
 216                         } else {
 217                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 218                                        pud->pud_uid);
 219                                 RETURN(-EACCES);
 220                         }
 221                 } else {
 222                         ucred->mu_identity = identity;
 223                         perm = mdt_identity_get_perm(ucred->mu_identity,
 224                                                      remote, peernid);
 225                 }
 226         }
 227
 228         /* find out the setuid/setgid attempt */
 229         setuid = (pud->pud_uid != pud->pud_fsuid);
 230         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 231                   (ucred->mu_identity &&
 232                   (pud->pud_gid != ucred->mu_identity->mi_gid)));
 233
 234         /* check permission of setuid */
 235         if (setuid && !(perm & CFS_SETUID_PERM)) {
 236                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 237                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 238                 GOTO(out, rc = -EACCES);
 239         }
 240
 241         /* check permission of setgid */
 242         if (setgid && !(perm & CFS_SETGID_PERM)) {
 243                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 244                        "from %s\n", pud->pud_uid, pud->pud_gid,
 245                        pud->pud_fsuid, pud->pud_fsgid,
 246                        ucred->mu_identity->mi_gid, libcfs_nid2str(peernid));
 247                 GOTO(out, rc = -EACCES);
 248         }
 249
 250         /*
 251          * NB: remote client not allowed to setgroups anyway.
 252          */
 253         if (!remote && perm & CFS_SETGRP_PERM) {
 254                 if (pud->pud_ngroups) {
 255                         /* setgroups for local client */
 256                         ucred->mu_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 257                         if (!ucred->mu_ginfo) {
 258                                 CERROR("failed to alloc %d groups\n",
 259                                        pud->pud_ngroups);
 260                                 GOTO(out, rc = -ENOMEM);
 261                         }
 262
 263                         lustre_groups_from_list(ucred->mu_ginfo,
 264                                                 pud->pud_groups);
 265                         lustre_groups_sort(ucred->mu_ginfo);
 266                 } else {
 267                         ucred->mu_ginfo = NULL;
 268                 }
 269         } else {
 270                 ucred->mu_suppgids[0] = -1;
 271                 ucred->mu_suppgids[1] = -1;
 272                 ucred->mu_ginfo = NULL;
 273         }
 274
 275         ucred->mu_uid   = pud->pud_uid;
 276         ucred->mu_gid   = pud->pud_gid;
 277         ucred->mu_fsuid = pud->pud_fsuid;
 278         ucred->mu_fsgid = pud->pud_fsgid;
 279
 280         /* process root_squash here. */
 281         mdt_root_squash(info, peernid);
 282
 283         /* remove fs privilege for non-root user. */
 284         if (ucred->mu_fsuid)
 285                 ucred->mu_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 286         else
 287                 ucred->mu_cap = pud->pud_cap;
 288         if (remote && !(perm & CFS_RMTOWN_PERM))
 289                 ucred->mu_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 290                                    CFS_CAP_CHOWN_MASK);
 291         ucred->mu_valid = UCRED_NEW;
 292
 293         EXIT;
 294
 295 out:
 296         if (rc) {
 297                 if (ucred->mu_ginfo) {
 298                         cfs_put_group_info(ucred->mu_ginfo);
 299                         ucred->mu_ginfo = NULL;
 300                 }
 301                 if (ucred->mu_identity) {
 302                         mdt_identity_put(mdt->mdt_identity_cache,
 303                                          ucred->mu_identity);
 304                         ucred->mu_identity = NULL;
 305                 }
 306         }
 307
 308         return rc;
 309 }
 310
 311 int mdt_check_ucred(struct mdt_thread_info *info)
 312 {
 313         struct ptlrpc_request   *req = mdt_info_req(info);
 314         struct mdt_device       *mdt = info->mti_mdt;
 315         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 316         struct md_ucred         *ucred = mdt_ucred(info);
 317         struct md_identity      *identity = NULL;
 318         lnet_nid_t               peernid = req->rq_peer.nid;
 319         __u32                    perm = 0;
 320         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 321         int                      setuid;
 322         int                      setgid;
 323         int                      rc = 0;
 324
 325         ENTRY;
 326
 327         if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
 328                 RETURN(0);
 329
 330         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 331                 RETURN(0);
 332
 333         /* sanity check: if we use strong authentication, we expect the
 334          * uid which client claimed is true */
 335         if (remote) {
 336                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 337                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 338                         RETURN(-EACCES);
 339                 }
 340
 341                 if (ptlrpc_user_desc_do_idmap(req, pud))
 342                         RETURN(-EACCES);
 343
 344                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 345                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 346                                "while client claims %u:%u/%u:%u\n",
 347                                libcfs_nid2str(peernid), req->rq_auth_uid,
 348                                req->rq_auth_mapped_uid,
 349                                pud->pud_uid, pud->pud_gid,
 350                                pud->pud_fsuid, pud->pud_fsgid);
 351                         RETURN(-EACCES);
 352                 }
 353         } else {
 354                 if (req->rq_auth_uid != pud->pud_uid) {
 355                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 356                                "while client claims %u:%u/%u:%u\n",
 357                                libcfs_nid2str(peernid), req->rq_auth_uid,
 358                                pud->pud_uid, pud->pud_gid,
 359                                pud->pud_fsuid, pud->pud_fsgid);
 360                         RETURN(-EACCES);
 361                 }
 362         }
 363
 364         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 365                 if (remote) {
 366                         CDEBUG(D_SEC, "remote client must run with identity_get "
 367                                "enabled!\n");
 368                         RETURN(-EACCES);
 369                 }
 370                 RETURN(0);
 371         }
 372
 373         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 374         if (IS_ERR(identity)) {
 375                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 376                              !remote)) {
 377                         RETURN(0);
 378                 } else {
 379                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 380                                pud->pud_uid);
 381                         RETURN(-EACCES);
 382                }
 383         }
 384
 385         perm = mdt_identity_get_perm(identity, remote, peernid);
 386         /* find out the setuid/setgid attempt */
 387         setuid = (pud->pud_uid != pud->pud_fsuid);
 388         setgid = (pud->pud_gid != pud->pud_fsgid ||
 389                   pud->pud_gid != identity->mi_gid);
 390
 391         /* check permission of setuid */
 392         if (setuid && !(perm & CFS_SETUID_PERM)) {
 393                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 394                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 395                 GOTO(out, rc = -EACCES);
 396         }
 397
 398         /* check permission of setgid */
 399         if (setgid && !(perm & CFS_SETGID_PERM)) {
 400                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 401                        "from %s\n", pud->pud_uid, pud->pud_gid,
 402                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 403                        libcfs_nid2str(peernid));
 404                 GOTO(out, rc = -EACCES);
 405         }
 406
 407         EXIT;
 408
 409 out:
 410         mdt_identity_put(mdt->mdt_identity_cache, identity);
 411         return rc;
 412 }
 413
 414 static int old_init_ucred(struct mdt_thread_info *info,
 415                           struct mdt_body *body)
 416 {
 417         struct md_ucred *uc = mdt_ucred(info);
 418         struct mdt_device  *mdt = info->mti_mdt;
 419         struct md_identity *identity = NULL;
 420
 421         ENTRY;
 422
 423         uc->mu_valid = UCRED_INVALID;
 424         uc->mu_o_uid = uc->mu_uid = body->uid;
 425         uc->mu_o_gid = uc->mu_gid = body->gid;
 426         uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
 427         uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
 428         uc->mu_suppgids[0] = body->suppgid;
 429         uc->mu_suppgids[1] = -1;
 430         uc->mu_ginfo = NULL;
 431         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 432                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 433                                             uc->mu_fsuid);
 434                 if (IS_ERR(identity)) {
 435                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 436                                 identity = NULL;
 437                         } else {
 438                                 CDEBUG(D_SEC, "Deny access without identity: "
 439                                        "uid %u\n", uc->mu_fsuid);
 440                                 RETURN(-EACCES);
 441                         }
 442                 }
 443         }
 444         uc->mu_identity = identity;
 445
 446         /* process root_squash here. */
 447         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 448
 449         /* remove fs privilege for non-root user. */
 450         if (uc->mu_fsuid)
 451                 uc->mu_cap = body->capability & ~CFS_CAP_FS_MASK;
 452         else
 453                 uc->mu_cap = body->capability;
 454         uc->mu_valid = UCRED_OLD;
 455
 456         RETURN(0);
 457 }
 458
 459 static int old_init_ucred_reint(struct mdt_thread_info *info)
 460 {
 461         struct md_ucred *uc = mdt_ucred(info);
 462         struct mdt_device  *mdt = info->mti_mdt;
 463         struct md_identity *identity = NULL;
 464
 465         ENTRY;
 466
 467         uc->mu_valid = UCRED_INVALID;
 468         uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
 469         uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
 470         uc->mu_ginfo = NULL;
 471         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 472                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 473                                             uc->mu_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->mu_fsuid);
 480                                 RETURN(-EACCES);
 481                         }
 482                 }
 483         }
 484         uc->mu_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->mu_fsuid)
 491                 uc->mu_cap &= ~CFS_CAP_FS_MASK;
 492         uc->mu_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 md_ucred       *uc  = mdt_ucred(info);
 501
 502         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 503                 return 0;
 504
 505         mdt_exit_ucred(info);
 506
 507         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 508                 return old_init_ucred(info, body);
 509         else
 510                 return new_init_ucred(info, BODY_INIT, body);
 511 }
 512
 513 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 514 {
 515         struct ptlrpc_request *req = mdt_info_req(info);
 516         struct md_ucred       *uc  = mdt_ucred(info);
 517
 518         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 519                 return 0;
 520
 521         mdt_exit_ucred(info);
 522
 523         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 524                 return old_init_ucred_reint(info);
 525         else
 526                 return new_init_ucred(info, REC_INIT, NULL);
 527 }
 528
 529 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 530 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 531 {
 532         const struct lov_ost_data_v1 *lod;
 533         int                           i;
 534         __u16                         count;
 535
 536         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 537
 538         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
 539                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 540                le32_to_cpu(lmm->lmm_pattern));
 541         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 542                le32_to_cpu(lmm->lmm_stripe_size), count);
 543         if (count == LOV_ALL_STRIPES)
 544                 return;
 545         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 546         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++)
 547                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 548                        i, le32_to_cpu(lod->l_ost_idx),
 549                        le64_to_cpu(lod->l_object_seq),
 550                        le64_to_cpu(lod->l_object_id));
 551 }
 552
 553 /* Shrink and/or grow reply buffers */
 554 int mdt_fix_reply(struct mdt_thread_info *info)
 555 {
 556         struct req_capsule *pill = info->mti_pill;
 557         struct mdt_body    *body;
 558         int                md_size, md_packed = 0;
 559         int                acl_size;
 560         int                rc = 0;
 561         ENTRY;
 562
 563         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 564         LASSERT(body != NULL);
 565
 566         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 567                 md_size = body->eadatasize;
 568         else
 569                 md_size = 0;
 570
 571         acl_size = body->aclsize;
 572
 573         /* this replay - not send info to client */
 574         if (info->mti_spec.no_create == 1) {
 575                 md_size = 0;
 576                 acl_size = 0;
 577         }
 578
 579         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 580                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 581                         md_size, acl_size,
 582                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 583                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 584 /*
 585             &RMF_MDT_BODY,
 586             &RMF_MDT_MD,
 587             &RMF_ACL, or &RMF_LOGCOOKIES
 588 (optional)  &RMF_CAPA1,
 589 (optional)  &RMF_CAPA2,
 590 (optional)  something else
 591 */
 592
 593         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 594          * buffers before growing it */
 595         if (info->mti_attr.ma_big_lmm_used) {
 596                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 597                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 598                                                  RCL_SERVER);
 599                 LASSERT(md_packed > 0);
 600                 /* buffer must be allocated separately */
 601                 LASSERT(info->mti_attr.ma_lmm !=
 602                         req_capsule_server_get(pill, &RMF_MDT_MD));
 603                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 604                 /* free big lmm if md_size is not needed */
 605                 if (md_size == 0)
 606                         info->mti_attr.ma_big_lmm_used = 0;
 607         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 608                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 609         }
 610
 611         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 612                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 613         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 614                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 615                                    acl_size, RCL_SERVER);
 616
 617         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 618             !(body->valid & OBD_MD_FLMDSCAPA))
 619                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 620
 621         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 622             !(body->valid & OBD_MD_FLOSSCAPA))
 623                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 624
 625         /*
 626          * Some more field should be shrinked if needed.
 627          * This should be done by those who added fields to reply message.
 628          */
 629
 630         /* Grow MD buffer if needed finally */
 631         if (info->mti_attr.ma_big_lmm_used) {
 632                 void *lmm;
 633
 634                 LASSERT(md_size > md_packed);
 635                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 636                        md_size - md_packed);
 637                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 638                 if (rc) {
 639                         /* we can't answer with proper LOV EA, drop flags,
 640                          * the rc is also returned so this request is
 641                          * considered as failed */
 642                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 643                         /* don't return transno along with error */
 644                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 645                 } else {
 646                         /* now we need to pack right LOV EA */
 647                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 648                         LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 649                                                      RCL_SERVER) ==
 650                                 info->mti_attr.ma_lmm_size);
 651                         memcpy(lmm, info->mti_attr.ma_lmm,
 652                                info->mti_attr.ma_lmm_size);
 653                 }
 654                 /* update mdt_max_mdsize so clients will be aware about that */
 655                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 656                         info->mti_mdt->mdt_max_mdsize =
 657                                                     info->mti_attr.ma_lmm_size;
 658                 info->mti_attr.ma_big_lmm_used = 0;
 659         }
 660         RETURN(rc);
 661 }
 662
 663
 664 /* if object is dying, pack the lov/llog data,
 665  * parameter info->mti_attr should be valid at this point! */
 666 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 667                            const struct md_attr *ma)
 668 {
 669         struct mdt_body       *repbody;
 670         const struct lu_attr *la = &ma->ma_attr;
 671         int rc;
 672         ENTRY;
 673
 674         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 675         LASSERT(repbody != NULL);
 676
 677         if (ma->ma_valid & MA_INODE)
 678                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 679
 680         if (ma->ma_valid & MA_LOV) {
 681                 __u32 mode;
 682
 683                 if (mdt_object_exists(mo) < 0)
 684                         /* If it is a remote object, and we do not retrieve
 685                          * EA back unlink reg file*/
 686                         mode = S_IFREG;
 687                 else
 688                         mode = lu_object_attr(&mo->mot_obj.mo_lu);
 689
 690                 LASSERT(ma->ma_lmm_size);
 691                 mdt_dump_lmm(D_INFO, ma->ma_lmm);
 692                 repbody->eadatasize = ma->ma_lmm_size;
 693                 if (S_ISREG(mode))
 694                         repbody->valid |= OBD_MD_FLEASIZE;
 695                 else if (S_ISDIR(mode))
 696                         repbody->valid |= OBD_MD_FLDIREA;
 697                 else
 698                         LBUG();
 699         }
 700
 701         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 702                 repbody->aclsize = ma->ma_cookie_size;
 703                 repbody->valid |= OBD_MD_FLCOOKIE;
 704         }
 705
 706         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 707             info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
 708             repbody->valid & OBD_MD_FLEASIZE) {
 709                 struct lustre_capa *capa;
 710
 711                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 712                 LASSERT(capa);
 713                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 714                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 715                 if (rc)
 716                         RETURN(rc);
 717
 718                 repbody->valid |= OBD_MD_FLOSSCAPA;
 719         }
 720
 721         RETURN(0);
 722 }
 723
 724 static inline unsigned int attr_unpack(__u64 sa_valid) {
 725         unsigned int ia_valid = 0;
 726
 727         if (sa_valid & MDS_ATTR_MODE)
 728                 ia_valid |= ATTR_MODE;
 729         if (sa_valid & MDS_ATTR_UID)
 730                 ia_valid |= ATTR_UID;
 731         if (sa_valid & MDS_ATTR_GID)
 732                 ia_valid |= ATTR_GID;
 733         if (sa_valid & MDS_ATTR_SIZE)
 734                 ia_valid |= ATTR_SIZE;
 735         if (sa_valid & MDS_ATTR_ATIME)
 736                 ia_valid |= ATTR_ATIME;
 737         if (sa_valid & MDS_ATTR_MTIME)
 738                 ia_valid |= ATTR_MTIME;
 739         if (sa_valid & MDS_ATTR_CTIME)
 740                 ia_valid |= ATTR_CTIME;
 741         if (sa_valid & MDS_ATTR_ATIME_SET)
 742                 ia_valid |= ATTR_ATIME_SET;
 743         if (sa_valid & MDS_ATTR_MTIME_SET)
 744                 ia_valid |= ATTR_MTIME_SET;
 745         if (sa_valid & MDS_ATTR_FORCE)
 746                 ia_valid |= ATTR_FORCE;
 747         if (sa_valid & MDS_ATTR_ATTR_FLAG)
 748                 ia_valid |= ATTR_ATTR_FLAG;
 749         if (sa_valid & MDS_ATTR_KILL_SUID)
 750                 ia_valid |=  ATTR_KILL_SUID;
 751         if (sa_valid & MDS_ATTR_KILL_SGID)
 752                 ia_valid |= ATTR_KILL_SGID;
 753         if (sa_valid & MDS_ATTR_CTIME_SET)
 754                 ia_valid |= ATTR_CTIME_SET;
 755         if (sa_valid & MDS_ATTR_FROM_OPEN)
 756                 ia_valid |= ATTR_FROM_OPEN;
 757         if (sa_valid & MDS_ATTR_BLOCKS)
 758                 ia_valid |= ATTR_BLOCKS;
 759         if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
 760                 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
 761         return ia_valid;
 762 }
 763
 764 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 765                                   struct md_attr *ma)
 766 {
 767         __u64 out;
 768
 769         out = 0;
 770         if (in & ATTR_MODE)
 771                 out |= LA_MODE;
 772         if (in & ATTR_UID)
 773                 out |= LA_UID;
 774         if (in & ATTR_GID)
 775                 out |= LA_GID;
 776         if (in & ATTR_SIZE)
 777                 out |= LA_SIZE;
 778         if (in & ATTR_BLOCKS)
 779                 out |= LA_BLOCKS;
 780
 781         if (in & ATTR_FROM_OPEN)
 782                 rr->rr_flags |= MRF_OPEN_TRUNC;
 783
 784         if (in & ATTR_ATIME_SET)
 785                 out |= LA_ATIME;
 786
 787         if (in & ATTR_CTIME_SET)
 788                 out |= LA_CTIME;
 789
 790         if (in & ATTR_MTIME_SET)
 791                 out |= LA_MTIME;
 792
 793         if (in & ATTR_ATTR_FLAG)
 794                 out |= LA_FLAGS;
 795
 796         if (in & ATTR_KILL_SUID)
 797                 out |= LA_KILL_SUID;
 798
 799         if (in & ATTR_KILL_SGID)
 800                 out |= LA_KILL_SGID;
 801
 802         if (in & MDS_OPEN_OWNEROVERRIDE)
 803                 ma->ma_attr_flags |= MDS_OPEN_OWNEROVERRIDE;
 804
 805         if (in & ATTR_FORCE)
 806                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 807
 808         /*XXX need ATTR_RAW?*/
 809         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 810                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 811                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 812                 ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
 813                 ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
 814         if (in != 0)
 815                 CERROR("Unknown attr bits: "LPX64"\n", in);
 816         return out;
 817 }
 818 /* unpacking */
 819
 820 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 821 {
 822         struct md_ucred         *uc  = mdt_ucred(info);
 823         struct md_attr          *ma = &info->mti_attr;
 824         struct lu_attr          *la = &ma->ma_attr;
 825         struct req_capsule      *pill = info->mti_pill;
 826         struct mdt_reint_record *rr = &info->mti_rr;
 827         struct mdt_rec_setattr  *rec;
 828         ENTRY;
 829
 830         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 831         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 832         if (rec == NULL)
 833                 RETURN(-EFAULT);
 834
 835         uc->mu_fsuid = rec->sa_fsuid;
 836         uc->mu_fsgid = rec->sa_fsgid;
 837         uc->mu_cap   = rec->sa_cap;
 838         uc->mu_suppgids[0] = rec->sa_suppgid;
 839         uc->mu_suppgids[1] = -1;
 840
 841         rr->rr_fid1 = &rec->sa_fid;
 842         la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
 843         la->la_mode  = rec->sa_mode;
 844         la->la_flags = rec->sa_attr_flags;
 845         la->la_uid   = rec->sa_uid;
 846         la->la_gid   = rec->sa_gid;
 847         la->la_size  = rec->sa_size;
 848         la->la_blocks = rec->sa_blocks;
 849         la->la_ctime = rec->sa_ctime;
 850         la->la_atime = rec->sa_atime;
 851         la->la_mtime = rec->sa_mtime;
 852         ma->ma_valid = MA_INODE;
 853
 854         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 855                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 856                                  req_capsule_client_get(pill, &RMF_CAPA1));
 857
 858         RETURN(0);
 859 }
 860
 861 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 862 {
 863         struct req_capsule *pill = info->mti_pill;
 864         ENTRY;
 865
 866         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 867                 info->mti_ioepoch =
 868                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 869         else
 870                 info->mti_ioepoch = NULL;
 871         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 872 }
 873
 874 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 875         struct req_capsule      *pill = info->mti_pill;
 876
 877         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 878                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 879                 if (info->mti_dlm_req == NULL)
 880                         RETURN(-EFAULT);
 881         }
 882
 883         RETURN(0);
 884 }
 885
 886 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 887 {
 888         struct mdt_reint_record *rr = &info->mti_rr;
 889         struct md_attr          *ma = &info->mti_attr;
 890         struct req_capsule      *pill = info->mti_pill;
 891         int rc;
 892         ENTRY;
 893
 894         rc = mdt_setattr_unpack_rec(info);
 895         if (rc)
 896                 RETURN(rc);
 897
 898         /* Epoch may be absent */
 899         mdt_ioepoch_unpack(info);
 900
 901         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 902                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 903                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 904                                                         RCL_CLIENT);
 905                 ma->ma_lmm_size = rr->rr_eadatalen;
 906                 if (ma->ma_lmm_size > 0) {
 907                         ma->ma_lmm = (void *)rr->rr_eadata;
 908                         ma->ma_valid |= MA_LOV;
 909                 }
 910         }
 911
 912         ma->ma_cookie_size = req_capsule_get_size(pill, &RMF_LOGCOOKIES,
 913                                                   RCL_CLIENT);
 914         if (ma->ma_cookie_size) {
 915                 ma->ma_cookie = req_capsule_client_get(pill, &RMF_LOGCOOKIES);
 916                 ma->ma_valid |= MA_COOKIE;
 917         }
 918
 919         rc = mdt_dlmreq_unpack(info);
 920         RETURN(rc);
 921 }
 922
 923 int mdt_close_unpack(struct mdt_thread_info *info)
 924 {
 925         int rc;
 926         ENTRY;
 927
 928         rc = mdt_ioepoch_unpack(info);
 929         if (rc)
 930                 RETURN(rc);
 931
 932         RETURN(mdt_setattr_unpack_rec(info));
 933 }
 934
 935 static int mdt_create_unpack(struct mdt_thread_info *info)
 936 {
 937         struct md_ucred         *uc  = mdt_ucred(info);
 938         struct mdt_rec_create   *rec;
 939         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 940         struct mdt_reint_record *rr = &info->mti_rr;
 941         struct req_capsule      *pill = info->mti_pill;
 942         struct md_op_spec       *sp = &info->mti_spec;
 943         int rc;
 944         ENTRY;
 945
 946         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 947         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 948         if (rec == NULL)
 949                 RETURN(-EFAULT);
 950
 951         uc->mu_fsuid = rec->cr_fsuid;
 952         uc->mu_fsgid = rec->cr_fsgid;
 953         uc->mu_cap   = rec->cr_cap;
 954         uc->mu_suppgids[0] = rec->cr_suppgid1;
 955         uc->mu_suppgids[1] = -1;
 956
 957         rr->rr_fid1 = &rec->cr_fid1;
 958         rr->rr_fid2 = &rec->cr_fid2;
 959         attr->la_mode = rec->cr_mode;
 960         attr->la_rdev  = rec->cr_rdev;
 961         attr->la_uid   = rec->cr_fsuid;
 962         attr->la_gid   = rec->cr_fsgid;
 963         attr->la_ctime = rec->cr_time;
 964         attr->la_mtime = rec->cr_time;
 965         attr->la_atime = rec->cr_time;
 966         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
 967                          LA_CTIME | LA_MTIME | LA_ATIME;
 968         memset(&sp->u, 0, sizeof(sp->u));
 969         sp->sp_cr_flags = get_mrc_cr_flags(rec);
 970         sp->sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
 971         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
 972
 973         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 974                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 975                                  req_capsule_client_get(pill, &RMF_CAPA1));
 976         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 977
 978         if (!info->mti_cross_ref) {
 979                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 980                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
 981                                                       RCL_CLIENT) - 1;
 982                 LASSERT(rr->rr_name && rr->rr_namelen > 0);
 983         } else {
 984                 rr->rr_name = NULL;
 985                 rr->rr_namelen = 0;
 986         }
 987
 988 #ifdef CONFIG_FS_POSIX_ACL
 989         if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
 990                 if (S_ISDIR(attr->la_mode))
 991                         sp->u.sp_pfid = rr->rr_fid1;
 992                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 993                 LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
 994                                                   RCL_CLIENT));
 995                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 996                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 997                                                         RCL_CLIENT);
 998                 sp->u.sp_ea.eadata = rr->rr_eadata;
 999                 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1000                 sp->u.sp_ea.fid = rr->rr_fid1;
1001                 RETURN(0);
1002         }
1003 #endif
1004         if (S_ISDIR(attr->la_mode)) {
1005                 /* pass parent fid for cross-ref cases */
1006                 sp->u.sp_pfid = rr->rr_fid1;
1007                 if (sp->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1008                         /* create salve object req, need
1009                          * unpack split ea here
1010                          */
1011                        req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SLAVE);
1012                        LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
1013                                                          RCL_CLIENT));
1014                        rr->rr_eadata = req_capsule_client_get(pill,
1015                                                               &RMF_EADATA);
1016                        rr->rr_eadatalen = req_capsule_get_size(pill,
1017                                                                &RMF_EADATA,
1018                                                                RCL_CLIENT);
1019                        sp->u.sp_ea.eadata = rr->rr_eadata;
1020                        sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1021                        sp->u.sp_ea.fid = rr->rr_fid1;
1022                        RETURN(0);
1023                 }
1024                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1025         } else if (S_ISLNK(attr->la_mode)) {
1026                 const char *tgt = NULL;
1027
1028                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1029                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1030                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1031                         sp->u.sp_symname = tgt;
1032                 }
1033                 if (tgt == NULL)
1034                         RETURN(-EFAULT);
1035         } else {
1036                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1037         }
1038         rc = mdt_dlmreq_unpack(info);
1039         RETURN(rc);
1040 }
1041
1042 static int mdt_link_unpack(struct mdt_thread_info *info)
1043 {
1044         struct md_ucred         *uc  = mdt_ucred(info);
1045         struct mdt_rec_link     *rec;
1046         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1047         struct mdt_reint_record *rr = &info->mti_rr;
1048         struct req_capsule      *pill = info->mti_pill;
1049         int rc;
1050         ENTRY;
1051
1052         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1053         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1054         if (rec == NULL)
1055                 RETURN(-EFAULT);
1056
1057         uc->mu_fsuid = rec->lk_fsuid;
1058         uc->mu_fsgid = rec->lk_fsgid;
1059         uc->mu_cap   = rec->lk_cap;
1060         uc->mu_suppgids[0] = rec->lk_suppgid1;
1061         uc->mu_suppgids[1] = rec->lk_suppgid2;
1062
1063         attr->la_uid = rec->lk_fsuid;
1064         attr->la_gid = rec->lk_fsgid;
1065         rr->rr_fid1 = &rec->lk_fid1;
1066         rr->rr_fid2 = &rec->lk_fid2;
1067         attr->la_ctime = rec->lk_time;
1068         attr->la_mtime = rec->lk_time;
1069         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1070
1071         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1072                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1073                                  req_capsule_client_get(pill, &RMF_CAPA1));
1074         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1075                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1076                                  req_capsule_client_get(pill, &RMF_CAPA2));
1077
1078         info->mti_spec.sp_ck_split = !!(rec->lk_bias & MDS_CHECK_SPLIT);
1079         info->mti_cross_ref = !!(rec->lk_bias & MDS_CROSS_REF);
1080         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1081         if (rr->rr_name == NULL)
1082                 RETURN(-EFAULT);
1083         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1084         if (!info->mti_cross_ref)
1085                 LASSERT(rr->rr_namelen > 0);
1086
1087         rc = mdt_dlmreq_unpack(info);
1088         RETURN(rc);
1089 }
1090
1091 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1092 {
1093         struct md_ucred         *uc  = mdt_ucred(info);
1094         struct mdt_rec_unlink   *rec;
1095         struct md_attr          *ma = &info->mti_attr;
1096         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1097         struct mdt_reint_record *rr = &info->mti_rr;
1098         struct req_capsule      *pill = info->mti_pill;
1099         int rc;
1100         ENTRY;
1101
1102         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1103         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1104         if (rec == NULL)
1105                 RETURN(-EFAULT);
1106
1107         uc->mu_fsuid = rec->ul_fsuid;
1108         uc->mu_fsgid = rec->ul_fsgid;
1109         uc->mu_cap   = rec->ul_cap;
1110         uc->mu_suppgids[0] = rec->ul_suppgid1;
1111         uc->mu_suppgids[1] = -1;
1112
1113         attr->la_uid = rec->ul_fsuid;
1114         attr->la_gid = rec->ul_fsgid;
1115         rr->rr_fid1 = &rec->ul_fid1;
1116         rr->rr_fid2 = &rec->ul_fid2;
1117         attr->la_ctime = rec->ul_time;
1118         attr->la_mtime = rec->ul_time;
1119         attr->la_mode  = rec->ul_mode;
1120         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1121
1122         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1123                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1124                                  req_capsule_client_get(pill, &RMF_CAPA1));
1125
1126         info->mti_cross_ref = !!(rec->ul_bias & MDS_CROSS_REF);
1127         if (!info->mti_cross_ref) {
1128                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1129                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1130                 if (rr->rr_name == NULL || rr->rr_namelen == 0)
1131                         RETURN(-EFAULT);
1132         } else {
1133                 rr->rr_name = NULL;
1134                 rr->rr_namelen = 0;
1135         }
1136         info->mti_spec.sp_ck_split = !!(rec->ul_bias & MDS_CHECK_SPLIT);
1137         if (rec->ul_bias & MDS_VTX_BYPASS)
1138                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1139         else
1140                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1141
1142         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1143         /* last unlink need LOV EA sent back */
1144         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1145
1146         rc = mdt_dlmreq_unpack(info);
1147         RETURN(rc);
1148 }
1149
1150 static int mdt_rename_unpack(struct mdt_thread_info *info)
1151 {
1152         struct md_ucred         *uc = mdt_ucred(info);
1153         struct mdt_rec_rename   *rec;
1154         struct md_attr          *ma = &info->mti_attr;
1155         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1156         struct mdt_reint_record *rr = &info->mti_rr;
1157         struct req_capsule      *pill = info->mti_pill;
1158         int rc;
1159         ENTRY;
1160
1161         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1162         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1163         if (rec == NULL)
1164                 RETURN(-EFAULT);
1165
1166         uc->mu_fsuid = rec->rn_fsuid;
1167         uc->mu_fsgid = rec->rn_fsgid;
1168         uc->mu_cap   = rec->rn_cap;
1169         uc->mu_suppgids[0] = rec->rn_suppgid1;
1170         uc->mu_suppgids[1] = rec->rn_suppgid2;
1171
1172         attr->la_uid = rec->rn_fsuid;
1173         attr->la_gid = rec->rn_fsgid;
1174         rr->rr_fid1 = &rec->rn_fid1;
1175         rr->rr_fid2 = &rec->rn_fid2;
1176         attr->la_ctime = rec->rn_time;
1177         attr->la_mtime = rec->rn_time;
1178         /* rename_tgt contains the mode already */
1179         attr->la_mode = rec->rn_mode;
1180         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1181
1182         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1183                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1184                                  req_capsule_client_get(pill, &RMF_CAPA1));
1185         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1186                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1187                                  req_capsule_client_get(pill, &RMF_CAPA2));
1188
1189         info->mti_spec.sp_ck_split = !!(rec->rn_bias & MDS_CHECK_SPLIT);
1190         info->mti_cross_ref = !!(rec->rn_bias & MDS_CROSS_REF);
1191         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1192         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1193         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1194                 RETURN(-EFAULT);
1195         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1196         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1197         if (!info->mti_cross_ref)
1198                 LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1199         if (rec->rn_bias & MDS_VTX_BYPASS)
1200                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1201         else
1202                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1203
1204         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1205         /* rename may contain unlink so we might need LOV EA sent back */
1206         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1207
1208         rc = mdt_dlmreq_unpack(info);
1209         RETURN(rc);
1210 }
1211
1212 static int mdt_open_unpack(struct mdt_thread_info *info)
1213 {
1214         struct md_ucred         *uc = mdt_ucred(info);
1215         struct mdt_rec_create   *rec;
1216         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1217         struct req_capsule      *pill = info->mti_pill;
1218         struct mdt_reint_record *rr   = &info->mti_rr;
1219         struct ptlrpc_request   *req  = mdt_info_req(info);
1220         struct md_op_spec       *sp   = &info->mti_spec;
1221         ENTRY;
1222
1223         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1224         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1225         if (rec == NULL)
1226                 RETURN(-EFAULT);
1227
1228         uc->mu_fsuid = rec->cr_fsuid;
1229         uc->mu_fsgid = rec->cr_fsgid;
1230         uc->mu_cap   = rec->cr_cap;
1231         uc->mu_suppgids[0] = rec->cr_suppgid1;
1232         uc->mu_suppgids[1] = rec->cr_suppgid2;
1233
1234         rr->rr_fid1   = &rec->cr_fid1;
1235         rr->rr_fid2   = &rec->cr_fid2;
1236         rr->rr_handle = &rec->cr_old_handle;
1237         attr->la_mode = rec->cr_mode;
1238         attr->la_rdev  = rec->cr_rdev;
1239         attr->la_uid   = rec->cr_fsuid;
1240         attr->la_gid   = rec->cr_fsgid;
1241         attr->la_ctime = rec->cr_time;
1242         attr->la_mtime = rec->cr_time;
1243         attr->la_atime = rec->cr_time;
1244         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1245                          LA_CTIME | LA_MTIME | LA_ATIME;
1246         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1247         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1248         /* Do not trigger ASSERTION if client miss to set such flags. */
1249         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1250                 RETURN(-EPROTO);
1251         info->mti_replayepoch = rec->cr_ioepoch;
1252
1253         info->mti_spec.sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
1254         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1255
1256         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1257                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1258                                  req_capsule_client_get(pill, &RMF_CAPA1));
1259         if (req_is_replay(req) &&
1260             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1261 #if 0
1262                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1263                                  req_capsule_client_get(pill, &RMF_CAPA2));
1264 #else
1265                 /*
1266                  * FIXME: capa in replay open request might have expired,
1267                  * bypass capa check. Security hole?
1268                  */
1269                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1270                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1271 #endif
1272         }
1273
1274         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1275         if (rr->rr_name == NULL)
1276                 RETURN(-EFAULT);
1277         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1278
1279         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1280                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1281                                                         RCL_CLIENT);
1282                 if (rr->rr_eadatalen > 0) {
1283                         rr->rr_eadata = req_capsule_client_get(pill,
1284                                                                &RMF_EADATA);
1285                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1286                         sp->u.sp_ea.eadata = rr->rr_eadata;
1287                         sp->no_create = !!req_is_replay(req);
1288                 }
1289
1290                 /*
1291                  * Client default md_size may be 0 right after client start,
1292                  * until all osc are connected, set here just some reasonable
1293                  * value to prevent misbehavior.
1294                  */
1295                 if (rr->rr_eadatalen == 0 &&
1296                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1297                         rr->rr_eadatalen = MIN_MD_SIZE;        }
1298
1299         RETURN(0);
1300 }
1301
1302 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1303 {
1304         struct mdt_reint_record   *rr   = &info->mti_rr;
1305         struct md_ucred           *uc   = mdt_ucred(info);
1306         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1307         struct req_capsule        *pill = info->mti_pill;
1308         struct mdt_rec_setxattr   *rec;
1309         ENTRY;
1310
1311
1312         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1313                          sizeof(struct mdt_rec_reint));
1314
1315         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1316         if (rec == NULL)
1317                 RETURN(-EFAULT);
1318
1319         uc->mu_fsuid  = rec->sx_fsuid;
1320         uc->mu_fsgid  = rec->sx_fsgid;
1321         uc->mu_cap    = rec->sx_cap;
1322         uc->mu_suppgids[0] = rec->sx_suppgid1;
1323         uc->mu_suppgids[1] = -1;
1324
1325         rr->rr_opcode = rec->sx_opcode;
1326         rr->rr_fid1   = &rec->sx_fid;
1327         attr->la_valid = rec->sx_valid;
1328         attr->la_ctime = rec->sx_time;
1329         attr->la_size = rec->sx_size;
1330         attr->la_flags = rec->sx_flags;
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         else
1336                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1337
1338         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1339         if (rr->rr_name == NULL)
1340                 RETURN(-EFAULT);
1341         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1342         LASSERT(rr->rr_namelen > 0);
1343
1344         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1345                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1346                                                         RCL_CLIENT);
1347                 if (rr->rr_eadatalen > 0) {
1348                         rr->rr_eadata = req_capsule_client_get(pill,
1349                                                                &RMF_EADATA);
1350                         if (rr->rr_eadata == NULL)
1351                                 RETURN(-EFAULT);
1352                 } else {
1353                         rr->rr_eadata = NULL;
1354                 }
1355         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1356                 CDEBUG(D_INFO, "no xattr data supplied\n");
1357                 RETURN(-EFAULT);
1358         }
1359
1360         RETURN(0);
1361 }
1362
1363
1364 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1365
1366 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1367         [REINT_SETATTR]  = mdt_setattr_unpack,
1368         [REINT_CREATE]   = mdt_create_unpack,
1369         [REINT_LINK]     = mdt_link_unpack,
1370         [REINT_UNLINK]   = mdt_unlink_unpack,
1371         [REINT_RENAME]   = mdt_rename_unpack,
1372         [REINT_OPEN]     = mdt_open_unpack,
1373         [REINT_SETXATTR] = mdt_setxattr_unpack
1374 };
1375
1376 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1377 {
1378         int rc;
1379         ENTRY;
1380
1381         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1382         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1383                 info->mti_rr.rr_opcode = op;
1384                 rc = mdt_reint_unpackers[op](info);
1385         } else {
1386                 CERROR("Unexpected opcode %d\n", op);
1387                 rc = -EFAULT;
1388         }
1389         RETURN(rc);
1390 }