lustre/mdt/mdt_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/mdt/mdt_lib.c
   5  *  Lustre Metadata Target (mdt) request unpacking helper.
   6  *
   7  *  Copyright (c) 2006 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *   Author: Mike Shaver <shaver@clusterfs.com>
  12  *   Author: Nikita Danilov <nikita@clusterfs.com>
  13  *   Author: Huang Hua <huanghua@clusterfs.com>
  14  *
  15  *
  16  *   This file is part of the Lustre file system, http://www.lustre.org
  17  *   Lustre is a trademark of Cluster File Systems, Inc.
  18  *
  19  *   You may have signed or agreed to another license before downloading
  20  *   this software.  If so, you are bound by the terms and conditions
  21  *   of that agreement, and the following does not apply to you.  See the
  22  *   LICENSE file included with this distribution for more information.
  23  *
  24  *   If you did not agree to a different license, then this copy of Lustre
  25  *   is open source software; you can redistribute it and/or modify it
  26  *   under the terms of version 2 of the GNU General Public License as
  27  *   published by the Free Software Foundation.
  28  *
  29  *   In either case, Lustre is distributed in the hope that it will be
  30  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  31  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  32  *   license text for more details.
  33  */
  34
  35
  36 #ifndef EXPORT_SYMTAB
  37 # define EXPORT_SYMTAB
  38 #endif
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include "mdt_internal.h"
  42
  43
  44 typedef enum ucred_init_type {
  45         BODY_INIT       = 0,
  46         REC_INIT        = 1,
  47 } ucred_init_type_t;
  48
  49 int groups_from_list(struct group_info *ginfo, gid_t *glist)
  50 {
  51         int i;
  52         int count = ginfo->ngroups;
  53
  54         /* fill group_info from gid array */
  55         for (i = 0; i < ginfo->nblocks; i++) {
  56                 int cp_count = min(NGROUPS_PER_BLOCK, count);
  57                 int off = i * NGROUPS_PER_BLOCK;
  58                 int len = cp_count * sizeof(*glist);
  59
  60                 if (memcpy(ginfo->blocks[i], glist + off, len))
  61                         return -EFAULT;
  62
  63                 count -= cp_count;
  64         }
  65         return 0;
  66 }
  67
  68 /* groups_sort() is copied from linux kernel! */
  69 /* a simple shell-metzner sort */
  70 void groups_sort(struct group_info *group_info)
  71 {
  72         int base, max, stride;
  73         int gidsetsize = group_info->ngroups;
  74
  75         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
  76                 ; /* nothing */
  77         stride /= 3;
  78
  79         while (stride) {
  80                 max = gidsetsize - stride;
  81                 for (base = 0; base < max; base++) {
  82                         int left = base;
  83                         int right = left + stride;
  84                         gid_t tmp = GROUP_AT(group_info, right);
  85
  86                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
  87                                 GROUP_AT(group_info, right) =
  88                                     GROUP_AT(group_info, left);
  89                                 right = left;
  90                                 left -= stride;
  91                         }
  92                         GROUP_AT(group_info, right) = tmp;
  93                 }
  94                 stride /= 3;
  95         }
  96 }
  97
  98 void mdt_exit_ucred(struct mdt_thread_info *info)
  99 {
 100         struct md_ucred   *uc  = mdt_ucred(info);
 101         struct mdt_device *mdt = info->mti_mdt;
 102
 103         if (uc->mu_valid != UCRED_INIT) {
 104                 uc->mu_suppgids[0] = uc->mu_suppgids[1] = -1;
 105                 if (uc->mu_ginfo) {
 106                         groups_free(uc->mu_ginfo);
 107                         uc->mu_ginfo = NULL;
 108                 }
 109                 if (uc->mu_identity) {
 110                         mdt_identity_put(mdt->mdt_identity_cache,
 111                                          uc->mu_identity);
 112                         uc->mu_identity = NULL;
 113                 }
 114                 uc->mu_valid = UCRED_INIT;
 115         }
 116 }
 117
 118 static int old_init_ucred(struct mdt_thread_info *info,
 119                           struct mdt_body *body)
 120 {
 121         struct md_ucred     *uc  = mdt_ucred(info);
 122         struct mdt_device   *mdt = info->mti_mdt;
 123         struct mdt_identity *identity = NULL;
 124
 125         ENTRY;
 126
 127         uc->mu_valid = UCRED_INVALID;
 128
 129         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 130                 /* get identity info of this user */
 131                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 132                                             body->fsuid);
 133                 if (!identity) {
 134                         CERROR("Deny access without identity: uid %d\n",
 135                                body->fsuid);
 136                         RETURN(-EACCES);
 137                 }
 138         }
 139
 140         uc->mu_valid = UCRED_OLD;
 141         uc->mu_squash = SQUASH_NONE;
 142         uc->mu_o_uid = uc->mu_uid = body->uid;
 143         uc->mu_o_gid = uc->mu_gid = body->gid;
 144         uc->mu_o_fsuid = uc->mu_fsuid = body->fsuid;
 145         uc->mu_o_fsgid = uc->mu_fsgid = body->fsgid;
 146         uc->mu_suppgids[0] = body->suppgid;
 147         uc->mu_suppgids[1] = -1;
 148         uc->mu_cap = body->capability;
 149         uc->mu_ginfo = NULL;
 150         uc->mu_identity = identity;
 151
 152         RETURN(0);
 153 }
 154
 155 static int old_init_ucred_reint(struct mdt_thread_info *info)
 156 {
 157         struct md_ucred     *uc  = mdt_ucred(info);
 158         struct mdt_device   *mdt = info->mti_mdt;
 159         struct mdt_identity *identity = NULL;
 160
 161         ENTRY;
 162
 163         uc->mu_valid = UCRED_INVALID;
 164
 165         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 166                 /* get identity info of this user */
 167                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 168                                             uc->mu_fsuid);
 169                 if (!identity) {
 170                         CERROR("Deny access without identity: uid %d\n",
 171                                uc->mu_fsuid);
 172                         RETURN(-EACCES);
 173                 }
 174         }
 175
 176         uc->mu_valid = UCRED_OLD;
 177         uc->mu_squash = SQUASH_NONE;
 178         uc->mu_o_uid = uc->mu_o_fsuid = uc->mu_uid = uc->mu_fsuid;
 179         uc->mu_o_gid = uc->mu_o_fsgid = uc->mu_gid = uc->mu_fsgid;
 180         uc->mu_ginfo = NULL;
 181         uc->mu_identity = identity;
 182
 183         RETURN(0);
 184 }
 185
 186 static int nid_nosquash(struct mdt_device *mdt, lnet_nid_t nid)
 187 {
 188         struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
 189         int i;
 190
 191         for (i = 0; i < rsi->rsi_n_nosquash_nids; i++)
 192                 if ((rsi->rsi_nosquash_nids[i] == nid) ||
 193                     (rsi->rsi_nosquash_nids[i] == LNET_NID_ANY))
 194                         return 1;
 195
 196         return 0;
 197 }
 198
 199 static int mdt_squash_root(struct mdt_device *mdt, struct md_ucred *ucred,
 200                            struct ptlrpc_user_desc *pud, lnet_nid_t peernid)
 201 {
 202         struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
 203
 204         if (!rsi || (!rsi->rsi_uid && !rsi->rsi_gid) ||
 205             nid_nosquash(mdt, peernid))
 206                 return 0;
 207
 208         CDEBUG(D_SEC, "squash req from "LPX64":"
 209                "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", peernid,
 210                pud->pud_uid, pud->pud_gid,
 211                pud->pud_fsuid, pud->pud_fsgid, pud->pud_cap,
 212                pud->pud_uid ? pud->pud_uid : rsi->rsi_uid,
 213                pud->pud_uid ? pud->pud_gid : rsi->rsi_gid,
 214                pud->pud_fsuid ? pud->pud_fsuid : rsi->rsi_uid,
 215                pud->pud_fsuid ? pud->pud_fsgid : rsi->rsi_gid,
 216                pud->pud_cap & ~CAP_FS_MASK);
 217
 218         if (rsi->rsi_uid) {
 219                 if (!pud->pud_uid) {
 220                         ucred->mu_uid = rsi->rsi_uid;
 221                         ucred->mu_squash |= SQUASH_UID;
 222                 } else {
 223                         ucred->mu_uid = pud->pud_uid;
 224                 }
 225
 226                 if (!pud->pud_fsuid) {
 227                         ucred->mu_fsuid = rsi->rsi_uid;
 228                         ucred->mu_squash |= SQUASH_UID;
 229                 } else {
 230                         ucred->mu_fsuid = pud->pud_fsuid;
 231                 }
 232         } else {
 233                 ucred->mu_uid   = pud->pud_uid;
 234                 ucred->mu_fsuid = pud->pud_fsuid;
 235         }
 236
 237         if (rsi->rsi_gid) {
 238                 int i;
 239
 240                 if (!pud->pud_gid) {
 241                         ucred->mu_gid = rsi->rsi_gid;
 242                         ucred->mu_squash |= SQUASH_GID;
 243                 } else {
 244                         ucred->mu_gid = pud->pud_gid;
 245                 }
 246
 247                 if (!pud->pud_fsgid) {
 248                         ucred->mu_fsgid = rsi->rsi_gid;
 249                         ucred->mu_squash |= SQUASH_GID;
 250                 } else {
 251                         ucred->mu_fsgid = pud->pud_fsgid;
 252                 }
 253
 254                 for (i = 0; i < 2; i++) {
 255                         if (!ucred->mu_suppgids[i]) {
 256                                 ucred->mu_suppgids[i] = rsi->rsi_gid;
 257                                 ucred->mu_squash |= SQUASH_GID;
 258                         }
 259                 }
 260
 261                 for (i = 0; i < pud->pud_ngroups; i++) {
 262                         if (!pud->pud_groups[i]) {
 263                                 pud->pud_groups[i] = rsi->rsi_gid;
 264                                 ucred->mu_squash |= SQUASH_GID;
 265                         }
 266                 }
 267         } else {
 268                 ucred->mu_gid   = pud->pud_gid;
 269                 ucred->mu_fsgid = pud->pud_fsgid;
 270         }
 271
 272         if ((ucred->mu_squash & SQUASH_UID) || ucred->mu_fsuid)
 273                 ucred->mu_cap = (pud->pud_cap & ~CAP_FS_MASK);
 274         else
 275                 ucred->mu_cap = pud->pud_cap;
 276
 277         return 1;
 278 }
 279
 280 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 281                           void *buf)
 282 {
 283         struct ptlrpc_request   *req = mdt_info_req(info);
 284         struct mdt_export_data  *med = mdt_req2med(req);
 285         struct mdt_device       *mdt = info->mti_mdt;
 286         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 287         struct md_ucred         *ucred = mdt_ucred(info);
 288         struct mdt_identity     *identity = NULL;
 289         lnet_nid_t              peernid = req->rq_peer.nid;
 290         __u32                   setxid_perm = 0;
 291         int                     root_squashed = 0;
 292         int                     setuid;
 293         int                     setgid;
 294         int                     rc = 0;
 295
 296         ENTRY;
 297
 298         ucred->mu_valid = UCRED_INVALID;
 299
 300         if (req->rq_auth_gss && req->rq_auth_uid == INVALID_UID) {
 301                 CWARN("user not authenticated, deny access!\n");
 302                 RETURN(-EACCES);
 303         }
 304
 305         ucred->mu_o_uid   = pud->pud_uid;
 306         ucred->mu_o_gid   = pud->pud_gid;
 307         ucred->mu_o_fsuid = pud->pud_fsuid;
 308         ucred->mu_o_fsgid = pud->pud_fsgid;
 309
 310         if (type == BODY_INIT) {
 311                 struct mdt_body *body = (struct mdt_body *)buf;
 312
 313                 ucred->mu_suppgids[0] = body->suppgid;
 314                 ucred->mu_suppgids[1] = -1;
 315         }
 316
 317         /* sanity check: if we use strong authentication, we expect the
 318          * uid which client claimed is true */
 319         if (req->rq_auth_gss) {
 320                 if (med->med_rmtclient) {
 321                         if (ptlrpc_user_desc_do_idmap(req, pud))
 322                                 RETURN(-EACCES);
 323
 324                         if (req->rq_auth_mapped_uid != pud->pud_uid) {
 325                                 CERROR("remote client "LPU64": auth uid %u "
 326                                        "while client claim %u:%u/%u:%u\n",
 327                                        peernid, req->rq_auth_uid, pud->pud_uid,
 328                                        pud->pud_gid, pud->pud_fsuid,
 329                                        pud->pud_fsgid);
 330                                 RETURN(-EACCES);
 331                         }
 332                 } else {
 333                         if (req->rq_auth_uid != pud->pud_uid) {
 334                                 CERROR("local client "LPU64": auth uid %u "
 335                                        "while client claim %u:%u/%u:%u\n",
 336                                        peernid, req->rq_auth_uid, pud->pud_uid,
 337                                        pud->pud_gid, pud->pud_fsuid,
 338                                        pud->pud_fsgid);
 339                                 RETURN(-EACCES);
 340                         }
 341                 }
 342         }
 343
 344         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 345                 if (med->med_rmtclient) {
 346                         CERROR("remote client must run with identity_get "
 347                                "enabled!\n");
 348                         RETURN(-EACCES);
 349                 } else {
 350                         setxid_perm |= LUSTRE_SETGRP_PERM;
 351                         goto check_squash;
 352                 }
 353         }
 354
 355         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 356         if (!identity) {
 357                 CERROR("Deny access without identity: uid %d\n", pud->pud_uid);
 358                 RETURN(-EACCES);
 359         }
 360
 361         /* find out the setuid/setgid attempt */
 362         setuid = (pud->pud_uid != pud->pud_fsuid);
 363         setgid = (pud->pud_gid != pud->pud_fsgid ||
 364                   pud->pud_gid != identity->mi_gid);
 365
 366         setxid_perm = mdt_identity_get_setxid_perm(identity,
 367                                                    med->med_rmtclient,
 368                                                    peernid);
 369
 370         /* check permission of setuid */
 371         if (setuid && !(setxid_perm & LUSTRE_SETUID_PERM)) {
 372                 CWARN("mdt blocked setuid attempt (%u -> %u) from "
 373                       LPX64"\n", pud->pud_uid, pud->pud_fsuid, peernid);
 374                 GOTO(out, rc = -EACCES);
 375         }
 376
 377         /* check permission of setgid */
 378         if (setgid && !(setxid_perm & LUSTRE_SETGID_PERM)) {
 379                 CWARN("mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 380                       "from "LPX64"\n", pud->pud_uid, pud->pud_gid,
 381                       pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 382                       peernid);
 383                 GOTO(out, rc = -EACCES);
 384         }
 385
 386 check_squash:
 387         /* FIXME: The exact behavior of root_squash is not defined. */
 388         ucred->mu_squash = SQUASH_NONE;
 389         root_squashed = mdt_squash_root(mdt, ucred, pud, peernid);
 390         if (!root_squashed) {
 391                 ucred->mu_uid   = pud->pud_uid;
 392                 ucred->mu_gid   = pud->pud_gid;
 393                 ucred->mu_fsuid = pud->pud_fsuid;
 394                 ucred->mu_fsgid = pud->pud_fsgid;
 395                 /* remove fs privilege for non-root user */
 396                 if (pud->pud_fsuid)
 397                         ucred->mu_cap = (pud->pud_cap & ~CAP_FS_MASK);
 398                 else
 399                         ucred->mu_cap = pud->pud_cap;
 400         }
 401
 402         /*
 403          * NB: remote client not allowed to setgroups anyway.
 404          */
 405         if (!med->med_rmtclient && pud->pud_ngroups &&
 406             (setxid_perm & LUSTRE_SETGRP_PERM)) {
 407                 struct group_info *ginfo;
 408
 409                 /* setgroups for local client */
 410                 ginfo = groups_alloc(pud->pud_ngroups);
 411                 if (!ginfo) {
 412                         CERROR("failed to alloc %d groups\n",
 413                                pud->pud_ngroups);
 414                         GOTO(out, rc = -ENOMEM);
 415                 }
 416                 groups_from_list(ginfo, pud->pud_groups);
 417                 groups_sort(ginfo);
 418                 ucred->mu_ginfo = ginfo;
 419         } else {
 420                 ucred->mu_ginfo = NULL;
 421         }
 422
 423         ucred->mu_identity = identity;
 424         ucred->mu_valid = UCRED_NEW;
 425
 426 out:
 427         if (rc && identity)
 428                 mdt_identity_put(mdt->mdt_identity_cache, identity);
 429
 430         RETURN(rc);
 431 }
 432
 433 int mdt_check_ucred(struct mdt_thread_info *info)
 434 {
 435         struct ptlrpc_request   *req = mdt_info_req(info);
 436         struct mdt_export_data  *med = mdt_req2med(req);
 437         struct mdt_device       *mdt = info->mti_mdt;
 438         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 439         struct md_ucred         *ucred = mdt_ucred(info);
 440         struct mdt_identity     *identity;
 441         lnet_nid_t              peernid = req->rq_peer.nid;
 442
 443         ENTRY;
 444
 445         if ((ucred->mu_valid == UCRED_OLD) || (ucred->mu_valid == UCRED_NEW))
 446                 RETURN(0);
 447
 448         /* !rq_user_desc means null security, maybe inter-mds ops */
 449         if (!req->rq_user_desc)
 450                 RETURN(0);
 451
 452         if (req->rq_auth_gss && req->rq_auth_uid == INVALID_UID) {
 453                 CWARN("user not authenticated, deny access!\n");
 454                 RETURN(-EACCES);
 455         }
 456
 457         /* sanity check: if we use strong authentication, we expect the
 458          * uid which client claimed is true */
 459         if (req->rq_auth_gss) {
 460                 if (med->med_rmtclient) {
 461                         if (ptlrpc_user_desc_do_idmap(req, pud))
 462                                 RETURN(-EACCES);
 463
 464                         if (req->rq_auth_mapped_uid != pud->pud_uid) {
 465                                 CERROR("remote client "LPU64": auth uid %u "
 466                                        "while client claim %u:%u/%u:%u\n",
 467                                        peernid, req->rq_auth_uid, pud->pud_uid,
 468                                        pud->pud_gid, pud->pud_fsuid,
 469                                        pud->pud_fsgid);
 470                                 RETURN(-EACCES);
 471                         }
 472                 } else {
 473                         if (req->rq_auth_uid != pud->pud_uid) {
 474                                 CERROR("local client "LPU64": auth uid %u "
 475                                        "while client claim %u:%u/%u:%u\n",
 476                                        peernid, req->rq_auth_uid, pud->pud_uid,
 477                                        pud->pud_gid, pud->pud_fsuid,
 478                                        pud->pud_fsgid);
 479                                 RETURN(-EACCES);
 480                         }
 481                 }
 482         }
 483
 484         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 485                 if (med->med_rmtclient) {
 486                         CERROR("remote client must run with "
 487                                "identity_get enabled!\n");
 488                         RETURN(-EACCES);
 489                 } else {
 490                         RETURN(0);
 491                 }
 492         }
 493
 494         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 495         if (!identity) {
 496                 CERROR("Deny access without identity: uid %d\n", pud->pud_uid);
 497                 RETURN(-EACCES);
 498         }
 499
 500         mdt_identity_put(mdt->mdt_identity_cache, identity);
 501
 502         RETURN(0);
 503 }
 504
 505 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 506 {
 507         struct ptlrpc_request *req = mdt_info_req(info);
 508         struct md_ucred       *uc  = mdt_ucred(info);
 509
 510         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 511                 return 0;
 512
 513         mdt_exit_ucred(info);
 514
 515         /* !rq_user_desc means null security, maybe inter-mds ops */
 516         return req->rq_user_desc ? new_init_ucred(info, BODY_INIT, body) :
 517                                    old_init_ucred(info, body);
 518 }
 519
 520 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 521 {
 522         struct ptlrpc_request *req = mdt_info_req(info);
 523         struct md_ucred       *uc  = mdt_ucred(info);
 524
 525         if ((uc->mu_valid == UCRED_OLD) || (uc->mu_valid == UCRED_NEW))
 526                 return 0;
 527
 528         mdt_exit_ucred(info);
 529
 530         /* !rq_user_desc means null security, maybe inter-mds ops */
 531         return req->rq_user_desc ? new_init_ucred(info, REC_INIT, NULL) :
 532                                    old_init_ucred_reint(info);
 533 }
 534
 535 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 536 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm)
 537 {
 538         const struct lov_ost_data_v1 *lod;
 539         int i;
 540         __s16 stripe_count =
 541                 le16_to_cpu(((struct lov_user_md*)lmm)->lmm_stripe_count);
 542
 543         CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n",
 544                le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic),
 545                le32_to_cpu(lmm->lmm_pattern));
 546         CDEBUG(level,"stripe_size=0x%x, stripe_count=0x%x\n",
 547                le32_to_cpu(lmm->lmm_stripe_size),
 548                le32_to_cpu(lmm->lmm_stripe_count));
 549         LASSERT(stripe_count < (__s16)LOV_MAX_STRIPE_COUNT);
 550         for (i = 0, lod = lmm->lmm_objects; i < stripe_count; i++, lod++) {
 551                 CDEBUG(level, "stripe %u idx %u subobj "LPX64"/"LPX64"\n",
 552                        i, le32_to_cpu(lod->l_ost_idx),
 553                        le64_to_cpu(lod->l_object_gr),
 554                        le64_to_cpu(lod->l_object_id));
 555         }
 556 }
 557
 558 void mdt_shrink_reply(struct mdt_thread_info *info, int offset,
 559                       int mdscapa, int osscapa)
 560 {
 561         struct ptlrpc_request *req = mdt_info_req(info);
 562         struct mdt_body *body;
 563         int acl_size, md_size;
 564
 565         body = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 566         LASSERT(body != NULL);
 567
 568         if (body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE | OBD_MD_LINKNAME))
 569                 md_size = body->eadatasize;
 570         else
 571                 md_size = 0;
 572
 573         acl_size = body->aclsize;
 574
 575         CDEBUG(D_INFO, "Shrink to md_size %d cookie_size %d \n",
 576                md_size, acl_size);
 577
 578         lustre_shrink_reply(req, offset, md_size, 1);
 579         offset += !!md_size;
 580         lustre_shrink_reply(req, offset, acl_size, 1);
 581         offset += !!acl_size;
 582         if (mdscapa && !(body->valid & OBD_MD_FLMDSCAPA))
 583                 lustre_shrink_reply(req, offset, 0, 1);
 584         offset += mdscapa;
 585         if (osscapa && !(body->valid & OBD_MD_FLOSSCAPA))
 586                 lustre_shrink_reply(req, offset, 0, 0);
 587         offset += osscapa;
 588 }
 589
 590
 591 /* if object is dying, pack the lov/llog data,
 592  * parameter info->mti_attr should be valid at this point! */
 593 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 594                            const struct md_attr *ma)
 595 {
 596         struct mdt_body       *repbody;
 597         const struct lu_attr *la = &ma->ma_attr;
 598         ENTRY;
 599
 600         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 601         LASSERT(repbody != NULL);
 602
 603         if (ma->ma_valid & MA_INODE)
 604                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 605
 606         if (ma->ma_valid & MA_LOV) {
 607                 __u32 mode;
 608
 609                 if (mdt_object_exists(mo) < 0)
 610                         /* If it is a remote object, and we do not retrieve
 611                          * EA back unlink reg file*/
 612                         mode = S_IFREG;
 613                 else
 614                         mode = lu_object_attr(&mo->mot_obj.mo_lu);
 615
 616                 LASSERT(ma->ma_lmm_size);
 617                 mdt_dump_lmm(D_INFO, ma->ma_lmm);
 618                 repbody->eadatasize = ma->ma_lmm_size;
 619                 if (S_ISREG(mode))
 620                         repbody->valid |= OBD_MD_FLEASIZE;
 621                 else if (S_ISDIR(mode))
 622                         repbody->valid |= OBD_MD_FLDIREA;
 623                 else
 624                         LBUG();
 625         }
 626
 627         if (ma->ma_cookie_size && (ma->ma_valid & MA_COOKIE)) {
 628                 repbody->aclsize = ma->ma_cookie_size;
 629                 repbody->valid |= OBD_MD_FLCOOKIE;
 630         }
 631
 632         RETURN(0);
 633 }
 634
 635 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 636                                   struct md_attr *ma)
 637 {
 638         __u64 out;
 639
 640         out = 0;
 641         if (in & ATTR_MODE)
 642                 out |= LA_MODE;
 643         if (in & ATTR_UID)
 644                 out |= LA_UID;
 645         if (in & ATTR_GID)
 646                 out |= LA_GID;
 647         if (in & ATTR_SIZE)
 648                 out |= LA_SIZE;
 649         if (in & ATTR_BLOCKS)
 650                 out |= LA_BLOCKS;
 651
 652         if (in & ATTR_FROM_OPEN)
 653                 rr->rr_flags |= MRF_SETATTR_LOCKED;
 654
 655         if (in & ATTR_ATIME_SET)
 656                 out |= LA_ATIME;
 657
 658         if (in & ATTR_CTIME_SET)
 659                 out |= LA_CTIME;
 660
 661         if (in & ATTR_MTIME_SET)
 662                 out |= LA_MTIME;
 663
 664         if (in & ATTR_ATTR_FLAG)
 665                 out |= LA_FLAGS;
 666
 667         /*XXX need ATTR_RAW?*/
 668         in &= ~(ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_BLOCKS|
 669                 ATTR_ATIME|ATTR_MTIME|ATTR_CTIME|ATTR_FROM_OPEN|
 670                 ATTR_ATIME_SET|ATTR_CTIME_SET|ATTR_MTIME_SET|
 671                 ATTR_ATTR_FLAG|ATTR_RAW);
 672         if (in != 0)
 673                 CERROR("Unknown attr bits: %#llx\n", in);
 674         return out;
 675 }
 676 /* unpacking */
 677
 678 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 679 {
 680         struct md_ucred         *uc  = mdt_ucred(info);
 681         struct md_attr          *ma = &info->mti_attr;
 682         struct lu_attr          *la = &ma->ma_attr;
 683         struct req_capsule      *pill = &info->mti_pill;
 684         struct mdt_reint_record *rr = &info->mti_rr;
 685         struct mdt_rec_setattr  *rec;
 686         ENTRY;
 687
 688         rec = req_capsule_client_get(pill, &RMF_REC_SETATTR);
 689         if (rec == NULL)
 690                 RETURN(-EFAULT);
 691
 692         uc->mu_fsuid = rec->sa_fsuid;
 693         uc->mu_fsgid = rec->sa_fsgid;
 694         uc->mu_cap   = rec->sa_cap;
 695         uc->mu_suppgids[0] = rec->sa_suppgid;
 696         uc->mu_suppgids[1] = -1;
 697
 698         rr->rr_fid1 = &rec->sa_fid;
 699         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 700         la->la_mode  = rec->sa_mode;
 701         la->la_flags = rec->sa_attr_flags;
 702         la->la_uid   = rec->sa_uid;
 703         la->la_gid   = rec->sa_gid;
 704         la->la_size  = rec->sa_size;
 705         la->la_blocks = rec->sa_blocks;
 706         la->la_ctime = rec->sa_ctime;
 707         la->la_atime = rec->sa_atime;
 708         la->la_mtime = rec->sa_mtime;
 709         ma->ma_valid = MA_INODE;
 710
 711         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 712                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 713                                  req_capsule_client_get(pill, &RMF_CAPA1));
 714
 715         RETURN(0);
 716 }
 717
 718 static int mdt_epoch_unpack(struct mdt_thread_info *info)
 719 {
 720         struct req_capsule *pill = &info->mti_pill;
 721         ENTRY;
 722
 723         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 724                 info->mti_epoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 725         else
 726                 info->mti_epoch = NULL;
 727         RETURN(info->mti_epoch == NULL ? -EFAULT : 0);
 728 }
 729
 730 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 731 {
 732         struct md_attr          *ma = &info->mti_attr;
 733         struct req_capsule      *pill = &info->mti_pill;
 734         int rc;
 735         ENTRY;
 736
 737         rc = mdt_setattr_unpack_rec(info);
 738         if (rc)
 739                 RETURN(rc);
 740
 741         /* Epoch may be absent */
 742         mdt_epoch_unpack(info);
 743
 744         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 745                 ma->ma_lmm = req_capsule_client_get(pill, &RMF_EADATA);
 746                 ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_EADATA,
 747                                                        RCL_CLIENT);
 748                 ma->ma_valid |= MA_LOV;
 749         }
 750
 751         if (req_capsule_field_present(pill, &RMF_LOGCOOKIES, RCL_CLIENT)) {
 752                 ma->ma_cookie = req_capsule_client_get(pill,
 753                                                        &RMF_LOGCOOKIES);
 754                 ma->ma_cookie_size = req_capsule_get_size(pill,
 755                                                           &RMF_LOGCOOKIES,
 756                                                           RCL_CLIENT);
 757                 ma->ma_valid |= MA_COOKIE;
 758         }
 759
 760         RETURN(0);
 761 }
 762
 763 int mdt_close_unpack(struct mdt_thread_info *info)
 764 {
 765         int rc;
 766         ENTRY;
 767
 768         rc = mdt_epoch_unpack(info);
 769         if (rc)
 770                 RETURN(rc);
 771
 772         RETURN(mdt_setattr_unpack_rec(info));
 773 }
 774
 775 static int mdt_create_unpack(struct mdt_thread_info *info)
 776 {
 777         struct md_ucred         *uc  = mdt_ucred(info);
 778         struct mdt_rec_create   *rec;
 779         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 780         struct mdt_reint_record *rr = &info->mti_rr;
 781         struct req_capsule      *pill = &info->mti_pill;
 782         struct md_op_spec       *sp = &info->mti_spec;
 783         ENTRY;
 784
 785         rec = req_capsule_client_get(pill, &RMF_REC_CREATE);
 786         if (rec == NULL)
 787                 RETURN(-EFAULT);
 788
 789         uc->mu_fsuid = rec->cr_fsuid;
 790         uc->mu_fsgid = rec->cr_fsgid;
 791         uc->mu_cap   = rec->cr_cap;
 792         uc->mu_suppgids[0] = rec->cr_suppgid;
 793         uc->mu_suppgids[1] = -1;
 794
 795         rr->rr_fid1 = &rec->cr_fid1;
 796         rr->rr_fid2 = &rec->cr_fid2;
 797         attr->la_mode = rec->cr_mode;
 798         attr->la_rdev  = rec->cr_rdev;
 799         attr->la_uid   = rec->cr_fsuid;
 800         attr->la_gid   = rec->cr_fsgid;
 801         attr->la_ctime = rec->cr_time;
 802         attr->la_mtime = rec->cr_time;
 803         attr->la_atime = rec->cr_time;
 804         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID |
 805                          LA_CTIME | LA_MTIME | LA_ATIME;
 806         memset(&sp->u, 0, sizeof(sp->u));
 807         sp->sp_cr_flags = rec->cr_flags;
 808         sp->sp_ck_split = rec->cr_cksplit;
 809
 810         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 811                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 812                                  req_capsule_client_get(pill, &RMF_CAPA1));
 813         mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
 814
 815         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 816         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT);
 817
 818 #ifdef CONFIG_FS_POSIX_ACL
 819         if (sp->sp_cr_flags & MDS_CREATE_RMT_ACL) {
 820                 if (S_ISDIR(attr->la_mode))
 821                         sp->u.sp_pfid = rr->rr_fid1;
 822                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
 823                 LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
 824                                                   RCL_CLIENT));
 825                 sp->u.sp_ea.eadata = req_capsule_client_get(pill, &RMF_EADATA);
 826                 sp->u.sp_ea.eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 827                                                                 RCL_CLIENT);
 828                 RETURN(0);
 829         }
 830 #endif
 831         if (S_ISDIR(attr->la_mode)) {
 832                 /* pass parent fid for cross-ref cases */
 833                 sp->u.sp_pfid = rr->rr_fid1;
 834                 if (sp->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
 835                         /* create salve object req, need
 836                          * unpack split ea here
 837                          */
 838                        req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SLAVE);
 839                        LASSERT(req_capsule_field_present(pill, &RMF_EADATA,
 840                                                          RCL_CLIENT));
 841                        sp->u.sp_ea.eadata = req_capsule_client_get(pill,
 842                                                                    &RMF_EADATA);
 843                        sp->u.sp_ea.eadatalen = req_capsule_get_size(pill,
 844                                                                     &RMF_EADATA,
 845                                                                     RCL_CLIENT);
 846                        sp->u.sp_ea.fid = rr->rr_fid1;
 847                 }
 848         } else if (S_ISLNK(attr->la_mode)) {
 849                 const char *tgt = NULL;
 850
 851                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
 852                 if (req_capsule_field_present(pill, &RMF_SYMTGT, RCL_CLIENT)) {
 853                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
 854                         sp->u.sp_symname = tgt;
 855                 }
 856                 if (tgt == NULL)
 857                         RETURN(-EFAULT);
 858         }
 859         RETURN(0);
 860 }
 861
 862 static int mdt_link_unpack(struct mdt_thread_info *info)
 863 {
 864         struct md_ucred         *uc  = mdt_ucred(info);
 865         struct mdt_rec_link     *rec;
 866         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 867         struct mdt_reint_record *rr = &info->mti_rr;
 868         struct req_capsule      *pill = &info->mti_pill;
 869         ENTRY;
 870
 871         rec = req_capsule_client_get(pill, &RMF_REC_LINK);
 872         if (rec == NULL)
 873                 RETURN(-EFAULT);
 874
 875         uc->mu_fsuid = rec->lk_fsuid;
 876         uc->mu_fsgid = rec->lk_fsgid;
 877         uc->mu_cap   = rec->lk_cap;
 878         uc->mu_suppgids[0] = rec->lk_suppgid1;
 879         uc->mu_suppgids[1] = rec->lk_suppgid2;
 880
 881         attr->la_uid = rec->lk_fsuid;
 882         attr->la_gid = rec->lk_fsgid;
 883         rr->rr_fid1 = &rec->lk_fid1;
 884         rr->rr_fid2 = &rec->lk_fid2;
 885         attr->la_ctime = rec->lk_time;
 886         attr->la_mtime = rec->lk_time;
 887         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
 888
 889         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 890                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 891                                  req_capsule_client_get(pill, &RMF_CAPA1));
 892         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
 893                 mdt_set_capainfo(info, 1, rr->rr_fid2,
 894                                  req_capsule_client_get(pill, &RMF_CAPA2));
 895
 896         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 897         if (rr->rr_name == NULL)
 898                 RETURN(-EFAULT);
 899         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT);
 900         info->mti_spec.sp_ck_split = rec->lk_cksplit;
 901
 902         RETURN(0);
 903 }
 904
 905 static int mdt_unlink_unpack(struct mdt_thread_info *info)
 906 {
 907         struct md_ucred         *uc  = mdt_ucred(info);
 908         struct mdt_rec_unlink   *rec;
 909         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 910         struct mdt_reint_record *rr = &info->mti_rr;
 911         struct req_capsule      *pill = &info->mti_pill;
 912         ENTRY;
 913
 914         rec = req_capsule_client_get(pill, &RMF_REC_UNLINK);
 915         if (rec == NULL)
 916                 RETURN(-EFAULT);
 917
 918         uc->mu_fsuid = rec->ul_fsuid;
 919         uc->mu_fsgid = rec->ul_fsgid;
 920         uc->mu_cap   = rec->ul_cap;
 921         uc->mu_suppgids[0] = rec->ul_suppgid;
 922         uc->mu_suppgids[1] = -1;
 923
 924         attr->la_uid = rec->ul_fsuid;
 925         attr->la_gid = rec->ul_fsgid;
 926         rr->rr_fid1 = &rec->ul_fid1;
 927         rr->rr_fid2 = &rec->ul_fid2;
 928         attr->la_ctime = rec->ul_time;
 929         attr->la_mtime = rec->ul_time;
 930         attr->la_mode  = rec->ul_mode;
 931         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
 932
 933         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 934                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 935                                  req_capsule_client_get(pill, &RMF_CAPA1));
 936
 937         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 938         if (rr->rr_name == NULL)
 939                 RETURN(-EFAULT);
 940         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT);
 941         info->mti_spec.sp_ck_split = rec->ul_cksplit;
 942
 943         RETURN(0);
 944 }
 945
 946 static int mdt_rename_unpack(struct mdt_thread_info *info)
 947 {
 948         struct md_ucred         *uc = mdt_ucred(info);
 949         struct mdt_rec_rename   *rec;
 950         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 951         struct mdt_reint_record *rr = &info->mti_rr;
 952         struct req_capsule      *pill = &info->mti_pill;
 953         ENTRY;
 954
 955         rec = req_capsule_client_get(pill, &RMF_REC_RENAME);
 956         if (rec == NULL)
 957                 RETURN(-EFAULT);
 958
 959         uc->mu_fsuid = rec->rn_fsuid;
 960         uc->mu_fsgid = rec->rn_fsgid;
 961         uc->mu_cap   = rec->rn_cap;
 962         uc->mu_suppgids[0] = rec->rn_suppgid1;
 963         uc->mu_suppgids[1] = rec->rn_suppgid2;
 964
 965         attr->la_uid = rec->rn_fsuid;
 966         attr->la_gid = rec->rn_fsgid;
 967         rr->rr_fid1 = &rec->rn_fid1;
 968         rr->rr_fid2 = &rec->rn_fid2;
 969         attr->la_ctime = rec->rn_time;
 970         attr->la_mtime = rec->rn_time;
 971         /* rename_tgt contains the mode already */
 972         attr->la_mode = rec->rn_mode;
 973         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
 974
 975         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
 976                 mdt_set_capainfo(info, 0, rr->rr_fid1,
 977                                  req_capsule_client_get(pill, &RMF_CAPA1));
 978         if (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT))
 979                 mdt_set_capainfo(info, 1, rr->rr_fid2,
 980                                  req_capsule_client_get(pill, &RMF_CAPA2));
 981
 982         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
 983         rr->rr_tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
 984         if (rr->rr_name == NULL || rr->rr_tgt == NULL)
 985                 RETURN(-EFAULT);
 986         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT);
 987         rr->rr_tgtlen = req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT);
 988         info->mti_spec.sp_ck_split = rec->rn_cksplit;
 989
 990         RETURN(0);
 991 }
 992
 993 static int mdt_open_unpack(struct mdt_thread_info *info)
 994 {
 995         struct md_ucred         *uc = mdt_ucred(info);
 996         struct mdt_rec_create   *rec;
 997         struct lu_attr          *attr = &info->mti_attr.ma_attr;
 998         struct req_capsule      *pill = &info->mti_pill;
 999         struct mdt_reint_record *rr   = &info->mti_rr;
1000         struct ptlrpc_request   *req  = mdt_info_req(info);
1001         ENTRY;
1002
1003         rec = req_capsule_client_get(pill, &RMF_REC_CREATE);
1004         if (rec == NULL)
1005                 RETURN(-EFAULT);
1006
1007         uc->mu_fsuid = rec->cr_fsuid;
1008         uc->mu_fsgid = rec->cr_fsgid;
1009         uc->mu_cap   = rec->cr_cap;
1010         uc->mu_suppgids[0] = rec->cr_suppgid;
1011         uc->mu_suppgids[1] = -1;
1012
1013         rr->rr_fid1   = &rec->cr_fid1;
1014         rr->rr_fid2   = &rec->cr_fid2;
1015         attr->la_mode = rec->cr_mode;
1016         attr->la_rdev  = rec->cr_rdev;
1017         attr->la_uid   = rec->cr_fsuid;
1018         attr->la_gid   = rec->cr_fsgid;
1019         attr->la_ctime = rec->cr_time;
1020         attr->la_mtime = rec->cr_time;
1021         attr->la_atime = rec->cr_time;
1022         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1023                          LA_CTIME | LA_MTIME | LA_ATIME;
1024         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1025         info->mti_spec.sp_cr_flags = rec->cr_flags;
1026         info->mti_replayepoch = rec->cr_ioepoch;
1027
1028         info->mti_spec.sp_ck_split = rec->cr_cksplit;
1029
1030         if (req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1031                 mdt_set_capainfo(info, 0, rr->rr_fid1,
1032                                  req_capsule_client_get(pill, &RMF_CAPA1));
1033         if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
1034             (req_capsule_get_size(pill, &RMF_CAPA2, RCL_CLIENT)))
1035                 mdt_set_capainfo(info, 1, rr->rr_fid2,
1036                                  req_capsule_client_get(pill, &RMF_CAPA2));
1037
1038         rr->rr_name = req_capsule_client_get(pill, &RMF_NAME);
1039         if (rr->rr_name == NULL)
1040                 RETURN(-EFAULT);
1041         rr->rr_namelen = req_capsule_get_size(pill, &RMF_NAME, RCL_CLIENT);
1042
1043         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1044                 struct md_op_spec *sp = &info->mti_spec;
1045                 sp->u.sp_ea.eadata = req_capsule_client_get(pill,
1046                                                             &RMF_EADATA);
1047                 sp->u.sp_ea.eadatalen = req_capsule_get_size(pill,
1048                                                              &RMF_EADATA,
1049                                                              RCL_CLIENT);
1050                 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
1051                         sp->u.sp_ea.no_lov_create = 1;
1052         }
1053
1054         RETURN(0);
1055 }
1056
1057 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1058
1059 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1060         [REINT_SETATTR]  = mdt_setattr_unpack,
1061         [REINT_CREATE]   = mdt_create_unpack,
1062         [REINT_LINK]     = mdt_link_unpack,
1063         [REINT_UNLINK]   = mdt_unlink_unpack,
1064         [REINT_RENAME]   = mdt_rename_unpack,
1065         [REINT_OPEN]     = mdt_open_unpack
1066 };
1067
1068 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1069 {
1070         int rc;
1071         ENTRY;
1072
1073         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1074         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1075                 info->mti_rr.rr_opcode = op;
1076                 rc = mdt_reint_unpackers[op](info);
1077         } else {
1078                 CERROR("Unexpected opcode %d\n", op);
1079                 rc = -EFAULT;
1080         }
1081         RETURN(rc);
1082 }