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
  50 #ifndef EXPORT_SYMTAB
  51 # define EXPORT_SYMTAB
  52 #endif
  53 #define DEBUG_SUBSYSTEM S_MDS
  54
  55 #include "mdt_internal.h"
  56 #include <lnet/lib-lnet.h>
  57
  58
  59 typedef enum ucred_init_type {
  60         NONE_INIT       = 0,
  61         BODY_INIT       = 1,
  62         REC_INIT        = 2
  63 } ucred_init_type_t;
  64
  65 void mdt_exit_ucred(struct mdt_thread_info *info)
  66 {
  67         struct md_ucred   *uc  = mdt_ucred(info);
  68         struct mdt_device *mdt = info->mti_mdt;
  69
  70         if (uc->mu_valid != UCRED_INIT) {
  71                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
  72                 if (uc->mu_ginfo) {
  73                         cfs_put_group_info(uc->mu_ginfo);
  74                         uc->mu_ginfo = NULL;
  75                 }
  76                 if (uc->mu_identity) {
  77                         mdt_identity_put(mdt->mdt_identity_cache,
  78                                          uc->mu_identity);
  79                         uc->mu_identity = NULL;
  80                 }
  81                 uc->mu_valid = UCRED_INIT;
  82         }
  83 }
  84
  85 static int match_nosquash_list(cfs_rw_semaphore_t *sem,
  86                                cfs_list_t *nidlist,
  87                                lnet_nid_t peernid)
  88 {
  89         int rc;
  90         ENTRY;
  91         cfs_down_read(sem);
  92         rc = cfs_match_nid(peernid, nidlist);
  93         cfs_up_read(sem);
  94         RETURN(rc);
  95 }
  96
  97 /* root_squash for inter-MDS operations */
  98 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  99 {
 100         struct md_ucred *ucred = mdt_ucred(info);
 101         ENTRY;
 102
 103         if (!info->mti_mdt->mdt_squash_uid || ucred->mu_fsuid)
 104                 RETURN(0);
 105
 106         if (match_nosquash_list(&info->mti_mdt->mdt_squash_sem,
 107                                 &info->mti_mdt->mdt_nosquash_nids,
 108                                 peernid)) {
 109                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 110                        libcfs_nid2str(peernid));
 111                 RETURN(0);
 112         }
 113
 114         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 115                libcfs_nid2str(peernid),
 116                ucred->mu_fsuid, ucred->mu_fsgid, ucred->mu_cap,
 117                info->mti_mdt->mdt_squash_uid, info->mti_mdt->mdt_squash_gid,
 118                0);
 119
 120         ucred->mu_fsuid = info->mti_mdt->mdt_squash_uid;
 121         ucred->mu_fsgid = info->mti_mdt->mdt_squash_gid;
 122         ucred->mu_cap = 0;
 123         ucred->mu_suppgids[0] = -1;
 124         ucred->mu_suppgids[1] = -1;
 125
 126         RETURN(0);
 127 }
 128
 129 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 130                           void *buf)
 131 {
 132         struct ptlrpc_request   *req = mdt_info_req(info);
 133         struct mdt_device       *mdt = info->mti_mdt;
 134         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 135         struct md_ucred         *ucred = mdt_ucred(info);
 136         lnet_nid_t               peernid = req->rq_peer.nid;
 137         __u32                    perm = 0;
 138         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 139         int                      setuid;
 140         int                      setgid;
 141         int                      rc = 0;
 142
 143         ENTRY;
 144
 145         LASSERT(req->rq_auth_gss);
 146         LASSERT(!req->rq_auth_usr_mdt);
 147         LASSERT(req->rq_user_desc);
 148
 149         ucred->mu_valid = UCRED_INVALID;
 150
 151         ucred->mu_o_uid   = pud->pud_uid;
 152         ucred->mu_o_gid   = pud->pud_gid;
 153         ucred->mu_o_fsuid = pud->pud_fsuid;
 154         ucred->mu_o_fsgid = pud->pud_fsgid;
 155
 156         if (type == BODY_INIT) {
 157                 struct mdt_body *body = (struct mdt_body *)buf;
 158
 159                 ucred->mu_suppgids[0] = body->suppgid;
 160                 ucred->mu_suppgids[1] = -1;
 161         }
 162
 163         /* sanity check: we expect the uid which client claimed is true */
 164         if (remote) {
 165                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 166                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 167                         RETURN(-EACCES);
 168                 }
 169
 170                 if (ptlrpc_user_desc_do_idmap(req, pud))
 171                         RETURN(-EACCES);
 172
 173                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 174                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 175                                "while client claims %u:%u/%u:%u\n",
 176                                libcfs_nid2str(peernid), req->rq_auth_uid,
 177                                req->rq_auth_mapped_uid,
 178                                pud->pud_uid, pud->pud_gid,
 179                                pud->pud_fsuid, pud->pud_fsgid);
 180                         RETURN(-EACCES);
 181                 }
 182         } else {
 183                 if (req->rq_auth_uid != pud->pud_uid) {
 184                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 185                                "while client claims %u:%u/%u:%u\n",
 186                                libcfs_nid2str(peernid), req->rq_auth_uid,
 187                                pud->pud_uid, pud->pud_gid,
 188                                pud->pud_fsuid, pud->pud_fsgid);
 189                         RETURN(-EACCES);
 190                 }
 191         }
 192
 193         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 194                 if (remote) {
 195                         CDEBUG(D_SEC, "remote client must run with identity_get "
 196                                "enabled!\n");
 197                         RETURN(-EACCES);
 198                 } else {
 199                         ucred->mu_identity = NULL;
 200                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 201                                CFS_SETGRP_PERM;
 202                 }
 203         } else {
 204                 struct md_identity *identity;
 205
 206                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 207                                             pud->pud_uid);
 208                 if (IS_ERR(identity)) {
 209                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 210                                      !remote)) {
 211                                 ucred->mu_identity = NULL;
 212                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 213                                        CFS_SETGRP_PERM;
 214                         } else {
 215                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 216                                        pud->pud_uid);
 217                                 RETURN(-EACCES);
 218                         }
 219                 } else {
 220                         ucred->mu_identity = identity;
 221                         perm = mdt_identity_get_perm(ucred->mu_identity,
 222                                                      remote, peernid);
 223                 }
 224         }
 225
 226         /* find out the setuid/setgid attempt */
 227         setuid = (pud->pud_uid != pud->pud_fsuid);
 228         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 229                   (ucred->mu_identity &&
 230                   (pud->pud_gid != ucred->mu_identity->mi_gid)));
 231
 232         /* check permission of setuid */
 233         if (setuid && !(perm & CFS_SETUID_PERM)) {
 234                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 235                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 236                 GOTO(out, rc = -EACCES);
 237         }
 238
 239         /* check permission of setgid */
 240         if (setgid && !(perm & CFS_SETGID_PERM)) {
 241                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 242                        "from %s\n", pud->pud_uid, pud->pud_gid,
 243                        pud->pud_fsuid, pud->pud_fsgid,
 244                        ucred->mu_identity->mi_gid, libcfs_nid2str(peernid));
 245                 GOTO(out, rc = -EACCES);
 246         }
 247
 248         /*
 249          * NB: remote client not allowed to setgroups anyway.
 250          */
 251         if (!remote && perm & CFS_SETGRP_PERM) {
 252                 if (pud->pud_ngroups) {
 253                         /* setgroups for local client */
 254                         ucred->mu_ginfo = cfs_groups_alloc(pud->pud_ngroups);
 255                         if (!ucred->mu_ginfo) {
 256                                 CERROR("failed to alloc %d groups\n",
 257                                        pud->pud_ngroups);
 258                                 GOTO(out, rc = -ENOMEM);
 259                         }
 260
 261                         lustre_groups_from_list(ucred->mu_ginfo,
 262                                                 pud->pud_groups);
 263                         lustre_groups_sort(ucred->mu_ginfo);
 264                 } else {
 265                         ucred->mu_ginfo = NULL;
 266                 }
 267         } else {
 268                 ucred->mu_suppgids[0] = -1;
 269                 ucred->mu_suppgids[1] = -1;
 270                 ucred->mu_ginfo = NULL;
 271         }
 272
 273         ucred->mu_uid   = pud->pud_uid;
 274         ucred->mu_gid   = pud->pud_gid;
 275         ucred->mu_fsuid = pud->pud_fsuid;
 276         ucred->mu_fsgid = pud->pud_fsgid;
 277
 278         /* process root_squash here. */
 279         mdt_root_squash(info, peernid);
 280
 281         /* remove fs privilege for non-root user. */
 282         if (ucred->mu_fsuid)
 283                 ucred->mu_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 284         else
 285                 ucred->mu_cap = pud->pud_cap;
 286         if (remote && !(perm & CFS_RMTOWN_PERM))
 287                 ucred->mu_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 288                                    CFS_CAP_CHOWN_MASK);
 289         ucred->mu_valid = UCRED_NEW;
 290
 291         EXIT;
 292
 293 out:
 294         if (rc) {
 295                 if (ucred->mu_ginfo) {
 296                         cfs_put_group_info(ucred->mu_ginfo);
 297                         ucred->mu_ginfo = NULL;
 298                 }
 299                 if (ucred->mu_identity) {
 300                         mdt_identity_put(mdt->mdt_identity_cache,
 301                                          ucred->mu_identity);
 302                         ucred->mu_identity = NULL;
 303                 }
 304         }
 305
 306         return rc;
 307 }
 308
 309 int mdt_check_ucred(struct mdt_thread_info *info)
 310 {
 311         struct ptlrpc_request   *req = mdt_info_req(info);
 312         struct mdt_device       *mdt = info->mti_mdt;
 313         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 314         struct md_ucred         *ucred = mdt_ucred(info);
 315         struct md_identity      *identity = NULL;
 316         lnet_nid_t               peernid = req->rq_peer.nid;
 317         __u32                    perm = 0;
 318         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 319         int                      setuid;
 320         int                      setgid;
 321         int                      rc = 0;
 322
 323         ENTRY;
 324
 325         if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
 326                 RETURN(0);
 327
 328         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 329                 RETURN(0);
 330
 331         /* sanity check: if we use strong authentication, we expect the
 332          * uid which client claimed is true */
 333         if (remote) {
 334                 if (req->rq_auth_mapped_uid == INVALID_UID) {
 335                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 336                         RETURN(-EACCES);
 337                 }
 338
 339                 if (ptlrpc_user_desc_do_idmap(req, pud))
 340                         RETURN(-EACCES);
 341
 342                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 343                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 344                                "while client claims %u:%u/%u:%u\n",
 345                                libcfs_nid2str(peernid), req->rq_auth_uid,
 346                                req->rq_auth_mapped_uid,
 347                                pud->pud_uid, pud->pud_gid,
 348                                pud->pud_fsuid, pud->pud_fsgid);
 349                         RETURN(-EACCES);
 350                 }
 351         } else {
 352                 if (req->rq_auth_uid != pud->pud_uid) {
 353                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 354                                "while client claims %u:%u/%u:%u\n",
 355                                libcfs_nid2str(peernid), req->rq_auth_uid,
 356                                pud->pud_uid, pud->pud_gid,
 357                                pud->pud_fsuid, pud->pud_fsgid);
 358                         RETURN(-EACCES);
 359                 }
 360         }
 361
 362         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 363                 if (remote) {
 364                         CDEBUG(D_SEC, "remote client must run with identity_get "
 365                                "enabled!\n");
 366                         RETURN(-EACCES);
 367                 }
 368                 RETURN(0);
 369         }
 370
 371         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 372         if (IS_ERR(identity)) {
 373                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 374                              !remote)) {
 375                         RETURN(0);
 376                 } else {
 377                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 378                                pud->pud_uid);
 379                         RETURN(-EACCES);
 380                }
 381         }
 382
 383         perm = mdt_identity_get_perm(identity, remote, peernid);
 384         /* find out the setuid/setgid attempt */
 385         setuid = (pud->pud_uid != pud->pud_fsuid);
 386         setgid = (pud->pud_gid != pud->pud_fsgid ||
 387                   pud->pud_gid != identity->mi_gid);
 388
 389         /* check permission of setuid */
 390         if (setuid && !(perm & CFS_SETUID_PERM)) {
 391                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 392                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 393                 GOTO(out, rc = -EACCES);
 394         }
 395
 396         /* check permission of setgid */
 397         if (setgid && !(perm & CFS_SETGID_PERM)) {
 398                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 399                        "from %s\n", pud->pud_uid, pud->pud_gid,
 400                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 401                        libcfs_nid2str(peernid));
 402                 GOTO(out, rc = -EACCES);
 403         }
 404
 405         EXIT;
 406
 407 out:
 408         mdt_identity_put(mdt->mdt_identity_cache, identity);
 409         return rc;
 410 }
 411
 412 static int old_init_ucred(struct mdt_thread_info *info,
 413                           struct mdt_body *body)
 414 {
 415         struct md_ucred *uc = mdt_ucred(info);
 416         struct mdt_device  *mdt = info->mti_mdt;
 417         struct md_identity *identity = NULL;
 418
 419         ENTRY;
 420
 421         uc->mu_valid = UCRED_INVALID;
 422         uc->mu_o_uid = uc->mu_uid = body->uid;
 423         uc->mu_o_gid = uc->mu_gid = body->gid;
 424         uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
 425         uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
 426         uc->mu_suppgids[0] = body->suppgid;
 427         uc->mu_suppgids[1] = -1;
 428         uc->mu_ginfo = NULL;
 429         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 430                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 431                                             uc->mu_fsuid);
 432                 if (IS_ERR(identity)) {
 433                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 434                                 identity = NULL;
 435                         } else {
 436                                 CDEBUG(D_SEC, "Deny access without identity: "
 437                                        "uid %u\n", uc->mu_fsuid);
 438                                 RETURN(-EACCES);
 439                         }
 440                 }
 441         }
 442         uc->mu_identity = identity;
 443
 444         /* process root_squash here. */
 445         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 446
 447         /* remove fs privilege for non-root user. */
 448         if (uc->mu_fsuid)
 449                 uc->mu_cap = body->capability & ~CFS_CAP_FS_MASK;
 450         else
 451                 uc->mu_cap = body->capability;
 452         uc->mu_valid = UCRED_OLD;
 453
 454         RETURN(0);
 455 }
 456
 457 static int old_init_ucred_reint(struct mdt_thread_info *info)
 458 {
 459         struct md_ucred *uc = mdt_ucred(info);
 460         struct mdt_device  *mdt = info->mti_mdt;
 461         struct md_identity *identity = NULL;
 462
 463         ENTRY;
 464
 465         uc->mu_valid = UCRED_INVALID;
 466         uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
 467         uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
 468         uc->mu_ginfo = NULL;
 469         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 470                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 471                                             uc->mu_fsuid);
 472                 if (IS_ERR(identity)) {
 473                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 474                                 identity = NULL;
 475                         } else {
 476                                 CDEBUG(D_SEC, "Deny access without identity: "
 477                                        "uid %u\n", uc->mu_fsuid);
 478                                 RETURN(-EACCES);
 479                         }
 480                 }
 481         }
 482         uc->mu_identity = identity;
 483
 484         /* process root_squash here. */
 485         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 486
 487         /* remove fs privilege for non-root user. */
 488         if (uc->mu_fsuid)
 489                 uc->mu_cap &= ~CFS_CAP_FS_MASK;
 490         uc->mu_valid = UCRED_OLD;
 491
 492         RETURN(0);
 493 }
 494
 495 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 496 {
 497         struct ptlrpc_request *req = mdt_info_req(info);
 498         struct md_ucred       *uc  = mdt_ucred(info);
 499
 500         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 501                 return 0;
 502
 503         mdt_exit_ucred(info);
 504
 505         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 506                 return old_init_ucred(info, body);
 507         else
 508                 return new_init_ucred(info, BODY_INIT, body);
 509 }
 510
 511 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 512 {
 513         struct ptlrpc_request *req = mdt_info_req(info);
 514         struct md_ucred       *uc  = mdt_ucred(info);
 515
 516         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 517                 return 0;
 518
 519         mdt_exit_ucred(info);
 520
 521         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 522                 return old_init_ucred_reint(info);
 523         else
 524                 return new_init_ucred(info, REC_INIT, NULL);
 525 }
 526
 527 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 528 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 529 {
 530         const struct lov_ost_data_v1 *lod;
 531         int                           i;
 532         __u16                         count;
 533
 534         count = le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 535
 536         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
 537                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 538                le32_to_cpu(lmm->lmm_pattern));
 539         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 540                le32_to_cpu(lmm->lmm_stripe_size), count);
 541         if (count == LOV_ALL_STRIPES)
 542                 return;
 543         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 544         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++)
 545                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 546                        i, le32_to_cpu(lod->l_ost_idx),
 547                        le64_to_cpu(lod->l_object_seq),
 548                        le64_to_cpu(lod->l_object_id));
 549 }
 550
 551 /* Shrink and/or grow reply buffers */
 552 int mdt_fix_reply(struct mdt_thread_info *info)
 553 {
 554         struct req_capsule *pill = info->mti_pill;
 555         struct mdt_body    *body;
 556         int                md_size, md_packed = 0;
 557         int                acl_size;
 558         int                rc = 0;
 559         ENTRY;
 560
 561         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 562         LASSERT(body != NULL);
 563
 564         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 565                 md_size = body->eadatasize;
 566         else
 567                 md_size = 0;
 568
 569         acl_size = body->aclsize;
 570
 571         /* this replay - not send info to client */
 572         if (info->mti_spec.no_create == 1) {
 573                 md_size = 0;
 574                 acl_size = 0;
 575         }
 576
 577         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 578                         " MDSCAPA = %llx, OSSCAPA = %llx\n",
 579                         md_size, acl_size,
 580                         (unsigned long long)(body->valid & OBD_MD_FLMDSCAPA),
 581                         (unsigned long long)(body->valid & OBD_MD_FLOSSCAPA));
 582 /*
 583             &RMF_MDT_BODY,
 584             &RMF_MDT_MD,
 585             &RMF_ACL, or &RMF_LOGCOOKIES
 586 (optional)  &RMF_CAPA1,
 587 (optional)  &RMF_CAPA2,
 588 (optional)  something else
 589 */
 590
 591         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 592          * buffers before growing it */
 593         if (info->mti_attr.ma_big_lmm_used) {
 594                 LASSERT(req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER));
 595                 md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 596                                                  RCL_SERVER);
 597                 LASSERT(md_packed > 0);
 598                 /* buffer must be allocated separately */
 599                 LASSERT(info->mti_attr.ma_lmm !=
 600                         req_capsule_server_get(pill, &RMF_MDT_MD));
 601                 req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 602                 /* free big lmm if md_size is not needed */
 603                 if (md_size == 0)
 604                         info->mti_attr.ma_big_lmm_used = 0;
 605         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 606                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 607         }
 608
 609         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 610                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 611         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 612                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 613                                    acl_size, RCL_SERVER);
 614
 615         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 616             !(body->valid & OBD_MD_FLMDSCAPA))
 617                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 618
 619         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 620             !(body->valid & OBD_MD_FLOSSCAPA))
 621                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 622
 623         /*
 624          * Some more field should be shrinked if needed.
 625          * This should be done by those who added fields to reply message.
 626          */
 627
 628         /* Grow MD buffer if needed finally */
 629         if (info->mti_attr.ma_big_lmm_used) {
 630                 void *lmm;
 631
 632                 LASSERT(md_size > md_packed);
 633                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 634                        md_size - md_packed);
 635                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 636                 if (rc) {
 637                         /* we can't answer with proper LOV EA, drop flags,
 638                          * the rc is also returned so this request is
 639                          * considered as failed */
 640                         body->valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 641                         /* don't return transno along with error */
 642                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 643                 } else {
 644                         /* now we need to pack right LOV EA */
 645                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 646                         LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 647                                                      RCL_SERVER) ==
 648                                 info->mti_attr.ma_lmm_size);
 649                         memcpy(lmm, info->mti_attr.ma_lmm,
 650                                info->mti_attr.ma_lmm_size);
 651                 }
 652                 /* update mdt_max_mdsize so clients will be aware about that */
 653                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 654                         info->mti_mdt->mdt_max_mdsize =
 655                                                     info->mti_attr.ma_lmm_size;
 656                 info->mti_attr.ma_big_lmm_used = 0;
 657         }
 658         RETURN(rc);
 659 }
 660
 661
 662 /* if object is dying, pack the lov/llog data,
 663  * parameter info->mti_attr should be valid at this point! */
 664 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 665                            const struct md_attr *ma)
 666 {
 667         struct mdt_body       *repbody;
 668         const struct lu_attr *la = &ma->ma_attr;
 669         int rc;
 670         ENTRY;
 671
 672         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 673         LASSERT(repbody != NULL);
 674
 675         if (ma->ma_valid & MA_INODE)
 676                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 677
 678         if (ma->ma_valid & MA_LOV) {
 679                 __u32 mode;
 680
 681                 if (mdt_object_exists(mo) < 0)
 682                         /* If it is a remote object, and we do not retrieve
 683                          * EA back unlink reg file*/
 684                         mode = S_IFREG;
 685                 else
 686                         mode = lu_object_attr(&mo->mot_obj.mo_lu);
 687
 688                 LASSERT(ma->ma_lmm_size);
 689                 mdt_dump_lmm(D_INFO, ma->ma_lmm);
 690                 repbody->eadatasize = ma->ma_lmm_size;
 691                 if (S_ISREG(mode))
 692                         repbody->valid |= OBD_MD_FLEASIZE;
 693                 else if (S_ISDIR(mode))
 694                         repbody->valid |= OBD_MD_FLDIREA;
 695                 else
 696                         LBUG();
 697         }
 698
 699         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 700                 repbody->aclsize = ma->ma_cookie_size;
 701                 repbody->valid |= OBD_MD_FLCOOKIE;
 702         }
 703
 704         if (info->mti_mdt->mdt_opts.mo_oss_capa &&
 705             info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
 706             repbody->valid & OBD_MD_FLEASIZE) {
 707                 struct lustre_capa *capa;
 708
 709                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 710                 LASSERT(capa);
 711                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 712                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 713                 if (rc)
 714                         RETURN(rc);
 715
 716                 repbody->valid |= OBD_MD_FLOSSCAPA;
 717         }
 718
 719         RETURN(0);
 720 }
 721
 722 static inline unsigned int attr_unpack(__u64 sa_valid) {
 723         unsigned int ia_valid = 0;
 724
 725         if (sa_valid & MDS_ATTR_MODE)
 726                 ia_valid |= ATTR_MODE;
 727         if (sa_valid & MDS_ATTR_UID)
 728                 ia_valid |= ATTR_UID;
 729         if (sa_valid & MDS_ATTR_GID)
 730                 ia_valid |= ATTR_GID;
 731         if (sa_valid & MDS_ATTR_SIZE)
 732                 ia_valid |= ATTR_SIZE;
 733         if (sa_valid & MDS_ATTR_ATIME)
 734                 ia_valid |= ATTR_ATIME;
 735         if (sa_valid & MDS_ATTR_MTIME)
 736                 ia_valid |= ATTR_MTIME;
 737         if (sa_valid & MDS_ATTR_CTIME)
 738                 ia_valid |= ATTR_CTIME;
 739         if (sa_valid & MDS_ATTR_ATIME_SET)
 740                 ia_valid |= ATTR_ATIME_SET;
 741         if (sa_valid & MDS_ATTR_MTIME_SET)
 742                 ia_valid |= ATTR_MTIME_SET;
 743         if (sa_valid & MDS_ATTR_FORCE)
 744                 ia_valid |= ATTR_FORCE;
 745         if (sa_valid & MDS_ATTR_ATTR_FLAG)
 746                 ia_valid |= ATTR_ATTR_FLAG;
 747         if (sa_valid & MDS_ATTR_KILL_SUID)
 748                 ia_valid |=  ATTR_KILL_SUID;
 749         if (sa_valid & MDS_ATTR_KILL_SGID)
 750                 ia_valid |= ATTR_KILL_SGID;
 751         if (sa_valid & MDS_ATTR_CTIME_SET)
 752                 ia_valid |= ATTR_CTIME_SET;
 753         if (sa_valid & MDS_ATTR_FROM_OPEN)
 754                 ia_valid |= ATTR_FROM_OPEN;
 755         if (sa_valid & MDS_ATTR_BLOCKS)
 756                 ia_valid |= ATTR_BLOCKS;
 757         if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
 758                 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
 759         return ia_valid;
 760 }
 761
 762 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 763                                   struct md_attr *ma)
 764 {
 765         __u64 out;
 766
 767         out = 0;
 768         if (in & ATTR_MODE)
 769                 out |= LA_MODE;
 770         if (in & ATTR_UID)
 771                 out |= LA_UID;
 772         if (in & ATTR_GID)
 773                 out |= LA_GID;
 774         if (in & ATTR_SIZE)
 775                 out |= LA_SIZE;
 776         if (in & ATTR_BLOCKS)
 777                 out |= LA_BLOCKS;
 778
 779         if (in & ATTR_FROM_OPEN)
 780                 rr->rr_flags |= MRF_OPEN_TRUNC;
 781
 782         if (in & ATTR_ATIME_SET)
 783                 out |= LA_ATIME;
 784
 785         if (in & ATTR_CTIME_SET)
 786                 out |= LA_CTIME;
 787
 788         if (in & ATTR_MTIME_SET)
 789                 out |= LA_MTIME;
 790
 791         if (in & ATTR_ATTR_FLAG)
 792                 out |= LA_FLAGS;
 793
 794         if (in & ATTR_KILL_SUID)
 795                 out |= LA_KILL_SUID;
 796
 797         if (in & ATTR_KILL_SGID)
 798                 out |= LA_KILL_SGID;
 799
 800         if (in & MDS_OPEN_OWNEROVERRIDE)
 801                 ma->ma_attr_flags |= MDS_OPEN_OWNEROVERRIDE;
 802
 803         if (in & ATTR_FORCE)
 804                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 805
 806         /*XXX need ATTR_RAW?*/
 807         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 808                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 809                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 810                 ATTR_ATTR_FLAG|ATTR_RAW|MDS_OPEN_OWNEROVERRIDE|
 811                 ATTR_FORCE|ATTR_KILL_SUID|ATTR_KILL_SGID);
 812         if (in != 0)
 813                 CERROR("Unknown attr bits: "LPX64"\n", in);
 814         return out;
 815 }
 816 /* unpacking */
 817
 818 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 819 {
 820         struct md_ucred         *uc  = mdt_ucred(info);
 821         struct md_attr          *ma = &info->mti_attr;
 822         struct lu_attr          *la = &ma->ma_attr;
 823         struct req_capsule      *pill = info->mti_pill;
 824         struct mdt_reint_record *rr = &info->mti_rr;
 825         struct mdt_rec_setattr  *rec;
 826         ENTRY;
 827
 828         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 829         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 830         if (rec == NULL)
 831                 RETURN(-EFAULT);
 832
 833         uc->mu_fsuid = rec->sa_fsuid;
 834         uc->mu_fsgid = rec->sa_fsgid;
 835         uc->mu_cap   = rec->sa_cap;
 836         uc->mu_suppgids[0] = rec->sa_suppgid;
 837         uc->mu_suppgids[1] = -1;
 838
 839         rr->rr_fid1 = &rec->sa_fid;
 840         la->la_valid = mdt_attr_valid_xlate(attr_unpack(rec->sa_valid), rr, ma);
 841         la->la_mode  = rec->sa_mode;
 842         la->la_flags = rec->sa_attr_flags;
 843         la->la_uid   = rec->sa_uid;
 844         la->la_gid   = rec->sa_gid;
 845         la->la_size  = rec->sa_size;
 846         la->la_blocks = rec->sa_blocks;
 847         la->la_ctime = rec->sa_ctime;
 848         la->la_atime = rec->sa_atime;
 849         la->la_mtime = rec->sa_mtime;
 850         ma->ma_valid = MA_INODE;
 851
 852         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 853                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 854                                  req_capsule_client_get(pill, &RMF_CAPA1));
 855
 856         RETURN(0);
 857 }
 858
 859 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 860 {
 861         struct req_capsule *pill = info->mti_pill;
 862         ENTRY;
 863
 864         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 865                 info->mti_ioepoch =
 866                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 867         else
 868                 info->mti_ioepoch = NULL;
 869         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 870 }
 871
 872 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 873         struct req_capsule      *pill = info->mti_pill;
 874
 875         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 876                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 877                 if (info->mti_dlm_req == NULL)
 878                         RETURN(-EFAULT);
 879         }
 880
 881         RETURN(0);
 882 }
 883
 884 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 885 {
 886         struct mdt_reint_record *rr = &info->mti_rr;
 887         struct md_attr          *ma = &info->mti_attr;
 888         struct req_capsule      *pill = info->mti_pill;
 889         int rc;
 890         ENTRY;
 891
 892         rc = mdt_setattr_unpack_rec(info);
 893         if (rc)
 894                 RETURN(rc);
 895
 896         /* Epoch may be absent */
 897         mdt_ioepoch_unpack(info);
 898
 899         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 900                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 901                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 902                                                         RCL_CLIENT);
 903                 ma->ma_lmm_size = rr->rr_eadatalen;
 904                 if (ma->ma_lmm_size > 0) {
 905                         ma->ma_lmm = (void *)rr->rr_eadata;
 906                         ma->ma_valid |= MA_LOV;
 907                 }
 908         }
 909
 910         ma->ma_cookie_size = req_capsule_get_size(pill, &RMF_LOGCOOKIES,
 911                                                   RCL_CLIENT);
 912         if (ma->ma_cookie_size) {
 913                 ma->ma_cookie = req_capsule_client_get(pill, &RMF_LOGCOOKIES);
 914                 ma->ma_valid |= MA_COOKIE;
 915         }
 916
 917         rc = mdt_dlmreq_unpack(info);
 918         RETURN(rc);
 919 }
 920
 921 int mdt_close_unpack(struct mdt_thread_info *info)
 922 {
 923         int rc;
 924         ENTRY;
 925
 926         rc = mdt_ioepoch_unpack(info);
 927         if (rc)
 928                 RETURN(rc);
 929
 930         RETURN(mdt_setattr_unpack_rec(info));
 931 }
 932
 933 static int mdt_create_unpack(struct mdt_thread_info *info)
 934 {
 935         struct md_ucred         *uc  = mdt_ucred(info);
 936         struct mdt_rec_create   *rec;
 937         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 938         struct mdt_reint_record *rr = &info->mti_rr;
 939         struct req_capsule      *pill = info->mti_pill;
 940         struct md_op_spec       *sp = &info->mti_spec;
 941         int rc;
 942         ENTRY;
 943
 944         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
 945         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 946         if (rec == NULL)
 947                 RETURN(-EFAULT);
 948
 949         uc->mu_fsuid = rec->cr_fsuid;
 950         uc->mu_fsgid = rec->cr_fsgid;
 951         uc->mu_cap   = rec->cr_cap;
 952         uc->mu_suppgids[0] = rec->cr_suppgid1;
 953         uc->mu_suppgids[1] = -1;
 954
 955         rr->rr_fid1 = &rec->cr_fid1;
 956         rr->rr_fid2 = &rec->cr_fid2;
 957         attr->la_mode = rec->cr_mode;
 958         attr->la_rdev  = rec->cr_rdev;
 959         attr->la_uid   = rec->cr_fsuid;
 960         attr->la_gid   = rec->cr_fsgid;
 961         attr->la_ctime = rec->cr_time;
 962         attr->la_mtime = rec->cr_time;
 963         attr->la_atime = rec->cr_time;
 964         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
 965                          LA_CTIME | LA_MTIME | LA_ATIME;
 966         memset(&sp->u, 0, sizeof(sp->u));
 967         sp->sp_cr_flags = get_mrc_cr_flags(rec);
 968         sp->sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
 969         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
 970
 971         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 972                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 973                                  req_capsule_client_get(pill, &RMF_CAPA1));
 974         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 975
 976         if (!info->mti_cross_ref) {
 977                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 978                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME,
 979                                                       RCL_CLIENT) - 1;
 980                 LASSERT(rr->rr_name && rr->rr_namelen > 0);
 981         } else {
 982                 rr->rr_name = NULL;
 983                 rr->rr_namelen = 0;
 984         }
 985
 986 #ifdef CONFIG_FS_POSIX_ACL
 987         if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
 988                 if (S_ISDIR(attr->la_mode))
 989                         sp->u.sp_pfid = rr->rr_fid1;
 990                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 991                 LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
 992                                                   RCL_CLIENT));
 993                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 994                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 995                                                         RCL_CLIENT);
 996                 sp->u.sp_ea.eadata = rr->rr_eadata;
 997                 sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
 998                 sp->u.sp_ea.fid = rr->rr_fid1;
 999                 RETURN(0);
1000         }
1001 #endif
1002         if (S_ISDIR(attr->la_mode)) {
1003                 /* pass parent fid for cross-ref cases */
1004                 sp->u.sp_pfid = rr->rr_fid1;
1005                 if (sp->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1006                         /* create salve object req, need
1007                          * unpack split ea here
1008                          */
1009                        req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SLAVE);
1010                        LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
1011                                                          RCL_CLIENT));
1012                        rr->rr_eadata = req_capsule_client_get(pill,
1013                                                               &RMF_EADATA);
1014                        rr->rr_eadatalen = req_capsule_get_size(pill,
1015                                                                &RMF_EADATA,
1016                                                                RCL_CLIENT);
1017                        sp->u.sp_ea.eadata = rr->rr_eadata;
1018                        sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1019                        sp->u.sp_ea.fid = rr->rr_fid1;
1020                        RETURN(0);
1021                 }
1022                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1023         } else if (S_ISLNK(attr->la_mode)) {
1024                 const char *tgt = NULL;
1025
1026                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1027                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1028                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1029                         sp->u.sp_symname = tgt;
1030                 }
1031                 if (tgt == NULL)
1032                         RETURN(-EFAULT);
1033         } else {
1034                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1035         }
1036         rc = mdt_dlmreq_unpack(info);
1037         RETURN(rc);
1038 }
1039
1040 static int mdt_link_unpack(struct mdt_thread_info *info)
1041 {
1042         struct md_ucred         *uc  = mdt_ucred(info);
1043         struct mdt_rec_link     *rec;
1044         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1045         struct mdt_reint_record *rr = &info->mti_rr;
1046         struct req_capsule      *pill = info->mti_pill;
1047         int rc;
1048         ENTRY;
1049
1050         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1051         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1052         if (rec == NULL)
1053                 RETURN(-EFAULT);
1054
1055         uc->mu_fsuid = rec->lk_fsuid;
1056         uc->mu_fsgid = rec->lk_fsgid;
1057         uc->mu_cap   = rec->lk_cap;
1058         uc->mu_suppgids[0] = rec->lk_suppgid1;
1059         uc->mu_suppgids[1] = rec->lk_suppgid2;
1060
1061         attr->la_uid = rec->lk_fsuid;
1062         attr->la_gid = rec->lk_fsgid;
1063         rr->rr_fid1 = &rec->lk_fid1;
1064         rr->rr_fid2 = &rec->lk_fid2;
1065         attr->la_ctime = rec->lk_time;
1066         attr->la_mtime = rec->lk_time;
1067         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1068
1069         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1070                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1071                                  req_capsule_client_get(pill, &RMF_CAPA1));
1072         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1073                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1074                                  req_capsule_client_get(pill, &RMF_CAPA2));
1075
1076         info->mti_spec.sp_ck_split = !!(rec->lk_bias & MDS_CHECK_SPLIT);
1077         info->mti_cross_ref = !!(rec->lk_bias & MDS_CROSS_REF);
1078         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1079         if (rr->rr_name == NULL)
1080                 RETURN(-EFAULT);
1081         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1082         if (!info->mti_cross_ref)
1083                 LASSERT(rr->rr_namelen > 0);
1084
1085         rc = mdt_dlmreq_unpack(info);
1086         RETURN(rc);
1087 }
1088
1089 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1090 {
1091         struct md_ucred         *uc  = mdt_ucred(info);
1092         struct mdt_rec_unlink   *rec;
1093         struct md_attr          *ma = &info->mti_attr;
1094         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1095         struct mdt_reint_record *rr = &info->mti_rr;
1096         struct req_capsule      *pill = info->mti_pill;
1097         int rc;
1098         ENTRY;
1099
1100         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1101         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1102         if (rec == NULL)
1103                 RETURN(-EFAULT);
1104
1105         uc->mu_fsuid = rec->ul_fsuid;
1106         uc->mu_fsgid = rec->ul_fsgid;
1107         uc->mu_cap   = rec->ul_cap;
1108         uc->mu_suppgids[0] = rec->ul_suppgid1;
1109         uc->mu_suppgids[1] = -1;
1110
1111         attr->la_uid = rec->ul_fsuid;
1112         attr->la_gid = rec->ul_fsgid;
1113         rr->rr_fid1 = &rec->ul_fid1;
1114         rr->rr_fid2 = &rec->ul_fid2;
1115         attr->la_ctime = rec->ul_time;
1116         attr->la_mtime = rec->ul_time;
1117         attr->la_mode  = rec->ul_mode;
1118         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1119
1120         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1121                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1122                                  req_capsule_client_get(pill, &RMF_CAPA1));
1123
1124         info->mti_cross_ref = !!(rec->ul_bias & MDS_CROSS_REF);
1125         if (!info->mti_cross_ref) {
1126                 rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1127                 rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1128                 if (rr->rr_name == NULL || rr->rr_namelen == 0)
1129                         RETURN(-EFAULT);
1130         } else {
1131                 rr->rr_name = NULL;
1132                 rr->rr_namelen = 0;
1133         }
1134         info->mti_spec.sp_ck_split = !!(rec->ul_bias & MDS_CHECK_SPLIT);
1135         if (rec->ul_bias & MDS_VTX_BYPASS)
1136                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1137         else
1138                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1139
1140         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1141         /* last unlink need LOV EA sent back */
1142         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1143
1144         rc = mdt_dlmreq_unpack(info);
1145         RETURN(rc);
1146 }
1147
1148 static int mdt_rename_unpack(struct mdt_thread_info *info)
1149 {
1150         struct md_ucred         *uc = mdt_ucred(info);
1151         struct mdt_rec_rename   *rec;
1152         struct md_attr          *ma = &info->mti_attr;
1153         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1154         struct mdt_reint_record *rr = &info->mti_rr;
1155         struct req_capsule      *pill = info->mti_pill;
1156         int rc;
1157         ENTRY;
1158
1159         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1160         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1161         if (rec == NULL)
1162                 RETURN(-EFAULT);
1163
1164         uc->mu_fsuid = rec->rn_fsuid;
1165         uc->mu_fsgid = rec->rn_fsgid;
1166         uc->mu_cap   = rec->rn_cap;
1167         uc->mu_suppgids[0] = rec->rn_suppgid1;
1168         uc->mu_suppgids[1] = rec->rn_suppgid2;
1169
1170         attr->la_uid = rec->rn_fsuid;
1171         attr->la_gid = rec->rn_fsgid;
1172         rr->rr_fid1 = &rec->rn_fid1;
1173         rr->rr_fid2 = &rec->rn_fid2;
1174         attr->la_ctime = rec->rn_time;
1175         attr->la_mtime = rec->rn_time;
1176         /* rename_tgt contains the mode already */
1177         attr->la_mode = rec->rn_mode;
1178         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1179
1180         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1181                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1182                                  req_capsule_client_get(pill, &RMF_CAPA1));
1183         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1184                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1185                                  req_capsule_client_get(pill, &RMF_CAPA2));
1186
1187         info->mti_spec.sp_ck_split = !!(rec->rn_bias & MDS_CHECK_SPLIT);
1188         info->mti_cross_ref = !!(rec->rn_bias & MDS_CROSS_REF);
1189         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1190         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1191         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
1192                 RETURN(-EFAULT);
1193         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1194         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT) - 1;
1195         if (!info->mti_cross_ref)
1196                 LASSERT(rr->rr_namelen > 0 && rr->rr_tgtlen > 0);
1197         if (rec->rn_bias & MDS_VTX_BYPASS)
1198                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1199         else
1200                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1201
1202         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1203         /* rename may contain unlink so we might need LOV EA sent back */
1204         rr->rr_eadatalen = info->mti_mdt->mdt_max_mdsize;
1205
1206         rc = mdt_dlmreq_unpack(info);
1207         RETURN(rc);
1208 }
1209
1210 static int mdt_open_unpack(struct mdt_thread_info *info)
1211 {
1212         struct md_ucred         *uc = mdt_ucred(info);
1213         struct mdt_rec_create   *rec;
1214         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1215         struct req_capsule      *pill = info->mti_pill;
1216         struct mdt_reint_record *rr   = &info->mti_rr;
1217         struct ptlrpc_request   *req  = mdt_info_req(info);
1218         struct md_op_spec       *sp   = &info->mti_spec;
1219         ENTRY;
1220
1221         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1222         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1223         if (rec == NULL)
1224                 RETURN(-EFAULT);
1225
1226         uc->mu_fsuid = rec->cr_fsuid;
1227         uc->mu_fsgid = rec->cr_fsgid;
1228         uc->mu_cap   = rec->cr_cap;
1229         uc->mu_suppgids[0] = rec->cr_suppgid1;
1230         uc->mu_suppgids[1] = rec->cr_suppgid2;
1231
1232         rr->rr_fid1   = &rec->cr_fid1;
1233         rr->rr_fid2   = &rec->cr_fid2;
1234         rr->rr_handle = &rec->cr_old_handle;
1235         attr->la_mode = rec->cr_mode;
1236         attr->la_rdev  = rec->cr_rdev;
1237         attr->la_uid   = rec->cr_fsuid;
1238         attr->la_gid   = rec->cr_fsgid;
1239         attr->la_ctime = rec->cr_time;
1240         attr->la_mtime = rec->cr_time;
1241         attr->la_atime = rec->cr_time;
1242         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1243                          LA_CTIME | LA_MTIME | LA_ATIME;
1244         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1245         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1246         /* Do not trigger ASSERTION if client miss to set such flags. */
1247         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1248                 RETURN(-EPROTO);
1249         info->mti_replayepoch = rec->cr_ioepoch;
1250
1251         info->mti_spec.sp_ck_split = !!(rec->cr_bias & MDS_CHECK_SPLIT);
1252         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1253
1254         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1255                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1256                                  req_capsule_client_get(pill, &RMF_CAPA1));
1257         if (req_is_replay(req) &&
1258             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1259 #if 0
1260                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1261                                  req_capsule_client_get(pill, &RMF_CAPA2));
1262 #else
1263                 /*
1264                  * FIXME: capa in replay open request might have expired,
1265                  * bypass capa check. Security hole?
1266                  */
1267                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1268                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1269 #endif
1270         }
1271
1272         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1273         if (rr->rr_name == NULL)
1274                 RETURN(-EFAULT);
1275         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1276
1277         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1278                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1279                                                         RCL_CLIENT);
1280                 if (rr->rr_eadatalen > 0) {
1281                         rr->rr_eadata = req_capsule_client_get(pill,
1282                                                                &RMF_EADATA);
1283                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1284                         sp->u.sp_ea.eadata = rr->rr_eadata;
1285                         sp->no_create = !!req_is_replay(req);
1286                 }
1287
1288                 /*
1289                  * Client default md_size may be 0 right after client start,
1290                  * until all osc are connected, set here just some reasonable
1291                  * value to prevent misbehavior.
1292                  */
1293                 if (rr->rr_eadatalen == 0 &&
1294                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1295                         rr->rr_eadatalen = MIN_MD_SIZE;        }
1296
1297         RETURN(0);
1298 }
1299
1300 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1301 {
1302         struct mdt_reint_record   *rr   = &info->mti_rr;
1303         struct md_ucred           *uc   = mdt_ucred(info);
1304         struct lu_attr            *attr = &info->mti_attr.ma_attr;
1305         struct req_capsule        *pill = info->mti_pill;
1306         struct mdt_rec_setxattr   *rec;
1307         ENTRY;
1308
1309
1310         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1311                          sizeof(struct mdt_rec_reint));
1312
1313         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1314         if (rec == NULL)
1315                 RETURN(-EFAULT);
1316
1317         uc->mu_fsuid  = rec->sx_fsuid;
1318         uc->mu_fsgid  = rec->sx_fsgid;
1319         uc->mu_cap    = rec->sx_cap;
1320         uc->mu_suppgids[0] = rec->sx_suppgid1;
1321         uc->mu_suppgids[1] = -1;
1322
1323         rr->rr_opcode = rec->sx_opcode;
1324         rr->rr_fid1   = &rec->sx_fid;
1325         attr->la_valid = rec->sx_valid;
1326         attr->la_ctime = rec->sx_time;
1327         attr->la_size = rec->sx_size;
1328         attr->la_flags = rec->sx_flags;
1329
1330         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1331                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1332                                  req_capsule_client_get(pill, &RMF_CAPA1));
1333         else
1334                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1335
1336         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1337         if (rr->rr_name == NULL)
1338                 RETURN(-EFAULT);
1339         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT) - 1;
1340         LASSERT(rr->rr_namelen > 0);
1341
1342         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1343                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1344                                                         RCL_CLIENT);
1345                 if (rr->rr_eadatalen > 0) {
1346                         rr->rr_eadata = req_capsule_client_get(pill,
1347                                                                &RMF_EADATA);
1348                         if (rr->rr_eadata == NULL)
1349                                 RETURN(-EFAULT);
1350                 } else {
1351                         rr->rr_eadata = NULL;
1352                 }
1353         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1354                 CDEBUG(D_INFO, "no xattr data supplied\n");
1355                 RETURN(-EFAULT);
1356         }
1357
1358         RETURN(0);
1359 }
1360
1361
1362 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1363
1364 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1365         [REINT_SETATTR]  = mdt_setattr_unpack,
1366         [REINT_CREATE]   = mdt_create_unpack,
1367         [REINT_LINK]     = mdt_link_unpack,
1368         [REINT_UNLINK]   = mdt_unlink_unpack,
1369         [REINT_RENAME]   = mdt_rename_unpack,
1370         [REINT_OPEN]     = mdt_open_unpack,
1371         [REINT_SETXATTR] = mdt_setxattr_unpack
1372 };
1373
1374 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1375 {
1376         int rc;
1377         ENTRY;
1378
1379         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1380         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1381                 info->mti_rr.rr_opcode = op;
1382                 rc = mdt_reint_unpackers[op](info);
1383         } else {
1384                 CERROR("Unexpected opcode %d\n", op);
1385                 rc = -EFAULT;
1386         }
1387         RETURN(rc);
1388 }