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, false);
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, false);
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, false);
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, false);
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                                               false);
2262                         if (unlikely(rc != 0))
2263                                 RETURN(rc);
2264                 }
2265
2266                 /* step 2: lookup child's fid by name */
2267                 fid_zero(child_fid);
2268                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2269                                 child_fid, &info->mti_spec);
2270                 if (rc == -ENOENT)
2271                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2272
2273                 if (rc != 0)
2274                         GOTO(unlock_parent, rc);
2275
2276                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2277                                         child_fid);
2278                 if (unlikely(IS_ERR(child)))
2279                         GOTO(unlock_parent, rc = PTR_ERR(child));
2280         }
2281
2282         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2283
2284         /* step 3: lock child regardless if it is local or remote. */
2285         LASSERT(child);
2286
2287         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2288                 /* Here, lname is an encoded hash of on-disk name, and
2289                  * client is doing access without encryption key.
2290                  * So we need to compare name hash with the one in the request.
2291                  */
2292                 if (!find_name_matching_hash(info, lname, parent,
2293                                              child)) {
2294                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2295                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2296                         GOTO(out_child, rc = -ENOENT);
2297                 }
2298         }
2299
2300         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2301         if (!mdt_object_exists(child)) {
2302                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2303                                 &child->mot_obj,
2304                                 "Object doesn't exist!");
2305                 GOTO(out_child, rc = -ENOENT);
2306         }
2307
2308         rc = mdt_check_resent_lock(info, child, lhc);
2309         if (rc < 0) {
2310                 GOTO(out_child, rc);
2311         } else if (rc > 0) {
2312                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2313                     !mdt_object_remote(child)) {
2314                         struct md_attr *ma = &info->mti_attr;
2315
2316                         ma->ma_valid = 0;
2317                         ma->ma_need = MA_INODE;
2318                         rc = mdt_attr_get_complex(info, child, ma);
2319                         if (unlikely(rc != 0))
2320                                 GOTO(out_child, rc);
2321
2322                         /* If the file has not been changed for some time, we
2323                          * return not only a LOOKUP lock, but also an UPDATE
2324                          * lock and this might save us RPC on later STAT. For
2325                          * directories, it also let negative dentry cache start
2326                          * working for this dir. */
2327                         if (ma->ma_valid & MA_INODE &&
2328                             ma->ma_attr.la_valid & LA_CTIME &&
2329                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2330                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2331                                 child_bits |= MDS_INODELOCK_UPDATE;
2332                 }
2333
2334                 /* layout lock must be granted in a best-effort way
2335                  * for IT operations */
2336                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2337                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2338                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2339                         if (!CFS_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2340                             exp_connect_layout(info->mti_exp)) {
2341                                 /* try to grant layout lock for regular file. */
2342                                 try_bits = MDS_INODELOCK_LAYOUT;
2343                         }
2344                         /* Acquire DOM lock in advance for data-on-mdt file */
2345                         if (child != parent)
2346                                 try_bits |= MDS_INODELOCK_DOM;
2347                 }
2348
2349                 /*
2350                  * To avoid possible deadlock between batched statahead RPC
2351                  * and rename()/migrate() operation, it should use trylock to
2352                  * obtain the DLM PR ibits lock for file attributes in a
2353                  * batched statahead RPC. A failed trylock means that other
2354                  * users maybe modify the directory simultaneously as in current
2355                  * Lustre design the server only grants read lock to a client.
2356                  *
2357                  * When a trylock failed, the MDT reports the conflict with
2358                  * error code -EBUSY, and stops statahead immediately.
2359                  */
2360                 if (info->mti_batch_env) {
2361                         /*
2362                          * This is a sub stat-ahead request in a batched RPC.
2363                          * However, the @child is a remote object, we just
2364                          * return -EREMOTE here to forbid stat-ahead on it.
2365                          */
2366                         if (mdt_object_remote(child))
2367                                 GOTO(out_child, rc = -EREMOTE);
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, false);
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, false);
2385                         else
2386                                 rc = mdt_object_lock(info, child, lhc,
2387                                                      child_bits, LCK_PR, false);
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         bool cos_incompat = false;
2534         int rc;
2535         ENTRY;
2536
2537         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2538         if (IS_ERR(pobj))
2539                 GOTO(out, rc = PTR_ERR(pobj));
2540
2541         if (mdt_object_remote(pobj))
2542                 cos_incompat = true;
2543
2544         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2545         rc = mdt_parent_lock(info, pobj, parent_lh, name, LCK_PW, cos_incompat);
2546         if (rc != 0)
2547                 GOTO(put_parent, rc);
2548
2549         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2550                         name, child_fid, &info->mti_spec);
2551         if (rc != 0)
2552                 GOTO(unlock_parent, rc);
2553
2554         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2555                 GOTO(unlock_parent, rc = -EREMCHG);
2556
2557         child_lh = &info->mti_lh[MDT_LH_CHILD];
2558         rc = mdt_object_stripes_lock(info, pobj, obj, child_lh, einfo,
2559                                      MDS_INODELOCK_LOOKUP |
2560                                      MDS_INODELOCK_UPDATE,
2561                                      LCK_EX, cos_incompat);
2562         if (rc != 0)
2563                 GOTO(unlock_parent, rc);
2564
2565         if (atomic_read(&obj->mot_open_count)) {
2566                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2567                        PFID(mdt_object_fid(obj)));
2568                 GOTO(unlock_child, rc = -EBUSY);
2569         }
2570
2571         ma->ma_need = 0;
2572         ma->ma_valid = MA_INODE;
2573         ma->ma_attr.la_valid = LA_CTIME;
2574         ma->ma_attr.la_ctime = ctime;
2575
2576         mutex_lock(&obj->mot_lov_mutex);
2577
2578         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2579                         mdt_object_child(obj), name, ma, 0);
2580
2581         mutex_unlock(&obj->mot_lov_mutex);
2582
2583 unlock_child:
2584         mdt_object_stripes_unlock(info, obj, child_lh, einfo, 1);
2585 unlock_parent:
2586         mdt_object_unlock(info, pobj, parent_lh, 1);
2587 put_parent:
2588         mdt_object_put(info->mti_env, pobj);
2589 out:
2590         RETURN(rc);
2591 }
2592
2593 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2594                                         struct mdt_object *obj)
2595 {
2596         struct lu_ucred *uc = lu_ucred(info->mti_env);
2597         struct md_attr *ma = &info->mti_attr;
2598         struct lu_attr *la = &ma->ma_attr;
2599         int rc = 0;
2600         ENTRY;
2601
2602         ma->ma_need = MA_INODE;
2603         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2604         if (rc)
2605                 GOTO(out, rc);
2606
2607         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2608                         rc = -EACCES;
2609
2610         /* we want rbac roles to have precedence over any other
2611          * permission or capability checks
2612          */
2613         if (!uc->uc_rbac_byfid_ops)
2614                 RETURN(-EPERM);
2615         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2616                 RETURN(0);
2617         if (uc->uc_fsuid == la->la_uid) {
2618                 if ((la->la_mode & S_IWUSR) == 0)
2619                         rc = -EACCES;
2620         } else if (uc->uc_fsgid == la->la_gid) {
2621                 if ((la->la_mode & S_IWGRP) == 0)
2622                         rc = -EACCES;
2623         } else if ((la->la_mode & S_IWOTH) == 0) {
2624                         rc = -EACCES;
2625         }
2626
2627 out:
2628         RETURN(rc);
2629 }
2630
2631 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2632                          s64 ctime)
2633 {
2634         struct mdt_device *mdt = info->mti_mdt;
2635         struct mdt_object *obj = NULL;
2636         struct linkea_data ldata = { NULL };
2637         struct lu_buf *buf = &info->mti_big_buf;
2638         struct lu_name *name = &info->mti_name;
2639         struct lu_fid *pfid = &info->mti_tmp_fid1;
2640         struct link_ea_header *leh;
2641         struct link_ea_entry *lee;
2642         int reclen, count, rc = 0;
2643         ENTRY;
2644
2645         if (!fid_is_sane(fid))
2646                 GOTO(out, rc = -EINVAL);
2647
2648         if (!fid_is_namespace_visible(fid))
2649                 GOTO(out, rc = -EINVAL);
2650
2651         obj = mdt_object_find(info->mti_env, mdt, fid);
2652         if (IS_ERR(obj))
2653                 GOTO(out, rc = PTR_ERR(obj));
2654
2655         if (mdt_object_remote(obj))
2656                 GOTO(out, rc = -EREMOTE);
2657         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2658                 GOTO(out, rc = -ENOENT);
2659
2660         rc = mdt_rmfid_check_permission(info, obj);
2661         if (rc)
2662                 GOTO(out, rc);
2663
2664         /* take LinkEA */
2665         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2666         if (!buf->lb_buf)
2667                 GOTO(out, rc = -ENOMEM);
2668
2669         ldata.ld_buf = buf;
2670         rc = mdt_links_read(info, obj, &ldata);
2671         if (rc)
2672                 GOTO(out, rc);
2673
2674         leh = buf->lb_buf;
2675         lee = (struct link_ea_entry *)(leh + 1);
2676         for (count = 0; count < leh->leh_reccount; count++) {
2677                 /* remove every hardlink */
2678                 linkea_entry_unpack(lee, &reclen, name, pfid);
2679                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2680                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2681                 if (rc)
2682                         break;
2683         }
2684
2685 out:
2686         if (obj && !IS_ERR(obj))
2687                 mdt_object_put(info->mti_env, obj);
2688         if (info->mti_big_buf.lb_buf)
2689                 lu_buf_free(&info->mti_big_buf);
2690
2691         RETURN(rc);
2692 }
2693
2694 static int mdt_rmfid(struct tgt_session_info *tsi)
2695 {
2696         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2697         struct mdt_body *reqbody;
2698         struct lu_fid *fids, *rfids;
2699         int bufsize, rc;
2700         __u32 *rcs;
2701         int i, nr;
2702         ENTRY;
2703
2704         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2705         if (reqbody == NULL)
2706                 RETURN(-EPROTO);
2707         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2708                                        RCL_CLIENT);
2709         nr = bufsize / sizeof(struct lu_fid);
2710         if (nr * sizeof(struct lu_fid) != bufsize)
2711                 RETURN(-EINVAL);
2712         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2713                              RCL_SERVER, nr * sizeof(__u32));
2714         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2715                              RCL_SERVER, nr * sizeof(struct lu_fid));
2716         rc = req_capsule_server_pack(tsi->tsi_pill);
2717         if (rc)
2718                 GOTO(out, rc = err_serious(rc));
2719         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2720         if (fids == NULL)
2721                 RETURN(-EPROTO);
2722         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2723         LASSERT(rcs);
2724         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2725         LASSERT(rfids);
2726
2727         mdt_init_ucred(mti, reqbody);
2728         for (i = 0; i < nr; i++) {
2729                 rfids[i] = fids[i];
2730                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2731         }
2732         mdt_exit_ucred(mti);
2733
2734 out:
2735         RETURN(rc);
2736 }
2737
2738 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2739                          void *karg, void __user *uarg);
2740
2741 int mdt_io_set_info(struct tgt_session_info *tsi)
2742 {
2743         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2744         struct ost_body         *body = NULL, *repbody;
2745         void                    *key, *val = NULL;
2746         int                      keylen, vallen, rc = 0;
2747         bool                     is_grant_shrink;
2748
2749         ENTRY;
2750
2751         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2752         if (key == NULL) {
2753                 DEBUG_REQ(D_HA, req, "no set_info key");
2754                 RETURN(err_serious(-EFAULT));
2755         }
2756         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2757                                       RCL_CLIENT);
2758
2759         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2760         if (val == NULL) {
2761                 DEBUG_REQ(D_HA, req, "no set_info val");
2762                 RETURN(err_serious(-EFAULT));
2763         }
2764         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2765                                       RCL_CLIENT);
2766
2767         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2768         if (is_grant_shrink)
2769                 /* In this case the value is actually an RMF_OST_BODY, so we
2770                  * transmutate the type of this PTLRPC */
2771                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2772
2773         rc = req_capsule_server_pack(tsi->tsi_pill);
2774         if (rc < 0)
2775                 RETURN(rc);
2776
2777         if (is_grant_shrink) {
2778                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2779
2780                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2781                 *repbody = *body;
2782
2783                 /** handle grant shrink, similar to a read request */
2784                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2785                                        &repbody->oa);
2786         } else {
2787                 CERROR("%s: Unsupported key %s\n",
2788                        tgt_name(tsi->tsi_tgt), (char *)key);
2789                 rc = -EOPNOTSUPP;
2790         }
2791
2792         RETURN(rc);
2793 }
2794
2795
2796 static int mdt_set_info(struct tgt_session_info *tsi)
2797 {
2798         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2799         char                    *key;
2800         void                    *val;
2801         int                      keylen, vallen, rc = 0;
2802
2803         ENTRY;
2804
2805         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2806         if (key == NULL) {
2807                 DEBUG_REQ(D_HA, req, "no set_info key");
2808                 RETURN(err_serious(-EFAULT));
2809         }
2810
2811         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2812                                       RCL_CLIENT);
2813
2814         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2815         if (val == NULL) {
2816                 DEBUG_REQ(D_HA, req, "no set_info val");
2817                 RETURN(err_serious(-EFAULT));
2818         }
2819
2820         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2821                                       RCL_CLIENT);
2822
2823         /* Swab any part of val you need to here */
2824         if (KEY_IS(KEY_READ_ONLY)) {
2825                 spin_lock(&req->rq_export->exp_lock);
2826                 if (*(__u32 *)val)
2827                         *exp_connect_flags_ptr(req->rq_export) |=
2828                                 OBD_CONNECT_RDONLY;
2829                 else
2830                         *exp_connect_flags_ptr(req->rq_export) &=
2831                                 ~OBD_CONNECT_RDONLY;
2832                 spin_unlock(&req->rq_export->exp_lock);
2833         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2834                 struct changelog_setinfo *cs = val;
2835
2836                 if (vallen != sizeof(*cs)) {
2837                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2838                                tgt_name(tsi->tsi_tgt), vallen);
2839                         RETURN(-EINVAL);
2840                 }
2841                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2842                         __swab64s(&cs->cs_recno);
2843                         __swab32s(&cs->cs_id);
2844                 }
2845
2846                 if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
2847                         RETURN(-EACCES);
2848                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2849                                    vallen, val, NULL);
2850         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2851                 if (vallen > 0)
2852                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2853         } else {
2854                 RETURN(-EINVAL);
2855         }
2856         RETURN(rc);
2857 }
2858
2859 static int mdt_readpage(struct tgt_session_info *tsi)
2860 {
2861         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2862         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2863         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2864         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2865         struct mdt_body         *repbody;
2866         int                      rc;
2867         int                      i;
2868
2869         ENTRY;
2870
2871         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2872                 RETURN(err_serious(-ENOMEM));
2873
2874         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2875         if (repbody == NULL || reqbody == NULL)
2876                 RETURN(err_serious(-EFAULT));
2877
2878         /*
2879          * prepare @rdpg before calling lower layers and transfer itself. Here
2880          * reqbody->size contains offset of where to start to read and
2881          * reqbody->nlink contains number bytes to read.
2882          */
2883         rdpg->rp_hash = reqbody->mbo_size;
2884         if (rdpg->rp_hash != reqbody->mbo_size) {
2885                 CERROR("Invalid hash: %#llx != %#llx\n",
2886                        rdpg->rp_hash, reqbody->mbo_size);
2887                 RETURN(-EFAULT);
2888         }
2889
2890         rdpg->rp_attrs = reqbody->mbo_mode;
2891         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2892                 rdpg->rp_attrs |= LUDA_64BITHASH;
2893         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2894                                 exp_max_brw_size(tsi->tsi_exp));
2895         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2896                           PAGE_SHIFT;
2897         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2898         if (rdpg->rp_pages == NULL)
2899                 RETURN(-ENOMEM);
2900
2901         for (i = 0; i < rdpg->rp_npages; ++i) {
2902                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2903                 if (rdpg->rp_pages[i] == NULL)
2904                         GOTO(free_rdpg, rc = -ENOMEM);
2905         }
2906
2907         /* call lower layers to fill allocated pages with directory data */
2908         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2909         if (rc < 0)
2910                 GOTO(free_rdpg, rc);
2911
2912         /* send pages to client */
2913         rc = tgt_sendpage(tsi, rdpg, rc);
2914
2915         EXIT;
2916 free_rdpg:
2917
2918         for (i = 0; i < rdpg->rp_npages; i++)
2919                 if (rdpg->rp_pages[i] != NULL)
2920                         __free_page(rdpg->rp_pages[i]);
2921         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2922
2923         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2924                 RETURN(0);
2925
2926         return rc;
2927 }
2928
2929 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2930 {
2931         struct lu_ucred *uc = mdt_ucred_check(info);
2932         struct lu_attr *attr = &info->mti_attr.ma_attr;
2933
2934         if (uc == NULL)
2935                 return -EINVAL;
2936
2937         if (op != REINT_SETATTR) {
2938                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2939                         attr->la_uid = uc->uc_fsuid;
2940                 /* for S_ISGID, inherit gid from his parent, such work will be
2941                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2942                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2943                         attr->la_gid = uc->uc_fsgid;
2944         }
2945
2946         return 0;
2947 }
2948
2949 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2950 {
2951         return op == REINT_OPEN &&
2952              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2953 }
2954
2955 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2956 {
2957         struct req_capsule *pill = info->mti_pill;
2958
2959         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2960                                   RCL_SERVER) &&
2961             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2962                                   RCL_CLIENT)) {
2963                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2964                                          RCL_CLIENT) != 0)
2965                         /* pre-set size in server part with max size */
2966                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2967                                              RCL_SERVER,
2968                                              req_capsule_ptlreq(pill) ?
2969                                              OBD_MAX_DEFAULT_EA_SIZE :
2970                                              MAX_MD_SIZE_OLD);
2971                 else
2972                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2973                                              RCL_SERVER, 0);
2974         }
2975 }
2976
2977 static int mdt_reint_internal(struct mdt_thread_info *info,
2978                               struct mdt_lock_handle *lhc,
2979                               __u32 op)
2980 {
2981         struct req_capsule      *pill = info->mti_pill;
2982         struct mdt_body         *repbody;
2983         int                      rc = 0, rc2;
2984
2985         ENTRY;
2986
2987         rc = mdt_reint_unpack(info, op);
2988         if (rc != 0) {
2989                 CERROR("Can't unpack reint, rc %d\n", rc);
2990                 RETURN(err_serious(rc));
2991         }
2992
2993
2994         /* check if the file system is set to readonly. O_RDONLY open
2995          * is still allowed even the file system is set to readonly mode */
2996         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2997                 RETURN(err_serious(-EROFS));
2998
2999         /* for replay (no_create) lmm is not needed, client has it already */
3000         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3001                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3002                                      DEF_REP_MD_SIZE);
3003
3004         /* llog cookies are always 0, the field is kept for compatibility */
3005         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3006                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
3007
3008         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3009          * by default. If the target object has more ACL entries, then
3010          * enlarge the buffer when necessary. */
3011         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3012                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3013                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3014
3015         mdt_preset_secctx_size(info);
3016         mdt_preset_encctx_size(info);
3017
3018         rc = req_capsule_server_pack(pill);
3019         if (rc != 0) {
3020                 CERROR("Can't pack response, rc %d\n", rc);
3021                 RETURN(err_serious(rc));
3022         }
3023
3024         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
3025                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
3026                 LASSERT(repbody);
3027                 repbody->mbo_eadatasize = 0;
3028                 repbody->mbo_aclsize = 0;
3029         }
3030
3031         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
3032
3033         /* for replay no cookkie / lmm need, because client have this already */
3034         if (info->mti_spec.no_create)
3035                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3036                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
3037
3038         rc = mdt_init_ucred_reint(info);
3039         if (rc)
3040                 GOTO(out_shrink, rc);
3041
3042         rc = mdt_fix_attr_ucred(info, op);
3043         if (rc != 0)
3044                 GOTO(out_ucred, rc = err_serious(rc));
3045
3046         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
3047         if (rc < 0) {
3048                 GOTO(out_ucred, rc);
3049         } else if (rc == 1) {
3050                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
3051                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
3052                 GOTO(out_ucred, rc);
3053         }
3054         rc = mdt_reint_rec(info, lhc);
3055         EXIT;
3056 out_ucred:
3057         mdt_exit_ucred(info);
3058 out_shrink:
3059         mdt_client_compatibility(info);
3060
3061         rc2 = mdt_fix_reply(info);
3062         if (rc == 0)
3063                 rc = rc2;
3064
3065         /*
3066          * Data-on-MDT optimization - read data along with OPEN and return it
3067          * in reply when possible.
3068          */
3069         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
3070                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
3071                                           &lhc->mlh_reg_lh);
3072
3073         return rc;
3074 }
3075
3076 static long mdt_reint_opcode(struct ptlrpc_request *req,
3077                              const struct req_format **fmt)
3078 {
3079         struct mdt_device       *mdt;
3080         struct mdt_rec_reint    *rec;
3081         long                     opc;
3082
3083         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3084         if (rec != NULL) {
3085                 opc = rec->rr_opcode;
3086                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3087                 if (opc < REINT_MAX && fmt[opc] != NULL)
3088                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3089                 else {
3090                         mdt = mdt_exp2dev(req->rq_export);
3091                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3092                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3093                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3094                         opc = err_serious(-EFAULT);
3095                 }
3096         } else {
3097                 opc = err_serious(-EFAULT);
3098         }
3099         return opc;
3100 }
3101
3102 static int mdt_reint(struct tgt_session_info *tsi)
3103 {
3104         long opc;
3105         int  rc;
3106         static const struct req_format *reint_fmts[REINT_MAX] = {
3107                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3108                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3109                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3110                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3111                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3112                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3113                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3114                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3115                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3116                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3117         };
3118
3119         ENTRY;
3120
3121         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3122         if (opc >= 0) {
3123                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3124                 /*
3125                  * No lock possible here from client to pass it to reint code
3126                  * path.
3127                  */
3128                 rc = mdt_reint_internal(info, NULL, opc);
3129                 mdt_thread_info_fini(info);
3130         } else {
3131                 rc = opc;
3132         }
3133
3134         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3135         RETURN(rc);
3136 }
3137
3138 /* this should sync the whole device */
3139 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3140 {
3141         struct dt_device *dt = mdt->mdt_bottom;
3142         int rc;
3143         ENTRY;
3144
3145         rc = dt->dd_ops->dt_sync(env, dt);
3146         RETURN(rc);
3147 }
3148
3149 /* this should sync this object */
3150 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3151                            struct mdt_object *mo)
3152 {
3153         int rc = 0;
3154
3155         ENTRY;
3156
3157         if (!mdt_object_exists(mo)) {
3158                 CWARN("%s: non existing object "DFID": rc = %d\n",
3159                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3160                       -ESTALE);
3161                 RETURN(-ESTALE);
3162         }
3163
3164         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3165                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3166                 dt_obj_version_t version;
3167
3168                 version = dt_version_get(env, mdt_obj2dt(mo));
3169                 if (version > tgt->lut_obd->obd_last_committed)
3170                         rc = mo_object_sync(env, mdt_object_child(mo));
3171         } else {
3172                 rc = mo_object_sync(env, mdt_object_child(mo));
3173         }
3174
3175         RETURN(rc);
3176 }
3177
3178 static int mdt_sync(struct tgt_session_info *tsi)
3179 {
3180         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3181         struct req_capsule      *pill = tsi->tsi_pill;
3182         struct mdt_body         *body;
3183         ktime_t                  kstart = ktime_get();
3184         int                      rc;
3185
3186         ENTRY;
3187
3188         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3189                 RETURN(err_serious(-ENOMEM));
3190
3191         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3192                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3193         } else {
3194                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3195
3196                 if (unlikely(info->mti_object == NULL))
3197                         RETURN(-EPROTO);
3198
3199                 /* sync an object */
3200                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3201                                      info->mti_object);
3202                 if (rc == 0) {
3203                         const struct lu_fid *fid;
3204                         struct lu_attr *la = &info->mti_attr.ma_attr;
3205
3206                         info->mti_attr.ma_need = MA_INODE;
3207                         info->mti_attr.ma_valid = 0;
3208                         rc = mdt_attr_get_complex(info, info->mti_object,
3209                                                   &info->mti_attr);
3210                         if (rc == 0) {
3211                                 body = req_capsule_server_get(pill,
3212                                                               &RMF_MDT_BODY);
3213                                 fid = mdt_object_fid(info->mti_object);
3214                                 mdt_pack_attr2body(info, body, la, fid);
3215                         }
3216                 }
3217                 mdt_thread_info_fini(info);
3218         }
3219         if (rc == 0)
3220                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3221                                  ktime_us_delta(ktime_get(), kstart));
3222
3223         RETURN(rc);
3224 }
3225
3226 static int mdt_data_sync(struct tgt_session_info *tsi)
3227 {
3228         struct mdt_thread_info *info;
3229         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3230         struct ost_body *body = tsi->tsi_ost_body;
3231         struct ost_body *repbody;
3232         struct mdt_object *mo = NULL;
3233         struct md_attr *ma;
3234         int rc = 0;
3235
3236         ENTRY;
3237
3238         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3239
3240         /* if no fid is specified then do nothing,
3241          * device sync is done via MDS_SYNC */
3242         if (fid_is_zero(&tsi->tsi_fid))
3243                 RETURN(0);
3244
3245         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3246         if (IS_ERR(mo))
3247                 RETURN(PTR_ERR(mo));
3248
3249         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3250         if (rc)
3251                 GOTO(put, rc);
3252
3253         repbody->oa.o_oi = body->oa.o_oi;
3254         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3255
3256         info = tsi2mdt_info(tsi);
3257         ma = &info->mti_attr;
3258         ma->ma_need = MA_INODE;
3259         ma->ma_valid = 0;
3260         rc = mdt_attr_get_complex(info, mo, ma);
3261         if (rc == 0)
3262                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3263         else
3264                 rc = 0;
3265         mdt_thread_info_fini(info);
3266
3267         EXIT;
3268 put:
3269         if (mo != NULL)
3270                 mdt_object_put(tsi->tsi_env, mo);
3271         return rc;
3272 }
3273
3274 /*
3275  * Handle quota control requests to consult current usage/limit, but also
3276  * to configure quota enforcement
3277  */
3278 static int mdt_quotactl(struct tgt_session_info *tsi)
3279 {
3280         struct obd_export *exp  = tsi->tsi_exp;
3281         struct req_capsule *pill = tsi->tsi_pill;
3282         struct obd_quotactl *oqctl, *repoqc;
3283         int id, rc;
3284         struct mdt_device *mdt = mdt_exp2dev(exp);
3285         struct lu_device *qmt = mdt->mdt_qmt_dev;
3286         struct lu_nodemap *nodemap;
3287         ENTRY;
3288
3289         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3290         if (!oqctl)
3291                 RETURN(err_serious(-EPROTO));
3292
3293         rc = req_capsule_server_pack(pill);
3294         if (rc)
3295                 RETURN(err_serious(rc));
3296
3297         nodemap = nodemap_get_from_exp(exp);
3298         if (IS_ERR(nodemap))
3299                 RETURN(PTR_ERR(nodemap));
3300
3301         switch (oqctl->qc_cmd) {
3302                 /* master quotactl */
3303         case Q_SETINFO:
3304         case Q_SETQUOTA:
3305         case LUSTRE_Q_SETDEFAULT:
3306         case LUSTRE_Q_SETQUOTAPOOL:
3307         case LUSTRE_Q_SETINFOPOOL:
3308         case LUSTRE_Q_SETDEFAULT_POOL:
3309         case LUSTRE_Q_DELETEQID:
3310         case LUSTRE_Q_RESETQID:
3311                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3312                                           oqctl->qc_id))
3313                         GOTO(out_nodemap, rc = -EPERM);
3314                 fallthrough;
3315         case Q_GETINFO:
3316         case Q_GETQUOTA:
3317         case LUSTRE_Q_GETDEFAULT:
3318         case LUSTRE_Q_GETQUOTAPOOL:
3319         case LUSTRE_Q_GETINFOPOOL:
3320         case LUSTRE_Q_GETDEFAULT_POOL:
3321                 if (qmt == NULL)
3322                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3323                 /* slave quotactl */
3324                 fallthrough;
3325         case Q_GETOINFO:
3326         case Q_GETOQUOTA:
3327                 break;
3328         default:
3329                 rc = -EFAULT;
3330                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3331                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3332                 GOTO(out_nodemap, rc);
3333         }
3334
3335         id = oqctl->qc_id;
3336         switch (oqctl->qc_type) {
3337         case USRQUOTA:
3338                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3339                                     NODEMAP_CLIENT_TO_FS, id);
3340                 break;
3341         case GRPQUOTA:
3342                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3343                                     NODEMAP_CLIENT_TO_FS, id);
3344                 break;
3345         case PRJQUOTA:
3346                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3347                                     NODEMAP_CLIENT_TO_FS, id);
3348                 break;
3349         default:
3350                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3351         }
3352         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3353         if (repoqc == NULL)
3354                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3355
3356         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3357                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3358
3359         if (oqctl->qc_id != id)
3360                 swap(oqctl->qc_id, id);
3361
3362         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3363                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3364                         RETURN(-EINPROGRESS);
3365         }
3366
3367         switch (oqctl->qc_cmd) {
3368
3369         case Q_GETINFO:
3370         case Q_SETINFO:
3371         case Q_SETQUOTA:
3372         case Q_GETQUOTA:
3373         case LUSTRE_Q_SETDEFAULT:
3374         case LUSTRE_Q_GETDEFAULT:
3375         case LUSTRE_Q_SETQUOTAPOOL:
3376         case LUSTRE_Q_GETQUOTAPOOL:
3377         case LUSTRE_Q_SETINFOPOOL:
3378         case LUSTRE_Q_GETINFOPOOL:
3379         case LUSTRE_Q_SETDEFAULT_POOL:
3380         case LUSTRE_Q_GETDEFAULT_POOL:
3381         case LUSTRE_Q_DELETEQID:
3382         case LUSTRE_Q_RESETQID:
3383                 /* forward quotactl request to QMT */
3384                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3385                 break;
3386
3387         case Q_GETOINFO:
3388         case Q_GETOQUOTA:
3389                 /* slave quotactl */
3390                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3391                                    oqctl);
3392                 break;
3393
3394         default:
3395                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3396                 GOTO(out_nodemap, rc = -EFAULT);
3397         }
3398
3399         if (oqctl->qc_id != id)
3400                 swap(oqctl->qc_id, id);
3401
3402         QCTL_COPY_NO_PNAME(repoqc, oqctl);
3403         EXIT;
3404
3405 out_nodemap:
3406         nodemap_putref(nodemap);
3407
3408         return rc;
3409 }
3410
3411 /** clone llog ctxt from child (mdd)
3412  * This allows remote llog (replicator) access.
3413  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3414  * context was originally set up, or we can handle them directly.
3415  * I choose the latter, but that means I need any llog
3416  * contexts set up by child to be accessable by the mdt.  So we clone the
3417  * context into our context list here.
3418  */
3419 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3420                                int idx)
3421 {
3422         struct md_device  *next = mdt->mdt_child;
3423         struct llog_ctxt *ctxt;
3424         int rc;
3425
3426         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3427                 return 0;
3428
3429         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3430         if (rc || ctxt == NULL) {
3431                 return 0;
3432         }
3433
3434         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3435         if (rc)
3436                 CERROR("Can't set mdt ctxt %d\n", rc);
3437
3438         return rc;
3439 }
3440
3441 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3442                                  struct mdt_device *mdt, int idx)
3443 {
3444         struct llog_ctxt *ctxt;
3445
3446         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3447         if (ctxt == NULL)
3448                 return 0;
3449         /* Put once for the get we just did, and once for the clone */
3450         llog_ctxt_put(ctxt);
3451         llog_ctxt_put(ctxt);
3452         return 0;
3453 }
3454
3455 /*
3456  * sec context handlers
3457  */
3458 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3459 {
3460         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3461
3462         return 0;
3463 }
3464
3465 /*
3466  * quota request handlers
3467  */
3468 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3469 {
3470         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3471         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3472         int                      rc;
3473         ENTRY;
3474
3475         if (qmt == NULL)
3476                 RETURN(err_serious(-EOPNOTSUPP));
3477
3478         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3479         RETURN(rc);
3480 }
3481
3482 struct mdt_object *mdt_object_new(const struct lu_env *env,
3483                                   struct mdt_device *d,
3484                                   const struct lu_fid *f)
3485 {
3486         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3487         struct lu_object *o;
3488         struct mdt_object *m;
3489         ENTRY;
3490
3491         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3492         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3493         if (unlikely(IS_ERR(o)))
3494                 m = (struct mdt_object *)o;
3495         else
3496                 m = mdt_obj(o);
3497         RETURN(m);
3498 }
3499
3500 struct mdt_object *mdt_object_find(const struct lu_env *env,
3501                                    struct mdt_device *d,
3502                                    const struct lu_fid *f)
3503 {
3504         struct lu_object *o;
3505         struct mdt_object *m;
3506         ENTRY;
3507
3508         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3509         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3510         if (unlikely(IS_ERR(o)))
3511                 m = (struct mdt_object *)o;
3512         else
3513                 m = mdt_obj(o);
3514
3515         RETURN(m);
3516 }
3517
3518 /**
3519  * Asyncronous commit for mdt device.
3520  *
3521  * Pass asynchonous commit call down the MDS stack.
3522  *
3523  * \param env environment
3524  * \param mdt the mdt device
3525  */
3526 static void mdt_device_commit_async(const struct lu_env *env,
3527                                     struct mdt_device *mdt)
3528 {
3529         struct dt_device *dt = mdt->mdt_bottom;
3530         int rc;
3531         ENTRY;
3532
3533         rc = dt->dd_ops->dt_commit_async(env, dt);
3534         if (unlikely(rc != 0))
3535                 CWARN("%s: async commit start failed: rc = %d\n",
3536                       mdt_obd_name(mdt), rc);
3537         atomic_inc(&mdt->mdt_async_commit_count);
3538         EXIT;
3539 }
3540
3541 /**
3542  * Mark the lock as "synchonous".
3543  *
3544  * Mark the lock to deffer transaction commit to the unlock time.
3545  *
3546  * \param lock the lock to mark as "synchonous"
3547  *
3548  * \see mdt_is_lock_sync
3549  * \see mdt_save_lock
3550  */
3551 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3552 {
3553         lock->l_ast_data = (void*)1;
3554 }
3555
3556 /**
3557  * Check whehter the lock "synchonous" or not.
3558  *
3559  * \param lock the lock to check
3560  * \retval 1 the lock is "synchonous"
3561  * \retval 0 the lock isn't "synchronous"
3562  *
3563  * \see mdt_set_lock_sync
3564  * \see mdt_save_lock
3565  */
3566 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3567 {
3568         return lock->l_ast_data != NULL;
3569 }
3570
3571 /**
3572  * Blocking AST for mdt locks.
3573  *
3574  * Starts transaction commit if in case of COS lock conflict or
3575  * deffers such a commit to the mdt_save_lock.
3576  *
3577  * \param lock the lock which blocks a request or cancelling lock
3578  * \param desc unused
3579  * \param data unused
3580  * \param flag indicates whether this cancelling or blocking callback
3581  * \retval 0
3582  * \see ldlm_blocking_ast_nocheck
3583  */
3584 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3585                      void *data, int flag)
3586 {
3587         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3588         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3589         struct ldlm_cb_set_arg *arg = data;
3590         bool commit_async = false;
3591         int rc;
3592         ENTRY;
3593
3594         if (flag == LDLM_CB_CANCELING)
3595                 RETURN(0);
3596
3597         lock_res_and_lock(lock);
3598         if (lock->l_blocking_ast != mdt_blocking_ast) {
3599                 unlock_res_and_lock(lock);
3600                 RETURN(0);
3601         }
3602
3603         /* A blocking ast may be sent from ldlm_lock_decref_internal
3604          * when the last reference to a local lock was released and
3605          * during blocking event from ldlm_work_bl_ast_lock().
3606          * The 'data' parameter is l_ast_data in the first case and
3607          * callback arguments in the second one. Distinguish them by that.
3608          */
3609         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3610                 goto skip_cos_checks;
3611
3612         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3613                 if (mdt_cos_is_enabled(mdt)) {
3614                         if (!arg->bl_desc->bl_same_client)
3615                                 mdt_set_lock_sync(lock);
3616                 } else if (mdt_slc_is_enabled(mdt) &&
3617                            arg->bl_desc->bl_cos_incompat) {
3618                         mdt_set_lock_sync(lock);
3619                         /*
3620                          * we may do extra commit here, but there is a small
3621                          * window to miss a commit: lock was unlocked (saved),
3622                          * then a conflict lock queued and we come here, but
3623                          * REP-ACK not received, so lock was not converted to
3624                          * COS mode yet.
3625                          * Fortunately this window is quite small, so the
3626                          * extra commit should be rare (not to say distributed
3627                          * operation is rare too).
3628                          */
3629                         commit_async = true;
3630                 }
3631         } else if (lock->l_req_mode == LCK_COS) {
3632                 commit_async = true;
3633         }
3634
3635 skip_cos_checks:
3636         rc = ldlm_blocking_ast_nocheck(lock);
3637
3638         if (commit_async) {
3639                 struct lu_env env;
3640
3641                 rc = lu_env_init(&env, LCT_LOCAL);
3642                 if (unlikely(rc != 0))
3643                         CWARN("%s: lu_env initialization failed, cannot "
3644                               "start asynchronous commit: rc = %d\n",
3645                               obd->obd_name, rc);
3646                 else
3647                         mdt_device_commit_async(&env, mdt);
3648                 lu_env_fini(&env);
3649         }
3650         RETURN(rc);
3651 }
3652
3653 /*
3654  * Blocking AST for cross-MDT lock
3655  *
3656  * Discard lock from uncommitted_slc_locks and cancel it.
3657  *
3658  * \param lock  the lock which blocks a request or cancelling lock
3659  * \param desc  unused
3660  * \param data  unused
3661  * \param flag  indicates whether this cancelling or blocking callback
3662  * \retval      0 on success
3663  * \retval      negative number on error
3664  */
3665 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3666                             void *data, int flag)
3667 {
3668         int rc = 0;
3669         ENTRY;
3670
3671         switch (flag) {
3672         case LDLM_CB_BLOCKING: {
3673                 struct lustre_handle lockh;
3674
3675                 ldlm_lock2handle(lock, &lockh);
3676                 rc = ldlm_cli_cancel(&lockh,
3677                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3678                 if (rc < 0) {
3679                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3680                         RETURN(rc);
3681                 }
3682                 break;
3683         }
3684         case LDLM_CB_CANCELING: {
3685                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3686                 struct mdt_device *mdt =
3687                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3688
3689                 LDLM_DEBUG(lock, "Revoke remote lock");
3690
3691                 /* discard slc lock here so that it can be cleaned anytime,
3692                  * especially for cleanup_resource() */
3693                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3694
3695                 /* once we cache lock, l_ast_data is set to mdt_object */
3696                 if (lock->l_ast_data != NULL) {
3697                         struct mdt_object *mo = lock->l_ast_data;
3698                         struct lu_env env;
3699
3700                         rc = lu_env_init(&env, LCT_MD_THREAD);
3701                         if (unlikely(rc != 0)) {
3702                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3703                                       obd->obd_name, mo,
3704                                       PFID(mdt_object_fid(mo)), rc);
3705                                 RETURN(rc);
3706                         }
3707
3708                         if (lock->l_policy_data.l_inodebits.bits &
3709                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3710                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3711                                 mo->mot_cache_attr = 0;
3712                         }
3713                         mdt_object_put(&env, mo);
3714                         lu_env_fini(&env);
3715                 }
3716                 break;
3717         }
3718         default:
3719                 LBUG();
3720         }
3721
3722         RETURN(rc);
3723 }
3724
3725 int mdt_check_resent_lock(struct mdt_thread_info *info,
3726                           struct mdt_object *mo,
3727                           struct mdt_lock_handle *lhc)
3728 {
3729         /* the lock might already be gotten in ldlm_handle_enqueue() */
3730         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3731                 struct ptlrpc_request *req = mdt_info_req(info);
3732                 struct ldlm_lock      *lock;
3733
3734                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3735                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3736                 if (lock == NULL) {
3737                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3738                          * a resend case, however, it could be already destroyed
3739                          * due to client eviction or a raced cancel RPC. */
3740                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3741                                           lhc->mlh_reg_lh.cookie);
3742                         RETURN(-ESTALE);
3743                 }
3744
3745                 if (!fid_res_name_eq(mdt_object_fid(mo),
3746                                      &lock->l_resource->lr_name)) {
3747                         CWARN("%s: Although resent, but still not "
3748                               "get child lock:"DFID"\n",
3749                               info->mti_exp->exp_obd->obd_name,
3750                               PFID(mdt_object_fid(mo)));
3751                         LDLM_LOCK_PUT(lock);
3752                         RETURN(-EPROTO);
3753                 }
3754                 LDLM_LOCK_PUT(lock);
3755                 return 0;
3756         }
3757         return 1;
3758 }
3759
3760 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3761 {
3762         mdt_object_get(NULL, lock->l_ast_data);
3763 }
3764
3765 static int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3766                                       struct mdt_object *obj,
3767                                       struct lustre_handle *lh,
3768                                       enum ldlm_mode mode,
3769                                       union ldlm_policy_data *policy,
3770                                       struct ldlm_res_id *res_id,
3771                                       bool cache)
3772 {
3773         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3774         int rc;
3775
3776         LASSERT(mdt_object_remote(obj));
3777
3778         memset(einfo, 0, sizeof(*einfo));
3779         einfo->ei_type = LDLM_IBITS;
3780         einfo->ei_mode = mode;
3781         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3782         einfo->ei_cb_cp = ldlm_completion_ast;
3783         einfo->ei_enq_slave = 0;
3784         einfo->ei_res_id = res_id;
3785         einfo->ei_req_slot = 1;
3786         if (cache) {
3787                 /*
3788                  * if we cache lock, couple lock with mdt_object, so that object
3789                  * can be easily found in lock ASTs.
3790                  */
3791                 einfo->ei_cbdata = obj;
3792                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3793         }
3794
3795         rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), lh, einfo,
3796                             policy);
3797         if (rc) {
3798                 lh->cookie = 0ull;
3799                 return rc;
3800         }
3801
3802         /* other components like LFSCK can use lockless access
3803          * and populate cache, so we better invalidate it
3804          */
3805         if (policy->l_inodebits.bits &
3806             (MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR))
3807                 mo_invalidate(mti->mti_env, mdt_object_child(obj));
3808
3809         return 0;
3810 }
3811
3812 /*
3813  * Helper function to take PDO and hash lock.
3814  *
3815  * if \a pdo_lock is false, don't take PDO lock, this is case in rename.
3816  */
3817 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3818                         struct mdt_lock_handle *lh, const struct lu_name *name,
3819                         enum ldlm_mode mode, bool pdo_lock, bool cos_incompat)
3820 {
3821         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3822         union ldlm_policy_data *policy = &info->mti_policy;
3823         struct ldlm_res_id *res_id = &info->mti_res_id;
3824         /*
3825          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it is never going to
3826          * be sent to client and we do not want it slowed down due to possible
3827          * cancels.
3828          */
3829         __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3830         __u64 *cookie = NULL;
3831         int rc;
3832
3833         LASSERT(obj);
3834
3835         /* check for exists after object is locked */
3836         if (!mdt_object_exists(obj))
3837                 /* Non-existent object shouldn't have PDO lock */
3838                 return -ESTALE;
3839
3840         /* Non-dir object shouldn't have PDO lock */
3841         if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
3842                 return -ENOTDIR;
3843
3844         if (cos_incompat) {
3845                 LASSERT(mode == LCK_PW || mode == LCK_EX);
3846                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3847         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3848                 dlmflags |= LDLM_FL_COS_ENABLED;
3849         }
3850
3851         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
3852         policy->l_inodebits.try_bits = 0;
3853         policy->l_inodebits.li_gid = 0;
3854         fid_build_reg_res_name(mdt_object_fid(obj), res_id);
3855         if (info->mti_exp)
3856                 cookie = &info->mti_exp->exp_handle.h_cookie;
3857
3858         mdt_lock_pdo_init(lh, mode, name);
3859         mdt_lock_pdo_mode(info, obj, lh);
3860         if (lh->mlh_pdo_mode != LCK_NL) {
3861                 if (pdo_lock) {
3862                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3863                                           lh->mlh_pdo_mode, policy, res_id,
3864                                           dlmflags, cookie);
3865                         if (rc) {
3866                                 mdt_object_unlock(info, obj, lh, 1);
3867                                 return rc;
3868                         }
3869                 }
3870                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3871         }
3872
3873         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3874                           policy, res_id, dlmflags, cookie);
3875         if (rc)
3876                 mdt_object_unlock(info, obj, lh, 1);
3877         else if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK) &&
3878                    lh->mlh_pdo_hash != 0 &&
3879                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3880                 CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3881
3882         return rc;
3883 }
3884
3885 int mdt_object_lock_internal(struct mdt_thread_info *info,
3886                              struct mdt_object *obj, const struct lu_fid *fid,
3887                              struct mdt_lock_handle *lh, __u64 *ibits,
3888                              __u64 trybits, bool cache, bool cos_incompat)
3889 {
3890         union ldlm_policy_data *policy = &info->mti_policy;
3891         struct ldlm_res_id *res_id = &info->mti_res_id;
3892         struct lustre_handle *handle;
3893         int rc;
3894
3895         policy->l_inodebits.bits = *ibits;
3896         policy->l_inodebits.try_bits = trybits;
3897         policy->l_inodebits.li_gid = lh->mlh_gid;
3898         fid_build_reg_res_name(fid, res_id);
3899
3900         if (obj && mdt_object_remote(obj)) {
3901                 handle = &lh->mlh_rreg_lh;
3902                 LASSERT(!lustre_handle_is_used(handle));
3903                 LASSERT(lh->mlh_rreg_mode != LCK_MINMODE);
3904                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3905                 rc = mdt_remote_object_lock_try(info, obj, handle,
3906                                                 lh->mlh_rreg_mode, policy,
3907                                                 res_id, cache);
3908         } else {
3909                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3910                 /*
3911                  * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if
3912                  * it is going to be sent to client. If it is -
3913                  * mdt_intent_policy() path will fix it up and turn FL_LOCAL
3914                  * flag off.
3915                  */
3916                 __u64 dlmflags = LDLM_FL_ATOMIC_CB | LDLM_FL_LOCAL_ONLY;
3917                 __u64 *cookie = NULL;
3918
3919                 handle = &lh->mlh_reg_lh;
3920                 LASSERT(!lustre_handle_is_used(handle));
3921                 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3922                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3923
3924                 if (cos_incompat) {
3925                         LASSERT(lh->mlh_reg_mode == LCK_PW ||
3926                                 lh->mlh_reg_mode == LCK_EX);
3927                         dlmflags |= LDLM_FL_COS_INCOMPAT;
3928                 } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3929                         dlmflags |= LDLM_FL_COS_ENABLED;
3930                 }
3931
3932                 /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3933                 if (lh->mlh_type == MDT_REG_LOCK &&
3934                     lh->mlh_reg_mode == LCK_EX && *ibits == MDS_INODELOCK_OPEN)
3935                         dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3936
3937
3938                 if (info->mti_exp)
3939                         cookie = &info->mti_exp->exp_handle.h_cookie;
3940
3941                 rc = mdt_fid_lock(info->mti_env, ns, handle, lh->mlh_reg_mode,
3942                                   policy, res_id, dlmflags, cookie);
3943                 if (rc)
3944                         mdt_object_unlock(info, obj, lh, 1);
3945         }
3946
3947         if (rc == 0) {
3948                 struct ldlm_lock *lock;
3949
3950                 /* Return successfully acquired bits to a caller */
3951                 lock = ldlm_handle2lock(handle);
3952                 LASSERT(lock);
3953                 *ibits = lock->l_policy_data.l_inodebits.bits;
3954                 LDLM_LOCK_PUT(lock);
3955         }
3956
3957         return rc;
3958 }
3959
3960 /*
3961  * MDT object locking functions:
3962  * mdt_object_lock(): lock object, this is used in most places, and normally
3963  *      lock ibits doesn't contain LOOKUP, unless the caller knows it's not
3964  *      remote object.
3965  * mdt_object_check_lock(): lock object with LOOKUP and other ibits, it needs
3966  *      to check whether parent is on remote MDT, if so, take LOOKUP on parent
3967  *      MDT separately, and then lock other ibits on child object.
3968  * mdt_parent_lock(): take parent UPDATE lock with specific mode, if parent is
3969  *      local, take PDO lock by name hash, otherwise take regular lock.
3970  * mdt_object_stripes_lock(): lock object which should be local, and if it's a
3971  *      striped directory, lock its stripes, this is called in operations which
3972  *      modify both object and stripes.
3973  * mdt_object_lock_try(): lock object with trybits, the trybits contains
3974  *      optional inode lock bits that can be granted. This is called by
3975  *      getattr/open to fetch more inode lock bits to client, and is also called
3976  *      by dir migration to lock link parent in non-block mode to avoid
3977  *      deadlock.
3978  */
3979
3980 /**
3981  * lock object
3982  *
3983  * this is used to lock object in most places, and normally lock ibits doesn't
3984  * contain LOOKUP, unless the caller knows it's not remote object.
3985  *
3986  * \param info          struct mdt_thread_info
3987  * \param obj           object
3988  * \param lh            lock handle
3989  * \param ibits         MDS inode lock bits
3990  * \param mode          lock mode
3991  * \param cos_incompat  DNE COS incompatible
3992  *
3993  * \retval              0 on success, -ev on error.
3994  */
3995 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3996                     struct mdt_lock_handle *lh, __u64 ibits,
3997                     enum ldlm_mode mode, bool cos_incompat)
3998 {
3999         int rc;
4000
4001         ENTRY;
4002         mdt_lock_reg_init(lh, mode);
4003         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh,
4004                                       &ibits, 0, false, cos_incompat);
4005         RETURN(rc);
4006 }
4007
4008 /**
4009  * lock object with LOOKUP and other ibits
4010  *
4011  * it will check whether parent and child are on different MDTs, if so, take
4012  * LOOKUP lock on parent MDT, and lock other ibits on child MDT, otherwise lock
4013  * all ibits on child MDT. Note, parent and child shouldn't be both on remote
4014  * MDTs, in which case specific lock function should be used, and it's in
4015  * rename and migrate only.
4016  *
4017  * \param info          struct mdt_thread_info
4018  * \param parent        parent object
4019  * \param child         child object
4020  * \param lh            lock handle
4021  * \param ibits         MDS inode lock bits
4022  * \param mode          lock mode
4023  * \param cos_incompat  DNE COS incompatible
4024  *
4025  * \retval              0 on success, -ev on error.
4026  */
4027 int mdt_object_check_lock(struct mdt_thread_info *info,
4028                           struct mdt_object *parent, struct mdt_object *child,
4029                           struct mdt_lock_handle *lh, __u64 ibits,
4030                           enum ldlm_mode mode, bool cos_incompat)
4031 {
4032         int rc;
4033
4034         ENTRY;
4035         /* if LOOKUP ibit is not set, use mdt_object_lock() */
4036         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
4037         /* if only LOOKUP ibit is needed, use mdt_object_lookup_lock() */
4038         LASSERT(ibits != MDS_INODELOCK_LOOKUP);
4039         LASSERT(parent);
4040         /* @parent and @child shouldn't both be on remote MDTs */
4041         LASSERT(!(mdt_object_remote(parent) && mdt_object_remote(child)));
4042
4043         mdt_lock_reg_init(lh, mode);
4044         if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
4045                 __u64 lookup_ibits = MDS_INODELOCK_LOOKUP;
4046
4047                 rc = mdt_object_lock_internal(info, parent,
4048                                               mdt_object_fid(child), lh,
4049                                               &lookup_ibits, 0, false,
4050                                               cos_incompat);
4051                 if (rc)
4052                         RETURN(rc);
4053
4054                 ibits &= ~MDS_INODELOCK_LOOKUP;
4055         }
4056
4057         rc = mdt_object_lock_internal(info, child, mdt_object_fid(child), lh,
4058                                       &ibits, 0, false, cos_incompat);
4059         if (rc && !(ibits & MDS_INODELOCK_LOOKUP))
4060                 mdt_object_unlock(info, NULL, lh, 1);
4061
4062         RETURN(rc);
4063 }
4064
4065 /**
4066  * take parent UPDATE lock
4067  *
4068  * if parent is local, take PDO lock by name hash, otherwise take regular lock.
4069  *
4070  * \param info  struct mdt_thread_info
4071  * \param obj   parent object
4072  * \param lh    lock handle
4073  * \param lname child name
4074  * \param mode  lock mode
4075  * \param cos_incompat  DNE COS incompatible
4076  *
4077  * \retval      0 on success, -ev on error.
4078  */
4079 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4080                     struct mdt_lock_handle *lh, const struct lu_name *lname,
4081                     enum ldlm_mode mode, bool cos_incompat)
4082 {
4083         int rc;
4084
4085         ENTRY;
4086         LASSERT(obj && lname);
4087         if (mdt_object_remote(obj)) {
4088                 __u64 ibits = MDS_INODELOCK_UPDATE;
4089
4090                 mdt_lock_reg_init(lh, mode);
4091                 rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj),
4092                                               lh, &ibits, 0, false,
4093                                               cos_incompat);
4094         } else {
4095                 rc = mdt_object_pdo_lock(info, obj, lh, lname, mode, true,
4096                                          cos_incompat);
4097         }
4098         RETURN(rc);
4099 }
4100
4101 /**
4102  * lock object with trybits
4103  *
4104  * the trybits contains optional inode lock bits that can be granted. This is
4105  * called by getattr/open to fetch more inode lock bits to client, and is also
4106  * called by dir migration to lock link parent in non-block mode to avoid
4107  * deadlock.
4108  *
4109  * \param info          struct mdt_thread_info
4110  * \param obj           object
4111  * \param lh            lock handle
4112  * \param ibits         MDS inode lock bits
4113  * \param trybits       optional inode lock bits
4114  * \param mode          lock mode
4115  * \param cos_incompat  DNE COS incompatible
4116  *
4117  * \retval              0 on success, -ev on error.
4118  */
4119 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *obj,
4120                         struct mdt_lock_handle *lh, __u64 *ibits,
4121                         __u64 trybits, enum ldlm_mode mode, bool cos_incompat)
4122 {
4123         bool trylock_only = *ibits == 0;
4124         int rc;
4125
4126         ENTRY;
4127         LASSERT(!(*ibits & trybits));
4128         mdt_lock_reg_init(lh, mode);
4129         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh, ibits,
4130                                       trybits, false, cos_incompat);
4131         if (rc && trylock_only) { /* clear error for try ibits lock only */
4132                 LASSERT(*ibits == 0);
4133                 rc = 0;
4134         }
4135         RETURN(rc);
4136 }
4137
4138 /*
4139  * Helper function to take \a obj LOOKUP lock.
4140  *
4141  * Both \a pobj and \a obj may be located on remote MDTs.
4142  */
4143 int mdt_object_lookup_lock(struct mdt_thread_info *info,
4144                            struct mdt_object *pobj, struct mdt_object *obj,
4145                            struct mdt_lock_handle *lh, enum ldlm_mode mode,
4146                            bool cos_incompat)
4147 {
4148         __u64 ibits = MDS_INODELOCK_LOOKUP;
4149         int rc;
4150
4151         ENTRY;
4152         /* if @parent is NULL, it's on local MDT, and @child is remote,
4153          * this is case in getattr/unlink/open by name.
4154          */
4155         LASSERT(ergo(!pobj, mdt_object_remote(obj)));
4156         mdt_lock_reg_init(lh, mode);
4157         rc = mdt_object_lock_internal(info, pobj, mdt_object_fid(obj), lh,
4158                                       &ibits, 0, false, cos_incompat);
4159         RETURN(rc);
4160 }
4161
4162 /**
4163  * Save a lock within request object.
4164  *
4165  * Keep the lock referenced until whether client ACK or transaction
4166  * commit happens or release the lock immediately depending on input
4167  * parameters. If COS is ON, a write lock is converted to COS lock
4168  * before saving.
4169  *
4170  * \param info thead info object
4171  * \param h lock handle
4172  * \param mode lock mode
4173  * \param decref force immediate lock releasing
4174  */
4175 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
4176                           enum ldlm_mode mode, int decref)
4177 {
4178         struct tgt_session_info *tsi = info->mti_env->le_ses ?
4179                                        tgt_ses_info(info->mti_env) : NULL;
4180
4181         ENTRY;
4182
4183         if (lustre_handle_is_used(h)) {
4184                 bool has_trans = tsi && tsi->tsi_has_trans;
4185
4186                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
4187                         mdt_fid_unlock(h, mode);
4188                 } else {
4189                         struct mdt_device *mdt = info->mti_mdt;
4190                         struct ldlm_lock *lock = ldlm_handle2lock(h);
4191                         struct ptlrpc_request *req = mdt_info_req(info);
4192                         bool cos = mdt_cos_is_enabled(mdt);
4193                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
4194
4195                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
4196                                  h->cookie);
4197
4198                         /* there is no request if mdt_object_unlock() is called
4199                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
4200                         if (likely(req != NULL)) {
4201                                 LDLM_DEBUG(lock, "save lock request %p reply "
4202                                         "state %p transno %lld\n", req,
4203                                         req->rq_reply_state, req->rq_transno);
4204                                 if (cos) {
4205                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
4206                                         mode = LCK_COS;
4207                                 }
4208                                 if (req->rq_export->exp_disconnected)
4209                                         mdt_fid_unlock(h, mode);
4210                                 else
4211                                         ptlrpc_save_lock(req, h, mode, cos,
4212                                                          convert_lock);
4213                         } else {
4214                                 mdt_fid_unlock(h, mode);
4215                         }
4216
4217                         if (mdt_is_lock_sync(lock)) {
4218                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4219                                 mdt_device_commit_async(info->mti_env, mdt);
4220                         }
4221                         LDLM_LOCK_PUT(lock);
4222                 }
4223                 h->cookie = 0ull;
4224         }
4225
4226         EXIT;
4227 }
4228
4229 /**
4230  * Save cross-MDT lock in uncommitted_slc_locks
4231  *
4232  * Keep the lock referenced until transaction commit happens or release the lock
4233  * immediately depending on input parameters.
4234  *
4235  * \param info thead info object
4236  * \param h lock handle
4237  * \param mode lock mode
4238  * \param decref force immediate lock releasing
4239  */
4240 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4241                                  struct mdt_object *o, struct lustre_handle *h,
4242                                  enum ldlm_mode mode, int decref)
4243 {
4244         ENTRY;
4245
4246         if (lustre_handle_is_used(h)) {
4247                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4248                 struct ptlrpc_request *req = mdt_info_req(info);
4249
4250                 if (o != NULL &&
4251                     (lock->l_policy_data.l_inodebits.bits &
4252                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4253                         mo_invalidate(info->mti_env, mdt_object_child(o));
4254
4255                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4256                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4257                         ldlm_lock_decref_and_cancel(h, mode);
4258                         LDLM_LOCK_PUT(lock);
4259                 } else {
4260                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4261                                           req->rq_transno);
4262                         ldlm_lock_decref(h, mode);
4263                 }
4264                 h->cookie = 0ull;
4265         }
4266
4267         EXIT;
4268 }
4269
4270 /**
4271  * Unlock mdt object.
4272  *
4273  * Immeditely release the regular lock and the PDO lock or save the
4274  * lock in request and keep them referenced until client ACK or
4275  * transaction commit.
4276  *
4277  * \param info thread info object
4278  * \param o mdt object
4279  * \param lh mdt lock handle referencing regular and PDO locks
4280  * \param decref force immediate lock releasing
4281  *
4282  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4283  */
4284 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4285                        struct mdt_lock_handle *lh, int decref)
4286 {
4287         ENTRY;
4288
4289         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4290         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4291         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4292                              decref);
4293
4294         EXIT;
4295 }
4296
4297 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4298                                         const struct lu_fid *f,
4299                                         struct mdt_lock_handle *lh,
4300                                         __u64 ibits, enum ldlm_mode mode)
4301 {
4302         struct mdt_object *o;
4303
4304         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4305         if (!IS_ERR(o)) {
4306                 int rc;
4307
4308                 rc = mdt_object_lock(info, o, lh, ibits, mode, false);
4309                 if (rc != 0) {
4310                         mdt_object_put(info->mti_env, o);
4311                         o = ERR_PTR(rc);
4312                 }
4313         }
4314         return o;
4315 }
4316
4317 void mdt_object_unlock_put(struct mdt_thread_info * info,
4318                            struct mdt_object *o,
4319                            struct mdt_lock_handle *lh,
4320                            int decref)
4321 {
4322         mdt_object_unlock(info, o, lh, decref);
4323         mdt_object_put(info->mti_env, o);
4324 }
4325
4326 /*
4327  * Generic code handling requests that have struct mdt_body passed in:
4328  *
4329  *  - extract mdt_body from request and save it in @info, if present;
4330  *
4331  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4332  *  @info;
4333  *
4334  *  - if HAS_BODY flag is set for this request type check whether object
4335  *  actually exists on storage (lu_object_exists()).
4336  *
4337  */
4338 static int mdt_body_unpack(struct mdt_thread_info *info,
4339                            enum tgt_handler_flags flags)
4340 {
4341         const struct mdt_body    *body;
4342         struct mdt_object        *obj;
4343         const struct lu_env      *env;
4344         struct req_capsule       *pill;
4345         int                       rc;
4346         ENTRY;
4347
4348         env = info->mti_env;
4349         pill = info->mti_pill;
4350
4351         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4352         if (body == NULL)
4353                 RETURN(-EFAULT);
4354
4355         if (!(body->mbo_valid & OBD_MD_FLID))
4356                 RETURN(0);
4357
4358         if (!fid_is_sane(&body->mbo_fid1)) {
4359                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4360                 RETURN(-EINVAL);
4361         }
4362
4363         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4364         if (!IS_ERR(obj)) {
4365                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4366                         mdt_object_put(env, obj);
4367                         rc = -ENOENT;
4368                 } else {
4369                         info->mti_object = obj;
4370                         rc = 0;
4371                 }
4372         } else
4373                 rc = PTR_ERR(obj);
4374
4375         RETURN(rc);
4376 }
4377
4378 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4379                                    enum tgt_handler_flags flags)
4380 {
4381         struct req_capsule *pill = info->mti_pill;
4382         int rc;
4383
4384         ENTRY;
4385
4386         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4387                 rc = mdt_body_unpack(info, flags);
4388         else
4389                 rc = 0;
4390
4391         if (rc == 0 && (flags & HAS_REPLY)) {
4392                 /* Pack reply. */
4393                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4394                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4395                                              req_capsule_ptlreq(pill) ?
4396                                              DEF_REP_MD_SIZE : MAX_MD_SIZE_OLD);
4397
4398                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4399                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4400                                              RCL_SERVER, 0);
4401
4402                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4403                  * by default. If the target object has more ACL entries, then
4404                  * enlarge the buffer when necessary. */
4405                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4406                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4407                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4408
4409                 mdt_preset_secctx_size(info);
4410                 mdt_preset_encctx_size(info);
4411
4412                 rc = req_capsule_server_pack(pill);
4413                 if (rc)
4414                         CWARN("%s: cannot pack response: rc = %d\n",
4415                                       mdt_obd_name(info->mti_mdt), rc);
4416         }
4417         RETURN(rc);
4418 }
4419
4420 void mdt_thread_info_reset(struct mdt_thread_info *info)
4421 {
4422         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4423         info->mti_body = NULL;
4424         info->mti_dlm_req = NULL;
4425         info->mti_cross_ref = 0;
4426         info->mti_opdata = 0;
4427         info->mti_big_lmm_used = 0;
4428         info->mti_big_acl_used = 0;
4429         info->mti_som_strict = 0;
4430
4431         info->mti_spec.no_create = 0;
4432         info->mti_spec.sp_rm_entry = 0;
4433         info->mti_spec.sp_permitted = 0;
4434
4435         info->mti_spec.u.sp_ea.eadata = NULL;
4436         info->mti_spec.u.sp_ea.eadatalen = 0;
4437
4438         if (info->mti_batch_env && info->mti_object != NULL) {
4439                 mdt_object_put(info->mti_env, info->mti_object);
4440                 info->mti_object = NULL;
4441         }
4442 }
4443
4444 /*
4445  * Initialize fields of struct mdt_thread_info. Other fields are left in
4446  * uninitialized state, because it's too expensive to zero out whole
4447  * mdt_thread_info (> 1K) on each request arrival.
4448  */
4449 void mdt_thread_info_init(struct ptlrpc_request *req,
4450                           struct mdt_thread_info *info)
4451 {
4452         info->mti_pill = &req->rq_pill;
4453
4454         /* mdt device: it can be NULL while CONNECT */
4455         if (req->rq_export) {
4456                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4457                 info->mti_exp = req->rq_export;
4458         } else
4459                 info->mti_mdt = NULL;
4460         info->mti_env = req->rq_svc_thread->t_env;
4461         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4462         info->mti_big_buf = LU_BUF_NULL;
4463         info->mti_batch_env = 0;
4464         info->mti_object = NULL;
4465
4466         mdt_thread_info_reset(info);
4467 }
4468
4469 void mdt_thread_info_fini(struct mdt_thread_info *info)
4470 {
4471         int i;
4472
4473         if (info->mti_object != NULL) {
4474                 mdt_object_put(info->mti_env, info->mti_object);
4475                 info->mti_object = NULL;
4476         }
4477
4478         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4479                 mdt_lock_handle_assert(&info->mti_lh[i]);
4480         info->mti_env = NULL;
4481         info->mti_pill = NULL;
4482         info->mti_exp = NULL;
4483         info->mti_mdt = NULL;
4484
4485         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4486                 lu_buf_free(&info->mti_big_buf);
4487 }
4488
4489 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4490 {
4491         struct mdt_thread_info  *mti;
4492
4493         mti = mdt_th_info(tsi->tsi_env);
4494         LASSERT(mti != NULL);
4495
4496         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4497         if (tsi->tsi_corpus != NULL) {
4498                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4499                 lu_object_get(tsi->tsi_corpus);
4500         }
4501         mti->mti_body = tsi->tsi_mdt_body;
4502         mti->mti_dlm_req = tsi->tsi_dlm_req;
4503
4504         return mti;
4505 }
4506
4507 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4508 {
4509         if (CFS_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4510             cfs_fail_val ==
4511             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4512                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4513
4514         return tgt_connect(tsi);
4515 }
4516
4517 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4518                               struct mdt_thread_info *info,
4519                               struct ldlm_lock **lockp, __u64 flags)
4520 {
4521         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4522                                    lockp, flags);
4523 }
4524 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4525                           struct mdt_thread_info *info,
4526                           struct ldlm_lock **lockp, __u64 flags)
4527 {
4528         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4529                                lockp, flags);
4530 }
4531
4532 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4533                             struct ldlm_lock **lockp,
4534                             struct mdt_lock_handle *lh,
4535                             __u64 flags, int result)
4536 {
4537         struct ptlrpc_request  *req = mdt_info_req(info);
4538         struct ldlm_lock       *lock = *lockp;
4539         struct ldlm_lock       *new_lock;
4540
4541         /* If possible resent found a lock, @lh is set to its handle */
4542         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4543
4544         if (new_lock == NULL) {
4545                 if (flags & LDLM_FL_INTENT_ONLY) {
4546                         result = 0;
4547                 } else if (flags & LDLM_FL_RESENT) {
4548                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4549                          * resend case, however, it could be already destroyed
4550                          * due to client eviction or a raced cancel RPC.
4551                          */
4552                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4553                                           lh->mlh_reg_lh.cookie);
4554                         result = -ESTALE;
4555                 } else {
4556                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4557                                mdt_obd_name(info->mti_mdt),
4558                                lh->mlh_reg_lh.cookie, flags,
4559                                PFID(&info->mti_tmp_fid1),
4560                                PFID(&info->mti_tmp_fid2), result);
4561                         result = -ESTALE;
4562                 }
4563                 lh->mlh_reg_lh.cookie = 0;
4564                 RETURN(result);
4565         }
4566
4567         /*
4568          * If we've already given this lock to a client once, then we should
4569          * have no readers or writers.  Otherwise, we should have one reader
4570          * _or_ writer ref (which will be zeroed below) before returning the
4571          * lock to a client.
4572          */
4573         if (new_lock->l_export == req->rq_export) {
4574                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4575         } else {
4576                 LASSERT(new_lock->l_export == NULL);
4577                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4578         }
4579
4580         *lockp = new_lock;
4581
4582         if (new_lock->l_export == req->rq_export) {
4583                 /*
4584                  * Already gave this to the client, which means that we
4585                  * reconstructed a reply.
4586                  */
4587                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4588                         MSG_RESENT);
4589
4590                 LDLM_LOCK_RELEASE(new_lock);
4591                 lh->mlh_reg_lh.cookie = 0;
4592                 RETURN(ELDLM_LOCK_REPLACED);
4593         }
4594
4595         /*
4596          * Fixup the lock to be given to the client.
4597          */
4598         lock_res_and_lock(new_lock);
4599         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4600          * possible blocking AST. */
4601         while (new_lock->l_readers > 0) {
4602                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4603                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4604                 new_lock->l_readers--;
4605         }
4606         while (new_lock->l_writers > 0) {
4607                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4608                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4609                 new_lock->l_writers--;
4610         }
4611
4612         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4613         new_lock->l_blocking_ast = lock->l_blocking_ast;
4614         new_lock->l_completion_ast = lock->l_completion_ast;
4615         if (ldlm_has_dom(new_lock))
4616                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4617         new_lock->l_remote_handle = lock->l_remote_handle;
4618         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4619
4620         unlock_res_and_lock(new_lock);
4621
4622         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4623                      &new_lock->l_remote_handle,
4624                      &new_lock->l_exp_hash);
4625
4626         LDLM_LOCK_RELEASE(new_lock);
4627         lh->mlh_reg_lh.cookie = 0;
4628
4629         RETURN(ELDLM_LOCK_REPLACED);
4630 }
4631
4632 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4633                              struct ldlm_lock *new_lock,
4634                              struct mdt_lock_handle *lh, __u64 flags)
4635 {
4636         struct ptlrpc_request  *req = mdt_info_req(info);
4637         struct ldlm_request    *dlmreq;
4638
4639         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4640                 return;
4641
4642         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4643
4644         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4645          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4646          * initialized. */
4647         if (flags & LDLM_FL_RESENT) {
4648                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4649                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4650
4651                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4652                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4653                           lh->mlh_reg_lh.cookie);
4654                 return;
4655         }
4656
4657         /*
4658          * If the xid matches, then we know this is a resent request, and allow
4659          * it. (It's probably an OPEN, for which we don't send a lock.
4660          */
4661         if (req_can_reconstruct(req, NULL) != 0)
4662                 return;
4663
4664         /*
4665          * This remote handle isn't enqueued, so we never received or processed
4666          * this request.  Clear MSG_RESENT, because it can be handled like any
4667          * normal request now.
4668          */
4669         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4670
4671         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4672                   dlmreq->lock_handle[0].cookie);
4673 }
4674
4675 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4676                                struct mdt_thread_info *info,
4677                                struct ldlm_lock **lockp,
4678                                __u64 flags)
4679 {
4680         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4681         struct ldlm_reply      *ldlm_rep = NULL;
4682         int rc;
4683         ENTRY;
4684
4685         /*
4686          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4687          * (for the resend case) or a new lock. Below we will use it to
4688          * replace the original lock.
4689          */
4690         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4691         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4692                 rc = mdt_object_lock(info, info->mti_object, lhc,
4693                                      MDS_INODELOCK_XATTR, (*lockp)->l_req_mode,
4694                                      false);
4695                 if (rc)
4696                         return rc;
4697         }
4698
4699         rc = mdt_getxattr(info);
4700
4701         if (mdt_info_req(info)->rq_repmsg != NULL)
4702                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4703
4704         if (ldlm_rep == NULL ||
4705             CFS_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4706                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4707                 if (is_serious(rc))
4708                         RETURN(rc);
4709                 else
4710                         RETURN(err_serious(-EFAULT));
4711         }
4712
4713         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4714
4715         /* This is left for interop instead of adding a new interop flag.
4716          * LU-7433 */
4717 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4718         if (ldlm_rep->lock_policy_res2) {
4719                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4720                 RETURN(ELDLM_LOCK_ABORTED);
4721         }
4722 #endif
4723
4724         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4725         RETURN(rc);
4726 }
4727
4728 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4729                               struct mdt_thread_info *info,
4730                               struct ldlm_lock **lockp,
4731                               __u64 flags)
4732 {
4733         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4734         __u64                   child_bits;
4735         struct ldlm_reply      *ldlm_rep;
4736         struct mdt_body        *reqbody;
4737         struct mdt_body        *repbody;
4738         int                     rc, rc2;
4739         ENTRY;
4740
4741         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4742         LASSERT(reqbody);
4743
4744         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4745         LASSERT(repbody);
4746
4747         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4748         repbody->mbo_eadatasize = 0;
4749         repbody->mbo_aclsize = 0;
4750
4751         switch (it_opc) {
4752         case IT_LOOKUP:
4753                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4754                 break;
4755         case IT_GETATTR:
4756                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4757                              MDS_INODELOCK_PERM;
4758                 break;
4759         default:
4760                 CERROR("%s: unsupported intent %#x\n",
4761                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4762                 GOTO(out_shrink, rc = -EINVAL);
4763         }
4764
4765         rc = mdt_init_ucred(info, reqbody);
4766         if (rc)
4767                 GOTO(out_shrink, rc);
4768
4769         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4770         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4771
4772         /* Get lock from request for possible resent case. */
4773         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4774
4775         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4776         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4777
4778         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4779             ldlm_rep->lock_policy_res2 != -ENOKEY)
4780                 ldlm_rep->lock_policy_res2 = 0;
4781         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4782             ldlm_rep->lock_policy_res2) {
4783                 lhc->mlh_reg_lh.cookie = 0ull;
4784                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4785         }
4786
4787         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4788         EXIT;
4789 out_ucred:
4790         mdt_exit_ucred(info);
4791 out_shrink:
4792         mdt_client_compatibility(info);
4793         rc2 = mdt_fix_reply(info);
4794         if (rc == 0)
4795                 rc = rc2;
4796         return rc;
4797 }
4798
4799 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4800                              struct mdt_thread_info *info,
4801                              struct ldlm_lock **lockp,
4802                              __u64 flags)
4803 {
4804         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4805         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4806         struct layout_intent *intent;
4807         struct ldlm_reply *ldlm_rep;
4808         struct lu_fid *fid = &info->mti_tmp_fid2;
4809         struct mdt_object *obj = NULL;
4810         int layout_size = 0;
4811         struct lu_buf *buf = &layout.mlc_buf;
4812         int rc = 0;
4813
4814         ENTRY;
4815
4816         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4817
4818         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4819         if (intent == NULL)
4820                 RETURN(-EPROTO);
4821
4822         CDEBUG(D_INFO, DFID "got layout change request from client: "
4823                "opc:%u flags:%#x extent "DEXT"\n",
4824                PFID(fid), intent->li_opc, intent->li_flags,
4825                PEXT(&intent->li_extent));
4826
4827         switch (intent->li_opc) {
4828         case LAYOUT_INTENT_TRUNC:
4829         case LAYOUT_INTENT_WRITE:
4830                 layout.mlc_opc = MD_LAYOUT_WRITE;
4831                 layout.mlc_intent = intent;
4832                 break;
4833         case LAYOUT_INTENT_ACCESS:
4834                 break;
4835         case LAYOUT_INTENT_READ:
4836         case LAYOUT_INTENT_GLIMPSE:
4837         case LAYOUT_INTENT_RELEASE:
4838         case LAYOUT_INTENT_RESTORE:
4839                 CERROR("%s: Unsupported layout intent opc %d\n",
4840                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4841                 RETURN(-ENOTSUPP);
4842         default:
4843                 CERROR("%s: Unknown layout intent opc %d\n",
4844                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4845                 RETURN(-EINVAL);
4846         }
4847
4848         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4849         if (IS_ERR(obj))
4850                 RETURN(PTR_ERR(obj));
4851
4852         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4853                 /* if layout is going to be changed don't use the current EA
4854                  * size but the maximum one. That buffer will be shrinked
4855                  * to the actual size in req_capsule_shrink() before reply.
4856                  */
4857                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4858                         layout_size = info->mti_mdt->mdt_max_mdsize;
4859                 } else {
4860                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4861                         if (layout_size < 0)
4862                                 GOTO(out, rc = layout_size);
4863
4864                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4865                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4866                 }
4867                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4868                        mdt_obd_name(info->mti_mdt), layout_size);
4869         }
4870
4871         /*
4872          * set reply buffer size, so that ldlm_handle_enqueue0()->
4873          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4874          */
4875         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4876                              layout_size);
4877         rc = req_capsule_server_pack(info->mti_pill);
4878         if (rc)
4879                 GOTO(out, rc);
4880
4881         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4882         if (!ldlm_rep)
4883                 GOTO(out, rc = -EPROTO);
4884
4885         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4886
4887         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4888         if (layout.mlc_opc == MD_LAYOUT_NOP)
4889                 GOTO(out, rc = 0);
4890
4891         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4892         if (rc < 0)
4893                 GOTO(out, rc);
4894         if (rc == 1) {
4895                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4896                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4897                 GOTO(out, rc);
4898         }
4899
4900         buf->lb_buf = NULL;
4901         buf->lb_len = 0;
4902         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4903                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4904                                                      &RMF_EADATA);
4905                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4906                                                      &RMF_EADATA, RCL_CLIENT);
4907                 /*
4908                  * If it's a replay of layout write intent RPC, the client has
4909                  * saved the extended lovea when it get reply then.
4910                  */
4911                 if (buf->lb_len > 0)
4912                         mdt_fix_lov_magic(info, buf->lb_buf);
4913         }
4914
4915         /* Get lock from request for possible resent case. */
4916         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4917         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4918
4919         /*
4920          * Instantiate some layout components, if @buf contains lovea, then it's
4921          * a replay of the layout intent write RPC.
4922          */
4923         rc = mdt_layout_change(info, obj, lhc, &layout);
4924         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4925
4926         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4927                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4928                 if (rc == ELDLM_LOCK_REPLACED &&
4929                     (*lockp)->l_granted_mode == LCK_EX)
4930                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4931         }
4932
4933         EXIT;
4934 out:
4935         mdt_object_put(info->mti_env, obj);
4936         return rc;
4937 }
4938
4939 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4940                            struct mdt_thread_info *info,
4941                            struct ldlm_lock **lockp,
4942                            __u64 flags)
4943 {
4944         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4945         struct ldlm_reply      *rep = NULL;
4946         long                    opc;
4947         int                     rc;
4948         struct ptlrpc_request  *req = mdt_info_req(info);
4949
4950         static const struct req_format *intent_fmts[REINT_MAX] = {
4951                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4952                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4953         };
4954
4955         ENTRY;
4956
4957         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4958         if (opc < 0)
4959                 RETURN(opc);
4960
4961         /* Get lock from request for possible resent case. */
4962         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4963
4964         rc = mdt_reint_internal(info, lhc, opc);
4965
4966         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4967                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4968
4969         /* Check whether the reply has been packed successfully. */
4970         if (mdt_info_req(info)->rq_repmsg != NULL)
4971                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4972         if (rep == NULL) {
4973                 if (is_serious(rc))
4974                         RETURN(rc);
4975                 else
4976                         RETURN(err_serious(-EFAULT));
4977         }
4978
4979         /* MDC expects this in any case */
4980         if (rc != 0)
4981                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4982
4983         /* the open lock or the lock for cross-ref object should be
4984          * returned to the client */
4985         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4986             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4987                 rep->lock_policy_res2 = 0;
4988                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4989                 RETURN(rc);
4990         }
4991
4992         rep->lock_policy_res2 = clear_serious(rc);
4993
4994         if (rep->lock_policy_res2 == -ENOENT &&
4995             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4996             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4997                 rep->lock_policy_res2 = 0;
4998
4999         lhc->mlh_reg_lh.cookie = 0ull;
5000         if (rc == -ENOTCONN || rc == -ENODEV ||
5001             rc == -EOVERFLOW) { /**< if VBR failure then return error */
5002                 /*
5003                  * If it is the disconnect error (ENODEV & ENOCONN), the error
5004                  * will be returned by rq_status, and client at ptlrpc layer
5005                  * will detect this, then disconnect, reconnect the import
5006                  * immediately, instead of impacting the following the rpc.
5007                  */
5008                 RETURN(rc);
5009         }
5010         /*
5011          * For other cases, the error will be returned by intent, and client
5012          * will retrieve the result from intent.
5013          */
5014         RETURN(ELDLM_LOCK_ABORTED);
5015 }
5016
5017 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
5018                           struct mdt_thread_info *info,
5019                           struct ldlm_lock **lockp,
5020                           u64 flags /* LDLM_FL_* */)
5021 {
5022         struct req_capsule *pill = info->mti_pill;
5023         struct ptlrpc_request *req = mdt_info_req(info);
5024         const struct req_format *it_format;
5025         int (*it_handler)(enum ldlm_intent_flags,
5026                           struct mdt_thread_info *,
5027                           struct ldlm_lock **,
5028                           u64);
5029         enum tgt_handler_flags it_handler_flags = 0;
5030         struct ldlm_reply *rep;
5031         bool check_mdt_object = false;
5032         int rc;
5033         ENTRY;
5034
5035         switch (it_opc) {
5036         case IT_OPEN:
5037         case IT_OPEN|IT_CREAT:
5038                 /*
5039                  * OCREAT is not a IS_MUTABLE request since the file may
5040                  * already exist. We do the extra check of
5041                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
5042                  * really need to create the object.
5043                  */
5044                 it_format = &RQF_LDLM_INTENT;
5045                 it_handler = &mdt_intent_open;
5046                 break;
5047         case IT_GETATTR:
5048                 check_mdt_object = true;
5049                 fallthrough;
5050         case IT_LOOKUP:
5051                 it_format = &RQF_LDLM_INTENT_GETATTR;
5052                 it_handler = &mdt_intent_getattr;
5053                 it_handler_flags = HAS_REPLY;
5054                 break;
5055         case IT_GETXATTR:
5056                 check_mdt_object = true;
5057                 it_format = &RQF_LDLM_INTENT_GETXATTR;
5058                 it_handler = &mdt_intent_getxattr;
5059                 it_handler_flags = HAS_BODY;
5060                 break;
5061         case IT_LAYOUT:
5062                 it_format = &RQF_LDLM_INTENT_LAYOUT;
5063                 it_handler = &mdt_intent_layout;
5064                 break;
5065         case IT_GLIMPSE:
5066                 it_format = &RQF_LDLM_INTENT;
5067                 it_handler = &mdt_intent_glimpse;
5068                 break;
5069         case IT_BRW:
5070                 it_format = &RQF_LDLM_INTENT;
5071                 it_handler = &mdt_intent_brw;
5072                 break;
5073         case IT_QUOTA_DQACQ:
5074         case IT_QUOTA_CONN: {
5075                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
5076
5077                 if (qmt == NULL)
5078                         RETURN(-EOPNOTSUPP);
5079
5080                 if (mdt_rdonly(req->rq_export))
5081                         RETURN(-EROFS);
5082
5083                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
5084                 /* pass the request to quota master */
5085                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
5086                                                  mdt_info_req(info), lockp,
5087                                                  flags);
5088                 RETURN(rc);
5089         }
5090         default:
5091                 CERROR("%s: unknown intent code %#x\n",
5092                        mdt_obd_name(info->mti_mdt), it_opc);
5093                 RETURN(-EPROTO);
5094         }
5095
5096         if (!info->mti_batch_env)
5097                 req_capsule_extend(pill, it_format);
5098
5099         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
5100         if (rc < 0)
5101                 RETURN(rc);
5102
5103         if (unlikely(info->mti_object == NULL && check_mdt_object))
5104                 RETURN(-EPROTO);
5105
5106         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
5107                 RETURN(-EROFS);
5108
5109         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
5110
5111         /* execute policy */
5112         rc = (*it_handler)(it_opc, info, lockp, flags);
5113
5114         /* Check whether the reply has been packed successfully. */
5115         if (info->mti_batch_env || req->rq_repmsg != NULL) {
5116                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5117                 rep->lock_policy_res2 =
5118                         ptlrpc_status_hton(rep->lock_policy_res2);
5119         }
5120
5121         RETURN(rc);
5122 }
5123
5124 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
5125                                     enum ldlm_intent_flags it_opc)
5126 {
5127         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
5128
5129         /* update stats when IT code is known */
5130         if (srv_stats != NULL)
5131                 lprocfs_counter_incr(srv_stats,
5132                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
5133                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
5134 }
5135
5136 static int mdt_intent_policy(const struct lu_env *env,
5137                              struct ldlm_namespace *ns,
5138                              struct ldlm_lock **lockp,
5139                              void *req_cookie,
5140                              enum ldlm_mode mode,
5141                              __u64 flags, void *data)
5142 {
5143         struct tgt_session_info *tsi;
5144         struct mdt_thread_info  *info;
5145         struct ptlrpc_request   *req  =  req_cookie;
5146         struct ldlm_intent      *it;
5147         struct req_capsule      *pill;
5148         const struct ldlm_lock_desc *ldesc;
5149         int rc;
5150
5151         ENTRY;
5152
5153         LASSERT(req != NULL);
5154
5155         tsi = tgt_ses_info(env);
5156
5157         info = mdt_th_info(env);
5158         LASSERT(info != NULL);
5159
5160         /* Check whether it is a sub request processing in a batch request */
5161         if (info->mti_batch_env) {
5162                 pill = info->mti_pill;
5163                 LASSERT(pill == &info->mti_sub_pill);
5164         } else {
5165                 info = tsi2mdt_info(tsi);
5166                 pill = info->mti_pill;
5167         }
5168
5169         LASSERT(pill->rc_req == req);
5170         ldesc = &info->mti_dlm_req->lock_desc;
5171
5172         if (info->mti_batch_env ||
5173             req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
5174                 /*
5175                  * For batch processing environment, the request format has
5176                  * already been set.
5177                  */
5178                 if (!info->mti_batch_env)
5179                         req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
5180
5181                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
5182                 if (it != NULL) {
5183                         mdt_ptlrpc_stats_update(req, it->opc);
5184                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
5185                         if (rc == 0)
5186                                 rc = ELDLM_OK;
5187
5188                         /* Lock without inodebits makes no sense and will oops
5189                          * later in ldlm. Let's check it now to see if we have
5190                          * ibits corrupted somewhere in mdt_intent_opc().
5191                          * The case for client miss to set ibits has been
5192                          * processed by others. */
5193                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
5194                                 ldesc->l_policy_data.l_inodebits.bits != 0));
5195                 } else {
5196                         rc = err_serious(-EFAULT);
5197                 }
5198         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
5199                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
5200                 struct ldlm_reply *rep;
5201
5202                 /* No intent was provided but INTENT flag is set along with
5203                  * DOM bit, this is considered as GLIMPSE request.
5204                  * This logic is common for MDT and OST glimpse
5205                  */
5206                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
5207                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
5208                 /* Check whether the reply has been packed successfully. */
5209                 if (req->rq_repmsg != NULL) {
5210                         rep = req_capsule_server_get(info->mti_pill,
5211                                                      &RMF_DLM_REP);
5212                         rep->lock_policy_res2 =
5213                                 ptlrpc_status_hton(rep->lock_policy_res2);
5214                 }
5215         } else {
5216                 /* No intent was provided */
5217                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5218                 rc = req_capsule_server_pack(pill);
5219                 if (rc)
5220                         rc = err_serious(rc);
5221         }
5222
5223         if (!info->mti_batch_env)
5224                 mdt_thread_info_fini(info);
5225         RETURN(rc);
5226 }
5227
5228 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5229 {
5230         struct seq_server_site  *ss = mdt_seq_site(mdt);
5231
5232         if (ss->ss_node_id == 0)
5233                 return;
5234
5235         if (ss->ss_client_seq != NULL) {
5236                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5237                 ss->ss_client_seq->lcs_exp = NULL;
5238         }
5239
5240         if (ss->ss_server_fld != NULL) {
5241                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5242                 ss->ss_server_fld->lsf_control_exp = NULL;
5243         }
5244 }
5245
5246 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5247 {
5248         struct seq_server_site *ss = mdt_seq_site(mdt);
5249
5250         if (ss == NULL)
5251                 return;
5252
5253         if (ss->ss_server_seq != NULL)
5254                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5255 }
5256
5257 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5258 {
5259         mdt_seq_fini_cli(mdt);
5260         mdt_deregister_seq_exp(mdt);
5261
5262         return seq_site_fini(env, mdt_seq_site(mdt));
5263 }
5264
5265 /**
5266  * It will retrieve its FLDB entries from MDT0, and it only happens
5267  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5268  * and it needs to refresh FLDB from the MDT0.
5269  **/
5270 static int mdt_register_lwp_callback(void *data)
5271 {
5272         struct lu_env           env;
5273         struct mdt_device       *mdt = data;
5274         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5275         int                     rc;
5276         ENTRY;
5277
5278         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5279
5280         rc = lu_env_init(&env, LCT_MD_THREAD);
5281         if (rc < 0) {
5282                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5283                 RETURN(rc);
5284         }
5285
5286         /* Allocate new sequence now to avoid creating local transaction
5287          * in the normal transaction process */
5288         rc = seq_server_check_and_alloc_super(&env,
5289                                               mdt_seq_site(mdt)->ss_server_seq);
5290         if (rc < 0)
5291                 GOTO(out, rc);
5292
5293         if (fld->lsf_new) {
5294                 rc = fld_update_from_controller(&env, fld);
5295                 if (rc != 0) {
5296                         CERROR("%s: cannot update controller: rc = %d\n",
5297                                mdt_obd_name(mdt), rc);
5298                         GOTO(out, rc);
5299                 }
5300         }
5301 out:
5302         lu_env_fini(&env);
5303         RETURN(rc);
5304 }
5305
5306 static int mdt_register_seq_exp(struct mdt_device *mdt)
5307 {
5308         struct seq_server_site  *ss = mdt_seq_site(mdt);
5309         char                    *lwp_name = NULL;
5310         int                     rc;
5311
5312         if (ss->ss_node_id == 0)
5313                 return 0;
5314
5315         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5316         if (lwp_name == NULL)
5317                 GOTO(out_free, rc = -ENOMEM);
5318
5319         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5320         if (rc != 0)
5321                 GOTO(out_free, rc);
5322
5323         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5324                                       NULL, NULL);
5325         if (rc != 0)
5326                 GOTO(out_free, rc);
5327
5328         rc = lustre_register_lwp_item(lwp_name,
5329                                       &ss->ss_server_fld->lsf_control_exp,
5330                                       mdt_register_lwp_callback, mdt);
5331         if (rc != 0) {
5332                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5333                 ss->ss_client_seq->lcs_exp = NULL;
5334                 GOTO(out_free, rc);
5335         }
5336 out_free:
5337         if (lwp_name != NULL)
5338                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5339
5340         return rc;
5341 }
5342
5343 /*
5344  * Init client sequence manager which is used by local MDS to talk to sequence
5345  * controller on remote node.
5346  */
5347 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5348 {
5349         struct seq_server_site *ss = mdt_seq_site(mdt);
5350         char *prefix;
5351         ENTRY;
5352
5353         /* check if this is adding the first MDC and controller is not yet
5354          * initialized. */
5355         OBD_ALLOC_PTR(ss->ss_client_seq);
5356         if (ss->ss_client_seq == NULL)
5357                 RETURN(-ENOMEM);
5358
5359         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5360         if (prefix == NULL) {
5361                 OBD_FREE_PTR(ss->ss_client_seq);
5362                 ss->ss_client_seq = NULL;
5363                 RETURN(-ENOMEM);
5364         }
5365
5366         /* Note: seq_client_fini will be called in seq_site_fini */
5367         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5368         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5369                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5370                                                             NULL);
5371         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5372
5373         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5374 }
5375
5376 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5377 {
5378         struct seq_server_site  *ss;
5379         int                     rc;
5380         ENTRY;
5381
5382         ss = mdt_seq_site(mdt);
5383         /* init sequence controller server(MDT0) */
5384         if (ss->ss_node_id == 0) {
5385                 OBD_ALLOC_PTR(ss->ss_control_seq);
5386                 if (ss->ss_control_seq == NULL)
5387                         RETURN(-ENOMEM);
5388
5389                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5390                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5391                                      ss);
5392                 if (rc)
5393                         GOTO(out_seq_fini, rc);
5394         }
5395
5396         /* Init normal sequence server */
5397         OBD_ALLOC_PTR(ss->ss_server_seq);
5398         if (ss->ss_server_seq == NULL)
5399                 GOTO(out_seq_fini, rc = -ENOMEM);
5400
5401         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5402                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5403         if (rc)
5404                 GOTO(out_seq_fini, rc);
5405
5406         /* init seq client for seq server to talk to seq controller(MDT0) */
5407         rc = mdt_seq_init_cli(env, mdt);
5408         if (rc != 0)
5409                 GOTO(out_seq_fini, rc);
5410
5411         if (ss->ss_node_id != 0)
5412                 /* register controller export through lwp */
5413                 rc = mdt_register_seq_exp(mdt);
5414
5415         EXIT;
5416 out_seq_fini:
5417         if (rc)
5418                 mdt_seq_fini(env, mdt);
5419
5420         return rc;
5421 }
5422
5423 /*
5424  * FLD wrappers
5425  */
5426 static int mdt_fld_fini(const struct lu_env *env,
5427                         struct mdt_device *m)
5428 {
5429         struct seq_server_site *ss = mdt_seq_site(m);
5430         ENTRY;
5431
5432         if (ss && ss->ss_server_fld) {
5433                 fld_server_fini(env, ss->ss_server_fld);
5434                 OBD_FREE_PTR(ss->ss_server_fld);
5435                 ss->ss_server_fld = NULL;
5436         }
5437
5438         RETURN(0);
5439 }
5440
5441 static int mdt_fld_init(const struct lu_env *env,
5442                         const char *uuid,
5443                         struct mdt_device *m)
5444 {
5445         struct seq_server_site *ss;
5446         int rc;
5447         ENTRY;
5448
5449         ss = mdt_seq_site(m);
5450
5451         OBD_ALLOC_PTR(ss->ss_server_fld);
5452         if (ss->ss_server_fld == NULL)
5453                 RETURN(rc = -ENOMEM);
5454
5455         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5456                              LU_SEQ_RANGE_MDT);
5457         if (rc) {
5458                 OBD_FREE_PTR(ss->ss_server_fld);
5459                 ss->ss_server_fld = NULL;
5460                 RETURN(rc);
5461         }
5462
5463         RETURN(0);
5464 }
5465
5466 static void mdt_stack_pre_fini(const struct lu_env *env,
5467                            struct mdt_device *m, struct lu_device *top)
5468 {
5469         struct lustre_cfg_bufs  *bufs;
5470         struct lustre_cfg       *lcfg;
5471         struct mdt_thread_info  *info;
5472         ENTRY;
5473
5474         LASSERT(top);
5475
5476         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5477         LASSERT(info != NULL);
5478
5479         bufs = &info->mti_u.bufs;
5480
5481         LASSERT(m->mdt_child_exp);
5482         LASSERT(m->mdt_child_exp->exp_obd);
5483
5484         /* process cleanup, pass mdt obd name to get obd umount flags */
5485         /* XXX: this is needed because all layers are referenced by
5486          * objects (some of them are pinned by osd, for example *
5487          * the proper solution should be a model where object used
5488          * by osd only doesn't have mdt/mdd slices -bzzz */
5489         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5490         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5491         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5492         if (!lcfg)
5493                 RETURN_EXIT;
5494         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5495
5496         top->ld_ops->ldo_process_config(env, top, lcfg);
5497         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5498         EXIT;
5499 }
5500
5501 static void mdt_stack_fini(const struct lu_env *env,
5502                            struct mdt_device *m, struct lu_device *top)
5503 {
5504         struct obd_device       *obd = mdt2obd_dev(m);
5505         struct lustre_cfg_bufs  *bufs;
5506         struct lustre_cfg       *lcfg;
5507         struct mdt_thread_info  *info;
5508         char                     flags[3] = "";
5509         ENTRY;
5510
5511         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5512         LASSERT(info != NULL);
5513
5514         lu_dev_del_linkage(top->ld_site, top);
5515
5516         lu_site_purge(env, top->ld_site, -1);
5517
5518         bufs = &info->mti_u.bufs;
5519         /* process cleanup, pass mdt obd name to get obd umount flags */
5520         /* another purpose is to let all layers to release their objects */
5521         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5522         if (obd->obd_force)
5523                 strcat(flags, "F");
5524         if (obd->obd_fail)
5525                 strcat(flags, "A");
5526         lustre_cfg_bufs_set_string(bufs, 1, flags);
5527         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5528         if (!lcfg)
5529                 RETURN_EXIT;
5530         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5531
5532         LASSERT(top);
5533         top->ld_ops->ldo_process_config(env, top, lcfg);
5534         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5535
5536         lu_site_purge(env, top->ld_site, -1);
5537
5538         m->mdt_child = NULL;
5539         m->mdt_bottom = NULL;
5540
5541         obd_disconnect(m->mdt_child_exp);
5542         m->mdt_child_exp = NULL;
5543
5544         obd_disconnect(m->mdt_bottom_exp);
5545         m->mdt_child_exp = NULL;
5546 }
5547
5548 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5549                                const char *next, struct obd_export **exp)
5550 {
5551         struct obd_connect_data *data = NULL;
5552         struct obd_device       *obd;
5553         int                      rc;
5554         ENTRY;
5555
5556         OBD_ALLOC_PTR(data);
5557         if (data == NULL)
5558                 GOTO(out, rc = -ENOMEM);
5559
5560         obd = class_name2obd(next);
5561         if (obd == NULL) {
5562                 CERROR("%s: can't locate next device: %s\n",
5563                        mdt_obd_name(m), next);
5564                 GOTO(out, rc = -ENOTCONN);
5565         }
5566
5567         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5568         data->ocd_version = LUSTRE_VERSION_CODE;
5569
5570         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5571         if (rc) {
5572                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5573                        mdt_obd_name(m), next, rc);
5574                 GOTO(out, rc);
5575         }
5576
5577 out:
5578         if (data)
5579                 OBD_FREE_PTR(data);
5580         RETURN(rc);
5581 }
5582
5583 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5584                           struct lustre_cfg *cfg)
5585 {
5586         char                   *dev = lustre_cfg_string(cfg, 0);
5587         int                     rc, name_size, uuid_size;
5588         char                   *name, *uuid, *p;
5589         struct lustre_cfg_bufs *bufs;
5590         struct lustre_cfg      *lcfg;
5591         struct obd_device      *obd;
5592         struct lustre_profile  *lprof;
5593         struct lu_site         *site;
5594         ENTRY;
5595
5596         /* in 1.8 we had the only device in the stack - MDS.
5597          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5598          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5599          * to generate names and setup MDT, MDD. MDT will be using
5600          * generated name to connect to MDD. for MDD the next device
5601          * will be LOD with name taken from so called "profile" which
5602          * is generated by mount_option line
5603          *
5604          * 1.8 MGS generates config. commands like this:
5605          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5606          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5607          * 2.0 MGS generates config. commands like this:
5608          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5609          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5610          *                    3:lustre-MDT0000-mdtlov  4:f
5611          *
5612          * we generate MDD name from MDT one, just replacing T with D
5613          *
5614          * after all the preparations, the logical equivalent will be
5615          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5616          *                    3:lustre-MDT0000-mdtlov  4:f
5617          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5618          *                    3:lustre-MDD0000  4:f
5619          *
5620          *  notice we build the stack from down to top: MDD first, then MDT */
5621
5622         name_size = MAX_OBD_NAME;
5623         uuid_size = MAX_OBD_NAME;
5624
5625         OBD_ALLOC(name, name_size);
5626         OBD_ALLOC(uuid, uuid_size);
5627         if (name == NULL || uuid == NULL)
5628                 GOTO(cleanup_mem, rc = -ENOMEM);
5629
5630         OBD_ALLOC_PTR(bufs);
5631         if (!bufs)
5632                 GOTO(cleanup_mem, rc = -ENOMEM);
5633
5634         strcpy(name, dev);
5635         p = strstr(name, "-MDT");
5636         if (p == NULL)
5637                 GOTO(free_bufs, rc = -ENOMEM);
5638         p[3] = 'D';
5639
5640         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5641
5642         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5643         if (lprof == NULL || lprof->lp_dt == NULL) {
5644                 CERROR("can't find the profile: %s\n",
5645                        lustre_cfg_string(cfg, 0));
5646                 GOTO(free_bufs, rc = -EINVAL);
5647         }
5648
5649         lustre_cfg_bufs_reset(bufs, name);
5650         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5651         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5652         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5653
5654         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5655         if (!lcfg)
5656                 GOTO(put_profile, rc = -ENOMEM);
5657         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5658
5659         rc = class_attach(lcfg);
5660         if (rc)
5661                 GOTO(lcfg_cleanup, rc);
5662
5663         obd = class_name2obd(name);
5664         if (!obd) {
5665                 CERROR("Can not find obd %s (%s in config)\n",
5666                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5667                 GOTO(lcfg_cleanup, rc = -EINVAL);
5668         }
5669
5670         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5671
5672         lustre_cfg_bufs_reset(bufs, name);
5673         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5674         lustre_cfg_bufs_set_string(bufs, 2, dev);
5675         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5676
5677         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5678         if (!lcfg)
5679                 GOTO(class_detach, rc = -ENOMEM);
5680         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5681
5682         rc = class_setup(obd, lcfg);
5683         if (rc)
5684                 GOTO(class_detach, rc);
5685
5686         /* connect to MDD we just setup */
5687         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5688         if (rc)
5689                 GOTO(class_detach, rc);
5690
5691         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5692         LASSERT(site);
5693         LASSERT(mdt_lu_site(mdt) == NULL);
5694         mdt->mdt_lu_dev.ld_site = site;
5695         site->ls_top_dev = &mdt->mdt_lu_dev;
5696         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5697
5698         /* now connect to bottom OSD */
5699         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5700         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5701         if (rc)
5702                 GOTO(class_detach, rc);
5703         mdt->mdt_bottom =
5704                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5705
5706         rc = lu_env_refill((struct lu_env *)env);
5707         if (rc != 0)
5708                 CERROR("Failure to refill session: '%d'\n", rc);
5709
5710         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5711
5712         EXIT;
5713 class_detach:
5714         if (rc)
5715                 class_detach(obd, lcfg);
5716 lcfg_cleanup:
5717         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5718 put_profile:
5719         class_put_profile(lprof);
5720 free_bufs:
5721         OBD_FREE_PTR(bufs);
5722 cleanup_mem:
5723         if (name)
5724                 OBD_FREE(name, name_size);
5725         if (uuid)
5726                 OBD_FREE(uuid, uuid_size);
5727         RETURN(rc);
5728 }
5729
5730 /* setup quota master target on MDT0 */
5731 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5732                           struct lustre_cfg *cfg)
5733 {
5734         struct obd_device       *obd;
5735         char                    *dev = lustre_cfg_string(cfg, 0);
5736         char                    *qmtname, *uuid, *p;
5737         struct lustre_cfg_bufs  *bufs;
5738         struct lustre_cfg       *lcfg;
5739         struct lustre_profile   *lprof;
5740         struct obd_connect_data *data;
5741         int                      rc;
5742         ENTRY;
5743
5744         LASSERT(mdt->mdt_qmt_exp == NULL);
5745         LASSERT(mdt->mdt_qmt_dev == NULL);
5746
5747         /* quota master is on MDT0 only for now */
5748         if (mdt->mdt_seq_site.ss_node_id != 0)
5749                 RETURN(0);
5750
5751         /* MGS generates config commands which look as follows:
5752          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5753          *                    3:lustre-MDT0000-mdtlov  4:f
5754          *
5755          * We generate the QMT name from the MDT one, just replacing MD with QM
5756          * after all the preparations, the logical equivalent will be:
5757          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5758          *                    3:lustre-MDT0000-osd  4:f */
5759         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5760         OBD_ALLOC(uuid, UUID_MAX);
5761         OBD_ALLOC_PTR(bufs);
5762         OBD_ALLOC_PTR(data);
5763         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5764                 GOTO(cleanup_mem, rc = -ENOMEM);
5765
5766         strcpy(qmtname, dev);
5767         p = strstr(qmtname, "-MDT");
5768         if (p == NULL)
5769                 GOTO(cleanup_mem, rc = -ENOMEM);
5770         /* replace MD with QM */
5771         p[1] = 'Q';
5772         p[2] = 'M';
5773
5774         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5775
5776         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5777         if (lprof == NULL || lprof->lp_dt == NULL) {
5778                 CERROR("can't find profile for %s\n",
5779                        lustre_cfg_string(cfg, 0));
5780                 GOTO(cleanup_mem, rc = -EINVAL);
5781         }
5782
5783         lustre_cfg_bufs_reset(bufs, qmtname);
5784         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5785         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5786         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5787
5788         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5789         if (!lcfg)
5790                 GOTO(put_profile, rc = -ENOMEM);
5791         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5792
5793         rc = class_attach(lcfg);
5794         if (rc)
5795                 GOTO(lcfg_cleanup, rc);
5796
5797         obd = class_name2obd(qmtname);
5798         if (!obd) {
5799                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5800                        lustre_cfg_string(cfg, 0));
5801                 GOTO(lcfg_cleanup, rc = -EINVAL);
5802         }
5803
5804         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5805
5806         lustre_cfg_bufs_reset(bufs, qmtname);
5807         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5808         lustre_cfg_bufs_set_string(bufs, 2, dev);
5809
5810         /* for quota, the next device should be the OSD device */
5811         lustre_cfg_bufs_set_string(bufs, 3,
5812                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5813
5814         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5815         if (!lcfg)
5816                 GOTO(class_detach, rc = -ENOMEM);
5817         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5818
5819         rc = class_setup(obd, lcfg);
5820         if (rc)
5821                 GOTO(class_detach, rc);
5822
5823         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5824
5825         /* configure local quota objects */
5826         if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5827                 rc = -EBADF;
5828         else
5829                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5830                                                            &mdt->mdt_lu_dev,
5831                                                            mdt->mdt_qmt_dev);
5832         if (rc)
5833                 GOTO(class_cleanup, rc);
5834
5835         /* connect to quota master target */
5836         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5837         data->ocd_version = LUSTRE_VERSION_CODE;
5838         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5839                          data, NULL);
5840         if (rc) {
5841                 CERROR("cannot connect to quota master device %s (%d)\n",
5842                        qmtname, rc);
5843                 GOTO(class_cleanup, rc);
5844         }
5845
5846         EXIT;
5847 class_cleanup:
5848         if (rc) {
5849                 class_manual_cleanup(obd);
5850                 mdt->mdt_qmt_dev = NULL;
5851                 GOTO(lcfg_cleanup, rc);
5852         }
5853 class_detach:
5854         if (rc)
5855                 class_detach(obd, lcfg);
5856 lcfg_cleanup:
5857         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5858 put_profile:
5859         class_put_profile(lprof);
5860 cleanup_mem:
5861         if (bufs)
5862                 OBD_FREE_PTR(bufs);
5863         if (qmtname)
5864                 OBD_FREE(qmtname, MAX_OBD_NAME);
5865         if (uuid)
5866                 OBD_FREE(uuid, UUID_MAX);
5867         if (data)
5868                 OBD_FREE_PTR(data);
5869         return rc;
5870 }
5871
5872 /* Shutdown quota master target associated with mdt */
5873 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5874 {
5875         ENTRY;
5876
5877         if (mdt->mdt_qmt_exp == NULL)
5878                 RETURN_EXIT;
5879         LASSERT(mdt->mdt_qmt_dev != NULL);
5880
5881         /* the qmt automatically shuts down when the mdt disconnects */
5882         obd_disconnect(mdt->mdt_qmt_exp);
5883         mdt->mdt_qmt_exp = NULL;
5884         mdt->mdt_qmt_dev = NULL;
5885         EXIT;
5886 }
5887
5888 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5889  * for now. This will be removed along with converting rest of MDT code
5890  * to use tgt_session_info */
5891 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5892 {
5893         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5894         int                      rc;
5895
5896         if (unlikely(info->mti_object == NULL))
5897                 return -EPROTO;
5898
5899         rc = mdt_getxattr(info);
5900
5901         mdt_thread_info_fini(info);
5902         return rc;
5903 }
5904
5905 static int mdt_llog_open(struct tgt_session_info *tsi)
5906 {
5907         ENTRY;
5908
5909         if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
5910                 RETURN(err_serious(-EACCES));
5911
5912         RETURN(tgt_llog_open(tsi));
5913 }
5914
5915 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5916 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5917 #define OST_BRW_READ    OST_READ
5918 #define OST_BRW_WRITE   OST_WRITE
5919
5920 static struct tgt_handler mdt_tgt_handlers[] = {
5921 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5922                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5923                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5924 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5925                 0,                      MDS_DISCONNECT, tgt_disconnect,
5926                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5927 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5928                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5929                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5930 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5931 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5932 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5933 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5934                                                         mdt_getattr_name),
5935 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5936 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5937 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5938 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5939 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5940 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5941 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5942 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5943                                                         mdt_hsm_progress),
5944 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5945                                                         mdt_hsm_ct_register),
5946 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5947                                                         mdt_hsm_ct_unregister),
5948 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5949                                                         mdt_hsm_state_get),
5950 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5951                                                         mdt_hsm_state_set),
5952 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5953 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5954                                                         mdt_hsm_request),
5955 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5956             MDS_SWAP_LAYOUTS,
5957             mdt_swap_layouts),
5958 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5959 TGT_MDT_HDL(IS_MUTABLE,         MDS_BATCH,      mdt_batch),
5960 };
5961
5962 static struct tgt_handler mdt_io_ops[] = {
5963 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5964                                                         mdt_hp_brw),
5965 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5966                                                         mdt_hp_brw),
5967 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5968                                          OST_PUNCH,     mdt_punch_hdl,
5969                                                         mdt_hp_punch),
5970 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5971 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5972                                                         mdt_fallocate_hdl),
5973 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5974 TGT_RPC_HANDLER(OST_FIRST_OPC,
5975                 0,                      OST_SET_INFO,   mdt_io_set_info,
5976                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
5977 };
5978
5979 static struct tgt_handler mdt_sec_ctx_ops[] = {
5980 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5981 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5982 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5983 };
5984
5985 static struct tgt_handler mdt_quota_ops[] = {
5986 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5987 };
5988
5989 static struct tgt_handler mdt_llog_handlers[] = {
5990         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5991         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5992         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5993         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5994 };
5995
5996 static struct tgt_opc_slice mdt_common_slice[] = {
5997         {
5998                 .tos_opc_start  = MDS_FIRST_OPC,
5999                 .tos_opc_end    = MDS_LAST_OPC,
6000                 .tos_hs         = mdt_tgt_handlers
6001         },
6002         {
6003                 .tos_opc_start  = OBD_FIRST_OPC,
6004                 .tos_opc_end    = OBD_LAST_OPC,
6005                 .tos_hs         = tgt_obd_handlers
6006         },
6007         {
6008                 .tos_opc_start  = LDLM_FIRST_OPC,
6009                 .tos_opc_end    = LDLM_LAST_OPC,
6010                 .tos_hs         = tgt_dlm_handlers
6011         },
6012         {
6013                 .tos_opc_start  = SEC_FIRST_OPC,
6014                 .tos_opc_end    = SEC_LAST_OPC,
6015                 .tos_hs         = mdt_sec_ctx_ops
6016         },
6017         {
6018                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
6019                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
6020                 .tos_hs         = tgt_out_handlers
6021         },
6022         {
6023                 .tos_opc_start  = FLD_FIRST_OPC,
6024                 .tos_opc_end    = FLD_LAST_OPC,
6025                 .tos_hs         = fld_handlers
6026         },
6027         {
6028                 .tos_opc_start  = SEQ_FIRST_OPC,
6029                 .tos_opc_end    = SEQ_LAST_OPC,
6030                 .tos_hs         = seq_handlers
6031         },
6032         {
6033                 .tos_opc_start  = QUOTA_DQACQ,
6034                 .tos_opc_end    = QUOTA_LAST_OPC,
6035                 .tos_hs         = mdt_quota_ops
6036         },
6037         {
6038                 .tos_opc_start  = LLOG_FIRST_OPC,
6039                 .tos_opc_end    = LLOG_LAST_OPC,
6040                 .tos_hs         = mdt_llog_handlers
6041         },
6042         {
6043                 .tos_opc_start  = LFSCK_FIRST_OPC,
6044                 .tos_opc_end    = LFSCK_LAST_OPC,
6045                 .tos_hs         = tgt_lfsck_handlers
6046         },
6047         {
6048                 .tos_opc_start  = OST_FIRST_OPC,
6049                 .tos_opc_end    = OST_LAST_OPC,
6050                 .tos_hs         = mdt_io_ops
6051         },
6052         {
6053                 .tos_hs         = NULL
6054         }
6055 };
6056
6057 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
6058 {
6059         struct md_device *next = m->mdt_child;
6060         struct lu_device *d = &m->mdt_lu_dev;
6061         struct obd_device *obd = mdt2obd_dev(m);
6062         struct lfsck_stop stop;
6063
6064         ENTRY;
6065         stop.ls_status = LS_PAUSED;
6066         stop.ls_flags = 0;
6067         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
6068
6069         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
6070
6071         mdt_restriper_stop(m);
6072         ping_evictor_stop();
6073
6074         /* Remove the HSM /proc entry so the coordinator cannot be
6075          * restarted by a user while it's shutting down.
6076          */
6077         mdt_hsm_cdt_stop(m);
6078
6079         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
6080         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
6081
6082         if (m->mdt_namespace != NULL)
6083                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
6084                                           d->ld_obd->obd_force);
6085
6086         obd_exports_barrier(obd);
6087         obd_zombie_barrier();
6088
6089         mdt_quota_fini(env, m);
6090
6091         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
6092
6093         /* Calling the cleanup functions in the same order as in the mdt_init0
6094          * error path
6095          */
6096         mdt_tunables_fini(m);
6097
6098         target_recovery_fini(obd);
6099         upcall_cache_cleanup(m->mdt_identity_cache);
6100         m->mdt_identity_cache = NULL;
6101
6102         tgt_fini(env, &m->mdt_lut);
6103
6104         mdt_hsm_cdt_fini(m);
6105
6106         if (m->mdt_los != NULL) {
6107                 local_oid_storage_fini(env, m->mdt_los);
6108                 m->mdt_los = NULL;
6109         }
6110
6111         if (m->mdt_namespace != NULL) {
6112                 ldlm_namespace_free_post(m->mdt_namespace);
6113                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
6114         }
6115
6116         if (m->mdt_md_root != NULL) {
6117                 mdt_object_put(env, m->mdt_md_root);
6118                 m->mdt_md_root = NULL;
6119         }
6120
6121         mdt_seq_fini(env, m);
6122
6123         mdt_fld_fini(env, m);
6124
6125         /*
6126          * Finish the stack
6127          */
6128         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6129
6130         LASSERT(atomic_read(&d->ld_ref) == 0);
6131
6132         server_put_mount(mdt_obd_name(m), true);
6133
6134         EXIT;
6135 }
6136
6137 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
6138
6139 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
6140                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
6141 {
6142         const struct dt_device_param *dt_conf;
6143         struct mdt_thread_info *info;
6144         struct obd_device *obd;
6145         const char *dev = lustre_cfg_string(cfg, 0);
6146         const char *num = lustre_cfg_string(cfg, 2);
6147         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
6148         struct lustre_mount_info *lmi = NULL;
6149         struct lustre_sb_info *lsi;
6150         struct lu_site *s;
6151         struct seq_server_site *ss_site;
6152         const char *identity_upcall = "NONE";
6153         struct md_device *next;
6154         struct lu_fid fid;
6155         int rc;
6156         long node_id;
6157         mntopt_t mntopts;
6158         ENTRY;
6159
6160         lu_device_init(&m->mdt_lu_dev, ldt);
6161         /*
6162          * Environment (env) might be missing mdt_thread_key values at that
6163          * point, if device is allocated when mdt_thread_key is in QUIESCENT
6164          * mode.
6165          *
6166          * Usually device allocation path doesn't use module key values, but
6167          * mdt has to do a lot of work here, so allocate key value.
6168          */
6169         rc = lu_env_refill((struct lu_env *)env);
6170         if (rc != 0)
6171                 RETURN(rc);
6172
6173         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
6174         LASSERT(info != NULL);
6175
6176         obd = class_name2obd(dev);
6177         LASSERT(obd != NULL);
6178
6179         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
6180         m->mdt_opts.mo_evict_tgt_nids = 1;
6181         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
6182
6183         lmi = server_get_mount(dev);
6184         if (lmi == NULL) {
6185                 CERROR("Cannot get mount info for %s!\n", dev);
6186                 RETURN(-EFAULT);
6187         } else {
6188                 lsi = s2lsi(lmi->lmi_sb);
6189                 LASSERT(lsi->lsi_lmd);
6190                 /* CMD is supported only in IAM mode */
6191                 LASSERT(num);
6192                 rc = kstrtol(num, 10, &node_id);
6193                 if (rc)
6194                         RETURN(rc);
6195
6196                 obd_obt_init(obd);
6197                 if (test_bit(LMD_FLG_SKIP_LFSCK, lsi->lsi_lmd->lmd_flags))
6198                         m->mdt_skip_lfsck = 1;
6199         }
6200
6201         /* Just try to get a DoM lock by default. Otherwise, having a group
6202          * lock granted, it may get blocked for a long time. */
6203         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
6204         /* DoM files are read at open and data is packed in the reply */
6205         m->mdt_opts.mo_dom_read_open = 1;
6206
6207         m->mdt_squash.rsi_uid = 0;
6208         m->mdt_squash.rsi_gid = 0;
6209         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
6210         spin_lock_init(&m->mdt_squash.rsi_lock);
6211         spin_lock_init(&m->mdt_lock);
6212         m->mdt_enable_chprojid_gid = 0;
6213         m->mdt_enable_dir_migration = 1;
6214         m->mdt_enable_dir_restripe = 0;
6215         m->mdt_enable_dir_auto_split = 0;
6216         m->mdt_enable_parallel_rename_dir = 1;
6217         m->mdt_enable_parallel_rename_file = 1;
6218         m->mdt_enable_remote_dir = 1;
6219         m->mdt_enable_remote_dir_gid = 0;
6220         m->mdt_enable_remote_rename = 1;
6221         m->mdt_enable_striped_dir = 1;
6222         m->mdt_enable_dmv_implicit_inherit = 1;
6223         m->mdt_dir_restripe_nsonly = 1;
6224         m->mdt_max_mod_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
6225
6226         atomic_set(&m->mdt_mds_mds_conns, 0);
6227         atomic_set(&m->mdt_async_commit_count, 0);
6228         atomic_set(&m->mdt_dmv_old_client_count, 0);
6229
6230         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6231         m->mdt_lu_dev.ld_obd = obd;
6232         /* Set this lu_device to obd for error handling purposes. */
6233         obd->obd_lu_dev = &m->mdt_lu_dev;
6234
6235         /* init the stack */
6236         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6237         if (rc) {
6238                 CERROR("%s: Can't init device stack, rc %d\n",
6239                        mdt_obd_name(m), rc);
6240                 GOTO(err_lmi, rc);
6241         }
6242
6243         s = mdt_lu_site(m);
6244         ss_site = mdt_seq_site(m);
6245         s->ld_seq_site = ss_site;
6246         ss_site->ss_lu = s;
6247
6248         /* set server index */
6249         ss_site->ss_node_id = node_id;
6250
6251         /* failover is the default
6252          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6253          * assumed whose ss_node_id == 0 XXX
6254          * */
6255         obd->obd_replayable = 1;
6256         /* No connection accepted until configurations will finish */
6257         obd->obd_no_conn = 1;
6258
6259         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6260                 char *str = lustre_cfg_string(cfg, 4);
6261                 if (strchr(str, 'n')) {
6262                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6263                         obd->obd_replayable = 0;
6264                 }
6265         }
6266
6267         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6268         if (rc)
6269                 GOTO(err_fini_stack, rc);
6270
6271         rc = mdt_seq_init(env, m);
6272         if (rc)
6273                 GOTO(err_fini_fld, rc);
6274
6275         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6276                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6277         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6278                                               LDLM_NAMESPACE_SERVER,
6279                                               LDLM_NAMESPACE_GREEDY,
6280                                               LDLM_NS_TYPE_MDT);
6281         if (IS_ERR(m->mdt_namespace)) {
6282                 rc = PTR_ERR(m->mdt_namespace);
6283                 CERROR("%s: unable to create server namespace: rc = %d\n",
6284                        obd->obd_name, rc);
6285                 m->mdt_namespace = NULL;
6286                 GOTO(err_fini_seq, rc);
6287         }
6288
6289         m->mdt_namespace->ns_lvbp = m;
6290         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6291
6292         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6293         /* set obd_namespace for compatibility with old code */
6294         obd->obd_namespace = m->mdt_namespace;
6295
6296         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6297                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6298                       OBD_FAIL_MDS_ALL_REPLY_NET);
6299         if (rc)
6300                 GOTO(err_free_ns, rc);
6301
6302         /* Amount of available space excluded from granting and reserved
6303          * for metadata. It is a percentage of the total MDT size. */
6304         tgd->tgd_reserved_pcnt = 10;
6305
6306         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6307                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6308         else
6309                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6310
6311         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6312                 GOTO(err_tgt, rc = -ENOENT);
6313
6314         fid.f_seq = FID_SEQ_LOCAL_NAME;
6315         fid.f_oid = 1;
6316         fid.f_ver = 0;
6317         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6318         if (rc != 0)
6319                 GOTO(err_tgt, rc);
6320
6321         rc = mdt_hsm_cdt_init(m);
6322         if (rc != 0) {
6323                 CERROR("%s: error initializing coordinator, rc %d\n",
6324                        mdt_obd_name(m), rc);
6325                 GOTO(err_los_fini, rc);
6326         }
6327
6328         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6329
6330         next = m->mdt_child;
6331         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6332
6333         mntopts = dt_conf->ddp_mntopts;
6334
6335         if (mntopts & MNTOPT_USERXATTR)
6336                 m->mdt_opts.mo_user_xattr = 1;
6337         else
6338                 m->mdt_opts.mo_user_xattr = 0;
6339
6340         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6341
6342         if (mntopts & MNTOPT_ACL)
6343                 m->mdt_opts.mo_acl = 1;
6344         else
6345                 m->mdt_opts.mo_acl = 0;
6346
6347         m->mdt_opts.mo_enable_strict_som = 1;
6348
6349         /* XXX: to support suppgid for ACL, we enable identity_upcall
6350          * by default, otherwise, maybe got unexpected -EACCESS. */
6351         if (m->mdt_opts.mo_acl)
6352                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6353
6354         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6355                                                 identity_upcall,
6356                                                 &mdt_identity_upcall_cache_ops);
6357         if (IS_ERR(m->mdt_identity_cache)) {
6358                 rc = PTR_ERR(m->mdt_identity_cache);
6359                 m->mdt_identity_cache = NULL;
6360                 GOTO(err_free_hsm, rc);
6361         }
6362
6363         rc = mdt_tunables_init(m, dev);
6364         if (rc) {
6365                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6366                 GOTO(err_recovery, rc);
6367         }
6368
6369         rc = mdt_quota_init(env, m, cfg);
6370         if (rc)
6371                 GOTO(err_procfs, rc);
6372
6373         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6374         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6375                            "mdt_ldlm_client", m->mdt_ldlm_client);
6376
6377         ping_evictor_start();
6378
6379         /* recovery will be started upon mdt_prepare()
6380          * when the whole stack is complete and ready
6381          * to serve the requests */
6382
6383         /* Reduce the initial timeout on an MDS because it doesn't need such
6384          * a long timeout as an OST does. Adaptive timeouts will adjust this
6385          * value appropriately. */
6386         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6387                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6388
6389         if (test_bit(LMD_FLG_LOCAL_RECOV, lsi->lsi_lmd->lmd_flags))
6390                 m->mdt_lut.lut_local_recovery = 1;
6391
6392         rc = mdt_restriper_start(m);
6393         if (rc)
6394                 GOTO(err_ping_evictor, rc);
6395
6396         RETURN(0);
6397
6398 err_ping_evictor:
6399         ping_evictor_stop();
6400 err_procfs:
6401         mdt_tunables_fini(m);
6402 err_recovery:
6403         upcall_cache_cleanup(m->mdt_identity_cache);
6404         m->mdt_identity_cache = NULL;
6405 err_free_hsm:
6406         mdt_hsm_cdt_fini(m);
6407 err_los_fini:
6408         local_oid_storage_fini(env, m->mdt_los);
6409         m->mdt_los = NULL;
6410 err_tgt:
6411         /* keep recoverable clients */
6412         obd->obd_fail = 1;
6413         target_recovery_fini(obd);
6414         obd_exports_barrier(obd);
6415         obd_zombie_barrier();
6416         tgt_fini(env, &m->mdt_lut);
6417 err_free_ns:
6418         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6419         obd->obd_namespace = m->mdt_namespace = NULL;
6420 err_fini_seq:
6421         mdt_seq_fini(env, m);
6422 err_fini_fld:
6423         mdt_fld_fini(env, m);
6424 err_fini_stack:
6425         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6426 err_lmi:
6427         if (lmi)
6428                 server_put_mount(dev, true);
6429         return(rc);
6430 }
6431
6432 /* For interoperability, the left element is old parameter, the right one
6433  * is the new version of the parameter, if some parameter is deprecated,
6434  * the new version should be set as NULL. */
6435 static struct cfg_interop_param mdt_interop_param[] = {
6436         { "mdt.group_upcall",   NULL },
6437         { "mdt.quota_type",     NULL },
6438         { "mdd.quota_type",     NULL },
6439         { "mdt.som",            NULL },
6440         { "mdt.rootsquash",     "mdt.root_squash" },
6441         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6442         { NULL }
6443 };
6444
6445 /* used by MGS to process specific configurations */
6446 static int mdt_process_config(const struct lu_env *env,
6447                               struct lu_device *d, struct lustre_cfg *cfg)
6448 {
6449         struct mdt_device *m = mdt_dev(d);
6450         struct md_device *md_next = m->mdt_child;
6451         struct lu_device *next = md2lu_dev(md_next);
6452         int rc;
6453         ENTRY;
6454
6455         switch (cfg->lcfg_command) {
6456         case LCFG_PARAM: {
6457                 struct obd_device *obd = d->ld_obd;
6458                 /* For interoperability */
6459                 struct cfg_interop_param *ptr = NULL;
6460                 struct lustre_cfg *old_cfg = NULL;
6461                 char *param = NULL;
6462                 ssize_t count;
6463
6464                 param = lustre_cfg_string(cfg, 1);
6465                 if (param == NULL) {
6466                         CERROR("param is empty\n");
6467                         rc = -EINVAL;
6468                         break;
6469                 }
6470
6471                 ptr = class_find_old_param(param, mdt_interop_param);
6472                 if (ptr != NULL) {
6473                         if (ptr->new_param == NULL) {
6474                                 rc = 0;
6475                                 CWARN("For interoperability, skip this %s."
6476                                       " It is obsolete.\n", ptr->old_param);
6477                                 break;
6478                         }
6479
6480                         CWARN("Found old param %s, changed it to %s.\n",
6481                               ptr->old_param, ptr->new_param);
6482
6483                         old_cfg = cfg;
6484                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6485                         if (IS_ERR(cfg)) {
6486                                 rc = PTR_ERR(cfg);
6487                                 break;
6488                         }
6489                 }
6490
6491                 count = class_modify_config(cfg, PARAM_MDT,
6492                                             &obd->obd_kset.kobj);
6493                 if (count < 0) {
6494                         struct coordinator *cdt = &m->mdt_coordinator;
6495
6496                         /* is it an HSM var ? */
6497                         count = class_modify_config(cfg, PARAM_HSM,
6498                                                     &cdt->cdt_hsm_kobj);
6499                         if (count < 0)
6500                                 /* we don't understand; pass it on */
6501                                 rc = next->ld_ops->ldo_process_config(env, next,
6502                                                                       cfg);
6503                         else
6504                                 rc = count > 0 ? 0 : count;
6505                 } else {
6506                         rc = count > 0 ? 0 : count;
6507                 }
6508
6509                 if (old_cfg)
6510                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6511                                                      cfg->lcfg_buflens));
6512                 break;
6513         }
6514         default:
6515                 /* others are passed further */
6516                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6517                 break;
6518         }
6519         RETURN(rc);
6520 }
6521
6522 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6523                                           const struct lu_object_header *hdr,
6524                                           struct lu_device *d)
6525 {
6526         struct mdt_object *mo;
6527
6528         ENTRY;
6529
6530         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6531         if (mo != NULL) {
6532                 struct lu_object *o;
6533                 struct lu_object_header *h;
6534
6535                 o = &mo->mot_obj;
6536                 h = &mo->mot_header;
6537                 lu_object_header_init(h);
6538                 lu_object_init(o, h, d);
6539                 lu_object_add_top(h, o);
6540                 o->lo_ops = &mdt_obj_ops;
6541                 spin_lock_init(&mo->mot_write_lock);
6542                 mutex_init(&mo->mot_som_mutex);
6543                 mutex_init(&mo->mot_lov_mutex);
6544                 init_rwsem(&mo->mot_dom_sem);
6545                 init_rwsem(&mo->mot_open_sem);
6546                 atomic_set(&mo->mot_open_count, 0);
6547                 mo->mot_restripe_offset = 0;
6548                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6549                 mo->mot_lsom_size = 0;
6550                 mo->mot_lsom_blocks = 0;
6551                 mo->mot_lsom_inited = false;
6552                 RETURN(o);
6553         }
6554         RETURN(NULL);
6555 }
6556
6557 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6558                            const struct lu_object_conf *unused)
6559 {
6560         struct mdt_device *d = mdt_dev(o->lo_dev);
6561         struct lu_device  *under;
6562         struct lu_object  *below;
6563         int                rc = 0;
6564         ENTRY;
6565
6566         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6567                PFID(lu_object_fid(o)));
6568
6569         under = &d->mdt_child->md_lu_dev;
6570         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6571         if (below != NULL) {
6572                 lu_object_add(o, below);
6573         } else
6574                 rc = -ENOMEM;
6575
6576         RETURN(rc);
6577 }
6578
6579 static void mdt_object_free_rcu(struct rcu_head *head)
6580 {
6581         struct mdt_object *mo = container_of(head, struct mdt_object,
6582                                              mot_header.loh_rcu);
6583
6584         kmem_cache_free(mdt_object_kmem, mo);
6585 }
6586
6587 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6588 {
6589         struct mdt_object *mo = mdt_obj(o);
6590         struct lu_object_header *h;
6591         ENTRY;
6592
6593         h = o->lo_header;
6594         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6595                PFID(lu_object_fid(o)));
6596
6597         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6598         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6599
6600         lu_object_fini(o);
6601         lu_object_header_fini(h);
6602         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6603         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6604
6605         EXIT;
6606 }
6607
6608 static int mdt_object_print(const struct lu_env *env, void *cookie,
6609                             lu_printer_t p, const struct lu_object *o)
6610 {
6611         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6612
6613         return (*p)(env, cookie,
6614                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6615                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6616                     mdto->mot_cache_attr ? "cache_attr" : "",
6617                     mdto->mot_write_count);
6618 }
6619
6620 static int mdt_prepare(const struct lu_env *env,
6621                 struct lu_device *pdev,
6622                 struct lu_device *cdev)
6623 {
6624         struct mdt_device *mdt = mdt_dev(cdev);
6625         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6626         struct obd_device *obd = cdev->ld_obd;
6627         int rc;
6628
6629         ENTRY;
6630
6631         LASSERT(obd);
6632
6633         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6634         if (rc)
6635                 RETURN(rc);
6636
6637         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6638         if (rc)
6639                 RETURN(rc);
6640
6641         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6642         if (rc)
6643                 RETURN(rc);
6644
6645         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6646         /* The LFSCK instance is registered just now, so it must be there when
6647          * register the namespace to such instance. */
6648         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6649
6650         if (mdt->mdt_seq_site.ss_node_id == 0) {
6651                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6652                                                          &mdt->mdt_md_root_fid);
6653                 if (rc)
6654                         RETURN(rc);
6655         }
6656
6657         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6658
6659         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6660         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6661         LASSERT(obd->obd_no_conn);
6662         spin_lock(&obd->obd_dev_lock);
6663         obd->obd_no_conn = 0;
6664         spin_unlock(&obd->obd_dev_lock);
6665
6666         if (obd->obd_recovering == 0)
6667                 mdt_postrecov(env, mdt);
6668
6669         RETURN(rc);
6670 }
6671
6672 const struct lu_device_operations mdt_lu_ops = {
6673         .ldo_object_alloc   = mdt_object_alloc,
6674         .ldo_process_config = mdt_process_config,
6675         .ldo_prepare        = mdt_prepare,
6676 };
6677
6678 static const struct lu_object_operations mdt_obj_ops = {
6679         .loo_object_init    = mdt_object_init,
6680         .loo_object_free    = mdt_object_free,
6681         .loo_object_print   = mdt_object_print
6682 };
6683
6684 static int mdt_obd_set_info_async(const struct lu_env *env,
6685                                   struct obd_export *exp,
6686                                   __u32 keylen, void *key,
6687                                   __u32 vallen, void *val,
6688                                   struct ptlrpc_request_set *set)
6689 {
6690         int rc;
6691
6692         ENTRY;
6693
6694         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6695                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6696                 RETURN(rc);
6697         }
6698
6699         RETURN(0);
6700 }
6701
6702 static inline void mdt_enable_slc(struct mdt_device *mdt)
6703 {
6704         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6705                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6706 }
6707
6708 static inline void mdt_disable_slc(struct mdt_device *mdt)
6709 {
6710         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6711                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6712 }
6713
6714 /**
6715  * Match client and server connection feature flags.
6716  *
6717  * Compute the compatibility flags for a connection request based on
6718  * features mutually supported by client and server.
6719  *
6720  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6721  * must not be updated here, otherwise a partially initialized value may
6722  * be exposed. After the connection request is successfully processed,
6723  * the top-level MDT connect request handler atomically updates the export
6724  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6725  * reply. \see mdt_connect().
6726  *
6727  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6728  * full struct obd_connect_data. So care must be taken when accessing fields
6729  * that are not present in struct obd_connect_data_v1. See LU-16.
6730  *
6731  * \param exp   the obd_export associated with this client/target pair
6732  * \param mdt   the target device for the connection
6733  * \param data  stores data for this connect request
6734  *
6735  * \retval 0       success
6736  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6737  * \retval -EBADE  client and server feature requirements are incompatible
6738  */
6739 static int mdt_connect_internal(const struct lu_env *env,
6740                                 struct obd_export *exp,
6741                                 struct mdt_device *mdt,
6742                                 struct obd_connect_data *data, bool reconnect)
6743 {
6744         const char *obd_name = mdt_obd_name(mdt);
6745         LASSERT(data != NULL);
6746
6747         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6748
6749         if (mdt->mdt_bottom->dd_rdonly &&
6750             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6751             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6752                 RETURN(-EACCES);
6753
6754         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6755                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6756
6757         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6758
6759         if (!mdt->mdt_opts.mo_acl)
6760                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6761
6762         if (!mdt->mdt_opts.mo_user_xattr)
6763                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6764
6765         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6766                 data->ocd_brw_size = min(data->ocd_brw_size,
6767                                          mdt->mdt_brw_size);
6768                 if (data->ocd_brw_size == 0) {
6769                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6770                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6771                                "ocd_brw_size unexpectedly zero, network data "
6772                                "corruption? Refusing to connect this client\n",
6773                                obd_name, exp->exp_client_uuid.uuid,
6774                                exp, data->ocd_connect_flags, data->ocd_version,
6775                                data->ocd_grant, data->ocd_index);
6776                         return -EPROTO;
6777                 }
6778         }
6779
6780         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6781                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6782
6783                 /* client is reporting its page size, for future use */
6784                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6785                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6786                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6787                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6788                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6789                 /* ocd_grant_tax_kb is in 1K byte blocks */
6790                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6791                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6792         }
6793
6794         /* Save connect_data we have so far because tgt_grant_connect()
6795          * uses it to calculate grant, and we want to save the client
6796          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6797         exp->exp_connect_data = *data;
6798         if (OCD_HAS_FLAG(data, GRANT))
6799                 tgt_grant_connect(env, exp, data, !reconnect);
6800
6801         if (OCD_HAS_FLAG(data, MAXBYTES))
6802                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6803
6804         /* NB: Disregard the rule against updating
6805          * exp_connect_data.ocd_connect_flags in this case, since
6806          * tgt_client_new() needs to know if this is a lightweight
6807          * connection, and it is safe to expose this flag before
6808          * connection processing completes. */
6809         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6810                 spin_lock(&exp->exp_lock);
6811                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6812                 spin_unlock(&exp->exp_lock);
6813         }
6814
6815         data->ocd_version = LUSTRE_VERSION_CODE;
6816
6817         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6818                 CWARN("%s: MDS requires FID support, but client not\n",
6819                       obd_name);
6820                 return -EBADE;
6821         }
6822
6823         if (OCD_HAS_FLAG(data, PINGLESS)) {
6824                 if (ptlrpc_pinger_suppress_pings()) {
6825                         spin_lock(&exp->exp_obd->obd_dev_lock);
6826                         list_del_init(&exp->exp_obd_chain_timed);
6827                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6828                 } else {
6829                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6830                 }
6831         }
6832
6833         data->ocd_max_easize = mdt->mdt_max_ea_size;
6834
6835         /* NB: Disregard the rule against updating
6836          * exp_connect_data.ocd_connect_flags in this case, since
6837          * tgt_client_new() needs to know if this is client supports
6838          * multiple modify RPCs, and it is safe to expose this flag before
6839          * connection processing completes. */
6840         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6841                 if (mdt_max_mod_rpcs_changed(mdt))
6842                         /* The new mdt.*.max_mod_rpcs_in_flight parameter
6843                          * has not changed since initialization, but the
6844                          * deprecated module parameter was changed,
6845                          * so use that instead.
6846                          */
6847                         data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6848                 else
6849                         data->ocd_maxmodrpcs = mdt->mdt_max_mod_rpcs_in_flight;
6850                 spin_lock(&exp->exp_lock);
6851                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6852                 spin_unlock(&exp->exp_lock);
6853         }
6854
6855         if (OCD_HAS_FLAG(data, CKSUM)) {
6856                 __u32 cksum_types = data->ocd_cksum_types;
6857
6858                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6859
6860                 if (unlikely(data->ocd_cksum_types == 0)) {
6861                         CERROR("%s: Connect with checksum support but no "
6862                                "ocd_cksum_types is set\n",
6863                                exp->exp_obd->obd_name);
6864                         RETURN(-EPROTO);
6865                 }
6866
6867                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6868                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6869                        cksum_types, data->ocd_cksum_types);
6870         } else {
6871                 /* This client does not support OBD_CONNECT_CKSUM
6872                  * fall back to CRC32 */
6873                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6874                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6875                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6876         }
6877
6878         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6879             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6880                 atomic_inc(&mdt->mdt_mds_mds_conns);
6881                 mdt_enable_slc(mdt);
6882         }
6883
6884         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6885                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6886
6887         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
6888             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT)) {
6889                 atomic_inc(&mdt->mdt_dmv_old_client_count);
6890                 mdt->mdt_enable_dmv_implicit_inherit = 0;
6891         }
6892
6893         return 0;
6894 }
6895
6896 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6897                                    struct mdt_thread_info *info,
6898                                    struct mdt_file_data *mfd)
6899 {
6900         struct lu_context ses;
6901         int rc;
6902         ENTRY;
6903
6904         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6905         if (rc)
6906                 RETURN(rc);
6907
6908         env->le_ses = &ses;
6909         lu_context_enter(&ses);
6910
6911         mdt_ucred(info)->uc_valid = UCRED_OLD;
6912         /* do not let rbac interfere with dirty flag internal system event */
6913         mdt_ucred(info)->uc_rbac_file_perms = 1;
6914         mdt_ucred(info)->uc_rbac_dne_ops = 1;
6915         mdt_ucred(info)->uc_rbac_quota_ops = 1;
6916         mdt_ucred(info)->uc_rbac_byfid_ops = 1;
6917         mdt_ucred(info)->uc_rbac_chlg_ops = 1;
6918         mdt_ucred(info)->uc_rbac_fscrypt_admin = 1;
6919         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6920
6921         lu_context_exit(&ses);
6922         lu_context_fini(&ses);
6923         env->le_ses = NULL;
6924
6925         RETURN(rc);
6926 }
6927
6928 static int mdt_export_cleanup(struct obd_export *exp)
6929 {
6930         LIST_HEAD(closing_list);
6931         struct mdt_export_data  *med = &exp->exp_mdt_data;
6932         struct obd_device       *obd = exp->exp_obd;
6933         struct mdt_device       *mdt;
6934         struct mdt_thread_info  *info;
6935         struct lu_env            env;
6936         struct mdt_file_data    *mfd, *n;
6937         int rc = 0;
6938         ENTRY;
6939
6940         spin_lock(&med->med_open_lock);
6941         while (!list_empty(&med->med_open_head)) {
6942                 struct list_head *tmp = med->med_open_head.next;
6943                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6944
6945                 /* Remove mfd handle so it can't be found again.
6946                  * We are consuming the mfd_list reference here. */
6947                 class_handle_unhash(&mfd->mfd_open_handle);
6948                 list_move_tail(&mfd->mfd_list, &closing_list);
6949         }
6950         spin_unlock(&med->med_open_lock);
6951         mdt = mdt_dev(obd->obd_lu_dev);
6952         LASSERT(mdt != NULL);
6953
6954         rc = lu_env_init(&env, LCT_MD_THREAD);
6955         if (rc)
6956                 RETURN(rc);
6957
6958         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6959         LASSERT(info != NULL);
6960         memset(info, 0, sizeof *info);
6961         info->mti_env = &env;
6962         info->mti_mdt = mdt;
6963         info->mti_exp = exp;
6964
6965         if (!list_empty(&closing_list)) {
6966                 struct md_attr *ma = &info->mti_attr;
6967
6968                 /* Close any open files (which may also cause orphan
6969                  * unlinking). */
6970                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6971                         list_del_init(&mfd->mfd_list);
6972                         ma->ma_need = ma->ma_valid = 0;
6973
6974                         /* This file is being closed due to an eviction, it
6975                          * could have been modified and now dirty regarding to
6976                          * HSM archive, check this!
6977                          * The logic here is to mark a file dirty if there's a
6978                          * chance it was dirtied before the client was evicted,
6979                          * so that we don't have to wait for a release attempt
6980                          * before finding out the file was actually dirty and
6981                          * fail the release. Aggressively marking it dirty here
6982                          * will cause the policy engine to attempt to
6983                          * re-archive it; when rearchiving, we can compare the
6984                          * current version to the HSM data_version and make the
6985                          * archive request into a noop if it's not actually
6986                          * dirty.
6987                          */
6988                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6989                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6990
6991                         /* Don't unlink orphan on failover umount, LU-184 */
6992                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6993                             exp->exp_obd->obd_stopping) {
6994                                 ma->ma_valid = MA_FLAGS;
6995                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6996                         }
6997                         ma->ma_valid |= MA_FORCE_LOG;
6998                         mdt_mfd_close(info, mfd);
6999                 }
7000         }
7001         info->mti_mdt = NULL;
7002         /* cleanup client slot early */
7003         /* Do not erase record for recoverable client. */
7004         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
7005                 tgt_client_del(&env, exp);
7006         lu_env_fini(&env);
7007
7008         RETURN(rc);
7009 }
7010
7011 static int mdt_obd_disconnect(struct obd_export *exp)
7012 {
7013         struct obd_connect_data *data = &exp->exp_connect_data;
7014         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
7015         int rc;
7016
7017         ENTRY;
7018
7019         LASSERT(exp);
7020         class_export_get(exp);
7021
7022         if (!(exp->exp_flags & OBD_OPT_FORCE))
7023                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7024
7025         if (OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7026             atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
7027                 mdt_disable_slc(mdt);
7028
7029         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7030             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT) &&
7031             atomic_dec_and_test(&mdt->mdt_dmv_old_client_count))
7032                 mdt->mdt_enable_dmv_implicit_inherit = 1;
7033
7034         rc = server_disconnect_export(exp);
7035         if (rc != 0)
7036                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
7037
7038         tgt_grant_discard(exp);
7039
7040         rc = mdt_export_cleanup(exp);
7041         nodemap_del_member(exp);
7042         class_export_put(exp);
7043         RETURN(rc);
7044 }
7045
7046 /* mds_connect copy */
7047 static int mdt_obd_connect(const struct lu_env *env,
7048                            struct obd_export **exp, struct obd_device *obd,
7049                            struct obd_uuid *cluuid,
7050                            struct obd_connect_data *data,
7051                            void *localdata)
7052 {
7053         struct obd_export       *lexp;
7054         struct lustre_handle    conn = { 0 };
7055         struct mdt_device       *mdt;
7056         int                      rc;
7057         struct lnet_nid         *client_nid = localdata;
7058         ENTRY;
7059
7060         LASSERT(env != NULL);
7061         LASSERT(data != NULL);
7062
7063         if (!exp || !obd || !cluuid)
7064                 RETURN(-EINVAL);
7065
7066         mdt = mdt_dev(obd->obd_lu_dev);
7067
7068         /*
7069          * first, check whether the stack is ready to handle requests
7070          * XXX: probably not very appropriate method is used now
7071          *      at some point we should find a better one
7072          */
7073         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
7074             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
7075             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
7076                 rc = obd_get_info(env, mdt->mdt_child_exp,
7077                                   sizeof(KEY_OSP_CONNECTED),
7078                                   KEY_OSP_CONNECTED, NULL, NULL);
7079                 if (rc)
7080                         RETURN(-EAGAIN);
7081                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
7082         }
7083
7084         rc = class_connect(&conn, obd, cluuid);
7085         if (rc)
7086                 RETURN(rc);
7087
7088         lexp = class_conn2export(&conn);
7089         LASSERT(lexp != NULL);
7090
7091         if (nid_is_nid4(client_nid)) {
7092                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), lexp);
7093                 if (rc != 0 && rc != -EEXIST)
7094                         GOTO(out, rc);
7095         }
7096
7097         rc = mdt_connect_internal(env, lexp, mdt, data, false);
7098         if (rc == 0) {
7099                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
7100
7101                 LASSERT(lcd);
7102                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
7103                 rc = tgt_client_new(env, lexp);
7104                 if (rc == 0)
7105                         mdt_export_stats_init(obd, lexp, localdata);
7106         }
7107 out:
7108         if (rc != 0) {
7109                 class_disconnect(lexp);
7110                 nodemap_del_member(lexp);
7111                 *exp = NULL;
7112         } else {
7113                 *exp = lexp;
7114                 /* Because we do not want this export to be evicted by pinger,
7115                  * let's not add this export to the timed chain list. */
7116                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
7117                         spin_lock(&lexp->exp_obd->obd_dev_lock);
7118                         list_del_init(&lexp->exp_obd_chain_timed);
7119                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
7120                 }
7121         }
7122
7123         RETURN(rc);
7124 }
7125
7126 static int mdt_obd_reconnect(const struct lu_env *env,
7127                              struct obd_export *exp, struct obd_device *obd,
7128                              struct obd_uuid *cluuid,
7129                              struct obd_connect_data *data,
7130                              void *localdata)
7131 {
7132         struct lnet_nid *client_nid = localdata;
7133         int rc;
7134         ENTRY;
7135
7136         if (exp == NULL || obd == NULL || cluuid == NULL)
7137                 RETURN(-EINVAL);
7138
7139         if (nid_is_nid4(client_nid)) {
7140                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), exp);
7141                 if (rc != 0 && rc != -EEXIST)
7142                         RETURN(rc);
7143         }
7144
7145         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
7146                                   true);
7147         if (rc == 0)
7148                 mdt_export_stats_init(obd, exp, localdata);
7149         else
7150                 nodemap_del_member(exp);
7151
7152         RETURN(rc);
7153 }
7154
7155 /* FIXME: Can we avoid using these two interfaces? */
7156 static int mdt_init_export(struct obd_export *exp)
7157 {
7158         struct mdt_export_data *med = &exp->exp_mdt_data;
7159         int                     rc;
7160         ENTRY;
7161
7162         INIT_LIST_HEAD(&med->med_open_head);
7163         spin_lock_init(&med->med_open_lock);
7164         spin_lock(&exp->exp_lock);
7165         exp->exp_connecting = 1;
7166         spin_unlock(&exp->exp_lock);
7167
7168         OBD_ALLOC(exp->exp_used_slots,
7169                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7170         if (exp->exp_used_slots == NULL)
7171                 RETURN(-ENOMEM);
7172
7173         /* self-export doesn't need client data and ldlm initialization */
7174         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7175                                      &exp->exp_client_uuid)))
7176                 RETURN(0);
7177
7178         rc = tgt_client_alloc(exp);
7179         if (rc)
7180                 GOTO(err, rc);
7181
7182         rc = ldlm_init_export(exp);
7183         if (rc)
7184                 GOTO(err_free, rc);
7185
7186         RETURN(rc);
7187
7188 err_free:
7189         tgt_client_free(exp);
7190 err:
7191         OBD_FREE(exp->exp_used_slots,
7192                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7193         exp->exp_used_slots = NULL;
7194
7195         CERROR("%s: Failed to initialize export: rc = %d\n",
7196                exp->exp_obd->obd_name, rc);
7197         return rc;
7198 }
7199
7200 static int mdt_destroy_export(struct obd_export *exp)
7201 {
7202         ENTRY;
7203
7204         target_destroy_export(exp);
7205         if (exp->exp_used_slots)
7206                 OBD_FREE(exp->exp_used_slots,
7207                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7208
7209         /* destroy can be called from failed obd_setup, so
7210          * checking uuid is safer than obd_self_export */
7211         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7212                                      &exp->exp_client_uuid)))
7213                 RETURN(0);
7214
7215         ldlm_destroy_export(exp);
7216         tgt_client_free(exp);
7217
7218         LASSERT(list_empty(&exp->exp_outstanding_replies));
7219         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
7220
7221         /*
7222          * discard grants once we're sure no more
7223          * interaction with the client is possible
7224          */
7225         tgt_grant_discard(exp);
7226         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
7227                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
7228
7229         if (!(exp->exp_flags & OBD_OPT_FORCE))
7230                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7231
7232         RETURN(0);
7233 }
7234
7235 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
7236                    struct linkea_data *ldata)
7237 {
7238         int rc;
7239
7240         LASSERT(ldata->ld_buf->lb_buf != NULL);
7241
7242         if (!mdt_object_exists(mdt_obj))
7243                 return -ENODATA;
7244
7245         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7246                           ldata->ld_buf, XATTR_NAME_LINK);
7247         if (rc == -ERANGE) {
7248                 /* Buf was too small, figure out what we need. */
7249                 lu_buf_free(ldata->ld_buf);
7250                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7251                                   ldata->ld_buf, XATTR_NAME_LINK);
7252                 if (rc < 0)
7253                         return rc;
7254                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7255                 if (ldata->ld_buf->lb_buf == NULL)
7256                         return -ENOMEM;
7257                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7258                                   ldata->ld_buf, XATTR_NAME_LINK);
7259         }
7260         if (rc < 0)
7261                 return rc;
7262
7263         return linkea_init_with_rec(ldata);
7264 }
7265
7266 /**
7267  * Given an MDT object, try to look up the full path to the object.
7268  * Part of the MDT layer implementation of lfs fid2path.
7269  *
7270  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7271  * \param[in]     obj   Object to do path lookup of
7272  * \param[in,out] fp    User-provided struct to store path information
7273  * \param[in]     root_fid Root FID of current path should reach
7274  *
7275  * \retval 0 Lookup successful, path information stored in fp
7276  * \retval -EAGAIN Lookup failed, usually because object is being moved
7277  * \retval negative errno if there was a problem
7278  */
7279 static int mdt_path_current(struct mdt_thread_info *info,
7280                             struct mdt_object *obj,
7281                             struct getinfo_fid2path *fp,
7282                             struct lu_fid *root_fid)
7283 {
7284         struct mdt_device *mdt = info->mti_mdt;
7285         struct lu_name *tmpname = &info->mti_name;
7286         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7287         struct lu_buf *buf = &info->mti_big_buf;
7288         struct linkea_data ldata = { NULL };
7289         bool first = true;
7290         struct mdt_object *mdt_obj;
7291         struct link_ea_header *leh;
7292         struct link_ea_entry *lee;
7293         bool worthchecking = true;
7294         bool needsfid = false;
7295         bool supported = false;
7296         int isenc = -1;
7297         char *ptr;
7298         int reclen;
7299         int rc = 0;
7300
7301         ENTRY;
7302
7303         /* temp buffer for path element, the buffer will be finally freed
7304          * in mdt_thread_info_fini */
7305         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7306         if (buf->lb_buf == NULL)
7307                 RETURN(-ENOMEM);
7308
7309         ldata.ld_buf = buf;
7310         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7311         *ptr = 0;
7312         --ptr;
7313         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7314
7315         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7316                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7317                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7318                         GOTO(out, rc = -ENOENT);
7319
7320                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7321                         mdt_obj = obj;
7322                         mdt_object_get(info->mti_env, mdt_obj);
7323                 } else {
7324                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7325                         if (IS_ERR(mdt_obj))
7326                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7327                 }
7328
7329                 if (!mdt_object_exists(mdt_obj)) {
7330                         mdt_object_put(info->mti_env, mdt_obj);
7331                         GOTO(out, rc = -ENOENT);
7332                 }
7333
7334                 if (mdt_object_remote(mdt_obj)) {
7335                         mdt_object_put(info->mti_env, mdt_obj);
7336                         GOTO(remote_out, rc = -EREMOTE);
7337                 }
7338
7339                 if (worthchecking) {
7340                         /* we need to know if the FID being
7341                          * looked up is encrypted
7342                          */
7343                         struct lu_attr la = { 0 };
7344                         struct dt_object *dt = mdt_obj2dt(mdt_obj);
7345
7346                         if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7347                                 dt_attr_get(info->mti_env, dt, &la);
7348                         if (la.la_valid & LA_FLAGS &&
7349                             la.la_flags & LUSTRE_ENCRYPT_FL) {
7350                                 if (!supported && mdt_info_req(info) &&
7351                                     !exp_connect_encrypt_fid2path(
7352                                             mdt_info_req(info)->rq_export)) {
7353                                         /* client does not support fid2path
7354                                          * for encrypted files
7355                                          */
7356                                         mdt_object_put(info->mti_env, mdt_obj);
7357                                         GOTO(out, rc = -ENODATA);
7358                                 } else {
7359                                         supported = true;
7360                                 }
7361                                 needsfid = true;
7362                                 if (isenc == -1)
7363                                         isenc = 1;
7364                         } else {
7365                                 worthchecking = false;
7366                                 needsfid = false;
7367                                 if (isenc == -1)
7368                                         isenc = 0;
7369                         }
7370                 }
7371
7372                 rc = mdt_links_read(info, mdt_obj, &ldata);
7373                 if (rc != 0) {
7374                         mdt_object_put(info->mti_env, mdt_obj);
7375                         GOTO(out, rc);
7376                 }
7377
7378                 leh = buf->lb_buf;
7379                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7380                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7381                 /* If set, use link #linkno for path lookup, otherwise use
7382                    link #0.  Only do this for the final path element. */
7383                 if (first && fp->gf_linkno < leh->leh_reccount) {
7384                         int count;
7385                         for (count = 0; count < fp->gf_linkno; count++) {
7386                                 lee = (struct link_ea_entry *)
7387                                      ((char *)lee + reclen);
7388                                 linkea_entry_unpack(lee, &reclen, tmpname,
7389                                                     tmpfid);
7390                         }
7391                         if (fp->gf_linkno < leh->leh_reccount - 1)
7392                                 /* indicate to user there are more links */
7393                                 fp->gf_linkno++;
7394                 }
7395
7396                 /* Check if it is slave stripes */
7397                 rc = mdt_is_dir_stripe(info, mdt_obj);
7398                 mdt_object_put(info->mti_env, mdt_obj);
7399                 if (rc < 0)
7400                         GOTO(out, rc);
7401                 if (rc == 1) {
7402                         fp->gf_fid = *tmpfid;
7403                         continue;
7404                 }
7405
7406                 /* Pack the name in the end of the buffer */
7407                 ptr -= tmpname->ln_namelen;
7408                 if (ptr - 1 <= fp->gf_u.gf_path)
7409                         GOTO(out, rc = -ENAMETOOLONG);
7410                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7411                 if (needsfid) {
7412                         /* Pack FID before file name, so that client can build
7413                          * encoded/digested form.
7414                          */
7415                         char fidstr[FID_LEN + 1];
7416
7417                         snprintf(fidstr, sizeof(fidstr), DFID,
7418                                  PFID(&fp->gf_fid));
7419                         ptr -= strlen(fidstr);
7420                         if (ptr - 1 <= fp->gf_u.gf_path)
7421                                 GOTO(out, rc = -ENAMETOOLONG);
7422                         strncpy(ptr, fidstr, strlen(fidstr));
7423                 }
7424                 *(--ptr) = '/';
7425
7426                 /* keep the last resolved fid to the client, so the
7427                  * client will build the left path on another MDT for
7428                  * remote object */
7429                 fp->gf_fid = *tmpfid;
7430
7431                 first = false;
7432         }
7433
7434         /* non-zero will be treated as an error */
7435         rc = 0;
7436
7437 remote_out:
7438         if (isenc != 1)
7439                 ptr++; /* skip leading / unless this is an encrypted file */
7440         memmove(fp->gf_u.gf_path, ptr,
7441                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7442
7443 out:
7444         RETURN(rc);
7445 }
7446
7447 /**
7448  * Given an MDT object, use mdt_path_current to get the path.
7449  * Essentially a wrapper to retry mdt_path_current a set number of times
7450  * if -EAGAIN is returned (usually because an object is being moved).
7451  *
7452  * Part of the MDT layer implementation of lfs fid2path.
7453  *
7454  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7455  * \param[in]     obj   Object to do path lookup of
7456  * \param[in,out] fp    User-provided struct for arguments and to store path
7457  *                      information
7458  *
7459  * \retval 0 Lookup successful, path information stored in fp
7460  * \retval negative errno if there was a problem
7461  */
7462 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7463                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7464 {
7465         struct mdt_device       *mdt = info->mti_mdt;
7466         int                     tries = 3;
7467         int                     rc = -EAGAIN;
7468         ENTRY;
7469
7470         if (fp->gf_pathlen < 3)
7471                 RETURN(-EOVERFLOW);
7472
7473         if (root_fid == NULL)
7474                 root_fid = &mdt->mdt_md_root_fid;
7475
7476         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7477                 fp->gf_u.gf_path[0] = '\0';
7478                 RETURN(0);
7479         }
7480
7481         /* Retry multiple times in case file is being moved */
7482         while (tries-- && rc == -EAGAIN)
7483                 rc = mdt_path_current(info, obj, fp, root_fid);
7484
7485         RETURN(rc);
7486 }
7487
7488 /**
7489  * Get the full path of the provided FID, as of changelog record recno.
7490  *
7491  * This checks sanity and looks up object for user provided FID
7492  * before calling the actual path lookup code.
7493  *
7494  * Part of the MDT layer implementation of lfs fid2path.
7495  *
7496  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7497  * \param[in,out] fp    User-provided struct for arguments and to store path
7498  *                      information
7499  *
7500  * \retval 0 Lookup successful, path information and recno stored in fp
7501  * \retval -ENOENT, object does not exist
7502  * \retval negative errno if there was a problem
7503  */
7504 static int mdt_fid2path(struct mdt_thread_info *info,
7505                         struct lu_fid *root_fid,
7506                         struct getinfo_fid2path *fp)
7507 {
7508         struct mdt_device *mdt = info->mti_mdt;
7509         struct mdt_object *obj;
7510         int    rc;
7511         ENTRY;
7512
7513         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7514                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7515
7516         if (!fid_is_sane(&fp->gf_fid))
7517                 RETURN(-EINVAL);
7518
7519         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7520                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7521                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7522                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7523                 RETURN(-EINVAL);
7524         }
7525
7526         /* return error if client-provided root fid
7527          * is not the one stored in the export
7528          */
7529         if (root_fid && !fid_is_zero(&info->mti_exp->exp_root_fid) &&
7530             !lu_fid_eq(root_fid, &info->mti_exp->exp_root_fid)) {
7531                 CDEBUG(D_INFO,
7532                        "%s: root fid from client "DFID" but "DFID" stored in export\n",
7533                        mdt_obd_name(mdt), PFID(root_fid),
7534                        PFID(&info->mti_exp->exp_root_fid));
7535                 RETURN(-EXDEV);
7536         }
7537
7538         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7539         if (IS_ERR(obj)) {
7540                 rc = PTR_ERR(obj);
7541                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7542                        PFID(&fp->gf_fid), rc);
7543                 RETURN(rc);
7544         }
7545
7546         if (mdt_object_remote(obj))
7547                 rc = -EREMOTE;
7548         else if (!mdt_object_exists(obj))
7549                 rc = -ENOENT;
7550         else
7551                 rc = 0;
7552
7553         if (rc < 0) {
7554                 mdt_object_put(info->mti_env, obj);
7555                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7556                        PFID(&fp->gf_fid), rc);
7557                 RETURN(rc);
7558         }
7559
7560         rc = mdt_path(info, obj, fp, root_fid);
7561
7562         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7563                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7564                fp->gf_recno, fp->gf_linkno);
7565
7566         mdt_object_put(info->mti_env, obj);
7567
7568         RETURN(rc);
7569 }
7570
7571 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7572                             void *val, int vallen)
7573 {
7574         struct getinfo_fid2path *fpout, *fpin;
7575         struct lu_fid *root_fid = NULL;
7576         int rc = 0;
7577
7578         fpin = key + round_up(sizeof(KEY_FID2PATH), 8);
7579         fpout = val;
7580
7581         if (req_capsule_req_need_swab(info->mti_pill))
7582                 lustre_swab_fid2path(fpin);
7583
7584         memcpy(fpout, fpin, sizeof(*fpin));
7585         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7586                 RETURN(-EINVAL);
7587
7588         if (keylen >= round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*fpin) +
7589                       sizeof(struct lu_fid)) {
7590                 /* client sent its root FID, which is normally fileset FID */
7591                 root_fid = fpin->gf_u.gf_root_fid;
7592                 if (req_capsule_req_need_swab(info->mti_pill))
7593                         lustre_swab_lu_fid(root_fid);
7594
7595                 if (root_fid != NULL && !fid_is_sane(root_fid))
7596                         RETURN(-EINVAL);
7597         }
7598
7599         rc = mdt_fid2path(info, root_fid, fpout);
7600         RETURN(rc);
7601 }
7602
7603 int mdt_get_info(struct tgt_session_info *tsi)
7604 {
7605         char    *key;
7606         int      keylen;
7607         __u32   *vallen;
7608         void    *valout;
7609         int      rc;
7610
7611         ENTRY;
7612
7613         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7614         if (key == NULL) {
7615                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7616                 RETURN(err_serious(-EFAULT));
7617         }
7618         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7619                                       RCL_CLIENT);
7620
7621         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7622         if (vallen == NULL) {
7623                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7624                                 tgt_name(tsi->tsi_tgt));
7625                 RETURN(err_serious(-EFAULT));
7626         }
7627
7628         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7629                              *vallen);
7630         rc = req_capsule_server_pack(tsi->tsi_pill);
7631         if (rc)
7632                 RETURN(err_serious(rc));
7633
7634         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7635         if (valout == NULL) {
7636                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7637                                 tgt_name(tsi->tsi_tgt));
7638                 RETURN(err_serious(-EFAULT));
7639         }
7640
7641         if (KEY_IS(KEY_FID2PATH)) {
7642                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7643
7644                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7645                 mdt_thread_info_fini(info);
7646         } else {
7647                 rc = -EINVAL;
7648         }
7649         RETURN(rc);
7650 }
7651
7652 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7653 {
7654         struct obd_ioctl_data *data = karg;
7655         struct lu_fid *fid;
7656         __u64 version;
7657         struct mdt_object *obj;
7658         struct mdt_lock_handle  *lh;
7659         int rc;
7660         ENTRY;
7661
7662         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7663             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7664                 RETURN(-EINVAL);
7665
7666         fid = (struct lu_fid *)data->ioc_inlbuf1;
7667
7668         if (!fid_is_sane(fid))
7669                 RETURN(-EINVAL);
7670
7671         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7672
7673         lh = &mti->mti_lh[MDT_LH_PARENT];
7674         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE, LCK_CR);
7675         if (IS_ERR(obj))
7676                 RETURN(PTR_ERR(obj));
7677
7678         if (mdt_object_remote(obj)) {
7679                 rc = -EREMOTE;
7680                 /**
7681                  * before calling version get the correct MDS should be
7682                  * fid, this is error to find remote object here
7683                  */
7684                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7685         } else if (!mdt_object_exists(obj)) {
7686                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7687                 rc = -ENOENT;
7688         } else {
7689                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7690                *(__u64 *)data->ioc_inlbuf2 = version;
7691                 rc = 0;
7692         }
7693         mdt_object_unlock_put(mti, obj, lh, 1);
7694         RETURN(rc);
7695 }
7696
7697 /* ioctls on obd dev */
7698 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7699                          void *karg, void __user *uarg)
7700 {
7701         struct obd_device *obd = exp->exp_obd;
7702         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7703         struct dt_device *dt = mdt->mdt_bottom;
7704         struct obd_ioctl_data *data;
7705         struct lu_env env;
7706         int rc;
7707
7708         ENTRY;
7709         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
7710                obd->obd_name, cmd, len, karg, uarg);
7711
7712         rc = lu_env_init(&env, LCT_MD_THREAD);
7713         if (rc)
7714                 RETURN(rc);
7715
7716         /* handle commands that don't use @karg first */
7717         switch (cmd) {
7718         case OBD_IOC_SYNC:
7719                 rc = mdt_device_sync(&env, mdt);
7720                 GOTO(out, rc);
7721         case OBD_IOC_SET_READONLY:
7722                 rc = dt_sync(&env, dt);
7723                 if (rc == 0)
7724                         rc = dt_ro(&env, dt);
7725                 GOTO(out, rc);
7726         }
7727
7728         if (unlikely(karg == NULL)) {
7729                 OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc = -EINVAL);
7730                 GOTO(out, rc);
7731         }
7732         data = karg;
7733
7734         switch (cmd) {
7735         case OBD_IOC_ABORT_RECOVERY: {
7736                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7737                         LCONSOLE_WARN("%s: Aborting MDT recovery\n",
7738                                       obd->obd_name);
7739                         obd->obd_abort_mdt_recovery = 1;
7740                         wake_up(&obd->obd_next_transno_waitq);
7741                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7742                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7743                         LCONSOLE_WARN("%s: Aborting client recovery\n",
7744                                       obd->obd_name);
7745                         obd->obd_abort_recovery = 1;
7746                         target_stop_recovery_thread(obd);
7747                 }
7748                 rc = 0;
7749                 break;
7750         }
7751         case OBD_IOC_CHANGELOG_REG:
7752         case OBD_IOC_CHANGELOG_DEREG:
7753         case OBD_IOC_CHANGELOG_CLEAR:
7754         case OBD_IOC_LLOG_PRINT:
7755         case OBD_IOC_LLOG_CANCEL:
7756                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env, mdt->mdt_child,
7757                                                            cmd, len, karg);
7758                 break;
7759         case OBD_IOC_START_LFSCK: {
7760                 struct md_device *next = mdt->mdt_child;
7761                 struct lfsck_start_param lsp;
7762
7763                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7764                 lsp.lsp_index_valid = 0;
7765                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7766                 break;
7767         }
7768         case OBD_IOC_STOP_LFSCK: {
7769                 struct md_device *next = mdt->mdt_child;
7770                 struct lfsck_stop stop;
7771
7772                 stop.ls_status = LS_STOPPED;
7773                 /* Old lfsck utils may pass NULL @stop. */
7774                 if (data->ioc_inlbuf1 == NULL)
7775                         stop.ls_flags = 0;
7776                 else
7777                         stop.ls_flags =
7778                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7779
7780                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7781                 break;
7782         }
7783         case OBD_IOC_QUERY_LFSCK: {
7784                 struct md_device *next = mdt->mdt_child;
7785
7786                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7787                                                  data->ioc_inlbuf1);
7788                 break;
7789         }
7790         case OBD_IOC_GET_OBJ_VERSION: {
7791                 struct mdt_thread_info *mti;
7792
7793                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7794                 memset(mti, 0, sizeof(*mti));
7795                 mti->mti_env = &env;
7796                 mti->mti_mdt = mdt;
7797                 mti->mti_exp = exp;
7798
7799                 rc = mdt_ioc_version_get(mti, karg);
7800                 break;
7801         }
7802         case OBD_IOC_CATLOGLIST: {
7803                 struct mdt_thread_info *mti;
7804
7805                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7806                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7807                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7808                                        &mti->mti_tmp_fid1);
7809                 break;
7810         }
7811         default:
7812                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
7813                 break;
7814         }
7815 out:
7816         lu_env_fini(&env);
7817         RETURN(rc);
7818 }
7819
7820 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7821 {
7822         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7823         int rc;
7824         ENTRY;
7825
7826         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7827                 struct lfsck_start_param lsp;
7828
7829                 lsp.lsp_start = NULL;
7830                 lsp.lsp_index_valid = 0;
7831                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7832                                                            OBD_IOC_START_LFSCK,
7833                                                            0, &lsp);
7834                 if (rc != 0 && rc != -EALREADY)
7835                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7836                               mdt_obd_name(mdt), rc);
7837         }
7838
7839         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7840         RETURN(rc);
7841 }
7842
7843 static int mdt_obd_postrecov(struct obd_device *obd)
7844 {
7845         struct lu_env env;
7846         int rc;
7847
7848         rc = lu_env_init(&env, LCT_MD_THREAD);
7849         if (rc)
7850                 RETURN(rc);
7851         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7852         lu_env_fini(&env);
7853         return rc;
7854 }
7855
7856 static const struct obd_ops mdt_obd_device_ops = {
7857         .o_owner          = THIS_MODULE,
7858         .o_set_info_async = mdt_obd_set_info_async,
7859         .o_connect        = mdt_obd_connect,
7860         .o_reconnect      = mdt_obd_reconnect,
7861         .o_disconnect     = mdt_obd_disconnect,
7862         .o_init_export    = mdt_init_export,
7863         .o_destroy_export = mdt_destroy_export,
7864         .o_iocontrol      = mdt_iocontrol,
7865         .o_postrecov      = mdt_obd_postrecov,
7866         /* Data-on-MDT IO methods */
7867         .o_preprw         = mdt_obd_preprw,
7868         .o_commitrw       = mdt_obd_commitrw,
7869 };
7870
7871 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7872                                          struct lu_device *d)
7873 {
7874         struct mdt_device *m = mdt_dev(d);
7875         ENTRY;
7876
7877         mdt_fini(env, m);
7878         RETURN(NULL);
7879 }
7880
7881 static struct lu_device *mdt_device_free(const struct lu_env *env,
7882                                          struct lu_device *d)
7883 {
7884         struct mdt_device *m = mdt_dev(d);
7885         ENTRY;
7886
7887         lu_device_fini(&m->mdt_lu_dev);
7888         OBD_FREE_PTR(m);
7889
7890         RETURN(NULL);
7891 }
7892
7893 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7894                                           struct lu_device_type *t,
7895                                           struct lustre_cfg *cfg)
7896 {
7897         struct lu_device  *l;
7898         struct mdt_device *m;
7899
7900         OBD_ALLOC_PTR(m);
7901         if (m != NULL) {
7902                 int rc;
7903
7904                 l = &m->mdt_lu_dev;
7905                 rc = mdt_init0(env, m, t, cfg);
7906                 if (rc != 0) {
7907                         mdt_device_free(env, l);
7908                         l = ERR_PTR(rc);
7909                         return l;
7910                 }
7911         } else
7912                 l = ERR_PTR(-ENOMEM);
7913         return l;
7914 }
7915
7916 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7917 LU_KEY_INIT(mdt, struct mdt_thread_info);
7918
7919 static void mdt_key_fini(const struct lu_context *ctx,
7920                          struct lu_context_key *key, void* data)
7921 {
7922         struct mdt_thread_info *info = data;
7923
7924         if (info->mti_big_lmm) {
7925                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7926                 info->mti_big_lmm = NULL;
7927                 info->mti_big_lmmsize = 0;
7928         }
7929
7930         if (info->mti_big_acl) {
7931                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7932                 info->mti_big_acl = NULL;
7933                 info->mti_big_aclsize = 0;
7934         }
7935
7936         OBD_FREE_PTR(info);
7937 }
7938
7939 /* context key: mdt_thread_key */
7940 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7941
7942 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7943 {
7944         return lu_ucred(info->mti_env);
7945 }
7946
7947 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7948 {
7949         return lu_ucred_check(info->mti_env);
7950 }
7951
7952 /**
7953  * Enable/disable COS (Commit On Sharing).
7954  *
7955  * Set/Clear the COS flag in mdt options.
7956  *
7957  * \param mdt mdt device
7958  * \param val 0 disables COS, other values enable COS
7959  */
7960 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7961 {
7962         struct lu_env env;
7963         int rc;
7964
7965         mdt->mdt_opts.mo_cos = val;
7966         rc = lu_env_init(&env, LCT_LOCAL);
7967         if (unlikely(rc != 0)) {
7968                 CWARN("%s: lu_env initialization failed, cannot "
7969                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7970                 return;
7971         }
7972         mdt_device_sync(&env, mdt);
7973         lu_env_fini(&env);
7974 }
7975
7976 /**
7977  * Check COS (Commit On Sharing) status.
7978  *
7979  * Return COS flag status.
7980  *
7981  * \param mdt mdt device
7982  */
7983 int mdt_cos_is_enabled(struct mdt_device *mdt)
7984 {
7985         return mdt->mdt_opts.mo_cos != 0;
7986 }
7987
7988 static const struct lu_device_type_operations mdt_device_type_ops = {
7989         .ldto_device_alloc = mdt_device_alloc,
7990         .ldto_device_free  = mdt_device_free,
7991         .ldto_device_fini  = mdt_device_fini
7992 };
7993
7994 static struct lu_device_type mdt_device_type = {
7995         .ldt_tags     = LU_DEVICE_MD,
7996         .ldt_name     = LUSTRE_MDT_NAME,
7997         .ldt_ops      = &mdt_device_type_ops,
7998         .ldt_ctx_tags = LCT_MD_THREAD
7999 };
8000
8001 static int __init mdt_init(void)
8002 {
8003         int rc;
8004
8005         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
8006                      FID_NOBRACE_LEN + 1);
8007         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
8008                      FID_LEN + 1);
8009         rc = lu_kmem_init(mdt_caches);
8010         if (rc)
8011                 return rc;
8012
8013         rc = mds_mod_init();
8014         if (rc)
8015                 GOTO(lu_fini, rc);
8016
8017         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
8018                                  LUSTRE_MDT_NAME, &mdt_device_type);
8019         if (rc)
8020                 GOTO(mds_fini, rc);
8021 lu_fini:
8022         if (rc)
8023                 lu_kmem_fini(mdt_caches);
8024 mds_fini:
8025         if (rc)
8026                 mds_mod_exit();
8027         return rc;
8028 }
8029
8030 static void __exit mdt_exit(void)
8031 {
8032         class_unregister_type(LUSTRE_MDT_NAME);
8033         mds_mod_exit();
8034         lu_kmem_fini(mdt_caches);
8035 }
8036
8037 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8038 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
8039 MODULE_VERSION(LUSTRE_VERSION_STRING);
8040 MODULE_LICENSE("GPL");
8041
8042 module_init(mdt_init);
8043 module_exit(mdt_exit);