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, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/mdt/mdt_handler.c
  32  *
  33  * Lustre Metadata Target (mdt) request handler
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Mike Shaver <shaver@clusterfs.com>
  39  * Author: Nikita Danilov <nikita@clusterfs.com>
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  * Author: Yury Umanets <umka@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47 #include <linux/pagemap.h>
  48
  49 #include <dt_object.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_export.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_lfsck.h>
  54 #include <lustre_log.h>
  55 #include <lustre_nodemap.h>
  56 #include <lustre_mds.h>
  57 #include <uapi/linux/lustre/lustre_param.h>
  58 #include <lustre_quota.h>
  59 #include <lustre_swab.h>
  60 #include <lustre_lmv.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock)
 167 {
 168         mdt_lock_reg_init(lh, lock->l_req_mode);
 169         if (lock->l_req_mode == LCK_GROUP)
 170                 lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid;
 171 }
 172
 173 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 174                        const struct lu_name *lname)
 175 {
 176         lh->mlh_reg_mode = lock_mode;
 177         lh->mlh_pdo_mode = LCK_MINMODE;
 178         lh->mlh_rreg_mode = lock_mode;
 179         lh->mlh_type = MDT_PDO_LOCK;
 180
 181         if (lu_name_is_valid(lname)) {
 182                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 183                                                      lname->ln_namelen);
 184                 /* XXX Workaround for LU-2856
 185                  *
 186                  * Zero is a valid return value of full_name_hash, but
 187                  * several users of mlh_pdo_hash assume a non-zero
 188                  * hash value. We therefore map zero onto an
 189                  * arbitrary, but consistent value (1) to avoid
 190                  * problems further down the road. */
 191                 if (unlikely(lh->mlh_pdo_hash == 0))
 192                         lh->mlh_pdo_hash = 1;
 193         } else {
 194                 lh->mlh_pdo_hash = 0;
 195         }
 196 }
 197
 198 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 199                               struct mdt_lock_handle *lh)
 200 {
 201         mdl_mode_t mode;
 202         ENTRY;
 203
 204         /*
 205          * Any dir access needs couple of locks:
 206          *
 207          * 1) on part of dir we gonna take lookup/modify;
 208          *
 209          * 2) on whole dir to protect it from concurrent splitting and/or to
 210          * flush client's cache for readdir().
 211          *
 212          * so, for a given mode and object this routine decides what lock mode
 213          * to use for lock #2:
 214          *
 215          * 1) if caller's gonna lookup in dir then we need to protect dir from
 216          * being splitted only - LCK_CR
 217          *
 218          * 2) if caller's gonna modify dir then we need to protect dir from
 219          * being splitted and to flush cache - LCK_CW
 220          *
 221          * 3) if caller's gonna modify dir and that dir seems ready for
 222          * splitting then we need to protect it from any type of access
 223          * (lookup/modify/split) - LCK_EX --bzzz
 224          */
 225
 226         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 227         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 228
 229         /*
 230          * Ask underlaying level its opinion about preferable PDO lock mode
 231          * having access type passed as regular lock mode:
 232          *
 233          * - MDL_MINMODE means that lower layer does not want to specify lock
 234          * mode;
 235          *
 236          * - MDL_NL means that no PDO lock should be taken. This is used in some
 237          * cases. Say, for non-splittable directories no need to use PDO locks
 238          * at all.
 239          */
 240         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 241                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 242
 243         if (mode != MDL_MINMODE) {
 244                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 245         } else {
 246                 /*
 247                  * Lower layer does not want to specify locking mode. We do it
 248                  * our selves. No special protection is needed, just flush
 249                  * client's cache on modification and allow concurrent
 250                  * mondification.
 251                  */
 252                 switch (lh->mlh_reg_mode) {
 253                 case LCK_EX:
 254                         lh->mlh_pdo_mode = LCK_EX;
 255                         break;
 256                 case LCK_PR:
 257                         lh->mlh_pdo_mode = LCK_CR;
 258                         break;
 259                 case LCK_PW:
 260                         lh->mlh_pdo_mode = LCK_CW;
 261                         break;
 262                 default:
 263                         CERROR("Not expected lock type (0x%x)\n",
 264                                (int)lh->mlh_reg_mode);
 265                         LBUG();
 266                 }
 267         }
 268
 269         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 270         EXIT;
 271 }
 272
 273 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 274                               struct lu_fid *fid)
 275 {
 276         struct mdt_device *mdt = info->mti_mdt;
 277         struct lu_name *lname = &info->mti_name;
 278         const char *start = fileset;
 279         char *filename = info->mti_filename;
 280         struct mdt_object *parent;
 281         u32 mode;
 282         int rc = 0;
 283
 284         LASSERT(!info->mti_cross_ref);
 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 && start != NULL && *start != '\0') {
 294                 const char *s1 = start;
 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                 start = 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(filename, s1, lname->ln_namelen);
 322                 filename[lname->ln_namelen] = '\0';
 323                 lname->ln_name = filename;
 324
 325                 parent = mdt_object_find(info->mti_env, mdt, fid);
 326                 if (IS_ERR(parent)) {
 327                         rc = PTR_ERR(parent);
 328                         break;
 329                 }
 330                 /* Only got the fid of this obj by name */
 331                 fid_zero(fid);
 332                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 333                                 fid, &info->mti_spec);
 334                 mdt_object_put(info->mti_env, parent);
 335         }
 336         if (!rc) {
 337                 parent = mdt_object_find(info->mti_env, mdt, fid);
 338                 if (IS_ERR(parent))
 339                         rc = PTR_ERR(parent);
 340                 else {
 341                         mode = lu_object_attr(&parent->mot_obj);
 342                         if (!S_ISDIR(mode)) {
 343                                 rc = -ENOTDIR;
 344                         } else if (mdt_is_remote_object(info, parent, parent)) {
 345                                 if (!mdt->mdt_enable_remote_subdir_mount) {
 346                                         rc = -EREMOTE;
 347                                         LCONSOLE_WARN("%s: subdir mount '%s' refused because 'enable_remote_subdir_mount=0': rc = %d\n",
 348                                                       mdt_obd_name(mdt),
 349                                                       fileset, rc);
 350                                 } else {
 351                                         LCONSOLE_INFO("%s: subdir mount '%s' is remote and may be slow\n",
 352                                                       mdt_obd_name(mdt),
 353                                                       fileset);
 354                                 }
 355                         }
 356                         mdt_object_put(info->mti_env, parent);
 357                 }
 358         }
 359
 360         return rc;
 361 }
 362
 363 static int mdt_get_root(struct tgt_session_info *tsi)
 364 {
 365         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 366         struct mdt_device       *mdt = info->mti_mdt;
 367         struct mdt_body         *repbody;
 368         char                    *fileset = NULL, *buffer = NULL;
 369         int                      rc;
 370         struct obd_export       *exp = info->mti_exp;
 371         char                    *nodemap_fileset;
 372
 373         ENTRY;
 374
 375         rc = mdt_check_ucred(info);
 376         if (rc)
 377                 GOTO(out, rc = err_serious(rc));
 378
 379         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 380                 GOTO(out, rc = err_serious(-ENOMEM));
 381
 382         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 383         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 384                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 385                 if (fileset == NULL)
 386                         GOTO(out, rc = err_serious(-EFAULT));
 387         }
 388
 389         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 390         if (nodemap_fileset && nodemap_fileset[0]) {
 391                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 392                 if (fileset) {
 393                         /* consider fileset from client as a sub-fileset
 394                          * of the nodemap one */
 395                         OBD_ALLOC(buffer, PATH_MAX + 1);
 396                         if (buffer == NULL)
 397                                 GOTO(out, rc = err_serious(-ENOMEM));
 398                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 399                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 400                                 GOTO(out, rc = err_serious(-EINVAL));
 401                         fileset = buffer;
 402                 } else {
 403                         /* enforce fileset as specified in the nodemap */
 404                         fileset = nodemap_fileset;
 405                 }
 406         }
 407
 408         if (fileset) {
 409                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 410                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 411                 if (rc < 0)
 412                         GOTO(out, rc = err_serious(rc));
 413         } else {
 414                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 415         }
 416         repbody->mbo_valid |= OBD_MD_FLID;
 417
 418         EXIT;
 419 out:
 420         mdt_thread_info_fini(info);
 421         if (buffer)
 422                 OBD_FREE(buffer, PATH_MAX+1);
 423         return rc;
 424 }
 425
 426 static int mdt_statfs(struct tgt_session_info *tsi)
 427 {
 428         struct ptlrpc_request *req = tgt_ses_req(tsi);
 429         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 430         struct mdt_device *mdt = info->mti_mdt;
 431         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 432         struct md_device *next = mdt->mdt_child;
 433         struct ptlrpc_service_part *svcpt;
 434         struct obd_statfs *osfs;
 435         struct mdt_body *reqbody = NULL;
 436         struct mdt_statfs_cache *msf;
 437         ktime_t kstart = ktime_get();
 438         int current_blockbits;
 439         int rc;
 440
 441         ENTRY;
 442
 443         svcpt = req->rq_rqbd->rqbd_svcpt;
 444
 445         /* This will trigger a watchdog timeout */
 446         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 447                          (MDT_SERVICE_WATCHDOG_FACTOR *
 448                           at_get(&svcpt->scp_at_estimate)) + 1);
 449
 450         rc = mdt_check_ucred(info);
 451         if (rc)
 452                 GOTO(out, rc = err_serious(rc));
 453
 454         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 455                 GOTO(out, rc = err_serious(-ENOMEM));
 456
 457         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 458         if (!osfs)
 459                 GOTO(out, rc = -EPROTO);
 460
 461         if (mdt_is_sum_statfs_client(req->rq_export) &&
 462                 lustre_packed_msg_size(req->rq_reqmsg) ==
 463                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 464                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 465                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 466                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 467         }
 468
 469         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 470                 msf = &mdt->mdt_sum_osfs;
 471         else
 472                 msf = &mdt->mdt_osfs;
 473
 474         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 475                         /** statfs data is too old, get up-to-date one */
 476                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 477                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 478                                                               next, osfs);
 479                         else
 480                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 481                                                osfs);
 482                         if (rc)
 483                                 GOTO(out, rc);
 484                         spin_lock(&mdt->mdt_lock);
 485                         msf->msf_osfs = *osfs;
 486                         msf->msf_age = ktime_get_seconds();
 487                         spin_unlock(&mdt->mdt_lock);
 488         } else {
 489                         /** use cached statfs data */
 490                         spin_lock(&mdt->mdt_lock);
 491                         *osfs = msf->msf_osfs;
 492                         spin_unlock(&mdt->mdt_lock);
 493         }
 494
 495         /* tgd_blockbit is recordsize bits set during mkfs.
 496          * This once set does not change. However, 'zfs set'
 497          * can be used to change the MDT blocksize. Instead
 498          * of using cached value of 'tgd_blockbit' always
 499          * calculate the blocksize bits which may have
 500          * changed.
 501          */
 502         current_blockbits = fls64(osfs->os_bsize) - 1;
 503
 504         /* at least try to account for cached pages.  its still racy and
 505          * might be under-reporting if clients haven't announced their
 506          * caches with brw recently */
 507         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 508                " pending %llu free %llu avail %llu\n",
 509                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 510                tgd->tgd_tot_pending,
 511                osfs->os_bfree << current_blockbits,
 512                osfs->os_bavail << current_blockbits);
 513
 514         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 515                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 516                                    osfs->os_bsize - 1) >> current_blockbits));
 517
 518         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 519         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 520                "%llu objects: %llu free; state %x\n",
 521                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 522                osfs->os_files, osfs->os_ffree, osfs->os_state);
 523
 524         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 525             current_blockbits > COMPAT_BSIZE_SHIFT) {
 526                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 527                  * should not see a block size > page size, otherwise
 528                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 529                  * block size which is the biggest block size known to work
 530                  * with all client's page size. */
 531                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 532                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 533                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 534                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 535         }
 536         if (rc == 0)
 537                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 538                                  ktime_us_delta(ktime_get(), kstart));
 539 out:
 540         mdt_thread_info_fini(info);
 541         RETURN(rc);
 542 }
 543
 544 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 545 {
 546         struct lov_comp_md_v1 *comp_v1;
 547         struct lov_mds_md *v1;
 548         __u32 off;
 549         __u32 dom_stripesize = 0;
 550         int i;
 551         bool has_ost_stripes = false;
 552
 553         ENTRY;
 554
 555         if (is_dom_only)
 556                 *is_dom_only = 0;
 557
 558         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 559                 RETURN(0);
 560
 561         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 562         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 563         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 564
 565         /* Fast check for DoM entry with no mirroring, should be the first */
 566         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 567             lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
 568                 RETURN(0);
 569
 570         /* check all entries otherwise */
 571         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 572                 struct lov_comp_md_entry_v1 *lcme;
 573
 574                 lcme = &comp_v1->lcm_entries[i];
 575                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 576                         continue;
 577
 578                 off = le32_to_cpu(lcme->lcme_offset);
 579                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 580
 581                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) ==
 582                     LOV_PATTERN_MDT)
 583                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 584                 else
 585                         has_ost_stripes = true;
 586
 587                 if (dom_stripesize && has_ost_stripes)
 588                         RETURN(dom_stripesize);
 589         }
 590         /* DoM-only case exits here */
 591         if (is_dom_only && dom_stripesize)
 592                 *is_dom_only = 1;
 593         RETURN(dom_stripesize);
 594 }
 595
 596 /**
 597  * Pack size attributes into the reply.
 598  */
 599 int mdt_pack_size2body(struct mdt_thread_info *info,
 600                         const struct lu_fid *fid, struct lustre_handle *lh)
 601 {
 602         struct mdt_body *b;
 603         struct md_attr *ma = &info->mti_attr;
 604         __u32 dom_stripe;
 605         bool dom_lock = false;
 606
 607         ENTRY;
 608
 609         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 610
 611         if (!S_ISREG(ma->ma_attr.la_mode) ||
 612             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 613                 RETURN(-ENODATA);
 614
 615         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 616         /* no DoM stripe, no size in reply */
 617         if (!dom_stripe)
 618                 RETURN(-ENOENT);
 619
 620         if (lustre_handle_is_used(lh)) {
 621                 struct ldlm_lock *lock;
 622
 623                 lock = ldlm_handle2lock(lh);
 624                 if (lock != NULL) {
 625                         dom_lock = ldlm_has_dom(lock);
 626                         LDLM_LOCK_PUT(lock);
 627                 }
 628         }
 629
 630         /* no DoM lock, no size in reply */
 631         if (!dom_lock)
 632                 RETURN(0);
 633
 634         /* Either DoM lock exists or LMM has only DoM stripe then
 635          * return size on body. */
 636         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 637
 638         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 639         RETURN(0);
 640 }
 641
 642 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 643 /*
 644  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 645  *
 646  * \param       info    thread info object
 647  * \param       repbody reply to pack ACLs into
 648  * \param       o       mdt object of file to examine
 649  * \param       nodemap nodemap of client to reply to
 650  * \retval      0       success
 651  * \retval      -errno  error getting or parsing ACL from disk
 652  */
 653 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 654                       struct mdt_object *o, struct lu_nodemap *nodemap)
 655 {
 656         const struct lu_env     *env = info->mti_env;
 657         struct md_object        *next = mdt_object_child(o);
 658         struct lu_buf           *buf = &info->mti_buf;
 659         struct mdt_device       *mdt = info->mti_mdt;
 660         struct req_capsule *pill = info->mti_pill;
 661         int rc;
 662
 663         ENTRY;
 664
 665         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 666         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 667         if (buf->lb_len == 0)
 668                 RETURN(0);
 669
 670         LASSERT(!info->mti_big_acl_used);
 671 again:
 672         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 673         if (rc < 0) {
 674                 if (rc == -ENODATA) {
 675                         repbody->mbo_aclsize = 0;
 676                         repbody->mbo_valid |= OBD_MD_FLACL;
 677                         rc = 0;
 678                 } else if (rc == -EOPNOTSUPP) {
 679                         rc = 0;
 680                 } else if (rc == -ERANGE) {
 681                         if (exp_connect_large_acl(info->mti_exp) &&
 682                             !info->mti_big_acl_used) {
 683                                 if (info->mti_big_acl == NULL) {
 684                                         info->mti_big_aclsize =
 685                                                         min_t(unsigned int,
 686                                                               mdt->mdt_max_ea_size,
 687                                                               XATTR_SIZE_MAX);
 688                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 689                                                         info->mti_big_aclsize);
 690                                         if (info->mti_big_acl == NULL) {
 691                                                 info->mti_big_aclsize = 0;
 692                                                 CERROR("%s: unable to grow "
 693                                                        DFID" ACL buffer\n",
 694                                                        mdt_obd_name(mdt),
 695                                                        PFID(mdt_object_fid(o)));
 696                                                 RETURN(-ENOMEM);
 697                                         }
 698                                 }
 699
 700                                 CDEBUG(D_INODE, "%s: grow the "DFID
 701                                        " ACL buffer to size %d\n",
 702                                        mdt_obd_name(mdt),
 703                                        PFID(mdt_object_fid(o)),
 704                                        info->mti_big_aclsize);
 705
 706                                 buf->lb_buf = info->mti_big_acl;
 707                                 buf->lb_len = info->mti_big_aclsize;
 708                                 info->mti_big_acl_used = 1;
 709                                 goto again;
 710                         }
 711                         /* FS has ACL bigger that our limits */
 712                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
 713                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
 714                                info->mti_big_aclsize);
 715                         rc = -E2BIG;
 716                 } else {
 717                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 718                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 719                 }
 720         } else {
 721                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 722                                      rc, NODEMAP_FS_TO_CLIENT);
 723                 /* if all ACLs mapped out, rc is still >= 0 */
 724                 if (rc < 0) {
 725                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 726                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 727                                PFID(mdt_object_fid(o)), rc);
 728                         repbody->mbo_aclsize = 0;
 729                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 730                 } else {
 731                         repbody->mbo_aclsize = rc;
 732                         repbody->mbo_valid |= OBD_MD_FLACL;
 733                         rc = 0;
 734                 }
 735         }
 736
 737         RETURN(rc);
 738 }
 739 #endif
 740
 741 /* XXX Look into layout in MDT layer. */
 742 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 743 {
 744         struct lov_comp_md_v1   *comp_v1;
 745         struct lov_mds_md       *v1;
 746         int                      i;
 747
 748         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 749                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 750
 751                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 752                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 753                                 comp_v1->lcm_entries[i].lcme_offset);
 754                         /* We don't support partial release for now */
 755                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 756                                 return false;
 757                 }
 758                 return true;
 759         } else {
 760                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 761                         true : false;
 762         }
 763 }
 764
 765 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 766                         const struct lu_attr *attr, const struct lu_fid *fid)
 767 {
 768         struct md_attr *ma = &info->mti_attr;
 769         struct obd_export *exp = info->mti_exp;
 770         struct lu_nodemap *nodemap = NULL;
 771
 772         LASSERT(ma->ma_valid & MA_INODE);
 773
 774         if (attr->la_valid & LA_ATIME) {
 775                 b->mbo_atime = attr->la_atime;
 776                 b->mbo_valid |= OBD_MD_FLATIME;
 777         }
 778         if (attr->la_valid & LA_MTIME) {
 779                 b->mbo_mtime = attr->la_mtime;
 780                 b->mbo_valid |= OBD_MD_FLMTIME;
 781         }
 782         if (attr->la_valid & LA_CTIME) {
 783                 b->mbo_ctime = attr->la_ctime;
 784                 b->mbo_valid |= OBD_MD_FLCTIME;
 785         }
 786         if (attr->la_valid & LA_BTIME) {
 787                 b->mbo_btime = attr->la_btime;
 788                 b->mbo_valid |= OBD_MD_FLBTIME;
 789         }
 790         if (attr->la_valid & LA_FLAGS) {
 791                 b->mbo_flags = attr->la_flags;
 792                 b->mbo_valid |= OBD_MD_FLFLAGS;
 793         }
 794         if (attr->la_valid & LA_NLINK) {
 795                 b->mbo_nlink = attr->la_nlink;
 796                 b->mbo_valid |= OBD_MD_FLNLINK;
 797         }
 798         if (attr->la_valid & (LA_UID|LA_GID)) {
 799                 nodemap = nodemap_get_from_exp(exp);
 800                 if (IS_ERR(nodemap))
 801                         goto out;
 802         }
 803         if (attr->la_valid & LA_UID) {
 804                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 805                                             NODEMAP_FS_TO_CLIENT,
 806                                             attr->la_uid);
 807                 b->mbo_valid |= OBD_MD_FLUID;
 808         }
 809         if (attr->la_valid & LA_GID) {
 810                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 811                                             NODEMAP_FS_TO_CLIENT,
 812                                             attr->la_gid);
 813                 b->mbo_valid |= OBD_MD_FLGID;
 814         }
 815
 816         if (attr->la_valid & LA_PROJID) {
 817                 /* TODO, nodemap for project id */
 818                 b->mbo_projid = attr->la_projid;
 819                 b->mbo_valid |= OBD_MD_FLPROJID;
 820         }
 821
 822         b->mbo_mode = attr->la_mode;
 823         if (attr->la_valid & LA_MODE)
 824                 b->mbo_valid |= OBD_MD_FLMODE;
 825         if (attr->la_valid & LA_TYPE)
 826                 b->mbo_valid |= OBD_MD_FLTYPE;
 827
 828         if (fid != NULL) {
 829                 b->mbo_fid1 = *fid;
 830                 b->mbo_valid |= OBD_MD_FLID;
 831                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 832                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 833         }
 834
 835         if (!(attr->la_valid & LA_TYPE))
 836                 return;
 837
 838         b->mbo_rdev   = attr->la_rdev;
 839         b->mbo_size   = attr->la_size;
 840         b->mbo_blocks = attr->la_blocks;
 841
 842         if (!S_ISREG(attr->la_mode)) {
 843                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 844         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 845                 /* means no objects are allocated on osts. */
 846                 LASSERT(!(ma->ma_valid & MA_LOV));
 847                 /* just ignore blocks occupied by extend attributes on MDS */
 848                 b->mbo_blocks = 0;
 849                 /* if no object is allocated on osts, the size on mds is valid.
 850                  * b=22272 */
 851                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 852         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 853                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 854                         /* A released file stores its size on MDS. */
 855                         /* But return 1 block for released file, unless tools
 856                          * like tar will consider it fully sparse. (LU-3864)
 857                          */
 858                         if (unlikely(b->mbo_size == 0))
 859                                 b->mbo_blocks = 0;
 860                         else
 861                                 b->mbo_blocks = 1;
 862                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 863                 } else if (info->mti_som_valid) { /* som is valid */
 864                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 865                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 866                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 867                         b->mbo_size = ma->ma_som.ms_size;
 868                         b->mbo_blocks = ma->ma_som.ms_blocks;
 869                 }
 870         }
 871
 872         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 873                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 874                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 875                        PFID(fid), (unsigned long long)b->mbo_size);
 876
 877 out:
 878         if (!IS_ERR_OR_NULL(nodemap))
 879                 nodemap_putref(nodemap);
 880 }
 881
 882 static inline int mdt_body_has_lov(const struct lu_attr *la,
 883                                    const struct mdt_body *body)
 884 {
 885         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 886                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 887 }
 888
 889 void mdt_client_compatibility(struct mdt_thread_info *info)
 890 {
 891         struct mdt_body       *body;
 892         struct ptlrpc_request *req = mdt_info_req(info);
 893         struct obd_export     *exp = req->rq_export;
 894         struct md_attr        *ma = &info->mti_attr;
 895         struct lu_attr        *la = &ma->ma_attr;
 896         ENTRY;
 897
 898         if (exp_connect_layout(exp))
 899                 /* the client can deal with 16-bit lmm_stripe_count */
 900                 RETURN_EXIT;
 901
 902         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 903
 904         if (!mdt_body_has_lov(la, body))
 905                 RETURN_EXIT;
 906
 907         /* now we have a reply with a lov for a client not compatible with the
 908          * layout lock so we have to clean the layout generation number */
 909         if (S_ISREG(la->la_mode))
 910                 ma->ma_lmm->lmm_layout_gen = 0;
 911         EXIT;
 912 }
 913
 914 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 915                                    struct mdt_object *o)
 916 {
 917         const struct lu_env *env = info->mti_env;
 918         int rc, rc2;
 919
 920         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 921                           XATTR_NAME_LOV);
 922
 923         if (rc == -ENODATA)
 924                 rc = 0;
 925
 926         if (rc < 0)
 927                 goto out;
 928
 929         /* Is it a directory? Let's check for the LMV as well */
 930         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 931                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 932                                    XATTR_NAME_LMV);
 933
 934                 if (rc2 == -ENODATA)
 935                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 936                                            &LU_BUF_NULL,
 937                                            XATTR_NAME_DEFAULT_LMV);
 938
 939                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 940                         rc = rc2;
 941         }
 942
 943 out:
 944         return rc;
 945 }
 946
 947 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 948                       const char *name)
 949 {
 950         const struct lu_env *env = info->mti_env;
 951         int rc;
 952         ENTRY;
 953
 954         LASSERT(info->mti_big_lmm_used == 0);
 955         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 956         if (rc < 0)
 957                 RETURN(rc);
 958
 959         /* big_lmm may need to be grown */
 960         if (info->mti_big_lmmsize < rc) {
 961                 int size = size_roundup_power2(rc);
 962
 963                 if (info->mti_big_lmmsize > 0) {
 964                         /* free old buffer */
 965                         LASSERT(info->mti_big_lmm);
 966                         OBD_FREE_LARGE(info->mti_big_lmm,
 967                                        info->mti_big_lmmsize);
 968                         info->mti_big_lmm = NULL;
 969                         info->mti_big_lmmsize = 0;
 970                 }
 971
 972                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 973                 if (info->mti_big_lmm == NULL)
 974                         RETURN(-ENOMEM);
 975                 info->mti_big_lmmsize = size;
 976         }
 977         LASSERT(info->mti_big_lmmsize >= rc);
 978
 979         info->mti_buf.lb_buf = info->mti_big_lmm;
 980         info->mti_buf.lb_len = info->mti_big_lmmsize;
 981         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 982
 983         RETURN(rc);
 984 }
 985
 986 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 987                      struct md_attr *ma, const char *name)
 988 {
 989         struct md_object *next = mdt_object_child(o);
 990         struct lu_buf    *buf = &info->mti_buf;
 991         int rc;
 992
 993         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 994                 buf->lb_buf = ma->ma_lmm;
 995                 buf->lb_len = ma->ma_lmm_size;
 996                 LASSERT(!(ma->ma_valid & MA_LOV));
 997         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 998                 buf->lb_buf = ma->ma_lmv;
 999                 buf->lb_len = ma->ma_lmv_size;
1000                 LASSERT(!(ma->ma_valid & MA_LMV));
1001         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1002                 buf->lb_buf = ma->ma_default_lmv;
1003                 buf->lb_len = ma->ma_default_lmv_size;
1004                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1005         } else {
1006                 return -EINVAL;
1007         }
1008
1009         LASSERT(buf->lb_buf);
1010
1011         rc = mo_xattr_get(info->mti_env, next, buf, name);
1012         if (rc > 0) {
1013
1014 got:
1015                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1016                         if (info->mti_big_lmm_used)
1017                                 ma->ma_lmm = info->mti_big_lmm;
1018
1019                         /* NOT return LOV EA with hole to old client. */
1020                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1021                                      LOV_PATTERN_F_HOLE) &&
1022                             !(exp_connect_flags(info->mti_exp) &
1023                               OBD_CONNECT_LFSCK)) {
1024                                 return -EIO;
1025                         } else {
1026                                 ma->ma_lmm_size = rc;
1027                                 ma->ma_valid |= MA_LOV;
1028                         }
1029                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1030                         if (info->mti_big_lmm_used)
1031                                 ma->ma_lmv = info->mti_big_lmm;
1032
1033                         ma->ma_lmv_size = rc;
1034                         ma->ma_valid |= MA_LMV;
1035                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1036                         ma->ma_default_lmv_size = rc;
1037                         ma->ma_valid |= MA_LMV_DEF;
1038                 }
1039
1040                 /* Update mdt_max_mdsize so all clients will be aware that */
1041                 if (info->mti_mdt->mdt_max_mdsize < rc)
1042                         info->mti_mdt->mdt_max_mdsize = rc;
1043
1044                 rc = 0;
1045         } else if (rc == -ENODATA) {
1046                 /* no LOV EA */
1047                 rc = 0;
1048         } else if (rc == -ERANGE) {
1049                 /* Default LMV has fixed size, so it must be able to fit
1050                  * in the original buffer */
1051                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1052                         return rc;
1053                 rc = mdt_big_xattr_get(info, o, name);
1054                 if (rc > 0) {
1055                         info->mti_big_lmm_used = 1;
1056                         goto got;
1057                 }
1058         }
1059
1060         return rc;
1061 }
1062
1063 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1064                    struct md_attr *ma, const char *name)
1065 {
1066         int rc;
1067
1068         if (!info->mti_big_lmm) {
1069                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1070                 if (!info->mti_big_lmm)
1071                         return -ENOMEM;
1072                 info->mti_big_lmmsize = PAGE_SIZE;
1073         }
1074
1075         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1076                 ma->ma_lmm = info->mti_big_lmm;
1077                 ma->ma_lmm_size = info->mti_big_lmmsize;
1078                 ma->ma_valid &= ~MA_LOV;
1079         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1080                 ma->ma_lmv = info->mti_big_lmm;
1081                 ma->ma_lmv_size = info->mti_big_lmmsize;
1082                 ma->ma_valid &= ~MA_LMV;
1083         } else {
1084                 LBUG();
1085         }
1086
1087         LASSERT(!info->mti_big_lmm_used);
1088         rc = __mdt_stripe_get(info, o, ma, name);
1089         /* since big_lmm is always used here, clear 'used' flag to avoid
1090          * assertion in mdt_big_xattr_get().
1091          */
1092         info->mti_big_lmm_used = 0;
1093
1094         return rc;
1095 }
1096
1097 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1098                       struct lu_fid *pfid)
1099 {
1100         struct lu_buf           *buf = &info->mti_buf;
1101         struct link_ea_header   *leh;
1102         struct link_ea_entry    *lee;
1103         int                      rc;
1104         ENTRY;
1105
1106         buf->lb_buf = info->mti_big_lmm;
1107         buf->lb_len = info->mti_big_lmmsize;
1108         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1109                           buf, XATTR_NAME_LINK);
1110         /* ignore errors, MA_PFID won't be set and it is
1111          * up to the caller to treat this as an error */
1112         if (rc == -ERANGE || buf->lb_len == 0) {
1113                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1114                 buf->lb_buf = info->mti_big_lmm;
1115                 buf->lb_len = info->mti_big_lmmsize;
1116         }
1117
1118         if (rc < 0)
1119                 RETURN(rc);
1120         if (rc < sizeof(*leh)) {
1121                 CERROR("short LinkEA on "DFID": rc = %d\n",
1122                        PFID(mdt_object_fid(o)), rc);
1123                 RETURN(-ENODATA);
1124         }
1125
1126         leh = (struct link_ea_header *) buf->lb_buf;
1127         lee = (struct link_ea_entry *)(leh + 1);
1128         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1129                 leh->leh_magic = LINK_EA_MAGIC;
1130                 leh->leh_reccount = __swab32(leh->leh_reccount);
1131                 leh->leh_len = __swab64(leh->leh_len);
1132         }
1133         if (leh->leh_magic != LINK_EA_MAGIC)
1134                 RETURN(-EINVAL);
1135         if (leh->leh_reccount == 0)
1136                 RETURN(-ENODATA);
1137
1138         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1139         fid_be_to_cpu(pfid, pfid);
1140
1141         RETURN(0);
1142 }
1143
1144 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1145                            struct lu_fid *pfid, struct lu_name *lname)
1146 {
1147         struct lu_buf *buf = &info->mti_buf;
1148         struct link_ea_header *leh;
1149         struct link_ea_entry *lee;
1150         int reclen;
1151         int rc;
1152
1153         buf->lb_buf = info->mti_xattr_buf;
1154         buf->lb_len = sizeof(info->mti_xattr_buf);
1155         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1156                           XATTR_NAME_LINK);
1157         if (rc == -ERANGE) {
1158                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1159                 buf->lb_buf = info->mti_big_lmm;
1160                 buf->lb_len = info->mti_big_lmmsize;
1161         }
1162         if (rc < 0)
1163                 return rc;
1164
1165         if (rc < sizeof(*leh)) {
1166                 CERROR("short LinkEA on "DFID": rc = %d\n",
1167                        PFID(mdt_object_fid(o)), rc);
1168                 return -ENODATA;
1169         }
1170
1171         leh = (struct link_ea_header *)buf->lb_buf;
1172         lee = (struct link_ea_entry *)(leh + 1);
1173         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1174                 leh->leh_magic = LINK_EA_MAGIC;
1175                 leh->leh_reccount = __swab32(leh->leh_reccount);
1176                 leh->leh_len = __swab64(leh->leh_len);
1177         }
1178         if (leh->leh_magic != LINK_EA_MAGIC)
1179                 return -EINVAL;
1180
1181         if (leh->leh_reccount == 0)
1182                 return -ENODATA;
1183
1184         linkea_entry_unpack(lee, &reclen, lname, pfid);
1185
1186         return 0;
1187 }
1188
1189 int mdt_attr_get_complex(struct mdt_thread_info *info,
1190                          struct mdt_object *o, struct md_attr *ma)
1191 {
1192         const struct lu_env *env = info->mti_env;
1193         struct md_object    *next = mdt_object_child(o);
1194         struct lu_buf       *buf = &info->mti_buf;
1195         int                  need = ma->ma_need;
1196         int                  rc = 0, rc2;
1197         u32                  mode;
1198         ENTRY;
1199
1200         ma->ma_valid = 0;
1201
1202         if (mdt_object_exists(o) == 0)
1203                 GOTO(out, rc = -ENOENT);
1204         mode = lu_object_attr(&next->mo_lu);
1205
1206         if (need & MA_INODE) {
1207                 ma->ma_need = MA_INODE;
1208                 rc = mo_attr_get(env, next, ma);
1209                 if (rc)
1210                         GOTO(out, rc);
1211
1212                 if (S_ISREG(mode))
1213                         (void) mdt_get_som(info, o, ma);
1214                 ma->ma_valid |= MA_INODE;
1215         }
1216
1217         if (need & MA_PFID) {
1218                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1219                 if (rc == 0)
1220                         ma->ma_valid |= MA_PFID;
1221                 /* ignore this error, parent fid is not mandatory */
1222                 rc = 0;
1223         }
1224
1225         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1226                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1227                 if (rc)
1228                         GOTO(out, rc);
1229         }
1230
1231         if (need & MA_LMV && S_ISDIR(mode)) {
1232                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1233                 if (rc != 0)
1234                         GOTO(out, rc);
1235         }
1236
1237         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1238                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1239                 if (rc != 0)
1240                         GOTO(out, rc);
1241         }
1242
1243         /*
1244          * In the handle of MA_INODE, we may already get the SOM attr.
1245          */
1246         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1247                 rc = mdt_get_som(info, o, ma);
1248                 if (rc != 0)
1249                         GOTO(out, rc);
1250         }
1251
1252         if (need & MA_HSM && S_ISREG(mode)) {
1253                 buf->lb_buf = info->mti_xattr_buf;
1254                 buf->lb_len = sizeof(info->mti_xattr_buf);
1255                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1256                              sizeof(info->mti_xattr_buf));
1257                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1258                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1259                 if (rc2 == 0)
1260                         ma->ma_valid |= MA_HSM;
1261                 else if (rc2 < 0 && rc2 != -ENODATA)
1262                         GOTO(out, rc = rc2);
1263         }
1264
1265 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1266         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1267                 buf->lb_buf = ma->ma_acl;
1268                 buf->lb_len = ma->ma_acl_size;
1269                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1270                 if (rc2 > 0) {
1271                         ma->ma_acl_size = rc2;
1272                         ma->ma_valid |= MA_ACL_DEF;
1273                 } else if (rc2 == -ENODATA) {
1274                         /* no ACLs */
1275                         ma->ma_acl_size = 0;
1276                 } else
1277                         GOTO(out, rc = rc2);
1278         }
1279 #endif
1280 out:
1281         ma->ma_need = need;
1282         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1283                rc, ma->ma_valid, ma->ma_lmm);
1284         RETURN(rc);
1285 }
1286
1287 static int mdt_getattr_internal(struct mdt_thread_info *info,
1288                                 struct mdt_object *o, int ma_need)
1289 {
1290         struct mdt_device *mdt = info->mti_mdt;
1291         struct md_object *next = mdt_object_child(o);
1292         const struct mdt_body *reqbody = info->mti_body;
1293         struct ptlrpc_request *req = mdt_info_req(info);
1294         struct md_attr *ma = &info->mti_attr;
1295         struct lu_attr *la = &ma->ma_attr;
1296         struct req_capsule *pill = info->mti_pill;
1297         const struct lu_env *env = info->mti_env;
1298         struct mdt_body *repbody;
1299         struct lu_buf *buffer = &info->mti_buf;
1300         struct obd_export *exp = info->mti_exp;
1301         ktime_t kstart = ktime_get();
1302         int rc;
1303
1304         ENTRY;
1305
1306         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1307                 RETURN(err_serious(-ENOMEM));
1308
1309         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1310
1311         ma->ma_valid = 0;
1312
1313         if (mdt_object_remote(o)) {
1314                 /* This object is located on remote node.*/
1315                 /* Return -ENOTSUPP for old client */
1316                 if (!mdt_is_dne_client(req->rq_export))
1317                         GOTO(out, rc = -ENOTSUPP);
1318
1319                 repbody->mbo_fid1 = *mdt_object_fid(o);
1320                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1321                 GOTO(out, rc = 0);
1322         }
1323
1324         if (reqbody->mbo_eadatasize > 0) {
1325                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1326                 if (buffer->lb_buf == NULL)
1327                         GOTO(out, rc = -EPROTO);
1328                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1329                                                       RCL_SERVER);
1330         } else {
1331                 buffer->lb_buf = NULL;
1332                 buffer->lb_len = 0;
1333                 ma_need &= ~(MA_LOV | MA_LMV);
1334                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1335                        mdt_obd_name(info->mti_mdt),
1336                        req->rq_export->exp_client_uuid.uuid);
1337         }
1338
1339         /* from 2.12.58 intent_getattr pack default LMV in reply */
1340         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1341             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1342                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1343             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1344                                   RCL_SERVER)) {
1345                 ma->ma_lmv = buffer->lb_buf;
1346                 ma->ma_lmv_size = buffer->lb_len;
1347                 ma->ma_default_lmv = req_capsule_server_get(pill,
1348                                                 &RMF_DEFAULT_MDT_MD);
1349                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1350                                                 &RMF_DEFAULT_MDT_MD,
1351                                                 RCL_SERVER);
1352                 ma->ma_need = MA_INODE;
1353                 if (ma->ma_lmv_size > 0)
1354                         ma->ma_need |= MA_LMV;
1355                 if (ma->ma_default_lmv_size > 0)
1356                         ma->ma_need |= MA_LMV_DEF;
1357         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1358                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1359                 /* If it is dir and client require MEA, then we got MEA */
1360                 /* Assumption: MDT_MD size is enough for lmv size. */
1361                 ma->ma_lmv = buffer->lb_buf;
1362                 ma->ma_lmv_size = buffer->lb_len;
1363                 ma->ma_need = MA_INODE;
1364                 if (ma->ma_lmv_size > 0) {
1365                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1366                                 ma->ma_need |= MA_LMV;
1367                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1368                                 ma->ma_need |= MA_LMV_DEF;
1369                                 ma->ma_default_lmv = buffer->lb_buf;
1370                                 ma->ma_lmv = NULL;
1371                                 ma->ma_default_lmv_size = buffer->lb_len;
1372                                 ma->ma_lmv_size = 0;
1373                         }
1374                 }
1375         } else {
1376                 ma->ma_lmm = buffer->lb_buf;
1377                 ma->ma_lmm_size = buffer->lb_len;
1378                 ma->ma_need = MA_INODE | MA_HSM;
1379                 if (ma->ma_lmm_size > 0) {
1380                         ma->ma_need |= MA_LOV;
1381                         /* Older clients may crash if they getattr overstriped
1382                          * files
1383                          */
1384                         if (!exp_connect_overstriping(exp) &&
1385                             mdt_lmm_is_overstriping(ma->ma_lmm))
1386                                 RETURN(-EOPNOTSUPP);
1387                 }
1388         }
1389
1390         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1391             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1392             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1393                 /* get default stripe info for this dir. */
1394                 ma->ma_need |= MA_LOV_DEF;
1395         }
1396         ma->ma_need |= ma_need;
1397
1398         rc = mdt_attr_get_complex(info, o, ma);
1399         if (unlikely(rc)) {
1400                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1401                              "%s: getattr error for "DFID": rc = %d\n",
1402                              mdt_obd_name(info->mti_mdt),
1403                              PFID(mdt_object_fid(o)), rc);
1404                 RETURN(rc);
1405         }
1406
1407         /* if file is released, check if a restore is running */
1408         if (ma->ma_valid & MA_HSM) {
1409                 repbody->mbo_valid |= OBD_MD_TSTATE;
1410                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1411                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1412                         repbody->mbo_t_state = MS_RESTORE;
1413         }
1414
1415         if (unlikely(!(ma->ma_valid & MA_INODE)))
1416                 RETURN(-EFAULT);
1417
1418         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1419
1420         if (mdt_body_has_lov(la, reqbody)) {
1421                 u32 stripe_count = 1;
1422
1423                 if (ma->ma_valid & MA_LOV) {
1424                         LASSERT(ma->ma_lmm_size);
1425                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1426                         if (S_ISDIR(la->la_mode))
1427                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1428                         else
1429                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1430                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1431                 }
1432                 if (ma->ma_valid & MA_LMV) {
1433                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1434                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1435
1436                         /* Return -ENOTSUPP for old client */
1437                         if (!mdt_is_striped_client(req->rq_export))
1438                                 RETURN(-ENOTSUPP);
1439
1440                         LASSERT(S_ISDIR(la->la_mode));
1441                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1442                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1443                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1444
1445                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1446                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1447                                 mdt_restripe_migrate_add(info, o);
1448                         else if (magic == LMV_MAGIC_V1 &&
1449                                  lmv_is_restriping(lmv))
1450                                 mdt_restripe_update_add(info, o);
1451                 }
1452                 if (ma->ma_valid & MA_LMV_DEF) {
1453                         /* Return -ENOTSUPP for old client */
1454                         if (!mdt_is_striped_client(req->rq_export))
1455                                 RETURN(-ENOTSUPP);
1456                         LASSERT(S_ISDIR(la->la_mode));
1457                         /*
1458                          * when ll_dir_getstripe() gets default LMV, it
1459                          * checks mbo_eadatasize.
1460                          */
1461                         if (!(ma->ma_valid & MA_LMV))
1462                                 repbody->mbo_eadatasize =
1463                                         ma->ma_default_lmv_size;
1464                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1465                                                OBD_MD_DEFAULT_MEA);
1466                 }
1467                 CDEBUG(D_VFSTRACE,
1468                        "dirent count %llu stripe count %u MDT count %d\n",
1469                        ma->ma_attr.la_dirent_count, stripe_count,
1470                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1471                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1472                     ma->ma_attr.la_dirent_count >
1473                         mdt->mdt_restriper.mdr_dir_split_count &&
1474                     !fid_is_root(mdt_object_fid(o)) &&
1475                     mdt->mdt_enable_dir_auto_split &&
1476                     !o->mot_restriping &&
1477                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1)
1478                         mdt_auto_split_add(info, o);
1479         } else if (S_ISLNK(la->la_mode) &&
1480                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1481                 buffer->lb_buf = ma->ma_lmm;
1482                 /* eadatasize from client includes NULL-terminator, so
1483                  * there is no need to read it */
1484                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1485                 rc = mo_readlink(env, next, buffer);
1486                 if (unlikely(rc <= 0)) {
1487                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1488                                mdt_obd_name(info->mti_mdt),
1489                                PFID(mdt_object_fid(o)), rc);
1490                         rc = -EFAULT;
1491                 } else {
1492                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1493
1494                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1495                                 rc -= 2;
1496                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1497                         /* we need to report back size with NULL-terminator
1498                          * because client expects that */
1499                         repbody->mbo_eadatasize = rc + 1;
1500                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1501                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1502                                        "on "DFID ", expected %d\n",
1503                                        mdt_obd_name(info->mti_mdt),
1504                                        rc, PFID(mdt_object_fid(o)),
1505                                        reqbody->mbo_eadatasize - 1);
1506                         /* NULL terminate */
1507                         ((char *)ma->ma_lmm)[rc] = 0;
1508
1509                         /* If the total CDEBUG() size is larger than a page, it
1510                          * will print a warning to the console, avoid this by
1511                          * printing just the last part of the symlink. */
1512                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1513                                print_limit < rc ? "..." : "", print_limit,
1514                                (char *)ma->ma_lmm + rc - print_limit, rc);
1515                         rc = 0;
1516                 }
1517         }
1518
1519         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1520                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1521                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1522                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1523                        repbody->mbo_max_mdsize);
1524         }
1525
1526 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1527         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1528                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1529                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1530                 if (IS_ERR(nodemap))
1531                         RETURN(PTR_ERR(nodemap));
1532
1533                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1534                 nodemap_putref(nodemap);
1535         }
1536 #endif
1537
1538 out:
1539         if (rc == 0)
1540                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1541                                  ktime_us_delta(ktime_get(), kstart));
1542
1543         RETURN(rc);
1544 }
1545
1546 static int mdt_getattr(struct tgt_session_info *tsi)
1547 {
1548         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1549         struct mdt_object       *obj = info->mti_object;
1550         struct req_capsule      *pill = info->mti_pill;
1551         struct mdt_body         *reqbody;
1552         struct mdt_body         *repbody;
1553         int rc, rc2;
1554         ENTRY;
1555
1556         if (unlikely(info->mti_object == NULL))
1557                 RETURN(-EPROTO);
1558
1559         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1560         LASSERT(reqbody);
1561         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1562
1563         /* Special case for Data-on-MDT files to get data version */
1564         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1565                 rc = mdt_data_version_get(tsi);
1566                 GOTO(out, rc);
1567         }
1568
1569         /* Unlike intent case where we need to pre-fill out buffers early on
1570          * in intent policy for ldlm reasons, here we can have a much better
1571          * guess at EA size by just reading it from disk.
1572          * Exceptions are readdir and (missing) directory striping */
1573         /* Readlink */
1574         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1575                 /* No easy way to know how long is the symlink, but it cannot
1576                  * be more than PATH_MAX, so we allocate +1 */
1577                 rc = PATH_MAX + 1;
1578         /* A special case for fs ROOT: getattr there might fetch
1579          * default EA for entire fs, not just for this dir!
1580          */
1581         } else if (lu_fid_eq(mdt_object_fid(obj),
1582                              &info->mti_mdt->mdt_md_root_fid) &&
1583                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1584                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1585                                                                  MDS_GETATTR)) {
1586                 /* Should the default strping be bigger, mdt_fix_reply
1587                  * will reallocate */
1588                 rc = DEF_REP_MD_SIZE;
1589         } else {
1590                 /* Read the actual EA size from disk */
1591                 rc = mdt_attr_get_eabuf_size(info, obj);
1592         }
1593
1594         if (rc < 0)
1595                 GOTO(out, rc = err_serious(rc));
1596
1597         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1598
1599         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1600          * by default. If the target object has more ACL entries, then
1601          * enlarge the buffer when necessary. */
1602         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1603                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1604
1605         rc = req_capsule_server_pack(pill);
1606         if (unlikely(rc != 0))
1607                 GOTO(out, rc = err_serious(rc));
1608
1609         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1610         LASSERT(repbody != NULL);
1611         repbody->mbo_eadatasize = 0;
1612         repbody->mbo_aclsize = 0;
1613
1614         rc = mdt_check_ucred(info);
1615         if (unlikely(rc))
1616                 GOTO(out_shrink, rc);
1617
1618         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1619
1620         rc = mdt_getattr_internal(info, obj, 0);
1621         EXIT;
1622 out_shrink:
1623         mdt_client_compatibility(info);
1624         rc2 = mdt_fix_reply(info);
1625         if (rc == 0)
1626                 rc = rc2;
1627 out:
1628         mdt_thread_info_fini(info);
1629         return rc;
1630 }
1631
1632 /**
1633  * Handler of layout intent RPC requiring the layout modification
1634  *
1635  * \param[in]  info     thread environment
1636  * \param[in]  obj      object
1637  * \param[out] lhc      object ldlm lock handle
1638  * \param[in]  layout   layout change descriptor
1639  *
1640  * \retval 0    on success
1641  * \retval < 0  error code
1642  */
1643 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1644                       struct mdt_lock_handle *lhc,
1645                       struct md_layout_change *layout)
1646 {
1647         int rc;
1648
1649         ENTRY;
1650
1651         if (!mdt_object_exists(obj))
1652                 RETURN(-ENOENT);
1653
1654         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1655                 RETURN(-EINVAL);
1656
1657         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1658                            MAY_WRITE);
1659         if (rc)
1660                 RETURN(rc);
1661
1662         rc = mdt_check_resent_lock(info, obj, lhc);
1663         if (rc < 0)
1664                 RETURN(rc);
1665
1666         if (rc > 0) {
1667                 /* not resent */
1668                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1669
1670                 /* take layout lock to prepare layout change */
1671                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1672                         lockpart |= MDS_INODELOCK_UPDATE;
1673
1674                 mdt_lock_handle_init(lhc);
1675                 mdt_lock_reg_init(lhc, LCK_EX);
1676                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1677                 if (rc)
1678                         RETURN(rc);
1679         }
1680
1681         mutex_lock(&obj->mot_som_mutex);
1682         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1683         mutex_unlock(&obj->mot_som_mutex);
1684
1685         if (rc)
1686                 mdt_object_unlock(info, obj, lhc, 1);
1687
1688         RETURN(rc);
1689 }
1690
1691 /**
1692  * Exchange MOF_LOV_CREATED flags between two objects after a
1693  * layout swap. No assumption is made on whether o1 or o2 have
1694  * created objects or not.
1695  *
1696  * \param[in,out] o1    First swap layout object
1697  * \param[in,out] o2    Second swap layout object
1698  */
1699 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1700 {
1701         unsigned int o1_lov_created = o1->mot_lov_created;
1702
1703         mutex_lock(&o1->mot_lov_mutex);
1704         mutex_lock(&o2->mot_lov_mutex);
1705
1706         o1->mot_lov_created = o2->mot_lov_created;
1707         o2->mot_lov_created = o1_lov_created;
1708
1709         mutex_unlock(&o2->mot_lov_mutex);
1710         mutex_unlock(&o1->mot_lov_mutex);
1711 }
1712
1713 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1714 {
1715         struct mdt_thread_info  *info;
1716         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1717         struct obd_export       *exp = req->rq_export;
1718         struct mdt_object       *o1, *o2, *o;
1719         struct mdt_lock_handle  *lh1, *lh2;
1720         struct mdc_swap_layouts *msl;
1721         int                      rc;
1722         ENTRY;
1723
1724         /* client does not support layout lock, so layout swaping
1725          * is disabled.
1726          * FIXME: there is a problem for old clients which don't support
1727          * layout lock yet. If those clients have already opened the file
1728          * they won't be notified at all so that old layout may still be
1729          * used to do IO. This can be fixed after file release is landed by
1730          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1731         if (!exp_connect_layout(exp))
1732                 RETURN(-EOPNOTSUPP);
1733
1734         info = tsi2mdt_info(tsi);
1735         if (unlikely(info->mti_object == NULL))
1736                 RETURN(-EPROTO);
1737
1738         if (info->mti_dlm_req != NULL)
1739                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1740
1741         o1 = info->mti_object;
1742         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1743                                 &info->mti_body->mbo_fid2);
1744         if (IS_ERR(o))
1745                 GOTO(out, rc = PTR_ERR(o));
1746
1747         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1748                 GOTO(put, rc = -ENOENT);
1749
1750         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1751         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1752                 GOTO(put, rc);
1753
1754         if (rc < 0)
1755                 swap(o1, o2);
1756
1757         /* permission check. Make sure the calling process having permission
1758          * to write both files. */
1759         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1760                            MAY_WRITE);
1761         if (rc < 0)
1762                 GOTO(put, rc);
1763
1764         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1765                            MAY_WRITE);
1766         if (rc < 0)
1767                 GOTO(put, rc);
1768
1769         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1770         if (msl == NULL)
1771                 GOTO(put, rc = -EPROTO);
1772
1773         lh1 = &info->mti_lh[MDT_LH_NEW];
1774         mdt_lock_reg_init(lh1, LCK_EX);
1775         lh2 = &info->mti_lh[MDT_LH_OLD];
1776         mdt_lock_reg_init(lh2, LCK_EX);
1777
1778         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1779                              MDS_INODELOCK_XATTR);
1780         if (rc < 0)
1781                 GOTO(put, rc);
1782
1783         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1784                              MDS_INODELOCK_XATTR);
1785         if (rc < 0)
1786                 GOTO(unlock1, rc);
1787
1788         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1789                              mdt_object_child(o2), msl->msl_flags);
1790         if (rc < 0)
1791                 GOTO(unlock2, rc);
1792
1793         mdt_swap_lov_flag(o1, o2);
1794
1795 unlock2:
1796         mdt_object_unlock(info, o2, lh2, rc);
1797 unlock1:
1798         mdt_object_unlock(info, o1, lh1, rc);
1799 put:
1800         mdt_object_put(info->mti_env, o);
1801 out:
1802         mdt_thread_info_fini(info);
1803         RETURN(rc);
1804 }
1805
1806 static int mdt_raw_lookup(struct mdt_thread_info *info,
1807                           struct mdt_object *parent,
1808                           const struct lu_name *lname,
1809                           struct ldlm_reply *ldlm_rep)
1810 {
1811         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1812         int              rc;
1813         ENTRY;
1814
1815         LASSERT(!info->mti_cross_ref);
1816
1817         /* Only got the fid of this obj by name */
1818         fid_zero(child_fid);
1819         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1820                         lname, child_fid, &info->mti_spec);
1821         if (rc == 0) {
1822                 struct mdt_body *repbody;
1823
1824                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1825                 repbody->mbo_fid1 = *child_fid;
1826                 repbody->mbo_valid = OBD_MD_FLID;
1827                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1828         } else if (rc == -ENOENT) {
1829                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1830         }
1831
1832         RETURN(rc);
1833 }
1834
1835 /*
1836  * UPDATE lock should be taken against parent, and be released before exit;
1837  * child_bits lock should be taken against child, and be returned back:
1838  *            (1)normal request should release the child lock;
1839  *            (2)intent request will grant the lock to client.
1840  */
1841 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1842                                  struct mdt_lock_handle *lhc,
1843                                  __u64 child_bits,
1844                                  struct ldlm_reply *ldlm_rep)
1845 {
1846         struct ptlrpc_request *req = mdt_info_req(info);
1847         struct mdt_body *reqbody = NULL;
1848         struct mdt_object *parent = info->mti_object;
1849         struct mdt_object *child = NULL;
1850         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1851         struct lu_name *lname = NULL;
1852         struct mdt_lock_handle *lhp = NULL;
1853         struct ldlm_lock *lock;
1854         struct req_capsule *pill = info->mti_pill;
1855         __u64 try_bits = 0;
1856         bool is_resent;
1857         int ma_need = 0;
1858         int rc;
1859
1860         ENTRY;
1861
1862         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1863         LASSERT(ergo(is_resent,
1864                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1865
1866         if (parent == NULL)
1867                 RETURN(-ENOENT);
1868
1869         if (info->mti_cross_ref) {
1870                 /* Only getattr on the child. Parent is on another node. */
1871                 mdt_set_disposition(info, ldlm_rep,
1872                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1873                 child = parent;
1874                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1875                        "ldlm_rep = %p\n",
1876                        PFID(mdt_object_fid(child)), ldlm_rep);
1877
1878                 rc = mdt_check_resent_lock(info, child, lhc);
1879                 if (rc < 0) {
1880                         RETURN(rc);
1881                 } else if (rc > 0) {
1882                         mdt_lock_handle_init(lhc);
1883                         mdt_lock_reg_init(lhc, LCK_PR);
1884
1885                         /*
1886                          * Object's name entry is on another MDS, it will
1887                          * request PERM lock only because LOOKUP lock is owned
1888                          * by the MDS where name entry resides.
1889                          *
1890                          * TODO: it should try layout lock too. - Jinshan
1891                          */
1892                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1893                                         MDS_INODELOCK_LAYOUT);
1894                         child_bits |= MDS_INODELOCK_PERM;
1895
1896                         rc = mdt_object_lock(info, child, lhc, child_bits);
1897                         if (rc < 0)
1898                                 RETURN(rc);
1899                 }
1900
1901                 /* Finally, we can get attr for child. */
1902                 if (!mdt_object_exists(child)) {
1903                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1904                                         &child->mot_obj,
1905                                         "remote object doesn't exist.");
1906                         mdt_object_unlock(info, child, lhc, 1);
1907                         RETURN(-ENOENT);
1908                 }
1909
1910                 rc = mdt_getattr_internal(info, child, 0);
1911                 if (unlikely(rc != 0)) {
1912                         mdt_object_unlock(info, child, lhc, 1);
1913                         RETURN(rc);
1914                 }
1915
1916                 rc = mdt_pack_secctx_in_reply(info, child);
1917                 if (unlikely(rc)) {
1918                         mdt_object_unlock(info, child, lhc, 1);
1919                         RETURN(rc);
1920                 }
1921
1922                 rc = mdt_pack_encctx_in_reply(info, child);
1923                 if (unlikely(rc))
1924                         mdt_object_unlock(info, child, lhc, 1);
1925                 RETURN(rc);
1926         }
1927
1928         lname = &info->mti_name;
1929         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1930
1931         if (lu_name_is_valid(lname)) {
1932                 if (mdt_object_remote(parent)) {
1933                         CERROR("%s: parent "DFID" is on remote target\n",
1934                                mdt_obd_name(info->mti_mdt),
1935                                PFID(mdt_object_fid(parent)));
1936                         RETURN(-EPROTO);
1937                 }
1938
1939                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1940                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1941                        PNAME(lname), ldlm_rep);
1942         } else {
1943                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1944                 if (unlikely(reqbody == NULL))
1945                         RETURN(err_serious(-EPROTO));
1946
1947                 *child_fid = reqbody->mbo_fid2;
1948                 if (unlikely(!fid_is_sane(child_fid)))
1949                         RETURN(err_serious(-EINVAL));
1950
1951                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1952                         mdt_object_get(info->mti_env, parent);
1953                         child = parent;
1954                 } else {
1955                         child = mdt_object_find(info->mti_env, info->mti_mdt,
1956                                                 child_fid);
1957                         if (IS_ERR(child))
1958                                 RETURN(PTR_ERR(child));
1959                 }
1960
1961                 if (mdt_object_remote(child)) {
1962                         CERROR("%s: child "DFID" is on remote target\n",
1963                                mdt_obd_name(info->mti_mdt),
1964                                PFID(mdt_object_fid(child)));
1965                         GOTO(out_child, rc = -EPROTO);
1966                 }
1967
1968                 /* don't fetch LOOKUP lock if it's remote object */
1969                 rc = mdt_is_remote_object(info, parent, child);
1970                 if (rc < 0)
1971                         GOTO(out_child, rc);
1972                 if (rc)
1973                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1974
1975                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1976                        "ldlm_rep = %p\n",
1977                        PFID(mdt_object_fid(parent)),
1978                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1979         }
1980
1981         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1982
1983         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1984                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1985                                 &parent->mot_obj,
1986                                 "Parent doesn't exist!");
1987                 GOTO(out_child, rc = -ESTALE);
1988         }
1989
1990         if (lu_name_is_valid(lname)) {
1991                 /* Always allow to lookup ".." */
1992                 if (unlikely(lname->ln_namelen == 2 &&
1993                              lname->ln_name[0] == '.' &&
1994                              lname->ln_name[1] == '.'))
1995                         info->mti_spec.sp_permitted = 1;
1996
1997                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1998                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1999
2000                         RETURN(rc);
2001                 }
2002
2003                 /* step 1: lock parent only if parent is a directory */
2004                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2005                         lhp = &info->mti_lh[MDT_LH_PARENT];
2006                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
2007                         rc = mdt_object_lock(info, parent, lhp,
2008                                              MDS_INODELOCK_UPDATE);
2009                         if (unlikely(rc != 0))
2010                                 RETURN(rc);
2011                 }
2012
2013                 /* step 2: lookup child's fid by name */
2014                 fid_zero(child_fid);
2015                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2016                                 child_fid, &info->mti_spec);
2017                 if (rc == -ENOENT)
2018                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2019
2020                 if (rc != 0)
2021                         GOTO(unlock_parent, rc);
2022
2023                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2024                                         child_fid);
2025                 if (unlikely(IS_ERR(child)))
2026                         GOTO(unlock_parent, rc = PTR_ERR(child));
2027         }
2028
2029         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2030
2031         /* step 3: lock child regardless if it is local or remote. */
2032         LASSERT(child);
2033
2034         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2035         if (!mdt_object_exists(child)) {
2036                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2037                                 &child->mot_obj,
2038                                 "Object doesn't exist!");
2039                 GOTO(out_child, rc = -ENOENT);
2040         }
2041
2042         rc = mdt_check_resent_lock(info, child, lhc);
2043         if (rc < 0) {
2044                 GOTO(out_child, rc);
2045         } else if (rc > 0) {
2046                 mdt_lock_handle_init(lhc);
2047                 mdt_lock_reg_init(lhc, LCK_PR);
2048
2049                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2050                     !mdt_object_remote(child)) {
2051                         struct md_attr *ma = &info->mti_attr;
2052
2053                         ma->ma_valid = 0;
2054                         ma->ma_need = MA_INODE;
2055                         rc = mdt_attr_get_complex(info, child, ma);
2056                         if (unlikely(rc != 0))
2057                                 GOTO(out_child, rc);
2058
2059                         /* If the file has not been changed for some time, we
2060                          * return not only a LOOKUP lock, but also an UPDATE
2061                          * lock and this might save us RPC on later STAT. For
2062                          * directories, it also let negative dentry cache start
2063                          * working for this dir. */
2064                         if (ma->ma_valid & MA_INODE &&
2065                             ma->ma_attr.la_valid & LA_CTIME &&
2066                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2067                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2068                                 child_bits |= MDS_INODELOCK_UPDATE;
2069                 }
2070
2071                 /* layout lock must be granted in a best-effort way
2072                  * for IT operations */
2073                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2074                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2075                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2076                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2077                             exp_connect_layout(info->mti_exp)) {
2078                                 /* try to grant layout lock for regular file. */
2079                                 try_bits = MDS_INODELOCK_LAYOUT;
2080                         }
2081                         /* Acquire DOM lock in advance for data-on-mdt file */
2082                         if (child != parent)
2083                                 try_bits |= MDS_INODELOCK_DOM;
2084                 }
2085
2086                 if (try_bits != 0) {
2087                         /* try layout lock, it may fail to be granted due to
2088                          * contention at LOOKUP or UPDATE */
2089                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2090                                                  try_bits, false);
2091                         if (child_bits & MDS_INODELOCK_LAYOUT)
2092                                 ma_need |= MA_LOV;
2093                 } else {
2094                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2095                          * client will enqueue the lock to the remote MDT */
2096                         if (mdt_object_remote(child))
2097                                 child_bits &= ~MDS_INODELOCK_UPDATE;
2098                         rc = mdt_object_lock(info, child, lhc, child_bits);
2099                 }
2100                 if (unlikely(rc != 0))
2101                         GOTO(out_child, rc);
2102         }
2103
2104         /* finally, we can get attr for child. */
2105         rc = mdt_getattr_internal(info, child, ma_need);
2106         if (unlikely(rc != 0)) {
2107                 mdt_object_unlock(info, child, lhc, 1);
2108                 GOTO(out_child, rc);
2109         }
2110
2111         rc = mdt_pack_secctx_in_reply(info, child);
2112         if (unlikely(rc)) {
2113                 mdt_object_unlock(info, child, lhc, 1);
2114                 GOTO(out_child, rc);
2115         }
2116
2117         rc = mdt_pack_encctx_in_reply(info, child);
2118         if (unlikely(rc)) {
2119                 mdt_object_unlock(info, child, lhc, 1);
2120                 GOTO(out_child, rc);
2121         }
2122
2123         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2124         if (lock) {
2125                 /* Debugging code. */
2126                 LDLM_DEBUG(lock, "Returning lock to client");
2127                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2128                                          &lock->l_resource->lr_name),
2129                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2130                          PLDLMRES(lock->l_resource),
2131                          PFID(mdt_object_fid(child)));
2132
2133                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2134                     !mdt_object_remote(child) && child != parent) {
2135                         mdt_object_put(info->mti_env, child);
2136                         rc = mdt_pack_size2body(info, child_fid,
2137                                                 &lhc->mlh_reg_lh);
2138                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2139                                 /* DOM lock was taken in advance but this is
2140                                  * not DoM file. Drop the lock.
2141                                  */
2142                                 lock_res_and_lock(lock);
2143                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2144                                 unlock_res_and_lock(lock);
2145                         }
2146                         LDLM_LOCK_PUT(lock);
2147                         GOTO(unlock_parent, rc = 0);
2148                 }
2149                 LDLM_LOCK_PUT(lock);
2150         }
2151
2152         EXIT;
2153 out_child:
2154         if (child)
2155                 mdt_object_put(info->mti_env, child);
2156 unlock_parent:
2157         if (lhp)
2158                 mdt_object_unlock(info, parent, lhp, 1);
2159         return rc;
2160 }
2161
2162 /* normal handler: should release the child lock */
2163 static int mdt_getattr_name(struct tgt_session_info *tsi)
2164 {
2165         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2166         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2167         struct mdt_body *reqbody;
2168         struct mdt_body *repbody;
2169         int rc, rc2;
2170
2171         ENTRY;
2172
2173         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2174         LASSERT(reqbody != NULL);
2175         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2176         LASSERT(repbody != NULL);
2177
2178         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2179         repbody->mbo_eadatasize = 0;
2180         repbody->mbo_aclsize = 0;
2181
2182         rc = mdt_init_ucred(info, reqbody);
2183         if (unlikely(rc))
2184                 GOTO(out_shrink, rc);
2185
2186         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2187         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2188                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2189                 lhc->mlh_reg_lh.cookie = 0;
2190         }
2191         mdt_exit_ucred(info);
2192         EXIT;
2193 out_shrink:
2194         mdt_client_compatibility(info);
2195         rc2 = mdt_fix_reply(info);
2196         if (rc == 0)
2197                 rc = rc2;
2198         mdt_thread_info_fini(info);
2199         return rc;
2200 }
2201
2202 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2203                             const struct lu_fid *pfid,
2204                             const struct lu_name *name,
2205                             struct mdt_object *obj, s64 ctime)
2206 {
2207         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2208         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2209         struct mdt_device *mdt = info->mti_mdt;
2210         struct md_attr *ma = &info->mti_attr;
2211         struct mdt_lock_handle *parent_lh;
2212         struct mdt_lock_handle *child_lh;
2213         struct mdt_object *pobj;
2214         bool cos_incompat = false;
2215         int rc;
2216         ENTRY;
2217
2218         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2219         if (IS_ERR(pobj))
2220                 GOTO(out, rc = PTR_ERR(pobj));
2221
2222         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2223         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2224         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2225         if (rc != 0)
2226                 GOTO(put_parent, rc);
2227
2228         if (mdt_object_remote(pobj))
2229                 cos_incompat = true;
2230
2231         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2232                         name, child_fid, &info->mti_spec);
2233         if (rc != 0)
2234                 GOTO(unlock_parent, rc);
2235
2236         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2237                 GOTO(unlock_parent, rc = -EREMCHG);
2238
2239         child_lh = &info->mti_lh[MDT_LH_CHILD];
2240         mdt_lock_reg_init(child_lh, LCK_EX);
2241         rc = mdt_reint_striped_lock(info, obj, child_lh,
2242                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2243                                     einfo, cos_incompat);
2244         if (rc != 0)
2245                 GOTO(unlock_parent, rc);
2246
2247         if (atomic_read(&obj->mot_open_count)) {
2248                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2249                        PFID(mdt_object_fid(obj)));
2250                 GOTO(unlock_child, rc = -EBUSY);
2251         }
2252
2253         ma->ma_need = 0;
2254         ma->ma_valid = MA_INODE;
2255         ma->ma_attr.la_valid = LA_CTIME;
2256         ma->ma_attr.la_ctime = ctime;
2257
2258         mutex_lock(&obj->mot_lov_mutex);
2259
2260         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2261                         mdt_object_child(obj), name, ma, 0);
2262
2263         mutex_unlock(&obj->mot_lov_mutex);
2264
2265 unlock_child:
2266         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2267 unlock_parent:
2268         mdt_object_unlock(info, pobj, parent_lh, 1);
2269 put_parent:
2270         mdt_object_put(info->mti_env, pobj);
2271 out:
2272         RETURN(rc);
2273 }
2274
2275 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2276                                         struct mdt_object *obj)
2277 {
2278         struct lu_ucred *uc = lu_ucred(info->mti_env);
2279         struct md_attr *ma = &info->mti_attr;
2280         struct lu_attr *la = &ma->ma_attr;
2281         int rc = 0;
2282         ENTRY;
2283
2284         ma->ma_need = MA_INODE;
2285         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2286         if (rc)
2287                 GOTO(out, rc);
2288
2289         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2290                         rc = -EACCES;
2291
2292         if (md_capable(uc, CAP_DAC_OVERRIDE))
2293                 RETURN(0);
2294         if (uc->uc_fsuid == la->la_uid) {
2295                 if ((la->la_mode & S_IWUSR) == 0)
2296                         rc = -EACCES;
2297         } else if (uc->uc_fsgid == la->la_gid) {
2298                 if ((la->la_mode & S_IWGRP) == 0)
2299                         rc = -EACCES;
2300         } else if ((la->la_mode & S_IWOTH) == 0) {
2301                         rc = -EACCES;
2302         }
2303
2304 out:
2305         RETURN(rc);
2306 }
2307
2308 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2309                          s64 ctime)
2310 {
2311         struct mdt_device *mdt = info->mti_mdt;
2312         struct mdt_object *obj = NULL;
2313         struct linkea_data ldata = { NULL };
2314         struct lu_buf *buf = &info->mti_big_buf;
2315         struct lu_name *name = &info->mti_name;
2316         struct lu_fid *pfid = &info->mti_tmp_fid1;
2317         struct link_ea_header *leh;
2318         struct link_ea_entry *lee;
2319         int reclen, count, rc = 0;
2320         ENTRY;
2321
2322         if (!fid_is_sane(fid))
2323                 GOTO(out, rc = -EINVAL);
2324
2325         if (!fid_is_namespace_visible(fid))
2326                 GOTO(out, rc = -EINVAL);
2327
2328         obj = mdt_object_find(info->mti_env, mdt, fid);
2329         if (IS_ERR(obj))
2330                 GOTO(out, rc = PTR_ERR(obj));
2331
2332         if (mdt_object_remote(obj))
2333                 GOTO(out, rc = -EREMOTE);
2334         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2335                 GOTO(out, rc = -ENOENT);
2336
2337         rc = mdt_rmfid_check_permission(info, obj);
2338         if (rc)
2339                 GOTO(out, rc);
2340
2341         /* take LinkEA */
2342         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2343         if (!buf->lb_buf)
2344                 GOTO(out, rc = -ENOMEM);
2345
2346         ldata.ld_buf = buf;
2347         rc = mdt_links_read(info, obj, &ldata);
2348         if (rc)
2349                 GOTO(out, rc);
2350
2351         leh = buf->lb_buf;
2352         lee = (struct link_ea_entry *)(leh + 1);
2353         for (count = 0; count < leh->leh_reccount; count++) {
2354                 /* remove every hardlink */
2355                 linkea_entry_unpack(lee, &reclen, name, pfid);
2356                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2357                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2358                 if (rc)
2359                         break;
2360         }
2361
2362 out:
2363         if (obj && !IS_ERR(obj))
2364                 mdt_object_put(info->mti_env, obj);
2365         if (info->mti_big_buf.lb_buf)
2366                 lu_buf_free(&info->mti_big_buf);
2367
2368         RETURN(rc);
2369 }
2370
2371 static int mdt_rmfid(struct tgt_session_info *tsi)
2372 {
2373         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2374         struct mdt_body *reqbody;
2375         struct lu_fid *fids, *rfids;
2376         int bufsize, rc;
2377         __u32 *rcs;
2378         int i, nr;
2379         ENTRY;
2380
2381         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2382         if (reqbody == NULL)
2383                 RETURN(-EPROTO);
2384         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2385                                        RCL_CLIENT);
2386         nr = bufsize / sizeof(struct lu_fid);
2387         if (nr * sizeof(struct lu_fid) != bufsize)
2388                 RETURN(-EINVAL);
2389         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2390                              RCL_SERVER, nr * sizeof(__u32));
2391         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2392                              RCL_SERVER, nr * sizeof(struct lu_fid));
2393         rc = req_capsule_server_pack(tsi->tsi_pill);
2394         if (rc)
2395                 GOTO(out, rc = err_serious(rc));
2396         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2397         if (fids == NULL)
2398                 RETURN(-EPROTO);
2399         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2400         LASSERT(rcs);
2401         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2402         LASSERT(rfids);
2403
2404         mdt_init_ucred(mti, reqbody);
2405         for (i = 0; i < nr; i++) {
2406                 rfids[i] = fids[i];
2407                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2408         }
2409         mdt_exit_ucred(mti);
2410
2411 out:
2412         RETURN(rc);
2413 }
2414
2415 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2416                          void *karg, void __user *uarg);
2417
2418 static int mdt_set_info(struct tgt_session_info *tsi)
2419 {
2420         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2421         char                    *key;
2422         void                    *val;
2423         int                      keylen, vallen, rc = 0;
2424
2425         ENTRY;
2426
2427         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2428         if (key == NULL) {
2429                 DEBUG_REQ(D_HA, req, "no set_info key");
2430                 RETURN(err_serious(-EFAULT));
2431         }
2432
2433         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2434                                       RCL_CLIENT);
2435
2436         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2437         if (val == NULL) {
2438                 DEBUG_REQ(D_HA, req, "no set_info val");
2439                 RETURN(err_serious(-EFAULT));
2440         }
2441
2442         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2443                                       RCL_CLIENT);
2444
2445         /* Swab any part of val you need to here */
2446         if (KEY_IS(KEY_READ_ONLY)) {
2447                 spin_lock(&req->rq_export->exp_lock);
2448                 if (*(__u32 *)val)
2449                         *exp_connect_flags_ptr(req->rq_export) |=
2450                                 OBD_CONNECT_RDONLY;
2451                 else
2452                         *exp_connect_flags_ptr(req->rq_export) &=
2453                                 ~OBD_CONNECT_RDONLY;
2454                 spin_unlock(&req->rq_export->exp_lock);
2455         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2456                 struct changelog_setinfo *cs = val;
2457
2458                 if (vallen != sizeof(*cs)) {
2459                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2460                                tgt_name(tsi->tsi_tgt), vallen);
2461                         RETURN(-EINVAL);
2462                 }
2463                 if (ptlrpc_req_need_swab(req)) {
2464                         __swab64s(&cs->cs_recno);
2465                         __swab32s(&cs->cs_id);
2466                 }
2467
2468                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2469                         RETURN(-EACCES);
2470                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2471                                    vallen, val, NULL);
2472         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2473                 if (vallen > 0)
2474                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2475         } else {
2476                 RETURN(-EINVAL);
2477         }
2478         RETURN(rc);
2479 }
2480
2481 static int mdt_readpage(struct tgt_session_info *tsi)
2482 {
2483         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2484         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2485         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2486         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2487         struct mdt_body         *repbody;
2488         int                      rc;
2489         int                      i;
2490
2491         ENTRY;
2492
2493         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2494                 RETURN(err_serious(-ENOMEM));
2495
2496         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2497         if (repbody == NULL || reqbody == NULL)
2498                 RETURN(err_serious(-EFAULT));
2499
2500         /*
2501          * prepare @rdpg before calling lower layers and transfer itself. Here
2502          * reqbody->size contains offset of where to start to read and
2503          * reqbody->nlink contains number bytes to read.
2504          */
2505         rdpg->rp_hash = reqbody->mbo_size;
2506         if (rdpg->rp_hash != reqbody->mbo_size) {
2507                 CERROR("Invalid hash: %#llx != %#llx\n",
2508                        rdpg->rp_hash, reqbody->mbo_size);
2509                 RETURN(-EFAULT);
2510         }
2511
2512         rdpg->rp_attrs = reqbody->mbo_mode;
2513         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2514                 rdpg->rp_attrs |= LUDA_64BITHASH;
2515         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2516                                 exp_max_brw_size(tsi->tsi_exp));
2517         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2518                           PAGE_SHIFT;
2519         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2520         if (rdpg->rp_pages == NULL)
2521                 RETURN(-ENOMEM);
2522
2523         for (i = 0; i < rdpg->rp_npages; ++i) {
2524                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2525                 if (rdpg->rp_pages[i] == NULL)
2526                         GOTO(free_rdpg, rc = -ENOMEM);
2527         }
2528
2529         /* call lower layers to fill allocated pages with directory data */
2530         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2531         if (rc < 0)
2532                 GOTO(free_rdpg, rc);
2533
2534         /* send pages to client */
2535         rc = tgt_sendpage(tsi, rdpg, rc);
2536
2537         EXIT;
2538 free_rdpg:
2539
2540         for (i = 0; i < rdpg->rp_npages; i++)
2541                 if (rdpg->rp_pages[i] != NULL)
2542                         __free_page(rdpg->rp_pages[i]);
2543         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2544
2545         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2546                 RETURN(0);
2547
2548         return rc;
2549 }
2550
2551 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2552 {
2553         struct lu_ucred *uc = mdt_ucred_check(info);
2554         struct lu_attr *attr = &info->mti_attr.ma_attr;
2555
2556         if (uc == NULL)
2557                 return -EINVAL;
2558
2559         if (op != REINT_SETATTR) {
2560                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2561                         attr->la_uid = uc->uc_fsuid;
2562                 /* for S_ISGID, inherit gid from his parent, such work will be
2563                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2564                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2565                         attr->la_gid = uc->uc_fsgid;
2566         }
2567
2568         return 0;
2569 }
2570
2571 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2572 {
2573         return op == REINT_OPEN &&
2574              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2575 }
2576
2577 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2578 {
2579         struct req_capsule *pill = info->mti_pill;
2580
2581         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2582                                   RCL_SERVER) &&
2583             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2584                                   RCL_CLIENT)) {
2585                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2586                                          RCL_CLIENT) != 0)
2587                         /* pre-set size in server part with max size */
2588                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2589                                              RCL_SERVER,
2590                                              OBD_MAX_DEFAULT_EA_SIZE);
2591                 else
2592                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2593                                              RCL_SERVER, 0);
2594         }
2595 }
2596
2597 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
2598 {
2599         struct req_capsule *pill = info->mti_pill;
2600
2601         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
2602                                   RCL_SERVER))
2603                 /* pre-set size in server part with max size */
2604                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
2605                                      RCL_SERVER,
2606                                      info->mti_mdt->mdt_max_mdsize);
2607 }
2608
2609 static int mdt_reint_internal(struct mdt_thread_info *info,
2610                               struct mdt_lock_handle *lhc,
2611                               __u32 op)
2612 {
2613         struct req_capsule      *pill = info->mti_pill;
2614         struct mdt_body         *repbody;
2615         int                      rc = 0, rc2;
2616
2617         ENTRY;
2618
2619         rc = mdt_reint_unpack(info, op);
2620         if (rc != 0) {
2621                 CERROR("Can't unpack reint, rc %d\n", rc);
2622                 RETURN(err_serious(rc));
2623         }
2624
2625
2626         /* check if the file system is set to readonly. O_RDONLY open
2627          * is still allowed even the file system is set to readonly mode */
2628         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2629                 RETURN(err_serious(-EROFS));
2630
2631         /* for replay (no_create) lmm is not needed, client has it already */
2632         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2633                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2634                                      DEF_REP_MD_SIZE);
2635
2636         /* llog cookies are always 0, the field is kept for compatibility */
2637         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2638                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2639
2640         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2641          * by default. If the target object has more ACL entries, then
2642          * enlarge the buffer when necessary. */
2643         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2644                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2645                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2646
2647         mdt_preset_secctx_size(info);
2648         mdt_preset_encctx_size(info);
2649
2650         rc = req_capsule_server_pack(pill);
2651         if (rc != 0) {
2652                 CERROR("Can't pack response, rc %d\n", rc);
2653                 RETURN(err_serious(rc));
2654         }
2655
2656         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2657                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2658                 LASSERT(repbody);
2659                 repbody->mbo_eadatasize = 0;
2660                 repbody->mbo_aclsize = 0;
2661         }
2662
2663         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2664
2665         /* for replay no cookkie / lmm need, because client have this already */
2666         if (info->mti_spec.no_create)
2667                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2668                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2669
2670         rc = mdt_init_ucred_reint(info);
2671         if (rc)
2672                 GOTO(out_shrink, rc);
2673
2674         rc = mdt_fix_attr_ucred(info, op);
2675         if (rc != 0)
2676                 GOTO(out_ucred, rc = err_serious(rc));
2677
2678         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2679         if (rc < 0) {
2680                 GOTO(out_ucred, rc);
2681         } else if (rc == 1) {
2682                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2683                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2684                 GOTO(out_ucred, rc);
2685         }
2686         rc = mdt_reint_rec(info, lhc);
2687         EXIT;
2688 out_ucred:
2689         mdt_exit_ucred(info);
2690 out_shrink:
2691         mdt_client_compatibility(info);
2692
2693         rc2 = mdt_fix_reply(info);
2694         if (rc == 0)
2695                 rc = rc2;
2696
2697         /*
2698          * Data-on-MDT optimization - read data along with OPEN and return it
2699          * in reply when possible.
2700          */
2701         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2702                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2703                                           &lhc->mlh_reg_lh);
2704
2705         return rc;
2706 }
2707
2708 static long mdt_reint_opcode(struct ptlrpc_request *req,
2709                              const struct req_format **fmt)
2710 {
2711         struct mdt_device       *mdt;
2712         struct mdt_rec_reint    *rec;
2713         long                     opc;
2714
2715         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2716         if (rec != NULL) {
2717                 opc = rec->rr_opcode;
2718                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2719                 if (opc < REINT_MAX && fmt[opc] != NULL)
2720                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2721                 else {
2722                         mdt = mdt_exp2dev(req->rq_export);
2723                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2724                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2725                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2726                         opc = err_serious(-EFAULT);
2727                 }
2728         } else {
2729                 opc = err_serious(-EFAULT);
2730         }
2731         return opc;
2732 }
2733
2734 static int mdt_reint(struct tgt_session_info *tsi)
2735 {
2736         long opc;
2737         int  rc;
2738         static const struct req_format *reint_fmts[REINT_MAX] = {
2739                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2740                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2741                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2742                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2743                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2744                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2745                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2746                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2747                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2748                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2749         };
2750
2751         ENTRY;
2752
2753         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2754         if (opc >= 0) {
2755                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2756                 /*
2757                  * No lock possible here from client to pass it to reint code
2758                  * path.
2759                  */
2760                 rc = mdt_reint_internal(info, NULL, opc);
2761                 mdt_thread_info_fini(info);
2762         } else {
2763                 rc = opc;
2764         }
2765
2766         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2767         RETURN(rc);
2768 }
2769
2770 /* this should sync the whole device */
2771 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2772 {
2773         struct dt_device *dt = mdt->mdt_bottom;
2774         int rc;
2775         ENTRY;
2776
2777         rc = dt->dd_ops->dt_sync(env, dt);
2778         RETURN(rc);
2779 }
2780
2781 /* this should sync this object */
2782 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2783                            struct mdt_object *mo)
2784 {
2785         int rc = 0;
2786
2787         ENTRY;
2788
2789         if (!mdt_object_exists(mo)) {
2790                 CWARN("%s: non existing object "DFID": rc = %d\n",
2791                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2792                       -ESTALE);
2793                 RETURN(-ESTALE);
2794         }
2795
2796         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2797                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2798                 dt_obj_version_t version;
2799
2800                 version = dt_version_get(env, mdt_obj2dt(mo));
2801                 if (version > tgt->lut_obd->obd_last_committed)
2802                         rc = mo_object_sync(env, mdt_object_child(mo));
2803         } else {
2804                 rc = mo_object_sync(env, mdt_object_child(mo));
2805         }
2806
2807         RETURN(rc);
2808 }
2809
2810 static int mdt_sync(struct tgt_session_info *tsi)
2811 {
2812         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2813         struct req_capsule      *pill = tsi->tsi_pill;
2814         struct mdt_body         *body;
2815         ktime_t                  kstart = ktime_get();
2816         int                      rc;
2817
2818         ENTRY;
2819
2820         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2821                 RETURN(err_serious(-ENOMEM));
2822
2823         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2824                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2825         } else {
2826                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2827
2828                 if (unlikely(info->mti_object == NULL))
2829                         RETURN(-EPROTO);
2830
2831                 /* sync an object */
2832                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2833                                      info->mti_object);
2834                 if (rc == 0) {
2835                         const struct lu_fid *fid;
2836                         struct lu_attr *la = &info->mti_attr.ma_attr;
2837
2838                         info->mti_attr.ma_need = MA_INODE;
2839                         info->mti_attr.ma_valid = 0;
2840                         rc = mdt_attr_get_complex(info, info->mti_object,
2841                                                   &info->mti_attr);
2842                         if (rc == 0) {
2843                                 body = req_capsule_server_get(pill,
2844                                                               &RMF_MDT_BODY);
2845                                 fid = mdt_object_fid(info->mti_object);
2846                                 mdt_pack_attr2body(info, body, la, fid);
2847                         }
2848                 }
2849                 mdt_thread_info_fini(info);
2850         }
2851         if (rc == 0)
2852                 mdt_counter_incr(req, LPROC_MDT_SYNC,
2853                                  ktime_us_delta(ktime_get(), kstart));
2854
2855         RETURN(rc);
2856 }
2857
2858 static int mdt_data_sync(struct tgt_session_info *tsi)
2859 {
2860         struct mdt_thread_info *info;
2861         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2862         struct ost_body *body = tsi->tsi_ost_body;
2863         struct ost_body *repbody;
2864         struct mdt_object *mo = NULL;
2865         struct md_attr *ma;
2866         int rc = 0;
2867
2868         ENTRY;
2869
2870         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2871
2872         /* if no fid is specified then do nothing,
2873          * device sync is done via MDS_SYNC */
2874         if (fid_is_zero(&tsi->tsi_fid))
2875                 RETURN(0);
2876
2877         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2878         if (IS_ERR(mo))
2879                 RETURN(PTR_ERR(mo));
2880
2881         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2882         if (rc)
2883                 GOTO(put, rc);
2884
2885         repbody->oa.o_oi = body->oa.o_oi;
2886         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2887
2888         info = tsi2mdt_info(tsi);
2889         ma = &info->mti_attr;
2890         ma->ma_need = MA_INODE;
2891         ma->ma_valid = 0;
2892         rc = mdt_attr_get_complex(info, mo, ma);
2893         if (rc == 0)
2894                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2895         else
2896                 rc = 0;
2897         mdt_thread_info_fini(info);
2898
2899         EXIT;
2900 put:
2901         if (mo != NULL)
2902                 mdt_object_put(tsi->tsi_env, mo);
2903         return rc;
2904 }
2905
2906 /*
2907  * Handle quota control requests to consult current usage/limit, but also
2908  * to configure quota enforcement
2909  */
2910 static int mdt_quotactl(struct tgt_session_info *tsi)
2911 {
2912         struct obd_export *exp  = tsi->tsi_exp;
2913         struct req_capsule *pill = tsi->tsi_pill;
2914         struct obd_quotactl *oqctl, *repoqc;
2915         int id, rc;
2916         struct mdt_device *mdt = mdt_exp2dev(exp);
2917         struct lu_device *qmt = mdt->mdt_qmt_dev;
2918         struct lu_nodemap *nodemap;
2919         ENTRY;
2920
2921         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2922         if (!oqctl)
2923                 RETURN(err_serious(-EPROTO));
2924
2925         rc = req_capsule_server_pack(pill);
2926         if (rc)
2927                 RETURN(err_serious(rc));
2928
2929         nodemap = nodemap_get_from_exp(exp);
2930         if (IS_ERR(nodemap))
2931                 RETURN(PTR_ERR(nodemap));
2932
2933         switch (oqctl->qc_cmd) {
2934                 /* master quotactl */
2935         case Q_SETINFO:
2936         case Q_SETQUOTA:
2937         case LUSTRE_Q_SETDEFAULT:
2938         case LUSTRE_Q_SETQUOTAPOOL:
2939         case LUSTRE_Q_SETINFOPOOL:
2940                 if (!nodemap_can_setquota(nodemap))
2941                         GOTO(out_nodemap, rc = -EPERM);
2942                 /* fallthrough */
2943         case Q_GETINFO:
2944         case Q_GETQUOTA:
2945         case LUSTRE_Q_GETDEFAULT:
2946         case LUSTRE_Q_GETQUOTAPOOL:
2947         case LUSTRE_Q_GETINFOPOOL:
2948                 if (qmt == NULL)
2949                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2950                 /* slave quotactl */
2951                 /* fallthrough */
2952         case Q_GETOINFO:
2953         case Q_GETOQUOTA:
2954                 break;
2955         default:
2956                 rc = -EFAULT;
2957                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
2958                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
2959                 GOTO(out_nodemap, rc);
2960         }
2961
2962         id = oqctl->qc_id;
2963         switch (oqctl->qc_type) {
2964         case USRQUOTA:
2965                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2966                                     NODEMAP_CLIENT_TO_FS, id);
2967                 break;
2968         case GRPQUOTA:
2969                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2970                                     NODEMAP_CLIENT_TO_FS, id);
2971                 break;
2972         case PRJQUOTA:
2973                 /* todo: check/map project id */
2974                 id = oqctl->qc_id;
2975                 break;
2976         default:
2977                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2978         }
2979         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2980         if (repoqc == NULL)
2981                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2982
2983         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2984                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2985
2986         if (oqctl->qc_id != id)
2987                 swap(oqctl->qc_id, id);
2988
2989         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2990                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2991                         RETURN(-EINPROGRESS);
2992         }
2993
2994         switch (oqctl->qc_cmd) {
2995
2996         case Q_GETINFO:
2997         case Q_SETINFO:
2998         case Q_SETQUOTA:
2999         case Q_GETQUOTA:
3000         case LUSTRE_Q_SETDEFAULT:
3001         case LUSTRE_Q_GETDEFAULT:
3002         case LUSTRE_Q_SETQUOTAPOOL:
3003         case LUSTRE_Q_GETQUOTAPOOL:
3004         case LUSTRE_Q_SETINFOPOOL:
3005         case LUSTRE_Q_GETINFOPOOL:
3006                 /* forward quotactl request to QMT */
3007                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3008                 break;
3009
3010         case Q_GETOINFO:
3011         case Q_GETOQUOTA:
3012                 /* slave quotactl */
3013                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3014                                    oqctl);
3015                 break;
3016
3017         default:
3018                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3019                 GOTO(out_nodemap, rc = -EFAULT);
3020         }
3021
3022         if (oqctl->qc_id != id)
3023                 swap(oqctl->qc_id, id);
3024
3025         QCTL_COPY(repoqc, oqctl);
3026         EXIT;
3027
3028 out_nodemap:
3029         nodemap_putref(nodemap);
3030
3031         return rc;
3032 }
3033
3034 /** clone llog ctxt from child (mdd)
3035  * This allows remote llog (replicator) access.
3036  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3037  * context was originally set up, or we can handle them directly.
3038  * I choose the latter, but that means I need any llog
3039  * contexts set up by child to be accessable by the mdt.  So we clone the
3040  * context into our context list here.
3041  */
3042 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3043                                int idx)
3044 {
3045         struct md_device  *next = mdt->mdt_child;
3046         struct llog_ctxt *ctxt;
3047         int rc;
3048
3049         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3050                 return 0;
3051
3052         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3053         if (rc || ctxt == NULL) {
3054                 return 0;
3055         }
3056
3057         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3058         if (rc)
3059                 CERROR("Can't set mdt ctxt %d\n", rc);
3060
3061         return rc;
3062 }
3063
3064 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3065                                  struct mdt_device *mdt, int idx)
3066 {
3067         struct llog_ctxt *ctxt;
3068
3069         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3070         if (ctxt == NULL)
3071                 return 0;
3072         /* Put once for the get we just did, and once for the clone */
3073         llog_ctxt_put(ctxt);
3074         llog_ctxt_put(ctxt);
3075         return 0;
3076 }
3077
3078 /*
3079  * sec context handlers
3080  */
3081 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3082 {
3083         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3084
3085         return 0;
3086 }
3087
3088 /*
3089  * quota request handlers
3090  */
3091 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3092 {
3093         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3094         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3095         int                      rc;
3096         ENTRY;
3097
3098         if (qmt == NULL)
3099                 RETURN(err_serious(-EOPNOTSUPP));
3100
3101         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3102         RETURN(rc);
3103 }
3104
3105 struct mdt_object *mdt_object_new(const struct lu_env *env,
3106                                   struct mdt_device *d,
3107                                   const struct lu_fid *f)
3108 {
3109         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3110         struct lu_object *o;
3111         struct mdt_object *m;
3112         ENTRY;
3113
3114         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3115         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3116         if (unlikely(IS_ERR(o)))
3117                 m = (struct mdt_object *)o;
3118         else
3119                 m = mdt_obj(o);
3120         RETURN(m);
3121 }
3122
3123 struct mdt_object *mdt_object_find(const struct lu_env *env,
3124                                    struct mdt_device *d,
3125                                    const struct lu_fid *f)
3126 {
3127         struct lu_object *o;
3128         struct mdt_object *m;
3129         ENTRY;
3130
3131         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3132         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3133         if (unlikely(IS_ERR(o)))
3134                 m = (struct mdt_object *)o;
3135         else
3136                 m = mdt_obj(o);
3137
3138         RETURN(m);
3139 }
3140
3141 /**
3142  * Asyncronous commit for mdt device.
3143  *
3144  * Pass asynchonous commit call down the MDS stack.
3145  *
3146  * \param env environment
3147  * \param mdt the mdt device
3148  */
3149 static void mdt_device_commit_async(const struct lu_env *env,
3150                                     struct mdt_device *mdt)
3151 {
3152         struct dt_device *dt = mdt->mdt_bottom;
3153         int rc;
3154         ENTRY;
3155
3156         rc = dt->dd_ops->dt_commit_async(env, dt);
3157         if (unlikely(rc != 0))
3158                 CWARN("%s: async commit start failed: rc = %d\n",
3159                       mdt_obd_name(mdt), rc);
3160         atomic_inc(&mdt->mdt_async_commit_count);
3161         EXIT;
3162 }
3163
3164 /**
3165  * Mark the lock as "synchonous".
3166  *
3167  * Mark the lock to deffer transaction commit to the unlock time.
3168  *
3169  * \param lock the lock to mark as "synchonous"
3170  *
3171  * \see mdt_is_lock_sync
3172  * \see mdt_save_lock
3173  */
3174 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3175 {
3176         lock->l_ast_data = (void*)1;
3177 }
3178
3179 /**
3180  * Check whehter the lock "synchonous" or not.
3181  *
3182  * \param lock the lock to check
3183  * \retval 1 the lock is "synchonous"
3184  * \retval 0 the lock isn't "synchronous"
3185  *
3186  * \see mdt_set_lock_sync
3187  * \see mdt_save_lock
3188  */
3189 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3190 {
3191         return lock->l_ast_data != NULL;
3192 }
3193
3194 /**
3195  * Blocking AST for mdt locks.
3196  *
3197  * Starts transaction commit if in case of COS lock conflict or
3198  * deffers such a commit to the mdt_save_lock.
3199  *
3200  * \param lock the lock which blocks a request or cancelling lock
3201  * \param desc unused
3202  * \param data unused
3203  * \param flag indicates whether this cancelling or blocking callback
3204  * \retval 0
3205  * \see ldlm_blocking_ast_nocheck
3206  */
3207 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3208                      void *data, int flag)
3209 {
3210         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3211         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3212         struct ldlm_cb_set_arg *arg = data;
3213         bool commit_async = false;
3214         int rc;
3215         ENTRY;
3216
3217         if (flag == LDLM_CB_CANCELING)
3218                 RETURN(0);
3219
3220         lock_res_and_lock(lock);
3221         if (lock->l_blocking_ast != mdt_blocking_ast) {
3222                 unlock_res_and_lock(lock);
3223                 RETURN(0);
3224         }
3225
3226         /* A blocking ast may be sent from ldlm_lock_decref_internal
3227          * when the last reference to a local lock was released and
3228          * during blocking event from ldlm_work_bl_ast_lock().
3229          * The 'data' parameter is l_ast_data in the first case and
3230          * callback arguments in the second one. Distinguish them by that.
3231          */
3232         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3233                 goto skip_cos_checks;
3234
3235         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3236                 if (mdt_cos_is_enabled(mdt)) {
3237                         if (!arg->bl_desc->bl_same_client)
3238                                 mdt_set_lock_sync(lock);
3239                 } else if (mdt_slc_is_enabled(mdt) &&
3240                            arg->bl_desc->bl_cos_incompat) {
3241                         mdt_set_lock_sync(lock);
3242                         /*
3243                          * we may do extra commit here, but there is a small
3244                          * window to miss a commit: lock was unlocked (saved),
3245                          * then a conflict lock queued and we come here, but
3246                          * REP-ACK not received, so lock was not converted to
3247                          * COS mode yet.
3248                          * Fortunately this window is quite small, so the
3249                          * extra commit should be rare (not to say distributed
3250                          * operation is rare too).
3251                          */
3252                         commit_async = true;
3253                 }
3254         } else if (lock->l_req_mode == LCK_COS) {
3255                 commit_async = true;
3256         }
3257
3258 skip_cos_checks:
3259         rc = ldlm_blocking_ast_nocheck(lock);
3260
3261         if (commit_async) {
3262                 struct lu_env env;
3263
3264                 rc = lu_env_init(&env, LCT_LOCAL);
3265                 if (unlikely(rc != 0))
3266                         CWARN("%s: lu_env initialization failed, cannot "
3267                               "start asynchronous commit: rc = %d\n",
3268                               obd->obd_name, rc);
3269                 else
3270                         mdt_device_commit_async(&env, mdt);
3271                 lu_env_fini(&env);
3272         }
3273         RETURN(rc);
3274 }
3275
3276 /*
3277  * Blocking AST for cross-MDT lock
3278  *
3279  * Discard lock from uncommitted_slc_locks and cancel it.
3280  *
3281  * \param lock  the lock which blocks a request or cancelling lock
3282  * \param desc  unused
3283  * \param data  unused
3284  * \param flag  indicates whether this cancelling or blocking callback
3285  * \retval      0 on success
3286  * \retval      negative number on error
3287  */
3288 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3289                             void *data, int flag)
3290 {
3291         int rc = 0;
3292         ENTRY;
3293
3294         switch (flag) {
3295         case LDLM_CB_BLOCKING: {
3296                 struct lustre_handle lockh;
3297
3298                 ldlm_lock2handle(lock, &lockh);
3299                 rc = ldlm_cli_cancel(&lockh,
3300                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3301                 if (rc < 0) {
3302                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3303                         RETURN(rc);
3304                 }
3305                 break;
3306         }
3307         case LDLM_CB_CANCELING: {
3308                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3309                 struct mdt_device *mdt =
3310                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3311
3312                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3313
3314                 /* discard slc lock here so that it can be cleaned anytime,
3315                  * especially for cleanup_resource() */
3316                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3317
3318                 /* once we cache lock, l_ast_data is set to mdt_object */
3319                 if (lock->l_ast_data != NULL) {
3320                         struct mdt_object *mo = lock->l_ast_data;
3321                         struct lu_env env;
3322
3323                         rc = lu_env_init(&env, LCT_MD_THREAD);
3324                         if (unlikely(rc != 0)) {
3325                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3326                                       obd->obd_name, mo,
3327                                       PFID(mdt_object_fid(mo)), rc);
3328                                 RETURN(rc);
3329                         }
3330
3331                         if (lock->l_policy_data.l_inodebits.bits &
3332                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3333                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3334                                 mo->mot_cache_attr = 0;
3335                         }
3336                         mdt_object_put(&env, mo);
3337                         lu_env_fini(&env);
3338                 }
3339                 break;
3340         }
3341         default:
3342                 LBUG();
3343         }
3344
3345         RETURN(rc);
3346 }
3347
3348 int mdt_check_resent_lock(struct mdt_thread_info *info,
3349                           struct mdt_object *mo,
3350                           struct mdt_lock_handle *lhc)
3351 {
3352         /* the lock might already be gotten in ldlm_handle_enqueue() */
3353         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3354                 struct ptlrpc_request *req = mdt_info_req(info);
3355                 struct ldlm_lock      *lock;
3356
3357                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3358                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3359                 if (lock == NULL) {
3360                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3361                          * a resend case, however, it could be already destroyed
3362                          * due to client eviction or a raced cancel RPC. */
3363                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3364                                           lhc->mlh_reg_lh.cookie);
3365                         RETURN(-ESTALE);
3366                 }
3367
3368                 if (!fid_res_name_eq(mdt_object_fid(mo),
3369                                      &lock->l_resource->lr_name)) {
3370                         CWARN("%s: Although resent, but still not "
3371                               "get child lock:"DFID"\n",
3372                               info->mti_exp->exp_obd->obd_name,
3373                               PFID(mdt_object_fid(mo)));
3374                         LDLM_LOCK_PUT(lock);
3375                         RETURN(-EPROTO);
3376                 }
3377                 LDLM_LOCK_PUT(lock);
3378                 return 0;
3379         }
3380         return 1;
3381 }
3382
3383 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3384 {
3385         mdt_object_get(NULL, lock->l_ast_data);
3386 }
3387
3388 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3389                                struct mdt_object *o, const struct lu_fid *fid,
3390                                struct lustre_handle *lh, enum ldlm_mode mode,
3391                                __u64 *ibits, __u64 trybits, bool cache)
3392 {
3393         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3394         union ldlm_policy_data *policy = &mti->mti_policy;
3395         struct ldlm_res_id *res_id = &mti->mti_res_id;
3396         int rc = 0;
3397         ENTRY;
3398
3399         LASSERT(mdt_object_remote(o));
3400
3401         fid_build_reg_res_name(fid, res_id);
3402
3403         memset(einfo, 0, sizeof(*einfo));
3404         einfo->ei_type = LDLM_IBITS;
3405         einfo->ei_mode = mode;
3406         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3407         einfo->ei_cb_cp = ldlm_completion_ast;
3408         einfo->ei_enq_slave = 0;
3409         einfo->ei_res_id = res_id;
3410
3411         if (cache) {
3412                 /*
3413                  * if we cache lock, couple lock with mdt_object, so that object
3414                  * can be easily found in lock ASTs.
3415                  */
3416                 einfo->ei_cbdata = o;
3417                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3418         }
3419
3420         memset(policy, 0, sizeof(*policy));
3421         policy->l_inodebits.bits = *ibits;
3422         policy->l_inodebits.try_bits = trybits;
3423
3424         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3425                             policy);
3426
3427         /* Return successfully acquired bits to a caller */
3428         if (rc == 0) {
3429                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3430
3431                 LASSERT(lock);
3432                 *ibits = lock->l_policy_data.l_inodebits.bits;
3433                 LDLM_LOCK_PUT(lock);
3434         }
3435         RETURN(rc);
3436 }
3437
3438 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3439                            const struct lu_fid *fid, struct lustre_handle *lh,
3440                            enum ldlm_mode mode, __u64 ibits, bool cache)
3441 {
3442         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3443                                           cache);
3444 }
3445
3446 int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
3447                           struct mdt_lock_handle *lh, __u64 *ibits,
3448                           __u64 trybits, bool cos_incompat)
3449 {
3450         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3451         union ldlm_policy_data *policy = &info->mti_policy;
3452         struct ldlm_res_id *res_id = &info->mti_res_id;
3453         __u64 dlmflags = 0, *cookie = NULL;
3454         int rc;
3455         ENTRY;
3456
3457         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3458         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3459         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3460         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3461
3462         if (cos_incompat) {
3463                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3464                         lh->mlh_reg_mode == LCK_EX);
3465                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3466         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3467                 dlmflags |= LDLM_FL_COS_ENABLED;
3468         }
3469
3470         /* Only enqueue LOOKUP lock for remote object */
3471         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3472
3473         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3474         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3475             *ibits == MDS_INODELOCK_OPEN)
3476                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3477
3478         if (lh->mlh_type == MDT_PDO_LOCK) {
3479                 /* check for exists after object is locked */
3480                 if (mdt_object_exists(o) == 0) {
3481                         /* Non-existent object shouldn't have PDO lock */
3482                         RETURN(-ESTALE);
3483                 } else {
3484                         /* Non-dir object shouldn't have PDO lock */
3485                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3486                                 RETURN(-ENOTDIR);
3487                 }
3488         }
3489
3490         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3491         dlmflags |= LDLM_FL_ATOMIC_CB;
3492
3493         if (info->mti_exp)
3494                 cookie = &info->mti_exp->exp_handle.h_cookie;
3495
3496         /*
3497          * Take PDO lock on whole directory and build correct @res_id for lock
3498          * on part of directory.
3499          */
3500         if (lh->mlh_pdo_hash != 0) {
3501                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3502                 mdt_lock_pdo_mode(info, o, lh);
3503                 if (lh->mlh_pdo_mode != LCK_NL) {
3504                         /*
3505                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3506                          * is never going to be sent to client and we do not
3507                          * want it slowed down due to possible cancels.
3508                          */
3509                         policy->l_inodebits.bits =
3510                                 *ibits & MDS_INODELOCK_UPDATE;
3511                         policy->l_inodebits.try_bits =
3512                                 trybits & MDS_INODELOCK_UPDATE;
3513                         /* at least one of them should be set */
3514                         LASSERT(policy->l_inodebits.bits |
3515                                 policy->l_inodebits.try_bits);
3516                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3517                                           lh->mlh_pdo_mode, policy, res_id,
3518                                           dlmflags, cookie);
3519                         if (unlikely(rc != 0))
3520                                 GOTO(out_unlock, rc);
3521                 }
3522
3523                 /*
3524                  * Finish res_id initializing by name hash marking part of
3525                  * directory which is taking modification.
3526                  */
3527                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3528         }
3529
3530         policy->l_inodebits.bits = *ibits;
3531         policy->l_inodebits.try_bits = trybits;
3532         policy->l_inodebits.li_gid = lh->mlh_gid;
3533
3534         /*
3535          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3536          * going to be sent to client. If it is - mdt_intent_policy() path will
3537          * fix it up and turn FL_LOCAL flag off.
3538          */
3539         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3540                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3541                           cookie);
3542 out_unlock:
3543         if (rc != 0)
3544                 mdt_object_unlock(info, o, lh, 1);
3545         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3546                    lh->mlh_pdo_hash != 0 &&
3547                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3548                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3549
3550         /* Return successfully acquired bits to a caller */
3551         if (rc == 0) {
3552                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3553
3554                 LASSERT(lock);
3555                 *ibits = lock->l_policy_data.l_inodebits.bits;
3556                 LDLM_LOCK_PUT(lock);
3557         }
3558         RETURN(rc);
3559 }
3560
3561 static int
3562 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3563                          struct mdt_lock_handle *lh, __u64 *ibits,
3564                          __u64 trybits, bool cos_incompat)
3565 {
3566         struct mdt_lock_handle *local_lh = NULL;
3567         int rc;
3568         ENTRY;
3569
3570         if (!mdt_object_remote(o)) {
3571                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3572                                            cos_incompat);
3573                 RETURN(rc);
3574         }
3575
3576         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3577         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3578                     MDS_INODELOCK_XATTR);
3579
3580         /* Only enqueue LOOKUP lock for remote object */
3581         if (*ibits & MDS_INODELOCK_LOOKUP) {
3582                 __u64 local = MDS_INODELOCK_LOOKUP;
3583
3584                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3585                                            cos_incompat);
3586                 if (rc != ELDLM_OK)
3587                         RETURN(rc);
3588
3589                 local_lh = lh;
3590         }
3591
3592         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3593                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3594                  * enqueue RPC can only request 1 lock, to avoid extra
3595                  * RPC, so it will instead enqueue EX lock for remote
3596                  * object anyway XXX*/
3597                 if (lh->mlh_type == MDT_PDO_LOCK &&
3598                     lh->mlh_pdo_hash != 0) {
3599                         CDEBUG(D_INFO,
3600                                "%s: "DFID" convert PDO lock to EX lock.\n",
3601                                mdt_obd_name(info->mti_mdt),
3602                                PFID(mdt_object_fid(o)));
3603                         lh->mlh_pdo_hash = 0;
3604                         lh->mlh_rreg_mode = LCK_EX;
3605                         lh->mlh_type = MDT_REG_LOCK;
3606                 }
3607
3608                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3609                                                 &lh->mlh_rreg_lh,
3610                                                 lh->mlh_rreg_mode,
3611                                                 ibits, trybits, false);
3612                 if (rc != ELDLM_OK) {
3613                         if (local_lh != NULL)
3614                                 mdt_object_unlock(info, o, local_lh, rc);
3615                         RETURN(rc);
3616                 }
3617         }
3618
3619         /* other components like LFSCK can use lockless access
3620          * and populate cache, so we better invalidate it */
3621         mo_invalidate(info->mti_env, mdt_object_child(o));
3622
3623         RETURN(0);
3624 }
3625
3626 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3627                     struct mdt_lock_handle *lh, __u64 ibits)
3628 {
3629         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3630 }
3631
3632 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3633                           struct mdt_lock_handle *lh, __u64 ibits,
3634                           bool cos_incompat)
3635 {
3636         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3637         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3638                                         cos_incompat);
3639 }
3640
3641 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3642                         struct mdt_lock_handle *lh, __u64 *ibits,
3643                         __u64 trybits, bool cos_incompat)
3644 {
3645         bool trylock_only = *ibits == 0;
3646         int rc;
3647
3648         LASSERT(!(*ibits & trybits));
3649         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3650                                       cos_incompat);
3651         if (rc && trylock_only) { /* clear error for try ibits lock only */
3652                 LASSERT(*ibits == 0);
3653                 rc = 0;
3654         }
3655         return rc;
3656 }
3657
3658 /**
3659  * Save a lock within request object.
3660  *
3661  * Keep the lock referenced until whether client ACK or transaction
3662  * commit happens or release the lock immediately depending on input
3663  * parameters. If COS is ON, a write lock is converted to COS lock
3664  * before saving.
3665  *
3666  * \param info thead info object
3667  * \param h lock handle
3668  * \param mode lock mode
3669  * \param decref force immediate lock releasing
3670  */
3671 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3672                    enum ldlm_mode mode, int decref)
3673 {
3674         ENTRY;
3675
3676         if (lustre_handle_is_used(h)) {
3677                 if (decref || !info->mti_has_trans ||
3678                     !(mode & (LCK_PW | LCK_EX))) {
3679                         mdt_fid_unlock(h, mode);
3680                 } else {
3681                         struct mdt_device *mdt = info->mti_mdt;
3682                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3683                         struct ptlrpc_request *req = mdt_info_req(info);
3684                         bool cos = mdt_cos_is_enabled(mdt);
3685                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3686
3687                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3688                                  h->cookie);
3689
3690                         /* there is no request if mdt_object_unlock() is called
3691                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3692                         if (likely(req != NULL)) {
3693                                 LDLM_DEBUG(lock, "save lock request %p reply "
3694                                         "state %p transno %lld\n", req,
3695                                         req->rq_reply_state, req->rq_transno);
3696                                 if (cos) {
3697                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3698                                         mode = LCK_COS;
3699                                 }
3700                                 if (req->rq_export->exp_disconnected)
3701                                         mdt_fid_unlock(h, mode);
3702                                 else
3703                                         ptlrpc_save_lock(req, h, mode, cos,
3704                                                          convert_lock);
3705                         } else {
3706                                 mdt_fid_unlock(h, mode);
3707                         }
3708                         if (mdt_is_lock_sync(lock)) {
3709                                 CDEBUG(D_HA, "found sync-lock,"
3710                                        " async commit started\n");
3711                                 mdt_device_commit_async(info->mti_env,
3712                                                         mdt);
3713                         }
3714                         LDLM_LOCK_PUT(lock);
3715                 }
3716                 h->cookie = 0ull;
3717         }
3718
3719         EXIT;
3720 }
3721
3722 /**
3723  * Save cross-MDT lock in uncommitted_slc_locks
3724  *
3725  * Keep the lock referenced until transaction commit happens or release the lock
3726  * immediately depending on input parameters.
3727  *
3728  * \param info thead info object
3729  * \param h lock handle
3730  * \param mode lock mode
3731  * \param decref force immediate lock releasing
3732  */
3733 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3734                                  struct mdt_object *o, struct lustre_handle *h,
3735                                  enum ldlm_mode mode, int decref)
3736 {
3737         ENTRY;
3738
3739         if (lustre_handle_is_used(h)) {
3740                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3741                 struct ptlrpc_request *req = mdt_info_req(info);
3742
3743                 if (o != NULL &&
3744                     (lock->l_policy_data.l_inodebits.bits &
3745                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3746                         mo_invalidate(info->mti_env, mdt_object_child(o));
3747
3748                 if (decref || !info->mti_has_trans || !req ||
3749                     !(mode & (LCK_PW | LCK_EX))) {
3750                         ldlm_lock_decref_and_cancel(h, mode);
3751                         LDLM_LOCK_PUT(lock);
3752                 } else {
3753                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3754                                           req->rq_transno);
3755                         ldlm_lock_decref(h, mode);
3756                 }
3757                 h->cookie = 0ull;
3758         }
3759
3760         EXIT;
3761 }
3762
3763 /**
3764  * Unlock mdt object.
3765  *
3766  * Immeditely release the regular lock and the PDO lock or save the
3767  * lock in request and keep them referenced until client ACK or
3768  * transaction commit.
3769  *
3770  * \param info thread info object
3771  * \param o mdt object
3772  * \param lh mdt lock handle referencing regular and PDO locks
3773  * \param decref force immediate lock releasing
3774  *
3775  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3776  */
3777 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3778                        struct mdt_lock_handle *lh, int decref)
3779 {
3780         ENTRY;
3781
3782         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3783         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3784         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3785                              decref);
3786
3787         EXIT;
3788 }
3789
3790 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3791                                         const struct lu_fid *f,
3792                                         struct mdt_lock_handle *lh,
3793                                         __u64 ibits)
3794 {
3795         struct mdt_object *o;
3796
3797         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3798         if (!IS_ERR(o)) {
3799                 int rc;
3800
3801                 rc = mdt_object_lock(info, o, lh, ibits);
3802                 if (rc != 0) {
3803                         mdt_object_put(info->mti_env, o);
3804                         o = ERR_PTR(rc);
3805                 }
3806         }
3807         return o;
3808 }
3809
3810 void mdt_object_unlock_put(struct mdt_thread_info * info,
3811                            struct mdt_object * o,
3812                            struct mdt_lock_handle *lh,
3813                            int decref)
3814 {
3815         mdt_object_unlock(info, o, lh, decref);
3816         mdt_object_put(info->mti_env, o);
3817 }
3818
3819 /*
3820  * Generic code handling requests that have struct mdt_body passed in:
3821  *
3822  *  - extract mdt_body from request and save it in @info, if present;
3823  *
3824  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3825  *  @info;
3826  *
3827  *  - if HAS_BODY flag is set for this request type check whether object
3828  *  actually exists on storage (lu_object_exists()).
3829  *
3830  */
3831 static int mdt_body_unpack(struct mdt_thread_info *info,
3832                            enum tgt_handler_flags flags)
3833 {
3834         const struct mdt_body    *body;
3835         struct mdt_object        *obj;
3836         const struct lu_env      *env;
3837         struct req_capsule       *pill;
3838         int                       rc;
3839         ENTRY;
3840
3841         env = info->mti_env;
3842         pill = info->mti_pill;
3843
3844         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3845         if (body == NULL)
3846                 RETURN(-EFAULT);
3847
3848         if (!(body->mbo_valid & OBD_MD_FLID))
3849                 RETURN(0);
3850
3851         if (!fid_is_sane(&body->mbo_fid1)) {
3852                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3853                 RETURN(-EINVAL);
3854         }
3855
3856         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3857         if (!IS_ERR(obj)) {
3858                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3859                         mdt_object_put(env, obj);
3860                         rc = -ENOENT;
3861                 } else {
3862                         info->mti_object = obj;
3863                         rc = 0;
3864                 }
3865         } else
3866                 rc = PTR_ERR(obj);
3867
3868         RETURN(rc);
3869 }
3870
3871 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3872                                    enum tgt_handler_flags flags)
3873 {
3874         struct req_capsule *pill = info->mti_pill;
3875         int rc;
3876
3877         ENTRY;
3878
3879         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3880                 rc = mdt_body_unpack(info, flags);
3881         else
3882                 rc = 0;
3883
3884         if (rc == 0 && (flags & HAS_REPLY)) {
3885                 /* Pack reply. */
3886                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3887                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3888                                              DEF_REP_MD_SIZE);
3889                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3890                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3891                                              RCL_SERVER, 0);
3892
3893                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3894                  * by default. If the target object has more ACL entries, then
3895                  * enlarge the buffer when necessary. */
3896                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3897                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3898                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3899
3900                 mdt_preset_secctx_size(info);
3901                 mdt_preset_encctx_size(info);
3902
3903                 rc = req_capsule_server_pack(pill);
3904                 if (rc)
3905                         CWARN("%s: cannot pack response: rc = %d\n",
3906                                       mdt_obd_name(info->mti_mdt), rc);
3907         }
3908         RETURN(rc);
3909 }
3910
3911 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3912 {
3913         lh->mlh_type = MDT_NUL_LOCK;
3914         lh->mlh_reg_lh.cookie = 0ull;
3915         lh->mlh_reg_mode = LCK_MINMODE;
3916         lh->mlh_pdo_lh.cookie = 0ull;
3917         lh->mlh_pdo_mode = LCK_MINMODE;
3918         lh->mlh_rreg_lh.cookie = 0ull;
3919         lh->mlh_rreg_mode = LCK_MINMODE;
3920 }
3921
3922 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3923 {
3924         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3925         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3926 }
3927
3928 /*
3929  * Initialize fields of struct mdt_thread_info. Other fields are left in
3930  * uninitialized state, because it's too expensive to zero out whole
3931  * mdt_thread_info (> 1K) on each request arrival.
3932  */
3933 void mdt_thread_info_init(struct ptlrpc_request *req,
3934                           struct mdt_thread_info *info)
3935 {
3936         int i;
3937
3938         info->mti_pill = &req->rq_pill;
3939
3940         /* lock handle */
3941         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3942                 mdt_lock_handle_init(&info->mti_lh[i]);
3943
3944         /* mdt device: it can be NULL while CONNECT */
3945         if (req->rq_export) {
3946                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3947                 info->mti_exp = req->rq_export;
3948         } else
3949                 info->mti_mdt = NULL;
3950         info->mti_env = req->rq_svc_thread->t_env;
3951         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3952
3953         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3954         info->mti_big_buf = LU_BUF_NULL;
3955         info->mti_body = NULL;
3956         info->mti_object = NULL;
3957         info->mti_dlm_req = NULL;
3958         info->mti_has_trans = 0;
3959         info->mti_cross_ref = 0;
3960         info->mti_opdata = 0;
3961         info->mti_big_lmm_used = 0;
3962         info->mti_big_acl_used = 0;
3963         info->mti_som_valid = 0;
3964
3965         info->mti_spec.no_create = 0;
3966         info->mti_spec.sp_rm_entry = 0;
3967         info->mti_spec.sp_permitted = 0;
3968
3969         info->mti_spec.u.sp_ea.eadata = NULL;
3970         info->mti_spec.u.sp_ea.eadatalen = 0;
3971 }
3972
3973 void mdt_thread_info_fini(struct mdt_thread_info *info)
3974 {
3975         int i;
3976
3977         if (info->mti_object != NULL) {
3978                 mdt_object_put(info->mti_env, info->mti_object);
3979                 info->mti_object = NULL;
3980         }
3981
3982         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3983                 mdt_lock_handle_fini(&info->mti_lh[i]);
3984         info->mti_env = NULL;
3985         info->mti_pill = NULL;
3986         info->mti_exp = NULL;
3987         info->mti_mdt = NULL;
3988
3989         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3990                 lu_buf_free(&info->mti_big_buf);
3991 }
3992
3993 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3994 {
3995         struct mdt_thread_info  *mti;
3996
3997         mti = mdt_th_info(tsi->tsi_env);
3998         LASSERT(mti != NULL);
3999
4000         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4001         if (tsi->tsi_corpus != NULL) {
4002                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4003                 lu_object_get(tsi->tsi_corpus);
4004         }
4005         mti->mti_body = tsi->tsi_mdt_body;
4006         mti->mti_dlm_req = tsi->tsi_dlm_req;
4007
4008         return mti;
4009 }
4010
4011 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4012 {
4013         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4014             cfs_fail_val ==
4015             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4016                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4017
4018         return tgt_connect(tsi);
4019 }
4020
4021 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4022                               struct mdt_thread_info *info,
4023                               struct ldlm_lock **lockp, __u64 flags)
4024 {
4025         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4026                                    lockp, flags);
4027 }
4028 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4029                           struct mdt_thread_info *info,
4030                           struct ldlm_lock **lockp, __u64 flags)
4031 {
4032         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4033                                lockp, flags);
4034 }
4035
4036 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4037                             struct ldlm_lock **lockp,
4038                             struct mdt_lock_handle *lh,
4039                             __u64 flags, int result)
4040 {
4041         struct ptlrpc_request  *req = mdt_info_req(info);
4042         struct ldlm_lock       *lock = *lockp;
4043         struct ldlm_lock       *new_lock;
4044
4045         /* If possible resent found a lock, @lh is set to its handle */
4046         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4047
4048         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
4049                 lh->mlh_reg_lh.cookie = 0;
4050                 RETURN(0);
4051         }
4052
4053         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
4054                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
4055                  * a resend case, however, it could be already destroyed
4056                  * due to client eviction or a raced cancel RPC. */
4057                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4058                                   lh->mlh_reg_lh.cookie);
4059                 lh->mlh_reg_lh.cookie = 0;
4060                 RETURN(-ESTALE);
4061         }
4062
4063         LASSERTF(new_lock != NULL,
4064                  "lockh %#llx flags %#llx : rc = %d\n",
4065                  lh->mlh_reg_lh.cookie, flags, result);
4066
4067         /*
4068          * If we've already given this lock to a client once, then we should
4069          * have no readers or writers.  Otherwise, we should have one reader
4070          * _or_ writer ref (which will be zeroed below) before returning the
4071          * lock to a client.
4072          */
4073         if (new_lock->l_export == req->rq_export) {
4074                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4075         } else {
4076                 LASSERT(new_lock->l_export == NULL);
4077                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4078         }
4079
4080         *lockp = new_lock;
4081
4082         if (new_lock->l_export == req->rq_export) {
4083                 /*
4084                  * Already gave this to the client, which means that we
4085                  * reconstructed a reply.
4086                  */
4087                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4088                         MSG_RESENT);
4089
4090                 LDLM_LOCK_RELEASE(new_lock);
4091                 lh->mlh_reg_lh.cookie = 0;
4092                 RETURN(ELDLM_LOCK_REPLACED);
4093         }
4094
4095         /*
4096          * Fixup the lock to be given to the client.
4097          */
4098         lock_res_and_lock(new_lock);
4099         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4100          * possible blocking AST. */
4101         while (new_lock->l_readers > 0) {
4102                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4103                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4104                 new_lock->l_readers--;
4105         }
4106         while (new_lock->l_writers > 0) {
4107                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4108                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4109                 new_lock->l_writers--;
4110         }
4111
4112         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4113         new_lock->l_blocking_ast = lock->l_blocking_ast;
4114         new_lock->l_completion_ast = lock->l_completion_ast;
4115         if (ldlm_has_dom(new_lock))
4116                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4117         new_lock->l_remote_handle = lock->l_remote_handle;
4118         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4119
4120         unlock_res_and_lock(new_lock);
4121
4122         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4123                      &new_lock->l_remote_handle,
4124                      &new_lock->l_exp_hash);
4125
4126         LDLM_LOCK_RELEASE(new_lock);
4127         lh->mlh_reg_lh.cookie = 0;
4128
4129         RETURN(ELDLM_LOCK_REPLACED);
4130 }
4131
4132 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4133                              struct ldlm_lock *new_lock,
4134                              struct mdt_lock_handle *lh, __u64 flags)
4135 {
4136         struct ptlrpc_request  *req = mdt_info_req(info);
4137         struct ldlm_request    *dlmreq;
4138
4139         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4140                 return;
4141
4142         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4143
4144         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4145          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4146          * initialized. */
4147         if (flags & LDLM_FL_RESENT) {
4148                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4149                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4150
4151                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4152                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4153                           lh->mlh_reg_lh.cookie);
4154                 return;
4155         }
4156
4157         /*
4158          * If the xid matches, then we know this is a resent request, and allow
4159          * it. (It's probably an OPEN, for which we don't send a lock.
4160          */
4161         if (req_can_reconstruct(req, NULL) != 0)
4162                 return;
4163
4164         /*
4165          * This remote handle isn't enqueued, so we never received or processed
4166          * this request.  Clear MSG_RESENT, because it can be handled like any
4167          * normal request now.
4168          */
4169         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4170
4171         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4172                   dlmreq->lock_handle[0].cookie);
4173 }
4174
4175 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4176                                struct mdt_thread_info *info,
4177                                struct ldlm_lock **lockp,
4178                                __u64 flags)
4179 {
4180         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4181         struct ldlm_reply      *ldlm_rep = NULL;
4182         int rc;
4183         ENTRY;
4184
4185         /*
4186          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4187          * (for the resend case) or a new lock. Below we will use it to
4188          * replace the original lock.
4189          */
4190         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4191         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4192                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
4193                 rc = mdt_object_lock(info, info->mti_object, lhc,
4194                                      MDS_INODELOCK_XATTR);
4195                 if (rc)
4196                         return rc;
4197         }
4198
4199         rc = mdt_getxattr(info);
4200
4201         if (mdt_info_req(info)->rq_repmsg != NULL)
4202                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4203
4204         if (ldlm_rep == NULL ||
4205             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4206                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4207                 if (is_serious(rc))
4208                         RETURN(rc);
4209                 else
4210                         RETURN(err_serious(-EFAULT));
4211         }
4212
4213         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4214
4215         /* This is left for interop instead of adding a new interop flag.
4216          * LU-7433 */
4217 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4218         if (ldlm_rep->lock_policy_res2) {
4219                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4220                 RETURN(ELDLM_LOCK_ABORTED);
4221         }
4222 #endif
4223
4224         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4225         RETURN(rc);
4226 }
4227
4228 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4229                               struct mdt_thread_info *info,
4230                               struct ldlm_lock **lockp,
4231                               __u64 flags)
4232 {
4233         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4234         __u64                   child_bits;
4235         struct ldlm_reply      *ldlm_rep;
4236         struct mdt_body        *reqbody;
4237         struct mdt_body        *repbody;
4238         int                     rc, rc2;
4239         ENTRY;
4240
4241         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4242         LASSERT(reqbody);
4243
4244         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4245         LASSERT(repbody);
4246
4247         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4248         repbody->mbo_eadatasize = 0;
4249         repbody->mbo_aclsize = 0;
4250
4251         switch (it_opc) {
4252         case IT_LOOKUP:
4253                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4254                 break;
4255         case IT_GETATTR:
4256                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4257                              MDS_INODELOCK_PERM;
4258                 break;
4259         default:
4260                 CERROR("%s: unsupported intent %#x\n",
4261                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4262                 GOTO(out_shrink, rc = -EINVAL);
4263         }
4264
4265         rc = mdt_init_ucred(info, reqbody);
4266         if (rc)
4267                 GOTO(out_shrink, rc);
4268
4269         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4270         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4271
4272         /* Get lock from request for possible resent case. */
4273         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4274
4275         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4276         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4277
4278         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4279                 ldlm_rep->lock_policy_res2 = 0;
4280         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4281             ldlm_rep->lock_policy_res2) {
4282                 lhc->mlh_reg_lh.cookie = 0ull;
4283                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4284         }
4285
4286         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4287         EXIT;
4288 out_ucred:
4289         mdt_exit_ucred(info);
4290 out_shrink:
4291         mdt_client_compatibility(info);
4292         rc2 = mdt_fix_reply(info);
4293         if (rc == 0)
4294                 rc = rc2;
4295         return rc;
4296 }
4297
4298 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4299                              struct mdt_thread_info *info,
4300                              struct ldlm_lock **lockp,
4301                              __u64 flags)
4302 {
4303         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4304         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4305         struct layout_intent *intent;
4306         struct ldlm_reply *ldlm_rep;
4307         struct lu_fid *fid = &info->mti_tmp_fid2;
4308         struct mdt_object *obj = NULL;
4309         int layout_size = 0;
4310         struct lu_buf *buf = &layout.mlc_buf;
4311         int rc = 0;
4312
4313         ENTRY;
4314
4315         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4316
4317         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4318         if (intent == NULL)
4319                 RETURN(-EPROTO);
4320
4321         CDEBUG(D_INFO, DFID "got layout change request from client: "
4322                "opc:%u flags:%#x extent "DEXT"\n",
4323                PFID(fid), intent->li_opc, intent->li_flags,
4324                PEXT(&intent->li_extent));
4325
4326         switch (intent->li_opc) {
4327         case LAYOUT_INTENT_TRUNC:
4328         case LAYOUT_INTENT_WRITE:
4329                 layout.mlc_opc = MD_LAYOUT_WRITE;
4330                 layout.mlc_intent = intent;
4331                 break;
4332         case LAYOUT_INTENT_ACCESS:
4333                 break;
4334         case LAYOUT_INTENT_READ:
4335         case LAYOUT_INTENT_GLIMPSE:
4336         case LAYOUT_INTENT_RELEASE:
4337         case LAYOUT_INTENT_RESTORE:
4338                 CERROR("%s: Unsupported layout intent opc %d\n",
4339                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4340                 RETURN(-ENOTSUPP);
4341         default:
4342                 CERROR("%s: Unknown layout intent opc %d\n",
4343                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4344                 RETURN(-EINVAL);
4345         }
4346
4347         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4348         if (IS_ERR(obj))
4349                 RETURN(PTR_ERR(obj));
4350
4351         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4352                 /* if layout is going to be changed don't use the current EA
4353                  * size but the maximum one. That buffer will be shrinked
4354                  * to the actual size in req_capsule_shrink() before reply.
4355                  */
4356                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4357                         layout_size = info->mti_mdt->mdt_max_mdsize;
4358                 } else {
4359                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4360                         if (layout_size < 0)
4361                                 GOTO(out, rc = layout_size);
4362
4363                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4364                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4365                 }
4366                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4367                        mdt_obd_name(info->mti_mdt), layout_size);
4368         }
4369
4370         /*
4371          * set reply buffer size, so that ldlm_handle_enqueue0()->
4372          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4373          */
4374         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4375                              layout_size);
4376         rc = req_capsule_server_pack(info->mti_pill);
4377         if (rc)
4378                 GOTO(out, rc);
4379
4380         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4381         if (!ldlm_rep)
4382                 GOTO(out, rc = -EPROTO);
4383
4384         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4385
4386         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4387         if (layout.mlc_opc == MD_LAYOUT_NOP)
4388                 GOTO(out, rc = 0);
4389
4390         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4391         if (rc < 0)
4392                 GOTO(out, rc);
4393         if (rc == 1) {
4394                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4395                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4396                 GOTO(out, rc);
4397         }
4398
4399         buf->lb_buf = NULL;
4400         buf->lb_len = 0;
4401         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4402                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4403                                                      &RMF_EADATA);
4404                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4405                                                      &RMF_EADATA, RCL_CLIENT);
4406                 /*
4407                  * If it's a replay of layout write intent RPC, the client has
4408                  * saved the extended lovea when it get reply then.
4409                  */
4410                 if (buf->lb_len > 0)
4411                         mdt_fix_lov_magic(info, buf->lb_buf);
4412         }
4413
4414         /* Get lock from request for possible resent case. */
4415         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4416         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4417
4418         /*
4419          * Instantiate some layout components, if @buf contains lovea, then it's
4420          * a replay of the layout intent write RPC.
4421          */
4422         rc = mdt_layout_change(info, obj, lhc, &layout);
4423         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4424
4425         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4426                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4427                 if (rc == ELDLM_LOCK_REPLACED &&
4428                     (*lockp)->l_granted_mode == LCK_EX)
4429                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4430         }
4431
4432         EXIT;
4433 out:
4434         mdt_object_put(info->mti_env, obj);
4435         return rc;
4436 }
4437
4438 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4439                            struct mdt_thread_info *info,
4440                            struct ldlm_lock **lockp,
4441                            __u64 flags)
4442 {
4443         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4444         struct ldlm_reply      *rep = NULL;
4445         long                    opc;
4446         int                     rc;
4447         struct ptlrpc_request  *req = mdt_info_req(info);
4448
4449         static const struct req_format *intent_fmts[REINT_MAX] = {
4450                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4451                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4452         };
4453
4454         ENTRY;
4455
4456         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4457         if (opc < 0)
4458                 RETURN(opc);
4459
4460         /* Get lock from request for possible resent case. */
4461         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4462
4463         rc = mdt_reint_internal(info, lhc, opc);
4464
4465         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4466                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4467
4468         /* Check whether the reply has been packed successfully. */
4469         if (mdt_info_req(info)->rq_repmsg != NULL)
4470                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4471         if (rep == NULL) {
4472                 if (is_serious(rc))
4473                         RETURN(rc);
4474                 else
4475                         RETURN(err_serious(-EFAULT));
4476         }
4477
4478         /* MDC expects this in any case */
4479         if (rc != 0)
4480                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4481
4482         /* the open lock or the lock for cross-ref object should be
4483          * returned to the client */
4484         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4485             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4486                 rep->lock_policy_res2 = 0;
4487                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4488                 RETURN(rc);
4489         }
4490
4491         rep->lock_policy_res2 = clear_serious(rc);
4492
4493         if (rep->lock_policy_res2 == -ENOENT &&
4494             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4495             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4496                 rep->lock_policy_res2 = 0;
4497
4498         lhc->mlh_reg_lh.cookie = 0ull;
4499         if (rc == -ENOTCONN || rc == -ENODEV ||
4500             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4501                 /*
4502                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4503                  * will be returned by rq_status, and client at ptlrpc layer
4504                  * will detect this, then disconnect, reconnect the import
4505                  * immediately, instead of impacting the following the rpc.
4506                  */
4507                 RETURN(rc);
4508         }
4509         /*
4510          * For other cases, the error will be returned by intent, and client
4511          * will retrieve the result from intent.
4512          */
4513         RETURN(ELDLM_LOCK_ABORTED);
4514 }
4515
4516 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4517                           struct mdt_thread_info *info,
4518                           struct ldlm_lock **lockp,
4519                           u64 flags /* LDLM_FL_* */)
4520 {
4521         struct req_capsule *pill = info->mti_pill;
4522         struct ptlrpc_request *req = mdt_info_req(info);
4523         const struct req_format *it_format;
4524         int (*it_handler)(enum ldlm_intent_flags,
4525                           struct mdt_thread_info *,
4526                           struct ldlm_lock **,
4527                           u64);
4528         enum tgt_handler_flags it_handler_flags = 0;
4529         struct ldlm_reply *rep;
4530         bool check_mdt_object = false;
4531         int rc;
4532         ENTRY;
4533
4534         switch (it_opc) {
4535         case IT_OPEN:
4536         case IT_OPEN|IT_CREAT:
4537                 /*
4538                  * OCREAT is not a IS_MUTABLE request since the file may
4539                  * already exist. We do the extra check of
4540                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4541                  * really need to create the object.
4542                  */
4543                 it_format = &RQF_LDLM_INTENT;
4544                 it_handler = &mdt_intent_open;
4545                 break;
4546         case IT_GETATTR:
4547                 check_mdt_object = true;
4548                 /* fallthrough */
4549         case IT_LOOKUP:
4550                 it_format = &RQF_LDLM_INTENT_GETATTR;
4551                 it_handler = &mdt_intent_getattr;
4552                 it_handler_flags = HAS_REPLY;
4553                 break;
4554         case IT_GETXATTR:
4555                 check_mdt_object = true;
4556                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4557                 it_handler = &mdt_intent_getxattr;
4558                 it_handler_flags = HAS_BODY;
4559                 break;
4560         case IT_LAYOUT:
4561                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4562                 it_handler = &mdt_intent_layout;
4563                 break;
4564         case IT_GLIMPSE:
4565                 it_format = &RQF_LDLM_INTENT;
4566                 it_handler = &mdt_intent_glimpse;
4567                 break;
4568         case IT_BRW:
4569                 it_format = &RQF_LDLM_INTENT;
4570                 it_handler = &mdt_intent_brw;
4571                 break;
4572         case IT_QUOTA_DQACQ:
4573         case IT_QUOTA_CONN: {
4574                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4575
4576                 if (qmt == NULL)
4577                         RETURN(-EOPNOTSUPP);
4578
4579                 if (mdt_rdonly(req->rq_export))
4580                         RETURN(-EROFS);
4581
4582                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4583                 /* pass the request to quota master */
4584                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4585                                                  mdt_info_req(info), lockp,
4586                                                  flags);
4587                 RETURN(rc);
4588         }
4589         default:
4590                 CERROR("%s: unknown intent code %#x\n",
4591                        mdt_obd_name(info->mti_mdt), it_opc);
4592                 RETURN(-EPROTO);
4593         }
4594
4595         req_capsule_extend(pill, it_format);
4596
4597         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4598         if (rc < 0)
4599                 RETURN(rc);
4600
4601         if (unlikely(info->mti_object == NULL && check_mdt_object))
4602                 RETURN(-EPROTO);
4603
4604         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4605                 RETURN(-EROFS);
4606
4607         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4608
4609         /* execute policy */
4610         rc = (*it_handler)(it_opc, info, lockp, flags);
4611
4612         /* Check whether the reply has been packed successfully. */
4613         if (req->rq_repmsg != NULL) {
4614                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4615                 rep->lock_policy_res2 =
4616                         ptlrpc_status_hton(rep->lock_policy_res2);
4617         }
4618
4619         RETURN(rc);
4620 }
4621
4622 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4623                                     enum ldlm_intent_flags it_opc)
4624 {
4625         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4626
4627         /* update stats when IT code is known */
4628         if (srv_stats != NULL)
4629                 lprocfs_counter_incr(srv_stats,
4630                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4631                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4632 }
4633
4634 static int mdt_intent_policy(const struct lu_env *env,
4635                              struct ldlm_namespace *ns,
4636                              struct ldlm_lock **lockp,
4637                              void *req_cookie,
4638                              enum ldlm_mode mode,
4639                              __u64 flags, void *data)
4640 {
4641         struct tgt_session_info *tsi;
4642         struct mdt_thread_info  *info;
4643         struct ptlrpc_request   *req  =  req_cookie;
4644         struct ldlm_intent      *it;
4645         struct req_capsule      *pill;
4646         const struct ldlm_lock_desc *ldesc;
4647         int rc;
4648
4649         ENTRY;
4650
4651         LASSERT(req != NULL);
4652
4653         tsi = tgt_ses_info(env);
4654
4655         info = tsi2mdt_info(tsi);
4656         LASSERT(info != NULL);
4657         pill = info->mti_pill;
4658         LASSERT(pill->rc_req == req);
4659         ldesc = &info->mti_dlm_req->lock_desc;
4660
4661         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4662                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4663                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4664                 if (it != NULL) {
4665                         mdt_ptlrpc_stats_update(req, it->opc);
4666                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4667                         if (rc == 0)
4668                                 rc = ELDLM_OK;
4669
4670                         /* Lock without inodebits makes no sense and will oops
4671                          * later in ldlm. Let's check it now to see if we have
4672                          * ibits corrupted somewhere in mdt_intent_opc().
4673                          * The case for client miss to set ibits has been
4674                          * processed by others. */
4675                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4676                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4677                 } else {
4678                         rc = err_serious(-EFAULT);
4679                 }
4680         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
4681                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
4682                 struct ldlm_reply *rep;
4683
4684                 /* No intent was provided but INTENT flag is set along with
4685                  * DOM bit, this is considered as GLIMPSE request.
4686                  * This logic is common for MDT and OST glimpse
4687                  */
4688                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
4689                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
4690                 /* Check whether the reply has been packed successfully. */
4691                 if (req->rq_repmsg != NULL) {
4692                         rep = req_capsule_server_get(info->mti_pill,
4693                                                      &RMF_DLM_REP);
4694                         rep->lock_policy_res2 =
4695                                 ptlrpc_status_hton(rep->lock_policy_res2);
4696                 }
4697         } else {
4698                 /* No intent was provided */
4699                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4700                 rc = req_capsule_server_pack(pill);
4701                 if (rc)
4702                         rc = err_serious(rc);
4703         }
4704         mdt_thread_info_fini(info);
4705         RETURN(rc);
4706 }
4707
4708 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4709 {
4710         struct seq_server_site  *ss = mdt_seq_site(mdt);
4711
4712         if (ss->ss_node_id == 0)
4713                 return;
4714
4715         if (ss->ss_client_seq != NULL) {
4716                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4717                 ss->ss_client_seq->lcs_exp = NULL;
4718         }
4719
4720         if (ss->ss_server_fld != NULL) {
4721                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4722                 ss->ss_server_fld->lsf_control_exp = NULL;
4723         }
4724 }
4725
4726 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4727 {
4728         struct seq_server_site *ss = mdt_seq_site(mdt);
4729
4730         if (ss == NULL)
4731                 return;
4732
4733         if (ss->ss_server_seq != NULL)
4734                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4735 }
4736
4737 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4738 {
4739         mdt_seq_fini_cli(mdt);
4740         mdt_deregister_seq_exp(mdt);
4741
4742         return seq_site_fini(env, mdt_seq_site(mdt));
4743 }
4744
4745 /**
4746  * It will retrieve its FLDB entries from MDT0, and it only happens
4747  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4748  * and it needs to refresh FLDB from the MDT0.
4749  **/
4750 static int mdt_register_lwp_callback(void *data)
4751 {
4752         struct lu_env           env;
4753         struct mdt_device       *mdt = data;
4754         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4755         int                     rc;
4756         ENTRY;
4757
4758         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4759
4760         rc = lu_env_init(&env, LCT_MD_THREAD);
4761         if (rc < 0) {
4762                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4763                 RETURN(rc);
4764         }
4765
4766         /* Allocate new sequence now to avoid creating local transaction
4767          * in the normal transaction process */
4768         rc = seq_server_check_and_alloc_super(&env,
4769                                               mdt_seq_site(mdt)->ss_server_seq);
4770         if (rc < 0)
4771                 GOTO(out, rc);
4772
4773         if (fld->lsf_new) {
4774                 rc = fld_update_from_controller(&env, fld);
4775                 if (rc != 0) {
4776                         CERROR("%s: cannot update controller: rc = %d\n",
4777                                mdt_obd_name(mdt), rc);
4778                         GOTO(out, rc);
4779                 }
4780         }
4781 out:
4782         lu_env_fini(&env);
4783         RETURN(rc);
4784 }
4785
4786 static int mdt_register_seq_exp(struct mdt_device *mdt)
4787 {
4788         struct seq_server_site  *ss = mdt_seq_site(mdt);
4789         char                    *lwp_name = NULL;
4790         int                     rc;
4791
4792         if (ss->ss_node_id == 0)
4793                 return 0;
4794
4795         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4796         if (lwp_name == NULL)
4797                 GOTO(out_free, rc = -ENOMEM);
4798
4799         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4800         if (rc != 0)
4801                 GOTO(out_free, rc);
4802
4803         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4804                                       NULL, NULL);
4805         if (rc != 0)
4806                 GOTO(out_free, rc);
4807
4808         rc = lustre_register_lwp_item(lwp_name,
4809                                       &ss->ss_server_fld->lsf_control_exp,
4810                                       mdt_register_lwp_callback, mdt);
4811         if (rc != 0) {
4812                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4813                 ss->ss_client_seq->lcs_exp = NULL;
4814                 GOTO(out_free, rc);
4815         }
4816 out_free:
4817         if (lwp_name != NULL)
4818                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4819
4820         return rc;
4821 }
4822
4823 /*
4824  * Init client sequence manager which is used by local MDS to talk to sequence
4825  * controller on remote node.
4826  */
4827 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4828 {
4829         struct seq_server_site *ss = mdt_seq_site(mdt);
4830         char *prefix;
4831         ENTRY;
4832
4833         /* check if this is adding the first MDC and controller is not yet
4834          * initialized. */
4835         OBD_ALLOC_PTR(ss->ss_client_seq);
4836         if (ss->ss_client_seq == NULL)
4837                 RETURN(-ENOMEM);
4838
4839         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4840         if (prefix == NULL) {
4841                 OBD_FREE_PTR(ss->ss_client_seq);
4842                 ss->ss_client_seq = NULL;
4843                 RETURN(-ENOMEM);
4844         }
4845
4846         /* Note: seq_client_fini will be called in seq_site_fini */
4847         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4848         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4849                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4850                                                             NULL);
4851         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4852
4853         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
4854 }
4855
4856 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4857 {
4858         struct seq_server_site  *ss;
4859         int                     rc;
4860         ENTRY;
4861
4862         ss = mdt_seq_site(mdt);
4863         /* init sequence controller server(MDT0) */
4864         if (ss->ss_node_id == 0) {
4865                 OBD_ALLOC_PTR(ss->ss_control_seq);
4866                 if (ss->ss_control_seq == NULL)
4867                         RETURN(-ENOMEM);
4868
4869                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4870                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4871                                      ss);
4872                 if (rc)
4873                         GOTO(out_seq_fini, rc);
4874         }
4875
4876         /* Init normal sequence server */
4877         OBD_ALLOC_PTR(ss->ss_server_seq);
4878         if (ss->ss_server_seq == NULL)
4879                 GOTO(out_seq_fini, rc = -ENOMEM);
4880
4881         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4882                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4883         if (rc)
4884                 GOTO(out_seq_fini, rc);
4885
4886         /* init seq client for seq server to talk to seq controller(MDT0) */
4887         rc = mdt_seq_init_cli(env, mdt);
4888         if (rc != 0)
4889                 GOTO(out_seq_fini, rc);
4890
4891         if (ss->ss_node_id != 0)
4892                 /* register controller export through lwp */
4893                 rc = mdt_register_seq_exp(mdt);
4894
4895         EXIT;
4896 out_seq_fini:
4897         if (rc)
4898                 mdt_seq_fini(env, mdt);
4899
4900         return rc;
4901 }
4902
4903 /*
4904  * FLD wrappers
4905  */
4906 static int mdt_fld_fini(const struct lu_env *env,
4907                         struct mdt_device *m)
4908 {
4909         struct seq_server_site *ss = mdt_seq_site(m);
4910         ENTRY;
4911
4912         if (ss && ss->ss_server_fld) {
4913                 fld_server_fini(env, ss->ss_server_fld);
4914                 OBD_FREE_PTR(ss->ss_server_fld);
4915                 ss->ss_server_fld = NULL;
4916         }
4917
4918         RETURN(0);
4919 }
4920
4921 static int mdt_fld_init(const struct lu_env *env,
4922                         const char *uuid,
4923                         struct mdt_device *m)
4924 {
4925         struct seq_server_site *ss;
4926         int rc;
4927         ENTRY;
4928
4929         ss = mdt_seq_site(m);
4930
4931         OBD_ALLOC_PTR(ss->ss_server_fld);
4932         if (ss->ss_server_fld == NULL)
4933                 RETURN(rc = -ENOMEM);
4934
4935         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4936                              LU_SEQ_RANGE_MDT);
4937         if (rc) {
4938                 OBD_FREE_PTR(ss->ss_server_fld);
4939                 ss->ss_server_fld = NULL;
4940                 RETURN(rc);
4941         }
4942
4943         RETURN(0);
4944 }
4945
4946 static void mdt_stack_pre_fini(const struct lu_env *env,
4947                            struct mdt_device *m, struct lu_device *top)
4948 {
4949         struct lustre_cfg_bufs  *bufs;
4950         struct lustre_cfg       *lcfg;
4951         struct mdt_thread_info  *info;
4952         ENTRY;
4953
4954         LASSERT(top);
4955
4956         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4957         LASSERT(info != NULL);
4958
4959         bufs = &info->mti_u.bufs;
4960
4961         LASSERT(m->mdt_child_exp);
4962         LASSERT(m->mdt_child_exp->exp_obd);
4963
4964         /* process cleanup, pass mdt obd name to get obd umount flags */
4965         /* XXX: this is needed because all layers are referenced by
4966          * objects (some of them are pinned by osd, for example *
4967          * the proper solution should be a model where object used
4968          * by osd only doesn't have mdt/mdd slices -bzzz */
4969         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4970         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4971         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4972         if (!lcfg)
4973                 RETURN_EXIT;
4974         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4975
4976         top->ld_ops->ldo_process_config(env, top, lcfg);
4977         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4978         EXIT;
4979 }
4980
4981 static void mdt_stack_fini(const struct lu_env *env,
4982                            struct mdt_device *m, struct lu_device *top)
4983 {
4984         struct obd_device       *obd = mdt2obd_dev(m);
4985         struct lustre_cfg_bufs  *bufs;
4986         struct lustre_cfg       *lcfg;
4987         struct mdt_thread_info  *info;
4988         char                     flags[3] = "";
4989         ENTRY;
4990
4991         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4992         LASSERT(info != NULL);
4993
4994         lu_dev_del_linkage(top->ld_site, top);
4995
4996         lu_site_purge(env, top->ld_site, -1);
4997
4998         bufs = &info->mti_u.bufs;
4999         /* process cleanup, pass mdt obd name to get obd umount flags */
5000         /* another purpose is to let all layers to release their objects */
5001         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5002         if (obd->obd_force)
5003                 strcat(flags, "F");
5004         if (obd->obd_fail)
5005                 strcat(flags, "A");
5006         lustre_cfg_bufs_set_string(bufs, 1, flags);
5007         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5008         if (!lcfg)
5009                 RETURN_EXIT;
5010         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5011
5012         LASSERT(top);
5013         top->ld_ops->ldo_process_config(env, top, lcfg);
5014         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5015
5016         lu_site_purge(env, top->ld_site, -1);
5017
5018         m->mdt_child = NULL;
5019         m->mdt_bottom = NULL;
5020
5021         obd_disconnect(m->mdt_child_exp);
5022         m->mdt_child_exp = NULL;
5023
5024         obd_disconnect(m->mdt_bottom_exp);
5025         m->mdt_child_exp = NULL;
5026 }
5027
5028 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5029                                const char *next, struct obd_export **exp)
5030 {
5031         struct obd_connect_data *data = NULL;
5032         struct obd_device       *obd;
5033         int                      rc;
5034         ENTRY;
5035
5036         OBD_ALLOC_PTR(data);
5037         if (data == NULL)
5038                 GOTO(out, rc = -ENOMEM);
5039
5040         obd = class_name2obd(next);
5041         if (obd == NULL) {
5042                 CERROR("%s: can't locate next device: %s\n",
5043                        mdt_obd_name(m), next);
5044                 GOTO(out, rc = -ENOTCONN);
5045         }
5046
5047         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5048         data->ocd_version = LUSTRE_VERSION_CODE;
5049
5050         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5051         if (rc) {
5052                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5053                        mdt_obd_name(m), next, rc);
5054                 GOTO(out, rc);
5055         }
5056
5057 out:
5058         if (data)
5059                 OBD_FREE_PTR(data);
5060         RETURN(rc);
5061 }
5062
5063 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5064                           struct lustre_cfg *cfg)
5065 {
5066         char                   *dev = lustre_cfg_string(cfg, 0);
5067         int                     rc, name_size, uuid_size;
5068         char                   *name, *uuid, *p;
5069         struct lustre_cfg_bufs *bufs;
5070         struct lustre_cfg      *lcfg;
5071         struct obd_device      *obd;
5072         struct lustre_profile  *lprof;
5073         struct lu_site         *site;
5074         ENTRY;
5075
5076         /* in 1.8 we had the only device in the stack - MDS.
5077          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5078          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5079          * to generate names and setup MDT, MDD. MDT will be using
5080          * generated name to connect to MDD. for MDD the next device
5081          * will be LOD with name taken from so called "profile" which
5082          * is generated by mount_option line
5083          *
5084          * 1.8 MGS generates config. commands like this:
5085          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5086          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5087          * 2.0 MGS generates config. commands like this:
5088          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5089          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5090          *                    3:lustre-MDT0000-mdtlov  4:f
5091          *
5092          * we generate MDD name from MDT one, just replacing T with D
5093          *
5094          * after all the preparations, the logical equivalent will be
5095          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5096          *                    3:lustre-MDT0000-mdtlov  4:f
5097          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5098          *                    3:lustre-MDD0000  4:f
5099          *
5100          *  notice we build the stack from down to top: MDD first, then MDT */
5101
5102         name_size = MAX_OBD_NAME;
5103         uuid_size = MAX_OBD_NAME;
5104
5105         OBD_ALLOC(name, name_size);
5106         OBD_ALLOC(uuid, uuid_size);
5107         if (name == NULL || uuid == NULL)
5108                 GOTO(cleanup_mem, rc = -ENOMEM);
5109
5110         OBD_ALLOC_PTR(bufs);
5111         if (!bufs)
5112                 GOTO(cleanup_mem, rc = -ENOMEM);
5113
5114         strcpy(name, dev);
5115         p = strstr(name, "-MDT");
5116         if (p == NULL)
5117                 GOTO(free_bufs, rc = -ENOMEM);
5118         p[3] = 'D';
5119
5120         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5121
5122         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5123         if (lprof == NULL || lprof->lp_dt == NULL) {
5124                 CERROR("can't find the profile: %s\n",
5125                        lustre_cfg_string(cfg, 0));
5126                 GOTO(free_bufs, rc = -EINVAL);
5127         }
5128
5129         lustre_cfg_bufs_reset(bufs, name);
5130         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5131         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5132         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5133
5134         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5135         if (!lcfg)
5136                 GOTO(put_profile, rc = -ENOMEM);
5137         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5138
5139         rc = class_attach(lcfg);
5140         if (rc)
5141                 GOTO(lcfg_cleanup, rc);
5142
5143         obd = class_name2obd(name);
5144         if (!obd) {
5145                 CERROR("Can not find obd %s (%s in config)\n",
5146                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5147                 GOTO(lcfg_cleanup, rc = -EINVAL);
5148         }
5149
5150         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5151
5152         lustre_cfg_bufs_reset(bufs, name);
5153         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5154         lustre_cfg_bufs_set_string(bufs, 2, dev);
5155         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5156
5157         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5158         if (!lcfg)
5159                 GOTO(class_detach, rc = -ENOMEM);
5160         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5161
5162         rc = class_setup(obd, lcfg);
5163         if (rc)
5164                 GOTO(class_detach, rc);
5165
5166         /* connect to MDD we just setup */
5167         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5168         if (rc)
5169                 GOTO(class_detach, rc);
5170
5171         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5172         LASSERT(site);
5173         LASSERT(mdt_lu_site(mdt) == NULL);
5174         mdt->mdt_lu_dev.ld_site = site;
5175         site->ls_top_dev = &mdt->mdt_lu_dev;
5176         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5177
5178         /* now connect to bottom OSD */
5179         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5180         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5181         if (rc)
5182                 GOTO(class_detach, rc);
5183         mdt->mdt_bottom =
5184                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5185
5186         rc = lu_env_refill((struct lu_env *)env);
5187         if (rc != 0)
5188                 CERROR("Failure to refill session: '%d'\n", rc);
5189
5190         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5191
5192         EXIT;
5193 class_detach:
5194         if (rc)
5195                 class_detach(obd, lcfg);
5196 lcfg_cleanup:
5197         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5198 put_profile:
5199         class_put_profile(lprof);
5200 free_bufs:
5201         OBD_FREE_PTR(bufs);
5202 cleanup_mem:
5203         if (name)
5204                 OBD_FREE(name, name_size);
5205         if (uuid)
5206                 OBD_FREE(uuid, uuid_size);
5207         RETURN(rc);
5208 }
5209
5210 /* setup quota master target on MDT0 */
5211 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5212                           struct lustre_cfg *cfg)
5213 {
5214         struct obd_device       *obd;
5215         char                    *dev = lustre_cfg_string(cfg, 0);
5216         char                    *qmtname, *uuid, *p;
5217         struct lustre_cfg_bufs  *bufs;
5218         struct lustre_cfg       *lcfg;
5219         struct lustre_profile   *lprof;
5220         struct obd_connect_data *data;
5221         int                      rc;
5222         ENTRY;
5223
5224         LASSERT(mdt->mdt_qmt_exp == NULL);
5225         LASSERT(mdt->mdt_qmt_dev == NULL);
5226
5227         /* quota master is on MDT0 only for now */
5228         if (mdt->mdt_seq_site.ss_node_id != 0)
5229                 RETURN(0);
5230
5231         /* MGS generates config commands which look as follows:
5232          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5233          *                    3:lustre-MDT0000-mdtlov  4:f
5234          *
5235          * We generate the QMT name from the MDT one, just replacing MD with QM
5236          * after all the preparations, the logical equivalent will be:
5237          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5238          *                    3:lustre-MDT0000-osd  4:f */
5239         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5240         OBD_ALLOC(uuid, UUID_MAX);
5241         OBD_ALLOC_PTR(bufs);
5242         OBD_ALLOC_PTR(data);
5243         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5244                 GOTO(cleanup_mem, rc = -ENOMEM);
5245
5246         strcpy(qmtname, dev);
5247         p = strstr(qmtname, "-MDT");
5248         if (p == NULL)
5249                 GOTO(cleanup_mem, rc = -ENOMEM);
5250         /* replace MD with QM */
5251         p[1] = 'Q';
5252         p[2] = 'M';
5253
5254         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5255
5256         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5257         if (lprof == NULL || lprof->lp_dt == NULL) {
5258                 CERROR("can't find profile for %s\n",
5259                        lustre_cfg_string(cfg, 0));
5260                 GOTO(cleanup_mem, rc = -EINVAL);
5261         }
5262
5263         lustre_cfg_bufs_reset(bufs, qmtname);
5264         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5265         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5266         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5267
5268         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5269         if (!lcfg)
5270                 GOTO(put_profile, rc = -ENOMEM);
5271         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5272
5273         rc = class_attach(lcfg);
5274         if (rc)
5275                 GOTO(lcfg_cleanup, rc);
5276
5277         obd = class_name2obd(qmtname);
5278         if (!obd) {
5279                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5280                        lustre_cfg_string(cfg, 0));
5281                 GOTO(lcfg_cleanup, rc = -EINVAL);
5282         }
5283
5284         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5285
5286         lustre_cfg_bufs_reset(bufs, qmtname);
5287         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5288         lustre_cfg_bufs_set_string(bufs, 2, dev);
5289
5290         /* for quota, the next device should be the OSD device */
5291         lustre_cfg_bufs_set_string(bufs, 3,
5292                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5293
5294         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5295         if (!lcfg)
5296                 GOTO(class_detach, rc = -ENOMEM);
5297         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5298
5299         rc = class_setup(obd, lcfg);
5300         if (rc)
5301                 GOTO(class_detach, rc);
5302
5303         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5304
5305         /* configure local quota objects */
5306         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5307                 rc = -EBADF;
5308         else
5309                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5310                                                            &mdt->mdt_lu_dev,
5311                                                            mdt->mdt_qmt_dev);
5312         if (rc)
5313                 GOTO(class_cleanup, rc);
5314
5315         /* connect to quota master target */
5316         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5317         data->ocd_version = LUSTRE_VERSION_CODE;
5318         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5319                          data, NULL);
5320         if (rc) {
5321                 CERROR("cannot connect to quota master device %s (%d)\n",
5322                        qmtname, rc);
5323                 GOTO(class_cleanup, rc);
5324         }
5325
5326         EXIT;
5327 class_cleanup:
5328         if (rc) {
5329                 class_manual_cleanup(obd);
5330                 mdt->mdt_qmt_dev = NULL;
5331                 GOTO(lcfg_cleanup, rc);
5332         }
5333 class_detach:
5334         if (rc)
5335                 class_detach(obd, lcfg);
5336 lcfg_cleanup:
5337         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5338 put_profile:
5339         class_put_profile(lprof);
5340 cleanup_mem:
5341         if (bufs)
5342                 OBD_FREE_PTR(bufs);
5343         if (qmtname)
5344                 OBD_FREE(qmtname, MAX_OBD_NAME);
5345         if (uuid)
5346                 OBD_FREE(uuid, UUID_MAX);
5347         if (data)
5348                 OBD_FREE_PTR(data);
5349         return rc;
5350 }
5351
5352 /* Shutdown quota master target associated with mdt */
5353 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5354 {
5355         ENTRY;
5356
5357         if (mdt->mdt_qmt_exp == NULL)
5358                 RETURN_EXIT;
5359         LASSERT(mdt->mdt_qmt_dev != NULL);
5360
5361         /* the qmt automatically shuts down when the mdt disconnects */
5362         obd_disconnect(mdt->mdt_qmt_exp);
5363         mdt->mdt_qmt_exp = NULL;
5364         mdt->mdt_qmt_dev = NULL;
5365         EXIT;
5366 }
5367
5368 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5369  * for now. This will be removed along with converting rest of MDT code
5370  * to use tgt_session_info */
5371 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5372 {
5373         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5374         int                      rc;
5375
5376         if (unlikely(info->mti_object == NULL))
5377                 return -EPROTO;
5378
5379         rc = mdt_getxattr(info);
5380
5381         mdt_thread_info_fini(info);
5382         return rc;
5383 }
5384
5385 static int mdt_llog_open(struct tgt_session_info *tsi)
5386 {
5387         ENTRY;
5388
5389         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5390                 RETURN(err_serious(-EACCES));
5391
5392         RETURN(tgt_llog_open(tsi));
5393 }
5394
5395 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5396 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5397 #define OST_BRW_READ    OST_READ
5398 #define OST_BRW_WRITE   OST_WRITE
5399
5400 static struct tgt_handler mdt_tgt_handlers[] = {
5401 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5402                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5403                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5404 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5405                 0,                      MDS_DISCONNECT, tgt_disconnect,
5406                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5407 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5408                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5409                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5410 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5411 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5412 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5413 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5414                                                         mdt_getattr_name),
5415 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5416 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5417 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5418 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5419 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5420 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5421 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5422 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5423                                                         mdt_hsm_progress),
5424 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5425                                                         mdt_hsm_ct_register),
5426 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5427                                                         mdt_hsm_ct_unregister),
5428 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5429                                                         mdt_hsm_state_get),
5430 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5431                                                         mdt_hsm_state_set),
5432 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5433 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5434                                                         mdt_hsm_request),
5435 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5436             MDS_SWAP_LAYOUTS,
5437             mdt_swap_layouts),
5438 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5439 };
5440
5441 static struct tgt_handler mdt_io_ops[] = {
5442 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5443                                                         mdt_hp_brw),
5444 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5445                                                         mdt_hp_brw),
5446 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5447                                          OST_PUNCH,     mdt_punch_hdl,
5448                                                         mdt_hp_punch),
5449 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5450 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5451 };
5452
5453 static struct tgt_handler mdt_sec_ctx_ops[] = {
5454 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5455 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5456 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5457 };
5458
5459 static struct tgt_handler mdt_quota_ops[] = {
5460 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5461 };
5462
5463 static struct tgt_handler mdt_llog_handlers[] = {
5464         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5465         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5466         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5467         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5468 };
5469
5470 static struct tgt_opc_slice mdt_common_slice[] = {
5471         {
5472                 .tos_opc_start  = MDS_FIRST_OPC,
5473                 .tos_opc_end    = MDS_LAST_OPC,
5474                 .tos_hs         = mdt_tgt_handlers
5475         },
5476         {
5477                 .tos_opc_start  = OBD_FIRST_OPC,
5478                 .tos_opc_end    = OBD_LAST_OPC,
5479                 .tos_hs         = tgt_obd_handlers
5480         },
5481         {
5482                 .tos_opc_start  = LDLM_FIRST_OPC,
5483                 .tos_opc_end    = LDLM_LAST_OPC,
5484                 .tos_hs         = tgt_dlm_handlers
5485         },
5486         {
5487                 .tos_opc_start  = SEC_FIRST_OPC,
5488                 .tos_opc_end    = SEC_LAST_OPC,
5489                 .tos_hs         = mdt_sec_ctx_ops
5490         },
5491         {
5492                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5493                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5494                 .tos_hs         = tgt_out_handlers
5495         },
5496         {
5497                 .tos_opc_start  = FLD_FIRST_OPC,
5498                 .tos_opc_end    = FLD_LAST_OPC,
5499                 .tos_hs         = fld_handlers
5500         },
5501         {
5502                 .tos_opc_start  = SEQ_FIRST_OPC,
5503                 .tos_opc_end    = SEQ_LAST_OPC,
5504                 .tos_hs         = seq_handlers
5505         },
5506         {
5507                 .tos_opc_start  = QUOTA_DQACQ,
5508                 .tos_opc_end    = QUOTA_LAST_OPC,
5509                 .tos_hs         = mdt_quota_ops
5510         },
5511         {
5512                 .tos_opc_start  = LLOG_FIRST_OPC,
5513                 .tos_opc_end    = LLOG_LAST_OPC,
5514                 .tos_hs         = mdt_llog_handlers
5515         },
5516         {
5517                 .tos_opc_start  = LFSCK_FIRST_OPC,
5518                 .tos_opc_end    = LFSCK_LAST_OPC,
5519                 .tos_hs         = tgt_lfsck_handlers
5520         },
5521         {
5522                 .tos_opc_start  = OST_FIRST_OPC,
5523                 .tos_opc_end    = OST_LAST_OPC,
5524                 .tos_hs         = mdt_io_ops
5525         },
5526         {
5527                 .tos_hs         = NULL
5528         }
5529 };
5530
5531 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5532 {
5533         struct md_device *next = m->mdt_child;
5534         struct lu_device *d = &m->mdt_lu_dev;
5535         struct obd_device *obd = mdt2obd_dev(m);
5536         struct lfsck_stop stop;
5537
5538         ENTRY;
5539         stop.ls_status = LS_PAUSED;
5540         stop.ls_flags = 0;
5541         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5542
5543         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5544
5545         mdt_restriper_stop(m);
5546         ping_evictor_stop();
5547
5548         /* Remove the HSM /proc entry so the coordinator cannot be
5549          * restarted by a user while it's shutting down.
5550          */
5551         mdt_hsm_cdt_stop(m);
5552
5553         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5554         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5555
5556         if (m->mdt_namespace != NULL)
5557                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5558                                           d->ld_obd->obd_force);
5559
5560         obd_exports_barrier(obd);
5561         obd_zombie_barrier();
5562
5563         mdt_quota_fini(env, m);
5564
5565         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5566
5567         /* Calling the cleanup functions in the same order as in the mdt_init0
5568          * error path
5569          */
5570         mdt_tunables_fini(m);
5571
5572         target_recovery_fini(obd);
5573         upcall_cache_cleanup(m->mdt_identity_cache);
5574         m->mdt_identity_cache = NULL;
5575
5576         mdt_fs_cleanup(env, m);
5577
5578         tgt_fini(env, &m->mdt_lut);
5579
5580         mdt_hsm_cdt_fini(m);
5581
5582         if (m->mdt_los != NULL) {
5583                 local_oid_storage_fini(env, m->mdt_los);
5584                 m->mdt_los = NULL;
5585         }
5586
5587         if (m->mdt_namespace != NULL) {
5588                 ldlm_namespace_free_post(m->mdt_namespace);
5589                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5590         }
5591
5592         if (m->mdt_md_root != NULL) {
5593                 mdt_object_put(env, m->mdt_md_root);
5594                 m->mdt_md_root = NULL;
5595         }
5596
5597         mdt_seq_fini(env, m);
5598
5599         mdt_fld_fini(env, m);
5600
5601         /*
5602          * Finish the stack
5603          */
5604         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5605
5606         LASSERT(atomic_read(&d->ld_ref) == 0);
5607
5608         server_put_mount(mdt_obd_name(m), true);
5609
5610         EXIT;
5611 }
5612
5613 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5614
5615 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5616                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5617 {
5618         const struct dt_device_param *dt_conf;
5619         struct mdt_thread_info *info;
5620         struct obd_device *obd;
5621         const char *dev = lustre_cfg_string(cfg, 0);
5622         const char *num = lustre_cfg_string(cfg, 2);
5623         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5624         struct lustre_mount_info *lmi = NULL;
5625         struct lustre_sb_info *lsi;
5626         struct lu_site *s;
5627         struct seq_server_site *ss_site;
5628         const char *identity_upcall = "NONE";
5629         struct md_device *next;
5630         struct lu_fid fid;
5631         int rc;
5632         long node_id;
5633         mntopt_t mntopts;
5634         ENTRY;
5635
5636         lu_device_init(&m->mdt_lu_dev, ldt);
5637         /*
5638          * Environment (env) might be missing mdt_thread_key values at that
5639          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5640          * mode.
5641          *
5642          * Usually device allocation path doesn't use module key values, but
5643          * mdt has to do a lot of work here, so allocate key value.
5644          */
5645         rc = lu_env_refill((struct lu_env *)env);
5646         if (rc != 0)
5647                 RETURN(rc);
5648
5649         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5650         LASSERT(info != NULL);
5651
5652         obd = class_name2obd(dev);
5653         LASSERT(obd != NULL);
5654
5655         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5656         m->mdt_opts.mo_evict_tgt_nids = 1;
5657         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5658
5659         lmi = server_get_mount(dev);
5660         if (lmi == NULL) {
5661                 CERROR("Cannot get mount info for %s!\n", dev);
5662                 RETURN(-EFAULT);
5663         } else {
5664                 lsi = s2lsi(lmi->lmi_sb);
5665                 LASSERT(lsi->lsi_lmd);
5666                 /* CMD is supported only in IAM mode */
5667                 LASSERT(num);
5668                 rc = kstrtol(num, 10, &node_id);
5669                 if (rc)
5670                         RETURN(rc);
5671
5672                 obd->u.obt.obt_magic = OBT_MAGIC;
5673                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5674                         m->mdt_skip_lfsck = 1;
5675         }
5676
5677         /* Just try to get a DoM lock by default. Otherwise, having a group
5678          * lock granted, it may get blocked for a long time. */
5679         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
5680         /* DoM files are read at open and data is packed in the reply */
5681         m->mdt_opts.mo_dom_read_open = 1;
5682
5683         m->mdt_squash.rsi_uid = 0;
5684         m->mdt_squash.rsi_gid = 0;
5685         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5686         spin_lock_init(&m->mdt_squash.rsi_lock);
5687         spin_lock_init(&m->mdt_lock);
5688         m->mdt_enable_remote_dir = 1;
5689         m->mdt_enable_striped_dir = 1;
5690         m->mdt_enable_dir_migration = 1;
5691         m->mdt_enable_dir_restripe = 0;
5692         m->mdt_enable_dir_auto_split = 0;
5693         m->mdt_enable_remote_dir_gid = 0;
5694         m->mdt_enable_chprojid_gid = 0;
5695         m->mdt_enable_remote_rename = 1;
5696         m->mdt_dir_restripe_nsonly = 1;
5697         m->mdt_enable_remote_subdir_mount = 1;
5698
5699         atomic_set(&m->mdt_mds_mds_conns, 0);
5700         atomic_set(&m->mdt_async_commit_count, 0);
5701
5702         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5703         m->mdt_lu_dev.ld_obd = obd;
5704         /* Set this lu_device to obd for error handling purposes. */
5705         obd->obd_lu_dev = &m->mdt_lu_dev;
5706
5707         /* init the stack */
5708         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5709         if (rc) {
5710                 CERROR("%s: Can't init device stack, rc %d\n",
5711                        mdt_obd_name(m), rc);
5712                 GOTO(err_lmi, rc);
5713         }
5714
5715         s = mdt_lu_site(m);
5716         ss_site = mdt_seq_site(m);
5717         s->ld_seq_site = ss_site;
5718         ss_site->ss_lu = s;
5719
5720         /* set server index */
5721         ss_site->ss_node_id = node_id;
5722
5723         /* failover is the default
5724          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5725          * assumed whose ss_node_id == 0 XXX
5726          * */
5727         obd->obd_replayable = 1;
5728         /* No connection accepted until configurations will finish */
5729         obd->obd_no_conn = 1;
5730
5731         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5732                 char *str = lustre_cfg_string(cfg, 4);
5733                 if (strchr(str, 'n')) {
5734                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5735                         obd->obd_replayable = 0;
5736                 }
5737         }
5738
5739         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5740         if (rc)
5741                 GOTO(err_fini_stack, rc);
5742
5743         rc = mdt_seq_init(env, m);
5744         if (rc)
5745                 GOTO(err_fini_fld, rc);
5746
5747         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5748                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5749         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5750                                               LDLM_NAMESPACE_SERVER,
5751                                               LDLM_NAMESPACE_GREEDY,
5752                                               LDLM_NS_TYPE_MDT);
5753         if (IS_ERR(m->mdt_namespace)) {
5754                 rc = PTR_ERR(m->mdt_namespace);
5755                 CERROR("%s: unable to create server namespace: rc = %d\n",
5756                        obd->obd_name, rc);
5757                 m->mdt_namespace = NULL;
5758                 GOTO(err_fini_seq, rc);
5759         }
5760
5761         m->mdt_namespace->ns_lvbp = m;
5762         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5763
5764         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5765         /* set obd_namespace for compatibility with old code */
5766         obd->obd_namespace = m->mdt_namespace;
5767
5768         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5769                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5770                       OBD_FAIL_MDS_ALL_REPLY_NET);
5771         if (rc)
5772                 GOTO(err_free_ns, rc);
5773
5774         /* Amount of available space excluded from granting and reserved
5775          * for metadata. It is a percentage of the total MDT size. */
5776         tgd->tgd_reserved_pcnt = 10;
5777
5778         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5779                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5780         else
5781                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5782
5783         rc = mdt_fs_setup(env, m, obd, lsi);
5784         if (rc)
5785                 GOTO(err_tgt, rc);
5786
5787         fid.f_seq = FID_SEQ_LOCAL_NAME;
5788         fid.f_oid = 1;
5789         fid.f_ver = 0;
5790         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5791         if (rc != 0)
5792                 GOTO(err_fs_cleanup, rc);
5793
5794         rc = mdt_hsm_cdt_init(m);
5795         if (rc != 0) {
5796                 CERROR("%s: error initializing coordinator, rc %d\n",
5797                        mdt_obd_name(m), rc);
5798                 GOTO(err_los_fini, rc);
5799         }
5800
5801         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5802
5803         next = m->mdt_child;
5804         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5805
5806         mntopts = dt_conf->ddp_mntopts;
5807
5808         if (mntopts & MNTOPT_USERXATTR)
5809                 m->mdt_opts.mo_user_xattr = 1;
5810         else
5811                 m->mdt_opts.mo_user_xattr = 0;
5812
5813         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5814
5815         if (mntopts & MNTOPT_ACL)
5816                 m->mdt_opts.mo_acl = 1;
5817         else
5818                 m->mdt_opts.mo_acl = 0;
5819
5820         /* XXX: to support suppgid for ACL, we enable identity_upcall
5821          * by default, otherwise, maybe got unexpected -EACCESS. */
5822         if (m->mdt_opts.mo_acl)
5823                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5824
5825         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5826                                                 identity_upcall,
5827                                                 &mdt_identity_upcall_cache_ops);
5828         if (IS_ERR(m->mdt_identity_cache)) {
5829                 rc = PTR_ERR(m->mdt_identity_cache);
5830                 m->mdt_identity_cache = NULL;
5831                 GOTO(err_free_hsm, rc);
5832         }
5833
5834         rc = mdt_tunables_init(m, dev);
5835         if (rc) {
5836                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5837                 GOTO(err_recovery, rc);
5838         }
5839
5840         rc = mdt_quota_init(env, m, cfg);
5841         if (rc)
5842                 GOTO(err_procfs, rc);
5843
5844         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5845         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5846                            "mdt_ldlm_client", m->mdt_ldlm_client);
5847
5848         ping_evictor_start();
5849
5850         /* recovery will be started upon mdt_prepare()
5851          * when the whole stack is complete and ready
5852          * to serve the requests */
5853
5854         /* Reduce the initial timeout on an MDS because it doesn't need such
5855          * a long timeout as an OST does. Adaptive timeouts will adjust this
5856          * value appropriately. */
5857         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5858                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5859
5860         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
5861                 m->mdt_lut.lut_local_recovery = 1;
5862
5863         rc = mdt_restriper_start(m);
5864         if (rc)
5865                 GOTO(err_ping_evictor, rc);
5866
5867         RETURN(0);
5868
5869 err_ping_evictor:
5870         ping_evictor_stop();
5871 err_procfs:
5872         mdt_tunables_fini(m);
5873 err_recovery:
5874         upcall_cache_cleanup(m->mdt_identity_cache);
5875         m->mdt_identity_cache = NULL;
5876 err_free_hsm:
5877         mdt_hsm_cdt_fini(m);
5878 err_los_fini:
5879         local_oid_storage_fini(env, m->mdt_los);
5880         m->mdt_los = NULL;
5881 err_fs_cleanup:
5882         mdt_fs_cleanup(env, m);
5883 err_tgt:
5884         /* keep recoverable clients */
5885         obd->obd_fail = 1;
5886         target_recovery_fini(obd);
5887         obd_exports_barrier(obd);
5888         obd_zombie_barrier();
5889         tgt_fini(env, &m->mdt_lut);
5890 err_free_ns:
5891         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5892         obd->obd_namespace = m->mdt_namespace = NULL;
5893 err_fini_seq:
5894         mdt_seq_fini(env, m);
5895 err_fini_fld:
5896         mdt_fld_fini(env, m);
5897 err_fini_stack:
5898         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5899 err_lmi:
5900         if (lmi)
5901                 server_put_mount(dev, true);
5902         return(rc);
5903 }
5904
5905 /* For interoperability, the left element is old parameter, the right one
5906  * is the new version of the parameter, if some parameter is deprecated,
5907  * the new version should be set as NULL. */
5908 static struct cfg_interop_param mdt_interop_param[] = {
5909         { "mdt.group_upcall",   NULL },
5910         { "mdt.quota_type",     NULL },
5911         { "mdd.quota_type",     NULL },
5912         { "mdt.som",            NULL },
5913         { "mdt.rootsquash",     "mdt.root_squash" },
5914         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5915         { NULL }
5916 };
5917
5918 /* used by MGS to process specific configurations */
5919 static int mdt_process_config(const struct lu_env *env,
5920                               struct lu_device *d, struct lustre_cfg *cfg)
5921 {
5922         struct mdt_device *m = mdt_dev(d);
5923         struct md_device *md_next = m->mdt_child;
5924         struct lu_device *next = md2lu_dev(md_next);
5925         int rc;
5926         ENTRY;
5927
5928         switch (cfg->lcfg_command) {
5929         case LCFG_PARAM: {
5930                 struct obd_device *obd = d->ld_obd;
5931                 /* For interoperability */
5932                 struct cfg_interop_param *ptr = NULL;
5933                 struct lustre_cfg *old_cfg = NULL;
5934                 char *param = NULL;
5935                 ssize_t count;
5936
5937                 param = lustre_cfg_string(cfg, 1);
5938                 if (param == NULL) {
5939                         CERROR("param is empty\n");
5940                         rc = -EINVAL;
5941                         break;
5942                 }
5943
5944                 ptr = class_find_old_param(param, mdt_interop_param);
5945                 if (ptr != NULL) {
5946                         if (ptr->new_param == NULL) {
5947                                 rc = 0;
5948                                 CWARN("For interoperability, skip this %s."
5949                                       " It is obsolete.\n", ptr->old_param);
5950                                 break;
5951                         }
5952
5953                         CWARN("Found old param %s, changed it to %s.\n",
5954                               ptr->old_param, ptr->new_param);
5955
5956                         old_cfg = cfg;
5957                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5958                         if (IS_ERR(cfg)) {
5959                                 rc = PTR_ERR(cfg);
5960                                 break;
5961                         }
5962                 }
5963
5964                 count = class_modify_config(cfg, PARAM_MDT,
5965                                             &obd->obd_kset.kobj);
5966                 if (count < 0) {
5967                         struct coordinator *cdt = &m->mdt_coordinator;
5968
5969                         /* is it an HSM var ? */
5970                         count = class_modify_config(cfg, PARAM_HSM,
5971                                                     &cdt->cdt_hsm_kobj);
5972                         if (count < 0)
5973                                 /* we don't understand; pass it on */
5974                                 rc = next->ld_ops->ldo_process_config(env, next,
5975                                                                       cfg);
5976                         else
5977                                 rc = count > 0 ? 0 : count;
5978                 } else {
5979                         rc = count > 0 ? 0 : count;
5980                 }
5981
5982                 if (old_cfg)
5983                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5984                                                      cfg->lcfg_buflens));
5985                 break;
5986         }
5987         default:
5988                 /* others are passed further */
5989                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5990                 break;
5991         }
5992         RETURN(rc);
5993 }
5994
5995 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5996                                           const struct lu_object_header *hdr,
5997                                           struct lu_device *d)
5998 {
5999         struct mdt_object *mo;
6000
6001         ENTRY;
6002
6003         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6004         if (mo != NULL) {
6005                 struct lu_object *o;
6006                 struct lu_object_header *h;
6007
6008                 o = &mo->mot_obj;
6009                 h = &mo->mot_header;
6010                 lu_object_header_init(h);
6011                 lu_object_init(o, h, d);
6012                 lu_object_add_top(h, o);
6013                 o->lo_ops = &mdt_obj_ops;
6014                 spin_lock_init(&mo->mot_write_lock);
6015                 mutex_init(&mo->mot_som_mutex);
6016                 mutex_init(&mo->mot_lov_mutex);
6017                 init_rwsem(&mo->mot_dom_sem);
6018                 init_rwsem(&mo->mot_open_sem);
6019                 atomic_set(&mo->mot_open_count, 0);
6020                 mo->mot_restripe_offset = 0;
6021                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6022                 RETURN(o);
6023         }
6024         RETURN(NULL);
6025 }
6026
6027 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6028                            const struct lu_object_conf *unused)
6029 {
6030         struct mdt_device *d = mdt_dev(o->lo_dev);
6031         struct lu_device  *under;
6032         struct lu_object  *below;
6033         int                rc = 0;
6034         ENTRY;
6035
6036         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6037                PFID(lu_object_fid(o)));
6038
6039         under = &d->mdt_child->md_lu_dev;
6040         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6041         if (below != NULL) {
6042                 lu_object_add(o, below);
6043         } else
6044                 rc = -ENOMEM;
6045
6046         RETURN(rc);
6047 }
6048
6049 static void mdt_object_free_rcu(struct rcu_head *head)
6050 {
6051         struct mdt_object *mo = container_of(head, struct mdt_object,
6052                                              mot_header.loh_rcu);
6053
6054         kmem_cache_free(mdt_object_kmem, mo);
6055 }
6056
6057 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6058 {
6059         struct mdt_object *mo = mdt_obj(o);
6060         struct lu_object_header *h;
6061         ENTRY;
6062
6063         h = o->lo_header;
6064         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6065                PFID(lu_object_fid(o)));
6066
6067         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6068         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6069
6070         lu_object_fini(o);
6071         lu_object_header_fini(h);
6072         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6073         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6074
6075         EXIT;
6076 }
6077
6078 static int mdt_object_print(const struct lu_env *env, void *cookie,
6079                             lu_printer_t p, const struct lu_object *o)
6080 {
6081         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6082
6083         return (*p)(env, cookie,
6084                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6085                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6086                     mdto->mot_cache_attr ? "cache_attr" : "",
6087                     mdto->mot_write_count);
6088 }
6089
6090 static int mdt_prepare(const struct lu_env *env,
6091                 struct lu_device *pdev,
6092                 struct lu_device *cdev)
6093 {
6094         struct mdt_device *mdt = mdt_dev(cdev);
6095         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6096         struct obd_device *obd = cdev->ld_obd;
6097         int rc;
6098
6099         ENTRY;
6100
6101         LASSERT(obd);
6102
6103         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6104         if (rc)
6105                 RETURN(rc);
6106
6107         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6108         if (rc)
6109                 RETURN(rc);
6110
6111         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6112         if (rc)
6113                 RETURN(rc);
6114
6115         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6116         /* The LFSCK instance is registered just now, so it must be there when
6117          * register the namespace to such instance. */
6118         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6119
6120         if (mdt->mdt_seq_site.ss_node_id == 0) {
6121                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6122                                                          &mdt->mdt_md_root_fid);
6123                 if (rc)
6124                         RETURN(rc);
6125         }
6126
6127         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6128
6129         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6130         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6131         LASSERT(obd->obd_no_conn);
6132         spin_lock(&obd->obd_dev_lock);
6133         obd->obd_no_conn = 0;
6134         spin_unlock(&obd->obd_dev_lock);
6135
6136         if (obd->obd_recovering == 0)
6137                 mdt_postrecov(env, mdt);
6138
6139         RETURN(rc);
6140 }
6141
6142 const struct lu_device_operations mdt_lu_ops = {
6143         .ldo_object_alloc   = mdt_object_alloc,
6144         .ldo_process_config = mdt_process_config,
6145         .ldo_prepare        = mdt_prepare,
6146 };
6147
6148 static const struct lu_object_operations mdt_obj_ops = {
6149         .loo_object_init    = mdt_object_init,
6150         .loo_object_free    = mdt_object_free,
6151         .loo_object_print   = mdt_object_print
6152 };
6153
6154 static int mdt_obd_set_info_async(const struct lu_env *env,
6155                                   struct obd_export *exp,
6156                                   __u32 keylen, void *key,
6157                                   __u32 vallen, void *val,
6158                                   struct ptlrpc_request_set *set)
6159 {
6160         int rc;
6161
6162         ENTRY;
6163
6164         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6165                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6166                 RETURN(rc);
6167         }
6168
6169         RETURN(0);
6170 }
6171
6172 static inline void mdt_enable_slc(struct mdt_device *mdt)
6173 {
6174         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6175                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6176 }
6177
6178 static inline void mdt_disable_slc(struct mdt_device *mdt)
6179 {
6180         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6181                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6182 }
6183
6184 /**
6185  * Match client and server connection feature flags.
6186  *
6187  * Compute the compatibility flags for a connection request based on
6188  * features mutually supported by client and server.
6189  *
6190  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6191  * must not be updated here, otherwise a partially initialized value may
6192  * be exposed. After the connection request is successfully processed,
6193  * the top-level MDT connect request handler atomically updates the export
6194  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6195  * reply. \see mdt_connect().
6196  *
6197  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6198  * full struct obd_connect_data. So care must be taken when accessing fields
6199  * that are not present in struct obd_connect_data_v1. See LU-16.
6200  *
6201  * \param exp   the obd_export associated with this client/target pair
6202  * \param mdt   the target device for the connection
6203  * \param data  stores data for this connect request
6204  *
6205  * \retval 0       success
6206  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6207  * \retval -EBADE  client and server feature requirements are incompatible
6208  */
6209 static int mdt_connect_internal(const struct lu_env *env,
6210                                 struct obd_export *exp,
6211                                 struct mdt_device *mdt,
6212                                 struct obd_connect_data *data, bool reconnect)
6213 {
6214         const char *obd_name = mdt_obd_name(mdt);
6215         LASSERT(data != NULL);
6216
6217         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6218
6219         if (mdt->mdt_bottom->dd_rdonly &&
6220             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6221             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6222                 RETURN(-EACCES);
6223
6224         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6225                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6226
6227         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6228
6229         if (!mdt->mdt_opts.mo_acl)
6230                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6231
6232         if (!mdt->mdt_opts.mo_user_xattr)
6233                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6234
6235         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6236                 data->ocd_brw_size = min(data->ocd_brw_size,
6237                                          mdt->mdt_brw_size);
6238                 if (data->ocd_brw_size == 0) {
6239                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6240                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6241                                "ocd_brw_size unexpectedly zero, network data "
6242                                "corruption? Refusing to connect this client\n",
6243                                obd_name, exp->exp_client_uuid.uuid,
6244                                exp, data->ocd_connect_flags, data->ocd_version,
6245                                data->ocd_grant, data->ocd_index);
6246                         return -EPROTO;
6247                 }
6248         }
6249
6250         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6251                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6252
6253                 /* client is reporting its page size, for future use */
6254                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6255                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6256                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6257                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6258                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6259                 /* ocd_grant_tax_kb is in 1K byte blocks */
6260                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6261                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6262         }
6263
6264         /* Save connect_data we have so far because tgt_grant_connect()
6265          * uses it to calculate grant, and we want to save the client
6266          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6267         exp->exp_connect_data = *data;
6268         if (OCD_HAS_FLAG(data, GRANT))
6269                 tgt_grant_connect(env, exp, data, !reconnect);
6270
6271         if (OCD_HAS_FLAG(data, MAXBYTES))
6272                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6273
6274         /* NB: Disregard the rule against updating
6275          * exp_connect_data.ocd_connect_flags in this case, since
6276          * tgt_client_new() needs to know if this is a lightweight
6277          * connection, and it is safe to expose this flag before
6278          * connection processing completes. */
6279         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6280                 spin_lock(&exp->exp_lock);
6281                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6282                 spin_unlock(&exp->exp_lock);
6283         }
6284
6285         data->ocd_version = LUSTRE_VERSION_CODE;
6286
6287         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6288                 CWARN("%s: MDS requires FID support, but client not\n",
6289                       obd_name);
6290                 return -EBADE;
6291         }
6292
6293         if (OCD_HAS_FLAG(data, PINGLESS)) {
6294                 if (ptlrpc_pinger_suppress_pings()) {
6295                         spin_lock(&exp->exp_obd->obd_dev_lock);
6296                         list_del_init(&exp->exp_obd_chain_timed);
6297                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6298                 } else {
6299                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6300                 }
6301         }
6302
6303         data->ocd_max_easize = mdt->mdt_max_ea_size;
6304
6305         /* NB: Disregard the rule against updating
6306          * exp_connect_data.ocd_connect_flags in this case, since
6307          * tgt_client_new() needs to know if this is client supports
6308          * multiple modify RPCs, and it is safe to expose this flag before
6309          * connection processing completes. */
6310         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6311                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6312                 spin_lock(&exp->exp_lock);
6313                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6314                 spin_unlock(&exp->exp_lock);
6315         }
6316
6317         if (OCD_HAS_FLAG(data, CKSUM)) {
6318                 __u32 cksum_types = data->ocd_cksum_types;
6319
6320                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6321
6322                 if (unlikely(data->ocd_cksum_types == 0)) {
6323                         CERROR("%s: Connect with checksum support but no "
6324                                "ocd_cksum_types is set\n",
6325                                exp->exp_obd->obd_name);
6326                         RETURN(-EPROTO);
6327                 }
6328
6329                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6330                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6331                        cksum_types, data->ocd_cksum_types);
6332         } else {
6333                 /* This client does not support OBD_CONNECT_CKSUM
6334                  * fall back to CRC32 */
6335                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6336                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6337                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6338         }
6339
6340         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6341             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6342                 atomic_inc(&mdt->mdt_mds_mds_conns);
6343                 mdt_enable_slc(mdt);
6344         }
6345
6346         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6347                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6348
6349         return 0;
6350 }
6351
6352 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6353                                    struct mdt_thread_info *info,
6354                                    struct mdt_file_data *mfd)
6355 {
6356         struct lu_context ses;
6357         int rc;
6358         ENTRY;
6359
6360         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6361         if (rc)
6362                 RETURN(rc);
6363
6364         env->le_ses = &ses;
6365         lu_context_enter(&ses);
6366
6367         mdt_ucred(info)->uc_valid = UCRED_OLD;
6368         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6369
6370         lu_context_exit(&ses);
6371         lu_context_fini(&ses);
6372         env->le_ses = NULL;
6373
6374         RETURN(rc);
6375 }
6376
6377 static int mdt_export_cleanup(struct obd_export *exp)
6378 {
6379         LIST_HEAD(closing_list);
6380         struct mdt_export_data  *med = &exp->exp_mdt_data;
6381         struct obd_device       *obd = exp->exp_obd;
6382         struct mdt_device       *mdt;
6383         struct mdt_thread_info  *info;
6384         struct lu_env            env;
6385         struct mdt_file_data    *mfd, *n;
6386         int rc = 0;
6387         ENTRY;
6388
6389         spin_lock(&med->med_open_lock);
6390         while (!list_empty(&med->med_open_head)) {
6391                 struct list_head *tmp = med->med_open_head.next;
6392                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6393
6394                 /* Remove mfd handle so it can't be found again.
6395                  * We are consuming the mfd_list reference here. */
6396                 class_handle_unhash(&mfd->mfd_open_handle);
6397                 list_move_tail(&mfd->mfd_list, &closing_list);
6398         }
6399         spin_unlock(&med->med_open_lock);
6400         mdt = mdt_dev(obd->obd_lu_dev);
6401         LASSERT(mdt != NULL);
6402
6403         rc = lu_env_init(&env, LCT_MD_THREAD);
6404         if (rc)
6405                 RETURN(rc);
6406
6407         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6408         LASSERT(info != NULL);
6409         memset(info, 0, sizeof *info);
6410         info->mti_env = &env;
6411         info->mti_mdt = mdt;
6412         info->mti_exp = exp;
6413
6414         if (!list_empty(&closing_list)) {
6415                 struct md_attr *ma = &info->mti_attr;
6416
6417                 /* Close any open files (which may also cause orphan
6418                  * unlinking). */
6419                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6420                         list_del_init(&mfd->mfd_list);
6421                         ma->ma_need = ma->ma_valid = 0;
6422
6423                         /* This file is being closed due to an eviction, it
6424                          * could have been modified and now dirty regarding to
6425                          * HSM archive, check this!
6426                          * The logic here is to mark a file dirty if there's a
6427                          * chance it was dirtied before the client was evicted,
6428                          * so that we don't have to wait for a release attempt
6429                          * before finding out the file was actually dirty and
6430                          * fail the release. Aggressively marking it dirty here
6431                          * will cause the policy engine to attempt to
6432                          * re-archive it; when rearchiving, we can compare the
6433                          * current version to the HSM data_version and make the
6434                          * archive request into a noop if it's not actually
6435                          * dirty.
6436                          */
6437                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6438                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6439
6440                         /* Don't unlink orphan on failover umount, LU-184 */
6441                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6442                                 ma->ma_valid = MA_FLAGS;
6443                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6444                         }
6445                         ma->ma_valid |= MA_FORCE_LOG;
6446                         mdt_mfd_close(info, mfd);
6447                 }
6448         }
6449         info->mti_mdt = NULL;
6450         /* cleanup client slot early */
6451         /* Do not erase record for recoverable client. */
6452         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6453                 tgt_client_del(&env, exp);
6454         lu_env_fini(&env);
6455
6456         RETURN(rc);
6457 }
6458
6459 static int mdt_obd_disconnect(struct obd_export *exp)
6460 {
6461         int rc;
6462
6463         ENTRY;
6464
6465         LASSERT(exp);
6466         class_export_get(exp);
6467
6468         if (!(exp->exp_flags & OBD_OPT_FORCE))
6469                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6470
6471         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6472             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6473                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6474
6475                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6476                         mdt_disable_slc(mdt);
6477         }
6478
6479         rc = server_disconnect_export(exp);
6480         if (rc != 0)
6481                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6482
6483         tgt_grant_discard(exp);
6484
6485         rc = mdt_export_cleanup(exp);
6486         nodemap_del_member(exp);
6487         class_export_put(exp);
6488         RETURN(rc);
6489 }
6490
6491 /* mds_connect copy */
6492 static int mdt_obd_connect(const struct lu_env *env,
6493                            struct obd_export **exp, struct obd_device *obd,
6494                            struct obd_uuid *cluuid,
6495                            struct obd_connect_data *data,
6496                            void *localdata)
6497 {
6498         struct obd_export       *lexp;
6499         struct lustre_handle    conn = { 0 };
6500         struct mdt_device       *mdt;
6501         int                      rc;
6502         lnet_nid_t              *client_nid = localdata;
6503         ENTRY;
6504
6505         LASSERT(env != NULL);
6506         LASSERT(data != NULL);
6507
6508         if (!exp || !obd || !cluuid)
6509                 RETURN(-EINVAL);
6510
6511         mdt = mdt_dev(obd->obd_lu_dev);
6512
6513         /*
6514          * first, check whether the stack is ready to handle requests
6515          * XXX: probably not very appropriate method is used now
6516          *      at some point we should find a better one
6517          */
6518         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6519             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6520             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6521                 rc = obd_get_info(env, mdt->mdt_child_exp,
6522                                   sizeof(KEY_OSP_CONNECTED),
6523                                   KEY_OSP_CONNECTED, NULL, NULL);
6524                 if (rc)
6525                         RETURN(-EAGAIN);
6526                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6527         }
6528
6529         rc = class_connect(&conn, obd, cluuid);
6530         if (rc)
6531                 RETURN(rc);
6532
6533         lexp = class_conn2export(&conn);
6534         LASSERT(lexp != NULL);
6535
6536         rc = nodemap_add_member(*client_nid, lexp);
6537         if (rc != 0 && rc != -EEXIST)
6538                 GOTO(out, rc);
6539
6540         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6541         if (rc == 0) {
6542                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6543
6544                 LASSERT(lcd);
6545                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6546                 rc = tgt_client_new(env, lexp);
6547                 if (rc == 0)
6548                         mdt_export_stats_init(obd, lexp, localdata);
6549         }
6550 out:
6551         if (rc != 0) {
6552                 class_disconnect(lexp);
6553                 nodemap_del_member(lexp);
6554                 *exp = NULL;
6555         } else {
6556                 *exp = lexp;
6557                 /* Because we do not want this export to be evicted by pinger,
6558                  * let's not add this export to the timed chain list. */
6559                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6560                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6561                         list_del_init(&lexp->exp_obd_chain_timed);
6562                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6563                 }
6564         }
6565
6566         RETURN(rc);
6567 }
6568
6569 static int mdt_obd_reconnect(const struct lu_env *env,
6570                              struct obd_export *exp, struct obd_device *obd,
6571                              struct obd_uuid *cluuid,
6572                              struct obd_connect_data *data,
6573                              void *localdata)
6574 {
6575         lnet_nid_t             *client_nid = localdata;
6576         int                     rc;
6577         ENTRY;
6578
6579         if (exp == NULL || obd == NULL || cluuid == NULL)
6580                 RETURN(-EINVAL);
6581
6582         rc = nodemap_add_member(*client_nid, exp);
6583         if (rc != 0 && rc != -EEXIST)
6584                 RETURN(rc);
6585
6586         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6587                                   true);
6588         if (rc == 0)
6589                 mdt_export_stats_init(obd, exp, localdata);
6590         else
6591                 nodemap_del_member(exp);
6592
6593         RETURN(rc);
6594 }
6595
6596 /* FIXME: Can we avoid using these two interfaces? */
6597 static int mdt_init_export(struct obd_export *exp)
6598 {
6599         struct mdt_export_data *med = &exp->exp_mdt_data;
6600         int                     rc;
6601         ENTRY;
6602
6603         INIT_LIST_HEAD(&med->med_open_head);
6604         spin_lock_init(&med->med_open_lock);
6605         spin_lock(&exp->exp_lock);
6606         exp->exp_connecting = 1;
6607         spin_unlock(&exp->exp_lock);
6608
6609         OBD_ALLOC(exp->exp_used_slots,
6610                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6611         if (exp->exp_used_slots == NULL)
6612                 RETURN(-ENOMEM);
6613
6614         /* self-export doesn't need client data and ldlm initialization */
6615         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6616                                      &exp->exp_client_uuid)))
6617                 RETURN(0);
6618
6619         rc = tgt_client_alloc(exp);
6620         if (rc)
6621                 GOTO(err, rc);
6622
6623         rc = ldlm_init_export(exp);
6624         if (rc)
6625                 GOTO(err_free, rc);
6626
6627         RETURN(rc);
6628
6629 err_free:
6630         tgt_client_free(exp);
6631 err:
6632         OBD_FREE(exp->exp_used_slots,
6633                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6634         exp->exp_used_slots = NULL;
6635
6636         CERROR("%s: Failed to initialize export: rc = %d\n",
6637                exp->exp_obd->obd_name, rc);
6638         return rc;
6639 }
6640
6641 static int mdt_destroy_export(struct obd_export *exp)
6642 {
6643         ENTRY;
6644
6645         target_destroy_export(exp);
6646         if (exp->exp_used_slots)
6647                 OBD_FREE(exp->exp_used_slots,
6648                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6649
6650         /* destroy can be called from failed obd_setup, so
6651          * checking uuid is safer than obd_self_export */
6652         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6653                                      &exp->exp_client_uuid)))
6654                 RETURN(0);
6655
6656         ldlm_destroy_export(exp);
6657         tgt_client_free(exp);
6658
6659         LASSERT(list_empty(&exp->exp_outstanding_replies));
6660         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6661
6662         /*
6663          * discard grants once we're sure no more
6664          * interaction with the client is possible
6665          */
6666         tgt_grant_discard(exp);
6667         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6668                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6669
6670         if (!(exp->exp_flags & OBD_OPT_FORCE))
6671                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6672
6673         RETURN(0);
6674 }
6675
6676 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6677                    struct linkea_data *ldata)
6678 {
6679         int rc;
6680
6681         LASSERT(ldata->ld_buf->lb_buf != NULL);
6682
6683         if (!mdt_object_exists(mdt_obj))
6684                 return -ENODATA;
6685
6686         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6687                           ldata->ld_buf, XATTR_NAME_LINK);
6688         if (rc == -ERANGE) {
6689                 /* Buf was too small, figure out what we need. */
6690                 lu_buf_free(ldata->ld_buf);
6691                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6692                                   ldata->ld_buf, XATTR_NAME_LINK);
6693                 if (rc < 0)
6694                         return rc;
6695                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6696                 if (ldata->ld_buf->lb_buf == NULL)
6697                         return -ENOMEM;
6698                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6699                                   ldata->ld_buf, XATTR_NAME_LINK);
6700         }
6701         if (rc < 0)
6702                 return rc;
6703
6704         return linkea_init_with_rec(ldata);
6705 }
6706
6707 /**
6708  * Given an MDT object, try to look up the full path to the object.
6709  * Part of the MDT layer implementation of lfs fid2path.
6710  *
6711  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6712  * \param[in]     obj   Object to do path lookup of
6713  * \param[in,out] fp    User-provided struct to store path information
6714  * \param[in]     root_fid Root FID of current path should reach
6715  *
6716  * \retval 0 Lookup successful, path information stored in fp
6717  * \retval -EAGAIN Lookup failed, usually because object is being moved
6718  * \retval negative errno if there was a problem
6719  */
6720 static int mdt_path_current(struct mdt_thread_info *info,
6721                             struct mdt_object *obj,
6722                             struct getinfo_fid2path *fp,
6723                             struct lu_fid *root_fid)
6724 {
6725         struct mdt_device *mdt = info->mti_mdt;
6726         struct lu_name *tmpname = &info->mti_name;
6727         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
6728         struct lu_buf *buf = &info->mti_big_buf;
6729         struct md_attr *ma = &info->mti_attr;
6730         struct linkea_data ldata = { NULL };
6731         bool first = true;
6732         struct mdt_object *mdt_obj;
6733         struct link_ea_header *leh;
6734         struct link_ea_entry *lee;
6735         char *ptr;
6736         int reclen;
6737         int rc = 0;
6738
6739         ENTRY;
6740
6741         /* temp buffer for path element, the buffer will be finally freed
6742          * in mdt_thread_info_fini */
6743         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6744         if (buf->lb_buf == NULL)
6745                 RETURN(-ENOMEM);
6746
6747         ldata.ld_buf = buf;
6748         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6749         *ptr = 0;
6750         --ptr;
6751         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6752
6753         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6754                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6755                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6756                         GOTO(out, rc = -ENOENT);
6757
6758                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6759                         mdt_obj = obj;
6760                         mdt_object_get(info->mti_env, mdt_obj);
6761                 } else {
6762                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6763                         if (IS_ERR(mdt_obj))
6764                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6765                 }
6766
6767                 if (!mdt_object_exists(mdt_obj)) {
6768                         mdt_object_put(info->mti_env, mdt_obj);
6769                         GOTO(out, rc = -ENOENT);
6770                 }
6771
6772                 if (mdt_object_remote(mdt_obj)) {
6773                         mdt_object_put(info->mti_env, mdt_obj);
6774                         GOTO(remote_out, rc = -EREMOTE);
6775                 }
6776
6777                 rc = mdt_links_read(info, mdt_obj, &ldata);
6778                 if (rc != 0) {
6779                         mdt_object_put(info->mti_env, mdt_obj);
6780                         GOTO(out, rc);
6781                 }
6782
6783                 leh = buf->lb_buf;
6784                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6785                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6786                 /* If set, use link #linkno for path lookup, otherwise use
6787                    link #0.  Only do this for the final path element. */
6788                 if (first && fp->gf_linkno < leh->leh_reccount) {
6789                         int count;
6790                         for (count = 0; count < fp->gf_linkno; count++) {
6791                                 lee = (struct link_ea_entry *)
6792                                      ((char *)lee + reclen);
6793                                 linkea_entry_unpack(lee, &reclen, tmpname,
6794                                                     tmpfid);
6795                         }
6796                         if (fp->gf_linkno < leh->leh_reccount - 1)
6797                                 /* indicate to user there are more links */
6798                                 fp->gf_linkno++;
6799                 }
6800
6801                 /* Check if it is slave stripes */
6802                 rc = mdt_stripe_get(info, mdt_obj, ma, XATTR_NAME_LMV);
6803                 mdt_object_put(info->mti_env, mdt_obj);
6804                 if (rc < 0)
6805                         GOTO(out, rc);
6806
6807                 if (ma->ma_valid & MA_LMV) {
6808                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
6809
6810                         if (!lmv_is_sane2(lmv))
6811                                 GOTO(out, rc = -EBADF);
6812
6813                         /* For slave stripes, get its master */
6814                         if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE) {
6815                                 fp->gf_fid = *tmpfid;
6816                                 continue;
6817                         }
6818                 }
6819
6820                 /* Pack the name in the end of the buffer */
6821                 ptr -= tmpname->ln_namelen;
6822                 if (ptr - 1 <= fp->gf_u.gf_path)
6823                         GOTO(out, rc = -EOVERFLOW);
6824                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6825                 *(--ptr) = '/';
6826
6827                 /* keep the last resolved fid to the client, so the
6828                  * client will build the left path on another MDT for
6829                  * remote object */
6830                 fp->gf_fid = *tmpfid;
6831
6832                 first = false;
6833         }
6834
6835         /* non-zero will be treated as an error */
6836         rc = 0;
6837
6838 remote_out:
6839         ptr++; /* skip leading / */
6840         memmove(fp->gf_u.gf_path, ptr,
6841                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6842
6843 out:
6844         RETURN(rc);
6845 }
6846
6847 /**
6848  * Given an MDT object, use mdt_path_current to get the path.
6849  * Essentially a wrapper to retry mdt_path_current a set number of times
6850  * if -EAGAIN is returned (usually because an object is being moved).
6851  *
6852  * Part of the MDT layer implementation of lfs fid2path.
6853  *
6854  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6855  * \param[in]     obj   Object to do path lookup of
6856  * \param[in,out] fp    User-provided struct for arguments and to store path
6857  *                      information
6858  *
6859  * \retval 0 Lookup successful, path information stored in fp
6860  * \retval negative errno if there was a problem
6861  */
6862 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6863                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6864 {
6865         struct mdt_device       *mdt = info->mti_mdt;
6866         int                     tries = 3;
6867         int                     rc = -EAGAIN;
6868         ENTRY;
6869
6870         if (fp->gf_pathlen < 3)
6871                 RETURN(-EOVERFLOW);
6872
6873         if (root_fid == NULL)
6874                 root_fid = &mdt->mdt_md_root_fid;
6875
6876         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6877                 fp->gf_u.gf_path[0] = '\0';
6878                 RETURN(0);
6879         }
6880
6881         /* Retry multiple times in case file is being moved */
6882         while (tries-- && rc == -EAGAIN)
6883                 rc = mdt_path_current(info, obj, fp, root_fid);
6884
6885         RETURN(rc);
6886 }
6887
6888 /**
6889  * Get the full path of the provided FID, as of changelog record recno.
6890  *
6891  * This checks sanity and looks up object for user provided FID
6892  * before calling the actual path lookup code.
6893  *
6894  * Part of the MDT layer implementation of lfs fid2path.
6895  *
6896  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6897  * \param[in,out] fp    User-provided struct for arguments and to store path
6898  *                      information
6899  *
6900  * \retval 0 Lookup successful, path information and recno stored in fp
6901  * \retval -ENOENT, object does not exist
6902  * \retval negative errno if there was a problem
6903  */
6904 static int mdt_fid2path(struct mdt_thread_info *info,
6905                         struct lu_fid *root_fid,
6906                         struct getinfo_fid2path *fp)
6907 {
6908         struct mdt_device *mdt = info->mti_mdt;
6909         struct mdt_object *obj;
6910         int    rc;
6911         ENTRY;
6912
6913         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6914                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6915
6916         if (!fid_is_sane(&fp->gf_fid))
6917                 RETURN(-EINVAL);
6918
6919         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6920                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6921                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6922                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6923                 RETURN(-EINVAL);
6924         }
6925
6926         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6927         if (IS_ERR(obj)) {
6928                 rc = PTR_ERR(obj);
6929                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6930                        PFID(&fp->gf_fid), rc);
6931                 RETURN(rc);
6932         }
6933
6934         if (mdt_object_remote(obj))
6935                 rc = -EREMOTE;
6936         else if (!mdt_object_exists(obj))
6937                 rc = -ENOENT;
6938         else
6939                 rc = 0;
6940
6941         if (rc < 0) {
6942                 mdt_object_put(info->mti_env, obj);
6943                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6944                        PFID(&fp->gf_fid), rc);
6945                 RETURN(rc);
6946         }
6947
6948         rc = mdt_path(info, obj, fp, root_fid);
6949
6950         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6951                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6952                fp->gf_recno, fp->gf_linkno);
6953
6954         mdt_object_put(info->mti_env, obj);
6955
6956         RETURN(rc);
6957 }
6958
6959 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6960                             void *val, int vallen)
6961 {
6962         struct getinfo_fid2path *fpout, *fpin;
6963         struct lu_fid *root_fid = NULL;
6964         int rc = 0;
6965
6966         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6967         fpout = val;
6968
6969         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6970                 lustre_swab_fid2path(fpin);
6971
6972         memcpy(fpout, fpin, sizeof(*fpin));
6973         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6974                 RETURN(-EINVAL);
6975
6976         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6977                       sizeof(struct lu_fid)) {
6978                 /* client sent its root FID, which is normally fileset FID */
6979                 root_fid = fpin->gf_u.gf_root_fid;
6980                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6981                         lustre_swab_lu_fid(root_fid);
6982
6983                 if (root_fid != NULL && !fid_is_sane(root_fid))
6984                         RETURN(-EINVAL);
6985         }
6986
6987         rc = mdt_fid2path(info, root_fid, fpout);
6988         RETURN(rc);
6989 }
6990
6991 int mdt_get_info(struct tgt_session_info *tsi)
6992 {
6993         char    *key;
6994         int      keylen;
6995         __u32   *vallen;
6996         void    *valout;
6997         int      rc;
6998
6999         ENTRY;
7000
7001         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7002         if (key == NULL) {
7003                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7004                 RETURN(err_serious(-EFAULT));
7005         }
7006         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7007                                       RCL_CLIENT);
7008
7009         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7010         if (vallen == NULL) {
7011                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7012                                 tgt_name(tsi->tsi_tgt));
7013                 RETURN(err_serious(-EFAULT));
7014         }
7015
7016         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7017                              *vallen);
7018         rc = req_capsule_server_pack(tsi->tsi_pill);
7019         if (rc)
7020                 RETURN(err_serious(rc));
7021
7022         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7023         if (valout == NULL) {
7024                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7025                                 tgt_name(tsi->tsi_tgt));
7026                 RETURN(err_serious(-EFAULT));
7027         }
7028
7029         if (KEY_IS(KEY_FID2PATH)) {
7030                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7031
7032                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7033                 mdt_thread_info_fini(info);
7034         } else {
7035                 rc = -EINVAL;
7036         }
7037         RETURN(rc);
7038 }
7039
7040 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7041 {
7042         struct obd_ioctl_data *data = karg;
7043         struct lu_fid *fid;
7044         __u64 version;
7045         struct mdt_object *obj;
7046         struct mdt_lock_handle  *lh;
7047         int rc;
7048         ENTRY;
7049
7050         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7051             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7052                 RETURN(-EINVAL);
7053
7054         fid = (struct lu_fid *)data->ioc_inlbuf1;
7055
7056         if (!fid_is_sane(fid))
7057                 RETURN(-EINVAL);
7058
7059         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7060
7061         lh = &mti->mti_lh[MDT_LH_PARENT];
7062         mdt_lock_reg_init(lh, LCK_CR);
7063
7064         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
7065         if (IS_ERR(obj))
7066                 RETURN(PTR_ERR(obj));
7067
7068         if (mdt_object_remote(obj)) {
7069                 rc = -EREMOTE;
7070                 /**
7071                  * before calling version get the correct MDS should be
7072                  * fid, this is error to find remote object here
7073                  */
7074                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7075         } else if (!mdt_object_exists(obj)) {
7076                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7077                 rc = -ENOENT;
7078         } else {
7079                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7080                *(__u64 *)data->ioc_inlbuf2 = version;
7081                 rc = 0;
7082         }
7083         mdt_object_unlock_put(mti, obj, lh, 1);
7084         RETURN(rc);
7085 }
7086
7087 /* ioctls on obd dev */
7088 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7089                          void *karg, void __user *uarg)
7090 {
7091         struct lu_env      env;
7092         struct obd_device *obd = exp->exp_obd;
7093         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7094         struct dt_device  *dt = mdt->mdt_bottom;
7095         int rc;
7096
7097         ENTRY;
7098         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
7099         rc = lu_env_init(&env, LCT_MD_THREAD);
7100         if (rc)
7101                 RETURN(rc);
7102
7103         switch (cmd) {
7104         case OBD_IOC_SYNC:
7105                 rc = mdt_device_sync(&env, mdt);
7106                 break;
7107         case OBD_IOC_SET_READONLY:
7108                 rc = dt_sync(&env, dt);
7109                 if (rc == 0)
7110                         rc = dt_ro(&env, dt);
7111                 break;
7112         case OBD_IOC_ABORT_RECOVERY: {
7113                 struct obd_ioctl_data *data = karg;
7114
7115                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
7116                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT)
7117                         obd->obd_abort_recov_mdt = 1;
7118                 else /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7119                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7120                         obd->obd_abort_recovery = 1;
7121
7122                 target_stop_recovery_thread(obd);
7123                 rc = 0;
7124                 break;
7125         }
7126         case OBD_IOC_CHANGELOG_REG:
7127         case OBD_IOC_CHANGELOG_DEREG:
7128         case OBD_IOC_CHANGELOG_CLEAR:
7129                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
7130                                                            mdt->mdt_child,
7131                                                            cmd, len, karg);
7132                 break;
7133         case OBD_IOC_START_LFSCK: {
7134                 struct md_device *next = mdt->mdt_child;
7135                 struct obd_ioctl_data *data = karg;
7136                 struct lfsck_start_param lsp;
7137
7138                 if (unlikely(data == NULL)) {
7139                         rc = -EINVAL;
7140                         break;
7141                 }
7142
7143                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7144                 lsp.lsp_index_valid = 0;
7145                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7146                 break;
7147         }
7148         case OBD_IOC_STOP_LFSCK: {
7149                 struct md_device        *next = mdt->mdt_child;
7150                 struct obd_ioctl_data   *data = karg;
7151                 struct lfsck_stop        stop;
7152
7153                 stop.ls_status = LS_STOPPED;
7154                 /* Old lfsck utils may pass NULL @stop. */
7155                 if (data->ioc_inlbuf1 == NULL)
7156                         stop.ls_flags = 0;
7157                 else
7158                         stop.ls_flags =
7159                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7160
7161                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7162                 break;
7163         }
7164         case OBD_IOC_QUERY_LFSCK: {
7165                 struct md_device        *next = mdt->mdt_child;
7166                 struct obd_ioctl_data   *data = karg;
7167
7168                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7169                                                  data->ioc_inlbuf1);
7170                 break;
7171         }
7172         case OBD_IOC_GET_OBJ_VERSION: {
7173                 struct mdt_thread_info *mti;
7174                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7175                 memset(mti, 0, sizeof *mti);
7176                 mti->mti_env = &env;
7177                 mti->mti_mdt = mdt;
7178                 mti->mti_exp = exp;
7179
7180                 rc = mdt_ioc_version_get(mti, karg);
7181                 break;
7182         }
7183         case OBD_IOC_CATLOGLIST: {
7184                 struct mdt_thread_info *mti;
7185
7186                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7187                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7188                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7189                                        &mti->mti_tmp_fid1);
7190                 break;
7191          }
7192         default:
7193                 rc = -EOPNOTSUPP;
7194                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
7195                         mdt_obd_name(mdt), cmd, rc);
7196         }
7197
7198         lu_env_fini(&env);
7199         RETURN(rc);
7200 }
7201
7202 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7203 {
7204         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7205         int rc;
7206         ENTRY;
7207
7208         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7209                 struct lfsck_start_param lsp;
7210
7211                 lsp.lsp_start = NULL;
7212                 lsp.lsp_index_valid = 0;
7213                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7214                                                            OBD_IOC_START_LFSCK,
7215                                                            0, &lsp);
7216                 if (rc != 0 && rc != -EALREADY)
7217                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7218                               mdt_obd_name(mdt), rc);
7219         }
7220
7221         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7222         RETURN(rc);
7223 }
7224
7225 static int mdt_obd_postrecov(struct obd_device *obd)
7226 {
7227         struct lu_env env;
7228         int rc;
7229
7230         rc = lu_env_init(&env, LCT_MD_THREAD);
7231         if (rc)
7232                 RETURN(rc);
7233         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7234         lu_env_fini(&env);
7235         return rc;
7236 }
7237
7238 static const struct obd_ops mdt_obd_device_ops = {
7239         .o_owner          = THIS_MODULE,
7240         .o_set_info_async = mdt_obd_set_info_async,
7241         .o_connect        = mdt_obd_connect,
7242         .o_reconnect      = mdt_obd_reconnect,
7243         .o_disconnect     = mdt_obd_disconnect,
7244         .o_init_export    = mdt_init_export,
7245         .o_destroy_export = mdt_destroy_export,
7246         .o_iocontrol      = mdt_iocontrol,
7247         .o_postrecov      = mdt_obd_postrecov,
7248         /* Data-on-MDT IO methods */
7249         .o_preprw         = mdt_obd_preprw,
7250         .o_commitrw       = mdt_obd_commitrw,
7251 };
7252
7253 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7254                                          struct lu_device *d)
7255 {
7256         struct mdt_device *m = mdt_dev(d);
7257         ENTRY;
7258
7259         mdt_fini(env, m);
7260         RETURN(NULL);
7261 }
7262
7263 static struct lu_device *mdt_device_free(const struct lu_env *env,
7264                                          struct lu_device *d)
7265 {
7266         struct mdt_device *m = mdt_dev(d);
7267         ENTRY;
7268
7269         lu_device_fini(&m->mdt_lu_dev);
7270         OBD_FREE_PTR(m);
7271
7272         RETURN(NULL);
7273 }
7274
7275 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7276                                           struct lu_device_type *t,
7277                                           struct lustre_cfg *cfg)
7278 {
7279         struct lu_device  *l;
7280         struct mdt_device *m;
7281
7282         OBD_ALLOC_PTR(m);
7283         if (m != NULL) {
7284                 int rc;
7285
7286                 l = &m->mdt_lu_dev;
7287                 rc = mdt_init0(env, m, t, cfg);
7288                 if (rc != 0) {
7289                         mdt_device_free(env, l);
7290                         l = ERR_PTR(rc);
7291                         return l;
7292                 }
7293         } else
7294                 l = ERR_PTR(-ENOMEM);
7295         return l;
7296 }
7297
7298 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7299 LU_KEY_INIT(mdt, struct mdt_thread_info);
7300
7301 static void mdt_key_fini(const struct lu_context *ctx,
7302                          struct lu_context_key *key, void* data)
7303 {
7304         struct mdt_thread_info *info = data;
7305
7306         if (info->mti_big_lmm) {
7307                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7308                 info->mti_big_lmm = NULL;
7309                 info->mti_big_lmmsize = 0;
7310         }
7311
7312         if (info->mti_big_acl) {
7313                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7314                 info->mti_big_acl = NULL;
7315                 info->mti_big_aclsize = 0;
7316         }
7317
7318         OBD_FREE_PTR(info);
7319 }
7320
7321 /* context key: mdt_thread_key */
7322 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7323
7324 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7325 {
7326         return lu_ucred(info->mti_env);
7327 }
7328
7329 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7330 {
7331         return lu_ucred_check(info->mti_env);
7332 }
7333
7334 /**
7335  * Enable/disable COS (Commit On Sharing).
7336  *
7337  * Set/Clear the COS flag in mdt options.
7338  *
7339  * \param mdt mdt device
7340  * \param val 0 disables COS, other values enable COS
7341  */
7342 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7343 {
7344         struct lu_env env;
7345         int rc;
7346
7347         mdt->mdt_opts.mo_cos = val;
7348         rc = lu_env_init(&env, LCT_LOCAL);
7349         if (unlikely(rc != 0)) {
7350                 CWARN("%s: lu_env initialization failed, cannot "
7351                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7352                 return;
7353         }
7354         mdt_device_sync(&env, mdt);
7355         lu_env_fini(&env);
7356 }
7357
7358 /**
7359  * Check COS (Commit On Sharing) status.
7360  *
7361  * Return COS flag status.
7362  *
7363  * \param mdt mdt device
7364  */
7365 int mdt_cos_is_enabled(struct mdt_device *mdt)
7366 {
7367         return mdt->mdt_opts.mo_cos != 0;
7368 }
7369
7370 static const struct lu_device_type_operations mdt_device_type_ops = {
7371         .ldto_device_alloc = mdt_device_alloc,
7372         .ldto_device_free  = mdt_device_free,
7373         .ldto_device_fini  = mdt_device_fini
7374 };
7375
7376 static struct lu_device_type mdt_device_type = {
7377         .ldt_tags     = LU_DEVICE_MD,
7378         .ldt_name     = LUSTRE_MDT_NAME,
7379         .ldt_ops      = &mdt_device_type_ops,
7380         .ldt_ctx_tags = LCT_MD_THREAD
7381 };
7382
7383 static int __init mdt_init(void)
7384 {
7385         int rc;
7386
7387         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7388                      FID_NOBRACE_LEN + 1);
7389         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7390                      FID_LEN + 1);
7391         rc = lu_kmem_init(mdt_caches);
7392         if (rc)
7393                 return rc;
7394
7395         rc = mds_mod_init();
7396         if (rc)
7397                 GOTO(lu_fini, rc);
7398
7399         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
7400                                  LUSTRE_MDT_NAME, &mdt_device_type);
7401         if (rc)
7402                 GOTO(mds_fini, rc);
7403 lu_fini:
7404         if (rc)
7405                 lu_kmem_fini(mdt_caches);
7406 mds_fini:
7407         if (rc)
7408                 mds_mod_exit();
7409         return rc;
7410 }
7411
7412 static void __exit mdt_exit(void)
7413 {
7414         class_unregister_type(LUSTRE_MDT_NAME);
7415         mds_mod_exit();
7416         lu_kmem_fini(mdt_caches);
7417 }
7418
7419 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7420 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7421 MODULE_VERSION(LUSTRE_VERSION_STRING);
7422 MODULE_LICENSE("GPL");
7423
7424 module_init(mdt_init);
7425 module_exit(mdt_exit);