lustre/mdt/mdt_lib.c

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