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, 2014, 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 #include <lustre_nodemap.h>
  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                                struct list_head *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         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
  99         ENTRY;
 100
 101         LASSERT(ucred != NULL);
 102         if (!squash->rsi_uid || ucred->uc_fsuid)
 103                 RETURN(0);
 104
 105         if (match_nosquash_list(&squash->rsi_sem,
 106                                 &squash->rsi_nosquash_nids,
 107                                 peernid)) {
 108                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 109                        libcfs_nid2str(peernid));
 110                 RETURN(0);
 111         }
 112
 113         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 114                libcfs_nid2str(peernid),
 115                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 116                squash->rsi_uid, squash->rsi_gid, 0);
 117
 118         ucred->uc_fsuid = squash->rsi_uid;
 119         ucred->uc_fsgid = squash->rsi_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 void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 128 {
 129         struct ptlrpc_request   *req = mdt_info_req(info);
 130         const char              *jobid = mdt_req_get_jobid(req);
 131
 132         /* set jobid if specified. */
 133         if (jobid)
 134                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 135         else
 136                 uc->uc_jobid[0] = '\0';
 137 }
 138
 139 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 140                           void *buf)
 141 {
 142         struct ptlrpc_request   *req = mdt_info_req(info);
 143         struct mdt_device       *mdt = info->mti_mdt;
 144         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 145         struct lu_ucred         *ucred = mdt_ucred(info);
 146         lnet_nid_t               peernid = req->rq_peer.nid;
 147         __u32                    perm = 0;
 148         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 149         int                      setuid;
 150         int                      setgid;
 151         int                      rc = 0;
 152
 153         ENTRY;
 154
 155         LASSERT(req->rq_auth_gss);
 156         LASSERT(!req->rq_auth_usr_mdt);
 157         LASSERT(req->rq_user_desc);
 158         LASSERT(ucred != NULL);
 159
 160         ucred->uc_valid = UCRED_INVALID;
 161
 162         ucred->uc_o_uid   = pud->pud_uid;
 163         ucred->uc_o_gid   = pud->pud_gid;
 164         ucred->uc_o_fsuid = pud->pud_fsuid;
 165         ucred->uc_o_fsgid = pud->pud_fsgid;
 166
 167         if (type == BODY_INIT) {
 168                 struct mdt_body *body = (struct mdt_body *)buf;
 169
 170                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 171                 ucred->uc_suppgids[1] = -1;
 172         }
 173
 174         /* sanity check: we expect the uid which client claimed is true */
 175         if (remote) {
 176                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 177                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 178                         RETURN(-EACCES);
 179                 }
 180
 181                 if (ptlrpc_user_desc_do_idmap(req, pud))
 182                         RETURN(-EACCES);
 183
 184                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 185                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 186                                "while client claims %u:%u/%u:%u\n",
 187                                libcfs_nid2str(peernid), req->rq_auth_uid,
 188                                req->rq_auth_mapped_uid,
 189                                pud->pud_uid, pud->pud_gid,
 190                                pud->pud_fsuid, pud->pud_fsgid);
 191                         RETURN(-EACCES);
 192                 }
 193         } else {
 194                 if (req->rq_auth_uid != pud->pud_uid) {
 195                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 196                                "while client claims %u:%u/%u:%u\n",
 197                                libcfs_nid2str(peernid), req->rq_auth_uid,
 198                                pud->pud_uid, pud->pud_gid,
 199                                pud->pud_fsuid, pud->pud_fsgid);
 200                         RETURN(-EACCES);
 201                 }
 202         }
 203
 204         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 205                 if (remote) {
 206                         CDEBUG(D_SEC, "remote client must run with identity_get "
 207                                "enabled!\n");
 208                         RETURN(-EACCES);
 209                 } else {
 210                         ucred->uc_identity = NULL;
 211                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 212                                CFS_SETGRP_PERM;
 213                 }
 214         } else {
 215                 struct md_identity *identity;
 216
 217                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 218                                             pud->pud_uid);
 219                 if (IS_ERR(identity)) {
 220                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 221                                      !remote)) {
 222                                 ucred->uc_identity = NULL;
 223                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 224                                        CFS_SETGRP_PERM;
 225                         } else {
 226                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 227                                        pud->pud_uid);
 228                                 RETURN(-EACCES);
 229                         }
 230                 } else {
 231                         ucred->uc_identity = identity;
 232                         perm = mdt_identity_get_perm(ucred->uc_identity,
 233                                                      remote, peernid);
 234                 }
 235         }
 236
 237         /* find out the setuid/setgid attempt */
 238         setuid = (pud->pud_uid != pud->pud_fsuid);
 239         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 240                   (ucred->uc_identity &&
 241                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 242
 243         /* check permission of setuid */
 244         if (setuid && !(perm & CFS_SETUID_PERM)) {
 245                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 246                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 247                 GOTO(out, rc = -EACCES);
 248         }
 249
 250         /* check permission of setgid */
 251         if (setgid && !(perm & CFS_SETGID_PERM)) {
 252                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 253                        "from %s\n", pud->pud_uid, pud->pud_gid,
 254                        pud->pud_fsuid, pud->pud_fsgid,
 255                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 256                 GOTO(out, rc = -EACCES);
 257         }
 258
 259         /*
 260          * NB: remote client not allowed to setgroups anyway.
 261          */
 262         if (!remote && perm & CFS_SETGRP_PERM) {
 263                 if (pud->pud_ngroups) {
 264                         /* setgroups for local client */
 265                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 266                         if (!ucred->uc_ginfo) {
 267                                 CERROR("failed to alloc %d groups\n",
 268                                        pud->pud_ngroups);
 269                                 GOTO(out, rc = -ENOMEM);
 270                         }
 271
 272                         lustre_groups_from_list(ucred->uc_ginfo,
 273                                                 pud->pud_groups);
 274                         lustre_groups_sort(ucred->uc_ginfo);
 275                 } else {
 276                         ucred->uc_ginfo = NULL;
 277                 }
 278         } else {
 279                 ucred->uc_suppgids[0] = -1;
 280                 ucred->uc_suppgids[1] = -1;
 281                 ucred->uc_ginfo = NULL;
 282         }
 283
 284         ucred->uc_uid   = pud->pud_uid;
 285         ucred->uc_gid   = pud->pud_gid;
 286         ucred->uc_fsuid = pud->pud_fsuid;
 287         ucred->uc_fsgid = pud->pud_fsgid;
 288
 289         /* process root_squash here. */
 290         mdt_root_squash(info, peernid);
 291
 292         /* remove fs privilege for non-root user. */
 293         if (ucred->uc_fsuid)
 294                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 295         else
 296                 ucred->uc_cap = pud->pud_cap;
 297         if (remote && !(perm & CFS_RMTOWN_PERM))
 298                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 299                                    CFS_CAP_CHOWN_MASK);
 300         ucred->uc_valid = UCRED_NEW;
 301         ucred_set_jobid(info, ucred);
 302
 303         EXIT;
 304
 305 out:
 306         if (rc) {
 307                 if (ucred->uc_ginfo) {
 308                         put_group_info(ucred->uc_ginfo);
 309                         ucred->uc_ginfo = NULL;
 310                 }
 311                 if (ucred->uc_identity) {
 312                         mdt_identity_put(mdt->mdt_identity_cache,
 313                                          ucred->uc_identity);
 314                         ucred->uc_identity = NULL;
 315                 }
 316         }
 317
 318         return rc;
 319 }
 320
 321 int mdt_check_ucred(struct mdt_thread_info *info)
 322 {
 323         struct ptlrpc_request   *req = mdt_info_req(info);
 324         struct mdt_device       *mdt = info->mti_mdt;
 325         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 326         struct lu_ucred         *ucred = mdt_ucred(info);
 327         struct md_identity      *identity = NULL;
 328         lnet_nid_t               peernid = req->rq_peer.nid;
 329         __u32                    perm = 0;
 330         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 331         int                      setuid;
 332         int                      setgid;
 333         int                      rc = 0;
 334
 335         ENTRY;
 336
 337         LASSERT(ucred != NULL);
 338         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 339                 RETURN(0);
 340
 341         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 342                 RETURN(0);
 343
 344         /* sanity check: if we use strong authentication, we expect the
 345          * uid which client claimed is true */
 346         if (remote) {
 347                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 348                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 349                         RETURN(-EACCES);
 350                 }
 351
 352                 if (ptlrpc_user_desc_do_idmap(req, pud))
 353                         RETURN(-EACCES);
 354
 355                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 356                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 357                                "while client claims %u:%u/%u:%u\n",
 358                                libcfs_nid2str(peernid), req->rq_auth_uid,
 359                                req->rq_auth_mapped_uid,
 360                                pud->pud_uid, pud->pud_gid,
 361                                pud->pud_fsuid, pud->pud_fsgid);
 362                         RETURN(-EACCES);
 363                 }
 364         } else {
 365                 if (req->rq_auth_uid != pud->pud_uid) {
 366                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 367                                "while client claims %u:%u/%u:%u\n",
 368                                libcfs_nid2str(peernid), req->rq_auth_uid,
 369                                pud->pud_uid, pud->pud_gid,
 370                                pud->pud_fsuid, pud->pud_fsgid);
 371                         RETURN(-EACCES);
 372                 }
 373         }
 374
 375         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 376                 if (remote) {
 377                         CDEBUG(D_SEC, "remote client must run with identity_get "
 378                                "enabled!\n");
 379                         RETURN(-EACCES);
 380                 }
 381                 RETURN(0);
 382         }
 383
 384         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 385         if (IS_ERR(identity)) {
 386                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 387                              !remote)) {
 388                         RETURN(0);
 389                 } else {
 390                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 391                                pud->pud_uid);
 392                         RETURN(-EACCES);
 393                }
 394         }
 395
 396         perm = mdt_identity_get_perm(identity, remote, peernid);
 397         /* find out the setuid/setgid attempt */
 398         setuid = (pud->pud_uid != pud->pud_fsuid);
 399         setgid = (pud->pud_gid != pud->pud_fsgid ||
 400                   pud->pud_gid != identity->mi_gid);
 401
 402         /* check permission of setuid */
 403         if (setuid && !(perm & CFS_SETUID_PERM)) {
 404                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 405                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 406                 GOTO(out, rc = -EACCES);
 407         }
 408
 409         /* check permission of setgid */
 410         if (setgid && !(perm & CFS_SETGID_PERM)) {
 411                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 412                        "from %s\n", pud->pud_uid, pud->pud_gid,
 413                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 414                        libcfs_nid2str(peernid));
 415                 GOTO(out, rc = -EACCES);
 416         }
 417
 418         EXIT;
 419
 420 out:
 421         mdt_identity_put(mdt->mdt_identity_cache, identity);
 422         return rc;
 423 }
 424
 425 static void mdt_squash_nodemap_id(struct lu_ucred *ucred,
 426                                   struct lu_nodemap *nodemap)
 427 {
 428         if (ucred->uc_o_uid == nodemap->nm_squash_uid) {
 429                 ucred->uc_fsuid = nodemap->nm_squash_uid;
 430                 ucred->uc_fsgid = nodemap->nm_squash_gid;
 431                 ucred->uc_cap = 0;
 432                 ucred->uc_suppgids[0] = -1;
 433                 ucred->uc_suppgids[1] = -1;
 434         }
 435 }
 436
 437
 438 static int old_init_ucred(struct mdt_thread_info *info,
 439                           struct mdt_body *body)
 440 {
 441         struct lu_ucred         *uc = mdt_ucred(info);
 442         struct mdt_device       *mdt = info->mti_mdt;
 443         struct md_identity      *identity = NULL;
 444         struct lu_nodemap       *nodemap =
 445                 info->mti_exp->exp_target_data.ted_nodemap;
 446         ENTRY;
 447
 448         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 449                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 450         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 451                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 452         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 453                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 454         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 455                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 456
 457         LASSERT(uc != NULL);
 458         uc->uc_valid = UCRED_INVALID;
 459         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 460         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 461         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 462         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 463         uc->uc_suppgids[0] = body->mbo_suppgid;
 464         uc->uc_suppgids[1] = -1;
 465         uc->uc_ginfo = NULL;
 466         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 467                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 468                                             uc->uc_fsuid);
 469                 if (IS_ERR(identity)) {
 470                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 471                                 identity = NULL;
 472                         } else {
 473                                 CDEBUG(D_SEC, "Deny access without identity: "
 474                                        "uid %u\n", uc->uc_fsuid);
 475                                 RETURN(-EACCES);
 476                         }
 477                 }
 478         }
 479         uc->uc_identity = identity;
 480
 481         mdt_squash_nodemap_id(uc, nodemap);
 482
 483         /* process root_squash here. */
 484         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 485
 486         /* remove fs privilege for non-root user. */
 487         if (uc->uc_fsuid)
 488                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 489         else
 490                 uc->uc_cap = body->mbo_capability;
 491         uc->uc_valid = UCRED_OLD;
 492         ucred_set_jobid(info, uc);
 493
 494         RETURN(0);
 495 }
 496
 497 static int old_init_ucred_reint(struct mdt_thread_info *info)
 498 {
 499         struct lu_ucred         *uc = mdt_ucred(info);
 500         struct mdt_device       *mdt = info->mti_mdt;
 501         struct md_identity      *identity = NULL;
 502         struct lu_nodemap       *nodemap =
 503                 info->mti_exp->exp_target_data.ted_nodemap;
 504         ENTRY;
 505
 506         LASSERT(uc != NULL);
 507
 508         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 509                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 510         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 511                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 512
 513         uc->uc_valid = UCRED_INVALID;
 514         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 515         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 516         uc->uc_ginfo = NULL;
 517
 518         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 519                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 520                                             uc->uc_fsuid);
 521                 if (IS_ERR(identity)) {
 522                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 523                                 identity = NULL;
 524                         } else {
 525                                 CDEBUG(D_SEC, "Deny access without identity: "
 526                                        "uid %u\n", uc->uc_fsuid);
 527                                 RETURN(-EACCES);
 528                         }
 529                 }
 530         }
 531         uc->uc_identity = identity;
 532
 533         /* process root_squash here. */
 534         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 535
 536         /* remove fs privilege for non-root user. */
 537         if (uc->uc_fsuid)
 538                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 539         uc->uc_valid = UCRED_OLD;
 540         ucred_set_jobid(info, uc);
 541
 542         RETURN(0);
 543 }
 544
 545 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 546 {
 547         struct ptlrpc_request *req = mdt_info_req(info);
 548         struct lu_ucred       *uc  = mdt_ucred(info);
 549
 550         LASSERT(uc != NULL);
 551         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 552                 return 0;
 553
 554         mdt_exit_ucred(info);
 555
 556         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 557                 return old_init_ucred(info, body);
 558         else
 559                 return new_init_ucred(info, BODY_INIT, body);
 560 }
 561
 562 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 563 {
 564         struct ptlrpc_request *req = mdt_info_req(info);
 565         struct lu_ucred       *uc  = mdt_ucred(info);
 566
 567         LASSERT(uc != NULL);
 568         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 569                 return 0;
 570
 571         mdt_exit_ucred(info);
 572
 573         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 574                 return old_init_ucred_reint(info);
 575         else
 576                 return new_init_ucred(info, REC_INIT, NULL);
 577 }
 578
 579 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 580 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 581 {
 582         const struct lov_ost_data_v1    *lod;
 583         int                              i;
 584         __u16                            count;
 585
 586         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 587                 return;
 588
 589         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 590
 591         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 592                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 593                le32_to_cpu(lmm->lmm_pattern));
 594         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 595                le32_to_cpu(lmm->lmm_stripe_size), count);
 596
 597         /* If it's a directory or a released file, then there are
 598          * no actual objects to print, so bail out. */
 599         if (valid & OBD_MD_FLDIREA ||
 600             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 601                 return;
 602
 603         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 604         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 605                 struct ost_id oi;
 606
 607                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 608                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 609                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 610         }
 611 }
 612
 613 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 614 {
 615         const struct lmv_mds_md_v1 *lmm1;
 616         int                        i;
 617
 618         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 619                 return;
 620
 621         lmm1 = &lmv->lmv_md_v1;
 622         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 623                le32_to_cpu(lmm1->lmv_magic),
 624                le32_to_cpu(lmm1->lmv_master_mdt_index),
 625                le32_to_cpu(lmm1->lmv_stripe_count));
 626
 627         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 628                 return;
 629
 630         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 631                 struct lu_fid fid;
 632
 633                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 634                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 635         }
 636 }
 637
 638 /* Shrink and/or grow reply buffers */
 639 int mdt_fix_reply(struct mdt_thread_info *info)
 640 {
 641         struct req_capsule *pill = info->mti_pill;
 642         struct mdt_body    *body;
 643         int                md_size, md_packed = 0;
 644         int                acl_size;
 645         int                rc = 0;
 646         ENTRY;
 647
 648         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 649         LASSERT(body != NULL);
 650
 651         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 652                                OBD_MD_LINKNAME))
 653                 md_size = body->mbo_eadatasize;
 654         else
 655                 md_size = 0;
 656
 657         acl_size = body->mbo_aclsize;
 658
 659         /* this replay - not send info to client */
 660         if (info->mti_spec.no_create) {
 661                 md_size = 0;
 662                 acl_size = 0;
 663         }
 664
 665         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d"
 666                " MDSCAPA = %llx, OSSCAPA = %llx\n",
 667                md_size, acl_size,
 668                (unsigned long long)(body->mbo_valid & OBD_MD_FLMDSCAPA),
 669                (unsigned long long)(body->mbo_valid & OBD_MD_FLOSSCAPA));
 670 /*
 671             &RMF_MDT_BODY,
 672             &RMF_MDT_MD,
 673             &RMF_ACL, or &RMF_LOGCOOKIES
 674 (optional)  &RMF_CAPA1,
 675 (optional)  &RMF_CAPA2,
 676 (optional)  something else
 677 */
 678
 679         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 680          * buffers before growing it */
 681         if (info->mti_big_lmm_used) {
 682                 /* big_lmm buffer may be used even without packing the result
 683                  * into reply, just for internal server needs */
 684                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 685                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 686                                                          RCL_SERVER);
 687
 688                 /* free big lmm if md_size is not needed */
 689                 if (md_size == 0 || md_packed == 0) {
 690                         info->mti_big_lmm_used = 0;
 691                 } else {
 692                         /* buffer must be allocated separately */
 693                         LASSERT(info->mti_attr.ma_lmm !=
 694                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 695                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 696                 }
 697         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 698                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 699         }
 700
 701         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 702                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 703         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 704                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 705                                    acl_size, RCL_SERVER);
 706
 707         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 708             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 709                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 710
 711         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 712             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 713                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 714
 715         /*
 716          * Some more field should be shrinked if needed.
 717          * This should be done by those who added fields to reply message.
 718          */
 719
 720         /* Grow MD buffer if needed finally */
 721         if (info->mti_big_lmm_used) {
 722                 void *lmm;
 723
 724                 LASSERT(md_size > md_packed);
 725                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 726                        md_size - md_packed);
 727                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 728                 if (rc) {
 729                         /* we can't answer with proper LOV EA, drop flags,
 730                          * the rc is also returned so this request is
 731                          * considered as failed */
 732                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 733                         /* don't return transno along with error */
 734                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 735                 } else {
 736                         /* now we need to pack right LOV/LMV EA */
 737                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 738                         if (info->mti_attr.ma_valid & MA_LOV) {
 739                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 740                                                              RCL_SERVER) ==
 741                                                 info->mti_attr.ma_lmm_size);
 742                                 memcpy(lmm, info->mti_attr.ma_lmm,
 743                                        info->mti_attr.ma_lmm_size);
 744                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 745                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 746                                                              RCL_SERVER) ==
 747                                                 info->mti_attr.ma_lmv_size);
 748                                 memcpy(lmm, info->mti_attr.ma_lmv,
 749                                        info->mti_attr.ma_lmv_size);
 750                         }
 751                 }
 752                 /* update mdt_max_mdsize so clients will be aware about that */
 753                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 754                         info->mti_mdt->mdt_max_mdsize =
 755                                                     info->mti_attr.ma_lmm_size;
 756                 info->mti_big_lmm_used = 0;
 757         }
 758         RETURN(rc);
 759 }
 760
 761
 762 /* if object is dying, pack the lov/llog data,
 763  * parameter info->mti_attr should be valid at this point! */
 764 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 765                            const struct md_attr *ma)
 766 {
 767         struct mdt_body       *repbody;
 768         const struct lu_attr *la = &ma->ma_attr;
 769         int rc;
 770         ENTRY;
 771
 772         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 773         LASSERT(repbody != NULL);
 774
 775         if (ma->ma_valid & MA_INODE)
 776                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 777
 778         if (ma->ma_valid & MA_LOV) {
 779                 CERROR("No need in LOV EA upon unlink\n");
 780                 dump_stack();
 781         }
 782         repbody->mbo_eadatasize = 0;
 783
 784         if (info->mti_mdt->mdt_lut.lut_oss_capa &&
 785             exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
 786             repbody->mbo_valid & OBD_MD_FLEASIZE) {
 787                 struct lustre_capa *capa;
 788
 789                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 790                 LASSERT(capa);
 791                 capa->lc_opc = CAPA_OPC_OSS_DESTROY;
 792                 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
 793                 if (rc)
 794                         RETURN(rc);
 795
 796                 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
 797         }
 798
 799         RETURN(0);
 800 }
 801
 802 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 803                                   struct md_attr *ma)
 804 {
 805         __u64 out;
 806
 807         out = 0;
 808         if (in & MDS_ATTR_MODE)
 809                 out |= LA_MODE;
 810         if (in & MDS_ATTR_UID)
 811                 out |= LA_UID;
 812         if (in & MDS_ATTR_GID)
 813                 out |= LA_GID;
 814         if (in & MDS_ATTR_SIZE)
 815                 out |= LA_SIZE;
 816         if (in & MDS_ATTR_BLOCKS)
 817                 out |= LA_BLOCKS;
 818         if (in & MDS_ATTR_ATIME_SET)
 819                 out |= LA_ATIME;
 820         if (in & MDS_ATTR_CTIME_SET)
 821                 out |= LA_CTIME;
 822         if (in & MDS_ATTR_MTIME_SET)
 823                 out |= LA_MTIME;
 824         if (in & MDS_ATTR_ATTR_FLAG)
 825                 out |= LA_FLAGS;
 826         if (in & MDS_ATTR_KILL_SUID)
 827                 out |= LA_KILL_SUID;
 828         if (in & MDS_ATTR_KILL_SGID)
 829                 out |= LA_KILL_SGID;
 830
 831         if (in & MDS_ATTR_FROM_OPEN)
 832                 rr->rr_flags |= MRF_OPEN_TRUNC;
 833         if (in & MDS_OPEN_OWNEROVERRIDE)
 834                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 835         if (in & MDS_ATTR_FORCE)
 836                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 837
 838         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 839                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 840                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 841                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 842                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 843                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 844         if (in != 0)
 845                 CERROR("Unknown attr bits: "LPX64"\n", in);
 846         return out;
 847 }
 848
 849 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
 850                       const struct lu_fid *fid, struct lustre_capa *capa)
 851 {
 852         struct lu_capainfo *lci;
 853
 854         LASSERT(offset >= 0 && offset < LU_CAPAINFO_MAX);
 855         if (!info->mti_mdt->mdt_lut.lut_mds_capa ||
 856             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
 857                 return;
 858
 859         lci = lu_capainfo_get(info->mti_env);
 860         LASSERT(lci);
 861         lci->lci_fid[offset]  = *fid;
 862         lci->lci_capa[offset] = capa;
 863 }
 864
 865 #ifdef DEBUG_CAPA
 866 void mdt_dump_capainfo(struct mdt_thread_info *info)
 867 {
 868         struct lu_capainfo *lci = lu_capainfo_get(info->mti_env);
 869         int i;
 870
 871         if (lci == NULL)
 872                 return;
 873
 874         for (i = 0; i < LU_CAPAINFO_MAX; i++) {
 875                 if (lci->lci_capa[i] == NULL) {
 876                         CERROR("no capa for index %d "DFID"\n",
 877                                i, PFID(&lci->lci_fid[i]));
 878                         continue;
 879                 }
 880                 if (lci->lci_capa[i] == BYPASS_CAPA) {
 881                         CERROR("bypass for index %d "DFID"\n",
 882                                i, PFID(&lci->lci_fid[i]));
 883                         continue;
 884                 }
 885                 DEBUG_CAPA(D_ERROR, lci->lci_capa[i], "index %d", i);
 886         }
 887 }
 888 #endif /* DEBUG_CAPA */
 889
 890 /* unpacking */
 891
 892 int mdt_name_unpack(struct req_capsule *pill,
 893                     const struct req_msg_field *field,
 894                     struct lu_name *ln,
 895                     enum mdt_name_flags flags)
 896 {
 897         ln->ln_name = req_capsule_client_get(pill, field);
 898         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 899
 900         if (!lu_name_is_valid(ln)) {
 901                 ln->ln_name = NULL;
 902                 ln->ln_namelen = 0;
 903
 904                 return -EPROTO;
 905         }
 906
 907         if ((flags & MNF_FIX_ANON) &&
 908             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 909                 /* Newer (3.x) kernels use a name of "/" for the
 910                  * "anonymous" disconnected dentries from NFS
 911                  * filehandle conversion. See d_obtain_alias(). */
 912                 ln->ln_name = NULL;
 913                 ln->ln_namelen = 0;
 914         }
 915
 916         return 0;
 917 }
 918
 919 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 920 {
 921         struct lu_ucred         *uc = mdt_ucred(info);
 922         struct md_attr          *ma = &info->mti_attr;
 923         struct lu_attr          *la = &ma->ma_attr;
 924         struct req_capsule      *pill = info->mti_pill;
 925         struct mdt_reint_record *rr = &info->mti_rr;
 926         struct mdt_rec_setattr  *rec;
 927         struct lu_nodemap       *nodemap =
 928                 info->mti_exp->exp_target_data.ted_nodemap;
 929         ENTRY;
 930
 931         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 932         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 933         if (rec == NULL)
 934                 RETURN(-EFAULT);
 935
 936         /* This prior initialization is needed for old_init_ucred_reint() */
 937         uc->uc_fsuid = rec->sa_fsuid;
 938         uc->uc_fsgid = rec->sa_fsgid;
 939         uc->uc_cap   = rec->sa_cap;
 940         uc->uc_suppgids[0] = rec->sa_suppgid;
 941         uc->uc_suppgids[1] = -1;
 942
 943         rr->rr_fid1 = &rec->sa_fid;
 944         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 945         /*  If MDS_ATTR_xTIME is set without MDS_ATTR_xTIME_SET and
 946          *  the client does not have OBD_CONNECT_FULL20, convert it
 947          *  to LA_xTIME. LU-3036 */
 948         if (!(exp_connect_flags(info->mti_exp) & OBD_CONNECT_FULL20)) {
 949                 if (!(rec->sa_valid & MDS_ATTR_ATIME_SET) &&
 950                      (rec->sa_valid & MDS_ATTR_ATIME))
 951                         la->la_valid |= LA_ATIME;
 952                 if (!(rec->sa_valid & MDS_ATTR_MTIME_SET) &&
 953                      (rec->sa_valid & MDS_ATTR_MTIME))
 954                         la->la_valid |= LA_MTIME;
 955                 if (!(rec->sa_valid & MDS_ATTR_CTIME_SET) &&
 956                      (rec->sa_valid & MDS_ATTR_CTIME))
 957                         la->la_valid |= LA_CTIME;
 958         }
 959         la->la_mode  = rec->sa_mode;
 960         la->la_flags = rec->sa_attr_flags;
 961         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 962                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 963         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 964                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 965         la->la_size  = rec->sa_size;
 966         la->la_blocks = rec->sa_blocks;
 967         la->la_ctime = rec->sa_ctime;
 968         la->la_atime = rec->sa_atime;
 969         la->la_mtime = rec->sa_mtime;
 970         ma->ma_valid = MA_INODE;
 971
 972         if (rec->sa_bias & MDS_DATA_MODIFIED)
 973                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 974         else
 975                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 976
 977         if (rec->sa_bias & MDS_HSM_RELEASE)
 978                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 979         else
 980                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 981
 982         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 983                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 984                                  req_capsule_client_get(pill, &RMF_CAPA1));
 985
 986         RETURN(0);
 987 }
 988
 989 static int mdt_ioepoch_unpack(struct mdt_thread_info *info)
 990 {
 991         struct req_capsule *pill = info->mti_pill;
 992         ENTRY;
 993
 994         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 995                 info->mti_ioepoch =
 996                         req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 997         else
 998                 info->mti_ioepoch = NULL;
 999         RETURN(info->mti_ioepoch == NULL ? -EFAULT : 0);
1000 }
1001
1002 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1003         struct req_capsule      *pill = info->mti_pill;
1004
1005         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1006                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1007                 if (info->mti_dlm_req == NULL)
1008                         RETURN(-EFAULT);
1009         }
1010
1011         RETURN(0);
1012 }
1013
1014 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1015 {
1016         struct mdt_reint_record *rr = &info->mti_rr;
1017         struct md_attr          *ma = &info->mti_attr;
1018         struct req_capsule      *pill = info->mti_pill;
1019         int rc;
1020         ENTRY;
1021
1022         rc = mdt_setattr_unpack_rec(info);
1023         if (rc)
1024                 RETURN(rc);
1025
1026         /* Epoch may be absent */
1027         mdt_ioepoch_unpack(info);
1028
1029         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1030                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1031                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1032                                                         RCL_CLIENT);
1033                 if (rr->rr_eadatalen > 0) {
1034                         const struct lmv_user_md        *lum;
1035
1036                         lum = rr->rr_eadata;
1037                         /* Sigh ma_valid(from req) does not indicate whether
1038                          * it will set LOV/LMV EA, so we have to check magic */
1039                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1040                                 ma->ma_valid |= MA_LMV;
1041                                 ma->ma_lmv = (void *)rr->rr_eadata;
1042                                 ma->ma_lmv_size = rr->rr_eadatalen;
1043                         } else {
1044                                 ma->ma_valid |= MA_LOV;
1045                                 ma->ma_lmm = (void *)rr->rr_eadata;
1046                                 ma->ma_lmm_size = rr->rr_eadatalen;
1047                         }
1048                 }
1049         }
1050
1051         rc = mdt_dlmreq_unpack(info);
1052         RETURN(rc);
1053 }
1054
1055 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1056 {
1057         struct md_attr          *ma = &info->mti_attr;
1058         struct req_capsule      *pill = info->mti_pill;
1059         ENTRY;
1060
1061         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1062                 RETURN(0);
1063
1064         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1065
1066         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1067             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1068                 RETURN(-EFAULT);
1069
1070         RETURN(0);
1071 }
1072
1073 int mdt_close_unpack(struct mdt_thread_info *info)
1074 {
1075         int rc;
1076         ENTRY;
1077
1078         rc = mdt_ioepoch_unpack(info);
1079         if (rc)
1080                 RETURN(rc);
1081
1082         rc = mdt_setattr_unpack_rec(info);
1083         if (rc)
1084                 RETURN(rc);
1085
1086         rc = mdt_hsm_release_unpack(info);
1087         if (rc)
1088                 RETURN(rc);
1089
1090         RETURN(mdt_init_ucred_reint(info));
1091 }
1092
1093 static int mdt_create_unpack(struct mdt_thread_info *info)
1094 {
1095         struct lu_ucred         *uc  = mdt_ucred(info);
1096         struct mdt_rec_create   *rec;
1097         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1098         struct mdt_reint_record *rr = &info->mti_rr;
1099         struct req_capsule      *pill = info->mti_pill;
1100         struct md_op_spec       *sp = &info->mti_spec;
1101         int rc;
1102         ENTRY;
1103
1104         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1105         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1106         if (rec == NULL)
1107                 RETURN(-EFAULT);
1108
1109         /* This prior initialization is needed for old_init_ucred_reint() */
1110         uc->uc_fsuid = rec->cr_fsuid;
1111         uc->uc_fsgid = rec->cr_fsgid;
1112         uc->uc_cap   = rec->cr_cap;
1113         uc->uc_suppgids[0] = rec->cr_suppgid1;
1114         uc->uc_suppgids[1] = -1;
1115         uc->uc_umask = rec->cr_umask;
1116
1117         rr->rr_fid1 = &rec->cr_fid1;
1118         rr->rr_fid2 = &rec->cr_fid2;
1119         attr->la_mode = rec->cr_mode;
1120         attr->la_rdev  = rec->cr_rdev;
1121         attr->la_uid   = rec->cr_fsuid;
1122         attr->la_gid   = rec->cr_fsgid;
1123         attr->la_ctime = rec->cr_time;
1124         attr->la_mtime = rec->cr_time;
1125         attr->la_atime = rec->cr_time;
1126         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1127                          LA_CTIME | LA_MTIME | LA_ATIME;
1128         memset(&sp->u, 0, sizeof(sp->u));
1129         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1130
1131         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1132                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1133                                  req_capsule_client_get(pill, &RMF_CAPA1));
1134         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1135
1136         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1137         if (rc < 0)
1138                 RETURN(rc);
1139
1140         if (S_ISLNK(attr->la_mode)) {
1141                 const char *tgt = NULL;
1142
1143                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1144                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1145                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1146                         sp->u.sp_symname = tgt;
1147                 }
1148                 if (tgt == NULL)
1149                         RETURN(-EFAULT);
1150         } else {
1151                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1152                 if (S_ISDIR(attr->la_mode) &&
1153                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1154                         sp->u.sp_ea.eadata =
1155                                 req_capsule_client_get(pill, &RMF_EADATA);
1156                         sp->u.sp_ea.eadatalen =
1157                                 req_capsule_get_size(pill, &RMF_EADATA,
1158                                                      RCL_CLIENT);
1159                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1160                 }
1161         }
1162
1163         rc = mdt_dlmreq_unpack(info);
1164         RETURN(rc);
1165 }
1166
1167 static int mdt_link_unpack(struct mdt_thread_info *info)
1168 {
1169         struct lu_ucred         *uc  = mdt_ucred(info);
1170         struct mdt_rec_link     *rec;
1171         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1172         struct mdt_reint_record *rr = &info->mti_rr;
1173         struct req_capsule      *pill = info->mti_pill;
1174         int rc;
1175         ENTRY;
1176
1177         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1178         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1179         if (rec == NULL)
1180                 RETURN(-EFAULT);
1181
1182         /* This prior initialization is needed for old_init_ucred_reint() */
1183         uc->uc_fsuid = rec->lk_fsuid;
1184         uc->uc_fsgid = rec->lk_fsgid;
1185         uc->uc_cap   = rec->lk_cap;
1186         uc->uc_suppgids[0] = rec->lk_suppgid1;
1187         uc->uc_suppgids[1] = rec->lk_suppgid2;
1188
1189         attr->la_uid = rec->lk_fsuid;
1190         attr->la_gid = rec->lk_fsgid;
1191         rr->rr_fid1 = &rec->lk_fid1;
1192         rr->rr_fid2 = &rec->lk_fid2;
1193         attr->la_ctime = rec->lk_time;
1194         attr->la_mtime = rec->lk_time;
1195         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1196
1197         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1198                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1199                                  req_capsule_client_get(pill, &RMF_CAPA1));
1200         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1201                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1202                                  req_capsule_client_get(pill, &RMF_CAPA2));
1203
1204         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1205         if (rc < 0)
1206                 RETURN(rc);
1207
1208         rc = mdt_dlmreq_unpack(info);
1209
1210         RETURN(rc);
1211 }
1212
1213 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1214 {
1215         struct lu_ucred         *uc  = mdt_ucred(info);
1216         struct mdt_rec_unlink   *rec;
1217         struct md_attr          *ma = &info->mti_attr;
1218         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1219         struct mdt_reint_record *rr = &info->mti_rr;
1220         struct req_capsule      *pill = info->mti_pill;
1221         int rc;
1222         ENTRY;
1223
1224         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1225         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1226         if (rec == NULL)
1227                 RETURN(-EFAULT);
1228
1229         /* This prior initialization is needed for old_init_ucred_reint() */
1230         uc->uc_fsuid = rec->ul_fsuid;
1231         uc->uc_fsgid = rec->ul_fsgid;
1232         uc->uc_cap   = rec->ul_cap;
1233         uc->uc_suppgids[0] = rec->ul_suppgid1;
1234         uc->uc_suppgids[1] = -1;
1235
1236         attr->la_uid = rec->ul_fsuid;
1237         attr->la_gid = rec->ul_fsgid;
1238         rr->rr_fid1 = &rec->ul_fid1;
1239         rr->rr_fid2 = &rec->ul_fid2;
1240         attr->la_ctime = rec->ul_time;
1241         attr->la_mtime = rec->ul_time;
1242         attr->la_mode  = rec->ul_mode;
1243         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1244
1245         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1246                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1247                                  req_capsule_client_get(pill, &RMF_CAPA1));
1248
1249         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1250         if (rc < 0)
1251                 RETURN(rc);
1252
1253         if (rec->ul_bias & MDS_VTX_BYPASS)
1254                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1255         else
1256                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1257
1258         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1259
1260         rc = mdt_dlmreq_unpack(info);
1261         RETURN(rc);
1262 }
1263
1264 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1265 {
1266         info->mti_spec.sp_rm_entry = 1;
1267         return mdt_unlink_unpack(info);
1268 }
1269
1270 static int mdt_rename_unpack(struct mdt_thread_info *info)
1271 {
1272         struct lu_ucred         *uc = mdt_ucred(info);
1273         struct mdt_rec_rename   *rec;
1274         struct md_attr          *ma = &info->mti_attr;
1275         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1276         struct mdt_reint_record *rr = &info->mti_rr;
1277         struct req_capsule      *pill = info->mti_pill;
1278         int rc;
1279         ENTRY;
1280
1281         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1282         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1283         if (rec == NULL)
1284                 RETURN(-EFAULT);
1285
1286         /* This prior initialization is needed for old_init_ucred_reint() */
1287         uc->uc_fsuid = rec->rn_fsuid;
1288         uc->uc_fsgid = rec->rn_fsgid;
1289         uc->uc_cap   = rec->rn_cap;
1290         uc->uc_suppgids[0] = rec->rn_suppgid1;
1291         uc->uc_suppgids[1] = rec->rn_suppgid2;
1292
1293         attr->la_uid = rec->rn_fsuid;
1294         attr->la_gid = rec->rn_fsgid;
1295         rr->rr_fid1 = &rec->rn_fid1;
1296         rr->rr_fid2 = &rec->rn_fid2;
1297         attr->la_ctime = rec->rn_time;
1298         attr->la_mtime = rec->rn_time;
1299         /* rename_tgt contains the mode already */
1300         attr->la_mode = rec->rn_mode;
1301         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1302
1303         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1304                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1305                                  req_capsule_client_get(pill, &RMF_CAPA1));
1306         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
1307                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1308                                  req_capsule_client_get(pill, &RMF_CAPA2));
1309
1310         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1311         if (rc < 0)
1312                 RETURN(rc);
1313
1314         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1315         if (rc < 0)
1316                 RETURN(rc);
1317
1318         if (rec->rn_bias & MDS_VTX_BYPASS)
1319                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1320         else
1321                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1322
1323         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1324
1325
1326         rc = mdt_dlmreq_unpack(info);
1327
1328         RETURN(rc);
1329 }
1330
1331 /*
1332  * please see comment above LOV_MAGIC_V1_DEF
1333  */
1334 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1335 {
1336         struct mdt_reint_record *rr = &info->mti_rr;
1337         struct lov_user_md_v1   *v1;
1338
1339         v1 = (void *)rr->rr_eadata;
1340         LASSERT(v1);
1341
1342         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1343                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1344                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1345                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1346                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1347                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1348                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1349                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1350                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1351                 }
1352         }
1353 }
1354
1355 static int mdt_open_unpack(struct mdt_thread_info *info)
1356 {
1357         struct lu_ucred         *uc = mdt_ucred(info);
1358         struct mdt_rec_create   *rec;
1359         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1360         struct req_capsule      *pill = info->mti_pill;
1361         struct mdt_reint_record *rr   = &info->mti_rr;
1362         struct ptlrpc_request   *req  = mdt_info_req(info);
1363         struct md_op_spec       *sp   = &info->mti_spec;
1364         ENTRY;
1365
1366         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1367         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1368         if (rec == NULL)
1369                 RETURN(-EFAULT);
1370
1371         /* This prior initialization is needed for old_init_ucred_reint() */
1372         uc->uc_fsuid = rec->cr_fsuid;
1373         uc->uc_fsgid = rec->cr_fsgid;
1374         uc->uc_cap   = rec->cr_cap;
1375         uc->uc_suppgids[0] = rec->cr_suppgid1;
1376         uc->uc_suppgids[1] = rec->cr_suppgid2;
1377         uc->uc_umask = rec->cr_umask;
1378
1379         rr->rr_fid1   = &rec->cr_fid1;
1380         rr->rr_fid2   = &rec->cr_fid2;
1381         rr->rr_handle = &rec->cr_old_handle;
1382         attr->la_mode = rec->cr_mode;
1383         attr->la_rdev  = rec->cr_rdev;
1384         attr->la_uid   = rec->cr_fsuid;
1385         attr->la_gid   = rec->cr_fsgid;
1386         attr->la_ctime = rec->cr_time;
1387         attr->la_mtime = rec->cr_time;
1388         attr->la_atime = rec->cr_time;
1389         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1390                          LA_CTIME | LA_MTIME | LA_ATIME;
1391         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1392         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1393         /* Do not trigger ASSERTION if client miss to set such flags. */
1394         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1395                 RETURN(-EPROTO);
1396         info->mti_replayepoch = rec->cr_ioepoch;
1397
1398         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1399
1400         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1401                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1402                                  req_capsule_client_get(pill, &RMF_CAPA1));
1403         if (req_is_replay(req) &&
1404             req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)) {
1405 #if 0
1406                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1407                                  req_capsule_client_get(pill, &RMF_CAPA2));
1408 #else
1409                 /*
1410                  * FIXME: capa in replay open request might have expired,
1411                  * bypass capa check. Security hole?
1412                  */
1413                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1414                 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
1415 #endif
1416         }
1417
1418         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1419
1420         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1421                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1422                                                         RCL_CLIENT);
1423                 if (rr->rr_eadatalen > 0) {
1424                         rr->rr_eadata = req_capsule_client_get(pill,
1425                                                                &RMF_EADATA);
1426                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1427                         sp->u.sp_ea.eadata = rr->rr_eadata;
1428                         sp->no_create = !!req_is_replay(req);
1429                         mdt_fix_lov_magic(info);
1430                 }
1431
1432                 /*
1433                  * Client default md_size may be 0 right after client start,
1434                  * until all osc are connected, set here just some reasonable
1435                  * value to prevent misbehavior.
1436                  */
1437                 if (rr->rr_eadatalen == 0 &&
1438                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1439                         rr->rr_eadatalen = MIN_MD_SIZE;
1440         }
1441
1442         RETURN(0);
1443 }
1444
1445 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1446 {
1447         struct mdt_reint_record *rr     = &info->mti_rr;
1448         struct lu_ucred         *uc     = mdt_ucred(info);
1449         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1450         struct req_capsule      *pill   = info->mti_pill;
1451         struct mdt_rec_setxattr *rec;
1452         int                      rc;
1453         ENTRY;
1454
1455
1456         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1457                          sizeof(struct mdt_rec_reint));
1458
1459         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1460         if (rec == NULL)
1461                 RETURN(-EFAULT);
1462
1463         /* This prior initialization is needed for old_init_ucred_reint() */
1464         uc->uc_fsuid  = rec->sx_fsuid;
1465         uc->uc_fsgid  = rec->sx_fsgid;
1466         uc->uc_cap    = rec->sx_cap;
1467         uc->uc_suppgids[0] = rec->sx_suppgid1;
1468         uc->uc_suppgids[1] = -1;
1469
1470         rr->rr_opcode = rec->sx_opcode;
1471         rr->rr_fid1   = &rec->sx_fid;
1472         attr->la_valid = rec->sx_valid;
1473         attr->la_ctime = rec->sx_time;
1474         attr->la_size = rec->sx_size;
1475         attr->la_flags = rec->sx_flags;
1476
1477         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1478                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1479                                  req_capsule_client_get(pill, &RMF_CAPA1));
1480         else
1481                 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
1482
1483         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1484         if (rc < 0)
1485                 RETURN(rc);
1486
1487         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1488                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1489                                                         RCL_CLIENT);
1490                 if (rr->rr_eadatalen > 0) {
1491                         rr->rr_eadata = req_capsule_client_get(pill,
1492                                                                &RMF_EADATA);
1493                         if (rr->rr_eadata == NULL)
1494                                 RETURN(-EFAULT);
1495                 } else {
1496                         rr->rr_eadata = NULL;
1497                 }
1498         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1499                 CDEBUG(D_INFO, "no xattr data supplied\n");
1500                 RETURN(-EFAULT);
1501         }
1502
1503         if (mdt_dlmreq_unpack(info) < 0)
1504                 RETURN(-EPROTO);
1505
1506         RETURN(0);
1507 }
1508
1509
1510 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1511
1512 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1513         [REINT_SETATTR]  = mdt_setattr_unpack,
1514         [REINT_CREATE]   = mdt_create_unpack,
1515         [REINT_LINK]     = mdt_link_unpack,
1516         [REINT_UNLINK]   = mdt_unlink_unpack,
1517         [REINT_RENAME]   = mdt_rename_unpack,
1518         [REINT_OPEN]     = mdt_open_unpack,
1519         [REINT_SETXATTR] = mdt_setxattr_unpack,
1520         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1521         [REINT_MIGRATE]  = mdt_rename_unpack,
1522 };
1523
1524 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1525 {
1526         int rc;
1527         ENTRY;
1528
1529         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1530         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1531                 info->mti_rr.rr_opcode = op;
1532                 rc = mdt_reint_unpackers[op](info);
1533         } else {
1534                 CERROR("Unexpected opcode %d\n", op);
1535                 rc = -EFAULT;
1536         }
1537         RETURN(rc);
1538 }