lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_lib.c
  37  *
  38  * Lustre Metadata Target (mdt) request unpacking helper.
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Fan Yong <fanyong@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include "mdt_internal.h"
  52 #include <lnet/nidstr.h>
  53
  54 typedef enum ucred_init_type {
  55         NONE_INIT       = 0,
  56         BODY_INIT       = 1,
  57         REC_INIT        = 2
  58 } ucred_init_type_t;
  59
  60 void mdt_exit_ucred(struct mdt_thread_info *info)
  61 {
  62         struct lu_ucred   *uc  = mdt_ucred(info);
  63         struct mdt_device *mdt = info->mti_mdt;
  64
  65         LASSERT(uc != NULL);
  66         if (uc->uc_valid != UCRED_INIT) {
  67                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  68                 if (uc->uc_ginfo) {
  69                         put_group_info(uc->uc_ginfo);
  70                         uc->uc_ginfo = NULL;
  71                 }
  72                 if (uc->uc_identity) {
  73                         mdt_identity_put(mdt->mdt_identity_cache,
  74                                          uc->uc_identity);
  75                         uc->uc_identity = NULL;
  76                 }
  77                 uc->uc_valid = UCRED_INIT;
  78         }
  79 }
  80
  81 static int match_nosquash_list(struct rw_semaphore *sem,
  82                                struct list_head *nidlist,
  83                                lnet_nid_t peernid)
  84 {
  85         int rc;
  86         ENTRY;
  87         down_read(sem);
  88         rc = cfs_match_nid(peernid, nidlist);
  89         up_read(sem);
  90         RETURN(rc);
  91 }
  92
  93 /* root_squash for inter-MDS operations */
  94 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
  95 {
  96         struct lu_ucred *ucred = mdt_ucred(info);
  97         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
  98         ENTRY;
  99
 100         LASSERT(ucred != NULL);
 101         if (!squash->rsi_uid || ucred->uc_fsuid)
 102                 RETURN(0);
 103
 104         if (match_nosquash_list(&squash->rsi_sem,
 105                                 &squash->rsi_nosquash_nids,
 106                                 peernid)) {
 107                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 108                        libcfs_nid2str(peernid));
 109                 RETURN(0);
 110         }
 111
 112         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 113                libcfs_nid2str(peernid),
 114                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 115                squash->rsi_uid, squash->rsi_gid, 0);
 116
 117         ucred->uc_fsuid = squash->rsi_uid;
 118         ucred->uc_fsgid = squash->rsi_gid;
 119         ucred->uc_cap = 0;
 120         ucred->uc_suppgids[0] = -1;
 121         ucred->uc_suppgids[1] = -1;
 122
 123         RETURN(0);
 124 }
 125
 126 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 127 {
 128         struct ptlrpc_request   *req = mdt_info_req(info);
 129         const char              *jobid = mdt_req_get_jobid(req);
 130
 131         /* set jobid if specified. */
 132         if (jobid)
 133                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 134         else
 135                 uc->uc_jobid[0] = '\0';
 136 }
 137
 138 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 139                           void *buf)
 140 {
 141         struct ptlrpc_request   *req = mdt_info_req(info);
 142         struct mdt_device       *mdt = info->mti_mdt;
 143         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 144         struct lu_ucred         *ucred = mdt_ucred(info);
 145         lnet_nid_t               peernid = req->rq_peer.nid;
 146         __u32                    perm = 0;
 147         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 148         int                      setuid;
 149         int                      setgid;
 150         int                      rc = 0;
 151
 152         ENTRY;
 153
 154         LASSERT(req->rq_auth_gss);
 155         LASSERT(!req->rq_auth_usr_mdt);
 156         LASSERT(req->rq_user_desc);
 157         LASSERT(ucred != NULL);
 158
 159         ucred->uc_valid = UCRED_INVALID;
 160
 161         ucred->uc_o_uid   = pud->pud_uid;
 162         ucred->uc_o_gid   = pud->pud_gid;
 163         ucred->uc_o_fsuid = pud->pud_fsuid;
 164         ucred->uc_o_fsgid = pud->pud_fsgid;
 165
 166         if (type == BODY_INIT) {
 167                 struct mdt_body *body = (struct mdt_body *)buf;
 168
 169                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 170                 ucred->uc_suppgids[1] = -1;
 171         }
 172
 173         /* sanity check: we expect the uid which client claimed is true */
 174         if (remote) {
 175                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 176                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 177                         RETURN(-EACCES);
 178                 }
 179
 180                 if (ptlrpc_user_desc_do_idmap(req, pud))
 181                         RETURN(-EACCES);
 182
 183                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 184                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 185                                "while client claims %u:%u/%u:%u\n",
 186                                libcfs_nid2str(peernid), req->rq_auth_uid,
 187                                req->rq_auth_mapped_uid,
 188                                pud->pud_uid, pud->pud_gid,
 189                                pud->pud_fsuid, pud->pud_fsgid);
 190                         RETURN(-EACCES);
 191                 }
 192         } else {
 193                 if (req->rq_auth_uid != pud->pud_uid) {
 194                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 195                                "while client claims %u:%u/%u:%u\n",
 196                                libcfs_nid2str(peernid), req->rq_auth_uid,
 197                                pud->pud_uid, pud->pud_gid,
 198                                pud->pud_fsuid, pud->pud_fsgid);
 199                         RETURN(-EACCES);
 200                 }
 201         }
 202
 203         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 204                 if (remote) {
 205                         CDEBUG(D_SEC, "remote client must run with identity_get "
 206                                "enabled!\n");
 207                         RETURN(-EACCES);
 208                 } else {
 209                         ucred->uc_identity = NULL;
 210                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 211                                CFS_SETGRP_PERM;
 212                 }
 213         } else {
 214                 struct md_identity *identity;
 215
 216                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 217                                             pud->pud_uid);
 218                 if (IS_ERR(identity)) {
 219                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 220                                      !remote)) {
 221                                 ucred->uc_identity = NULL;
 222                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 223                                        CFS_SETGRP_PERM;
 224                         } else {
 225                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 226                                        pud->pud_uid);
 227                                 RETURN(-EACCES);
 228                         }
 229                 } else {
 230                         ucred->uc_identity = identity;
 231                         perm = mdt_identity_get_perm(ucred->uc_identity,
 232                                                      remote, peernid);
 233                 }
 234         }
 235
 236         /* find out the setuid/setgid attempt */
 237         setuid = (pud->pud_uid != pud->pud_fsuid);
 238         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 239                   (ucred->uc_identity &&
 240                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 241
 242         /* check permission of setuid */
 243         if (setuid && !(perm & CFS_SETUID_PERM)) {
 244                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 245                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 246                 GOTO(out, rc = -EACCES);
 247         }
 248
 249         /* check permission of setgid */
 250         if (setgid && !(perm & CFS_SETGID_PERM)) {
 251                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 252                        "from %s\n", pud->pud_uid, pud->pud_gid,
 253                        pud->pud_fsuid, pud->pud_fsgid,
 254                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 255                 GOTO(out, rc = -EACCES);
 256         }
 257
 258         /*
 259          * NB: remote client not allowed to setgroups anyway.
 260          */
 261         if (!remote && perm & CFS_SETGRP_PERM) {
 262                 if (pud->pud_ngroups) {
 263                         /* setgroups for local client */
 264                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 265                         if (!ucred->uc_ginfo) {
 266                                 CERROR("failed to alloc %d groups\n",
 267                                        pud->pud_ngroups);
 268                                 GOTO(out, rc = -ENOMEM);
 269                         }
 270
 271                         lustre_groups_from_list(ucred->uc_ginfo,
 272                                                 pud->pud_groups);
 273                         lustre_groups_sort(ucred->uc_ginfo);
 274                 } else {
 275                         ucred->uc_ginfo = NULL;
 276                 }
 277         } else {
 278                 ucred->uc_suppgids[0] = -1;
 279                 ucred->uc_suppgids[1] = -1;
 280                 ucred->uc_ginfo = NULL;
 281         }
 282
 283         ucred->uc_uid   = pud->pud_uid;
 284         ucred->uc_gid   = pud->pud_gid;
 285         ucred->uc_fsuid = pud->pud_fsuid;
 286         ucred->uc_fsgid = pud->pud_fsgid;
 287
 288         /* process root_squash here. */
 289         mdt_root_squash(info, peernid);
 290
 291         /* remove fs privilege for non-root user. */
 292         if (ucred->uc_fsuid)
 293                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 294         else
 295                 ucred->uc_cap = pud->pud_cap;
 296         if (remote && !(perm & CFS_RMTOWN_PERM))
 297                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 298                                    CFS_CAP_CHOWN_MASK);
 299         ucred->uc_valid = UCRED_NEW;
 300         ucred_set_jobid(info, ucred);
 301
 302         EXIT;
 303
 304 out:
 305         if (rc) {
 306                 if (ucred->uc_ginfo) {
 307                         put_group_info(ucred->uc_ginfo);
 308                         ucred->uc_ginfo = NULL;
 309                 }
 310                 if (ucred->uc_identity) {
 311                         mdt_identity_put(mdt->mdt_identity_cache,
 312                                          ucred->uc_identity);
 313                         ucred->uc_identity = NULL;
 314                 }
 315         }
 316
 317         return rc;
 318 }
 319
 320 int mdt_check_ucred(struct mdt_thread_info *info)
 321 {
 322         struct ptlrpc_request   *req = mdt_info_req(info);
 323         struct mdt_device       *mdt = info->mti_mdt;
 324         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 325         struct lu_ucred         *ucred = mdt_ucred(info);
 326         struct md_identity      *identity = NULL;
 327         lnet_nid_t               peernid = req->rq_peer.nid;
 328         __u32                    perm = 0;
 329         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 330         int                      setuid;
 331         int                      setgid;
 332         int                      rc = 0;
 333
 334         ENTRY;
 335
 336         LASSERT(ucred != NULL);
 337         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 338                 RETURN(0);
 339
 340         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 341                 RETURN(0);
 342
 343         /* sanity check: if we use strong authentication, we expect the
 344          * uid which client claimed is true */
 345         if (remote) {
 346                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 347                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 348                         RETURN(-EACCES);
 349                 }
 350
 351                 if (ptlrpc_user_desc_do_idmap(req, pud))
 352                         RETURN(-EACCES);
 353
 354                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 355                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 356                                "while client claims %u:%u/%u:%u\n",
 357                                libcfs_nid2str(peernid), req->rq_auth_uid,
 358                                req->rq_auth_mapped_uid,
 359                                pud->pud_uid, pud->pud_gid,
 360                                pud->pud_fsuid, pud->pud_fsgid);
 361                         RETURN(-EACCES);
 362                 }
 363         } else {
 364                 if (req->rq_auth_uid != pud->pud_uid) {
 365                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 366                                "while client claims %u:%u/%u:%u\n",
 367                                libcfs_nid2str(peernid), req->rq_auth_uid,
 368                                pud->pud_uid, pud->pud_gid,
 369                                pud->pud_fsuid, pud->pud_fsgid);
 370                         RETURN(-EACCES);
 371                 }
 372         }
 373
 374         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 375                 if (remote) {
 376                         CDEBUG(D_SEC, "remote client must run with identity_get "
 377                                "enabled!\n");
 378                         RETURN(-EACCES);
 379                 }
 380                 RETURN(0);
 381         }
 382
 383         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 384         if (IS_ERR(identity)) {
 385                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 386                              !remote)) {
 387                         RETURN(0);
 388                 } else {
 389                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 390                                pud->pud_uid);
 391                         RETURN(-EACCES);
 392                }
 393         }
 394
 395         perm = mdt_identity_get_perm(identity, remote, peernid);
 396         /* find out the setuid/setgid attempt */
 397         setuid = (pud->pud_uid != pud->pud_fsuid);
 398         setgid = (pud->pud_gid != pud->pud_fsgid ||
 399                   pud->pud_gid != identity->mi_gid);
 400
 401         /* check permission of setuid */
 402         if (setuid && !(perm & CFS_SETUID_PERM)) {
 403                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 404                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 405                 GOTO(out, rc = -EACCES);
 406         }
 407
 408         /* check permission of setgid */
 409         if (setgid && !(perm & CFS_SETGID_PERM)) {
 410                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 411                        "from %s\n", pud->pud_uid, pud->pud_gid,
 412                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 413                        libcfs_nid2str(peernid));
 414                 GOTO(out, rc = -EACCES);
 415         }
 416
 417         EXIT;
 418
 419 out:
 420         mdt_identity_put(mdt->mdt_identity_cache, identity);
 421         return rc;
 422 }
 423
 424 static int old_init_ucred(struct mdt_thread_info *info,
 425                           struct mdt_body *body)
 426 {
 427         struct lu_ucred *uc = mdt_ucred(info);
 428         struct mdt_device  *mdt = info->mti_mdt;
 429         struct md_identity *identity = NULL;
 430
 431         ENTRY;
 432
 433         LASSERT(uc != NULL);
 434         uc->uc_valid = UCRED_INVALID;
 435         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 436         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 437         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 438         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 439         uc->uc_suppgids[0] = body->mbo_suppgid;
 440         uc->uc_suppgids[1] = -1;
 441         uc->uc_ginfo = NULL;
 442         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 443                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 444                                             uc->uc_fsuid);
 445                 if (IS_ERR(identity)) {
 446                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 447                                 identity = NULL;
 448                         } else {
 449                                 CDEBUG(D_SEC, "Deny access without identity: "
 450                                        "uid %u\n", uc->uc_fsuid);
 451                                 RETURN(-EACCES);
 452                         }
 453                 }
 454         }
 455         uc->uc_identity = identity;
 456
 457         /* process root_squash here. */
 458         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 459
 460         /* remove fs privilege for non-root user. */
 461         if (uc->uc_fsuid)
 462                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 463         else
 464                 uc->uc_cap = body->mbo_capability;
 465         uc->uc_valid = UCRED_OLD;
 466         ucred_set_jobid(info, uc);
 467
 468         RETURN(0);
 469 }
 470
 471 static int old_init_ucred_reint(struct mdt_thread_info *info)
 472 {
 473         struct lu_ucred *uc = mdt_ucred(info);
 474         struct mdt_device  *mdt = info->mti_mdt;
 475         struct md_identity *identity = NULL;
 476
 477         ENTRY;
 478
 479         LASSERT(uc != NULL);
 480         uc->uc_valid = UCRED_INVALID;
 481         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 482         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 483         uc->uc_ginfo = NULL;
 484         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 485                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 486                                             uc->uc_fsuid);
 487                 if (IS_ERR(identity)) {
 488                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 489                                 identity = NULL;
 490                         } else {
 491                                 CDEBUG(D_SEC, "Deny access without identity: "
 492                                        "uid %u\n", uc->uc_fsuid);
 493                                 RETURN(-EACCES);
 494                         }
 495                 }
 496         }
 497         uc->uc_identity = identity;
 498
 499         /* process root_squash here. */
 500         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 501
 502         /* remove fs privilege for non-root user. */
 503         if (uc->uc_fsuid)
 504                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 505         uc->uc_valid = UCRED_OLD;
 506         ucred_set_jobid(info, uc);
 507
 508         RETURN(0);
 509 }
 510
 511 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 512 {
 513         struct ptlrpc_request *req = mdt_info_req(info);
 514         struct lu_ucred       *uc  = mdt_ucred(info);
 515
 516         LASSERT(uc != NULL);
 517         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 518                 return 0;
 519
 520         mdt_exit_ucred(info);
 521
 522         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 523                 return old_init_ucred(info, body);
 524         else
 525                 return new_init_ucred(info, BODY_INIT, body);
 526 }
 527
 528 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 529 {
 530         struct ptlrpc_request *req = mdt_info_req(info);
 531         struct lu_ucred       *uc  = mdt_ucred(info);
 532
 533         LASSERT(uc != NULL);
 534         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 535                 return 0;
 536
 537         mdt_exit_ucred(info);
 538
 539         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 540                 return old_init_ucred_reint(info);
 541         else
 542                 return new_init_ucred(info, REC_INIT, NULL);
 543 }
 544
 545 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 546 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 547 {
 548         const struct lov_ost_data_v1    *lod;
 549         int                              i;
 550         __u16                            count;
 551
 552         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 553                 return;
 554
 555         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 556
 557         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 558                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 559                le32_to_cpu(lmm->lmm_pattern));
 560         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 561                le32_to_cpu(lmm->lmm_stripe_size), count);
 562
 563         /* If it's a directory or a released file, then there are
 564          * no actual objects to print, so bail out. */
 565         if (valid & OBD_MD_FLDIREA ||
 566             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 567                 return;
 568
 569         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 570         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 571                 struct ost_id oi;
 572
 573                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 574                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 575                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 576         }
 577 }
 578
 579 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 580 {
 581         const struct lmv_mds_md_v1 *lmm1;
 582         int                        i;
 583
 584         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 585                 return;
 586
 587         lmm1 = &lmv->lmv_md_v1;
 588         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 589                le32_to_cpu(lmm1->lmv_magic),
 590                le32_to_cpu(lmm1->lmv_master_mdt_index),
 591                le32_to_cpu(lmm1->lmv_stripe_count));
 592
 593         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 594                 return;
 595
 596         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 597                 struct lu_fid fid;
 598
 599                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 600                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 601         }
 602 }
 603
 604 /* Shrink and/or grow reply buffers */
 605 int mdt_fix_reply(struct mdt_thread_info *info)
 606 {
 607         struct req_capsule *pill = info->mti_pill;
 608         struct mdt_body    *body;
 609         int                md_size, md_packed = 0;
 610         int                acl_size;
 611         int                rc = 0;
 612         ENTRY;
 613
 614         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 615         LASSERT(body != NULL);
 616
 617         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 618                                OBD_MD_LINKNAME))
 619                 md_size = body->mbo_eadatasize;
 620         else
 621                 md_size = 0;
 622
 623         acl_size = body->mbo_aclsize;
 624
 625         /* this replay - not send info to client */
 626         if (info->mti_spec.no_create) {
 627                 md_size = 0;
 628                 acl_size = 0;
 629         }
 630
 631         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 632                " MDSCAPA = %llx, OSSCAPA = %llx\n",
 633                md_size, acl_size,
 634                (unsigned long long)(body->mbo_valid & OBD_MD_FLMDSCAPA),
 635                (unsigned long long)(body->mbo_valid & OBD_MD_FLOSSCAPA));
 636 /*
 637             &RMF_MDT_BODY,
 638             &RMF_MDT_MD,
 639             &RMF_ACL, or &RMF_LOGCOOKIES
 640 (optional)  &RMF_CAPA1,
 641 (optional)  &RMF_CAPA2,
 642 (optional)  something else
 643 */
 644
 645         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 646          * buffers before growing it */
 647         if (info->mti_big_lmm_used) {
 648                 /* big_lmm buffer may be used even without packing the result
 649                  * into reply, just for internal server needs */
 650                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 651                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 652                                                          RCL_SERVER);
 653
 654                 /* free big lmm if md_size is not needed */
 655                 if (md_size == 0 || md_packed == 0) {
 656                         info->mti_big_lmm_used = 0;
 657                 } else {
 658                         /* buffer must be allocated separately */
 659                         LASSERT(info->mti_attr.ma_lmm !=
 660                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 661                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 662                 }
 663         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 664                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 665         }
 666
 667         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 668                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 669         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 670                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 671                                    acl_size, RCL_SERVER);
 672
 673         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 674             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 675                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 676
 677         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 678             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 679                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 680
 681         /*
 682          * Some more field should be shrinked if needed.
 683          * This should be done by those who added fields to reply message.
 684          */
 685
 686         /* Grow MD buffer if needed finally */
 687         if (info->mti_big_lmm_used) {
 688                 void *lmm;
 689
 690                 LASSERT(md_size > md_packed);
 691                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 692                        md_size - md_packed);
 693                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 694                 if (rc) {
 695                         /* we can't answer with proper LOV EA, drop flags,
 696                          * the rc is also returned so this request is
 697                          * considered as failed */
 698                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 699                         /* don't return transno along with error */
 700                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 701                 } else {
 702                         /* now we need to pack right LOV/LMV EA */
 703                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 704                         if (info->mti_attr.ma_valid & MA_LOV) {
 705                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 706                                                              RCL_SERVER) ==
 707                                                 info->mti_attr.ma_lmm_size);
 708                                 memcpy(lmm, info->mti_attr.ma_lmm,
 709                                        info->mti_attr.ma_lmm_size);
 710                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 711                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 712                                                              RCL_SERVER) ==
 713                                                 info->mti_attr.ma_lmv_size);
 714                                 memcpy(lmm, info->mti_attr.ma_lmv,
 715                                        info->mti_attr.ma_lmv_size);
 716                         }
 717                 }
 718                 /* update mdt_max_mdsize so clients will be aware about that */
 719                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 720                         info->mti_mdt->mdt_max_mdsize =
 721                                                     info->mti_attr.ma_lmm_size;
 722                 info->mti_big_lmm_used = 0;
 723         }
 724         RETURN(rc);
 725 }
 726
 727
 728 /* if object is dying, pack the lov/llog data,
 729  * parameter info->mti_attr should be valid at this point! */
 730 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 731                            const struct md_attr *ma)
 732 {
 733         struct mdt_body       *repbody;
 734         const struct lu_attr *la = &ma->ma_attr;
 735         int rc;
 736         ENTRY;
 737
 738         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 739         LASSERT(repbody != NULL);
 740
 741         if (ma->ma_valid & MA_INODE)
 742                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 743
 744         if (ma->ma_valid & MA_LOV) {
 745                 CERROR("No need in LOV EA upon unlink\n");
 746                 dump_stack();
 747         }
 748         repbody->mbo_eadatasize = 0;
 749
 750         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 751             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 752             repbody->mbo_valid & OBD_MD_FLEASIZE) {
 753                 struct lustre_capa *capa;
 754
 755                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 756                 LASSERT(capa);
 757                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 758                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 759                 if (rc)
 760                         RETURN(rc);
 761
 762                 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
 763         }
 764
 765         RETURN(0);
 766 }
 767
 768 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 769                                   struct md_attr *ma)
 770 {
 771         __u64 out;
 772
 773         out = 0;
 774         if (in & MDS_ATTR_MODE)
 775                 out |= LA_MODE;
 776         if (in & MDS_ATTR_UID)
 777                 out |= LA_UID;
 778         if (in & MDS_ATTR_GID)
 779                 out |= LA_GID;
 780         if (in & MDS_ATTR_SIZE)
 781                 out |= LA_SIZE;
 782         if (in & MDS_ATTR_BLOCKS)
 783                 out |= LA_BLOCKS;
 784         if (in & MDS_ATTR_ATIME_SET)
 785                 out |= LA_ATIME;
 786         if (in & MDS_ATTR_CTIME_SET)
 787                 out |= LA_CTIME;
 788         if (in & MDS_ATTR_MTIME_SET)
 789                 out |= LA_MTIME;
 790         if (in & MDS_ATTR_ATTR_FLAG)
 791                 out |= LA_FLAGS;
 792         if (in & MDS_ATTR_KILL_SUID)
 793                 out |= LA_KILL_SUID;
 794         if (in & MDS_ATTR_KILL_SGID)
 795                 out |= LA_KILL_SGID;
 796
 797         if (in & MDS_ATTR_FROM_OPEN)
 798                 rr->rr_flags |= MRF_OPEN_TRUNC;
 799         if (in & MDS_OPEN_OWNEROVERRIDE)
 800                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 801         if (in & MDS_ATTR_FORCE)
 802                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 803
 804         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 805                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 806                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 807                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 808                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 809                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 810         if (in != 0)
 811                 CERROR("Unknown attr bits: "LPX64"\n", in);
 812         return out;
 813 }
 814
 815 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 816                       const struct lu_fid *fid, struct lustre_capa *capa)
 817 {
 818         struct lu_capainfo *lci;
 819
 820         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 821         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 822             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 823                 return;
 824
 825         lci = lu_capainfo_get(info->mti_env);
 826         LASSERT(lci);
 827         lci->lci_fid[offset]  = *fid;
 828         lci->lci_capa[offset] = capa;
 829 }
 830
 831 #ifdef DEBUG_CAPA
 832 void mdt_dump_capainfo(struct mdt_thread_info *info)
 833 {
 834         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 835         int i;
 836
 837         if (lci == NULL)
 838                 return;
 839
 840         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 841                 if (lci->lci_capa[i] == NULL) {
 842                         CERROR("no capa for index %d "DFID"\n",
 843                                i, PFID(&lci->lci_fid[i]));
 844                         continue;
 845                 }
 846                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 847                         CERROR("bypass for index %d "DFID"\n",
 848                                i, PFID(&lci->lci_fid[i]));
 849                         continue;
 850                 }
 851                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 852         }
 853 }
 854 #endif /* DEBUG_CAPA */
 855
 856 /* unpacking */
 857
 858 int mdt_name_unpack(struct req_capsule *pill,
 859                     const struct req_msg_field *field,
 860                     struct lu_name *ln,
 861                     enum mdt_name_flags flags)
 862 {
 863         ln->ln_name = req_capsule_client_get(pill, field);
 864         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 865
 866         if (!lu_name_is_valid(ln)) {
 867                 ln->ln_name = NULL;
 868                 ln->ln_namelen = 0;
 869
 870                 return -EPROTO;
 871         }
 872
 873         if ((flags & MNF_FIX_ANON) &&
 874             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 875                 /* Newer (3.x) kernels use a name of "/" for the
 876                  * "anonymous" disconnected dentries from NFS
 877                  * filehandle conversion. See d_obtain_alias(). */
 878                 ln->ln_name = NULL;
 879                 ln->ln_namelen = 0;
 880         }
 881
 882         return 0;
 883 }
 884
 885 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 886 {
 887         struct lu_ucred         *uc  = mdt_ucred(info);
 888         struct md_attr          *ma = &info->mti_attr;
 889         struct lu_attr          *la = &ma->ma_attr;
 890         struct req_capsule      *pill = info->mti_pill;
 891         struct mdt_reint_record *rr = &info->mti_rr;
 892         struct mdt_rec_setattr  *rec;
 893         ENTRY;
 894
 895         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 896         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 897         if (rec == NULL)
 898                 RETURN(-EFAULT);
 899
 900         /* This prior initialization is needed for old_init_ucred_reint() */
 901         uc->uc_fsuid = rec->sa_fsuid;
 902         uc->uc_fsgid = rec->sa_fsgid;
 903         uc->uc_cap   = rec->sa_cap;
 904         uc->uc_suppgids[0] = rec->sa_suppgid;
 905         uc->uc_suppgids[1] = -1;
 906
 907         rr->rr_fid1 = &rec->sa_fid;
 908         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 909         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 910          *  the client does not have OBD_CONNECT_FULL20, convert it
 911          *  to LA_xTIME. LU-3036 */
 912         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 913                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 914                      (rec->sa_valid & MDS_ATTR_ATIME))
 915                         la->la_valid |= LA_ATIME;
 916                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 917                      (rec->sa_valid & MDS_ATTR_MTIME))
 918                         la->la_valid |= LA_MTIME;
 919                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 920                      (rec->sa_valid & MDS_ATTR_CTIME))
 921                         la->la_valid |= LA_CTIME;
 922         }
 923         la->la_mode  = rec->sa_mode;
 924         la->la_flags = rec->sa_attr_flags;
 925         la->la_uid   = rec->sa_uid;
 926         la->la_gid   = rec->sa_gid;
 927         la->la_size  = rec->sa_size;
 928         la->la_blocks = rec->sa_blocks;
 929         la->la_ctime = rec->sa_ctime;
 930         la->la_atime = rec->sa_atime;
 931         la->la_mtime = rec->sa_mtime;
 932         ma->ma_valid = MA_INODE;
 933
 934         if (rec->sa_bias & MDS_DATA_MODIFIED)
 935                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 936         else
 937                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 938
 939         if (rec->sa_bias & MDS_HSM_RELEASE)
 940                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 941         else
 942                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 943
 944         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 945                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 946                                  req_capsule_client_get(pill, &RMF_CAPA1));
 947
 948         RETURN(0);
 949 }
 950
 951 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 952 {
 953         struct req_capsule *pill = info->mti_pill;
 954         ENTRY;
 955
 956         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 957                 info->mti_ioepoch =
 958                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 959         else
 960                 info->mti_ioepoch = NULL;
 961         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
 962 }
 963
 964 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 965         struct req_capsule      *pill = info->mti_pill;
 966
 967         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 968                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 969                 if (info->mti_dlm_req == NULL)
 970                         RETURN(-EFAULT);
 971         }
 972
 973         RETURN(0);
 974 }
 975
 976 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 977 {
 978         struct mdt_reint_record *rr = &info->mti_rr;
 979         struct md_attr          *ma = &info->mti_attr;
 980         struct req_capsule      *pill = info->mti_pill;
 981         int rc;
 982         ENTRY;
 983
 984         rc = mdt_setattr_unpack_rec(info);
 985         if (rc)
 986                 RETURN(rc);
 987
 988         /* Epoch may be absent */
 989         mdt_ioepoch_unpack(info);
 990
 991         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 992                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 993                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 994                                                         RCL_CLIENT);
 995                 if (rr->rr_eadatalen > 0) {
 996                         const struct lmv_user_md        *lum;
 997
 998                         lum = rr->rr_eadata;
 999                         /* Sigh ma_valid(from req) does not indicate whether
1000                          * it will set LOV/LMV EA, so we have to check magic */
1001                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1002                                 ma->ma_valid |= MA_LMV;
1003                                 ma->ma_lmv = (void *)rr->rr_eadata;
1004                                 ma->ma_lmv_size = rr->rr_eadatalen;
1005                         } else {
1006                                 ma->ma_valid |= MA_LOV;
1007                                 ma->ma_lmm = (void *)rr->rr_eadata;
1008                                 ma->ma_lmm_size = rr->rr_eadatalen;
1009                         }
1010                 }
1011         }
1012
1013         rc = mdt_dlmreq_unpack(info);
1014         RETURN(rc);
1015 }
1016
1017 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1018 {
1019         struct md_attr          *ma = &info->mti_attr;
1020         struct req_capsule      *pill = info->mti_pill;
1021         ENTRY;
1022
1023         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1024                 RETURN(0);
1025
1026         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1027
1028         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1029             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1030                 RETURN(-EFAULT);
1031
1032         RETURN(0);
1033 }
1034
1035 int mdt_close_unpack(struct mdt_thread_info *info)
1036 {
1037         int rc;
1038         ENTRY;
1039
1040         rc = mdt_ioepoch_unpack(info);
1041         if (rc)
1042                 RETURN(rc);
1043
1044         rc = mdt_setattr_unpack_rec(info);
1045         if (rc)
1046                 RETURN(rc);
1047
1048         rc = mdt_hsm_release_unpack(info);
1049         if (rc)
1050                 RETURN(rc);
1051
1052         RETURN(mdt_init_ucred_reint(info));
1053 }
1054
1055 static int mdt_create_unpack(struct mdt_thread_info *info)
1056 {
1057         struct lu_ucred         *uc  = mdt_ucred(info);
1058         struct mdt_rec_create   *rec;
1059         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1060         struct mdt_reint_record *rr = &info->mti_rr;
1061         struct req_capsule      *pill = info->mti_pill;
1062         struct md_op_spec       *sp = &info->mti_spec;
1063         int rc;
1064         ENTRY;
1065
1066         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1067         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1068         if (rec == NULL)
1069                 RETURN(-EFAULT);
1070
1071         /* This prior initialization is needed for old_init_ucred_reint() */
1072         uc->uc_fsuid = rec->cr_fsuid;
1073         uc->uc_fsgid = rec->cr_fsgid;
1074         uc->uc_cap   = rec->cr_cap;
1075         uc->uc_suppgids[0] = rec->cr_suppgid1;
1076         uc->uc_suppgids[1] = -1;
1077         uc->uc_umask = rec->cr_umask;
1078
1079         rr->rr_fid1 = &rec->cr_fid1;
1080         rr->rr_fid2 = &rec->cr_fid2;
1081         attr->la_mode = rec->cr_mode;
1082         attr->la_rdev  = rec->cr_rdev;
1083         attr->la_uid   = rec->cr_fsuid;
1084         attr->la_gid   = rec->cr_fsgid;
1085         attr->la_ctime = rec->cr_time;
1086         attr->la_mtime = rec->cr_time;
1087         attr->la_atime = rec->cr_time;
1088         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1089                          LA_CTIME | LA_MTIME | LA_ATIME;
1090         memset(&sp->u, 0, sizeof(sp->u));
1091         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1092
1093         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1094                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1095                                  req_capsule_client_get(pill, &RMF_CAPA1));
1096         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1097
1098         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1099         if (rc < 0)
1100                 RETURN(rc);
1101
1102         if (S_ISLNK(attr->la_mode)) {
1103                 const char *tgt = NULL;
1104
1105                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1106                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1107                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1108                         sp->u.sp_symname = tgt;
1109                 }
1110                 if (tgt == NULL)
1111                         RETURN(-EFAULT);
1112         } else {
1113                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1114                 if (S_ISDIR(attr->la_mode) &&
1115                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1116                         sp->u.sp_ea.eadata =
1117                                 req_capsule_client_get(pill, &RMF_EADATA);
1118                         sp->u.sp_ea.eadatalen =
1119                                 req_capsule_get_size(pill, &RMF_EADATA,
1120                                                      RCL_CLIENT);
1121                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1122                 }
1123         }
1124
1125         rc = mdt_dlmreq_unpack(info);
1126         RETURN(rc);
1127 }
1128
1129 static int mdt_link_unpack(struct mdt_thread_info *info)
1130 {
1131         struct lu_ucred         *uc  = mdt_ucred(info);
1132         struct mdt_rec_link     *rec;
1133         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1134         struct mdt_reint_record *rr = &info->mti_rr;
1135         struct req_capsule      *pill = info->mti_pill;
1136         int rc;
1137         ENTRY;
1138
1139         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1140         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1141         if (rec == NULL)
1142                 RETURN(-EFAULT);
1143
1144         /* This prior initialization is needed for old_init_ucred_reint() */
1145         uc->uc_fsuid = rec->lk_fsuid;
1146         uc->uc_fsgid = rec->lk_fsgid;
1147         uc->uc_cap   = rec->lk_cap;
1148         uc->uc_suppgids[0] = rec->lk_suppgid1;
1149         uc->uc_suppgids[1] = rec->lk_suppgid2;
1150
1151         attr->la_uid = rec->lk_fsuid;
1152         attr->la_gid = rec->lk_fsgid;
1153         rr->rr_fid1 = &rec->lk_fid1;
1154         rr->rr_fid2 = &rec->lk_fid2;
1155         attr->la_ctime = rec->lk_time;
1156         attr->la_mtime = rec->lk_time;
1157         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1158
1159         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1160                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1161                                  req_capsule_client_get(pill, &RMF_CAPA1));
1162         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1163                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1164                                  req_capsule_client_get(pill, &RMF_CAPA2));
1165
1166         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1167         if (rc < 0)
1168                 RETURN(rc);
1169
1170         rc = mdt_dlmreq_unpack(info);
1171
1172         RETURN(rc);
1173 }
1174
1175 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1176 {
1177         struct lu_ucred         *uc  = mdt_ucred(info);
1178         struct mdt_rec_unlink   *rec;
1179         struct md_attr          *ma = &info->mti_attr;
1180         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1181         struct mdt_reint_record *rr = &info->mti_rr;
1182         struct req_capsule      *pill = info->mti_pill;
1183         int rc;
1184         ENTRY;
1185
1186         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1187         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1188         if (rec == NULL)
1189                 RETURN(-EFAULT);
1190
1191         /* This prior initialization is needed for old_init_ucred_reint() */
1192         uc->uc_fsuid = rec->ul_fsuid;
1193         uc->uc_fsgid = rec->ul_fsgid;
1194         uc->uc_cap   = rec->ul_cap;
1195         uc->uc_suppgids[0] = rec->ul_suppgid1;
1196         uc->uc_suppgids[1] = -1;
1197
1198         attr->la_uid = rec->ul_fsuid;
1199         attr->la_gid = rec->ul_fsgid;
1200         rr->rr_fid1 = &rec->ul_fid1;
1201         rr->rr_fid2 = &rec->ul_fid2;
1202         attr->la_ctime = rec->ul_time;
1203         attr->la_mtime = rec->ul_time;
1204         attr->la_mode  = rec->ul_mode;
1205         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1206
1207         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1208                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1209                                  req_capsule_client_get(pill, &RMF_CAPA1));
1210
1211         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1212         if (rc < 0)
1213                 RETURN(rc);
1214
1215         if (rec->ul_bias & MDS_VTX_BYPASS)
1216                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1217         else
1218                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1219
1220         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1221
1222         rc = mdt_dlmreq_unpack(info);
1223         RETURN(rc);
1224 }
1225
1226 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1227 {
1228         info->mti_spec.sp_rm_entry = 1;
1229         return mdt_unlink_unpack(info);
1230 }
1231
1232 static int mdt_rename_unpack(struct mdt_thread_info *info)
1233 {
1234         struct lu_ucred         *uc = mdt_ucred(info);
1235         struct mdt_rec_rename   *rec;
1236         struct md_attr          *ma = &info->mti_attr;
1237         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1238         struct mdt_reint_record *rr = &info->mti_rr;
1239         struct req_capsule      *pill = info->mti_pill;
1240         int rc;
1241         ENTRY;
1242
1243         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1244         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1245         if (rec == NULL)
1246                 RETURN(-EFAULT);
1247
1248         /* This prior initialization is needed for old_init_ucred_reint() */
1249         uc->uc_fsuid = rec->rn_fsuid;
1250         uc->uc_fsgid = rec->rn_fsgid;
1251         uc->uc_cap   = rec->rn_cap;
1252         uc->uc_suppgids[0] = rec->rn_suppgid1;
1253         uc->uc_suppgids[1] = rec->rn_suppgid2;
1254
1255         attr->la_uid = rec->rn_fsuid;
1256         attr->la_gid = rec->rn_fsgid;
1257         rr->rr_fid1 = &rec->rn_fid1;
1258         rr->rr_fid2 = &rec->rn_fid2;
1259         attr->la_ctime = rec->rn_time;
1260         attr->la_mtime = rec->rn_time;
1261         /* rename_tgt contains the mode already */
1262         attr->la_mode = rec->rn_mode;
1263         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1264
1265         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1266                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1267                                  req_capsule_client_get(pill, &RMF_CAPA1));
1268         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1269                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1270                                  req_capsule_client_get(pill, &RMF_CAPA2));
1271
1272         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1273         if (rc < 0)
1274                 RETURN(rc);
1275
1276         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1277         if (rc < 0)
1278                 RETURN(rc);
1279
1280         if (rec->rn_bias & MDS_VTX_BYPASS)
1281                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1282         else
1283                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1284
1285         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1286
1287
1288         rc = mdt_dlmreq_unpack(info);
1289
1290         RETURN(rc);
1291 }
1292
1293 /*
1294  * please see comment above LOV_MAGIC_V1_DEF
1295  */
1296 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1297 {
1298         struct mdt_reint_record *rr = &info->mti_rr;
1299         struct lov_user_md_v1   *v1;
1300
1301         v1 = (void *)rr->rr_eadata;
1302         LASSERT(v1);
1303
1304         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1305                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1306                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1307                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1308                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1309                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1310                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1311                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1312                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1313                 }
1314         }
1315 }
1316
1317 static int mdt_open_unpack(struct mdt_thread_info *info)
1318 {
1319         struct lu_ucred         *uc = mdt_ucred(info);
1320         struct mdt_rec_create   *rec;
1321         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1322         struct req_capsule      *pill = info->mti_pill;
1323         struct mdt_reint_record *rr   = &info->mti_rr;
1324         struct ptlrpc_request   *req  = mdt_info_req(info);
1325         struct md_op_spec       *sp   = &info->mti_spec;
1326         ENTRY;
1327
1328         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1329         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1330         if (rec == NULL)
1331                 RETURN(-EFAULT);
1332
1333         /* This prior initialization is needed for old_init_ucred_reint() */
1334         uc->uc_fsuid = rec->cr_fsuid;
1335         uc->uc_fsgid = rec->cr_fsgid;
1336         uc->uc_cap   = rec->cr_cap;
1337         uc->uc_suppgids[0] = rec->cr_suppgid1;
1338         uc->uc_suppgids[1] = rec->cr_suppgid2;
1339         uc->uc_umask = rec->cr_umask;
1340
1341         rr->rr_fid1   = &rec->cr_fid1;
1342         rr->rr_fid2   = &rec->cr_fid2;
1343         rr->rr_handle = &rec->cr_old_handle;
1344         attr->la_mode = rec->cr_mode;
1345         attr->la_rdev  = rec->cr_rdev;
1346         attr->la_uid   = rec->cr_fsuid;
1347         attr->la_gid   = rec->cr_fsgid;
1348         attr->la_ctime = rec->cr_time;
1349         attr->la_mtime = rec->cr_time;
1350         attr->la_atime = rec->cr_time;
1351         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1352                          LA_CTIME | LA_MTIME | LA_ATIME;
1353         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1354         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1355         /* Do not trigger ASSERTION if client miss to set such flags. */
1356         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1357                 RETURN(-EPROTO);
1358         info->mti_replayepoch = rec->cr_ioepoch;
1359
1360         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1361
1362         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1363                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1364                                  req_capsule_client_get(pill, &RMF_CAPA1));
1365         if (req_is_replay(req) &&
1366             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1367 #if 0
1368                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1369                                  req_capsule_client_get(pill, &RMF_CAPA2));
1370 #else
1371                 /*
1372                  * FIXME: capa in replay open request might have expired,
1373                  * bypass capa check. Security hole?
1374                  */
1375                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1376                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1377 #endif
1378         }
1379
1380         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1381
1382         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1383                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1384                                                         RCL_CLIENT);
1385                 if (rr->rr_eadatalen > 0) {
1386                         rr->rr_eadata = req_capsule_client_get(pill,
1387                                                                &RMF_EADATA);
1388                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1389                         sp->u.sp_ea.eadata = rr->rr_eadata;
1390                         sp->no_create = !!req_is_replay(req);
1391                         mdt_fix_lov_magic(info);
1392                 }
1393
1394                 /*
1395                  * Client default md_size may be 0 right after client start,
1396                  * until all osc are connected, set here just some reasonable
1397                  * value to prevent misbehavior.
1398                  */
1399                 if (rr->rr_eadatalen == 0 &&
1400                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1401                         rr->rr_eadatalen = MIN_MD_SIZE;
1402         }
1403
1404         RETURN(0);
1405 }
1406
1407 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1408 {
1409         struct mdt_reint_record *rr     = &info->mti_rr;
1410         struct lu_ucred         *uc     = mdt_ucred(info);
1411         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1412         struct req_capsule      *pill   = info->mti_pill;
1413         struct mdt_rec_setxattr *rec;
1414         int                      rc;
1415         ENTRY;
1416
1417
1418         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1419                          sizeof(struct mdt_rec_reint));
1420
1421         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1422         if (rec == NULL)
1423                 RETURN(-EFAULT);
1424
1425         /* This prior initialization is needed for old_init_ucred_reint() */
1426         uc->uc_fsuid  = rec->sx_fsuid;
1427         uc->uc_fsgid  = rec->sx_fsgid;
1428         uc->uc_cap    = rec->sx_cap;
1429         uc->uc_suppgids[0] = rec->sx_suppgid1;
1430         uc->uc_suppgids[1] = -1;
1431
1432         rr->rr_opcode = rec->sx_opcode;
1433         rr->rr_fid1   = &rec->sx_fid;
1434         attr->la_valid = rec->sx_valid;
1435         attr->la_ctime = rec->sx_time;
1436         attr->la_size = rec->sx_size;
1437         attr->la_flags = rec->sx_flags;
1438
1439         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1440                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1441                                  req_capsule_client_get(pill, &RMF_CAPA1));
1442         else
1443                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1444
1445         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1446         if (rc < 0)
1447                 RETURN(rc);
1448
1449         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1450                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1451                                                         RCL_CLIENT);
1452                 if (rr->rr_eadatalen > 0) {
1453                         rr->rr_eadata = req_capsule_client_get(pill,
1454                                                                &RMF_EADATA);
1455                         if (rr->rr_eadata == NULL)
1456                                 RETURN(-EFAULT);
1457                 } else {
1458                         rr->rr_eadata = NULL;
1459                 }
1460         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1461                 CDEBUG(D_INFO, "no xattr data supplied\n");
1462                 RETURN(-EFAULT);
1463         }
1464
1465         if (mdt_dlmreq_unpack(info) < 0)
1466                 RETURN(-EPROTO);
1467
1468         RETURN(0);
1469 }
1470
1471
1472 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1473
1474 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1475         [REINT_SETATTR]  = mdt_setattr_unpack,
1476         [REINT_CREATE]   = mdt_create_unpack,
1477         [REINT_LINK]     = mdt_link_unpack,
1478         [REINT_UNLINK]   = mdt_unlink_unpack,
1479         [REINT_RENAME]   = mdt_rename_unpack,
1480         [REINT_OPEN]     = mdt_open_unpack,
1481         [REINT_SETXATTR] = mdt_setxattr_unpack,
1482         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1483         [REINT_MIGRATE]  = mdt_rename_unpack,
1484 };
1485
1486 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1487 {
1488         int rc;
1489         ENTRY;
1490
1491         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1492         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1493                 info->mti_rr.rr_opcode = op;
1494                 rc = mdt_reint_unpackers[op](info);
1495         } else {
1496                 CERROR("Unexpected opcode %d\n", op);
1497                 rc = -EFAULT;
1498         }
1499         RETURN(rc);
1500 }