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         timeout_t at_est;
 497
 498         ENTRY;
 499
 500         svcpt = req->rq_rqbd->rqbd_svcpt;
 501
 502         /* This will trigger a watchdog timeout */
 503         at_est = obd_at_get(mdt->mdt_lu_dev.ld_obd, &svcpt->scp_at_estimate);
 504         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 505                          (MDT_SERVICE_WATCHDOG_FACTOR * at_est) + 1);
 506
 507         rc = mdt_check_ucred(info);
 508         if (rc)
 509                 GOTO(out, rc = err_serious(rc));
 510
 511         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 512                 GOTO(out, rc = err_serious(-ENOMEM));
 513
 514         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 515         if (!osfs)
 516                 GOTO(out, rc = -EPROTO);
 517
 518         if (mdt_is_sum_statfs_client(req->rq_export) &&
 519                 lustre_packed_msg_size(req->rq_reqmsg) ==
 520                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 521                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 522                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 523                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 524         }
 525
 526         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 527                 msf = &mdt->mdt_sum_osfs;
 528         else
 529                 msf = &mdt->mdt_osfs;
 530
 531         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 532                         /** statfs data is too old, get up-to-date one */
 533                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 534                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 535                                                               next, osfs);
 536                         else
 537                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 538                                                osfs);
 539                         if (rc)
 540                                 GOTO(out, rc);
 541                         spin_lock(&mdt->mdt_lock);
 542                         msf->msf_osfs = *osfs;
 543                         msf->msf_age = ktime_get_seconds();
 544                         spin_unlock(&mdt->mdt_lock);
 545         } else {
 546                         /** use cached statfs data */
 547                         spin_lock(&mdt->mdt_lock);
 548                         *osfs = msf->msf_osfs;
 549                         spin_unlock(&mdt->mdt_lock);
 550         }
 551
 552         /* tgd_blockbit is recordsize bits set during mkfs.
 553          * This once set does not change. However, 'zfs set'
 554          * can be used to change the MDT blocksize. Instead
 555          * of using cached value of 'tgd_blockbit' always
 556          * calculate the blocksize bits which may have
 557          * changed.
 558          */
 559         current_blockbits = fls64(osfs->os_bsize) - 1;
 560
 561         /* at least try to account for cached pages.  its still racy and
 562          * might be under-reporting if clients haven't announced their
 563          * caches with brw recently */
 564         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 565                " pending %llu free %llu avail %llu\n",
 566                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 567                tgd->tgd_tot_pending,
 568                osfs->os_bfree << current_blockbits,
 569                osfs->os_bavail << current_blockbits);
 570
 571         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 572                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 573                                    osfs->os_bsize - 1) >> current_blockbits));
 574
 575         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 576         if (mdt->mdt_lut.lut_no_create)
 577                 osfs->os_state |= OS_STATFS_NOCREATE;
 578         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 579                "%llu objects: %llu free; state %x\n",
 580                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 581                osfs->os_files, osfs->os_ffree, osfs->os_state);
 582
 583         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 584             current_blockbits > COMPAT_BSIZE_SHIFT) {
 585                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 586                  * should not see a block size > page size, otherwise
 587                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 588                  * block size which is the biggest block size known to work
 589                  * with all client's page size. */
 590                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 591                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 592                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 593                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 594         }
 595         if (rc == 0)
 596                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 597                                  ktime_us_delta(ktime_get(), kstart));
 598 out:
 599         mdt_thread_info_fini(info);
 600         RETURN(rc);
 601 }
 602
 603 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 604 {
 605         struct lov_comp_md_v1 *comp_v1;
 606         struct lov_mds_md *v1;
 607         __u32 off;
 608         __u32 dom_stripesize = 0;
 609         int i;
 610         bool has_ost_stripes = false;
 611
 612         ENTRY;
 613
 614         if (is_dom_only)
 615                 *is_dom_only = 0;
 616
 617         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 618                 RETURN(0);
 619
 620         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 621         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 622         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 623
 624         /* Fast check for DoM entry with no mirroring, should be the first */
 625         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 626             !(lov_pattern(le32_to_cpu(v1->lmm_pattern)) & LOV_PATTERN_MDT))
 627                 RETURN(0);
 628
 629         /* check all entries otherwise */
 630         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 631                 struct lov_comp_md_entry_v1 *lcme;
 632
 633                 lcme = &comp_v1->lcm_entries[i];
 634                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 635                         continue;
 636
 637                 off = le32_to_cpu(lcme->lcme_offset);
 638                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 639
 640                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) &
 641                     LOV_PATTERN_MDT)
 642                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 643                 else
 644                         has_ost_stripes = true;
 645
 646                 if (dom_stripesize && has_ost_stripes)
 647                         RETURN(dom_stripesize);
 648         }
 649         /* DoM-only case exits here */
 650         if (is_dom_only && dom_stripesize)
 651                 *is_dom_only = 1;
 652         RETURN(dom_stripesize);
 653 }
 654
 655 /**
 656  * Pack size attributes into the reply.
 657  */
 658 int mdt_pack_size2body(struct mdt_thread_info *info,
 659                         const struct lu_fid *fid, struct lustre_handle *lh)
 660 {
 661         struct mdt_body *b;
 662         struct md_attr *ma = &info->mti_attr;
 663         __u32 dom_stripe;
 664         bool dom_lock = false;
 665
 666         ENTRY;
 667
 668         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 669
 670         if (!S_ISREG(ma->ma_attr.la_mode) ||
 671             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 672                 RETURN(-ENODATA);
 673
 674         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 675         /* no DoM stripe, no size in reply */
 676         if (!dom_stripe)
 677                 RETURN(-ENOENT);
 678
 679         if (lustre_handle_is_used(lh)) {
 680                 struct ldlm_lock *lock;
 681
 682                 lock = ldlm_handle2lock(lh);
 683                 if (lock != NULL) {
 684                         dom_lock = ldlm_has_dom(lock);
 685                         LDLM_LOCK_PUT(lock);
 686                 }
 687         }
 688
 689         /* no DoM lock, no size in reply */
 690         if (!dom_lock)
 691                 RETURN(0);
 692
 693         /* Either DoM lock exists or LMM has only DoM stripe then
 694          * return size on body. */
 695         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 696
 697         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 698         RETURN(0);
 699 }
 700
 701 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 702 /*
 703  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 704  *
 705  * \param       info    thread info object
 706  * \param       repbody reply to pack ACLs into
 707  * \param       o       mdt object of file to examine
 708  * \param       nodemap nodemap of client to reply to
 709  * \retval      0       success
 710  * \retval      -errno  error getting or parsing ACL from disk
 711  */
 712 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 713                       struct mdt_object *o, struct lu_nodemap *nodemap)
 714 {
 715         const struct lu_env     *env = info->mti_env;
 716         struct md_object        *next = mdt_object_child(o);
 717         struct lu_buf           *buf = &info->mti_buf;
 718         struct mdt_device       *mdt = info->mti_mdt;
 719         struct req_capsule *pill = info->mti_pill;
 720         int rc;
 721
 722         ENTRY;
 723
 724         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 725         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 726         if (buf->lb_len == 0)
 727                 RETURN(0);
 728
 729         LASSERT(!info->mti_big_acl_used);
 730 again:
 731         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 732         if (rc < 0) {
 733                 if (rc == -ENODATA) {
 734                         repbody->mbo_aclsize = 0;
 735                         repbody->mbo_valid |= OBD_MD_FLACL;
 736                         rc = 0;
 737                 } else if (rc == -EOPNOTSUPP) {
 738                         rc = 0;
 739                 } else if (rc == -ERANGE) {
 740                         if (exp_connect_large_acl(info->mti_exp) &&
 741                             !info->mti_big_acl_used) {
 742                                 if (info->mti_big_acl == NULL) {
 743                                         info->mti_big_aclsize =
 744                                                         min_t(unsigned int,
 745                                                               mdt->mdt_max_ea_size,
 746                                                               XATTR_SIZE_MAX);
 747                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 748                                                         info->mti_big_aclsize);
 749                                         if (info->mti_big_acl == NULL) {
 750                                                 info->mti_big_aclsize = 0;
 751                                                 CERROR("%s: unable to grow "
 752                                                        DFID" ACL buffer\n",
 753                                                        mdt_obd_name(mdt),
 754                                                        PFID(mdt_object_fid(o)));
 755                                                 RETURN(-ENOMEM);
 756                                         }
 757                                 }
 758
 759                                 CDEBUG(D_INODE, "%s: grow the "DFID
 760                                        " ACL buffer to size %d\n",
 761                                        mdt_obd_name(mdt),
 762                                        PFID(mdt_object_fid(o)),
 763                                        info->mti_big_aclsize);
 764
 765                                 buf->lb_buf = info->mti_big_acl;
 766                                 buf->lb_len = info->mti_big_aclsize;
 767                                 info->mti_big_acl_used = 1;
 768                                 goto again;
 769                         }
 770                         /* FS has ACL bigger that our limits */
 771                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
 772                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
 773                                info->mti_big_aclsize);
 774                         rc = -E2BIG;
 775                 } else {
 776                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 777                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 778                 }
 779         } else {
 780                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 781                                      rc, NODEMAP_FS_TO_CLIENT);
 782                 /* if all ACLs mapped out, rc is still >= 0 */
 783                 if (rc < 0) {
 784                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 785                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 786                                PFID(mdt_object_fid(o)), rc);
 787                         repbody->mbo_aclsize = 0;
 788                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 789                 } else {
 790                         repbody->mbo_aclsize = rc;
 791                         repbody->mbo_valid |= OBD_MD_FLACL;
 792                         rc = 0;
 793                 }
 794         }
 795
 796         RETURN(rc);
 797 }
 798 #endif
 799
 800 /* XXX Look into layout in MDT layer. */
 801 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 802 {
 803         struct lov_comp_md_v1   *comp_v1;
 804         struct lov_mds_md       *v1;
 805         int                      i;
 806
 807         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 808                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 809
 810                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 811                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 812                                 comp_v1->lcm_entries[i].lcme_offset);
 813                         /* We don't support partial release for now */
 814                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 815                                 return false;
 816                 }
 817                 return true;
 818         } else {
 819                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 820                         true : false;
 821         }
 822 }
 823
 824 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 825                         const struct lu_attr *attr, const struct lu_fid *fid)
 826 {
 827         struct mdt_device *mdt = info->mti_mdt;
 828         struct obd_export *exp = info->mti_exp;
 829         struct md_attr *ma = &info->mti_attr;
 830         struct lu_nodemap *nodemap = NULL;
 831
 832         LASSERT(ma->ma_valid & MA_INODE);
 833
 834         if (attr->la_valid & LA_ATIME) {
 835                 b->mbo_atime = attr->la_atime;
 836                 b->mbo_valid |= OBD_MD_FLATIME;
 837         }
 838         if (attr->la_valid & LA_MTIME) {
 839                 b->mbo_mtime = attr->la_mtime;
 840                 b->mbo_valid |= OBD_MD_FLMTIME;
 841         }
 842         if (attr->la_valid & LA_CTIME) {
 843                 b->mbo_ctime = attr->la_ctime;
 844                 b->mbo_valid |= OBD_MD_FLCTIME;
 845         }
 846         if (attr->la_valid & LA_BTIME) {
 847                 b->mbo_btime = attr->la_btime;
 848                 b->mbo_valid |= OBD_MD_FLBTIME;
 849         }
 850         if (attr->la_valid & LA_FLAGS) {
 851                 b->mbo_flags = attr->la_flags;
 852                 b->mbo_valid |= OBD_MD_FLFLAGS;
 853         }
 854         if (attr->la_valid & LA_NLINK) {
 855                 b->mbo_nlink = attr->la_nlink;
 856                 b->mbo_valid |= OBD_MD_FLNLINK;
 857         }
 858         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
 859                 nodemap = nodemap_get_from_exp(exp);
 860                 if (IS_ERR(nodemap))
 861                         goto out;
 862         }
 863         if (attr->la_valid & LA_UID) {
 864                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 865                                             NODEMAP_FS_TO_CLIENT,
 866                                             attr->la_uid);
 867                 b->mbo_valid |= OBD_MD_FLUID;
 868         }
 869         if (attr->la_valid & LA_GID) {
 870                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 871                                             NODEMAP_FS_TO_CLIENT,
 872                                             attr->la_gid);
 873                 b->mbo_valid |= OBD_MD_FLGID;
 874         }
 875
 876         if (attr->la_valid & LA_PROJID) {
 877                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
 878                                                NODEMAP_FS_TO_CLIENT,
 879                                                attr->la_projid);
 880                 b->mbo_valid |= OBD_MD_FLPROJID;
 881         }
 882
 883         b->mbo_mode = attr->la_mode;
 884         if (attr->la_valid & LA_MODE)
 885                 b->mbo_valid |= OBD_MD_FLMODE;
 886         if (attr->la_valid & LA_TYPE)
 887                 b->mbo_valid |= OBD_MD_FLTYPE;
 888
 889         if (fid != NULL) {
 890                 b->mbo_fid1 = *fid;
 891                 b->mbo_valid |= OBD_MD_FLID;
 892                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 893                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 894         }
 895
 896         if (!(attr->la_valid & LA_TYPE))
 897                 return;
 898
 899         b->mbo_rdev   = attr->la_rdev;
 900         b->mbo_size   = attr->la_size;
 901         b->mbo_blocks = attr->la_blocks;
 902
 903         if (!S_ISREG(attr->la_mode)) {
 904                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 905         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 906                 /* means no objects are allocated on osts. */
 907                 LASSERT(!(ma->ma_valid & MA_LOV));
 908                 /* just ignore blocks occupied by extend attributes on MDS */
 909                 b->mbo_blocks = 0;
 910                 /* if no object is allocated on osts, the size on mds is valid.
 911                  * b=22272 */
 912                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 913         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 914                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 915                         /* A released file stores its size on MDS. */
 916                         /* But return 1 block for released file, unless tools
 917                          * like tar will consider it fully sparse. (LU-3864)
 918                          */
 919                         if (unlikely(b->mbo_size == 0))
 920                                 b->mbo_blocks = 0;
 921                         else
 922                                 b->mbo_blocks = 1;
 923                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 924                 } else if (info->mti_som_strict &&
 925                            mdt->mdt_opts.mo_enable_strict_som) {
 926                         /* use SOM for size*/
 927                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 928                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 929                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 930                         b->mbo_size = ma->ma_som.ms_size;
 931                         b->mbo_blocks = ma->ma_som.ms_blocks;
 932                 }
 933         }
 934
 935         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 936                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 937                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 938                        PFID(fid), (unsigned long long)b->mbo_size);
 939
 940 out:
 941         if (!IS_ERR_OR_NULL(nodemap))
 942                 nodemap_putref(nodemap);
 943 }
 944
 945 static inline int mdt_body_has_lov(const struct lu_attr *la,
 946                                    const struct mdt_body *body)
 947 {
 948         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 949                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 950 }
 951
 952 void mdt_client_compatibility(struct mdt_thread_info *info)
 953 {
 954         struct mdt_body       *body;
 955         struct ptlrpc_request *req = mdt_info_req(info);
 956         struct obd_export     *exp = req->rq_export;
 957         struct md_attr        *ma = &info->mti_attr;
 958         struct lu_attr        *la = &ma->ma_attr;
 959         ENTRY;
 960
 961         if (exp_connect_layout(exp))
 962                 /* the client can deal with 16-bit lmm_stripe_count */
 963                 RETURN_EXIT;
 964
 965         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 966
 967         if (!mdt_body_has_lov(la, body))
 968                 RETURN_EXIT;
 969
 970         /* now we have a reply with a lov for a client not compatible with the
 971          * layout lock so we have to clean the layout generation number */
 972         if (S_ISREG(la->la_mode))
 973                 ma->ma_lmm->lmm_layout_gen = 0;
 974         EXIT;
 975 }
 976
 977 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 978                                    struct mdt_object *o)
 979 {
 980         const struct lu_env *env = info->mti_env;
 981         int rc, rc2;
 982
 983         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 984                           XATTR_NAME_LOV);
 985
 986         if (rc == -ENODATA)
 987                 rc = 0;
 988
 989         if (rc < 0)
 990                 goto out;
 991
 992         /* Is it a directory? Let's check for the LMV as well */
 993         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 994                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 995                                    XATTR_NAME_LMV);
 996
 997                 if (rc2 == -ENODATA)
 998                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 999                                            &LU_BUF_NULL,
1000                                            XATTR_NAME_DEFAULT_LMV);
1001
1002                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
1003                         rc = rc2;
1004         }
1005
1006 out:
1007         return rc;
1008 }
1009
1010 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
1011                       const char *name)
1012 {
1013         const struct lu_env *env = info->mti_env;
1014         int rc;
1015         ENTRY;
1016
1017         LASSERT(info->mti_big_lmm_used == 0);
1018         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
1019         if (rc < 0)
1020                 RETURN(rc);
1021
1022         /* big_lmm may need to be grown */
1023         if (info->mti_big_lmmsize < rc) {
1024                 int size = size_roundup_power2(rc);
1025
1026                 if (info->mti_big_lmmsize > 0) {
1027                         /* free old buffer */
1028                         LASSERT(info->mti_big_lmm);
1029                         OBD_FREE_LARGE(info->mti_big_lmm,
1030                                        info->mti_big_lmmsize);
1031                         info->mti_big_lmm = NULL;
1032                         info->mti_big_lmmsize = 0;
1033                 }
1034
1035                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1036                 if (info->mti_big_lmm == NULL)
1037                         RETURN(-ENOMEM);
1038                 info->mti_big_lmmsize = size;
1039         }
1040         LASSERT(info->mti_big_lmmsize >= rc);
1041
1042         info->mti_buf.lb_buf = info->mti_big_lmm;
1043         info->mti_buf.lb_len = info->mti_big_lmmsize;
1044         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1045
1046         RETURN(rc);
1047 }
1048
1049 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1050                      struct md_attr *ma, const char *name)
1051 {
1052         struct md_object *next = mdt_object_child(o);
1053         struct lu_buf    *buf = &info->mti_buf;
1054         int rc;
1055
1056         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1057                 buf->lb_buf = ma->ma_lmm;
1058                 buf->lb_len = ma->ma_lmm_size;
1059                 LASSERT(!(ma->ma_valid & MA_LOV));
1060         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1061                 buf->lb_buf = ma->ma_lmv;
1062                 buf->lb_len = ma->ma_lmv_size;
1063                 LASSERT(!(ma->ma_valid & MA_LMV));
1064         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1065                 buf->lb_buf = ma->ma_default_lmv;
1066                 buf->lb_len = ma->ma_default_lmv_size;
1067                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1068         } else {
1069                 return -EINVAL;
1070         }
1071
1072         LASSERT(buf->lb_buf);
1073
1074         if (!mdt_object_exists(o))
1075                 return -ENOENT;
1076
1077         if (mdt_object_remote(o) && S_ISDIR(lu_object_attr(&o->mot_obj)))
1078                 /* force reload layout for remote dir in case layout changed */
1079                 mo_invalidate(info->mti_env, mdt_object_child(o));
1080
1081         rc = mo_xattr_get(info->mti_env, next, buf, name);
1082         if (rc > 0) {
1083
1084 got:
1085                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1086                         if (info->mti_big_lmm_used)
1087                                 ma->ma_lmm = info->mti_big_lmm;
1088
1089                         /* NOT return LOV EA with hole to old client. */
1090                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1091                                      LOV_PATTERN_F_HOLE) &&
1092                             !(exp_connect_flags(info->mti_exp) &
1093                               OBD_CONNECT_LFSCK)) {
1094                                 return -EIO;
1095                         } else {
1096                                 ma->ma_lmm_size = rc;
1097                                 ma->ma_valid |= MA_LOV;
1098                         }
1099                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1100                         if (info->mti_big_lmm_used)
1101                                 ma->ma_lmv = info->mti_big_lmm;
1102
1103                         ma->ma_lmv_size = rc;
1104                         ma->ma_valid |= MA_LMV;
1105                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1106                         ma->ma_default_lmv_size = rc;
1107                         ma->ma_valid |= MA_LMV_DEF;
1108                 }
1109
1110                 /* Update mdt_max_mdsize so all clients will be aware that */
1111                 if (info->mti_mdt->mdt_max_mdsize < rc)
1112                         info->mti_mdt->mdt_max_mdsize = rc;
1113
1114                 rc = 0;
1115         } else if (rc == -ENODATA) {
1116                 /* no LOV EA */
1117                 rc = 0;
1118         } else if (rc == -ERANGE) {
1119                 /* Default LMV has fixed size, so it must be able to fit
1120                  * in the original buffer */
1121                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1122                         return rc;
1123                 rc = mdt_big_xattr_get(info, o, name);
1124                 if (rc > 0) {
1125                         info->mti_big_lmm_used = 1;
1126                         goto got;
1127                 }
1128         }
1129
1130         return rc;
1131 }
1132
1133 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1134                    struct md_attr *ma, const char *name)
1135 {
1136         int rc;
1137
1138         if (!info->mti_big_lmm) {
1139                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1140                 if (!info->mti_big_lmm)
1141                         return -ENOMEM;
1142                 info->mti_big_lmmsize = PAGE_SIZE;
1143         }
1144
1145         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1146                 ma->ma_lmm = info->mti_big_lmm;
1147                 ma->ma_lmm_size = info->mti_big_lmmsize;
1148                 ma->ma_valid &= ~MA_LOV;
1149         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1150                 ma->ma_lmv = info->mti_big_lmm;
1151                 ma->ma_lmv_size = info->mti_big_lmmsize;
1152                 ma->ma_valid &= ~MA_LMV;
1153         } else {
1154                 LBUG();
1155         }
1156
1157         LASSERT(!info->mti_big_lmm_used);
1158         rc = __mdt_stripe_get(info, o, ma, name);
1159         /* since big_lmm is always used here, clear 'used' flag to avoid
1160          * assertion in mdt_big_xattr_get().
1161          */
1162         info->mti_big_lmm_used = 0;
1163
1164         return rc;
1165 }
1166
1167 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1168                       struct lu_fid *pfid)
1169 {
1170         struct lu_buf           *buf = &info->mti_buf;
1171         struct link_ea_header   *leh;
1172         struct link_ea_entry    *lee;
1173         int                      rc;
1174         ENTRY;
1175
1176         buf->lb_buf = info->mti_big_lmm;
1177         buf->lb_len = info->mti_big_lmmsize;
1178         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1179                           buf, XATTR_NAME_LINK);
1180         /* ignore errors, MA_PFID won't be set and it is
1181          * up to the caller to treat this as an error */
1182         if (rc == -ERANGE || buf->lb_len == 0) {
1183                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1184                 buf->lb_buf = info->mti_big_lmm;
1185                 buf->lb_len = info->mti_big_lmmsize;
1186         }
1187
1188         if (rc < 0)
1189                 RETURN(rc);
1190         if (rc < sizeof(*leh)) {
1191                 CERROR("short LinkEA on "DFID": rc = %d\n",
1192                        PFID(mdt_object_fid(o)), rc);
1193                 RETURN(-ENODATA);
1194         }
1195
1196         leh = (struct link_ea_header *) buf->lb_buf;
1197         lee = (struct link_ea_entry *)(leh + 1);
1198         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1199                 leh->leh_magic = LINK_EA_MAGIC;
1200                 leh->leh_reccount = __swab32(leh->leh_reccount);
1201                 leh->leh_len = __swab64(leh->leh_len);
1202         }
1203         if (leh->leh_magic != LINK_EA_MAGIC)
1204                 RETURN(-EINVAL);
1205         if (leh->leh_reccount == 0)
1206                 RETURN(-ENODATA);
1207
1208         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1209         fid_be_to_cpu(pfid, pfid);
1210
1211         RETURN(0);
1212 }
1213
1214 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1215                            struct lu_fid *pfid, struct lu_name *lname)
1216 {
1217         struct lu_buf *buf = &info->mti_buf;
1218         struct link_ea_header *leh;
1219         struct link_ea_entry *lee;
1220         int reclen;
1221         int rc;
1222
1223         buf->lb_buf = info->mti_xattr_buf;
1224         buf->lb_len = sizeof(info->mti_xattr_buf);
1225         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1226                           XATTR_NAME_LINK);
1227         if (rc == -ERANGE) {
1228                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1229                 buf->lb_buf = info->mti_big_lmm;
1230                 buf->lb_len = info->mti_big_lmmsize;
1231         }
1232         if (rc < 0)
1233                 return rc;
1234
1235         if (rc < sizeof(*leh)) {
1236                 CERROR("short LinkEA on "DFID": rc = %d\n",
1237                        PFID(mdt_object_fid(o)), rc);
1238                 return -ENODATA;
1239         }
1240
1241         leh = (struct link_ea_header *)buf->lb_buf;
1242         lee = (struct link_ea_entry *)(leh + 1);
1243         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1244                 leh->leh_magic = LINK_EA_MAGIC;
1245                 leh->leh_reccount = __swab32(leh->leh_reccount);
1246                 leh->leh_len = __swab64(leh->leh_len);
1247         }
1248         if (leh->leh_magic != LINK_EA_MAGIC)
1249                 return -EINVAL;
1250
1251         if (leh->leh_reccount == 0)
1252                 return -ENODATA;
1253
1254         linkea_entry_unpack(lee, &reclen, lname, pfid);
1255
1256         return 0;
1257 }
1258
1259 int mdt_attr_get_complex(struct mdt_thread_info *info,
1260                          struct mdt_object *o, struct md_attr *ma)
1261 {
1262         const struct lu_env *env = info->mti_env;
1263         struct md_object    *next = mdt_object_child(o);
1264         struct lu_buf       *buf = &info->mti_buf;
1265         int                  need = ma->ma_need;
1266         int                  rc = 0, rc2;
1267         u32                  mode;
1268         ENTRY;
1269
1270         ma->ma_valid = 0;
1271
1272         if (mdt_object_exists(o) == 0)
1273                 GOTO(out, rc = -ENOENT);
1274         mode = lu_object_attr(&next->mo_lu);
1275
1276         if (need & MA_INODE) {
1277                 ma->ma_need = MA_INODE;
1278                 rc = mo_attr_get(env, next, ma);
1279                 if (rc)
1280                         GOTO(out, rc);
1281
1282                 if (S_ISREG(mode))
1283                         (void) mdt_get_som(info, o, ma);
1284                 ma->ma_valid |= MA_INODE;
1285         }
1286
1287         if (need & MA_PFID) {
1288                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1289                 if (rc == 0)
1290                         ma->ma_valid |= MA_PFID;
1291                 /* ignore this error, parent fid is not mandatory */
1292                 rc = 0;
1293         }
1294
1295         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1296                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1297                 if (rc)
1298                         GOTO(out, rc);
1299         }
1300
1301         if (need & MA_LMV && S_ISDIR(mode)) {
1302                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1303                 if (rc != 0)
1304                         GOTO(out, rc);
1305         }
1306
1307         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1308                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1309                 if (rc != 0)
1310                         GOTO(out, rc);
1311         }
1312
1313         /*
1314          * In the handle of MA_INODE, we may already get the SOM attr.
1315          */
1316         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1317                 rc = mdt_get_som(info, o, ma);
1318                 if (rc != 0)
1319                         GOTO(out, rc);
1320         }
1321
1322         if (need & MA_HSM && S_ISREG(mode)) {
1323                 buf->lb_buf = info->mti_xattr_buf;
1324                 buf->lb_len = sizeof(info->mti_xattr_buf);
1325                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1326                              sizeof(info->mti_xattr_buf));
1327                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1328                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1329                 if (rc2 == 0)
1330                         ma->ma_valid |= MA_HSM;
1331                 else if (rc2 < 0 && rc2 != -ENODATA)
1332                         GOTO(out, rc = rc2);
1333         }
1334
1335 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1336         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1337                 buf->lb_buf = ma->ma_acl;
1338                 buf->lb_len = ma->ma_acl_size;
1339                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1340                 if (rc2 > 0) {
1341                         ma->ma_acl_size = rc2;
1342                         ma->ma_valid |= MA_ACL_DEF;
1343                 } else if (rc2 == -ENODATA) {
1344                         /* no ACLs */
1345                         ma->ma_acl_size = 0;
1346                 } else
1347                         GOTO(out, rc = rc2);
1348         }
1349 #endif
1350 out:
1351         ma->ma_need = need;
1352         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1353                rc, ma->ma_valid, ma->ma_lmm);
1354         RETURN(rc);
1355 }
1356
1357 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
1358 {
1359         struct req_capsule *pill = info->mti_pill;
1360
1361         ENTRY;
1362         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
1363                                   RCL_SERVER))
1364                 /* pre-set size in server part with max size */
1365                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
1366                                      RCL_SERVER,
1367                                      info->mti_mdt->mdt_max_mdsize);
1368         EXIT;
1369 }
1370
1371 static int mdt_getattr_internal(struct mdt_thread_info *info,
1372                                 struct mdt_object *o, int ma_need)
1373 {
1374         struct mdt_device *mdt = info->mti_mdt;
1375         struct md_object *next = mdt_object_child(o);
1376         const struct mdt_body *reqbody = info->mti_body;
1377         struct ptlrpc_request *req = mdt_info_req(info);
1378         struct md_attr *ma = &info->mti_attr;
1379         struct lu_attr *la = &ma->ma_attr;
1380         struct req_capsule *pill = info->mti_pill;
1381         const struct lu_env *env = info->mti_env;
1382         struct mdt_body *repbody;
1383         struct lu_buf *buffer = &info->mti_buf;
1384         struct obd_export *exp = info->mti_exp;
1385         ktime_t kstart = ktime_get();
1386         int rc;
1387
1388         ENTRY;
1389
1390         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1391                 RETURN(err_serious(-ENOMEM));
1392
1393         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1394
1395         ma->ma_valid = 0;
1396
1397         if (mdt_object_remote(o)) {
1398                 /* This object is located on remote node.*/
1399                 /* Return -ENOTSUPP for old client */
1400                 if (!mdt_is_dne_client(req->rq_export))
1401                         GOTO(out, rc = -ENOTSUPP);
1402
1403                 repbody->mbo_fid1 = *mdt_object_fid(o);
1404                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1405                 GOTO(out, rc = 0);
1406         }
1407
1408         if (reqbody->mbo_eadatasize > 0) {
1409                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1410                 if (buffer->lb_buf == NULL)
1411                         GOTO(out, rc = -EPROTO);
1412                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1413                                                       RCL_SERVER);
1414         } else {
1415                 buffer->lb_buf = NULL;
1416                 buffer->lb_len = 0;
1417                 ma_need &= ~(MA_LOV | MA_LMV);
1418                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1419                        mdt_obd_name(info->mti_mdt),
1420                        req->rq_export->exp_client_uuid.uuid);
1421         }
1422
1423         /* from 2.12.58 intent_getattr pack default LMV in reply */
1424         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1425             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1426                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1427             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1428                                   RCL_SERVER)) {
1429                 ma->ma_lmv = buffer->lb_buf;
1430                 ma->ma_lmv_size = buffer->lb_len;
1431                 ma->ma_default_lmv = req_capsule_server_get(pill,
1432                                                 &RMF_DEFAULT_MDT_MD);
1433                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1434                                                 &RMF_DEFAULT_MDT_MD,
1435                                                 RCL_SERVER);
1436                 ma->ma_need = MA_INODE;
1437                 if (ma->ma_lmv_size > 0)
1438                         ma->ma_need |= MA_LMV;
1439                 if (ma->ma_default_lmv_size > 0)
1440                         ma->ma_need |= MA_LMV_DEF;
1441         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1442                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1443                 /* If it is dir and client require MEA, then we got MEA */
1444                 /* Assumption: MDT_MD size is enough for lmv size. */
1445                 ma->ma_lmv = buffer->lb_buf;
1446                 ma->ma_lmv_size = buffer->lb_len;
1447                 ma->ma_need = MA_INODE;
1448                 if (ma->ma_lmv_size > 0) {
1449                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1450                                 ma->ma_need |= MA_LMV;
1451                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1452                                 ma->ma_need |= MA_LMV_DEF;
1453                                 ma->ma_default_lmv = buffer->lb_buf;
1454                                 ma->ma_lmv = NULL;
1455                                 ma->ma_default_lmv_size = buffer->lb_len;
1456                                 ma->ma_lmv_size = 0;
1457                         }
1458                 }
1459         } else {
1460                 ma->ma_lmm = buffer->lb_buf;
1461                 ma->ma_lmm_size = buffer->lb_len;
1462                 ma->ma_need = MA_INODE | MA_HSM;
1463                 if (ma->ma_lmm_size > 0) {
1464                         ma->ma_need |= MA_LOV;
1465                         /* Older clients may crash if they getattr overstriped
1466                          * files
1467                          */
1468                         if (!exp_connect_overstriping(exp) &&
1469                             mdt_lmm_is_overstriping(ma->ma_lmm))
1470                                 RETURN(-EOPNOTSUPP);
1471                 }
1472         }
1473
1474         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1475             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1476             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1477                 /* get default stripe info for this dir. */
1478                 ma->ma_need |= MA_LOV_DEF;
1479         }
1480         ma->ma_need |= ma_need;
1481
1482         rc = mdt_attr_get_complex(info, o, ma);
1483         if (unlikely(rc)) {
1484                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1485                              "%s: getattr error for "DFID": rc = %d\n",
1486                              mdt_obd_name(info->mti_mdt),
1487                              PFID(mdt_object_fid(o)), rc);
1488                 RETURN(rc);
1489         }
1490
1491         /* return immutable attr on fscrypt metadata files
1492          * if fscrypt admin is not permitted
1493          */
1494         if (o->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD &&
1495             !mdt_ucred(info)->uc_rbac_fscrypt_admin)
1496                 la->la_flags |= LUSTRE_IMMUTABLE_FL;
1497
1498         /* if file is released, check if a restore is running */
1499         if (ma->ma_valid & MA_HSM) {
1500                 repbody->mbo_valid |= OBD_MD_TSTATE;
1501                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1502                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1503                         repbody->mbo_t_state = MS_RESTORE;
1504         }
1505
1506         if (unlikely(!(ma->ma_valid & MA_INODE)))
1507                 RETURN(-EFAULT);
1508
1509         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1510
1511         if (mdt_body_has_lov(la, reqbody)) {
1512                 u32 stripe_count = 1;
1513                 bool fixed_layout = false;
1514
1515                 if (ma->ma_valid & MA_LOV) {
1516                         LASSERT(ma->ma_lmm_size);
1517                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1518                         if (S_ISDIR(la->la_mode))
1519                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1520                         else
1521                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1522                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1523                 }
1524                 if (ma->ma_valid & MA_LMV) {
1525                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1526                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1527
1528                         /* Return -ENOTSUPP for old client */
1529                         if (!mdt_is_striped_client(req->rq_export))
1530                                 RETURN(-ENOTSUPP);
1531
1532                         LASSERT(S_ISDIR(la->la_mode));
1533                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1534                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1535                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1536
1537                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1538                         fixed_layout = lmv_is_fixed(lmv);
1539                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1540                                 mdt_restripe_migrate_add(info, o);
1541                         else if (magic == LMV_MAGIC_V1 &&
1542                                  lmv_is_restriping(lmv))
1543                                 mdt_restripe_update_add(info, o);
1544                 }
1545                 if (ma->ma_valid & MA_LMV_DEF) {
1546                         /* Return -ENOTSUPP for old client */
1547                         if (!mdt_is_striped_client(req->rq_export))
1548                                 RETURN(-ENOTSUPP);
1549                         LASSERT(S_ISDIR(la->la_mode));
1550                         /*
1551                          * when ll_dir_getstripe() gets default LMV, it
1552                          * checks mbo_eadatasize.
1553                          */
1554                         if (!(ma->ma_valid & MA_LMV))
1555                                 repbody->mbo_eadatasize =
1556                                         ma->ma_default_lmv_size;
1557                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1558                                                OBD_MD_DEFAULT_MEA);
1559                 }
1560                 CDEBUG(D_VFSTRACE,
1561                        "dirent count %llu stripe count %u MDT count %d\n",
1562                        ma->ma_attr.la_dirent_count, stripe_count,
1563                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1564                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1565                     ma->ma_attr.la_dirent_count >
1566                         mdt->mdt_restriper.mdr_dir_split_count &&
1567                     !fid_is_root(mdt_object_fid(o)) &&
1568                     mdt->mdt_enable_dir_auto_split &&
1569                     !o->mot_restriping &&
1570                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1571                     !fixed_layout)
1572                         mdt_auto_split_add(info, o);
1573         } else if (S_ISLNK(la->la_mode) &&
1574                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1575                 buffer->lb_buf = ma->ma_lmm;
1576                 /* eadatasize from client includes NULL-terminator, so
1577                  * there is no need to read it */
1578                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1579                 rc = mo_readlink(env, next, buffer);
1580                 if (unlikely(rc <= 0)) {
1581                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1582                                mdt_obd_name(info->mti_mdt),
1583                                PFID(mdt_object_fid(o)), rc);
1584                         rc = -EFAULT;
1585                 } else {
1586                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1587
1588                         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1589                                 rc -= 2;
1590                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1591                         /* we need to report back size with NULL-terminator
1592                          * because client expects that */
1593                         repbody->mbo_eadatasize = rc + 1;
1594                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1595                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1596                                        "on "DFID ", expected %d\n",
1597                                        mdt_obd_name(info->mti_mdt),
1598                                        rc, PFID(mdt_object_fid(o)),
1599                                        reqbody->mbo_eadatasize - 1);
1600                         /* NULL terminate */
1601                         ((char *)ma->ma_lmm)[rc] = 0;
1602
1603                         /* If the total CDEBUG() size is larger than a page, it
1604                          * will print a warning to the console, avoid this by
1605                          * printing just the last part of the symlink. */
1606                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1607                                print_limit < rc ? "..." : "", print_limit,
1608                                (char *)ma->ma_lmm + rc - print_limit, rc);
1609                         rc = 0;
1610                 }
1611         }
1612
1613         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1614                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1615                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1616                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1617                        repbody->mbo_max_mdsize);
1618         }
1619
1620 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1621         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1622                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1623                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1624                 if (IS_ERR(nodemap))
1625                         RETURN(PTR_ERR(nodemap));
1626
1627                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1628                 nodemap_putref(nodemap);
1629         }
1630 #endif
1631
1632 out:
1633         if (rc == 0)
1634                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1635                                  ktime_us_delta(ktime_get(), kstart));
1636
1637         RETURN(rc);
1638 }
1639
1640 static int mdt_getattr(struct tgt_session_info *tsi)
1641 {
1642         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1643         struct mdt_object       *obj = info->mti_object;
1644         struct req_capsule      *pill = info->mti_pill;
1645         struct mdt_body         *reqbody;
1646         struct mdt_body         *repbody;
1647         int rc, rc2;
1648         ENTRY;
1649
1650         if (unlikely(info->mti_object == NULL))
1651                 RETURN(-EPROTO);
1652
1653         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1654         LASSERT(reqbody);
1655         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1656
1657         /* Special case for Data-on-MDT files to get data version */
1658         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1659                 rc = mdt_data_version_get(tsi);
1660                 GOTO(out, rc);
1661         }
1662
1663         /* Unlike intent case where we need to pre-fill out buffers early on
1664          * in intent policy for ldlm reasons, here we can have a much better
1665          * guess at EA size by just reading it from disk.
1666          * Exceptions are readdir and (missing) directory striping */
1667         /* Readlink */
1668         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1669                 /* No easy way to know how long is the symlink, but it cannot
1670                  * be more than PATH_MAX, so we allocate +1 */
1671                 rc = PATH_MAX + 1;
1672         /* A special case for fs ROOT: getattr there might fetch
1673          * default EA for entire fs, not just for this dir!
1674          */
1675         } else if (lu_fid_eq(mdt_object_fid(obj),
1676                              &info->mti_mdt->mdt_md_root_fid) &&
1677                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1678                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1679                                                                  MDS_GETATTR)) {
1680                 /* Should the default strping be bigger, mdt_fix_reply
1681                  * will reallocate */
1682                 rc = DEF_REP_MD_SIZE;
1683         } else {
1684                 /* Read the actual EA size from disk */
1685                 rc = mdt_attr_get_eabuf_size(info, obj);
1686         }
1687
1688         if (rc < 0)
1689                 GOTO(out, rc = err_serious(rc));
1690
1691         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1692
1693         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1694          * by default. If the target object has more ACL entries, then
1695          * enlarge the buffer when necessary. */
1696         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1697                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1698         mdt_preset_encctx_size(info);
1699
1700         rc = req_capsule_server_pack(pill);
1701         if (unlikely(rc != 0))
1702                 GOTO(out, rc = err_serious(rc));
1703
1704         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1705         LASSERT(repbody != NULL);
1706         repbody->mbo_eadatasize = 0;
1707         repbody->mbo_aclsize = 0;
1708
1709         rc = mdt_check_ucred(info);
1710         if (unlikely(rc))
1711                 GOTO(out_shrink, rc);
1712
1713         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1714
1715         rc = mdt_init_ucred(info, reqbody);
1716         if (rc)
1717                 GOTO(out_shrink, rc);
1718
1719         rc = mdt_getattr_internal(info, obj, 0);
1720         if (unlikely(rc))
1721                 GOTO(out_ucred, rc);
1722
1723         rc = mdt_pack_encctx_in_reply(info, obj);
1724         EXIT;
1725 out_ucred:
1726         mdt_exit_ucred(info);
1727 out_shrink:
1728         mdt_client_compatibility(info);
1729         rc2 = mdt_fix_reply(info);
1730         if (rc == 0)
1731                 rc = rc2;
1732 out:
1733         mdt_thread_info_fini(info);
1734         return rc;
1735 }
1736
1737 /**
1738  * Handler of layout intent RPC requiring the layout modification
1739  *
1740  * \param[in]  info     thread environment
1741  * \param[in]  obj      object
1742  * \param[out] lhc      object ldlm lock handle
1743  * \param[in]  layout   layout change descriptor
1744  *
1745  * \retval 0    on success
1746  * \retval < 0  error code
1747  */
1748 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1749                       struct mdt_lock_handle *lhc,
1750                       struct md_layout_change *layout)
1751 {
1752         int rc;
1753
1754         ENTRY;
1755
1756         if (!mdt_object_exists(obj))
1757                 RETURN(-ENOENT);
1758
1759         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1760                 RETURN(-EINVAL);
1761
1762         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1763                            MAY_WRITE);
1764         if (rc)
1765                 RETURN(rc);
1766
1767         rc = mdt_check_resent_lock(info, obj, lhc);
1768         if (rc < 0)
1769                 RETURN(rc);
1770
1771         if (rc > 0) {
1772                 /* not resent */
1773                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1774
1775                 /* take layout lock to prepare layout change */
1776                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1777                         lockpart |= MDS_INODELOCK_UPDATE;
1778
1779                 rc = mdt_object_lock(info, obj, lhc, lockpart, LCK_EX);
1780                 if (rc)
1781                         RETURN(rc);
1782         }
1783
1784         mutex_lock(&obj->mot_som_mutex);
1785         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1786         mutex_unlock(&obj->mot_som_mutex);
1787
1788         if (rc)
1789                 mdt_object_unlock(info, obj, lhc, 1);
1790
1791         RETURN(rc);
1792 }
1793
1794 /**
1795  * Exchange MOF_LOV_CREATED flags between two objects after a
1796  * layout swap. No assumption is made on whether o1 or o2 have
1797  * created objects or not.
1798  *
1799  * \param[in,out] o1    First swap layout object
1800  * \param[in,out] o2    Second swap layout object
1801  */
1802 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1803 {
1804         unsigned int o1_lov_created = o1->mot_lov_created;
1805
1806         mutex_lock(&o1->mot_lov_mutex);
1807         mutex_lock(&o2->mot_lov_mutex);
1808
1809         o1->mot_lov_created = o2->mot_lov_created;
1810         o2->mot_lov_created = o1_lov_created;
1811
1812         mutex_unlock(&o2->mot_lov_mutex);
1813         mutex_unlock(&o1->mot_lov_mutex);
1814 }
1815
1816 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1817 {
1818         struct mdt_thread_info  *info;
1819         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1820         struct obd_export       *exp = req->rq_export;
1821         struct mdt_object       *o1, *o2, *o;
1822         struct mdt_lock_handle  *lh1, *lh2;
1823         struct mdc_swap_layouts *msl;
1824         int                      rc;
1825         ENTRY;
1826
1827         /* client does not support layout lock, so layout swaping
1828          * is disabled.
1829          * FIXME: there is a problem for old clients which don't support
1830          * layout lock yet. If those clients have already opened the file
1831          * they won't be notified at all so that old layout may still be
1832          * used to do IO. This can be fixed after file release is landed by
1833          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1834         if (!exp_connect_layout(exp))
1835                 RETURN(-EOPNOTSUPP);
1836
1837         info = tsi2mdt_info(tsi);
1838         if (unlikely(info->mti_object == NULL))
1839                 RETURN(-EPROTO);
1840
1841         if (info->mti_dlm_req != NULL)
1842                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1843
1844         o1 = info->mti_object;
1845         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1846                                 &info->mti_body->mbo_fid2);
1847         if (IS_ERR(o))
1848                 GOTO(out, rc = PTR_ERR(o));
1849
1850         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1851                 GOTO(put, rc = -ENOENT);
1852
1853         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1854         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1855                 GOTO(put, rc);
1856
1857         if (rc < 0)
1858                 swap(o1, o2);
1859
1860         /* permission check. Make sure the calling process having permission
1861          * to write both files. */
1862         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1863                            MAY_WRITE);
1864         if (rc < 0)
1865                 GOTO(put, rc);
1866
1867         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1868                            MAY_WRITE);
1869         if (rc < 0)
1870                 GOTO(put, rc);
1871
1872         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1873         if (msl == NULL)
1874                 GOTO(put, rc = -EPROTO);
1875
1876         lh1 = &info->mti_lh[MDT_LH_NEW];
1877         lh2 = &info->mti_lh[MDT_LH_OLD];
1878         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1879                              MDS_INODELOCK_XATTR, LCK_EX);
1880         if (rc < 0)
1881                 GOTO(put, rc);
1882
1883         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1884                              MDS_INODELOCK_XATTR, LCK_EX);
1885         if (rc < 0)
1886                 GOTO(unlock1, rc);
1887
1888         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1889                              mdt_object_child(o2), msl->msl_flags);
1890         if (rc < 0)
1891                 GOTO(unlock2, rc);
1892
1893         mdt_swap_lov_flag(o1, o2);
1894
1895 unlock2:
1896         mdt_object_unlock(info, o2, lh2, rc);
1897 unlock1:
1898         mdt_object_unlock(info, o1, lh1, rc);
1899 put:
1900         mdt_object_put(info->mti_env, o);
1901 out:
1902         mdt_thread_info_fini(info);
1903         RETURN(rc);
1904 }
1905
1906 static int mdt_raw_lookup(struct mdt_thread_info *info,
1907                           struct mdt_object *parent,
1908                           const struct lu_name *lname)
1909 {
1910         struct lu_fid *fid = &info->mti_tmp_fid1;
1911         struct mdt_body *repbody;
1912         bool is_dotdot = false;
1913         bool is_old_parent_stripe = false;
1914         bool is_new_parent_checked = false;
1915         int rc;
1916
1917         ENTRY;
1918
1919         LASSERT(!info->mti_cross_ref);
1920         /* Always allow to lookup ".." */
1921         if (lname->ln_namelen == 2 &&
1922             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1923                 info->mti_spec.sp_permitted = 1;
1924                 is_dotdot = true;
1925                 if (mdt_is_dir_stripe(info, parent) == 1)
1926                         is_old_parent_stripe = true;
1927         }
1928
1929         mdt_object_get(info->mti_env, parent);
1930 lookup:
1931         /* Only got the fid of this obj by name */
1932         fid_zero(fid);
1933         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1934                         &info->mti_spec);
1935         mdt_object_put(info->mti_env, parent);
1936         if (rc)
1937                 RETURN(rc);
1938
1939         /* getattr_name("..") should return master object FID for striped dir */
1940         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1941                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1942                 if (IS_ERR(parent))
1943                         RETURN(PTR_ERR(parent));
1944
1945                 /* old client getattr_name("..") with stripe FID */
1946                 if (unlikely(is_old_parent_stripe)) {
1947                         is_old_parent_stripe = false;
1948                         goto lookup;
1949                 }
1950
1951                 /* ".." may be a stripe */
1952                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1953                         is_new_parent_checked = true;
1954                         goto lookup;
1955                 }
1956
1957                 mdt_object_put(info->mti_env, parent);
1958         }
1959
1960         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1961         repbody->mbo_fid1 = *fid;
1962         repbody->mbo_valid = OBD_MD_FLID;
1963
1964         RETURN(rc);
1965 }
1966
1967 /**
1968  * Find name matching hash
1969  *
1970  * We search \a child LinkEA for a name whose hash matches \a lname
1971  * (it contains an encoded hash).
1972  *
1973  * \param info mdt thread info
1974  * \param lname encoded hash to find
1975  * \param parent parent object
1976  * \param child object to search with LinkEA
1977  *
1978  * \retval 1 match found
1979  * \retval 0 no match found
1980  * \retval -ev negative errno upon error
1981  */
1982 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
1983                             struct mdt_object *parent, struct mdt_object *child)
1984 {
1985         /* Here, lname is an encoded hash of on-disk name, and
1986          * client is doing access without encryption key.
1987          * So we need to get LinkEA, check parent fid is correct and
1988          * compare name hash with the one in the request.
1989          */
1990         struct lu_buf *buf = &info->mti_big_buf;
1991         struct lu_name name;
1992         struct lu_fid pfid;
1993         struct linkea_data ldata = { NULL };
1994         struct link_ea_header *leh;
1995         struct link_ea_entry *lee;
1996         struct lu_buf link = { 0 };
1997         char *hash;
1998         int reclen, count, rc;
1999
2000         ENTRY;
2001         if (lname->ln_namelen < LL_CRYPTO_BLOCK_SIZE)
2002                 RETURN(-EINVAL);
2003
2004         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2005         if (!buf->lb_buf)
2006                 RETURN(-ENOMEM);
2007
2008         ldata.ld_buf = buf;
2009         rc = mdt_links_read(info, child, &ldata);
2010         if (rc < 0)
2011                 RETURN(rc);
2012
2013         hash = kmalloc(lname->ln_namelen, GFP_NOFS);
2014         if (!hash)
2015                 RETURN(-ENOMEM);
2016         rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
2017
2018         leh = buf->lb_buf;
2019         lee = (struct link_ea_entry *)(leh + 1);
2020         for (count = 0; count < leh->leh_reccount; count++) {
2021                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
2022                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
2023                         lu_buf_check_and_alloc(&link, name.ln_namelen);
2024                         if (!link.lb_buf)
2025                                 GOTO(out_match, rc = -ENOMEM);
2026                         rc = critical_decode(name.ln_name, name.ln_namelen,
2027                                              link.lb_buf);
2028
2029                         if (memcmp(LLCRYPT_EXTRACT_DIGEST(link.lb_buf, rc),
2030                                    hash, LL_CRYPTO_BLOCK_SIZE) == 0) {
2031                                 *lname = name;
2032                                 break;
2033                         }
2034                 }
2035                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2036         }
2037         if (count == leh->leh_reccount)
2038                 rc = 0;
2039         else
2040                 rc = 1;
2041
2042 out_match:
2043         lu_buf_free(&link);
2044         kfree(hash);
2045
2046         RETURN(rc);
2047 }
2048
2049 /*
2050  * UPDATE lock should be taken against parent, and be released before exit;
2051  * child_bits lock should be taken against child, and be returned back:
2052  *            (1)normal request should release the child lock;
2053  *            (2)intent request will grant the lock to client.
2054  */
2055 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2056                                  struct mdt_lock_handle *lhc,
2057                                  __u64 child_bits,
2058                                  struct ldlm_reply *ldlm_rep)
2059 {
2060         struct ptlrpc_request *req = mdt_info_req(info);
2061         struct mdt_body *reqbody = NULL;
2062         struct mdt_object *parent = info->mti_object;
2063         struct mdt_object *child = NULL;
2064         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2065         struct lu_name *lname = NULL;
2066         struct mdt_lock_handle *lhp = NULL;
2067         struct ldlm_lock *lock;
2068         struct req_capsule *pill = info->mti_pill;
2069         bool fscrypt_md = false;
2070         __u64 try_bits = 0;
2071         bool is_resent;
2072         int ma_need = 0;
2073         int rc;
2074
2075         ENTRY;
2076
2077         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2078         LASSERT(ergo(is_resent,
2079                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2080
2081         if (parent == NULL)
2082                 RETURN(-ENOENT);
2083
2084         if (info->mti_cross_ref) {
2085                 /* Only getattr on the child. Parent is on another node. */
2086                 mdt_set_disposition(info, ldlm_rep,
2087                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2088                 child = parent;
2089                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
2090                        "ldlm_rep = %p\n",
2091                        PFID(mdt_object_fid(child)), ldlm_rep);
2092
2093                 rc = mdt_check_resent_lock(info, child, lhc);
2094                 if (rc < 0) {
2095                         RETURN(rc);
2096                 } else if (rc > 0) {
2097                         /*
2098                          * Object's name entry is on another MDS, it will
2099                          * request PERM lock only because LOOKUP lock is owned
2100                          * by the MDS where name entry resides.
2101                          *
2102                          * TODO: it should try layout lock too. - Jinshan
2103                          */
2104                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2105                                         MDS_INODELOCK_LAYOUT);
2106                         child_bits |= MDS_INODELOCK_PERM;
2107                         rc = mdt_object_lock(info, child, lhc, child_bits,
2108                                              LCK_PR);
2109                         if (rc < 0)
2110                                 RETURN(rc);
2111                 }
2112
2113                 /* Finally, we can get attr for child. */
2114                 if (!mdt_object_exists(child)) {
2115                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2116                                         &child->mot_obj,
2117                                         "remote object doesn't exist.");
2118                         mdt_object_unlock(info, child, lhc, 1);
2119                         RETURN(-ENOENT);
2120                 }
2121
2122                 rc = mdt_getattr_internal(info, child, 0);
2123                 if (unlikely(rc != 0)) {
2124                         mdt_object_unlock(info, child, lhc, 1);
2125                         RETURN(rc);
2126                 }
2127
2128                 rc = mdt_pack_secctx_in_reply(info, child);
2129                 if (unlikely(rc)) {
2130                         mdt_object_unlock(info, child, lhc, 1);
2131                         RETURN(rc);
2132                 }
2133
2134                 rc = mdt_pack_encctx_in_reply(info, child);
2135                 if (unlikely(rc))
2136                         mdt_object_unlock(info, child, lhc, 1);
2137                 RETURN(rc);
2138         }
2139
2140         lname = &info->mti_name;
2141         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2142
2143         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2144                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2145                 if (unlikely(reqbody == NULL))
2146                         RETURN(err_serious(-EPROTO));
2147
2148                 *child_fid = reqbody->mbo_fid2;
2149                 if (unlikely(!fid_is_sane(child_fid)))
2150                         RETURN(err_serious(-EINVAL));
2151
2152                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2153                         mdt_object_get(info->mti_env, parent);
2154                         child = parent;
2155                 } else {
2156                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2157                                                 child_fid);
2158                         if (IS_ERR(child))
2159                                 RETURN(PTR_ERR(child));
2160                 }
2161
2162                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2163                        "ldlm_rep = %p\n",
2164                        PFID(mdt_object_fid(parent)),
2165                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2166         } else if (lu_name_is_valid(lname)) {
2167                 if (mdt_object_remote(parent)) {
2168                         CERROR("%s: parent "DFID" is on remote target\n",
2169                                mdt_obd_name(info->mti_mdt),
2170                                PFID(mdt_object_fid(parent)));
2171                         RETURN(-EPROTO);
2172                 }
2173
2174                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
2175                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
2176                        PNAME(lname), ldlm_rep);
2177
2178                 if (parent->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD ||
2179                     (fid_is_root(mdt_object_fid(parent)) &&
2180                      lname->ln_namelen == strlen(dot_fscrypt_name) &&
2181                      strncmp(lname->ln_name, dot_fscrypt_name,
2182                              lname->ln_namelen) == 0))
2183                         fscrypt_md = true;
2184         } else {
2185                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2186                 if (unlikely(reqbody == NULL))
2187                         RETURN(err_serious(-EPROTO));
2188
2189                 *child_fid = reqbody->mbo_fid2;
2190                 if (unlikely(!fid_is_sane(child_fid)))
2191                         RETURN(err_serious(-EINVAL));
2192
2193                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2194                         mdt_object_get(info->mti_env, parent);
2195                         child = parent;
2196                 } else {
2197                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2198                                                 child_fid);
2199                         if (IS_ERR(child))
2200                                 RETURN(PTR_ERR(child));
2201                 }
2202
2203                 if (mdt_object_remote(child)) {
2204                         CERROR("%s: child "DFID" is on remote target\n",
2205                                mdt_obd_name(info->mti_mdt),
2206                                PFID(mdt_object_fid(child)));
2207                         GOTO(out_child, rc = -EPROTO);
2208                 }
2209
2210                 /* don't fetch LOOKUP lock if it's remote object */
2211                 rc = mdt_is_remote_object(info, parent, child);
2212                 if (rc < 0)
2213                         GOTO(out_child, rc);
2214                 if (rc)
2215                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2216
2217                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2218                        "ldlm_rep = %p\n",
2219                        PFID(mdt_object_fid(parent)),
2220                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2221         }
2222
2223         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2224
2225         if (unlikely(!mdt_object_exists(parent)) &&
2226             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2227             lu_name_is_valid(lname)) {
2228                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2229                                 &parent->mot_obj,
2230                                 "Parent doesn't exist!");
2231                 GOTO(out_child, rc = -ESTALE);
2232         }
2233
2234         if (!child && is_resent) {
2235                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2236                 if (lock == NULL) {
2237                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2238                          * a resend case, however, it could be already destroyed
2239                          * due to client eviction or a raced cancel RPC.
2240                          */
2241                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2242                                           lhc->mlh_reg_lh.cookie);
2243                         RETURN(-ESTALE);
2244                 }
2245                 fid_extract_from_res_name(child_fid,
2246                                           &lock->l_resource->lr_name);
2247                 LDLM_LOCK_PUT(lock);
2248                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2249                                         child_fid);
2250                 if (IS_ERR(child))
2251                         RETURN(PTR_ERR(child));
2252         } else if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2253             lu_name_is_valid(lname)) {
2254                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2255                         rc = mdt_raw_lookup(info, parent, lname);
2256
2257                         RETURN(rc);
2258                 }
2259
2260                 /* step 1: lock parent only if parent is a directory */
2261                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2262                         lhp = &info->mti_lh[MDT_LH_PARENT];
2263                         rc = mdt_parent_lock(info, parent, lhp, lname, LCK_PR);
2264                         if (unlikely(rc != 0))
2265                                 RETURN(rc);
2266                 }
2267
2268                 /* step 2: lookup child's fid by name */
2269                 fid_zero(child_fid);
2270                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2271                                 child_fid, &info->mti_spec);
2272                 if (rc == -ENOENT)
2273                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2274
2275                 if (rc != 0)
2276                         GOTO(unlock_parent, rc);
2277
2278                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2279                                         child_fid);
2280                 if (unlikely(IS_ERR(child)))
2281                         GOTO(unlock_parent, rc = PTR_ERR(child));
2282         }
2283
2284         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2285
2286         /* step 3: lock child regardless if it is local or remote. */
2287         LASSERT(child);
2288
2289         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2290                 /* Here, lname is an encoded hash of on-disk name, and
2291                  * client is doing access without encryption key.
2292                  * So we need to compare name hash with the one in the request.
2293                  */
2294                 if (!find_name_matching_hash(info, lname, parent,
2295                                              child)) {
2296                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2297                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2298                         GOTO(out_child, rc = -ENOENT);
2299                 }
2300         }
2301
2302         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2303         if (!mdt_object_exists(child)) {
2304                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2305                                 &child->mot_obj,
2306                                 "Object doesn't exist!");
2307                 GOTO(out_child, rc = -ENOENT);
2308         }
2309
2310         rc = mdt_check_resent_lock(info, child, lhc);
2311         if (rc < 0) {
2312                 GOTO(out_child, rc);
2313         } else if (rc > 0) {
2314                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2315                     !mdt_object_remote(child)) {
2316                         struct md_attr *ma = &info->mti_attr;
2317
2318                         ma->ma_valid = 0;
2319                         ma->ma_need = MA_INODE;
2320                         rc = mdt_attr_get_complex(info, child, ma);
2321                         if (unlikely(rc != 0))
2322                                 GOTO(out_child, rc);
2323
2324                         /* If the file has not been changed for some time, we
2325                          * return not only a LOOKUP lock, but also an UPDATE
2326                          * lock and this might save us RPC on later STAT. For
2327                          * directories, it also let negative dentry cache start
2328                          * working for this dir. */
2329                         if (ma->ma_valid & MA_INODE &&
2330                             ma->ma_attr.la_valid & LA_CTIME &&
2331                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2332                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2333                                 child_bits |= MDS_INODELOCK_UPDATE;
2334                 }
2335
2336                 /* layout lock must be granted in a best-effort way
2337                  * for IT operations */
2338                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2339                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2340                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2341                         if (!CFS_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2342                             exp_connect_layout(info->mti_exp)) {
2343                                 /* try to grant layout lock for regular file. */
2344                                 try_bits = MDS_INODELOCK_LAYOUT;
2345                         }
2346                         /* Acquire DOM lock in advance for data-on-mdt file */
2347                         if (child != parent)
2348                                 try_bits |= MDS_INODELOCK_DOM;
2349                 }
2350
2351                 /*
2352                  * To avoid possible deadlock between batched statahead RPC
2353                  * and rename()/migrate() operation, it should use trylock to
2354                  * obtain the DLM PR ibits lock for file attributes in a
2355                  * batched statahead RPC. A failed trylock means that other
2356                  * users maybe modify the directory simultaneously as in current
2357                  * Lustre design the server only grants read lock to a client.
2358                  *
2359                  * When a trylock failed, the MDT reports the conflict with
2360                  * error code -EBUSY, and stops statahead immediately.
2361                  */
2362                 if (info->mti_batch_env) {
2363                         /*
2364                          * This is a sub stat-ahead request in a batched RPC.
2365                          * However, the @child is a remote object, we just
2366                          * return -EREMOTE here to forbid stat-ahead on it.
2367                          */
2368                         if (mdt_object_remote(child))
2369                                 GOTO(out_child, rc = -EREMOTE);
2370
2371                         try_bits |= child_bits;
2372                         child_bits = 0;
2373                 }
2374
2375                 if (try_bits != 0) {
2376                         /* try layout lock, it may fail to be granted due to
2377                          * contention at LOOKUP or UPDATE */
2378                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2379                                                  try_bits, LCK_PR);
2380                         if (child_bits & MDS_INODELOCK_LAYOUT)
2381                                 ma_need |= MA_LOV;
2382                 } else {
2383                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2384                          * client will enqueue the lock to the remote MDT */
2385                         if (mdt_object_remote(child))
2386                                 rc = mdt_object_lookup_lock(info, NULL, child,
2387                                                             lhc, LCK_PR);
2388                         else
2389                                 rc = mdt_object_lock(info, child, lhc,
2390                                                      child_bits, LCK_PR);
2391                 }
2392                 if (unlikely(rc != 0))
2393                         GOTO(out_child, rc);
2394                 if (info->mti_batch_env && child_bits == 0) {
2395                         if (!is_resent)
2396                                 mdt_object_unlock(info, child, lhc, 1);
2397                         GOTO(out_child, rc = -EBUSY);
2398                 }
2399         }
2400
2401         if (fscrypt_md)
2402                 child->mot_obj.lo_header->loh_attr |= LOHA_FSCRYPT_MD;
2403
2404         /* finally, we can get attr for child. */
2405         rc = mdt_getattr_internal(info, child, ma_need);
2406         if (unlikely(rc != 0)) {
2407                 if (!is_resent)
2408                         mdt_object_unlock(info, child, lhc, 1);
2409                 GOTO(out_child, rc);
2410         }
2411
2412         rc = mdt_pack_secctx_in_reply(info, child);
2413         if (unlikely(rc)) {
2414                 if (!is_resent)
2415                         mdt_object_unlock(info, child, lhc, 1);
2416                 GOTO(out_child, rc);
2417         }
2418
2419         rc = mdt_pack_encctx_in_reply(info, child);
2420         if (unlikely(rc)) {
2421                 if (!is_resent)
2422                         mdt_object_unlock(info, child, lhc, 1);
2423                 GOTO(out_child, rc);
2424         }
2425
2426         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2427         if (lock) {
2428                 /* Debugging code. */
2429                 LDLM_DEBUG(lock, "Returning lock to client");
2430                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2431                                          &lock->l_resource->lr_name),
2432                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2433                          PLDLMRES(lock->l_resource),
2434                          PFID(mdt_object_fid(child)));
2435
2436                 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2437                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2438                                 CFS_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2439                                                  req->rq_deadline -
2440                                                  req->rq_arrival_time.tv_sec +
2441                                                  cfs_fail_val ?: 3);
2442                         /* Put the lock to the waiting list and force the cancel */
2443                         ldlm_set_ast_sent(lock);
2444                 }
2445
2446                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2447                     !mdt_object_remote(child) && child != parent) {
2448                         mdt_object_put(info->mti_env, child);
2449                         rc = mdt_pack_size2body(info, child_fid,
2450                                                 &lhc->mlh_reg_lh);
2451                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2452                                 /* DOM lock was taken in advance but this is
2453                                  * not DoM file. Drop the lock.
2454                                  */
2455                                 lock_res_and_lock(lock);
2456                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2457                                 unlock_res_and_lock(lock);
2458                         }
2459                         LDLM_LOCK_PUT(lock);
2460                         GOTO(unlock_parent, rc = 0);
2461                 }
2462                 LDLM_LOCK_PUT(lock);
2463         }
2464
2465         EXIT;
2466 out_child:
2467         if (child)
2468                 mdt_object_put(info->mti_env, child);
2469 unlock_parent:
2470         if (lhp)
2471                 mdt_object_unlock(info, parent, lhp, 1);
2472         if (rc == -ENOENT) {
2473                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2474                  * client if trying to access an encrypted file
2475                  */
2476                 int rc2 = mdt_check_enc(info, parent);
2477
2478                 if (rc2)
2479                         rc = rc2;
2480         }
2481         return rc;
2482 }
2483
2484 /* normal handler: should release the child lock */
2485 static int mdt_getattr_name(struct tgt_session_info *tsi)
2486 {
2487         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2488         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2489         struct mdt_body *reqbody;
2490         struct mdt_body *repbody;
2491         int rc, rc2;
2492
2493         ENTRY;
2494
2495         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2496         LASSERT(reqbody != NULL);
2497         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2498         LASSERT(repbody != NULL);
2499
2500         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2501         repbody->mbo_eadatasize = 0;
2502         repbody->mbo_aclsize = 0;
2503
2504         rc = mdt_init_ucred(info, reqbody);
2505         if (unlikely(rc))
2506                 GOTO(out_shrink, rc);
2507
2508         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2509         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2510                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2511                 lhc->mlh_reg_lh.cookie = 0;
2512         }
2513         mdt_exit_ucred(info);
2514         EXIT;
2515 out_shrink:
2516         mdt_client_compatibility(info);
2517         rc2 = mdt_fix_reply(info);
2518         if (rc == 0)
2519                 rc = rc2;
2520         mdt_thread_info_fini(info);
2521         return rc;
2522 }
2523
2524 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2525                             const struct lu_fid *pfid,
2526                             const struct lu_name *name,
2527                             struct mdt_object *obj, s64 ctime)
2528 {
2529         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2530         struct ldlm_enqueue_info *einfo = &info->mti_einfo;
2531         struct mdt_device *mdt = info->mti_mdt;
2532         struct md_attr *ma = &info->mti_attr;
2533         struct mdt_lock_handle *parent_lh;
2534         struct mdt_lock_handle *child_lh;
2535         struct mdt_object *pobj;
2536         int rc;
2537         ENTRY;
2538
2539         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2540         if (IS_ERR(pobj))
2541                 GOTO(out, rc = PTR_ERR(pobj));
2542
2543         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2544         rc = mdt_parent_lock(info, pobj, parent_lh, name, LCK_PW);
2545         if (rc != 0)
2546                 GOTO(put_parent, rc);
2547
2548         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2549                         name, child_fid, &info->mti_spec);
2550         if (rc != 0)
2551                 GOTO(unlock_parent, rc);
2552
2553         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2554                 GOTO(unlock_parent, rc = -EREMCHG);
2555
2556         child_lh = &info->mti_lh[MDT_LH_CHILD];
2557         rc = mdt_object_stripes_lock(info, pobj, obj, child_lh, einfo,
2558                                      MDS_INODELOCK_LOOKUP |
2559                                      MDS_INODELOCK_UPDATE, LCK_EX);
2560         if (rc != 0)
2561                 GOTO(unlock_parent, rc);
2562
2563         if (atomic_read(&obj->mot_open_count)) {
2564                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2565                        PFID(mdt_object_fid(obj)));
2566                 GOTO(unlock_child, rc = -EBUSY);
2567         }
2568
2569         ma->ma_need = 0;
2570         ma->ma_valid = MA_INODE;
2571         ma->ma_attr.la_valid = LA_CTIME;
2572         ma->ma_attr.la_ctime = ctime;
2573
2574         mutex_lock(&obj->mot_lov_mutex);
2575
2576         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2577                         mdt_object_child(obj), name, ma, 0);
2578
2579         mutex_unlock(&obj->mot_lov_mutex);
2580
2581 unlock_child:
2582         mdt_object_stripes_unlock(info, obj, child_lh, einfo, 1);
2583 unlock_parent:
2584         mdt_object_unlock(info, pobj, parent_lh, 1);
2585 put_parent:
2586         mdt_object_put(info->mti_env, pobj);
2587 out:
2588         RETURN(rc);
2589 }
2590
2591 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2592                                         struct mdt_object *obj)
2593 {
2594         struct lu_ucred *uc = lu_ucred(info->mti_env);
2595         struct md_attr *ma = &info->mti_attr;
2596         struct lu_attr *la = &ma->ma_attr;
2597         int rc = 0;
2598         ENTRY;
2599
2600         ma->ma_need = MA_INODE;
2601         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2602         if (rc)
2603                 GOTO(out, rc);
2604
2605         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2606                         rc = -EACCES;
2607
2608         /* we want rbac roles to have precedence over any other
2609          * permission or capability checks
2610          */
2611         if (!uc->uc_rbac_byfid_ops)
2612                 RETURN(-EPERM);
2613         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2614                 RETURN(0);
2615         if (uc->uc_fsuid == la->la_uid) {
2616                 if ((la->la_mode & S_IWUSR) == 0)
2617                         rc = -EACCES;
2618         } else if (uc->uc_fsgid == la->la_gid) {
2619                 if ((la->la_mode & S_IWGRP) == 0)
2620                         rc = -EACCES;
2621         } else if ((la->la_mode & S_IWOTH) == 0) {
2622                         rc = -EACCES;
2623         }
2624
2625 out:
2626         RETURN(rc);
2627 }
2628
2629 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2630                          s64 ctime)
2631 {
2632         struct mdt_device *mdt = info->mti_mdt;
2633         struct mdt_object *obj = NULL;
2634         struct linkea_data ldata = { NULL };
2635         struct lu_buf *buf = &info->mti_big_buf;
2636         struct lu_name *name = &info->mti_name;
2637         struct lu_fid *pfid = &info->mti_tmp_fid1;
2638         struct link_ea_header *leh;
2639         struct link_ea_entry *lee;
2640         int reclen, count, rc = 0;
2641         ENTRY;
2642
2643         if (!fid_is_sane(fid))
2644                 GOTO(out, rc = -EINVAL);
2645
2646         if (!fid_is_namespace_visible(fid))
2647                 GOTO(out, rc = -EINVAL);
2648
2649         obj = mdt_object_find(info->mti_env, mdt, fid);
2650         if (IS_ERR(obj))
2651                 GOTO(out, rc = PTR_ERR(obj));
2652
2653         if (mdt_object_remote(obj))
2654                 GOTO(out, rc = -EREMOTE);
2655         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2656                 GOTO(out, rc = -ENOENT);
2657
2658         rc = mdt_rmfid_check_permission(info, obj);
2659         if (rc)
2660                 GOTO(out, rc);
2661
2662         /* take LinkEA */
2663         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2664         if (!buf->lb_buf)
2665                 GOTO(out, rc = -ENOMEM);
2666
2667         ldata.ld_buf = buf;
2668         rc = mdt_links_read(info, obj, &ldata);
2669         if (rc)
2670                 GOTO(out, rc);
2671
2672         leh = buf->lb_buf;
2673         lee = (struct link_ea_entry *)(leh + 1);
2674         for (count = 0; count < leh->leh_reccount; count++) {
2675                 /* remove every hardlink */
2676                 linkea_entry_unpack(lee, &reclen, name, pfid);
2677                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2678                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2679                 if (rc)
2680                         break;
2681         }
2682
2683 out:
2684         if (obj && !IS_ERR(obj))
2685                 mdt_object_put(info->mti_env, obj);
2686         if (info->mti_big_buf.lb_buf)
2687                 lu_buf_free(&info->mti_big_buf);
2688
2689         RETURN(rc);
2690 }
2691
2692 static int mdt_rmfid(struct tgt_session_info *tsi)
2693 {
2694         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2695         struct mdt_body *reqbody;
2696         struct lu_fid *fids, *rfids;
2697         int bufsize, rc;
2698         __u32 *rcs;
2699         int i, nr;
2700         ENTRY;
2701
2702         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2703         if (reqbody == NULL)
2704                 RETURN(-EPROTO);
2705         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2706                                        RCL_CLIENT);
2707         nr = bufsize / sizeof(struct lu_fid);
2708         if (nr * sizeof(struct lu_fid) != bufsize)
2709                 RETURN(-EINVAL);
2710         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2711                              RCL_SERVER, nr * sizeof(__u32));
2712         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2713                              RCL_SERVER, nr * sizeof(struct lu_fid));
2714         rc = req_capsule_server_pack(tsi->tsi_pill);
2715         if (rc)
2716                 GOTO(out, rc = err_serious(rc));
2717         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2718         if (fids == NULL)
2719                 RETURN(-EPROTO);
2720         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2721         LASSERT(rcs);
2722         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2723         LASSERT(rfids);
2724
2725         mdt_init_ucred(mti, reqbody);
2726         for (i = 0; i < nr; i++) {
2727                 rfids[i] = fids[i];
2728                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2729         }
2730         mdt_exit_ucred(mti);
2731
2732 out:
2733         RETURN(rc);
2734 }
2735
2736 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2737                          void *karg, void __user *uarg);
2738
2739 int mdt_io_set_info(struct tgt_session_info *tsi)
2740 {
2741         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2742         struct ost_body         *body = NULL, *repbody;
2743         void                    *key, *val = NULL;
2744         int                      keylen, vallen, rc = 0;
2745         bool                     is_grant_shrink;
2746
2747         ENTRY;
2748
2749         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2750         if (key == NULL) {
2751                 DEBUG_REQ(D_HA, req, "no set_info key");
2752                 RETURN(err_serious(-EFAULT));
2753         }
2754         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2755                                       RCL_CLIENT);
2756
2757         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2758         if (val == NULL) {
2759                 DEBUG_REQ(D_HA, req, "no set_info val");
2760                 RETURN(err_serious(-EFAULT));
2761         }
2762         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2763                                       RCL_CLIENT);
2764
2765         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2766         if (is_grant_shrink)
2767                 /* In this case the value is actually an RMF_OST_BODY, so we
2768                  * transmutate the type of this PTLRPC */
2769                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2770
2771         rc = req_capsule_server_pack(tsi->tsi_pill);
2772         if (rc < 0)
2773                 RETURN(rc);
2774
2775         if (is_grant_shrink) {
2776                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2777
2778                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2779                 *repbody = *body;
2780
2781                 /** handle grant shrink, similar to a read request */
2782                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2783                                        &repbody->oa);
2784         } else {
2785                 CERROR("%s: Unsupported key %s\n",
2786                        tgt_name(tsi->tsi_tgt), (char *)key);
2787                 rc = -EOPNOTSUPP;
2788         }
2789
2790         RETURN(rc);
2791 }
2792
2793
2794 static int mdt_set_info(struct tgt_session_info *tsi)
2795 {
2796         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2797         char                    *key;
2798         void                    *val;
2799         int                      keylen, vallen, rc = 0;
2800
2801         ENTRY;
2802
2803         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2804         if (key == NULL) {
2805                 DEBUG_REQ(D_HA, req, "no set_info key");
2806                 RETURN(err_serious(-EFAULT));
2807         }
2808
2809         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2810                                       RCL_CLIENT);
2811
2812         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2813         if (val == NULL) {
2814                 DEBUG_REQ(D_HA, req, "no set_info val");
2815                 RETURN(err_serious(-EFAULT));
2816         }
2817
2818         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2819                                       RCL_CLIENT);
2820
2821         /* Swab any part of val you need to here */
2822         if (KEY_IS(KEY_READ_ONLY)) {
2823                 spin_lock(&req->rq_export->exp_lock);
2824                 if (*(__u32 *)val)
2825                         *exp_connect_flags_ptr(req->rq_export) |=
2826                                 OBD_CONNECT_RDONLY;
2827                 else
2828                         *exp_connect_flags_ptr(req->rq_export) &=
2829                                 ~OBD_CONNECT_RDONLY;
2830                 spin_unlock(&req->rq_export->exp_lock);
2831         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2832                 struct changelog_setinfo *cs = val;
2833
2834                 if (vallen != sizeof(*cs)) {
2835                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2836                                tgt_name(tsi->tsi_tgt), vallen);
2837                         RETURN(-EINVAL);
2838                 }
2839                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2840                         __swab64s(&cs->cs_recno);
2841                         __swab32s(&cs->cs_id);
2842                 }
2843
2844                 if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
2845                         RETURN(-EACCES);
2846                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2847                                    vallen, val, NULL);
2848         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2849                 if (vallen > 0)
2850                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2851         } else {
2852                 RETURN(-EINVAL);
2853         }
2854         RETURN(rc);
2855 }
2856
2857 static int mdt_readpage(struct tgt_session_info *tsi)
2858 {
2859         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2860         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2861         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2862         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2863         struct mdt_body         *repbody;
2864         int                      rc;
2865         int                      i;
2866
2867         ENTRY;
2868
2869         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2870                 RETURN(err_serious(-ENOMEM));
2871
2872         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2873         if (repbody == NULL || reqbody == NULL)
2874                 RETURN(err_serious(-EFAULT));
2875
2876         /*
2877          * prepare @rdpg before calling lower layers and transfer itself. Here
2878          * reqbody->size contains offset of where to start to read and
2879          * reqbody->nlink contains number bytes to read.
2880          */
2881         rdpg->rp_hash = reqbody->mbo_size;
2882         if (rdpg->rp_hash != reqbody->mbo_size) {
2883                 CERROR("Invalid hash: %#llx != %#llx\n",
2884                        rdpg->rp_hash, reqbody->mbo_size);
2885                 RETURN(-EFAULT);
2886         }
2887
2888         rdpg->rp_attrs = reqbody->mbo_mode;
2889         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2890                 rdpg->rp_attrs |= LUDA_64BITHASH;
2891         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2892                                 exp_max_brw_size(tsi->tsi_exp));
2893         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2894                           PAGE_SHIFT;
2895         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2896         if (rdpg->rp_pages == NULL)
2897                 RETURN(-ENOMEM);
2898
2899         for (i = 0; i < rdpg->rp_npages; ++i) {
2900                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2901                 if (rdpg->rp_pages[i] == NULL)
2902                         GOTO(free_rdpg, rc = -ENOMEM);
2903         }
2904
2905         /* call lower layers to fill allocated pages with directory data */
2906         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2907         if (rc < 0)
2908                 GOTO(free_rdpg, rc);
2909
2910         /* send pages to client */
2911         rc = tgt_sendpage(tsi, rdpg, rc);
2912
2913         EXIT;
2914 free_rdpg:
2915
2916         for (i = 0; i < rdpg->rp_npages; i++)
2917                 if (rdpg->rp_pages[i] != NULL)
2918                         __free_page(rdpg->rp_pages[i]);
2919         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2920
2921         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2922                 RETURN(0);
2923
2924         return rc;
2925 }
2926
2927 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2928 {
2929         struct lu_ucred *uc = mdt_ucred_check(info);
2930         struct lu_attr *attr = &info->mti_attr.ma_attr;
2931
2932         if (uc == NULL)
2933                 return -EINVAL;
2934
2935         if (op != REINT_SETATTR) {
2936                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2937                         attr->la_uid = uc->uc_fsuid;
2938                 /* for S_ISGID, inherit gid from his parent, such work will be
2939                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2940                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2941                         attr->la_gid = uc->uc_fsgid;
2942         }
2943
2944         return 0;
2945 }
2946
2947 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2948 {
2949         return op == REINT_OPEN &&
2950              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2951 }
2952
2953 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2954 {
2955         struct req_capsule *pill = info->mti_pill;
2956
2957         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2958                                   RCL_SERVER) &&
2959             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2960                                   RCL_CLIENT)) {
2961                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2962                                          RCL_CLIENT) != 0)
2963                         /* pre-set size in server part with max size */
2964                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2965                                              RCL_SERVER,
2966                                              req_capsule_ptlreq(pill) ?
2967                                              OBD_MAX_DEFAULT_EA_SIZE :
2968                                              MAX_MD_SIZE_OLD);
2969                 else
2970                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2971                                              RCL_SERVER, 0);
2972         }
2973 }
2974
2975 static int mdt_reint_internal(struct mdt_thread_info *info,
2976                               struct mdt_lock_handle *lhc,
2977                               __u32 op)
2978 {
2979         struct req_capsule      *pill = info->mti_pill;
2980         struct mdt_body         *repbody;
2981         int                      rc = 0, rc2;
2982
2983         ENTRY;
2984
2985         rc = mdt_reint_unpack(info, op);
2986         if (rc != 0) {
2987                 CERROR("Can't unpack reint, rc %d\n", rc);
2988                 RETURN(err_serious(rc));
2989         }
2990
2991
2992         /* check if the file system is set to readonly. O_RDONLY open
2993          * is still allowed even the file system is set to readonly mode */
2994         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2995                 RETURN(err_serious(-EROFS));
2996
2997         /* for replay (no_create) lmm is not needed, client has it already */
2998         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2999                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3000                                      DEF_REP_MD_SIZE);
3001
3002         /* llog cookies are always 0, the field is kept for compatibility */
3003         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3004                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
3005
3006         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3007          * by default. If the target object has more ACL entries, then
3008          * enlarge the buffer when necessary. */
3009         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3010                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3011                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3012
3013         mdt_preset_secctx_size(info);
3014         mdt_preset_encctx_size(info);
3015
3016         rc = req_capsule_server_pack(pill);
3017         if (rc != 0) {
3018                 CERROR("Can't pack response, rc %d\n", rc);
3019                 RETURN(err_serious(rc));
3020         }
3021
3022         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
3023                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
3024                 LASSERT(repbody);
3025                 repbody->mbo_eadatasize = 0;
3026                 repbody->mbo_aclsize = 0;
3027         }
3028
3029         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
3030
3031         /* for replay no cookkie / lmm need, because client have this already */
3032         if (info->mti_spec.no_create)
3033                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3034                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
3035
3036         rc = mdt_init_ucred_reint(info);
3037         if (rc)
3038                 GOTO(out_shrink, rc);
3039
3040         rc = mdt_fix_attr_ucred(info, op);
3041         if (rc != 0)
3042                 GOTO(out_ucred, rc = err_serious(rc));
3043
3044         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
3045         if (rc < 0) {
3046                 GOTO(out_ucred, rc);
3047         } else if (rc == 1) {
3048                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
3049                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
3050                 GOTO(out_ucred, rc);
3051         }
3052         rc = mdt_reint_rec(info, lhc);
3053         EXIT;
3054 out_ucred:
3055         mdt_exit_ucred(info);
3056 out_shrink:
3057         mdt_client_compatibility(info);
3058
3059         rc2 = mdt_fix_reply(info);
3060         if (rc == 0)
3061                 rc = rc2;
3062
3063         /*
3064          * Data-on-MDT optimization - read data along with OPEN and return it
3065          * in reply when possible.
3066          */
3067         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
3068                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
3069                                           &lhc->mlh_reg_lh);
3070
3071         return rc;
3072 }
3073
3074 static long mdt_reint_opcode(struct ptlrpc_request *req,
3075                              const struct req_format **fmt)
3076 {
3077         struct mdt_device       *mdt;
3078         struct mdt_rec_reint    *rec;
3079         long                     opc;
3080
3081         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3082         if (rec != NULL) {
3083                 opc = rec->rr_opcode;
3084                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3085                 if (opc < REINT_MAX && fmt[opc] != NULL)
3086                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3087                 else {
3088                         mdt = mdt_exp2dev(req->rq_export);
3089                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3090                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3091                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3092                         opc = err_serious(-EFAULT);
3093                 }
3094         } else {
3095                 opc = err_serious(-EFAULT);
3096         }
3097         return opc;
3098 }
3099
3100 static int mdt_reint(struct tgt_session_info *tsi)
3101 {
3102         long opc;
3103         int  rc;
3104         static const struct req_format *reint_fmts[REINT_MAX] = {
3105                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3106                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3107                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3108                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3109                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3110                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3111                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3112                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3113                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3114                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3115         };
3116
3117         ENTRY;
3118
3119         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3120         if (opc >= 0) {
3121                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3122                 /*
3123                  * No lock possible here from client to pass it to reint code
3124                  * path.
3125                  */
3126                 rc = mdt_reint_internal(info, NULL, opc);
3127                 mdt_thread_info_fini(info);
3128         } else {
3129                 rc = opc;
3130         }
3131
3132         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3133         RETURN(rc);
3134 }
3135
3136 /* this should sync the whole device */
3137 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3138 {
3139         struct dt_device *dt = mdt->mdt_bottom;
3140         int rc;
3141         ENTRY;
3142
3143         rc = dt->dd_ops->dt_sync(env, dt);
3144         RETURN(rc);
3145 }
3146
3147 /* this should sync this object */
3148 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3149                            struct mdt_object *mo)
3150 {
3151         int rc = 0;
3152
3153         ENTRY;
3154
3155         if (!mdt_object_exists(mo)) {
3156                 CWARN("%s: non existing object "DFID": rc = %d\n",
3157                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3158                       -ESTALE);
3159                 RETURN(-ESTALE);
3160         }
3161
3162         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3163                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3164                 dt_obj_version_t version;
3165
3166                 version = dt_version_get(env, mdt_obj2dt(mo));
3167                 if (version > tgt->lut_obd->obd_last_committed)
3168                         rc = mo_object_sync(env, mdt_object_child(mo));
3169         } else {
3170                 rc = mo_object_sync(env, mdt_object_child(mo));
3171         }
3172
3173         RETURN(rc);
3174 }
3175
3176 static int mdt_sync(struct tgt_session_info *tsi)
3177 {
3178         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3179         struct req_capsule      *pill = tsi->tsi_pill;
3180         struct mdt_body         *body;
3181         ktime_t                  kstart = ktime_get();
3182         int                      rc;
3183
3184         ENTRY;
3185
3186         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3187                 RETURN(err_serious(-ENOMEM));
3188
3189         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3190                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3191         } else {
3192                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3193
3194                 if (unlikely(info->mti_object == NULL))
3195                         RETURN(-EPROTO);
3196
3197                 /* sync an object */
3198                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3199                                      info->mti_object);
3200                 if (rc == 0) {
3201                         const struct lu_fid *fid;
3202                         struct lu_attr *la = &info->mti_attr.ma_attr;
3203
3204                         info->mti_attr.ma_need = MA_INODE;
3205                         info->mti_attr.ma_valid = 0;
3206                         rc = mdt_attr_get_complex(info, info->mti_object,
3207                                                   &info->mti_attr);
3208                         if (rc == 0) {
3209                                 body = req_capsule_server_get(pill,
3210                                                               &RMF_MDT_BODY);
3211                                 fid = mdt_object_fid(info->mti_object);
3212                                 mdt_pack_attr2body(info, body, la, fid);
3213                         }
3214                 }
3215                 mdt_thread_info_fini(info);
3216         }
3217         if (rc == 0)
3218                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3219                                  ktime_us_delta(ktime_get(), kstart));
3220
3221         RETURN(rc);
3222 }
3223
3224 static int mdt_data_sync(struct tgt_session_info *tsi)
3225 {
3226         struct mdt_thread_info *info;
3227         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3228         struct ost_body *body = tsi->tsi_ost_body;
3229         struct ost_body *repbody;
3230         struct mdt_object *mo = NULL;
3231         struct md_attr *ma;
3232         int rc = 0;
3233
3234         ENTRY;
3235
3236         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3237
3238         /* if no fid is specified then do nothing,
3239          * device sync is done via MDS_SYNC */
3240         if (fid_is_zero(&tsi->tsi_fid))
3241                 RETURN(0);
3242
3243         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3244         if (IS_ERR(mo))
3245                 RETURN(PTR_ERR(mo));
3246
3247         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3248         if (rc)
3249                 GOTO(put, rc);
3250
3251         repbody->oa.o_oi = body->oa.o_oi;
3252         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3253
3254         info = tsi2mdt_info(tsi);
3255         ma = &info->mti_attr;
3256         ma->ma_need = MA_INODE;
3257         ma->ma_valid = 0;
3258         rc = mdt_attr_get_complex(info, mo, ma);
3259         if (rc == 0)
3260                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3261         else
3262                 rc = 0;
3263         mdt_thread_info_fini(info);
3264
3265         EXIT;
3266 put:
3267         if (mo != NULL)
3268                 mdt_object_put(tsi->tsi_env, mo);
3269         return rc;
3270 }
3271
3272 /*
3273  * Handle quota control requests to consult current usage/limit, but also
3274  * to configure quota enforcement
3275  */
3276 static int mdt_quotactl(struct tgt_session_info *tsi)
3277 {
3278         struct obd_export *exp  = tsi->tsi_exp;
3279         struct req_capsule *pill = tsi->tsi_pill;
3280         struct obd_quotactl *oqctl, *repoqc;
3281         int id, rc;
3282         struct mdt_device *mdt = mdt_exp2dev(exp);
3283         struct lu_device *qmt = mdt->mdt_qmt_dev;
3284         struct lu_nodemap *nodemap;
3285         ENTRY;
3286
3287         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3288         if (!oqctl)
3289                 RETURN(err_serious(-EPROTO));
3290
3291         rc = req_capsule_server_pack(pill);
3292         if (rc)
3293                 RETURN(err_serious(rc));
3294
3295         nodemap = nodemap_get_from_exp(exp);
3296         if (IS_ERR(nodemap))
3297                 RETURN(PTR_ERR(nodemap));
3298
3299         switch (oqctl->qc_cmd) {
3300                 /* master quotactl */
3301         case Q_SETINFO:
3302         case Q_SETQUOTA:
3303         case LUSTRE_Q_SETDEFAULT:
3304         case LUSTRE_Q_SETQUOTAPOOL:
3305         case LUSTRE_Q_SETINFOPOOL:
3306         case LUSTRE_Q_SETDEFAULT_POOL:
3307         case LUSTRE_Q_DELETEQID:
3308         case LUSTRE_Q_RESETQID:
3309                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3310                                           oqctl->qc_id))
3311                         GOTO(out_nodemap, rc = -EPERM);
3312                 fallthrough;
3313         case Q_GETINFO:
3314         case Q_GETQUOTA:
3315         case LUSTRE_Q_GETDEFAULT:
3316         case LUSTRE_Q_GETQUOTAPOOL:
3317         case LUSTRE_Q_GETINFOPOOL:
3318         case LUSTRE_Q_GETDEFAULT_POOL:
3319                 if (qmt == NULL)
3320                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3321                 /* slave quotactl */
3322                 fallthrough;
3323         case Q_GETOINFO:
3324         case Q_GETOQUOTA:
3325                 break;
3326         default:
3327                 rc = -EFAULT;
3328                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3329                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3330                 GOTO(out_nodemap, rc);
3331         }
3332
3333         id = oqctl->qc_id;
3334         switch (oqctl->qc_type) {
3335         case USRQUOTA:
3336                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3337                                     NODEMAP_CLIENT_TO_FS, id);
3338                 break;
3339         case GRPQUOTA:
3340                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3341                                     NODEMAP_CLIENT_TO_FS, id);
3342                 break;
3343         case PRJQUOTA:
3344                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3345                                     NODEMAP_CLIENT_TO_FS, id);
3346                 break;
3347         default:
3348                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3349         }
3350         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3351         if (repoqc == NULL)
3352                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3353
3354         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3355                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3356
3357         if (oqctl->qc_id != id)
3358                 swap(oqctl->qc_id, id);
3359
3360         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3361                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3362                         RETURN(-EINPROGRESS);
3363         }
3364
3365         switch (oqctl->qc_cmd) {
3366
3367         case Q_GETINFO:
3368         case Q_SETINFO:
3369         case Q_SETQUOTA:
3370         case Q_GETQUOTA:
3371         case LUSTRE_Q_SETDEFAULT:
3372         case LUSTRE_Q_GETDEFAULT:
3373         case LUSTRE_Q_SETQUOTAPOOL:
3374         case LUSTRE_Q_GETQUOTAPOOL:
3375         case LUSTRE_Q_SETINFOPOOL:
3376         case LUSTRE_Q_GETINFOPOOL:
3377         case LUSTRE_Q_SETDEFAULT_POOL:
3378         case LUSTRE_Q_GETDEFAULT_POOL:
3379         case LUSTRE_Q_DELETEQID:
3380         case LUSTRE_Q_RESETQID:
3381                 /* forward quotactl request to QMT */
3382                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3383                 break;
3384
3385         case Q_GETOINFO:
3386         case Q_GETOQUOTA:
3387                 /* slave quotactl */
3388                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3389                                    oqctl);
3390                 break;
3391
3392         default:
3393                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3394                 GOTO(out_nodemap, rc = -EFAULT);
3395         }
3396
3397         if (oqctl->qc_id != id)
3398                 swap(oqctl->qc_id, id);
3399
3400         QCTL_COPY_NO_PNAME(repoqc, oqctl);
3401         EXIT;
3402
3403 out_nodemap:
3404         nodemap_putref(nodemap);
3405
3406         return rc;
3407 }
3408
3409 /** clone llog ctxt from child (mdd)
3410  * This allows remote llog (replicator) access.
3411  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3412  * context was originally set up, or we can handle them directly.
3413  * I choose the latter, but that means I need any llog
3414  * contexts set up by child to be accessable by the mdt.  So we clone the
3415  * context into our context list here.
3416  */
3417 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3418                                int idx)
3419 {
3420         struct md_device  *next = mdt->mdt_child;
3421         struct llog_ctxt *ctxt;
3422         int rc;
3423
3424         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3425                 return 0;
3426
3427         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3428         if (rc || ctxt == NULL) {
3429                 return 0;
3430         }
3431
3432         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3433         if (rc)
3434                 CERROR("Can't set mdt ctxt %d\n", rc);
3435
3436         return rc;
3437 }
3438
3439 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3440                                  struct mdt_device *mdt, int idx)
3441 {
3442         struct llog_ctxt *ctxt;
3443
3444         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3445         if (ctxt == NULL)
3446                 return 0;
3447         /* Put once for the get we just did, and once for the clone */
3448         llog_ctxt_put(ctxt);
3449         llog_ctxt_put(ctxt);
3450         return 0;
3451 }
3452
3453 /*
3454  * sec context handlers
3455  */
3456 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3457 {
3458         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3459
3460         return 0;
3461 }
3462
3463 /*
3464  * quota request handlers
3465  */
3466 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3467 {
3468         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3469         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3470         int                      rc;
3471         ENTRY;
3472
3473         if (qmt == NULL)
3474                 RETURN(err_serious(-EOPNOTSUPP));
3475
3476         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3477         RETURN(rc);
3478 }
3479
3480 struct mdt_object *mdt_object_new(const struct lu_env *env,
3481                                   struct mdt_device *d,
3482                                   const struct lu_fid *f)
3483 {
3484         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3485         struct lu_object *o;
3486         struct mdt_object *m;
3487         ENTRY;
3488
3489         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3490         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3491         if (unlikely(IS_ERR(o)))
3492                 m = (struct mdt_object *)o;
3493         else
3494                 m = mdt_obj(o);
3495         RETURN(m);
3496 }
3497
3498 struct mdt_object *mdt_object_find(const struct lu_env *env,
3499                                    struct mdt_device *d,
3500                                    const struct lu_fid *f)
3501 {
3502         struct lu_object *o;
3503         struct mdt_object *m;
3504         ENTRY;
3505
3506         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3507         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3508         if (unlikely(IS_ERR(o)))
3509                 m = (struct mdt_object *)o;
3510         else
3511                 m = mdt_obj(o);
3512
3513         RETURN(m);
3514 }
3515
3516 /**
3517  * Asyncronous commit for mdt device.
3518  *
3519  * Pass asynchonous commit call down the MDS stack.
3520  *
3521  * \param env environment
3522  * \param mdt the mdt device
3523  */
3524 static void mdt_device_commit_async(const struct lu_env *env,
3525                                     struct mdt_device *mdt)
3526 {
3527         struct dt_device *dt = mdt->mdt_bottom;
3528         int rc;
3529         ENTRY;
3530
3531         rc = dt->dd_ops->dt_commit_async(env, dt);
3532         if (unlikely(rc != 0))
3533                 CWARN("%s: async commit start failed: rc = %d\n",
3534                       mdt_obd_name(mdt), rc);
3535         atomic_inc(&mdt->mdt_async_commit_count);
3536         EXIT;
3537 }
3538
3539 /**
3540  * Mark the lock as "synchonous".
3541  *
3542  * Mark the lock to deffer transaction commit to the unlock time.
3543  *
3544  * \param lock the lock to mark as "synchonous"
3545  *
3546  * \see mdt_is_lock_sync
3547  * \see mdt_save_lock
3548  */
3549 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3550 {
3551         lock->l_ast_data = (void*)1;
3552 }
3553
3554 /**
3555  * Check whehter the lock "synchonous" or not.
3556  *
3557  * \param lock the lock to check
3558  * \retval 1 the lock is "synchonous"
3559  * \retval 0 the lock isn't "synchronous"
3560  *
3561  * \see mdt_set_lock_sync
3562  * \see mdt_save_lock
3563  */
3564 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3565 {
3566         return lock->l_ast_data != NULL;
3567 }
3568
3569 /**
3570  * Blocking AST for mdt locks.
3571  *
3572  * Starts transaction commit if in case of COS lock conflict or
3573  * deffers such a commit to the mdt_save_lock.
3574  *
3575  * \param lock the lock which blocks a request or cancelling lock
3576  * \param desc unused
3577  * \param data unused
3578  * \param flag indicates whether this cancelling or blocking callback
3579  * \retval 0
3580  * \see ldlm_blocking_ast_nocheck
3581  */
3582 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3583                      void *data, int flag)
3584 {
3585         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3586         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3587         struct ldlm_cb_set_arg *arg = data;
3588         bool commit_async = false;
3589         int rc;
3590
3591         ENTRY;
3592         if (flag == LDLM_CB_CANCELING)
3593                 RETURN(0);
3594
3595         lock_res_and_lock(lock);
3596         if (lock->l_blocking_ast != mdt_blocking_ast) {
3597                 unlock_res_and_lock(lock);
3598                 RETURN(0);
3599         }
3600
3601         /* A blocking ast may be sent from ldlm_lock_decref_internal
3602          * when the last reference to a local lock was released and
3603          * during blocking event from ldlm_work_bl_ast_lock().
3604          * The 'data' parameter is l_ast_data in the first case and
3605          * callback arguments in the second one. Distinguish them by that.
3606          */
3607         if (data && data != lock->l_ast_data && arg->bl_desc) {
3608                 if (lock->l_req_mode & (LCK_COS | LCK_TXN))
3609                         commit_async = true;
3610                 else if ((lock->l_req_mode & (LCK_PW | LCK_EX)) &&
3611                          ((mdt_cos_is_enabled(mdt) &&
3612                              !arg->bl_desc->bl_same_client) ||
3613                           (mdt_slc_is_enabled(mdt) &&
3614                              arg->bl_desc->bl_txn_dependent)))
3615                         mdt_set_lock_sync(lock);
3616         }
3617
3618         rc = ldlm_blocking_ast_nocheck(lock);
3619
3620         if (commit_async) {
3621                 struct lu_env env;
3622
3623                 rc = lu_env_init(&env, LCT_LOCAL);
3624                 if (unlikely(rc != 0))
3625                         CWARN("%s: lu_env initialization failed, cannot "
3626                               "start asynchronous commit: rc = %d\n",
3627                               obd->obd_name, rc);
3628                 else
3629                         mdt_device_commit_async(&env, mdt);
3630                 lu_env_fini(&env);
3631         }
3632         RETURN(rc);
3633 }
3634
3635 /*
3636  * Blocking AST for cross-MDT lock
3637  *
3638  * Discard lock from uncommitted_slc_locks and cancel it.
3639  *
3640  * \param lock  the lock which blocks a request or cancelling lock
3641  * \param desc  unused
3642  * \param data  unused
3643  * \param flag  indicates whether this cancelling or blocking callback
3644  * \retval      0 on success
3645  * \retval      negative number on error
3646  */
3647 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3648                             void *data, int flag)
3649 {
3650         int rc = 0;
3651         ENTRY;
3652
3653         switch (flag) {
3654         case LDLM_CB_BLOCKING: {
3655                 struct lustre_handle lockh;
3656
3657                 ldlm_lock2handle(lock, &lockh);
3658                 rc = ldlm_cli_cancel(&lockh,
3659                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3660                 if (rc < 0) {
3661                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3662                         RETURN(rc);
3663                 }
3664                 break;
3665         }
3666         case LDLM_CB_CANCELING: {
3667                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3668                 struct mdt_device *mdt =
3669                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3670
3671                 LDLM_DEBUG(lock, "Revoke remote lock");
3672
3673                 /* discard slc lock here so that it can be cleaned anytime,
3674                  * especially for cleanup_resource() */
3675                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3676
3677                 /* once we cache lock, l_ast_data is set to mdt_object */
3678                 if (lock->l_ast_data != NULL) {
3679                         struct mdt_object *mo = lock->l_ast_data;
3680                         struct lu_env env;
3681
3682                         rc = lu_env_init(&env, LCT_MD_THREAD);
3683                         if (unlikely(rc != 0)) {
3684                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3685                                       obd->obd_name, mo,
3686                                       PFID(mdt_object_fid(mo)), rc);
3687                                 RETURN(rc);
3688                         }
3689
3690                         if (lock->l_policy_data.l_inodebits.bits &
3691                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3692                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3693                                 mo->mot_cache_attr = 0;
3694                         }
3695                         mdt_object_put(&env, mo);
3696                         lu_env_fini(&env);
3697                 }
3698                 break;
3699         }
3700         default:
3701                 LBUG();
3702         }
3703
3704         RETURN(rc);
3705 }
3706
3707 int mdt_check_resent_lock(struct mdt_thread_info *info,
3708                           struct mdt_object *mo,
3709                           struct mdt_lock_handle *lhc)
3710 {
3711         /* the lock might already be gotten in ldlm_handle_enqueue() */
3712         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3713                 struct ptlrpc_request *req = mdt_info_req(info);
3714                 struct ldlm_lock      *lock;
3715
3716                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3717                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3718                 if (lock == NULL) {
3719                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3720                          * a resend case, however, it could be already destroyed
3721                          * due to client eviction or a raced cancel RPC. */
3722                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3723                                           lhc->mlh_reg_lh.cookie);
3724                         RETURN(-ESTALE);
3725                 }
3726
3727                 if (!fid_res_name_eq(mdt_object_fid(mo),
3728                                      &lock->l_resource->lr_name)) {
3729                         CWARN("%s: Although resent, but still not "
3730                               "get child lock:"DFID"\n",
3731                               info->mti_exp->exp_obd->obd_name,
3732                               PFID(mdt_object_fid(mo)));
3733                         LDLM_LOCK_PUT(lock);
3734                         RETURN(-EPROTO);
3735                 }
3736                 LDLM_LOCK_PUT(lock);
3737                 return 0;
3738         }
3739         return 1;
3740 }
3741
3742 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3743 {
3744         mdt_object_get(NULL, lock->l_ast_data);
3745 }
3746
3747 static int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3748                                       struct mdt_object *obj,
3749                                       struct lustre_handle *lh,
3750                                       enum ldlm_mode mode,
3751                                       union ldlm_policy_data *policy,
3752                                       struct ldlm_res_id *res_id,
3753                                       bool cache)
3754 {
3755         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3756         int rc;
3757
3758         LASSERT(mdt_object_remote(obj));
3759
3760         memset(einfo, 0, sizeof(*einfo));
3761         einfo->ei_type = LDLM_IBITS;
3762         einfo->ei_mode = mode;
3763         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3764         einfo->ei_cb_cp = ldlm_completion_ast;
3765         einfo->ei_enq_slave = 0;
3766         einfo->ei_res_id = res_id;
3767         einfo->ei_req_slot = 1;
3768         if (cache) {
3769                 /*
3770                  * if we cache lock, couple lock with mdt_object, so that object
3771                  * can be easily found in lock ASTs.
3772                  */
3773                 einfo->ei_cbdata = obj;
3774                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3775         }
3776
3777         rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), lh, einfo,
3778                             policy);
3779         if (rc) {
3780                 lh->cookie = 0ull;
3781                 return rc;
3782         }
3783
3784         /* other components like LFSCK can use lockless access
3785          * and populate cache, so we better invalidate it
3786          */
3787         if (policy->l_inodebits.bits &
3788             (MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR))
3789                 mo_invalidate(mti->mti_env, mdt_object_child(obj));
3790
3791         return 0;
3792 }
3793
3794 /*
3795  * Helper function to take PDO and hash lock.
3796  *
3797  * if \a pdo_lock is false, don't take PDO lock, this is case in rename.
3798  */
3799 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3800                         struct mdt_lock_handle *lh, const struct lu_name *name,
3801                         enum ldlm_mode mode, bool pdo_lock)
3802 {
3803         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3804         union ldlm_policy_data *policy = &info->mti_policy;
3805         struct ldlm_res_id *res_id = &info->mti_res_id;
3806         /*
3807          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it is never going to
3808          * be sent to client and we do not want it slowed down due to possible
3809          * cancels.
3810          */
3811         __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3812         __u64 *cookie = NULL;
3813         int rc;
3814
3815         LASSERT(obj);
3816
3817         /* check for exists after object is locked */
3818         if (!mdt_object_exists(obj))
3819                 /* Non-existent object shouldn't have PDO lock */
3820                 return -ESTALE;
3821
3822         /* Non-dir object shouldn't have PDO lock */
3823         if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
3824                 return -ENOTDIR;
3825
3826         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
3827         policy->l_inodebits.try_bits = 0;
3828         policy->l_inodebits.li_gid = 0;
3829         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3830         fid_build_reg_res_name(mdt_object_fid(obj), res_id);
3831         if (info->mti_exp)
3832                 cookie = &info->mti_exp->exp_handle.h_cookie;
3833
3834         mdt_lock_pdo_init(lh, mode, name);
3835         mdt_lock_pdo_mode(info, obj, lh);
3836         if (lh->mlh_pdo_mode != LCK_NL) {
3837                 if (pdo_lock) {
3838                         if (mdt_object_remote(obj)) {
3839                                 rc = mdt_remote_object_lock_try(info, obj,
3840                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3841                                         policy, res_id, false);
3842                                 lh->mlh_pdo_remote = 1;
3843                         } else {
3844                                 rc = mdt_fid_lock(info->mti_env, ns,
3845                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3846                                         policy, res_id, dlmflags, cookie);
3847                         }
3848                         if (rc) {
3849                                 mdt_object_unlock(info, obj, lh, 1);
3850                                 return rc;
3851                         }
3852                 }
3853                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3854         }
3855
3856         if (mdt_object_remote(obj))
3857                 rc = mdt_remote_object_lock_try(info, obj, &lh->mlh_rreg_lh,
3858                         lh->mlh_rreg_mode, policy, res_id, false);
3859         else
3860                 rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh,
3861                         lh->mlh_reg_mode, policy, res_id, dlmflags, cookie);
3862         if (rc)
3863                 mdt_object_unlock(info, obj, lh, 1);
3864         else if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK) &&
3865                    lh->mlh_pdo_hash != 0 &&
3866                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3867                 CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3868
3869         return rc;
3870 }
3871
3872 int mdt_object_lock_internal(struct mdt_thread_info *info,
3873                              struct mdt_object *obj, const struct lu_fid *fid,
3874                              struct mdt_lock_handle *lh, __u64 *ibits,
3875                              __u64 trybits, bool cache)
3876 {
3877         union ldlm_policy_data *policy = &info->mti_policy;
3878         struct ldlm_res_id *res_id = &info->mti_res_id;
3879         struct lustre_handle *handle;
3880         int rc;
3881
3882         policy->l_inodebits.bits = *ibits;
3883         policy->l_inodebits.try_bits = trybits;
3884         policy->l_inodebits.li_gid = lh->mlh_gid;
3885         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3886         fid_build_reg_res_name(fid, res_id);
3887
3888         if (obj && mdt_object_remote(obj)) {
3889                 handle = &lh->mlh_rreg_lh;
3890                 LASSERT(!lustre_handle_is_used(handle));
3891                 LASSERT(lh->mlh_rreg_mode != LCK_MINMODE);
3892                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3893                 rc = mdt_remote_object_lock_try(info, obj, handle,
3894                                                 lh->mlh_rreg_mode, policy,
3895                                                 res_id, cache);
3896         } else {
3897                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3898                 /*
3899                  * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if
3900                  * it is going to be sent to client. If it is -
3901                  * mdt_intent_policy() path will fix it up and turn FL_LOCAL
3902                  * flag off.
3903                  */
3904                 __u64 dlmflags = LDLM_FL_ATOMIC_CB | LDLM_FL_LOCAL_ONLY;
3905                 __u64 *cookie = NULL;
3906
3907                 handle = &lh->mlh_reg_lh;
3908                 LASSERT(!lustre_handle_is_used(handle));
3909                 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3910                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3911
3912                 /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3913                 if (lh->mlh_type == MDT_REG_LOCK &&
3914                     lh->mlh_reg_mode == LCK_EX && *ibits == MDS_INODELOCK_OPEN)
3915                         dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3916
3917
3918                 if (info->mti_exp)
3919                         cookie = &info->mti_exp->exp_handle.h_cookie;
3920
3921                 rc = mdt_fid_lock(info->mti_env, ns, handle, lh->mlh_reg_mode,
3922                                   policy, res_id, dlmflags, cookie);
3923                 if (rc)
3924                         mdt_object_unlock(info, obj, lh, 1);
3925         }
3926
3927         if (rc == 0) {
3928                 struct ldlm_lock *lock;
3929
3930                 /* Return successfully acquired bits to a caller */
3931                 lock = ldlm_handle2lock(handle);
3932                 LASSERT(lock);
3933                 *ibits = lock->l_policy_data.l_inodebits.bits;
3934                 LDLM_LOCK_PUT(lock);
3935         }
3936
3937         return rc;
3938 }
3939
3940 /*
3941  * MDT object locking functions:
3942  * mdt_object_lock(): lock object, this is used in most places, and normally
3943  *      lock ibits doesn't contain LOOKUP, unless the caller knows it's not
3944  *      remote object.
3945  * mdt_object_check_lock(): lock object with LOOKUP and other ibits, it needs
3946  *      to check whether parent is on remote MDT, if so, take LOOKUP on parent
3947  *      MDT separately, and then lock other ibits on child object.
3948  * mdt_parent_lock(): take parent UPDATE lock with specific mode, if parent is
3949  *      local, take PDO lock by name hash, otherwise take regular lock.
3950  * mdt_object_stripes_lock(): lock object which should be local, and if it's a
3951  *      striped directory, lock its stripes, this is called in operations which
3952  *      modify both object and stripes.
3953  * mdt_object_lock_try(): lock object with trybits, the trybits contains
3954  *      optional inode lock bits that can be granted. This is called by
3955  *      getattr/open to fetch more inode lock bits to client, and is also called
3956  *      by dir migration to lock link parent in non-block mode to avoid
3957  *      deadlock.
3958  */
3959
3960 /**
3961  * lock object
3962  *
3963  * this is used to lock object in most places, and normally lock ibits doesn't
3964  * contain LOOKUP, unless the caller knows it's not remote object.
3965  *
3966  * \param info          struct mdt_thread_info
3967  * \param obj           object
3968  * \param lh            lock handle
3969  * \param ibits         MDS inode lock bits
3970  * \param mode          lock mode
3971  *
3972  * \retval              0 on success, -ev on error.
3973  */
3974 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3975                     struct mdt_lock_handle *lh, __u64 ibits,
3976                     enum ldlm_mode mode)
3977 {
3978         int rc;
3979
3980         ENTRY;
3981         mdt_lock_reg_init(lh, mode);
3982         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh,
3983                                       &ibits, 0, false);
3984         RETURN(rc);
3985 }
3986
3987 /**
3988  * lock object with LOOKUP and other ibits
3989  *
3990  * it will check whether parent and child are on different MDTs, if so, take
3991  * LOOKUP lock on parent MDT, and lock other ibits on child MDT, otherwise lock
3992  * all ibits on child MDT. Note, parent and child shouldn't be both on remote
3993  * MDTs, in which case specific lock function should be used, and it's in
3994  * rename and migrate only.
3995  *
3996  * \param info          struct mdt_thread_info
3997  * \param parent        parent object
3998  * \param child         child object
3999  * \param lh            lock handle
4000  * \param ibits         MDS inode lock bits
4001  * \param mode          lock mode
4002  *
4003  * \retval              0 on success, -ev on error.
4004  */
4005 int mdt_object_check_lock(struct mdt_thread_info *info,
4006                           struct mdt_object *parent, struct mdt_object *child,
4007                           struct mdt_lock_handle *lh, __u64 ibits,
4008                           enum ldlm_mode mode)
4009 {
4010         int rc;
4011
4012         ENTRY;
4013         /* if LOOKUP ibit is not set, use mdt_object_lock() */
4014         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
4015         /* if only LOOKUP ibit is needed, use mdt_object_lookup_lock() */
4016         LASSERT(ibits != MDS_INODELOCK_LOOKUP);
4017         LASSERT(parent);
4018         /* @parent and @child shouldn't both be on remote MDTs */
4019         LASSERT(!(mdt_object_remote(parent) && mdt_object_remote(child)));
4020
4021         mdt_lock_reg_init(lh, mode);
4022         if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
4023                 __u64 lookup_ibits = MDS_INODELOCK_LOOKUP;
4024
4025                 rc = mdt_object_lock_internal(info, parent,
4026                                               mdt_object_fid(child), lh,
4027                                               &lookup_ibits, 0, false);
4028                 if (rc)
4029                         RETURN(rc);
4030
4031                 ibits &= ~MDS_INODELOCK_LOOKUP;
4032         }
4033
4034         rc = mdt_object_lock_internal(info, child, mdt_object_fid(child), lh,
4035                                       &ibits, 0, false);
4036         if (rc && !(ibits & MDS_INODELOCK_LOOKUP))
4037                 mdt_object_unlock(info, NULL, lh, 1);
4038
4039         RETURN(rc);
4040 }
4041
4042 /**
4043  * take parent UPDATE lock
4044  *
4045  * if parent is local or mode is LCK_PW, take PDO lock, otherwise take regular
4046  * lock.
4047  *
4048  * \param info  struct mdt_thread_info
4049  * \param obj   parent object
4050  * \param lh    lock handle
4051  * \param lname child name
4052  * \param mode  lock mode
4053  *
4054  * \retval      0 on success, -ev on error.
4055  */
4056 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4057                     struct mdt_lock_handle *lh, const struct lu_name *lname,
4058                     enum ldlm_mode mode)
4059 {
4060         int rc;
4061
4062         ENTRY;
4063         LASSERT(obj && lname);
4064         LASSERT(mode == LCK_PW || mode == LCK_PR);
4065         if (mdt_object_remote(obj) && mode == LCK_PR) {
4066                 __u64 ibits = MDS_INODELOCK_UPDATE;
4067
4068                 mdt_lock_reg_init(lh, mode);
4069                 rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj),
4070                                               lh, &ibits, 0, false);
4071         } else {
4072                 rc = mdt_object_pdo_lock(info, obj, lh, lname, mode, true);
4073         }
4074         RETURN(rc);
4075 }
4076
4077 /**
4078  * lock object with trybits
4079  *
4080  * the trybits contains optional inode lock bits that can be granted. This is
4081  * called by getattr/open to fetch more inode lock bits to client, and is also
4082  * called by dir migration to lock link parent in non-block mode to avoid
4083  * deadlock.
4084  *
4085  * \param info          struct mdt_thread_info
4086  * \param obj           object
4087  * \param lh            lock handle
4088  * \param ibits         MDS inode lock bits
4089  * \param trybits       optional inode lock bits
4090  * \param mode          lock mode
4091  *
4092  * \retval              0 on success, -ev on error.
4093  */
4094 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *obj,
4095                         struct mdt_lock_handle *lh, __u64 *ibits,
4096                         __u64 trybits, enum ldlm_mode mode)
4097 {
4098         bool trylock_only = *ibits == 0;
4099         int rc;
4100
4101         ENTRY;
4102         LASSERT(!(*ibits & trybits));
4103         mdt_lock_reg_init(lh, mode);
4104         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh, ibits,
4105                                       trybits, false);
4106         if (rc && trylock_only) { /* clear error for try ibits lock only */
4107                 LASSERT(*ibits == 0);
4108                 rc = 0;
4109         }
4110         RETURN(rc);
4111 }
4112
4113 /*
4114  * Helper function to take \a obj LOOKUP lock.
4115  *
4116  * Both \a pobj and \a obj may be located on remote MDTs.
4117  */
4118 int mdt_object_lookup_lock(struct mdt_thread_info *info,
4119                            struct mdt_object *pobj, struct mdt_object *obj,
4120                            struct mdt_lock_handle *lh, enum ldlm_mode mode)
4121 {
4122         __u64 ibits = MDS_INODELOCK_LOOKUP;
4123         int rc;
4124
4125         ENTRY;
4126         /* if @parent is NULL, it's on local MDT, and @child is remote,
4127          * this is case in getattr/unlink/open by name.
4128          */
4129         LASSERT(ergo(!pobj, mdt_object_remote(obj)));
4130         mdt_lock_reg_init(lh, mode);
4131         rc = mdt_object_lock_internal(info, pobj, mdt_object_fid(obj), lh,
4132                                       &ibits, 0, false);
4133         RETURN(rc);
4134 }
4135
4136 /**
4137  * Save a lock within request object.
4138  *
4139  * Keep the lock referenced until whether client ACK or transaction
4140  * commit happens or release the lock immediately depending on input
4141  * parameters. If COS is ON, a write lock is converted to COS lock
4142  * before saving.
4143  *
4144  * \param info thead info object
4145  * \param h lock handle
4146  * \param mode lock mode
4147  * \param decref force immediate lock releasing
4148  */
4149 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
4150                           enum ldlm_mode mode, int decref)
4151 {
4152         struct tgt_session_info *tsi = info->mti_env->le_ses ?
4153                                        tgt_ses_info(info->mti_env) : NULL;
4154
4155         ENTRY;
4156
4157         if (lustre_handle_is_used(h)) {
4158                 bool has_trans = tsi && tsi->tsi_has_trans;
4159
4160                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
4161                         mdt_fid_unlock(h, mode);
4162                 } else {
4163                         struct mdt_device *mdt = info->mti_mdt;
4164                         struct ldlm_lock *lock = ldlm_handle2lock(h);
4165                         struct ptlrpc_request *req = mdt_info_req(info);
4166                         bool no_ack = false;
4167
4168                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
4169                                  h->cookie);
4170
4171                         /* there is no request if mdt_object_unlock() is called
4172                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
4173                         if (likely(req != NULL)) {
4174                                 LDLM_DEBUG(lock, "save lock request %p reply "
4175                                         "state %p transno %lld", req,
4176                                         req->rq_reply_state, req->rq_transno);
4177                                 if (mdt_cos_is_enabled(mdt)) {
4178                                         mode = LCK_COS;
4179                                         no_ack = true;
4180                                         ldlm_lock_mode_downgrade(lock, mode);
4181                                 } else if (mdt_slc_is_enabled(mdt)) {
4182                                         no_ack = true;
4183                                         if (mode != LCK_TXN) {
4184                                                 mode = LCK_TXN;
4185                                                 ldlm_lock_mode_downgrade(lock,
4186                                                                          mode);
4187                                         }
4188                                 }
4189                                 if (req->rq_export->exp_disconnected)
4190                                         mdt_fid_unlock(h, mode);
4191                                 else
4192                                         ptlrpc_save_lock(req, h, mode, no_ack);
4193                         } else {
4194                                 mdt_fid_unlock(h, mode);
4195                         }
4196
4197                         if (mdt_is_lock_sync(lock)) {
4198                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4199                                 mdt_device_commit_async(info->mti_env, mdt);
4200                         }
4201                         LDLM_LOCK_PUT(lock);
4202                 }
4203                 h->cookie = 0ull;
4204         }
4205
4206         EXIT;
4207 }
4208
4209 /**
4210  * Save cross-MDT lock in uncommitted_slc_locks
4211  *
4212  * Keep the lock referenced until transaction commit happens or release the lock
4213  * immediately depending on input parameters.
4214  *
4215  * \param info thead info object
4216  * \param h lock handle
4217  * \param mode lock mode
4218  * \param decref force immediate lock releasing
4219  */
4220 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4221                                  struct mdt_object *o, struct lustre_handle *h,
4222                                  enum ldlm_mode mode, int decref)
4223 {
4224         ENTRY;
4225
4226         if (lustre_handle_is_used(h)) {
4227                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4228                 struct ptlrpc_request *req = mdt_info_req(info);
4229
4230                 if (o != NULL &&
4231                     (lock->l_policy_data.l_inodebits.bits &
4232                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4233                         mo_invalidate(info->mti_env, mdt_object_child(o));
4234
4235                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4236                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4237                         ldlm_lock_decref_and_cancel(h, mode);
4238                         LDLM_LOCK_PUT(lock);
4239                 } else {
4240                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4241                                           req->rq_transno);
4242                         ldlm_lock_decref(h, mode);
4243                 }
4244                 h->cookie = 0ull;
4245         }
4246
4247         EXIT;
4248 }
4249
4250 /**
4251  * Unlock mdt object.
4252  *
4253  * Immeditely release the regular lock and the PDO lock or save the
4254  * lock in request and keep them referenced until client ACK or
4255  * transaction commit.
4256  *
4257  * \param info thread info object
4258  * \param o mdt object
4259  * \param lh mdt lock handle referencing regular and PDO locks
4260  * \param decref force immediate lock releasing
4261  *
4262  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4263  */
4264 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4265                        struct mdt_lock_handle *lh, int decref)
4266 {
4267         ENTRY;
4268
4269         if (lh->mlh_pdo_remote)
4270                 mdt_save_remote_lock(info, o, &lh->mlh_pdo_lh,
4271                                      lh->mlh_pdo_mode, decref);
4272         else
4273                 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4274         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4275         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4276                              decref);
4277
4278         EXIT;
4279 }
4280
4281 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4282                                         const struct lu_fid *f,
4283                                         struct mdt_lock_handle *lh,
4284                                         __u64 ibits, enum ldlm_mode mode)
4285 {
4286         struct mdt_object *o;
4287
4288         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4289         if (!IS_ERR(o)) {
4290                 int rc;
4291
4292                 rc = mdt_object_lock(info, o, lh, ibits, mode);
4293                 if (rc != 0) {
4294                         mdt_object_put(info->mti_env, o);
4295                         o = ERR_PTR(rc);
4296                 }
4297         }
4298         return o;
4299 }
4300
4301 void mdt_object_unlock_put(struct mdt_thread_info * info,
4302                            struct mdt_object *o,
4303                            struct mdt_lock_handle *lh,
4304                            int decref)
4305 {
4306         mdt_object_unlock(info, o, lh, decref);
4307         mdt_object_put(info->mti_env, o);
4308 }
4309
4310 /*
4311  * Generic code handling requests that have struct mdt_body passed in:
4312  *
4313  *  - extract mdt_body from request and save it in @info, if present;
4314  *
4315  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4316  *  @info;
4317  *
4318  *  - if HAS_BODY flag is set for this request type check whether object
4319  *  actually exists on storage (lu_object_exists()).
4320  *
4321  */
4322 static int mdt_body_unpack(struct mdt_thread_info *info,
4323                            enum tgt_handler_flags flags)
4324 {
4325         const struct mdt_body    *body;
4326         struct mdt_object        *obj;
4327         const struct lu_env      *env;
4328         struct req_capsule       *pill;
4329         int                       rc;
4330         ENTRY;
4331
4332         env = info->mti_env;
4333         pill = info->mti_pill;
4334
4335         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4336         if (body == NULL)
4337                 RETURN(-EFAULT);
4338
4339         if (!(body->mbo_valid & OBD_MD_FLID))
4340                 RETURN(0);
4341
4342         if (!fid_is_sane(&body->mbo_fid1)) {
4343                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4344                 RETURN(-EINVAL);
4345         }
4346
4347         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4348         if (!IS_ERR(obj)) {
4349                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4350                         mdt_object_put(env, obj);
4351                         rc = -ENOENT;
4352                 } else {
4353                         info->mti_object = obj;
4354                         rc = 0;
4355                 }
4356         } else
4357                 rc = PTR_ERR(obj);
4358
4359         RETURN(rc);
4360 }
4361
4362 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4363                                    enum tgt_handler_flags flags)
4364 {
4365         struct req_capsule *pill = info->mti_pill;
4366         int rc;
4367
4368         ENTRY;
4369
4370         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4371                 rc = mdt_body_unpack(info, flags);
4372         else
4373                 rc = 0;
4374
4375         if (rc == 0 && (flags & HAS_REPLY)) {
4376                 /* Pack reply. */
4377                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4378                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4379                                              req_capsule_ptlreq(pill) ?
4380                                              DEF_REP_MD_SIZE : MAX_MD_SIZE_OLD);
4381
4382                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4383                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4384                                              RCL_SERVER, 0);
4385
4386                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4387                  * by default. If the target object has more ACL entries, then
4388                  * enlarge the buffer when necessary. */
4389                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4390                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4391                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4392
4393                 mdt_preset_secctx_size(info);
4394                 mdt_preset_encctx_size(info);
4395
4396                 rc = req_capsule_server_pack(pill);
4397                 if (rc)
4398                         CWARN("%s: cannot pack response: rc = %d\n",
4399                                       mdt_obd_name(info->mti_mdt), rc);
4400         }
4401         RETURN(rc);
4402 }
4403
4404 void mdt_thread_info_reset(struct mdt_thread_info *info)
4405 {
4406         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4407         info->mti_body = NULL;
4408         info->mti_dlm_req = NULL;
4409         info->mti_cross_ref = 0;
4410         info->mti_opdata = 0;
4411         info->mti_big_lmm_used = 0;
4412         info->mti_big_acl_used = 0;
4413         info->mti_som_strict = 0;
4414
4415         info->mti_spec.no_create = 0;
4416         info->mti_spec.sp_rm_entry = 0;
4417         info->mti_spec.sp_permitted = 0;
4418
4419         info->mti_spec.u.sp_ea.eadata = NULL;
4420         info->mti_spec.u.sp_ea.eadatalen = 0;
4421
4422         if (info->mti_batch_env && info->mti_object != NULL) {
4423                 mdt_object_put(info->mti_env, info->mti_object);
4424                 info->mti_object = NULL;
4425         }
4426 }
4427
4428 /*
4429  * Initialize fields of struct mdt_thread_info. Other fields are left in
4430  * uninitialized state, because it's too expensive to zero out whole
4431  * mdt_thread_info (> 1K) on each request arrival.
4432  */
4433 void mdt_thread_info_init(struct ptlrpc_request *req,
4434                           struct mdt_thread_info *info)
4435 {
4436         info->mti_pill = &req->rq_pill;
4437
4438         /* mdt device: it can be NULL while CONNECT */
4439         if (req->rq_export) {
4440                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4441                 info->mti_exp = req->rq_export;
4442         } else
4443                 info->mti_mdt = NULL;
4444         info->mti_env = req->rq_svc_thread->t_env;
4445         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4446         info->mti_big_buf = LU_BUF_NULL;
4447         info->mti_batch_env = 0;
4448         info->mti_object = NULL;
4449
4450         mdt_thread_info_reset(info);
4451 }
4452
4453 void mdt_thread_info_fini(struct mdt_thread_info *info)
4454 {
4455         int i;
4456
4457         if (info->mti_object != NULL) {
4458                 mdt_object_put(info->mti_env, info->mti_object);
4459                 info->mti_object = NULL;
4460         }
4461
4462         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4463                 mdt_lock_handle_assert(&info->mti_lh[i]);
4464         info->mti_env = NULL;
4465         info->mti_pill = NULL;
4466         info->mti_exp = NULL;
4467         info->mti_mdt = NULL;
4468
4469         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4470                 lu_buf_free(&info->mti_big_buf);
4471 }
4472
4473 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4474 {
4475         struct mdt_thread_info  *mti;
4476
4477         mti = mdt_th_info(tsi->tsi_env);
4478         LASSERT(mti != NULL);
4479
4480         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4481         if (tsi->tsi_corpus != NULL) {
4482                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4483                 lu_object_get(tsi->tsi_corpus);
4484         }
4485         mti->mti_body = tsi->tsi_mdt_body;
4486         mti->mti_dlm_req = tsi->tsi_dlm_req;
4487
4488         return mti;
4489 }
4490
4491 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4492 {
4493         if (CFS_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4494             cfs_fail_val ==
4495             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4496                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4497
4498         return tgt_connect(tsi);
4499 }
4500
4501 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4502                               struct mdt_thread_info *info,
4503                               struct ldlm_lock **lockp, __u64 flags)
4504 {
4505         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4506                                    lockp, flags);
4507 }
4508 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4509                           struct mdt_thread_info *info,
4510                           struct ldlm_lock **lockp, __u64 flags)
4511 {
4512         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4513                                lockp, flags);
4514 }
4515
4516 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4517                             struct ldlm_lock **lockp,
4518                             struct mdt_lock_handle *lh,
4519                             __u64 flags, int result)
4520 {
4521         struct ptlrpc_request  *req = mdt_info_req(info);
4522         struct ldlm_lock       *lock = *lockp;
4523         struct ldlm_lock       *new_lock;
4524
4525         /* If possible resent found a lock, @lh is set to its handle */
4526         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4527
4528         if (new_lock == NULL) {
4529                 if (flags & LDLM_FL_INTENT_ONLY) {
4530                         result = 0;
4531                 } else if (flags & LDLM_FL_RESENT) {
4532                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4533                          * resend case, however, it could be already destroyed
4534                          * due to client eviction or a raced cancel RPC.
4535                          */
4536                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4537                                           lh->mlh_reg_lh.cookie);
4538                         result = -ESTALE;
4539                 } else {
4540                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4541                                mdt_obd_name(info->mti_mdt),
4542                                lh->mlh_reg_lh.cookie, flags,
4543                                PFID(&info->mti_tmp_fid1),
4544                                PFID(&info->mti_tmp_fid2), result);
4545                         result = -ESTALE;
4546                 }
4547                 lh->mlh_reg_lh.cookie = 0;
4548                 RETURN(result);
4549         }
4550
4551         /*
4552          * If we've already given this lock to a client once, then we should
4553          * have no readers or writers.  Otherwise, we should have one reader
4554          * _or_ writer ref (which will be zeroed below) before returning the
4555          * lock to a client.
4556          */
4557         if (new_lock->l_export == req->rq_export) {
4558                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4559         } else {
4560                 LASSERT(new_lock->l_export == NULL);
4561                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4562         }
4563
4564         *lockp = new_lock;
4565
4566         if (new_lock->l_export == req->rq_export) {
4567                 /*
4568                  * Already gave this to the client, which means that we
4569                  * reconstructed a reply.
4570                  */
4571                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4572                         MSG_RESENT);
4573
4574                 LDLM_LOCK_RELEASE(new_lock);
4575                 lh->mlh_reg_lh.cookie = 0;
4576                 RETURN(ELDLM_LOCK_REPLACED);
4577         }
4578
4579         /*
4580          * Fixup the lock to be given to the client.
4581          */
4582         lock_res_and_lock(new_lock);
4583         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4584          * possible blocking AST. */
4585         while (new_lock->l_readers > 0) {
4586                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4587                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4588                 new_lock->l_readers--;
4589         }
4590         while (new_lock->l_writers > 0) {
4591                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4592                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4593                 new_lock->l_writers--;
4594         }
4595
4596         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4597         new_lock->l_blocking_ast = lock->l_blocking_ast;
4598         new_lock->l_completion_ast = lock->l_completion_ast;
4599         if (ldlm_has_dom(new_lock))
4600                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4601         new_lock->l_remote_handle = lock->l_remote_handle;
4602         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4603
4604         unlock_res_and_lock(new_lock);
4605
4606         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4607                      &new_lock->l_remote_handle,
4608                      &new_lock->l_exp_hash);
4609
4610         LDLM_LOCK_RELEASE(new_lock);
4611         lh->mlh_reg_lh.cookie = 0;
4612
4613         RETURN(ELDLM_LOCK_REPLACED);
4614 }
4615
4616 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4617                              struct ldlm_lock *new_lock,
4618                              struct mdt_lock_handle *lh, __u64 flags)
4619 {
4620         struct ptlrpc_request  *req = mdt_info_req(info);
4621         struct ldlm_request    *dlmreq;
4622
4623         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4624                 return;
4625
4626         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4627
4628         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4629          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4630          * initialized. */
4631         if (flags & LDLM_FL_RESENT) {
4632                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4633                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4634
4635                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4636                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4637                           lh->mlh_reg_lh.cookie);
4638                 return;
4639         }
4640
4641         /*
4642          * If the xid matches, then we know this is a resent request, and allow
4643          * it. (It's probably an OPEN, for which we don't send a lock.
4644          */
4645         if (req_can_reconstruct(req, NULL) != 0)
4646                 return;
4647
4648         /*
4649          * This remote handle isn't enqueued, so we never received or processed
4650          * this request.  Clear MSG_RESENT, because it can be handled like any
4651          * normal request now.
4652          */
4653         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4654
4655         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4656                   dlmreq->lock_handle[0].cookie);
4657 }
4658
4659 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4660                                struct mdt_thread_info *info,
4661                                struct ldlm_lock **lockp,
4662                                __u64 flags)
4663 {
4664         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4665         struct ldlm_reply      *ldlm_rep = NULL;
4666         int rc;
4667         ENTRY;
4668
4669         /*
4670          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4671          * (for the resend case) or a new lock. Below we will use it to
4672          * replace the original lock.
4673          */
4674         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4675         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4676                 rc = mdt_object_lock(info, info->mti_object, lhc,
4677                                      MDS_INODELOCK_XATTR, (*lockp)->l_req_mode);
4678                 if (rc)
4679                         return rc;
4680         }
4681
4682         rc = mdt_getxattr(info);
4683
4684         if (mdt_info_req(info)->rq_repmsg != NULL)
4685                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4686
4687         if (ldlm_rep == NULL ||
4688             CFS_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4689                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4690                 if (is_serious(rc))
4691                         RETURN(rc);
4692                 else
4693                         RETURN(err_serious(-EFAULT));
4694         }
4695
4696         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4697
4698         /* This is left for interop instead of adding a new interop flag.
4699          * LU-7433 */
4700 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4701         if (ldlm_rep->lock_policy_res2) {
4702                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4703                 RETURN(ELDLM_LOCK_ABORTED);
4704         }
4705 #endif
4706
4707         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4708         RETURN(rc);
4709 }
4710
4711 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4712                               struct mdt_thread_info *info,
4713                               struct ldlm_lock **lockp,
4714                               __u64 flags)
4715 {
4716         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4717         __u64                   child_bits;
4718         struct ldlm_reply      *ldlm_rep;
4719         struct mdt_body        *reqbody;
4720         struct mdt_body        *repbody;
4721         int                     rc, rc2;
4722         ENTRY;
4723
4724         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4725         LASSERT(reqbody);
4726
4727         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4728         LASSERT(repbody);
4729
4730         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4731         repbody->mbo_eadatasize = 0;
4732         repbody->mbo_aclsize = 0;
4733
4734         switch (it_opc) {
4735         case IT_LOOKUP:
4736                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4737                 break;
4738         case IT_GETATTR:
4739                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4740                              MDS_INODELOCK_PERM;
4741                 break;
4742         default:
4743                 CERROR("%s: unsupported intent %#x\n",
4744                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4745                 GOTO(out_shrink, rc = -EINVAL);
4746         }
4747
4748         rc = mdt_init_ucred(info, reqbody);
4749         if (rc)
4750                 GOTO(out_shrink, rc);
4751
4752         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4753         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4754
4755         /* Get lock from request for possible resent case. */
4756         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4757
4758         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4759         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4760
4761         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4762             ldlm_rep->lock_policy_res2 != -ENOKEY)
4763                 ldlm_rep->lock_policy_res2 = 0;
4764         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4765             ldlm_rep->lock_policy_res2) {
4766                 lhc->mlh_reg_lh.cookie = 0ull;
4767                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4768         }
4769
4770         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4771         EXIT;
4772 out_ucred:
4773         mdt_exit_ucred(info);
4774 out_shrink:
4775         mdt_client_compatibility(info);
4776         rc2 = mdt_fix_reply(info);
4777         if (rc == 0)
4778                 rc = rc2;
4779         return rc;
4780 }
4781
4782 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4783                              struct mdt_thread_info *info,
4784                              struct ldlm_lock **lockp,
4785                              __u64 flags)
4786 {
4787         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4788         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4789         struct layout_intent *intent;
4790         struct ldlm_reply *ldlm_rep;
4791         struct lu_fid *fid = &info->mti_tmp_fid2;
4792         struct mdt_object *obj = NULL;
4793         int layout_size = 0;
4794         struct lu_buf *buf = &layout.mlc_buf;
4795         int rc = 0;
4796
4797         ENTRY;
4798
4799         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4800
4801         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4802         if (intent == NULL)
4803                 RETURN(-EPROTO);
4804
4805         CDEBUG(D_INFO, DFID "got layout change request from client: "
4806                "opc:%u flags:%#x extent "DEXT"\n",
4807                PFID(fid), intent->li_opc, intent->li_flags,
4808                PEXT(&intent->li_extent));
4809
4810         switch (intent->li_opc) {
4811         case LAYOUT_INTENT_TRUNC:
4812         case LAYOUT_INTENT_WRITE:
4813         case LAYOUT_INTENT_PCCRO_SET:
4814         case LAYOUT_INTENT_PCCRO_CLEAR:
4815                 layout.mlc_opc = MD_LAYOUT_WRITE;
4816                 layout.mlc_intent = intent;
4817                 break;
4818         case LAYOUT_INTENT_ACCESS:
4819                 break;
4820         case LAYOUT_INTENT_READ:
4821         case LAYOUT_INTENT_GLIMPSE:
4822         case LAYOUT_INTENT_RELEASE:
4823         case LAYOUT_INTENT_RESTORE:
4824                 CERROR("%s: Unsupported layout intent opc %d\n",
4825                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4826                 RETURN(-ENOTSUPP);
4827         default:
4828                 CERROR("%s: Unknown layout intent opc %d\n",
4829                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4830                 RETURN(-EINVAL);
4831         }
4832
4833         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4834         if (IS_ERR(obj))
4835                 RETURN(PTR_ERR(obj));
4836
4837         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4838                 /* if layout is going to be changed don't use the current EA
4839                  * size but the maximum one. That buffer will be shrinked
4840                  * to the actual size in req_capsule_shrink() before reply.
4841                  */
4842                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4843                         layout_size = info->mti_mdt->mdt_max_mdsize;
4844                 } else {
4845                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4846                         if (layout_size < 0)
4847                                 GOTO(out, rc = layout_size);
4848
4849                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4850                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4851                 }
4852                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4853                        mdt_obd_name(info->mti_mdt), layout_size);
4854         }
4855
4856         /*
4857          * set reply buffer size, so that ldlm_handle_enqueue0()->
4858          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4859          */
4860         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4861                              layout_size);
4862         rc = req_capsule_server_pack(info->mti_pill);
4863         if (rc)
4864                 GOTO(out, rc);
4865
4866         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4867         if (!ldlm_rep)
4868                 GOTO(out, rc = -EPROTO);
4869
4870         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4871
4872         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4873         if (layout.mlc_opc == MD_LAYOUT_NOP)
4874                 GOTO(out, rc = 0);
4875
4876         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4877         if (rc < 0)
4878                 GOTO(out, rc);
4879         if (rc == 1) {
4880                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4881                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4882                 GOTO(out, rc);
4883         }
4884
4885         buf->lb_buf = NULL;
4886         buf->lb_len = 0;
4887         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4888                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4889                                                      &RMF_EADATA);
4890                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4891                                                      &RMF_EADATA, RCL_CLIENT);
4892                 /*
4893                  * If it's a replay of layout write intent RPC, the client has
4894                  * saved the extended lovea when it get reply then.
4895                  */
4896                 if (buf->lb_len > 0)
4897                         mdt_fix_lov_magic(info, buf->lb_buf);
4898         }
4899
4900         /* Get lock from request for possible resent case. */
4901         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4902         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4903
4904         /*
4905          * Instantiate some layout components, if @buf contains lovea, then it's
4906          * a replay of the layout intent write RPC.
4907          */
4908         rc = mdt_layout_change(info, obj, lhc, &layout);
4909         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4910
4911         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4912                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4913                 if (rc == ELDLM_LOCK_REPLACED &&
4914                     (*lockp)->l_granted_mode == LCK_EX)
4915                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4916         }
4917
4918         EXIT;
4919 out:
4920         mdt_object_put(info->mti_env, obj);
4921         return rc;
4922 }
4923
4924 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4925                            struct mdt_thread_info *info,
4926                            struct ldlm_lock **lockp,
4927                            __u64 flags)
4928 {
4929         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4930         struct ldlm_reply      *rep = NULL;
4931         long                    opc;
4932         int                     rc;
4933         struct ptlrpc_request  *req = mdt_info_req(info);
4934
4935         static const struct req_format *intent_fmts[REINT_MAX] = {
4936                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4937                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4938         };
4939
4940         ENTRY;
4941
4942         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4943         if (opc < 0)
4944                 RETURN(opc);
4945
4946         /* Get lock from request for possible resent case. */
4947         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4948
4949         rc = mdt_reint_internal(info, lhc, opc);
4950
4951         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4952                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4953
4954         /* Check whether the reply has been packed successfully. */
4955         if (mdt_info_req(info)->rq_repmsg != NULL)
4956                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4957         if (rep == NULL) {
4958                 if (is_serious(rc))
4959                         RETURN(rc);
4960                 else
4961                         RETURN(err_serious(-EFAULT));
4962         }
4963
4964         /* MDC expects this in any case */
4965         if (rc != 0)
4966                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4967
4968         /* the open lock or the lock for cross-ref object should be
4969          * returned to the client */
4970         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4971             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4972                 rep->lock_policy_res2 = 0;
4973                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4974                 RETURN(rc);
4975         }
4976
4977         rep->lock_policy_res2 = clear_serious(rc);
4978
4979         if (rep->lock_policy_res2 == -ENOENT &&
4980             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4981             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4982                 rep->lock_policy_res2 = 0;
4983
4984         lhc->mlh_reg_lh.cookie = 0ull;
4985         if (rc == -ENOTCONN || rc == -ENODEV ||
4986             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4987                 /*
4988                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4989                  * will be returned by rq_status, and client at ptlrpc layer
4990                  * will detect this, then disconnect, reconnect the import
4991                  * immediately, instead of impacting the following the rpc.
4992                  */
4993                 RETURN(rc);
4994         }
4995         /*
4996          * For other cases, the error will be returned by intent, and client
4997          * will retrieve the result from intent.
4998          */
4999         RETURN(ELDLM_LOCK_ABORTED);
5000 }
5001
5002 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
5003                           struct mdt_thread_info *info,
5004                           struct ldlm_lock **lockp,
5005                           u64 flags /* LDLM_FL_* */)
5006 {
5007         struct req_capsule *pill = info->mti_pill;
5008         struct ptlrpc_request *req = mdt_info_req(info);
5009         const struct req_format *it_format;
5010         int (*it_handler)(enum ldlm_intent_flags,
5011                           struct mdt_thread_info *,
5012                           struct ldlm_lock **,
5013                           u64);
5014         enum tgt_handler_flags it_handler_flags = 0;
5015         struct ldlm_reply *rep;
5016         bool check_mdt_object = false;
5017         int rc;
5018         ENTRY;
5019
5020         switch (it_opc) {
5021         case IT_OPEN:
5022         case IT_OPEN|IT_CREAT:
5023                 /*
5024                  * OCREAT is not a IS_MUTABLE request since the file may
5025                  * already exist. We do the extra check of
5026                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
5027                  * really need to create the object.
5028                  */
5029                 it_format = &RQF_LDLM_INTENT;
5030                 it_handler = &mdt_intent_open;
5031                 break;
5032         case IT_GETATTR:
5033                 check_mdt_object = true;
5034                 fallthrough;
5035         case IT_LOOKUP:
5036                 it_format = &RQF_LDLM_INTENT_GETATTR;
5037                 it_handler = &mdt_intent_getattr;
5038                 it_handler_flags = HAS_REPLY;
5039                 break;
5040         case IT_GETXATTR:
5041                 check_mdt_object = true;
5042                 it_format = &RQF_LDLM_INTENT_GETXATTR;
5043                 it_handler = &mdt_intent_getxattr;
5044                 it_handler_flags = HAS_BODY;
5045                 break;
5046         case IT_LAYOUT:
5047                 it_format = &RQF_LDLM_INTENT_LAYOUT;
5048                 it_handler = &mdt_intent_layout;
5049                 break;
5050         case IT_GLIMPSE:
5051                 it_format = &RQF_LDLM_INTENT;
5052                 it_handler = &mdt_intent_glimpse;
5053                 break;
5054         case IT_BRW:
5055                 it_format = &RQF_LDLM_INTENT;
5056                 it_handler = &mdt_intent_brw;
5057                 break;
5058         case IT_QUOTA_DQACQ:
5059         case IT_QUOTA_CONN: {
5060                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
5061
5062                 if (qmt == NULL)
5063                         RETURN(-EOPNOTSUPP);
5064
5065                 if (mdt_rdonly(req->rq_export))
5066                         RETURN(-EROFS);
5067
5068                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
5069                 /* pass the request to quota master */
5070                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
5071                                                  mdt_info_req(info), lockp,
5072                                                  flags);
5073                 RETURN(rc);
5074         }
5075         default:
5076                 CERROR("%s: unknown intent code %#x\n",
5077                        mdt_obd_name(info->mti_mdt), it_opc);
5078                 RETURN(-EPROTO);
5079         }
5080
5081         if (!info->mti_batch_env)
5082                 req_capsule_extend(pill, it_format);
5083
5084         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
5085         if (rc < 0)
5086                 RETURN(rc);
5087
5088         if (unlikely(info->mti_object == NULL && check_mdt_object))
5089                 RETURN(-EPROTO);
5090
5091         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
5092                 RETURN(-EROFS);
5093
5094         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
5095
5096         /* execute policy */
5097         rc = (*it_handler)(it_opc, info, lockp, flags);
5098
5099         /* Check whether the reply has been packed successfully. */
5100         if (info->mti_batch_env || req->rq_repmsg != NULL) {
5101                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5102                 rep->lock_policy_res2 =
5103                         ptlrpc_status_hton(rep->lock_policy_res2);
5104         }
5105
5106         RETURN(rc);
5107 }
5108
5109 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
5110                                     enum ldlm_intent_flags it_opc)
5111 {
5112         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
5113
5114         /* update stats when IT code is known */
5115         if (srv_stats != NULL)
5116                 lprocfs_counter_incr(srv_stats,
5117                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
5118                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
5119 }
5120
5121 static int mdt_intent_policy(const struct lu_env *env,
5122                              struct ldlm_namespace *ns,
5123                              struct ldlm_lock **lockp,
5124                              void *req_cookie,
5125                              enum ldlm_mode mode,
5126                              __u64 flags, void *data)
5127 {
5128         struct tgt_session_info *tsi;
5129         struct mdt_thread_info  *info;
5130         struct ptlrpc_request   *req  =  req_cookie;
5131         struct ldlm_intent      *it;
5132         struct req_capsule      *pill;
5133         const struct ldlm_lock_desc *ldesc;
5134         int rc;
5135
5136         ENTRY;
5137
5138         LASSERT(req != NULL);
5139
5140         tsi = tgt_ses_info(env);
5141
5142         info = mdt_th_info(env);
5143         LASSERT(info != NULL);
5144
5145         /* Check whether it is a sub request processing in a batch request */
5146         if (info->mti_batch_env) {
5147                 pill = info->mti_pill;
5148                 LASSERT(pill == &info->mti_sub_pill);
5149         } else {
5150                 info = tsi2mdt_info(tsi);
5151                 pill = info->mti_pill;
5152         }
5153
5154         LASSERT(pill->rc_req == req);
5155         ldesc = &info->mti_dlm_req->lock_desc;
5156
5157         if (info->mti_batch_env ||
5158             req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
5159                 /*
5160                  * For batch processing environment, the request format has
5161                  * already been set.
5162                  */
5163                 if (!info->mti_batch_env)
5164                         req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
5165
5166                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
5167                 if (it != NULL) {
5168                         mdt_ptlrpc_stats_update(req, it->opc);
5169                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
5170                         if (rc == 0)
5171                                 rc = ELDLM_OK;
5172
5173                         /* Lock without inodebits makes no sense and will oops
5174                          * later in ldlm. Let's check it now to see if we have
5175                          * ibits corrupted somewhere in mdt_intent_opc().
5176                          * The case for client miss to set ibits has been
5177                          * processed by others. */
5178                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
5179                                 ldesc->l_policy_data.l_inodebits.bits != 0));
5180                 } else {
5181                         rc = err_serious(-EFAULT);
5182                 }
5183         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
5184                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
5185                 struct ldlm_reply *rep;
5186
5187                 /* No intent was provided but INTENT flag is set along with
5188                  * DOM bit, this is considered as GLIMPSE request.
5189                  * This logic is common for MDT and OST glimpse
5190                  */
5191                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
5192                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
5193                 /* Check whether the reply has been packed successfully. */
5194                 if (req->rq_repmsg != NULL) {
5195                         rep = req_capsule_server_get(info->mti_pill,
5196                                                      &RMF_DLM_REP);
5197                         rep->lock_policy_res2 =
5198                                 ptlrpc_status_hton(rep->lock_policy_res2);
5199                 }
5200         } else {
5201                 /* No intent was provided */
5202                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5203                 rc = req_capsule_server_pack(pill);
5204                 if (rc)
5205                         rc = err_serious(rc);
5206         }
5207
5208         if (!info->mti_batch_env)
5209                 mdt_thread_info_fini(info);
5210         RETURN(rc);
5211 }
5212
5213 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5214 {
5215         struct seq_server_site  *ss = mdt_seq_site(mdt);
5216
5217         if (ss->ss_node_id == 0)
5218                 return;
5219
5220         if (ss->ss_client_seq != NULL) {
5221                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5222                 ss->ss_client_seq->lcs_exp = NULL;
5223         }
5224
5225         if (ss->ss_server_fld != NULL) {
5226                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5227                 ss->ss_server_fld->lsf_control_exp = NULL;
5228         }
5229 }
5230
5231 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5232 {
5233         struct seq_server_site *ss = mdt_seq_site(mdt);
5234
5235         if (ss == NULL)
5236                 return;
5237
5238         if (ss->ss_server_seq != NULL)
5239                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5240 }
5241
5242 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5243 {
5244         mdt_seq_fini_cli(mdt);
5245         mdt_deregister_seq_exp(mdt);
5246
5247         return seq_site_fini(env, mdt_seq_site(mdt));
5248 }
5249
5250 /**
5251  * It will retrieve its FLDB entries from MDT0, and it only happens
5252  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5253  * and it needs to refresh FLDB from the MDT0.
5254  **/
5255 static int mdt_register_lwp_callback(void *data)
5256 {
5257         struct lu_env           env;
5258         struct mdt_device       *mdt = data;
5259         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5260         int                     rc;
5261         ENTRY;
5262
5263         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5264
5265         rc = lu_env_init(&env, LCT_MD_THREAD);
5266         if (rc < 0) {
5267                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5268                 RETURN(rc);
5269         }
5270
5271         /* Allocate new sequence now to avoid creating local transaction
5272          * in the normal transaction process */
5273         rc = seq_server_check_and_alloc_super(&env,
5274                                               mdt_seq_site(mdt)->ss_server_seq);
5275         if (rc < 0)
5276                 GOTO(out, rc);
5277
5278         if (fld->lsf_new) {
5279                 rc = fld_update_from_controller(&env, fld);
5280                 if (rc != 0) {
5281                         CERROR("%s: cannot update controller: rc = %d\n",
5282                                mdt_obd_name(mdt), rc);
5283                         GOTO(out, rc);
5284                 }
5285         }
5286 out:
5287         lu_env_fini(&env);
5288         RETURN(rc);
5289 }
5290
5291 static int mdt_register_seq_exp(struct mdt_device *mdt)
5292 {
5293         struct seq_server_site  *ss = mdt_seq_site(mdt);
5294         char                    *lwp_name = NULL;
5295         int                     rc;
5296
5297         if (ss->ss_node_id == 0)
5298                 return 0;
5299
5300         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5301         if (lwp_name == NULL)
5302                 GOTO(out_free, rc = -ENOMEM);
5303
5304         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5305         if (rc != 0)
5306                 GOTO(out_free, rc);
5307
5308         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5309                                       NULL, NULL);
5310         if (rc != 0)
5311                 GOTO(out_free, rc);
5312
5313         rc = lustre_register_lwp_item(lwp_name,
5314                                       &ss->ss_server_fld->lsf_control_exp,
5315                                       mdt_register_lwp_callback, mdt);
5316         if (rc != 0) {
5317                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5318                 ss->ss_client_seq->lcs_exp = NULL;
5319                 GOTO(out_free, rc);
5320         }
5321 out_free:
5322         if (lwp_name != NULL)
5323                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5324
5325         return rc;
5326 }
5327
5328 /*
5329  * Init client sequence manager which is used by local MDS to talk to sequence
5330  * controller on remote node.
5331  */
5332 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5333 {
5334         struct seq_server_site *ss = mdt_seq_site(mdt);
5335         char *prefix;
5336         ENTRY;
5337
5338         /* check if this is adding the first MDC and controller is not yet
5339          * initialized. */
5340         OBD_ALLOC_PTR(ss->ss_client_seq);
5341         if (ss->ss_client_seq == NULL)
5342                 RETURN(-ENOMEM);
5343
5344         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5345         if (prefix == NULL) {
5346                 OBD_FREE_PTR(ss->ss_client_seq);
5347                 ss->ss_client_seq = NULL;
5348                 RETURN(-ENOMEM);
5349         }
5350
5351         /* Note: seq_client_fini will be called in seq_site_fini */
5352         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5353         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5354                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5355                                                             NULL);
5356         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5357
5358         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5359 }
5360
5361 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5362 {
5363         struct seq_server_site  *ss;
5364         int                     rc;
5365         ENTRY;
5366
5367         ss = mdt_seq_site(mdt);
5368         /* init sequence controller server(MDT0) */
5369         if (ss->ss_node_id == 0) {
5370                 OBD_ALLOC_PTR(ss->ss_control_seq);
5371                 if (ss->ss_control_seq == NULL)
5372                         RETURN(-ENOMEM);
5373
5374                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5375                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5376                                      ss);
5377                 if (rc)
5378                         GOTO(out_seq_fini, rc);
5379         }
5380
5381         /* Init normal sequence server */
5382         OBD_ALLOC_PTR(ss->ss_server_seq);
5383         if (ss->ss_server_seq == NULL)
5384                 GOTO(out_seq_fini, rc = -ENOMEM);
5385
5386         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5387                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5388         if (rc)
5389                 GOTO(out_seq_fini, rc);
5390
5391         /* init seq client for seq server to talk to seq controller(MDT0) */
5392         rc = mdt_seq_init_cli(env, mdt);
5393         if (rc != 0)
5394                 GOTO(out_seq_fini, rc);
5395
5396         if (ss->ss_node_id != 0)
5397                 /* register controller export through lwp */
5398                 rc = mdt_register_seq_exp(mdt);
5399
5400         EXIT;
5401 out_seq_fini:
5402         if (rc)
5403                 mdt_seq_fini(env, mdt);
5404
5405         return rc;
5406 }
5407
5408 /*
5409  * FLD wrappers
5410  */
5411 static int mdt_fld_fini(const struct lu_env *env,
5412                         struct mdt_device *m)
5413 {
5414         struct seq_server_site *ss = mdt_seq_site(m);
5415         ENTRY;
5416
5417         if (ss && ss->ss_server_fld) {
5418                 fld_server_fini(env, ss->ss_server_fld);
5419                 OBD_FREE_PTR(ss->ss_server_fld);
5420                 ss->ss_server_fld = NULL;
5421         }
5422
5423         RETURN(0);
5424 }
5425
5426 static int mdt_fld_init(const struct lu_env *env,
5427                         const char *uuid,
5428                         struct mdt_device *m)
5429 {
5430         struct seq_server_site *ss;
5431         int rc;
5432         ENTRY;
5433
5434         ss = mdt_seq_site(m);
5435
5436         OBD_ALLOC_PTR(ss->ss_server_fld);
5437         if (ss->ss_server_fld == NULL)
5438                 RETURN(rc = -ENOMEM);
5439
5440         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5441                              LU_SEQ_RANGE_MDT);
5442         if (rc) {
5443                 OBD_FREE_PTR(ss->ss_server_fld);
5444                 ss->ss_server_fld = NULL;
5445                 RETURN(rc);
5446         }
5447
5448         RETURN(0);
5449 }
5450
5451 static void mdt_stack_pre_fini(const struct lu_env *env,
5452                            struct mdt_device *m, struct lu_device *top)
5453 {
5454         struct lustre_cfg_bufs  *bufs;
5455         struct lustre_cfg       *lcfg;
5456         struct mdt_thread_info  *info;
5457         ENTRY;
5458
5459         LASSERT(top);
5460
5461         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5462         LASSERT(info != NULL);
5463
5464         bufs = &info->mti_u.bufs;
5465
5466         LASSERT(m->mdt_child_exp);
5467         LASSERT(m->mdt_child_exp->exp_obd);
5468
5469         /* process cleanup, pass mdt obd name to get obd umount flags */
5470         /* XXX: this is needed because all layers are referenced by
5471          * objects (some of them are pinned by osd, for example *
5472          * the proper solution should be a model where object used
5473          * by osd only doesn't have mdt/mdd slices -bzzz */
5474         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5475         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5476         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5477         if (!lcfg)
5478                 RETURN_EXIT;
5479         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5480
5481         top->ld_ops->ldo_process_config(env, top, lcfg);
5482         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5483         EXIT;
5484 }
5485
5486 static void mdt_stack_fini(const struct lu_env *env,
5487                            struct mdt_device *m, struct lu_device *top)
5488 {
5489         struct obd_device       *obd = mdt2obd_dev(m);
5490         struct lustre_cfg_bufs  *bufs;
5491         struct lustre_cfg       *lcfg;
5492         struct mdt_thread_info  *info;
5493         char                     flags[3] = "";
5494         ENTRY;
5495
5496         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5497         LASSERT(info != NULL);
5498
5499         lu_dev_del_linkage(top->ld_site, top);
5500
5501         lu_site_purge(env, top->ld_site, -1);
5502
5503         bufs = &info->mti_u.bufs;
5504         /* process cleanup, pass mdt obd name to get obd umount flags */
5505         /* another purpose is to let all layers to release their objects */
5506         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5507         if (obd->obd_force)
5508                 strcat(flags, "F");
5509         if (obd->obd_fail)
5510                 strcat(flags, "A");
5511         lustre_cfg_bufs_set_string(bufs, 1, flags);
5512         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5513         if (!lcfg)
5514                 RETURN_EXIT;
5515         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5516
5517         LASSERT(top);
5518         top->ld_ops->ldo_process_config(env, top, lcfg);
5519         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5520
5521         lu_site_purge(env, top->ld_site, -1);
5522
5523         m->mdt_child = NULL;
5524         m->mdt_bottom = NULL;
5525
5526         obd_disconnect(m->mdt_child_exp);
5527         m->mdt_child_exp = NULL;
5528
5529         obd_disconnect(m->mdt_bottom_exp);
5530         m->mdt_child_exp = NULL;
5531 }
5532
5533 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5534                                const char *next, struct obd_export **exp)
5535 {
5536         struct obd_connect_data *data = NULL;
5537         struct obd_device       *obd;
5538         int                      rc;
5539         ENTRY;
5540
5541         OBD_ALLOC_PTR(data);
5542         if (data == NULL)
5543                 GOTO(out, rc = -ENOMEM);
5544
5545         obd = class_name2obd(next);
5546         if (obd == NULL) {
5547                 CERROR("%s: can't locate next device: %s\n",
5548                        mdt_obd_name(m), next);
5549                 GOTO(out, rc = -ENOTCONN);
5550         }
5551
5552         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5553         data->ocd_version = LUSTRE_VERSION_CODE;
5554
5555         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5556         if (rc) {
5557                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5558                        mdt_obd_name(m), next, rc);
5559                 GOTO(out, rc);
5560         }
5561
5562 out:
5563         if (data)
5564                 OBD_FREE_PTR(data);
5565         RETURN(rc);
5566 }
5567
5568 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5569                           struct lustre_cfg *cfg)
5570 {
5571         char                   *dev = lustre_cfg_string(cfg, 0);
5572         int                     rc, name_size, uuid_size;
5573         char                   *name, *uuid, *p;
5574         struct lustre_cfg_bufs *bufs;
5575         struct lustre_cfg      *lcfg;
5576         struct obd_device      *obd;
5577         struct lustre_profile  *lprof;
5578         struct lu_site         *site;
5579         ENTRY;
5580
5581         /* in 1.8 we had the only device in the stack - MDS.
5582          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5583          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5584          * to generate names and setup MDT, MDD. MDT will be using
5585          * generated name to connect to MDD. for MDD the next device
5586          * will be LOD with name taken from so called "profile" which
5587          * is generated by mount_option line
5588          *
5589          * 1.8 MGS generates config. commands like this:
5590          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5591          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5592          * 2.0 MGS generates config. commands like this:
5593          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5594          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5595          *                    3:lustre-MDT0000-mdtlov  4:f
5596          *
5597          * we generate MDD name from MDT one, just replacing T with D
5598          *
5599          * after all the preparations, the logical equivalent will be
5600          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5601          *                    3:lustre-MDT0000-mdtlov  4:f
5602          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5603          *                    3:lustre-MDD0000  4:f
5604          *
5605          *  notice we build the stack from down to top: MDD first, then MDT */
5606
5607         name_size = MAX_OBD_NAME;
5608         uuid_size = MAX_OBD_NAME;
5609
5610         OBD_ALLOC(name, name_size);
5611         OBD_ALLOC(uuid, uuid_size);
5612         if (name == NULL || uuid == NULL)
5613                 GOTO(cleanup_mem, rc = -ENOMEM);
5614
5615         OBD_ALLOC_PTR(bufs);
5616         if (!bufs)
5617                 GOTO(cleanup_mem, rc = -ENOMEM);
5618
5619         strcpy(name, dev);
5620         p = strstr(name, "-MDT");
5621         if (p == NULL)
5622                 GOTO(free_bufs, rc = -ENOMEM);
5623         p[3] = 'D';
5624
5625         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5626
5627         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5628         if (lprof == NULL || lprof->lp_dt == NULL) {
5629                 CERROR("can't find the profile: %s\n",
5630                        lustre_cfg_string(cfg, 0));
5631                 GOTO(free_bufs, rc = -EINVAL);
5632         }
5633
5634         lustre_cfg_bufs_reset(bufs, name);
5635         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5636         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5637         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5638
5639         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5640         if (!lcfg)
5641                 GOTO(put_profile, rc = -ENOMEM);
5642         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5643
5644         rc = class_attach(lcfg);
5645         if (rc)
5646                 GOTO(lcfg_cleanup, rc);
5647
5648         obd = class_name2obd(name);
5649         if (!obd) {
5650                 CERROR("Can not find obd %s (%s in config)\n",
5651                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5652                 GOTO(lcfg_cleanup, rc = -EINVAL);
5653         }
5654
5655         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5656
5657         lustre_cfg_bufs_reset(bufs, name);
5658         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5659         lustre_cfg_bufs_set_string(bufs, 2, dev);
5660         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5661
5662         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5663         if (!lcfg)
5664                 GOTO(class_detach, rc = -ENOMEM);
5665         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5666
5667         rc = class_setup(obd, lcfg);
5668         if (rc)
5669                 GOTO(class_detach, rc);
5670
5671         /* connect to MDD we just setup */
5672         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5673         if (rc)
5674                 GOTO(class_detach, rc);
5675
5676         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5677         LASSERT(site);
5678         LASSERT(mdt_lu_site(mdt) == NULL);
5679         mdt->mdt_lu_dev.ld_site = site;
5680         site->ls_top_dev = &mdt->mdt_lu_dev;
5681         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5682
5683         /* now connect to bottom OSD */
5684         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5685         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5686         if (rc)
5687                 GOTO(class_detach, rc);
5688         mdt->mdt_bottom =
5689                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5690
5691         rc = lu_env_refill((struct lu_env *)env);
5692         if (rc != 0)
5693                 CERROR("Failure to refill session: '%d'\n", rc);
5694
5695         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5696
5697         EXIT;
5698 class_detach:
5699         if (rc)
5700                 class_detach(obd, lcfg);
5701 lcfg_cleanup:
5702         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5703 put_profile:
5704         class_put_profile(lprof);
5705 free_bufs:
5706         OBD_FREE_PTR(bufs);
5707 cleanup_mem:
5708         if (name)
5709                 OBD_FREE(name, name_size);
5710         if (uuid)
5711                 OBD_FREE(uuid, uuid_size);
5712         RETURN(rc);
5713 }
5714
5715 /* setup quota master target on MDT0 */
5716 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5717                           struct lustre_cfg *cfg)
5718 {
5719         struct obd_device       *obd;
5720         char                    *dev = lustre_cfg_string(cfg, 0);
5721         char                    *qmtname, *uuid, *p;
5722         struct lustre_cfg_bufs  *bufs;
5723         struct lustre_cfg       *lcfg;
5724         struct lustre_profile   *lprof;
5725         struct obd_connect_data *data;
5726         int                      rc;
5727         ENTRY;
5728
5729         LASSERT(mdt->mdt_qmt_exp == NULL);
5730         LASSERT(mdt->mdt_qmt_dev == NULL);
5731
5732         /* quota master is on MDT0 only for now */
5733         if (mdt->mdt_seq_site.ss_node_id != 0)
5734                 RETURN(0);
5735
5736         /* MGS generates config commands which look as follows:
5737          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5738          *                    3:lustre-MDT0000-mdtlov  4:f
5739          *
5740          * We generate the QMT name from the MDT one, just replacing MD with QM
5741          * after all the preparations, the logical equivalent will be:
5742          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5743          *                    3:lustre-MDT0000-osd  4:f */
5744         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5745         OBD_ALLOC(uuid, UUID_MAX);
5746         OBD_ALLOC_PTR(bufs);
5747         OBD_ALLOC_PTR(data);
5748         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5749                 GOTO(cleanup_mem, rc = -ENOMEM);
5750
5751         strcpy(qmtname, dev);
5752         p = strstr(qmtname, "-MDT");
5753         if (p == NULL)
5754                 GOTO(cleanup_mem, rc = -ENOMEM);
5755         /* replace MD with QM */
5756         p[1] = 'Q';
5757         p[2] = 'M';
5758
5759         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5760
5761         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5762         if (lprof == NULL || lprof->lp_dt == NULL) {
5763                 CERROR("can't find profile for %s\n",
5764                        lustre_cfg_string(cfg, 0));
5765                 GOTO(cleanup_mem, rc = -EINVAL);
5766         }
5767
5768         lustre_cfg_bufs_reset(bufs, qmtname);
5769         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5770         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5771         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5772
5773         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5774         if (!lcfg)
5775                 GOTO(put_profile, rc = -ENOMEM);
5776         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5777
5778         rc = class_attach(lcfg);
5779         if (rc)
5780                 GOTO(lcfg_cleanup, rc);
5781
5782         obd = class_name2obd(qmtname);
5783         if (!obd) {
5784                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5785                        lustre_cfg_string(cfg, 0));
5786                 GOTO(lcfg_cleanup, rc = -EINVAL);
5787         }
5788
5789         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5790
5791         lustre_cfg_bufs_reset(bufs, qmtname);
5792         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5793         lustre_cfg_bufs_set_string(bufs, 2, dev);
5794
5795         /* for quota, the next device should be the OSD device */
5796         lustre_cfg_bufs_set_string(bufs, 3,
5797                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5798
5799         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5800         if (!lcfg)
5801                 GOTO(class_detach, rc = -ENOMEM);
5802         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5803
5804         rc = class_setup(obd, lcfg);
5805         if (rc)
5806                 GOTO(class_detach, rc);
5807
5808         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5809
5810         /* configure local quota objects */
5811         if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5812                 rc = -EBADF;
5813         else
5814                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5815                                                            &mdt->mdt_lu_dev,
5816                                                            mdt->mdt_qmt_dev);
5817         if (rc)
5818                 GOTO(class_cleanup, rc);
5819
5820         /* connect to quota master target */
5821         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5822         data->ocd_version = LUSTRE_VERSION_CODE;
5823         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5824                          data, NULL);
5825         if (rc) {
5826                 CERROR("cannot connect to quota master device %s (%d)\n",
5827                        qmtname, rc);
5828                 GOTO(class_cleanup, rc);
5829         }
5830
5831         EXIT;
5832 class_cleanup:
5833         if (rc) {
5834                 class_manual_cleanup(obd);
5835                 mdt->mdt_qmt_dev = NULL;
5836                 GOTO(lcfg_cleanup, rc);
5837         }
5838 class_detach:
5839         if (rc)
5840                 class_detach(obd, lcfg);
5841 lcfg_cleanup:
5842         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5843 put_profile:
5844         class_put_profile(lprof);
5845 cleanup_mem:
5846         if (bufs)
5847                 OBD_FREE_PTR(bufs);
5848         if (qmtname)
5849                 OBD_FREE(qmtname, MAX_OBD_NAME);
5850         if (uuid)
5851                 OBD_FREE(uuid, UUID_MAX);
5852         if (data)
5853                 OBD_FREE_PTR(data);
5854         return rc;
5855 }
5856
5857 /* Shutdown quota master target associated with mdt */
5858 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5859 {
5860         ENTRY;
5861
5862         if (mdt->mdt_qmt_exp == NULL)
5863                 RETURN_EXIT;
5864         LASSERT(mdt->mdt_qmt_dev != NULL);
5865
5866         /* the qmt automatically shuts down when the mdt disconnects */
5867         obd_disconnect(mdt->mdt_qmt_exp);
5868         mdt->mdt_qmt_exp = NULL;
5869         mdt->mdt_qmt_dev = NULL;
5870         EXIT;
5871 }
5872
5873 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5874  * for now. This will be removed along with converting rest of MDT code
5875  * to use tgt_session_info */
5876 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5877 {
5878         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5879         int                      rc;
5880
5881         if (unlikely(info->mti_object == NULL))
5882                 return -EPROTO;
5883
5884         rc = mdt_getxattr(info);
5885
5886         mdt_thread_info_fini(info);
5887         return rc;
5888 }
5889
5890 static int mdt_llog_open(struct tgt_session_info *tsi)
5891 {
5892         ENTRY;
5893
5894         if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
5895                 RETURN(err_serious(-EACCES));
5896
5897         RETURN(tgt_llog_open(tsi));
5898 }
5899
5900 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5901 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5902 #define OST_BRW_READ    OST_READ
5903 #define OST_BRW_WRITE   OST_WRITE
5904
5905 static struct tgt_handler mdt_tgt_handlers[] = {
5906 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5907                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5908                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5909 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5910                 0,                      MDS_DISCONNECT, tgt_disconnect,
5911                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5912 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5913                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5914                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5915 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5916 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5917 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5918 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5919                                                         mdt_getattr_name),
5920 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5921 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5922 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5923 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5924 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5925 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5926 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5927 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5928                                                         mdt_hsm_progress),
5929 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5930                                                         mdt_hsm_ct_register),
5931 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5932                                                         mdt_hsm_ct_unregister),
5933 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5934                                                         mdt_hsm_state_get),
5935 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5936                                                         mdt_hsm_state_set),
5937 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5938 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5939                                                         mdt_hsm_request),
5940 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5941             MDS_SWAP_LAYOUTS,
5942             mdt_swap_layouts),
5943 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5944 TGT_MDT_HDL(IS_MUTABLE,         MDS_BATCH,      mdt_batch),
5945 };
5946
5947 static struct tgt_handler mdt_io_ops[] = {
5948 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5949                                                         mdt_hp_brw),
5950 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5951                                                         mdt_hp_brw),
5952 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5953                                          OST_PUNCH,     mdt_punch_hdl,
5954                                                         mdt_hp_punch),
5955 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5956 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5957                                                         mdt_fallocate_hdl),
5958 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5959 TGT_RPC_HANDLER(OST_FIRST_OPC,
5960                 0,                      OST_SET_INFO,   mdt_io_set_info,
5961                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
5962 };
5963
5964 static struct tgt_handler mdt_sec_ctx_ops[] = {
5965 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5966 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5967 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5968 };
5969
5970 static struct tgt_handler mdt_quota_ops[] = {
5971 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5972 };
5973
5974 static struct tgt_handler mdt_llog_handlers[] = {
5975         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5976         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5977         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5978         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5979 };
5980
5981 static struct tgt_opc_slice mdt_common_slice[] = {
5982         {
5983                 .tos_opc_start  = MDS_FIRST_OPC,
5984                 .tos_opc_end    = MDS_LAST_OPC,
5985                 .tos_hs         = mdt_tgt_handlers
5986         },
5987         {
5988                 .tos_opc_start  = OBD_FIRST_OPC,
5989                 .tos_opc_end    = OBD_LAST_OPC,
5990                 .tos_hs         = tgt_obd_handlers
5991         },
5992         {
5993                 .tos_opc_start  = LDLM_FIRST_OPC,
5994                 .tos_opc_end    = LDLM_LAST_OPC,
5995                 .tos_hs         = tgt_dlm_handlers
5996         },
5997         {
5998                 .tos_opc_start  = SEC_FIRST_OPC,
5999                 .tos_opc_end    = SEC_LAST_OPC,
6000                 .tos_hs         = mdt_sec_ctx_ops
6001         },
6002         {
6003                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
6004                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
6005                 .tos_hs         = tgt_out_handlers
6006         },
6007         {
6008                 .tos_opc_start  = FLD_FIRST_OPC,
6009                 .tos_opc_end    = FLD_LAST_OPC,
6010                 .tos_hs         = fld_handlers
6011         },
6012         {
6013                 .tos_opc_start  = SEQ_FIRST_OPC,
6014                 .tos_opc_end    = SEQ_LAST_OPC,
6015                 .tos_hs         = seq_handlers
6016         },
6017         {
6018                 .tos_opc_start  = QUOTA_DQACQ,
6019                 .tos_opc_end    = QUOTA_LAST_OPC,
6020                 .tos_hs         = mdt_quota_ops
6021         },
6022         {
6023                 .tos_opc_start  = LLOG_FIRST_OPC,
6024                 .tos_opc_end    = LLOG_LAST_OPC,
6025                 .tos_hs         = mdt_llog_handlers
6026         },
6027         {
6028                 .tos_opc_start  = LFSCK_FIRST_OPC,
6029                 .tos_opc_end    = LFSCK_LAST_OPC,
6030                 .tos_hs         = tgt_lfsck_handlers
6031         },
6032         {
6033                 .tos_opc_start  = OST_FIRST_OPC,
6034                 .tos_opc_end    = OST_LAST_OPC,
6035                 .tos_hs         = mdt_io_ops
6036         },
6037         {
6038                 .tos_hs         = NULL
6039         }
6040 };
6041
6042 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
6043 {
6044         struct md_device *next = m->mdt_child;
6045         struct lu_device *d = &m->mdt_lu_dev;
6046         struct obd_device *obd = mdt2obd_dev(m);
6047         struct lfsck_stop stop;
6048
6049         ENTRY;
6050         stop.ls_status = LS_PAUSED;
6051         stop.ls_flags = 0;
6052         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
6053
6054         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
6055
6056         mdt_restriper_stop(m);
6057         ping_evictor_stop();
6058
6059         /* Remove the HSM /proc entry so the coordinator cannot be
6060          * restarted by a user while it's shutting down.
6061          */
6062         mdt_hsm_cdt_stop(m);
6063
6064         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
6065         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
6066
6067         if (m->mdt_namespace != NULL)
6068                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
6069                                           d->ld_obd->obd_force);
6070
6071         obd_exports_barrier(obd);
6072         obd_zombie_barrier();
6073
6074         mdt_quota_fini(env, m);
6075
6076         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
6077
6078         /* Calling the cleanup functions in the same order as in the mdt_init0
6079          * error path
6080          */
6081         mdt_tunables_fini(m);
6082
6083         target_recovery_fini(obd);
6084         upcall_cache_cleanup(m->mdt_identity_cache);
6085         m->mdt_identity_cache = NULL;
6086
6087         tgt_fini(env, &m->mdt_lut);
6088
6089         mdt_hsm_cdt_fini(m);
6090
6091         if (m->mdt_los != NULL) {
6092                 local_oid_storage_fini(env, m->mdt_los);
6093                 m->mdt_los = NULL;
6094         }
6095
6096         if (m->mdt_namespace != NULL) {
6097                 ldlm_namespace_free_post(m->mdt_namespace);
6098                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
6099         }
6100
6101         if (m->mdt_md_root != NULL) {
6102                 mdt_object_put(env, m->mdt_md_root);
6103                 m->mdt_md_root = NULL;
6104         }
6105
6106         mdt_seq_fini(env, m);
6107
6108         mdt_fld_fini(env, m);
6109
6110         /*
6111          * Finish the stack
6112          */
6113         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6114
6115         LASSERT(atomic_read(&d->ld_ref) == 0);
6116
6117         server_put_mount(mdt_obd_name(m), true);
6118
6119         EXIT;
6120 }
6121
6122 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
6123
6124 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
6125                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
6126 {
6127         const struct dt_device_param *dt_conf;
6128         struct mdt_thread_info *info;
6129         struct obd_device *obd;
6130         const char *dev = lustre_cfg_string(cfg, 0);
6131         const char *num = lustre_cfg_string(cfg, 2);
6132         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
6133         struct lustre_mount_info *lmi = NULL;
6134         struct lustre_sb_info *lsi;
6135         struct lu_site *s;
6136         struct seq_server_site *ss_site;
6137         const char *identity_upcall = "NONE";
6138         struct md_device *next;
6139         struct lu_fid fid;
6140         int rc;
6141         long node_id;
6142         mntopt_t mntopts;
6143         ENTRY;
6144
6145         lu_device_init(&m->mdt_lu_dev, ldt);
6146         /*
6147          * Environment (env) might be missing mdt_thread_key values at that
6148          * point, if device is allocated when mdt_thread_key is in QUIESCENT
6149          * mode.
6150          *
6151          * Usually device allocation path doesn't use module key values, but
6152          * mdt has to do a lot of work here, so allocate key value.
6153          */
6154         rc = lu_env_refill((struct lu_env *)env);
6155         if (rc != 0)
6156                 RETURN(rc);
6157
6158         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
6159         LASSERT(info != NULL);
6160
6161         obd = class_name2obd(dev);
6162         LASSERT(obd != NULL);
6163
6164         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
6165         m->mdt_opts.mo_evict_tgt_nids = 1;
6166         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
6167
6168         lmi = server_get_mount(dev);
6169         if (lmi == NULL) {
6170                 CERROR("Cannot get mount info for %s!\n", dev);
6171                 RETURN(-EFAULT);
6172         } else {
6173                 lsi = s2lsi(lmi->lmi_sb);
6174                 LASSERT(lsi->lsi_lmd);
6175                 /* CMD is supported only in IAM mode */
6176                 LASSERT(num);
6177                 rc = kstrtol(num, 10, &node_id);
6178                 if (rc)
6179                         RETURN(rc);
6180
6181                 obd_obt_init(obd);
6182                 if (test_bit(LMD_FLG_SKIP_LFSCK, lsi->lsi_lmd->lmd_flags))
6183                         m->mdt_skip_lfsck = 1;
6184                 if (test_bit(LMD_FLG_NO_CREATE, lsi->lsi_lmd->lmd_flags))
6185                         m->mdt_lut.lut_no_create = 1;
6186         }
6187
6188         /* Just try to get a DoM lock by default. Otherwise, having a group
6189          * lock granted, it may get blocked for a long time. */
6190         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
6191         /* DoM files are read at open and data is packed in the reply */
6192         m->mdt_opts.mo_dom_read_open = 1;
6193
6194         m->mdt_squash.rsi_uid = 0;
6195         m->mdt_squash.rsi_gid = 0;
6196         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
6197         spin_lock_init(&m->mdt_squash.rsi_lock);
6198         spin_lock_init(&m->mdt_lock);
6199         m->mdt_enable_chprojid_gid = 0;
6200         m->mdt_enable_dir_migration = 1;
6201         m->mdt_enable_dir_restripe = 0;
6202         m->mdt_enable_dir_auto_split = 0;
6203         m->mdt_enable_parallel_rename_dir = 1;
6204         m->mdt_enable_parallel_rename_file = 1;
6205         m->mdt_enable_remote_dir = 1;
6206         m->mdt_enable_remote_dir_gid = 0;
6207         m->mdt_enable_remote_rename = 1;
6208         m->mdt_enable_striped_dir = 1;
6209         m->mdt_enable_dmv_implicit_inherit = 1;
6210         m->mdt_dir_restripe_nsonly = 1;
6211         m->mdt_max_mod_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
6212
6213         atomic_set(&m->mdt_mds_mds_conns, 0);
6214         atomic_set(&m->mdt_async_commit_count, 0);
6215         atomic_set(&m->mdt_dmv_old_client_count, 0);
6216
6217         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6218         m->mdt_lu_dev.ld_obd = obd;
6219         /* Set this lu_device to obd for error handling purposes. */
6220         obd->obd_lu_dev = &m->mdt_lu_dev;
6221
6222         strncpy(m->mdt_job_xattr, XATTR_NAME_JOB_DEFAULT, XATTR_JOB_MAX_LEN);
6223
6224         /* init the stack */
6225         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6226         if (rc) {
6227                 CERROR("%s: Can't init device stack, rc %d\n",
6228                        mdt_obd_name(m), rc);
6229                 GOTO(err_lmi, rc);
6230         }
6231
6232         s = mdt_lu_site(m);
6233         ss_site = mdt_seq_site(m);
6234         s->ld_seq_site = ss_site;
6235         ss_site->ss_lu = s;
6236
6237         /* set server index */
6238         ss_site->ss_node_id = node_id;
6239
6240         /* failover is the default
6241          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6242          * assumed whose ss_node_id == 0 XXX
6243          * */
6244         obd->obd_replayable = 1;
6245         /* No connection accepted until configurations will finish */
6246         obd->obd_no_conn = 1;
6247
6248         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6249                 char *str = lustre_cfg_string(cfg, 4);
6250                 if (strchr(str, 'n')) {
6251                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6252                         obd->obd_replayable = 0;
6253                 }
6254         }
6255
6256         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6257         if (rc)
6258                 GOTO(err_fini_stack, rc);
6259
6260         rc = mdt_seq_init(env, m);
6261         if (rc)
6262                 GOTO(err_fini_fld, rc);
6263
6264         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6265                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6266         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6267                                               LDLM_NAMESPACE_SERVER,
6268                                               LDLM_NAMESPACE_GREEDY,
6269                                               LDLM_NS_TYPE_MDT);
6270         if (IS_ERR(m->mdt_namespace)) {
6271                 rc = PTR_ERR(m->mdt_namespace);
6272                 CERROR("%s: unable to create server namespace: rc = %d\n",
6273                        obd->obd_name, rc);
6274                 m->mdt_namespace = NULL;
6275                 GOTO(err_fini_seq, rc);
6276         }
6277
6278         m->mdt_namespace->ns_lvbp = m;
6279         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6280
6281         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6282         /* set obd_namespace for compatibility with old code */
6283         obd->obd_namespace = m->mdt_namespace;
6284
6285         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6286                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6287                       OBD_FAIL_MDS_ALL_REPLY_NET);
6288         if (rc)
6289                 GOTO(err_free_ns, rc);
6290
6291         /* Amount of available space excluded from granting and reserved
6292          * for metadata. It is a percentage of the total MDT size. */
6293         tgd->tgd_reserved_pcnt = 10;
6294
6295         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6296                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6297         else
6298                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6299
6300         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6301                 GOTO(err_tgt, rc = -ENOENT);
6302
6303         fid.f_seq = FID_SEQ_LOCAL_NAME;
6304         fid.f_oid = 1;
6305         fid.f_ver = 0;
6306         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6307         if (rc != 0)
6308                 GOTO(err_tgt, rc);
6309
6310         rc = mdt_hsm_cdt_init(m);
6311         if (rc != 0) {
6312                 CERROR("%s: error initializing coordinator, rc %d\n",
6313                        mdt_obd_name(m), rc);
6314                 GOTO(err_los_fini, rc);
6315         }
6316
6317         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6318
6319         next = m->mdt_child;
6320         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6321
6322         mntopts = dt_conf->ddp_mntopts;
6323
6324         if (mntopts & MNTOPT_USERXATTR)
6325                 m->mdt_opts.mo_user_xattr = 1;
6326         else
6327                 m->mdt_opts.mo_user_xattr = 0;
6328
6329         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6330
6331         if (mntopts & MNTOPT_ACL)
6332                 m->mdt_opts.mo_acl = 1;
6333         else
6334                 m->mdt_opts.mo_acl = 0;
6335
6336         m->mdt_opts.mo_enable_strict_som = 1;
6337
6338         /* XXX: to support suppgid for ACL, we enable identity_upcall
6339          * by default, otherwise, maybe got unexpected -EACCESS. */
6340         if (m->mdt_opts.mo_acl)
6341                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6342
6343         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6344                                                 identity_upcall,
6345                                                 UC_IDCACHE_HASH_SIZE,
6346                                                 1200, /* entry expire: 20 mn */
6347                                                 30, /* acquire expire: 30 s */
6348                                                 true, /* acquire can replay */
6349                                                 &mdt_identity_upcall_cache_ops);
6350         if (IS_ERR(m->mdt_identity_cache)) {
6351                 rc = PTR_ERR(m->mdt_identity_cache);
6352                 m->mdt_identity_cache = NULL;
6353                 GOTO(err_free_hsm, rc);
6354         }
6355
6356         rc = mdt_tunables_init(m, dev);
6357         if (rc) {
6358                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6359                 GOTO(err_recovery, rc);
6360         }
6361
6362         rc = mdt_quota_init(env, m, cfg);
6363         if (rc)
6364                 GOTO(err_procfs, rc);
6365
6366         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6367         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6368                            "mdt_ldlm_client", m->mdt_ldlm_client);
6369
6370         ping_evictor_start();
6371
6372         /* recovery will be started upon mdt_prepare()
6373          * when the whole stack is complete and ready
6374          * to serve the requests */
6375
6376         /* Reduce the initial timeout on an MDS because it doesn't need such
6377          * a long timeout as an OST does. Adaptive timeouts will adjust this
6378          * value appropriately. */
6379         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6380                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6381
6382         if (test_bit(LMD_FLG_LOCAL_RECOV, lsi->lsi_lmd->lmd_flags))
6383                 m->mdt_lut.lut_local_recovery = 1;
6384
6385         rc = mdt_restriper_start(m);
6386         if (rc)
6387                 GOTO(err_ping_evictor, rc);
6388
6389         RETURN(0);
6390
6391 err_ping_evictor:
6392         ping_evictor_stop();
6393 err_procfs:
6394         mdt_tunables_fini(m);
6395 err_recovery:
6396         upcall_cache_cleanup(m->mdt_identity_cache);
6397         m->mdt_identity_cache = NULL;
6398 err_free_hsm:
6399         mdt_hsm_cdt_fini(m);
6400 err_los_fini:
6401         local_oid_storage_fini(env, m->mdt_los);
6402         m->mdt_los = NULL;
6403 err_tgt:
6404         /* keep recoverable clients */
6405         obd->obd_fail = 1;
6406         target_recovery_fini(obd);
6407         obd_exports_barrier(obd);
6408         obd_zombie_barrier();
6409         tgt_fini(env, &m->mdt_lut);
6410 err_free_ns:
6411         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6412         obd->obd_namespace = m->mdt_namespace = NULL;
6413 err_fini_seq:
6414         mdt_seq_fini(env, m);
6415 err_fini_fld:
6416         mdt_fld_fini(env, m);
6417 err_fini_stack:
6418         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6419 err_lmi:
6420         if (lmi)
6421                 server_put_mount(dev, true);
6422         return(rc);
6423 }
6424
6425 /* For interoperability, the left element is old parameter, the right one
6426  * is the new version of the parameter, if some parameter is deprecated,
6427  * the new version should be set as NULL. */
6428 static struct cfg_interop_param mdt_interop_param[] = {
6429         { "mdt.group_upcall",   NULL },
6430         { "mdt.quota_type",     NULL },
6431         { "mdd.quota_type",     NULL },
6432         { "mdt.som",            NULL },
6433         { "mdt.rootsquash",     "mdt.root_squash" },
6434         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6435         { NULL }
6436 };
6437
6438 /* used by MGS to process specific configurations */
6439 static int mdt_process_config(const struct lu_env *env,
6440                               struct lu_device *d, struct lustre_cfg *cfg)
6441 {
6442         struct mdt_device *m = mdt_dev(d);
6443         struct md_device *md_next = m->mdt_child;
6444         struct lu_device *next = md2lu_dev(md_next);
6445         int rc;
6446         ENTRY;
6447
6448         switch (cfg->lcfg_command) {
6449         case LCFG_PARAM: {
6450                 struct obd_device *obd = d->ld_obd;
6451                 /* For interoperability */
6452                 struct cfg_interop_param *ptr = NULL;
6453                 struct lustre_cfg *old_cfg = NULL;
6454                 char *param = NULL;
6455                 ssize_t count;
6456
6457                 param = lustre_cfg_string(cfg, 1);
6458                 if (param == NULL) {
6459                         CERROR("param is empty\n");
6460                         rc = -EINVAL;
6461                         break;
6462                 }
6463
6464                 ptr = class_find_old_param(param, mdt_interop_param);
6465                 if (ptr != NULL) {
6466                         if (ptr->new_param == NULL) {
6467                                 rc = 0;
6468                                 CWARN("For interoperability, skip this %s."
6469                                       " It is obsolete.\n", ptr->old_param);
6470                                 break;
6471                         }
6472
6473                         CWARN("Found old param %s, changed it to %s.\n",
6474                               ptr->old_param, ptr->new_param);
6475
6476                         old_cfg = cfg;
6477                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6478                         if (IS_ERR(cfg)) {
6479                                 rc = PTR_ERR(cfg);
6480                                 break;
6481                         }
6482                 }
6483
6484                 count = class_modify_config(cfg, PARAM_MDT,
6485                                             &obd->obd_kset.kobj);
6486                 if (count < 0) {
6487                         struct coordinator *cdt = &m->mdt_coordinator;
6488
6489                         /* is it an HSM var ? */
6490                         count = class_modify_config(cfg, PARAM_HSM,
6491                                                     &cdt->cdt_hsm_kobj);
6492                         if (count < 0)
6493                                 /* we don't understand; pass it on */
6494                                 rc = next->ld_ops->ldo_process_config(env, next,
6495                                                                       cfg);
6496                         else
6497                                 rc = count > 0 ? 0 : count;
6498                 } else {
6499                         rc = count > 0 ? 0 : count;
6500                 }
6501
6502                 if (old_cfg)
6503                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6504                                                      cfg->lcfg_buflens));
6505                 break;
6506         }
6507         default:
6508                 /* others are passed further */
6509                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6510                 break;
6511         }
6512         RETURN(rc);
6513 }
6514
6515 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6516                                           const struct lu_object_header *hdr,
6517                                           struct lu_device *d)
6518 {
6519         struct mdt_object *mo;
6520
6521         ENTRY;
6522
6523         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6524         if (mo != NULL) {
6525                 struct lu_object *o;
6526                 struct lu_object_header *h;
6527
6528                 o = &mo->mot_obj;
6529                 h = &mo->mot_header;
6530                 lu_object_header_init(h);
6531                 lu_object_init(o, h, d);
6532                 lu_object_add_top(h, o);
6533                 o->lo_ops = &mdt_obj_ops;
6534                 spin_lock_init(&mo->mot_write_lock);
6535                 mutex_init(&mo->mot_som_mutex);
6536                 mutex_init(&mo->mot_lov_mutex);
6537                 init_rwsem(&mo->mot_dom_sem);
6538                 init_rwsem(&mo->mot_open_sem);
6539                 atomic_set(&mo->mot_open_count, 0);
6540                 mo->mot_restripe_offset = 0;
6541                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6542                 mo->mot_lsom_size = 0;
6543                 mo->mot_lsom_blocks = 0;
6544                 mo->mot_lsom_inited = false;
6545                 RETURN(o);
6546         }
6547         RETURN(NULL);
6548 }
6549
6550 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6551                            const struct lu_object_conf *unused)
6552 {
6553         struct mdt_device *d = mdt_dev(o->lo_dev);
6554         struct lu_device  *under;
6555         struct lu_object  *below;
6556         int                rc = 0;
6557         ENTRY;
6558
6559         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6560                PFID(lu_object_fid(o)));
6561
6562         under = &d->mdt_child->md_lu_dev;
6563         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6564         if (below != NULL) {
6565                 lu_object_add(o, below);
6566         } else
6567                 rc = -ENOMEM;
6568
6569         RETURN(rc);
6570 }
6571
6572 static void mdt_object_free_rcu(struct rcu_head *head)
6573 {
6574         struct mdt_object *mo = container_of(head, struct mdt_object,
6575                                              mot_header.loh_rcu);
6576
6577         kmem_cache_free(mdt_object_kmem, mo);
6578 }
6579
6580 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6581 {
6582         struct mdt_object *mo = mdt_obj(o);
6583         struct lu_object_header *h;
6584         ENTRY;
6585
6586         h = o->lo_header;
6587         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6588                PFID(lu_object_fid(o)));
6589
6590         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6591         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6592
6593         lu_object_fini(o);
6594         lu_object_header_fini(h);
6595         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6596         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6597
6598         EXIT;
6599 }
6600
6601 static int mdt_object_print(const struct lu_env *env, void *cookie,
6602                             lu_printer_t p, const struct lu_object *o)
6603 {
6604         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6605
6606         return (*p)(env, cookie,
6607                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6608                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6609                     mdto->mot_cache_attr ? "cache_attr" : "",
6610                     mdto->mot_write_count);
6611 }
6612
6613 static int mdt_prepare(const struct lu_env *env,
6614                 struct lu_device *pdev,
6615                 struct lu_device *cdev)
6616 {
6617         struct mdt_device *mdt = mdt_dev(cdev);
6618         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6619         struct obd_device *obd = cdev->ld_obd;
6620         int rc;
6621
6622         ENTRY;
6623
6624         LASSERT(obd);
6625
6626         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6627         if (rc)
6628                 RETURN(rc);
6629
6630         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6631         if (rc)
6632                 RETURN(rc);
6633
6634         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6635         if (rc)
6636                 RETURN(rc);
6637
6638         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6639         /* The LFSCK instance is registered just now, so it must be there when
6640          * register the namespace to such instance. */
6641         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6642
6643         if (mdt->mdt_seq_site.ss_node_id == 0) {
6644                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6645                                                          &mdt->mdt_md_root_fid);
6646                 if (rc)
6647                         RETURN(rc);
6648         }
6649
6650         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6651
6652         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6653         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6654         LASSERT(obd->obd_no_conn);
6655         spin_lock(&obd->obd_dev_lock);
6656         obd->obd_no_conn = 0;
6657         spin_unlock(&obd->obd_dev_lock);
6658
6659         if (obd->obd_recovering == 0)
6660                 mdt_postrecov(env, mdt);
6661
6662         RETURN(rc);
6663 }
6664
6665 const struct lu_device_operations mdt_lu_ops = {
6666         .ldo_object_alloc   = mdt_object_alloc,
6667         .ldo_process_config = mdt_process_config,
6668         .ldo_prepare        = mdt_prepare,
6669 };
6670
6671 static const struct lu_object_operations mdt_obj_ops = {
6672         .loo_object_init    = mdt_object_init,
6673         .loo_object_free    = mdt_object_free,
6674         .loo_object_print   = mdt_object_print
6675 };
6676
6677 static int mdt_obd_set_info_async(const struct lu_env *env,
6678                                   struct obd_export *exp,
6679                                   __u32 keylen, void *key,
6680                                   __u32 vallen, void *val,
6681                                   struct ptlrpc_request_set *set)
6682 {
6683         int rc;
6684
6685         ENTRY;
6686
6687         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6688                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6689                 RETURN(rc);
6690         }
6691
6692         RETURN(0);
6693 }
6694
6695 static inline void mdt_enable_slc(struct mdt_device *mdt)
6696 {
6697         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6698                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6699 }
6700
6701 static inline void mdt_disable_slc(struct mdt_device *mdt)
6702 {
6703         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6704                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6705 }
6706
6707 /**
6708  * Match client and server connection feature flags.
6709  *
6710  * Compute the compatibility flags for a connection request based on
6711  * features mutually supported by client and server.
6712  *
6713  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6714  * must not be updated here, otherwise a partially initialized value may
6715  * be exposed. After the connection request is successfully processed,
6716  * the top-level MDT connect request handler atomically updates the export
6717  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6718  * reply. \see mdt_connect().
6719  *
6720  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6721  * full struct obd_connect_data. So care must be taken when accessing fields
6722  * that are not present in struct obd_connect_data_v1. See LU-16.
6723  *
6724  * \param exp   the obd_export associated with this client/target pair
6725  * \param mdt   the target device for the connection
6726  * \param data  stores data for this connect request
6727  *
6728  * \retval 0       success
6729  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6730  * \retval -EBADE  client and server feature requirements are incompatible
6731  */
6732 static int mdt_connect_internal(const struct lu_env *env,
6733                                 struct obd_export *exp,
6734                                 struct mdt_device *mdt,
6735                                 struct obd_connect_data *data, bool reconnect)
6736 {
6737         const char *obd_name = mdt_obd_name(mdt);
6738         LASSERT(data != NULL);
6739
6740         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6741
6742         if (mdt->mdt_bottom->dd_rdonly &&
6743             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6744             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6745                 RETURN(-EACCES);
6746
6747         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6748                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6749
6750         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6751
6752         if (!mdt->mdt_opts.mo_acl)
6753                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6754
6755         if (!mdt->mdt_opts.mo_user_xattr)
6756                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6757
6758         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6759                 data->ocd_brw_size = min(data->ocd_brw_size,
6760                                          mdt->mdt_brw_size);
6761                 if (data->ocd_brw_size == 0) {
6762                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6763                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6764                                "ocd_brw_size unexpectedly zero, network data "
6765                                "corruption? Refusing to connect this client\n",
6766                                obd_name, exp->exp_client_uuid.uuid,
6767                                exp, data->ocd_connect_flags, data->ocd_version,
6768                                data->ocd_grant, data->ocd_index);
6769                         return -EPROTO;
6770                 }
6771         }
6772
6773         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6774                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6775
6776                 /* client is reporting its page size, for future use */
6777                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6778                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6779                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6780                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6781                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6782                 /* ocd_grant_tax_kb is in 1K byte blocks */
6783                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6784                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6785         }
6786
6787         /* Save connect_data we have so far because tgt_grant_connect()
6788          * uses it to calculate grant, and we want to save the client
6789          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6790         exp->exp_connect_data = *data;
6791         if (OCD_HAS_FLAG(data, GRANT))
6792                 tgt_grant_connect(env, exp, data, !reconnect);
6793
6794         if (OCD_HAS_FLAG(data, MAXBYTES))
6795                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6796
6797         /* NB: Disregard the rule against updating
6798          * exp_connect_data.ocd_connect_flags in this case, since
6799          * tgt_client_new() needs to know if this is a lightweight
6800          * connection, and it is safe to expose this flag before
6801          * connection processing completes. */
6802         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6803                 spin_lock(&exp->exp_lock);
6804                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6805                 spin_unlock(&exp->exp_lock);
6806         }
6807
6808         data->ocd_version = LUSTRE_VERSION_CODE;
6809
6810         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6811                 CWARN("%s: MDS requires FID support, but client not\n",
6812                       obd_name);
6813                 return -EBADE;
6814         }
6815
6816         if (OCD_HAS_FLAG(data, PINGLESS)) {
6817                 if (ptlrpc_pinger_suppress_pings()) {
6818                         spin_lock(&exp->exp_obd->obd_dev_lock);
6819                         list_del_init(&exp->exp_obd_chain_timed);
6820                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6821                 } else {
6822                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6823                 }
6824         }
6825
6826         data->ocd_max_easize = mdt->mdt_max_ea_size;
6827
6828         /* NB: Disregard the rule against updating
6829          * exp_connect_data.ocd_connect_flags in this case, since
6830          * tgt_client_new() needs to know if this is client supports
6831          * multiple modify RPCs, and it is safe to expose this flag before
6832          * connection processing completes. */
6833         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6834                 if (mdt_max_mod_rpcs_changed(mdt))
6835                         /* The new mdt.*.max_mod_rpcs_in_flight parameter
6836                          * has not changed since initialization, but the
6837                          * deprecated module parameter was changed,
6838                          * so use that instead.
6839                          */
6840                         data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6841                 else
6842                         data->ocd_maxmodrpcs = mdt->mdt_max_mod_rpcs_in_flight;
6843                 spin_lock(&exp->exp_lock);
6844                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6845                 spin_unlock(&exp->exp_lock);
6846         }
6847
6848         if (OCD_HAS_FLAG(data, CKSUM)) {
6849                 __u32 cksum_types = data->ocd_cksum_types;
6850
6851                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6852
6853                 if (unlikely(data->ocd_cksum_types == 0)) {
6854                         CERROR("%s: Connect with checksum support but no "
6855                                "ocd_cksum_types is set\n",
6856                                exp->exp_obd->obd_name);
6857                         RETURN(-EPROTO);
6858                 }
6859
6860                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6861                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6862                        cksum_types, data->ocd_cksum_types);
6863         } else {
6864                 /* This client does not support OBD_CONNECT_CKSUM
6865                  * fall back to CRC32 */
6866                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6867                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6868                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6869         }
6870
6871         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6872             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6873                 atomic_inc(&mdt->mdt_mds_mds_conns);
6874                 mdt_enable_slc(mdt);
6875         }
6876
6877         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6878                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6879
6880         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
6881             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT)) {
6882                 atomic_inc(&mdt->mdt_dmv_old_client_count);
6883                 mdt->mdt_enable_dmv_implicit_inherit = 0;
6884         }
6885
6886         return 0;
6887 }
6888
6889 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6890                                    struct mdt_thread_info *info,
6891                                    struct mdt_file_data *mfd)
6892 {
6893         struct lu_context ses;
6894         int rc;
6895         ENTRY;
6896
6897         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6898         if (rc)
6899                 RETURN(rc);
6900
6901         env->le_ses = &ses;
6902         lu_context_enter(&ses);
6903
6904         mdt_ucred(info)->uc_valid = UCRED_OLD;
6905         /* do not let rbac interfere with dirty flag internal system event */
6906         mdt_ucred(info)->uc_rbac_file_perms = 1;
6907         mdt_ucred(info)->uc_rbac_dne_ops = 1;
6908         mdt_ucred(info)->uc_rbac_quota_ops = 1;
6909         mdt_ucred(info)->uc_rbac_byfid_ops = 1;
6910         mdt_ucred(info)->uc_rbac_chlg_ops = 1;
6911         mdt_ucred(info)->uc_rbac_fscrypt_admin = 1;
6912         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6913
6914         lu_context_exit(&ses);
6915         lu_context_fini(&ses);
6916         env->le_ses = NULL;
6917
6918         RETURN(rc);
6919 }
6920
6921 static int mdt_export_cleanup(struct obd_export *exp)
6922 {
6923         LIST_HEAD(closing_list);
6924         struct mdt_export_data  *med = &exp->exp_mdt_data;
6925         struct obd_device       *obd = exp->exp_obd;
6926         struct mdt_device       *mdt;
6927         struct mdt_thread_info  *info;
6928         struct lu_env            env;
6929         struct mdt_file_data    *mfd, *n;
6930         int rc = 0;
6931         ENTRY;
6932
6933         spin_lock(&med->med_open_lock);
6934         while (!list_empty(&med->med_open_head)) {
6935                 struct list_head *tmp = med->med_open_head.next;
6936                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6937
6938                 /* Remove mfd handle so it can't be found again.
6939                  * We are consuming the mfd_list reference here. */
6940                 class_handle_unhash(&mfd->mfd_open_handle);
6941                 list_move_tail(&mfd->mfd_list, &closing_list);
6942         }
6943         spin_unlock(&med->med_open_lock);
6944         mdt = mdt_dev(obd->obd_lu_dev);
6945         LASSERT(mdt != NULL);
6946
6947         rc = lu_env_init(&env, LCT_MD_THREAD);
6948         if (rc)
6949                 RETURN(rc);
6950
6951         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6952         LASSERT(info != NULL);
6953         memset(info, 0, sizeof *info);
6954         info->mti_env = &env;
6955         info->mti_mdt = mdt;
6956         info->mti_exp = exp;
6957
6958         if (!list_empty(&closing_list)) {
6959                 struct md_attr *ma = &info->mti_attr;
6960
6961                 /* Close any open files (which may also cause orphan
6962                  * unlinking). */
6963                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6964                         list_del_init(&mfd->mfd_list);
6965                         ma->ma_need = ma->ma_valid = 0;
6966
6967                         /* This file is being closed due to an eviction, it
6968                          * could have been modified and now dirty regarding to
6969                          * HSM archive, check this!
6970                          * The logic here is to mark a file dirty if there's a
6971                          * chance it was dirtied before the client was evicted,
6972                          * so that we don't have to wait for a release attempt
6973                          * before finding out the file was actually dirty and
6974                          * fail the release. Aggressively marking it dirty here
6975                          * will cause the policy engine to attempt to
6976                          * re-archive it; when rearchiving, we can compare the
6977                          * current version to the HSM data_version and make the
6978                          * archive request into a noop if it's not actually
6979                          * dirty.
6980                          */
6981                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6982                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6983
6984                         /* Don't unlink orphan on failover umount, LU-184 */
6985                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6986                             exp->exp_obd->obd_stopping) {
6987                                 ma->ma_valid = MA_FLAGS;
6988                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6989                         }
6990                         ma->ma_valid |= MA_FORCE_LOG;
6991                         mdt_mfd_close(info, mfd);
6992                 }
6993         }
6994         info->mti_mdt = NULL;
6995         /* cleanup client slot early */
6996         /* Do not erase record for recoverable client. */
6997         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6998                 tgt_client_del(&env, exp);
6999         lu_env_fini(&env);
7000
7001         RETURN(rc);
7002 }
7003
7004 static int mdt_obd_disconnect(struct obd_export *exp)
7005 {
7006         struct obd_connect_data *data = &exp->exp_connect_data;
7007         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
7008         int rc;
7009
7010         ENTRY;
7011
7012         LASSERT(exp);
7013         class_export_get(exp);
7014
7015         if (!(exp->exp_flags & OBD_OPT_FORCE))
7016                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7017
7018         if (OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7019             atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
7020                 mdt_disable_slc(mdt);
7021
7022         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7023             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT) &&
7024             atomic_dec_and_test(&mdt->mdt_dmv_old_client_count))
7025                 mdt->mdt_enable_dmv_implicit_inherit = 1;
7026
7027         rc = server_disconnect_export(exp);
7028         if (rc != 0)
7029                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
7030
7031         tgt_grant_discard(exp);
7032
7033         rc = mdt_export_cleanup(exp);
7034         nodemap_del_member(exp);
7035         class_export_put(exp);
7036         RETURN(rc);
7037 }
7038
7039 /* mds_connect copy */
7040 static int mdt_obd_connect(const struct lu_env *env,
7041                            struct obd_export **exp, struct obd_device *obd,
7042                            struct obd_uuid *cluuid,
7043                            struct obd_connect_data *data,
7044                            void *localdata)
7045 {
7046         struct obd_export       *lexp;
7047         struct lustre_handle    conn = { 0 };
7048         struct mdt_device       *mdt;
7049         int                      rc;
7050         struct lnet_nid         *client_nid = localdata;
7051         ENTRY;
7052
7053         LASSERT(env != NULL);
7054         LASSERT(data != NULL);
7055
7056         if (!exp || !obd || !cluuid)
7057                 RETURN(-EINVAL);
7058
7059         mdt = mdt_dev(obd->obd_lu_dev);
7060
7061         /*
7062          * first, check whether the stack is ready to handle requests
7063          * XXX: probably not very appropriate method is used now
7064          *      at some point we should find a better one
7065          */
7066         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
7067             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
7068             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
7069                 rc = obd_get_info(env, mdt->mdt_child_exp,
7070                                   sizeof(KEY_OSP_CONNECTED),
7071                                   KEY_OSP_CONNECTED, NULL, NULL);
7072                 if (rc)
7073                         RETURN(-EAGAIN);
7074                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
7075         }
7076
7077         rc = class_connect(&conn, obd, cluuid);
7078         if (rc)
7079                 RETURN(rc);
7080
7081         lexp = class_conn2export(&conn);
7082         LASSERT(lexp != NULL);
7083
7084         if (nid_is_nid4(client_nid)) {
7085                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), lexp);
7086                 if (rc != 0 && rc != -EEXIST)
7087                         GOTO(out, rc);
7088         }
7089
7090         rc = mdt_connect_internal(env, lexp, mdt, data, false);
7091         if (rc == 0) {
7092                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
7093
7094                 LASSERT(lcd);
7095                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
7096                 rc = tgt_client_new(env, lexp);
7097                 if (rc == 0)
7098                         mdt_export_stats_init(obd, lexp, localdata);
7099         }
7100 out:
7101         if (rc != 0) {
7102                 class_disconnect(lexp);
7103                 nodemap_del_member(lexp);
7104                 *exp = NULL;
7105         } else {
7106                 *exp = lexp;
7107                 /* Because we do not want this export to be evicted by pinger,
7108                  * let's not add this export to the timed chain list. */
7109                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
7110                         spin_lock(&lexp->exp_obd->obd_dev_lock);
7111                         list_del_init(&lexp->exp_obd_chain_timed);
7112                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
7113                 }
7114         }
7115
7116         RETURN(rc);
7117 }
7118
7119 static int mdt_obd_reconnect(const struct lu_env *env,
7120                              struct obd_export *exp, struct obd_device *obd,
7121                              struct obd_uuid *cluuid,
7122                              struct obd_connect_data *data,
7123                              void *localdata)
7124 {
7125         struct lnet_nid *client_nid = localdata;
7126         int rc;
7127         ENTRY;
7128
7129         if (exp == NULL || obd == NULL || cluuid == NULL)
7130                 RETURN(-EINVAL);
7131
7132         if (nid_is_nid4(client_nid)) {
7133                 rc = nodemap_add_member(lnet_nid_to_nid4(client_nid), exp);
7134                 if (rc != 0 && rc != -EEXIST)
7135                         RETURN(rc);
7136         }
7137
7138         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
7139                                   true);
7140         if (rc == 0)
7141                 mdt_export_stats_init(obd, exp, localdata);
7142         else
7143                 nodemap_del_member(exp);
7144
7145         RETURN(rc);
7146 }
7147
7148 /* FIXME: Can we avoid using these two interfaces? */
7149 static int mdt_init_export(struct obd_export *exp)
7150 {
7151         struct mdt_export_data *med = &exp->exp_mdt_data;
7152         int                     rc;
7153         ENTRY;
7154
7155         INIT_LIST_HEAD(&med->med_open_head);
7156         spin_lock_init(&med->med_open_lock);
7157         spin_lock(&exp->exp_lock);
7158         exp->exp_connecting = 1;
7159         spin_unlock(&exp->exp_lock);
7160
7161         OBD_ALLOC(exp->exp_used_slots,
7162                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7163         if (exp->exp_used_slots == NULL)
7164                 RETURN(-ENOMEM);
7165
7166         /* self-export doesn't need client data and ldlm initialization */
7167         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7168                                      &exp->exp_client_uuid)))
7169                 RETURN(0);
7170
7171         rc = tgt_client_alloc(exp);
7172         if (rc)
7173                 GOTO(err, rc);
7174
7175         rc = ldlm_init_export(exp);
7176         if (rc)
7177                 GOTO(err_free, rc);
7178
7179         RETURN(rc);
7180
7181 err_free:
7182         tgt_client_free(exp);
7183 err:
7184         OBD_FREE(exp->exp_used_slots,
7185                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7186         exp->exp_used_slots = NULL;
7187
7188         CERROR("%s: Failed to initialize export: rc = %d\n",
7189                exp->exp_obd->obd_name, rc);
7190         return rc;
7191 }
7192
7193 static int mdt_destroy_export(struct obd_export *exp)
7194 {
7195         ENTRY;
7196
7197         target_destroy_export(exp);
7198         if (exp->exp_used_slots)
7199                 OBD_FREE(exp->exp_used_slots,
7200                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7201
7202         /* destroy can be called from failed obd_setup, so
7203          * checking uuid is safer than obd_self_export */
7204         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7205                                      &exp->exp_client_uuid)))
7206                 RETURN(0);
7207
7208         ldlm_destroy_export(exp);
7209         tgt_client_free(exp);
7210
7211         LASSERT(list_empty(&exp->exp_outstanding_replies));
7212         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
7213
7214         /*
7215          * discard grants once we're sure no more
7216          * interaction with the client is possible
7217          */
7218         tgt_grant_discard(exp);
7219         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
7220                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
7221
7222         if (!(exp->exp_flags & OBD_OPT_FORCE))
7223                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7224
7225         RETURN(0);
7226 }
7227
7228 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
7229                    struct linkea_data *ldata)
7230 {
7231         int rc;
7232
7233         LASSERT(ldata->ld_buf->lb_buf != NULL);
7234
7235         if (!mdt_object_exists(mdt_obj))
7236                 return -ENODATA;
7237
7238         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7239                           ldata->ld_buf, XATTR_NAME_LINK);
7240         if (rc == -ERANGE) {
7241                 /* Buf was too small, figure out what we need. */
7242                 lu_buf_free(ldata->ld_buf);
7243                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7244                                   ldata->ld_buf, XATTR_NAME_LINK);
7245                 if (rc < 0)
7246                         return rc;
7247                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7248                 if (ldata->ld_buf->lb_buf == NULL)
7249                         return -ENOMEM;
7250                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7251                                   ldata->ld_buf, XATTR_NAME_LINK);
7252         }
7253         if (rc < 0)
7254                 return rc;
7255
7256         return linkea_init_with_rec(ldata);
7257 }
7258
7259 /**
7260  * Given an MDT object, try to look up the full path to the object.
7261  * Part of the MDT layer implementation of lfs fid2path.
7262  *
7263  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7264  * \param[in]     obj   Object to do path lookup of
7265  * \param[in,out] fp    User-provided struct to store path information
7266  * \param[in]     root_fid Root FID of current path should reach
7267  *
7268  * \retval 0 Lookup successful, path information stored in fp
7269  * \retval -EAGAIN Lookup failed, usually because object is being moved
7270  * \retval negative errno if there was a problem
7271  */
7272 static int mdt_path_current(struct mdt_thread_info *info,
7273                             struct mdt_object *obj,
7274                             struct getinfo_fid2path *fp,
7275                             struct lu_fid *root_fid)
7276 {
7277         struct mdt_device *mdt = info->mti_mdt;
7278         struct lu_name *tmpname = &info->mti_name;
7279         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7280         struct lu_buf *buf = &info->mti_big_buf;
7281         struct linkea_data ldata = { NULL };
7282         bool first = true;
7283         struct mdt_object *mdt_obj;
7284         struct link_ea_header *leh;
7285         struct link_ea_entry *lee;
7286         bool worthchecking = true;
7287         bool needsfid = false;
7288         bool supported = false;
7289         int isenc = -1;
7290         char *ptr;
7291         int reclen;
7292         int rc = 0;
7293
7294         ENTRY;
7295
7296         /* temp buffer for path element, the buffer will be finally freed
7297          * in mdt_thread_info_fini */
7298         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7299         if (buf->lb_buf == NULL)
7300                 RETURN(-ENOMEM);
7301
7302         ldata.ld_buf = buf;
7303         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7304         *ptr = 0;
7305         --ptr;
7306         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7307
7308         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7309                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7310                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7311                         GOTO(out, rc = -ENOENT);
7312
7313                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7314                         mdt_obj = obj;
7315                         mdt_object_get(info->mti_env, mdt_obj);
7316                 } else {
7317                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7318                         if (IS_ERR(mdt_obj))
7319                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7320                 }
7321
7322                 if (!mdt_object_exists(mdt_obj)) {
7323                         mdt_object_put(info->mti_env, mdt_obj);
7324                         GOTO(out, rc = -ENOENT);
7325                 }
7326
7327                 if (mdt_object_remote(mdt_obj)) {
7328                         mdt_object_put(info->mti_env, mdt_obj);
7329                         GOTO(remote_out, rc = -EREMOTE);
7330                 }
7331
7332                 if (worthchecking) {
7333                         /* we need to know if the FID being
7334                          * looked up is encrypted
7335                          */
7336                         struct lu_attr la = { 0 };
7337                         struct dt_object *dt = mdt_obj2dt(mdt_obj);
7338
7339                         if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7340                                 dt_attr_get(info->mti_env, dt, &la);
7341                         if (la.la_valid & LA_FLAGS &&
7342                             la.la_flags & LUSTRE_ENCRYPT_FL) {
7343                                 if (!supported && mdt_info_req(info) &&
7344                                     !exp_connect_encrypt_fid2path(
7345                                             mdt_info_req(info)->rq_export)) {
7346                                         /* client does not support fid2path
7347                                          * for encrypted files
7348                                          */
7349                                         mdt_object_put(info->mti_env, mdt_obj);
7350                                         GOTO(out, rc = -ENODATA);
7351                                 } else {
7352                                         supported = true;
7353                                 }
7354                                 needsfid = true;
7355                                 if (isenc == -1)
7356                                         isenc = 1;
7357                         } else {
7358                                 worthchecking = false;
7359                                 needsfid = false;
7360                                 if (isenc == -1)
7361                                         isenc = 0;
7362                         }
7363                 }
7364
7365                 rc = mdt_links_read(info, mdt_obj, &ldata);
7366                 if (rc != 0) {
7367                         mdt_object_put(info->mti_env, mdt_obj);
7368                         GOTO(out, rc);
7369                 }
7370
7371                 leh = buf->lb_buf;
7372                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7373                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7374                 /* If set, use link #linkno for path lookup, otherwise use
7375                    link #0.  Only do this for the final path element. */
7376                 if (first && fp->gf_linkno < leh->leh_reccount) {
7377                         int count;
7378                         for (count = 0; count < fp->gf_linkno; count++) {
7379                                 lee = (struct link_ea_entry *)
7380                                      ((char *)lee + reclen);
7381                                 linkea_entry_unpack(lee, &reclen, tmpname,
7382                                                     tmpfid);
7383                         }
7384                         if (fp->gf_linkno < leh->leh_reccount - 1)
7385                                 /* indicate to user there are more links */
7386                                 fp->gf_linkno++;
7387                 }
7388
7389                 /* Check if it is slave stripes */
7390                 rc = mdt_is_dir_stripe(info, mdt_obj);
7391                 mdt_object_put(info->mti_env, mdt_obj);
7392                 if (rc < 0)
7393                         GOTO(out, rc);
7394                 if (rc == 1) {
7395                         fp->gf_fid = *tmpfid;
7396                         continue;
7397                 }
7398
7399                 /* Pack the name in the end of the buffer */
7400                 ptr -= tmpname->ln_namelen;
7401                 if (ptr - 1 <= fp->gf_u.gf_path)
7402                         GOTO(out, rc = -ENAMETOOLONG);
7403                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7404                 if (needsfid) {
7405                         /* Pack FID before file name, so that client can build
7406                          * encoded/digested form.
7407                          */
7408                         char fidstr[FID_LEN + 1];
7409
7410                         snprintf(fidstr, sizeof(fidstr), DFID,
7411                                  PFID(&fp->gf_fid));
7412                         ptr -= strlen(fidstr);
7413                         if (ptr - 1 <= fp->gf_u.gf_path)
7414                                 GOTO(out, rc = -ENAMETOOLONG);
7415                         strncpy(ptr, fidstr, strlen(fidstr));
7416                 }
7417                 *(--ptr) = '/';
7418
7419                 /* keep the last resolved fid to the client, so the
7420                  * client will build the left path on another MDT for
7421                  * remote object */
7422                 fp->gf_fid = *tmpfid;
7423
7424                 first = false;
7425         }
7426
7427         /* non-zero will be treated as an error */
7428         rc = 0;
7429
7430 remote_out:
7431         if (isenc != 1)
7432                 ptr++; /* skip leading / unless this is an encrypted file */
7433         memmove(fp->gf_u.gf_path, ptr,
7434                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7435
7436 out:
7437         RETURN(rc);
7438 }
7439
7440 /**
7441  * Given an MDT object, use mdt_path_current to get the path.
7442  * Essentially a wrapper to retry mdt_path_current a set number of times
7443  * if -EAGAIN is returned (usually because an object is being moved).
7444  *
7445  * Part of the MDT layer implementation of lfs fid2path.
7446  *
7447  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7448  * \param[in]     obj   Object to do path lookup of
7449  * \param[in,out] fp    User-provided struct for arguments and to store path
7450  *                      information
7451  *
7452  * \retval 0 Lookup successful, path information stored in fp
7453  * \retval negative errno if there was a problem
7454  */
7455 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7456                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7457 {
7458         struct mdt_device       *mdt = info->mti_mdt;
7459         int                     tries = 3;
7460         int                     rc = -EAGAIN;
7461         ENTRY;
7462
7463         if (fp->gf_pathlen < 3)
7464                 RETURN(-EOVERFLOW);
7465
7466         if (root_fid == NULL)
7467                 root_fid = &mdt->mdt_md_root_fid;
7468
7469         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7470                 fp->gf_u.gf_path[0] = '\0';
7471                 RETURN(0);
7472         }
7473
7474         /* Retry multiple times in case file is being moved */
7475         while (tries-- && rc == -EAGAIN)
7476                 rc = mdt_path_current(info, obj, fp, root_fid);
7477
7478         RETURN(rc);
7479 }
7480
7481 /**
7482  * Get the full path of the provided FID, as of changelog record recno.
7483  *
7484  * This checks sanity and looks up object for user provided FID
7485  * before calling the actual path lookup code.
7486  *
7487  * Part of the MDT layer implementation of lfs fid2path.
7488  *
7489  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7490  * \param[in,out] fp    User-provided struct for arguments and to store path
7491  *                      information
7492  *
7493  * \retval 0 Lookup successful, path information and recno stored in fp
7494  * \retval -ENOENT, object does not exist
7495  * \retval negative errno if there was a problem
7496  */
7497 static int mdt_fid2path(struct mdt_thread_info *info,
7498                         struct lu_fid *root_fid,
7499                         struct getinfo_fid2path *fp)
7500 {
7501         struct mdt_device *mdt = info->mti_mdt;
7502         struct mdt_object *obj;
7503         int    rc;
7504         ENTRY;
7505
7506         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7507                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7508
7509         if (!fid_is_sane(&fp->gf_fid))
7510                 RETURN(-EINVAL);
7511
7512         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7513                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7514                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7515                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7516                 RETURN(-EINVAL);
7517         }
7518
7519         /* return error if client-provided root fid
7520          * is not the one stored in the export
7521          */
7522         if (root_fid && !fid_is_zero(&info->mti_exp->exp_root_fid) &&
7523             !lu_fid_eq(root_fid, &info->mti_exp->exp_root_fid)) {
7524                 CDEBUG(D_INFO,
7525                        "%s: root fid from client "DFID" but "DFID" stored in export\n",
7526                        mdt_obd_name(mdt), PFID(root_fid),
7527                        PFID(&info->mti_exp->exp_root_fid));
7528                 RETURN(-EXDEV);
7529         }
7530
7531         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7532         if (IS_ERR(obj)) {
7533                 rc = PTR_ERR(obj);
7534                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7535                        PFID(&fp->gf_fid), rc);
7536                 RETURN(rc);
7537         }
7538
7539         if (mdt_object_remote(obj))
7540                 rc = -EREMOTE;
7541         else if (!mdt_object_exists(obj))
7542                 rc = -ENOENT;
7543         else
7544                 rc = 0;
7545
7546         if (rc < 0) {
7547                 mdt_object_put(info->mti_env, obj);
7548                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7549                        PFID(&fp->gf_fid), rc);
7550                 RETURN(rc);
7551         }
7552
7553         rc = mdt_path(info, obj, fp, root_fid);
7554
7555         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7556                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7557                fp->gf_recno, fp->gf_linkno);
7558
7559         mdt_object_put(info->mti_env, obj);
7560
7561         RETURN(rc);
7562 }
7563
7564 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7565                             void *val, int vallen)
7566 {
7567         struct getinfo_fid2path *fpout, *fpin;
7568         struct lu_fid *root_fid = NULL;
7569         int rc = 0;
7570
7571         fpin = key + round_up(sizeof(KEY_FID2PATH), 8);
7572         fpout = val;
7573
7574         if (req_capsule_req_need_swab(info->mti_pill))
7575                 lustre_swab_fid2path(fpin);
7576
7577         memcpy(fpout, fpin, sizeof(*fpin));
7578         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7579                 RETURN(-EINVAL);
7580
7581         if (keylen >= round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*fpin) +
7582                       sizeof(struct lu_fid)) {
7583                 /* client sent its root FID, which is normally fileset FID */
7584                 root_fid = fpin->gf_u.gf_root_fid;
7585                 if (req_capsule_req_need_swab(info->mti_pill))
7586                         lustre_swab_lu_fid(root_fid);
7587
7588                 if (root_fid != NULL && !fid_is_sane(root_fid))
7589                         RETURN(-EINVAL);
7590         }
7591
7592         rc = mdt_fid2path(info, root_fid, fpout);
7593         RETURN(rc);
7594 }
7595
7596 int mdt_get_info(struct tgt_session_info *tsi)
7597 {
7598         char    *key;
7599         int      keylen;
7600         __u32   *vallen;
7601         void    *valout;
7602         int      rc;
7603
7604         ENTRY;
7605
7606         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7607         if (key == NULL) {
7608                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7609                 RETURN(err_serious(-EFAULT));
7610         }
7611         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7612                                       RCL_CLIENT);
7613
7614         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7615         if (vallen == NULL) {
7616                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7617                                 tgt_name(tsi->tsi_tgt));
7618                 RETURN(err_serious(-EFAULT));
7619         }
7620
7621         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7622                              *vallen);
7623         rc = req_capsule_server_pack(tsi->tsi_pill);
7624         if (rc)
7625                 RETURN(err_serious(rc));
7626
7627         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7628         if (valout == NULL) {
7629                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7630                                 tgt_name(tsi->tsi_tgt));
7631                 RETURN(err_serious(-EFAULT));
7632         }
7633
7634         if (KEY_IS(KEY_FID2PATH)) {
7635                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7636
7637                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7638                 mdt_thread_info_fini(info);
7639         } else {
7640                 rc = -EINVAL;
7641         }
7642         RETURN(rc);
7643 }
7644
7645 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7646 {
7647         struct obd_ioctl_data *data = karg;
7648         struct lu_fid *fid;
7649         __u64 version;
7650         struct mdt_object *obj;
7651         struct mdt_lock_handle  *lh;
7652         int rc;
7653         ENTRY;
7654
7655         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7656             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7657                 RETURN(-EINVAL);
7658
7659         fid = (struct lu_fid *)data->ioc_inlbuf1;
7660
7661         if (!fid_is_sane(fid))
7662                 RETURN(-EINVAL);
7663
7664         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7665
7666         lh = &mti->mti_lh[MDT_LH_PARENT];
7667         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE, LCK_CR);
7668         if (IS_ERR(obj))
7669                 RETURN(PTR_ERR(obj));
7670
7671         if (mdt_object_remote(obj)) {
7672                 rc = -EREMOTE;
7673                 /**
7674                  * before calling version get the correct MDS should be
7675                  * fid, this is error to find remote object here
7676                  */
7677                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7678         } else if (!mdt_object_exists(obj)) {
7679                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7680                 rc = -ENOENT;
7681         } else {
7682                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7683                *(__u64 *)data->ioc_inlbuf2 = version;
7684                 rc = 0;
7685         }
7686         mdt_object_unlock_put(mti, obj, lh, 1);
7687         RETURN(rc);
7688 }
7689
7690 /* ioctls on obd dev */
7691 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7692                          void *karg, void __user *uarg)
7693 {
7694         struct obd_device *obd = exp->exp_obd;
7695         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7696         struct dt_device *dt = mdt->mdt_bottom;
7697         struct obd_ioctl_data *data;
7698         struct lu_env env;
7699         int rc;
7700
7701         ENTRY;
7702         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
7703                obd->obd_name, cmd, len, karg, uarg);
7704
7705         rc = lu_env_init(&env, LCT_MD_THREAD);
7706         if (rc)
7707                 RETURN(rc);
7708
7709         /* handle commands that don't use @karg first */
7710         switch (cmd) {
7711         case OBD_IOC_SYNC:
7712                 rc = mdt_device_sync(&env, mdt);
7713                 GOTO(out, rc);
7714         case OBD_IOC_SET_READONLY:
7715                 rc = dt_sync(&env, dt);
7716                 if (rc == 0)
7717                         rc = dt_ro(&env, dt);
7718                 GOTO(out, rc);
7719         }
7720
7721         if (unlikely(karg == NULL)) {
7722                 OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc = -EINVAL);
7723                 GOTO(out, rc);
7724         }
7725         data = karg;
7726
7727         switch (cmd) {
7728         case OBD_IOC_ABORT_RECOVERY: {
7729                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7730                         LCONSOLE_WARN("%s: Aborting MDT recovery\n",
7731                                       obd->obd_name);
7732                         obd->obd_abort_mdt_recovery = 1;
7733                         wake_up(&obd->obd_next_transno_waitq);
7734                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7735                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7736                         LCONSOLE_WARN("%s: Aborting client recovery\n",
7737                                       obd->obd_name);
7738                         obd->obd_abort_recovery = 1;
7739                         target_stop_recovery_thread(obd);
7740                 }
7741                 rc = 0;
7742                 break;
7743         }
7744         case OBD_IOC_CHANGELOG_REG:
7745         case OBD_IOC_CHANGELOG_DEREG:
7746         case OBD_IOC_CHANGELOG_CLEAR:
7747         case OBD_IOC_LLOG_PRINT:
7748         case OBD_IOC_LLOG_CANCEL:
7749                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env, mdt->mdt_child,
7750                                                            cmd, len, karg);
7751                 break;
7752         case OBD_IOC_START_LFSCK: {
7753                 struct md_device *next = mdt->mdt_child;
7754                 struct lfsck_start_param lsp;
7755
7756                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7757                 lsp.lsp_index_valid = 0;
7758                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7759                 break;
7760         }
7761         case OBD_IOC_STOP_LFSCK: {
7762                 struct md_device *next = mdt->mdt_child;
7763                 struct lfsck_stop stop;
7764
7765                 stop.ls_status = LS_STOPPED;
7766                 /* Old lfsck utils may pass NULL @stop. */
7767                 if (data->ioc_inlbuf1 == NULL)
7768                         stop.ls_flags = 0;
7769                 else
7770                         stop.ls_flags =
7771                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7772
7773                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7774                 break;
7775         }
7776         case OBD_IOC_QUERY_LFSCK: {
7777                 struct md_device *next = mdt->mdt_child;
7778
7779                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7780                                                  data->ioc_inlbuf1);
7781                 break;
7782         }
7783         case OBD_IOC_GET_OBJ_VERSION: {
7784                 struct mdt_thread_info *mti;
7785
7786                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7787                 memset(mti, 0, sizeof(*mti));
7788                 mti->mti_env = &env;
7789                 mti->mti_mdt = mdt;
7790                 mti->mti_exp = exp;
7791
7792                 rc = mdt_ioc_version_get(mti, karg);
7793                 break;
7794         }
7795         case OBD_IOC_CATLOGLIST: {
7796                 struct mdt_thread_info *mti;
7797
7798                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7799                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7800                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7801                                        &mti->mti_tmp_fid1);
7802                 break;
7803         }
7804         default:
7805                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
7806                 break;
7807         }
7808 out:
7809         lu_env_fini(&env);
7810         RETURN(rc);
7811 }
7812
7813 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7814 {
7815         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7816         int rc;
7817         ENTRY;
7818
7819         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7820                 struct lfsck_start_param lsp;
7821
7822                 lsp.lsp_start = NULL;
7823                 lsp.lsp_index_valid = 0;
7824                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7825                                                            OBD_IOC_START_LFSCK,
7826                                                            0, &lsp);
7827                 if (rc != 0 && rc != -EALREADY)
7828                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7829                               mdt_obd_name(mdt), rc);
7830         }
7831
7832         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7833         RETURN(rc);
7834 }
7835
7836 static int mdt_obd_postrecov(struct obd_device *obd)
7837 {
7838         struct lu_env env;
7839         int rc;
7840
7841         rc = lu_env_init(&env, LCT_MD_THREAD);
7842         if (rc)
7843                 RETURN(rc);
7844         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7845         lu_env_fini(&env);
7846         return rc;
7847 }
7848
7849 static const struct obd_ops mdt_obd_device_ops = {
7850         .o_owner          = THIS_MODULE,
7851         .o_set_info_async = mdt_obd_set_info_async,
7852         .o_connect        = mdt_obd_connect,
7853         .o_reconnect      = mdt_obd_reconnect,
7854         .o_disconnect     = mdt_obd_disconnect,
7855         .o_init_export    = mdt_init_export,
7856         .o_destroy_export = mdt_destroy_export,
7857         .o_iocontrol      = mdt_iocontrol,
7858         .o_postrecov      = mdt_obd_postrecov,
7859         /* Data-on-MDT IO methods */
7860         .o_preprw         = mdt_obd_preprw,
7861         .o_commitrw       = mdt_obd_commitrw,
7862 };
7863
7864 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7865                                          struct lu_device *d)
7866 {
7867         struct mdt_device *m = mdt_dev(d);
7868         ENTRY;
7869
7870         mdt_fini(env, m);
7871         RETURN(NULL);
7872 }
7873
7874 static struct lu_device *mdt_device_free(const struct lu_env *env,
7875                                          struct lu_device *d)
7876 {
7877         struct mdt_device *m = mdt_dev(d);
7878         ENTRY;
7879
7880         lu_device_fini(&m->mdt_lu_dev);
7881         OBD_FREE_PTR(m);
7882
7883         RETURN(NULL);
7884 }
7885
7886 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7887                                           struct lu_device_type *t,
7888                                           struct lustre_cfg *cfg)
7889 {
7890         struct lu_device  *l;
7891         struct mdt_device *m;
7892
7893         OBD_ALLOC_PTR(m);
7894         if (m != NULL) {
7895                 int rc;
7896
7897                 l = &m->mdt_lu_dev;
7898                 rc = mdt_init0(env, m, t, cfg);
7899                 if (rc != 0) {
7900                         mdt_device_free(env, l);
7901                         l = ERR_PTR(rc);
7902                         return l;
7903                 }
7904         } else
7905                 l = ERR_PTR(-ENOMEM);
7906         return l;
7907 }
7908
7909 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7910 LU_KEY_INIT(mdt, struct mdt_thread_info);
7911
7912 static void mdt_key_fini(const struct lu_context *ctx,
7913                          struct lu_context_key *key, void* data)
7914 {
7915         struct mdt_thread_info *info = data;
7916
7917         if (info->mti_big_lmm) {
7918                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7919                 info->mti_big_lmm = NULL;
7920                 info->mti_big_lmmsize = 0;
7921         }
7922
7923         if (info->mti_big_acl) {
7924                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7925                 info->mti_big_acl = NULL;
7926                 info->mti_big_aclsize = 0;
7927         }
7928
7929         OBD_FREE_PTR(info);
7930 }
7931
7932 /* context key: mdt_thread_key */
7933 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7934
7935 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7936 {
7937         return lu_ucred(info->mti_env);
7938 }
7939
7940 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7941 {
7942         return lu_ucred_check(info->mti_env);
7943 }
7944
7945 /**
7946  * Enable/disable COS (Commit On Sharing).
7947  *
7948  * Set/Clear the COS flag in mdt options.
7949  *
7950  * \param mdt mdt device
7951  * \param val 0 disables COS, other values enable COS
7952  */
7953 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7954 {
7955         struct lu_env env;
7956         int rc;
7957
7958         mdt->mdt_opts.mo_cos = val;
7959         rc = lu_env_init(&env, LCT_LOCAL);
7960         if (unlikely(rc != 0)) {
7961                 CWARN("%s: lu_env initialization failed, cannot "
7962                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7963                 return;
7964         }
7965         mdt_device_sync(&env, mdt);
7966         lu_env_fini(&env);
7967 }
7968
7969 /**
7970  * Check COS (Commit On Sharing) status.
7971  *
7972  * Return COS flag status.
7973  *
7974  * \param mdt mdt device
7975  */
7976 int mdt_cos_is_enabled(struct mdt_device *mdt)
7977 {
7978         return mdt->mdt_opts.mo_cos != 0;
7979 }
7980
7981 static const struct lu_device_type_operations mdt_device_type_ops = {
7982         .ldto_device_alloc = mdt_device_alloc,
7983         .ldto_device_free  = mdt_device_free,
7984         .ldto_device_fini  = mdt_device_fini
7985 };
7986
7987 static struct lu_device_type mdt_device_type = {
7988         .ldt_tags     = LU_DEVICE_MD,
7989         .ldt_name     = LUSTRE_MDT_NAME,
7990         .ldt_ops      = &mdt_device_type_ops,
7991         .ldt_ctx_tags = LCT_MD_THREAD
7992 };
7993
7994 static int __init mdt_init(void)
7995 {
7996         int rc;
7997
7998         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7999                      FID_NOBRACE_LEN + 1);
8000         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
8001                      FID_LEN + 1);
8002
8003         rc = libcfs_setup();
8004         if (rc)
8005                 return rc;
8006
8007         rc = lu_kmem_init(mdt_caches);
8008         if (rc)
8009                 return rc;
8010
8011         rc = mds_mod_init();
8012         if (rc)
8013                 GOTO(lu_fini, rc);
8014
8015         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
8016                                  LUSTRE_MDT_NAME, &mdt_device_type);
8017         if (rc)
8018                 GOTO(mds_fini, rc);
8019 lu_fini:
8020         if (rc)
8021                 lu_kmem_fini(mdt_caches);
8022 mds_fini:
8023         if (rc)
8024                 mds_mod_exit();
8025         return rc;
8026 }
8027
8028 static void __exit mdt_exit(void)
8029 {
8030         class_unregister_type(LUSTRE_MDT_NAME);
8031         mds_mod_exit();
8032         lu_kmem_fini(mdt_caches);
8033 }
8034
8035 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8036 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
8037 MODULE_VERSION(LUSTRE_VERSION_STRING);
8038 MODULE_LICENSE("GPL");
8039
8040 module_init(mdt_init);
8041 module_exit(mdt_exit);