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         struct md_attr        *ma  = &info->mti_attr;
 551
 552         LASSERT(uc != NULL);
 553         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 554                 return 0;
 555
 556         /* LU-5564: for normal close request, skip permission check */
 557         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 558             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 559                 uc->uc_cap |= CFS_CAP_FS_MASK;
 560
 561         mdt_exit_ucred(info);
 562
 563         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 564                 return old_init_ucred_reint(info);
 565         else
 566                 return new_init_ucred(info, REC_INIT, NULL, true);
 567 }
 568
 569 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 570 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 571 {
 572         const struct lov_ost_data_v1    *lod;
 573         int                              i;
 574         __u16                            count;
 575
 576         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 577                 return;
 578
 579         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 580
 581         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 582                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 583                le32_to_cpu(lmm->lmm_pattern));
 584         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 585                le32_to_cpu(lmm->lmm_stripe_size), count);
 586
 587         /* If it's a directory or a released file, then there are
 588          * no actual objects to print, so bail out. */
 589         if (valid & OBD_MD_FLDIREA ||
 590             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 591                 return;
 592
 593         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 594         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 595                 struct ost_id oi;
 596
 597                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 598                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 599                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 600         }
 601 }
 602
 603 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 604 {
 605         const struct lmv_mds_md_v1 *lmm1;
 606         int                        i;
 607
 608         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 609                 return;
 610
 611         lmm1 = &lmv->lmv_md_v1;
 612         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 613                le32_to_cpu(lmm1->lmv_magic),
 614                le32_to_cpu(lmm1->lmv_master_mdt_index),
 615                le32_to_cpu(lmm1->lmv_stripe_count));
 616
 617         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 618                 return;
 619
 620         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 621                 struct lu_fid fid;
 622
 623                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 624                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 625         }
 626 }
 627
 628 /* Shrink and/or grow reply buffers */
 629 int mdt_fix_reply(struct mdt_thread_info *info)
 630 {
 631         struct req_capsule *pill = info->mti_pill;
 632         struct mdt_body    *body;
 633         int                md_size, md_packed = 0;
 634         int                acl_size;
 635         int                rc = 0;
 636         ENTRY;
 637
 638         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 639         LASSERT(body != NULL);
 640
 641         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 642                                OBD_MD_LINKNAME))
 643                 md_size = body->mbo_eadatasize;
 644         else
 645                 md_size = 0;
 646
 647         acl_size = body->mbo_aclsize;
 648
 649         /* this replay - not send info to client */
 650         if (info->mti_spec.no_create) {
 651                 md_size = 0;
 652                 acl_size = 0;
 653         }
 654
 655         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 656                md_size, acl_size);
 657 /*
 658             &RMF_MDT_BODY,
 659             &RMF_MDT_MD,
 660             &RMF_ACL, or &RMF_LOGCOOKIES
 661 (optional)  &RMF_CAPA1,
 662 (optional)  &RMF_CAPA2,
 663 (optional)  something else
 664 */
 665
 666         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 667          * buffers before growing it */
 668         if (info->mti_big_lmm_used) {
 669                 /* big_lmm buffer may be used even without packing the result
 670                  * into reply, just for internal server needs */
 671                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 672                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 673                                                          RCL_SERVER);
 674
 675                 /* free big lmm if md_size is not needed */
 676                 if (md_size == 0 || md_packed == 0) {
 677                         info->mti_big_lmm_used = 0;
 678                 } else {
 679                         /* buffer must be allocated separately */
 680                         LASSERT(info->mti_attr.ma_lmm !=
 681                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 682                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 683                 }
 684         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 685                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 686         }
 687
 688         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 689                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 690         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 691                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 692                                    acl_size, RCL_SERVER);
 693
 694         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 695             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 696                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 697
 698         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 699             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 700                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 701
 702         /*
 703          * Some more field should be shrinked if needed.
 704          * This should be done by those who added fields to reply message.
 705          */
 706
 707         /* Grow MD buffer if needed finally */
 708         if (info->mti_big_lmm_used) {
 709                 void *lmm;
 710
 711                 LASSERT(md_size > md_packed);
 712                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 713                        md_size - md_packed);
 714                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 715                 if (rc) {
 716                         /* we can't answer with proper LOV EA, drop flags,
 717                          * the rc is also returned so this request is
 718                          * considered as failed */
 719                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 720                         /* don't return transno along with error */
 721                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 722                 } else {
 723                         /* now we need to pack right LOV/LMV EA */
 724                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 725                         if (info->mti_attr.ma_valid & MA_LOV) {
 726                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 727                                                              RCL_SERVER) ==
 728                                                 info->mti_attr.ma_lmm_size);
 729                                 memcpy(lmm, info->mti_attr.ma_lmm,
 730                                        info->mti_attr.ma_lmm_size);
 731                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 732                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 733                                                              RCL_SERVER) ==
 734                                                 info->mti_attr.ma_lmv_size);
 735                                 memcpy(lmm, info->mti_attr.ma_lmv,
 736                                        info->mti_attr.ma_lmv_size);
 737                         }
 738                 }
 739                 /* update mdt_max_mdsize so clients will be aware about that */
 740                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 741                         info->mti_mdt->mdt_max_mdsize =
 742                                                     info->mti_attr.ma_lmm_size;
 743                 info->mti_big_lmm_used = 0;
 744         }
 745         RETURN(rc);
 746 }
 747
 748
 749 /* if object is dying, pack the lov/llog data,
 750  * parameter info->mti_attr should be valid at this point! */
 751 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 752                            const struct md_attr *ma)
 753 {
 754         struct mdt_body       *repbody;
 755         const struct lu_attr *la = &ma->ma_attr;
 756         ENTRY;
 757
 758         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 759         LASSERT(repbody != NULL);
 760
 761         if (ma->ma_valid & MA_INODE)
 762                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 763
 764         if (ma->ma_valid & MA_LOV) {
 765                 CERROR("No need in LOV EA upon unlink\n");
 766                 dump_stack();
 767         }
 768         repbody->mbo_eadatasize = 0;
 769
 770         RETURN(0);
 771 }
 772
 773 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 774                                   struct md_attr *ma)
 775 {
 776         __u64 out;
 777
 778         out = 0;
 779         if (in & MDS_ATTR_MODE)
 780                 out |= LA_MODE;
 781         if (in & MDS_ATTR_UID)
 782                 out |= LA_UID;
 783         if (in & MDS_ATTR_GID)
 784                 out |= LA_GID;
 785         if (in & MDS_ATTR_SIZE)
 786                 out |= LA_SIZE;
 787         if (in & MDS_ATTR_BLOCKS)
 788                 out |= LA_BLOCKS;
 789         if (in & MDS_ATTR_ATIME_SET)
 790                 out |= LA_ATIME;
 791         if (in & MDS_ATTR_CTIME_SET)
 792                 out |= LA_CTIME;
 793         if (in & MDS_ATTR_MTIME_SET)
 794                 out |= LA_MTIME;
 795         if (in & MDS_ATTR_ATTR_FLAG)
 796                 out |= LA_FLAGS;
 797         if (in & MDS_ATTR_KILL_SUID)
 798                 out |= LA_KILL_SUID;
 799         if (in & MDS_ATTR_KILL_SGID)
 800                 out |= LA_KILL_SGID;
 801
 802         if (in & MDS_ATTR_FROM_OPEN)
 803                 rr->rr_flags |= MRF_OPEN_TRUNC;
 804         if (in & MDS_OPEN_OWNEROVERRIDE)
 805                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 806         if (in & MDS_ATTR_FORCE)
 807                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 808
 809         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 810                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 811                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 812                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 813                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 814                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 815         if (in != 0)
 816                 CERROR("Unknown attr bits: "LPX64"\n", in);
 817         return out;
 818 }
 819
 820 /* unpacking */
 821
 822 int mdt_name_unpack(struct req_capsule *pill,
 823                     const struct req_msg_field *field,
 824                     struct lu_name *ln,
 825                     enum mdt_name_flags flags)
 826 {
 827         ln->ln_name = req_capsule_client_get(pill, field);
 828         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 829
 830         if (!lu_name_is_valid(ln)) {
 831                 ln->ln_name = NULL;
 832                 ln->ln_namelen = 0;
 833
 834                 return -EPROTO;
 835         }
 836
 837         if ((flags & MNF_FIX_ANON) &&
 838             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 839                 /* Newer (3.x) kernels use a name of "/" for the
 840                  * "anonymous" disconnected dentries from NFS
 841                  * filehandle conversion. See d_obtain_alias(). */
 842                 ln->ln_name = NULL;
 843                 ln->ln_namelen = 0;
 844         }
 845
 846         return 0;
 847 }
 848
 849 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 850 {
 851         struct lu_ucred         *uc = mdt_ucred(info);
 852         struct md_attr          *ma = &info->mti_attr;
 853         struct lu_attr          *la = &ma->ma_attr;
 854         struct req_capsule      *pill = info->mti_pill;
 855         struct mdt_reint_record *rr = &info->mti_rr;
 856         struct mdt_rec_setattr  *rec;
 857         struct lu_nodemap       *nodemap;
 858         ENTRY;
 859
 860         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 861         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 862         if (rec == NULL)
 863                 RETURN(-EFAULT);
 864
 865         /* This prior initialization is needed for old_init_ucred_reint() */
 866         uc->uc_fsuid = rec->sa_fsuid;
 867         uc->uc_fsgid = rec->sa_fsgid;
 868         uc->uc_cap   = rec->sa_cap;
 869         uc->uc_suppgids[0] = rec->sa_suppgid;
 870         uc->uc_suppgids[1] = -1;
 871
 872         rr->rr_fid1 = &rec->sa_fid;
 873         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 874         la->la_mode  = rec->sa_mode;
 875         la->la_flags = rec->sa_attr_flags;
 876
 877         nodemap = nodemap_get_from_exp(info->mti_exp);
 878         if (IS_ERR(nodemap))
 879                 RETURN(PTR_ERR(nodemap));
 880
 881         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 882                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 883         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 884                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 885         nodemap_putref(nodemap);
 886
 887         la->la_size  = rec->sa_size;
 888         la->la_blocks = rec->sa_blocks;
 889         la->la_ctime = rec->sa_ctime;
 890         la->la_atime = rec->sa_atime;
 891         la->la_mtime = rec->sa_mtime;
 892         ma->ma_valid = MA_INODE;
 893
 894         if (rec->sa_bias & MDS_DATA_MODIFIED)
 895                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 896         else
 897                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 898
 899         if (rec->sa_bias & MDS_HSM_RELEASE)
 900                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 901         else
 902                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 903
 904         if (rec->sa_bias & MDS_CLOSE_LAYOUT_SWAP)
 905                 ma->ma_attr_flags |= MDS_CLOSE_LAYOUT_SWAP;
 906         else
 907                 ma->ma_attr_flags &= ~MDS_CLOSE_LAYOUT_SWAP;
 908
 909         RETURN(0);
 910 }
 911
 912 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 913 {
 914         struct req_capsule *pill = info->mti_pill;
 915         struct mdt_ioepoch *ioepoch;
 916         ENTRY;
 917
 918         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 919                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 920         else
 921                 ioepoch = NULL;
 922
 923         if (ioepoch == NULL)
 924                 RETURN(-EPROTO);
 925
 926         info->mti_close_handle = ioepoch->mio_handle;
 927
 928         RETURN(0);
 929 }
 930
 931 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 932         struct req_capsule      *pill = info->mti_pill;
 933
 934         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 935                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 936                 if (info->mti_dlm_req == NULL)
 937                         RETURN(-EFAULT);
 938         }
 939
 940         RETURN(0);
 941 }
 942
 943 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 944 {
 945         struct mdt_reint_record *rr = &info->mti_rr;
 946         struct md_attr          *ma = &info->mti_attr;
 947         struct req_capsule      *pill = info->mti_pill;
 948         int rc;
 949         ENTRY;
 950
 951         rc = mdt_setattr_unpack_rec(info);
 952         if (rc)
 953                 RETURN(rc);
 954
 955         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 956                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 957                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 958                                                         RCL_CLIENT);
 959                 if (rr->rr_eadatalen > 0) {
 960                         const struct lmv_user_md        *lum;
 961
 962                         lum = rr->rr_eadata;
 963                         /* Sigh ma_valid(from req) does not indicate whether
 964                          * it will set LOV/LMV EA, so we have to check magic */
 965                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
 966                                 ma->ma_valid |= MA_LMV;
 967                                 ma->ma_lmv = (void *)rr->rr_eadata;
 968                                 ma->ma_lmv_size = rr->rr_eadatalen;
 969                         } else {
 970                                 ma->ma_valid |= MA_LOV;
 971                                 ma->ma_lmm = (void *)rr->rr_eadata;
 972                                 ma->ma_lmm_size = rr->rr_eadatalen;
 973                         }
 974                 }
 975         }
 976
 977         rc = mdt_dlmreq_unpack(info);
 978         RETURN(rc);
 979 }
 980
 981 static int mdt_intent_close_unpack(struct mdt_thread_info *info)
 982 {
 983         struct md_attr          *ma = &info->mti_attr;
 984         struct req_capsule      *pill = info->mti_pill;
 985         ENTRY;
 986
 987         if (!(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 988                 RETURN(0);
 989
 990         req_capsule_extend(pill, &RQF_MDS_INTENT_CLOSE);
 991
 992         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
 993             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
 994                 RETURN(-EFAULT);
 995
 996         RETURN(0);
 997 }
 998
 999 int mdt_close_unpack(struct mdt_thread_info *info)
1000 {
1001         int rc;
1002         ENTRY;
1003
1004         rc = mdt_close_handle_unpack(info);
1005         if (rc)
1006                 RETURN(rc);
1007
1008         rc = mdt_setattr_unpack_rec(info);
1009         if (rc)
1010                 RETURN(rc);
1011
1012         rc = mdt_intent_close_unpack(info);
1013         if (rc)
1014                 RETURN(rc);
1015
1016         RETURN(mdt_init_ucred_reint(info));
1017 }
1018
1019 static int mdt_create_unpack(struct mdt_thread_info *info)
1020 {
1021         struct lu_ucred         *uc  = mdt_ucred(info);
1022         struct mdt_rec_create   *rec;
1023         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1024         struct mdt_reint_record *rr = &info->mti_rr;
1025         struct req_capsule      *pill = info->mti_pill;
1026         struct md_op_spec       *sp = &info->mti_spec;
1027         int rc;
1028         ENTRY;
1029
1030         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1031         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1032         if (rec == NULL)
1033                 RETURN(-EFAULT);
1034
1035         /* This prior initialization is needed for old_init_ucred_reint() */
1036         uc->uc_fsuid = rec->cr_fsuid;
1037         uc->uc_fsgid = rec->cr_fsgid;
1038         uc->uc_cap   = rec->cr_cap;
1039         uc->uc_suppgids[0] = rec->cr_suppgid1;
1040         uc->uc_suppgids[1] = -1;
1041         uc->uc_umask = rec->cr_umask;
1042
1043         rr->rr_fid1 = &rec->cr_fid1;
1044         rr->rr_fid2 = &rec->cr_fid2;
1045         attr->la_mode = rec->cr_mode;
1046         attr->la_rdev  = rec->cr_rdev;
1047         attr->la_uid   = rec->cr_fsuid;
1048         attr->la_gid   = rec->cr_fsgid;
1049         attr->la_ctime = rec->cr_time;
1050         attr->la_mtime = rec->cr_time;
1051         attr->la_atime = rec->cr_time;
1052         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1053                          LA_CTIME | LA_MTIME | LA_ATIME;
1054         memset(&sp->u, 0, sizeof(sp->u));
1055         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1056
1057         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1058         if (rc < 0)
1059                 RETURN(rc);
1060
1061         if (S_ISLNK(attr->la_mode)) {
1062                 const char *tgt = NULL;
1063
1064                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1065                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1066                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1067                         sp->u.sp_symname = tgt;
1068                 }
1069                 if (tgt == NULL)
1070                         RETURN(-EFAULT);
1071         } else {
1072                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1073                 if (S_ISDIR(attr->la_mode) &&
1074                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1075                         sp->u.sp_ea.eadata =
1076                                 req_capsule_client_get(pill, &RMF_EADATA);
1077                         sp->u.sp_ea.eadatalen =
1078                                 req_capsule_get_size(pill, &RMF_EADATA,
1079                                                      RCL_CLIENT);
1080                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1081                 }
1082         }
1083
1084         rc = mdt_dlmreq_unpack(info);
1085         RETURN(rc);
1086 }
1087
1088 static int mdt_link_unpack(struct mdt_thread_info *info)
1089 {
1090         struct lu_ucred         *uc  = mdt_ucred(info);
1091         struct mdt_rec_link     *rec;
1092         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1093         struct mdt_reint_record *rr = &info->mti_rr;
1094         struct req_capsule      *pill = info->mti_pill;
1095         int rc;
1096         ENTRY;
1097
1098         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1099         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1100         if (rec == NULL)
1101                 RETURN(-EFAULT);
1102
1103         /* This prior initialization is needed for old_init_ucred_reint() */
1104         uc->uc_fsuid = rec->lk_fsuid;
1105         uc->uc_fsgid = rec->lk_fsgid;
1106         uc->uc_cap   = rec->lk_cap;
1107         uc->uc_suppgids[0] = rec->lk_suppgid1;
1108         uc->uc_suppgids[1] = rec->lk_suppgid2;
1109
1110         attr->la_uid = rec->lk_fsuid;
1111         attr->la_gid = rec->lk_fsgid;
1112         rr->rr_fid1 = &rec->lk_fid1;
1113         rr->rr_fid2 = &rec->lk_fid2;
1114         attr->la_ctime = rec->lk_time;
1115         attr->la_mtime = rec->lk_time;
1116         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1117
1118         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1119         if (rc < 0)
1120                 RETURN(rc);
1121
1122         rc = mdt_dlmreq_unpack(info);
1123
1124         RETURN(rc);
1125 }
1126
1127 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1128 {
1129         struct lu_ucred         *uc  = mdt_ucred(info);
1130         struct mdt_rec_unlink   *rec;
1131         struct md_attr          *ma = &info->mti_attr;
1132         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1133         struct mdt_reint_record *rr = &info->mti_rr;
1134         struct req_capsule      *pill = info->mti_pill;
1135         int rc;
1136         ENTRY;
1137
1138         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1139         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1140         if (rec == NULL)
1141                 RETURN(-EFAULT);
1142
1143         /* This prior initialization is needed for old_init_ucred_reint() */
1144         uc->uc_fsuid = rec->ul_fsuid;
1145         uc->uc_fsgid = rec->ul_fsgid;
1146         uc->uc_cap   = rec->ul_cap;
1147         uc->uc_suppgids[0] = rec->ul_suppgid1;
1148         uc->uc_suppgids[1] = -1;
1149
1150         attr->la_uid = rec->ul_fsuid;
1151         attr->la_gid = rec->ul_fsgid;
1152         rr->rr_fid1 = &rec->ul_fid1;
1153         rr->rr_fid2 = &rec->ul_fid2;
1154         attr->la_ctime = rec->ul_time;
1155         attr->la_mtime = rec->ul_time;
1156         attr->la_mode  = rec->ul_mode;
1157         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1158
1159         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1160         if (rc < 0)
1161                 RETURN(rc);
1162
1163         if (rec->ul_bias & MDS_VTX_BYPASS)
1164                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1165         else
1166                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1167
1168         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1169
1170         rc = mdt_dlmreq_unpack(info);
1171         RETURN(rc);
1172 }
1173
1174 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1175 {
1176         info->mti_spec.sp_rm_entry = 1;
1177         return mdt_unlink_unpack(info);
1178 }
1179
1180 static int mdt_rename_unpack(struct mdt_thread_info *info)
1181 {
1182         struct lu_ucred         *uc = mdt_ucred(info);
1183         struct mdt_rec_rename   *rec;
1184         struct md_attr          *ma = &info->mti_attr;
1185         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1186         struct mdt_reint_record *rr = &info->mti_rr;
1187         struct req_capsule      *pill = info->mti_pill;
1188         int rc;
1189         ENTRY;
1190
1191         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1192         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1193         if (rec == NULL)
1194                 RETURN(-EFAULT);
1195
1196         /* This prior initialization is needed for old_init_ucred_reint() */
1197         uc->uc_fsuid = rec->rn_fsuid;
1198         uc->uc_fsgid = rec->rn_fsgid;
1199         uc->uc_cap   = rec->rn_cap;
1200         uc->uc_suppgids[0] = rec->rn_suppgid1;
1201         uc->uc_suppgids[1] = rec->rn_suppgid2;
1202
1203         attr->la_uid = rec->rn_fsuid;
1204         attr->la_gid = rec->rn_fsgid;
1205         rr->rr_fid1 = &rec->rn_fid1;
1206         rr->rr_fid2 = &rec->rn_fid2;
1207         attr->la_ctime = rec->rn_time;
1208         attr->la_mtime = rec->rn_time;
1209         /* rename_tgt contains the mode already */
1210         attr->la_mode = rec->rn_mode;
1211         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1212
1213         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1214         if (rc < 0)
1215                 RETURN(rc);
1216
1217         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1218         if (rc < 0)
1219                 RETURN(rc);
1220
1221         if (rec->rn_bias & MDS_VTX_BYPASS)
1222                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1223         else
1224                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1225
1226         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1227                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1228                 rc = mdt_close_handle_unpack(info);
1229                 if (rc < 0)
1230                         RETURN(rc);
1231                 info->mti_spec.sp_migrate_close = 1;
1232         }
1233
1234         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1235
1236
1237         rc = mdt_dlmreq_unpack(info);
1238
1239         RETURN(rc);
1240 }
1241
1242 /*
1243  * please see comment above LOV_MAGIC_V1_DEF
1244  */
1245 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1246 {
1247         struct mdt_reint_record *rr = &info->mti_rr;
1248         struct lov_user_md_v1   *v1;
1249
1250         v1 = (void *)rr->rr_eadata;
1251         LASSERT(v1);
1252
1253         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1254                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1255                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1256                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1257                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1258                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1259                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1260                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1261                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1262                 }
1263         }
1264 }
1265
1266 static int mdt_open_unpack(struct mdt_thread_info *info)
1267 {
1268         struct lu_ucred         *uc = mdt_ucred(info);
1269         struct mdt_rec_create   *rec;
1270         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1271         struct req_capsule      *pill = info->mti_pill;
1272         struct mdt_reint_record *rr   = &info->mti_rr;
1273         struct ptlrpc_request   *req  = mdt_info_req(info);
1274         struct md_op_spec       *sp   = &info->mti_spec;
1275         ENTRY;
1276
1277         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1278         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1279         if (rec == NULL)
1280                 RETURN(-EFAULT);
1281
1282         /* This prior initialization is needed for old_init_ucred_reint() */
1283         uc->uc_fsuid = rec->cr_fsuid;
1284         uc->uc_fsgid = rec->cr_fsgid;
1285         uc->uc_cap   = rec->cr_cap;
1286         uc->uc_suppgids[0] = rec->cr_suppgid1;
1287         uc->uc_suppgids[1] = rec->cr_suppgid2;
1288         uc->uc_umask = rec->cr_umask;
1289
1290         rr->rr_fid1   = &rec->cr_fid1;
1291         rr->rr_fid2   = &rec->cr_fid2;
1292         rr->rr_handle = &rec->cr_old_handle;
1293         attr->la_mode = rec->cr_mode;
1294         attr->la_rdev  = rec->cr_rdev;
1295         attr->la_uid   = rec->cr_fsuid;
1296         attr->la_gid   = rec->cr_fsgid;
1297         attr->la_ctime = rec->cr_time;
1298         attr->la_mtime = rec->cr_time;
1299         attr->la_atime = rec->cr_time;
1300         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1301                          LA_CTIME | LA_MTIME | LA_ATIME;
1302         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1303         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1304         /* Do not trigger ASSERTION if client miss to set such flags. */
1305         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1306                 RETURN(-EPROTO);
1307
1308         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1309
1310         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1311
1312         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1313                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1314                                                         RCL_CLIENT);
1315                 if (rr->rr_eadatalen > 0) {
1316                         rr->rr_eadata = req_capsule_client_get(pill,
1317                                                                &RMF_EADATA);
1318                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1319                         sp->u.sp_ea.eadata = rr->rr_eadata;
1320                         sp->no_create = !!req_is_replay(req);
1321                         mdt_fix_lov_magic(info);
1322                 }
1323
1324                 /*
1325                  * Client default md_size may be 0 right after client start,
1326                  * until all osc are connected, set here just some reasonable
1327                  * value to prevent misbehavior.
1328                  */
1329                 if (rr->rr_eadatalen == 0 &&
1330                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1331                         rr->rr_eadatalen = MIN_MD_SIZE;
1332         }
1333
1334         RETURN(0);
1335 }
1336
1337 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1338 {
1339         struct mdt_reint_record *rr     = &info->mti_rr;
1340         struct lu_ucred         *uc     = mdt_ucred(info);
1341         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1342         struct req_capsule      *pill   = info->mti_pill;
1343         struct mdt_rec_setxattr *rec;
1344         int                      rc;
1345         ENTRY;
1346
1347
1348         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1349                          sizeof(struct mdt_rec_reint));
1350
1351         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1352         if (rec == NULL)
1353                 RETURN(-EFAULT);
1354
1355         /* This prior initialization is needed for old_init_ucred_reint() */
1356         uc->uc_fsuid  = rec->sx_fsuid;
1357         uc->uc_fsgid  = rec->sx_fsgid;
1358         uc->uc_cap    = rec->sx_cap;
1359         uc->uc_suppgids[0] = rec->sx_suppgid1;
1360         uc->uc_suppgids[1] = -1;
1361
1362         rr->rr_opcode = rec->sx_opcode;
1363         rr->rr_fid1   = &rec->sx_fid;
1364         attr->la_valid = rec->sx_valid;
1365         attr->la_ctime = rec->sx_time;
1366         attr->la_size = rec->sx_size;
1367         attr->la_flags = rec->sx_flags;
1368
1369         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1370         if (rc < 0)
1371                 RETURN(rc);
1372
1373         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1374                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1375                                                         RCL_CLIENT);
1376                 if (rr->rr_eadatalen > 0) {
1377                         rr->rr_eadata = req_capsule_client_get(pill,
1378                                                                &RMF_EADATA);
1379                         if (rr->rr_eadata == NULL)
1380                                 RETURN(-EFAULT);
1381                 } else {
1382                         rr->rr_eadata = NULL;
1383                 }
1384         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1385                 CDEBUG(D_INFO, "no xattr data supplied\n");
1386                 RETURN(-EFAULT);
1387         }
1388
1389         if (mdt_dlmreq_unpack(info) < 0)
1390                 RETURN(-EPROTO);
1391
1392         RETURN(0);
1393 }
1394
1395
1396 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1397
1398 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1399         [REINT_SETATTR]  = mdt_setattr_unpack,
1400         [REINT_CREATE]   = mdt_create_unpack,
1401         [REINT_LINK]     = mdt_link_unpack,
1402         [REINT_UNLINK]   = mdt_unlink_unpack,
1403         [REINT_RENAME]   = mdt_rename_unpack,
1404         [REINT_OPEN]     = mdt_open_unpack,
1405         [REINT_SETXATTR] = mdt_setxattr_unpack,
1406         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1407         [REINT_MIGRATE]  = mdt_rename_unpack,
1408 };
1409
1410 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1411 {
1412         int rc;
1413         ENTRY;
1414
1415         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1416         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1417                 info->mti_rr.rr_opcode = op;
1418                 rc = mdt_reint_unpackers[op](info);
1419         } else {
1420                 CERROR("Unexpected opcode %d\n", op);
1421                 rc = -EFAULT;
1422         }
1423         RETURN(rc);
1424 }