lustre/mdt/mdt_handler.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) 2010, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_net.h>
  57 #include <lustre_nodemap.h>
  58 #include <lustre_mds.h>
  59 #include <lustre_param.h>
  60 #include <lustre_quota.h>
  61 #include <lustre_swab.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 module_param(max_mod_rpcs_per_client, uint, 0644);
  73 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 270                               struct lu_fid *fid)
 271 {
 272         struct mdt_device *mdt = info->mti_mdt;
 273         struct lu_name *lname = &info->mti_name;
 274         char *name = NULL;
 275         struct mdt_object *parent;
 276         u32 mode;
 277         int rc = 0;
 278
 279         LASSERT(!info->mti_cross_ref);
 280
 281         OBD_ALLOC(name, NAME_MAX + 1);
 282         if (name == NULL)
 283                 return -ENOMEM;
 284         lname->ln_name = name;
 285
 286         /*
 287          * We may want to allow this to mount a completely separate
 288          * fileset from the MDT in the future, but keeping it to
 289          * ROOT/ only for now avoid potential security issues.
 290          */
 291         *fid = mdt->mdt_md_root_fid;
 292
 293         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 294                 const char *s1 = fileset;
 295                 const char *s2;
 296
 297                 while (*++s1 == '/')
 298                         ;
 299                 s2 = s1;
 300                 while (*s2 != '/' && *s2 != '\0')
 301                         s2++;
 302
 303                 if (s2 == s1)
 304                         break;
 305
 306                 fileset = s2;
 307
 308                 lname->ln_namelen = s2 - s1;
 309                 if (lname->ln_namelen > NAME_MAX) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 /* reject .. as a path component */
 315                 if (lname->ln_namelen == 2 &&
 316                     strncmp(s1, "..", 2) == 0) {
 317                         rc = -EINVAL;
 318                         break;
 319                 }
 320
 321                 strncpy(name, s1, lname->ln_namelen);
 322                 name[lname->ln_namelen] = '\0';
 323
 324                 parent = mdt_object_find(info->mti_env, mdt, fid);
 325                 if (IS_ERR(parent)) {
 326                         rc = PTR_ERR(parent);
 327                         break;
 328                 }
 329                 /* Only got the fid of this obj by name */
 330                 fid_zero(fid);
 331                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 332                                 fid, &info->mti_spec);
 333                 mdt_object_put(info->mti_env, parent);
 334         }
 335         if (!rc) {
 336                 parent = mdt_object_find(info->mti_env, mdt, fid);
 337                 if (IS_ERR(parent))
 338                         rc = PTR_ERR(parent);
 339                 else {
 340                         mode = lu_object_attr(&parent->mot_obj);
 341                         mdt_object_put(info->mti_env, parent);
 342                         if (!S_ISDIR(mode))
 343                                 rc = -ENOTDIR;
 344                 }
 345         }
 346
 347         OBD_FREE(name, NAME_MAX + 1);
 348
 349         return rc;
 350 }
 351
 352 static int mdt_get_root(struct tgt_session_info *tsi)
 353 {
 354         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 355         struct mdt_device       *mdt = info->mti_mdt;
 356         struct mdt_body         *repbody;
 357         char                    *fileset = NULL, *buffer = NULL;
 358         int                      rc;
 359         struct obd_export       *exp = info->mti_exp;
 360         char                    *nodemap_fileset;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376         }
 377
 378         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 379         if (nodemap_fileset && nodemap_fileset[0]) {
 380                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 381                 if (fileset) {
 382                         /* consider fileset from client as a sub-fileset
 383                          * of the nodemap one */
 384                         OBD_ALLOC(buffer, PATH_MAX + 1);
 385                         if (buffer == NULL)
 386                                 GOTO(out, rc = err_serious(-ENOMEM));
 387                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 388                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 389                                 GOTO(out, rc = err_serious(-EINVAL));
 390                         fileset = buffer;
 391                 } else {
 392                         /* enforce fileset as specified in the nodemap */
 393                         fileset = nodemap_fileset;
 394                 }
 395         }
 396
 397         if (fileset) {
 398                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 399                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 400                 if (rc < 0)
 401                         GOTO(out, rc = err_serious(rc));
 402         } else {
 403                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 404         }
 405         repbody->mbo_valid |= OBD_MD_FLID;
 406
 407         EXIT;
 408 out:
 409         mdt_thread_info_fini(info);
 410         if (buffer)
 411                 OBD_FREE(buffer, PATH_MAX+1);
 412         return rc;
 413 }
 414
 415 static int mdt_statfs(struct tgt_session_info *tsi)
 416 {
 417         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 418         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 419         struct md_device                *next = info->mti_mdt->mdt_child;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         /** statfs information are cached in the mdt_device */
 445         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 446                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 447                 /** statfs data is too old, get up-to-date one */
 448                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 449                 if (rc)
 450                         GOTO(out, rc);
 451                 spin_lock(&info->mti_mdt->mdt_lock);
 452                 info->mti_mdt->mdt_osfs = *osfs;
 453                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 454                 spin_unlock(&info->mti_mdt->mdt_lock);
 455         } else {
 456                 /** use cached statfs data */
 457                 spin_lock(&info->mti_mdt->mdt_lock);
 458                 *osfs = info->mti_mdt->mdt_osfs;
 459                 spin_unlock(&info->mti_mdt->mdt_lock);
 460         }
 461
 462         if (rc == 0)
 463                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 464 out:
 465         mdt_thread_info_fini(info);
 466         RETURN(rc);
 467 }
 468
 469 #ifdef CONFIG_FS_POSIX_ACL
 470 /*
 471  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 472  *
 473  * \param       info    thread info object
 474  * \param       repbody reply to pack ACLs into
 475  * \param       o       mdt object of file to examine
 476  * \param       nodemap nodemap of client to reply to
 477  * \retval      0       success
 478  * \retval      -errno  error getting or parsing ACL from disk
 479  */
 480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 481                       struct mdt_object *o, struct lu_nodemap *nodemap)
 482 {
 483         const struct lu_env     *env = info->mti_env;
 484         struct md_object        *next = mdt_object_child(o);
 485         struct lu_buf           *buf = &info->mti_buf;
 486         int rc;
 487
 488         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 489         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 490                                            RCL_SERVER);
 491         if (buf->lb_len == 0)
 492                 return 0;
 493
 494         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 495         if (rc < 0) {
 496                 if (rc == -ENODATA) {
 497                         repbody->mbo_aclsize = 0;
 498                         repbody->mbo_valid |= OBD_MD_FLACL;
 499                         rc = 0;
 500                 } else if (rc == -EOPNOTSUPP) {
 501                         rc = 0;
 502                 } else {
 503                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 504                                mdt_obd_name(info->mti_mdt),
 505                                PFID(mdt_object_fid(o)), rc);
 506                 }
 507         } else {
 508                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 509                                      rc, NODEMAP_FS_TO_CLIENT);
 510                 /* if all ACLs mapped out, rc is still >= 0 */
 511                 if (rc < 0) {
 512                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 513                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 514                                PFID(mdt_object_fid(o)), rc);
 515                 } else {
 516                         repbody->mbo_aclsize = rc;
 517                         repbody->mbo_valid |= OBD_MD_FLACL;
 518                         rc = 0;
 519                 }
 520         }
 521         return rc;
 522 }
 523 #endif
 524
 525 /* XXX Look into layout in MDT layer. */
 526 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 527 {
 528         struct lov_comp_md_v1   *comp_v1;
 529         struct lov_mds_md       *v1;
 530         int                      i;
 531
 532         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 533                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 534
 535                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 536                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 537                                 comp_v1->lcm_entries[i].lcme_offset);
 538                         /* We don't support partial release for now */
 539                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 540                                 return false;
 541                 }
 542                 return true;
 543         } else {
 544                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 545                         true : false;
 546         }
 547 }
 548
 549 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 550                         const struct lu_attr *attr, const struct lu_fid *fid)
 551 {
 552         struct md_attr *ma = &info->mti_attr;
 553         struct obd_export *exp = info->mti_exp;
 554         struct lu_nodemap *nodemap = NULL;
 555
 556         LASSERT(ma->ma_valid & MA_INODE);
 557
 558         if (attr->la_valid & LA_ATIME) {
 559                 b->mbo_atime = attr->la_atime;
 560                 b->mbo_valid |= OBD_MD_FLATIME;
 561         }
 562         if (attr->la_valid & LA_MTIME) {
 563                 b->mbo_mtime = attr->la_mtime;
 564                 b->mbo_valid |= OBD_MD_FLMTIME;
 565         }
 566         if (attr->la_valid & LA_CTIME) {
 567                 b->mbo_ctime = attr->la_ctime;
 568                 b->mbo_valid |= OBD_MD_FLCTIME;
 569         }
 570         if (attr->la_valid & LA_FLAGS) {
 571                 b->mbo_flags = attr->la_flags;
 572                 b->mbo_valid |= OBD_MD_FLFLAGS;
 573         }
 574         if (attr->la_valid & LA_NLINK) {
 575                 b->mbo_nlink = attr->la_nlink;
 576                 b->mbo_valid |= OBD_MD_FLNLINK;
 577         }
 578         if (attr->la_valid & (LA_UID|LA_GID)) {
 579                 nodemap = nodemap_get_from_exp(exp);
 580                 if (IS_ERR(nodemap))
 581                         goto out;
 582         }
 583         if (attr->la_valid & LA_UID) {
 584                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 585                                             NODEMAP_FS_TO_CLIENT,
 586                                             attr->la_uid);
 587                 b->mbo_valid |= OBD_MD_FLUID;
 588         }
 589         if (attr->la_valid & LA_GID) {
 590                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 591                                             NODEMAP_FS_TO_CLIENT,
 592                                             attr->la_gid);
 593                 b->mbo_valid |= OBD_MD_FLGID;
 594         }
 595
 596         if (attr->la_valid & LA_PROJID) {
 597                 /* TODO, nodemap for project id */
 598                 b->mbo_projid = attr->la_projid;
 599                 b->mbo_valid |= OBD_MD_FLPROJID;
 600         }
 601
 602         b->mbo_mode = attr->la_mode;
 603         if (attr->la_valid & LA_MODE)
 604                 b->mbo_valid |= OBD_MD_FLMODE;
 605         if (attr->la_valid & LA_TYPE)
 606                 b->mbo_valid |= OBD_MD_FLTYPE;
 607
 608         if (fid != NULL) {
 609                 b->mbo_fid1 = *fid;
 610                 b->mbo_valid |= OBD_MD_FLID;
 611                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 612                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 613         }
 614
 615         if (!(attr->la_valid & LA_TYPE))
 616                 return;
 617
 618         b->mbo_rdev   = attr->la_rdev;
 619         b->mbo_size   = attr->la_size;
 620         b->mbo_blocks = attr->la_blocks;
 621
 622         if (!S_ISREG(attr->la_mode)) {
 623                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 624         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 625                 /* means no objects are allocated on osts. */
 626                 LASSERT(!(ma->ma_valid & MA_LOV));
 627                 /* just ignore blocks occupied by extend attributes on MDS */
 628                 b->mbo_blocks = 0;
 629                 /* if no object is allocated on osts, the size on mds is valid.
 630                  * b=22272 */
 631                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 632         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 633                    mdt_hsm_is_released(ma->ma_lmm)) {
 634                 /* A released file stores its size on MDS. */
 635                 /* But return 1 block for released file, unless tools like tar
 636                  * will consider it fully sparse. (LU-3864)
 637                  */
 638                 if (unlikely(b->mbo_size == 0))
 639                         b->mbo_blocks = 0;
 640                 else
 641                         b->mbo_blocks = 1;
 642                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 643         }
 644
 645         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 646                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 647                        PFID(fid), (unsigned long long)b->mbo_size);
 648
 649 out:
 650         if (!IS_ERR_OR_NULL(nodemap))
 651                 nodemap_putref(nodemap);
 652 }
 653
 654 static inline int mdt_body_has_lov(const struct lu_attr *la,
 655                                    const struct mdt_body *body)
 656 {
 657         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 658                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 659 }
 660
 661 void mdt_client_compatibility(struct mdt_thread_info *info)
 662 {
 663         struct mdt_body       *body;
 664         struct ptlrpc_request *req = mdt_info_req(info);
 665         struct obd_export     *exp = req->rq_export;
 666         struct md_attr        *ma = &info->mti_attr;
 667         struct lu_attr        *la = &ma->ma_attr;
 668         ENTRY;
 669
 670         if (exp_connect_layout(exp))
 671                 /* the client can deal with 16-bit lmm_stripe_count */
 672                 RETURN_EXIT;
 673
 674         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 675
 676         if (!mdt_body_has_lov(la, body))
 677                 RETURN_EXIT;
 678
 679         /* now we have a reply with a lov for a client not compatible with the
 680          * layout lock so we have to clean the layout generation number */
 681         if (S_ISREG(la->la_mode))
 682                 ma->ma_lmm->lmm_layout_gen = 0;
 683         EXIT;
 684 }
 685
 686 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 687                                    struct mdt_object *o)
 688 {
 689         const struct lu_env *env = info->mti_env;
 690         int rc, rc2;
 691
 692         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 693                           XATTR_NAME_LOV);
 694
 695         if (rc == -ENODATA)
 696                 rc = 0;
 697
 698         if (rc < 0)
 699                 goto out;
 700
 701         /* Is it a directory? Let's check for the LMV as well */
 702         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 703                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 704                                    XATTR_NAME_LMV);
 705
 706                 if (rc2 == -ENODATA)
 707                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 708                                            &LU_BUF_NULL,
 709                                            XATTR_NAME_DEFAULT_LMV);
 710
 711                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 712                         rc = rc2;
 713         }
 714
 715 out:
 716         return rc;
 717 }
 718
 719 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 720                       const char *name)
 721 {
 722         const struct lu_env *env = info->mti_env;
 723         int rc;
 724         ENTRY;
 725
 726         LASSERT(info->mti_big_lmm_used == 0);
 727         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 728         if (rc < 0)
 729                 RETURN(rc);
 730
 731         /* big_lmm may need to be grown */
 732         if (info->mti_big_lmmsize < rc) {
 733                 int size = size_roundup_power2(rc);
 734
 735                 if (info->mti_big_lmmsize > 0) {
 736                         /* free old buffer */
 737                         LASSERT(info->mti_big_lmm);
 738                         OBD_FREE_LARGE(info->mti_big_lmm,
 739                                        info->mti_big_lmmsize);
 740                         info->mti_big_lmm = NULL;
 741                         info->mti_big_lmmsize = 0;
 742                 }
 743
 744                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 745                 if (info->mti_big_lmm == NULL)
 746                         RETURN(-ENOMEM);
 747                 info->mti_big_lmmsize = size;
 748         }
 749         LASSERT(info->mti_big_lmmsize >= rc);
 750
 751         info->mti_buf.lb_buf = info->mti_big_lmm;
 752         info->mti_buf.lb_len = info->mti_big_lmmsize;
 753         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 754
 755         RETURN(rc);
 756 }
 757
 758 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 759                    struct md_attr *ma, const char *name)
 760 {
 761         struct md_object *next = mdt_object_child(o);
 762         struct lu_buf    *buf = &info->mti_buf;
 763         int rc;
 764
 765         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 766                 buf->lb_buf = ma->ma_lmm;
 767                 buf->lb_len = ma->ma_lmm_size;
 768                 LASSERT(!(ma->ma_valid & MA_LOV));
 769         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 770                 buf->lb_buf = ma->ma_lmv;
 771                 buf->lb_len = ma->ma_lmv_size;
 772                 LASSERT(!(ma->ma_valid & MA_LMV));
 773         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 774                 buf->lb_buf = ma->ma_lmv;
 775                 buf->lb_len = ma->ma_lmv_size;
 776                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 777         } else {
 778                 return -EINVAL;
 779         }
 780
 781         rc = mo_xattr_get(info->mti_env, next, buf, name);
 782         if (rc > 0) {
 783
 784 got:
 785                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 786                         if (info->mti_big_lmm_used)
 787                                 ma->ma_lmm = info->mti_big_lmm;
 788
 789                         /* NOT return LOV EA with hole to old client. */
 790                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 791                                      LOV_PATTERN_F_HOLE) &&
 792                             !(exp_connect_flags(info->mti_exp) &
 793                               OBD_CONNECT_LFSCK)) {
 794                                 return -EIO;
 795                         } else {
 796                                 ma->ma_lmm_size = rc;
 797                                 ma->ma_valid |= MA_LOV;
 798                         }
 799                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 800                         if (info->mti_big_lmm_used)
 801                                 ma->ma_lmv = info->mti_big_lmm;
 802
 803                         ma->ma_lmv_size = rc;
 804                         ma->ma_valid |= MA_LMV;
 805                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 806                         ma->ma_lmv_size = rc;
 807                         ma->ma_valid |= MA_LMV_DEF;
 808                 }
 809
 810                 /* Update mdt_max_mdsize so all clients will be aware that */
 811                 if (info->mti_mdt->mdt_max_mdsize < rc)
 812                         info->mti_mdt->mdt_max_mdsize = rc;
 813
 814                 rc = 0;
 815         } else if (rc == -ENODATA) {
 816                 /* no LOV EA */
 817                 rc = 0;
 818         } else if (rc == -ERANGE) {
 819                 /* Default LMV has fixed size, so it must be able to fit
 820                  * in the original buffer */
 821                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 822                         return rc;
 823                 rc = mdt_big_xattr_get(info, o, name);
 824                 if (rc > 0) {
 825                         info->mti_big_lmm_used = 1;
 826                         goto got;
 827                 }
 828         }
 829
 830         return rc;
 831 }
 832
 833 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 834                              struct mdt_object *o, struct lu_fid *pfid)
 835 {
 836         struct lu_buf           *buf = &info->mti_buf;
 837         struct link_ea_header   *leh;
 838         struct link_ea_entry    *lee;
 839         int                      rc;
 840         ENTRY;
 841
 842         buf->lb_buf = info->mti_big_lmm;
 843         buf->lb_len = info->mti_big_lmmsize;
 844         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 845                           buf, XATTR_NAME_LINK);
 846         /* ignore errors, MA_PFID won't be set and it is
 847          * up to the caller to treat this as an error */
 848         if (rc == -ERANGE || buf->lb_len == 0) {
 849                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 850                 buf->lb_buf = info->mti_big_lmm;
 851                 buf->lb_len = info->mti_big_lmmsize;
 852         }
 853
 854         if (rc < 0)
 855                 RETURN(rc);
 856         if (rc < sizeof(*leh)) {
 857                 CERROR("short LinkEA on "DFID": rc = %d\n",
 858                        PFID(mdt_object_fid(o)), rc);
 859                 RETURN(-ENODATA);
 860         }
 861
 862         leh = (struct link_ea_header *) buf->lb_buf;
 863         lee = (struct link_ea_entry *)(leh + 1);
 864         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 865                 leh->leh_magic = LINK_EA_MAGIC;
 866                 leh->leh_reccount = __swab32(leh->leh_reccount);
 867                 leh->leh_len = __swab64(leh->leh_len);
 868         }
 869         if (leh->leh_magic != LINK_EA_MAGIC)
 870                 RETURN(-EINVAL);
 871         if (leh->leh_reccount == 0)
 872                 RETURN(-ENODATA);
 873
 874         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 875         fid_be_to_cpu(pfid, pfid);
 876
 877         RETURN(0);
 878 }
 879
 880 int mdt_attr_get_complex(struct mdt_thread_info *info,
 881                          struct mdt_object *o, struct md_attr *ma)
 882 {
 883         const struct lu_env *env = info->mti_env;
 884         struct md_object    *next = mdt_object_child(o);
 885         struct lu_buf       *buf = &info->mti_buf;
 886         int                  need = ma->ma_need;
 887         int                  rc = 0, rc2;
 888         u32                  mode;
 889         ENTRY;
 890
 891         ma->ma_valid = 0;
 892
 893         if (mdt_object_exists(o) == 0)
 894                 GOTO(out, rc = -ENOENT);
 895         mode = lu_object_attr(&next->mo_lu);
 896
 897         if (need & MA_INODE) {
 898                 ma->ma_need = MA_INODE;
 899                 rc = mo_attr_get(env, next, ma);
 900                 if (rc)
 901                         GOTO(out, rc);
 902                 ma->ma_valid |= MA_INODE;
 903         }
 904
 905         if (need & MA_PFID) {
 906                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 907                 if (rc == 0)
 908                         ma->ma_valid |= MA_PFID;
 909                 /* ignore this error, parent fid is not mandatory */
 910                 rc = 0;
 911         }
 912
 913         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 914                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 915                 if (rc)
 916                         GOTO(out, rc);
 917         }
 918
 919         if (need & MA_LMV && S_ISDIR(mode)) {
 920                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 921                 if (rc != 0)
 922                         GOTO(out, rc);
 923         }
 924
 925         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 926                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 927                 if (rc != 0)
 928                         GOTO(out, rc);
 929         }
 930
 931         if (need & MA_HSM && S_ISREG(mode)) {
 932                 buf->lb_buf = info->mti_xattr_buf;
 933                 buf->lb_len = sizeof(info->mti_xattr_buf);
 934                 CLASSERT(sizeof(struct hsm_attrs) <=
 935                          sizeof(info->mti_xattr_buf));
 936                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 937                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 938                 if (rc2 == 0)
 939                         ma->ma_valid |= MA_HSM;
 940                 else if (rc2 < 0 && rc2 != -ENODATA)
 941                         GOTO(out, rc = rc2);
 942         }
 943
 944 #ifdef CONFIG_FS_POSIX_ACL
 945         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 946                 buf->lb_buf = ma->ma_acl;
 947                 buf->lb_len = ma->ma_acl_size;
 948                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 949                 if (rc2 > 0) {
 950                         ma->ma_acl_size = rc2;
 951                         ma->ma_valid |= MA_ACL_DEF;
 952                 } else if (rc2 == -ENODATA) {
 953                         /* no ACLs */
 954                         ma->ma_acl_size = 0;
 955                 } else
 956                         GOTO(out, rc = rc2);
 957         }
 958 #endif
 959 out:
 960         ma->ma_need = need;
 961         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
 962                rc, ma->ma_valid, ma->ma_lmm);
 963         RETURN(rc);
 964 }
 965
 966 static int mdt_getattr_internal(struct mdt_thread_info *info,
 967                                 struct mdt_object *o, int ma_need)
 968 {
 969         struct md_object        *next = mdt_object_child(o);
 970         const struct mdt_body   *reqbody = info->mti_body;
 971         struct ptlrpc_request   *req = mdt_info_req(info);
 972         struct md_attr          *ma = &info->mti_attr;
 973         struct lu_attr          *la = &ma->ma_attr;
 974         struct req_capsule      *pill = info->mti_pill;
 975         const struct lu_env     *env = info->mti_env;
 976         struct mdt_body         *repbody;
 977         struct lu_buf           *buffer = &info->mti_buf;
 978         struct obd_export       *exp = info->mti_exp;
 979         int                      rc;
 980         ENTRY;
 981
 982         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 983                 RETURN(err_serious(-ENOMEM));
 984
 985         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 986
 987         ma->ma_valid = 0;
 988
 989         if (mdt_object_remote(o)) {
 990                 /* This object is located on remote node.*/
 991                 /* Return -ENOTSUPP for old client */
 992                 if (!mdt_is_dne_client(req->rq_export))
 993                         GOTO(out, rc = -ENOTSUPP);
 994
 995                 repbody->mbo_fid1 = *mdt_object_fid(o);
 996                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 997                 GOTO(out, rc = 0);
 998         }
 999
1000         if (reqbody->mbo_eadatasize > 0) {
1001                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1002                 if (buffer->lb_buf == NULL)
1003                         GOTO(out, rc = -EPROTO);
1004                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1005                                                       RCL_SERVER);
1006         } else {
1007                 buffer->lb_buf = NULL;
1008                 buffer->lb_len = 0;
1009                 ma_need &= ~(MA_LOV | MA_LMV);
1010                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1011                        mdt_obd_name(info->mti_mdt),
1012                        req->rq_export->exp_client_uuid.uuid);
1013         }
1014
1015         /* If it is dir object and client require MEA, then we got MEA */
1016         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1017             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1018                 /* Assumption: MDT_MD size is enough for lmv size. */
1019                 ma->ma_lmv = buffer->lb_buf;
1020                 ma->ma_lmv_size = buffer->lb_len;
1021                 ma->ma_need = MA_INODE;
1022                 if (ma->ma_lmv_size > 0) {
1023                         if (reqbody->mbo_valid & OBD_MD_MEA)
1024                                 ma->ma_need |= MA_LMV;
1025                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1026                                 ma->ma_need |= MA_LMV_DEF;
1027                 }
1028         } else {
1029                 ma->ma_lmm = buffer->lb_buf;
1030                 ma->ma_lmm_size = buffer->lb_len;
1031                 ma->ma_need = MA_INODE | MA_HSM;
1032                 if (ma->ma_lmm_size > 0)
1033                         ma->ma_need |= MA_LOV;
1034         }
1035
1036         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1037             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1038             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1039                 /* get default stripe info for this dir. */
1040                 ma->ma_need |= MA_LOV_DEF;
1041         }
1042         ma->ma_need |= ma_need;
1043
1044         rc = mdt_attr_get_complex(info, o, ma);
1045         if (unlikely(rc)) {
1046                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1047                        "%s: getattr error for "DFID": rc = %d\n",
1048                        mdt_obd_name(info->mti_mdt),
1049                        PFID(mdt_object_fid(o)), rc);
1050                 RETURN(rc);
1051         }
1052
1053         /* if file is released, check if a restore is running */
1054         if (ma->ma_valid & MA_HSM) {
1055                 repbody->mbo_valid |= OBD_MD_TSTATE;
1056                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1057                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1058                         repbody->mbo_t_state = MS_RESTORE;
1059         }
1060
1061         if (likely(ma->ma_valid & MA_INODE))
1062                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1063         else
1064                 RETURN(-EFAULT);
1065
1066         if (mdt_body_has_lov(la, reqbody)) {
1067                 if (ma->ma_valid & MA_LOV) {
1068                         LASSERT(ma->ma_lmm_size);
1069                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1070                         if (S_ISDIR(la->la_mode))
1071                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1072                         else
1073                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1074                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1075                 }
1076                 if (ma->ma_valid & MA_LMV) {
1077                         /* Return -ENOTSUPP for old client */
1078                         if (!mdt_is_striped_client(req->rq_export))
1079                                 RETURN(-ENOTSUPP);
1080
1081                         LASSERT(S_ISDIR(la->la_mode));
1082                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1083                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1084                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1085                 }
1086                 if (ma->ma_valid & MA_LMV_DEF) {
1087                         /* Return -ENOTSUPP for old client */
1088                         if (!mdt_is_striped_client(req->rq_export))
1089                                 RETURN(-ENOTSUPP);
1090                         LASSERT(S_ISDIR(la->la_mode));
1091                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1092                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1093                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1094                                                OBD_MD_DEFAULT_MEA);
1095                 }
1096         } else if (S_ISLNK(la->la_mode) &&
1097                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1098                 buffer->lb_buf = ma->ma_lmm;
1099                 /* eadatasize from client includes NULL-terminator, so
1100                  * there is no need to read it */
1101                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1102                 rc = mo_readlink(env, next, buffer);
1103                 if (unlikely(rc <= 0)) {
1104                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1105                                mdt_obd_name(info->mti_mdt),
1106                                PFID(mdt_object_fid(o)), rc);
1107                         rc = -EFAULT;
1108                 } else {
1109                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1110
1111                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1112                                 rc -= 2;
1113                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1114                         /* we need to report back size with NULL-terminator
1115                          * because client expects that */
1116                         repbody->mbo_eadatasize = rc + 1;
1117                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1118                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1119                                        "on "DFID ", expected %d\n",
1120                                        mdt_obd_name(info->mti_mdt),
1121                                        rc, PFID(mdt_object_fid(o)),
1122                                        reqbody->mbo_eadatasize - 1);
1123                         /* NULL terminate */
1124                         ((char *)ma->ma_lmm)[rc] = 0;
1125
1126                         /* If the total CDEBUG() size is larger than a page, it
1127                          * will print a warning to the console, avoid this by
1128                          * printing just the last part of the symlink. */
1129                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1130                                print_limit < rc ? "..." : "", print_limit,
1131                                (char *)ma->ma_lmm + rc - print_limit, rc);
1132                         rc = 0;
1133                 }
1134         }
1135
1136         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1137                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1138                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1139                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1140                        repbody->mbo_max_mdsize);
1141         }
1142
1143 #ifdef CONFIG_FS_POSIX_ACL
1144         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1145                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1146                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1147                 if (IS_ERR(nodemap))
1148                         RETURN(PTR_ERR(nodemap));
1149
1150                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1151                 nodemap_putref(nodemap);
1152         }
1153 #endif
1154
1155 out:
1156         if (rc == 0)
1157                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1158
1159         RETURN(rc);
1160 }
1161
1162 static int mdt_getattr(struct tgt_session_info *tsi)
1163 {
1164         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1165         struct mdt_object       *obj = info->mti_object;
1166         struct req_capsule      *pill = info->mti_pill;
1167         struct mdt_body         *reqbody;
1168         struct mdt_body         *repbody;
1169         int rc, rc2;
1170         ENTRY;
1171
1172         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1173         LASSERT(reqbody);
1174         LASSERT(obj != NULL);
1175         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1176
1177         /* Unlike intent case where we need to pre-fill out buffers early on
1178          * in intent policy for ldlm reasons, here we can have a much better
1179          * guess at EA size by just reading it from disk.
1180          * Exceptions are readdir and (missing) directory striping */
1181         /* Readlink */
1182         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1183                 /* No easy way to know how long is the symlink, but it cannot
1184                  * be more than PATH_MAX, so we allocate +1 */
1185                 rc = PATH_MAX + 1;
1186
1187         /* A special case for fs ROOT: getattr there might fetch
1188          * default EA for entire fs, not just for this dir!
1189          */
1190         } else if (lu_fid_eq(mdt_object_fid(obj),
1191                              &info->mti_mdt->mdt_md_root_fid) &&
1192                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1193                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1194                                                                  MDS_GETATTR)) {
1195                 /* Should the default strping be bigger, mdt_fix_reply
1196                  * will reallocate */
1197                 rc = DEF_REP_MD_SIZE;
1198         } else {
1199                 /* Read the actual EA size from disk */
1200                 rc = mdt_attr_get_eabuf_size(info, obj);
1201         }
1202
1203         if (rc < 0)
1204                 GOTO(out, rc = err_serious(rc));
1205
1206         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1207
1208         rc = req_capsule_server_pack(pill);
1209         if (unlikely(rc != 0))
1210                 GOTO(out, rc = err_serious(rc));
1211
1212         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1213         LASSERT(repbody != NULL);
1214         repbody->mbo_eadatasize = 0;
1215         repbody->mbo_aclsize = 0;
1216
1217         rc = mdt_check_ucred(info);
1218         if (unlikely(rc))
1219                 GOTO(out_shrink, rc);
1220
1221         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1222
1223         rc = mdt_getattr_internal(info, obj, 0);
1224         EXIT;
1225 out_shrink:
1226         mdt_client_compatibility(info);
1227         rc2 = mdt_fix_reply(info);
1228         if (rc == 0)
1229                 rc = rc2;
1230 out:
1231         mdt_thread_info_fini(info);
1232         return rc;
1233 }
1234
1235 /**
1236  * Handler of layout intent RPC requiring the layout modification
1237  *
1238  * \param[in] info      thread environment
1239  * \param[in] obj       object
1240  * \param[in] layout    layout intent
1241  * \param[in] buf       buffer containing client's lovea, could be empty
1242  *
1243  * \retval 0    on success
1244  * \retval < 0  error code
1245  */
1246 static int mdt_layout_change(struct mdt_thread_info *info,
1247                              struct mdt_object *obj,
1248                              struct layout_intent *layout,
1249                              const struct lu_buf *buf)
1250 {
1251         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1252         int rc;
1253         ENTRY;
1254
1255         CDEBUG(D_INFO, "got layout change request from client: "
1256                "opc:%u flags:%#x extent[%#llx,%#llx)\n",
1257                layout->li_opc, layout->li_flags,
1258                layout->li_start, layout->li_end);
1259         if (layout->li_start >= layout->li_end) {
1260                 CERROR("Recieved an invalid layout change range [%llu, %llu) "
1261                        "for "DFID"\n", layout->li_start, layout->li_end,
1262                        PFID(mdt_object_fid(obj)));
1263                 RETURN(-EINVAL);
1264         }
1265
1266         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1267                 GOTO(out, rc = -EINVAL);
1268
1269         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1270                            MAY_WRITE);
1271         if (rc)
1272                 GOTO(out, rc);
1273
1274         /* take layout lock to prepare layout change */
1275         mdt_lock_reg_init(lh, LCK_EX);
1276         rc = mdt_object_lock(info, obj, lh,
1277                              MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1278         if (rc)
1279                 GOTO(out, rc);
1280
1281         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout,
1282                               buf);
1283
1284         mdt_object_unlock(info, obj, lh, 1);
1285 out:
1286         RETURN(rc);
1287 }
1288
1289 /**
1290  * Exchange MOF_LOV_CREATED flags between two objects after a
1291  * layout swap. No assumption is made on whether o1 or o2 have
1292  * created objects or not.
1293  *
1294  * \param[in,out] o1    First swap layout object
1295  * \param[in,out] o2    Second swap layout object
1296  */
1297 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1298 {
1299         unsigned int o1_lov_created = o1->mot_lov_created;
1300
1301         mutex_lock(&o1->mot_lov_mutex);
1302         mutex_lock(&o2->mot_lov_mutex);
1303
1304         o1->mot_lov_created = o2->mot_lov_created;
1305         o2->mot_lov_created = o1_lov_created;
1306
1307         mutex_unlock(&o2->mot_lov_mutex);
1308         mutex_unlock(&o1->mot_lov_mutex);
1309 }
1310
1311 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1312 {
1313         struct mdt_thread_info  *info;
1314         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1315         struct obd_export       *exp = req->rq_export;
1316         struct mdt_object       *o1, *o2, *o;
1317         struct mdt_lock_handle  *lh1, *lh2;
1318         struct mdc_swap_layouts *msl;
1319         int                      rc;
1320         ENTRY;
1321
1322         /* client does not support layout lock, so layout swaping
1323          * is disabled.
1324          * FIXME: there is a problem for old clients which don't support
1325          * layout lock yet. If those clients have already opened the file
1326          * they won't be notified at all so that old layout may still be
1327          * used to do IO. This can be fixed after file release is landed by
1328          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1329         if (!exp_connect_layout(exp))
1330                 RETURN(-EOPNOTSUPP);
1331
1332         info = tsi2mdt_info(tsi);
1333
1334         if (info->mti_dlm_req != NULL)
1335                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1336
1337         o1 = info->mti_object;
1338         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1339                                 &info->mti_body->mbo_fid2);
1340         if (IS_ERR(o))
1341                 GOTO(out, rc = PTR_ERR(o));
1342
1343         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1344                 GOTO(put, rc = -ENOENT);
1345
1346         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1347         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1348                 GOTO(put, rc);
1349
1350         if (rc < 0)
1351                 swap(o1, o2);
1352
1353         /* permission check. Make sure the calling process having permission
1354          * to write both files. */
1355         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1356                                 MAY_WRITE);
1357         if (rc < 0)
1358                 GOTO(put, rc);
1359
1360         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1361                                 MAY_WRITE);
1362         if (rc < 0)
1363                 GOTO(put, rc);
1364
1365         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1366         if (msl == NULL)
1367                 GOTO(put, rc = -EPROTO);
1368
1369         lh1 = &info->mti_lh[MDT_LH_NEW];
1370         mdt_lock_reg_init(lh1, LCK_EX);
1371         lh2 = &info->mti_lh[MDT_LH_OLD];
1372         mdt_lock_reg_init(lh2, LCK_EX);
1373
1374         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1375                              MDS_INODELOCK_XATTR);
1376         if (rc < 0)
1377                 GOTO(put, rc);
1378
1379         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1380                              MDS_INODELOCK_XATTR);
1381         if (rc < 0)
1382                 GOTO(unlock1, rc);
1383
1384         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1385                              mdt_object_child(o2), msl->msl_flags);
1386         if (rc < 0)
1387                 GOTO(unlock2, rc);
1388
1389         mdt_swap_lov_flag(o1, o2);
1390
1391 unlock2:
1392         mdt_object_unlock(info, o2, lh2, rc);
1393 unlock1:
1394         mdt_object_unlock(info, o1, lh1, rc);
1395 put:
1396         mdt_object_put(info->mti_env, o);
1397 out:
1398         mdt_thread_info_fini(info);
1399         RETURN(rc);
1400 }
1401
1402 static int mdt_raw_lookup(struct mdt_thread_info *info,
1403                           struct mdt_object *parent,
1404                           const struct lu_name *lname,
1405                           struct ldlm_reply *ldlm_rep)
1406 {
1407         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1408         int              rc;
1409         ENTRY;
1410
1411         LASSERT(!info->mti_cross_ref);
1412
1413         /* Only got the fid of this obj by name */
1414         fid_zero(child_fid);
1415         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1416                         lname, child_fid, &info->mti_spec);
1417         if (rc == 0) {
1418                 struct mdt_body *repbody;
1419
1420                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1421                 repbody->mbo_fid1 = *child_fid;
1422                 repbody->mbo_valid = OBD_MD_FLID;
1423                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1424         } else if (rc == -ENOENT) {
1425                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1426         }
1427
1428         RETURN(rc);
1429 }
1430
1431 /*
1432  * UPDATE lock should be taken against parent, and be released before exit;
1433  * child_bits lock should be taken against child, and be returned back:
1434  *            (1)normal request should release the child lock;
1435  *            (2)intent request will grant the lock to client.
1436  */
1437 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1438                                  struct mdt_lock_handle *lhc,
1439                                  __u64 child_bits,
1440                                  struct ldlm_reply *ldlm_rep)
1441 {
1442         struct ptlrpc_request  *req       = mdt_info_req(info);
1443         struct mdt_body        *reqbody   = NULL;
1444         struct mdt_object      *parent    = info->mti_object;
1445         struct mdt_object      *child;
1446         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1447         struct lu_name         *lname     = NULL;
1448         struct mdt_lock_handle *lhp       = NULL;
1449         struct ldlm_lock       *lock;
1450         bool                    is_resent;
1451         bool                    try_layout;
1452         int                     ma_need = 0;
1453         int                     rc;
1454         ENTRY;
1455
1456         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1457         LASSERT(ergo(is_resent,
1458                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1459
1460         if (parent == NULL)
1461                 RETURN(-ENOENT);
1462
1463         if (info->mti_cross_ref) {
1464                 /* Only getattr on the child. Parent is on another node. */
1465                 mdt_set_disposition(info, ldlm_rep,
1466                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1467                 child = parent;
1468                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1469                        "ldlm_rep = %p\n",
1470                        PFID(mdt_object_fid(child)), ldlm_rep);
1471
1472                 rc = mdt_check_resent_lock(info, child, lhc);
1473                 if (rc < 0) {
1474                         RETURN(rc);
1475                 } else if (rc > 0) {
1476                         mdt_lock_handle_init(lhc);
1477                         mdt_lock_reg_init(lhc, LCK_PR);
1478
1479                         /*
1480                          * Object's name is on another MDS, no lookup or layout
1481                          * lock is needed here but update lock is.
1482                          */
1483                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1484                                         MDS_INODELOCK_LAYOUT);
1485                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1486
1487                         rc = mdt_object_lock(info, child, lhc, child_bits);
1488                         if (rc < 0)
1489                                 RETURN(rc);
1490                 }
1491
1492                 /* Finally, we can get attr for child. */
1493                 if (!mdt_object_exists(child)) {
1494                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1495                                         &child->mot_obj,
1496                                         "remote object doesn't exist.");
1497                         mdt_object_unlock(info, child, lhc, 1);
1498                         RETURN(-ENOENT);
1499                 }
1500
1501                 rc = mdt_getattr_internal(info, child, 0);
1502                 if (unlikely(rc != 0))
1503                         mdt_object_unlock(info, child, lhc, 1);
1504
1505                 RETURN(rc);
1506         }
1507
1508         lname = &info->mti_name;
1509         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1510
1511         if (lu_name_is_valid(lname)) {
1512                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1513                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1514                        PNAME(lname), ldlm_rep);
1515         } else {
1516                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1517                 if (unlikely(reqbody == NULL))
1518                         RETURN(err_serious(-EPROTO));
1519
1520                 *child_fid = reqbody->mbo_fid2;
1521
1522                 if (unlikely(!fid_is_sane(child_fid)))
1523                         RETURN(err_serious(-EINVAL));
1524
1525                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1526                        "ldlm_rep = %p\n",
1527                        PFID(mdt_object_fid(parent)),
1528                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1529         }
1530
1531         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1532
1533         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1534                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1535                                 &parent->mot_obj,
1536                                 "Parent doesn't exist!");
1537                 RETURN(-ESTALE);
1538         }
1539
1540         if (mdt_object_remote(parent)) {
1541                 CERROR("%s: parent "DFID" is on remote target\n",
1542                        mdt_obd_name(info->mti_mdt),
1543                        PFID(mdt_object_fid(parent)));
1544                 RETURN(-EIO);
1545         }
1546
1547         if (lu_name_is_valid(lname)) {
1548                 /* Always allow to lookup ".." */
1549                 if (unlikely(lname->ln_namelen == 2 &&
1550                              lname->ln_name[0] == '.' &&
1551                              lname->ln_name[1] == '.'))
1552                         info->mti_spec.sp_permitted = 1;
1553
1554                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1555                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1556
1557                         RETURN(rc);
1558                 }
1559
1560                 /* step 1: lock parent only if parent is a directory */
1561                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1562                         lhp = &info->mti_lh[MDT_LH_PARENT];
1563                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1564                         rc = mdt_object_lock(info, parent, lhp,
1565                                              MDS_INODELOCK_UPDATE);
1566                         if (unlikely(rc != 0))
1567                                 RETURN(rc);
1568                 }
1569
1570                 /* step 2: lookup child's fid by name */
1571                 fid_zero(child_fid);
1572                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1573                                 child_fid, &info->mti_spec);
1574                 if (rc == -ENOENT)
1575                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1576
1577                 if (rc != 0)
1578                         GOTO(out_parent, rc);
1579         }
1580
1581         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1582
1583         /*
1584          *step 3: find the child object by fid & lock it.
1585          *        regardless if it is local or remote.
1586          *
1587          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1588          *      set parent dir fid the same as child fid in getattr by fid case
1589          *      we should not lu_object_find() the object again, could lead
1590          *      to hung if there is a concurrent unlink destroyed the object.
1591          */
1592         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1593                 mdt_object_get(info->mti_env, parent);
1594                 child = parent;
1595         } else {
1596                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1597                                         child_fid);
1598         }
1599
1600         if (unlikely(IS_ERR(child)))
1601                 GOTO(out_parent, rc = PTR_ERR(child));
1602
1603         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1604         if (!mdt_object_exists(child)) {
1605                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1606                                 &child->mot_obj,
1607                                 "Object doesn't exist!");
1608                 GOTO(out_child, rc = -ENOENT);
1609         }
1610
1611         rc = mdt_check_resent_lock(info, child, lhc);
1612         if (rc < 0) {
1613                 GOTO(out_child, rc);
1614         } else if (rc > 0) {
1615                 mdt_lock_handle_init(lhc);
1616                 mdt_lock_reg_init(lhc, LCK_PR);
1617                 try_layout = false;
1618
1619                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1620                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1621                         struct md_attr *ma = &info->mti_attr;
1622
1623                         ma->ma_valid = 0;
1624                         ma->ma_need = MA_INODE;
1625                         rc = mdt_attr_get_complex(info, child, ma);
1626                         if (unlikely(rc != 0))
1627                                 GOTO(out_child, rc);
1628
1629                         /* If the file has not been changed for some time, we
1630                          * return not only a LOOKUP lock, but also an UPDATE
1631                          * lock and this might save us RPC on later STAT. For
1632                          * directories, it also let negative dentry cache start
1633                          * working for this dir. */
1634                         if (ma->ma_valid & MA_INODE &&
1635                             ma->ma_attr.la_valid & LA_CTIME &&
1636                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1637                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1638                                 child_bits |= MDS_INODELOCK_UPDATE;
1639                 }
1640
1641                 /* layout lock must be granted in a best-effort way
1642                  * for IT operations */
1643                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1644                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1645                     exp_connect_layout(info->mti_exp) &&
1646                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1647                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1648                         /* try to grant layout lock for regular file. */
1649                         try_layout = true;
1650                 }
1651
1652                 rc = 0;
1653                 if (try_layout) {
1654                         child_bits |= MDS_INODELOCK_LAYOUT;
1655                         /* try layout lock, it may fail to be granted due to
1656                          * contention at LOOKUP or UPDATE */
1657                         if (!mdt_object_lock_try(info, child, lhc,
1658                                                  child_bits)) {
1659                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1660                                 LASSERT(child_bits != 0);
1661                                 rc = mdt_object_lock(info, child, lhc,
1662                                                      child_bits);
1663                         } else {
1664                                 ma_need |= MA_LOV;
1665                         }
1666                 } else {
1667                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1668                          * client will enqueue the lock to the remote MDT */
1669                         if (mdt_object_remote(child))
1670                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1671                         rc = mdt_object_lock(info, child, lhc, child_bits);
1672                 }
1673                 if (unlikely(rc != 0))
1674                         GOTO(out_child, rc);
1675         }
1676
1677         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1678
1679         /* finally, we can get attr for child. */
1680         rc = mdt_getattr_internal(info, child, ma_need);
1681         if (unlikely(rc != 0)) {
1682                 mdt_object_unlock(info, child, lhc, 1);
1683         } else if (lock) {
1684                 /* Debugging code. */
1685                 LDLM_DEBUG(lock, "Returning lock to client");
1686                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1687                                          &lock->l_resource->lr_name),
1688                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1689                          PLDLMRES(lock->l_resource),
1690                          PFID(mdt_object_fid(child)));
1691         }
1692         if (lock)
1693                 LDLM_LOCK_PUT(lock);
1694
1695         EXIT;
1696 out_child:
1697         mdt_object_put(info->mti_env, child);
1698 out_parent:
1699         if (lhp)
1700                 mdt_object_unlock(info, parent, lhp, 1);
1701         return rc;
1702 }
1703
1704 /* normal handler: should release the child lock */
1705 static int mdt_getattr_name(struct tgt_session_info *tsi)
1706 {
1707         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1708         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1709         struct mdt_body        *reqbody;
1710         struct mdt_body        *repbody;
1711         int rc, rc2;
1712         ENTRY;
1713
1714         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1715         LASSERT(reqbody != NULL);
1716         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1717         LASSERT(repbody != NULL);
1718
1719         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1720         repbody->mbo_eadatasize = 0;
1721         repbody->mbo_aclsize = 0;
1722
1723         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1724         if (unlikely(rc))
1725                 GOTO(out_shrink, rc);
1726
1727         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1728         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1729                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1730                 lhc->mlh_reg_lh.cookie = 0;
1731         }
1732         mdt_exit_ucred(info);
1733         EXIT;
1734 out_shrink:
1735         mdt_client_compatibility(info);
1736         rc2 = mdt_fix_reply(info);
1737         if (rc == 0)
1738                 rc = rc2;
1739         mdt_thread_info_fini(info);
1740         return rc;
1741 }
1742
1743 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1744                          void *karg, void __user *uarg);
1745
1746 static int mdt_set_info(struct tgt_session_info *tsi)
1747 {
1748         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1749         char                    *key;
1750         void                    *val;
1751         int                      keylen, vallen, rc = 0;
1752
1753         ENTRY;
1754
1755         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1756         if (key == NULL) {
1757                 DEBUG_REQ(D_HA, req, "no set_info key");
1758                 RETURN(err_serious(-EFAULT));
1759         }
1760
1761         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1762                                       RCL_CLIENT);
1763
1764         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1765         if (val == NULL) {
1766                 DEBUG_REQ(D_HA, req, "no set_info val");
1767                 RETURN(err_serious(-EFAULT));
1768         }
1769
1770         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1771                                       RCL_CLIENT);
1772
1773         /* Swab any part of val you need to here */
1774         if (KEY_IS(KEY_READ_ONLY)) {
1775                 spin_lock(&req->rq_export->exp_lock);
1776                 if (*(__u32 *)val)
1777                         *exp_connect_flags_ptr(req->rq_export) |=
1778                                 OBD_CONNECT_RDONLY;
1779                 else
1780                         *exp_connect_flags_ptr(req->rq_export) &=
1781                                 ~OBD_CONNECT_RDONLY;
1782                 spin_unlock(&req->rq_export->exp_lock);
1783         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1784                 struct changelog_setinfo *cs = val;
1785
1786                 if (vallen != sizeof(*cs)) {
1787                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1788                                tgt_name(tsi->tsi_tgt), vallen);
1789                         RETURN(-EINVAL);
1790                 }
1791                 if (ptlrpc_req_need_swab(req)) {
1792                         __swab64s(&cs->cs_recno);
1793                         __swab32s(&cs->cs_id);
1794                 }
1795
1796                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1797                                    vallen, val, NULL);
1798         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1799                 if (vallen > 0)
1800                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1801         } else {
1802                 RETURN(-EINVAL);
1803         }
1804         RETURN(rc);
1805 }
1806
1807 static int mdt_readpage(struct tgt_session_info *tsi)
1808 {
1809         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1810         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1811         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1812         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1813         struct mdt_body         *repbody;
1814         int                      rc;
1815         int                      i;
1816
1817         ENTRY;
1818
1819         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1820                 RETURN(err_serious(-ENOMEM));
1821
1822         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1823         if (repbody == NULL || reqbody == NULL)
1824                 RETURN(err_serious(-EFAULT));
1825
1826         /*
1827          * prepare @rdpg before calling lower layers and transfer itself. Here
1828          * reqbody->size contains offset of where to start to read and
1829          * reqbody->nlink contains number bytes to read.
1830          */
1831         rdpg->rp_hash = reqbody->mbo_size;
1832         if (rdpg->rp_hash != reqbody->mbo_size) {
1833                 CERROR("Invalid hash: %#llx != %#llx\n",
1834                        rdpg->rp_hash, reqbody->mbo_size);
1835                 RETURN(-EFAULT);
1836         }
1837
1838         rdpg->rp_attrs = reqbody->mbo_mode;
1839         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1840                 rdpg->rp_attrs |= LUDA_64BITHASH;
1841         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1842                                 exp_max_brw_size(tsi->tsi_exp));
1843         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1844                           PAGE_SHIFT;
1845         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1846         if (rdpg->rp_pages == NULL)
1847                 RETURN(-ENOMEM);
1848
1849         for (i = 0; i < rdpg->rp_npages; ++i) {
1850                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1851                 if (rdpg->rp_pages[i] == NULL)
1852                         GOTO(free_rdpg, rc = -ENOMEM);
1853         }
1854
1855         /* call lower layers to fill allocated pages with directory data */
1856         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1857         if (rc < 0)
1858                 GOTO(free_rdpg, rc);
1859
1860         /* send pages to client */
1861         rc = tgt_sendpage(tsi, rdpg, rc);
1862
1863         EXIT;
1864 free_rdpg:
1865
1866         for (i = 0; i < rdpg->rp_npages; i++)
1867                 if (rdpg->rp_pages[i] != NULL)
1868                         __free_page(rdpg->rp_pages[i]);
1869         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1870
1871         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1872                 RETURN(0);
1873
1874         return rc;
1875 }
1876
1877 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1878 {
1879         struct lu_ucred *uc = mdt_ucred_check(info);
1880         struct lu_attr *attr = &info->mti_attr.ma_attr;
1881
1882         if (uc == NULL)
1883                 return -EINVAL;
1884
1885         if (op != REINT_SETATTR) {
1886                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1887                         attr->la_uid = uc->uc_fsuid;
1888                 /* for S_ISGID, inherit gid from his parent, such work will be
1889                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1890                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1891                         attr->la_gid = uc->uc_fsgid;
1892         }
1893
1894         return 0;
1895 }
1896
1897 static int mdt_reint_internal(struct mdt_thread_info *info,
1898                               struct mdt_lock_handle *lhc,
1899                               __u32 op)
1900 {
1901         struct req_capsule      *pill = info->mti_pill;
1902         struct mdt_body         *repbody;
1903         int                      rc = 0, rc2;
1904
1905         ENTRY;
1906
1907         rc = mdt_reint_unpack(info, op);
1908         if (rc != 0) {
1909                 CERROR("Can't unpack reint, rc %d\n", rc);
1910                 RETURN(err_serious(rc));
1911         }
1912
1913         /* for replay (no_create) lmm is not needed, client has it already */
1914         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1915                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1916                                      DEF_REP_MD_SIZE);
1917
1918         /* llog cookies are always 0, the field is kept for compatibility */
1919         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1920                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1921
1922         rc = req_capsule_server_pack(pill);
1923         if (rc != 0) {
1924                 CERROR("Can't pack response, rc %d\n", rc);
1925                 RETURN(err_serious(rc));
1926         }
1927
1928         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1929                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1930                 LASSERT(repbody);
1931                 repbody->mbo_eadatasize = 0;
1932                 repbody->mbo_aclsize = 0;
1933         }
1934
1935         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1936
1937         /* for replay no cookkie / lmm need, because client have this already */
1938         if (info->mti_spec.no_create)
1939                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1940                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1941
1942         rc = mdt_init_ucred_reint(info);
1943         if (rc)
1944                 GOTO(out_shrink, rc);
1945
1946         rc = mdt_fix_attr_ucred(info, op);
1947         if (rc != 0)
1948                 GOTO(out_ucred, rc = err_serious(rc));
1949
1950         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1951         if (rc < 0) {
1952                 GOTO(out_ucred, rc);
1953         } else if (rc == 1) {
1954                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1955                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1956                 GOTO(out_ucred, rc);
1957         }
1958         rc = mdt_reint_rec(info, lhc);
1959         EXIT;
1960 out_ucred:
1961         mdt_exit_ucred(info);
1962 out_shrink:
1963         mdt_client_compatibility(info);
1964         rc2 = mdt_fix_reply(info);
1965         if (rc == 0)
1966                 rc = rc2;
1967         return rc;
1968 }
1969
1970 static long mdt_reint_opcode(struct ptlrpc_request *req,
1971                              const struct req_format **fmt)
1972 {
1973         struct mdt_device       *mdt;
1974         struct mdt_rec_reint    *rec;
1975         long                     opc;
1976
1977         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1978         if (rec != NULL) {
1979                 opc = rec->rr_opcode;
1980                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1981                 if (opc < REINT_MAX && fmt[opc] != NULL)
1982                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1983                 else {
1984                         mdt = mdt_exp2dev(req->rq_export);
1985                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1986                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1987                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1988                         opc = err_serious(-EFAULT);
1989                 }
1990         } else {
1991                 opc = err_serious(-EFAULT);
1992         }
1993         return opc;
1994 }
1995
1996 static int mdt_reint(struct tgt_session_info *tsi)
1997 {
1998         long opc;
1999         int  rc;
2000         static const struct req_format *reint_fmts[REINT_MAX] = {
2001                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2002                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2003                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2004                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2005                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2006                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2007                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2008                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2009                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2010         };
2011
2012         ENTRY;
2013
2014         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2015         if (opc >= 0) {
2016                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2017                 /*
2018                  * No lock possible here from client to pass it to reint code
2019                  * path.
2020                  */
2021                 rc = mdt_reint_internal(info, NULL, opc);
2022                 mdt_thread_info_fini(info);
2023         } else {
2024                 rc = opc;
2025         }
2026
2027         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2028         RETURN(rc);
2029 }
2030
2031 /* this should sync the whole device */
2032 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2033 {
2034         struct dt_device *dt = mdt->mdt_bottom;
2035         int rc;
2036         ENTRY;
2037
2038         rc = dt->dd_ops->dt_sync(env, dt);
2039         RETURN(rc);
2040 }
2041
2042 /* this should sync this object */
2043 static int mdt_object_sync(struct mdt_thread_info *info)
2044 {
2045         struct md_object *next;
2046         int rc;
2047         ENTRY;
2048
2049         if (!mdt_object_exists(info->mti_object)) {
2050                 CWARN("%s: non existing object "DFID": rc = %d\n",
2051                       mdt_obd_name(info->mti_mdt),
2052                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
2053                 RETURN(-ESTALE);
2054         }
2055         next = mdt_object_child(info->mti_object);
2056         rc = mo_object_sync(info->mti_env, next);
2057
2058         RETURN(rc);
2059 }
2060
2061 static int mdt_sync(struct tgt_session_info *tsi)
2062 {
2063         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2064         struct req_capsule      *pill = tsi->tsi_pill;
2065         struct mdt_body         *body;
2066         int                      rc;
2067
2068         ENTRY;
2069
2070         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2071                 RETURN(err_serious(-ENOMEM));
2072
2073         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2074                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2075         } else {
2076                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2077
2078                 /* sync an object */
2079                 rc = mdt_object_sync(info);
2080                 if (rc == 0) {
2081                         const struct lu_fid *fid;
2082                         struct lu_attr *la = &info->mti_attr.ma_attr;
2083
2084                         info->mti_attr.ma_need = MA_INODE;
2085                         info->mti_attr.ma_valid = 0;
2086                         rc = mdt_attr_get_complex(info, info->mti_object,
2087                                                   &info->mti_attr);
2088                         if (rc == 0) {
2089                                 body = req_capsule_server_get(pill,
2090                                                               &RMF_MDT_BODY);
2091                                 fid = mdt_object_fid(info->mti_object);
2092                                 mdt_pack_attr2body(info, body, la, fid);
2093                         }
2094                 }
2095                 mdt_thread_info_fini(info);
2096         }
2097         if (rc == 0)
2098                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2099
2100         RETURN(rc);
2101 }
2102
2103 /*
2104  * Handle quota control requests to consult current usage/limit, but also
2105  * to configure quota enforcement
2106  */
2107 static int mdt_quotactl(struct tgt_session_info *tsi)
2108 {
2109         struct obd_export       *exp  = tsi->tsi_exp;
2110         struct req_capsule      *pill = tsi->tsi_pill;
2111         struct obd_quotactl     *oqctl, *repoqc;
2112         int                      id, rc;
2113         struct mdt_device       *mdt = mdt_exp2dev(exp);
2114         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2115         struct lu_nodemap       *nodemap;
2116         ENTRY;
2117
2118         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2119         if (oqctl == NULL)
2120                 RETURN(err_serious(-EPROTO));
2121
2122         rc = req_capsule_server_pack(pill);
2123         if (rc)
2124                 RETURN(err_serious(rc));
2125
2126         nodemap = nodemap_get_from_exp(exp);
2127         if (IS_ERR(nodemap))
2128                 RETURN(PTR_ERR(nodemap));
2129
2130         switch (oqctl->qc_cmd) {
2131                 /* master quotactl */
2132         case Q_SETINFO:
2133         case Q_SETQUOTA:
2134                 if (!nodemap_can_setquota(nodemap))
2135                         GOTO(out_nodemap, rc = -EPERM);
2136         case Q_GETINFO:
2137         case Q_GETQUOTA:
2138                 if (qmt == NULL)
2139                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2140                 /* slave quotactl */
2141         case Q_GETOINFO:
2142         case Q_GETOQUOTA:
2143                 break;
2144         default:
2145                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2146                 GOTO(out_nodemap, rc = -EFAULT);
2147         }
2148
2149         id = oqctl->qc_id;
2150         switch (oqctl->qc_type) {
2151         case USRQUOTA:
2152                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2153                                     NODEMAP_CLIENT_TO_FS, id);
2154                 break;
2155         case GRPQUOTA:
2156                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2157                                     NODEMAP_CLIENT_TO_FS, id);
2158                 break;
2159         case PRJQUOTA:
2160                 /* todo: check/map project id */
2161                 id = oqctl->qc_id;
2162                 break;
2163         default:
2164                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2165         }
2166         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2167         if (repoqc == NULL)
2168                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2169
2170         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2171                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2172
2173         if (oqctl->qc_id != id)
2174                 swap(oqctl->qc_id, id);
2175
2176         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2177                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2178                         RETURN(-EINPROGRESS);
2179         }
2180
2181         switch (oqctl->qc_cmd) {
2182
2183         case Q_GETINFO:
2184         case Q_SETINFO:
2185         case Q_SETQUOTA:
2186         case Q_GETQUOTA:
2187                 /* forward quotactl request to QMT */
2188                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2189                 break;
2190
2191         case Q_GETOINFO:
2192         case Q_GETOQUOTA:
2193                 /* slave quotactl */
2194                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2195                                    oqctl);
2196                 break;
2197
2198         default:
2199                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2200                 GOTO(out_nodemap, rc = -EFAULT);
2201         }
2202
2203         if (oqctl->qc_id != id)
2204                 swap(oqctl->qc_id, id);
2205
2206         *repoqc = *oqctl;
2207
2208         EXIT;
2209
2210 out_nodemap:
2211         nodemap_putref(nodemap);
2212
2213         return rc;
2214 }
2215
2216 /** clone llog ctxt from child (mdd)
2217  * This allows remote llog (replicator) access.
2218  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2219  * context was originally set up, or we can handle them directly.
2220  * I choose the latter, but that means I need any llog
2221  * contexts set up by child to be accessable by the mdt.  So we clone the
2222  * context into our context list here.
2223  */
2224 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2225                                int idx)
2226 {
2227         struct md_device  *next = mdt->mdt_child;
2228         struct llog_ctxt *ctxt;
2229         int rc;
2230
2231         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2232                 return 0;
2233
2234         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2235         if (rc || ctxt == NULL) {
2236                 return 0;
2237         }
2238
2239         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2240         if (rc)
2241                 CERROR("Can't set mdt ctxt %d\n", rc);
2242
2243         return rc;
2244 }
2245
2246 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2247                                  struct mdt_device *mdt, int idx)
2248 {
2249         struct llog_ctxt *ctxt;
2250
2251         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2252         if (ctxt == NULL)
2253                 return 0;
2254         /* Put once for the get we just did, and once for the clone */
2255         llog_ctxt_put(ctxt);
2256         llog_ctxt_put(ctxt);
2257         return 0;
2258 }
2259
2260 /*
2261  * sec context handlers
2262  */
2263 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2264 {
2265         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2266
2267         return 0;
2268 }
2269
2270 /*
2271  * quota request handlers
2272  */
2273 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2274 {
2275         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2276         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2277         int                      rc;
2278         ENTRY;
2279
2280         if (qmt == NULL)
2281                 RETURN(err_serious(-EOPNOTSUPP));
2282
2283         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2284         RETURN(rc);
2285 }
2286
2287 struct mdt_object *mdt_object_new(const struct lu_env *env,
2288                                   struct mdt_device *d,
2289                                   const struct lu_fid *f)
2290 {
2291         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2292         struct lu_object *o;
2293         struct mdt_object *m;
2294         ENTRY;
2295
2296         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2297         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2298         if (unlikely(IS_ERR(o)))
2299                 m = (struct mdt_object *)o;
2300         else
2301                 m = mdt_obj(o);
2302         RETURN(m);
2303 }
2304
2305 struct mdt_object *mdt_object_find(const struct lu_env *env,
2306                                    struct mdt_device *d,
2307                                    const struct lu_fid *f)
2308 {
2309         struct lu_object *o;
2310         struct mdt_object *m;
2311         ENTRY;
2312
2313         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2314         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2315         if (unlikely(IS_ERR(o)))
2316                 m = (struct mdt_object *)o;
2317         else
2318                 m = mdt_obj(o);
2319
2320         RETURN(m);
2321 }
2322
2323 /**
2324  * Asyncronous commit for mdt device.
2325  *
2326  * Pass asynchonous commit call down the MDS stack.
2327  *
2328  * \param env environment
2329  * \param mdt the mdt device
2330  */
2331 static void mdt_device_commit_async(const struct lu_env *env,
2332                                     struct mdt_device *mdt)
2333 {
2334         struct dt_device *dt = mdt->mdt_bottom;
2335         int rc;
2336         ENTRY;
2337
2338         rc = dt->dd_ops->dt_commit_async(env, dt);
2339         if (unlikely(rc != 0))
2340                 CWARN("%s: async commit start failed: rc = %d\n",
2341                       mdt_obd_name(mdt), rc);
2342         atomic_inc(&mdt->mdt_async_commit_count);
2343         EXIT;
2344 }
2345
2346 /**
2347  * Mark the lock as "synchonous".
2348  *
2349  * Mark the lock to deffer transaction commit to the unlock time.
2350  *
2351  * \param lock the lock to mark as "synchonous"
2352  *
2353  * \see mdt_is_lock_sync
2354  * \see mdt_save_lock
2355  */
2356 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2357 {
2358         lock->l_ast_data = (void*)1;
2359 }
2360
2361 /**
2362  * Check whehter the lock "synchonous" or not.
2363  *
2364  * \param lock the lock to check
2365  * \retval 1 the lock is "synchonous"
2366  * \retval 0 the lock isn't "synchronous"
2367  *
2368  * \see mdt_set_lock_sync
2369  * \see mdt_save_lock
2370  */
2371 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2372 {
2373         return lock->l_ast_data != NULL;
2374 }
2375
2376 /**
2377  * Blocking AST for mdt locks.
2378  *
2379  * Starts transaction commit if in case of COS lock conflict or
2380  * deffers such a commit to the mdt_save_lock.
2381  *
2382  * \param lock the lock which blocks a request or cancelling lock
2383  * \param desc unused
2384  * \param data unused
2385  * \param flag indicates whether this cancelling or blocking callback
2386  * \retval 0
2387  * \see ldlm_blocking_ast_nocheck
2388  */
2389 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2390                      void *data, int flag)
2391 {
2392         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2393         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2394         int rc;
2395         ENTRY;
2396
2397         if (flag == LDLM_CB_CANCELING)
2398                 RETURN(0);
2399
2400         lock_res_and_lock(lock);
2401         if (lock->l_blocking_ast != mdt_blocking_ast) {
2402                 unlock_res_and_lock(lock);
2403                 RETURN(0);
2404         }
2405         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2406             lock->l_blocking_lock != NULL) {
2407                 if (mdt_cos_is_enabled(mdt) &&
2408                     lock->l_client_cookie !=
2409                     lock->l_blocking_lock->l_client_cookie)
2410                         mdt_set_lock_sync(lock);
2411                 else if (mdt_slc_is_enabled(mdt) &&
2412                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2413                         mdt_set_lock_sync(lock);
2414         }
2415         rc = ldlm_blocking_ast_nocheck(lock);
2416
2417         /* There is no lock conflict if l_blocking_lock == NULL,
2418          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2419          * when the last reference to a local lock was released */
2420         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2421                 struct lu_env env;
2422
2423                 rc = lu_env_init(&env, LCT_LOCAL);
2424                 if (unlikely(rc != 0))
2425                         CWARN("%s: lu_env initialization failed, cannot "
2426                               "start asynchronous commit: rc = %d\n",
2427                               obd->obd_name, rc);
2428                 else
2429                         mdt_device_commit_async(&env, mdt);
2430                 lu_env_fini(&env);
2431         }
2432         RETURN(rc);
2433 }
2434
2435 /*
2436  * Blocking AST for cross-MDT lock
2437  *
2438  * Discard lock from uncommitted_slc_locks and cancel it.
2439  *
2440  * \param lock  the lock which blocks a request or cancelling lock
2441  * \param desc  unused
2442  * \param data  unused
2443  * \param flag  indicates whether this cancelling or blocking callback
2444  * \retval      0 on success
2445  * \retval      negative number on error
2446  */
2447 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2448                             void *data, int flag)
2449 {
2450         int rc = 0;
2451         ENTRY;
2452
2453         switch (flag) {
2454         case LDLM_CB_BLOCKING: {
2455                 struct lustre_handle lockh;
2456
2457                 ldlm_lock2handle(lock, &lockh);
2458                 rc = ldlm_cli_cancel(&lockh,
2459                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2460                 if (rc < 0) {
2461                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2462                         RETURN(rc);
2463                 }
2464                 break;
2465         }
2466         case LDLM_CB_CANCELING: {
2467                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2468                 struct mdt_device *mdt =
2469                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2470
2471                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2472
2473                 /* discard slc lock here so that it can be cleaned anytime,
2474                  * especially for cleanup_resource() */
2475                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2476
2477                 /* once we cache lock, l_ast_data is set to mdt_object */
2478                 if (lock->l_ast_data != NULL) {
2479                         struct mdt_object *mo = lock->l_ast_data;
2480                         struct lu_env env;
2481
2482                         rc = lu_env_init(&env, LCT_MD_THREAD);
2483                         if (unlikely(rc != 0)) {
2484                                 CWARN("%s: lu_env initialization failed, object"
2485                                       "%p "DFID" is leaked!\n",
2486                                       obd->obd_name, mo,
2487                                       PFID(mdt_object_fid(mo)));
2488                                 RETURN(rc);
2489                         }
2490
2491                         if (lock->l_policy_data.l_inodebits.bits &
2492                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2493                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2494                                 mo->mot_cache_attr = 0;
2495                         }
2496                         mdt_object_put(&env, mo);
2497                         lu_env_fini(&env);
2498                 }
2499                 break;
2500         }
2501         default:
2502                 LBUG();
2503         }
2504
2505         RETURN(rc);
2506 }
2507
2508 int mdt_check_resent_lock(struct mdt_thread_info *info,
2509                           struct mdt_object *mo,
2510                           struct mdt_lock_handle *lhc)
2511 {
2512         /* the lock might already be gotten in ldlm_handle_enqueue() */
2513         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2514                 struct ptlrpc_request *req = mdt_info_req(info);
2515                 struct ldlm_lock      *lock;
2516
2517                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2518                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2519                 if (lock == NULL) {
2520                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2521                          * a resend case, however, it could be already destroyed
2522                          * due to client eviction or a raced cancel RPC. */
2523                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2524                                           lhc->mlh_reg_lh.cookie);
2525                         RETURN(-ESTALE);
2526                 }
2527
2528                 if (!fid_res_name_eq(mdt_object_fid(mo),
2529                                      &lock->l_resource->lr_name)) {
2530                         CWARN("%s: Although resent, but still not "
2531                               "get child lock:"DFID"\n",
2532                               info->mti_exp->exp_obd->obd_name,
2533                               PFID(mdt_object_fid(mo)));
2534                         LDLM_LOCK_PUT(lock);
2535                         RETURN(-EPROTO);
2536                 }
2537                 LDLM_LOCK_PUT(lock);
2538                 return 0;
2539         }
2540         return 1;
2541 }
2542
2543 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2544                            const struct lu_fid *fid, struct lustre_handle *lh,
2545                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2546                            bool cache)
2547 {
2548         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2549         union ldlm_policy_data *policy = &mti->mti_policy;
2550         struct ldlm_res_id *res_id = &mti->mti_res_id;
2551         int rc = 0;
2552         ENTRY;
2553
2554         LASSERT(mdt_object_remote(o));
2555
2556         fid_build_reg_res_name(fid, res_id);
2557
2558         memset(einfo, 0, sizeof(*einfo));
2559         einfo->ei_type = LDLM_IBITS;
2560         einfo->ei_mode = mode;
2561         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2562         einfo->ei_cb_cp = ldlm_completion_ast;
2563         einfo->ei_enq_slave = 0;
2564         einfo->ei_res_id = res_id;
2565         if (nonblock)
2566                 einfo->ei_nonblock = 1;
2567         if (cache) {
2568                 /*
2569                  * if we cache lock, couple lock with mdt_object, so that object
2570                  * can be easily found in lock ASTs.
2571                  */
2572                 mdt_object_get(mti->mti_env, o);
2573                 einfo->ei_cbdata = o;
2574         }
2575
2576         memset(policy, 0, sizeof(*policy));
2577         policy->l_inodebits.bits = ibits;
2578
2579         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2580                             policy);
2581         if (rc < 0 && cache) {
2582                 mdt_object_put(mti->mti_env, o);
2583                 einfo->ei_cbdata = NULL;
2584         }
2585         RETURN(rc);
2586 }
2587
2588 static int mdt_object_local_lock(struct mdt_thread_info *info,
2589                                  struct mdt_object *o,
2590                                  struct mdt_lock_handle *lh, __u64 ibits,
2591                                  bool nonblock, bool cos_incompat)
2592 {
2593         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2594         union ldlm_policy_data *policy = &info->mti_policy;
2595         struct ldlm_res_id *res_id = &info->mti_res_id;
2596         __u64 dlmflags = 0;
2597         int rc;
2598         ENTRY;
2599
2600         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2601         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2602         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2603         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2604
2605         if (cos_incompat) {
2606                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2607                         lh->mlh_reg_mode == LCK_EX);
2608                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2609         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2610                 dlmflags |= LDLM_FL_COS_ENABLED;
2611         }
2612
2613         /* Only enqueue LOOKUP lock for remote object */
2614         if (mdt_object_remote(o))
2615                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2616
2617         if (lh->mlh_type == MDT_PDO_LOCK) {
2618                 /* check for exists after object is locked */
2619                 if (mdt_object_exists(o) == 0) {
2620                         /* Non-existent object shouldn't have PDO lock */
2621                         RETURN(-ESTALE);
2622                 } else {
2623                         /* Non-dir object shouldn't have PDO lock */
2624                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2625                                 RETURN(-ENOTDIR);
2626                 }
2627         }
2628
2629         memset(policy, 0, sizeof(*policy));
2630         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2631
2632         dlmflags |= LDLM_FL_ATOMIC_CB;
2633         if (nonblock)
2634                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2635
2636         /*
2637          * Take PDO lock on whole directory and build correct @res_id for lock
2638          * on part of directory.
2639          */
2640         if (lh->mlh_pdo_hash != 0) {
2641                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2642                 mdt_lock_pdo_mode(info, o, lh);
2643                 if (lh->mlh_pdo_mode != LCK_NL) {
2644                         /*
2645                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2646                          * is never going to be sent to client and we do not
2647                          * want it slowed down due to possible cancels.
2648                          */
2649                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2650                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2651                                           policy, res_id, dlmflags,
2652                                           info->mti_exp == NULL ? NULL :
2653                                           &info->mti_exp->exp_handle.h_cookie);
2654                         if (unlikely(rc != 0))
2655                                 GOTO(out_unlock, rc);
2656                 }
2657
2658                 /*
2659                  * Finish res_id initializing by name hash marking part of
2660                  * directory which is taking modification.
2661                  */
2662                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2663         }
2664
2665         policy->l_inodebits.bits = ibits;
2666
2667         /*
2668          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2669          * going to be sent to client. If it is - mdt_intent_policy() path will
2670          * fix it up and turn FL_LOCAL flag off.
2671          */
2672         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2673                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2674                           info->mti_exp == NULL ? NULL :
2675                           &info->mti_exp->exp_handle.h_cookie);
2676 out_unlock:
2677         if (rc != 0)
2678                 mdt_object_unlock(info, o, lh, 1);
2679         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2680                    lh->mlh_pdo_hash != 0 &&
2681                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2682                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2683
2684         RETURN(rc);
2685 }
2686
2687 static int
2688 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2689                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2690                          bool cos_incompat)
2691 {
2692         struct mdt_lock_handle *local_lh = NULL;
2693         int rc;
2694         ENTRY;
2695
2696         if (!mdt_object_remote(o)) {
2697                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2698                                            cos_incompat);
2699                 RETURN(rc);
2700         }
2701
2702         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2703         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2704                    MDS_INODELOCK_XATTR);
2705
2706         /* Only enqueue LOOKUP lock for remote object */
2707         if (ibits & MDS_INODELOCK_LOOKUP) {
2708                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2709                                            nonblock, cos_incompat);
2710                 if (rc != ELDLM_OK)
2711                         RETURN(rc);
2712
2713                 local_lh = lh;
2714         }
2715
2716         if (ibits & MDS_INODELOCK_UPDATE) {
2717                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2718                  * enqueue RPC can only request 1 lock, to avoid extra
2719                  * RPC, so it will instead enqueue EX lock for remote
2720                  * object anyway XXX*/
2721                 if (lh->mlh_type == MDT_PDO_LOCK &&
2722                     lh->mlh_pdo_hash != 0) {
2723                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2724                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2725                                PFID(mdt_object_fid(o)));
2726                         lh->mlh_pdo_hash = 0;
2727                         lh->mlh_rreg_mode = LCK_EX;
2728                         lh->mlh_type = MDT_REG_LOCK;
2729                 }
2730                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2731                                             &lh->mlh_rreg_lh,
2732                                             lh->mlh_rreg_mode,
2733                                             MDS_INODELOCK_UPDATE, nonblock,
2734                                             false);
2735                 if (rc != ELDLM_OK) {
2736                         if (local_lh != NULL)
2737                                 mdt_object_unlock(info, o, local_lh, rc);
2738                         RETURN(rc);
2739                 }
2740         }
2741
2742         RETURN(0);
2743 }
2744
2745 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2746                     struct mdt_lock_handle *lh, __u64 ibits)
2747 {
2748         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2749 }
2750
2751 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2752                           struct mdt_lock_handle *lh, __u64 ibits,
2753                           bool cos_incompat)
2754 {
2755         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2756         return mdt_object_lock_internal(info, o, lh, ibits, false,
2757                                         cos_incompat);
2758 }
2759
2760 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2761                         struct mdt_lock_handle *lh, __u64 ibits)
2762 {
2763         struct mdt_lock_handle tmp = *lh;
2764         int rc;
2765
2766         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2767         if (rc == 0)
2768                 *lh = tmp;
2769
2770         return rc == 0;
2771 }
2772
2773 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2774                               struct mdt_object *o, struct mdt_lock_handle *lh,
2775                               __u64 ibits, bool cos_incompat)
2776 {
2777         struct mdt_lock_handle tmp = *lh;
2778         int rc;
2779
2780         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2781         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2782         if (rc == 0)
2783                 *lh = tmp;
2784
2785         return rc == 0;
2786 }
2787
2788 /**
2789  * Save a lock within request object.
2790  *
2791  * Keep the lock referenced until whether client ACK or transaction
2792  * commit happens or release the lock immediately depending on input
2793  * parameters. If COS is ON, a write lock is converted to COS lock
2794  * before saving.
2795  *
2796  * \param info thead info object
2797  * \param h lock handle
2798  * \param mode lock mode
2799  * \param decref force immediate lock releasing
2800  */
2801 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2802                           enum ldlm_mode mode, int decref)
2803 {
2804         ENTRY;
2805
2806         if (lustre_handle_is_used(h)) {
2807                 if (decref || !info->mti_has_trans ||
2808                     !(mode & (LCK_PW | LCK_EX))) {
2809                         mdt_fid_unlock(h, mode);
2810                 } else {
2811                         struct mdt_device *mdt = info->mti_mdt;
2812                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2813                         struct ptlrpc_request *req = mdt_info_req(info);
2814                         int cos;
2815
2816                         cos = (mdt_cos_is_enabled(mdt) ||
2817                                mdt_slc_is_enabled(mdt));
2818
2819                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2820                                  h->cookie);
2821
2822                         /* there is no request if mdt_object_unlock() is called
2823                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2824                         if (likely(req != NULL)) {
2825                                 CDEBUG(D_HA, "request = %p reply state = %p"
2826                                        " transno = %lld\n", req,
2827                                        req->rq_reply_state, req->rq_transno);
2828                                 if (cos) {
2829                                         ldlm_lock_downgrade(lock, LCK_COS);
2830                                         mode = LCK_COS;
2831                                 }
2832                                 ptlrpc_save_lock(req, h, mode, cos);
2833                         } else {
2834                                 mdt_fid_unlock(h, mode);
2835                         }
2836                         if (mdt_is_lock_sync(lock)) {
2837                                 CDEBUG(D_HA, "found sync-lock,"
2838                                        " async commit started\n");
2839                                 mdt_device_commit_async(info->mti_env,
2840                                                         mdt);
2841                         }
2842                         LDLM_LOCK_PUT(lock);
2843                 }
2844                 h->cookie = 0ull;
2845         }
2846
2847         EXIT;
2848 }
2849
2850 /**
2851  * Save cross-MDT lock in uncommitted_slc_locks
2852  *
2853  * Keep the lock referenced until transaction commit happens or release the lock
2854  * immediately depending on input parameters.
2855  *
2856  * \param info thead info object
2857  * \param h lock handle
2858  * \param mode lock mode
2859  * \param decref force immediate lock releasing
2860  */
2861 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2862                                  struct mdt_object *o, struct lustre_handle *h,
2863                                  enum ldlm_mode mode, int decref)
2864 {
2865         ENTRY;
2866
2867         if (lustre_handle_is_used(h)) {
2868                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2869
2870                 if (o != NULL &&
2871                     (lock->l_policy_data.l_inodebits.bits &
2872                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2873                         mo_invalidate(info->mti_env, mdt_object_child(o));
2874
2875                 if (decref || !info->mti_has_trans ||
2876                     !(mode & (LCK_PW | LCK_EX))) {
2877                         ldlm_lock_decref_and_cancel(h, mode);
2878                         LDLM_LOCK_PUT(lock);
2879                 } else {
2880                         struct ptlrpc_request *req = mdt_info_req(info);
2881
2882                         LASSERT(req != NULL);
2883                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
2884                                           req->rq_transno);
2885                         ldlm_lock_decref(h, mode);
2886                 }
2887                 h->cookie = 0ull;
2888         }
2889
2890         EXIT;
2891 }
2892
2893 /**
2894  * Unlock mdt object.
2895  *
2896  * Immeditely release the regular lock and the PDO lock or save the
2897  * lock in request and keep them referenced until client ACK or
2898  * transaction commit.
2899  *
2900  * \param info thread info object
2901  * \param o mdt object
2902  * \param lh mdt lock handle referencing regular and PDO locks
2903  * \param decref force immediate lock releasing
2904  *
2905  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
2906  */
2907 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2908                        struct mdt_lock_handle *lh, int decref)
2909 {
2910         ENTRY;
2911
2912         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2913         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2914         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2915                              decref);
2916
2917         EXIT;
2918 }
2919
2920 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2921                                         const struct lu_fid *f,
2922                                         struct mdt_lock_handle *lh,
2923                                         __u64 ibits)
2924 {
2925         struct mdt_object *o;
2926
2927         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2928         if (!IS_ERR(o)) {
2929                 int rc;
2930
2931                 rc = mdt_object_lock(info, o, lh, ibits);
2932                 if (rc != 0) {
2933                         mdt_object_put(info->mti_env, o);
2934                         o = ERR_PTR(rc);
2935                 }
2936         }
2937         return o;
2938 }
2939
2940 void mdt_object_unlock_put(struct mdt_thread_info * info,
2941                            struct mdt_object * o,
2942                            struct mdt_lock_handle *lh,
2943                            int decref)
2944 {
2945         mdt_object_unlock(info, o, lh, decref);
2946         mdt_object_put(info->mti_env, o);
2947 }
2948
2949 /*
2950  * Generic code handling requests that have struct mdt_body passed in:
2951  *
2952  *  - extract mdt_body from request and save it in @info, if present;
2953  *
2954  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2955  *  @info;
2956  *
2957  *  - if HABEO_CORPUS flag is set for this request type check whether object
2958  *  actually exists on storage (lu_object_exists()).
2959  *
2960  */
2961 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2962 {
2963         const struct mdt_body    *body;
2964         struct mdt_object        *obj;
2965         const struct lu_env      *env;
2966         struct req_capsule       *pill;
2967         int                       rc;
2968         ENTRY;
2969
2970         env = info->mti_env;
2971         pill = info->mti_pill;
2972
2973         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2974         if (body == NULL)
2975                 RETURN(-EFAULT);
2976
2977         if (!(body->mbo_valid & OBD_MD_FLID))
2978                 RETURN(0);
2979
2980         if (!fid_is_sane(&body->mbo_fid1)) {
2981                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2982                 RETURN(-EINVAL);
2983         }
2984
2985         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2986         if (!IS_ERR(obj)) {
2987                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2988                         mdt_object_put(env, obj);
2989                         rc = -ENOENT;
2990                 } else {
2991                         info->mti_object = obj;
2992                         rc = 0;
2993                 }
2994         } else
2995                 rc = PTR_ERR(obj);
2996
2997         RETURN(rc);
2998 }
2999
3000 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
3001 {
3002         struct req_capsule *pill = info->mti_pill;
3003         int rc;
3004         ENTRY;
3005
3006         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3007                 rc = mdt_body_unpack(info, flags);
3008         else
3009                 rc = 0;
3010
3011         if (rc == 0 && (flags & HABEO_REFERO)) {
3012                 /* Pack reply. */
3013                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3014                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3015                                              DEF_REP_MD_SIZE);
3016                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3017                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3018                                              RCL_SERVER, 0);
3019
3020                 rc = req_capsule_server_pack(pill);
3021         }
3022         RETURN(rc);
3023 }
3024
3025 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3026 {
3027         lh->mlh_type = MDT_NUL_LOCK;
3028         lh->mlh_reg_lh.cookie = 0ull;
3029         lh->mlh_reg_mode = LCK_MINMODE;
3030         lh->mlh_pdo_lh.cookie = 0ull;
3031         lh->mlh_pdo_mode = LCK_MINMODE;
3032         lh->mlh_rreg_lh.cookie = 0ull;
3033         lh->mlh_rreg_mode = LCK_MINMODE;
3034 }
3035
3036 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3037 {
3038         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3039         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3040 }
3041
3042 /*
3043  * Initialize fields of struct mdt_thread_info. Other fields are left in
3044  * uninitialized state, because it's too expensive to zero out whole
3045  * mdt_thread_info (> 1K) on each request arrival.
3046  */
3047 void mdt_thread_info_init(struct ptlrpc_request *req,
3048                           struct mdt_thread_info *info)
3049 {
3050         int i;
3051
3052         info->mti_pill = &req->rq_pill;
3053
3054         /* lock handle */
3055         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3056                 mdt_lock_handle_init(&info->mti_lh[i]);
3057
3058         /* mdt device: it can be NULL while CONNECT */
3059         if (req->rq_export) {
3060                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3061                 info->mti_exp = req->rq_export;
3062         } else
3063                 info->mti_mdt = NULL;
3064         info->mti_env = req->rq_svc_thread->t_env;
3065         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3066
3067         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3068         info->mti_big_buf = LU_BUF_NULL;
3069         info->mti_body = NULL;
3070         info->mti_object = NULL;
3071         info->mti_dlm_req = NULL;
3072         info->mti_has_trans = 0;
3073         info->mti_cross_ref = 0;
3074         info->mti_opdata = 0;
3075         info->mti_big_lmm_used = 0;
3076
3077         info->mti_spec.no_create = 0;
3078         info->mti_spec.sp_rm_entry = 0;
3079         info->mti_spec.sp_permitted = 0;
3080         info->mti_spec.sp_migrate_close = 0;
3081
3082         info->mti_spec.u.sp_ea.eadata = NULL;
3083         info->mti_spec.u.sp_ea.eadatalen = 0;
3084 }
3085
3086 void mdt_thread_info_fini(struct mdt_thread_info *info)
3087 {
3088         int i;
3089
3090         if (info->mti_object != NULL) {
3091                 mdt_object_put(info->mti_env, info->mti_object);
3092                 info->mti_object = NULL;
3093         }
3094
3095         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3096                 mdt_lock_handle_fini(&info->mti_lh[i]);
3097         info->mti_env = NULL;
3098         info->mti_pill = NULL;
3099         info->mti_exp = NULL;
3100
3101         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3102                 lu_buf_free(&info->mti_big_buf);
3103 }
3104
3105 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3106 {
3107         struct mdt_thread_info  *mti;
3108
3109         mti = mdt_th_info(tsi->tsi_env);
3110         LASSERT(mti != NULL);
3111
3112         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3113         if (tsi->tsi_corpus != NULL) {
3114                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3115                 lu_object_get(tsi->tsi_corpus);
3116         }
3117         mti->mti_body = tsi->tsi_mdt_body;
3118         mti->mti_dlm_req = tsi->tsi_dlm_req;
3119
3120         return mti;
3121 }
3122
3123 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3124 {
3125         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3126             cfs_fail_val ==
3127             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3128                 set_current_state(TASK_UNINTERRUPTIBLE);
3129                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3130         }
3131
3132         return tgt_connect(tsi);
3133 }
3134
3135 enum mdt_it_code {
3136         MDT_IT_OPEN,
3137         MDT_IT_OCREAT,
3138         MDT_IT_CREATE,
3139         MDT_IT_GETATTR,
3140         MDT_IT_READDIR,
3141         MDT_IT_LOOKUP,
3142         MDT_IT_UNLINK,
3143         MDT_IT_TRUNC,
3144         MDT_IT_GETXATTR,
3145         MDT_IT_LAYOUT,
3146         MDT_IT_QUOTA,
3147         MDT_IT_NR
3148 };
3149
3150 static int mdt_intent_getattr(enum mdt_it_code opcode,
3151                               struct mdt_thread_info *info,
3152                               struct ldlm_lock **,
3153                               __u64);
3154
3155 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3156                                 struct mdt_thread_info *info,
3157                                 struct ldlm_lock **lockp,
3158                                 __u64 flags);
3159
3160 static int mdt_intent_layout(enum mdt_it_code opcode,
3161                              struct mdt_thread_info *info,
3162                              struct ldlm_lock **,
3163                              __u64);
3164 static int mdt_intent_reint(enum mdt_it_code opcode,
3165                             struct mdt_thread_info *info,
3166                             struct ldlm_lock **,
3167                             __u64);
3168
3169 static struct mdt_it_flavor {
3170         const struct req_format *it_fmt;
3171         __u32                    it_flags;
3172         int                    (*it_act)(enum mdt_it_code ,
3173                                          struct mdt_thread_info *,
3174                                          struct ldlm_lock **,
3175                                          __u64);
3176         long                     it_reint;
3177 } mdt_it_flavor[] = {
3178         [MDT_IT_OPEN]     = {
3179                 .it_fmt   = &RQF_LDLM_INTENT,
3180                 /*.it_flags = HABEO_REFERO,*/
3181                 .it_flags = 0,
3182                 .it_act   = mdt_intent_reint,
3183                 .it_reint = REINT_OPEN
3184         },
3185         [MDT_IT_OCREAT]   = {
3186                 .it_fmt   = &RQF_LDLM_INTENT,
3187                 /*
3188                  * OCREAT is not a MUTABOR request as if the file
3189                  * already exists.
3190                  * We do the extra check of OBD_CONNECT_RDONLY in
3191                  * mdt_reint_open() when we really need to create
3192                  * the object.
3193                  */
3194                 .it_flags = 0,
3195                 .it_act   = mdt_intent_reint,
3196                 .it_reint = REINT_OPEN
3197         },
3198         [MDT_IT_CREATE]   = {
3199                 .it_fmt   = &RQF_LDLM_INTENT,
3200                 .it_flags = MUTABOR,
3201                 .it_act   = mdt_intent_reint,
3202                 .it_reint = REINT_CREATE
3203         },
3204         [MDT_IT_GETATTR]  = {
3205                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3206                 .it_flags = HABEO_REFERO,
3207                 .it_act   = mdt_intent_getattr
3208         },
3209         [MDT_IT_READDIR]  = {
3210                 .it_fmt   = NULL,
3211                 .it_flags = 0,
3212                 .it_act   = NULL
3213         },
3214         [MDT_IT_LOOKUP]   = {
3215                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3216                 .it_flags = HABEO_REFERO,
3217                 .it_act   = mdt_intent_getattr
3218         },
3219         [MDT_IT_UNLINK]   = {
3220                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3221                 .it_flags = MUTABOR,
3222                 .it_act   = NULL,
3223                 .it_reint = REINT_UNLINK
3224         },
3225         [MDT_IT_TRUNC]    = {
3226                 .it_fmt   = NULL,
3227                 .it_flags = MUTABOR,
3228                 .it_act   = NULL
3229         },
3230         [MDT_IT_GETXATTR] = {
3231                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3232                 .it_flags = HABEO_CORPUS,
3233                 .it_act   = mdt_intent_getxattr
3234         },
3235         [MDT_IT_LAYOUT] = {
3236                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3237                 .it_flags = 0,
3238                 .it_act   = mdt_intent_layout
3239         }
3240 };
3241
3242 static int
3243 mdt_intent_lock_replace(struct mdt_thread_info *info,
3244                         struct ldlm_lock **lockp,
3245                         struct mdt_lock_handle *lh,
3246                         __u64 flags, int result)
3247 {
3248         struct ptlrpc_request  *req = mdt_info_req(info);
3249         struct ldlm_lock       *lock = *lockp;
3250         struct ldlm_lock       *new_lock;
3251
3252         /* If possible resent found a lock, @lh is set to its handle */
3253         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3254
3255         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3256                 lh->mlh_reg_lh.cookie = 0;
3257                 RETURN(0);
3258         }
3259
3260         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3261                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3262                  * a resend case, however, it could be already destroyed
3263                  * due to client eviction or a raced cancel RPC. */
3264                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3265                                   lh->mlh_reg_lh.cookie);
3266                 lh->mlh_reg_lh.cookie = 0;
3267                 RETURN(-ESTALE);
3268         }
3269
3270         LASSERTF(new_lock != NULL,
3271                  "lockh %#llx flags %#llx : rc = %d\n",
3272                  lh->mlh_reg_lh.cookie, flags, result);
3273
3274         /*
3275          * If we've already given this lock to a client once, then we should
3276          * have no readers or writers.  Otherwise, we should have one reader
3277          * _or_ writer ref (which will be zeroed below) before returning the
3278          * lock to a client.
3279          */
3280         if (new_lock->l_export == req->rq_export) {
3281                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3282         } else {
3283                 LASSERT(new_lock->l_export == NULL);
3284                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3285         }
3286
3287         *lockp = new_lock;
3288
3289         if (new_lock->l_export == req->rq_export) {
3290                 /*
3291                  * Already gave this to the client, which means that we
3292                  * reconstructed a reply.
3293                  */
3294                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3295                         MSG_RESENT);
3296
3297                 LDLM_LOCK_RELEASE(new_lock);
3298                 lh->mlh_reg_lh.cookie = 0;
3299                 RETURN(ELDLM_LOCK_REPLACED);
3300         }
3301
3302         /*
3303          * Fixup the lock to be given to the client.
3304          */
3305         lock_res_and_lock(new_lock);
3306         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3307          * possible blocking AST. */
3308         while (new_lock->l_readers > 0) {
3309                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3310                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3311                 new_lock->l_readers--;
3312         }
3313         while (new_lock->l_writers > 0) {
3314                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3315                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3316                 new_lock->l_writers--;
3317         }
3318
3319         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3320         new_lock->l_blocking_ast = lock->l_blocking_ast;
3321         new_lock->l_completion_ast = lock->l_completion_ast;
3322         new_lock->l_remote_handle = lock->l_remote_handle;
3323         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3324
3325         unlock_res_and_lock(new_lock);
3326
3327         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3328                      &new_lock->l_remote_handle,
3329                      &new_lock->l_exp_hash);
3330
3331         LDLM_LOCK_RELEASE(new_lock);
3332         lh->mlh_reg_lh.cookie = 0;
3333
3334         RETURN(ELDLM_LOCK_REPLACED);
3335 }
3336
3337 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3338                                     struct ldlm_lock *new_lock,
3339                                     struct mdt_lock_handle *lh,
3340                                     __u64 flags)
3341 {
3342         struct ptlrpc_request  *req = mdt_info_req(info);
3343         struct ldlm_request    *dlmreq;
3344
3345         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3346                 return;
3347
3348         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3349
3350         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3351          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3352          * initialized. */
3353         if (flags & LDLM_FL_RESENT) {
3354                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3355                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3356
3357                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3358                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3359                           lh->mlh_reg_lh.cookie);
3360                 return;
3361         }
3362
3363         /*
3364          * If the xid matches, then we know this is a resent request, and allow
3365          * it. (It's probably an OPEN, for which we don't send a lock.
3366          */
3367         if (req_can_reconstruct(req, NULL))
3368                 return;
3369
3370         /*
3371          * This remote handle isn't enqueued, so we never received or processed
3372          * this request.  Clear MSG_RESENT, because it can be handled like any
3373          * normal request now.
3374          */
3375         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3376
3377         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3378                   dlmreq->lock_handle[0].cookie);
3379 }
3380
3381 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3382                                 struct mdt_thread_info *info,
3383                                 struct ldlm_lock **lockp,
3384                                 __u64 flags)
3385 {
3386         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3387         struct ldlm_reply      *ldlm_rep = NULL;
3388         int rc;
3389         ENTRY;
3390
3391         /*
3392          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3393          * (for the resend case) or a new lock. Below we will use it to
3394          * replace the original lock.
3395          */
3396         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3397         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3398                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3399                 rc = mdt_object_lock(info, info->mti_object, lhc,
3400                                      MDS_INODELOCK_XATTR);
3401                 if (rc)
3402                         return rc;
3403         }
3404
3405         rc = mdt_getxattr(info);
3406
3407         if (mdt_info_req(info)->rq_repmsg != NULL)
3408                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3409
3410         if (ldlm_rep == NULL ||
3411             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3412                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3413                 RETURN(err_serious(-EFAULT));
3414         }
3415
3416         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3417
3418         /* This is left for interop instead of adding a new interop flag.
3419          * LU-7433 */
3420 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3421         if (ldlm_rep->lock_policy_res2) {
3422                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3423                 RETURN(ELDLM_LOCK_ABORTED);
3424         }
3425 #endif
3426
3427         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3428         RETURN(rc);
3429 }
3430
3431 static int mdt_intent_getattr(enum mdt_it_code opcode,
3432                               struct mdt_thread_info *info,
3433                               struct ldlm_lock **lockp,
3434                               __u64 flags)
3435 {
3436         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3437         __u64                   child_bits;
3438         struct ldlm_reply      *ldlm_rep;
3439         struct mdt_body        *reqbody;
3440         struct mdt_body        *repbody;
3441         int                     rc, rc2;
3442         ENTRY;
3443
3444         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3445         LASSERT(reqbody);
3446
3447         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3448         LASSERT(repbody);
3449
3450         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3451         repbody->mbo_eadatasize = 0;
3452         repbody->mbo_aclsize = 0;
3453
3454         switch (opcode) {
3455         case MDT_IT_LOOKUP:
3456                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3457                 break;
3458         case MDT_IT_GETATTR:
3459                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3460                              MDS_INODELOCK_PERM;
3461                 break;
3462         default:
3463                 CERROR("Unsupported intent (%d)\n", opcode);
3464                 GOTO(out_shrink, rc = -EINVAL);
3465         }
3466
3467         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3468         if (rc)
3469                 GOTO(out_shrink, rc);
3470
3471         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3472         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3473
3474         /* Get lock from request for possible resent case. */
3475         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3476
3477         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3478         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3479
3480         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3481                 ldlm_rep->lock_policy_res2 = 0;
3482         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3483             ldlm_rep->lock_policy_res2) {
3484                 lhc->mlh_reg_lh.cookie = 0ull;
3485                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3486         }
3487
3488         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3489         EXIT;
3490 out_ucred:
3491         mdt_exit_ucred(info);
3492 out_shrink:
3493         mdt_client_compatibility(info);
3494         rc2 = mdt_fix_reply(info);
3495         if (rc == 0)
3496                 rc = rc2;
3497         return rc;
3498 }
3499
3500 static int mdt_intent_layout(enum mdt_it_code opcode,
3501                              struct mdt_thread_info *info,
3502                              struct ldlm_lock **lockp,
3503                              __u64 flags)
3504 {
3505         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3506         struct layout_intent *layout;
3507         struct lu_fid *fid;
3508         struct mdt_object *obj = NULL;
3509         bool layout_change = false;
3510         int layout_size = 0;
3511         int rc = 0;
3512         ENTRY;
3513
3514         if (opcode != MDT_IT_LAYOUT) {
3515                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3516                         opcode);
3517                 RETURN(-EINVAL);
3518         }
3519
3520         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3521         if (layout == NULL)
3522                 RETURN(-EPROTO);
3523
3524         switch (layout->li_opc) {
3525         case LAYOUT_INTENT_TRUNC:
3526         case LAYOUT_INTENT_WRITE:
3527                 layout_change = true;
3528                 break;
3529         case LAYOUT_INTENT_ACCESS:
3530                 break;
3531         case LAYOUT_INTENT_READ:
3532         case LAYOUT_INTENT_GLIMPSE:
3533         case LAYOUT_INTENT_RELEASE:
3534         case LAYOUT_INTENT_RESTORE:
3535                 CERROR("%s: Unsupported layout intent opc %d\n",
3536                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3537                 rc = -ENOTSUPP;
3538                 break;
3539         default:
3540                 CERROR("%s: Unknown layout intent opc %d\n",
3541                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3542                 rc = -EINVAL;
3543                 break;
3544         }
3545         if (rc < 0)
3546                 RETURN(rc);
3547
3548         fid = &info->mti_tmp_fid2;
3549         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3550
3551         /* Get lock from request for possible resent case. */
3552         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3553
3554         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3555         if (IS_ERR(obj))
3556                 GOTO(out, rc = PTR_ERR(obj));
3557
3558         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3559                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3560                 if (layout_size < 0)
3561                         GOTO(out_obj, rc = layout_size);
3562
3563                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3564                         info->mti_mdt->mdt_max_mdsize = layout_size;
3565         }
3566
3567         /*
3568          * set reply buffer size, so that ldlm_handle_enqueue0()->
3569          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3570          */
3571         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3572         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3573                              layout_size);
3574         rc = req_capsule_server_pack(info->mti_pill);
3575         if (rc)
3576                 GOTO(out_obj, rc);
3577
3578
3579         if (layout_change) {
3580                 struct lu_buf *buf = &info->mti_buf;
3581
3582                 /**
3583                  * mdt_layout_change is a reint operation, when the request
3584                  * is resent, layout write shouldn't reprocess it again.
3585                  */
3586                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3587                 if (rc)
3588                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3589
3590                 /**
3591                  * There is another resent case: the client's job has been
3592                  * done by another client, referring lod_declare_layout_change
3593                  * -EALREADY case, and it became a operation w/o transaction,
3594                  * so we should not do the layout change, otherwise
3595                  * mdt_layout_change() will try to cancel the granted server
3596                  * CR lock whose remote counterpart is still in hold on the
3597                  * client, and a deadlock ensues.
3598                  */
3599                 rc = mdt_check_resent_lock(info, obj, lhc);
3600                 if (rc <= 0)
3601                         GOTO(out_obj, rc);
3602
3603                 buf->lb_buf = NULL;
3604                 buf->lb_len = 0;
3605                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3606                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3607                                         &RMF_EADATA);
3608                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3609                                         &RMF_EADATA, RCL_CLIENT);
3610                         /*
3611                          * If it's a replay of layout write intent RPC, the
3612                          * client has saved the extended lovea when
3613                          * it get reply then.
3614                          */
3615                         if (buf->lb_len > 0)
3616                                 mdt_fix_lov_magic(info, buf->lb_buf);
3617                 }
3618
3619                 /*
3620                  * Instantiate some layout components, if @buf contains
3621                  * lovea, then it's a replay of the layout intent write
3622                  * RPC.
3623                  */
3624                 rc = mdt_layout_change(info, obj, layout, buf);
3625                 if (rc)
3626                         GOTO(out_obj, rc);
3627         }
3628 out_obj:
3629         mdt_object_put(info->mti_env, obj);
3630
3631         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3632                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3633
3634 out:
3635         lhc->mlh_reg_lh.cookie = 0;
3636
3637         return rc;
3638 }
3639
3640 static int mdt_intent_reint(enum mdt_it_code opcode,
3641                             struct mdt_thread_info *info,
3642                             struct ldlm_lock **lockp,
3643                             __u64 flags)
3644 {
3645         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3646         struct ldlm_reply      *rep = NULL;
3647         long                    opc;
3648         int                     rc;
3649
3650         static const struct req_format *intent_fmts[REINT_MAX] = {
3651                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3652                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3653         };
3654
3655         ENTRY;
3656
3657         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3658         if (opc < 0)
3659                 RETURN(opc);
3660
3661         if (mdt_it_flavor[opcode].it_reint != opc) {
3662                 CERROR("Reint code %ld doesn't match intent: %d\n",
3663                        opc, opcode);
3664                 RETURN(err_serious(-EPROTO));
3665         }
3666
3667         /* Get lock from request for possible resent case. */
3668         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3669
3670         rc = mdt_reint_internal(info, lhc, opc);
3671
3672         /* Check whether the reply has been packed successfully. */
3673         if (mdt_info_req(info)->rq_repmsg != NULL)
3674                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3675         if (rep == NULL)
3676                 RETURN(err_serious(-EFAULT));
3677
3678         /* MDC expects this in any case */
3679         if (rc != 0)
3680                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3681
3682         /* the open lock or the lock for cross-ref object should be
3683          * returned to the client */
3684         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3685             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3686                 rep->lock_policy_res2 = 0;
3687                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3688                 RETURN(rc);
3689         }
3690
3691         rep->lock_policy_res2 = clear_serious(rc);
3692
3693         if (rep->lock_policy_res2 == -ENOENT &&
3694             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3695             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3696                 rep->lock_policy_res2 = 0;
3697
3698         lhc->mlh_reg_lh.cookie = 0ull;
3699         if (rc == -ENOTCONN || rc == -ENODEV ||
3700             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3701                 /*
3702                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3703                  * will be returned by rq_status, and client at ptlrpc layer
3704                  * will detect this, then disconnect, reconnect the import
3705                  * immediately, instead of impacting the following the rpc.
3706                  */
3707                 RETURN(rc);
3708         }
3709         /*
3710          * For other cases, the error will be returned by intent, and client
3711          * will retrieve the result from intent.
3712          */
3713         RETURN(ELDLM_LOCK_ABORTED);
3714 }
3715
3716 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3717 {
3718         int rc;
3719
3720         switch (itcode) {
3721         case IT_OPEN:
3722                 rc = MDT_IT_OPEN;
3723                 break;
3724         case IT_OPEN|IT_CREAT:
3725                 rc = MDT_IT_OCREAT;
3726                 break;
3727         case IT_CREAT:
3728                 rc = MDT_IT_CREATE;
3729                 break;
3730         case IT_READDIR:
3731                 rc = MDT_IT_READDIR;
3732                 break;
3733         case IT_GETATTR:
3734                 rc = MDT_IT_GETATTR;
3735                 break;
3736         case IT_LOOKUP:
3737                 rc = MDT_IT_LOOKUP;
3738                 break;
3739         case IT_UNLINK:
3740                 rc = MDT_IT_UNLINK;
3741                 break;
3742         case IT_TRUNC:
3743                 rc = MDT_IT_TRUNC;
3744                 break;
3745         case IT_GETXATTR:
3746                 rc = MDT_IT_GETXATTR;
3747                 break;
3748         case IT_LAYOUT:
3749                 rc = MDT_IT_LAYOUT;
3750                 break;
3751         case IT_QUOTA_DQACQ:
3752         case IT_QUOTA_CONN:
3753                 rc = MDT_IT_QUOTA;
3754                 break;
3755         default:
3756                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3757                 rc = -EINVAL;
3758                 break;
3759         }
3760         return rc;
3761 }
3762
3763 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3764                           struct mdt_thread_info *info,
3765                           struct ldlm_lock **lockp, __u64 flags)
3766 {
3767         struct req_capsule      *pill = info->mti_pill;
3768         struct ptlrpc_request   *req = mdt_info_req(info);
3769         struct mdt_it_flavor    *flv;
3770         int opc;
3771         int rc;
3772         ENTRY;
3773
3774         opc = mdt_intent_code(itopc);
3775         if (opc < 0)
3776                 RETURN(-EINVAL);
3777
3778         if (opc == MDT_IT_QUOTA) {
3779                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3780
3781                 if (qmt == NULL)
3782                         RETURN(-EOPNOTSUPP);
3783
3784                 if (mdt_rdonly(req->rq_export))
3785                         RETURN(-EROFS);
3786
3787                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3788                 /* pass the request to quota master */
3789                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3790                                                  mdt_info_req(info), lockp,
3791                                                  flags);
3792                 RETURN(rc);
3793         }
3794
3795         flv = &mdt_it_flavor[opc];
3796         if (flv->it_fmt != NULL)
3797                 req_capsule_extend(pill, flv->it_fmt);
3798
3799         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3800         if (rc < 0)
3801                 RETURN(rc);
3802
3803         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3804                 RETURN(-EROFS);
3805
3806         if (flv->it_act != NULL) {
3807                 struct ldlm_reply *rep;
3808
3809                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3810
3811                 /* execute policy */
3812                 rc = flv->it_act(opc, info, lockp, flags);
3813
3814                 /* Check whether the reply has been packed successfully. */
3815                 if (req->rq_repmsg != NULL) {
3816                         rep = req_capsule_server_get(info->mti_pill,
3817                                                      &RMF_DLM_REP);
3818                         rep->lock_policy_res2 =
3819                                 ptlrpc_status_hton(rep->lock_policy_res2);
3820                 }
3821         }
3822
3823         RETURN(rc);
3824 }
3825
3826 static int mdt_intent_policy(struct ldlm_namespace *ns,
3827                              struct ldlm_lock **lockp, void *req_cookie,
3828                              enum ldlm_mode mode, __u64 flags, void *data)
3829 {
3830         struct tgt_session_info *tsi;
3831         struct mdt_thread_info  *info;
3832         struct ptlrpc_request   *req  =  req_cookie;
3833         struct ldlm_intent      *it;
3834         struct req_capsule      *pill;
3835         int rc;
3836
3837         ENTRY;
3838
3839         LASSERT(req != NULL);
3840
3841         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3842
3843         info = tsi2mdt_info(tsi);
3844         LASSERT(info != NULL);
3845         pill = info->mti_pill;
3846         LASSERT(pill->rc_req == req);
3847
3848         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3849                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3850                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3851                 if (it != NULL) {
3852                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3853                         if (rc == 0)
3854                                 rc = ELDLM_OK;
3855
3856                         /* Lock without inodebits makes no sense and will oops
3857                          * later in ldlm. Let's check it now to see if we have
3858                          * ibits corrupted somewhere in mdt_intent_opc().
3859                          * The case for client miss to set ibits has been
3860                          * processed by others. */
3861                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3862                                         lr_type == LDLM_IBITS,
3863                                      info->mti_dlm_req->lock_desc.\
3864                                         l_policy_data.l_inodebits.bits != 0));
3865                 } else
3866                         rc = err_serious(-EFAULT);
3867         } else {
3868                 /* No intent was provided */
3869                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3870                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3871                 rc = req_capsule_server_pack(pill);
3872                 if (rc)
3873                         rc = err_serious(rc);
3874         }
3875         mdt_thread_info_fini(info);
3876         RETURN(rc);
3877 }
3878
3879 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3880 {
3881         struct seq_server_site  *ss = mdt_seq_site(mdt);
3882
3883         if (ss->ss_node_id == 0)
3884                 return;
3885
3886         if (ss->ss_client_seq != NULL) {
3887                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3888                 ss->ss_client_seq->lcs_exp = NULL;
3889         }
3890
3891         if (ss->ss_server_fld != NULL) {
3892                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3893                 ss->ss_server_fld->lsf_control_exp = NULL;
3894         }
3895 }
3896
3897 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3898 {
3899         struct seq_server_site *ss = mdt_seq_site(mdt);
3900
3901         if (ss == NULL)
3902                 return;
3903
3904         if (ss->ss_server_seq != NULL)
3905                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3906 }
3907
3908 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3909 {
3910         mdt_seq_fini_cli(mdt);
3911         mdt_deregister_seq_exp(mdt);
3912
3913         return seq_site_fini(env, mdt_seq_site(mdt));
3914 }
3915
3916 /**
3917  * It will retrieve its FLDB entries from MDT0, and it only happens
3918  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3919  * and it needs to refresh FLDB from the MDT0.
3920  **/
3921 static int mdt_register_lwp_callback(void *data)
3922 {
3923         struct lu_env           env;
3924         struct mdt_device       *mdt = data;
3925         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3926         int                     rc;
3927         ENTRY;
3928
3929         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3930
3931         rc = lu_env_init(&env, LCT_MD_THREAD);
3932         if (rc < 0) {
3933                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3934                 RETURN(rc);
3935         }
3936
3937         /* Allocate new sequence now to avoid creating local transaction
3938          * in the normal transaction process */
3939         rc = seq_server_check_and_alloc_super(&env,
3940                                               mdt_seq_site(mdt)->ss_server_seq);
3941         if (rc < 0)
3942                 GOTO(out, rc);
3943
3944         if (fld->lsf_new) {
3945                 rc = fld_update_from_controller(&env, fld);
3946                 if (rc != 0) {
3947                         CERROR("%s: cannot update controller: rc = %d\n",
3948                                mdt_obd_name(mdt), rc);
3949                         GOTO(out, rc);
3950                 }
3951         }
3952 out:
3953         lu_env_fini(&env);
3954         RETURN(rc);
3955 }
3956
3957 static int mdt_register_seq_exp(struct mdt_device *mdt)
3958 {
3959         struct seq_server_site  *ss = mdt_seq_site(mdt);
3960         char                    *lwp_name = NULL;
3961         int                     rc;
3962
3963         if (ss->ss_node_id == 0)
3964                 return 0;
3965
3966         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3967         if (lwp_name == NULL)
3968                 GOTO(out_free, rc = -ENOMEM);
3969
3970         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3971         if (rc != 0)
3972                 GOTO(out_free, rc);
3973
3974         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3975                                       NULL, NULL);
3976         if (rc != 0)
3977                 GOTO(out_free, rc);
3978
3979         rc = lustre_register_lwp_item(lwp_name,
3980                                       &ss->ss_server_fld->lsf_control_exp,
3981                                       mdt_register_lwp_callback, mdt);
3982         if (rc != 0) {
3983                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3984                 ss->ss_client_seq->lcs_exp = NULL;
3985                 GOTO(out_free, rc);
3986         }
3987 out_free:
3988         if (lwp_name != NULL)
3989                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3990
3991         return rc;
3992 }
3993
3994 /*
3995  * Init client sequence manager which is used by local MDS to talk to sequence
3996  * controller on remote node.
3997  */
3998 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3999 {
4000         struct seq_server_site  *ss = mdt_seq_site(mdt);
4001         int                     rc;
4002         char                    *prefix;
4003         ENTRY;
4004
4005         /* check if this is adding the first MDC and controller is not yet
4006          * initialized. */
4007         OBD_ALLOC_PTR(ss->ss_client_seq);
4008         if (ss->ss_client_seq == NULL)
4009                 RETURN(-ENOMEM);
4010
4011         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4012         if (prefix == NULL) {
4013                 OBD_FREE_PTR(ss->ss_client_seq);
4014                 ss->ss_client_seq = NULL;
4015                 RETURN(-ENOMEM);
4016         }
4017
4018         /* Note: seq_client_fini will be called in seq_site_fini */
4019         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4020         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4021                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4022                                                             NULL);
4023         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4024         if (rc != 0) {
4025                 OBD_FREE_PTR(ss->ss_client_seq);
4026                 ss->ss_client_seq = NULL;
4027                 RETURN(rc);
4028         }
4029
4030         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4031
4032         RETURN(rc);
4033 }
4034
4035 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4036 {
4037         struct seq_server_site  *ss;
4038         int                     rc;
4039         ENTRY;
4040
4041         ss = mdt_seq_site(mdt);
4042         /* init sequence controller server(MDT0) */
4043         if (ss->ss_node_id == 0) {
4044                 OBD_ALLOC_PTR(ss->ss_control_seq);
4045                 if (ss->ss_control_seq == NULL)
4046                         RETURN(-ENOMEM);
4047
4048                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4049                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4050                                      ss);
4051                 if (rc)
4052                         GOTO(out_seq_fini, rc);
4053         }
4054
4055         /* Init normal sequence server */
4056         OBD_ALLOC_PTR(ss->ss_server_seq);
4057         if (ss->ss_server_seq == NULL)
4058                 GOTO(out_seq_fini, rc = -ENOMEM);
4059
4060         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4061                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4062         if (rc)
4063                 GOTO(out_seq_fini, rc);
4064
4065         /* init seq client for seq server to talk to seq controller(MDT0) */
4066         rc = mdt_seq_init_cli(env, mdt);
4067         if (rc != 0)
4068                 GOTO(out_seq_fini, rc);
4069
4070         if (ss->ss_node_id != 0)
4071                 /* register controller export through lwp */
4072                 rc = mdt_register_seq_exp(mdt);
4073
4074         EXIT;
4075 out_seq_fini:
4076         if (rc)
4077                 mdt_seq_fini(env, mdt);
4078
4079         return rc;
4080 }
4081
4082 /*
4083  * FLD wrappers
4084  */
4085 static int mdt_fld_fini(const struct lu_env *env,
4086                         struct mdt_device *m)
4087 {
4088         struct seq_server_site *ss = mdt_seq_site(m);
4089         ENTRY;
4090
4091         if (ss && ss->ss_server_fld) {
4092                 fld_server_fini(env, ss->ss_server_fld);
4093                 OBD_FREE_PTR(ss->ss_server_fld);
4094                 ss->ss_server_fld = NULL;
4095         }
4096
4097         RETURN(0);
4098 }
4099
4100 static int mdt_fld_init(const struct lu_env *env,
4101                         const char *uuid,
4102                         struct mdt_device *m)
4103 {
4104         struct seq_server_site *ss;
4105         int rc;
4106         ENTRY;
4107
4108         ss = mdt_seq_site(m);
4109
4110         OBD_ALLOC_PTR(ss->ss_server_fld);
4111         if (ss->ss_server_fld == NULL)
4112                 RETURN(rc = -ENOMEM);
4113
4114         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4115                              LU_SEQ_RANGE_MDT);
4116         if (rc) {
4117                 OBD_FREE_PTR(ss->ss_server_fld);
4118                 ss->ss_server_fld = NULL;
4119                 RETURN(rc);
4120         }
4121
4122         RETURN(0);
4123 }
4124
4125 static void mdt_stack_pre_fini(const struct lu_env *env,
4126                            struct mdt_device *m, struct lu_device *top)
4127 {
4128         struct lustre_cfg_bufs  *bufs;
4129         struct lustre_cfg       *lcfg;
4130         struct mdt_thread_info  *info;
4131         ENTRY;
4132
4133         LASSERT(top);
4134
4135         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4136         LASSERT(info != NULL);
4137
4138         bufs = &info->mti_u.bufs;
4139
4140         LASSERT(m->mdt_child_exp);
4141         LASSERT(m->mdt_child_exp->exp_obd);
4142
4143         /* process cleanup, pass mdt obd name to get obd umount flags */
4144         /* XXX: this is needed because all layers are referenced by
4145          * objects (some of them are pinned by osd, for example *
4146          * the proper solution should be a model where object used
4147          * by osd only doesn't have mdt/mdd slices -bzzz */
4148         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4149         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4150         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4151         if (lcfg == NULL)
4152                 RETURN_EXIT;
4153
4154         top->ld_ops->ldo_process_config(env, top, lcfg);
4155         lustre_cfg_free(lcfg);
4156         EXIT;
4157 }
4158
4159 static void mdt_stack_fini(const struct lu_env *env,
4160                            struct mdt_device *m, struct lu_device *top)
4161 {
4162         struct obd_device       *obd = mdt2obd_dev(m);
4163         struct lustre_cfg_bufs  *bufs;
4164         struct lustre_cfg       *lcfg;
4165         struct mdt_thread_info  *info;
4166         char                     flags[3] = "";
4167         ENTRY;
4168
4169         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4170         LASSERT(info != NULL);
4171
4172         lu_dev_del_linkage(top->ld_site, top);
4173
4174         lu_site_purge(env, top->ld_site, -1);
4175
4176         bufs = &info->mti_u.bufs;
4177         /* process cleanup, pass mdt obd name to get obd umount flags */
4178         /* another purpose is to let all layers to release their objects */
4179         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4180         if (obd->obd_force)
4181                 strcat(flags, "F");
4182         if (obd->obd_fail)
4183                 strcat(flags, "A");
4184         lustre_cfg_bufs_set_string(bufs, 1, flags);
4185         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4186         if (lcfg == NULL)
4187                 RETURN_EXIT;
4188
4189         LASSERT(top);
4190         top->ld_ops->ldo_process_config(env, top, lcfg);
4191         lustre_cfg_free(lcfg);
4192
4193         lu_site_purge(env, top->ld_site, -1);
4194
4195         m->mdt_child = NULL;
4196         m->mdt_bottom = NULL;
4197
4198         obd_disconnect(m->mdt_child_exp);
4199         m->mdt_child_exp = NULL;
4200
4201         obd_disconnect(m->mdt_bottom_exp);
4202         m->mdt_child_exp = NULL;
4203 }
4204
4205 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4206                                const char *next, struct obd_export **exp)
4207 {
4208         struct obd_connect_data *data = NULL;
4209         struct obd_device       *obd;
4210         int                      rc;
4211         ENTRY;
4212
4213         OBD_ALLOC_PTR(data);
4214         if (data == NULL)
4215                 GOTO(out, rc = -ENOMEM);
4216
4217         obd = class_name2obd(next);
4218         if (obd == NULL) {
4219                 CERROR("%s: can't locate next device: %s\n",
4220                        mdt_obd_name(m), next);
4221                 GOTO(out, rc = -ENOTCONN);
4222         }
4223
4224         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4225         data->ocd_version = LUSTRE_VERSION_CODE;
4226
4227         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4228         if (rc) {
4229                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4230                        mdt_obd_name(m), next, rc);
4231                 GOTO(out, rc);
4232         }
4233
4234 out:
4235         if (data)
4236                 OBD_FREE_PTR(data);
4237         RETURN(rc);
4238 }
4239
4240 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4241                           struct lustre_cfg *cfg)
4242 {
4243         char                   *dev = lustre_cfg_string(cfg, 0);
4244         int                     rc, name_size, uuid_size;
4245         char                   *name, *uuid, *p;
4246         struct lustre_cfg_bufs *bufs;
4247         struct lustre_cfg      *lcfg;
4248         struct obd_device      *obd;
4249         struct lustre_profile  *lprof;
4250         struct lu_site         *site;
4251         ENTRY;
4252
4253         /* in 1.8 we had the only device in the stack - MDS.
4254          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4255          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4256          * to generate names and setup MDT, MDD. MDT will be using
4257          * generated name to connect to MDD. for MDD the next device
4258          * will be LOD with name taken from so called "profile" which
4259          * is generated by mount_option line
4260          *
4261          * 1.8 MGS generates config. commands like this:
4262          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4263          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4264          * 2.0 MGS generates config. commands like this:
4265          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4266          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4267          *                    3:lustre-MDT0000-mdtlov  4:f
4268          *
4269          * we generate MDD name from MDT one, just replacing T with D
4270          *
4271          * after all the preparations, the logical equivalent will be
4272          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4273          *                    3:lustre-MDT0000-mdtlov  4:f
4274          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4275          *                    3:lustre-MDD0000  4:f
4276          *
4277          *  notice we build the stack from down to top: MDD first, then MDT */
4278
4279         name_size = MAX_OBD_NAME;
4280         uuid_size = MAX_OBD_NAME;
4281
4282         OBD_ALLOC(name, name_size);
4283         OBD_ALLOC(uuid, uuid_size);
4284         if (name == NULL || uuid == NULL)
4285                 GOTO(cleanup_mem, rc = -ENOMEM);
4286
4287         OBD_ALLOC_PTR(bufs);
4288         if (!bufs)
4289                 GOTO(cleanup_mem, rc = -ENOMEM);
4290
4291         strcpy(name, dev);
4292         p = strstr(name, "-MDT");
4293         if (p == NULL)
4294                 GOTO(free_bufs, rc = -ENOMEM);
4295         p[3] = 'D';
4296
4297         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4298
4299         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4300         if (lprof == NULL || lprof->lp_dt == NULL) {
4301                 CERROR("can't find the profile: %s\n",
4302                        lustre_cfg_string(cfg, 0));
4303                 GOTO(free_bufs, rc = -EINVAL);
4304         }
4305
4306         lustre_cfg_bufs_reset(bufs, name);
4307         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4308         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4309         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4310
4311         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4312         if (lcfg == NULL)
4313                 GOTO(put_profile, rc = -ENOMEM);
4314
4315         rc = class_attach(lcfg);
4316         if (rc)
4317                 GOTO(lcfg_cleanup, rc);
4318
4319         obd = class_name2obd(name);
4320         if (!obd) {
4321                 CERROR("Can not find obd %s (%s in config)\n",
4322                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4323                 GOTO(lcfg_cleanup, rc = -EINVAL);
4324         }
4325
4326         lustre_cfg_free(lcfg);
4327
4328         lustre_cfg_bufs_reset(bufs, name);
4329         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4330         lustre_cfg_bufs_set_string(bufs, 2, dev);
4331         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4332
4333         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4334         if (lcfg == NULL)
4335                 GOTO(class_detach, rc = -ENOMEM);
4336
4337         rc = class_setup(obd, lcfg);
4338         if (rc)
4339                 GOTO(class_detach, rc);
4340
4341         /* connect to MDD we just setup */
4342         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4343         if (rc)
4344                 GOTO(class_detach, rc);
4345
4346         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4347         LASSERT(site);
4348         LASSERT(mdt_lu_site(mdt) == NULL);
4349         mdt->mdt_lu_dev.ld_site = site;
4350         site->ls_top_dev = &mdt->mdt_lu_dev;
4351         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4352
4353         /* now connect to bottom OSD */
4354         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4355         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4356         if (rc)
4357                 GOTO(class_detach, rc);
4358         mdt->mdt_bottom =
4359                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4360
4361         rc = lu_env_refill((struct lu_env *)env);
4362         if (rc != 0)
4363                 CERROR("Failure to refill session: '%d'\n", rc);
4364
4365         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4366
4367         EXIT;
4368 class_detach:
4369         if (rc)
4370                 class_detach(obd, lcfg);
4371 lcfg_cleanup:
4372         lustre_cfg_free(lcfg);
4373 put_profile:
4374         class_put_profile(lprof);
4375 free_bufs:
4376         OBD_FREE_PTR(bufs);
4377 cleanup_mem:
4378         if (name)
4379                 OBD_FREE(name, name_size);
4380         if (uuid)
4381                 OBD_FREE(uuid, uuid_size);
4382         RETURN(rc);
4383 }
4384
4385 /* setup quota master target on MDT0 */
4386 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4387                           struct lustre_cfg *cfg)
4388 {
4389         struct obd_device       *obd;
4390         char                    *dev = lustre_cfg_string(cfg, 0);
4391         char                    *qmtname, *uuid, *p;
4392         struct lustre_cfg_bufs  *bufs;
4393         struct lustre_cfg       *lcfg;
4394         struct lustre_profile   *lprof;
4395         struct obd_connect_data *data;
4396         int                      rc;
4397         ENTRY;
4398
4399         LASSERT(mdt->mdt_qmt_exp == NULL);
4400         LASSERT(mdt->mdt_qmt_dev == NULL);
4401
4402         /* quota master is on MDT0 only for now */
4403         if (mdt->mdt_seq_site.ss_node_id != 0)
4404                 RETURN(0);
4405
4406         /* MGS generates config commands which look as follows:
4407          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4408          *                    3:lustre-MDT0000-mdtlov  4:f
4409          *
4410          * We generate the QMT name from the MDT one, just replacing MD with QM
4411          * after all the preparations, the logical equivalent will be:
4412          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4413          *                    3:lustre-MDT0000-osd  4:f */
4414         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4415         OBD_ALLOC(uuid, UUID_MAX);
4416         OBD_ALLOC_PTR(bufs);
4417         OBD_ALLOC_PTR(data);
4418         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4419                 GOTO(cleanup_mem, rc = -ENOMEM);
4420
4421         strcpy(qmtname, dev);
4422         p = strstr(qmtname, "-MDT");
4423         if (p == NULL)
4424                 GOTO(cleanup_mem, rc = -ENOMEM);
4425         /* replace MD with QM */
4426         p[1] = 'Q';
4427         p[2] = 'M';
4428
4429         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4430
4431         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4432         if (lprof == NULL || lprof->lp_dt == NULL) {
4433                 CERROR("can't find profile for %s\n",
4434                        lustre_cfg_string(cfg, 0));
4435                 GOTO(cleanup_mem, rc = -EINVAL);
4436         }
4437
4438         lustre_cfg_bufs_reset(bufs, qmtname);
4439         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4440         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4441         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4442
4443         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4444         if (lcfg == NULL)
4445                 GOTO(put_profile, rc = -ENOMEM);
4446
4447         rc = class_attach(lcfg);
4448         if (rc)
4449                 GOTO(lcfg_cleanup, rc);
4450
4451         obd = class_name2obd(qmtname);
4452         if (!obd) {
4453                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4454                        lustre_cfg_string(cfg, 0));
4455                 GOTO(lcfg_cleanup, rc = -EINVAL);
4456         }
4457
4458         lustre_cfg_free(lcfg);
4459
4460         lustre_cfg_bufs_reset(bufs, qmtname);
4461         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4462         lustre_cfg_bufs_set_string(bufs, 2, dev);
4463
4464         /* for quota, the next device should be the OSD device */
4465         lustre_cfg_bufs_set_string(bufs, 3,
4466                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4467
4468         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4469         if (lcfg == NULL)
4470                 GOTO(class_detach, rc = -ENOMEM);
4471
4472         rc = class_setup(obd, lcfg);
4473         if (rc)
4474                 GOTO(class_detach, rc);
4475
4476         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4477
4478         /* configure local quota objects */
4479         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4480                                                    &mdt->mdt_lu_dev,
4481                                                    mdt->mdt_qmt_dev);
4482         if (rc)
4483                 GOTO(class_cleanup, rc);
4484
4485         /* connect to quota master target */
4486         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4487         data->ocd_version = LUSTRE_VERSION_CODE;
4488         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4489                          data, NULL);
4490         if (rc) {
4491                 CERROR("cannot connect to quota master device %s (%d)\n",
4492                        qmtname, rc);
4493                 GOTO(class_cleanup, rc);
4494         }
4495
4496         EXIT;
4497 class_cleanup:
4498         if (rc) {
4499                 class_manual_cleanup(obd);
4500                 mdt->mdt_qmt_dev = NULL;
4501         }
4502 class_detach:
4503         if (rc)
4504                 class_detach(obd, lcfg);
4505 lcfg_cleanup:
4506         lustre_cfg_free(lcfg);
4507 put_profile:
4508         class_put_profile(lprof);
4509 cleanup_mem:
4510         if (bufs)
4511                 OBD_FREE_PTR(bufs);
4512         if (qmtname)
4513                 OBD_FREE(qmtname, MAX_OBD_NAME);
4514         if (uuid)
4515                 OBD_FREE(uuid, UUID_MAX);
4516         if (data)
4517                 OBD_FREE_PTR(data);
4518         return rc;
4519 }
4520
4521 /* Shutdown quota master target associated with mdt */
4522 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4523 {
4524         ENTRY;
4525
4526         if (mdt->mdt_qmt_exp == NULL)
4527                 RETURN_EXIT;
4528         LASSERT(mdt->mdt_qmt_dev != NULL);
4529
4530         /* the qmt automatically shuts down when the mdt disconnects */
4531         obd_disconnect(mdt->mdt_qmt_exp);
4532         mdt->mdt_qmt_exp = NULL;
4533         mdt->mdt_qmt_dev = NULL;
4534         EXIT;
4535 }
4536
4537 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4538  * for now. This will be removed along with converting rest of MDT code
4539  * to use tgt_session_info */
4540 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4541 {
4542         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4543         int                      rc;
4544
4545         rc = mdt_getxattr(info);
4546
4547         mdt_thread_info_fini(info);
4548         return rc;
4549 }
4550
4551 static struct tgt_handler mdt_tgt_handlers[] = {
4552 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4553                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4554                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4555 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4556                 0,                      MDS_DISCONNECT, tgt_disconnect,
4557                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4558 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4559                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4560                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4561 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4562 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4563 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4564 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4565                                                         mdt_getattr_name),
4566 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4567 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4568 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4569 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4570 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4571 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4572 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4573 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4574                                                         mdt_hsm_progress),
4575 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4576                                                         mdt_hsm_ct_register),
4577 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4578                                                         mdt_hsm_ct_unregister),
4579 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4580                                                         mdt_hsm_state_get),
4581 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4582                                                         mdt_hsm_state_set),
4583 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4584 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4585                                                         mdt_hsm_request),
4586 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4587             MDS_SWAP_LAYOUTS,
4588             mdt_swap_layouts),
4589 };
4590
4591 static struct tgt_handler mdt_sec_ctx_ops[] = {
4592 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4593 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4594 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4595 };
4596
4597 static struct tgt_handler mdt_quota_ops[] = {
4598 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4599 };
4600
4601 static struct tgt_opc_slice mdt_common_slice[] = {
4602         {
4603                 .tos_opc_start  = MDS_FIRST_OPC,
4604                 .tos_opc_end    = MDS_LAST_OPC,
4605                 .tos_hs         = mdt_tgt_handlers
4606         },
4607         {
4608                 .tos_opc_start  = OBD_FIRST_OPC,
4609                 .tos_opc_end    = OBD_LAST_OPC,
4610                 .tos_hs         = tgt_obd_handlers
4611         },
4612         {
4613                 .tos_opc_start  = LDLM_FIRST_OPC,
4614                 .tos_opc_end    = LDLM_LAST_OPC,
4615                 .tos_hs         = tgt_dlm_handlers
4616         },
4617         {
4618                 .tos_opc_start  = SEC_FIRST_OPC,
4619                 .tos_opc_end    = SEC_LAST_OPC,
4620                 .tos_hs         = mdt_sec_ctx_ops
4621         },
4622         {
4623                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4624                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4625                 .tos_hs         = tgt_out_handlers
4626         },
4627         {
4628                 .tos_opc_start  = FLD_FIRST_OPC,
4629                 .tos_opc_end    = FLD_LAST_OPC,
4630                 .tos_hs         = fld_handlers
4631         },
4632         {
4633                 .tos_opc_start  = SEQ_FIRST_OPC,
4634                 .tos_opc_end    = SEQ_LAST_OPC,
4635                 .tos_hs         = seq_handlers
4636         },
4637         {
4638                 .tos_opc_start  = QUOTA_DQACQ,
4639                 .tos_opc_end    = QUOTA_LAST_OPC,
4640                 .tos_hs         = mdt_quota_ops
4641         },
4642         {
4643                 .tos_opc_start  = LLOG_FIRST_OPC,
4644                 .tos_opc_end    = LLOG_LAST_OPC,
4645                 .tos_hs         = tgt_llog_handlers
4646         },
4647         {
4648                 .tos_opc_start  = LFSCK_FIRST_OPC,
4649                 .tos_opc_end    = LFSCK_LAST_OPC,
4650                 .tos_hs         = tgt_lfsck_handlers
4651         },
4652
4653         {
4654                 .tos_hs         = NULL
4655         }
4656 };
4657
4658 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4659 {
4660         struct md_device *next = m->mdt_child;
4661         struct lu_device *d = &m->mdt_lu_dev;
4662         struct obd_device *obd = mdt2obd_dev(m);
4663         struct lfsck_stop stop;
4664
4665         ENTRY;
4666         stop.ls_status = LS_PAUSED;
4667         stop.ls_flags = 0;
4668         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4669
4670         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4671         ping_evictor_stop();
4672
4673         if (m->mdt_opts.mo_coordinator)
4674                 mdt_hsm_cdt_stop(m);
4675
4676         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4677         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4678
4679         if (m->mdt_namespace != NULL)
4680                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4681                                           d->ld_obd->obd_force);
4682
4683         obd_exports_barrier(obd);
4684         obd_zombie_barrier();
4685
4686         mdt_quota_fini(env, m);
4687
4688         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4689
4690         /* Calling the cleanup functions in the same order as in the mdt_init0
4691          * error path
4692          */
4693         mdt_procfs_fini(m);
4694
4695         target_recovery_fini(obd);
4696         upcall_cache_cleanup(m->mdt_identity_cache);
4697         m->mdt_identity_cache = NULL;
4698
4699         mdt_fs_cleanup(env, m);
4700
4701         tgt_fini(env, &m->mdt_lut);
4702
4703         mdt_hsm_cdt_fini(m);
4704
4705         if (m->mdt_namespace != NULL) {
4706                 ldlm_namespace_free_post(m->mdt_namespace);
4707                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4708         }
4709
4710         if (m->mdt_md_root != NULL) {
4711                 mdt_object_put(env, m->mdt_md_root);
4712                 m->mdt_md_root = NULL;
4713         }
4714
4715         mdt_seq_fini(env, m);
4716
4717         mdt_fld_fini(env, m);
4718
4719         /*
4720          * Finish the stack
4721          */
4722         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4723
4724         LASSERT(atomic_read(&d->ld_ref) == 0);
4725
4726         server_put_mount(mdt_obd_name(m), true);
4727
4728         EXIT;
4729 }
4730
4731 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4732
4733 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4734                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4735 {
4736         struct mdt_thread_info    *info;
4737         struct obd_device         *obd;
4738         const char                *dev = lustre_cfg_string(cfg, 0);
4739         const char                *num = lustre_cfg_string(cfg, 2);
4740         struct lustre_mount_info  *lmi = NULL;
4741         struct lustre_sb_info     *lsi;
4742         struct lu_site            *s;
4743         struct seq_server_site    *ss_site;
4744         const char                *identity_upcall = "NONE";
4745         struct md_device          *next;
4746         int                        rc;
4747         long                       node_id;
4748         mntopt_t                   mntopts;
4749         ENTRY;
4750
4751         lu_device_init(&m->mdt_lu_dev, ldt);
4752         /*
4753          * Environment (env) might be missing mdt_thread_key values at that
4754          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4755          * mode.
4756          *
4757          * Usually device allocation path doesn't use module key values, but
4758          * mdt has to do a lot of work here, so allocate key value.
4759          */
4760         rc = lu_env_refill((struct lu_env *)env);
4761         if (rc != 0)
4762                 RETURN(rc);
4763
4764         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4765         LASSERT(info != NULL);
4766
4767         obd = class_name2obd(dev);
4768         LASSERT(obd != NULL);
4769
4770         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4771         m->mdt_opts.mo_evict_tgt_nids = 1;
4772         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4773
4774         /* default is coordinator off, it is started through conf_param
4775          * or /proc */
4776         m->mdt_opts.mo_coordinator = 0;
4777
4778         lmi = server_get_mount(dev);
4779         if (lmi == NULL) {
4780                 CERROR("Cannot get mount info for %s!\n", dev);
4781                 RETURN(-EFAULT);
4782         } else {
4783                 lsi = s2lsi(lmi->lmi_sb);
4784                 /* CMD is supported only in IAM mode */
4785                 LASSERT(num);
4786                 node_id = simple_strtol(num, NULL, 10);
4787                 obd->u.obt.obt_magic = OBT_MAGIC;
4788                 if (lsi->lsi_lmd != NULL &&
4789                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4790                         m->mdt_skip_lfsck = 1;
4791         }
4792
4793         m->mdt_squash.rsi_uid = 0;
4794         m->mdt_squash.rsi_gid = 0;
4795         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4796         init_rwsem(&m->mdt_squash.rsi_sem);
4797         spin_lock_init(&m->mdt_lock);
4798         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4799         m->mdt_enable_remote_dir = 0;
4800         m->mdt_enable_remote_dir_gid = 0;
4801
4802         atomic_set(&m->mdt_mds_mds_conns, 0);
4803         atomic_set(&m->mdt_async_commit_count, 0);
4804
4805         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4806         m->mdt_lu_dev.ld_obd = obd;
4807         /* Set this lu_device to obd for error handling purposes. */
4808         obd->obd_lu_dev = &m->mdt_lu_dev;
4809
4810         /* init the stack */
4811         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4812         if (rc) {
4813                 CERROR("%s: Can't init device stack, rc %d\n",
4814                        mdt_obd_name(m), rc);
4815                 GOTO(err_lmi, rc);
4816         }
4817
4818         s = mdt_lu_site(m);
4819         ss_site = mdt_seq_site(m);
4820         s->ld_seq_site = ss_site;
4821         ss_site->ss_lu = s;
4822
4823         /* set server index */
4824         ss_site->ss_node_id = node_id;
4825
4826         /* failover is the default
4827          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4828          * assumed whose ss_node_id == 0 XXX
4829          * */
4830         obd->obd_replayable = 1;
4831         /* No connection accepted until configurations will finish */
4832         obd->obd_no_conn = 1;
4833
4834         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4835                 char *str = lustre_cfg_string(cfg, 4);
4836                 if (strchr(str, 'n')) {
4837                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4838                         obd->obd_replayable = 0;
4839                 }
4840         }
4841
4842         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4843         if (rc)
4844                 GOTO(err_fini_stack, rc);
4845
4846         rc = mdt_seq_init(env, m);
4847         if (rc)
4848                 GOTO(err_fini_fld, rc);
4849
4850         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4851                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4852         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4853                                               LDLM_NAMESPACE_SERVER,
4854                                               LDLM_NAMESPACE_GREEDY,
4855                                               LDLM_NS_TYPE_MDT);
4856         if (m->mdt_namespace == NULL)
4857                 GOTO(err_fini_seq, rc = -ENOMEM);
4858
4859         m->mdt_namespace->ns_lvbp = m;
4860         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4861
4862         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4863         /* set obd_namespace for compatibility with old code */
4864         obd->obd_namespace = m->mdt_namespace;
4865
4866         rc = mdt_hsm_cdt_init(m);
4867         if (rc != 0) {
4868                 CERROR("%s: error initializing coordinator, rc %d\n",
4869                        mdt_obd_name(m), rc);
4870                 GOTO(err_free_ns, rc);
4871         }
4872
4873         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4874                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4875                       OBD_FAIL_MDS_ALL_REPLY_NET);
4876         if (rc)
4877                 GOTO(err_free_hsm, rc);
4878
4879         rc = mdt_fs_setup(env, m, obd, lsi);
4880         if (rc)
4881                 GOTO(err_tgt, rc);
4882
4883         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
4884
4885         next = m->mdt_child;
4886         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4887                                          &mntopts);
4888         if (rc)
4889                 GOTO(err_fs_cleanup, rc);
4890
4891         if (mntopts & MNTOPT_USERXATTR)
4892                 m->mdt_opts.mo_user_xattr = 1;
4893         else
4894                 m->mdt_opts.mo_user_xattr = 0;
4895
4896         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4897         if (rc)
4898                 GOTO(err_fs_cleanup, rc);
4899
4900         if (mntopts & MNTOPT_ACL)
4901                 m->mdt_opts.mo_acl = 1;
4902         else
4903                 m->mdt_opts.mo_acl = 0;
4904
4905         /* XXX: to support suppgid for ACL, we enable identity_upcall
4906          * by default, otherwise, maybe got unexpected -EACCESS. */
4907         if (m->mdt_opts.mo_acl)
4908                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4909
4910         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4911                                                 identity_upcall,
4912                                                 &mdt_identity_upcall_cache_ops);
4913         if (IS_ERR(m->mdt_identity_cache)) {
4914                 rc = PTR_ERR(m->mdt_identity_cache);
4915                 m->mdt_identity_cache = NULL;
4916                 GOTO(err_fs_cleanup, rc);
4917         }
4918
4919         rc = mdt_procfs_init(m, dev);
4920         if (rc) {
4921                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4922                 GOTO(err_recovery, rc);
4923         }
4924
4925         rc = mdt_quota_init(env, m, cfg);
4926         if (rc)
4927                 GOTO(err_procfs, rc);
4928
4929         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4930         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4931                            "mdt_ldlm_client", m->mdt_ldlm_client);
4932
4933         ping_evictor_start();
4934
4935         /* recovery will be started upon mdt_prepare()
4936          * when the whole stack is complete and ready
4937          * to serve the requests */
4938
4939         /* Reduce the initial timeout on an MDS because it doesn't need such
4940          * a long timeout as an OST does. Adaptive timeouts will adjust this
4941          * value appropriately. */
4942         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4943                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4944
4945         RETURN(0);
4946 err_procfs:
4947         mdt_procfs_fini(m);
4948 err_recovery:
4949         target_recovery_fini(obd);
4950         upcall_cache_cleanup(m->mdt_identity_cache);
4951         m->mdt_identity_cache = NULL;
4952 err_fs_cleanup:
4953         mdt_fs_cleanup(env, m);
4954 err_tgt:
4955         tgt_fini(env, &m->mdt_lut);
4956 err_free_hsm:
4957         mdt_hsm_cdt_fini(m);
4958 err_free_ns:
4959         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4960         obd->obd_namespace = m->mdt_namespace = NULL;
4961 err_fini_seq:
4962         mdt_seq_fini(env, m);
4963 err_fini_fld:
4964         mdt_fld_fini(env, m);
4965 err_fini_stack:
4966         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4967 err_lmi:
4968         if (lmi)
4969                 server_put_mount(dev, true);
4970         return(rc);
4971 }
4972
4973 /* For interoperability, the left element is old parameter, the right one
4974  * is the new version of the parameter, if some parameter is deprecated,
4975  * the new version should be set as NULL. */
4976 static struct cfg_interop_param mdt_interop_param[] = {
4977         { "mdt.group_upcall",   NULL },
4978         { "mdt.quota_type",     NULL },
4979         { "mdd.quota_type",     NULL },
4980         { "mdt.som",            NULL },
4981         { "mdt.rootsquash",     "mdt.root_squash" },
4982         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4983         { NULL }
4984 };
4985
4986 /* used by MGS to process specific configurations */
4987 static int mdt_process_config(const struct lu_env *env,
4988                               struct lu_device *d, struct lustre_cfg *cfg)
4989 {
4990         struct mdt_device *m = mdt_dev(d);
4991         struct md_device *md_next = m->mdt_child;
4992         struct lu_device *next = md2lu_dev(md_next);
4993         int rc;
4994         ENTRY;
4995
4996         switch (cfg->lcfg_command) {
4997         case LCFG_PARAM: {
4998                 struct obd_device          *obd = d->ld_obd;
4999
5000                 /* For interoperability */
5001                 struct cfg_interop_param   *ptr = NULL;
5002                 struct lustre_cfg          *old_cfg = NULL;
5003                 char                       *param = NULL;
5004
5005                 param = lustre_cfg_string(cfg, 1);
5006                 if (param == NULL) {
5007                         CERROR("param is empty\n");
5008                         rc = -EINVAL;
5009                         break;
5010                 }
5011
5012                 ptr = class_find_old_param(param, mdt_interop_param);
5013                 if (ptr != NULL) {
5014                         if (ptr->new_param == NULL) {
5015                                 rc = 0;
5016                                 CWARN("For interoperability, skip this %s."
5017                                       " It is obsolete.\n", ptr->old_param);
5018                                 break;
5019                         }
5020
5021                         CWARN("Found old param %s, changed it to %s.\n",
5022                               ptr->old_param, ptr->new_param);
5023
5024                         old_cfg = cfg;
5025                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5026                         if (IS_ERR(cfg)) {
5027                                 rc = PTR_ERR(cfg);
5028                                 break;
5029                         }
5030                 }
5031
5032                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5033                                               cfg, obd);
5034                 if (rc > 0 || rc == -ENOSYS) {
5035                         /* is it an HSM var ? */
5036                         rc = class_process_proc_param(PARAM_HSM,
5037                                                       hsm_cdt_get_proc_vars(),
5038                                                       cfg, obd);
5039                         if (rc > 0 || rc == -ENOSYS)
5040                                 /* we don't understand; pass it on */
5041                                 rc = next->ld_ops->ldo_process_config(env, next,
5042                                                                       cfg);
5043                 }
5044
5045                 if (old_cfg != NULL)
5046                         lustre_cfg_free(cfg);
5047
5048                 break;
5049         }
5050         default:
5051                 /* others are passed further */
5052                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5053                 break;
5054         }
5055         RETURN(rc);
5056 }
5057
5058 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5059                                           const struct lu_object_header *hdr,
5060                                           struct lu_device *d)
5061 {
5062         struct mdt_object *mo;
5063
5064         ENTRY;
5065
5066         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5067         if (mo != NULL) {
5068                 struct lu_object *o;
5069                 struct lu_object_header *h;
5070
5071                 o = &mo->mot_obj;
5072                 h = &mo->mot_header;
5073                 lu_object_header_init(h);
5074                 lu_object_init(o, h, d);
5075                 lu_object_add_top(h, o);
5076                 o->lo_ops = &mdt_obj_ops;
5077                 spin_lock_init(&mo->mot_write_lock);
5078                 mutex_init(&mo->mot_lov_mutex);
5079                 init_rwsem(&mo->mot_open_sem);
5080                 atomic_set(&mo->mot_open_count, 0);
5081                 RETURN(o);
5082         }
5083         RETURN(NULL);
5084 }
5085
5086 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5087                            const struct lu_object_conf *unused)
5088 {
5089         struct mdt_device *d = mdt_dev(o->lo_dev);
5090         struct lu_device  *under;
5091         struct lu_object  *below;
5092         int                rc = 0;
5093         ENTRY;
5094
5095         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5096                PFID(lu_object_fid(o)));
5097
5098         under = &d->mdt_child->md_lu_dev;
5099         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5100         if (below != NULL) {
5101                 lu_object_add(o, below);
5102         } else
5103                 rc = -ENOMEM;
5104
5105         RETURN(rc);
5106 }
5107
5108 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5109 {
5110         struct mdt_object *mo = mdt_obj(o);
5111         struct lu_object_header *h;
5112         ENTRY;
5113
5114         h = o->lo_header;
5115         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5116                PFID(lu_object_fid(o)));
5117
5118         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5119         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5120
5121         lu_object_fini(o);
5122         lu_object_header_fini(h);
5123         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5124
5125         EXIT;
5126 }
5127
5128 static int mdt_object_print(const struct lu_env *env, void *cookie,
5129                             lu_printer_t p, const struct lu_object *o)
5130 {
5131         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5132
5133         return (*p)(env, cookie,
5134                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5135                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5136                     mdto->mot_cache_attr ? "cache_attr" : "",
5137                     mdto->mot_write_count);
5138 }
5139
5140 static int mdt_prepare(const struct lu_env *env,
5141                 struct lu_device *pdev,
5142                 struct lu_device *cdev)
5143 {
5144         struct mdt_device *mdt = mdt_dev(cdev);
5145         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5146         struct obd_device *obd = cdev->ld_obd;
5147         int rc;
5148
5149         ENTRY;
5150
5151         LASSERT(obd);
5152
5153         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5154         if (rc)
5155                 RETURN(rc);
5156
5157         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5158         if (rc)
5159                 RETURN(rc);
5160
5161         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5162         if (rc)
5163                 RETURN(rc);
5164
5165         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5166         /* The LFSCK instance is registered just now, so it must be there when
5167          * register the namespace to such instance. */
5168         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5169
5170         if (mdt->mdt_seq_site.ss_node_id == 0) {
5171                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5172                                                          &mdt->mdt_md_root_fid);
5173                 if (rc)
5174                         RETURN(rc);
5175         }
5176
5177         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5178
5179         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5180         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5181         LASSERT(obd->obd_no_conn);
5182         spin_lock(&obd->obd_dev_lock);
5183         obd->obd_no_conn = 0;
5184         spin_unlock(&obd->obd_dev_lock);
5185
5186         if (obd->obd_recovering == 0)
5187                 mdt_postrecov(env, mdt);
5188
5189         RETURN(rc);
5190 }
5191
5192 const struct lu_device_operations mdt_lu_ops = {
5193         .ldo_object_alloc   = mdt_object_alloc,
5194         .ldo_process_config = mdt_process_config,
5195         .ldo_prepare        = mdt_prepare,
5196 };
5197
5198 static const struct lu_object_operations mdt_obj_ops = {
5199         .loo_object_init    = mdt_object_init,
5200         .loo_object_free    = mdt_object_free,
5201         .loo_object_print   = mdt_object_print
5202 };
5203
5204 static int mdt_obd_set_info_async(const struct lu_env *env,
5205                                   struct obd_export *exp,
5206                                   __u32 keylen, void *key,
5207                                   __u32 vallen, void *val,
5208                                   struct ptlrpc_request_set *set)
5209 {
5210         int rc;
5211
5212         ENTRY;
5213
5214         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5215                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5216                 RETURN(rc);
5217         }
5218
5219         RETURN(0);
5220 }
5221
5222 /**
5223  * Match client and server connection feature flags.
5224  *
5225  * Compute the compatibility flags for a connection request based on
5226  * features mutually supported by client and server.
5227  *
5228  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5229  * must not be updated here, otherwise a partially initialized value may
5230  * be exposed. After the connection request is successfully processed,
5231  * the top-level MDT connect request handler atomically updates the export
5232  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5233  * reply. \see mdt_connect().
5234  *
5235  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5236  * full struct obd_connect_data. So care must be taken when accessing fields
5237  * that are not present in struct obd_connect_data_v1. See LU-16.
5238  *
5239  * \param exp   the obd_export associated with this client/target pair
5240  * \param mdt   the target device for the connection
5241  * \param data  stores data for this connect request
5242  *
5243  * \retval 0       success
5244  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5245  * \retval -EBADE  client and server feature requirements are incompatible
5246  */
5247 static int mdt_connect_internal(struct obd_export *exp,
5248                                 struct mdt_device *mdt,
5249                                 struct obd_connect_data *data)
5250 {
5251         LASSERT(data != NULL);
5252
5253         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5254
5255         if (mdt->mdt_bottom->dd_rdonly &&
5256             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5257             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5258                 RETURN(-EACCES);
5259
5260         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5261                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5262
5263         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5264
5265         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5266             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5267                 CWARN("%s: client %s does not support ibits lock, either "
5268                       "very old or an invalid client: flags %#llx\n",
5269                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5270                       data->ocd_connect_flags);
5271                 return -EBADE;
5272         }
5273
5274         if (!mdt->mdt_opts.mo_acl)
5275                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5276
5277         if (!mdt->mdt_opts.mo_user_xattr)
5278                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5279
5280         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5281                 data->ocd_brw_size = min(data->ocd_brw_size,
5282                                          (__u32)MD_MAX_BRW_SIZE);
5283                 if (data->ocd_brw_size == 0) {
5284                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
5285                                " ocd_version: %x ocd_grant: %d "
5286                                "ocd_index: %u ocd_brw_size is "
5287                                "unexpectedly zero, network data "
5288                                "corruption? Refusing connection of this"
5289                                " client\n",
5290                                mdt_obd_name(mdt),
5291                                exp->exp_client_uuid.uuid,
5292                                exp, data->ocd_connect_flags, data->ocd_version,
5293                                data->ocd_grant, data->ocd_index);
5294                         return -EPROTO;
5295                 }
5296         }
5297
5298         /* NB: Disregard the rule against updating
5299          * exp_connect_data.ocd_connect_flags in this case, since
5300          * tgt_client_new() needs to know if this is a lightweight
5301          * connection, and it is safe to expose this flag before
5302          * connection processing completes. */
5303         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5304                 spin_lock(&exp->exp_lock);
5305                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5306                 spin_unlock(&exp->exp_lock);
5307         }
5308
5309         data->ocd_version = LUSTRE_VERSION_CODE;
5310
5311         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5312                 CWARN("%s: MDS requires FID support, but client not\n",
5313                       mdt_obd_name(mdt));
5314                 return -EBADE;
5315         }
5316
5317         if (OCD_HAS_FLAG(data, PINGLESS)) {
5318                 if (ptlrpc_pinger_suppress_pings()) {
5319                         spin_lock(&exp->exp_obd->obd_dev_lock);
5320                         list_del_init(&exp->exp_obd_chain_timed);
5321                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5322                 } else {
5323                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5324                 }
5325         }
5326
5327         data->ocd_max_easize = mdt->mdt_max_ea_size;
5328
5329         /* NB: Disregard the rule against updating
5330          * exp_connect_data.ocd_connect_flags in this case, since
5331          * tgt_client_new() needs to know if this is client supports
5332          * multiple modify RPCs, and it is safe to expose this flag before
5333          * connection processing completes. */
5334         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5335                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5336                 spin_lock(&exp->exp_lock);
5337                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5338                 spin_unlock(&exp->exp_lock);
5339         }
5340
5341         return 0;
5342 }
5343
5344 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5345                                    struct mdt_thread_info *info,
5346                                    struct mdt_file_data *mfd)
5347 {
5348         struct lu_context ses;
5349         int rc;
5350         ENTRY;
5351
5352         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5353         if (rc)
5354                 RETURN(rc);
5355
5356         env->le_ses = &ses;
5357         lu_context_enter(&ses);
5358
5359         mdt_ucred(info)->uc_valid = UCRED_OLD;
5360         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5361
5362         lu_context_exit(&ses);
5363         lu_context_fini(&ses);
5364         env->le_ses = NULL;
5365
5366         RETURN(rc);
5367 }
5368
5369 static int mdt_export_cleanup(struct obd_export *exp)
5370 {
5371         struct list_head         closing_list;
5372         struct mdt_export_data  *med = &exp->exp_mdt_data;
5373         struct obd_device       *obd = exp->exp_obd;
5374         struct mdt_device       *mdt;
5375         struct mdt_thread_info  *info;
5376         struct lu_env            env;
5377         struct mdt_file_data    *mfd, *n;
5378         int rc = 0;
5379         ENTRY;
5380
5381         INIT_LIST_HEAD(&closing_list);
5382         spin_lock(&med->med_open_lock);
5383         while (!list_empty(&med->med_open_head)) {
5384                 struct list_head *tmp = med->med_open_head.next;
5385                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5386
5387                 /* Remove mfd handle so it can't be found again.
5388                  * We are consuming the mfd_list reference here. */
5389                 class_handle_unhash(&mfd->mfd_handle);
5390                 list_move_tail(&mfd->mfd_list, &closing_list);
5391         }
5392         spin_unlock(&med->med_open_lock);
5393         mdt = mdt_dev(obd->obd_lu_dev);
5394         LASSERT(mdt != NULL);
5395
5396         rc = lu_env_init(&env, LCT_MD_THREAD);
5397         if (rc)
5398                 RETURN(rc);
5399
5400         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5401         LASSERT(info != NULL);
5402         memset(info, 0, sizeof *info);
5403         info->mti_env = &env;
5404         info->mti_mdt = mdt;
5405         info->mti_exp = exp;
5406
5407         if (!list_empty(&closing_list)) {
5408                 struct md_attr *ma = &info->mti_attr;
5409
5410                 /* Close any open files (which may also cause orphan
5411                  * unlinking). */
5412                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5413                         list_del_init(&mfd->mfd_list);
5414                         ma->ma_need = ma->ma_valid = 0;
5415
5416                         /* This file is being closed due to an eviction, it
5417                          * could have been modified and now dirty regarding to
5418                          * HSM archive, check this!
5419                          * The logic here is to mark a file dirty if there's a
5420                          * chance it was dirtied before the client was evicted,
5421                          * so that we don't have to wait for a release attempt
5422                          * before finding out the file was actually dirty and
5423                          * fail the release. Aggressively marking it dirty here
5424                          * will cause the policy engine to attempt to
5425                          * re-archive it; when rearchiving, we can compare the
5426                          * current version to the HSM data_version and make the
5427                          * archive request into a noop if it's not actually
5428                          * dirty.
5429                          */
5430                         if (mfd->mfd_mode & FMODE_WRITE)
5431                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5432
5433                         /* Don't unlink orphan on failover umount, LU-184 */
5434                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5435                                 ma->ma_valid = MA_FLAGS;
5436                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5437                         }
5438                         mdt_mfd_close(info, mfd);
5439                 }
5440         }
5441         info->mti_mdt = NULL;
5442         /* cleanup client slot early */
5443         /* Do not erase record for recoverable client. */
5444         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5445                 tgt_client_del(&env, exp);
5446         lu_env_fini(&env);
5447
5448         RETURN(rc);
5449 }
5450
5451 static inline void mdt_enable_slc(struct mdt_device *mdt)
5452 {
5453         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5454                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5455 }
5456
5457 static inline void mdt_disable_slc(struct mdt_device *mdt)
5458 {
5459         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5460                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5461 }
5462
5463 static int mdt_obd_disconnect(struct obd_export *exp)
5464 {
5465         int rc;
5466         ENTRY;
5467
5468         LASSERT(exp);
5469         class_export_get(exp);
5470
5471         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5472             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5473                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5474
5475                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5476                         mdt_disable_slc(mdt);
5477         }
5478
5479         rc = server_disconnect_export(exp);
5480         if (rc != 0)
5481                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5482
5483         rc = mdt_export_cleanup(exp);
5484         nodemap_del_member(exp);
5485         class_export_put(exp);
5486         RETURN(rc);
5487 }
5488
5489 /* mds_connect copy */
5490 static int mdt_obd_connect(const struct lu_env *env,
5491                            struct obd_export **exp, struct obd_device *obd,
5492                            struct obd_uuid *cluuid,
5493                            struct obd_connect_data *data,
5494                            void *localdata)
5495 {
5496         struct obd_export       *lexp;
5497         struct lustre_handle    conn = { 0 };
5498         struct mdt_device       *mdt;
5499         int                      rc;
5500         lnet_nid_t              *client_nid = localdata;
5501         ENTRY;
5502
5503         LASSERT(env != NULL);
5504         LASSERT(data != NULL);
5505
5506         if (!exp || !obd || !cluuid)
5507                 RETURN(-EINVAL);
5508
5509         mdt = mdt_dev(obd->obd_lu_dev);
5510
5511         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5512             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5513                 atomic_inc(&mdt->mdt_mds_mds_conns);
5514                 mdt_enable_slc(mdt);
5515         }
5516
5517         /*
5518          * first, check whether the stack is ready to handle requests
5519          * XXX: probably not very appropriate method is used now
5520          *      at some point we should find a better one
5521          */
5522         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5523             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5524             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5525                 rc = obd_get_info(env, mdt->mdt_child_exp,
5526                                   sizeof(KEY_OSP_CONNECTED),
5527                                   KEY_OSP_CONNECTED, NULL, NULL);
5528                 if (rc)
5529                         RETURN(-EAGAIN);
5530                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5531         }
5532
5533         rc = class_connect(&conn, obd, cluuid);
5534         if (rc)
5535                 RETURN(rc);
5536
5537         lexp = class_conn2export(&conn);
5538         LASSERT(lexp != NULL);
5539
5540         rc = nodemap_add_member(*client_nid, lexp);
5541         if (rc != 0 && rc != -EEXIST)
5542                 GOTO(out, rc);
5543
5544         rc = mdt_connect_internal(lexp, mdt, data);
5545         if (rc == 0) {
5546                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5547
5548                 LASSERT(lcd);
5549                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5550                 rc = tgt_client_new(env, lexp);
5551                 if (rc == 0)
5552                         mdt_export_stats_init(obd, lexp, localdata);
5553         }
5554 out:
5555         if (rc != 0) {
5556                 class_disconnect(lexp);
5557                 nodemap_del_member(lexp);
5558                 *exp = NULL;
5559         } else {
5560                 *exp = lexp;
5561                 /* Because we do not want this export to be evicted by pinger,
5562                  * let's not add this export to the timed chain list. */
5563                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5564                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5565                         list_del_init(&lexp->exp_obd_chain_timed);
5566                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5567                 }
5568         }
5569
5570         RETURN(rc);
5571 }
5572
5573 static int mdt_obd_reconnect(const struct lu_env *env,
5574                              struct obd_export *exp, struct obd_device *obd,
5575                              struct obd_uuid *cluuid,
5576                              struct obd_connect_data *data,
5577                              void *localdata)
5578 {
5579         lnet_nid_t             *client_nid = localdata;
5580         int                     rc;
5581         ENTRY;
5582
5583         if (exp == NULL || obd == NULL || cluuid == NULL)
5584                 RETURN(-EINVAL);
5585
5586         rc = nodemap_add_member(*client_nid, exp);
5587         if (rc != 0 && rc != -EEXIST)
5588                 RETURN(rc);
5589
5590         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5591         if (rc == 0)
5592                 mdt_export_stats_init(obd, exp, localdata);
5593         else
5594                 nodemap_del_member(exp);
5595
5596         RETURN(rc);
5597 }
5598
5599 /* FIXME: Can we avoid using these two interfaces? */
5600 static int mdt_init_export(struct obd_export *exp)
5601 {
5602         struct mdt_export_data *med = &exp->exp_mdt_data;
5603         int                     rc;
5604         ENTRY;
5605
5606         INIT_LIST_HEAD(&med->med_open_head);
5607         spin_lock_init(&med->med_open_lock);
5608         spin_lock(&exp->exp_lock);
5609         exp->exp_connecting = 1;
5610         spin_unlock(&exp->exp_lock);
5611
5612         /* self-export doesn't need client data and ldlm initialization */
5613         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5614                                      &exp->exp_client_uuid)))
5615                 RETURN(0);
5616
5617         rc = tgt_client_alloc(exp);
5618         if (rc)
5619                 GOTO(err, rc);
5620
5621         rc = ldlm_init_export(exp);
5622         if (rc)
5623                 GOTO(err_free, rc);
5624
5625         RETURN(rc);
5626
5627 err_free:
5628         tgt_client_free(exp);
5629 err:
5630         CERROR("%s: Failed to initialize export: rc = %d\n",
5631                exp->exp_obd->obd_name, rc);
5632         return rc;
5633 }
5634
5635 static int mdt_destroy_export(struct obd_export *exp)
5636 {
5637         ENTRY;
5638
5639         target_destroy_export(exp);
5640         /* destroy can be called from failed obd_setup, so
5641          * checking uuid is safer than obd_self_export */
5642         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5643                                      &exp->exp_client_uuid)))
5644                 RETURN(0);
5645
5646         ldlm_destroy_export(exp);
5647         tgt_client_free(exp);
5648
5649         LASSERT(list_empty(&exp->exp_outstanding_replies));
5650         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5651
5652         RETURN(0);
5653 }
5654
5655 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5656                    struct linkea_data *ldata)
5657 {
5658         int rc;
5659
5660         LASSERT(ldata->ld_buf->lb_buf != NULL);
5661
5662         if (!mdt_object_exists(mdt_obj))
5663                 return -ENODATA;
5664
5665         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5666                           ldata->ld_buf, XATTR_NAME_LINK);
5667         if (rc == -ERANGE) {
5668                 /* Buf was too small, figure out what we need. */
5669                 lu_buf_free(ldata->ld_buf);
5670                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5671                                   ldata->ld_buf, XATTR_NAME_LINK);
5672                 if (rc < 0)
5673                         return rc;
5674                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5675                 if (ldata->ld_buf->lb_buf == NULL)
5676                         return -ENOMEM;
5677                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5678                                   ldata->ld_buf, XATTR_NAME_LINK);
5679         }
5680         if (rc < 0)
5681                 return rc;
5682
5683         return linkea_init_with_rec(ldata);
5684 }
5685
5686 /**
5687  * Given an MDT object, try to look up the full path to the object.
5688  * Part of the MDT layer implementation of lfs fid2path.
5689  *
5690  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5691  * \param[in]     obj   Object to do path lookup of
5692  * \param[in,out] fp    User-provided struct to store path information
5693  * \param[in]     root_fid Root FID of current path should reach
5694  *
5695  * \retval 0 Lookup successful, path information stored in fp
5696  * \retval -EAGAIN Lookup failed, usually because object is being moved
5697  * \retval negative errno if there was a problem
5698  */
5699 static int mdt_path_current(struct mdt_thread_info *info,
5700                             struct mdt_object *obj,
5701                             struct getinfo_fid2path *fp,
5702                             struct lu_fid *root_fid)
5703 {
5704         struct mdt_device       *mdt = info->mti_mdt;
5705         struct mdt_object       *mdt_obj;
5706         struct link_ea_header   *leh;
5707         struct link_ea_entry    *lee;
5708         struct lu_name          *tmpname = &info->mti_name;
5709         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5710         struct lu_buf           *buf = &info->mti_big_buf;
5711         char                    *ptr;
5712         int                     reclen;
5713         struct linkea_data      ldata = { NULL };
5714         int                     rc = 0;
5715         bool                    first = true;
5716         ENTRY;
5717
5718         /* temp buffer for path element, the buffer will be finally freed
5719          * in mdt_thread_info_fini */
5720         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5721         if (buf->lb_buf == NULL)
5722                 RETURN(-ENOMEM);
5723
5724         ldata.ld_buf = buf;
5725         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5726         *ptr = 0;
5727         --ptr;
5728         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5729
5730         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5731                 struct lu_buf           lmv_buf;
5732
5733                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5734                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5735                         GOTO(out, rc = -ENOENT);
5736
5737                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5738                         mdt_obj = obj;
5739                         mdt_object_get(info->mti_env, mdt_obj);
5740                 } else {
5741                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5742                         if (IS_ERR(mdt_obj))
5743                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5744                 }
5745
5746                 if (!mdt_object_exists(mdt_obj)) {
5747                         mdt_object_put(info->mti_env, mdt_obj);
5748                         GOTO(out, rc = -ENOENT);
5749                 }
5750
5751                 if (mdt_object_remote(mdt_obj)) {
5752                         mdt_object_put(info->mti_env, mdt_obj);
5753                         GOTO(remote_out, rc = -EREMOTE);
5754                 }
5755
5756                 rc = mdt_links_read(info, mdt_obj, &ldata);
5757                 if (rc != 0) {
5758                         mdt_object_put(info->mti_env, mdt_obj);
5759                         GOTO(out, rc);
5760                 }
5761
5762                 leh = buf->lb_buf;
5763                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5764                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5765                 /* If set, use link #linkno for path lookup, otherwise use
5766                    link #0.  Only do this for the final path element. */
5767                 if (first && fp->gf_linkno < leh->leh_reccount) {
5768                         int count;
5769                         for (count = 0; count < fp->gf_linkno; count++) {
5770                                 lee = (struct link_ea_entry *)
5771                                      ((char *)lee + reclen);
5772                                 linkea_entry_unpack(lee, &reclen, tmpname,
5773                                                     tmpfid);
5774                         }
5775                         if (fp->gf_linkno < leh->leh_reccount - 1)
5776                                 /* indicate to user there are more links */
5777                                 fp->gf_linkno++;
5778                 }
5779
5780                 lmv_buf.lb_buf = info->mti_xattr_buf;
5781                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5782                 /* Check if it is slave stripes */
5783                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5784                                   &lmv_buf, XATTR_NAME_LMV);
5785                 mdt_object_put(info->mti_env, mdt_obj);
5786                 if (rc > 0) {
5787                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5788
5789                         /* For slave stripes, get its master */
5790                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5791                                 fp->gf_fid = *tmpfid;
5792                                 continue;
5793                         }
5794                 } else if (rc < 0 && rc != -ENODATA) {
5795                         GOTO(out, rc);
5796                 }
5797
5798                 rc = 0;
5799
5800                 /* Pack the name in the end of the buffer */
5801                 ptr -= tmpname->ln_namelen;
5802                 if (ptr - 1 <= fp->gf_u.gf_path)
5803                         GOTO(out, rc = -EOVERFLOW);
5804                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5805                 *(--ptr) = '/';
5806
5807                 /* keep the last resolved fid to the client, so the
5808                  * client will build the left path on another MDT for
5809                  * remote object */
5810                 fp->gf_fid = *tmpfid;
5811
5812                 first = false;
5813         }
5814
5815 remote_out:
5816         ptr++; /* skip leading / */
5817         memmove(fp->gf_u.gf_path, ptr,
5818                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5819
5820 out:
5821         RETURN(rc);
5822 }
5823
5824 /**
5825  * Given an MDT object, use mdt_path_current to get the path.
5826  * Essentially a wrapper to retry mdt_path_current a set number of times
5827  * if -EAGAIN is returned (usually because an object is being moved).
5828  *
5829  * Part of the MDT layer implementation of lfs fid2path.
5830  *
5831  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5832  * \param[in]     obj   Object to do path lookup of
5833  * \param[in,out] fp    User-provided struct for arguments and to store path
5834  *                      information
5835  *
5836  * \retval 0 Lookup successful, path information stored in fp
5837  * \retval negative errno if there was a problem
5838  */
5839 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5840                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5841 {
5842         struct mdt_device       *mdt = info->mti_mdt;
5843         int                     tries = 3;
5844         int                     rc = -EAGAIN;
5845         ENTRY;
5846
5847         if (fp->gf_pathlen < 3)
5848                 RETURN(-EOVERFLOW);
5849
5850         if (root_fid == NULL)
5851                 root_fid = &mdt->mdt_md_root_fid;
5852
5853         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5854                 fp->gf_u.gf_path[0] = '\0';
5855                 RETURN(0);
5856         }
5857
5858         /* Retry multiple times in case file is being moved */
5859         while (tries-- && rc == -EAGAIN)
5860                 rc = mdt_path_current(info, obj, fp, root_fid);
5861
5862         RETURN(rc);
5863 }
5864
5865 /**
5866  * Get the full path of the provided FID, as of changelog record recno.
5867  *
5868  * This checks sanity and looks up object for user provided FID
5869  * before calling the actual path lookup code.
5870  *
5871  * Part of the MDT layer implementation of lfs fid2path.
5872  *
5873  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5874  * \param[in,out] fp    User-provided struct for arguments and to store path
5875  *                      information
5876  *
5877  * \retval 0 Lookup successful, path information and recno stored in fp
5878  * \retval -ENOENT, object does not exist
5879  * \retval negative errno if there was a problem
5880  */
5881 static int mdt_fid2path(struct mdt_thread_info *info,
5882                         struct lu_fid *root_fid,
5883                         struct getinfo_fid2path *fp)
5884 {
5885         struct mdt_device *mdt = info->mti_mdt;
5886         struct mdt_object *obj;
5887         int    rc;
5888         ENTRY;
5889
5890         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
5891                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5892
5893         if (!fid_is_sane(&fp->gf_fid))
5894                 RETURN(-EINVAL);
5895
5896         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5897                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
5898                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
5899                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5900                 RETURN(-EINVAL);
5901         }
5902
5903         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5904         if (IS_ERR(obj)) {
5905                 rc = PTR_ERR(obj);
5906                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5907                        PFID(&fp->gf_fid), rc);
5908                 RETURN(rc);
5909         }
5910
5911         if (mdt_object_remote(obj))
5912                 rc = -EREMOTE;
5913         else if (!mdt_object_exists(obj))
5914                 rc = -ENOENT;
5915         else
5916                 rc = 0;
5917
5918         if (rc < 0) {
5919                 mdt_object_put(info->mti_env, obj);
5920                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5921                        PFID(&fp->gf_fid), rc);
5922                 RETURN(rc);
5923         }
5924
5925         rc = mdt_path(info, obj, fp, root_fid);
5926
5927         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
5928                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5929                fp->gf_recno, fp->gf_linkno);
5930
5931         mdt_object_put(info->mti_env, obj);
5932
5933         RETURN(rc);
5934 }
5935
5936 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5937                             void *val, int vallen)
5938 {
5939         struct getinfo_fid2path *fpout, *fpin;
5940         struct lu_fid *root_fid = NULL;
5941         int rc = 0;
5942
5943         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5944         fpout = val;
5945
5946         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5947                 lustre_swab_fid2path(fpin);
5948
5949         memcpy(fpout, fpin, sizeof(*fpin));
5950         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5951                 RETURN(-EINVAL);
5952
5953         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5954                       sizeof(struct lu_fid)) {
5955                 /* client sent its root FID, which is normally fileset FID */
5956                 root_fid = fpin->gf_u.gf_root_fid;
5957                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5958                         lustre_swab_lu_fid(root_fid);
5959
5960                 if (root_fid != NULL && !fid_is_sane(root_fid))
5961                         RETURN(-EINVAL);
5962         }
5963
5964         rc = mdt_fid2path(info, root_fid, fpout);
5965         RETURN(rc);
5966 }
5967
5968 int mdt_get_info(struct tgt_session_info *tsi)
5969 {
5970         char    *key;
5971         int      keylen;
5972         __u32   *vallen;
5973         void    *valout;
5974         int      rc;
5975
5976         ENTRY;
5977
5978         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5979         if (key == NULL) {
5980                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5981                 RETURN(err_serious(-EFAULT));
5982         }
5983         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5984                                       RCL_CLIENT);
5985
5986         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5987         if (vallen == NULL) {
5988                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5989                                 tgt_name(tsi->tsi_tgt));
5990                 RETURN(err_serious(-EFAULT));
5991         }
5992
5993         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5994                              *vallen);
5995         rc = req_capsule_server_pack(tsi->tsi_pill);
5996         if (rc)
5997                 RETURN(err_serious(rc));
5998
5999         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6000         if (valout == NULL) {
6001                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6002                                 tgt_name(tsi->tsi_tgt));
6003                 RETURN(err_serious(-EFAULT));
6004         }
6005
6006         if (KEY_IS(KEY_FID2PATH)) {
6007                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6008
6009                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6010                 mdt_thread_info_fini(info);
6011         } else {
6012                 rc = -EINVAL;
6013         }
6014         RETURN(rc);
6015 }
6016
6017 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6018 {
6019         struct obd_ioctl_data *data = karg;
6020         struct lu_fid *fid;
6021         __u64 version;
6022         struct mdt_object *obj;
6023         struct mdt_lock_handle  *lh;
6024         int rc;
6025         ENTRY;
6026
6027         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6028             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6029                 RETURN(-EINVAL);
6030
6031         fid = (struct lu_fid *)data->ioc_inlbuf1;
6032
6033         if (!fid_is_sane(fid))
6034                 RETURN(-EINVAL);
6035
6036         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6037
6038         lh = &mti->mti_lh[MDT_LH_PARENT];
6039         mdt_lock_reg_init(lh, LCK_CR);
6040
6041         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6042         if (IS_ERR(obj))
6043                 RETURN(PTR_ERR(obj));
6044
6045         if (mdt_object_remote(obj)) {
6046                 rc = -EREMOTE;
6047                 /**
6048                  * before calling version get the correct MDS should be
6049                  * fid, this is error to find remote object here
6050                  */
6051                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6052         } else if (!mdt_object_exists(obj)) {
6053                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6054                 rc = -ENOENT;
6055         } else {
6056                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6057                *(__u64 *)data->ioc_inlbuf2 = version;
6058                 rc = 0;
6059         }
6060         mdt_object_unlock_put(mti, obj, lh, 1);
6061         RETURN(rc);
6062 }
6063
6064 /* ioctls on obd dev */
6065 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6066                          void *karg, void __user *uarg)
6067 {
6068         struct lu_env      env;
6069         struct obd_device *obd = exp->exp_obd;
6070         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6071         struct dt_device  *dt = mdt->mdt_bottom;
6072         int rc;
6073
6074         ENTRY;
6075         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6076         rc = lu_env_init(&env, LCT_MD_THREAD);
6077         if (rc)
6078                 RETURN(rc);
6079
6080         switch (cmd) {
6081         case OBD_IOC_SYNC:
6082                 rc = mdt_device_sync(&env, mdt);
6083                 break;
6084         case OBD_IOC_SET_READONLY:
6085                 rc = dt_sync(&env, dt);
6086                 if (rc == 0)
6087                         rc = dt_ro(&env, dt);
6088                 break;
6089         case OBD_IOC_ABORT_RECOVERY:
6090                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6091                 obd->obd_abort_recovery = 1;
6092                 target_stop_recovery_thread(obd);
6093                 rc = 0;
6094                 break;
6095         case OBD_IOC_CHANGELOG_REG:
6096         case OBD_IOC_CHANGELOG_DEREG:
6097         case OBD_IOC_CHANGELOG_CLEAR:
6098                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6099                                                            mdt->mdt_child,
6100                                                            cmd, len, karg);
6101                 break;
6102         case OBD_IOC_START_LFSCK: {
6103                 struct md_device *next = mdt->mdt_child;
6104                 struct obd_ioctl_data *data = karg;
6105                 struct lfsck_start_param lsp;
6106
6107                 if (unlikely(data == NULL)) {
6108                         rc = -EINVAL;
6109                         break;
6110                 }
6111
6112                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6113                 lsp.lsp_index_valid = 0;
6114                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6115                 break;
6116         }
6117         case OBD_IOC_STOP_LFSCK: {
6118                 struct md_device        *next = mdt->mdt_child;
6119                 struct obd_ioctl_data   *data = karg;
6120                 struct lfsck_stop        stop;
6121
6122                 stop.ls_status = LS_STOPPED;
6123                 /* Old lfsck utils may pass NULL @stop. */
6124                 if (data->ioc_inlbuf1 == NULL)
6125                         stop.ls_flags = 0;
6126                 else
6127                         stop.ls_flags =
6128                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6129
6130                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6131                 break;
6132         }
6133         case OBD_IOC_QUERY_LFSCK: {
6134                 struct md_device        *next = mdt->mdt_child;
6135                 struct obd_ioctl_data   *data = karg;
6136
6137                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6138                                                  data->ioc_inlbuf1);
6139                 break;
6140         }
6141         case OBD_IOC_GET_OBJ_VERSION: {
6142                 struct mdt_thread_info *mti;
6143                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6144                 memset(mti, 0, sizeof *mti);
6145                 mti->mti_env = &env;
6146                 mti->mti_mdt = mdt;
6147                 mti->mti_exp = exp;
6148
6149                 rc = mdt_ioc_version_get(mti, karg);
6150                 break;
6151         }
6152         case OBD_IOC_CATLOGLIST: {
6153                 struct mdt_thread_info *mti;
6154
6155                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6156                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6157                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6158                                        &mti->mti_tmp_fid1);
6159                 break;
6160          }
6161         default:
6162                 rc = -EOPNOTSUPP;
6163                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6164                         mdt_obd_name(mdt), cmd, rc);
6165         }
6166
6167         lu_env_fini(&env);
6168         RETURN(rc);
6169 }
6170
6171 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6172 {
6173         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6174         int rc;
6175         ENTRY;
6176
6177         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6178                 struct lfsck_start_param lsp;
6179
6180                 lsp.lsp_start = NULL;
6181                 lsp.lsp_index_valid = 0;
6182                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6183                                                            OBD_IOC_START_LFSCK,
6184                                                            0, &lsp);
6185                 if (rc != 0 && rc != -EALREADY)
6186                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6187                               mdt_obd_name(mdt), rc);
6188         }
6189
6190         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6191         RETURN(rc);
6192 }
6193
6194 static int mdt_obd_postrecov(struct obd_device *obd)
6195 {
6196         struct lu_env env;
6197         int rc;
6198
6199         rc = lu_env_init(&env, LCT_MD_THREAD);
6200         if (rc)
6201                 RETURN(rc);
6202         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6203         lu_env_fini(&env);
6204         return rc;
6205 }
6206
6207 static struct obd_ops mdt_obd_device_ops = {
6208         .o_owner          = THIS_MODULE,
6209         .o_set_info_async = mdt_obd_set_info_async,
6210         .o_connect        = mdt_obd_connect,
6211         .o_reconnect      = mdt_obd_reconnect,
6212         .o_disconnect     = mdt_obd_disconnect,
6213         .o_init_export    = mdt_init_export,
6214         .o_destroy_export = mdt_destroy_export,
6215         .o_iocontrol      = mdt_iocontrol,
6216         .o_postrecov      = mdt_obd_postrecov,
6217 };
6218
6219 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6220                                          struct lu_device *d)
6221 {
6222         struct mdt_device *m = mdt_dev(d);
6223         ENTRY;
6224
6225         mdt_fini(env, m);
6226         RETURN(NULL);
6227 }
6228
6229 static struct lu_device *mdt_device_free(const struct lu_env *env,
6230                                          struct lu_device *d)
6231 {
6232         struct mdt_device *m = mdt_dev(d);
6233         ENTRY;
6234
6235         lu_device_fini(&m->mdt_lu_dev);
6236         OBD_FREE_PTR(m);
6237
6238         RETURN(NULL);
6239 }
6240
6241 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6242                                           struct lu_device_type *t,
6243                                           struct lustre_cfg *cfg)
6244 {
6245         struct lu_device  *l;
6246         struct mdt_device *m;
6247
6248         OBD_ALLOC_PTR(m);
6249         if (m != NULL) {
6250                 int rc;
6251
6252                 l = &m->mdt_lu_dev;
6253                 rc = mdt_init0(env, m, t, cfg);
6254                 if (rc != 0) {
6255                         mdt_device_free(env, l);
6256                         l = ERR_PTR(rc);
6257                         return l;
6258                 }
6259         } else
6260                 l = ERR_PTR(-ENOMEM);
6261         return l;
6262 }
6263
6264 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6265 LU_KEY_INIT(mdt, struct mdt_thread_info);
6266
6267 static void mdt_key_fini(const struct lu_context *ctx,
6268                          struct lu_context_key *key, void* data)
6269 {
6270         struct mdt_thread_info *info = data;
6271
6272         if (info->mti_big_lmm) {
6273                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6274                 info->mti_big_lmm = NULL;
6275                 info->mti_big_lmmsize = 0;
6276         }
6277         OBD_FREE_PTR(info);
6278 }
6279
6280 /* context key: mdt_thread_key */
6281 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6282
6283 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6284 {
6285         return lu_ucred(info->mti_env);
6286 }
6287
6288 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6289 {
6290         return lu_ucred_check(info->mti_env);
6291 }
6292
6293 /**
6294  * Enable/disable COS (Commit On Sharing).
6295  *
6296  * Set/Clear the COS flag in mdt options.
6297  *
6298  * \param mdt mdt device
6299  * \param val 0 disables COS, other values enable COS
6300  */
6301 void mdt_enable_cos(struct mdt_device *mdt, int val)
6302 {
6303         struct lu_env env;
6304         int rc;
6305
6306         mdt->mdt_opts.mo_cos = !!val;
6307         rc = lu_env_init(&env, LCT_LOCAL);
6308         if (unlikely(rc != 0)) {
6309                 CWARN("%s: lu_env initialization failed, cannot "
6310                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6311                 return;
6312         }
6313         mdt_device_sync(&env, mdt);
6314         lu_env_fini(&env);
6315 }
6316
6317 /**
6318  * Check COS (Commit On Sharing) status.
6319  *
6320  * Return COS flag status.
6321  *
6322  * \param mdt mdt device
6323  */
6324 int mdt_cos_is_enabled(struct mdt_device *mdt)
6325 {
6326         return mdt->mdt_opts.mo_cos != 0;
6327 }
6328
6329 static struct lu_device_type_operations mdt_device_type_ops = {
6330         .ldto_device_alloc = mdt_device_alloc,
6331         .ldto_device_free  = mdt_device_free,
6332         .ldto_device_fini  = mdt_device_fini
6333 };
6334
6335 static struct lu_device_type mdt_device_type = {
6336         .ldt_tags     = LU_DEVICE_MD,
6337         .ldt_name     = LUSTRE_MDT_NAME,
6338         .ldt_ops      = &mdt_device_type_ops,
6339         .ldt_ctx_tags = LCT_MD_THREAD
6340 };
6341
6342 static int __init mdt_init(void)
6343 {
6344         int rc;
6345
6346         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6347                  FID_NOBRACE_LEN + 1);
6348         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6349                  FID_LEN + 1);
6350         rc = lu_kmem_init(mdt_caches);
6351         if (rc)
6352                 return rc;
6353
6354         rc = mds_mod_init();
6355         if (rc)
6356                 GOTO(lu_fini, rc);
6357
6358         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6359                                  LUSTRE_MDT_NAME, &mdt_device_type);
6360         if (rc)
6361                 GOTO(mds_fini, rc);
6362 lu_fini:
6363         if (rc)
6364                 lu_kmem_fini(mdt_caches);
6365 mds_fini:
6366         if (rc)
6367                 mds_mod_exit();
6368         return rc;
6369 }
6370
6371 static void __exit mdt_exit(void)
6372 {
6373         class_unregister_type(LUSTRE_MDT_NAME);
6374         mds_mod_exit();
6375         lu_kmem_fini(mdt_caches);
6376 }
6377
6378 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6379 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6380 MODULE_VERSION(LUSTRE_VERSION_STRING);
6381 MODULE_LICENSE("GPL");
6382
6383 module_init(mdt_init);
6384 module_exit(mdt_exit);