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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_lib.c
  33  *
  34  * Lustre Metadata Target (mdt) request unpacking helper.
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Fan Yong <fanyong@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include "mdt_internal.h"
  52 #include <uapi/linux/lnet/nidstr.h>
  53 #include <lustre_nodemap.h>
  54
  55 typedef enum ucred_init_type {
  56         NONE_INIT       = 0,
  57         BODY_INIT       = 1,
  58         REC_INIT        = 2
  59 } ucred_init_type_t;
  60
  61 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
  62 {
  63         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
  64 }
  65
  66 void mdt_exit_ucred(struct mdt_thread_info *info)
  67 {
  68         struct lu_ucred   *uc  = mdt_ucred(info);
  69         struct mdt_device *mdt = info->mti_mdt;
  70
  71         LASSERT(uc != NULL);
  72         if (uc->uc_valid != UCRED_INIT) {
  73                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  74                 if (uc->uc_ginfo) {
  75                         put_group_info(uc->uc_ginfo);
  76                         uc->uc_ginfo = NULL;
  77                 }
  78                 if (uc->uc_identity) {
  79                         mdt_identity_put(mdt->mdt_identity_cache,
  80                                          uc->uc_identity);
  81                         uc->uc_identity = NULL;
  82                 }
  83                 uc->uc_valid = UCRED_INIT;
  84         }
  85 }
  86
  87 static int match_nosquash_list(struct rw_semaphore *sem,
  88                                struct list_head *nidlist,
  89                                lnet_nid_t peernid)
  90 {
  91         int rc;
  92         ENTRY;
  93         down_read(sem);
  94         rc = cfs_match_nid(peernid, nidlist);
  95         up_read(sem);
  96         RETURN(rc);
  97 }
  98
  99 /* root_squash for inter-MDS operations */
 100 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 101 {
 102         struct lu_ucred *ucred = mdt_ucred(info);
 103         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 104         ENTRY;
 105
 106         LASSERT(ucred != NULL);
 107         if (!squash->rsi_uid || ucred->uc_fsuid)
 108                 RETURN(0);
 109
 110         if (match_nosquash_list(&squash->rsi_sem,
 111                                 &squash->rsi_nosquash_nids,
 112                                 peernid)) {
 113                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 114                        libcfs_nid2str(peernid));
 115                 RETURN(0);
 116         }
 117
 118         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 119                libcfs_nid2str(peernid),
 120                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 121                squash->rsi_uid, squash->rsi_gid, 0);
 122
 123         ucred->uc_fsuid = squash->rsi_uid;
 124         ucred->uc_fsgid = squash->rsi_gid;
 125         ucred->uc_cap = 0;
 126         ucred->uc_suppgids[0] = -1;
 127         ucred->uc_suppgids[1] = -1;
 128
 129         RETURN(0);
 130 }
 131
 132 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 133 {
 134         struct ptlrpc_request   *req = mdt_info_req(info);
 135         const char              *jobid = mdt_req_get_jobid(req);
 136
 137         /* set jobid if specified. */
 138         if (jobid)
 139                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 140         else
 141                 uc->uc_jobid[0] = '\0';
 142 }
 143
 144 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 145                           void *buf, bool drop_fs_cap)
 146 {
 147         struct ptlrpc_request   *req = mdt_info_req(info);
 148         struct mdt_device       *mdt = info->mti_mdt;
 149         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 150         struct lu_ucred         *ucred = mdt_ucred(info);
 151         lnet_nid_t               peernid = req->rq_peer.nid;
 152         __u32                    perm = 0;
 153         int                      setuid;
 154         int                      setgid;
 155         int                      rc = 0;
 156
 157         ENTRY;
 158
 159         LASSERT(req->rq_auth_gss);
 160         LASSERT(!req->rq_auth_usr_mdt);
 161         LASSERT(req->rq_user_desc);
 162         LASSERT(ucred != NULL);
 163
 164         ucred->uc_valid = UCRED_INVALID;
 165
 166         ucred->uc_o_uid   = pud->pud_uid;
 167         ucred->uc_o_gid   = pud->pud_gid;
 168         ucred->uc_o_fsuid = pud->pud_fsuid;
 169         ucred->uc_o_fsgid = pud->pud_fsgid;
 170
 171         if (type == BODY_INIT) {
 172                 struct mdt_body *body = (struct mdt_body *)buf;
 173
 174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 175                 ucred->uc_suppgids[1] = -1;
 176         }
 177
 178         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 179             req->rq_auth_uid != pud->pud_uid) {
 180                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 181                        "while client claims %u:%u/%u:%u\n",
 182                        libcfs_nid2str(peernid), req->rq_auth_uid,
 183                        pud->pud_uid, pud->pud_gid,
 184                        pud->pud_fsuid, pud->pud_fsgid);
 185                 RETURN(-EACCES);
 186         }
 187
 188         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 189                 ucred->uc_identity = NULL;
 190                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 191         } else {
 192                 struct md_identity *identity;
 193
 194                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 195                                             pud->pud_uid);
 196                 if (IS_ERR(identity)) {
 197                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 198                                 ucred->uc_identity = NULL;
 199                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 200                                        CFS_SETGRP_PERM;
 201                         } else {
 202                                 CDEBUG(D_SEC,
 203                                        "Deny access without identity: uid %u\n",
 204                                        pud->pud_uid);
 205                                 RETURN(-EACCES);
 206                         }
 207                 } else {
 208                         ucred->uc_identity = identity;
 209                         perm = mdt_identity_get_perm(ucred->uc_identity,
 210                                                      peernid);
 211                 }
 212         }
 213
 214         /* find out the setuid/setgid attempt */
 215         setuid = (pud->pud_uid != pud->pud_fsuid);
 216         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 217                   (ucred->uc_identity &&
 218                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 219
 220         /* check permission of setuid */
 221         if (setuid && !(perm & CFS_SETUID_PERM)) {
 222                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 223                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 224                 GOTO(out, rc = -EACCES);
 225         }
 226
 227         /* check permission of setgid */
 228         if (setgid && !(perm & CFS_SETGID_PERM)) {
 229                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 230                        "from %s\n", pud->pud_uid, pud->pud_gid,
 231                        pud->pud_fsuid, pud->pud_fsgid,
 232                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 233                 GOTO(out, rc = -EACCES);
 234         }
 235
 236         if (perm & CFS_SETGRP_PERM) {
 237                 if (pud->pud_ngroups) {
 238                         /* setgroups for local client */
 239                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 240                         if (!ucred->uc_ginfo) {
 241                                 CERROR("failed to alloc %d groups\n",
 242                                        pud->pud_ngroups);
 243                                 GOTO(out, rc = -ENOMEM);
 244                         }
 245
 246                         lustre_groups_from_list(ucred->uc_ginfo,
 247                                                 pud->pud_groups);
 248                         lustre_groups_sort(ucred->uc_ginfo);
 249                 } else {
 250                         ucred->uc_ginfo = NULL;
 251                 }
 252         } else {
 253                 ucred->uc_suppgids[0] = -1;
 254                 ucred->uc_suppgids[1] = -1;
 255                 ucred->uc_ginfo = NULL;
 256         }
 257
 258         ucred->uc_uid   = pud->pud_uid;
 259         ucred->uc_gid   = pud->pud_gid;
 260         ucred->uc_fsuid = pud->pud_fsuid;
 261         ucred->uc_fsgid = pud->pud_fsgid;
 262
 263         /* process root_squash here. */
 264         mdt_root_squash(info, peernid);
 265
 266         /* remove fs privilege for non-root user. */
 267         if (ucred->uc_fsuid && drop_fs_cap)
 268                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 269         else
 270                 ucred->uc_cap = pud->pud_cap;
 271         ucred->uc_valid = UCRED_NEW;
 272         ucred_set_jobid(info, ucred);
 273
 274         EXIT;
 275
 276 out:
 277         if (rc) {
 278                 if (ucred->uc_ginfo) {
 279                         put_group_info(ucred->uc_ginfo);
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282                 if (ucred->uc_identity) {
 283                         mdt_identity_put(mdt->mdt_identity_cache,
 284                                          ucred->uc_identity);
 285                         ucred->uc_identity = NULL;
 286                 }
 287         }
 288
 289         return rc;
 290 }
 291
 292 /**
 293  * Check whether allow the client to set supplementary group IDs or not.
 294  *
 295  * \param[in] info      pointer to the thread context
 296  * \param[in] uc        pointer to the RPC user descriptor
 297  *
 298  * \retval              true if allow to set supplementary group IDs
 299  * \retval              false for other cases
 300  */
 301 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 302 {
 303         __u32 perm;
 304
 305         /* 1. If identity_upcall is disabled,
 306          *    permit local client to do anything. */
 307         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 308                 return true;
 309
 310         /* 2. If fail to get related identities, then forbid any client to
 311          *    set supplementary group IDs. */
 312         if (uc->uc_identity == NULL)
 313                 return false;
 314
 315         /* 3. Check the permission in the identities. */
 316         perm = mdt_identity_get_perm(uc->uc_identity,
 317                                      mdt_info_req(info)->rq_peer.nid);
 318         if (perm & CFS_SETGRP_PERM)
 319                 return true;
 320
 321         return false;
 322 }
 323
 324 int mdt_check_ucred(struct mdt_thread_info *info)
 325 {
 326         struct ptlrpc_request   *req = mdt_info_req(info);
 327         struct mdt_device       *mdt = info->mti_mdt;
 328         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 329         struct lu_ucred         *ucred = mdt_ucred(info);
 330         struct md_identity      *identity = NULL;
 331         lnet_nid_t               peernid = req->rq_peer.nid;
 332         __u32                    perm = 0;
 333         int                      setuid;
 334         int                      setgid;
 335         int                      rc = 0;
 336
 337         ENTRY;
 338
 339         LASSERT(ucred != NULL);
 340         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 341                 RETURN(0);
 342
 343         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 344                 RETURN(0);
 345
 346         /* sanity check: if we use strong authentication, we expect the
 347          * uid which client claimed is true */
 348         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 349             req->rq_auth_uid != pud->pud_uid) {
 350                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 351                        "while client claims %u:%u/%u:%u\n",
 352                        libcfs_nid2str(peernid), req->rq_auth_uid,
 353                        pud->pud_uid, pud->pud_gid,
 354                        pud->pud_fsuid, pud->pud_fsgid);
 355                 RETURN(-EACCES);
 356         }
 357
 358         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 359                 RETURN(0);
 360
 361         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 362         if (IS_ERR(identity)) {
 363                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 364                         RETURN(0);
 365                 } else {
 366                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 367                                pud->pud_uid);
 368                         RETURN(-EACCES);
 369                 }
 370         }
 371
 372         perm = mdt_identity_get_perm(identity, peernid);
 373         /* find out the setuid/setgid attempt */
 374         setuid = (pud->pud_uid != pud->pud_fsuid);
 375         setgid = (pud->pud_gid != pud->pud_fsgid ||
 376                   pud->pud_gid != identity->mi_gid);
 377
 378         /* check permission of setuid */
 379         if (setuid && !(perm & CFS_SETUID_PERM)) {
 380                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 381                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 382                 GOTO(out, rc = -EACCES);
 383         }
 384
 385         /* check permission of setgid */
 386         if (setgid && !(perm & CFS_SETGID_PERM)) {
 387                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 388                        "from %s\n", pud->pud_uid, pud->pud_gid,
 389                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 390                        libcfs_nid2str(peernid));
 391                 GOTO(out, rc = -EACCES);
 392         }
 393
 394         EXIT;
 395
 396 out:
 397         mdt_identity_put(mdt->mdt_identity_cache, identity);
 398         return rc;
 399 }
 400
 401 static int old_init_ucred_common(struct mdt_thread_info *info,
 402                                  struct lu_nodemap *nodemap,
 403                                  bool drop_fs_cap)
 404 {
 405         struct lu_ucred         *uc = mdt_ucred(info);
 406         struct mdt_device       *mdt = info->mti_mdt;
 407         struct md_identity      *identity = NULL;
 408
 409         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 410                 /* deny access before we get identity ref */
 411                 if (nodemap->nmf_deny_unknown)
 412                         RETURN(-EACCES);
 413
 414                 uc->uc_fsuid = nodemap->nm_squash_uid;
 415                 uc->uc_fsgid = nodemap->nm_squash_gid;
 416                 uc->uc_cap = 0;
 417                 uc->uc_suppgids[0] = -1;
 418                 uc->uc_suppgids[1] = -1;
 419         }
 420
 421         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 422                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 423                                             uc->uc_fsuid);
 424                 if (IS_ERR(identity)) {
 425                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 426                                      uc->uc_cap & CFS_CAP_FS_MASK)) {
 427                                 identity = NULL;
 428                         } else {
 429                                 CDEBUG(D_SEC, "Deny access without identity: "
 430                                        "uid %u\n", uc->uc_fsuid);
 431                                 RETURN(-EACCES);
 432                         }
 433                 }
 434         }
 435         uc->uc_identity = identity;
 436
 437         /* process root_squash here. */
 438         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 439
 440         /* remove fs privilege for non-root user. */
 441         if (uc->uc_fsuid && drop_fs_cap)
 442                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 443         uc->uc_valid = UCRED_OLD;
 444         ucred_set_jobid(info, uc);
 445
 446         EXIT;
 447
 448         return 0;
 449 }
 450
 451 static int old_init_ucred(struct mdt_thread_info *info,
 452                           struct mdt_body *body, bool drop_fs_cap)
 453 {
 454         struct lu_ucred *uc = mdt_ucred(info);
 455         struct lu_nodemap *nodemap;
 456         int rc;
 457         ENTRY;
 458
 459         nodemap = nodemap_get_from_exp(info->mti_exp);
 460         if (IS_ERR(nodemap))
 461                 RETURN(PTR_ERR(nodemap));
 462
 463         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 464                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 465         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 466                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 467         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 468                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 469         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 470                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 471
 472         LASSERT(uc != NULL);
 473         uc->uc_valid = UCRED_INVALID;
 474         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 475         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 476         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 477         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 478         uc->uc_suppgids[0] = body->mbo_suppgid;
 479         uc->uc_suppgids[1] = -1;
 480         uc->uc_ginfo = NULL;
 481         uc->uc_cap = body->mbo_capability;
 482
 483         rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
 484         nodemap_putref(nodemap);
 485
 486         RETURN(rc);
 487 }
 488
 489 static int old_init_ucred_reint(struct mdt_thread_info *info)
 490 {
 491         struct lu_ucred *uc = mdt_ucred(info);
 492         struct lu_nodemap *nodemap;
 493         int rc;
 494         ENTRY;
 495
 496         nodemap = nodemap_get_from_exp(info->mti_exp);
 497         if (IS_ERR(nodemap))
 498                 RETURN(PTR_ERR(nodemap));
 499
 500         LASSERT(uc != NULL);
 501
 502         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 503                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 504         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 505                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 506
 507         uc->uc_valid = UCRED_INVALID;
 508         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 509         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 510         uc->uc_ginfo = NULL;
 511
 512         rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
 513         nodemap_putref(nodemap);
 514
 515         RETURN(rc);
 516 }
 517
 518 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 519                                    struct mdt_body *body,
 520                                    bool drop_fs_cap)
 521 {
 522         struct ptlrpc_request   *req = mdt_info_req(info);
 523         struct lu_ucred         *uc  = mdt_ucred(info);
 524
 525         LASSERT(uc != NULL);
 526         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 527                 return 0;
 528
 529         mdt_exit_ucred(info);
 530
 531         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 532                 return old_init_ucred(info, body, drop_fs_cap);
 533         else
 534                 return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
 535 }
 536
 537 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 538 {
 539         return __mdt_init_ucred(info, body, true);
 540 }
 541
 542 /* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
 543  * doesn't set correct fsuid explicitely, but raise capability to allow
 544  * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
 545  * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
 546  * will fail permission check. */
 547 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
 548                                   struct mdt_body *body)
 549 {
 550         return __mdt_init_ucred(info, body, false);
 551 }
 552
 553 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 554 {
 555         struct ptlrpc_request *req = mdt_info_req(info);
 556         struct lu_ucred       *uc  = mdt_ucred(info);
 557         struct md_attr        *ma  = &info->mti_attr;
 558
 559         LASSERT(uc != NULL);
 560         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 561                 return 0;
 562
 563         /* LU-5564: for normal close request, skip permission check */
 564         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 565             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP)))
 566                 uc->uc_cap |= CFS_CAP_FS_MASK;
 567
 568         mdt_exit_ucred(info);
 569
 570         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 571                 return old_init_ucred_reint(info);
 572         else
 573                 return new_init_ucred(info, REC_INIT, NULL, true);
 574 }
 575
 576 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 577 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 578 {
 579         const struct lov_ost_data_v1 *lod;
 580         __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
 581         __u16 count;
 582         int i;
 583
 584         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 585                 return;
 586
 587         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 588                POSTID(&lmm->lmm_oi), lmm_magic,
 589                le32_to_cpu(lmm->lmm_pattern));
 590
 591         /* No support for compount layouts yet */
 592         if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
 593                 return;
 594
 595         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 596         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 597                le32_to_cpu(lmm->lmm_stripe_size), count);
 598
 599         /* If it's a directory or a released file, then there are
 600          * no actual objects to print, so bail out. */
 601         if (valid & OBD_MD_FLDIREA ||
 602             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 603                 return;
 604
 605         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 606         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 607                 struct ost_id oi;
 608
 609                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 610                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 611                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 612         }
 613 }
 614
 615 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 616 {
 617         const struct lmv_mds_md_v1 *lmm1;
 618         int                        i;
 619
 620         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 621                 return;
 622
 623         lmm1 = &lmv->lmv_md_v1;
 624         CDEBUG(level,
 625                "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
 626                le32_to_cpu(lmm1->lmv_magic),
 627                le32_to_cpu(lmm1->lmv_master_mdt_index),
 628                le32_to_cpu(lmm1->lmv_stripe_count),
 629                le32_to_cpu(lmm1->lmv_hash_type));
 630
 631         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 632                 return;
 633
 634         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 635                 struct lu_fid fid;
 636
 637                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 638                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 639         }
 640 }
 641
 642 /* Shrink and/or grow reply buffers */
 643 int mdt_fix_reply(struct mdt_thread_info *info)
 644 {
 645         struct req_capsule *pill = info->mti_pill;
 646         struct mdt_body    *body;
 647         int                md_size, md_packed = 0;
 648         int                acl_size;
 649         int                rc = 0;
 650         ENTRY;
 651
 652         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 653         LASSERT(body != NULL);
 654
 655         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 656                                OBD_MD_LINKNAME))
 657                 md_size = body->mbo_eadatasize;
 658         else
 659                 md_size = 0;
 660
 661         acl_size = body->mbo_aclsize;
 662
 663         /* this replay - not send info to client */
 664         if (info->mti_spec.no_create) {
 665                 md_size = 0;
 666                 acl_size = 0;
 667         }
 668
 669         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 670                md_size, acl_size);
 671 /*
 672             &RMF_MDT_BODY,
 673             &RMF_MDT_MD,
 674             &RMF_ACL, or &RMF_LOGCOOKIES
 675 (optional)  &RMF_CAPA1,
 676 (optional)  &RMF_CAPA2,
 677 (optional)  something else
 678 */
 679
 680         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 681          * buffers before growing it */
 682         if (info->mti_big_lmm_used) {
 683                 /* big_lmm buffer may be used even without packing the result
 684                  * into reply, just for internal server needs */
 685                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 686                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 687                                                          RCL_SERVER);
 688
 689                 /* free big lmm if md_size is not needed */
 690                 if (md_size == 0 || md_packed == 0) {
 691                         info->mti_big_lmm_used = 0;
 692                 } else {
 693                         /* buffer must be allocated separately */
 694                         LASSERT(info->mti_attr.ma_lmm !=
 695                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 696                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 697                 }
 698         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 699                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 700         }
 701
 702         if (info->mti_big_acl_used) {
 703                 if (acl_size == 0)
 704                         info->mti_big_acl_used = 0;
 705                 else
 706                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 707         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 708                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 709         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 710                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 711         }
 712
 713         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 714             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 715                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 716
 717         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 718             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 719                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 720
 721         /*
 722          * Some more field should be shrinked if needed.
 723          * This should be done by those who added fields to reply message.
 724          */
 725
 726         /* Grow MD buffer if needed finally */
 727         if (info->mti_big_lmm_used) {
 728                 void *lmm;
 729
 730                 LASSERT(md_size > md_packed);
 731                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 732                        md_size - md_packed);
 733                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 734                 if (rc) {
 735                         /* we can't answer with proper LOV EA, drop flags,
 736                          * the rc is also returned so this request is
 737                          * considered as failed */
 738                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 739                         /* don't return transno along with error */
 740                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 741                 } else {
 742                         /* now we need to pack right LOV/LMV EA */
 743                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 744                         if (info->mti_attr.ma_valid & MA_LOV) {
 745                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 746                                                              RCL_SERVER) ==
 747                                                 info->mti_attr.ma_lmm_size);
 748                                 memcpy(lmm, info->mti_attr.ma_lmm,
 749                                        info->mti_attr.ma_lmm_size);
 750                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 751                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 752                                                              RCL_SERVER) ==
 753                                                 info->mti_attr.ma_lmv_size);
 754                                 memcpy(lmm, info->mti_attr.ma_lmv,
 755                                        info->mti_attr.ma_lmv_size);
 756                         }
 757                 }
 758
 759                 /* update mdt_max_mdsize so clients will be aware about that */
 760                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 761                         info->mti_mdt->mdt_max_mdsize =
 762                                                 info->mti_attr.ma_lmm_size;
 763                 info->mti_big_lmm_used = 0;
 764         }
 765
 766         if (info->mti_big_acl_used) {
 767                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 768                        acl_size);
 769
 770                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 771                 if (rc) {
 772                         body->mbo_valid &= ~OBD_MD_FLACL;
 773                 } else {
 774                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 775
 776                         memcpy(acl, info->mti_big_acl, acl_size);
 777                 }
 778
 779                 info->mti_big_acl_used = 0;
 780         }
 781
 782         RETURN(rc);
 783 }
 784
 785
 786 /* if object is dying, pack the lov/llog data,
 787  * parameter info->mti_attr should be valid at this point!
 788  * Also implements RAoLU policy */
 789 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 790                            struct md_attr *ma)
 791 {
 792         struct mdt_body *repbody = NULL;
 793         const struct lu_attr *la = &ma->ma_attr;
 794         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 795         int rc;
 796         __u64 need = 0;
 797         struct hsm_action_item hai = {
 798                 .hai_len = sizeof(hai),
 799                 .hai_action = HSMA_REMOVE,
 800                 .hai_extent.length = -1,
 801                 .hai_cookie = 0,
 802                 .hai_gid = 0,
 803         };
 804         __u64 compound_id;
 805         int archive_id;
 806
 807         ENTRY;
 808
 809         if (mdt_info_req(info) != NULL) {
 810                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 811                 LASSERT(repbody != NULL);
 812         } else {
 813                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 814         }
 815
 816         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 817                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 818
 819         if (ma->ma_valid & MA_LOV) {
 820                 CERROR("No need in LOV EA upon unlink\n");
 821                 dump_stack();
 822         }
 823         if (repbody != NULL)
 824                 repbody->mbo_eadatasize = 0;
 825
 826         /* Only check unlinked and archived if RAoLU and upon last close */
 827         if (!cdt->cdt_remove_archive_on_last_unlink ||
 828             atomic_read(&mo->mot_open_count) != 0)
 829                 RETURN(0);
 830
 831         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 832         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 833                 RETURN(0);
 834
 835         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 836                 RETURN(0);
 837
 838         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 839         if (need != 0) {
 840                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 841                  * try setting them */
 842                 ma->ma_need |= need;
 843                 rc = mdt_attr_get_complex(info, mo, ma);
 844                 if (rc) {
 845                         CERROR("%s: unable to fetch missing attributes of"
 846                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 847                                PFID(mdt_object_fid(mo)), rc);
 848                         RETURN(0);
 849                 }
 850
 851                 if (need & MA_INODE) {
 852                         if (ma->ma_valid & MA_INODE) {
 853                                 if (ma->ma_attr.la_nlink != 0)
 854                                         RETURN(0);
 855                         } else {
 856                                 RETURN(0);
 857                         }
 858                 }
 859
 860                 if (need & MA_HSM) {
 861                         if (ma->ma_valid & MA_HSM) {
 862                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 863                                         RETURN(0);
 864                         } else {
 865                                 RETURN(0);
 866                         }
 867                 }
 868         }
 869
 870         /* RAoLU policy is active, last close on file has occured,
 871          * file is unlinked, file is archived, so create remove request
 872          * for copytool!
 873          * If CDT is not running, requests will be logged for later. */
 874         compound_id = atomic_inc_return(&cdt->cdt_compound_id);
 875         if (ma->ma_hsm.mh_arch_id != 0)
 876                 archive_id = ma->ma_hsm.mh_arch_id;
 877         else
 878                 archive_id = cdt->cdt_default_archive_id;
 879
 880         hai.hai_fid = *mdt_object_fid(mo);
 881
 882         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt,
 883                                   compound_id, archive_id, 0, &hai);
 884         if (rc)
 885                 CERROR("%s: unable to add HSM remove request for "DFID
 886                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
 887                        PFID(mdt_object_fid(mo)), rc);
 888
 889         RETURN(0);
 890 }
 891
 892 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 893                                   struct md_attr *ma)
 894 {
 895         __u64 out;
 896
 897         out = 0;
 898         if (in & MDS_ATTR_MODE)
 899                 out |= LA_MODE;
 900         if (in & MDS_ATTR_UID)
 901                 out |= LA_UID;
 902         if (in & MDS_ATTR_GID)
 903                 out |= LA_GID;
 904         if (in & MDS_ATTR_SIZE)
 905                 out |= LA_SIZE;
 906         if (in & MDS_ATTR_BLOCKS)
 907                 out |= LA_BLOCKS;
 908         if (in & MDS_ATTR_ATIME_SET)
 909                 out |= LA_ATIME;
 910         if (in & MDS_ATTR_CTIME_SET)
 911                 out |= LA_CTIME;
 912         if (in & MDS_ATTR_MTIME_SET)
 913                 out |= LA_MTIME;
 914         if (in & MDS_ATTR_ATTR_FLAG)
 915                 out |= LA_FLAGS;
 916         if (in & MDS_ATTR_KILL_SUID)
 917                 out |= LA_KILL_SUID;
 918         if (in & MDS_ATTR_KILL_SGID)
 919                 out |= LA_KILL_SGID;
 920         if (in & MDS_ATTR_PROJID)
 921                 out |= LA_PROJID;
 922
 923         if (in & MDS_ATTR_FROM_OPEN)
 924                 rr->rr_flags |= MRF_OPEN_TRUNC;
 925         if (in & MDS_OPEN_OWNEROVERRIDE)
 926                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 927         if (in & MDS_ATTR_FORCE)
 928                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 929
 930         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
 931                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 932                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 933                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 934                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 935                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 936         if (in != 0)
 937                 CERROR("Unknown attr bits: %#llx\n", in);
 938         return out;
 939 }
 940
 941 /* unpacking */
 942
 943 int mdt_name_unpack(struct req_capsule *pill,
 944                     const struct req_msg_field *field,
 945                     struct lu_name *ln,
 946                     enum mdt_name_flags flags)
 947 {
 948         ln->ln_name = req_capsule_client_get(pill, field);
 949         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 950
 951         if (!lu_name_is_valid(ln)) {
 952                 ln->ln_name = NULL;
 953                 ln->ln_namelen = 0;
 954
 955                 return -EPROTO;
 956         }
 957
 958         if ((flags & MNF_FIX_ANON) &&
 959             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 960                 /* Newer (3.x) kernels use a name of "/" for the
 961                  * "anonymous" disconnected dentries from NFS
 962                  * filehandle conversion. See d_obtain_alias(). */
 963                 ln->ln_name = NULL;
 964                 ln->ln_namelen = 0;
 965         }
 966
 967         return 0;
 968 }
 969
 970 static int mdt_file_secctx_unpack(struct req_capsule *pill,
 971                                   const char **secctx_name,
 972                                   void **secctx, size_t *secctx_size)
 973 {
 974         const char *name;
 975         size_t name_size;
 976
 977         *secctx_name = NULL;
 978         *secctx = NULL;
 979         *secctx_size = 0;
 980
 981         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
 982             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
 983                 return 0;
 984
 985         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
 986                                          RCL_CLIENT);
 987         if (name_size == 0)
 988                 return 0;
 989
 990         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
 991         if (strnlen(name, name_size) != name_size - 1)
 992                 return -EPROTO;
 993
 994         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
 995             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
 996                 return -EPROTO;
 997
 998         *secctx_name = name;
 999         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1000         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1001
1002         return 0;
1003 }
1004
1005 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1006 {
1007         struct lu_ucred *uc = mdt_ucred(info);
1008         struct md_attr *ma = &info->mti_attr;
1009         struct lu_attr *la = &ma->ma_attr;
1010         struct req_capsule *pill = info->mti_pill;
1011         struct mdt_reint_record *rr = &info->mti_rr;
1012         struct mdt_rec_setattr *rec;
1013         struct lu_nodemap *nodemap;
1014
1015         ENTRY;
1016
1017         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1018         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1019         if (rec == NULL)
1020                 RETURN(-EFAULT);
1021
1022         /* This prior initialization is needed for old_init_ucred_reint() */
1023         uc->uc_fsuid = rec->sa_fsuid;
1024         uc->uc_fsgid = rec->sa_fsgid;
1025         uc->uc_cap   = rec->sa_cap;
1026         uc->uc_suppgids[0] = rec->sa_suppgid;
1027         uc->uc_suppgids[1] = -1;
1028
1029         rr->rr_fid1 = &rec->sa_fid;
1030         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1031         la->la_mode  = rec->sa_mode;
1032         la->la_flags = rec->sa_attr_flags;
1033
1034         nodemap = nodemap_get_from_exp(info->mti_exp);
1035         if (IS_ERR(nodemap))
1036                 RETURN(PTR_ERR(nodemap));
1037
1038         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
1039                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1040         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
1041                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1042         la->la_projid = rec->sa_projid;
1043         nodemap_putref(nodemap);
1044
1045         la->la_size  = rec->sa_size;
1046         la->la_blocks = rec->sa_blocks;
1047         la->la_ctime = rec->sa_ctime;
1048         la->la_atime = rec->sa_atime;
1049         la->la_mtime = rec->sa_mtime;
1050         ma->ma_valid = MA_INODE;
1051
1052         if (rec->sa_bias & MDS_DATA_MODIFIED)
1053                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
1054         else
1055                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
1056
1057         ma->ma_attr_flags &= ~MDS_CLOSE_INTENT;
1058         ma->ma_attr_flags |= rec->sa_bias & MDS_CLOSE_INTENT;
1059         RETURN(0);
1060 }
1061
1062 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1063 {
1064         struct req_capsule *pill = info->mti_pill;
1065         struct mdt_ioepoch *ioepoch;
1066         ENTRY;
1067
1068         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1069                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1070         else
1071                 ioepoch = NULL;
1072
1073         if (ioepoch == NULL)
1074                 RETURN(-EPROTO);
1075
1076         info->mti_close_handle = ioepoch->mio_handle;
1077
1078         RETURN(0);
1079 }
1080
1081 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1082         struct req_capsule      *pill = info->mti_pill;
1083
1084         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1085                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1086                 if (info->mti_dlm_req == NULL)
1087                         RETURN(-EFAULT);
1088         }
1089
1090         RETURN(0);
1091 }
1092
1093 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1094 {
1095         struct mdt_reint_record *rr = &info->mti_rr;
1096         struct md_attr          *ma = &info->mti_attr;
1097         struct req_capsule      *pill = info->mti_pill;
1098         int rc;
1099         ENTRY;
1100
1101         rc = mdt_setattr_unpack_rec(info);
1102         if (rc)
1103                 RETURN(rc);
1104
1105         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1106                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1107                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1108                                                         RCL_CLIENT);
1109                 if (rr->rr_eadatalen > 0) {
1110                         const struct lmv_user_md        *lum;
1111
1112                         lum = rr->rr_eadata;
1113                         /* Sigh ma_valid(from req) does not indicate whether
1114                          * it will set LOV/LMV EA, so we have to check magic */
1115                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1116                                 ma->ma_valid |= MA_LMV;
1117                                 ma->ma_lmv = (void *)rr->rr_eadata;
1118                                 ma->ma_lmv_size = rr->rr_eadatalen;
1119                         } else {
1120                                 ma->ma_valid |= MA_LOV;
1121                                 ma->ma_lmm = (void *)rr->rr_eadata;
1122                                 ma->ma_lmm_size = rr->rr_eadatalen;
1123                         }
1124                 }
1125         }
1126
1127         rc = mdt_dlmreq_unpack(info);
1128         RETURN(rc);
1129 }
1130
1131 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1132 {
1133         struct md_attr          *ma = &info->mti_attr;
1134         struct req_capsule      *pill = info->mti_pill;
1135         ENTRY;
1136
1137         if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1138                 RETURN(0);
1139
1140         req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1141
1142         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1143             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1144                 RETURN(-EFAULT);
1145
1146         RETURN(0);
1147 }
1148
1149 int mdt_close_unpack(struct mdt_thread_info *info)
1150 {
1151         int rc;
1152         ENTRY;
1153
1154         rc = mdt_close_handle_unpack(info);
1155         if (rc)
1156                 RETURN(rc);
1157
1158         rc = mdt_setattr_unpack_rec(info);
1159         if (rc)
1160                 RETURN(rc);
1161
1162         rc = mdt_close_intent_unpack(info);
1163         if (rc)
1164                 RETURN(rc);
1165
1166         RETURN(mdt_init_ucred_reint(info));
1167 }
1168
1169 static int mdt_create_unpack(struct mdt_thread_info *info)
1170 {
1171         struct lu_ucred *uc  = mdt_ucred(info);
1172         struct mdt_rec_create *rec;
1173         struct lu_attr *attr = &info->mti_attr.ma_attr;
1174         struct mdt_reint_record *rr = &info->mti_rr;
1175         struct req_capsule *pill = info->mti_pill;
1176         struct md_op_spec *sp = &info->mti_spec;
1177         int rc;
1178
1179         ENTRY;
1180
1181         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1182         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1183         if (rec == NULL)
1184                 RETURN(-EFAULT);
1185
1186         /* This prior initialization is needed for old_init_ucred_reint() */
1187         uc->uc_fsuid = rec->cr_fsuid;
1188         uc->uc_fsgid = rec->cr_fsgid;
1189         uc->uc_cap   = rec->cr_cap;
1190         uc->uc_suppgids[0] = rec->cr_suppgid1;
1191         uc->uc_suppgids[1] = -1;
1192         uc->uc_umask = rec->cr_umask;
1193
1194         rr->rr_fid1 = &rec->cr_fid1;
1195         rr->rr_fid2 = &rec->cr_fid2;
1196         attr->la_mode = rec->cr_mode;
1197         attr->la_rdev  = rec->cr_rdev;
1198         attr->la_uid   = rec->cr_fsuid;
1199         attr->la_gid   = rec->cr_fsgid;
1200         attr->la_ctime = rec->cr_time;
1201         attr->la_mtime = rec->cr_time;
1202         attr->la_atime = rec->cr_time;
1203         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1204                          LA_CTIME | LA_MTIME | LA_ATIME;
1205         memset(&sp->u, 0, sizeof(sp->u));
1206         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1207
1208         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1209         if (rc < 0)
1210                 RETURN(rc);
1211
1212         if (S_ISLNK(attr->la_mode)) {
1213                 const char *tgt = NULL;
1214
1215                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1216                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1217                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1218                         sp->u.sp_symname = tgt;
1219                 }
1220                 if (tgt == NULL)
1221                         RETURN(-EFAULT);
1222         } else {
1223                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1224                 if (S_ISDIR(attr->la_mode) &&
1225                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1226                         sp->u.sp_ea.eadata =
1227                                 req_capsule_client_get(pill, &RMF_EADATA);
1228                         sp->u.sp_ea.eadatalen =
1229                                 req_capsule_get_size(pill, &RMF_EADATA,
1230                                                      RCL_CLIENT);
1231                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1232                 }
1233         }
1234
1235         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1236                                     &sp->sp_cr_file_secctx,
1237                                     &sp->sp_cr_file_secctx_size);
1238         if (rc < 0)
1239                 RETURN(rc);
1240
1241         rc = mdt_dlmreq_unpack(info);
1242         RETURN(rc);
1243 }
1244
1245 static int mdt_link_unpack(struct mdt_thread_info *info)
1246 {
1247         struct lu_ucred *uc  = mdt_ucred(info);
1248         struct mdt_rec_link *rec;
1249         struct lu_attr *attr = &info->mti_attr.ma_attr;
1250         struct mdt_reint_record *rr = &info->mti_rr;
1251         struct req_capsule *pill = info->mti_pill;
1252         int rc;
1253
1254         ENTRY;
1255
1256         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1257         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1258         if (rec == NULL)
1259                 RETURN(-EFAULT);
1260
1261         /* This prior initialization is needed for old_init_ucred_reint() */
1262         uc->uc_fsuid = rec->lk_fsuid;
1263         uc->uc_fsgid = rec->lk_fsgid;
1264         uc->uc_cap   = rec->lk_cap;
1265         uc->uc_suppgids[0] = rec->lk_suppgid1;
1266         uc->uc_suppgids[1] = rec->lk_suppgid2;
1267
1268         attr->la_uid = rec->lk_fsuid;
1269         attr->la_gid = rec->lk_fsgid;
1270         rr->rr_fid1 = &rec->lk_fid1;
1271         rr->rr_fid2 = &rec->lk_fid2;
1272         attr->la_ctime = rec->lk_time;
1273         attr->la_mtime = rec->lk_time;
1274         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1275
1276         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1277         if (rc < 0)
1278                 RETURN(rc);
1279
1280         rc = mdt_dlmreq_unpack(info);
1281
1282         RETURN(rc);
1283 }
1284
1285 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1286 {
1287         struct lu_ucred *uc  = mdt_ucred(info);
1288         struct mdt_rec_unlink *rec;
1289         struct md_attr *ma = &info->mti_attr;
1290         struct lu_attr *attr = &info->mti_attr.ma_attr;
1291         struct mdt_reint_record *rr = &info->mti_rr;
1292         struct req_capsule *pill = info->mti_pill;
1293         int rc;
1294
1295         ENTRY;
1296
1297         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1298         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1299         if (rec == NULL)
1300                 RETURN(-EFAULT);
1301
1302         /* This prior initialization is needed for old_init_ucred_reint() */
1303         uc->uc_fsuid = rec->ul_fsuid;
1304         uc->uc_fsgid = rec->ul_fsgid;
1305         uc->uc_cap   = rec->ul_cap;
1306         uc->uc_suppgids[0] = rec->ul_suppgid1;
1307         uc->uc_suppgids[1] = -1;
1308
1309         attr->la_uid = rec->ul_fsuid;
1310         attr->la_gid = rec->ul_fsgid;
1311         rr->rr_fid1 = &rec->ul_fid1;
1312         rr->rr_fid2 = &rec->ul_fid2;
1313         attr->la_ctime = rec->ul_time;
1314         attr->la_mtime = rec->ul_time;
1315         attr->la_mode  = rec->ul_mode;
1316         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1317
1318         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1319         if (rc < 0)
1320                 RETURN(rc);
1321
1322         if (rec->ul_bias & MDS_VTX_BYPASS)
1323                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1324         else
1325                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1326
1327         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1328
1329         rc = mdt_dlmreq_unpack(info);
1330         RETURN(rc);
1331 }
1332
1333 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1334 {
1335         info->mti_spec.sp_rm_entry = 1;
1336         return mdt_unlink_unpack(info);
1337 }
1338
1339 static int mdt_rename_unpack(struct mdt_thread_info *info)
1340 {
1341         struct lu_ucred *uc = mdt_ucred(info);
1342         struct mdt_rec_rename *rec;
1343         struct md_attr *ma = &info->mti_attr;
1344         struct lu_attr *attr = &info->mti_attr.ma_attr;
1345         struct mdt_reint_record *rr = &info->mti_rr;
1346         struct req_capsule *pill = info->mti_pill;
1347         int rc;
1348
1349         ENTRY;
1350
1351         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1352         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1353         if (rec == NULL)
1354                 RETURN(-EFAULT);
1355
1356         /* This prior initialization is needed for old_init_ucred_reint() */
1357         uc->uc_fsuid = rec->rn_fsuid;
1358         uc->uc_fsgid = rec->rn_fsgid;
1359         uc->uc_cap   = rec->rn_cap;
1360         uc->uc_suppgids[0] = rec->rn_suppgid1;
1361         uc->uc_suppgids[1] = rec->rn_suppgid2;
1362
1363         attr->la_uid = rec->rn_fsuid;
1364         attr->la_gid = rec->rn_fsgid;
1365         rr->rr_fid1 = &rec->rn_fid1;
1366         rr->rr_fid2 = &rec->rn_fid2;
1367         attr->la_ctime = rec->rn_time;
1368         attr->la_mtime = rec->rn_time;
1369         /* rename_tgt contains the mode already */
1370         attr->la_mode = rec->rn_mode;
1371         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1372
1373         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1374         if (rc < 0)
1375                 RETURN(rc);
1376
1377         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1378         if (rc < 0)
1379                 RETURN(rc);
1380
1381         if (rec->rn_bias & MDS_VTX_BYPASS)
1382                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1383         else
1384                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1385
1386         if (rec->rn_bias & MDS_RENAME_MIGRATE) {
1387                 req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
1388                 rc = mdt_close_handle_unpack(info);
1389                 if (rc < 0)
1390                         RETURN(rc);
1391                 info->mti_spec.sp_migrate_close = 1;
1392         }
1393
1394         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1395
1396
1397         rc = mdt_dlmreq_unpack(info);
1398
1399         RETURN(rc);
1400 }
1401
1402 /*
1403  * please see comment above LOV_MAGIC_V1_DEFINED
1404  */
1405 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1406 {
1407         struct lov_user_md_v1   *v1 = eadata;
1408
1409         LASSERT(v1);
1410
1411         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1412                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1413                         v1->lmm_magic |= LOV_MAGIC_DEFINED;
1414                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1415                          __swab32(LOV_MAGIC_MAGIC))
1416                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1417         }
1418 }
1419
1420 static int mdt_open_unpack(struct mdt_thread_info *info)
1421 {
1422         struct lu_ucred *uc = mdt_ucred(info);
1423         struct mdt_rec_create *rec;
1424         struct lu_attr *attr = &info->mti_attr.ma_attr;
1425         struct req_capsule *pill = info->mti_pill;
1426         struct mdt_reint_record *rr = &info->mti_rr;
1427         struct ptlrpc_request *req = mdt_info_req(info);
1428         struct md_op_spec *sp = &info->mti_spec;
1429         int rc;
1430         ENTRY;
1431
1432         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1433         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1434         if (rec == NULL)
1435                 RETURN(-EFAULT);
1436
1437         /* This prior initialization is needed for old_init_ucred_reint() */
1438         uc->uc_fsuid = rec->cr_fsuid;
1439         uc->uc_fsgid = rec->cr_fsgid;
1440         uc->uc_cap   = rec->cr_cap;
1441         uc->uc_suppgids[0] = rec->cr_suppgid1;
1442         uc->uc_suppgids[1] = rec->cr_suppgid2;
1443         uc->uc_umask = rec->cr_umask;
1444
1445         rr->rr_fid1   = &rec->cr_fid1;
1446         rr->rr_fid2   = &rec->cr_fid2;
1447         rr->rr_handle = &rec->cr_old_handle;
1448         attr->la_mode = rec->cr_mode;
1449         attr->la_rdev  = rec->cr_rdev;
1450         attr->la_uid   = rec->cr_fsuid;
1451         attr->la_gid   = rec->cr_fsgid;
1452         attr->la_ctime = rec->cr_time;
1453         attr->la_mtime = rec->cr_time;
1454         attr->la_atime = rec->cr_time;
1455         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1456                          LA_CTIME | LA_MTIME | LA_ATIME;
1457         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1458         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1459         /* Do not trigger ASSERTION if client miss to set such flags. */
1460         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1461                 RETURN(-EPROTO);
1462
1463         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1464
1465         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1466
1467         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1468                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1469                                                         RCL_CLIENT);
1470                 if (rr->rr_eadatalen > 0) {
1471                         rr->rr_eadata = req_capsule_client_get(pill,
1472                                                                &RMF_EADATA);
1473                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1474                         sp->u.sp_ea.eadata = rr->rr_eadata;
1475                         sp->no_create = !!req_is_replay(req);
1476                         mdt_fix_lov_magic(info, rr->rr_eadata);
1477                 }
1478
1479                 /*
1480                  * Client default md_size may be 0 right after client start,
1481                  * until all osc are connected, set here just some reasonable
1482                  * value to prevent misbehavior.
1483                  */
1484                 if (rr->rr_eadatalen == 0 &&
1485                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1486                         rr->rr_eadatalen = MIN_MD_SIZE;
1487         }
1488
1489         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1490                                     &sp->sp_cr_file_secctx,
1491                                     &sp->sp_cr_file_secctx_size);
1492
1493         RETURN(rc);
1494 }
1495
1496 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1497 {
1498         struct mdt_reint_record *rr = &info->mti_rr;
1499         struct lu_ucred *uc = mdt_ucred(info);
1500         struct lu_attr *attr = &info->mti_attr.ma_attr;
1501         struct req_capsule *pill = info->mti_pill;
1502         struct mdt_rec_setxattr *rec;
1503         int rc;
1504         ENTRY;
1505
1506
1507         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1508                  sizeof(struct mdt_rec_reint));
1509
1510         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1511         if (rec == NULL)
1512                 RETURN(-EFAULT);
1513
1514         /* This prior initialization is needed for old_init_ucred_reint() */
1515         uc->uc_fsuid  = rec->sx_fsuid;
1516         uc->uc_fsgid  = rec->sx_fsgid;
1517         uc->uc_cap    = rec->sx_cap;
1518         uc->uc_suppgids[0] = rec->sx_suppgid1;
1519         uc->uc_suppgids[1] = -1;
1520
1521         rr->rr_opcode = rec->sx_opcode;
1522         rr->rr_fid1   = &rec->sx_fid;
1523         attr->la_valid = rec->sx_valid;
1524         attr->la_ctime = rec->sx_time;
1525         attr->la_size = rec->sx_size;
1526         attr->la_flags = rec->sx_flags;
1527
1528         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1529         if (rc < 0)
1530                 RETURN(rc);
1531
1532         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1533                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1534                                                         RCL_CLIENT);
1535                 if (rr->rr_eadatalen > 0) {
1536                         rr->rr_eadata = req_capsule_client_get(pill,
1537                                                                &RMF_EADATA);
1538                         if (rr->rr_eadata == NULL)
1539                                 RETURN(-EFAULT);
1540                 } else {
1541                         rr->rr_eadata = NULL;
1542                 }
1543         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1544                 CDEBUG(D_INFO, "no xattr data supplied\n");
1545                 RETURN(-EFAULT);
1546         }
1547
1548         if (mdt_dlmreq_unpack(info) < 0)
1549                 RETURN(-EPROTO);
1550
1551         RETURN(0);
1552 }
1553
1554 static int mdt_resync_unpack(struct mdt_thread_info *info)
1555 {
1556         struct req_capsule      *pill = info->mti_pill;
1557         struct mdt_reint_record *rr   = &info->mti_rr;
1558         struct lu_ucred         *uc     = mdt_ucred(info);
1559         struct mdt_rec_resync   *rec;
1560         ENTRY;
1561
1562         CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
1563         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1564         if (rec == NULL)
1565                 RETURN(-EFAULT);
1566
1567         /* This prior initialization is needed for old_init_ucred_reint() */
1568         uc->uc_fsuid = rec->rs_fsuid;
1569         uc->uc_fsgid = rec->rs_fsgid;
1570         uc->uc_cap   = rec->rs_cap;
1571
1572         rr->rr_fid1   = &rec->rs_fid;
1573
1574         /* cookie doesn't need to be swapped but it has been swapped
1575          * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1576          * restoring. */
1577         if (ptlrpc_req_need_swab(mdt_info_req(info)))
1578                 __swab64s(&rec->rs_handle.cookie);
1579         rr->rr_handle = &rec->rs_handle;
1580
1581         RETURN(mdt_dlmreq_unpack(info));
1582 }
1583
1584 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1585
1586 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1587         [REINT_SETATTR]  = mdt_setattr_unpack,
1588         [REINT_CREATE]   = mdt_create_unpack,
1589         [REINT_LINK]     = mdt_link_unpack,
1590         [REINT_UNLINK]   = mdt_unlink_unpack,
1591         [REINT_RENAME]   = mdt_rename_unpack,
1592         [REINT_OPEN]     = mdt_open_unpack,
1593         [REINT_SETXATTR] = mdt_setxattr_unpack,
1594         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1595         [REINT_MIGRATE]  = mdt_rename_unpack,
1596         [REINT_RESYNC]   = mdt_resync_unpack,
1597 };
1598
1599 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1600 {
1601         int rc;
1602         ENTRY;
1603
1604         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1605         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1606                 info->mti_rr.rr_opcode = op;
1607                 rc = mdt_reint_unpackers[op](info);
1608         } else {
1609                 CERROR("Unexpected opcode %d\n", op);
1610                 rc = -EFAULT;
1611         }
1612         RETURN(rc);
1613 }