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, 2015, 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 <linux/user_namespace.h>
  52 #ifdef HAVE_UIDGID_HEADER
  53 # include <linux/uidgid.h>
  54 #endif
  55 #include "mdt_internal.h"
  56 #include <lnet/nidstr.h>
  57 #include <lustre_nodemap.h>
  58
  59 typedef enum ucred_init_type {
  60         NONE_INIT       = 0,
  61         BODY_INIT       = 1,
  62         REC_INIT        = 2
  63 } ucred_init_type_t;
  64
  65 void mdt_exit_ucred(struct mdt_thread_info *info)
  66 {
  67         struct lu_ucred   *uc  = mdt_ucred(info);
  68         struct mdt_device *mdt = info->mti_mdt;
  69
  70         LASSERT(uc != NULL);
  71         if (uc->uc_valid != UCRED_INIT) {
  72                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  73                 if (uc->uc_ginfo) {
  74                         put_group_info(uc->uc_ginfo);
  75                         uc->uc_ginfo = NULL;
  76                 }
  77                 if (uc->uc_identity) {
  78                         mdt_identity_put(mdt->mdt_identity_cache,
  79                                          uc->uc_identity);
  80                         uc->uc_identity = NULL;
  81                 }
  82                 uc->uc_valid = UCRED_INIT;
  83         }
  84 }
  85
  86 static int match_nosquash_list(struct rw_semaphore *sem,
  87                                struct list_head *nidlist,
  88                                lnet_nid_t peernid)
  89 {
  90         int rc;
  91         ENTRY;
  92         down_read(sem);
  93         rc = cfs_match_nid(peernid, nidlist);
  94         up_read(sem);
  95         RETURN(rc);
  96 }
  97
  98 /* root_squash for inter-MDS operations */
  99 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 100 {
 101         struct lu_ucred *ucred = mdt_ucred(info);
 102         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 103         ENTRY;
 104
 105         LASSERT(ucred != NULL);
 106         if (!squash->rsi_uid || ucred->uc_fsuid)
 107                 RETURN(0);
 108
 109         if (match_nosquash_list(&squash->rsi_sem,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nid2str(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nid2str(peernid),
 119                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 120                squash->rsi_uid, squash->rsi_gid, 0);
 121
 122         ucred->uc_fsuid = squash->rsi_uid;
 123         ucred->uc_fsgid = squash->rsi_gid;
 124         ucred->uc_cap = 0;
 125         ucred->uc_suppgids[0] = -1;
 126         ucred->uc_suppgids[1] = -1;
 127
 128         RETURN(0);
 129 }
 130
 131 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 132 {
 133         struct ptlrpc_request   *req = mdt_info_req(info);
 134         const char              *jobid = mdt_req_get_jobid(req);
 135
 136         /* set jobid if specified. */
 137         if (jobid)
 138                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 139         else
 140                 uc->uc_jobid[0] = '\0';
 141 }
 142
 143 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 144                           void *buf, bool drop_fs_cap)
 145 {
 146         struct ptlrpc_request   *req = mdt_info_req(info);
 147         struct mdt_device       *mdt = info->mti_mdt;
 148         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 149         struct lu_ucred         *ucred = mdt_ucred(info);
 150         lnet_nid_t               peernid = req->rq_peer.nid;
 151         __u32                    perm = 0;
 152         int                      setuid;
 153         int                      setgid;
 154         int                      rc = 0;
 155
 156         ENTRY;
 157
 158         LASSERT(req->rq_auth_gss);
 159         LASSERT(!req->rq_auth_usr_mdt);
 160         LASSERT(req->rq_user_desc);
 161         LASSERT(ucred != NULL);
 162
 163         ucred->uc_valid = UCRED_INVALID;
 164
 165         ucred->uc_o_uid   = pud->pud_uid;
 166         ucred->uc_o_gid   = pud->pud_gid;
 167         ucred->uc_o_fsuid = pud->pud_fsuid;
 168         ucred->uc_o_fsgid = pud->pud_fsgid;
 169
 170         if (type == BODY_INIT) {
 171                 struct mdt_body *body = (struct mdt_body *)buf;
 172
 173                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 174                 ucred->uc_suppgids[1] = -1;
 175         }
 176
 177         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 178             req->rq_auth_uid != pud->pud_uid) {
 179                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 180                        "while client claims %u:%u/%u:%u\n",
 181                        libcfs_nid2str(peernid), req->rq_auth_uid,
 182                        pud->pud_uid, pud->pud_gid,
 183                        pud->pud_fsuid, pud->pud_fsgid);
 184                 RETURN(-EACCES);
 185         }
 186
 187         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 188                 ucred->uc_identity = NULL;
 189                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 190         } else {
 191                 struct md_identity *identity;
 192
 193                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 194                                             pud->pud_uid);
 195                 if (IS_ERR(identity)) {
 196                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 197                                 ucred->uc_identity = NULL;
 198                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 199                                        CFS_SETGRP_PERM;
 200                         } else {
 201                                 CDEBUG(D_SEC,
 202                                        "Deny access without identity: uid %u\n",
 203                                        pud->pud_uid);
 204                                 RETURN(-EACCES);
 205                         }
 206                 } else {
 207                         ucred->uc_identity = identity;
 208                         perm = mdt_identity_get_perm(ucred->uc_identity,
 209                                                      peernid);
 210                 }
 211         }
 212
 213         /* find out the setuid/setgid attempt */
 214         setuid = (pud->pud_uid != pud->pud_fsuid);
 215         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 216                   (ucred->uc_identity &&
 217                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 218
 219         /* check permission of setuid */
 220         if (setuid && !(perm & CFS_SETUID_PERM)) {
 221                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 222                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 223                 GOTO(out, rc = -EACCES);
 224         }
 225
 226         /* check permission of setgid */
 227         if (setgid && !(perm & CFS_SETGID_PERM)) {
 228                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 229                        "from %s\n", pud->pud_uid, pud->pud_gid,
 230                        pud->pud_fsuid, pud->pud_fsgid,
 231                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 232                 GOTO(out, rc = -EACCES);
 233         }
 234
 235         if (perm & CFS_SETGRP_PERM) {
 236                 if (pud->pud_ngroups) {
 237                         /* setgroups for local client */
 238                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 239                         if (!ucred->uc_ginfo) {
 240                                 CERROR("failed to alloc %d groups\n",
 241                                        pud->pud_ngroups);
 242                                 GOTO(out, rc = -ENOMEM);
 243                         }
 244
 245                         lustre_groups_from_list(ucred->uc_ginfo,
 246                                                 pud->pud_groups);
 247                         lustre_groups_sort(ucred->uc_ginfo);
 248                 } else {
 249                         ucred->uc_ginfo = NULL;
 250                 }
 251         } else {
 252                 ucred->uc_suppgids[0] = -1;
 253                 ucred->uc_suppgids[1] = -1;
 254                 ucred->uc_ginfo = NULL;
 255         }
 256
 257         ucred->uc_uid   = pud->pud_uid;
 258         ucred->uc_gid   = pud->pud_gid;
 259         ucred->uc_fsuid = pud->pud_fsuid;
 260         ucred->uc_fsgid = pud->pud_fsgid;
 261
 262         /* process root_squash here. */
 263         mdt_root_squash(info, peernid);
 264
 265         /* remove fs privilege for non-root user. */
 266         if (ucred->uc_fsuid && drop_fs_cap)
 267                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 268         else
 269                 ucred->uc_cap = pud->pud_cap;
 270         ucred->uc_valid = UCRED_NEW;
 271         ucred_set_jobid(info, ucred);
 272
 273         EXIT;
 274
 275 out:
 276         if (rc) {
 277                 if (ucred->uc_ginfo) {
 278                         put_group_info(ucred->uc_ginfo);
 279                         ucred->uc_ginfo = NULL;
 280                 }
 281                 if (ucred->uc_identity) {
 282                         mdt_identity_put(mdt->mdt_identity_cache,
 283                                          ucred->uc_identity);
 284                         ucred->uc_identity = NULL;
 285                 }
 286         }
 287
 288         return rc;
 289 }
 290
 291 /**
 292  * Check whether allow the client to set supplementary group IDs or not.
 293  *
 294  * \param[in] info      pointer to the thread context
 295  * \param[in] uc        pointer to the RPC user descriptor
 296  *
 297  * \retval              true if allow to set supplementary group IDs
 298  * \retval              false for other cases
 299  */
 300 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 301 {
 302         __u32 perm;
 303
 304         /* 1. If identity_upcall is disabled,
 305          *    permit local client to do anything. */
 306         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 307                 return true;
 308
 309         /* 2. If fail to get related identities, then forbid any client to
 310          *    set supplementary group IDs. */
 311         if (uc->uc_identity == NULL)
 312                 return false;
 313
 314         /* 3. Check the permission in the identities. */
 315         perm = mdt_identity_get_perm(uc->uc_identity,
 316                                      mdt_info_req(info)->rq_peer.nid);
 317         if (perm & CFS_SETGRP_PERM)
 318                 return true;
 319
 320         return false;
 321 }
 322
 323 int mdt_check_ucred(struct mdt_thread_info *info)
 324 {
 325         struct ptlrpc_request   *req = mdt_info_req(info);
 326         struct mdt_device       *mdt = info->mti_mdt;
 327         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 328         struct lu_ucred         *ucred = mdt_ucred(info);
 329         struct md_identity      *identity = NULL;
 330         lnet_nid_t               peernid = req->rq_peer.nid;
 331         __u32                    perm = 0;
 332         int                      setuid;
 333         int                      setgid;
 334         int                      rc = 0;
 335
 336         ENTRY;
 337
 338         LASSERT(ucred != NULL);
 339         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 340                 RETURN(0);
 341
 342         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 343                 RETURN(0);
 344
 345         /* sanity check: if we use strong authentication, we expect the
 346          * uid which client claimed is true */
 347         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 348             req->rq_auth_uid != pud->pud_uid) {
 349                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 350                        "while client claims %u:%u/%u:%u\n",
 351                        libcfs_nid2str(peernid), req->rq_auth_uid,
 352                        pud->pud_uid, pud->pud_gid,
 353                        pud->pud_fsuid, pud->pud_fsgid);
 354                 RETURN(-EACCES);
 355         }
 356
 357         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 358                 RETURN(0);
 359
 360         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 361         if (IS_ERR(identity)) {
 362                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 363                         RETURN(0);
 364                 } else {
 365                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 366                                pud->pud_uid);
 367                         RETURN(-EACCES);
 368                 }
 369         }
 370
 371         perm = mdt_identity_get_perm(identity, peernid);
 372         /* find out the setuid/setgid attempt */
 373         setuid = (pud->pud_uid != pud->pud_fsuid);
 374         setgid = (pud->pud_gid != pud->pud_fsgid ||
 375                   pud->pud_gid != identity->mi_gid);
 376
 377         /* check permission of setuid */
 378         if (setuid && !(perm & CFS_SETUID_PERM)) {
 379                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 380                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 381                 GOTO(out, rc = -EACCES);
 382         }
 383
 384         /* check permission of setgid */
 385         if (setgid && !(perm & CFS_SETGID_PERM)) {
 386                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 387                        "from %s\n", pud->pud_uid, pud->pud_gid,
 388                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 389                        libcfs_nid2str(peernid));
 390                 GOTO(out, rc = -EACCES);
 391         }
 392
 393         EXIT;
 394
 395 out:
 396         mdt_identity_put(mdt->mdt_identity_cache, identity);
 397         return rc;
 398 }
 399
 400 static int old_init_ucred_common(struct mdt_thread_info *info,
 401                                  struct lu_nodemap *nodemap,
 402                                  bool drop_fs_cap)
 403 {
 404         struct lu_ucred         *uc = mdt_ucred(info);
 405         struct mdt_device       *mdt = info->mti_mdt;
 406         struct md_identity      *identity = NULL;
 407
 408         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 409                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 410                                             uc->uc_fsuid);
 411                 if (IS_ERR(identity)) {
 412                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 413                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 414                                 identity = NULL;
 415                         } else {
 416                                 CDEBUG(D_SEC, "Deny access without identity: "
 417                                        "uid %u\n", uc->uc_fsuid);
 418                                 RETURN(-EACCES);
 419                         }
 420                 }
 421         }
 422         uc->uc_identity = identity;
 423
 424         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 425                 uc->uc_fsuid = nodemap->nm_squash_uid;
 426                 uc->uc_fsgid = nodemap->nm_squash_gid;
 427                 uc->uc_cap = 0;
 428                 uc->uc_suppgids[0] = -1;
 429                 uc->uc_suppgids[1] = -1;
 430         }
 431
 432         /* process root_squash here. */
 433         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 434
 435         /* remove fs privilege for non-root user. */
 436         if (uc->uc_fsuid && drop_fs_cap)
 437                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 438         uc->uc_valid = UCRED_OLD;
 439         ucred_set_jobid(info, uc);
 440
 441         return 0;
 442 }
 443
 444 static int old_init_ucred(struct mdt_thread_info *info,
 445                           struct mdt_body *body, bool drop_fs_cap)
 446 {
 447         struct lu_ucred *uc = mdt_ucred(info);
 448         struct lu_nodemap *nodemap;
 449         int rc;
 450         ENTRY;
 451
 452         nodemap = nodemap_get_from_exp(info->mti_exp);
 453         if (IS_ERR(nodemap))
 454                 RETURN(PTR_ERR(nodemap));
 455
 456         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 457                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 458         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 459                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 460         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 461                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 462         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 463                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 464
 465         LASSERT(uc != NULL);
 466         uc->uc_valid = UCRED_INVALID;
 467         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 468         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 469         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 470         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 471         uc->uc_suppgids[0] = body->mbo_suppgid;
 472         uc->uc_suppgids[1] = -1;
 473         uc->uc_ginfo = NULL;
 474         uc->uc_cap = body->mbo_capability;
 475
 476         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 477         nodemap_putref(nodemap);
 478
 479         RETURN(rc);
 480 }
 481
 482 static int old_init_ucred_reint(struct mdt_thread_info *info)
 483 {
 484         struct lu_ucred *uc = mdt_ucred(info);
 485         struct lu_nodemap *nodemap;
 486         int rc;
 487         ENTRY;
 488
 489         nodemap = nodemap_get_from_exp(info->mti_exp);
 490         if (IS_ERR(nodemap))
 491                 RETURN(PTR_ERR(nodemap));
 492
 493         LASSERT(uc != NULL);
 494
 495         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 496                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 497         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 498                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 499
 500         uc->uc_valid = UCRED_INVALID;
 501         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 502         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 503         uc->uc_ginfo = NULL;
 504
 505         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 506         nodemap_putref(nodemap);
 507
 508         RETURN(rc);
 509 }
 510
 511 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 512                                    struct mdt_body *body,
 513                                    bool drop_fs_cap)
 514 {
 515         struct ptlrpc_request   *req = mdt_info_req(info);
 516         struct lu_ucred         *uc  = mdt_ucred(info);
 517
 518         LASSERT(uc != NULL);
 519         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 520                 return 0;
 521
 522         mdt_exit_ucred(info);
 523
 524         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 525                 return old_init_ucred(info, body, drop_fs_cap);
 526         else
 527                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 528 }
 529
 530 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 531 {
 532         return __mdt_init_ucred(info, body, true);
 533 }
 534
 535 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 536  * doesn't set correct fsuid explicitely, but raise capability to allow
 537  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 538  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 539  * will fail permission check. */
 540 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 541                                   struct mdt_body *body)
 542 {
 543         return __mdt_init_ucred(info, body, false);
 544 }
 545
 546 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 547 {
 548         struct ptlrpc_request *req = mdt_info_req(info);
 549         struct lu_ucred       *uc  = mdt_ucred(info);
 550
 551         LASSERT(uc != NULL);
 552         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 553                 return 0;
 554
 555         mdt_exit_ucred(info);
 556
 557         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 558                 return old_init_ucred_reint(info);
 559         else
 560                 return new_init_ucred(info, REC_INIT, NULL, true);
 561 }
 562
 563 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 564 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 565 {
 566         const struct lov_ost_data_v1    *lod;
 567         int                              i;
 568         __u16                            count;
 569
 570         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 571                 return;
 572
 573         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 574
 575         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 576                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 577                le32_to_cpu(lmm->lmm_pattern));
 578         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 579                le32_to_cpu(lmm->lmm_stripe_size), count);
 580
 581         /* If it's a directory or a released file, then there are
 582          * no actual objects to print, so bail out. */
 583         if (valid & OBD_MD_FLDIREA ||
 584             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 585                 return;
 586
 587         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 588         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 589                 struct ost_id oi;
 590
 591                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 592                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 593                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 594         }
 595 }
 596
 597 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 598 {
 599         const struct lmv_mds_md_v1 *lmm1;
 600         int                        i;
 601
 602         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 603                 return;
 604
 605         lmm1 = &lmv->lmv_md_v1;
 606         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 607                le32_to_cpu(lmm1->lmv_magic),
 608                le32_to_cpu(lmm1->lmv_master_mdt_index),
 609                le32_to_cpu(lmm1->lmv_stripe_count));
 610
 611         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 612                 return;
 613
 614         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 615                 struct lu_fid fid;
 616
 617                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 618                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 619         }
 620 }
 621
 622 /* Shrink and/or grow reply buffers */
 623 int mdt_fix_reply(struct mdt_thread_info *info)
 624 {
 625         struct req_capsule *pill = info->mti_pill;
 626         struct mdt_body    *body;
 627         int                md_size, md_packed = 0;
 628         int                acl_size;
 629         int                rc = 0;
 630         ENTRY;
 631
 632         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 633         LASSERT(body != NULL);
 634
 635         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 636                                OBD_MD_LINKNAME))
 637                 md_size = body->mbo_eadatasize;
 638         else
 639                 md_size = 0;
 640
 641         acl_size = body->mbo_aclsize;
 642
 643         /* this replay - not send info to client */
 644         if (info->mti_spec.no_create) {
 645                 md_size = 0;
 646                 acl_size = 0;
 647         }
 648
 649         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 650                md_size, acl_size);
 651 /*
 652             &RMF_MDT_BODY,
 653             &RMF_MDT_MD,
 654             &RMF_ACL, or &RMF_LOGCOOKIES
 655 (optional)  &RMF_CAPA1,
 656 (optional)  &RMF_CAPA2,
 657 (optional)  something else
 658 */
 659
 660         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 661          * buffers before growing it */
 662         if (info->mti_big_lmm_used) {
 663                 /* big_lmm buffer may be used even without packing the result
 664                  * into reply, just for internal server needs */
 665                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 666                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 667                                                          RCL_SERVER);
 668
 669                 /* free big lmm if md_size is not needed */
 670                 if (md_size == 0 || md_packed == 0) {
 671                         info->mti_big_lmm_used = 0;
 672                 } else {
 673                         /* buffer must be allocated separately */
 674                         LASSERT(info->mti_attr.ma_lmm !=
 675                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 676                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 677                 }
 678         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 679                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 680         }
 681
 682         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 683                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 684         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 685                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 686                                    acl_size, RCL_SERVER);
 687
 688         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 689             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 690                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 691
 692         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 693             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 694                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 695
 696         /*
 697          * Some more field should be shrinked if needed.
 698          * This should be done by those who added fields to reply message.
 699          */
 700
 701         /* Grow MD buffer if needed finally */
 702         if (info->mti_big_lmm_used) {
 703                 void *lmm;
 704
 705                 LASSERT(md_size > md_packed);
 706                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 707                        md_size - md_packed);
 708                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 709                 if (rc) {
 710                         /* we can't answer with proper LOV EA, drop flags,
 711                          * the rc is also returned so this request is
 712                          * considered as failed */
 713                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 714                         /* don't return transno along with error */
 715                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 716                 } else {
 717                         /* now we need to pack right LOV/LMV EA */
 718                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 719                         if (info->mti_attr.ma_valid & MA_LOV) {
 720                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 721                                                              RCL_SERVER) ==
 722                                                 info->mti_attr.ma_lmm_size);
 723                                 memcpy(lmm, info->mti_attr.ma_lmm,
 724                                        info->mti_attr.ma_lmm_size);
 725                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 726                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 727                                                              RCL_SERVER) ==
 728                                                 info->mti_attr.ma_lmv_size);
 729                                 memcpy(lmm, info->mti_attr.ma_lmv,
 730                                        info->mti_attr.ma_lmv_size);
 731                         }
 732                 }
 733                 /* update mdt_max_mdsize so clients will be aware about that */
 734                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 735                         info->mti_mdt->mdt_max_mdsize =
 736                                                     info->mti_attr.ma_lmm_size;
 737                 info->mti_big_lmm_used = 0;
 738         }
 739         RETURN(rc);
 740 }
 741
 742
 743 /* if object is dying, pack the lov/llog data,
 744  * parameter info->mti_attr should be valid at this point! */
 745 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 746                            const struct md_attr *ma)
 747 {
 748         struct mdt_body       *repbody;
 749         const struct lu_attr *la = &ma->ma_attr;
 750         ENTRY;
 751
 752         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 753         LASSERT(repbody != NULL);
 754
 755         if (ma->ma_valid & MA_INODE)
 756                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 757
 758         if (ma->ma_valid & MA_LOV) {
 759                 CERROR("No need in LOV EA upon unlink\n");
 760                 dump_stack();
 761         }
 762         repbody->mbo_eadatasize = 0;
 763
 764         RETURN(0);
 765 }
 766
 767 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 768                                   struct md_attr *ma)
 769 {
 770         __u64 out;
 771
 772         out = 0;
 773         if (in & MDS_ATTR_MODE)
 774                 out |= LA_MODE;
 775         if (in & MDS_ATTR_UID)
 776                 out |= LA_UID;
 777         if (in & MDS_ATTR_GID)
 778                 out |= LA_GID;
 779         if (in & MDS_ATTR_SIZE)
 780                 out |= LA_SIZE;
 781         if (in & MDS_ATTR_BLOCKS)
 782                 out |= LA_BLOCKS;
 783         if (in & MDS_ATTR_ATIME_SET)
 784                 out |= LA_ATIME;
 785         if (in & MDS_ATTR_CTIME_SET)
 786                 out |= LA_CTIME;
 787         if (in & MDS_ATTR_MTIME_SET)
 788                 out |= LA_MTIME;
 789         if (in & MDS_ATTR_ATTR_FLAG)
 790                 out |= LA_FLAGS;
 791         if (in & MDS_ATTR_KILL_SUID)
 792                 out |= LA_KILL_SUID;
 793         if (in & MDS_ATTR_KILL_SGID)
 794                 out |= LA_KILL_SGID;
 795
 796         if (in & MDS_ATTR_FROM_OPEN)
 797                 rr->rr_flags |= MRF_OPEN_TRUNC;
 798         if (in & MDS_OPEN_OWNEROVERRIDE)
 799                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 800         if (in & MDS_ATTR_FORCE)
 801                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 802
 803         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 804                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 805                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 806                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 807                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 808                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 809         if (in != 0)
 810                 CERROR("Unknown attr bits: "LPX64"\n", in);
 811         return out;
 812 }
 813
 814 /* unpacking */
 815
 816 int mdt_name_unpack(struct req_capsule *pill,
 817                     const struct req_msg_field *field,
 818                     struct lu_name *ln,
 819                     enum mdt_name_flags flags)
 820 {
 821         ln->ln_name = req_capsule_client_get(pill, field);
 822         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 823
 824         if (!lu_name_is_valid(ln)) {
 825                 ln->ln_name = NULL;
 826                 ln->ln_namelen = 0;
 827
 828                 return -EPROTO;
 829         }
 830
 831         if ((flags & MNF_FIX_ANON) &&
 832             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 833                 /* Newer (3.x) kernels use a name of "/" for the
 834                  * "anonymous" disconnected dentries from NFS
 835                  * filehandle conversion. See d_obtain_alias(). */
 836                 ln->ln_name = NULL;
 837                 ln->ln_namelen = 0;
 838         }
 839
 840         return 0;
 841 }
 842
 843 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 844 {
 845         struct lu_ucred         *uc = mdt_ucred(info);
 846         struct md_attr          *ma = &info->mti_attr;
 847         struct lu_attr          *la = &ma->ma_attr;
 848         struct req_capsule      *pill = info->mti_pill;
 849         struct mdt_reint_record *rr = &info->mti_rr;
 850         struct mdt_rec_setattr  *rec;
 851         struct lu_nodemap       *nodemap;
 852         ENTRY;
 853
 854         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 855         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 856         if (rec == NULL)
 857                 RETURN(-EFAULT);
 858
 859         /* This prior initialization is needed for old_init_ucred_reint() */
 860         uc->uc_fsuid = rec->sa_fsuid;
 861         uc->uc_fsgid = rec->sa_fsgid;
 862         uc->uc_cap   = rec->sa_cap;
 863         uc->uc_suppgids[0] = rec->sa_suppgid;
 864         uc->uc_suppgids[1] = -1;
 865
 866         rr->rr_fid1 = &rec->sa_fid;
 867         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 868         la->la_mode  = rec->sa_mode;
 869         la->la_flags = rec->sa_attr_flags;
 870
 871         nodemap = nodemap_get_from_exp(info->mti_exp);
 872         if (IS_ERR(nodemap))
 873                 RETURN(PTR_ERR(nodemap));
 874
 875         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 876                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 877         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 878                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 879         nodemap_putref(nodemap);
 880
 881         la->la_size  = rec->sa_size;
 882         la->la_blocks = rec->sa_blocks;
 883         la->la_ctime = rec->sa_ctime;
 884         la->la_atime = rec->sa_atime;
 885         la->la_mtime = rec->sa_mtime;
 886         ma->ma_valid = MA_INODE;
 887
 888         if (rec->sa_bias & MDS_DATA_MODIFIED)
 889                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 890         else
 891                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 892
 893         if (rec->sa_bias & MDS_HSM_RELEASE)
 894                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 895         else
 896                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 897
 898         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 899                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 900         else
 901                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 902
 903         RETURN(0);
 904 }
 905
 906 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 907 {
 908         struct req_capsule *pill = info->mti_pill;
 909         struct mdt_ioepoch *ioepoch;
 910         ENTRY;
 911
 912         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 913                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 914         else
 915                 ioepoch = NULL;
 916
 917         if (ioepoch == NULL)
 918                 RETURN(-EPROTO);
 919
 920         info->mti_close_handle = ioepoch->mio_handle;
 921
 922         RETURN(0);
 923 }
 924
 925 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 926         struct req_capsule      *pill = info->mti_pill;
 927
 928         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 929                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 930                 if (info->mti_dlm_req == NULL)
 931                         RETURN(-EFAULT);
 932         }
 933
 934         RETURN(0);
 935 }
 936
 937 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 938 {
 939         struct mdt_reint_record *rr = &info->mti_rr;
 940         struct md_attr          *ma = &info->mti_attr;
 941         struct req_capsule      *pill = info->mti_pill;
 942         int rc;
 943         ENTRY;
 944
 945         rc = mdt_setattr_unpack_rec(info);
 946         if (rc)
 947                 RETURN(rc);
 948
 949         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 950                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 951                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 952                                                         RCL_CLIENT);
 953                 if (rr->rr_eadatalen > 0) {
 954                         const struct lmv_user_md        *lum;
 955
 956                         lum = rr->rr_eadata;
 957                         /* Sigh ma_valid(from req) does not indicate whether
 958                          * it will set LOV/LMV EA, so we have to check magic */
 959                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 960                                 ma->ma_valid |= MA_LMV;
 961                                 ma->ma_lmv = (void *)rr->rr_eadata;
 962                                 ma->ma_lmv_size = rr->rr_eadatalen;
 963                         } else {
 964                                 ma->ma_valid |= MA_LOV;
 965                                 ma->ma_lmm = (void *)rr->rr_eadata;
 966                                 ma->ma_lmm_size = rr->rr_eadatalen;
 967                         }
 968                 }
 969         }
 970
 971         rc = mdt_dlmreq_unpack(info);
 972         RETURN(rc);
 973 }
 974
 975 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
 976 {
 977         struct md_attr          *ma = &info->mti_attr;
 978         struct req_capsule      *pill = info->mti_pill;
 979         ENTRY;
 980
 981         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 982                 RETURN(0);
 983
 984         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
 985
 986         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
 987             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
 988                 RETURN(-EFAULT);
 989
 990         RETURN(0);
 991 }
 992
 993 int mdt_close_unpack(struct mdt_thread_info *info)
 994 {
 995         int rc;
 996         ENTRY;
 997
 998         rc = mdt_close_handle_unpack(info);
 999         if (rc)
1000                 RETURN(rc);
1001
1002         rc = mdt_setattr_unpack_rec(info);
1003         if (rc)
1004                 RETURN(rc);
1005
1006         rc = mdt_intent_close_unpack(info);
1007         if (rc)
1008                 RETURN(rc);
1009
1010         RETURN(mdt_init_ucred_reint(info));
1011 }
1012
1013 static int mdt_create_unpack(struct mdt_thread_info *info)
1014 {
1015         struct lu_ucred         *uc  = mdt_ucred(info);
1016         struct mdt_rec_create   *rec;
1017         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1018         struct mdt_reint_record *rr = &info->mti_rr;
1019         struct req_capsule      *pill = info->mti_pill;
1020         struct md_op_spec       *sp = &info->mti_spec;
1021         int rc;
1022         ENTRY;
1023
1024         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1025         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1026         if (rec == NULL)
1027                 RETURN(-EFAULT);
1028
1029         /* This prior initialization is needed for old_init_ucred_reint() */
1030         uc->uc_fsuid = rec->cr_fsuid;
1031         uc->uc_fsgid = rec->cr_fsgid;
1032         uc->uc_cap   = rec->cr_cap;
1033         uc->uc_suppgids[0] = rec->cr_suppgid1;
1034         uc->uc_suppgids[1] = -1;
1035         uc->uc_umask = rec->cr_umask;
1036
1037         rr->rr_fid1 = &rec->cr_fid1;
1038         rr->rr_fid2 = &rec->cr_fid2;
1039         attr->la_mode = rec->cr_mode;
1040         attr->la_rdev  = rec->cr_rdev;
1041         attr->la_uid   = rec->cr_fsuid;
1042         attr->la_gid   = rec->cr_fsgid;
1043         attr->la_ctime = rec->cr_time;
1044         attr->la_mtime = rec->cr_time;
1045         attr->la_atime = rec->cr_time;
1046         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1047                          LA_CTIME | LA_MTIME | LA_ATIME;
1048         memset(&sp->u, 0, sizeof(sp->u));
1049         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1050
1051         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1052         if (rc < 0)
1053                 RETURN(rc);
1054
1055         if (S_ISLNK(attr->la_mode)) {
1056                 const char *tgt = NULL;
1057
1058                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1059                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1060                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1061                         sp->u.sp_symname = tgt;
1062                 }
1063                 if (tgt == NULL)
1064                         RETURN(-EFAULT);
1065         } else {
1066                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1067                 if (S_ISDIR(attr->la_mode) &&
1068                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1069                         sp->u.sp_ea.eadata =
1070                                 req_capsule_client_get(pill, &RMF_EADATA);
1071                         sp->u.sp_ea.eadatalen =
1072                                 req_capsule_get_size(pill, &RMF_EADATA,
1073                                                      RCL_CLIENT);
1074                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1075                 }
1076         }
1077
1078         rc = mdt_dlmreq_unpack(info);
1079         RETURN(rc);
1080 }
1081
1082 static int mdt_link_unpack(struct mdt_thread_info *info)
1083 {
1084         struct lu_ucred         *uc  = mdt_ucred(info);
1085         struct mdt_rec_link     *rec;
1086         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1087         struct mdt_reint_record *rr = &info->mti_rr;
1088         struct req_capsule      *pill = info->mti_pill;
1089         int rc;
1090         ENTRY;
1091
1092         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1093         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1094         if (rec == NULL)
1095                 RETURN(-EFAULT);
1096
1097         /* This prior initialization is needed for old_init_ucred_reint() */
1098         uc->uc_fsuid = rec->lk_fsuid;
1099         uc->uc_fsgid = rec->lk_fsgid;
1100         uc->uc_cap   = rec->lk_cap;
1101         uc->uc_suppgids[0] = rec->lk_suppgid1;
1102         uc->uc_suppgids[1] = rec->lk_suppgid2;
1103
1104         attr->la_uid = rec->lk_fsuid;
1105         attr->la_gid = rec->lk_fsgid;
1106         rr->rr_fid1 = &rec->lk_fid1;
1107         rr->rr_fid2 = &rec->lk_fid2;
1108         attr->la_ctime = rec->lk_time;
1109         attr->la_mtime = rec->lk_time;
1110         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1111
1112         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1113         if (rc < 0)
1114                 RETURN(rc);
1115
1116         rc = mdt_dlmreq_unpack(info);
1117
1118         RETURN(rc);
1119 }
1120
1121 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1122 {
1123         struct lu_ucred         *uc  = mdt_ucred(info);
1124         struct mdt_rec_unlink   *rec;
1125         struct md_attr          *ma = &info->mti_attr;
1126         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1127         struct mdt_reint_record *rr = &info->mti_rr;
1128         struct req_capsule      *pill = info->mti_pill;
1129         int rc;
1130         ENTRY;
1131
1132         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1133         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1134         if (rec == NULL)
1135                 RETURN(-EFAULT);
1136
1137         /* This prior initialization is needed for old_init_ucred_reint() */
1138         uc->uc_fsuid = rec->ul_fsuid;
1139         uc->uc_fsgid = rec->ul_fsgid;
1140         uc->uc_cap   = rec->ul_cap;
1141         uc->uc_suppgids[0] = rec->ul_suppgid1;
1142         uc->uc_suppgids[1] = -1;
1143
1144         attr->la_uid = rec->ul_fsuid;
1145         attr->la_gid = rec->ul_fsgid;
1146         rr->rr_fid1 = &rec->ul_fid1;
1147         rr->rr_fid2 = &rec->ul_fid2;
1148         attr->la_ctime = rec->ul_time;
1149         attr->la_mtime = rec->ul_time;
1150         attr->la_mode  = rec->ul_mode;
1151         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1152
1153         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1154         if (rc < 0)
1155                 RETURN(rc);
1156
1157         if (rec->ul_bias & MDS_VTX_BYPASS)
1158                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1159         else
1160                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1161
1162         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1163
1164         rc = mdt_dlmreq_unpack(info);
1165         RETURN(rc);
1166 }
1167
1168 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1169 {
1170         info->mti_spec.sp_rm_entry = 1;
1171         return mdt_unlink_unpack(info);
1172 }
1173
1174 static int mdt_rename_unpack(struct mdt_thread_info *info)
1175 {
1176         struct lu_ucred         *uc = mdt_ucred(info);
1177         struct mdt_rec_rename   *rec;
1178         struct md_attr          *ma = &info->mti_attr;
1179         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1180         struct mdt_reint_record *rr = &info->mti_rr;
1181         struct req_capsule      *pill = info->mti_pill;
1182         int rc;
1183         ENTRY;
1184
1185         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1186         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1187         if (rec == NULL)
1188                 RETURN(-EFAULT);
1189
1190         /* This prior initialization is needed for old_init_ucred_reint() */
1191         uc->uc_fsuid = rec->rn_fsuid;
1192         uc->uc_fsgid = rec->rn_fsgid;
1193         uc->uc_cap   = rec->rn_cap;
1194         uc->uc_suppgids[0] = rec->rn_suppgid1;
1195         uc->uc_suppgids[1] = rec->rn_suppgid2;
1196
1197         attr->la_uid = rec->rn_fsuid;
1198         attr->la_gid = rec->rn_fsgid;
1199         rr->rr_fid1 = &rec->rn_fid1;
1200         rr->rr_fid2 = &rec->rn_fid2;
1201         attr->la_ctime = rec->rn_time;
1202         attr->la_mtime = rec->rn_time;
1203         /* rename_tgt contains the mode already */
1204         attr->la_mode = rec->rn_mode;
1205         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1206
1207         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1208         if (rc < 0)
1209                 RETURN(rc);
1210
1211         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1212         if (rc < 0)
1213                 RETURN(rc);
1214
1215         if (rec->rn_bias & MDS_VTX_BYPASS)
1216                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1217         else
1218                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1219
1220         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1221                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1222                 rc = mdt_close_handle_unpack(info);
1223                 if (rc < 0)
1224                         RETURN(rc);
1225                 info->mti_spec.sp_migrate_close = 1;
1226         }
1227
1228         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1229
1230
1231         rc = mdt_dlmreq_unpack(info);
1232
1233         RETURN(rc);
1234 }
1235
1236 /*
1237  * please see comment above LOV_MAGIC_V1_DEF
1238  */
1239 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1240 {
1241         struct mdt_reint_record *rr = &info->mti_rr;
1242         struct lov_user_md_v1   *v1;
1243
1244         v1 = (void *)rr->rr_eadata;
1245         LASSERT(v1);
1246
1247         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1248                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1249                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1250                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1251                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1252                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1253                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1254                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1255                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1256                 }
1257         }
1258 }
1259
1260 static int mdt_open_unpack(struct mdt_thread_info *info)
1261 {
1262         struct lu_ucred         *uc = mdt_ucred(info);
1263         struct mdt_rec_create   *rec;
1264         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1265         struct req_capsule      *pill = info->mti_pill;
1266         struct mdt_reint_record *rr   = &info->mti_rr;
1267         struct ptlrpc_request   *req  = mdt_info_req(info);
1268         struct md_op_spec       *sp   = &info->mti_spec;
1269         ENTRY;
1270
1271         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1272         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1273         if (rec == NULL)
1274                 RETURN(-EFAULT);
1275
1276         /* This prior initialization is needed for old_init_ucred_reint() */
1277         uc->uc_fsuid = rec->cr_fsuid;
1278         uc->uc_fsgid = rec->cr_fsgid;
1279         uc->uc_cap   = rec->cr_cap;
1280         uc->uc_suppgids[0] = rec->cr_suppgid1;
1281         uc->uc_suppgids[1] = rec->cr_suppgid2;
1282         uc->uc_umask = rec->cr_umask;
1283
1284         rr->rr_fid1   = &rec->cr_fid1;
1285         rr->rr_fid2   = &rec->cr_fid2;
1286         rr->rr_handle = &rec->cr_old_handle;
1287         attr->la_mode = rec->cr_mode;
1288         attr->la_rdev  = rec->cr_rdev;
1289         attr->la_uid   = rec->cr_fsuid;
1290         attr->la_gid   = rec->cr_fsgid;
1291         attr->la_ctime = rec->cr_time;
1292         attr->la_mtime = rec->cr_time;
1293         attr->la_atime = rec->cr_time;
1294         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1295                          LA_CTIME | LA_MTIME | LA_ATIME;
1296         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1297         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1298         /* Do not trigger ASSERTION if client miss to set such flags. */
1299         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1300                 RETURN(-EPROTO);
1301
1302         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1303
1304         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1305
1306         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1307                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1308                                                         RCL_CLIENT);
1309                 if (rr->rr_eadatalen > 0) {
1310                         rr->rr_eadata = req_capsule_client_get(pill,
1311                                                                &RMF_EADATA);
1312                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1313                         sp->u.sp_ea.eadata = rr->rr_eadata;
1314                         sp->no_create = !!req_is_replay(req);
1315                         mdt_fix_lov_magic(info);
1316                 }
1317
1318                 /*
1319                  * Client default md_size may be 0 right after client start,
1320                  * until all osc are connected, set here just some reasonable
1321                  * value to prevent misbehavior.
1322                  */
1323                 if (rr->rr_eadatalen == 0 &&
1324                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1325                         rr->rr_eadatalen = MIN_MD_SIZE;
1326         }
1327
1328         RETURN(0);
1329 }
1330
1331 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1332 {
1333         struct mdt_reint_record *rr     = &info->mti_rr;
1334         struct lu_ucred         *uc     = mdt_ucred(info);
1335         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1336         struct req_capsule      *pill   = info->mti_pill;
1337         struct mdt_rec_setxattr *rec;
1338         int                      rc;
1339         ENTRY;
1340
1341
1342         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1343                          sizeof(struct mdt_rec_reint));
1344
1345         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1346         if (rec == NULL)
1347                 RETURN(-EFAULT);
1348
1349         /* This prior initialization is needed for old_init_ucred_reint() */
1350         uc->uc_fsuid  = rec->sx_fsuid;
1351         uc->uc_fsgid  = rec->sx_fsgid;
1352         uc->uc_cap    = rec->sx_cap;
1353         uc->uc_suppgids[0] = rec->sx_suppgid1;
1354         uc->uc_suppgids[1] = -1;
1355
1356         rr->rr_opcode = rec->sx_opcode;
1357         rr->rr_fid1   = &rec->sx_fid;
1358         attr->la_valid = rec->sx_valid;
1359         attr->la_ctime = rec->sx_time;
1360         attr->la_size = rec->sx_size;
1361         attr->la_flags = rec->sx_flags;
1362
1363         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1364         if (rc < 0)
1365                 RETURN(rc);
1366
1367         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1368                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1369                                                         RCL_CLIENT);
1370                 if (rr->rr_eadatalen > 0) {
1371                         rr->rr_eadata = req_capsule_client_get(pill,
1372                                                                &RMF_EADATA);
1373                         if (rr->rr_eadata == NULL)
1374                                 RETURN(-EFAULT);
1375                 } else {
1376                         rr->rr_eadata = NULL;
1377                 }
1378         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1379                 CDEBUG(D_INFO, "no xattr data supplied\n");
1380                 RETURN(-EFAULT);
1381         }
1382
1383         if (mdt_dlmreq_unpack(info) < 0)
1384                 RETURN(-EPROTO);
1385
1386         RETURN(0);
1387 }
1388
1389
1390 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1391
1392 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1393         [REINT_SETATTR]  = mdt_setattr_unpack,
1394         [REINT_CREATE]   = mdt_create_unpack,
1395         [REINT_LINK]     = mdt_link_unpack,
1396         [REINT_UNLINK]   = mdt_unlink_unpack,
1397         [REINT_RENAME]   = mdt_rename_unpack,
1398         [REINT_OPEN]     = mdt_open_unpack,
1399         [REINT_SETXATTR] = mdt_setxattr_unpack,
1400         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1401         [REINT_MIGRATE]  = mdt_rename_unpack,
1402 };
1403
1404 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1405 {
1406         int rc;
1407         ENTRY;
1408
1409         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1410         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1411                 info->mti_rr.rr_opcode = op;
1412                 rc = mdt_reint_unpackers[op](info);
1413         } else {
1414                 CERROR("Unexpected opcode %d\n", op);
1415                 rc = -EFAULT;
1416         }
1417         RETURN(rc);
1418 }