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 #include <lustre_crypto.h>
  67
  68 #include "mdt_internal.h"
  69
  70 #if OBD_OCD_VERSION(3, 0, 53, 0) > LUSTRE_VERSION_CODE
  71 static int mdt_max_mod_rpcs_per_client_set(const char *val,
  72                                 cfs_kernel_param_arg_t *kp)
  73 {
  74         unsigned int num;
  75         int rc;
  76
  77         rc = kstrtouint(val, 0, &num);
  78         if (rc < 0)
  79                 return rc;
  80
  81         if (num < 1 || num > OBD_MAX_RIF_MAX)
  82                 return -EINVAL;
  83
  84         CWARN("max_mod_rpcs_per_client is deprecated, set mdt.*.max_mod_rpcs_in_flight parameter instead\n");
  85
  86         max_mod_rpcs_per_client = num;
  87         return 0;
  88 }
  89 static const struct kernel_param_ops
  90                                 param_ops_max_mod_rpcs_per_client = {
  91         .set = mdt_max_mod_rpcs_per_client_set,
  92         .get = param_get_uint,
  93 };
  94
  95 #define param_check_max_mod_rpcs_per_client(name, p) \
  96                 __param_check(name, p, unsigned int)
  97
  98 module_param_cb(max_mod_rpcs_per_client,
  99                                 &param_ops_max_mod_rpcs_per_client,
 100                                 &max_mod_rpcs_per_client, 0644);
 101
 102 MODULE_PARM_DESC(max_mod_rpcs_per_client,
 103         "maximum number of modify RPCs in flight allowed per client (Deprecated)");
 104 #endif
 105
 106 mdl_mode_t mdt_mdl_lock_modes[] = {
 107         [LCK_MINMODE] = MDL_MINMODE,
 108         [LCK_EX]      = MDL_EX,
 109         [LCK_PW]      = MDL_PW,
 110         [LCK_PR]      = MDL_PR,
 111         [LCK_CW]      = MDL_CW,
 112         [LCK_CR]      = MDL_CR,
 113         [LCK_NL]      = MDL_NL,
 114         [LCK_GROUP]   = MDL_GROUP
 115 };
 116
 117 enum ldlm_mode mdt_dlm_lock_modes[] = {
 118         [MDL_MINMODE]   = LCK_MINMODE,
 119         [MDL_EX]        = LCK_EX,
 120         [MDL_PW]        = LCK_PW,
 121         [MDL_PR]        = LCK_PR,
 122         [MDL_CW]        = LCK_CW,
 123         [MDL_CR]        = LCK_CR,
 124         [MDL_NL]        = LCK_NL,
 125         [MDL_GROUP]     = LCK_GROUP
 126 };
 127
 128 static struct mdt_device *mdt_dev(struct lu_device *d);
 129
 130 static const struct lu_object_operations mdt_obj_ops;
 131
 132 /* Slab for MDT object allocation */
 133 static struct kmem_cache *mdt_object_kmem;
 134
 135 /* For HSM restore handles */
 136 struct kmem_cache *mdt_hsm_cdt_kmem;
 137
 138 /* For HSM request handles */
 139 struct kmem_cache *mdt_hsm_car_kmem;
 140
 141 static struct lu_kmem_descr mdt_caches[] = {
 142         {
 143                 .ckd_cache = &mdt_object_kmem,
 144                 .ckd_name  = "mdt_obj",
 145                 .ckd_size  = sizeof(struct mdt_object)
 146         },
 147         {
 148                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 149                 .ckd_name       = "mdt_cdt_restore_handle",
 150                 .ckd_size       = sizeof(struct cdt_restore_handle)
 151         },
 152         {
 153                 .ckd_cache      = &mdt_hsm_car_kmem,
 154                 .ckd_name       = "mdt_cdt_agent_req",
 155                 .ckd_size       = sizeof(struct cdt_agent_req)
 156         },
 157         {
 158                 .ckd_cache = NULL
 159         }
 160 };
 161
 162 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 163 {
 164         if (!rep)
 165                 return 0;
 166         return rep->lock_policy_res1 & op_flag;
 167 }
 168
 169 void mdt_clear_disposition(struct mdt_thread_info *info,
 170                            struct ldlm_reply *rep, __u64 op_flag)
 171 {
 172         if (info) {
 173                 info->mti_opdata &= ~op_flag;
 174                 tgt_opdata_clear(info->mti_env, op_flag);
 175         }
 176         if (rep)
 177                 rep->lock_policy_res1 &= ~op_flag;
 178 }
 179
 180 void mdt_set_disposition(struct mdt_thread_info *info,
 181                          struct ldlm_reply *rep, __u64 op_flag)
 182 {
 183         if (info) {
 184                 info->mti_opdata |= op_flag;
 185                 tgt_opdata_set(info->mti_env, op_flag);
 186         }
 187         if (rep)
 188                 rep->lock_policy_res1 |= op_flag;
 189 }
 190
 191 /* assert lock is unlocked before reuse */
 192 static inline void mdt_lock_handle_assert(struct mdt_lock_handle *lh)
 193 {
 194         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
 195         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
 196         LASSERT(!lustre_handle_is_used(&lh->mlh_rreg_lh));
 197 }
 198
 199 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 200 {
 201         mdt_lock_handle_assert(lh);
 202         lh->mlh_pdo_hash = 0;
 203         lh->mlh_reg_mode = lm;
 204         lh->mlh_rreg_mode = lm;
 205         lh->mlh_type = MDT_REG_LOCK;
 206 }
 207
 208 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock)
 209 {
 210         mdt_lock_reg_init(lh, lock->l_req_mode);
 211         if (lock->l_req_mode == LCK_GROUP)
 212                 lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid;
 213 }
 214
 215 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 216                        const struct lu_name *lname)
 217 {
 218         mdt_lock_handle_assert(lh);
 219         lh->mlh_reg_mode = lock_mode;
 220         lh->mlh_pdo_mode = LCK_MINMODE;
 221         lh->mlh_rreg_mode = lock_mode;
 222         lh->mlh_type = MDT_PDO_LOCK;
 223
 224         if (lu_name_is_valid(lname)) {
 225                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 226                                                      lname->ln_namelen);
 227                 /* XXX Workaround for LU-2856
 228                  *
 229                  * Zero is a valid return value of full_name_hash, but
 230                  * several users of mlh_pdo_hash assume a non-zero
 231                  * hash value. We therefore map zero onto an
 232                  * arbitrary, but consistent value (1) to avoid
 233                  * problems further down the road. */
 234                 if (unlikely(lh->mlh_pdo_hash == 0))
 235                         lh->mlh_pdo_hash = 1;
 236         } else {
 237                 lh->mlh_pdo_hash = 0;
 238         }
 239 }
 240
 241 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 242                               struct mdt_lock_handle *lh)
 243 {
 244         mdl_mode_t mode;
 245         ENTRY;
 246
 247         /*
 248          * Any dir access needs couple of locks:
 249          *
 250          * 1) on part of dir we gonna take lookup/modify;
 251          *
 252          * 2) on whole dir to protect it from concurrent splitting and/or to
 253          * flush client's cache for readdir().
 254          *
 255          * so, for a given mode and object this routine decides what lock mode
 256          * to use for lock #2:
 257          *
 258          * 1) if caller's gonna lookup in dir then we need to protect dir from
 259          * being splitted only - LCK_CR
 260          *
 261          * 2) if caller's gonna modify dir then we need to protect dir from
 262          * being splitted and to flush cache - LCK_CW
 263          *
 264          * 3) if caller's gonna modify dir and that dir seems ready for
 265          * splitting then we need to protect it from any type of access
 266          * (lookup/modify/split) - LCK_EX --bzzz
 267          */
 268
 269         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 270         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 271
 272         /*
 273          * Ask underlaying level its opinion about preferable PDO lock mode
 274          * having access type passed as regular lock mode:
 275          *
 276          * - MDL_MINMODE means that lower layer does not want to specify lock
 277          * mode;
 278          *
 279          * - MDL_NL means that no PDO lock should be taken. This is used in some
 280          * cases. Say, for non-splittable directories no need to use PDO locks
 281          * at all.
 282          */
 283         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 284                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 285
 286         if (mode != MDL_MINMODE) {
 287                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 288         } else {
 289                 /*
 290                  * Lower layer does not want to specify locking mode. We do it
 291                  * our selves. No special protection is needed, just flush
 292                  * client's cache on modification and allow concurrent
 293                  * mondification.
 294                  */
 295                 switch (lh->mlh_reg_mode) {
 296                 case LCK_EX:
 297                         lh->mlh_pdo_mode = LCK_EX;
 298                         break;
 299                 case LCK_PR:
 300                         lh->mlh_pdo_mode = LCK_CR;
 301                         break;
 302                 case LCK_PW:
 303                         lh->mlh_pdo_mode = LCK_CW;
 304                         break;
 305                 default:
 306                         CERROR("Not expected lock type (0x%x)\n",
 307                                (int)lh->mlh_reg_mode);
 308                         LBUG();
 309                 }
 310         }
 311
 312         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 313         EXIT;
 314 }
 315
 316 /**
 317  * Check whether \a o is directory stripe object.
 318  *
 319  * \param[in]  info     thread environment
 320  * \param[in]  o        MDT object
 321  *
 322  * \retval 1    is directory stripe.
 323  * \retval 0    isn't directory stripe.
 324  * \retval < 1  error code
 325  */
 326 static int mdt_is_dir_stripe(struct mdt_thread_info *info,
 327                                 struct mdt_object *o)
 328 {
 329         struct md_attr *ma = &info->mti_attr;
 330         struct lmv_mds_md_v1 *lmv;
 331         int rc;
 332
 333         rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 334         if (rc < 0)
 335                 return rc;
 336
 337         if (!(ma->ma_valid & MA_LMV))
 338                 return 0;
 339
 340         lmv = &ma->ma_lmv->lmv_md_v1;
 341
 342         if (!lmv_is_sane2(lmv))
 343                 return -EBADF;
 344
 345         if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE)
 346                 return 1;
 347
 348         return 0;
 349 }
 350
 351 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 352                               struct lu_fid *fid)
 353 {
 354         struct mdt_device *mdt = info->mti_mdt;
 355         struct lu_name *lname = &info->mti_name;
 356         const char *start = fileset;
 357         char *filename = info->mti_filename;
 358         struct mdt_object *obj;
 359         int rc = 0;
 360
 361         LASSERT(!info->mti_cross_ref);
 362
 363         /*
 364          * We may want to allow this to mount a completely separate
 365          * fileset from the MDT in the future, but keeping it to
 366          * ROOT/ only for now avoid potential security issues.
 367          */
 368         *fid = mdt->mdt_md_root_fid;
 369
 370         while (rc == 0 && start != NULL && *start != '\0') {
 371                 const char *s1 = start;
 372                 const char *s2;
 373
 374                 while (*++s1 == '/')
 375                         ;
 376                 s2 = s1;
 377                 while (*s2 != '/' && *s2 != '\0')
 378                         s2++;
 379
 380                 if (s2 == s1)
 381                         break;
 382
 383                 start = s2;
 384
 385                 lname->ln_namelen = s2 - s1;
 386                 if (lname->ln_namelen > NAME_MAX) {
 387                         rc = -EINVAL;
 388                         break;
 389                 }
 390
 391                 /* reject .. as a path component */
 392                 if (lname->ln_namelen == 2 && strncmp(s1, "..", 2) == 0) {
 393                         rc = -EINVAL;
 394                         break;
 395                 }
 396
 397                 strncpy(filename, s1, lname->ln_namelen);
 398                 filename[lname->ln_namelen] = '\0';
 399                 lname->ln_name = filename;
 400
 401                 obj = mdt_object_find(info->mti_env, mdt, fid);
 402                 if (IS_ERR(obj)) {
 403                         rc = PTR_ERR(obj);
 404                         break;
 405                 }
 406                 /* Only got the fid of this obj by name */
 407                 fid_zero(fid);
 408                 rc = mdo_lookup(info->mti_env, mdt_object_child(obj), lname,
 409                                 fid, &info->mti_spec);
 410                 if (!rc && !S_ISDIR(lu_object_attr(&obj->mot_obj)))
 411                         rc = -ENOTDIR;
 412                 mdt_object_put(info->mti_env, obj);
 413         }
 414
 415         return rc;
 416 }
 417
 418 static int mdt_get_root(struct tgt_session_info *tsi)
 419 {
 420         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 421         struct mdt_device       *mdt = info->mti_mdt;
 422         struct mdt_body         *repbody;
 423         char                    *fileset = NULL, *buffer = NULL;
 424         int                      rc;
 425         struct obd_export       *exp = info->mti_exp;
 426         char                    *nodemap_fileset;
 427
 428         ENTRY;
 429
 430         rc = mdt_check_ucred(info);
 431         if (rc)
 432                 GOTO(out, rc = err_serious(rc));
 433
 434         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 435                 GOTO(out, rc = err_serious(-ENOMEM));
 436
 437         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 438         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 439                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 440                 if (fileset == NULL)
 441                         GOTO(out, rc = err_serious(-EFAULT));
 442         }
 443
 444         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 445         if (nodemap_fileset && nodemap_fileset[0]) {
 446                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 447                 if (fileset) {
 448                         /* consider fileset from client as a sub-fileset
 449                          * of the nodemap one */
 450                         OBD_ALLOC(buffer, PATH_MAX + 1);
 451                         if (buffer == NULL)
 452                                 GOTO(out, rc = err_serious(-ENOMEM));
 453                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 454                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 455                                 GOTO(out, rc = err_serious(-EINVAL));
 456                         fileset = buffer;
 457                 } else {
 458                         /* enforce fileset as specified in the nodemap */
 459                         fileset = nodemap_fileset;
 460                 }
 461         }
 462
 463         if (fileset) {
 464                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 465                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 466                 if (rc < 0)
 467                         GOTO(out, rc = err_serious(rc));
 468         } else {
 469                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 470         }
 471         exp->exp_root_fid = repbody->mbo_fid1;
 472         repbody->mbo_valid |= OBD_MD_FLID;
 473
 474         EXIT;
 475 out:
 476         mdt_thread_info_fini(info);
 477         if (buffer)
 478                 OBD_FREE(buffer, PATH_MAX+1);
 479         return rc;
 480 }
 481
 482 static int mdt_statfs(struct tgt_session_info *tsi)
 483 {
 484         struct ptlrpc_request *req = tgt_ses_req(tsi);
 485         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 486         struct mdt_device *mdt = info->mti_mdt;
 487         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 488         struct md_device *next = mdt->mdt_child;
 489         struct ptlrpc_service_part *svcpt;
 490         struct obd_statfs *osfs;
 491         struct mdt_body *reqbody = NULL;
 492         struct mdt_statfs_cache *msf;
 493         ktime_t kstart = ktime_get();
 494         int current_blockbits;
 495         int rc;
 496
 497         ENTRY;
 498
 499         svcpt = req->rq_rqbd->rqbd_svcpt;
 500
 501         /* This will trigger a watchdog timeout */
 502         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 503                          (MDT_SERVICE_WATCHDOG_FACTOR *
 504                           at_get(&svcpt->scp_at_estimate)) + 1);
 505
 506         rc = mdt_check_ucred(info);
 507         if (rc)
 508                 GOTO(out, rc = err_serious(rc));
 509
 510         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 511                 GOTO(out, rc = err_serious(-ENOMEM));
 512
 513         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 514         if (!osfs)
 515                 GOTO(out, rc = -EPROTO);
 516
 517         if (mdt_is_sum_statfs_client(req->rq_export) &&
 518                 lustre_packed_msg_size(req->rq_reqmsg) ==
 519                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 520                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 521                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 522                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 523         }
 524
 525         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 526                 msf = &mdt->mdt_sum_osfs;
 527         else
 528                 msf = &mdt->mdt_osfs;
 529
 530         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 531                         /** statfs data is too old, get up-to-date one */
 532                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 533                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 534                                                               next, osfs);
 535                         else
 536                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 537                                                osfs);
 538                         if (rc)
 539                                 GOTO(out, rc);
 540                         spin_lock(&mdt->mdt_lock);
 541                         msf->msf_osfs = *osfs;
 542                         msf->msf_age = ktime_get_seconds();
 543                         spin_unlock(&mdt->mdt_lock);
 544         } else {
 545                         /** use cached statfs data */
 546                         spin_lock(&mdt->mdt_lock);
 547                         *osfs = msf->msf_osfs;
 548                         spin_unlock(&mdt->mdt_lock);
 549         }
 550
 551         /* tgd_blockbit is recordsize bits set during mkfs.
 552          * This once set does not change. However, 'zfs set'
 553          * can be used to change the MDT blocksize. Instead
 554          * of using cached value of 'tgd_blockbit' always
 555          * calculate the blocksize bits which may have
 556          * changed.
 557          */
 558         current_blockbits = fls64(osfs->os_bsize) - 1;
 559
 560         /* at least try to account for cached pages.  its still racy and
 561          * might be under-reporting if clients haven't announced their
 562          * caches with brw recently */
 563         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 564                " pending %llu free %llu avail %llu\n",
 565                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 566                tgd->tgd_tot_pending,
 567                osfs->os_bfree << current_blockbits,
 568                osfs->os_bavail << current_blockbits);
 569
 570         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 571                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 572                                    osfs->os_bsize - 1) >> current_blockbits));
 573
 574         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 575         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 576                "%llu objects: %llu free; state %x\n",
 577                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 578                osfs->os_files, osfs->os_ffree, osfs->os_state);
 579
 580         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 581             current_blockbits > COMPAT_BSIZE_SHIFT) {
 582                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 583                  * should not see a block size > page size, otherwise
 584                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 585                  * block size which is the biggest block size known to work
 586                  * with all client's page size. */
 587                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 588                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 589                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 590                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 591         }
 592         if (rc == 0)
 593                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 594                                  ktime_us_delta(ktime_get(), kstart));
 595 out:
 596         mdt_thread_info_fini(info);
 597         RETURN(rc);
 598 }
 599
 600 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 601 {
 602         struct lov_comp_md_v1 *comp_v1;
 603         struct lov_mds_md *v1;
 604         __u32 off;
 605         __u32 dom_stripesize = 0;
 606         int i;
 607         bool has_ost_stripes = false;
 608
 609         ENTRY;
 610
 611         if (is_dom_only)
 612                 *is_dom_only = 0;
 613
 614         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 615                 RETURN(0);
 616
 617         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 618         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 619         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 620
 621         /* Fast check for DoM entry with no mirroring, should be the first */
 622         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 623             !(lov_pattern(le32_to_cpu(v1->lmm_pattern)) & LOV_PATTERN_MDT))
 624                 RETURN(0);
 625
 626         /* check all entries otherwise */
 627         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 628                 struct lov_comp_md_entry_v1 *lcme;
 629
 630                 lcme = &comp_v1->lcm_entries[i];
 631                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 632                         continue;
 633
 634                 off = le32_to_cpu(lcme->lcme_offset);
 635                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 636
 637                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) &
 638                     LOV_PATTERN_MDT)
 639                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 640                 else
 641                         has_ost_stripes = true;
 642
 643                 if (dom_stripesize && has_ost_stripes)
 644                         RETURN(dom_stripesize);
 645         }
 646         /* DoM-only case exits here */
 647         if (is_dom_only && dom_stripesize)
 648                 *is_dom_only = 1;
 649         RETURN(dom_stripesize);
 650 }
 651
 652 /**
 653  * Pack size attributes into the reply.
 654  */
 655 int mdt_pack_size2body(struct mdt_thread_info *info,
 656                         const struct lu_fid *fid, struct lustre_handle *lh)
 657 {
 658         struct mdt_body *b;
 659         struct md_attr *ma = &info->mti_attr;
 660         __u32 dom_stripe;
 661         bool dom_lock = false;
 662
 663         ENTRY;
 664
 665         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 666
 667         if (!S_ISREG(ma->ma_attr.la_mode) ||
 668             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 669                 RETURN(-ENODATA);
 670
 671         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 672         /* no DoM stripe, no size in reply */
 673         if (!dom_stripe)
 674                 RETURN(-ENOENT);
 675
 676         if (lustre_handle_is_used(lh)) {
 677                 struct ldlm_lock *lock;
 678
 679                 lock = ldlm_handle2lock(lh);
 680                 if (lock != NULL) {
 681                         dom_lock = ldlm_has_dom(lock);
 682                         LDLM_LOCK_PUT(lock);
 683                 }
 684         }
 685
 686         /* no DoM lock, no size in reply */
 687         if (!dom_lock)
 688                 RETURN(0);
 689
 690         /* Either DoM lock exists or LMM has only DoM stripe then
 691          * return size on body. */
 692         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 693
 694         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 695         RETURN(0);
 696 }
 697
 698 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 699 /*
 700  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 701  *
 702  * \param       info    thread info object
 703  * \param       repbody reply to pack ACLs into
 704  * \param       o       mdt object of file to examine
 705  * \param       nodemap nodemap of client to reply to
 706  * \retval      0       success
 707  * \retval      -errno  error getting or parsing ACL from disk
 708  */
 709 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 710                       struct mdt_object *o, struct lu_nodemap *nodemap)
 711 {
 712         const struct lu_env     *env = info->mti_env;
 713         struct md_object        *next = mdt_object_child(o);
 714         struct lu_buf           *buf = &info->mti_buf;
 715         struct mdt_device       *mdt = info->mti_mdt;
 716         struct req_capsule *pill = info->mti_pill;
 717         int rc;
 718
 719         ENTRY;
 720
 721         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 722         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 723         if (buf->lb_len == 0)
 724                 RETURN(0);
 725
 726         LASSERT(!info->mti_big_acl_used);
 727 again:
 728         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 729         if (rc < 0) {
 730                 if (rc == -ENODATA) {
 731                         repbody->mbo_aclsize = 0;
 732                         repbody->mbo_valid |= OBD_MD_FLACL;
 733                         rc = 0;
 734                 } else if (rc == -EOPNOTSUPP) {
 735                         rc = 0;
 736                 } else if (rc == -ERANGE) {
 737                         if (exp_connect_large_acl(info->mti_exp) &&
 738                             !info->mti_big_acl_used) {
 739                                 if (info->mti_big_acl == NULL) {
 740                                         info->mti_big_aclsize =
 741                                                         min_t(unsigned int,
 742                                                               mdt->mdt_max_ea_size,
 743                                                               XATTR_SIZE_MAX);
 744                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 745                                                         info->mti_big_aclsize);
 746                                         if (info->mti_big_acl == NULL) {
 747                                                 info->mti_big_aclsize = 0;
 748                                                 CERROR("%s: unable to grow "
 749                                                        DFID" ACL buffer\n",
 750                                                        mdt_obd_name(mdt),
 751                                                        PFID(mdt_object_fid(o)));
 752                                                 RETURN(-ENOMEM);
 753                                         }
 754                                 }
 755
 756                                 CDEBUG(D_INODE, "%s: grow the "DFID
 757                                        " ACL buffer to size %d\n",
 758                                        mdt_obd_name(mdt),
 759                                        PFID(mdt_object_fid(o)),
 760                                        info->mti_big_aclsize);
 761
 762                                 buf->lb_buf = info->mti_big_acl;
 763                                 buf->lb_len = info->mti_big_aclsize;
 764                                 info->mti_big_acl_used = 1;
 765                                 goto again;
 766                         }
 767                         /* FS has ACL bigger that our limits */
 768                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
 769                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
 770                                info->mti_big_aclsize);
 771                         rc = -E2BIG;
 772                 } else {
 773                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 774                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 775                 }
 776         } else {
 777                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 778                                      rc, NODEMAP_FS_TO_CLIENT);
 779                 /* if all ACLs mapped out, rc is still >= 0 */
 780                 if (rc < 0) {
 781                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 782                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 783                                PFID(mdt_object_fid(o)), rc);
 784                         repbody->mbo_aclsize = 0;
 785                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 786                 } else {
 787                         repbody->mbo_aclsize = rc;
 788                         repbody->mbo_valid |= OBD_MD_FLACL;
 789                         rc = 0;
 790                 }
 791         }
 792
 793         RETURN(rc);
 794 }
 795 #endif
 796
 797 /* XXX Look into layout in MDT layer. */
 798 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 799 {
 800         struct lov_comp_md_v1   *comp_v1;
 801         struct lov_mds_md       *v1;
 802         int                      i;
 803
 804         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 805                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 806
 807                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 808                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 809                                 comp_v1->lcm_entries[i].lcme_offset);
 810                         /* We don't support partial release for now */
 811                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 812                                 return false;
 813                 }
 814                 return true;
 815         } else {
 816                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 817                         true : false;
 818         }
 819 }
 820
 821 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 822                         const struct lu_attr *attr, const struct lu_fid *fid)
 823 {
 824         struct mdt_device *mdt = info->mti_mdt;
 825         struct obd_export *exp = info->mti_exp;
 826         struct md_attr *ma = &info->mti_attr;
 827         struct lu_nodemap *nodemap = NULL;
 828
 829         LASSERT(ma->ma_valid & MA_INODE);
 830
 831         if (attr->la_valid & LA_ATIME) {
 832                 b->mbo_atime = attr->la_atime;
 833                 b->mbo_valid |= OBD_MD_FLATIME;
 834         }
 835         if (attr->la_valid & LA_MTIME) {
 836                 b->mbo_mtime = attr->la_mtime;
 837                 b->mbo_valid |= OBD_MD_FLMTIME;
 838         }
 839         if (attr->la_valid & LA_CTIME) {
 840                 b->mbo_ctime = attr->la_ctime;
 841                 b->mbo_valid |= OBD_MD_FLCTIME;
 842         }
 843         if (attr->la_valid & LA_BTIME) {
 844                 b->mbo_btime = attr->la_btime;
 845                 b->mbo_valid |= OBD_MD_FLBTIME;
 846         }
 847         if (attr->la_valid & LA_FLAGS) {
 848                 b->mbo_flags = attr->la_flags;
 849                 b->mbo_valid |= OBD_MD_FLFLAGS;
 850         }
 851         if (attr->la_valid & LA_NLINK) {
 852                 b->mbo_nlink = attr->la_nlink;
 853                 b->mbo_valid |= OBD_MD_FLNLINK;
 854         }
 855         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
 856                 nodemap = nodemap_get_from_exp(exp);
 857                 if (IS_ERR(nodemap))
 858                         goto out;
 859         }
 860         if (attr->la_valid & LA_UID) {
 861                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 862                                             NODEMAP_FS_TO_CLIENT,
 863                                             attr->la_uid);
 864                 b->mbo_valid |= OBD_MD_FLUID;
 865         }
 866         if (attr->la_valid & LA_GID) {
 867                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 868                                             NODEMAP_FS_TO_CLIENT,
 869                                             attr->la_gid);
 870                 b->mbo_valid |= OBD_MD_FLGID;
 871         }
 872
 873         if (attr->la_valid & LA_PROJID) {
 874                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
 875                                                NODEMAP_FS_TO_CLIENT,
 876                                                attr->la_projid);
 877                 b->mbo_valid |= OBD_MD_FLPROJID;
 878         }
 879
 880         b->mbo_mode = attr->la_mode;
 881         if (attr->la_valid & LA_MODE)
 882                 b->mbo_valid |= OBD_MD_FLMODE;
 883         if (attr->la_valid & LA_TYPE)
 884                 b->mbo_valid |= OBD_MD_FLTYPE;
 885
 886         if (fid != NULL) {
 887                 b->mbo_fid1 = *fid;
 888                 b->mbo_valid |= OBD_MD_FLID;
 889                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 890                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 891         }
 892
 893         if (!(attr->la_valid & LA_TYPE))
 894                 return;
 895
 896         b->mbo_rdev   = attr->la_rdev;
 897         b->mbo_size   = attr->la_size;
 898         b->mbo_blocks = attr->la_blocks;
 899
 900         if (!S_ISREG(attr->la_mode)) {
 901                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 902         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 903                 /* means no objects are allocated on osts. */
 904                 LASSERT(!(ma->ma_valid & MA_LOV));
 905                 /* just ignore blocks occupied by extend attributes on MDS */
 906                 b->mbo_blocks = 0;
 907                 /* if no object is allocated on osts, the size on mds is valid.
 908                  * b=22272 */
 909                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 910         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 911                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 912                         /* A released file stores its size on MDS. */
 913                         /* But return 1 block for released file, unless tools
 914                          * like tar will consider it fully sparse. (LU-3864)
 915                          */
 916                         if (unlikely(b->mbo_size == 0))
 917                                 b->mbo_blocks = 0;
 918                         else
 919                                 b->mbo_blocks = 1;
 920                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 921                 } else if (info->mti_som_strict &&
 922                            mdt->mdt_opts.mo_enable_strict_som) {
 923                         /* use SOM for size*/
 924                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 925                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 926                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 927                         b->mbo_size = ma->ma_som.ms_size;
 928                         b->mbo_blocks = ma->ma_som.ms_blocks;
 929                 }
 930         }
 931
 932         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 933                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 934                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 935                        PFID(fid), (unsigned long long)b->mbo_size);
 936
 937 out:
 938         if (!IS_ERR_OR_NULL(nodemap))
 939                 nodemap_putref(nodemap);
 940 }
 941
 942 static inline int mdt_body_has_lov(const struct lu_attr *la,
 943                                    const struct mdt_body *body)
 944 {
 945         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 946                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 947 }
 948
 949 void mdt_client_compatibility(struct mdt_thread_info *info)
 950 {
 951         struct mdt_body       *body;
 952         struct ptlrpc_request *req = mdt_info_req(info);
 953         struct obd_export     *exp = req->rq_export;
 954         struct md_attr        *ma = &info->mti_attr;
 955         struct lu_attr        *la = &ma->ma_attr;
 956         ENTRY;
 957
 958         if (exp_connect_layout(exp))
 959                 /* the client can deal with 16-bit lmm_stripe_count */
 960                 RETURN_EXIT;
 961
 962         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 963
 964         if (!mdt_body_has_lov(la, body))
 965                 RETURN_EXIT;
 966
 967         /* now we have a reply with a lov for a client not compatible with the
 968          * layout lock so we have to clean the layout generation number */
 969         if (S_ISREG(la->la_mode))
 970                 ma->ma_lmm->lmm_layout_gen = 0;
 971         EXIT;
 972 }
 973
 974 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 975                                    struct mdt_object *o)
 976 {
 977         const struct lu_env *env = info->mti_env;
 978         int rc, rc2;
 979
 980         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 981                           XATTR_NAME_LOV);
 982
 983         if (rc == -ENODATA)
 984                 rc = 0;
 985
 986         if (rc < 0)
 987                 goto out;
 988
 989         /* Is it a directory? Let's check for the LMV as well */
 990         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 991                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 992                                    XATTR_NAME_LMV);
 993
 994                 if (rc2 == -ENODATA)
 995                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 996                                            &LU_BUF_NULL,
 997                                            XATTR_NAME_DEFAULT_LMV);
 998
 999                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
1000                         rc = rc2;
1001         }
1002
1003 out:
1004         return rc;
1005 }
1006
1007 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
1008                       const char *name)
1009 {
1010         const struct lu_env *env = info->mti_env;
1011         int rc;
1012         ENTRY;
1013
1014         LASSERT(info->mti_big_lmm_used == 0);
1015         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
1016         if (rc < 0)
1017                 RETURN(rc);
1018
1019         /* big_lmm may need to be grown */
1020         if (info->mti_big_lmmsize < rc) {
1021                 int size = size_roundup_power2(rc);
1022
1023                 if (info->mti_big_lmmsize > 0) {
1024                         /* free old buffer */
1025                         LASSERT(info->mti_big_lmm);
1026                         OBD_FREE_LARGE(info->mti_big_lmm,
1027                                        info->mti_big_lmmsize);
1028                         info->mti_big_lmm = NULL;
1029                         info->mti_big_lmmsize = 0;
1030                 }
1031
1032                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1033                 if (info->mti_big_lmm == NULL)
1034                         RETURN(-ENOMEM);
1035                 info->mti_big_lmmsize = size;
1036         }
1037         LASSERT(info->mti_big_lmmsize >= rc);
1038
1039         info->mti_buf.lb_buf = info->mti_big_lmm;
1040         info->mti_buf.lb_len = info->mti_big_lmmsize;
1041         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1042
1043         RETURN(rc);
1044 }
1045
1046 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1047                      struct md_attr *ma, const char *name)
1048 {
1049         struct md_object *next = mdt_object_child(o);
1050         struct lu_buf    *buf = &info->mti_buf;
1051         int rc;
1052
1053         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1054                 buf->lb_buf = ma->ma_lmm;
1055                 buf->lb_len = ma->ma_lmm_size;
1056                 LASSERT(!(ma->ma_valid & MA_LOV));
1057         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1058                 buf->lb_buf = ma->ma_lmv;
1059                 buf->lb_len = ma->ma_lmv_size;
1060                 LASSERT(!(ma->ma_valid & MA_LMV));
1061         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1062                 buf->lb_buf = ma->ma_default_lmv;
1063                 buf->lb_len = ma->ma_default_lmv_size;
1064                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1065         } else {
1066                 return -EINVAL;
1067         }
1068
1069         LASSERT(buf->lb_buf);
1070
1071         if (!mdt_object_exists(o))
1072                 return -ENOENT;
1073
1074         if (mdt_object_remote(o) && S_ISDIR(lu_object_attr(&o->mot_obj)))
1075                 /* force reload layout for remote dir in case layout changed */
1076                 mo_invalidate(info->mti_env, mdt_object_child(o));
1077
1078         rc = mo_xattr_get(info->mti_env, next, buf, name);
1079         if (rc > 0) {
1080
1081 got:
1082                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1083                         if (info->mti_big_lmm_used)
1084                                 ma->ma_lmm = info->mti_big_lmm;
1085
1086                         /* NOT return LOV EA with hole to old client. */
1087                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1088                                      LOV_PATTERN_F_HOLE) &&
1089                             !(exp_connect_flags(info->mti_exp) &
1090                               OBD_CONNECT_LFSCK)) {
1091                                 return -EIO;
1092                         } else {
1093                                 ma->ma_lmm_size = rc;
1094                                 ma->ma_valid |= MA_LOV;
1095                         }
1096                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1097                         if (info->mti_big_lmm_used)
1098                                 ma->ma_lmv = info->mti_big_lmm;
1099
1100                         ma->ma_lmv_size = rc;
1101                         ma->ma_valid |= MA_LMV;
1102                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1103                         ma->ma_default_lmv_size = rc;
1104                         ma->ma_valid |= MA_LMV_DEF;
1105                 }
1106
1107                 /* Update mdt_max_mdsize so all clients will be aware that */
1108                 if (info->mti_mdt->mdt_max_mdsize < rc)
1109                         info->mti_mdt->mdt_max_mdsize = rc;
1110
1111                 rc = 0;
1112         } else if (rc == -ENODATA) {
1113                 /* no LOV EA */
1114                 rc = 0;
1115         } else if (rc == -ERANGE) {
1116                 /* Default LMV has fixed size, so it must be able to fit
1117                  * in the original buffer */
1118                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1119                         return rc;
1120                 rc = mdt_big_xattr_get(info, o, name);
1121                 if (rc > 0) {
1122                         info->mti_big_lmm_used = 1;
1123                         goto got;
1124                 }
1125         }
1126
1127         return rc;
1128 }
1129
1130 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1131                    struct md_attr *ma, const char *name)
1132 {
1133         int rc;
1134
1135         if (!info->mti_big_lmm) {
1136                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1137                 if (!info->mti_big_lmm)
1138                         return -ENOMEM;
1139                 info->mti_big_lmmsize = PAGE_SIZE;
1140         }
1141
1142         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1143                 ma->ma_lmm = info->mti_big_lmm;
1144                 ma->ma_lmm_size = info->mti_big_lmmsize;
1145                 ma->ma_valid &= ~MA_LOV;
1146         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1147                 ma->ma_lmv = info->mti_big_lmm;
1148                 ma->ma_lmv_size = info->mti_big_lmmsize;
1149                 ma->ma_valid &= ~MA_LMV;
1150         } else {
1151                 LBUG();
1152         }
1153
1154         LASSERT(!info->mti_big_lmm_used);
1155         rc = __mdt_stripe_get(info, o, ma, name);
1156         /* since big_lmm is always used here, clear 'used' flag to avoid
1157          * assertion in mdt_big_xattr_get().
1158          */
1159         info->mti_big_lmm_used = 0;
1160
1161         return rc;
1162 }
1163
1164 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1165                       struct lu_fid *pfid)
1166 {
1167         struct lu_buf           *buf = &info->mti_buf;
1168         struct link_ea_header   *leh;
1169         struct link_ea_entry    *lee;
1170         int                      rc;
1171         ENTRY;
1172
1173         buf->lb_buf = info->mti_big_lmm;
1174         buf->lb_len = info->mti_big_lmmsize;
1175         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1176                           buf, XATTR_NAME_LINK);
1177         /* ignore errors, MA_PFID won't be set and it is
1178          * up to the caller to treat this as an error */
1179         if (rc == -ERANGE || buf->lb_len == 0) {
1180                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1181                 buf->lb_buf = info->mti_big_lmm;
1182                 buf->lb_len = info->mti_big_lmmsize;
1183         }
1184
1185         if (rc < 0)
1186                 RETURN(rc);
1187         if (rc < sizeof(*leh)) {
1188                 CERROR("short LinkEA on "DFID": rc = %d\n",
1189                        PFID(mdt_object_fid(o)), rc);
1190                 RETURN(-ENODATA);
1191         }
1192
1193         leh = (struct link_ea_header *) buf->lb_buf;
1194         lee = (struct link_ea_entry *)(leh + 1);
1195         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1196                 leh->leh_magic = LINK_EA_MAGIC;
1197                 leh->leh_reccount = __swab32(leh->leh_reccount);
1198                 leh->leh_len = __swab64(leh->leh_len);
1199         }
1200         if (leh->leh_magic != LINK_EA_MAGIC)
1201                 RETURN(-EINVAL);
1202         if (leh->leh_reccount == 0)
1203                 RETURN(-ENODATA);
1204
1205         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1206         fid_be_to_cpu(pfid, pfid);
1207
1208         RETURN(0);
1209 }
1210
1211 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1212                            struct lu_fid *pfid, struct lu_name *lname)
1213 {
1214         struct lu_buf *buf = &info->mti_buf;
1215         struct link_ea_header *leh;
1216         struct link_ea_entry *lee;
1217         int reclen;
1218         int rc;
1219
1220         buf->lb_buf = info->mti_xattr_buf;
1221         buf->lb_len = sizeof(info->mti_xattr_buf);
1222         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1223                           XATTR_NAME_LINK);
1224         if (rc == -ERANGE) {
1225                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1226                 buf->lb_buf = info->mti_big_lmm;
1227                 buf->lb_len = info->mti_big_lmmsize;
1228         }
1229         if (rc < 0)
1230                 return rc;
1231
1232         if (rc < sizeof(*leh)) {
1233                 CERROR("short LinkEA on "DFID": rc = %d\n",
1234                        PFID(mdt_object_fid(o)), rc);
1235                 return -ENODATA;
1236         }
1237
1238         leh = (struct link_ea_header *)buf->lb_buf;
1239         lee = (struct link_ea_entry *)(leh + 1);
1240         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1241                 leh->leh_magic = LINK_EA_MAGIC;
1242                 leh->leh_reccount = __swab32(leh->leh_reccount);
1243                 leh->leh_len = __swab64(leh->leh_len);
1244         }
1245         if (leh->leh_magic != LINK_EA_MAGIC)
1246                 return -EINVAL;
1247
1248         if (leh->leh_reccount == 0)
1249                 return -ENODATA;
1250
1251         linkea_entry_unpack(lee, &reclen, lname, pfid);
1252
1253         return 0;
1254 }
1255
1256 int mdt_attr_get_complex(struct mdt_thread_info *info,
1257                          struct mdt_object *o, struct md_attr *ma)
1258 {
1259         const struct lu_env *env = info->mti_env;
1260         struct md_object    *next = mdt_object_child(o);
1261         struct lu_buf       *buf = &info->mti_buf;
1262         int                  need = ma->ma_need;
1263         int                  rc = 0, rc2;
1264         u32                  mode;
1265         ENTRY;
1266
1267         ma->ma_valid = 0;
1268
1269         if (mdt_object_exists(o) == 0)
1270                 GOTO(out, rc = -ENOENT);
1271         mode = lu_object_attr(&next->mo_lu);
1272
1273         if (need & MA_INODE) {
1274                 ma->ma_need = MA_INODE;
1275                 rc = mo_attr_get(env, next, ma);
1276                 if (rc)
1277                         GOTO(out, rc);
1278
1279                 if (S_ISREG(mode))
1280                         (void) mdt_get_som(info, o, ma);
1281                 ma->ma_valid |= MA_INODE;
1282         }
1283
1284         if (need & MA_PFID) {
1285                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1286                 if (rc == 0)
1287                         ma->ma_valid |= MA_PFID;
1288                 /* ignore this error, parent fid is not mandatory */
1289                 rc = 0;
1290         }
1291
1292         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1293                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1294                 if (rc)
1295                         GOTO(out, rc);
1296         }
1297
1298         if (need & MA_LMV && S_ISDIR(mode)) {
1299                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1300                 if (rc != 0)
1301                         GOTO(out, rc);
1302         }
1303
1304         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1305                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1306                 if (rc != 0)
1307                         GOTO(out, rc);
1308         }
1309
1310         /*
1311          * In the handle of MA_INODE, we may already get the SOM attr.
1312          */
1313         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1314                 rc = mdt_get_som(info, o, ma);
1315                 if (rc != 0)
1316                         GOTO(out, rc);
1317         }
1318
1319         if (need & MA_HSM && S_ISREG(mode)) {
1320                 buf->lb_buf = info->mti_xattr_buf;
1321                 buf->lb_len = sizeof(info->mti_xattr_buf);
1322                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1323                              sizeof(info->mti_xattr_buf));
1324                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1325                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1326                 if (rc2 == 0)
1327                         ma->ma_valid |= MA_HSM;
1328                 else if (rc2 < 0 && rc2 != -ENODATA)
1329                         GOTO(out, rc = rc2);
1330         }
1331
1332 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1333         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1334                 buf->lb_buf = ma->ma_acl;
1335                 buf->lb_len = ma->ma_acl_size;
1336                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1337                 if (rc2 > 0) {
1338                         ma->ma_acl_size = rc2;
1339                         ma->ma_valid |= MA_ACL_DEF;
1340                 } else if (rc2 == -ENODATA) {
1341                         /* no ACLs */
1342                         ma->ma_acl_size = 0;
1343                 } else
1344                         GOTO(out, rc = rc2);
1345         }
1346 #endif
1347 out:
1348         ma->ma_need = need;
1349         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1350                rc, ma->ma_valid, ma->ma_lmm);
1351         RETURN(rc);
1352 }
1353
1354 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
1355 {
1356         struct req_capsule *pill = info->mti_pill;
1357
1358         ENTRY;
1359         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
1360                                   RCL_SERVER))
1361                 /* pre-set size in server part with max size */
1362                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
1363                                      RCL_SERVER,
1364                                      info->mti_mdt->mdt_max_mdsize);
1365         EXIT;
1366 }
1367
1368 static int mdt_getattr_internal(struct mdt_thread_info *info,
1369                                 struct mdt_object *o, int ma_need)
1370 {
1371         struct mdt_device *mdt = info->mti_mdt;
1372         struct md_object *next = mdt_object_child(o);
1373         const struct mdt_body *reqbody = info->mti_body;
1374         struct ptlrpc_request *req = mdt_info_req(info);
1375         struct md_attr *ma = &info->mti_attr;
1376         struct lu_attr *la = &ma->ma_attr;
1377         struct req_capsule *pill = info->mti_pill;
1378         const struct lu_env *env = info->mti_env;
1379         struct mdt_body *repbody;
1380         struct lu_buf *buffer = &info->mti_buf;
1381         struct obd_export *exp = info->mti_exp;
1382         ktime_t kstart = ktime_get();
1383         int rc;
1384
1385         ENTRY;
1386
1387         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1388                 RETURN(err_serious(-ENOMEM));
1389
1390         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1391
1392         ma->ma_valid = 0;
1393
1394         if (mdt_object_remote(o)) {
1395                 /* This object is located on remote node.*/
1396                 /* Return -ENOTSUPP for old client */
1397                 if (!mdt_is_dne_client(req->rq_export))
1398                         GOTO(out, rc = -ENOTSUPP);
1399
1400                 repbody->mbo_fid1 = *mdt_object_fid(o);
1401                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1402                 GOTO(out, rc = 0);
1403         }
1404
1405         if (reqbody->mbo_eadatasize > 0) {
1406                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1407                 if (buffer->lb_buf == NULL)
1408                         GOTO(out, rc = -EPROTO);
1409                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1410                                                       RCL_SERVER);
1411         } else {
1412                 buffer->lb_buf = NULL;
1413                 buffer->lb_len = 0;
1414                 ma_need &= ~(MA_LOV | MA_LMV);
1415                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1416                        mdt_obd_name(info->mti_mdt),
1417                        req->rq_export->exp_client_uuid.uuid);
1418         }
1419
1420         /* from 2.12.58 intent_getattr pack default LMV in reply */
1421         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1422             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1423                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1424             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1425                                   RCL_SERVER)) {
1426                 ma->ma_lmv = buffer->lb_buf;
1427                 ma->ma_lmv_size = buffer->lb_len;
1428                 ma->ma_default_lmv = req_capsule_server_get(pill,
1429                                                 &RMF_DEFAULT_MDT_MD);
1430                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1431                                                 &RMF_DEFAULT_MDT_MD,
1432                                                 RCL_SERVER);
1433                 ma->ma_need = MA_INODE;
1434                 if (ma->ma_lmv_size > 0)
1435                         ma->ma_need |= MA_LMV;
1436                 if (ma->ma_default_lmv_size > 0)
1437                         ma->ma_need |= MA_LMV_DEF;
1438         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1439                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1440                 /* If it is dir and client require MEA, then we got MEA */
1441                 /* Assumption: MDT_MD size is enough for lmv size. */
1442                 ma->ma_lmv = buffer->lb_buf;
1443                 ma->ma_lmv_size = buffer->lb_len;
1444                 ma->ma_need = MA_INODE;
1445                 if (ma->ma_lmv_size > 0) {
1446                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1447                                 ma->ma_need |= MA_LMV;
1448                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1449                                 ma->ma_need |= MA_LMV_DEF;
1450                                 ma->ma_default_lmv = buffer->lb_buf;
1451                                 ma->ma_lmv = NULL;
1452                                 ma->ma_default_lmv_size = buffer->lb_len;
1453                                 ma->ma_lmv_size = 0;
1454                         }
1455                 }
1456         } else {
1457                 ma->ma_lmm = buffer->lb_buf;
1458                 ma->ma_lmm_size = buffer->lb_len;
1459                 ma->ma_need = MA_INODE | MA_HSM;
1460                 if (ma->ma_lmm_size > 0) {
1461                         ma->ma_need |= MA_LOV;
1462                         /* Older clients may crash if they getattr overstriped
1463                          * files
1464                          */
1465                         if (!exp_connect_overstriping(exp) &&
1466                             mdt_lmm_is_overstriping(ma->ma_lmm))
1467                                 RETURN(-EOPNOTSUPP);
1468                 }
1469         }
1470
1471         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1472             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1473             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1474                 /* get default stripe info for this dir. */
1475                 ma->ma_need |= MA_LOV_DEF;
1476         }
1477         ma->ma_need |= ma_need;
1478
1479         rc = mdt_attr_get_complex(info, o, ma);
1480         if (unlikely(rc)) {
1481                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1482                              "%s: getattr error for "DFID": rc = %d\n",
1483                              mdt_obd_name(info->mti_mdt),
1484                              PFID(mdt_object_fid(o)), rc);
1485                 RETURN(rc);
1486         }
1487
1488         /* return immutable attr on fscrypt metadata files
1489          * if fscrypt admin is not permitted
1490          */
1491         if (o->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD &&
1492             !mdt_ucred(info)->uc_rbac_fscrypt_admin)
1493                 la->la_flags |= LUSTRE_IMMUTABLE_FL;
1494
1495         /* if file is released, check if a restore is running */
1496         if (ma->ma_valid & MA_HSM) {
1497                 repbody->mbo_valid |= OBD_MD_TSTATE;
1498                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1499                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1500                         repbody->mbo_t_state = MS_RESTORE;
1501         }
1502
1503         if (unlikely(!(ma->ma_valid & MA_INODE)))
1504                 RETURN(-EFAULT);
1505
1506         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1507
1508         if (mdt_body_has_lov(la, reqbody)) {
1509                 u32 stripe_count = 1;
1510                 bool fixed_layout = false;
1511
1512                 if (ma->ma_valid & MA_LOV) {
1513                         LASSERT(ma->ma_lmm_size);
1514                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1515                         if (S_ISDIR(la->la_mode))
1516                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1517                         else
1518                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1519                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1520                 }
1521                 if (ma->ma_valid & MA_LMV) {
1522                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1523                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1524
1525                         /* Return -ENOTSUPP for old client */
1526                         if (!mdt_is_striped_client(req->rq_export))
1527                                 RETURN(-ENOTSUPP);
1528
1529                         LASSERT(S_ISDIR(la->la_mode));
1530                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1531                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1532                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1533
1534                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1535                         fixed_layout = lmv_is_fixed(lmv);
1536                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1537                                 mdt_restripe_migrate_add(info, o);
1538                         else if (magic == LMV_MAGIC_V1 &&
1539                                  lmv_is_restriping(lmv))
1540                                 mdt_restripe_update_add(info, o);
1541                 }
1542                 if (ma->ma_valid & MA_LMV_DEF) {
1543                         /* Return -ENOTSUPP for old client */
1544                         if (!mdt_is_striped_client(req->rq_export))
1545                                 RETURN(-ENOTSUPP);
1546                         LASSERT(S_ISDIR(la->la_mode));
1547                         /*
1548                          * when ll_dir_getstripe() gets default LMV, it
1549                          * checks mbo_eadatasize.
1550                          */
1551                         if (!(ma->ma_valid & MA_LMV))
1552                                 repbody->mbo_eadatasize =
1553                                         ma->ma_default_lmv_size;
1554                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1555                                                OBD_MD_DEFAULT_MEA);
1556                 }
1557                 CDEBUG(D_VFSTRACE,
1558                        "dirent count %llu stripe count %u MDT count %d\n",
1559                        ma->ma_attr.la_dirent_count, stripe_count,
1560                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1561                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1562                     ma->ma_attr.la_dirent_count >
1563                         mdt->mdt_restriper.mdr_dir_split_count &&
1564                     !fid_is_root(mdt_object_fid(o)) &&
1565                     mdt->mdt_enable_dir_auto_split &&
1566                     !o->mot_restriping &&
1567                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1568                     !fixed_layout)
1569                         mdt_auto_split_add(info, o);
1570         } else if (S_ISLNK(la->la_mode) &&
1571                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1572                 buffer->lb_buf = ma->ma_lmm;
1573                 /* eadatasize from client includes NULL-terminator, so
1574                  * there is no need to read it */
1575                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1576                 rc = mo_readlink(env, next, buffer);
1577                 if (unlikely(rc <= 0)) {
1578                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1579                                mdt_obd_name(info->mti_mdt),
1580                                PFID(mdt_object_fid(o)), rc);
1581                         rc = -EFAULT;
1582                 } else {
1583                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1584
1585                         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1586                                 rc -= 2;
1587                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1588                         /* we need to report back size with NULL-terminator
1589                          * because client expects that */
1590                         repbody->mbo_eadatasize = rc + 1;
1591                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1592                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1593                                        "on "DFID ", expected %d\n",
1594                                        mdt_obd_name(info->mti_mdt),
1595                                        rc, PFID(mdt_object_fid(o)),
1596                                        reqbody->mbo_eadatasize - 1);
1597                         /* NULL terminate */
1598                         ((char *)ma->ma_lmm)[rc] = 0;
1599
1600                         /* If the total CDEBUG() size is larger than a page, it
1601                          * will print a warning to the console, avoid this by
1602                          * printing just the last part of the symlink. */
1603                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1604                                print_limit < rc ? "..." : "", print_limit,
1605                                (char *)ma->ma_lmm + rc - print_limit, rc);
1606                         rc = 0;
1607                 }
1608         }
1609
1610         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1611                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1612                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1613                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1614                        repbody->mbo_max_mdsize);
1615         }
1616
1617 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1618         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1619                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1620                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1621                 if (IS_ERR(nodemap))
1622                         RETURN(PTR_ERR(nodemap));
1623
1624                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1625                 nodemap_putref(nodemap);
1626         }
1627 #endif
1628
1629 out:
1630         if (rc == 0)
1631                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1632                                  ktime_us_delta(ktime_get(), kstart));
1633
1634         RETURN(rc);
1635 }
1636
1637 static int mdt_getattr(struct tgt_session_info *tsi)
1638 {
1639         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1640         struct mdt_object       *obj = info->mti_object;
1641         struct req_capsule      *pill = info->mti_pill;
1642         struct mdt_body         *reqbody;
1643         struct mdt_body         *repbody;
1644         int rc, rc2;
1645         ENTRY;
1646
1647         if (unlikely(info->mti_object == NULL))
1648                 RETURN(-EPROTO);
1649
1650         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1651         LASSERT(reqbody);
1652         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1653
1654         /* Special case for Data-on-MDT files to get data version */
1655         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1656                 rc = mdt_data_version_get(tsi);
1657                 GOTO(out, rc);
1658         }
1659
1660         /* Unlike intent case where we need to pre-fill out buffers early on
1661          * in intent policy for ldlm reasons, here we can have a much better
1662          * guess at EA size by just reading it from disk.
1663          * Exceptions are readdir and (missing) directory striping */
1664         /* Readlink */
1665         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1666                 /* No easy way to know how long is the symlink, but it cannot
1667                  * be more than PATH_MAX, so we allocate +1 */
1668                 rc = PATH_MAX + 1;
1669         /* A special case for fs ROOT: getattr there might fetch
1670          * default EA for entire fs, not just for this dir!
1671          */
1672         } else if (lu_fid_eq(mdt_object_fid(obj),
1673                              &info->mti_mdt->mdt_md_root_fid) &&
1674                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1675                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1676                                                                  MDS_GETATTR)) {
1677                 /* Should the default strping be bigger, mdt_fix_reply
1678                  * will reallocate */
1679                 rc = DEF_REP_MD_SIZE;
1680         } else {
1681                 /* Read the actual EA size from disk */
1682                 rc = mdt_attr_get_eabuf_size(info, obj);
1683         }
1684
1685         if (rc < 0)
1686                 GOTO(out, rc = err_serious(rc));
1687
1688         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1689
1690         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1691          * by default. If the target object has more ACL entries, then
1692          * enlarge the buffer when necessary. */
1693         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1694                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1695         mdt_preset_encctx_size(info);
1696
1697         rc = req_capsule_server_pack(pill);
1698         if (unlikely(rc != 0))
1699                 GOTO(out, rc = err_serious(rc));
1700
1701         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1702         LASSERT(repbody != NULL);
1703         repbody->mbo_eadatasize = 0;
1704         repbody->mbo_aclsize = 0;
1705
1706         rc = mdt_check_ucred(info);
1707         if (unlikely(rc))
1708                 GOTO(out_shrink, rc);
1709
1710         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1711
1712         rc = mdt_init_ucred(info, reqbody);
1713         if (rc)
1714                 GOTO(out_shrink, rc);
1715
1716         rc = mdt_getattr_internal(info, obj, 0);
1717         if (unlikely(rc))
1718                 GOTO(out_ucred, rc);
1719
1720         rc = mdt_pack_encctx_in_reply(info, obj);
1721         EXIT;
1722 out_ucred:
1723         mdt_exit_ucred(info);
1724 out_shrink:
1725         mdt_client_compatibility(info);
1726         rc2 = mdt_fix_reply(info);
1727         if (rc == 0)
1728                 rc = rc2;
1729 out:
1730         mdt_thread_info_fini(info);
1731         return rc;
1732 }
1733
1734 /**
1735  * Handler of layout intent RPC requiring the layout modification
1736  *
1737  * \param[in]  info     thread environment
1738  * \param[in]  obj      object
1739  * \param[out] lhc      object ldlm lock handle
1740  * \param[in]  layout   layout change descriptor
1741  *
1742  * \retval 0    on success
1743  * \retval < 0  error code
1744  */
1745 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1746                       struct mdt_lock_handle *lhc,
1747                       struct md_layout_change *layout)
1748 {
1749         int rc;
1750
1751         ENTRY;
1752
1753         if (!mdt_object_exists(obj))
1754                 RETURN(-ENOENT);
1755
1756         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1757                 RETURN(-EINVAL);
1758
1759         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1760                            MAY_WRITE);
1761         if (rc)
1762                 RETURN(rc);
1763
1764         rc = mdt_check_resent_lock(info, obj, lhc);
1765         if (rc < 0)
1766                 RETURN(rc);
1767
1768         if (rc > 0) {
1769                 /* not resent */
1770                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1771
1772                 /* take layout lock to prepare layout change */
1773                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1774                         lockpart |= MDS_INODELOCK_UPDATE;
1775
1776                 rc = mdt_object_lock(info, obj, lhc, lockpart, LCK_EX);
1777                 if (rc)
1778                         RETURN(rc);
1779         }
1780
1781         mutex_lock(&obj->mot_som_mutex);
1782         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1783         mutex_unlock(&obj->mot_som_mutex);
1784
1785         if (rc)
1786                 mdt_object_unlock(info, obj, lhc, 1);
1787
1788         RETURN(rc);
1789 }
1790
1791 /**
1792  * Exchange MOF_LOV_CREATED flags between two objects after a
1793  * layout swap. No assumption is made on whether o1 or o2 have
1794  * created objects or not.
1795  *
1796  * \param[in,out] o1    First swap layout object
1797  * \param[in,out] o2    Second swap layout object
1798  */
1799 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1800 {
1801         unsigned int o1_lov_created = o1->mot_lov_created;
1802
1803         mutex_lock(&o1->mot_lov_mutex);
1804         mutex_lock(&o2->mot_lov_mutex);
1805
1806         o1->mot_lov_created = o2->mot_lov_created;
1807         o2->mot_lov_created = o1_lov_created;
1808
1809         mutex_unlock(&o2->mot_lov_mutex);
1810         mutex_unlock(&o1->mot_lov_mutex);
1811 }
1812
1813 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1814 {
1815         struct mdt_thread_info  *info;
1816         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1817         struct obd_export       *exp = req->rq_export;
1818         struct mdt_object       *o1, *o2, *o;
1819         struct mdt_lock_handle  *lh1, *lh2;
1820         struct mdc_swap_layouts *msl;
1821         int                      rc;
1822         ENTRY;
1823
1824         /* client does not support layout lock, so layout swaping
1825          * is disabled.
1826          * FIXME: there is a problem for old clients which don't support
1827          * layout lock yet. If those clients have already opened the file
1828          * they won't be notified at all so that old layout may still be
1829          * used to do IO. This can be fixed after file release is landed by
1830          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1831         if (!exp_connect_layout(exp))
1832                 RETURN(-EOPNOTSUPP);
1833
1834         info = tsi2mdt_info(tsi);
1835         if (unlikely(info->mti_object == NULL))
1836                 RETURN(-EPROTO);
1837
1838         if (info->mti_dlm_req != NULL)
1839                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1840
1841         o1 = info->mti_object;
1842         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1843                                 &info->mti_body->mbo_fid2);
1844         if (IS_ERR(o))
1845                 GOTO(out, rc = PTR_ERR(o));
1846
1847         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1848                 GOTO(put, rc = -ENOENT);
1849
1850         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1851         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1852                 GOTO(put, rc);
1853
1854         if (rc < 0)
1855                 swap(o1, o2);
1856
1857         /* permission check. Make sure the calling process having permission
1858          * to write both files. */
1859         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1860                            MAY_WRITE);
1861         if (rc < 0)
1862                 GOTO(put, rc);
1863
1864         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1865                            MAY_WRITE);
1866         if (rc < 0)
1867                 GOTO(put, rc);
1868
1869         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1870         if (msl == NULL)
1871                 GOTO(put, rc = -EPROTO);
1872
1873         lh1 = &info->mti_lh[MDT_LH_NEW];
1874         lh2 = &info->mti_lh[MDT_LH_OLD];
1875         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1876                              MDS_INODELOCK_XATTR, LCK_EX);
1877         if (rc < 0)
1878                 GOTO(put, rc);
1879
1880         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1881                              MDS_INODELOCK_XATTR, LCK_EX);
1882         if (rc < 0)
1883                 GOTO(unlock1, rc);
1884
1885         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1886                              mdt_object_child(o2), msl->msl_flags);
1887         if (rc < 0)
1888                 GOTO(unlock2, rc);
1889
1890         mdt_swap_lov_flag(o1, o2);
1891
1892 unlock2:
1893         mdt_object_unlock(info, o2, lh2, rc);
1894 unlock1:
1895         mdt_object_unlock(info, o1, lh1, rc);
1896 put:
1897         mdt_object_put(info->mti_env, o);
1898 out:
1899         mdt_thread_info_fini(info);
1900         RETURN(rc);
1901 }
1902
1903 static int mdt_raw_lookup(struct mdt_thread_info *info,
1904                           struct mdt_object *parent,
1905                           const struct lu_name *lname)
1906 {
1907         struct lu_fid *fid = &info->mti_tmp_fid1;
1908         struct mdt_body *repbody;
1909         bool is_dotdot = false;
1910         bool is_old_parent_stripe = false;
1911         bool is_new_parent_checked = false;
1912         int rc;
1913
1914         ENTRY;
1915
1916         LASSERT(!info->mti_cross_ref);
1917         /* Always allow to lookup ".." */
1918         if (lname->ln_namelen == 2 &&
1919             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1920                 info->mti_spec.sp_permitted = 1;
1921                 is_dotdot = true;
1922                 if (mdt_is_dir_stripe(info, parent) == 1)
1923                         is_old_parent_stripe = true;
1924         }
1925
1926         mdt_object_get(info->mti_env, parent);
1927 lookup:
1928         /* Only got the fid of this obj by name */
1929         fid_zero(fid);
1930         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1931                         &info->mti_spec);
1932         mdt_object_put(info->mti_env, parent);
1933         if (rc)
1934                 RETURN(rc);
1935
1936         /* getattr_name("..") should return master object FID for striped dir */
1937         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1938                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1939                 if (IS_ERR(parent))
1940                         RETURN(PTR_ERR(parent));
1941
1942                 /* old client getattr_name("..") with stripe FID */
1943                 if (unlikely(is_old_parent_stripe)) {
1944                         is_old_parent_stripe = false;
1945                         goto lookup;
1946                 }
1947
1948                 /* ".." may be a stripe */
1949                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1950                         is_new_parent_checked = true;
1951                         goto lookup;
1952                 }
1953
1954                 mdt_object_put(info->mti_env, parent);
1955         }
1956
1957         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1958         repbody->mbo_fid1 = *fid;
1959         repbody->mbo_valid = OBD_MD_FLID;
1960
1961         RETURN(rc);
1962 }
1963
1964 /**
1965  * Find name matching hash
1966  *
1967  * We search \a child LinkEA for a name whose hash matches \a lname
1968  * (it contains an encoded hash).
1969  *
1970  * \param info mdt thread info
1971  * \param lname encoded hash to find
1972  * \param parent parent object
1973  * \param child object to search with LinkEA
1974  *
1975  * \retval 1 match found
1976  * \retval 0 no match found
1977  * \retval -ev negative errno upon error
1978  */
1979 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
1980                             struct mdt_object *parent, struct mdt_object *child)
1981 {
1982         /* Here, lname is an encoded hash of on-disk name, and
1983          * client is doing access without encryption key.
1984          * So we need to get LinkEA, check parent fid is correct and
1985          * compare name hash with the one in the request.
1986          */
1987         struct lu_buf *buf = &info->mti_big_buf;
1988         struct lu_name name;
1989         struct lu_fid pfid;
1990         struct linkea_data ldata = { NULL };
1991         struct link_ea_header *leh;
1992         struct link_ea_entry *lee;
1993         struct lu_buf link = { 0 };
1994         char *hash;
1995         int reclen, count, rc;
1996
1997         ENTRY;
1998         if (lname->ln_namelen < LL_CRYPTO_BLOCK_SIZE)
1999                 RETURN(-EINVAL);
2000
2001         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2002         if (!buf->lb_buf)
2003                 RETURN(-ENOMEM);
2004
2005         ldata.ld_buf = buf;
2006         rc = mdt_links_read(info, child, &ldata);
2007         if (rc < 0)
2008                 RETURN(rc);
2009
2010         hash = kmalloc(lname->ln_namelen, GFP_NOFS);
2011         if (!hash)
2012                 RETURN(-ENOMEM);
2013         rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
2014
2015         leh = buf->lb_buf;
2016         lee = (struct link_ea_entry *)(leh + 1);
2017         for (count = 0; count < leh->leh_reccount; count++) {
2018                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
2019                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
2020                         lu_buf_check_and_alloc(&link, name.ln_namelen);
2021                         if (!link.lb_buf)
2022                                 GOTO(out_match, rc = -ENOMEM);
2023                         rc = critical_decode(name.ln_name, name.ln_namelen,
2024                                              link.lb_buf);
2025
2026                         if (memcmp(LLCRYPT_EXTRACT_DIGEST(link.lb_buf, rc),
2027                                    hash, LL_CRYPTO_BLOCK_SIZE) == 0) {
2028                                 *lname = name;
2029                                 break;
2030                         }
2031                 }
2032                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2033         }
2034         if (count == leh->leh_reccount)
2035                 rc = 0;
2036         else
2037                 rc = 1;
2038
2039 out_match:
2040         lu_buf_free(&link);
2041         kfree(hash);
2042
2043         RETURN(rc);
2044 }
2045
2046 /*
2047  * UPDATE lock should be taken against parent, and be released before exit;
2048  * child_bits lock should be taken against child, and be returned back:
2049  *            (1)normal request should release the child lock;
2050  *            (2)intent request will grant the lock to client.
2051  */
2052 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2053                                  struct mdt_lock_handle *lhc,
2054                                  __u64 child_bits,
2055                                  struct ldlm_reply *ldlm_rep)
2056 {
2057         struct ptlrpc_request *req = mdt_info_req(info);
2058         struct mdt_body *reqbody = NULL;
2059         struct mdt_object *parent = info->mti_object;
2060         struct mdt_object *child = NULL;
2061         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2062         struct lu_name *lname = NULL;
2063         struct mdt_lock_handle *lhp = NULL;
2064         struct ldlm_lock *lock;
2065         struct req_capsule *pill = info->mti_pill;
2066         bool fscrypt_md = false;
2067         __u64 try_bits = 0;
2068         bool is_resent;
2069         int ma_need = 0;
2070         int rc;
2071
2072         ENTRY;
2073
2074         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2075         LASSERT(ergo(is_resent,
2076                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2077
2078         if (parent == NULL)
2079                 RETURN(-ENOENT);
2080
2081         if (info->mti_cross_ref) {
2082                 /* Only getattr on the child. Parent is on another node. */
2083                 mdt_set_disposition(info, ldlm_rep,
2084                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2085                 child = parent;
2086                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
2087                        "ldlm_rep = %p\n",
2088                        PFID(mdt_object_fid(child)), ldlm_rep);
2089
2090                 rc = mdt_check_resent_lock(info, child, lhc);
2091                 if (rc < 0) {
2092                         RETURN(rc);
2093                 } else if (rc > 0) {
2094                         /*
2095                          * Object's name entry is on another MDS, it will
2096                          * request PERM lock only because LOOKUP lock is owned
2097                          * by the MDS where name entry resides.
2098                          *
2099                          * TODO: it should try layout lock too. - Jinshan
2100                          */
2101                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2102                                         MDS_INODELOCK_LAYOUT);
2103                         child_bits |= MDS_INODELOCK_PERM;
2104                         rc = mdt_object_lock(info, child, lhc, child_bits,
2105                                              LCK_PR);
2106                         if (rc < 0)
2107                                 RETURN(rc);
2108                 }
2109
2110                 /* Finally, we can get attr for child. */
2111                 if (!mdt_object_exists(child)) {
2112                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2113                                         &child->mot_obj,
2114                                         "remote object doesn't exist.");
2115                         mdt_object_unlock(info, child, lhc, 1);
2116                         RETURN(-ENOENT);
2117                 }
2118
2119                 rc = mdt_getattr_internal(info, child, 0);
2120                 if (unlikely(rc != 0)) {
2121                         mdt_object_unlock(info, child, lhc, 1);
2122                         RETURN(rc);
2123                 }
2124
2125                 rc = mdt_pack_secctx_in_reply(info, child);
2126                 if (unlikely(rc)) {
2127                         mdt_object_unlock(info, child, lhc, 1);
2128                         RETURN(rc);
2129                 }
2130
2131                 rc = mdt_pack_encctx_in_reply(info, child);
2132                 if (unlikely(rc))
2133                         mdt_object_unlock(info, child, lhc, 1);
2134                 RETURN(rc);
2135         }
2136
2137         lname = &info->mti_name;
2138         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2139
2140         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2141                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2142                 if (unlikely(reqbody == NULL))
2143                         RETURN(err_serious(-EPROTO));
2144
2145                 *child_fid = reqbody->mbo_fid2;
2146                 if (unlikely(!fid_is_sane(child_fid)))
2147                         RETURN(err_serious(-EINVAL));
2148
2149                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2150                         mdt_object_get(info->mti_env, parent);
2151                         child = parent;
2152                 } else {
2153                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2154                                                 child_fid);
2155                         if (IS_ERR(child))
2156                                 RETURN(PTR_ERR(child));
2157                 }
2158
2159                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2160                        "ldlm_rep = %p\n",
2161                        PFID(mdt_object_fid(parent)),
2162                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2163         } else if (lu_name_is_valid(lname)) {
2164                 if (mdt_object_remote(parent)) {
2165                         CERROR("%s: parent "DFID" is on remote target\n",
2166                                mdt_obd_name(info->mti_mdt),
2167                                PFID(mdt_object_fid(parent)));
2168                         RETURN(-EPROTO);
2169                 }
2170
2171                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
2172                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
2173                        PNAME(lname), ldlm_rep);
2174
2175                 if (parent->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD ||
2176                     (fid_is_root(mdt_object_fid(parent)) &&
2177                      lname->ln_namelen == strlen(dot_fscrypt_name) &&
2178                      strncmp(lname->ln_name, dot_fscrypt_name,
2179                              lname->ln_namelen) == 0))
2180                         fscrypt_md = true;
2181         } else {
2182                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2183                 if (unlikely(reqbody == NULL))
2184                         RETURN(err_serious(-EPROTO));
2185
2186                 *child_fid = reqbody->mbo_fid2;
2187                 if (unlikely(!fid_is_sane(child_fid)))
2188                         RETURN(err_serious(-EINVAL));
2189
2190                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2191                         mdt_object_get(info->mti_env, parent);
2192                         child = parent;
2193                 } else {
2194                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2195                                                 child_fid);
2196                         if (IS_ERR(child))
2197                                 RETURN(PTR_ERR(child));
2198                 }
2199
2200                 if (mdt_object_remote(child)) {
2201                         CERROR("%s: child "DFID" is on remote target\n",
2202                                mdt_obd_name(info->mti_mdt),
2203                                PFID(mdt_object_fid(child)));
2204                         GOTO(out_child, rc = -EPROTO);
2205                 }
2206
2207                 /* don't fetch LOOKUP lock if it's remote object */
2208                 rc = mdt_is_remote_object(info, parent, child);
2209                 if (rc < 0)
2210                         GOTO(out_child, rc);
2211                 if (rc)
2212                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2213
2214                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2215                        "ldlm_rep = %p\n",
2216                        PFID(mdt_object_fid(parent)),
2217                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2218         }
2219
2220         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2221
2222         if (unlikely(!mdt_object_exists(parent)) &&
2223             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2224             lu_name_is_valid(lname)) {
2225                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2226                                 &parent->mot_obj,
2227                                 "Parent doesn't exist!");
2228                 GOTO(out_child, rc = -ESTALE);
2229         }
2230
2231         if (!child && is_resent) {
2232                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2233                 if (lock == NULL) {
2234                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2235                          * a resend case, however, it could be already destroyed
2236                          * due to client eviction or a raced cancel RPC.
2237                          */
2238                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2239                                           lhc->mlh_reg_lh.cookie);
2240                         RETURN(-ESTALE);
2241                 }
2242                 fid_extract_from_res_name(child_fid,
2243                                           &lock->l_resource->lr_name);
2244                 LDLM_LOCK_PUT(lock);
2245                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2246                                         child_fid);
2247                 if (IS_ERR(child))
2248                         RETURN(PTR_ERR(child));
2249         } else if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2250             lu_name_is_valid(lname)) {
2251                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2252                         rc = mdt_raw_lookup(info, parent, lname);
2253
2254                         RETURN(rc);
2255                 }
2256
2257                 /* step 1: lock parent only if parent is a directory */
2258                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2259                         lhp = &info->mti_lh[MDT_LH_PARENT];
2260                         rc = mdt_parent_lock(info, parent, lhp, lname, LCK_PR);
2261                         if (unlikely(rc != 0))
2262                                 RETURN(rc);
2263                 }
2264
2265                 /* step 2: lookup child's fid by name */
2266                 fid_zero(child_fid);
2267                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2268                                 child_fid, &info->mti_spec);
2269                 if (rc == -ENOENT)
2270                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2271
2272                 if (rc != 0)
2273                         GOTO(unlock_parent, rc);
2274
2275                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2276                                         child_fid);
2277                 if (unlikely(IS_ERR(child)))
2278                         GOTO(unlock_parent, rc = PTR_ERR(child));
2279         }
2280
2281         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2282
2283         /* step 3: lock child regardless if it is local or remote. */
2284         LASSERT(child);
2285
2286         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2287                 /* Here, lname is an encoded hash of on-disk name, and
2288                  * client is doing access without encryption key.
2289                  * So we need to compare name hash with the one in the request.
2290                  */
2291                 if (!find_name_matching_hash(info, lname, parent,
2292                                              child)) {
2293                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2294                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2295                         GOTO(out_child, rc = -ENOENT);
2296                 }
2297         }
2298
2299         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2300         if (!mdt_object_exists(child)) {
2301                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2302                                 &child->mot_obj,
2303                                 "Object doesn't exist!");
2304                 GOTO(out_child, rc = -ENOENT);
2305         }
2306
2307         rc = mdt_check_resent_lock(info, child, lhc);
2308         if (rc < 0) {
2309                 GOTO(out_child, rc);
2310         } else if (rc > 0) {
2311                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2312                     !mdt_object_remote(child)) {
2313                         struct md_attr *ma = &info->mti_attr;
2314
2315                         ma->ma_valid = 0;
2316                         ma->ma_need = MA_INODE;
2317                         rc = mdt_attr_get_complex(info, child, ma);
2318                         if (unlikely(rc != 0))
2319                                 GOTO(out_child, rc);
2320
2321                         /* If the file has not been changed for some time, we
2322                          * return not only a LOOKUP lock, but also an UPDATE
2323                          * lock and this might save us RPC on later STAT. For
2324                          * directories, it also let negative dentry cache start
2325                          * working for this dir. */
2326                         if (ma->ma_valid & MA_INODE &&
2327                             ma->ma_attr.la_valid & LA_CTIME &&
2328                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2329                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2330                                 child_bits |= MDS_INODELOCK_UPDATE;
2331                 }
2332
2333                 /* layout lock must be granted in a best-effort way
2334                  * for IT operations */
2335                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2336                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2337                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2338                         if (!CFS_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2339                             exp_connect_layout(info->mti_exp)) {
2340                                 /* try to grant layout lock for regular file. */
2341                                 try_bits = MDS_INODELOCK_LAYOUT;
2342                         }
2343                         /* Acquire DOM lock in advance for data-on-mdt file */
2344                         if (child != parent)
2345                                 try_bits |= MDS_INODELOCK_DOM;
2346                 }
2347
2348                 /*
2349                  * To avoid possible deadlock between batched statahead RPC
2350                  * and rename()/migrate() operation, it should use trylock to
2351                  * obtain the DLM PR ibits lock for file attributes in a
2352                  * batched statahead RPC. A failed trylock means that other
2353                  * users maybe modify the directory simultaneously as in current
2354                  * Lustre design the server only grants read lock to a client.
2355                  *
2356                  * When a trylock failed, the MDT reports the conflict with
2357                  * error code -EBUSY, and stops statahead immediately.
2358                  */
2359                 if (info->mti_batch_env) {
2360                         /*
2361                          * This is a sub stat-ahead request in a batched RPC.
2362                          * However, the @child is a remote object, we just
2363                          * return -EREMOTE here to forbid stat-ahead on it.
2364                          */
2365                         if (mdt_object_remote(child))
2366                                 GOTO(out_child, rc = -EREMOTE);
2367
2368                         try_bits |= child_bits;
2369                         child_bits = 0;
2370                 }
2371
2372                 if (try_bits != 0) {
2373                         /* try layout lock, it may fail to be granted due to
2374                          * contention at LOOKUP or UPDATE */
2375                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2376                                                  try_bits, LCK_PR);
2377                         if (child_bits & MDS_INODELOCK_LAYOUT)
2378                                 ma_need |= MA_LOV;
2379                 } else {
2380                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2381                          * client will enqueue the lock to the remote MDT */
2382                         if (mdt_object_remote(child))
2383                                 rc = mdt_object_lookup_lock(info, NULL, child,
2384                                                             lhc, LCK_PR);
2385                         else
2386                                 rc = mdt_object_lock(info, child, lhc,
2387                                                      child_bits, LCK_PR);
2388                 }
2389                 if (unlikely(rc != 0))
2390                         GOTO(out_child, rc);
2391                 if (info->mti_batch_env && child_bits == 0) {
2392                         if (!is_resent)
2393                                 mdt_object_unlock(info, child, lhc, 1);
2394                         GOTO(out_child, rc = -EBUSY);
2395                 }
2396         }
2397
2398         if (fscrypt_md)
2399                 child->mot_obj.lo_header->loh_attr |= LOHA_FSCRYPT_MD;
2400
2401         /* finally, we can get attr for child. */
2402         rc = mdt_getattr_internal(info, child, ma_need);
2403         if (unlikely(rc != 0)) {
2404                 if (!is_resent)
2405                         mdt_object_unlock(info, child, lhc, 1);
2406                 GOTO(out_child, rc);
2407         }
2408
2409         rc = mdt_pack_secctx_in_reply(info, child);
2410         if (unlikely(rc)) {
2411                 if (!is_resent)
2412                         mdt_object_unlock(info, child, lhc, 1);
2413                 GOTO(out_child, rc);
2414         }
2415
2416         rc = mdt_pack_encctx_in_reply(info, child);
2417         if (unlikely(rc)) {
2418                 if (!is_resent)
2419                         mdt_object_unlock(info, child, lhc, 1);
2420                 GOTO(out_child, rc);
2421         }
2422
2423         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2424         if (lock) {
2425                 /* Debugging code. */
2426                 LDLM_DEBUG(lock, "Returning lock to client");
2427                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2428                                          &lock->l_resource->lr_name),
2429                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2430                          PLDLMRES(lock->l_resource),
2431                          PFID(mdt_object_fid(child)));
2432
2433                 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2434                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2435                                 CFS_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2436                                                  req->rq_deadline -
2437                                                  req->rq_arrival_time.tv_sec +
2438                                                  cfs_fail_val ?: 3);
2439                         /* Put the lock to the waiting list and force the cancel */
2440                         ldlm_set_ast_sent(lock);
2441                 }
2442
2443                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2444                     !mdt_object_remote(child) && child != parent) {
2445                         mdt_object_put(info->mti_env, child);
2446                         rc = mdt_pack_size2body(info, child_fid,
2447                                                 &lhc->mlh_reg_lh);
2448                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2449                                 /* DOM lock was taken in advance but this is
2450                                  * not DoM file. Drop the lock.
2451                                  */
2452                                 lock_res_and_lock(lock);
2453                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2454                                 unlock_res_and_lock(lock);
2455                         }
2456                         LDLM_LOCK_PUT(lock);
2457                         GOTO(unlock_parent, rc = 0);
2458                 }
2459                 LDLM_LOCK_PUT(lock);
2460         }
2461
2462         EXIT;
2463 out_child:
2464         if (child)
2465                 mdt_object_put(info->mti_env, child);
2466 unlock_parent:
2467         if (lhp)
2468                 mdt_object_unlock(info, parent, lhp, 1);
2469         if (rc == -ENOENT) {
2470                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2471                  * client if trying to access an encrypted file
2472                  */
2473                 int rc2 = mdt_check_enc(info, parent);
2474
2475                 if (rc2)
2476                         rc = rc2;
2477         }
2478         return rc;
2479 }
2480
2481 /* normal handler: should release the child lock */
2482 static int mdt_getattr_name(struct tgt_session_info *tsi)
2483 {
2484         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2485         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2486         struct mdt_body *reqbody;
2487         struct mdt_body *repbody;
2488         int rc, rc2;
2489
2490         ENTRY;
2491
2492         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2493         LASSERT(reqbody != NULL);
2494         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2495         LASSERT(repbody != NULL);
2496
2497         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2498         repbody->mbo_eadatasize = 0;
2499         repbody->mbo_aclsize = 0;
2500
2501         rc = mdt_init_ucred(info, reqbody);
2502         if (unlikely(rc))
2503                 GOTO(out_shrink, rc);
2504
2505         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2506         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2507                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2508                 lhc->mlh_reg_lh.cookie = 0;
2509         }
2510         mdt_exit_ucred(info);
2511         EXIT;
2512 out_shrink:
2513         mdt_client_compatibility(info);
2514         rc2 = mdt_fix_reply(info);
2515         if (rc == 0)
2516                 rc = rc2;
2517         mdt_thread_info_fini(info);
2518         return rc;
2519 }
2520
2521 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2522                             const struct lu_fid *pfid,
2523                             const struct lu_name *name,
2524                             struct mdt_object *obj, s64 ctime)
2525 {
2526         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2527         struct ldlm_enqueue_info *einfo = &info->mti_einfo;
2528         struct mdt_device *mdt = info->mti_mdt;
2529         struct md_attr *ma = &info->mti_attr;
2530         struct mdt_lock_handle *parent_lh;
2531         struct mdt_lock_handle *child_lh;
2532         struct mdt_object *pobj;
2533         int rc;
2534         ENTRY;
2535
2536         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2537         if (IS_ERR(pobj))
2538                 GOTO(out, rc = PTR_ERR(pobj));
2539
2540         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2541         rc = mdt_parent_lock(info, pobj, parent_lh, name, LCK_PW);
2542         if (rc != 0)
2543                 GOTO(put_parent, rc);
2544
2545         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2546                         name, child_fid, &info->mti_spec);
2547         if (rc != 0)
2548                 GOTO(unlock_parent, rc);
2549
2550         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2551                 GOTO(unlock_parent, rc = -EREMCHG);
2552
2553         child_lh = &info->mti_lh[MDT_LH_CHILD];
2554         rc = mdt_object_stripes_lock(info, pobj, obj, child_lh, einfo,
2555                                      MDS_INODELOCK_LOOKUP |
2556                                      MDS_INODELOCK_UPDATE, LCK_EX);
2557         if (rc != 0)
2558                 GOTO(unlock_parent, rc);
2559
2560         if (atomic_read(&obj->mot_open_count)) {
2561                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2562                        PFID(mdt_object_fid(obj)));
2563                 GOTO(unlock_child, rc = -EBUSY);
2564         }
2565
2566         ma->ma_need = 0;
2567         ma->ma_valid = MA_INODE;
2568         ma->ma_attr.la_valid = LA_CTIME;
2569         ma->ma_attr.la_ctime = ctime;
2570
2571         mutex_lock(&obj->mot_lov_mutex);
2572
2573         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2574                         mdt_object_child(obj), name, ma, 0);
2575
2576         mutex_unlock(&obj->mot_lov_mutex);
2577
2578 unlock_child:
2579         mdt_object_stripes_unlock(info, obj, child_lh, einfo, 1);
2580 unlock_parent:
2581         mdt_object_unlock(info, pobj, parent_lh, 1);
2582 put_parent:
2583         mdt_object_put(info->mti_env, pobj);
2584 out:
2585         RETURN(rc);
2586 }
2587
2588 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2589                                         struct mdt_object *obj)
2590 {
2591         struct lu_ucred *uc = lu_ucred(info->mti_env);
2592         struct md_attr *ma = &info->mti_attr;
2593         struct lu_attr *la = &ma->ma_attr;
2594         int rc = 0;
2595         ENTRY;
2596
2597         ma->ma_need = MA_INODE;
2598         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2599         if (rc)
2600                 GOTO(out, rc);
2601
2602         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2603                         rc = -EACCES;
2604
2605         /* we want rbac roles to have precedence over any other
2606          * permission or capability checks
2607          */
2608         if (!uc->uc_rbac_byfid_ops)
2609                 RETURN(-EPERM);
2610         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2611                 RETURN(0);
2612         if (uc->uc_fsuid == la->la_uid) {
2613                 if ((la->la_mode & S_IWUSR) == 0)
2614                         rc = -EACCES;
2615         } else if (uc->uc_fsgid == la->la_gid) {
2616                 if ((la->la_mode & S_IWGRP) == 0)
2617                         rc = -EACCES;
2618         } else if ((la->la_mode & S_IWOTH) == 0) {
2619                         rc = -EACCES;
2620         }
2621
2622 out:
2623         RETURN(rc);
2624 }
2625
2626 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2627                          s64 ctime)
2628 {
2629         struct mdt_device *mdt = info->mti_mdt;
2630         struct mdt_object *obj = NULL;
2631         struct linkea_data ldata = { NULL };
2632         struct lu_buf *buf = &info->mti_big_buf;
2633         struct lu_name *name = &info->mti_name;
2634         struct lu_fid *pfid = &info->mti_tmp_fid1;
2635         struct link_ea_header *leh;
2636         struct link_ea_entry *lee;
2637         int reclen, count, rc = 0;
2638         ENTRY;
2639
2640         if (!fid_is_sane(fid))
2641                 GOTO(out, rc = -EINVAL);
2642
2643         if (!fid_is_namespace_visible(fid))
2644                 GOTO(out, rc = -EINVAL);
2645
2646         obj = mdt_object_find(info->mti_env, mdt, fid);
2647         if (IS_ERR(obj))
2648                 GOTO(out, rc = PTR_ERR(obj));
2649
2650         if (mdt_object_remote(obj))
2651                 GOTO(out, rc = -EREMOTE);
2652         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2653                 GOTO(out, rc = -ENOENT);
2654
2655         rc = mdt_rmfid_check_permission(info, obj);
2656         if (rc)
2657                 GOTO(out, rc);
2658
2659         /* take LinkEA */
2660         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2661         if (!buf->lb_buf)
2662                 GOTO(out, rc = -ENOMEM);
2663
2664         ldata.ld_buf = buf;
2665         rc = mdt_links_read(info, obj, &ldata);
2666         if (rc)
2667                 GOTO(out, rc);
2668
2669         leh = buf->lb_buf;
2670         lee = (struct link_ea_entry *)(leh + 1);
2671         for (count = 0; count < leh->leh_reccount; count++) {
2672                 /* remove every hardlink */
2673                 linkea_entry_unpack(lee, &reclen, name, pfid);
2674                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2675                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2676                 if (rc)
2677                         break;
2678         }
2679
2680 out:
2681         if (obj && !IS_ERR(obj))
2682                 mdt_object_put(info->mti_env, obj);
2683         if (info->mti_big_buf.lb_buf)
2684                 lu_buf_free(&info->mti_big_buf);
2685
2686         RETURN(rc);
2687 }
2688
2689 static int mdt_rmfid(struct tgt_session_info *tsi)
2690 {
2691         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2692         struct mdt_body *reqbody;
2693         struct lu_fid *fids, *rfids;
2694         int bufsize, rc;
2695         __u32 *rcs;
2696         int i, nr;
2697         ENTRY;
2698
2699         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2700         if (reqbody == NULL)
2701                 RETURN(-EPROTO);
2702         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2703                                        RCL_CLIENT);
2704         nr = bufsize / sizeof(struct lu_fid);
2705         if (nr * sizeof(struct lu_fid) != bufsize)
2706                 RETURN(-EINVAL);
2707         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2708                              RCL_SERVER, nr * sizeof(__u32));
2709         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2710                              RCL_SERVER, nr * sizeof(struct lu_fid));
2711         rc = req_capsule_server_pack(tsi->tsi_pill);
2712         if (rc)
2713                 GOTO(out, rc = err_serious(rc));
2714         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2715         if (fids == NULL)
2716                 RETURN(-EPROTO);
2717         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2718         LASSERT(rcs);
2719         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2720         LASSERT(rfids);
2721
2722         mdt_init_ucred(mti, reqbody);
2723         for (i = 0; i < nr; i++) {
2724                 rfids[i] = fids[i];
2725                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2726         }
2727         mdt_exit_ucred(mti);
2728
2729 out:
2730         RETURN(rc);
2731 }
2732
2733 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2734                          void *karg, void __user *uarg);
2735
2736 int mdt_io_set_info(struct tgt_session_info *tsi)
2737 {
2738         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2739         struct ost_body         *body = NULL, *repbody;
2740         void                    *key, *val = NULL;
2741         int                      keylen, vallen, rc = 0;
2742         bool                     is_grant_shrink;
2743
2744         ENTRY;
2745
2746         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2747         if (key == NULL) {
2748                 DEBUG_REQ(D_HA, req, "no set_info key");
2749                 RETURN(err_serious(-EFAULT));
2750         }
2751         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2752                                       RCL_CLIENT);
2753
2754         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2755         if (val == NULL) {
2756                 DEBUG_REQ(D_HA, req, "no set_info val");
2757                 RETURN(err_serious(-EFAULT));
2758         }
2759         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2760                                       RCL_CLIENT);
2761
2762         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2763         if (is_grant_shrink)
2764                 /* In this case the value is actually an RMF_OST_BODY, so we
2765                  * transmutate the type of this PTLRPC */
2766                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2767
2768         rc = req_capsule_server_pack(tsi->tsi_pill);
2769         if (rc < 0)
2770                 RETURN(rc);
2771
2772         if (is_grant_shrink) {
2773                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2774
2775                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2776                 *repbody = *body;
2777
2778                 /** handle grant shrink, similar to a read request */
2779                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2780                                        &repbody->oa);
2781         } else {
2782                 CERROR("%s: Unsupported key %s\n",
2783                        tgt_name(tsi->tsi_tgt), (char *)key);
2784                 rc = -EOPNOTSUPP;
2785         }
2786
2787         RETURN(rc);
2788 }
2789
2790
2791 static int mdt_set_info(struct tgt_session_info *tsi)
2792 {
2793         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2794         char                    *key;
2795         void                    *val;
2796         int                      keylen, vallen, rc = 0;
2797
2798         ENTRY;
2799
2800         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2801         if (key == NULL) {
2802                 DEBUG_REQ(D_HA, req, "no set_info key");
2803                 RETURN(err_serious(-EFAULT));
2804         }
2805
2806         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2807                                       RCL_CLIENT);
2808
2809         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2810         if (val == NULL) {
2811                 DEBUG_REQ(D_HA, req, "no set_info val");
2812                 RETURN(err_serious(-EFAULT));
2813         }
2814
2815         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2816                                       RCL_CLIENT);
2817
2818         /* Swab any part of val you need to here */
2819         if (KEY_IS(KEY_READ_ONLY)) {
2820                 spin_lock(&req->rq_export->exp_lock);
2821                 if (*(__u32 *)val)
2822                         *exp_connect_flags_ptr(req->rq_export) |=
2823                                 OBD_CONNECT_RDONLY;
2824                 else
2825                         *exp_connect_flags_ptr(req->rq_export) &=
2826                                 ~OBD_CONNECT_RDONLY;
2827                 spin_unlock(&req->rq_export->exp_lock);
2828         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2829                 struct changelog_setinfo *cs = val;
2830
2831                 if (vallen != sizeof(*cs)) {
2832                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2833                                tgt_name(tsi->tsi_tgt), vallen);
2834                         RETURN(-EINVAL);
2835                 }
2836                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2837                         __swab64s(&cs->cs_recno);
2838                         __swab32s(&cs->cs_id);
2839                 }
2840
2841                 if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
2842                         RETURN(-EACCES);
2843                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2844                                    vallen, val, NULL);
2845         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2846                 if (vallen > 0)
2847                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2848         } else {
2849                 RETURN(-EINVAL);
2850         }
2851         RETURN(rc);
2852 }
2853
2854 static int mdt_readpage(struct tgt_session_info *tsi)
2855 {
2856         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2857         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2858         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2859         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2860         struct mdt_body         *repbody;
2861         int                      rc;
2862         int                      i;
2863
2864         ENTRY;
2865
2866         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2867                 RETURN(err_serious(-ENOMEM));
2868
2869         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2870         if (repbody == NULL || reqbody == NULL)
2871                 RETURN(err_serious(-EFAULT));
2872
2873         /*
2874          * prepare @rdpg before calling lower layers and transfer itself. Here
2875          * reqbody->size contains offset of where to start to read and
2876          * reqbody->nlink contains number bytes to read.
2877          */
2878         rdpg->rp_hash = reqbody->mbo_size;
2879         if (rdpg->rp_hash != reqbody->mbo_size) {
2880                 CERROR("Invalid hash: %#llx != %#llx\n",
2881                        rdpg->rp_hash, reqbody->mbo_size);
2882                 RETURN(-EFAULT);
2883         }
2884
2885         rdpg->rp_attrs = reqbody->mbo_mode;
2886         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2887                 rdpg->rp_attrs |= LUDA_64BITHASH;
2888         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2889                                 exp_max_brw_size(tsi->tsi_exp));
2890         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2891                           PAGE_SHIFT;
2892         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2893         if (rdpg->rp_pages == NULL)
2894                 RETURN(-ENOMEM);
2895
2896         for (i = 0; i < rdpg->rp_npages; ++i) {
2897                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2898                 if (rdpg->rp_pages[i] == NULL)
2899                         GOTO(free_rdpg, rc = -ENOMEM);
2900         }
2901
2902         /* call lower layers to fill allocated pages with directory data */
2903         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2904         if (rc < 0)
2905                 GOTO(free_rdpg, rc);
2906
2907         /* send pages to client */
2908         rc = tgt_sendpage(tsi, rdpg, rc);
2909
2910         EXIT;
2911 free_rdpg:
2912
2913         for (i = 0; i < rdpg->rp_npages; i++)
2914                 if (rdpg->rp_pages[i] != NULL)
2915                         __free_page(rdpg->rp_pages[i]);
2916         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2917
2918         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2919                 RETURN(0);
2920
2921         return rc;
2922 }
2923
2924 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2925 {
2926         struct lu_ucred *uc = mdt_ucred_check(info);
2927         struct lu_attr *attr = &info->mti_attr.ma_attr;
2928
2929         if (uc == NULL)
2930                 return -EINVAL;
2931
2932         if (op != REINT_SETATTR) {
2933                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2934                         attr->la_uid = uc->uc_fsuid;
2935                 /* for S_ISGID, inherit gid from his parent, such work will be
2936                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2937                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2938                         attr->la_gid = uc->uc_fsgid;
2939         }
2940
2941         return 0;
2942 }
2943
2944 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2945 {
2946         return op == REINT_OPEN &&
2947              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2948 }
2949
2950 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2951 {
2952         struct req_capsule *pill = info->mti_pill;
2953
2954         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2955                                   RCL_SERVER) &&
2956             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2957                                   RCL_CLIENT)) {
2958                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2959                                          RCL_CLIENT) != 0)
2960                         /* pre-set size in server part with max size */
2961                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2962                                              RCL_SERVER,
2963                                              req_capsule_ptlreq(pill) ?
2964                                              OBD_MAX_DEFAULT_EA_SIZE :
2965                                              MAX_MD_SIZE_OLD);
2966                 else
2967                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2968                                              RCL_SERVER, 0);
2969         }
2970 }
2971
2972 static int mdt_reint_internal(struct mdt_thread_info *info,
2973                               struct mdt_lock_handle *lhc,
2974                               __u32 op)
2975 {
2976         struct req_capsule      *pill = info->mti_pill;
2977         struct mdt_body         *repbody;
2978         int                      rc = 0, rc2;
2979
2980         ENTRY;
2981
2982         rc = mdt_reint_unpack(info, op);
2983         if (rc != 0) {
2984                 CERROR("Can't unpack reint, rc %d\n", rc);
2985                 RETURN(err_serious(rc));
2986         }
2987
2988
2989         /* check if the file system is set to readonly. O_RDONLY open
2990          * is still allowed even the file system is set to readonly mode */
2991         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2992                 RETURN(err_serious(-EROFS));
2993
2994         /* for replay (no_create) lmm is not needed, client has it already */
2995         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2996                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2997                                      DEF_REP_MD_SIZE);
2998
2999         /* llog cookies are always 0, the field is kept for compatibility */
3000         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3001                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
3002
3003         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3004          * by default. If the target object has more ACL entries, then
3005          * enlarge the buffer when necessary. */
3006         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3007                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3008                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3009
3010         mdt_preset_secctx_size(info);
3011         mdt_preset_encctx_size(info);
3012
3013         rc = req_capsule_server_pack(pill);
3014         if (rc != 0) {
3015                 CERROR("Can't pack response, rc %d\n", rc);
3016                 RETURN(err_serious(rc));
3017         }
3018
3019         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
3020                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
3021                 LASSERT(repbody);
3022                 repbody->mbo_eadatasize = 0;
3023                 repbody->mbo_aclsize = 0;
3024         }
3025
3026         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
3027
3028         /* for replay no cookkie / lmm need, because client have this already */
3029         if (info->mti_spec.no_create)
3030                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3031                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
3032
3033         rc = mdt_init_ucred_reint(info);
3034         if (rc)
3035                 GOTO(out_shrink, rc);
3036
3037         rc = mdt_fix_attr_ucred(info, op);
3038         if (rc != 0)
3039                 GOTO(out_ucred, rc = err_serious(rc));
3040
3041         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
3042         if (rc < 0) {
3043                 GOTO(out_ucred, rc);
3044         } else if (rc == 1) {
3045                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
3046                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
3047                 GOTO(out_ucred, rc);
3048         }
3049         rc = mdt_reint_rec(info, lhc);
3050         EXIT;
3051 out_ucred:
3052         mdt_exit_ucred(info);
3053 out_shrink:
3054         mdt_client_compatibility(info);
3055
3056         rc2 = mdt_fix_reply(info);
3057         if (rc == 0)
3058                 rc = rc2;
3059
3060         /*
3061          * Data-on-MDT optimization - read data along with OPEN and return it
3062          * in reply when possible.
3063          */
3064         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
3065                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
3066                                           &lhc->mlh_reg_lh);
3067
3068         return rc;
3069 }
3070
3071 static long mdt_reint_opcode(struct ptlrpc_request *req,
3072                              const struct req_format **fmt)
3073 {
3074         struct mdt_device       *mdt;
3075         struct mdt_rec_reint    *rec;
3076         long                     opc;
3077
3078         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3079         if (rec != NULL) {
3080                 opc = rec->rr_opcode;
3081                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3082                 if (opc < REINT_MAX && fmt[opc] != NULL)
3083                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3084                 else {
3085                         mdt = mdt_exp2dev(req->rq_export);
3086                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3087                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3088                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3089                         opc = err_serious(-EFAULT);
3090                 }
3091         } else {
3092                 opc = err_serious(-EFAULT);
3093         }
3094         return opc;
3095 }
3096
3097 static int mdt_reint(struct tgt_session_info *tsi)
3098 {
3099         long opc;
3100         int  rc;
3101         static const struct req_format *reint_fmts[REINT_MAX] = {
3102                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3103                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3104                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3105                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3106                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3107                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3108                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3109                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3110                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3111                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3112         };
3113
3114         ENTRY;
3115
3116         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3117         if (opc >= 0) {
3118                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3119                 /*
3120                  * No lock possible here from client to pass it to reint code
3121                  * path.
3122                  */
3123                 rc = mdt_reint_internal(info, NULL, opc);
3124                 mdt_thread_info_fini(info);
3125         } else {
3126                 rc = opc;
3127         }
3128
3129         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3130         RETURN(rc);
3131 }
3132
3133 /* this should sync the whole device */
3134 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3135 {
3136         struct dt_device *dt = mdt->mdt_bottom;
3137         int rc;
3138         ENTRY;
3139
3140         rc = dt->dd_ops->dt_sync(env, dt);
3141         RETURN(rc);
3142 }
3143
3144 /* this should sync this object */
3145 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3146                            struct mdt_object *mo)
3147 {
3148         int rc = 0;
3149
3150         ENTRY;
3151
3152         if (!mdt_object_exists(mo)) {
3153                 CWARN("%s: non existing object "DFID": rc = %d\n",
3154                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3155                       -ESTALE);
3156                 RETURN(-ESTALE);
3157         }
3158
3159         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3160                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3161                 dt_obj_version_t version;
3162
3163                 version = dt_version_get(env, mdt_obj2dt(mo));
3164                 if (version > tgt->lut_obd->obd_last_committed)
3165                         rc = mo_object_sync(env, mdt_object_child(mo));
3166         } else {
3167                 rc = mo_object_sync(env, mdt_object_child(mo));
3168         }
3169
3170         RETURN(rc);
3171 }
3172
3173 static int mdt_sync(struct tgt_session_info *tsi)
3174 {
3175         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3176         struct req_capsule      *pill = tsi->tsi_pill;
3177         struct mdt_body         *body;
3178         ktime_t                  kstart = ktime_get();
3179         int                      rc;
3180
3181         ENTRY;
3182
3183         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3184                 RETURN(err_serious(-ENOMEM));
3185
3186         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3187                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3188         } else {
3189                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3190
3191                 if (unlikely(info->mti_object == NULL))
3192                         RETURN(-EPROTO);
3193
3194                 /* sync an object */
3195                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3196                                      info->mti_object);
3197                 if (rc == 0) {
3198                         const struct lu_fid *fid;
3199                         struct lu_attr *la = &info->mti_attr.ma_attr;
3200
3201                         info->mti_attr.ma_need = MA_INODE;
3202                         info->mti_attr.ma_valid = 0;
3203                         rc = mdt_attr_get_complex(info, info->mti_object,
3204                                                   &info->mti_attr);
3205                         if (rc == 0) {
3206                                 body = req_capsule_server_get(pill,
3207                                                               &RMF_MDT_BODY);
3208                                 fid = mdt_object_fid(info->mti_object);
3209                                 mdt_pack_attr2body(info, body, la, fid);
3210                         }
3211                 }
3212                 mdt_thread_info_fini(info);
3213         }
3214         if (rc == 0)
3215                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3216                                  ktime_us_delta(ktime_get(), kstart));
3217
3218         RETURN(rc);
3219 }
3220
3221 static int mdt_data_sync(struct tgt_session_info *tsi)
3222 {
3223         struct mdt_thread_info *info;
3224         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3225         struct ost_body *body = tsi->tsi_ost_body;
3226         struct ost_body *repbody;
3227         struct mdt_object *mo = NULL;
3228         struct md_attr *ma;
3229         int rc = 0;
3230
3231         ENTRY;
3232
3233         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3234
3235         /* if no fid is specified then do nothing,
3236          * device sync is done via MDS_SYNC */
3237         if (fid_is_zero(&tsi->tsi_fid))
3238                 RETURN(0);
3239
3240         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3241         if (IS_ERR(mo))
3242                 RETURN(PTR_ERR(mo));
3243
3244         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3245         if (rc)
3246                 GOTO(put, rc);
3247
3248         repbody->oa.o_oi = body->oa.o_oi;
3249         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3250
3251         info = tsi2mdt_info(tsi);
3252         ma = &info->mti_attr;
3253         ma->ma_need = MA_INODE;
3254         ma->ma_valid = 0;
3255         rc = mdt_attr_get_complex(info, mo, ma);
3256         if (rc == 0)
3257                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3258         else
3259                 rc = 0;
3260         mdt_thread_info_fini(info);
3261
3262         EXIT;
3263 put:
3264         if (mo != NULL)
3265                 mdt_object_put(tsi->tsi_env, mo);
3266         return rc;
3267 }
3268
3269 /*
3270  * Handle quota control requests to consult current usage/limit, but also
3271  * to configure quota enforcement
3272  */
3273 static int mdt_quotactl(struct tgt_session_info *tsi)
3274 {
3275         struct obd_export *exp  = tsi->tsi_exp;
3276         struct req_capsule *pill = tsi->tsi_pill;
3277         struct obd_quotactl *oqctl, *repoqc;
3278         int id, rc;
3279         struct mdt_device *mdt = mdt_exp2dev(exp);
3280         struct lu_device *qmt = mdt->mdt_qmt_dev;
3281         struct lu_nodemap *nodemap;
3282         ENTRY;
3283
3284         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3285         if (!oqctl)
3286                 RETURN(err_serious(-EPROTO));
3287
3288         rc = req_capsule_server_pack(pill);
3289         if (rc)
3290                 RETURN(err_serious(rc));
3291
3292         nodemap = nodemap_get_from_exp(exp);
3293         if (IS_ERR(nodemap))
3294                 RETURN(PTR_ERR(nodemap));
3295
3296         switch (oqctl->qc_cmd) {
3297                 /* master quotactl */
3298         case Q_SETINFO:
3299         case Q_SETQUOTA:
3300         case LUSTRE_Q_SETDEFAULT:
3301         case LUSTRE_Q_SETQUOTAPOOL:
3302         case LUSTRE_Q_SETINFOPOOL:
3303         case LUSTRE_Q_SETDEFAULT_POOL:
3304         case LUSTRE_Q_DELETEQID:
3305         case LUSTRE_Q_RESETQID:
3306                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3307                                           oqctl->qc_id))
3308                         GOTO(out_nodemap, rc = -EPERM);
3309                 fallthrough;
3310         case Q_GETINFO:
3311         case Q_GETQUOTA:
3312         case LUSTRE_Q_GETDEFAULT:
3313         case LUSTRE_Q_GETQUOTAPOOL:
3314         case LUSTRE_Q_GETINFOPOOL:
3315         case LUSTRE_Q_GETDEFAULT_POOL:
3316                 if (qmt == NULL)
3317                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3318                 /* slave quotactl */
3319                 fallthrough;
3320         case Q_GETOINFO:
3321         case Q_GETOQUOTA:
3322                 break;
3323         default:
3324                 rc = -EFAULT;
3325                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3326                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3327                 GOTO(out_nodemap, rc);
3328         }
3329
3330         id = oqctl->qc_id;
3331         switch (oqctl->qc_type) {
3332         case USRQUOTA:
3333                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3334                                     NODEMAP_CLIENT_TO_FS, id);
3335                 break;
3336         case GRPQUOTA:
3337                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3338                                     NODEMAP_CLIENT_TO_FS, id);
3339                 break;
3340         case PRJQUOTA:
3341                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3342                                     NODEMAP_CLIENT_TO_FS, id);
3343                 break;
3344         default:
3345                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3346         }
3347         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3348         if (repoqc == NULL)
3349                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3350
3351         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3352                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3353
3354         if (oqctl->qc_id != id)
3355                 swap(oqctl->qc_id, id);
3356
3357         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3358                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3359                         RETURN(-EINPROGRESS);
3360         }
3361
3362         switch (oqctl->qc_cmd) {
3363
3364         case Q_GETINFO:
3365         case Q_SETINFO:
3366         case Q_SETQUOTA:
3367         case Q_GETQUOTA:
3368         case LUSTRE_Q_SETDEFAULT:
3369         case LUSTRE_Q_GETDEFAULT:
3370         case LUSTRE_Q_SETQUOTAPOOL:
3371         case LUSTRE_Q_GETQUOTAPOOL:
3372         case LUSTRE_Q_SETINFOPOOL:
3373         case LUSTRE_Q_GETINFOPOOL:
3374         case LUSTRE_Q_SETDEFAULT_POOL:
3375         case LUSTRE_Q_GETDEFAULT_POOL:
3376         case LUSTRE_Q_DELETEQID:
3377         case LUSTRE_Q_RESETQID:
3378                 /* forward quotactl request to QMT */
3379                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3380                 break;
3381
3382         case Q_GETOINFO:
3383         case Q_GETOQUOTA:
3384                 /* slave quotactl */
3385                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3386                                    oqctl);
3387                 break;
3388
3389         default:
3390                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3391                 GOTO(out_nodemap, rc = -EFAULT);
3392         }
3393
3394         if (oqctl->qc_id != id)
3395                 swap(oqctl->qc_id, id);
3396
3397         QCTL_COPY_NO_PNAME(repoqc, oqctl);
3398         EXIT;
3399
3400 out_nodemap:
3401         nodemap_putref(nodemap);
3402
3403         return rc;
3404 }
3405
3406 /** clone llog ctxt from child (mdd)
3407  * This allows remote llog (replicator) access.
3408  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3409  * context was originally set up, or we can handle them directly.
3410  * I choose the latter, but that means I need any llog
3411  * contexts set up by child to be accessable by the mdt.  So we clone the
3412  * context into our context list here.
3413  */
3414 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3415                                int idx)
3416 {
3417         struct md_device  *next = mdt->mdt_child;
3418         struct llog_ctxt *ctxt;
3419         int rc;
3420
3421         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3422                 return 0;
3423
3424         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3425         if (rc || ctxt == NULL) {
3426                 return 0;
3427         }
3428
3429         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3430         if (rc)
3431                 CERROR("Can't set mdt ctxt %d\n", rc);
3432
3433         return rc;
3434 }
3435
3436 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3437                                  struct mdt_device *mdt, int idx)
3438 {
3439         struct llog_ctxt *ctxt;
3440
3441         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3442         if (ctxt == NULL)
3443                 return 0;
3444         /* Put once for the get we just did, and once for the clone */
3445         llog_ctxt_put(ctxt);
3446         llog_ctxt_put(ctxt);
3447         return 0;
3448 }
3449
3450 /*
3451  * sec context handlers
3452  */
3453 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3454 {
3455         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3456
3457         return 0;
3458 }
3459
3460 /*
3461  * quota request handlers
3462  */
3463 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3464 {
3465         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3466         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3467         int                      rc;
3468         ENTRY;
3469
3470         if (qmt == NULL)
3471                 RETURN(err_serious(-EOPNOTSUPP));
3472
3473         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3474         RETURN(rc);
3475 }
3476
3477 struct mdt_object *mdt_object_new(const struct lu_env *env,
3478                                   struct mdt_device *d,
3479                                   const struct lu_fid *f)
3480 {
3481         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3482         struct lu_object *o;
3483         struct mdt_object *m;
3484         ENTRY;
3485
3486         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3487         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3488         if (unlikely(IS_ERR(o)))
3489                 m = (struct mdt_object *)o;
3490         else
3491                 m = mdt_obj(o);
3492         RETURN(m);
3493 }
3494
3495 struct mdt_object *mdt_object_find(const struct lu_env *env,
3496                                    struct mdt_device *d,
3497                                    const struct lu_fid *f)
3498 {
3499         struct lu_object *o;
3500         struct mdt_object *m;
3501         ENTRY;
3502
3503         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3504         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3505         if (unlikely(IS_ERR(o)))
3506                 m = (struct mdt_object *)o;
3507         else
3508                 m = mdt_obj(o);
3509
3510         RETURN(m);
3511 }
3512
3513 /**
3514  * Asyncronous commit for mdt device.
3515  *
3516  * Pass asynchonous commit call down the MDS stack.
3517  *
3518  * \param env environment
3519  * \param mdt the mdt device
3520  */
3521 static void mdt_device_commit_async(const struct lu_env *env,
3522                                     struct mdt_device *mdt)
3523 {
3524         struct dt_device *dt = mdt->mdt_bottom;
3525         int rc;
3526         ENTRY;
3527
3528         rc = dt->dd_ops->dt_commit_async(env, dt);
3529         if (unlikely(rc != 0))
3530                 CWARN("%s: async commit start failed: rc = %d\n",
3531                       mdt_obd_name(mdt), rc);
3532         atomic_inc(&mdt->mdt_async_commit_count);
3533         EXIT;
3534 }
3535
3536 /**
3537  * Mark the lock as "synchonous".
3538  *
3539  * Mark the lock to deffer transaction commit to the unlock time.
3540  *
3541  * \param lock the lock to mark as "synchonous"
3542  *
3543  * \see mdt_is_lock_sync
3544  * \see mdt_save_lock
3545  */
3546 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3547 {
3548         lock->l_ast_data = (void*)1;
3549 }
3550
3551 /**
3552  * Check whehter the lock "synchonous" or not.
3553  *
3554  * \param lock the lock to check
3555  * \retval 1 the lock is "synchonous"
3556  * \retval 0 the lock isn't "synchronous"
3557  *
3558  * \see mdt_set_lock_sync
3559  * \see mdt_save_lock
3560  */
3561 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3562 {
3563         return lock->l_ast_data != NULL;
3564 }
3565
3566 /**
3567  * Blocking AST for mdt locks.
3568  *
3569  * Starts transaction commit if in case of COS lock conflict or
3570  * deffers such a commit to the mdt_save_lock.
3571  *
3572  * \param lock the lock which blocks a request or cancelling lock
3573  * \param desc unused
3574  * \param data unused
3575  * \param flag indicates whether this cancelling or blocking callback
3576  * \retval 0
3577  * \see ldlm_blocking_ast_nocheck
3578  */
3579 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3580                      void *data, int flag)
3581 {
3582         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3583         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3584         struct ldlm_cb_set_arg *arg = data;
3585         bool commit_async = false;
3586         int rc;
3587
3588         ENTRY;
3589         if (flag == LDLM_CB_CANCELING)
3590                 RETURN(0);
3591
3592         lock_res_and_lock(lock);
3593         if (lock->l_blocking_ast != mdt_blocking_ast) {
3594                 unlock_res_and_lock(lock);
3595                 RETURN(0);
3596         }
3597
3598         /* A blocking ast may be sent from ldlm_lock_decref_internal
3599          * when the last reference to a local lock was released and
3600          * during blocking event from ldlm_work_bl_ast_lock().
3601          * The 'data' parameter is l_ast_data in the first case and
3602          * callback arguments in the second one. Distinguish them by that.
3603          */
3604         if (data && data != lock->l_ast_data && arg->bl_desc) {
3605                 if (lock->l_req_mode & (LCK_COS | LCK_TXN))
3606                         commit_async = true;
3607                 else if ((lock->l_req_mode & (LCK_PW | LCK_EX)) &&
3608                          ((mdt_cos_is_enabled(mdt) &&
3609                              !arg->bl_desc->bl_same_client) ||
3610                           (mdt_slc_is_enabled(mdt) &&
3611                              arg->bl_desc->bl_txn_dependent)))
3612                         mdt_set_lock_sync(lock);
3613         }
3614
3615         rc = ldlm_blocking_ast_nocheck(lock);
3616
3617         if (commit_async) {
3618                 struct lu_env env;
3619
3620                 rc = lu_env_init(&env, LCT_LOCAL);
3621                 if (unlikely(rc != 0))
3622                         CWARN("%s: lu_env initialization failed, cannot "
3623                               "start asynchronous commit: rc = %d\n",
3624                               obd->obd_name, rc);
3625                 else
3626                         mdt_device_commit_async(&env, mdt);
3627                 lu_env_fini(&env);
3628         }
3629         RETURN(rc);
3630 }
3631
3632 /*
3633  * Blocking AST for cross-MDT lock
3634  *
3635  * Discard lock from uncommitted_slc_locks and cancel it.
3636  *
3637  * \param lock  the lock which blocks a request or cancelling lock
3638  * \param desc  unused
3639  * \param data  unused
3640  * \param flag  indicates whether this cancelling or blocking callback
3641  * \retval      0 on success
3642  * \retval      negative number on error
3643  */
3644 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3645                             void *data, int flag)
3646 {
3647         int rc = 0;
3648         ENTRY;
3649
3650         switch (flag) {
3651         case LDLM_CB_BLOCKING: {
3652                 struct lustre_handle lockh;
3653
3654                 ldlm_lock2handle(lock, &lockh);
3655                 rc = ldlm_cli_cancel(&lockh,
3656                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3657                 if (rc < 0) {
3658                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3659                         RETURN(rc);
3660                 }
3661                 break;
3662         }
3663         case LDLM_CB_CANCELING: {
3664                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3665                 struct mdt_device *mdt =
3666                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3667
3668                 LDLM_DEBUG(lock, "Revoke remote lock");
3669
3670                 /* discard slc lock here so that it can be cleaned anytime,
3671                  * especially for cleanup_resource() */
3672                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3673
3674                 /* once we cache lock, l_ast_data is set to mdt_object */
3675                 if (lock->l_ast_data != NULL) {
3676                         struct mdt_object *mo = lock->l_ast_data;
3677                         struct lu_env env;
3678
3679                         rc = lu_env_init(&env, LCT_MD_THREAD);
3680                         if (unlikely(rc != 0)) {
3681                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3682                                       obd->obd_name, mo,
3683                                       PFID(mdt_object_fid(mo)), rc);
3684                                 RETURN(rc);
3685                         }
3686
3687                         if (lock->l_policy_data.l_inodebits.bits &
3688                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3689                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3690                                 mo->mot_cache_attr = 0;
3691                         }
3692                         mdt_object_put(&env, mo);
3693                         lu_env_fini(&env);
3694                 }
3695                 break;
3696         }
3697         default:
3698                 LBUG();
3699         }
3700
3701         RETURN(rc);
3702 }
3703
3704 int mdt_check_resent_lock(struct mdt_thread_info *info,
3705                           struct mdt_object *mo,
3706                           struct mdt_lock_handle *lhc)
3707 {
3708         /* the lock might already be gotten in ldlm_handle_enqueue() */
3709         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3710                 struct ptlrpc_request *req = mdt_info_req(info);
3711                 struct ldlm_lock      *lock;
3712
3713                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3714                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3715                 if (lock == NULL) {
3716                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3717                          * a resend case, however, it could be already destroyed
3718                          * due to client eviction or a raced cancel RPC. */
3719                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3720                                           lhc->mlh_reg_lh.cookie);
3721                         RETURN(-ESTALE);
3722                 }
3723
3724                 if (!fid_res_name_eq(mdt_object_fid(mo),
3725                                      &lock->l_resource->lr_name)) {
3726                         CWARN("%s: Although resent, but still not "
3727                               "get child lock:"DFID"\n",
3728                               info->mti_exp->exp_obd->obd_name,
3729                               PFID(mdt_object_fid(mo)));
3730                         LDLM_LOCK_PUT(lock);
3731                         RETURN(-EPROTO);
3732                 }
3733                 LDLM_LOCK_PUT(lock);
3734                 return 0;
3735         }
3736         return 1;
3737 }
3738
3739 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3740 {
3741         mdt_object_get(NULL, lock->l_ast_data);
3742 }
3743
3744 static int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3745                                       struct mdt_object *obj,
3746                                       struct lustre_handle *lh,
3747                                       enum ldlm_mode mode,
3748                                       union ldlm_policy_data *policy,
3749                                       struct ldlm_res_id *res_id,
3750                                       bool cache)
3751 {
3752         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3753         int rc;
3754
3755         LASSERT(mdt_object_remote(obj));
3756
3757         memset(einfo, 0, sizeof(*einfo));
3758         einfo->ei_type = LDLM_IBITS;
3759         einfo->ei_mode = mode;
3760         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3761         einfo->ei_cb_cp = ldlm_completion_ast;
3762         einfo->ei_enq_slave = 0;
3763         einfo->ei_res_id = res_id;
3764         einfo->ei_req_slot = 1;
3765         if (cache) {
3766                 /*
3767                  * if we cache lock, couple lock with mdt_object, so that object
3768                  * can be easily found in lock ASTs.
3769                  */
3770                 einfo->ei_cbdata = obj;
3771                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3772         }
3773
3774         rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), lh, einfo,
3775                             policy);
3776         if (rc) {
3777                 lh->cookie = 0ull;
3778                 return rc;
3779         }
3780
3781         /* other components like LFSCK can use lockless access
3782          * and populate cache, so we better invalidate it
3783          */
3784         if (policy->l_inodebits.bits &
3785             (MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR))
3786                 mo_invalidate(mti->mti_env, mdt_object_child(obj));
3787
3788         return 0;
3789 }
3790
3791 /*
3792  * Helper function to take PDO and hash lock.
3793  *
3794  * if \a pdo_lock is false, don't take PDO lock, this is case in rename.
3795  */
3796 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3797                         struct mdt_lock_handle *lh, const struct lu_name *name,
3798                         enum ldlm_mode mode, bool pdo_lock)
3799 {
3800         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3801         union ldlm_policy_data *policy = &info->mti_policy;
3802         struct ldlm_res_id *res_id = &info->mti_res_id;
3803         /*
3804          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it is never going to
3805          * be sent to client and we do not want it slowed down due to possible
3806          * cancels.
3807          */
3808         __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3809         __u64 *cookie = NULL;
3810         int rc;
3811
3812         LASSERT(obj);
3813
3814         /* check for exists after object is locked */
3815         if (!mdt_object_exists(obj))
3816                 /* Non-existent object shouldn't have PDO lock */
3817                 return -ESTALE;
3818
3819         /* Non-dir object shouldn't have PDO lock */
3820         if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
3821                 return -ENOTDIR;
3822
3823         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
3824         policy->l_inodebits.try_bits = 0;
3825         policy->l_inodebits.li_gid = 0;
3826         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3827         fid_build_reg_res_name(mdt_object_fid(obj), res_id);
3828         if (info->mti_exp)
3829                 cookie = &info->mti_exp->exp_handle.h_cookie;
3830
3831         mdt_lock_pdo_init(lh, mode, name);
3832         mdt_lock_pdo_mode(info, obj, lh);
3833         if (lh->mlh_pdo_mode != LCK_NL) {
3834                 if (pdo_lock) {
3835                         if (mdt_object_remote(obj)) {
3836                                 rc = mdt_remote_object_lock_try(info, obj,
3837                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3838                                         policy, res_id, false);
3839                                 lh->mlh_pdo_remote = 1;
3840                         } else {
3841                                 rc = mdt_fid_lock(info->mti_env, ns,
3842                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3843                                         policy, res_id, dlmflags, cookie);
3844                         }
3845                         if (rc) {
3846                                 mdt_object_unlock(info, obj, lh, 1);
3847                                 return rc;
3848                         }
3849                 }
3850                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3851         }
3852
3853         if (mdt_object_remote(obj))
3854                 rc = mdt_remote_object_lock_try(info, obj, &lh->mlh_rreg_lh,
3855                         lh->mlh_rreg_mode, policy, res_id, false);
3856         else
3857                 rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh,
3858                         lh->mlh_reg_mode, policy, res_id, dlmflags, cookie);
3859         if (rc)
3860                 mdt_object_unlock(info, obj, lh, 1);
3861         else if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK) &&
3862                    lh->mlh_pdo_hash != 0 &&
3863                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3864                 CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3865
3866         return rc;
3867 }
3868
3869 int mdt_object_lock_internal(struct mdt_thread_info *info,
3870                              struct mdt_object *obj, const struct lu_fid *fid,
3871                              struct mdt_lock_handle *lh, __u64 *ibits,
3872                              __u64 trybits, bool cache)
3873 {
3874         union ldlm_policy_data *policy = &info->mti_policy;
3875         struct ldlm_res_id *res_id = &info->mti_res_id;
3876         struct lustre_handle *handle;
3877         int rc;
3878
3879         policy->l_inodebits.bits = *ibits;
3880         policy->l_inodebits.try_bits = trybits;
3881         policy->l_inodebits.li_gid = lh->mlh_gid;
3882         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3883         fid_build_reg_res_name(fid, res_id);
3884
3885         if (obj && mdt_object_remote(obj)) {
3886                 handle = &lh->mlh_rreg_lh;
3887                 LASSERT(!lustre_handle_is_used(handle));
3888                 LASSERT(lh->mlh_rreg_mode != LCK_MINMODE);
3889                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3890                 rc = mdt_remote_object_lock_try(info, obj, handle,
3891                                                 lh->mlh_rreg_mode, policy,
3892                                                 res_id, cache);
3893         } else {
3894                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3895                 /*
3896                  * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if
3897                  * it is going to be sent to client. If it is -
3898                  * mdt_intent_policy() path will fix it up and turn FL_LOCAL
3899                  * flag off.
3900                  */
3901                 __u64 dlmflags = LDLM_FL_ATOMIC_CB | LDLM_FL_LOCAL_ONLY;
3902                 __u64 *cookie = NULL;
3903
3904                 handle = &lh->mlh_reg_lh;
3905                 LASSERT(!lustre_handle_is_used(handle));
3906                 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3907                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3908
3909                 /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3910                 if (lh->mlh_type == MDT_REG_LOCK &&
3911                     lh->mlh_reg_mode == LCK_EX && *ibits == MDS_INODELOCK_OPEN)
3912                         dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3913
3914
3915                 if (info->mti_exp)
3916                         cookie = &info->mti_exp->exp_handle.h_cookie;
3917
3918                 rc = mdt_fid_lock(info->mti_env, ns, handle, lh->mlh_reg_mode,
3919                                   policy, res_id, dlmflags, cookie);
3920                 if (rc)
3921                         mdt_object_unlock(info, obj, lh, 1);
3922         }
3923
3924         if (rc == 0) {
3925                 struct ldlm_lock *lock;
3926
3927                 /* Return successfully acquired bits to a caller */
3928                 lock = ldlm_handle2lock(handle);
3929                 LASSERT(lock);
3930                 *ibits = lock->l_policy_data.l_inodebits.bits;
3931                 LDLM_LOCK_PUT(lock);
3932         }
3933
3934         return rc;
3935 }
3936
3937 /*
3938  * MDT object locking functions:
3939  * mdt_object_lock(): lock object, this is used in most places, and normally
3940  *      lock ibits doesn't contain LOOKUP, unless the caller knows it's not
3941  *      remote object.
3942  * mdt_object_check_lock(): lock object with LOOKUP and other ibits, it needs
3943  *      to check whether parent is on remote MDT, if so, take LOOKUP on parent
3944  *      MDT separately, and then lock other ibits on child object.
3945  * mdt_parent_lock(): take parent UPDATE lock with specific mode, if parent is
3946  *      local, take PDO lock by name hash, otherwise take regular lock.
3947  * mdt_object_stripes_lock(): lock object which should be local, and if it's a
3948  *      striped directory, lock its stripes, this is called in operations which
3949  *      modify both object and stripes.
3950  * mdt_object_lock_try(): lock object with trybits, the trybits contains
3951  *      optional inode lock bits that can be granted. This is called by
3952  *      getattr/open to fetch more inode lock bits to client, and is also called
3953  *      by dir migration to lock link parent in non-block mode to avoid
3954  *      deadlock.
3955  */
3956
3957 /**
3958  * lock object
3959  *
3960  * this is used to lock object in most places, and normally lock ibits doesn't
3961  * contain LOOKUP, unless the caller knows it's not remote object.
3962  *
3963  * \param info          struct mdt_thread_info
3964  * \param obj           object
3965  * \param lh            lock handle
3966  * \param ibits         MDS inode lock bits
3967  * \param mode          lock mode
3968  *
3969  * \retval              0 on success, -ev on error.
3970  */
3971 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3972                     struct mdt_lock_handle *lh, __u64 ibits,
3973                     enum ldlm_mode mode)
3974 {
3975         int rc;
3976
3977         ENTRY;
3978         mdt_lock_reg_init(lh, mode);
3979         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh,
3980                                       &ibits, 0, false);
3981         RETURN(rc);
3982 }
3983
3984 /**
3985  * lock object with LOOKUP and other ibits
3986  *
3987  * it will check whether parent and child are on different MDTs, if so, take
3988  * LOOKUP lock on parent MDT, and lock other ibits on child MDT, otherwise lock
3989  * all ibits on child MDT. Note, parent and child shouldn't be both on remote
3990  * MDTs, in which case specific lock function should be used, and it's in
3991  * rename and migrate only.
3992  *
3993  * \param info          struct mdt_thread_info
3994  * \param parent        parent object
3995  * \param child         child object
3996  * \param lh            lock handle
3997  * \param ibits         MDS inode lock bits
3998  * \param mode          lock mode
3999  *
4000  * \retval              0 on success, -ev on error.
4001  */
4002 int mdt_object_check_lock(struct mdt_thread_info *info,
4003                           struct mdt_object *parent, struct mdt_object *child,
4004                           struct mdt_lock_handle *lh, __u64 ibits,
4005                           enum ldlm_mode mode)
4006 {
4007         int rc;
4008
4009         ENTRY;
4010         /* if LOOKUP ibit is not set, use mdt_object_lock() */
4011         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
4012         /* if only LOOKUP ibit is needed, use mdt_object_lookup_lock() */
4013         LASSERT(ibits != MDS_INODELOCK_LOOKUP);
4014         LASSERT(parent);
4015         /* @parent and @child shouldn't both be on remote MDTs */
4016         LASSERT(!(mdt_object_remote(parent) && mdt_object_remote(child)));
4017
4018         mdt_lock_reg_init(lh, mode);
4019         if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
4020                 __u64 lookup_ibits = MDS_INODELOCK_LOOKUP;
4021
4022                 rc = mdt_object_lock_internal(info, parent,
4023                                               mdt_object_fid(child), lh,
4024                                               &lookup_ibits, 0, false);
4025                 if (rc)
4026                         RETURN(rc);
4027
4028                 ibits &= ~MDS_INODELOCK_LOOKUP;
4029         }
4030
4031         rc = mdt_object_lock_internal(info, child, mdt_object_fid(child), lh,
4032                                       &ibits, 0, false);
4033         if (rc && !(ibits & MDS_INODELOCK_LOOKUP))
4034                 mdt_object_unlock(info, NULL, lh, 1);
4035
4036         RETURN(rc);
4037 }
4038
4039 /**
4040  * take parent UPDATE lock
4041  *
4042  * if parent is local or mode is LCK_PW, take PDO lock, otherwise take regular
4043  * lock.
4044  *
4045  * \param info  struct mdt_thread_info
4046  * \param obj   parent object
4047  * \param lh    lock handle
4048  * \param lname child name
4049  * \param mode  lock mode
4050  *
4051  * \retval      0 on success, -ev on error.
4052  */
4053 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4054                     struct mdt_lock_handle *lh, const struct lu_name *lname,
4055                     enum ldlm_mode mode)
4056 {
4057         int rc;
4058
4059         ENTRY;
4060         LASSERT(obj && lname);
4061         LASSERT(mode == LCK_PW || mode == LCK_PR);
4062         if (mdt_object_remote(obj) && mode == LCK_PR) {
4063                 __u64 ibits = MDS_INODELOCK_UPDATE;
4064
4065                 mdt_lock_reg_init(lh, mode);
4066                 rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj),
4067                                               lh, &ibits, 0, false);
4068         } else {
4069                 rc = mdt_object_pdo_lock(info, obj, lh, lname, mode, true);
4070         }
4071         RETURN(rc);
4072 }
4073
4074 /**
4075  * lock object with trybits
4076  *
4077  * the trybits contains optional inode lock bits that can be granted. This is
4078  * called by getattr/open to fetch more inode lock bits to client, and is also
4079  * called by dir migration to lock link parent in non-block mode to avoid
4080  * deadlock.
4081  *
4082  * \param info          struct mdt_thread_info
4083  * \param obj           object
4084  * \param lh            lock handle
4085  * \param ibits         MDS inode lock bits
4086  * \param trybits       optional inode lock bits
4087  * \param mode          lock mode
4088  *
4089  * \retval              0 on success, -ev on error.
4090  */
4091 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *obj,
4092                         struct mdt_lock_handle *lh, __u64 *ibits,
4093                         __u64 trybits, enum ldlm_mode mode)
4094 {
4095         bool trylock_only = *ibits == 0;
4096         int rc;
4097
4098         ENTRY;
4099         LASSERT(!(*ibits & trybits));
4100         mdt_lock_reg_init(lh, mode);
4101         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh, ibits,
4102                                       trybits, false);
4103         if (rc && trylock_only) { /* clear error for try ibits lock only */
4104                 LASSERT(*ibits == 0);
4105                 rc = 0;
4106         }
4107         RETURN(rc);
4108 }
4109
4110 /*
4111  * Helper function to take \a obj LOOKUP lock.
4112  *
4113  * Both \a pobj and \a obj may be located on remote MDTs.
4114  */
4115 int mdt_object_lookup_lock(struct mdt_thread_info *info,
4116                            struct mdt_object *pobj, struct mdt_object *obj,
4117                            struct mdt_lock_handle *lh, enum ldlm_mode mode)
4118 {
4119         __u64 ibits = MDS_INODELOCK_LOOKUP;
4120         int rc;
4121
4122         ENTRY;
4123         /* if @parent is NULL, it's on local MDT, and @child is remote,
4124          * this is case in getattr/unlink/open by name.
4125          */
4126         LASSERT(ergo(!pobj, mdt_object_remote(obj)));
4127         mdt_lock_reg_init(lh, mode);
4128         rc = mdt_object_lock_internal(info, pobj, mdt_object_fid(obj), lh,
4129                                       &ibits, 0, false);
4130         RETURN(rc);
4131 }
4132
4133 /**
4134  * Save a lock within request object.
4135  *
4136  * Keep the lock referenced until whether client ACK or transaction
4137  * commit happens or release the lock immediately depending on input
4138  * parameters. If COS is ON, a write lock is converted to COS lock
4139  * before saving.
4140  *
4141  * \param info thead info object
4142  * \param h lock handle
4143  * \param mode lock mode
4144  * \param decref force immediate lock releasing
4145  */
4146 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
4147                           enum ldlm_mode mode, int decref)
4148 {
4149         struct tgt_session_info *tsi = info->mti_env->le_ses ?
4150                                        tgt_ses_info(info->mti_env) : NULL;
4151
4152         ENTRY;
4153
4154         if (lustre_handle_is_used(h)) {
4155                 bool has_trans = tsi && tsi->tsi_has_trans;
4156
4157                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
4158                         mdt_fid_unlock(h, mode);
4159                 } else {
4160                         struct mdt_device *mdt = info->mti_mdt;
4161                         struct ldlm_lock *lock = ldlm_handle2lock(h);
4162                         struct ptlrpc_request *req = mdt_info_req(info);
4163                         bool no_ack = false;
4164
4165                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
4166                                  h->cookie);
4167
4168                         /* there is no request if mdt_object_unlock() is called
4169                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
4170                         if (likely(req != NULL)) {
4171                                 LDLM_DEBUG(lock, "save lock request %p reply "
4172                                         "state %p transno %lld\n", req,
4173                                         req->rq_reply_state, req->rq_transno);
4174                                 if (mdt_cos_is_enabled(mdt)) {
4175                                         mode = LCK_COS;
4176                                         no_ack = true;
4177                                         ldlm_lock_mode_downgrade(lock, mode);
4178                                 } else if (mdt_slc_is_enabled(mdt)) {
4179                                         no_ack = true;
4180                                         if (mode != LCK_TXN) {
4181                                                 mode = LCK_TXN;
4182                                                 ldlm_lock_mode_downgrade(lock,
4183                                                                          mode);
4184                                         }
4185                                 }
4186                                 if (req->rq_export->exp_disconnected)
4187                                         mdt_fid_unlock(h, mode);
4188                                 else
4189                                         ptlrpc_save_lock(req, h, mode, no_ack);
4190                         } else {
4191                                 mdt_fid_unlock(h, mode);
4192                         }
4193
4194                         if (mdt_is_lock_sync(lock)) {
4195                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4196                                 mdt_device_commit_async(info->mti_env, mdt);
4197                         }
4198                         LDLM_LOCK_PUT(lock);
4199                 }
4200                 h->cookie = 0ull;
4201         }
4202
4203         EXIT;
4204 }
4205
4206 /**
4207  * Save cross-MDT lock in uncommitted_slc_locks
4208  *
4209  * Keep the lock referenced until transaction commit happens or release the lock
4210  * immediately depending on input parameters.
4211  *
4212  * \param info thead info object
4213  * \param h lock handle
4214  * \param mode lock mode
4215  * \param decref force immediate lock releasing
4216  */
4217 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4218                                  struct mdt_object *o, struct lustre_handle *h,
4219                                  enum ldlm_mode mode, int decref)
4220 {
4221         ENTRY;
4222
4223         if (lustre_handle_is_used(h)) {
4224                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4225                 struct ptlrpc_request *req = mdt_info_req(info);
4226
4227                 if (o != NULL &&
4228                     (lock->l_policy_data.l_inodebits.bits &
4229                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4230                         mo_invalidate(info->mti_env, mdt_object_child(o));
4231
4232                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4233                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4234                         ldlm_lock_decref_and_cancel(h, mode);
4235                         LDLM_LOCK_PUT(lock);
4236                 } else {
4237                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4238                                           req->rq_transno);
4239                         ldlm_lock_decref(h, mode);
4240                 }
4241                 h->cookie = 0ull;
4242         }
4243
4244         EXIT;
4245 }
4246
4247 /**
4248  * Unlock mdt object.
4249  *
4250  * Immeditely release the regular lock and the PDO lock or save the
4251  * lock in request and keep them referenced until client ACK or
4252  * transaction commit.
4253  *
4254  * \param info thread info object
4255  * \param o mdt object
4256  * \param lh mdt lock handle referencing regular and PDO locks
4257  * \param decref force immediate lock releasing
4258  *
4259  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4260  */
4261 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4262                        struct mdt_lock_handle *lh, int decref)
4263 {
4264         ENTRY;
4265
4266         if (lh->mlh_pdo_remote)
4267                 mdt_save_remote_lock(info, o, &lh->mlh_pdo_lh,
4268                                      lh->mlh_pdo_mode, decref);
4269         else
4270                 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4271         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4272         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4273                              decref);
4274
4275         EXIT;
4276 }
4277
4278 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4279                                         const struct lu_fid *f,
4280                                         struct mdt_lock_handle *lh,
4281                                         __u64 ibits, enum ldlm_mode mode)
4282 {
4283         struct mdt_object *o;
4284
4285         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4286         if (!IS_ERR(o)) {
4287                 int rc;
4288
4289                 rc = mdt_object_lock(info, o, lh, ibits, mode);
4290                 if (rc != 0) {
4291                         mdt_object_put(info->mti_env, o);
4292                         o = ERR_PTR(rc);
4293                 }
4294         }
4295         return o;
4296 }
4297
4298 void mdt_object_unlock_put(struct mdt_thread_info * info,
4299                            struct mdt_object *o,
4300                            struct mdt_lock_handle *lh,
4301                            int decref)
4302 {
4303         mdt_object_unlock(info, o, lh, decref);
4304         mdt_object_put(info->mti_env, o);
4305 }
4306
4307 /*
4308  * Generic code handling requests that have struct mdt_body passed in:
4309  *
4310  *  - extract mdt_body from request and save it in @info, if present;
4311  *
4312  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4313  *  @info;
4314  *
4315  *  - if HAS_BODY flag is set for this request type check whether object
4316  *  actually exists on storage (lu_object_exists()).
4317  *
4318  */
4319 static int mdt_body_unpack(struct mdt_thread_info *info,
4320                            enum tgt_handler_flags flags)
4321 {
4322         const struct mdt_body    *body;
4323         struct mdt_object        *obj;
4324         const struct lu_env      *env;
4325         struct req_capsule       *pill;
4326         int                       rc;
4327         ENTRY;
4328
4329         env = info->mti_env;
4330         pill = info->mti_pill;
4331
4332         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4333         if (body == NULL)
4334                 RETURN(-EFAULT);
4335
4336         if (!(body->mbo_valid & OBD_MD_FLID))
4337                 RETURN(0);
4338
4339         if (!fid_is_sane(&body->mbo_fid1)) {
4340                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4341                 RETURN(-EINVAL);
4342         }
4343
4344         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4345         if (!IS_ERR(obj)) {
4346                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4347                         mdt_object_put(env, obj);
4348                         rc = -ENOENT;
4349                 } else {
4350                         info->mti_object = obj;
4351                         rc = 0;
4352                 }
4353         } else
4354                 rc = PTR_ERR(obj);
4355
4356         RETURN(rc);
4357 }
4358
4359 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4360                                    enum tgt_handler_flags flags)
4361 {
4362         struct req_capsule *pill = info->mti_pill;
4363         int rc;
4364
4365         ENTRY;
4366
4367         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4368                 rc = mdt_body_unpack(info, flags);
4369         else
4370                 rc = 0;
4371
4372         if (rc == 0 && (flags & HAS_REPLY)) {
4373                 /* Pack reply. */
4374                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4375                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4376                                              req_capsule_ptlreq(pill) ?
4377                                              DEF_REP_MD_SIZE : MAX_MD_SIZE_OLD);
4378
4379                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4380                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4381                                              RCL_SERVER, 0);
4382
4383                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4384                  * by default. If the target object has more ACL entries, then
4385                  * enlarge the buffer when necessary. */
4386                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4387                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4388                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4389
4390                 mdt_preset_secctx_size(info);
4391                 mdt_preset_encctx_size(info);
4392
4393                 rc = req_capsule_server_pack(pill);
4394                 if (rc)
4395                         CWARN("%s: cannot pack response: rc = %d\n",
4396                                       mdt_obd_name(info->mti_mdt), rc);
4397         }
4398         RETURN(rc);
4399 }
4400
4401 void mdt_thread_info_reset(struct mdt_thread_info *info)
4402 {
4403         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4404         info->mti_body = NULL;
4405         info->mti_dlm_req = NULL;
4406         info->mti_cross_ref = 0;
4407         info->mti_opdata = 0;
4408         info->mti_big_lmm_used = 0;
4409         info->mti_big_acl_used = 0;
4410         info->mti_som_strict = 0;
4411
4412         info->mti_spec.no_create = 0;
4413         info->mti_spec.sp_rm_entry = 0;
4414         info->mti_spec.sp_permitted = 0;
4415
4416         info->mti_spec.u.sp_ea.eadata = NULL;
4417         info->mti_spec.u.sp_ea.eadatalen = 0;
4418
4419         if (info->mti_batch_env && info->mti_object != NULL) {
4420                 mdt_object_put(info->mti_env, info->mti_object);
4421                 info->mti_object = NULL;
4422         }
4423 }
4424
4425 /*
4426  * Initialize fields of struct mdt_thread_info. Other fields are left in
4427  * uninitialized state, because it's too expensive to zero out whole
4428  * mdt_thread_info (> 1K) on each request arrival.
4429  */
4430 void mdt_thread_info_init(struct ptlrpc_request *req,
4431                           struct mdt_thread_info *info)
4432 {
4433         info->mti_pill = &req->rq_pill;
4434
4435         /* mdt device: it can be NULL while CONNECT */
4436         if (req->rq_export) {
4437                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4438                 info->mti_exp = req->rq_export;
4439         } else
4440                 info->mti_mdt = NULL;
4441         info->mti_env = req->rq_svc_thread->t_env;
4442         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4443         info->mti_big_buf = LU_BUF_NULL;
4444         info->mti_batch_env = 0;
4445         info->mti_object = NULL;
4446
4447         mdt_thread_info_reset(info);
4448 }
4449
4450 void mdt_thread_info_fini(struct mdt_thread_info *info)
4451 {
4452         int i;
4453
4454         if (info->mti_object != NULL) {
4455                 mdt_object_put(info->mti_env, info->mti_object);
4456                 info->mti_object = NULL;
4457         }
4458
4459         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4460                 mdt_lock_handle_assert(&info->mti_lh[i]);
4461         info->mti_env = NULL;
4462         info->mti_pill = NULL;
4463         info->mti_exp = NULL;
4464         info->mti_mdt = NULL;
4465
4466         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4467                 lu_buf_free(&info->mti_big_buf);
4468 }
4469
4470 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4471 {
4472         struct mdt_thread_info  *mti;
4473
4474         mti = mdt_th_info(tsi->tsi_env);
4475         LASSERT(mti != NULL);
4476
4477         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4478         if (tsi->tsi_corpus != NULL) {
4479                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4480                 lu_object_get(tsi->tsi_corpus);
4481         }
4482         mti->mti_body = tsi->tsi_mdt_body;
4483         mti->mti_dlm_req = tsi->tsi_dlm_req;
4484
4485         return mti;
4486 }
4487
4488 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4489 {
4490         if (CFS_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4491             cfs_fail_val ==
4492             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4493                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4494
4495         return tgt_connect(tsi);
4496 }
4497
4498 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4499                               struct mdt_thread_info *info,
4500                               struct ldlm_lock **lockp, __u64 flags)
4501 {
4502         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4503                                    lockp, flags);
4504 }
4505 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4506                           struct mdt_thread_info *info,
4507                           struct ldlm_lock **lockp, __u64 flags)
4508 {
4509         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4510                                lockp, flags);
4511 }
4512
4513 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4514                             struct ldlm_lock **lockp,
4515                             struct mdt_lock_handle *lh,
4516                             __u64 flags, int result)
4517 {
4518         struct ptlrpc_request  *req = mdt_info_req(info);
4519         struct ldlm_lock       *lock = *lockp;
4520         struct ldlm_lock       *new_lock;
4521
4522         /* If possible resent found a lock, @lh is set to its handle */
4523         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4524
4525         if (new_lock == NULL) {
4526                 if (flags & LDLM_FL_INTENT_ONLY) {
4527                         result = 0;
4528                 } else if (flags & LDLM_FL_RESENT) {
4529                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4530                          * resend case, however, it could be already destroyed
4531                          * due to client eviction or a raced cancel RPC.
4532                          */
4533                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4534                                           lh->mlh_reg_lh.cookie);
4535                         result = -ESTALE;
4536                 } else {
4537                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4538                                mdt_obd_name(info->mti_mdt),
4539                                lh->mlh_reg_lh.cookie, flags,
4540                                PFID(&info->mti_tmp_fid1),
4541                                PFID(&info->mti_tmp_fid2), result);
4542                         result = -ESTALE;
4543                 }
4544                 lh->mlh_reg_lh.cookie = 0;
4545                 RETURN(result);
4546         }
4547
4548         /*
4549          * If we've already given this lock to a client once, then we should
4550          * have no readers or writers.  Otherwise, we should have one reader
4551          * _or_ writer ref (which will be zeroed below) before returning the
4552          * lock to a client.
4553          */
4554         if (new_lock->l_export == req->rq_export) {
4555                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4556         } else {
4557                 LASSERT(new_lock->l_export == NULL);
4558                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4559         }
4560
4561         *lockp = new_lock;
4562
4563         if (new_lock->l_export == req->rq_export) {
4564                 /*
4565                  * Already gave this to the client, which means that we
4566                  * reconstructed a reply.
4567                  */
4568                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4569                         MSG_RESENT);
4570
4571                 LDLM_LOCK_RELEASE(new_lock);
4572                 lh->mlh_reg_lh.cookie = 0;
4573                 RETURN(ELDLM_LOCK_REPLACED);
4574         }
4575
4576         /*
4577          * Fixup the lock to be given to the client.
4578          */
4579         lock_res_and_lock(new_lock);
4580         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4581          * possible blocking AST. */
4582         while (new_lock->l_readers > 0) {
4583                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4584                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4585                 new_lock->l_readers--;
4586         }
4587         while (new_lock->l_writers > 0) {
4588                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4589                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4590                 new_lock->l_writers--;
4591         }
4592
4593         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4594         new_lock->l_blocking_ast = lock->l_blocking_ast;
4595         new_lock->l_completion_ast = lock->l_completion_ast;
4596         if (ldlm_has_dom(new_lock))
4597                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4598         new_lock->l_remote_handle = lock->l_remote_handle;
4599         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4600
4601         unlock_res_and_lock(new_lock);
4602
4603         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4604                      &new_lock->l_remote_handle,
4605                      &new_lock->l_exp_hash);
4606
4607         LDLM_LOCK_RELEASE(new_lock);
4608         lh->mlh_reg_lh.cookie = 0;
4609
4610         RETURN(ELDLM_LOCK_REPLACED);
4611 }
4612
4613 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4614                              struct ldlm_lock *new_lock,
4615                              struct mdt_lock_handle *lh, __u64 flags)
4616 {
4617         struct ptlrpc_request  *req = mdt_info_req(info);
4618         struct ldlm_request    *dlmreq;
4619
4620         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4621                 return;
4622
4623         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4624
4625         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4626          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4627          * initialized. */
4628         if (flags & LDLM_FL_RESENT) {
4629                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4630                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4631
4632                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4633                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4634                           lh->mlh_reg_lh.cookie);
4635                 return;
4636         }
4637
4638         /*
4639          * If the xid matches, then we know this is a resent request, and allow
4640          * it. (It's probably an OPEN, for which we don't send a lock.
4641          */
4642         if (req_can_reconstruct(req, NULL) != 0)
4643                 return;
4644
4645         /*
4646          * This remote handle isn't enqueued, so we never received or processed
4647          * this request.  Clear MSG_RESENT, because it can be handled like any
4648          * normal request now.
4649          */
4650         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4651
4652         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4653                   dlmreq->lock_handle[0].cookie);
4654 }
4655
4656 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4657                                struct mdt_thread_info *info,
4658                                struct ldlm_lock **lockp,
4659                                __u64 flags)
4660 {
4661         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4662         struct ldlm_reply      *ldlm_rep = NULL;
4663         int rc;
4664         ENTRY;
4665
4666         /*
4667          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4668          * (for the resend case) or a new lock. Below we will use it to
4669          * replace the original lock.
4670          */
4671         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4672         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4673                 rc = mdt_object_lock(info, info->mti_object, lhc,
4674                                      MDS_INODELOCK_XATTR, (*lockp)->l_req_mode);
4675                 if (rc)
4676                         return rc;
4677         }
4678
4679         rc = mdt_getxattr(info);
4680
4681         if (mdt_info_req(info)->rq_repmsg != NULL)
4682                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4683
4684         if (ldlm_rep == NULL ||
4685             CFS_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4686                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4687                 if (is_serious(rc))
4688                         RETURN(rc);
4689                 else
4690                         RETURN(err_serious(-EFAULT));
4691         }
4692
4693         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4694
4695         /* This is left for interop instead of adding a new interop flag.
4696          * LU-7433 */
4697 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4698         if (ldlm_rep->lock_policy_res2) {
4699                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4700                 RETURN(ELDLM_LOCK_ABORTED);
4701         }
4702 #endif
4703
4704         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4705         RETURN(rc);
4706 }
4707
4708 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4709                               struct mdt_thread_info *info,
4710                               struct ldlm_lock **lockp,
4711                               __u64 flags)
4712 {
4713         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4714         __u64                   child_bits;
4715         struct ldlm_reply      *ldlm_rep;
4716         struct mdt_body        *reqbody;
4717         struct mdt_body        *repbody;
4718         int                     rc, rc2;
4719         ENTRY;
4720
4721         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4722         LASSERT(reqbody);
4723
4724         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4725         LASSERT(repbody);
4726
4727         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4728         repbody->mbo_eadatasize = 0;
4729         repbody->mbo_aclsize = 0;
4730
4731         switch (it_opc) {
4732         case IT_LOOKUP:
4733                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4734                 break;
4735         case IT_GETATTR:
4736                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4737                              MDS_INODELOCK_PERM;
4738                 break;
4739         default:
4740                 CERROR("%s: unsupported intent %#x\n",
4741                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4742                 GOTO(out_shrink, rc = -EINVAL);
4743         }
4744
4745         rc = mdt_init_ucred(info, reqbody);
4746         if (rc)
4747                 GOTO(out_shrink, rc);
4748
4749         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4750         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4751
4752         /* Get lock from request for possible resent case. */
4753         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4754
4755         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4756         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4757
4758         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4759             ldlm_rep->lock_policy_res2 != -ENOKEY)
4760                 ldlm_rep->lock_policy_res2 = 0;
4761         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4762             ldlm_rep->lock_policy_res2) {
4763                 lhc->mlh_reg_lh.cookie = 0ull;
4764                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4765         }
4766
4767         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4768         EXIT;
4769 out_ucred:
4770         mdt_exit_ucred(info);
4771 out_shrink:
4772         mdt_client_compatibility(info);
4773         rc2 = mdt_fix_reply(info);
4774         if (rc == 0)
4775                 rc = rc2;
4776         return rc;
4777 }
4778
4779 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4780                              struct mdt_thread_info *info,
4781                              struct ldlm_lock **lockp,
4782                              __u64 flags)
4783 {
4784         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4785         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4786         struct layout_intent *intent;
4787         struct ldlm_reply *ldlm_rep;
4788         struct lu_fid *fid = &info->mti_tmp_fid2;
4789         struct mdt_object *obj = NULL;
4790         int layout_size = 0;
4791         struct lu_buf *buf = &layout.mlc_buf;
4792         int rc = 0;
4793
4794         ENTRY;
4795
4796         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4797
4798         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4799         if (intent == NULL)
4800                 RETURN(-EPROTO);
4801
4802         CDEBUG(D_INFO, DFID "got layout change request from client: "
4803                "opc:%u flags:%#x extent "DEXT"\n",
4804                PFID(fid), intent->li_opc, intent->li_flags,
4805                PEXT(&intent->li_extent));
4806
4807         switch (intent->li_opc) {
4808         case LAYOUT_INTENT_TRUNC:
4809         case LAYOUT_INTENT_WRITE:
4810         case LAYOUT_INTENT_PCCRO_SET:
4811         case LAYOUT_INTENT_PCCRO_CLEAR:
4812                 layout.mlc_opc = MD_LAYOUT_WRITE;
4813                 layout.mlc_intent = intent;
4814                 break;
4815         case LAYOUT_INTENT_ACCESS:
4816                 break;
4817         case LAYOUT_INTENT_READ:
4818         case LAYOUT_INTENT_GLIMPSE:
4819         case LAYOUT_INTENT_RELEASE:
4820         case LAYOUT_INTENT_RESTORE:
4821                 CERROR("%s: Unsupported layout intent opc %d\n",
4822                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4823                 RETURN(-ENOTSUPP);
4824         default:
4825                 CERROR("%s: Unknown layout intent opc %d\n",
4826                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4827                 RETURN(-EINVAL);
4828         }
4829
4830         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4831         if (IS_ERR(obj))
4832                 RETURN(PTR_ERR(obj));
4833
4834         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4835                 /* if layout is going to be changed don't use the current EA
4836                  * size but the maximum one. That buffer will be shrinked
4837                  * to the actual size in req_capsule_shrink() before reply.
4838                  */
4839                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4840                         layout_size = info->mti_mdt->mdt_max_mdsize;
4841                 } else {
4842                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4843                         if (layout_size < 0)
4844                                 GOTO(out, rc = layout_size);
4845
4846                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4847                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4848                 }
4849                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4850                        mdt_obd_name(info->mti_mdt), layout_size);
4851         }
4852
4853         /*
4854          * set reply buffer size, so that ldlm_handle_enqueue0()->
4855          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4856          */
4857         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4858                              layout_size);
4859         rc = req_capsule_server_pack(info->mti_pill);
4860         if (rc)
4861                 GOTO(out, rc);
4862
4863         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4864         if (!ldlm_rep)
4865                 GOTO(out, rc = -EPROTO);
4866
4867         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4868
4869         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4870         if (layout.mlc_opc == MD_LAYOUT_NOP)
4871                 GOTO(out, rc = 0);
4872
4873         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4874         if (rc < 0)
4875                 GOTO(out, rc);
4876         if (rc == 1) {
4877                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4878                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4879                 GOTO(out, rc);
4880         }
4881
4882         buf->lb_buf = NULL;
4883         buf->lb_len = 0;
4884         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4885                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4886                                                      &RMF_EADATA);
4887                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4888                                                      &RMF_EADATA, RCL_CLIENT);
4889                 /*
4890                  * If it's a replay of layout write intent RPC, the client has
4891                  * saved the extended lovea when it get reply then.
4892                  */
4893                 if (buf->lb_len > 0)
4894                         mdt_fix_lov_magic(info, buf->lb_buf);
4895         }
4896
4897         /* Get lock from request for possible resent case. */
4898         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4899         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4900
4901         /*
4902          * Instantiate some layout components, if @buf contains lovea, then it's
4903          * a replay of the layout intent write RPC.
4904          */
4905         rc = mdt_layout_change(info, obj, lhc, &layout);
4906         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4907
4908         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4909                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4910                 if (rc == ELDLM_LOCK_REPLACED &&
4911                     (*lockp)->l_granted_mode == LCK_EX)
4912                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4913         }
4914
4915         EXIT;
4916 out:
4917         mdt_object_put(info->mti_env, obj);
4918         return rc;
4919 }
4920
4921 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4922                            struct mdt_thread_info *info,
4923                            struct ldlm_lock **lockp,
4924                            __u64 flags)
4925 {
4926         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4927         struct ldlm_reply      *rep = NULL;
4928         long                    opc;
4929         int                     rc;
4930         struct ptlrpc_request  *req = mdt_info_req(info);
4931
4932         static const struct req_format *intent_fmts[REINT_MAX] = {
4933                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4934                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4935         };
4936
4937         ENTRY;
4938
4939         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4940         if (opc < 0)
4941                 RETURN(opc);
4942
4943         /* Get lock from request for possible resent case. */
4944         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4945
4946         rc = mdt_reint_internal(info, lhc, opc);
4947
4948         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4949                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4950
4951         /* Check whether the reply has been packed successfully. */
4952         if (mdt_info_req(info)->rq_repmsg != NULL)
4953                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4954         if (rep == NULL) {
4955                 if (is_serious(rc))
4956                         RETURN(rc);
4957                 else
4958                         RETURN(err_serious(-EFAULT));
4959         }
4960
4961         /* MDC expects this in any case */
4962         if (rc != 0)
4963                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4964
4965         /* the open lock or the lock for cross-ref object should be
4966          * returned to the client */
4967         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4968             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4969                 rep->lock_policy_res2 = 0;
4970                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4971                 RETURN(rc);
4972         }
4973
4974         rep->lock_policy_res2 = clear_serious(rc);
4975
4976         if (rep->lock_policy_res2 == -ENOENT &&
4977             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4978             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4979                 rep->lock_policy_res2 = 0;
4980
4981         lhc->mlh_reg_lh.cookie = 0ull;
4982         if (rc == -ENOTCONN || rc == -ENODEV ||
4983             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4984                 /*
4985                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4986                  * will be returned by rq_status, and client at ptlrpc layer
4987                  * will detect this, then disconnect, reconnect the import
4988                  * immediately, instead of impacting the following the rpc.
4989                  */
4990                 RETURN(rc);
4991         }
4992         /*
4993          * For other cases, the error will be returned by intent, and client
4994          * will retrieve the result from intent.
4995          */
4996         RETURN(ELDLM_LOCK_ABORTED);
4997 }
4998
4999 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
5000                           struct mdt_thread_info *info,
5001                           struct ldlm_lock **lockp,
5002                           u64 flags /* LDLM_FL_* */)
5003 {
5004         struct req_capsule *pill = info->mti_pill;
5005         struct ptlrpc_request *req = mdt_info_req(info);
5006         const struct req_format *it_format;
5007         int (*it_handler)(enum ldlm_intent_flags,
5008                           struct mdt_thread_info *,
5009                           struct ldlm_lock **,
5010                           u64);
5011         enum tgt_handler_flags it_handler_flags = 0;
5012         struct ldlm_reply *rep;
5013         bool check_mdt_object = false;
5014         int rc;
5015         ENTRY;
5016
5017         switch (it_opc) {
5018         case IT_OPEN:
5019         case IT_OPEN|IT_CREAT:
5020                 /*
5021                  * OCREAT is not a IS_MUTABLE request since the file may
5022                  * already exist. We do the extra check of
5023                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
5024                  * really need to create the object.
5025                  */
5026                 it_format = &RQF_LDLM_INTENT;
5027                 it_handler = &mdt_intent_open;
5028                 break;
5029         case IT_GETATTR:
5030                 check_mdt_object = true;
5031                 fallthrough;
5032         case IT_LOOKUP:
5033                 it_format = &RQF_LDLM_INTENT_GETATTR;
5034                 it_handler = &mdt_intent_getattr;
5035                 it_handler_flags = HAS_REPLY;
5036                 break;
5037         case IT_GETXATTR:
5038                 check_mdt_object = true;
5039                 it_format = &RQF_LDLM_INTENT_GETXATTR;
5040                 it_handler = &mdt_intent_getxattr;
5041                 it_handler_flags = HAS_BODY;
5042                 break;
5043         case IT_LAYOUT:
5044                 it_format = &RQF_LDLM_INTENT_LAYOUT;
5045                 it_handler = &mdt_intent_layout;
5046                 break;
5047         case IT_GLIMPSE:
5048                 it_format = &RQF_LDLM_INTENT;
5049                 it_handler = &mdt_intent_glimpse;
5050                 break;
5051         case IT_BRW:
5052                 it_format = &RQF_LDLM_INTENT;
5053                 it_handler = &mdt_intent_brw;
5054                 break;
5055         case IT_QUOTA_DQACQ:
5056         case IT_QUOTA_CONN: {
5057                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
5058
5059                 if (qmt == NULL)
5060                         RETURN(-EOPNOTSUPP);
5061
5062                 if (mdt_rdonly(req->rq_export))
5063                         RETURN(-EROFS);
5064
5065                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
5066                 /* pass the request to quota master */
5067                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
5068                                                  mdt_info_req(info), lockp,
5069                                                  flags);
5070                 RETURN(rc);
5071         }
5072         default:
5073                 CERROR("%s: unknown intent code %#x\n",
5074                        mdt_obd_name(info->mti_mdt), it_opc);
5075                 RETURN(-EPROTO);
5076         }
5077
5078         if (!info->mti_batch_env)
5079                 req_capsule_extend(pill, it_format);
5080
5081         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
5082         if (rc < 0)
5083                 RETURN(rc);
5084
5085         if (unlikely(info->mti_object == NULL && check_mdt_object))
5086                 RETURN(-EPROTO);
5087
5088         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
5089                 RETURN(-EROFS);
5090
5091         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
5092
5093         /* execute policy */
5094         rc = (*it_handler)(it_opc, info, lockp, flags);
5095
5096         /* Check whether the reply has been packed successfully. */
5097         if (info->mti_batch_env || req->rq_repmsg != NULL) {
5098                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5099                 rep->lock_policy_res2 =
5100                         ptlrpc_status_hton(rep->lock_policy_res2);
5101         }
5102
5103         RETURN(rc);
5104 }
5105
5106 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
5107                                     enum ldlm_intent_flags it_opc)
5108 {
5109         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
5110
5111         /* update stats when IT code is known */
5112         if (srv_stats != NULL)
5113                 lprocfs_counter_incr(srv_stats,
5114                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
5115                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
5116 }
5117
5118 static int mdt_intent_policy(const struct lu_env *env,
5119                              struct ldlm_namespace *ns,
5120                              struct ldlm_lock **lockp,
5121                              void *req_cookie,
5122                              enum ldlm_mode mode,
5123                              __u64 flags, void *data)
5124 {
5125         struct tgt_session_info *tsi;
5126         struct mdt_thread_info  *info;
5127         struct ptlrpc_request   *req  =  req_cookie;
5128         struct ldlm_intent      *it;
5129         struct req_capsule      *pill;
5130         const struct ldlm_lock_desc *ldesc;
5131         int rc;
5132
5133         ENTRY;
5134
5135         LASSERT(req != NULL);
5136
5137         tsi = tgt_ses_info(env);
5138
5139         info = mdt_th_info(env);
5140         LASSERT(info != NULL);
5141
5142         /* Check whether it is a sub request processing in a batch request */
5143         if (info->mti_batch_env) {
5144                 pill = info->mti_pill;
5145                 LASSERT(pill == &info->mti_sub_pill);
5146         } else {
5147                 info = tsi2mdt_info(tsi);
5148                 pill = info->mti_pill;
5149         }
5150
5151         LASSERT(pill->rc_req == req);
5152         ldesc = &info->mti_dlm_req->lock_desc;
5153
5154         if (info->mti_batch_env ||
5155             req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
5156                 /*
5157                  * For batch processing environment, the request format has
5158                  * already been set.
5159                  */
5160                 if (!info->mti_batch_env)
5161                         req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
5162
5163                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
5164                 if (it != NULL) {
5165                         mdt_ptlrpc_stats_update(req, it->opc);
5166                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
5167                         if (rc == 0)
5168                                 rc = ELDLM_OK;
5169
5170                         /* Lock without inodebits makes no sense and will oops
5171                          * later in ldlm. Let's check it now to see if we have
5172                          * ibits corrupted somewhere in mdt_intent_opc().
5173                          * The case for client miss to set ibits has been
5174                          * processed by others. */
5175                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
5176                                 ldesc->l_policy_data.l_inodebits.bits != 0));
5177                 } else {
5178                         rc = err_serious(-EFAULT);
5179                 }
5180         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
5181                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
5182                 struct ldlm_reply *rep;
5183
5184                 /* No intent was provided but INTENT flag is set along with
5185                  * DOM bit, this is considered as GLIMPSE request.
5186                  * This logic is common for MDT and OST glimpse
5187                  */
5188                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
5189                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
5190                 /* Check whether the reply has been packed successfully. */
5191                 if (req->rq_repmsg != NULL) {
5192                         rep = req_capsule_server_get(info->mti_pill,
5193                                                      &RMF_DLM_REP);
5194                         rep->lock_policy_res2 =
5195                                 ptlrpc_status_hton(rep->lock_policy_res2);
5196                 }
5197         } else {
5198                 /* No intent was provided */
5199                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5200                 rc = req_capsule_server_pack(pill);
5201                 if (rc)
5202                         rc = err_serious(rc);
5203         }
5204
5205         if (!info->mti_batch_env)
5206                 mdt_thread_info_fini(info);
5207         RETURN(rc);
5208 }
5209
5210 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5211 {
5212         struct seq_server_site  *ss = mdt_seq_site(mdt);
5213
5214         if (ss->ss_node_id == 0)
5215                 return;
5216
5217         if (ss->ss_client_seq != NULL) {
5218                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5219                 ss->ss_client_seq->lcs_exp = NULL;
5220         }
5221
5222         if (ss->ss_server_fld != NULL) {
5223                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5224                 ss->ss_server_fld->lsf_control_exp = NULL;
5225         }
5226 }
5227
5228 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5229 {
5230         struct seq_server_site *ss = mdt_seq_site(mdt);
5231
5232         if (ss == NULL)
5233                 return;
5234
5235         if (ss->ss_server_seq != NULL)
5236                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5237 }
5238
5239 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5240 {
5241         mdt_seq_fini_cli(mdt);
5242         mdt_deregister_seq_exp(mdt);
5243
5244         return seq_site_fini(env, mdt_seq_site(mdt));
5245 }
5246
5247 /**
5248  * It will retrieve its FLDB entries from MDT0, and it only happens
5249  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5250  * and it needs to refresh FLDB from the MDT0.
5251  **/
5252 static int mdt_register_lwp_callback(void *data)
5253 {
5254         struct lu_env           env;
5255         struct mdt_device       *mdt = data;
5256         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5257         int                     rc;
5258         ENTRY;
5259
5260         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5261
5262         rc = lu_env_init(&env, LCT_MD_THREAD);
5263         if (rc < 0) {
5264                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5265                 RETURN(rc);
5266         }
5267
5268         /* Allocate new sequence now to avoid creating local transaction
5269          * in the normal transaction process */
5270         rc = seq_server_check_and_alloc_super(&env,
5271                                               mdt_seq_site(mdt)->ss_server_seq);
5272         if (rc < 0)
5273                 GOTO(out, rc);
5274
5275         if (fld->lsf_new) {
5276                 rc = fld_update_from_controller(&env, fld);
5277                 if (rc != 0) {
5278                         CERROR("%s: cannot update controller: rc = %d\n",
5279                                mdt_obd_name(mdt), rc);
5280                         GOTO(out, rc);
5281                 }
5282         }
5283 out:
5284         lu_env_fini(&env);
5285         RETURN(rc);
5286 }
5287
5288 static int mdt_register_seq_exp(struct mdt_device *mdt)
5289 {
5290         struct seq_server_site  *ss = mdt_seq_site(mdt);
5291         char                    *lwp_name = NULL;
5292         int                     rc;
5293
5294         if (ss->ss_node_id == 0)
5295                 return 0;
5296
5297         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5298         if (lwp_name == NULL)
5299                 GOTO(out_free, rc = -ENOMEM);
5300
5301         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5302         if (rc != 0)
5303                 GOTO(out_free, rc);
5304
5305         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5306                                       NULL, NULL);
5307         if (rc != 0)
5308                 GOTO(out_free, rc);
5309
5310         rc = lustre_register_lwp_item(lwp_name,
5311                                       &ss->ss_server_fld->lsf_control_exp,
5312                                       mdt_register_lwp_callback, mdt);
5313         if (rc != 0) {
5314                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5315                 ss->ss_client_seq->lcs_exp = NULL;
5316                 GOTO(out_free, rc);
5317         }
5318 out_free:
5319         if (lwp_name != NULL)
5320                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5321
5322         return rc;
5323 }
5324
5325 /*
5326  * Init client sequence manager which is used by local MDS to talk to sequence
5327  * controller on remote node.
5328  */
5329 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5330 {
5331         struct seq_server_site *ss = mdt_seq_site(mdt);
5332         char *prefix;
5333         ENTRY;
5334
5335         /* check if this is adding the first MDC and controller is not yet
5336          * initialized. */
5337         OBD_ALLOC_PTR(ss->ss_client_seq);
5338         if (ss->ss_client_seq == NULL)
5339                 RETURN(-ENOMEM);
5340
5341         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5342         if (prefix == NULL) {
5343                 OBD_FREE_PTR(ss->ss_client_seq);
5344                 ss->ss_client_seq = NULL;
5345                 RETURN(-ENOMEM);
5346         }
5347
5348         /* Note: seq_client_fini will be called in seq_site_fini */
5349         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5350         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5351                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5352                                                             NULL);
5353         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5354
5355         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5356 }
5357
5358 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5359 {
5360         struct seq_server_site  *ss;
5361         int                     rc;
5362         ENTRY;
5363
5364         ss = mdt_seq_site(mdt);
5365         /* init sequence controller server(MDT0) */
5366         if (ss->ss_node_id == 0) {
5367                 OBD_ALLOC_PTR(ss->ss_control_seq);
5368                 if (ss->ss_control_seq == NULL)
5369                         RETURN(-ENOMEM);
5370
5371                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5372                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5373                                      ss);
5374                 if (rc)
5375                         GOTO(out_seq_fini, rc);
5376         }
5377
5378         /* Init normal sequence server */
5379         OBD_ALLOC_PTR(ss->ss_server_seq);
5380         if (ss->ss_server_seq == NULL)
5381                 GOTO(out_seq_fini, rc = -ENOMEM);
5382
5383         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5384                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5385         if (rc)
5386                 GOTO(out_seq_fini, rc);
5387
5388         /* init seq client for seq server to talk to seq controller(MDT0) */
5389         rc = mdt_seq_init_cli(env, mdt);
5390         if (rc != 0)
5391                 GOTO(out_seq_fini, rc);
5392
5393         if (ss->ss_node_id != 0)
5394                 /* register controller export through lwp */
5395                 rc = mdt_register_seq_exp(mdt);
5396
5397         EXIT;
5398 out_seq_fini:
5399         if (rc)
5400                 mdt_seq_fini(env, mdt);
5401
5402         return rc;
5403 }
5404
5405 /*
5406  * FLD wrappers
5407  */
5408 static int mdt_fld_fini(const struct lu_env *env,
5409                         struct mdt_device *m)
5410 {
5411         struct seq_server_site *ss = mdt_seq_site(m);
5412         ENTRY;
5413
5414         if (ss && ss->ss_server_fld) {
5415                 fld_server_fini(env, ss->ss_server_fld);
5416                 OBD_FREE_PTR(ss->ss_server_fld);
5417                 ss->ss_server_fld = NULL;
5418         }
5419
5420         RETURN(0);
5421 }
5422
5423 static int mdt_fld_init(const struct lu_env *env,
5424                         const char *uuid,
5425                         struct mdt_device *m)
5426 {
5427         struct seq_server_site *ss;
5428         int rc;
5429         ENTRY;
5430
5431         ss = mdt_seq_site(m);
5432
5433         OBD_ALLOC_PTR(ss->ss_server_fld);
5434         if (ss->ss_server_fld == NULL)
5435                 RETURN(rc = -ENOMEM);
5436
5437         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5438                              LU_SEQ_RANGE_MDT);
5439         if (rc) {
5440                 OBD_FREE_PTR(ss->ss_server_fld);
5441                 ss->ss_server_fld = NULL;
5442                 RETURN(rc);
5443         }
5444
5445         RETURN(0);
5446 }
5447
5448 static void mdt_stack_pre_fini(const struct lu_env *env,
5449                            struct mdt_device *m, struct lu_device *top)
5450 {
5451         struct lustre_cfg_bufs  *bufs;
5452         struct lustre_cfg       *lcfg;
5453         struct mdt_thread_info  *info;
5454         ENTRY;
5455
5456         LASSERT(top);
5457
5458         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5459         LASSERT(info != NULL);
5460
5461         bufs = &info->mti_u.bufs;
5462
5463         LASSERT(m->mdt_child_exp);
5464         LASSERT(m->mdt_child_exp->exp_obd);
5465
5466         /* process cleanup, pass mdt obd name to get obd umount flags */
5467         /* XXX: this is needed because all layers are referenced by
5468          * objects (some of them are pinned by osd, for example *
5469          * the proper solution should be a model where object used
5470          * by osd only doesn't have mdt/mdd slices -bzzz */
5471         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5472         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5473         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5474         if (!lcfg)
5475                 RETURN_EXIT;
5476         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5477
5478         top->ld_ops->ldo_process_config(env, top, lcfg);
5479         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5480         EXIT;
5481 }
5482
5483 static void mdt_stack_fini(const struct lu_env *env,
5484                            struct mdt_device *m, struct lu_device *top)
5485 {
5486         struct obd_device       *obd = mdt2obd_dev(m);
5487         struct lustre_cfg_bufs  *bufs;
5488         struct lustre_cfg       *lcfg;
5489         struct mdt_thread_info  *info;
5490         char                     flags[3] = "";
5491         ENTRY;
5492
5493         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5494         LASSERT(info != NULL);
5495
5496         lu_dev_del_linkage(top->ld_site, top);
5497
5498         lu_site_purge(env, top->ld_site, -1);
5499
5500         bufs = &info->mti_u.bufs;
5501         /* process cleanup, pass mdt obd name to get obd umount flags */
5502         /* another purpose is to let all layers to release their objects */
5503         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5504         if (obd->obd_force)
5505                 strcat(flags, "F");
5506         if (obd->obd_fail)
5507                 strcat(flags, "A");
5508         lustre_cfg_bufs_set_string(bufs, 1, flags);
5509         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5510         if (!lcfg)
5511                 RETURN_EXIT;
5512         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5513
5514         LASSERT(top);
5515         top->ld_ops->ldo_process_config(env, top, lcfg);
5516         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5517
5518         lu_site_purge(env, top->ld_site, -1);
5519
5520         m->mdt_child = NULL;
5521         m->mdt_bottom = NULL;
5522
5523         obd_disconnect(m->mdt_child_exp);
5524         m->mdt_child_exp = NULL;
5525
5526         obd_disconnect(m->mdt_bottom_exp);
5527         m->mdt_child_exp = NULL;
5528 }
5529
5530 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5531                                const char *next, struct obd_export **exp)
5532 {
5533         struct obd_connect_data *data = NULL;
5534         struct obd_device       *obd;
5535         int                      rc;
5536         ENTRY;
5537
5538         OBD_ALLOC_PTR(data);
5539         if (data == NULL)
5540                 GOTO(out, rc = -ENOMEM);
5541
5542         obd = class_name2obd(next);
5543         if (obd == NULL) {
5544                 CERROR("%s: can't locate next device: %s\n",
5545                        mdt_obd_name(m), next);
5546                 GOTO(out, rc = -ENOTCONN);
5547         }
5548
5549         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5550         data->ocd_version = LUSTRE_VERSION_CODE;
5551
5552         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5553         if (rc) {
5554                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5555                        mdt_obd_name(m), next, rc);
5556                 GOTO(out, rc);
5557         }
5558
5559 out:
5560         if (data)
5561                 OBD_FREE_PTR(data);
5562         RETURN(rc);
5563 }
5564
5565 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5566                           struct lustre_cfg *cfg)
5567 {
5568         char                   *dev = lustre_cfg_string(cfg, 0);
5569         int                     rc, name_size, uuid_size;
5570         char                   *name, *uuid, *p;
5571         struct lustre_cfg_bufs *bufs;
5572         struct lustre_cfg      *lcfg;
5573         struct obd_device      *obd;
5574         struct lustre_profile  *lprof;
5575         struct lu_site         *site;
5576         ENTRY;
5577
5578         /* in 1.8 we had the only device in the stack - MDS.
5579          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5580          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5581          * to generate names and setup MDT, MDD. MDT will be using
5582          * generated name to connect to MDD. for MDD the next device
5583          * will be LOD with name taken from so called "profile" which
5584          * is generated by mount_option line
5585          *
5586          * 1.8 MGS generates config. commands like this:
5587          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5588          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5589          * 2.0 MGS generates config. commands like this:
5590          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5591          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5592          *                    3:lustre-MDT0000-mdtlov  4:f
5593          *
5594          * we generate MDD name from MDT one, just replacing T with D
5595          *
5596          * after all the preparations, the logical equivalent will be
5597          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5598          *                    3:lustre-MDT0000-mdtlov  4:f
5599          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5600          *                    3:lustre-MDD0000  4:f
5601          *
5602          *  notice we build the stack from down to top: MDD first, then MDT */
5603
5604         name_size = MAX_OBD_NAME;
5605         uuid_size = MAX_OBD_NAME;
5606
5607         OBD_ALLOC(name, name_size);
5608         OBD_ALLOC(uuid, uuid_size);
5609         if (name == NULL || uuid == NULL)
5610                 GOTO(cleanup_mem, rc = -ENOMEM);
5611
5612         OBD_ALLOC_PTR(bufs);
5613         if (!bufs)
5614                 GOTO(cleanup_mem, rc = -ENOMEM);
5615
5616         strcpy(name, dev);
5617         p = strstr(name, "-MDT");
5618         if (p == NULL)
5619                 GOTO(free_bufs, rc = -ENOMEM);
5620         p[3] = 'D';
5621
5622         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5623
5624         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5625         if (lprof == NULL || lprof->lp_dt == NULL) {
5626                 CERROR("can't find the profile: %s\n",
5627                        lustre_cfg_string(cfg, 0));
5628                 GOTO(free_bufs, rc = -EINVAL);
5629         }
5630
5631         lustre_cfg_bufs_reset(bufs, name);
5632         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5633         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5634         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5635
5636         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5637         if (!lcfg)
5638                 GOTO(put_profile, rc = -ENOMEM);
5639         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5640
5641         rc = class_attach(lcfg);
5642         if (rc)
5643                 GOTO(lcfg_cleanup, rc);
5644
5645         obd = class_name2obd(name);
5646         if (!obd) {
5647                 CERROR("Can not find obd %s (%s in config)\n",
5648                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5649                 GOTO(lcfg_cleanup, rc = -EINVAL);
5650         }
5651
5652         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5653
5654         lustre_cfg_bufs_reset(bufs, name);
5655         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5656         lustre_cfg_bufs_set_string(bufs, 2, dev);
5657         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5658
5659         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5660         if (!lcfg)
5661                 GOTO(class_detach, rc = -ENOMEM);
5662         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5663
5664         rc = class_setup(obd, lcfg);
5665         if (rc)
5666                 GOTO(class_detach, rc);
5667
5668         /* connect to MDD we just setup */
5669         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5670         if (rc)
5671                 GOTO(class_detach, rc);
5672
5673         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5674         LASSERT(site);
5675         LASSERT(mdt_lu_site(mdt) == NULL);
5676         mdt->mdt_lu_dev.ld_site = site;
5677         site->ls_top_dev = &mdt->mdt_lu_dev;
5678         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5679
5680         /* now connect to bottom OSD */
5681         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5682         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5683         if (rc)
5684                 GOTO(class_detach, rc);
5685         mdt->mdt_bottom =
5686                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5687
5688         rc = lu_env_refill((struct lu_env *)env);
5689         if (rc != 0)
5690                 CERROR("Failure to refill session: '%d'\n", rc);
5691
5692         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5693
5694         EXIT;
5695 class_detach:
5696         if (rc)
5697                 class_detach(obd, lcfg);
5698 lcfg_cleanup:
5699         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5700 put_profile:
5701         class_put_profile(lprof);
5702 free_bufs:
5703         OBD_FREE_PTR(bufs);
5704 cleanup_mem:
5705         if (name)
5706                 OBD_FREE(name, name_size);
5707         if (uuid)
5708                 OBD_FREE(uuid, uuid_size);
5709         RETURN(rc);
5710 }
5711
5712 /* setup quota master target on MDT0 */
5713 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5714                           struct lustre_cfg *cfg)
5715 {
5716         struct obd_device       *obd;
5717         char                    *dev = lustre_cfg_string(cfg, 0);
5718         char                    *qmtname, *uuid, *p;
5719         struct lustre_cfg_bufs  *bufs;
5720         struct lustre_cfg       *lcfg;
5721         struct lustre_profile   *lprof;
5722         struct obd_connect_data *data;
5723         int                      rc;
5724         ENTRY;
5725
5726         LASSERT(mdt->mdt_qmt_exp == NULL);
5727         LASSERT(mdt->mdt_qmt_dev == NULL);
5728
5729         /* quota master is on MDT0 only for now */
5730         if (mdt->mdt_seq_site.ss_node_id != 0)
5731                 RETURN(0);
5732
5733         /* MGS generates config commands which look as follows:
5734          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5735          *                    3:lustre-MDT0000-mdtlov  4:f
5736          *
5737          * We generate the QMT name from the MDT one, just replacing MD with QM
5738          * after all the preparations, the logical equivalent will be:
5739          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5740          *                    3:lustre-MDT0000-osd  4:f */
5741         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5742         OBD_ALLOC(uuid, UUID_MAX);
5743         OBD_ALLOC_PTR(bufs);
5744         OBD_ALLOC_PTR(data);
5745         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5746                 GOTO(cleanup_mem, rc = -ENOMEM);
5747
5748         strcpy(qmtname, dev);
5749         p = strstr(qmtname, "-MDT");
5750         if (p == NULL)
5751                 GOTO(cleanup_mem, rc = -ENOMEM);
5752         /* replace MD with QM */
5753         p[1] = 'Q';
5754         p[2] = 'M';
5755
5756         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5757
5758         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5759         if (lprof == NULL || lprof->lp_dt == NULL) {
5760                 CERROR("can't find profile for %s\n",
5761                        lustre_cfg_string(cfg, 0));
5762                 GOTO(cleanup_mem, rc = -EINVAL);
5763         }
5764
5765         lustre_cfg_bufs_reset(bufs, qmtname);
5766         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5767         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5768         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5769
5770         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5771         if (!lcfg)
5772                 GOTO(put_profile, rc = -ENOMEM);
5773         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5774
5775         rc = class_attach(lcfg);
5776         if (rc)
5777                 GOTO(lcfg_cleanup, rc);
5778
5779         obd = class_name2obd(qmtname);
5780         if (!obd) {
5781                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5782                        lustre_cfg_string(cfg, 0));
5783                 GOTO(lcfg_cleanup, rc = -EINVAL);
5784         }
5785
5786         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5787
5788         lustre_cfg_bufs_reset(bufs, qmtname);
5789         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5790         lustre_cfg_bufs_set_string(bufs, 2, dev);
5791
5792         /* for quota, the next device should be the OSD device */
5793         lustre_cfg_bufs_set_string(bufs, 3,
5794                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5795
5796         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5797         if (!lcfg)
5798                 GOTO(class_detach, rc = -ENOMEM);
5799         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5800
5801         rc = class_setup(obd, lcfg);
5802         if (rc)
5803                 GOTO(class_detach, rc);
5804
5805         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5806
5807         /* configure local quota objects */
5808         if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5809                 rc = -EBADF;
5810         else
5811                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5812                                                            &mdt->mdt_lu_dev,
5813                                                            mdt->mdt_qmt_dev);
5814         if (rc)
5815                 GOTO(class_cleanup, rc);
5816
5817         /* connect to quota master target */
5818         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5819         data->ocd_version = LUSTRE_VERSION_CODE;
5820         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5821                          data, NULL);
5822         if (rc) {
5823                 CERROR("cannot connect to quota master device %s (%d)\n",
5824                        qmtname, rc);
5825                 GOTO(class_cleanup, rc);
5826         }
5827
5828         EXIT;
5829 class_cleanup:
5830         if (rc) {
5831                 class_manual_cleanup(obd);
5832                 mdt->mdt_qmt_dev = NULL;
5833                 GOTO(lcfg_cleanup, rc);
5834         }
5835 class_detach:
5836         if (rc)
5837                 class_detach(obd, lcfg);
5838 lcfg_cleanup:
5839         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5840 put_profile:
5841         class_put_profile(lprof);
5842 cleanup_mem:
5843         if (bufs)
5844                 OBD_FREE_PTR(bufs);
5845         if (qmtname)
5846                 OBD_FREE(qmtname, MAX_OBD_NAME);
5847         if (uuid)
5848                 OBD_FREE(uuid, UUID_MAX);
5849         if (data)
5850                 OBD_FREE_PTR(data);
5851         return rc;
5852 }
5853
5854 /* Shutdown quota master target associated with mdt */
5855 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5856 {
5857         ENTRY;
5858
5859         if (mdt->mdt_qmt_exp == NULL)
5860                 RETURN_EXIT;
5861         LASSERT(mdt->mdt_qmt_dev != NULL);
5862
5863         /* the qmt automatically shuts down when the mdt disconnects */
5864         obd_disconnect(mdt->mdt_qmt_exp);
5865         mdt->mdt_qmt_exp = NULL;
5866         mdt->mdt_qmt_dev = NULL;
5867         EXIT;
5868 }
5869
5870 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5871  * for now. This will be removed along with converting rest of MDT code
5872  * to use tgt_session_info */
5873 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5874 {
5875         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5876         int                      rc;
5877
5878         if (unlikely(info->mti_object == NULL))
5879                 return -EPROTO;
5880
5881         rc = mdt_getxattr(info);
5882
5883         mdt_thread_info_fini(info);
5884         return rc;
5885 }
5886
5887 static int mdt_llog_open(struct tgt_session_info *tsi)
5888 {
5889         ENTRY;
5890
5891         if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
5892                 RETURN(err_serious(-EACCES));
5893
5894         RETURN(tgt_llog_open(tsi));
5895 }
5896
5897 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5898 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5899 #define OST_BRW_READ    OST_READ
5900 #define OST_BRW_WRITE   OST_WRITE
5901
5902 static struct tgt_handler mdt_tgt_handlers[] = {
5903 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5904                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5905                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5906 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5907                 0,                      MDS_DISCONNECT, tgt_disconnect,
5908                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5909 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5910                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5911                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5912 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5913 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5914 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5915 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5916                                                         mdt_getattr_name),
5917 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5918 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5919 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5920 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5921 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5922 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5923 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5924 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5925                                                         mdt_hsm_progress),
5926 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5927                                                         mdt_hsm_ct_register),
5928 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5929                                                         mdt_hsm_ct_unregister),
5930 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5931                                                         mdt_hsm_state_get),
5932 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5933                                                         mdt_hsm_state_set),
5934 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5935 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5936                                                         mdt_hsm_request),
5937 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5938             MDS_SWAP_LAYOUTS,
5939             mdt_swap_layouts),
5940 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5941 TGT_MDT_HDL(IS_MUTABLE,         MDS_BATCH,      mdt_batch),
5942 };
5943
5944 static struct tgt_handler mdt_io_ops[] = {
5945 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5946                                                         mdt_hp_brw),
5947 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5948                                                         mdt_hp_brw),
5949 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5950                                          OST_PUNCH,     mdt_punch_hdl,
5951                                                         mdt_hp_punch),
5952 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5953 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5954                                                         mdt_fallocate_hdl),
5955 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5956 TGT_RPC_HANDLER(OST_FIRST_OPC,
5957                 0,                      OST_SET_INFO,   mdt_io_set_info,
5958                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
5959 };
5960
5961 static struct tgt_handler mdt_sec_ctx_ops[] = {
5962 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5963 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5964 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5965 };
5966
5967 static struct tgt_handler mdt_quota_ops[] = {
5968 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5969 };
5970
5971 static struct tgt_handler mdt_llog_handlers[] = {
5972         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5973         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5974         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5975         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5976 };
5977
5978 static struct tgt_opc_slice mdt_common_slice[] = {
5979         {
5980                 .tos_opc_start  = MDS_FIRST_OPC,
5981                 .tos_opc_end    = MDS_LAST_OPC,
5982                 .tos_hs         = mdt_tgt_handlers
5983         },
5984         {
5985                 .tos_opc_start  = OBD_FIRST_OPC,
5986                 .tos_opc_end    = OBD_LAST_OPC,
5987                 .tos_hs         = tgt_obd_handlers
5988         },
5989         {
5990                 .tos_opc_start  = LDLM_FIRST_OPC,
5991                 .tos_opc_end    = LDLM_LAST_OPC,
5992                 .tos_hs         = tgt_dlm_handlers
5993         },
5994         {
5995                 .tos_opc_start  = SEC_FIRST_OPC,
5996                 .tos_opc_end    = SEC_LAST_OPC,
5997                 .tos_hs         = mdt_sec_ctx_ops
5998         },
5999         {
6000                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
6001                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
6002                 .tos_hs         = tgt_out_handlers
6003         },
6004         {
6005                 .tos_opc_start  = FLD_FIRST_OPC,
6006                 .tos_opc_end    = FLD_LAST_OPC,
6007                 .tos_hs         = fld_handlers
6008         },
6009         {
6010                 .tos_opc_start  = SEQ_FIRST_OPC,
6011                 .tos_opc_end    = SEQ_LAST_OPC,
6012                 .tos_hs         = seq_handlers
6013         },
6014         {
6015                 .tos_opc_start  = QUOTA_DQACQ,
6016                 .tos_opc_end    = QUOTA_LAST_OPC,
6017                 .tos_hs         = mdt_quota_ops
6018         },
6019         {
6020                 .tos_opc_start  = LLOG_FIRST_OPC,
6021                 .tos_opc_end    = LLOG_LAST_OPC,
6022                 .tos_hs         = mdt_llog_handlers
6023         },
6024         {
6025                 .tos_opc_start  = LFSCK_FIRST_OPC,
6026                 .tos_opc_end    = LFSCK_LAST_OPC,
6027                 .tos_hs         = tgt_lfsck_handlers
6028         },
6029         {
6030                 .tos_opc_start  = OST_FIRST_OPC,
6031                 .tos_opc_end    = OST_LAST_OPC,
6032                 .tos_hs         = mdt_io_ops
6033         },
6034         {
6035                 .tos_hs         = NULL
6036         }
6037 };
6038
6039 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
6040 {
6041         struct md_device *next = m->mdt_child;
6042         struct lu_device *d = &m->mdt_lu_dev;
6043         struct obd_device *obd = mdt2obd_dev(m);
6044         struct lfsck_stop stop;
6045
6046         ENTRY;
6047         stop.ls_status = LS_PAUSED;
6048         stop.ls_flags = 0;
6049         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
6050
6051         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
6052
6053         mdt_restriper_stop(m);
6054         ping_evictor_stop();
6055
6056         /* Remove the HSM /proc entry so the coordinator cannot be
6057          * restarted by a user while it's shutting down.
6058          */
6059         mdt_hsm_cdt_stop(m);
6060
6061         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
6062         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
6063
6064         if (m->mdt_namespace != NULL)
6065                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
6066                                           d->ld_obd->obd_force);
6067
6068         obd_exports_barrier(obd);
6069         obd_zombie_barrier();
6070
6071         mdt_quota_fini(env, m);
6072
6073         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
6074
6075         /* Calling the cleanup functions in the same order as in the mdt_init0
6076          * error path
6077          */
6078         mdt_tunables_fini(m);
6079
6080         target_recovery_fini(obd);
6081         upcall_cache_cleanup(m->mdt_identity_cache);
6082         m->mdt_identity_cache = NULL;
6083
6084         tgt_fini(env, &m->mdt_lut);
6085
6086         mdt_hsm_cdt_fini(m);
6087
6088         if (m->mdt_los != NULL) {
6089                 local_oid_storage_fini(env, m->mdt_los);
6090                 m->mdt_los = NULL;
6091         }
6092
6093         if (m->mdt_namespace != NULL) {
6094                 ldlm_namespace_free_post(m->mdt_namespace);
6095                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
6096         }
6097
6098         if (m->mdt_md_root != NULL) {
6099                 mdt_object_put(env, m->mdt_md_root);
6100                 m->mdt_md_root = NULL;
6101         }
6102
6103         mdt_seq_fini(env, m);
6104
6105         mdt_fld_fini(env, m);
6106
6107         /*
6108          * Finish the stack
6109          */
6110         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6111
6112         LASSERT(atomic_read(&d->ld_ref) == 0);
6113
6114         server_put_mount(mdt_obd_name(m), true);
6115
6116         EXIT;
6117 }
6118
6119 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
6120
6121 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
6122                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
6123 {
6124         const struct dt_device_param *dt_conf;
6125         struct mdt_thread_info *info;
6126         struct obd_device *obd;
6127         const char *dev = lustre_cfg_string(cfg, 0);
6128         const char *num = lustre_cfg_string(cfg, 2);
6129         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
6130         struct lustre_mount_info *lmi = NULL;
6131         struct lustre_sb_info *lsi;
6132         struct lu_site *s;
6133         struct seq_server_site *ss_site;
6134         const char *identity_upcall = "NONE";
6135         struct md_device *next;
6136         struct lu_fid fid;
6137         int rc;
6138         long node_id;
6139         mntopt_t mntopts;
6140         ENTRY;
6141
6142         lu_device_init(&m->mdt_lu_dev, ldt);
6143         /*
6144          * Environment (env) might be missing mdt_thread_key values at that
6145          * point, if device is allocated when mdt_thread_key is in QUIESCENT
6146          * mode.
6147          *
6148          * Usually device allocation path doesn't use module key values, but
6149          * mdt has to do a lot of work here, so allocate key value.
6150          */
6151         rc = lu_env_refill((struct lu_env *)env);
6152         if (rc != 0)
6153                 RETURN(rc);
6154
6155         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
6156         LASSERT(info != NULL);
6157
6158         obd = class_name2obd(dev);
6159         LASSERT(obd != NULL);
6160
6161         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
6162         m->mdt_opts.mo_evict_tgt_nids = 1;
6163         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
6164
6165         lmi = server_get_mount(dev);
6166         if (lmi == NULL) {
6167                 CERROR("Cannot get mount info for %s!\n", dev);
6168                 RETURN(-EFAULT);
6169         } else {
6170                 lsi = s2lsi(lmi->lmi_sb);
6171                 LASSERT(lsi->lsi_lmd);
6172                 /* CMD is supported only in IAM mode */
6173                 LASSERT(num);
6174                 rc = kstrtol(num, 10, &node_id);
6175                 if (rc)
6176                         RETURN(rc);
6177
6178                 obd_obt_init(obd);
6179                 if (test_bit(LMD_FLG_SKIP_LFSCK, lsi->lsi_lmd->lmd_flags))
6180                         m->mdt_skip_lfsck = 1;
6181         }
6182
6183         /* Just try to get a DoM lock by default. Otherwise, having a group
6184          * lock granted, it may get blocked for a long time. */
6185         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
6186         /* DoM files are read at open and data is packed in the reply */
6187         m->mdt_opts.mo_dom_read_open = 1;
6188
6189         m->mdt_squash.rsi_uid = 0;
6190         m->mdt_squash.rsi_gid = 0;
6191         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
6192         spin_lock_init(&m->mdt_squash.rsi_lock);
6193         spin_lock_init(&m->mdt_lock);
6194         m->mdt_enable_chprojid_gid = 0;
6195         m->mdt_enable_dir_migration = 1;
6196         m->mdt_enable_dir_restripe = 0;
6197         m->mdt_enable_dir_auto_split = 0;
6198         m->mdt_enable_parallel_rename_dir = 1;
6199         m->mdt_enable_parallel_rename_file = 1;
6200         m->mdt_enable_remote_dir = 1;
6201         m->mdt_enable_remote_dir_gid = 0;
6202         m->mdt_enable_remote_rename = 1;
6203         m->mdt_enable_striped_dir = 1;
6204         m->mdt_enable_dmv_implicit_inherit = 1;
6205         m->mdt_dir_restripe_nsonly = 1;
6206         m->mdt_max_mod_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
6207
6208         atomic_set(&m->mdt_mds_mds_conns, 0);
6209         atomic_set(&m->mdt_async_commit_count, 0);
6210         atomic_set(&m->mdt_dmv_old_client_count, 0);
6211
6212         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6213         m->mdt_lu_dev.ld_obd = obd;
6214         /* Set this lu_device to obd for error handling purposes. */
6215         obd->obd_lu_dev = &m->mdt_lu_dev;
6216
6217         strncpy(m->mdt_job_xattr, XATTR_NAME_JOB_DEFAULT, XATTR_JOB_MAX_LEN);
6218
6219         /* init the stack */
6220         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6221         if (rc) {
6222                 CERROR("%s: Can't init device stack, rc %d\n",
6223                        mdt_obd_name(m), rc);
6224                 GOTO(err_lmi, rc);
6225         }
6226
6227         s = mdt_lu_site(m);
6228         ss_site = mdt_seq_site(m);
6229         s->ld_seq_site = ss_site;
6230         ss_site->ss_lu = s;
6231
6232         /* set server index */
6233         ss_site->ss_node_id = node_id;
6234
6235         /* failover is the default
6236          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6237          * assumed whose ss_node_id == 0 XXX
6238          * */
6239         obd->obd_replayable = 1;
6240         /* No connection accepted until configurations will finish */
6241         obd->obd_no_conn = 1;
6242
6243         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6244                 char *str = lustre_cfg_string(cfg, 4);
6245                 if (strchr(str, 'n')) {
6246                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6247                         obd->obd_replayable = 0;
6248                 }
6249         }
6250
6251         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6252         if (rc)
6253                 GOTO(err_fini_stack, rc);
6254
6255         rc = mdt_seq_init(env, m);
6256         if (rc)
6257                 GOTO(err_fini_fld, rc);
6258
6259         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6260                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6261         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6262                                               LDLM_NAMESPACE_SERVER,
6263                                               LDLM_NAMESPACE_GREEDY,
6264                                               LDLM_NS_TYPE_MDT);
6265         if (IS_ERR(m->mdt_namespace)) {
6266                 rc = PTR_ERR(m->mdt_namespace);
6267                 CERROR("%s: unable to create server namespace: rc = %d\n",
6268                        obd->obd_name, rc);
6269                 m->mdt_namespace = NULL;
6270                 GOTO(err_fini_seq, rc);
6271         }
6272
6273         m->mdt_namespace->ns_lvbp = m;
6274         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6275
6276         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6277         /* set obd_namespace for compatibility with old code */
6278         obd->obd_namespace = m->mdt_namespace;
6279
6280         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6281                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6282                       OBD_FAIL_MDS_ALL_REPLY_NET);
6283         if (rc)
6284                 GOTO(err_free_ns, rc);
6285
6286         /* Amount of available space excluded from granting and reserved
6287          * for metadata. It is a percentage of the total MDT size. */
6288         tgd->tgd_reserved_pcnt = 10;
6289
6290         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6291                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6292         else
6293                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6294
6295         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6296                 GOTO(err_tgt, rc = -ENOENT);
6297
6298         fid.f_seq = FID_SEQ_LOCAL_NAME;
6299         fid.f_oid = 1;
6300         fid.f_ver = 0;
6301         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6302         if (rc != 0)
6303                 GOTO(err_tgt, rc);
6304
6305         rc = mdt_hsm_cdt_init(m);
6306         if (rc != 0) {
6307                 CERROR("%s: error initializing coordinator, rc %d\n",
6308                        mdt_obd_name(m), rc);
6309                 GOTO(err_los_fini, rc);
6310         }
6311
6312         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6313
6314         next = m->mdt_child;
6315         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6316
6317         mntopts = dt_conf->ddp_mntopts;
6318
6319         if (mntopts & MNTOPT_USERXATTR)
6320                 m->mdt_opts.mo_user_xattr = 1;
6321         else
6322                 m->mdt_opts.mo_user_xattr = 0;
6323
6324         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6325
6326         if (mntopts & MNTOPT_ACL)
6327                 m->mdt_opts.mo_acl = 1;
6328         else
6329                 m->mdt_opts.mo_acl = 0;
6330
6331         m->mdt_opts.mo_enable_strict_som = 1;
6332
6333         /* XXX: to support suppgid for ACL, we enable identity_upcall
6334          * by default, otherwise, maybe got unexpected -EACCESS. */
6335         if (m->mdt_opts.mo_acl)
6336                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6337
6338         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6339                                                 identity_upcall,
6340                                                 UC_IDCACHE_HASH_SIZE,
6341                                                 &mdt_identity_upcall_cache_ops);
6342         if (IS_ERR(m->mdt_identity_cache)) {
6343                 rc = PTR_ERR(m->mdt_identity_cache);
6344                 m->mdt_identity_cache = NULL;
6345                 GOTO(err_free_hsm, rc);
6346         }
6347
6348         rc = mdt_tunables_init(m, dev);
6349         if (rc) {
6350                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6351                 GOTO(err_recovery, rc);
6352         }
6353
6354         rc = mdt_quota_init(env, m, cfg);
6355         if (rc)
6356                 GOTO(err_procfs, rc);
6357
6358         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6359         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6360                            "mdt_ldlm_client", m->mdt_ldlm_client);
6361
6362         ping_evictor_start();
6363
6364         /* recovery will be started upon mdt_prepare()
6365          * when the whole stack is complete and ready
6366          * to serve the requests */
6367
6368         /* Reduce the initial timeout on an MDS because it doesn't need such
6369          * a long timeout as an OST does. Adaptive timeouts will adjust this
6370          * value appropriately. */
6371         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6372                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6373
6374         if (test_bit(LMD_FLG_LOCAL_RECOV, lsi->lsi_lmd->lmd_flags))
6375                 m->mdt_lut.lut_local_recovery = 1;
6376
6377         rc = mdt_restriper_start(m);
6378         if (rc)
6379                 GOTO(err_ping_evictor, rc);
6380
6381         RETURN(0);
6382
6383 err_ping_evictor:
6384         ping_evictor_stop();
6385 err_procfs:
6386         mdt_tunables_fini(m);
6387 err_recovery:
6388         upcall_cache_cleanup(m->mdt_identity_cache);
6389         m->mdt_identity_cache = NULL;
6390 err_free_hsm:
6391         mdt_hsm_cdt_fini(m);
6392 err_los_fini:
6393         local_oid_storage_fini(env, m->mdt_los);
6394         m->mdt_los = NULL;
6395 err_tgt:
6396         /* keep recoverable clients */
6397         obd->obd_fail = 1;
6398         target_recovery_fini(obd);
6399         obd_exports_barrier(obd);
6400         obd_zombie_barrier();
6401         tgt_fini(env, &m->mdt_lut);
6402 err_free_ns:
6403         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6404         obd->obd_namespace = m->mdt_namespace = NULL;
6405 err_fini_seq:
6406         mdt_seq_fini(env, m);
6407 err_fini_fld:
6408         mdt_fld_fini(env, m);
6409 err_fini_stack:
6410         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6411 err_lmi:
6412         if (lmi)
6413                 server_put_mount(dev, true);
6414         return(rc);
6415 }
6416
6417 /* For interoperability, the left element is old parameter, the right one
6418  * is the new version of the parameter, if some parameter is deprecated,
6419  * the new version should be set as NULL. */
6420 static struct cfg_interop_param mdt_interop_param[] = {
6421         { "mdt.group_upcall",   NULL },
6422         { "mdt.quota_type",     NULL },
6423         { "mdd.quota_type",     NULL },
6424         { "mdt.som",            NULL },
6425         { "mdt.rootsquash",     "mdt.root_squash" },
6426         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6427         { NULL }
6428 };
6429
6430 /* used by MGS to process specific configurations */
6431 static int mdt_process_config(const struct lu_env *env,
6432                               struct lu_device *d, struct lustre_cfg *cfg)
6433 {
6434         struct mdt_device *m = mdt_dev(d);
6435         struct md_device *md_next = m->mdt_child;
6436         struct lu_device *next = md2lu_dev(md_next);
6437         int rc;
6438         ENTRY;
6439
6440         switch (cfg->lcfg_command) {
6441         case LCFG_PARAM: {
6442                 struct obd_device *obd = d->ld_obd;
6443                 /* For interoperability */
6444                 struct cfg_interop_param *ptr = NULL;
6445                 struct lustre_cfg *old_cfg = NULL;
6446                 char *param = NULL;
6447                 ssize_t count;
6448
6449                 param = lustre_cfg_string(cfg, 1);
6450                 if (param == NULL) {
6451                         CERROR("param is empty\n");
6452                         rc = -EINVAL;
6453                         break;
6454                 }
6455
6456                 ptr = class_find_old_param(param, mdt_interop_param);
6457                 if (ptr != NULL) {
6458                         if (ptr->new_param == NULL) {
6459                                 rc = 0;
6460                                 CWARN("For interoperability, skip this %s."
6461                                       " It is obsolete.\n", ptr->old_param);
6462                                 break;
6463                         }
6464
6465                         CWARN("Found old param %s, changed it to %s.\n",
6466                               ptr->old_param, ptr->new_param);
6467
6468                         old_cfg = cfg;
6469                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6470                         if (IS_ERR(cfg)) {
6471                                 rc = PTR_ERR(cfg);
6472                                 break;
6473                         }
6474                 }
6475
6476                 count = class_modify_config(cfg, PARAM_MDT,
6477                                             &obd->obd_kset.kobj);
6478                 if (count < 0) {
6479                         struct coordinator *cdt = &m->mdt_coordinator;
6480
6481                         /* is it an HSM var ? */
6482                         count = class_modify_config(cfg, PARAM_HSM,
6483                                                     &cdt->cdt_hsm_kobj);
6484                         if (count < 0)
6485                                 /* we don't understand; pass it on */
6486                                 rc = next->ld_ops->ldo_process_config(env, next,
6487                                                                       cfg);
6488                         else
6489                                 rc = count > 0 ? 0 : count;
6490                 } else {
6491                         rc = count > 0 ? 0 : count;
6492                 }
6493
6494                 if (old_cfg)
6495                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6496                                                      cfg->lcfg_buflens));
6497                 break;
6498         }
6499         default:
6500                 /* others are passed further */
6501                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6502                 break;
6503         }
6504         RETURN(rc);
6505 }
6506
6507 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6508                                           const struct lu_object_header *hdr,
6509                                           struct lu_device *d)
6510 {
6511         struct mdt_object *mo;
6512
6513         ENTRY;
6514
6515         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6516         if (mo != NULL) {
6517                 struct lu_object *o;
6518                 struct lu_object_header *h;
6519
6520                 o = &mo->mot_obj;
6521                 h = &mo->mot_header;
6522                 lu_object_header_init(h);
6523                 lu_object_init(o, h, d);
6524                 lu_object_add_top(h, o);
6525                 o->lo_ops = &mdt_obj_ops;
6526                 spin_lock_init(&mo->mot_write_lock);
6527                 mutex_init(&mo->mot_som_mutex);
6528                 mutex_init(&mo->mot_lov_mutex);
6529                 init_rwsem(&mo->mot_dom_sem);
6530                 init_rwsem(&mo->mot_open_sem);
6531                 atomic_set(&mo->mot_open_count, 0);
6532                 mo->mot_restripe_offset = 0;
6533                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6534                 mo->mot_lsom_size = 0;
6535                 mo->mot_lsom_blocks = 0;
6536                 mo->mot_lsom_inited = false;
6537                 RETURN(o);
6538         }
6539         RETURN(NULL);
6540 }
6541
6542 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6543                            const struct lu_object_conf *unused)
6544 {
6545         struct mdt_device *d = mdt_dev(o->lo_dev);
6546         struct lu_device  *under;
6547         struct lu_object  *below;
6548         int                rc = 0;
6549         ENTRY;
6550
6551         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6552                PFID(lu_object_fid(o)));
6553
6554         under = &d->mdt_child->md_lu_dev;
6555         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6556         if (below != NULL) {
6557                 lu_object_add(o, below);
6558         } else
6559                 rc = -ENOMEM;
6560
6561         RETURN(rc);
6562 }
6563
6564 static void mdt_object_free_rcu(struct rcu_head *head)
6565 {
6566         struct mdt_object *mo = container_of(head, struct mdt_object,
6567                                              mot_header.loh_rcu);
6568
6569         kmem_cache_free(mdt_object_kmem, mo);
6570 }
6571
6572 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6573 {
6574         struct mdt_object *mo = mdt_obj(o);
6575         struct lu_object_header *h;
6576         ENTRY;
6577
6578         h = o->lo_header;
6579         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6580                PFID(lu_object_fid(o)));
6581
6582         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6583         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6584
6585         lu_object_fini(o);
6586         lu_object_header_fini(h);
6587         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6588         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6589
6590         EXIT;
6591 }
6592
6593 static int mdt_object_print(const struct lu_env *env, void *cookie,
6594                             lu_printer_t p, const struct lu_object *o)
6595 {
6596         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6597
6598         return (*p)(env, cookie,
6599                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6600                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6601                     mdto->mot_cache_attr ? "cache_attr" : "",
6602                     mdto->mot_write_count);
6603 }
6604
6605 static int mdt_prepare(const struct lu_env *env,
6606                 struct lu_device *pdev,
6607                 struct lu_device *cdev)
6608 {
6609         struct mdt_device *mdt = mdt_dev(cdev);
6610         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6611         struct obd_device *obd = cdev->ld_obd;
6612         int rc;
6613
6614         ENTRY;
6615
6616         LASSERT(obd);
6617
6618         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6619         if (rc)
6620                 RETURN(rc);
6621
6622         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6623         if (rc)
6624                 RETURN(rc);
6625
6626         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6627         if (rc)
6628                 RETURN(rc);
6629
6630         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6631         /* The LFSCK instance is registered just now, so it must be there when
6632          * register the namespace to such instance. */
6633         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6634
6635         if (mdt->mdt_seq_site.ss_node_id == 0) {
6636                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6637                                                          &mdt->mdt_md_root_fid);
6638                 if (rc)
6639                         RETURN(rc);
6640         }
6641
6642         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6643
6644         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6645         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6646         LASSERT(obd->obd_no_conn);
6647         spin_lock(&obd->obd_dev_lock);
6648         obd->obd_no_conn = 0;
6649         spin_unlock(&obd->obd_dev_lock);
6650
6651         if (obd->obd_recovering == 0)
6652                 mdt_postrecov(env, mdt);
6653
6654         RETURN(rc);
6655 }
6656
6657 const struct lu_device_operations mdt_lu_ops = {
6658         .ldo_object_alloc   = mdt_object_alloc,
6659         .ldo_process_config = mdt_process_config,
6660         .ldo_prepare        = mdt_prepare,
6661 };
6662
6663 static const struct lu_object_operations mdt_obj_ops = {
6664         .loo_object_init    = mdt_object_init,
6665         .loo_object_free    = mdt_object_free,
6666         .loo_object_print   = mdt_object_print
6667 };
6668
6669 static int mdt_obd_set_info_async(const struct lu_env *env,
6670                                   struct obd_export *exp,
6671                                   __u32 keylen, void *key,
6672                                   __u32 vallen, void *val,
6673                                   struct ptlrpc_request_set *set)
6674 {
6675         int rc;
6676
6677         ENTRY;
6678
6679         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6680                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6681                 RETURN(rc);
6682         }
6683
6684         RETURN(0);
6685 }
6686
6687 static inline void mdt_enable_slc(struct mdt_device *mdt)
6688 {
6689         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6690                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6691 }
6692
6693 static inline void mdt_disable_slc(struct mdt_device *mdt)
6694 {
6695         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6696                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6697 }
6698
6699 /**
6700  * Match client and server connection feature flags.
6701  *
6702  * Compute the compatibility flags for a connection request based on
6703  * features mutually supported by client and server.
6704  *
6705  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6706  * must not be updated here, otherwise a partially initialized value may
6707  * be exposed. After the connection request is successfully processed,
6708  * the top-level MDT connect request handler atomically updates the export
6709  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6710  * reply. \see mdt_connect().
6711  *
6712  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6713  * full struct obd_connect_data. So care must be taken when accessing fields
6714  * that are not present in struct obd_connect_data_v1. See LU-16.
6715  *
6716  * \param exp   the obd_export associated with this client/target pair
6717  * \param mdt   the target device for the connection
6718  * \param data  stores data for this connect request
6719  *
6720  * \retval 0       success
6721  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6722  * \retval -EBADE  client and server feature requirements are incompatible
6723  */
6724 static int mdt_connect_internal(const struct lu_env *env,
6725                                 struct obd_export *exp,
6726                                 struct mdt_device *mdt,
6727                                 struct obd_connect_data *data, bool reconnect)
6728 {
6729         const char *obd_name = mdt_obd_name(mdt);
6730         LASSERT(data != NULL);
6731
6732         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6733
6734         if (mdt->mdt_bottom->dd_rdonly &&
6735             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6736             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6737                 RETURN(-EACCES);
6738
6739         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6740                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6741
6742         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6743
6744         if (!mdt->mdt_opts.mo_acl)
6745                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6746
6747         if (!mdt->mdt_opts.mo_user_xattr)
6748                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6749
6750         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6751                 data->ocd_brw_size = min(data->ocd_brw_size,
6752                                          mdt->mdt_brw_size);
6753                 if (data->ocd_brw_size == 0) {
6754                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6755                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6756                                "ocd_brw_size unexpectedly zero, network data "
6757                                "corruption? Refusing to connect this client\n",
6758                                obd_name, exp->exp_client_uuid.uuid,
6759                                exp, data->ocd_connect_flags, data->ocd_version,
6760                                data->ocd_grant, data->ocd_index);
6761                         return -EPROTO;
6762                 }
6763         }
6764
6765         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6766                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6767
6768                 /* client is reporting its page size, for future use */
6769                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6770                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6771                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6772                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6773                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6774                 /* ocd_grant_tax_kb is in 1K byte blocks */
6775                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6776                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6777         }
6778
6779         /* Save connect_data we have so far because tgt_grant_connect()
6780          * uses it to calculate grant, and we want to save the client
6781          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6782         exp->exp_connect_data = *data;
6783         if (OCD_HAS_FLAG(data, GRANT))
6784                 tgt_grant_connect(env, exp, data, !reconnect);
6785
6786         if (OCD_HAS_FLAG(data, MAXBYTES))
6787                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6788
6789         /* NB: Disregard the rule against updating
6790          * exp_connect_data.ocd_connect_flags in this case, since
6791          * tgt_client_new() needs to know if this is a lightweight
6792          * connection, and it is safe to expose this flag before
6793          * connection processing completes. */
6794         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6795                 spin_lock(&exp->exp_lock);
6796                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6797                 spin_unlock(&exp->exp_lock);
6798         }
6799
6800         data->ocd_version = LUSTRE_VERSION_CODE;
6801
6802         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6803                 CWARN("%s: MDS requires FID support, but client not\n",
6804                       obd_name);
6805                 return -EBADE;
6806         }
6807
6808         if (OCD_HAS_FLAG(data, PINGLESS)) {
6809                 if (ptlrpc_pinger_suppress_pings()) {
6810                         spin_lock(&exp->exp_obd->obd_dev_lock);
6811                         list_del_init(&exp->exp_obd_chain_timed);
6812                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6813                 } else {
6814                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6815                 }
6816         }
6817
6818         data->ocd_max_easize = mdt->mdt_max_ea_size;
6819
6820         /* NB: Disregard the rule against updating
6821          * exp_connect_data.ocd_connect_flags in this case, since
6822          * tgt_client_new() needs to know if this is client supports
6823          * multiple modify RPCs, and it is safe to expose this flag before
6824          * connection processing completes. */
6825         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6826                 if (mdt_max_mod_rpcs_changed(mdt))
6827                         /* The new mdt.*.max_mod_rpcs_in_flight parameter
6828                          * has not changed since initialization, but the
6829                          * deprecated module parameter was changed,
6830                          * so use that instead.
6831                          */
6832                         data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6833                 else
6834                         data->ocd_maxmodrpcs = mdt->mdt_max_mod_rpcs_in_flight;
6835                 spin_lock(&exp->exp_lock);
6836                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6837                 spin_unlock(&exp->exp_lock);
6838         }
6839
6840         if (OCD_HAS_FLAG(data, CKSUM)) {
6841                 __u32 cksum_types = data->ocd_cksum_types;
6842
6843                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6844
6845                 if (unlikely(data->ocd_cksum_types == 0)) {
6846                         CERROR("%s: Connect with checksum support but no "
6847                                "ocd_cksum_types is set\n",
6848                                exp->exp_obd->obd_name);
6849                         RETURN(-EPROTO);
6850                 }
6851
6852                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6853                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6854                        cksum_types, data->ocd_cksum_types);
6855         } else {
6856                 /* This client does not support OBD_CONNECT_CKSUM
6857                  * fall back to CRC32 */
6858                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6859                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6860                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6861         }
6862
6863         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6864             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6865                 atomic_inc(&mdt->mdt_mds_mds_conns);
6866                 mdt_enable_slc(mdt);
6867         }
6868
6869         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6870                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6871
6872         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
6873             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT)) {
6874                 atomic_inc(&mdt->mdt_dmv_old_client_count);
6875                 mdt->mdt_enable_dmv_implicit_inherit = 0;
6876         }
6877
6878         return 0;
6879 }
6880
6881 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6882                                    struct mdt_thread_info *info,
6883                                    struct mdt_file_data *mfd)
6884 {
6885         struct lu_context ses;
6886         int rc;
6887         ENTRY;
6888
6889         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6890         if (rc)
6891                 RETURN(rc);
6892
6893         env->le_ses = &ses;
6894         lu_context_enter(&ses);
6895
6896         mdt_ucred(info)->uc_valid = UCRED_OLD;
6897         /* do not let rbac interfere with dirty flag internal system event */
6898         mdt_ucred(info)->uc_rbac_file_perms = 1;
6899         mdt_ucred(info)->uc_rbac_dne_ops = 1;
6900         mdt_ucred(info)->uc_rbac_quota_ops = 1;
6901         mdt_ucred(info)->uc_rbac_byfid_ops = 1;
6902         mdt_ucred(info)->uc_rbac_chlg_ops = 1;
6903         mdt_ucred(info)->uc_rbac_fscrypt_admin = 1;
6904         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6905
6906         lu_context_exit(&ses);
6907         lu_context_fini(&ses);
6908         env->le_ses = NULL;
6909
6910         RETURN(rc);
6911 }
6912
6913 static int mdt_export_cleanup(struct obd_export *exp)
6914 {
6915         LIST_HEAD(closing_list);
6916         struct mdt_export_data  *med = &exp->exp_mdt_data;
6917         struct obd_device       *obd = exp->exp_obd;
6918         struct mdt_device       *mdt;
6919         struct mdt_thread_info  *info;
6920         struct lu_env            env;
6921         struct mdt_file_data    *mfd, *n;
6922         int rc = 0;
6923         ENTRY;
6924
6925         spin_lock(&med->med_open_lock);
6926         while (!list_empty(&med->med_open_head)) {
6927                 struct list_head *tmp = med->med_open_head.next;
6928                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6929
6930                 /* Remove mfd handle so it can't be found again.
6931                  * We are consuming the mfd_list reference here. */
6932                 class_handle_unhash(&mfd->mfd_open_handle);
6933                 list_move_tail(&mfd->mfd_list, &closing_list);
6934         }
6935         spin_unlock(&med->med_open_lock);
6936         mdt = mdt_dev(obd->obd_lu_dev);
6937         LASSERT(mdt != NULL);
6938
6939         rc = lu_env_init(&env, LCT_MD_THREAD);
6940         if (rc)
6941                 RETURN(rc);
6942
6943         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6944         LASSERT(info != NULL);
6945         memset(info, 0, sizeof *info);
6946         info->mti_env = &env;
6947         info->mti_mdt = mdt;
6948         info->mti_exp = exp;
6949
6950         if (!list_empty(&closing_list)) {
6951                 struct md_attr *ma = &info->mti_attr;
6952
6953                 /* Close any open files (which may also cause orphan
6954                  * unlinking). */
6955                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6956                         list_del_init(&mfd->mfd_list);
6957                         ma->ma_need = ma->ma_valid = 0;
6958
6959                         /* This file is being closed due to an eviction, it
6960                          * could have been modified and now dirty regarding to
6961                          * HSM archive, check this!
6962                          * The logic here is to mark a file dirty if there's a
6963                          * chance it was dirtied before the client was evicted,
6964                          * so that we don't have to wait for a release attempt
6965                          * before finding out the file was actually dirty and
6966                          * fail the release. Aggressively marking it dirty here
6967                          * will cause the policy engine to attempt to
6968                          * re-archive it; when rearchiving, we can compare the
6969                          * current version to the HSM data_version and make the
6970                          * archive request into a noop if it's not actually
6971                          * dirty.
6972                          */
6973                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6974                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6975
6976                         /* Don't unlink orphan on failover umount, LU-184 */
6977                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6978                             exp->exp_obd->obd_stopping) {
6979                                 ma->ma_valid = MA_FLAGS;
6980                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6981                         }
6982                         ma->ma_valid |= MA_FORCE_LOG;
6983                         mdt_mfd_close(info, mfd);
6984                 }
6985         }
6986         info->mti_mdt = NULL;
6987         /* cleanup client slot early */
6988         /* Do not erase record for recoverable client. */
6989         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6990                 tgt_client_del(&env, exp);
6991         lu_env_fini(&env);
6992
6993         RETURN(rc);
6994 }
6995
6996 static int mdt_obd_disconnect(struct obd_export *exp)
6997 {
6998         struct obd_connect_data *data = &exp->exp_connect_data;
6999         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
7000         int rc;
7001
7002         ENTRY;
7003
7004         LASSERT(exp);
7005         class_export_get(exp);
7006
7007         if (!(exp->exp_flags & OBD_OPT_FORCE))
7008                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7009
7010         if (OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7011             atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
7012                 mdt_disable_slc(mdt);
7013
7014         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7015             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT) &&
7016             atomic_dec_and_test(&mdt->mdt_dmv_old_client_count))
7017                 mdt->mdt_enable_dmv_implicit_inherit = 1;
7018
7019         rc = server_disconnect_export(exp);
7020         if (rc != 0)
7021                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
7022
7023         tgt_grant_discard(exp);
7024
7025         rc = mdt_export_cleanup(exp);
7026         nodemap_del_member(exp);
7027         class_export_put(exp);
7028         RETURN(rc);
7029 }
7030
7031 /* mds_connect copy */
7032 static int mdt_obd_connect(const struct lu_env *env,
7033                            struct obd_export **exp, struct obd_device *obd,
7034                            struct obd_uuid *cluuid,
7035                            struct obd_connect_data *data,
7036                            void *localdata)
7037 {
7038         struct obd_export       *lexp;
7039         struct lustre_handle    conn = { 0 };
7040         struct mdt_device       *mdt;
7041         int                      rc;
7042         struct lnet_nid         *client_nid = localdata;
7043         ENTRY;
7044
7045         LASSERT(env != NULL);
7046         LASSERT(data != NULL);
7047
7048         if (!exp || !obd || !cluuid)
7049                 RETURN(-EINVAL);
7050
7051         mdt = mdt_dev(obd->obd_lu_dev);
7052
7053         /*
7054          * first, check whether the stack is ready to handle requests
7055          * XXX: probably not very appropriate method is used now
7056          *      at some point we should find a better one
7057          */
7058         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
7059             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
7060             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
7061                 rc = obd_get_info(env, mdt->mdt_child_exp,
7062                                   sizeof(KEY_OSP_CONNECTED),
7063                                   KEY_OSP_CONNECTED, NULL, NULL);
7064                 if (rc)
7065                         RETURN(-EAGAIN);
7066                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
7067         }
7068
7069         rc = class_connect(&conn, obd, cluuid);
7070         if (rc)
7071                 RETURN(rc);
7072
7073         lexp = class_conn2export(&conn);
7074         LASSERT(lexp != NULL);
7075
7076         if (nid_is_nid4(client_nid)) {
7077                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), lexp);
7078                 if (rc != 0 && rc != -EEXIST)
7079                         GOTO(out, rc);
7080         }
7081
7082         rc = mdt_connect_internal(env, lexp, mdt, data, false);
7083         if (rc == 0) {
7084                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
7085
7086                 LASSERT(lcd);
7087                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
7088                 rc = tgt_client_new(env, lexp);
7089                 if (rc == 0)
7090                         mdt_export_stats_init(obd, lexp, localdata);
7091         }
7092 out:
7093         if (rc != 0) {
7094                 class_disconnect(lexp);
7095                 nodemap_del_member(lexp);
7096                 *exp = NULL;
7097         } else {
7098                 *exp = lexp;
7099                 /* Because we do not want this export to be evicted by pinger,
7100                  * let's not add this export to the timed chain list. */
7101                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
7102                         spin_lock(&lexp->exp_obd->obd_dev_lock);
7103                         list_del_init(&lexp->exp_obd_chain_timed);
7104                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
7105                 }
7106         }
7107
7108         RETURN(rc);
7109 }
7110
7111 static int mdt_obd_reconnect(const struct lu_env *env,
7112                              struct obd_export *exp, struct obd_device *obd,
7113                              struct obd_uuid *cluuid,
7114                              struct obd_connect_data *data,
7115                              void *localdata)
7116 {
7117         struct lnet_nid *client_nid = localdata;
7118         int rc;
7119         ENTRY;
7120
7121         if (exp == NULL || obd == NULL || cluuid == NULL)
7122                 RETURN(-EINVAL);
7123
7124         if (nid_is_nid4(client_nid)) {
7125                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), exp);
7126                 if (rc != 0 && rc != -EEXIST)
7127                         RETURN(rc);
7128         }
7129
7130         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
7131                                   true);
7132         if (rc == 0)
7133                 mdt_export_stats_init(obd, exp, localdata);
7134         else
7135                 nodemap_del_member(exp);
7136
7137         RETURN(rc);
7138 }
7139
7140 /* FIXME: Can we avoid using these two interfaces? */
7141 static int mdt_init_export(struct obd_export *exp)
7142 {
7143         struct mdt_export_data *med = &exp->exp_mdt_data;
7144         int                     rc;
7145         ENTRY;
7146
7147         INIT_LIST_HEAD(&med->med_open_head);
7148         spin_lock_init(&med->med_open_lock);
7149         spin_lock(&exp->exp_lock);
7150         exp->exp_connecting = 1;
7151         spin_unlock(&exp->exp_lock);
7152
7153         OBD_ALLOC(exp->exp_used_slots,
7154                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7155         if (exp->exp_used_slots == NULL)
7156                 RETURN(-ENOMEM);
7157
7158         /* self-export doesn't need client data and ldlm initialization */
7159         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7160                                      &exp->exp_client_uuid)))
7161                 RETURN(0);
7162
7163         rc = tgt_client_alloc(exp);
7164         if (rc)
7165                 GOTO(err, rc);
7166
7167         rc = ldlm_init_export(exp);
7168         if (rc)
7169                 GOTO(err_free, rc);
7170
7171         RETURN(rc);
7172
7173 err_free:
7174         tgt_client_free(exp);
7175 err:
7176         OBD_FREE(exp->exp_used_slots,
7177                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7178         exp->exp_used_slots = NULL;
7179
7180         CERROR("%s: Failed to initialize export: rc = %d\n",
7181                exp->exp_obd->obd_name, rc);
7182         return rc;
7183 }
7184
7185 static int mdt_destroy_export(struct obd_export *exp)
7186 {
7187         ENTRY;
7188
7189         target_destroy_export(exp);
7190         if (exp->exp_used_slots)
7191                 OBD_FREE(exp->exp_used_slots,
7192                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7193
7194         /* destroy can be called from failed obd_setup, so
7195          * checking uuid is safer than obd_self_export */
7196         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7197                                      &exp->exp_client_uuid)))
7198                 RETURN(0);
7199
7200         ldlm_destroy_export(exp);
7201         tgt_client_free(exp);
7202
7203         LASSERT(list_empty(&exp->exp_outstanding_replies));
7204         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
7205
7206         /*
7207          * discard grants once we're sure no more
7208          * interaction with the client is possible
7209          */
7210         tgt_grant_discard(exp);
7211         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
7212                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
7213
7214         if (!(exp->exp_flags & OBD_OPT_FORCE))
7215                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7216
7217         RETURN(0);
7218 }
7219
7220 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
7221                    struct linkea_data *ldata)
7222 {
7223         int rc;
7224
7225         LASSERT(ldata->ld_buf->lb_buf != NULL);
7226
7227         if (!mdt_object_exists(mdt_obj))
7228                 return -ENODATA;
7229
7230         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7231                           ldata->ld_buf, XATTR_NAME_LINK);
7232         if (rc == -ERANGE) {
7233                 /* Buf was too small, figure out what we need. */
7234                 lu_buf_free(ldata->ld_buf);
7235                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7236                                   ldata->ld_buf, XATTR_NAME_LINK);
7237                 if (rc < 0)
7238                         return rc;
7239                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7240                 if (ldata->ld_buf->lb_buf == NULL)
7241                         return -ENOMEM;
7242                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7243                                   ldata->ld_buf, XATTR_NAME_LINK);
7244         }
7245         if (rc < 0)
7246                 return rc;
7247
7248         return linkea_init_with_rec(ldata);
7249 }
7250
7251 /**
7252  * Given an MDT object, try to look up the full path to the object.
7253  * Part of the MDT layer implementation of lfs fid2path.
7254  *
7255  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7256  * \param[in]     obj   Object to do path lookup of
7257  * \param[in,out] fp    User-provided struct to store path information
7258  * \param[in]     root_fid Root FID of current path should reach
7259  *
7260  * \retval 0 Lookup successful, path information stored in fp
7261  * \retval -EAGAIN Lookup failed, usually because object is being moved
7262  * \retval negative errno if there was a problem
7263  */
7264 static int mdt_path_current(struct mdt_thread_info *info,
7265                             struct mdt_object *obj,
7266                             struct getinfo_fid2path *fp,
7267                             struct lu_fid *root_fid)
7268 {
7269         struct mdt_device *mdt = info->mti_mdt;
7270         struct lu_name *tmpname = &info->mti_name;
7271         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7272         struct lu_buf *buf = &info->mti_big_buf;
7273         struct linkea_data ldata = { NULL };
7274         bool first = true;
7275         struct mdt_object *mdt_obj;
7276         struct link_ea_header *leh;
7277         struct link_ea_entry *lee;
7278         bool worthchecking = true;
7279         bool needsfid = false;
7280         bool supported = false;
7281         int isenc = -1;
7282         char *ptr;
7283         int reclen;
7284         int rc = 0;
7285
7286         ENTRY;
7287
7288         /* temp buffer for path element, the buffer will be finally freed
7289          * in mdt_thread_info_fini */
7290         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7291         if (buf->lb_buf == NULL)
7292                 RETURN(-ENOMEM);
7293
7294         ldata.ld_buf = buf;
7295         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7296         *ptr = 0;
7297         --ptr;
7298         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7299
7300         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7301                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7302                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7303                         GOTO(out, rc = -ENOENT);
7304
7305                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7306                         mdt_obj = obj;
7307                         mdt_object_get(info->mti_env, mdt_obj);
7308                 } else {
7309                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7310                         if (IS_ERR(mdt_obj))
7311                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7312                 }
7313
7314                 if (!mdt_object_exists(mdt_obj)) {
7315                         mdt_object_put(info->mti_env, mdt_obj);
7316                         GOTO(out, rc = -ENOENT);
7317                 }
7318
7319                 if (mdt_object_remote(mdt_obj)) {
7320                         mdt_object_put(info->mti_env, mdt_obj);
7321                         GOTO(remote_out, rc = -EREMOTE);
7322                 }
7323
7324                 if (worthchecking) {
7325                         /* we need to know if the FID being
7326                          * looked up is encrypted
7327                          */
7328                         struct lu_attr la = { 0 };
7329                         struct dt_object *dt = mdt_obj2dt(mdt_obj);
7330
7331                         if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7332                                 dt_attr_get(info->mti_env, dt, &la);
7333                         if (la.la_valid & LA_FLAGS &&
7334                             la.la_flags & LUSTRE_ENCRYPT_FL) {
7335                                 if (!supported && mdt_info_req(info) &&
7336                                     !exp_connect_encrypt_fid2path(
7337                                             mdt_info_req(info)->rq_export)) {
7338                                         /* client does not support fid2path
7339                                          * for encrypted files
7340                                          */
7341                                         mdt_object_put(info->mti_env, mdt_obj);
7342                                         GOTO(out, rc = -ENODATA);
7343                                 } else {
7344                                         supported = true;
7345                                 }
7346                                 needsfid = true;
7347                                 if (isenc == -1)
7348                                         isenc = 1;
7349                         } else {
7350                                 worthchecking = false;
7351                                 needsfid = false;
7352                                 if (isenc == -1)
7353                                         isenc = 0;
7354                         }
7355                 }
7356
7357                 rc = mdt_links_read(info, mdt_obj, &ldata);
7358                 if (rc != 0) {
7359                         mdt_object_put(info->mti_env, mdt_obj);
7360                         GOTO(out, rc);
7361                 }
7362
7363                 leh = buf->lb_buf;
7364                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7365                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7366                 /* If set, use link #linkno for path lookup, otherwise use
7367                    link #0.  Only do this for the final path element. */
7368                 if (first && fp->gf_linkno < leh->leh_reccount) {
7369                         int count;
7370                         for (count = 0; count < fp->gf_linkno; count++) {
7371                                 lee = (struct link_ea_entry *)
7372                                      ((char *)lee + reclen);
7373                                 linkea_entry_unpack(lee, &reclen, tmpname,
7374                                                     tmpfid);
7375                         }
7376                         if (fp->gf_linkno < leh->leh_reccount - 1)
7377                                 /* indicate to user there are more links */
7378                                 fp->gf_linkno++;
7379                 }
7380
7381                 /* Check if it is slave stripes */
7382                 rc = mdt_is_dir_stripe(info, mdt_obj);
7383                 mdt_object_put(info->mti_env, mdt_obj);
7384                 if (rc < 0)
7385                         GOTO(out, rc);
7386                 if (rc == 1) {
7387                         fp->gf_fid = *tmpfid;
7388                         continue;
7389                 }
7390
7391                 /* Pack the name in the end of the buffer */
7392                 ptr -= tmpname->ln_namelen;
7393                 if (ptr - 1 <= fp->gf_u.gf_path)
7394                         GOTO(out, rc = -ENAMETOOLONG);
7395                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7396                 if (needsfid) {
7397                         /* Pack FID before file name, so that client can build
7398                          * encoded/digested form.
7399                          */
7400                         char fidstr[FID_LEN + 1];
7401
7402                         snprintf(fidstr, sizeof(fidstr), DFID,
7403                                  PFID(&fp->gf_fid));
7404                         ptr -= strlen(fidstr);
7405                         if (ptr - 1 <= fp->gf_u.gf_path)
7406                                 GOTO(out, rc = -ENAMETOOLONG);
7407                         strncpy(ptr, fidstr, strlen(fidstr));
7408                 }
7409                 *(--ptr) = '/';
7410
7411                 /* keep the last resolved fid to the client, so the
7412                  * client will build the left path on another MDT for
7413                  * remote object */
7414                 fp->gf_fid = *tmpfid;
7415
7416                 first = false;
7417         }
7418
7419         /* non-zero will be treated as an error */
7420         rc = 0;
7421
7422 remote_out:
7423         if (isenc != 1)
7424                 ptr++; /* skip leading / unless this is an encrypted file */
7425         memmove(fp->gf_u.gf_path, ptr,
7426                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7427
7428 out:
7429         RETURN(rc);
7430 }
7431
7432 /**
7433  * Given an MDT object, use mdt_path_current to get the path.
7434  * Essentially a wrapper to retry mdt_path_current a set number of times
7435  * if -EAGAIN is returned (usually because an object is being moved).
7436  *
7437  * Part of the MDT layer implementation of lfs fid2path.
7438  *
7439  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7440  * \param[in]     obj   Object to do path lookup of
7441  * \param[in,out] fp    User-provided struct for arguments and to store path
7442  *                      information
7443  *
7444  * \retval 0 Lookup successful, path information stored in fp
7445  * \retval negative errno if there was a problem
7446  */
7447 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7448                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7449 {
7450         struct mdt_device       *mdt = info->mti_mdt;
7451         int                     tries = 3;
7452         int                     rc = -EAGAIN;
7453         ENTRY;
7454
7455         if (fp->gf_pathlen < 3)
7456                 RETURN(-EOVERFLOW);
7457
7458         if (root_fid == NULL)
7459                 root_fid = &mdt->mdt_md_root_fid;
7460
7461         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7462                 fp->gf_u.gf_path[0] = '\0';
7463                 RETURN(0);
7464         }
7465
7466         /* Retry multiple times in case file is being moved */
7467         while (tries-- && rc == -EAGAIN)
7468                 rc = mdt_path_current(info, obj, fp, root_fid);
7469
7470         RETURN(rc);
7471 }
7472
7473 /**
7474  * Get the full path of the provided FID, as of changelog record recno.
7475  *
7476  * This checks sanity and looks up object for user provided FID
7477  * before calling the actual path lookup code.
7478  *
7479  * Part of the MDT layer implementation of lfs fid2path.
7480  *
7481  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7482  * \param[in,out] fp    User-provided struct for arguments and to store path
7483  *                      information
7484  *
7485  * \retval 0 Lookup successful, path information and recno stored in fp
7486  * \retval -ENOENT, object does not exist
7487  * \retval negative errno if there was a problem
7488  */
7489 static int mdt_fid2path(struct mdt_thread_info *info,
7490                         struct lu_fid *root_fid,
7491                         struct getinfo_fid2path *fp)
7492 {
7493         struct mdt_device *mdt = info->mti_mdt;
7494         struct mdt_object *obj;
7495         int    rc;
7496         ENTRY;
7497
7498         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7499                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7500
7501         if (!fid_is_sane(&fp->gf_fid))
7502                 RETURN(-EINVAL);
7503
7504         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7505                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7506                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7507                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7508                 RETURN(-EINVAL);
7509         }
7510
7511         /* return error if client-provided root fid
7512          * is not the one stored in the export
7513          */
7514         if (root_fid && !fid_is_zero(&info->mti_exp->exp_root_fid) &&
7515             !lu_fid_eq(root_fid, &info->mti_exp->exp_root_fid)) {
7516                 CDEBUG(D_INFO,
7517                        "%s: root fid from client "DFID" but "DFID" stored in export\n",
7518                        mdt_obd_name(mdt), PFID(root_fid),
7519                        PFID(&info->mti_exp->exp_root_fid));
7520                 RETURN(-EXDEV);
7521         }
7522
7523         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7524         if (IS_ERR(obj)) {
7525                 rc = PTR_ERR(obj);
7526                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7527                        PFID(&fp->gf_fid), rc);
7528                 RETURN(rc);
7529         }
7530
7531         if (mdt_object_remote(obj))
7532                 rc = -EREMOTE;
7533         else if (!mdt_object_exists(obj))
7534                 rc = -ENOENT;
7535         else
7536                 rc = 0;
7537
7538         if (rc < 0) {
7539                 mdt_object_put(info->mti_env, obj);
7540                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7541                        PFID(&fp->gf_fid), rc);
7542                 RETURN(rc);
7543         }
7544
7545         rc = mdt_path(info, obj, fp, root_fid);
7546
7547         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7548                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7549                fp->gf_recno, fp->gf_linkno);
7550
7551         mdt_object_put(info->mti_env, obj);
7552
7553         RETURN(rc);
7554 }
7555
7556 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7557                             void *val, int vallen)
7558 {
7559         struct getinfo_fid2path *fpout, *fpin;
7560         struct lu_fid *root_fid = NULL;
7561         int rc = 0;
7562
7563         fpin = key + round_up(sizeof(KEY_FID2PATH), 8);
7564         fpout = val;
7565
7566         if (req_capsule_req_need_swab(info->mti_pill))
7567                 lustre_swab_fid2path(fpin);
7568
7569         memcpy(fpout, fpin, sizeof(*fpin));
7570         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7571                 RETURN(-EINVAL);
7572
7573         if (keylen >= round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*fpin) +
7574                       sizeof(struct lu_fid)) {
7575                 /* client sent its root FID, which is normally fileset FID */
7576                 root_fid = fpin->gf_u.gf_root_fid;
7577                 if (req_capsule_req_need_swab(info->mti_pill))
7578                         lustre_swab_lu_fid(root_fid);
7579
7580                 if (root_fid != NULL && !fid_is_sane(root_fid))
7581                         RETURN(-EINVAL);
7582         }
7583
7584         rc = mdt_fid2path(info, root_fid, fpout);
7585         RETURN(rc);
7586 }
7587
7588 int mdt_get_info(struct tgt_session_info *tsi)
7589 {
7590         char    *key;
7591         int      keylen;
7592         __u32   *vallen;
7593         void    *valout;
7594         int      rc;
7595
7596         ENTRY;
7597
7598         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7599         if (key == NULL) {
7600                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7601                 RETURN(err_serious(-EFAULT));
7602         }
7603         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7604                                       RCL_CLIENT);
7605
7606         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7607         if (vallen == NULL) {
7608                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7609                                 tgt_name(tsi->tsi_tgt));
7610                 RETURN(err_serious(-EFAULT));
7611         }
7612
7613         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7614                              *vallen);
7615         rc = req_capsule_server_pack(tsi->tsi_pill);
7616         if (rc)
7617                 RETURN(err_serious(rc));
7618
7619         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7620         if (valout == NULL) {
7621                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7622                                 tgt_name(tsi->tsi_tgt));
7623                 RETURN(err_serious(-EFAULT));
7624         }
7625
7626         if (KEY_IS(KEY_FID2PATH)) {
7627                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7628
7629                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7630                 mdt_thread_info_fini(info);
7631         } else {
7632                 rc = -EINVAL;
7633         }
7634         RETURN(rc);
7635 }
7636
7637 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7638 {
7639         struct obd_ioctl_data *data = karg;
7640         struct lu_fid *fid;
7641         __u64 version;
7642         struct mdt_object *obj;
7643         struct mdt_lock_handle  *lh;
7644         int rc;
7645         ENTRY;
7646
7647         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7648             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7649                 RETURN(-EINVAL);
7650
7651         fid = (struct lu_fid *)data->ioc_inlbuf1;
7652
7653         if (!fid_is_sane(fid))
7654                 RETURN(-EINVAL);
7655
7656         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7657
7658         lh = &mti->mti_lh[MDT_LH_PARENT];
7659         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE, LCK_CR);
7660         if (IS_ERR(obj))
7661                 RETURN(PTR_ERR(obj));
7662
7663         if (mdt_object_remote(obj)) {
7664                 rc = -EREMOTE;
7665                 /**
7666                  * before calling version get the correct MDS should be
7667                  * fid, this is error to find remote object here
7668                  */
7669                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7670         } else if (!mdt_object_exists(obj)) {
7671                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7672                 rc = -ENOENT;
7673         } else {
7674                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7675                *(__u64 *)data->ioc_inlbuf2 = version;
7676                 rc = 0;
7677         }
7678         mdt_object_unlock_put(mti, obj, lh, 1);
7679         RETURN(rc);
7680 }
7681
7682 /* ioctls on obd dev */
7683 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7684                          void *karg, void __user *uarg)
7685 {
7686         struct obd_device *obd = exp->exp_obd;
7687         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7688         struct dt_device *dt = mdt->mdt_bottom;
7689         struct obd_ioctl_data *data;
7690         struct lu_env env;
7691         int rc;
7692
7693         ENTRY;
7694         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
7695                obd->obd_name, cmd, len, karg, uarg);
7696
7697         rc = lu_env_init(&env, LCT_MD_THREAD);
7698         if (rc)
7699                 RETURN(rc);
7700
7701         /* handle commands that don't use @karg first */
7702         switch (cmd) {
7703         case OBD_IOC_SYNC:
7704                 rc = mdt_device_sync(&env, mdt);
7705                 GOTO(out, rc);
7706         case OBD_IOC_SET_READONLY:
7707                 rc = dt_sync(&env, dt);
7708                 if (rc == 0)
7709                         rc = dt_ro(&env, dt);
7710                 GOTO(out, rc);
7711         }
7712
7713         if (unlikely(karg == NULL)) {
7714                 OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc = -EINVAL);
7715                 GOTO(out, rc);
7716         }
7717         data = karg;
7718
7719         switch (cmd) {
7720         case OBD_IOC_ABORT_RECOVERY: {
7721                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7722                         LCONSOLE_WARN("%s: Aborting MDT recovery\n",
7723                                       obd->obd_name);
7724                         obd->obd_abort_mdt_recovery = 1;
7725                         wake_up(&obd->obd_next_transno_waitq);
7726                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7727                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7728                         LCONSOLE_WARN("%s: Aborting client recovery\n",
7729                                       obd->obd_name);
7730                         obd->obd_abort_recovery = 1;
7731                         target_stop_recovery_thread(obd);
7732                 }
7733                 rc = 0;
7734                 break;
7735         }
7736         case OBD_IOC_CHANGELOG_REG:
7737         case OBD_IOC_CHANGELOG_DEREG:
7738         case OBD_IOC_CHANGELOG_CLEAR:
7739         case OBD_IOC_LLOG_PRINT:
7740         case OBD_IOC_LLOG_CANCEL:
7741                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env, mdt->mdt_child,
7742                                                            cmd, len, karg);
7743                 break;
7744         case OBD_IOC_START_LFSCK: {
7745                 struct md_device *next = mdt->mdt_child;
7746                 struct lfsck_start_param lsp;
7747
7748                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7749                 lsp.lsp_index_valid = 0;
7750                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7751                 break;
7752         }
7753         case OBD_IOC_STOP_LFSCK: {
7754                 struct md_device *next = mdt->mdt_child;
7755                 struct lfsck_stop stop;
7756
7757                 stop.ls_status = LS_STOPPED;
7758                 /* Old lfsck utils may pass NULL @stop. */
7759                 if (data->ioc_inlbuf1 == NULL)
7760                         stop.ls_flags = 0;
7761                 else
7762                         stop.ls_flags =
7763                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7764
7765                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7766                 break;
7767         }
7768         case OBD_IOC_QUERY_LFSCK: {
7769                 struct md_device *next = mdt->mdt_child;
7770
7771                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7772                                                  data->ioc_inlbuf1);
7773                 break;
7774         }
7775         case OBD_IOC_GET_OBJ_VERSION: {
7776                 struct mdt_thread_info *mti;
7777
7778                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7779                 memset(mti, 0, sizeof(*mti));
7780                 mti->mti_env = &env;
7781                 mti->mti_mdt = mdt;
7782                 mti->mti_exp = exp;
7783
7784                 rc = mdt_ioc_version_get(mti, karg);
7785                 break;
7786         }
7787         case OBD_IOC_CATLOGLIST: {
7788                 struct mdt_thread_info *mti;
7789
7790                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7791                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7792                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7793                                        &mti->mti_tmp_fid1);
7794                 break;
7795         }
7796         default:
7797                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
7798                 break;
7799         }
7800 out:
7801         lu_env_fini(&env);
7802         RETURN(rc);
7803 }
7804
7805 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7806 {
7807         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7808         int rc;
7809         ENTRY;
7810
7811         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7812                 struct lfsck_start_param lsp;
7813
7814                 lsp.lsp_start = NULL;
7815                 lsp.lsp_index_valid = 0;
7816                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7817                                                            OBD_IOC_START_LFSCK,
7818                                                            0, &lsp);
7819                 if (rc != 0 && rc != -EALREADY)
7820                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7821                               mdt_obd_name(mdt), rc);
7822         }
7823
7824         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7825         RETURN(rc);
7826 }
7827
7828 static int mdt_obd_postrecov(struct obd_device *obd)
7829 {
7830         struct lu_env env;
7831         int rc;
7832
7833         rc = lu_env_init(&env, LCT_MD_THREAD);
7834         if (rc)
7835                 RETURN(rc);
7836         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7837         lu_env_fini(&env);
7838         return rc;
7839 }
7840
7841 static const struct obd_ops mdt_obd_device_ops = {
7842         .o_owner          = THIS_MODULE,
7843         .o_set_info_async = mdt_obd_set_info_async,
7844         .o_connect        = mdt_obd_connect,
7845         .o_reconnect      = mdt_obd_reconnect,
7846         .o_disconnect     = mdt_obd_disconnect,
7847         .o_init_export    = mdt_init_export,
7848         .o_destroy_export = mdt_destroy_export,
7849         .o_iocontrol      = mdt_iocontrol,
7850         .o_postrecov      = mdt_obd_postrecov,
7851         /* Data-on-MDT IO methods */
7852         .o_preprw         = mdt_obd_preprw,
7853         .o_commitrw       = mdt_obd_commitrw,
7854 };
7855
7856 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7857                                          struct lu_device *d)
7858 {
7859         struct mdt_device *m = mdt_dev(d);
7860         ENTRY;
7861
7862         mdt_fini(env, m);
7863         RETURN(NULL);
7864 }
7865
7866 static struct lu_device *mdt_device_free(const struct lu_env *env,
7867                                          struct lu_device *d)
7868 {
7869         struct mdt_device *m = mdt_dev(d);
7870         ENTRY;
7871
7872         lu_device_fini(&m->mdt_lu_dev);
7873         OBD_FREE_PTR(m);
7874
7875         RETURN(NULL);
7876 }
7877
7878 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7879                                           struct lu_device_type *t,
7880                                           struct lustre_cfg *cfg)
7881 {
7882         struct lu_device  *l;
7883         struct mdt_device *m;
7884
7885         OBD_ALLOC_PTR(m);
7886         if (m != NULL) {
7887                 int rc;
7888
7889                 l = &m->mdt_lu_dev;
7890                 rc = mdt_init0(env, m, t, cfg);
7891                 if (rc != 0) {
7892                         mdt_device_free(env, l);
7893                         l = ERR_PTR(rc);
7894                         return l;
7895                 }
7896         } else
7897                 l = ERR_PTR(-ENOMEM);
7898         return l;
7899 }
7900
7901 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7902 LU_KEY_INIT(mdt, struct mdt_thread_info);
7903
7904 static void mdt_key_fini(const struct lu_context *ctx,
7905                          struct lu_context_key *key, void* data)
7906 {
7907         struct mdt_thread_info *info = data;
7908
7909         if (info->mti_big_lmm) {
7910                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7911                 info->mti_big_lmm = NULL;
7912                 info->mti_big_lmmsize = 0;
7913         }
7914
7915         if (info->mti_big_acl) {
7916                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7917                 info->mti_big_acl = NULL;
7918                 info->mti_big_aclsize = 0;
7919         }
7920
7921         OBD_FREE_PTR(info);
7922 }
7923
7924 /* context key: mdt_thread_key */
7925 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7926
7927 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7928 {
7929         return lu_ucred(info->mti_env);
7930 }
7931
7932 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7933 {
7934         return lu_ucred_check(info->mti_env);
7935 }
7936
7937 /**
7938  * Enable/disable COS (Commit On Sharing).
7939  *
7940  * Set/Clear the COS flag in mdt options.
7941  *
7942  * \param mdt mdt device
7943  * \param val 0 disables COS, other values enable COS
7944  */
7945 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7946 {
7947         struct lu_env env;
7948         int rc;
7949
7950         mdt->mdt_opts.mo_cos = val;
7951         rc = lu_env_init(&env, LCT_LOCAL);
7952         if (unlikely(rc != 0)) {
7953                 CWARN("%s: lu_env initialization failed, cannot "
7954                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7955                 return;
7956         }
7957         mdt_device_sync(&env, mdt);
7958         lu_env_fini(&env);
7959 }
7960
7961 /**
7962  * Check COS (Commit On Sharing) status.
7963  *
7964  * Return COS flag status.
7965  *
7966  * \param mdt mdt device
7967  */
7968 int mdt_cos_is_enabled(struct mdt_device *mdt)
7969 {
7970         return mdt->mdt_opts.mo_cos != 0;
7971 }
7972
7973 static const struct lu_device_type_operations mdt_device_type_ops = {
7974         .ldto_device_alloc = mdt_device_alloc,
7975         .ldto_device_free  = mdt_device_free,
7976         .ldto_device_fini  = mdt_device_fini
7977 };
7978
7979 static struct lu_device_type mdt_device_type = {
7980         .ldt_tags     = LU_DEVICE_MD,
7981         .ldt_name     = LUSTRE_MDT_NAME,
7982         .ldt_ops      = &mdt_device_type_ops,
7983         .ldt_ctx_tags = LCT_MD_THREAD
7984 };
7985
7986 static int __init mdt_init(void)
7987 {
7988         int rc;
7989
7990         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7991                      FID_NOBRACE_LEN + 1);
7992         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7993                      FID_LEN + 1);
7994         rc = lu_kmem_init(mdt_caches);
7995         if (rc)
7996                 return rc;
7997
7998         rc = mds_mod_init();
7999         if (rc)
8000                 GOTO(lu_fini, rc);
8001
8002         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
8003                                  LUSTRE_MDT_NAME, &mdt_device_type);
8004         if (rc)
8005                 GOTO(mds_fini, rc);
8006 lu_fini:
8007         if (rc)
8008                 lu_kmem_fini(mdt_caches);
8009 mds_fini:
8010         if (rc)
8011                 mds_mod_exit();
8012         return rc;
8013 }
8014
8015 static void __exit mdt_exit(void)
8016 {
8017         class_unregister_type(LUSTRE_MDT_NAME);
8018         mds_mod_exit();
8019         lu_kmem_fini(mdt_caches);
8020 }
8021
8022 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8023 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
8024 MODULE_VERSION(LUSTRE_VERSION_STRING);
8025 MODULE_LICENSE("GPL");
8026
8027 module_init(mdt_init);
8028 module_exit(mdt_exit);