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, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/mdt/mdt_lib.c
  32  *
  33  * Lustre Metadata Target (mdt) request unpacking helper.
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Mike Shaver <shaver@clusterfs.com>
  39  * Author: Nikita Danilov <nikita@clusterfs.com>
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  * Author: Fan Yong <fanyong@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/user_namespace.h>
  47 #include <linux/uidgid.h>
  48
  49 #include "mdt_internal.h"
  50 #include <uapi/linux/lnet/nidstr.h>
  51 #include <lustre_nodemap.h>
  52
  53 typedef enum ucred_init_type {
  54         NONE_INIT       = 0,
  55         BODY_INIT       = 1,
  56         REC_INIT        = 2
  57 } ucred_init_type_t;
  58
  59 static __u64 get_mrc_cr_flags(struct mdt_rec_create *mrc)
  60 {
  61         return (__u64)(mrc->cr_flags_l) | ((__u64)mrc->cr_flags_h << 32);
  62 }
  63
  64 void mdt_exit_ucred(struct mdt_thread_info *info)
  65 {
  66         struct lu_ucred   *uc  = mdt_ucred(info);
  67         struct mdt_device *mdt = info->mti_mdt;
  68
  69         LASSERT(uc != NULL);
  70         if (uc->uc_valid != UCRED_INIT) {
  71                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  72                 if (uc->uc_ginfo) {
  73                         put_group_info(uc->uc_ginfo);
  74                         uc->uc_ginfo = NULL;
  75                 }
  76                 if (uc->uc_identity) {
  77                         mdt_identity_put(mdt->mdt_identity_cache,
  78                                          uc->uc_identity);
  79                         uc->uc_identity = NULL;
  80                 }
  81                 uc->uc_valid = UCRED_INIT;
  82         }
  83 }
  84
  85 static int match_nosquash_list(struct spinlock *rsi_lock,
  86                                struct list_head *nidlist,
  87                                struct lnet_nid *peernid)
  88 {
  89         int rc;
  90         ENTRY;
  91         spin_lock(rsi_lock);
  92         rc = cfs_match_nid(peernid, nidlist);
  93         spin_unlock(rsi_lock);
  94         RETURN(rc);
  95 }
  96
  97 /* root_squash for inter-MDS operations */
  98 static int mdt_root_squash(struct mdt_thread_info *info,
  99                            struct lnet_nid *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_lock,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nidstr(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nidstr(peernid), ucred->uc_fsuid, ucred->uc_fsgid,
 119                (u32)ll_capability_u32(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 = CAP_EMPTY_SET;
 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 void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
 144 {
 145         if (info && info->mti_exp && info->mti_exp->exp_connection)
 146                 uc->uc_nid = lnet_nid_to_nid4(
 147                         &info->mti_exp->exp_connection->c_peer.nid);
 148         else
 149                 uc->uc_nid = LNET_NID_ANY;
 150 }
 151
 152 static void ucred_set_audit_enabled(struct mdt_thread_info *info,
 153                                     struct lu_ucred *uc)
 154 {
 155         struct lu_nodemap *nodemap = NULL;
 156         bool audit = true;
 157
 158         if (info && info->mti_exp) {
 159                 nodemap = nodemap_get_from_exp(info->mti_exp);
 160                 if (nodemap && !IS_ERR(nodemap)) {
 161                         audit = nodemap->nmf_enable_audit;
 162                         nodemap_putref(nodemap);
 163                 }
 164         }
 165
 166         uc->uc_enable_audit = audit;
 167 }
 168
 169 static void ucred_set_rbac_roles(struct mdt_thread_info *info,
 170                                  struct lu_ucred *uc)
 171 {
 172         struct lu_nodemap *nodemap = NULL;
 173         enum nodemap_rbac_roles rbac = NODEMAP_RBAC_ALL;
 174
 175         if (info && info->mti_exp) {
 176                 nodemap = nodemap_get_from_exp(info->mti_exp);
 177                 if (!IS_ERR_OR_NULL(nodemap)) {
 178                         rbac = nodemap->nmf_rbac;
 179                         nodemap_putref(nodemap);
 180                 }
 181         }
 182
 183         uc->uc_rbac_file_perms = !!(rbac & NODEMAP_RBAC_FILE_PERMS);
 184         uc->uc_rbac_dne_ops = !!(rbac & NODEMAP_RBAC_DNE_OPS);
 185         uc->uc_rbac_quota_ops = !!(rbac & NODEMAP_RBAC_QUOTA_OPS);
 186         uc->uc_rbac_byfid_ops = !!(rbac & NODEMAP_RBAC_BYFID_OPS);
 187         uc->uc_rbac_chlg_ops = !!(rbac & NODEMAP_RBAC_CHLG_OPS);
 188         uc->uc_rbac_fscrypt_admin = !!(rbac & NODEMAP_RBAC_FSCRYPT_ADMIN);
 189 }
 190
 191 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 192                           void *buf)
 193 {
 194         struct ptlrpc_request *req = mdt_info_req(info);
 195         struct mdt_device *mdt = info->mti_mdt;
 196         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 197         struct lu_ucred *ucred = mdt_ucred(info);
 198         struct lu_nodemap *nodemap;
 199         struct lnet_nid peernid = req->rq_peer.nid;
 200         __u32 perm = 0;
 201         int setuid;
 202         int setgid;
 203         bool is_nm_gid_squashed = false;
 204         int rc = 0;
 205
 206         ENTRY;
 207
 208         LASSERT(req->rq_auth_gss);
 209         LASSERT(!req->rq_auth_usr_mdt);
 210         LASSERT(req->rq_user_desc);
 211         LASSERT(ucred != NULL);
 212
 213         ucred->uc_valid = UCRED_INVALID;
 214
 215         nodemap = nodemap_get_from_exp(info->mti_exp);
 216         if (IS_ERR(nodemap))
 217                 RETURN(PTR_ERR(nodemap));
 218
 219         pud->pud_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 220                                        NODEMAP_CLIENT_TO_FS, pud->pud_uid);
 221         pud->pud_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 222                                        NODEMAP_CLIENT_TO_FS, pud->pud_gid);
 223         pud->pud_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 224                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsuid);
 225         pud->pud_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 226                                        NODEMAP_CLIENT_TO_FS, pud->pud_fsgid);
 227
 228         ucred->uc_o_uid = pud->pud_uid;
 229         ucred->uc_o_gid = pud->pud_gid;
 230         ucred->uc_o_fsuid = pud->pud_fsuid;
 231         ucred->uc_o_fsgid = pud->pud_fsgid;
 232
 233         if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
 234                 /* deny access before we get identity ref */
 235                 if (nodemap->nmf_deny_unknown) {
 236                         nodemap_putref(nodemap);
 237                         RETURN(-EACCES);
 238                 }
 239
 240                 ucred->uc_suppgids[0] = -1;
 241                 ucred->uc_suppgids[1] = -1;
 242         }
 243
 244         if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
 245                 is_nm_gid_squashed = true;
 246
 247         nodemap_putref(nodemap);
 248
 249         if (type == BODY_INIT) {
 250                 struct mdt_body *body = (struct mdt_body *)buf;
 251
 252                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 253                 ucred->uc_suppgids[1] = -1;
 254         }
 255
 256         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 257             req->rq_auth_uid != pud->pud_uid) {
 258                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 259                        "while client claims %u:%u/%u:%u\n",
 260                        libcfs_nidstr(&peernid), req->rq_auth_uid,
 261                        pud->pud_uid, pud->pud_gid,
 262                        pud->pud_fsuid, pud->pud_fsgid);
 263                 RETURN(-EACCES);
 264         }
 265
 266         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 267                 ucred->uc_identity = NULL;
 268                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM | CFS_SETGRP_PERM;
 269         } else {
 270                 struct md_identity *identity;
 271
 272                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 273                                             pud->pud_uid);
 274                 if (IS_ERR(identity)) {
 275                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 276                                 ucred->uc_identity = NULL;
 277                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 278                                        CFS_SETGRP_PERM;
 279                         } else {
 280                                 CDEBUG(D_SEC,
 281                                        "Deny access without identity: uid %u\n",
 282                                        pud->pud_uid);
 283                                 RETURN(-EACCES);
 284                         }
 285                 } else {
 286                         ucred->uc_identity = identity;
 287                         perm = mdt_identity_get_perm(ucred->uc_identity,
 288                                                      &peernid);
 289                 }
 290         }
 291
 292         /* find out the setuid/setgid attempt */
 293         setuid = (pud->pud_uid != pud->pud_fsuid);
 294         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 295                   (ucred->uc_identity &&
 296                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 297
 298         /* check permission of setuid */
 299         if (setuid && !(perm & CFS_SETUID_PERM)) {
 300                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 301                        pud->pud_uid, pud->pud_fsuid, libcfs_nidstr(&peernid));
 302                 GOTO(out, rc = -EACCES);
 303         }
 304
 305         /* check permission of setgid */
 306         if (setgid && !(perm & CFS_SETGID_PERM)) {
 307                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 308                        "from %s\n", pud->pud_uid, pud->pud_gid,
 309                        pud->pud_fsuid, pud->pud_fsgid,
 310                        ucred->uc_identity->mi_gid, libcfs_nidstr(&peernid));
 311                 GOTO(out, rc = -EACCES);
 312         }
 313
 314         if (perm & CFS_SETGRP_PERM) {
 315                 /* only set groups if GID is not squashed */
 316                 if (pud->pud_ngroups && !is_nm_gid_squashed) {
 317                         /* setgroups for local client */
 318                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 319                         if (!ucred->uc_ginfo) {
 320                                 CERROR("failed to alloc %d groups\n",
 321                                        pud->pud_ngroups);
 322                                 GOTO(out, rc = -ENOMEM);
 323                         }
 324
 325                         lustre_groups_from_list(ucred->uc_ginfo,
 326                                                 pud->pud_groups);
 327                         lustre_groups_sort(ucred->uc_ginfo);
 328                 } else {
 329                         ucred->uc_suppgids[0] = -1;
 330                         ucred->uc_suppgids[1] = -1;
 331                         ucred->uc_ginfo = NULL;
 332                 }
 333         } else {
 334                 ucred->uc_suppgids[0] = -1;
 335                 ucred->uc_suppgids[1] = -1;
 336                 ucred->uc_ginfo = NULL;
 337         }
 338
 339         ucred->uc_uid = pud->pud_uid;
 340         ucred->uc_gid = pud->pud_gid;
 341
 342         ucred->uc_cap = CAP_EMPTY_SET;
 343         if (!nodemap || ucred->uc_o_uid != nodemap->nm_squash_uid)
 344                 ll_set_capability_u32(&ucred->uc_cap, pud->pud_cap);
 345
 346         ucred->uc_fsuid = pud->pud_fsuid;
 347         ucred->uc_fsgid = pud->pud_fsgid;
 348
 349         /* process root_squash here. */
 350         mdt_root_squash(info, &peernid);
 351
 352         if (ucred->uc_fsuid) {
 353                 if (!cap_issubset(ucred->uc_cap, mdt->mdt_enable_cap_mask))
 354                         CDEBUG(D_SEC, "%s: drop capabilities %llx for NID %s\n",
 355                                mdt_obd_name(mdt),
 356 #ifdef CAP_FOR_EACH_U32
 357                                ucred->uc_cap.cap[0] |
 358                                ((u64)ucred->uc_cap.cap[1] << 32),
 359 #else
 360                                ucred->uc_cap.val,
 361 #endif
 362                                libcfs_nidstr(&mdt_info_req(info)->rq_peer.nid));
 363                 ucred->uc_cap = cap_intersect(ucred->uc_cap,
 364                                               mdt->mdt_enable_cap_mask);
 365         }
 366
 367         ucred->uc_valid = UCRED_NEW;
 368         ucred_set_jobid(info, ucred);
 369         ucred_set_nid(info, ucred);
 370         ucred_set_audit_enabled(info, ucred);
 371         ucred_set_rbac_roles(info, ucred);
 372
 373         EXIT;
 374
 375 out:
 376         if (rc) {
 377                 if (ucred->uc_ginfo) {
 378                         put_group_info(ucred->uc_ginfo);
 379                         ucred->uc_ginfo = NULL;
 380                 }
 381                 if (ucred->uc_identity) {
 382                         mdt_identity_put(mdt->mdt_identity_cache,
 383                                          ucred->uc_identity);
 384                         ucred->uc_identity = NULL;
 385                 }
 386         }
 387
 388         return rc;
 389 }
 390
 391 /**
 392  * Check whether allow the client to set supplementary group IDs or not.
 393  *
 394  * \param[in] info      pointer to the thread context
 395  * \param[in] uc        pointer to the RPC user descriptor
 396  *
 397  * \retval              true if allow to set supplementary group IDs
 398  * \retval              false for other cases
 399  */
 400 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 401 {
 402         __u32 perm;
 403
 404         /* 1. If identity_upcall is disabled,
 405          *    permit local client to do anything. */
 406         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache))
 407                 return true;
 408
 409         /* 2. If fail to get related identities, then forbid any client to
 410          *    set supplementary group IDs. */
 411         if (uc->uc_identity == NULL)
 412                 return false;
 413
 414         /* 3. Check the permission in the identities. */
 415         perm = mdt_identity_get_perm(
 416                 uc->uc_identity,
 417                 &mdt_info_req(info)->rq_peer.nid);
 418         if (perm & CFS_SETGRP_PERM)
 419                 return true;
 420
 421         return false;
 422 }
 423
 424 int mdt_check_ucred(struct mdt_thread_info *info)
 425 {
 426         struct ptlrpc_request   *req = mdt_info_req(info);
 427         struct mdt_device       *mdt = info->mti_mdt;
 428         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 429         struct lu_ucred         *ucred = mdt_ucred(info);
 430         struct md_identity      *identity = NULL;
 431         struct lnet_nid          peernid = req->rq_peer.nid;
 432         __u32                    perm = 0;
 433         int                      setuid;
 434         int                      setgid;
 435         int                      rc = 0;
 436
 437         ENTRY;
 438
 439         LASSERT(ucred != NULL);
 440         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 441                 RETURN(0);
 442
 443         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 444                 RETURN(0);
 445
 446         /* sanity check: if we use strong authentication, we expect the
 447          * uid which client claimed is true */
 448         if (!flvr_is_rootonly(req->rq_flvr.sf_rpc) &&
 449             req->rq_auth_uid != pud->pud_uid) {
 450                 CDEBUG(D_SEC, "local client %s: auth uid %u "
 451                        "while client claims %u:%u/%u:%u\n",
 452                        libcfs_nidstr(&peernid), req->rq_auth_uid,
 453                        pud->pud_uid, pud->pud_gid,
 454                        pud->pud_fsuid, pud->pud_fsgid);
 455                 RETURN(-EACCES);
 456         }
 457
 458         if (is_identity_get_disabled(mdt->mdt_identity_cache))
 459                 RETURN(0);
 460
 461         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 462         if (IS_ERR(identity)) {
 463                 if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 464                         RETURN(0);
 465                 } else {
 466                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 467                                pud->pud_uid);
 468                         RETURN(-EACCES);
 469                 }
 470         }
 471
 472         perm = mdt_identity_get_perm(identity, &peernid);
 473         /* find out the setuid/setgid attempt */
 474         setuid = (pud->pud_uid != pud->pud_fsuid);
 475         setgid = (pud->pud_gid != pud->pud_fsgid ||
 476                   pud->pud_gid != identity->mi_gid);
 477
 478         /* check permission of setuid */
 479         if (setuid && !(perm & CFS_SETUID_PERM)) {
 480                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 481                        pud->pud_uid, pud->pud_fsuid, libcfs_nidstr(&peernid));
 482                 GOTO(out, rc = -EACCES);
 483         }
 484
 485         /* check permission of setgid */
 486         if (setgid && !(perm & CFS_SETGID_PERM)) {
 487                 CDEBUG(D_SEC,
 488                        "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) from %s\n",
 489                        pud->pud_uid, pud->pud_gid,
 490                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 491                        libcfs_nidstr(&peernid));
 492                 GOTO(out, rc = -EACCES);
 493         }
 494
 495         EXIT;
 496
 497 out:
 498         mdt_identity_put(mdt->mdt_identity_cache, identity);
 499         return rc;
 500 }
 501
 502 static int old_init_ucred_common(struct mdt_thread_info *info,
 503                                  struct lu_nodemap *nodemap)
 504 {
 505         struct lu_ucred         *uc = mdt_ucred(info);
 506         struct mdt_device       *mdt = info->mti_mdt;
 507         struct md_identity      *identity = NULL;
 508
 509         if (nodemap && uc->uc_o_uid == nodemap->nm_squash_uid) {
 510                 /* deny access before we get identity ref */
 511                 if (nodemap->nmf_deny_unknown)
 512                         RETURN(-EACCES);
 513
 514                 uc->uc_cap = CAP_EMPTY_SET;
 515                 uc->uc_suppgids[0] = -1;
 516                 uc->uc_suppgids[1] = -1;
 517         }
 518
 519         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 520                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 521                                             uc->uc_fsuid);
 522                 if (IS_ERR(identity)) {
 523                         if (unlikely(PTR_ERR(identity) == -EREMCHG ||
 524                                      cap_raised(uc->uc_cap,
 525                                                 CAP_DAC_READ_SEARCH))) {
 526                                 identity = NULL;
 527                         } else {
 528                                 CDEBUG(D_SEC, "Deny access without identity: "
 529                                        "uid %u\n", uc->uc_fsuid);
 530                                 RETURN(-EACCES);
 531                         }
 532                 }
 533         }
 534         uc->uc_identity = identity;
 535
 536         /* process root_squash here. */
 537         mdt_root_squash(info,
 538                         &mdt_info_req(info)->rq_peer.nid);
 539
 540         if (uc->uc_fsuid) {
 541                 if (!cap_issubset(uc->uc_cap, mdt->mdt_enable_cap_mask))
 542                         CDEBUG(D_SEC, "%s: drop capabilities %llx for NID %s\n",
 543                                mdt_obd_name(mdt),
 544 #ifdef CAP_FOR_EACH_U32
 545                                uc->uc_cap.cap[0] | ((u64)uc->uc_cap.cap[1]<<32),
 546 #else
 547                                uc->uc_cap.val,
 548 #endif
 549                                libcfs_nidstr(&mdt_info_req(info)->rq_peer.nid));
 550                 uc->uc_cap = cap_intersect(uc->uc_cap,mdt->mdt_enable_cap_mask);
 551         }
 552         uc->uc_valid = UCRED_OLD;
 553         ucred_set_jobid(info, uc);
 554         ucred_set_nid(info, uc);
 555         ucred_set_audit_enabled(info, uc);
 556         ucred_set_rbac_roles(info, uc);
 557
 558         EXIT;
 559
 560         return 0;
 561 }
 562
 563 static int old_init_ucred(struct mdt_thread_info *info,
 564                           struct mdt_body *body)
 565 {
 566         struct lu_ucred *uc = mdt_ucred(info);
 567         struct lu_nodemap *nodemap;
 568         int rc;
 569         ENTRY;
 570
 571         nodemap = nodemap_get_from_exp(info->mti_exp);
 572         if (IS_ERR(nodemap))
 573                 RETURN(PTR_ERR(nodemap));
 574
 575         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 576                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 577         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 578                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 579         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 580                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 581         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 582                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 583
 584         LASSERT(uc != NULL);
 585         uc->uc_valid = UCRED_INVALID;
 586         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 587         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 588         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 589         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 590         uc->uc_suppgids[0] = body->mbo_suppgid;
 591         uc->uc_suppgids[1] = -1;
 592         uc->uc_ginfo = NULL;
 593         uc->uc_cap = CAP_EMPTY_SET;
 594         ll_set_capability_u32(&uc->uc_cap, body->mbo_capability);
 595
 596         rc = old_init_ucred_common(info, nodemap);
 597         nodemap_putref(nodemap);
 598
 599         RETURN(rc);
 600 }
 601
 602 static int old_init_ucred_reint(struct mdt_thread_info *info)
 603 {
 604         struct lu_ucred *uc = mdt_ucred(info);
 605         struct lu_nodemap *nodemap;
 606         int rc;
 607         ENTRY;
 608
 609         nodemap = nodemap_get_from_exp(info->mti_exp);
 610         if (IS_ERR(nodemap))
 611                 RETURN(PTR_ERR(nodemap));
 612
 613         LASSERT(uc != NULL);
 614
 615         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 616                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 617         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 618                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 619
 620         uc->uc_valid = UCRED_INVALID;
 621         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 622         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 623         uc->uc_ginfo = NULL;
 624
 625         rc = old_init_ucred_common(info, nodemap);
 626         nodemap_putref(nodemap);
 627
 628         RETURN(rc);
 629 }
 630
 631 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
 632                                    struct mdt_body *body)
 633 {
 634         struct ptlrpc_request   *req = mdt_info_req(info);
 635         struct lu_ucred         *uc  = mdt_ucred(info);
 636
 637         LASSERT(uc != NULL);
 638         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 639                 return 0;
 640
 641         mdt_exit_ucred(info);
 642
 643         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 644                 return old_init_ucred(info, body);
 645         else
 646                 return new_init_ucred(info, BODY_INIT, body);
 647 }
 648
 649 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 650 {
 651         return __mdt_init_ucred(info, body);
 652 }
 653
 654 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 655 {
 656         struct ptlrpc_request *req = mdt_info_req(info);
 657         struct lu_ucred       *uc  = mdt_ucred(info);
 658         struct md_attr        *ma  = &info->mti_attr;
 659
 660         LASSERT(uc != NULL);
 661         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 662                 return 0;
 663
 664         /* LU-5564: for normal close request, skip permission check */
 665         if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE &&
 666             !(ma->ma_attr_flags & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP))) {
 667                 cap_raise_nfsd_set(uc->uc_cap, CAP_FULL_SET);
 668                 cap_raise_fs_set(uc->uc_cap, CAP_FULL_SET);
 669         }
 670
 671         mdt_exit_ucred(info);
 672
 673         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 674                 return old_init_ucred_reint(info);
 675         else
 676                 return new_init_ucred(info, REC_INIT, NULL);
 677 }
 678
 679 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 680 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 681 {
 682         const struct lov_ost_data_v1 *lod;
 683         __u32 lmm_magic = le32_to_cpu(lmm->lmm_magic);
 684         __u16 count;
 685         int i;
 686
 687         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 688                 return;
 689
 690         CDEBUG_LIMIT(level, "objid="DOSTID" magic=0x%08X pattern=%#X\n",
 691                      POSTID(&lmm->lmm_oi), lmm_magic,
 692                      le32_to_cpu(lmm->lmm_pattern));
 693
 694         /* No support for compound layouts yet */
 695         if (lmm_magic != LOV_MAGIC_V1 && lmm_magic != LOV_MAGIC_V3)
 696                 return;
 697
 698         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 699         CDEBUG_LIMIT(level, "stripe_size=0x%x, %sstripe_count=0x%x\n",
 700                      le32_to_cpu(lmm->lmm_stripe_size),
 701                      count > LOV_MAX_STRIPE_COUNT ? "bad " : "", count);
 702
 703         /* If it's a directory or a released file, then there are
 704          * no actual objects to print, so bail out. */
 705         if (valid & OBD_MD_FLDIREA ||
 706             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 707                 return;
 708
 709         if (unlikely(count > LOV_MAX_STRIPE_COUNT))
 710                 return;
 711
 712         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 713                 struct ost_id oi;
 714
 715                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 716                 CDEBUG_LIMIT(level, "stripe %u idx %u subobj "DOSTID"\n",
 717                              i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 718         }
 719 }
 720
 721 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 722 {
 723         const struct lmv_mds_md_v1 *lmm1;
 724         const struct lmv_foreign_md *lfm;
 725         int                        i;
 726
 727         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 728                 return;
 729
 730         /* foreign LMV case */
 731         lfm = &lmv->lmv_foreign_md;
 732         if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
 733                 CDEBUG_LIMIT(level,
 734                              "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
 735                              le32_to_cpu(lfm->lfm_magic),
 736                              le32_to_cpu(lfm->lfm_length),
 737                              le32_to_cpu(lfm->lfm_type),
 738                              le32_to_cpu(lfm->lfm_flags),
 739                              le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
 740                 return;
 741         }
 742
 743         lmm1 = &lmv->lmv_md_v1;
 744         CDEBUG(level,
 745                "magic 0x%08X, master %#X stripe_count %d hash_type %#x\n",
 746                le32_to_cpu(lmm1->lmv_magic),
 747                le32_to_cpu(lmm1->lmv_master_mdt_index),
 748                le32_to_cpu(lmm1->lmv_stripe_count),
 749                le32_to_cpu(lmm1->lmv_hash_type));
 750
 751         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 752                 return;
 753
 754         if (le32_to_cpu(lmm1->lmv_stripe_count) > LMV_MAX_STRIPE_COUNT)
 755                 return;
 756
 757         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 758                 struct lu_fid fid;
 759
 760                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 761                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 762         }
 763 }
 764
 765 /* Shrink and/or grow reply buffers */
 766 int mdt_fix_reply(struct mdt_thread_info *info)
 767 {
 768         struct req_capsule *pill = info->mti_pill;
 769         struct mdt_body    *body;
 770         int                md_size, md_packed = 0;
 771         int                acl_size;
 772         int                rc = 0;
 773         ENTRY;
 774
 775         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 776         LASSERT(body != NULL);
 777
 778         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 779                                OBD_MD_LINKNAME))
 780                 md_size = body->mbo_eadatasize;
 781         else
 782                 md_size = 0;
 783
 784         acl_size = body->mbo_aclsize;
 785
 786         /* this replay - not send info to client */
 787         if (info->mti_spec.no_create) {
 788                 md_size = 0;
 789                 acl_size = 0;
 790         }
 791
 792         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 793                md_size, acl_size);
 794 /*
 795             &RMF_MDT_BODY,
 796             &RMF_MDT_MD,
 797             &RMF_ACL, or &RMF_LOGCOOKIES
 798 (optional)  &RMF_CAPA1,
 799 (optional)  &RMF_CAPA2,
 800 (optional)  something else
 801 */
 802
 803         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 804          * buffers before growing it */
 805         if (info->mti_big_lmm_used) {
 806                 /* big_lmm buffer may be used even without packing the result
 807                  * into reply, just for internal server needs */
 808                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 809                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 810                                                          RCL_SERVER);
 811
 812                 /* free big lmm if md_size is not needed */
 813                 if (md_size == 0 || md_packed == 0) {
 814                         info->mti_big_lmm_used = 0;
 815                 } else {
 816                         /* buffer must be allocated separately */
 817                         LASSERT(info->mti_attr.ma_lmm !=
 818                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 819                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 820                 }
 821         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 822                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 823         }
 824
 825         if (info->mti_big_acl_used) {
 826                 if (acl_size == 0)
 827                         info->mti_big_acl_used = 0;
 828                 else
 829                         req_capsule_shrink(pill, &RMF_ACL, 0, RCL_SERVER);
 830         } else if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER)) {
 831                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 832         } else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
 833                 req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
 834         }
 835
 836         /* Shrink optional SECCTX buffer if it is not used */
 837         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
 838             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
 839             !(body->mbo_valid & OBD_MD_SECCTX))
 840                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
 841
 842         /* Shrink optional ENCCTX buffer if it is not used */
 843         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
 844             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
 845             !(body->mbo_valid & OBD_MD_ENCCTX))
 846                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
 847
 848         /* Shrink optional default LMV buffer if it is not used */
 849         if (req_capsule_has_field(pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER) &&
 850             req_capsule_get_size(pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER) != 0 &&
 851             !(body->mbo_valid & OBD_MD_DEFAULT_MEA))
 852                 req_capsule_shrink(pill, &RMF_DEFAULT_MDT_MD, 0, RCL_SERVER);
 853
 854         /*
 855          * Some more field should be shrinked if needed.
 856          * This should be done by those who added fields to reply message.
 857          */
 858
 859         /* Grow MD buffer if needed finally */
 860         if (info->mti_big_lmm_used) {
 861                 void *lmm;
 862
 863                 LASSERT(md_size > md_packed);
 864                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 865                        md_size - md_packed);
 866
 867                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 868                 if (rc) {
 869                         /* we can't answer with proper LOV EA, drop flags,
 870                          * the rc is also returned so this request is
 871                          * considered as failed */
 872                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 873                         /* don't return transno along with error */
 874                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 875                 } else {
 876                         /* now we need to pack right LOV/LMV EA */
 877                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 878                         if (info->mti_attr.ma_valid & MA_LOV) {
 879                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 880                                                              RCL_SERVER) ==
 881                                                 info->mti_attr.ma_lmm_size);
 882                                 memcpy(lmm, info->mti_attr.ma_lmm,
 883                                        info->mti_attr.ma_lmm_size);
 884                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 885                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 886                                                              RCL_SERVER) ==
 887                                                 info->mti_attr.ma_lmv_size);
 888                                 memcpy(lmm, info->mti_attr.ma_lmv,
 889                                        info->mti_attr.ma_lmv_size);
 890                         }
 891                 }
 892
 893                 /* update mdt_max_mdsize so clients will be aware about that */
 894                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 895                         info->mti_mdt->mdt_max_mdsize =
 896                                                 info->mti_attr.ma_lmm_size;
 897                 info->mti_big_lmm_used = 0;
 898         }
 899
 900         if (info->mti_big_acl_used) {
 901                 CDEBUG(D_INFO, "Enlarge reply ACL buffer to %d bytes\n",
 902                        acl_size);
 903
 904                 rc = req_capsule_server_grow(pill, &RMF_ACL, acl_size);
 905                 if (rc) {
 906                         body->mbo_valid &= ~OBD_MD_FLACL;
 907                 } else {
 908                         void *acl = req_capsule_server_get(pill, &RMF_ACL);
 909
 910                         memcpy(acl, info->mti_big_acl, acl_size);
 911                 }
 912
 913                 info->mti_big_acl_used = 0;
 914         }
 915
 916         RETURN(rc);
 917 }
 918
 919
 920 /* if object is dying, pack the lov/llog data,
 921  * parameter info->mti_attr should be valid at this point!
 922  * Also implements RAoLU policy */
 923 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 924                            struct md_attr *ma)
 925 {
 926         struct mdt_body *repbody = NULL;
 927         const struct lu_attr *la = &ma->ma_attr;
 928         struct coordinator *cdt = &info->mti_mdt->mdt_coordinator;
 929         int rc;
 930         __u64 need = 0;
 931         struct hsm_action_item hai = {
 932                 .hai_len = sizeof(hai),
 933                 .hai_action = HSMA_REMOVE,
 934                 .hai_extent.length = -1,
 935                 .hai_cookie = 0,
 936                 .hai_gid = 0,
 937         };
 938         int archive_id;
 939
 940         ENTRY;
 941
 942         if (mdt_info_req(info) != NULL) {
 943                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 944                 LASSERT(repbody != NULL);
 945         } else {
 946                 CDEBUG(D_INFO, "not running in a request/reply context\n");
 947         }
 948
 949         if ((ma->ma_valid & MA_INODE) && repbody != NULL)
 950                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 951
 952         if (ma->ma_valid & MA_LOV) {
 953                 CERROR("No need in LOV EA upon unlink\n");
 954                 dump_stack();
 955         }
 956         if (repbody != NULL)
 957                 repbody->mbo_eadatasize = 0;
 958
 959         /* Only check unlinked and archived if RAoLU and upon last close */
 960         if (!cdt->cdt_remove_archive_on_last_unlink ||
 961             atomic_read(&mo->mot_open_count) != 0)
 962                 RETURN(0);
 963
 964         /* mdt_attr_get_complex will clear ma_valid, so check here first */
 965         if ((ma->ma_valid & MA_INODE) && (ma->ma_attr.la_nlink != 0))
 966                 RETURN(0);
 967
 968         if ((ma->ma_valid & MA_HSM) && (!(ma->ma_hsm.mh_flags & HS_EXISTS)))
 969                 RETURN(0);
 970
 971         need |= (MA_INODE | MA_HSM) & ~ma->ma_valid;
 972         if (need != 0) {
 973                 /* ma->ma_valid is missing either MA_INODE, MA_HSM, or both,
 974                  * try setting them */
 975                 ma->ma_need |= need;
 976                 rc = mdt_attr_get_complex(info, mo, ma);
 977                 if (rc) {
 978                         CERROR("%s: unable to fetch missing attributes of"
 979                                DFID": rc=%d\n", mdt_obd_name(info->mti_mdt),
 980                                PFID(mdt_object_fid(mo)), rc);
 981                         RETURN(0);
 982                 }
 983
 984                 if (need & MA_INODE) {
 985                         if (ma->ma_valid & MA_INODE) {
 986                                 if (ma->ma_attr.la_nlink != 0)
 987                                         RETURN(0);
 988                         } else {
 989                                 RETURN(0);
 990                         }
 991                 }
 992
 993                 if (need & MA_HSM) {
 994                         if (ma->ma_valid & MA_HSM) {
 995                                 if (!(ma->ma_hsm.mh_flags & HS_EXISTS))
 996                                         RETURN(0);
 997                         } else {
 998                                 RETURN(0);
 999                         }
1000                 }
1001         }
1002
1003         /* RAoLU policy is active, last close on file has occured,
1004          * file is unlinked, file is archived, so create remove request
1005          * for copytool!
1006          * If CDT is not running, requests will be logged for later. */
1007         if (ma->ma_hsm.mh_arch_id != 0)
1008                 archive_id = ma->ma_hsm.mh_arch_id;
1009         else
1010                 archive_id = cdt->cdt_default_archive_id;
1011
1012         hai.hai_fid = *mdt_object_fid(mo);
1013
1014         rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
1015                                   &hai);
1016         if (rc)
1017                 CERROR("%s: unable to add HSM remove request for "DFID
1018                        ": rc=%d\n", mdt_obd_name(info->mti_mdt),
1019                        PFID(mdt_object_fid(mo)), rc);
1020
1021         RETURN(0);
1022 }
1023
1024 static __u64 mdt_attr_valid_xlate(enum mds_attr_flags in,
1025                                   struct mdt_reint_record *rr,
1026                                   struct md_attr *ma)
1027 {
1028         __u64 out;
1029
1030         out = 0;
1031         if (in & MDS_ATTR_MODE)
1032                 out |= LA_MODE;
1033         if (in & MDS_ATTR_UID)
1034                 out |= LA_UID;
1035         if (in & MDS_ATTR_GID)
1036                 out |= LA_GID;
1037         if (in & MDS_ATTR_SIZE)
1038                 out |= LA_SIZE;
1039         if (in & MDS_ATTR_BLOCKS)
1040                 out |= LA_BLOCKS;
1041         if (in & MDS_ATTR_ATIME_SET)
1042                 out |= LA_ATIME;
1043         if (in & MDS_ATTR_CTIME_SET)
1044                 out |= LA_CTIME;
1045         if (in & MDS_ATTR_MTIME_SET)
1046                 out |= LA_MTIME;
1047         if (in & MDS_ATTR_ATTR_FLAG)
1048                 out |= LA_FLAGS;
1049         if (in & MDS_ATTR_KILL_SUID)
1050                 out |= LA_KILL_SUID;
1051         if (in & MDS_ATTR_KILL_SGID)
1052                 out |= LA_KILL_SGID;
1053         if (in & MDS_ATTR_PROJID)
1054                 out |= LA_PROJID;
1055         if (in & MDS_ATTR_LSIZE)
1056                 out |= LA_LSIZE;
1057         if (in & MDS_ATTR_LBLOCKS)
1058                 out |= LA_LBLOCKS;
1059
1060         if (in & MDS_ATTR_FROM_OPEN)
1061                 rr->rr_flags |= MRF_OPEN_TRUNC;
1062         if (in & MDS_ATTR_OVERRIDE)
1063                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
1064         if (in & MDS_ATTR_FORCE)
1065                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1066
1067         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID | MDS_ATTR_PROJID |
1068                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
1069                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
1070                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
1071                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
1072                 MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
1073                 MDS_ATTR_OVERRIDE);
1074         if (in != 0)
1075                 CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", (u64)in);
1076
1077         return out;
1078 }
1079
1080 /* unpacking */
1081
1082 int mdt_name_unpack(struct req_capsule *pill,
1083                     const struct req_msg_field *field,
1084                     struct lu_name *ln,
1085                     enum mdt_name_flags flags)
1086 {
1087         ln->ln_name = req_capsule_client_get(pill, field);
1088         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
1089
1090         if (!lu_name_is_valid(ln)) {
1091                 ln->ln_name = NULL;
1092                 ln->ln_namelen = 0;
1093
1094                 return -EPROTO;
1095         }
1096
1097         if ((flags & MNF_FIX_ANON) &&
1098             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
1099                 /* Newer (3.x) kernels use a name of "/" for the
1100                  * "anonymous" disconnected dentries from NFS
1101                  * filehandle conversion. See d_obtain_alias(). */
1102                 ln->ln_name = NULL;
1103                 ln->ln_namelen = 0;
1104         }
1105
1106         return 0;
1107 }
1108
1109 static int mdt_file_secctx_unpack(struct req_capsule *pill,
1110                                   const char **secctx_name,
1111                                   void **secctx, size_t *secctx_size)
1112 {
1113         const char *name;
1114         size_t name_size;
1115
1116         *secctx_name = NULL;
1117         *secctx = NULL;
1118         *secctx_size = 0;
1119
1120         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT) ||
1121             !req_capsule_field_present(pill, &RMF_FILE_SECCTX_NAME, RCL_CLIENT))
1122                 return 0;
1123
1124         name_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
1125                                          RCL_CLIENT);
1126         if (name_size == 0)
1127                 return 0;
1128
1129         if (name_size > XATTR_NAME_MAX + 1)
1130                 return -EPROTO;
1131
1132         name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1133         if (strnlen(name, name_size) != name_size - 1)
1134                 return -EPROTO;
1135
1136         if (!req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_CLIENT) ||
1137             !req_capsule_field_present(pill, &RMF_FILE_SECCTX, RCL_CLIENT))
1138                 return -EPROTO;
1139
1140         *secctx_name = name;
1141         *secctx = req_capsule_client_get(pill, &RMF_FILE_SECCTX);
1142         *secctx_size = req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_CLIENT);
1143
1144         return 0;
1145 }
1146
1147 static int mdt_file_encctx_unpack(struct req_capsule *pill,
1148                                   void **encctx, size_t *encctx_size)
1149 {
1150         *encctx = NULL;
1151         *encctx_size = 0;
1152
1153         if (!exp_connect_encrypt(pill->rc_req->rq_export))
1154                 return 0;
1155
1156         if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
1157             !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
1158                 return -EPROTO;
1159
1160         *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
1161         if (*encctx_size == 0)
1162                 return 0;
1163
1164         *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
1165
1166         return 0;
1167 }
1168
1169 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
1170 {
1171         struct lu_ucred *uc = mdt_ucred(info);
1172         struct md_attr *ma = &info->mti_attr;
1173         struct lu_attr *la = &ma->ma_attr;
1174         struct req_capsule *pill = info->mti_pill;
1175         struct mdt_reint_record *rr = &info->mti_rr;
1176         struct mdt_rec_setattr *rec;
1177         struct lu_nodemap *nodemap;
1178
1179         ENTRY;
1180
1181         BUILD_BUG_ON(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->sa_fsuid;
1188         uc->uc_fsgid = rec->sa_fsgid;
1189         uc->uc_cap = CAP_EMPTY_SET;
1190         ll_set_capability_u32(&uc->uc_cap, rec->sa_cap);
1191         uc->uc_suppgids[0] = rec->sa_suppgid;
1192         uc->uc_suppgids[1] = -1;
1193
1194         rr->rr_fid1 = &rec->sa_fid;
1195         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
1196         la->la_mode  = rec->sa_mode;
1197         la->la_flags = rec->sa_attr_flags;
1198
1199         nodemap = nodemap_get_from_exp(info->mti_exp);
1200         if (IS_ERR(nodemap))
1201                 RETURN(PTR_ERR(nodemap));
1202
1203         la->la_uid = nodemap_map_id(nodemap, NODEMAP_UID,
1204                                     NODEMAP_CLIENT_TO_FS, rec->sa_uid);
1205         la->la_gid = nodemap_map_id(nodemap, NODEMAP_GID,
1206                                     NODEMAP_CLIENT_TO_FS, rec->sa_gid);
1207         la->la_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
1208                                        NODEMAP_CLIENT_TO_FS, rec->sa_projid);
1209         nodemap_putref(nodemap);
1210
1211         la->la_size  = rec->sa_size;
1212         la->la_blocks = rec->sa_blocks;
1213         la->la_ctime = rec->sa_ctime;
1214         la->la_atime = rec->sa_atime;
1215         la->la_mtime = rec->sa_mtime;
1216         ma->ma_valid = MA_INODE;
1217
1218         ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
1219                                 MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
1220                                 MDS_PCC_ATTACH);
1221         RETURN(0);
1222 }
1223
1224 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
1225 {
1226         struct req_capsule *pill = info->mti_pill;
1227         struct mdt_ioepoch *ioepoch;
1228         ENTRY;
1229
1230         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
1231                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
1232         else
1233                 ioepoch = NULL;
1234
1235         if (ioepoch == NULL)
1236                 RETURN(-EPROTO);
1237
1238         info->mti_open_handle = ioepoch->mio_open_handle;
1239
1240         RETURN(0);
1241 }
1242
1243 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
1244         struct req_capsule      *pill = info->mti_pill;
1245
1246         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
1247                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
1248                 if (info->mti_dlm_req == NULL)
1249                         RETURN(-EFAULT);
1250         }
1251
1252         RETURN(0);
1253 }
1254
1255 static int mdt_setattr_unpack(struct mdt_thread_info *info)
1256 {
1257         struct mdt_reint_record *rr = &info->mti_rr;
1258         struct md_attr          *ma = &info->mti_attr;
1259         struct req_capsule      *pill = info->mti_pill;
1260         int rc;
1261         ENTRY;
1262
1263         rc = mdt_setattr_unpack_rec(info);
1264         if (rc)
1265                 RETURN(rc);
1266
1267         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1268                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
1269                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1270                                                         RCL_CLIENT);
1271
1272                 if (rr->rr_eadatalen > 0) {
1273                         const struct lmv_user_md        *lum;
1274
1275                         lum = rr->rr_eadata;
1276                         /* Sigh ma_valid(from req) does not indicate whether
1277                          * it will set LOV/LMV EA, so we have to check magic */
1278                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1279                                 ma->ma_valid |= MA_LMV;
1280                                 ma->ma_lmv = (void *)rr->rr_eadata;
1281                                 ma->ma_lmv_size = rr->rr_eadatalen;
1282                         } else {
1283                                 ma->ma_valid |= MA_LOV;
1284                                 ma->ma_lmm = (void *)rr->rr_eadata;
1285                                 ma->ma_lmm_size = rr->rr_eadatalen;
1286                         }
1287                 }
1288         }
1289
1290         rc = mdt_dlmreq_unpack(info);
1291         RETURN(rc);
1292 }
1293
1294 static int mdt_close_intent_unpack(struct mdt_thread_info *info)
1295 {
1296         struct md_attr          *ma = &info->mti_attr;
1297         struct req_capsule      *pill = info->mti_pill;
1298         ENTRY;
1299
1300         if (!(ma->ma_attr_flags & MDS_CLOSE_INTENT))
1301                 RETURN(0);
1302
1303         req_capsule_extend(pill, &RQF_MDS_CLOSE_INTENT);
1304
1305         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1306             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1307                 RETURN(-EFAULT);
1308
1309         RETURN(0);
1310 }
1311
1312 int mdt_close_unpack(struct mdt_thread_info *info)
1313 {
1314         int rc;
1315         ENTRY;
1316
1317         rc = mdt_close_handle_unpack(info);
1318         if (rc)
1319                 RETURN(rc);
1320
1321         rc = mdt_setattr_unpack_rec(info);
1322         if (rc)
1323                 RETURN(rc);
1324
1325         rc = mdt_close_intent_unpack(info);
1326         if (rc)
1327                 RETURN(rc);
1328
1329         RETURN(mdt_init_ucred_reint(info));
1330 }
1331
1332 static int mdt_create_unpack(struct mdt_thread_info *info)
1333 {
1334         struct lu_ucred *uc  = mdt_ucred(info);
1335         struct mdt_rec_create *rec;
1336         struct lu_attr *attr = &info->mti_attr.ma_attr;
1337         struct mdt_reint_record *rr = &info->mti_rr;
1338         struct req_capsule      *pill = info->mti_pill;
1339         struct md_op_spec       *sp = &info->mti_spec;
1340         int rc;
1341
1342         ENTRY;
1343
1344         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1345         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1346         if (rec == NULL)
1347                 RETURN(-EFAULT);
1348
1349         /* This prior initialization is needed for old_init_ucred_reint() */
1350         uc->uc_fsuid = rec->cr_fsuid;
1351         uc->uc_fsgid = rec->cr_fsgid;
1352         uc->uc_cap = CAP_EMPTY_SET;
1353         ll_set_capability_u32(&uc->uc_cap, rec->cr_cap);
1354         uc->uc_suppgids[0] = rec->cr_suppgid1;
1355         uc->uc_suppgids[1] = -1;
1356         uc->uc_umask = rec->cr_umask;
1357
1358         rr->rr_fid1 = &rec->cr_fid1;
1359         rr->rr_fid2 = &rec->cr_fid2;
1360         attr->la_mode = rec->cr_mode;
1361         attr->la_rdev  = rec->cr_rdev;
1362         attr->la_uid   = rec->cr_fsuid;
1363         attr->la_gid   = rec->cr_fsgid;
1364         attr->la_ctime = rec->cr_time;
1365         attr->la_mtime = rec->cr_time;
1366         attr->la_atime = rec->cr_time;
1367         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1368                 LA_CTIME | LA_MTIME | LA_ATIME;
1369         memset(&sp->u, 0, sizeof(sp->u));
1370         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1371
1372         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1373         if (rc < 0)
1374                 RETURN(rc);
1375
1376         if (S_ISLNK(attr->la_mode)) {
1377                 const char *tgt = NULL;
1378                 int sz;
1379
1380                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1381                 sz = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT);
1382                 if (sz) {
1383                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1384                         sp->u.sp_symname.ln_name = tgt;
1385                         sp->u.sp_symname.ln_namelen = sz - 1; /* skip NUL */
1386                 }
1387                 if (tgt == NULL)
1388                         RETURN(-EFAULT);
1389         } else {
1390                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_ACL);
1391                 if (S_ISDIR(attr->la_mode)) {
1392                         struct obd_export *exp = mdt_info_req(info)->rq_export;
1393
1394                         sp->sp_dmv_imp_inherit =
1395                                 info->mti_mdt->mdt_enable_dmv_implicit_inherit;
1396                         if (req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT)
1397                             > 0) {
1398                                 sp->u.sp_ea.eadata =
1399                                         req_capsule_client_get(pill,
1400                                                                &RMF_EADATA);
1401                                 sp->u.sp_ea.eadatalen =
1402                                         req_capsule_get_size(pill, &RMF_EADATA,
1403                                                              RCL_CLIENT);
1404                                 sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1405                         }
1406                         if (OCD_HAS_FLAG2(&exp->exp_connect_data,
1407                                           DMV_IMP_INHERIT)) {
1408                                 if ((sp->sp_cr_flags & MDS_OPEN_DEFAULT_LMV) &&
1409                                     !(sp->sp_cr_flags & MDS_OPEN_HAS_EA))
1410                                         RETURN(-EPROTO);
1411                         } else if (sp->sp_cr_flags & MDS_OPEN_DEFAULT_LMV) {
1412                                 RETURN(-EPROTO);
1413                         }
1414                 }
1415         }
1416
1417         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1418                                     &sp->sp_cr_file_secctx,
1419                                     &sp->sp_cr_file_secctx_size);
1420         if (rc < 0)
1421                 RETURN(rc);
1422
1423         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1424                                     &sp->sp_cr_file_encctx_size);
1425         if (rc < 0)
1426                 RETURN(rc);
1427
1428         rc = req_check_sepol(pill);
1429         if (rc)
1430                 RETURN(rc);
1431
1432         rc = mdt_dlmreq_unpack(info);
1433         RETURN(rc);
1434 }
1435
1436 static int mdt_link_unpack(struct mdt_thread_info *info)
1437 {
1438         struct lu_ucred *uc  = mdt_ucred(info);
1439         struct mdt_rec_link *rec;
1440         struct lu_attr *attr = &info->mti_attr.ma_attr;
1441         struct mdt_reint_record *rr = &info->mti_rr;
1442         struct req_capsule *pill = info->mti_pill;
1443         int rc;
1444
1445         ENTRY;
1446
1447         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1448         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1449         if (rec == NULL)
1450                 RETURN(-EFAULT);
1451
1452         /* This prior initialization is needed for old_init_ucred_reint() */
1453         uc->uc_fsuid = rec->lk_fsuid;
1454         uc->uc_fsgid = rec->lk_fsgid;
1455         uc->uc_cap = CAP_EMPTY_SET;
1456         ll_set_capability_u32(&uc->uc_cap, rec->lk_cap);
1457         uc->uc_suppgids[0] = rec->lk_suppgid1;
1458         uc->uc_suppgids[1] = rec->lk_suppgid2;
1459
1460         attr->la_uid = rec->lk_fsuid;
1461         attr->la_gid = rec->lk_fsgid;
1462         rr->rr_fid1 = &rec->lk_fid1;
1463         rr->rr_fid2 = &rec->lk_fid2;
1464         attr->la_ctime = rec->lk_time;
1465         attr->la_mtime = rec->lk_time;
1466         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1467
1468         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1469         if (rc < 0)
1470                 RETURN(rc);
1471
1472         rc = req_check_sepol(pill);
1473         if (rc)
1474                 RETURN(rc);
1475
1476         rc = mdt_dlmreq_unpack(info);
1477
1478         RETURN(rc);
1479 }
1480
1481 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1482 {
1483         struct lu_ucred *uc  = mdt_ucred(info);
1484         struct mdt_rec_unlink *rec;
1485         struct lu_attr *attr = &info->mti_attr.ma_attr;
1486         struct mdt_reint_record *rr = &info->mti_rr;
1487         struct req_capsule      *pill = info->mti_pill;
1488         int rc;
1489
1490         ENTRY;
1491
1492         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1493         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1494         if (rec == NULL)
1495                 RETURN(-EFAULT);
1496
1497         /* This prior initialization is needed for old_init_ucred_reint() */
1498         uc->uc_fsuid = rec->ul_fsuid;
1499         uc->uc_fsgid = rec->ul_fsgid;
1500         uc->uc_cap = CAP_EMPTY_SET;
1501         ll_set_capability_u32(&uc->uc_cap, rec->ul_cap);
1502         uc->uc_suppgids[0] = rec->ul_suppgid1;
1503         uc->uc_suppgids[1] = -1;
1504
1505         attr->la_uid = rec->ul_fsuid;
1506         attr->la_gid = rec->ul_fsgid;
1507         rr->rr_fid1 = &rec->ul_fid1;
1508         rr->rr_fid2 = &rec->ul_fid2;
1509         attr->la_ctime = rec->ul_time;
1510         attr->la_mtime = rec->ul_time;
1511         attr->la_mode  = rec->ul_mode;
1512         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1513         if (rec->ul_bias & MDS_FID_OP)
1514                 info->mti_spec.sp_cr_flags |= MDS_OP_WITH_FID;
1515         else
1516                 info->mti_spec.sp_cr_flags &= ~MDS_OP_WITH_FID;
1517
1518         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1519         if (rc < 0)
1520                 RETURN(rc);
1521
1522         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1523
1524         rc = req_check_sepol(pill);
1525         if (rc)
1526                 RETURN(rc);
1527
1528         rc = mdt_dlmreq_unpack(info);
1529         RETURN(rc);
1530 }
1531
1532 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1533 {
1534         info->mti_spec.sp_rm_entry = 1;
1535         return mdt_unlink_unpack(info);
1536 }
1537
1538 static int mdt_rename_unpack(struct mdt_thread_info *info)
1539 {
1540         struct lu_ucred *uc = mdt_ucred(info);
1541         struct mdt_rec_rename *rec;
1542         struct lu_attr *attr = &info->mti_attr.ma_attr;
1543         struct mdt_reint_record *rr = &info->mti_rr;
1544         struct req_capsule *pill = info->mti_pill;
1545         struct md_op_spec *spec = &info->mti_spec;
1546         int rc;
1547
1548         ENTRY;
1549
1550         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1551         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1552         if (rec == NULL)
1553                 RETURN(-EFAULT);
1554
1555         /* This prior initialization is needed for old_init_ucred_reint() */
1556         uc->uc_fsuid = rec->rn_fsuid;
1557         uc->uc_fsgid = rec->rn_fsgid;
1558         uc->uc_cap = CAP_EMPTY_SET;
1559         ll_set_capability_u32(&uc->uc_cap, rec->rn_cap);
1560         uc->uc_suppgids[0] = rec->rn_suppgid1;
1561         uc->uc_suppgids[1] = rec->rn_suppgid2;
1562
1563         attr->la_uid = rec->rn_fsuid;
1564         attr->la_gid = rec->rn_fsgid;
1565         rr->rr_fid1 = &rec->rn_fid1;
1566         rr->rr_fid2 = &rec->rn_fid2;
1567         attr->la_ctime = rec->rn_time;
1568         attr->la_mtime = rec->rn_time;
1569         /* rename_tgt contains the mode already */
1570         attr->la_mode = rec->rn_mode;
1571         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1572
1573         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1574         if (rc < 0)
1575                 RETURN(rc);
1576
1577         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1578         if (rc < 0)
1579                 RETURN(rc);
1580
1581         spec->no_create = !!req_is_replay(mdt_info_req(info));
1582
1583         rc = req_check_sepol(pill);
1584         if (rc)
1585                 RETURN(rc);
1586
1587         rc = mdt_dlmreq_unpack(info);
1588
1589         RETURN(rc);
1590 }
1591
1592 static int mdt_migrate_unpack(struct mdt_thread_info *info)
1593 {
1594         struct lu_ucred *uc = mdt_ucred(info);
1595         struct mdt_rec_rename *rec;
1596         struct lu_attr *attr = &info->mti_attr.ma_attr;
1597         struct mdt_reint_record *rr = &info->mti_rr;
1598         struct req_capsule *pill = info->mti_pill;
1599         struct md_op_spec *spec = &info->mti_spec;
1600         int rc;
1601
1602         ENTRY;
1603
1604         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1605         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1606         if (rec == NULL)
1607                 RETURN(-EFAULT);
1608
1609         /* This prior initialization is needed for old_init_ucred_reint() */
1610         uc->uc_fsuid = rec->rn_fsuid;
1611         uc->uc_fsgid = rec->rn_fsgid;
1612         uc->uc_cap = CAP_EMPTY_SET;
1613         ll_set_capability_u32(&uc->uc_cap, rec->rn_cap);
1614         uc->uc_suppgids[0] = rec->rn_suppgid1;
1615         uc->uc_suppgids[1] = rec->rn_suppgid2;
1616
1617         attr->la_uid = rec->rn_fsuid;
1618         attr->la_gid = rec->rn_fsgid;
1619         rr->rr_fid1 = &rec->rn_fid1;
1620         rr->rr_fid2 = &rec->rn_fid2;
1621         attr->la_ctime = rec->rn_time;
1622         attr->la_mtime = rec->rn_time;
1623         /* rename_tgt contains the mode already */
1624         attr->la_mode = rec->rn_mode;
1625         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1626         spec->sp_cr_flags = 0;
1627
1628         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1629         if (rc < 0)
1630                 RETURN(rc);
1631
1632         if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
1633                 rc = mdt_close_handle_unpack(info);
1634                 if (rc)
1635                         RETURN(rc);
1636
1637                 spec->sp_migrate_close = 1;
1638         } else {
1639                 spec->sp_migrate_close = 0;
1640         }
1641
1642         spec->sp_migrate_nsonly = !!(rec->rn_bias & MDS_MIGRATE_NSONLY);
1643
1644         /* lustre version > 2.11 migration packs lum */
1645         if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
1646                 if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1647                         rr->rr_eadatalen = req_capsule_get_size(pill,
1648                                                                 &RMF_EADATA,
1649                                                                 RCL_CLIENT);
1650
1651                         if (rr->rr_eadatalen > 0) {
1652                                 struct lmv_user_md_v1 *lmu;
1653
1654                                 lmu = req_capsule_client_get(pill, &RMF_EADATA);
1655                                 lmu->lum_hash_type |=
1656                                         cpu_to_le32(LMV_HASH_FLAG_FIXED);
1657                                 rr->rr_eadata = lmu;
1658                                 spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
1659                                 spec->u.sp_ea.eadata = rr->rr_eadata;
1660                                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1661                         }
1662                 } else {
1663                         /* old client doesn't provide lum. */
1664                         RETURN(-EOPNOTSUPP);
1665                 }
1666         }
1667
1668         spec->no_create = !!req_is_replay(mdt_info_req(info));
1669
1670         rc = mdt_dlmreq_unpack(info);
1671
1672         RETURN(rc);
1673 }
1674
1675 /*
1676  * please see comment above LOV_MAGIC_V1_DEFINED
1677  */
1678 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata)
1679 {
1680         struct lov_user_md_v1   *v1 = eadata;
1681
1682         LASSERT(v1);
1683
1684         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1685                 if ((v1->lmm_magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
1686                         v1->lmm_magic |= LOV_MAGIC_DEFINED;
1687                 else if ((v1->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
1688                          __swab32(LOV_MAGIC_MAGIC))
1689                         v1->lmm_magic |= __swab32(LOV_MAGIC_DEFINED);
1690         }
1691 }
1692
1693 static int mdt_open_unpack(struct mdt_thread_info *info)
1694 {
1695         struct lu_ucred *uc = mdt_ucred(info);
1696         struct mdt_rec_create *rec;
1697         struct lu_attr *attr = &info->mti_attr.ma_attr;
1698         struct req_capsule *pill = info->mti_pill;
1699         struct mdt_reint_record *rr = &info->mti_rr;
1700         struct ptlrpc_request *req = mdt_info_req(info);
1701         struct md_op_spec *sp = &info->mti_spec;
1702         int rc;
1703         ENTRY;
1704
1705         BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
1706                      sizeof(struct mdt_rec_reint));
1707         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1708         if (rec == NULL)
1709                 RETURN(-EFAULT);
1710
1711         /* This prior initialization is needed for old_init_ucred_reint() */
1712         uc->uc_fsuid = rec->cr_fsuid;
1713         uc->uc_fsgid = rec->cr_fsgid;
1714         uc->uc_cap = CAP_EMPTY_SET;
1715         ll_set_capability_u32(&uc->uc_cap, rec->cr_cap);
1716         uc->uc_suppgids[0] = rec->cr_suppgid1;
1717         uc->uc_suppgids[1] = rec->cr_suppgid2;
1718         uc->uc_umask = rec->cr_umask;
1719
1720         rr->rr_fid1   = &rec->cr_fid1;
1721         rr->rr_fid2   = &rec->cr_fid2;
1722         rr->rr_open_handle = &rec->cr_open_handle_old;
1723         attr->la_mode = rec->cr_mode;
1724         attr->la_rdev  = rec->cr_rdev;
1725         attr->la_uid   = rec->cr_fsuid;
1726         attr->la_gid   = rec->cr_fsgid;
1727         attr->la_ctime = rec->cr_time;
1728         attr->la_mtime = rec->cr_time;
1729         attr->la_atime = rec->cr_time;
1730         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1731                          LA_CTIME | LA_MTIME | LA_ATIME;
1732         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1733         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1734         /* Do not trigger ASSERTION if client miss to set such flags. */
1735         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1736                 RETURN(-EPROTO);
1737
1738         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1739
1740         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1741
1742         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1743                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1744                                                         RCL_CLIENT);
1745
1746                 if (rr->rr_eadatalen > 0) {
1747                         rr->rr_eadata = req_capsule_client_get(pill,
1748                                                                &RMF_EADATA);
1749                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1750                         sp->u.sp_ea.eadata = rr->rr_eadata;
1751                         sp->sp_archive_id = rec->cr_archive_id;
1752                         sp->no_create = !!req_is_replay(req);
1753                         mdt_fix_lov_magic(info, rr->rr_eadata);
1754                 }
1755
1756                 /*
1757                  * Client default md_size may be 0 right after client start,
1758                  * until all osc are connected, set here just some reasonable
1759                  * value to prevent misbehavior.
1760                  */
1761                 if (rr->rr_eadatalen == 0 &&
1762                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1763                         rr->rr_eadatalen = MIN_MD_SIZE;
1764         }
1765
1766         rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
1767                                     &sp->sp_cr_file_secctx,
1768                                     &sp->sp_cr_file_secctx_size);
1769         if (rc < 0)
1770                 RETURN(rc);
1771
1772         rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
1773                                     &sp->sp_cr_file_encctx_size);
1774         if (rc < 0)
1775                 RETURN(rc);
1776
1777         rc = req_check_sepol(pill);
1778         if (rc)
1779                 RETURN(rc);
1780
1781         RETURN(rc);
1782 }
1783
1784 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1785 {
1786         struct mdt_reint_record *rr = &info->mti_rr;
1787         struct lu_ucred *uc = mdt_ucred(info);
1788         struct lu_attr *attr = &info->mti_attr.ma_attr;
1789         struct req_capsule *pill = info->mti_pill;
1790         struct mdt_rec_setxattr *rec;
1791         int rc;
1792         ENTRY;
1793
1794
1795         BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
1796                      sizeof(struct mdt_rec_reint));
1797         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1798         if (rec == NULL)
1799                 RETURN(-EFAULT);
1800
1801         /* This prior initialization is needed for old_init_ucred_reint() */
1802         uc->uc_fsuid  = rec->sx_fsuid;
1803         uc->uc_fsgid  = rec->sx_fsgid;
1804         uc->uc_cap = CAP_EMPTY_SET;
1805         ll_set_capability_u32(&uc->uc_cap, rec->sx_cap);
1806         uc->uc_suppgids[0] = rec->sx_suppgid1;
1807         uc->uc_suppgids[1] = -1;
1808
1809         rr->rr_opcode = rec->sx_opcode;
1810         rr->rr_fid1   = &rec->sx_fid;
1811         attr->la_valid = rec->sx_valid;
1812         attr->la_ctime = rec->sx_time;
1813         attr->la_size = rec->sx_size;
1814         attr->la_flags = rec->sx_flags;
1815
1816         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1817         if (rc < 0)
1818                 RETURN(rc);
1819
1820         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1821                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1822                                                         RCL_CLIENT);
1823
1824                 if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
1825                         RETURN(-E2BIG);
1826
1827                 if (rr->rr_eadatalen > 0) {
1828                         rr->rr_eadata = req_capsule_client_get(pill,
1829                                                                &RMF_EADATA);
1830                         if (rr->rr_eadata == NULL)
1831                                 RETURN(-EFAULT);
1832                 } else {
1833                         rr->rr_eadata = NULL;
1834                 }
1835         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1836                 CDEBUG(D_INFO, "no xattr data supplied\n");
1837                 RETURN(-EFAULT);
1838         }
1839
1840         rc = req_check_sepol(pill);
1841         if (rc)
1842                 RETURN(rc);
1843
1844         if (mdt_dlmreq_unpack(info) < 0)
1845                 RETURN(-EPROTO);
1846
1847         RETURN(0);
1848 }
1849
1850 static int mdt_resync_unpack(struct mdt_thread_info *info)
1851 {
1852         struct req_capsule      *pill = info->mti_pill;
1853         struct mdt_reint_record *rr   = &info->mti_rr;
1854         struct lu_ucred         *uc     = mdt_ucred(info);
1855         struct mdt_rec_resync   *rec;
1856         ENTRY;
1857
1858         BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
1859         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1860         if (rec == NULL)
1861                 RETURN(-EFAULT);
1862
1863         /* This prior initialization is needed for old_init_ucred_reint() */
1864         uc->uc_fsuid = rec->rs_fsuid;
1865         uc->uc_fsgid = rec->rs_fsgid;
1866         uc->uc_cap = CAP_EMPTY_SET;
1867         ll_set_capability_u32(&uc->uc_cap, rec->rs_cap);
1868
1869         rr->rr_fid1   = &rec->rs_fid;
1870         rr->rr_mirror_id = rec->rs_mirror_id;
1871
1872         /* cookie doesn't need to be swapped but it has been swapped
1873          * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
1874          * restoring. */
1875         if (req_capsule_req_need_swab(pill))
1876                 __swab64s(&rec->rs_lease_handle.cookie);
1877         rr->rr_lease_handle = &rec->rs_lease_handle;
1878
1879         RETURN(mdt_dlmreq_unpack(info));
1880 }
1881
1882 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1883
1884 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1885         [REINT_SETATTR]  = mdt_setattr_unpack,
1886         [REINT_CREATE]   = mdt_create_unpack,
1887         [REINT_LINK]     = mdt_link_unpack,
1888         [REINT_UNLINK]   = mdt_unlink_unpack,
1889         [REINT_RENAME]   = mdt_rename_unpack,
1890         [REINT_OPEN]     = mdt_open_unpack,
1891         [REINT_SETXATTR] = mdt_setxattr_unpack,
1892         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1893         [REINT_MIGRATE]  = mdt_migrate_unpack,
1894         [REINT_RESYNC]   = mdt_resync_unpack,
1895 };
1896
1897 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1898 {
1899         int rc;
1900         ENTRY;
1901
1902         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1903         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1904                 info->mti_rr.rr_opcode = op;
1905                 rc = mdt_reint_unpackers[op](info);
1906         } else {
1907                 CERROR("Unexpected opcode %d\n", op);
1908                 rc = -EFAULT;
1909         }
1910         RETURN(rc);
1911 }
1912
1913 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1914                              struct mdt_object *child)
1915 {
1916         char *secctx_name;
1917         struct lu_buf *buffer;
1918         struct mdt_body *repbody;
1919         struct req_capsule *pill = info->mti_pill;
1920         int rc = 0;
1921
1922         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
1923             req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
1924                 secctx_name =
1925                         req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
1926                 buffer = &info->mti_buf;
1927
1928                 /* fill reply buffer with security context now */
1929                 buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
1930                                                       RCL_SERVER);
1931                 buffer->lb_buf = req_capsule_server_get(info->mti_pill,
1932                                                         &RMF_FILE_SECCTX);
1933                 rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
1934                                   buffer, secctx_name);
1935                 if (rc >= 0) {
1936                         CDEBUG(D_SEC,
1937                                "found security context of size %d for "DFID"\n",
1938                                rc, PFID(mdt_object_fid(child)));
1939
1940                         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1941                         repbody->mbo_valid |= OBD_MD_SECCTX;
1942                         if (rc < buffer->lb_len)
1943                                 req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
1944                                                    RCL_SERVER);
1945                         rc = 0;
1946                 } else {
1947                         CDEBUG(D_SEC,
1948                              "security context not found for "DFID": rc = %d\n",
1949                              PFID(mdt_object_fid(child)), rc);
1950                         req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
1951                                            RCL_SERVER);
1952                         /* handling -ENOENT is important because it may change
1953                          * object state in DNE env dropping LOHA_EXISTS flag,
1954                          * it is important to return that to the caller.
1955                          * Check LU-13115 for details.
1956                          */
1957                         if (rc != -ENOENT)
1958                                 rc = 0;
1959                 }
1960         }
1961         return rc;
1962 }
1963
1964 /* check whether two FIDs belong to different MDTs.
1965  * \retval      1 on different MDTs.
1966  *              0 on the same MDT.
1967  *              -ve on error
1968  */
1969 int mdt_fids_different_target(struct mdt_thread_info *info,
1970                               const struct lu_fid *fid1,
1971                               const struct lu_fid *fid2)
1972 {
1973         const struct lu_env *env = info->mti_env;
1974         struct mdt_device *mdt = info->mti_mdt;
1975         struct lu_seq_range *range = &info->mti_range;
1976         struct seq_server_site *ss;
1977         __u32 index1, index2;
1978         int rc;
1979
1980         if (fid_seq(fid1) == fid_seq(fid2))
1981                 return 0;
1982
1983         ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
1984
1985         range->lsr_flags = LU_SEQ_RANGE_MDT;
1986         rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
1987         if (rc)
1988                 return rc;
1989
1990         index1 = range->lsr_index;
1991
1992         rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
1993         if (rc)
1994                 return rc;
1995
1996         index2 = range->lsr_index;
1997
1998         return index1 != index2;
1999 }
2000
2001 /**
2002  * Check whether \a child is remote object on \a parent.
2003  *
2004  * \param[in]  info     thread environment
2005  * \param[in]  parent   parent object, it's the same as child object in
2006  *                      getattr_by_fid
2007  * \param[in]  child    child object
2008  *
2009  * \retval 1    is remote object.
2010  * \retval 0    isn't remote object.
2011  * \retval < 1  error code
2012  */
2013 int mdt_is_remote_object(struct mdt_thread_info *info,
2014                          struct mdt_object *parent,
2015                          struct mdt_object *child)
2016 {
2017         struct lu_buf *buf = &info->mti_big_buf;
2018         struct linkea_data ldata = { NULL };
2019         struct link_ea_header *leh;
2020         struct link_ea_entry *lee;
2021         struct lu_name name;
2022         struct lu_fid pfid;
2023         int reclen;
2024         int i;
2025         int rc;
2026
2027         ENTRY;
2028
2029         if (fid_is_root(mdt_object_fid(child)))
2030                 RETURN(0);
2031
2032         if (likely(parent != child)) {
2033                 if (mdt_object_remote(parent) ^ mdt_object_remote(child))
2034                         RETURN(1);
2035
2036                 if (!mdt_object_remote(parent) && !mdt_object_remote(child))
2037                         RETURN(0);
2038
2039                 rc = mdt_fids_different_target(info, mdt_object_fid(parent),
2040                                                mdt_object_fid(child));
2041                 RETURN(rc);
2042         }
2043
2044         /* client < 2.13.52 getattr_by_fid parent and child are the same */
2045         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2046         if (!buf->lb_buf)
2047                 RETURN(-ENOMEM);
2048
2049         ldata.ld_buf = buf;
2050         rc = mdt_links_read(info, child, &ldata);
2051         /* can't read linkea, just assume it's remote object */
2052         if (rc == -ENOENT || rc == -ENODATA)
2053                 RETURN(1);
2054         if (rc)
2055                 RETURN(rc);
2056
2057         leh = buf->lb_buf;
2058         lee = (struct link_ea_entry *)(leh + 1);
2059         for (i = 0; i < leh->leh_reccount; i++) {
2060                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
2061                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2062                 rc = mdt_fids_different_target(info, &pfid,
2063                                                mdt_object_fid(child));
2064                 if (rc)
2065                         RETURN(rc);
2066         }
2067
2068         RETURN(0);
2069 }
2070
2071 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
2072                              struct mdt_object *child)
2073 {
2074         struct lu_buf *buffer;
2075         struct mdt_body *repbody;
2076         struct req_capsule *pill = info->mti_pill;
2077         struct obd_export *exp = mdt_info_req(info)->rq_export;
2078         int rc = 0;
2079
2080         if (!exp_connect_encrypt(exp))
2081                 return rc;
2082
2083         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
2084             req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
2085                 struct lu_attr la = { 0 };
2086                 struct dt_object *dt = mdt_obj2dt(child);
2087
2088                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
2089                         dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
2090
2091                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
2092                         buffer = &info->mti_buf;
2093
2094                         /* fill reply buffer with encryption context now */
2095                         buffer->lb_len =
2096                                 req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
2097                                                      RCL_SERVER);
2098                         buffer->lb_buf =
2099                                 req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
2100                         rc = mo_xattr_get(info->mti_env,
2101                                           mdt_object_child(child),
2102                                           buffer,
2103                                           LL_XATTR_NAME_ENCRYPTION_CONTEXT);
2104                         if (unlikely(rc == -ENODATA))
2105                                 /* For compatibility with 2.14 */
2106                                 rc = mo_xattr_get(info->mti_env,
2107                                           mdt_object_child(child),
2108                                           buffer,
2109                                           LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD);
2110                         if (rc >= 0) {
2111                                 CDEBUG(D_SEC,
2112                                        "found encryption ctx of size %d for "DFID"\n",
2113                                        rc, PFID(mdt_object_fid(child)));
2114
2115                                 repbody = req_capsule_server_get(pill,
2116                                                                  &RMF_MDT_BODY);
2117                                 repbody->mbo_valid |= OBD_MD_ENCCTX;
2118                                 if (rc < buffer->lb_len)
2119                                         req_capsule_shrink(pill,
2120                                                            &RMF_FILE_ENCCTX, rc,
2121                                                            RCL_SERVER);
2122                                 rc = 0;
2123                         } else {
2124                                 CDEBUG(D_SEC,
2125                                        "encryption ctx not found for "DFID": rc = %d\n",
2126                                        PFID(mdt_object_fid(child)), rc);
2127                                 req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2128                                                    RCL_SERVER);
2129                                 /* handling -ENOENT is important because it may
2130                                  * change object state in DNE env dropping
2131                                  * LOHA_EXISTS flag, it is important to return
2132                                  * that to the caller.
2133                                  * Check LU-13115 for details.
2134                                  */
2135                                 if (rc != -ENOENT)
2136                                         rc = 0;
2137                         }
2138                 } else {
2139                         req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
2140                                            RCL_SERVER);
2141                 }
2142         }
2143         return rc;
2144 }