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  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         min_t(unsigned int,
 602                                                               mdt->mdt_max_ea_size,
 603                                                               XATTR_SIZE_MAX);
 604                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 605                                                         info->mti_big_aclsize);
 606                                         if (info->mti_big_acl == NULL) {
 607                                                 info->mti_big_aclsize = 0;
 608                                                 CERROR("%s: unable to grow "
 609                                                        DFID" ACL buffer\n",
 610                                                        mdt_obd_name(mdt),
 611                                                        PFID(mdt_object_fid(o)));
 612                                                 RETURN(-ENOMEM);
 613                                         }
 614                                 }
 615
 616                                 CDEBUG(D_INODE, "%s: grow the "DFID
 617                                        " ACL buffer to size %d\n",
 618                                        mdt_obd_name(mdt),
 619                                        PFID(mdt_object_fid(o)),
 620                                        info->mti_big_aclsize);
 621
 622                                 buf->lb_buf = info->mti_big_acl;
 623                                 buf->lb_len = info->mti_big_aclsize;
 624
 625                                 goto again;
 626                         }
 627
 628                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 629                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 630                 }
 631         } else {
 632                 int client;
 633                 int server;
 634                 int acl_buflen;
 635                 int lmm_buflen = 0;
 636                 int lmmsize = 0;
 637
 638                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 639                 if (acl_buflen >= rc)
 640                         goto map;
 641
 642                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 643                 client = ptlrpc_req_get_repsize(pill->rc_req);
 644                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 645                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 646                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 647                                                           RCL_SERVER);
 648                         lmmsize = repbody->mbo_eadatasize;
 649                 }
 650
 651                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 652                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 653                                "is not big enough with the ACL size %d (%d)\n",
 654                                mdt_obd_name(mdt), client, rc,
 655                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 656                         repbody->mbo_aclsize = 0;
 657                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 658                         RETURN(-ERANGE);
 659                 }
 660
 661 map:
 662                 if (buf->lb_buf == info->mti_big_acl)
 663                         info->mti_big_acl_used = 1;
 664
 665                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 666                                      rc, NODEMAP_FS_TO_CLIENT);
 667                 /* if all ACLs mapped out, rc is still >= 0 */
 668                 if (rc < 0) {
 669                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 670                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 671                                PFID(mdt_object_fid(o)), rc);
 672                         repbody->mbo_aclsize = 0;
 673                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 674                 } else {
 675                         repbody->mbo_aclsize = rc;
 676                         repbody->mbo_valid |= OBD_MD_FLACL;
 677                         rc = 0;
 678                 }
 679         }
 680
 681         RETURN(rc);
 682 }
 683 #endif
 684
 685 /* XXX Look into layout in MDT layer. */
 686 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 687 {
 688         struct lov_comp_md_v1   *comp_v1;
 689         struct lov_mds_md       *v1;
 690         int                      i;
 691
 692         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 693                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 694
 695                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 696                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 697                                 comp_v1->lcm_entries[i].lcme_offset);
 698                         /* We don't support partial release for now */
 699                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 700                                 return false;
 701                 }
 702                 return true;
 703         } else {
 704                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 705                         true : false;
 706         }
 707 }
 708
 709 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 710                         const struct lu_attr *attr, const struct lu_fid *fid)
 711 {
 712         struct md_attr *ma = &info->mti_attr;
 713         struct obd_export *exp = info->mti_exp;
 714         struct lu_nodemap *nodemap = NULL;
 715
 716         LASSERT(ma->ma_valid & MA_INODE);
 717
 718         if (attr->la_valid & LA_ATIME) {
 719                 b->mbo_atime = attr->la_atime;
 720                 b->mbo_valid |= OBD_MD_FLATIME;
 721         }
 722         if (attr->la_valid & LA_MTIME) {
 723                 b->mbo_mtime = attr->la_mtime;
 724                 b->mbo_valid |= OBD_MD_FLMTIME;
 725         }
 726         if (attr->la_valid & LA_CTIME) {
 727                 b->mbo_ctime = attr->la_ctime;
 728                 b->mbo_valid |= OBD_MD_FLCTIME;
 729         }
 730         if (attr->la_valid & LA_FLAGS) {
 731                 b->mbo_flags = attr->la_flags;
 732                 b->mbo_valid |= OBD_MD_FLFLAGS;
 733         }
 734         if (attr->la_valid & LA_NLINK) {
 735                 b->mbo_nlink = attr->la_nlink;
 736                 b->mbo_valid |= OBD_MD_FLNLINK;
 737         }
 738         if (attr->la_valid & (LA_UID|LA_GID)) {
 739                 nodemap = nodemap_get_from_exp(exp);
 740                 if (IS_ERR(nodemap))
 741                         goto out;
 742         }
 743         if (attr->la_valid & LA_UID) {
 744                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 745                                             NODEMAP_FS_TO_CLIENT,
 746                                             attr->la_uid);
 747                 b->mbo_valid |= OBD_MD_FLUID;
 748         }
 749         if (attr->la_valid & LA_GID) {
 750                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 751                                             NODEMAP_FS_TO_CLIENT,
 752                                             attr->la_gid);
 753                 b->mbo_valid |= OBD_MD_FLGID;
 754         }
 755
 756         if (attr->la_valid & LA_PROJID) {
 757                 /* TODO, nodemap for project id */
 758                 b->mbo_projid = attr->la_projid;
 759                 b->mbo_valid |= OBD_MD_FLPROJID;
 760         }
 761
 762         b->mbo_mode = attr->la_mode;
 763         if (attr->la_valid & LA_MODE)
 764                 b->mbo_valid |= OBD_MD_FLMODE;
 765         if (attr->la_valid & LA_TYPE)
 766                 b->mbo_valid |= OBD_MD_FLTYPE;
 767
 768         if (fid != NULL) {
 769                 b->mbo_fid1 = *fid;
 770                 b->mbo_valid |= OBD_MD_FLID;
 771                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 772                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 773         }
 774
 775         if (!(attr->la_valid & LA_TYPE))
 776                 return;
 777
 778         b->mbo_rdev   = attr->la_rdev;
 779         b->mbo_size   = attr->la_size;
 780         b->mbo_blocks = attr->la_blocks;
 781
 782         if (!S_ISREG(attr->la_mode)) {
 783                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 784         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 785                 /* means no objects are allocated on osts. */
 786                 LASSERT(!(ma->ma_valid & MA_LOV));
 787                 /* just ignore blocks occupied by extend attributes on MDS */
 788                 b->mbo_blocks = 0;
 789                 /* if no object is allocated on osts, the size on mds is valid.
 790                  * b=22272 */
 791                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 792         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 793                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 794                         /* A released file stores its size on MDS. */
 795                         /* But return 1 block for released file, unless tools
 796                          * like tar will consider it fully sparse. (LU-3864)
 797                          */
 798                         if (unlikely(b->mbo_size == 0))
 799                                 b->mbo_blocks = 0;
 800                         else
 801                                 b->mbo_blocks = 1;
 802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 803                 } else if (info->mti_som_valid) { /* som is valid */
 804                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 805                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 806                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 807                         b->mbo_size = ma->ma_som.ms_size;
 808                         b->mbo_blocks = ma->ma_som.ms_blocks;
 809                 }
 810         }
 811
 812         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 813                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 814                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 815                        PFID(fid), (unsigned long long)b->mbo_size);
 816
 817 out:
 818         if (!IS_ERR_OR_NULL(nodemap))
 819                 nodemap_putref(nodemap);
 820 }
 821
 822 static inline int mdt_body_has_lov(const struct lu_attr *la,
 823                                    const struct mdt_body *body)
 824 {
 825         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 826                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 827 }
 828
 829 void mdt_client_compatibility(struct mdt_thread_info *info)
 830 {
 831         struct mdt_body       *body;
 832         struct ptlrpc_request *req = mdt_info_req(info);
 833         struct obd_export     *exp = req->rq_export;
 834         struct md_attr        *ma = &info->mti_attr;
 835         struct lu_attr        *la = &ma->ma_attr;
 836         ENTRY;
 837
 838         if (exp_connect_layout(exp))
 839                 /* the client can deal with 16-bit lmm_stripe_count */
 840                 RETURN_EXIT;
 841
 842         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 843
 844         if (!mdt_body_has_lov(la, body))
 845                 RETURN_EXIT;
 846
 847         /* now we have a reply with a lov for a client not compatible with the
 848          * layout lock so we have to clean the layout generation number */
 849         if (S_ISREG(la->la_mode))
 850                 ma->ma_lmm->lmm_layout_gen = 0;
 851         EXIT;
 852 }
 853
 854 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 855                                    struct mdt_object *o)
 856 {
 857         const struct lu_env *env = info->mti_env;
 858         int rc, rc2;
 859
 860         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 861                           XATTR_NAME_LOV);
 862
 863         if (rc == -ENODATA)
 864                 rc = 0;
 865
 866         if (rc < 0)
 867                 goto out;
 868
 869         /* Is it a directory? Let's check for the LMV as well */
 870         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 871                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 872                                    XATTR_NAME_LMV);
 873
 874                 if (rc2 == -ENODATA)
 875                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 876                                            &LU_BUF_NULL,
 877                                            XATTR_NAME_DEFAULT_LMV);
 878
 879                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 880                         rc = rc2;
 881         }
 882
 883 out:
 884         return rc;
 885 }
 886
 887 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 888                       const char *name)
 889 {
 890         const struct lu_env *env = info->mti_env;
 891         int rc;
 892         ENTRY;
 893
 894         LASSERT(info->mti_big_lmm_used == 0);
 895         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 896         if (rc < 0)
 897                 RETURN(rc);
 898
 899         /* big_lmm may need to be grown */
 900         if (info->mti_big_lmmsize < rc) {
 901                 int size = size_roundup_power2(rc);
 902
 903                 if (info->mti_big_lmmsize > 0) {
 904                         /* free old buffer */
 905                         LASSERT(info->mti_big_lmm);
 906                         OBD_FREE_LARGE(info->mti_big_lmm,
 907                                        info->mti_big_lmmsize);
 908                         info->mti_big_lmm = NULL;
 909                         info->mti_big_lmmsize = 0;
 910                 }
 911
 912                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 913                 if (info->mti_big_lmm == NULL)
 914                         RETURN(-ENOMEM);
 915                 info->mti_big_lmmsize = size;
 916         }
 917         LASSERT(info->mti_big_lmmsize >= rc);
 918
 919         info->mti_buf.lb_buf = info->mti_big_lmm;
 920         info->mti_buf.lb_len = info->mti_big_lmmsize;
 921         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 922
 923         RETURN(rc);
 924 }
 925
 926 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 927                    struct md_attr *ma, const char *name)
 928 {
 929         struct md_object *next = mdt_object_child(o);
 930         struct lu_buf    *buf = &info->mti_buf;
 931         int rc;
 932
 933         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 934                 buf->lb_buf = ma->ma_lmm;
 935                 buf->lb_len = ma->ma_lmm_size;
 936                 LASSERT(!(ma->ma_valid & MA_LOV));
 937         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 938                 buf->lb_buf = ma->ma_lmv;
 939                 buf->lb_len = ma->ma_lmv_size;
 940                 LASSERT(!(ma->ma_valid & MA_LMV));
 941         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 942                 buf->lb_buf = ma->ma_default_lmv;
 943                 buf->lb_len = ma->ma_default_lmv_size;
 944                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 945         } else {
 946                 return -EINVAL;
 947         }
 948
 949         LASSERT(buf->lb_buf);
 950
 951         rc = mo_xattr_get(info->mti_env, next, buf, name);
 952         if (rc > 0) {
 953
 954 got:
 955                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 956                         if (info->mti_big_lmm_used)
 957                                 ma->ma_lmm = info->mti_big_lmm;
 958
 959                         /* NOT return LOV EA with hole to old client. */
 960                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 961                                      LOV_PATTERN_F_HOLE) &&
 962                             !(exp_connect_flags(info->mti_exp) &
 963                               OBD_CONNECT_LFSCK)) {
 964                                 return -EIO;
 965                         } else {
 966                                 ma->ma_lmm_size = rc;
 967                                 ma->ma_valid |= MA_LOV;
 968                         }
 969                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 970                         if (info->mti_big_lmm_used)
 971                                 ma->ma_lmv = info->mti_big_lmm;
 972
 973                         ma->ma_lmv_size = rc;
 974                         ma->ma_valid |= MA_LMV;
 975                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 976                         ma->ma_default_lmv_size = rc;
 977                         ma->ma_valid |= MA_LMV_DEF;
 978                 }
 979
 980                 /* Update mdt_max_mdsize so all clients will be aware that */
 981                 if (info->mti_mdt->mdt_max_mdsize < rc)
 982                         info->mti_mdt->mdt_max_mdsize = rc;
 983
 984                 rc = 0;
 985         } else if (rc == -ENODATA) {
 986                 /* no LOV EA */
 987                 rc = 0;
 988         } else if (rc == -ERANGE) {
 989                 /* Default LMV has fixed size, so it must be able to fit
 990                  * in the original buffer */
 991                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 992                         return rc;
 993                 rc = mdt_big_xattr_get(info, o, name);
 994                 if (rc > 0) {
 995                         info->mti_big_lmm_used = 1;
 996                         goto got;
 997                 }
 998         }
 999
1000         return rc;
1001 }
1002
1003 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1004                       struct lu_fid *pfid)
1005 {
1006         struct lu_buf           *buf = &info->mti_buf;
1007         struct link_ea_header   *leh;
1008         struct link_ea_entry    *lee;
1009         int                      rc;
1010         ENTRY;
1011
1012         buf->lb_buf = info->mti_big_lmm;
1013         buf->lb_len = info->mti_big_lmmsize;
1014         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1015                           buf, XATTR_NAME_LINK);
1016         /* ignore errors, MA_PFID won't be set and it is
1017          * up to the caller to treat this as an error */
1018         if (rc == -ERANGE || buf->lb_len == 0) {
1019                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1020                 buf->lb_buf = info->mti_big_lmm;
1021                 buf->lb_len = info->mti_big_lmmsize;
1022         }
1023
1024         if (rc < 0)
1025                 RETURN(rc);
1026         if (rc < sizeof(*leh)) {
1027                 CERROR("short LinkEA on "DFID": rc = %d\n",
1028                        PFID(mdt_object_fid(o)), rc);
1029                 RETURN(-ENODATA);
1030         }
1031
1032         leh = (struct link_ea_header *) buf->lb_buf;
1033         lee = (struct link_ea_entry *)(leh + 1);
1034         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1035                 leh->leh_magic = LINK_EA_MAGIC;
1036                 leh->leh_reccount = __swab32(leh->leh_reccount);
1037                 leh->leh_len = __swab64(leh->leh_len);
1038         }
1039         if (leh->leh_magic != LINK_EA_MAGIC)
1040                 RETURN(-EINVAL);
1041         if (leh->leh_reccount == 0)
1042                 RETURN(-ENODATA);
1043
1044         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1045         fid_be_to_cpu(pfid, pfid);
1046
1047         RETURN(0);
1048 }
1049
1050 int mdt_attr_get_complex(struct mdt_thread_info *info,
1051                          struct mdt_object *o, struct md_attr *ma)
1052 {
1053         const struct lu_env *env = info->mti_env;
1054         struct md_object    *next = mdt_object_child(o);
1055         struct lu_buf       *buf = &info->mti_buf;
1056         int                  need = ma->ma_need;
1057         int                  rc = 0, rc2;
1058         u32                  mode;
1059         ENTRY;
1060
1061         ma->ma_valid = 0;
1062
1063         if (mdt_object_exists(o) == 0)
1064                 GOTO(out, rc = -ENOENT);
1065         mode = lu_object_attr(&next->mo_lu);
1066
1067         if (need & MA_INODE) {
1068                 ma->ma_need = MA_INODE;
1069                 rc = mo_attr_get(env, next, ma);
1070                 if (rc)
1071                         GOTO(out, rc);
1072
1073                 if (S_ISREG(mode))
1074                         (void) mdt_get_som(info, o, ma);
1075                 ma->ma_valid |= MA_INODE;
1076         }
1077
1078         if (need & MA_PFID) {
1079                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1080                 if (rc == 0)
1081                         ma->ma_valid |= MA_PFID;
1082                 /* ignore this error, parent fid is not mandatory */
1083                 rc = 0;
1084         }
1085
1086         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1088                 if (rc)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1099                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1100                 if (rc != 0)
1101                         GOTO(out, rc);
1102         }
1103
1104         /*
1105          * In the handle of MA_INODE, we may already get the SOM attr.
1106          */
1107         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1108                 rc = mdt_get_som(info, o, ma);
1109                 if (rc != 0)
1110                         GOTO(out, rc);
1111         }
1112
1113         if (need & MA_HSM && S_ISREG(mode)) {
1114                 buf->lb_buf = info->mti_xattr_buf;
1115                 buf->lb_len = sizeof(info->mti_xattr_buf);
1116                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1117                              sizeof(info->mti_xattr_buf));
1118                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1119                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1120                 if (rc2 == 0)
1121                         ma->ma_valid |= MA_HSM;
1122                 else if (rc2 < 0 && rc2 != -ENODATA)
1123                         GOTO(out, rc = rc2);
1124         }
1125
1126 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1127         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1128                 buf->lb_buf = ma->ma_acl;
1129                 buf->lb_len = ma->ma_acl_size;
1130                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1131                 if (rc2 > 0) {
1132                         ma->ma_acl_size = rc2;
1133                         ma->ma_valid |= MA_ACL_DEF;
1134                 } else if (rc2 == -ENODATA) {
1135                         /* no ACLs */
1136                         ma->ma_acl_size = 0;
1137                 } else
1138                         GOTO(out, rc = rc2);
1139         }
1140 #endif
1141 out:
1142         ma->ma_need = need;
1143         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1144                rc, ma->ma_valid, ma->ma_lmm);
1145         RETURN(rc);
1146 }
1147
1148 static int mdt_getattr_internal(struct mdt_thread_info *info,
1149                                 struct mdt_object *o, int ma_need)
1150 {
1151         struct md_object        *next = mdt_object_child(o);
1152         const struct mdt_body   *reqbody = info->mti_body;
1153         struct ptlrpc_request   *req = mdt_info_req(info);
1154         struct md_attr          *ma = &info->mti_attr;
1155         struct lu_attr          *la = &ma->ma_attr;
1156         struct req_capsule      *pill = info->mti_pill;
1157         const struct lu_env     *env = info->mti_env;
1158         struct mdt_body         *repbody;
1159         struct lu_buf           *buffer = &info->mti_buf;
1160         struct obd_export       *exp = info->mti_exp;
1161         int                      rc;
1162         ENTRY;
1163
1164         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1165                 RETURN(err_serious(-ENOMEM));
1166
1167         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1168
1169         ma->ma_valid = 0;
1170
1171         if (mdt_object_remote(o)) {
1172                 /* This object is located on remote node.*/
1173                 /* Return -ENOTSUPP for old client */
1174                 if (!mdt_is_dne_client(req->rq_export))
1175                         GOTO(out, rc = -ENOTSUPP);
1176
1177                 repbody->mbo_fid1 = *mdt_object_fid(o);
1178                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1179                 GOTO(out, rc = 0);
1180         }
1181
1182         if (reqbody->mbo_eadatasize > 0) {
1183                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1184                 if (buffer->lb_buf == NULL)
1185                         GOTO(out, rc = -EPROTO);
1186                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1187                                                       RCL_SERVER);
1188         } else {
1189                 buffer->lb_buf = NULL;
1190                 buffer->lb_len = 0;
1191                 ma_need &= ~(MA_LOV | MA_LMV);
1192                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1193                        mdt_obd_name(info->mti_mdt),
1194                        req->rq_export->exp_client_uuid.uuid);
1195         }
1196
1197         /* from 2.12.58 intent_getattr pack default LMV in reply */
1198         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1199             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1200                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1201             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1202                                   RCL_SERVER)) {
1203                 ma->ma_lmv = buffer->lb_buf;
1204                 ma->ma_lmv_size = buffer->lb_len;
1205                 ma->ma_default_lmv = req_capsule_server_get(pill,
1206                                                 &RMF_DEFAULT_MDT_MD);
1207                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1208                                                 &RMF_DEFAULT_MDT_MD,
1209                                                 RCL_SERVER);
1210                 ma->ma_need = MA_INODE;
1211                 if (ma->ma_lmv_size > 0)
1212                         ma->ma_need |= MA_LMV;
1213                 if (ma->ma_default_lmv_size > 0)
1214                         ma->ma_need |= MA_LMV_DEF;
1215         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1216                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1217                 /* If it is dir and client require MEA, then we got MEA */
1218                 /* Assumption: MDT_MD size is enough for lmv size. */
1219                 ma->ma_lmv = buffer->lb_buf;
1220                 ma->ma_lmv_size = buffer->lb_len;
1221                 ma->ma_need = MA_INODE;
1222                 if (ma->ma_lmv_size > 0) {
1223                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1224                                 ma->ma_need |= MA_LMV;
1225                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1226                                 ma->ma_need |= MA_LMV_DEF;
1227                                 ma->ma_default_lmv = buffer->lb_buf;
1228                                 ma->ma_lmv = NULL;
1229                                 ma->ma_default_lmv_size = buffer->lb_len;
1230                                 ma->ma_lmv_size = 0;
1231                         }
1232                 }
1233         } else {
1234                 ma->ma_lmm = buffer->lb_buf;
1235                 ma->ma_lmm_size = buffer->lb_len;
1236                 ma->ma_need = MA_INODE | MA_HSM;
1237                 if (ma->ma_lmm_size > 0) {
1238                         ma->ma_need |= MA_LOV;
1239                         /* Older clients may crash if they getattr overstriped
1240                          * files
1241                          */
1242                         if (!exp_connect_overstriping(exp) &&
1243                             mdt_lmm_is_overstriping(ma->ma_lmm))
1244                                 RETURN(-EOPNOTSUPP);
1245                 }
1246         }
1247
1248         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1249             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1250             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1251                 /* get default stripe info for this dir. */
1252                 ma->ma_need |= MA_LOV_DEF;
1253         }
1254         ma->ma_need |= ma_need;
1255
1256         rc = mdt_attr_get_complex(info, o, ma);
1257         if (unlikely(rc)) {
1258                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1259                        "%s: getattr error for "DFID": rc = %d\n",
1260                        mdt_obd_name(info->mti_mdt),
1261                        PFID(mdt_object_fid(o)), rc);
1262                 RETURN(rc);
1263         }
1264
1265         /* if file is released, check if a restore is running */
1266         if (ma->ma_valid & MA_HSM) {
1267                 repbody->mbo_valid |= OBD_MD_TSTATE;
1268                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1269                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1270                         repbody->mbo_t_state = MS_RESTORE;
1271         }
1272
1273         if (likely(ma->ma_valid & MA_INODE))
1274                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1275         else
1276                 RETURN(-EFAULT);
1277
1278         if (mdt_body_has_lov(la, reqbody)) {
1279                 if (ma->ma_valid & MA_LOV) {
1280                         LASSERT(ma->ma_lmm_size);
1281                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1282                         if (S_ISDIR(la->la_mode))
1283                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1284                         else
1285                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1286                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1287                 }
1288                 if (ma->ma_valid & MA_LMV) {
1289                         /* Return -ENOTSUPP for old client */
1290                         if (!mdt_is_striped_client(req->rq_export))
1291                                 RETURN(-ENOTSUPP);
1292
1293                         LASSERT(S_ISDIR(la->la_mode));
1294                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1295                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1296                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1297                 }
1298                 if (ma->ma_valid & MA_LMV_DEF) {
1299                         /* Return -ENOTSUPP for old client */
1300                         if (!mdt_is_striped_client(req->rq_export))
1301                                 RETURN(-ENOTSUPP);
1302                         LASSERT(S_ISDIR(la->la_mode));
1303                         /*
1304                          * when ll_dir_getstripe() gets default LMV, it
1305                          * checks mbo_eadatasize.
1306                          */
1307                         if (!(ma->ma_valid & MA_LMV))
1308                                 repbody->mbo_eadatasize =
1309                                         ma->ma_default_lmv_size;
1310                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1311                                                OBD_MD_DEFAULT_MEA);
1312                 }
1313         } else if (S_ISLNK(la->la_mode) &&
1314                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1315                 buffer->lb_buf = ma->ma_lmm;
1316                 /* eadatasize from client includes NULL-terminator, so
1317                  * there is no need to read it */
1318                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1319                 rc = mo_readlink(env, next, buffer);
1320                 if (unlikely(rc <= 0)) {
1321                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1322                                mdt_obd_name(info->mti_mdt),
1323                                PFID(mdt_object_fid(o)), rc);
1324                         rc = -EFAULT;
1325                 } else {
1326                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1327
1328                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1329                                 rc -= 2;
1330                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1331                         /* we need to report back size with NULL-terminator
1332                          * because client expects that */
1333                         repbody->mbo_eadatasize = rc + 1;
1334                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1335                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1336                                        "on "DFID ", expected %d\n",
1337                                        mdt_obd_name(info->mti_mdt),
1338                                        rc, PFID(mdt_object_fid(o)),
1339                                        reqbody->mbo_eadatasize - 1);
1340                         /* NULL terminate */
1341                         ((char *)ma->ma_lmm)[rc] = 0;
1342
1343                         /* If the total CDEBUG() size is larger than a page, it
1344                          * will print a warning to the console, avoid this by
1345                          * printing just the last part of the symlink. */
1346                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1347                                print_limit < rc ? "..." : "", print_limit,
1348                                (char *)ma->ma_lmm + rc - print_limit, rc);
1349                         rc = 0;
1350                 }
1351         }
1352
1353         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1354                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1355                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1356                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1357                        repbody->mbo_max_mdsize);
1358         }
1359
1360 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1361         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1362                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1363                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1364                 if (IS_ERR(nodemap))
1365                         RETURN(PTR_ERR(nodemap));
1366
1367                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1368                 nodemap_putref(nodemap);
1369         }
1370 #endif
1371
1372 out:
1373         if (rc == 0)
1374                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1375
1376         RETURN(rc);
1377 }
1378
1379 static int mdt_getattr(struct tgt_session_info *tsi)
1380 {
1381         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1382         struct mdt_object       *obj = info->mti_object;
1383         struct req_capsule      *pill = info->mti_pill;
1384         struct mdt_body         *reqbody;
1385         struct mdt_body         *repbody;
1386         int rc, rc2;
1387         ENTRY;
1388
1389         if (unlikely(info->mti_object == NULL))
1390                 RETURN(-EPROTO);
1391
1392         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1393         LASSERT(reqbody);
1394         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1395
1396         /* Special case for Data-on-MDT files to get data version */
1397         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1398                 rc = mdt_data_version_get(tsi);
1399                 GOTO(out, rc);
1400         }
1401
1402         /* Unlike intent case where we need to pre-fill out buffers early on
1403          * in intent policy for ldlm reasons, here we can have a much better
1404          * guess at EA size by just reading it from disk.
1405          * Exceptions are readdir and (missing) directory striping */
1406         /* Readlink */
1407         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1408                 /* No easy way to know how long is the symlink, but it cannot
1409                  * be more than PATH_MAX, so we allocate +1 */
1410                 rc = PATH_MAX + 1;
1411         /* A special case for fs ROOT: getattr there might fetch
1412          * default EA for entire fs, not just for this dir!
1413          */
1414         } else if (lu_fid_eq(mdt_object_fid(obj),
1415                              &info->mti_mdt->mdt_md_root_fid) &&
1416                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1417                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1418                                                                  MDS_GETATTR)) {
1419                 /* Should the default strping be bigger, mdt_fix_reply
1420                  * will reallocate */
1421                 rc = DEF_REP_MD_SIZE;
1422         } else {
1423                 /* Read the actual EA size from disk */
1424                 rc = mdt_attr_get_eabuf_size(info, obj);
1425         }
1426
1427         if (rc < 0)
1428                 GOTO(out, rc = err_serious(rc));
1429
1430         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1431
1432         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1433          * by default. If the target object has more ACL entries, then
1434          * enlarge the buffer when necessary. */
1435         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1436                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1437
1438         rc = req_capsule_server_pack(pill);
1439         if (unlikely(rc != 0))
1440                 GOTO(out, rc = err_serious(rc));
1441
1442         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1443         LASSERT(repbody != NULL);
1444         repbody->mbo_eadatasize = 0;
1445         repbody->mbo_aclsize = 0;
1446
1447         rc = mdt_check_ucred(info);
1448         if (unlikely(rc))
1449                 GOTO(out_shrink, rc);
1450
1451         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1452
1453         rc = mdt_getattr_internal(info, obj, 0);
1454         EXIT;
1455 out_shrink:
1456         mdt_client_compatibility(info);
1457         rc2 = mdt_fix_reply(info);
1458         if (rc == 0)
1459                 rc = rc2;
1460 out:
1461         mdt_thread_info_fini(info);
1462         return rc;
1463 }
1464
1465 /**
1466  * Handler of layout intent RPC requiring the layout modification
1467  *
1468  * \param[in]  info     thread environment
1469  * \param[in]  obj      object
1470  * \param[out] lhc      object ldlm lock handle
1471  * \param[in]  layout   layout change descriptor
1472  *
1473  * \retval 0    on success
1474  * \retval < 0  error code
1475  */
1476 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1477                       struct mdt_lock_handle *lhc,
1478                       struct md_layout_change *layout)
1479 {
1480         int rc;
1481
1482         ENTRY;
1483
1484         if (!mdt_object_exists(obj))
1485                 RETURN(-ENOENT);
1486
1487         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1488                 RETURN(-EINVAL);
1489
1490         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1491                            MAY_WRITE);
1492         if (rc)
1493                 RETURN(rc);
1494
1495         rc = mdt_check_resent_lock(info, obj, lhc);
1496         if (rc < 0)
1497                 RETURN(rc);
1498
1499         if (rc > 0) {
1500                 /* not resent */
1501                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1502
1503                 /* take layout lock to prepare layout change */
1504                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1505                         lockpart |= MDS_INODELOCK_UPDATE;
1506
1507                 mdt_lock_handle_init(lhc);
1508                 mdt_lock_reg_init(lhc, LCK_EX);
1509                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1510                 if (rc)
1511                         RETURN(rc);
1512         }
1513
1514         mutex_lock(&obj->mot_som_mutex);
1515         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1516         mutex_unlock(&obj->mot_som_mutex);
1517
1518         if (rc)
1519                 mdt_object_unlock(info, obj, lhc, 1);
1520
1521         RETURN(rc);
1522 }
1523
1524 /**
1525  * Exchange MOF_LOV_CREATED flags between two objects after a
1526  * layout swap. No assumption is made on whether o1 or o2 have
1527  * created objects or not.
1528  *
1529  * \param[in,out] o1    First swap layout object
1530  * \param[in,out] o2    Second swap layout object
1531  */
1532 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1533 {
1534         unsigned int o1_lov_created = o1->mot_lov_created;
1535
1536         mutex_lock(&o1->mot_lov_mutex);
1537         mutex_lock(&o2->mot_lov_mutex);
1538
1539         o1->mot_lov_created = o2->mot_lov_created;
1540         o2->mot_lov_created = o1_lov_created;
1541
1542         mutex_unlock(&o2->mot_lov_mutex);
1543         mutex_unlock(&o1->mot_lov_mutex);
1544 }
1545
1546 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1547 {
1548         struct mdt_thread_info  *info;
1549         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1550         struct obd_export       *exp = req->rq_export;
1551         struct mdt_object       *o1, *o2, *o;
1552         struct mdt_lock_handle  *lh1, *lh2;
1553         struct mdc_swap_layouts *msl;
1554         int                      rc;
1555         ENTRY;
1556
1557         /* client does not support layout lock, so layout swaping
1558          * is disabled.
1559          * FIXME: there is a problem for old clients which don't support
1560          * layout lock yet. If those clients have already opened the file
1561          * they won't be notified at all so that old layout may still be
1562          * used to do IO. This can be fixed after file release is landed by
1563          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1564         if (!exp_connect_layout(exp))
1565                 RETURN(-EOPNOTSUPP);
1566
1567         info = tsi2mdt_info(tsi);
1568         if (unlikely(info->mti_object == NULL))
1569                 RETURN(-EPROTO);
1570
1571         if (info->mti_dlm_req != NULL)
1572                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1573
1574         o1 = info->mti_object;
1575         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1576                                 &info->mti_body->mbo_fid2);
1577         if (IS_ERR(o))
1578                 GOTO(out, rc = PTR_ERR(o));
1579
1580         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1581                 GOTO(put, rc = -ENOENT);
1582
1583         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1584         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1585                 GOTO(put, rc);
1586
1587         if (rc < 0)
1588                 swap(o1, o2);
1589
1590         /* permission check. Make sure the calling process having permission
1591          * to write both files. */
1592         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1593                            MAY_WRITE);
1594         if (rc < 0)
1595                 GOTO(put, rc);
1596
1597         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1598                            MAY_WRITE);
1599         if (rc < 0)
1600                 GOTO(put, rc);
1601
1602         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1603         if (msl == NULL)
1604                 GOTO(put, rc = -EPROTO);
1605
1606         lh1 = &info->mti_lh[MDT_LH_NEW];
1607         mdt_lock_reg_init(lh1, LCK_EX);
1608         lh2 = &info->mti_lh[MDT_LH_OLD];
1609         mdt_lock_reg_init(lh2, LCK_EX);
1610
1611         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1612                              MDS_INODELOCK_XATTR);
1613         if (rc < 0)
1614                 GOTO(put, rc);
1615
1616         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1617                              MDS_INODELOCK_XATTR);
1618         if (rc < 0)
1619                 GOTO(unlock1, rc);
1620
1621         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1622                              mdt_object_child(o2), msl->msl_flags);
1623         if (rc < 0)
1624                 GOTO(unlock2, rc);
1625
1626         mdt_swap_lov_flag(o1, o2);
1627
1628 unlock2:
1629         mdt_object_unlock(info, o2, lh2, rc);
1630 unlock1:
1631         mdt_object_unlock(info, o1, lh1, rc);
1632 put:
1633         mdt_object_put(info->mti_env, o);
1634 out:
1635         mdt_thread_info_fini(info);
1636         RETURN(rc);
1637 }
1638
1639 static int mdt_raw_lookup(struct mdt_thread_info *info,
1640                           struct mdt_object *parent,
1641                           const struct lu_name *lname,
1642                           struct ldlm_reply *ldlm_rep)
1643 {
1644         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1645         int              rc;
1646         ENTRY;
1647
1648         LASSERT(!info->mti_cross_ref);
1649
1650         /* Only got the fid of this obj by name */
1651         fid_zero(child_fid);
1652         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1653                         lname, child_fid, &info->mti_spec);
1654         if (rc == 0) {
1655                 struct mdt_body *repbody;
1656
1657                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1658                 repbody->mbo_fid1 = *child_fid;
1659                 repbody->mbo_valid = OBD_MD_FLID;
1660                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1661         } else if (rc == -ENOENT) {
1662                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1663         }
1664
1665         RETURN(rc);
1666 }
1667
1668 /*
1669  * UPDATE lock should be taken against parent, and be released before exit;
1670  * child_bits lock should be taken against child, and be returned back:
1671  *            (1)normal request should release the child lock;
1672  *            (2)intent request will grant the lock to client.
1673  */
1674 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1675                                  struct mdt_lock_handle *lhc,
1676                                  __u64 child_bits,
1677                                  struct ldlm_reply *ldlm_rep)
1678 {
1679         struct ptlrpc_request  *req = mdt_info_req(info);
1680         struct mdt_body        *reqbody = NULL;
1681         struct mdt_object      *parent = info->mti_object;
1682         struct mdt_object      *child;
1683         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1684         struct lu_name         *lname = NULL;
1685         struct mdt_lock_handle *lhp = NULL;
1686         struct ldlm_lock       *lock;
1687         struct req_capsule *pill = info->mti_pill;
1688         __u64 try_bits = 0;
1689         bool is_resent;
1690         int ma_need = 0;
1691         int rc;
1692
1693         ENTRY;
1694
1695         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1696         LASSERT(ergo(is_resent,
1697                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1698
1699         if (parent == NULL)
1700                 RETURN(-ENOENT);
1701
1702         if (info->mti_cross_ref) {
1703                 /* Only getattr on the child. Parent is on another node. */
1704                 mdt_set_disposition(info, ldlm_rep,
1705                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1706                 child = parent;
1707                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1708                        "ldlm_rep = %p\n",
1709                        PFID(mdt_object_fid(child)), ldlm_rep);
1710
1711                 rc = mdt_check_resent_lock(info, child, lhc);
1712                 if (rc < 0) {
1713                         RETURN(rc);
1714                 } else if (rc > 0) {
1715                         mdt_lock_handle_init(lhc);
1716                         mdt_lock_reg_init(lhc, LCK_PR);
1717
1718                         /*
1719                          * Object's name entry is on another MDS, it will
1720                          * request PERM lock only because LOOKUP lock is owned
1721                          * by the MDS where name entry resides.
1722                          *
1723                          * TODO: it should try layout lock too. - Jinshan
1724                          */
1725                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1726                                         MDS_INODELOCK_LAYOUT);
1727                         child_bits |= MDS_INODELOCK_PERM;
1728
1729                         rc = mdt_object_lock(info, child, lhc, child_bits);
1730                         if (rc < 0)
1731                                 RETURN(rc);
1732                 }
1733
1734                 /* Finally, we can get attr for child. */
1735                 if (!mdt_object_exists(child)) {
1736                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1737                                         &child->mot_obj,
1738                                         "remote object doesn't exist.");
1739                         mdt_object_unlock(info, child, lhc, 1);
1740                         RETURN(-ENOENT);
1741                 }
1742
1743                 rc = mdt_getattr_internal(info, child, 0);
1744                 if (unlikely(rc != 0)) {
1745                         mdt_object_unlock(info, child, lhc, 1);
1746                         RETURN(rc);
1747                 }
1748
1749                 rc = mdt_pack_secctx_in_reply(info, child);
1750                 if (unlikely(rc))
1751                         mdt_object_unlock(info, child, lhc, 1);
1752                 RETURN(rc);
1753         }
1754
1755         lname = &info->mti_name;
1756         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1757
1758         if (lu_name_is_valid(lname)) {
1759                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1760                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1761                        PNAME(lname), ldlm_rep);
1762         } else {
1763                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1764                 if (unlikely(reqbody == NULL))
1765                         RETURN(err_serious(-EPROTO));
1766
1767                 *child_fid = reqbody->mbo_fid2;
1768
1769                 if (unlikely(!fid_is_sane(child_fid)))
1770                         RETURN(err_serious(-EINVAL));
1771
1772                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1773                        "ldlm_rep = %p\n",
1774                        PFID(mdt_object_fid(parent)),
1775                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1776         }
1777
1778         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1779
1780         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1781                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1782                                 &parent->mot_obj,
1783                                 "Parent doesn't exist!");
1784                 RETURN(-ESTALE);
1785         }
1786
1787         if (mdt_object_remote(parent)) {
1788                 CERROR("%s: parent "DFID" is on remote target\n",
1789                        mdt_obd_name(info->mti_mdt),
1790                        PFID(mdt_object_fid(parent)));
1791                 RETURN(-EIO);
1792         }
1793
1794         if (lu_name_is_valid(lname)) {
1795                 /* Always allow to lookup ".." */
1796                 if (unlikely(lname->ln_namelen == 2 &&
1797                              lname->ln_name[0] == '.' &&
1798                              lname->ln_name[1] == '.'))
1799                         info->mti_spec.sp_permitted = 1;
1800
1801                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1802                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1803
1804                         RETURN(rc);
1805                 }
1806
1807                 /* step 1: lock parent only if parent is a directory */
1808                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1809                         lhp = &info->mti_lh[MDT_LH_PARENT];
1810                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1811                         rc = mdt_object_lock(info, parent, lhp,
1812                                              MDS_INODELOCK_UPDATE);
1813                         if (unlikely(rc != 0))
1814                                 RETURN(rc);
1815                 }
1816
1817                 /* step 2: lookup child's fid by name */
1818                 fid_zero(child_fid);
1819                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1820                                 child_fid, &info->mti_spec);
1821                 if (rc == -ENOENT)
1822                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1823
1824                 if (rc != 0)
1825                         GOTO(out_parent, rc);
1826         }
1827
1828         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1829
1830         /*
1831          *step 3: find the child object by fid & lock it.
1832          *        regardless if it is local or remote.
1833          *
1834          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1835          *      set parent dir fid the same as child fid in getattr by fid case
1836          *      we should not lu_object_find() the object again, could lead
1837          *      to hung if there is a concurrent unlink destroyed the object.
1838          */
1839         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1840                 mdt_object_get(info->mti_env, parent);
1841                 child = parent;
1842         } else {
1843                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1844                                         child_fid);
1845         }
1846
1847         if (unlikely(IS_ERR(child)))
1848                 GOTO(out_parent, rc = PTR_ERR(child));
1849
1850         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1851         if (!mdt_object_exists(child)) {
1852                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1853                                 &child->mot_obj,
1854                                 "Object doesn't exist!");
1855                 GOTO(out_child, rc = -ENOENT);
1856         }
1857
1858         rc = mdt_check_resent_lock(info, child, lhc);
1859         if (rc < 0) {
1860                 GOTO(out_child, rc);
1861         } else if (rc > 0) {
1862                 mdt_lock_handle_init(lhc);
1863                 mdt_lock_reg_init(lhc, LCK_PR);
1864
1865                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1866                     !mdt_object_remote(child)) {
1867                         struct md_attr *ma = &info->mti_attr;
1868
1869                         ma->ma_valid = 0;
1870                         ma->ma_need = MA_INODE;
1871                         rc = mdt_attr_get_complex(info, child, ma);
1872                         if (unlikely(rc != 0))
1873                                 GOTO(out_child, rc);
1874
1875                         /* If the file has not been changed for some time, we
1876                          * return not only a LOOKUP lock, but also an UPDATE
1877                          * lock and this might save us RPC on later STAT. For
1878                          * directories, it also let negative dentry cache start
1879                          * working for this dir. */
1880                         if (ma->ma_valid & MA_INODE &&
1881                             ma->ma_attr.la_valid & LA_CTIME &&
1882                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1883                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
1884                                 child_bits |= MDS_INODELOCK_UPDATE;
1885                 }
1886
1887                 /* layout lock must be granted in a best-effort way
1888                  * for IT operations */
1889                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1890                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1891                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1892                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1893                             exp_connect_layout(info->mti_exp)) {
1894                                 /* try to grant layout lock for regular file. */
1895                                 try_bits = MDS_INODELOCK_LAYOUT;
1896                         }
1897                         /* Acquire DOM lock in advance for data-on-mdt file */
1898                         if (child != parent)
1899                                 try_bits |= MDS_INODELOCK_DOM;
1900                 }
1901
1902                 if (try_bits != 0) {
1903                         /* try layout lock, it may fail to be granted due to
1904                          * contention at LOOKUP or UPDATE */
1905                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1906                                                  try_bits, false);
1907                         if (child_bits & MDS_INODELOCK_LAYOUT)
1908                                 ma_need |= MA_LOV;
1909                 } else {
1910                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1911                          * client will enqueue the lock to the remote MDT */
1912                         if (mdt_object_remote(child))
1913                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1914                         rc = mdt_object_lock(info, child, lhc, child_bits);
1915                 }
1916                 if (unlikely(rc != 0))
1917                         GOTO(out_child, rc);
1918         }
1919
1920         /* finally, we can get attr for child. */
1921         rc = mdt_getattr_internal(info, child, ma_need);
1922         if (unlikely(rc != 0)) {
1923                 mdt_object_unlock(info, child, lhc, 1);
1924                 GOTO(out_child, rc);
1925         }
1926
1927         rc = mdt_pack_secctx_in_reply(info, child);
1928         if (unlikely(rc)) {
1929                 mdt_object_unlock(info, child, lhc, 1);
1930                 GOTO(out_child, rc);
1931         }
1932
1933         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1934         if (lock) {
1935                 /* Debugging code. */
1936                 LDLM_DEBUG(lock, "Returning lock to client");
1937                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1938                                          &lock->l_resource->lr_name),
1939                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1940                          PLDLMRES(lock->l_resource),
1941                          PFID(mdt_object_fid(child)));
1942
1943                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1944                     !mdt_object_remote(child) && child != parent) {
1945                         mdt_object_put(info->mti_env, child);
1946                         rc = mdt_pack_size2body(info, child_fid,
1947                                                 &lhc->mlh_reg_lh);
1948                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1949                                 /* DOM lock was taken in advance but this is
1950                                  * not DoM file. Drop the lock.
1951                                  */
1952                                 lock_res_and_lock(lock);
1953                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1954                                 unlock_res_and_lock(lock);
1955                         }
1956                         LDLM_LOCK_PUT(lock);
1957                         GOTO(out_parent, rc = 0);
1958                 }
1959                 LDLM_LOCK_PUT(lock);
1960         }
1961
1962         EXIT;
1963 out_child:
1964         mdt_object_put(info->mti_env, child);
1965 out_parent:
1966         if (lhp)
1967                 mdt_object_unlock(info, parent, lhp, 1);
1968         return rc;
1969 }
1970
1971 /* normal handler: should release the child lock */
1972 static int mdt_getattr_name(struct tgt_session_info *tsi)
1973 {
1974         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1975         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1976         struct mdt_body        *reqbody;
1977         struct mdt_body        *repbody;
1978         int rc, rc2;
1979         ENTRY;
1980
1981         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1982         LASSERT(reqbody != NULL);
1983         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1984         LASSERT(repbody != NULL);
1985
1986         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1987         repbody->mbo_eadatasize = 0;
1988         repbody->mbo_aclsize = 0;
1989
1990         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1991         if (unlikely(rc))
1992                 GOTO(out_shrink, rc);
1993
1994         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1995         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1996                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1997                 lhc->mlh_reg_lh.cookie = 0;
1998         }
1999         mdt_exit_ucred(info);
2000         EXIT;
2001 out_shrink:
2002         mdt_client_compatibility(info);
2003         rc2 = mdt_fix_reply(info);
2004         if (rc == 0)
2005                 rc = rc2;
2006         mdt_thread_info_fini(info);
2007         return rc;
2008 }
2009
2010 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2011                             const struct lu_fid *pfid,
2012                             const struct lu_name *name,
2013                             struct mdt_object *obj, s64 ctime)
2014 {
2015         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2016         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2017         struct mdt_device *mdt = info->mti_mdt;
2018         struct md_attr *ma = &info->mti_attr;
2019         struct mdt_lock_handle *parent_lh;
2020         struct mdt_lock_handle *child_lh;
2021         struct mdt_object *pobj;
2022         bool cos_incompat = false;
2023         int rc;
2024         ENTRY;
2025
2026         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2027         if (IS_ERR(pobj))
2028                 GOTO(out, rc = PTR_ERR(pobj));
2029
2030         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2031         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2032         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2033         if (rc != 0)
2034                 GOTO(put_parent, rc);
2035
2036         if (mdt_object_remote(pobj))
2037                 cos_incompat = true;
2038
2039         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2040                         name, child_fid, &info->mti_spec);
2041         if (rc != 0)
2042                 GOTO(unlock_parent, rc);
2043
2044         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2045                 GOTO(unlock_parent, rc = -EREMCHG);
2046
2047         child_lh = &info->mti_lh[MDT_LH_CHILD];
2048         mdt_lock_reg_init(child_lh, LCK_EX);
2049         rc = mdt_reint_striped_lock(info, obj, child_lh,
2050                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2051                                     einfo, cos_incompat);
2052         if (rc != 0)
2053                 GOTO(unlock_parent, rc);
2054
2055         if (atomic_read(&obj->mot_open_count)) {
2056                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2057                        PFID(mdt_object_fid(obj)));
2058                 GOTO(unlock_child, rc = -EBUSY);
2059         }
2060
2061         ma->ma_need = 0;
2062         ma->ma_valid = MA_INODE;
2063         ma->ma_attr.la_valid = LA_CTIME;
2064         ma->ma_attr.la_ctime = ctime;
2065
2066         mutex_lock(&obj->mot_lov_mutex);
2067
2068         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2069                         mdt_object_child(obj), name, ma, 0);
2070
2071         mutex_unlock(&obj->mot_lov_mutex);
2072
2073 unlock_child:
2074         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2075 unlock_parent:
2076         mdt_object_unlock(info, pobj, parent_lh, 1);
2077 put_parent:
2078         mdt_object_put(info->mti_env, pobj);
2079 out:
2080         RETURN(rc);
2081 }
2082
2083 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2084                                         struct mdt_object *obj)
2085 {
2086         struct lu_ucred *uc = lu_ucred(info->mti_env);
2087         struct md_attr *ma = &info->mti_attr;
2088         struct lu_attr *la = &ma->ma_attr;
2089         int rc = 0;
2090         ENTRY;
2091
2092         ma->ma_need = MA_INODE;
2093         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2094         if (rc)
2095                 GOTO(out, rc);
2096
2097         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2098                         rc = -EACCES;
2099
2100         if (md_capable(uc, CFS_CAP_DAC_OVERRIDE))
2101                 RETURN(0);
2102         if (uc->uc_fsuid == la->la_uid) {
2103                 if ((la->la_mode & S_IWUSR) == 0)
2104                         rc = -EACCES;
2105         } else if (uc->uc_fsgid == la->la_gid) {
2106                 if ((la->la_mode & S_IWGRP) == 0)
2107                         rc = -EACCES;
2108         } else if ((la->la_mode & S_IWOTH) == 0) {
2109                         rc = -EACCES;
2110         }
2111
2112 out:
2113         RETURN(rc);
2114 }
2115
2116 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2117                          s64 ctime)
2118 {
2119         struct mdt_device *mdt = info->mti_mdt;
2120         struct mdt_object *obj = NULL;
2121         struct linkea_data ldata = { NULL };
2122         struct lu_buf *buf = &info->mti_big_buf;
2123         struct lu_name *name = &info->mti_name;
2124         struct lu_fid *pfid = &info->mti_tmp_fid1;
2125         struct link_ea_header *leh;
2126         struct link_ea_entry *lee;
2127         int reclen, count, rc = 0;
2128         ENTRY;
2129
2130         if (!fid_is_sane(fid))
2131                 GOTO(out, rc = -EINVAL);
2132
2133         if (!fid_is_namespace_visible(fid))
2134                 GOTO(out, rc = -EINVAL);
2135
2136         obj = mdt_object_find(info->mti_env, mdt, fid);
2137         if (IS_ERR(obj))
2138                 GOTO(out, rc = PTR_ERR(obj));
2139
2140         if (mdt_object_remote(obj))
2141                 GOTO(out, rc = -EREMOTE);
2142         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2143                 GOTO(out, rc = -ENOENT);
2144
2145         rc = mdt_rmfid_check_permission(info, obj);
2146         if (rc)
2147                 GOTO(out, rc);
2148
2149         /* take LinkEA */
2150         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2151         if (!buf->lb_buf)
2152                 GOTO(out, rc = -ENOMEM);
2153
2154         ldata.ld_buf = buf;
2155         rc = mdt_links_read(info, obj, &ldata);
2156         if (rc)
2157                 GOTO(out, rc);
2158
2159         leh = buf->lb_buf;
2160         lee = (struct link_ea_entry *)(leh + 1);
2161         for (count = 0; count < leh->leh_reccount; count++) {
2162                 /* remove every hardlink */
2163                 linkea_entry_unpack(lee, &reclen, name, pfid);
2164                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2165                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2166                 if (rc)
2167                         break;
2168         }
2169
2170 out:
2171         if (obj && !IS_ERR(obj))
2172                 mdt_object_put(info->mti_env, obj);
2173         if (info->mti_big_buf.lb_buf)
2174                 lu_buf_free(&info->mti_big_buf);
2175
2176         RETURN(rc);
2177 }
2178
2179 static int mdt_rmfid(struct tgt_session_info *tsi)
2180 {
2181         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2182         struct mdt_body *reqbody;
2183         struct lu_fid *fids, *rfids;
2184         int bufsize, rc;
2185         __u32 *rcs;
2186         int i, nr;
2187         ENTRY;
2188
2189         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2190         if (reqbody == NULL)
2191                 RETURN(-EPROTO);
2192         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2193                                        RCL_CLIENT);
2194         nr = bufsize / sizeof(struct lu_fid);
2195         if (nr * sizeof(struct lu_fid) != bufsize)
2196                 RETURN(-EINVAL);
2197         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2198                              RCL_SERVER, nr * sizeof(__u32));
2199         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2200                              RCL_SERVER, nr * sizeof(struct lu_fid));
2201         rc = req_capsule_server_pack(tsi->tsi_pill);
2202         if (rc)
2203                 GOTO(out, rc = err_serious(rc));
2204         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2205         if (fids == NULL)
2206                 RETURN(-EPROTO);
2207         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2208         LASSERT(rcs);
2209         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2210         LASSERT(rfids);
2211
2212         mdt_init_ucred(mti, reqbody);
2213         for (i = 0; i < nr; i++) {
2214                 rfids[i] = fids[i];
2215                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2216         }
2217         mdt_exit_ucred(mti);
2218
2219 out:
2220         RETURN(rc);
2221 }
2222
2223 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2224                          void *karg, void __user *uarg);
2225
2226 static int mdt_set_info(struct tgt_session_info *tsi)
2227 {
2228         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2229         char                    *key;
2230         void                    *val;
2231         int                      keylen, vallen, rc = 0;
2232
2233         ENTRY;
2234
2235         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2236         if (key == NULL) {
2237                 DEBUG_REQ(D_HA, req, "no set_info key");
2238                 RETURN(err_serious(-EFAULT));
2239         }
2240
2241         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2242                                       RCL_CLIENT);
2243
2244         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2245         if (val == NULL) {
2246                 DEBUG_REQ(D_HA, req, "no set_info val");
2247                 RETURN(err_serious(-EFAULT));
2248         }
2249
2250         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2251                                       RCL_CLIENT);
2252
2253         /* Swab any part of val you need to here */
2254         if (KEY_IS(KEY_READ_ONLY)) {
2255                 spin_lock(&req->rq_export->exp_lock);
2256                 if (*(__u32 *)val)
2257                         *exp_connect_flags_ptr(req->rq_export) |=
2258                                 OBD_CONNECT_RDONLY;
2259                 else
2260                         *exp_connect_flags_ptr(req->rq_export) &=
2261                                 ~OBD_CONNECT_RDONLY;
2262                 spin_unlock(&req->rq_export->exp_lock);
2263         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2264                 struct changelog_setinfo *cs = val;
2265
2266                 if (vallen != sizeof(*cs)) {
2267                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2268                                tgt_name(tsi->tsi_tgt), vallen);
2269                         RETURN(-EINVAL);
2270                 }
2271                 if (ptlrpc_req_need_swab(req)) {
2272                         __swab64s(&cs->cs_recno);
2273                         __swab32s(&cs->cs_id);
2274                 }
2275
2276                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2277                                    vallen, val, NULL);
2278         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2279                 if (vallen > 0)
2280                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2281         } else {
2282                 RETURN(-EINVAL);
2283         }
2284         RETURN(rc);
2285 }
2286
2287 static int mdt_readpage(struct tgt_session_info *tsi)
2288 {
2289         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2290         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2291         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2292         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2293         struct mdt_body         *repbody;
2294         int                      rc;
2295         int                      i;
2296
2297         ENTRY;
2298
2299         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2300                 RETURN(err_serious(-ENOMEM));
2301
2302         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2303         if (repbody == NULL || reqbody == NULL)
2304                 RETURN(err_serious(-EFAULT));
2305
2306         /*
2307          * prepare @rdpg before calling lower layers and transfer itself. Here
2308          * reqbody->size contains offset of where to start to read and
2309          * reqbody->nlink contains number bytes to read.
2310          */
2311         rdpg->rp_hash = reqbody->mbo_size;
2312         if (rdpg->rp_hash != reqbody->mbo_size) {
2313                 CERROR("Invalid hash: %#llx != %#llx\n",
2314                        rdpg->rp_hash, reqbody->mbo_size);
2315                 RETURN(-EFAULT);
2316         }
2317
2318         rdpg->rp_attrs = reqbody->mbo_mode;
2319         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2320                 rdpg->rp_attrs |= LUDA_64BITHASH;
2321         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2322                                 exp_max_brw_size(tsi->tsi_exp));
2323         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2324                           PAGE_SHIFT;
2325         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2326         if (rdpg->rp_pages == NULL)
2327                 RETURN(-ENOMEM);
2328
2329         for (i = 0; i < rdpg->rp_npages; ++i) {
2330                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2331                 if (rdpg->rp_pages[i] == NULL)
2332                         GOTO(free_rdpg, rc = -ENOMEM);
2333         }
2334
2335         /* call lower layers to fill allocated pages with directory data */
2336         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2337         if (rc < 0)
2338                 GOTO(free_rdpg, rc);
2339
2340         /* send pages to client */
2341         rc = tgt_sendpage(tsi, rdpg, rc);
2342
2343         EXIT;
2344 free_rdpg:
2345
2346         for (i = 0; i < rdpg->rp_npages; i++)
2347                 if (rdpg->rp_pages[i] != NULL)
2348                         __free_page(rdpg->rp_pages[i]);
2349         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2350
2351         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2352                 RETURN(0);
2353
2354         return rc;
2355 }
2356
2357 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2358 {
2359         struct lu_ucred *uc = mdt_ucred_check(info);
2360         struct lu_attr *attr = &info->mti_attr.ma_attr;
2361
2362         if (uc == NULL)
2363                 return -EINVAL;
2364
2365         if (op != REINT_SETATTR) {
2366                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2367                         attr->la_uid = uc->uc_fsuid;
2368                 /* for S_ISGID, inherit gid from his parent, such work will be
2369                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2370                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2371                         attr->la_gid = uc->uc_fsgid;
2372         }
2373
2374         return 0;
2375 }
2376
2377 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2378 {
2379         return op == REINT_OPEN &&
2380              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2381 }
2382
2383 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2384 {
2385         struct req_capsule *pill = info->mti_pill;
2386
2387         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2388                                   RCL_SERVER) &&
2389             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2390                                   RCL_CLIENT)) {
2391                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2392                                          RCL_CLIENT) != 0) {
2393                         /* pre-set size in server part with max size */
2394                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2395                                              RCL_SERVER,
2396                                              info->mti_mdt->mdt_max_ea_size);
2397                 } else {
2398                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2399                                              RCL_SERVER, 0);
2400                 }
2401         }
2402
2403 }
2404
2405 static int mdt_reint_internal(struct mdt_thread_info *info,
2406                               struct mdt_lock_handle *lhc,
2407                               __u32 op)
2408 {
2409         struct req_capsule      *pill = info->mti_pill;
2410         struct mdt_body         *repbody;
2411         int                      rc = 0, rc2;
2412
2413         ENTRY;
2414
2415         rc = mdt_reint_unpack(info, op);
2416         if (rc != 0) {
2417                 CERROR("Can't unpack reint, rc %d\n", rc);
2418                 RETURN(err_serious(rc));
2419         }
2420
2421
2422         /* check if the file system is set to readonly. O_RDONLY open
2423          * is still allowed even the file system is set to readonly mode */
2424         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2425                 RETURN(err_serious(-EROFS));
2426
2427         /* for replay (no_create) lmm is not needed, client has it already */
2428         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2429                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2430                                      DEF_REP_MD_SIZE);
2431
2432         /* llog cookies are always 0, the field is kept for compatibility */
2433         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2434                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2435
2436         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2437          * by default. If the target object has more ACL entries, then
2438          * enlarge the buffer when necessary. */
2439         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2440                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2441                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2442
2443         mdt_preset_secctx_size(info);
2444
2445         rc = req_capsule_server_pack(pill);
2446         if (rc != 0) {
2447                 CERROR("Can't pack response, rc %d\n", rc);
2448                 RETURN(err_serious(rc));
2449         }
2450
2451         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2452                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2453                 LASSERT(repbody);
2454                 repbody->mbo_eadatasize = 0;
2455                 repbody->mbo_aclsize = 0;
2456         }
2457
2458         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2459
2460         /* for replay no cookkie / lmm need, because client have this already */
2461         if (info->mti_spec.no_create)
2462                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2463                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2464
2465         rc = mdt_init_ucred_reint(info);
2466         if (rc)
2467                 GOTO(out_shrink, rc);
2468
2469         rc = mdt_fix_attr_ucred(info, op);
2470         if (rc != 0)
2471                 GOTO(out_ucred, rc = err_serious(rc));
2472
2473         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2474         if (rc < 0) {
2475                 GOTO(out_ucred, rc);
2476         } else if (rc == 1) {
2477                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2478                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2479                 GOTO(out_ucred, rc);
2480         }
2481         rc = mdt_reint_rec(info, lhc);
2482         EXIT;
2483 out_ucred:
2484         mdt_exit_ucred(info);
2485 out_shrink:
2486         mdt_client_compatibility(info);
2487
2488         rc2 = mdt_fix_reply(info);
2489         if (rc == 0)
2490                 rc = rc2;
2491
2492         /*
2493          * Data-on-MDT optimization - read data along with OPEN and return it
2494          * in reply. Do that only if we have both DOM and LAYOUT locks.
2495          */
2496         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2497             info->mti_attr.ma_lmm != NULL &&
2498             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2499                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2500                                           &lhc->mlh_reg_lh);
2501         }
2502
2503         return rc;
2504 }
2505
2506 static long mdt_reint_opcode(struct ptlrpc_request *req,
2507                              const struct req_format **fmt)
2508 {
2509         struct mdt_device       *mdt;
2510         struct mdt_rec_reint    *rec;
2511         long                     opc;
2512
2513         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2514         if (rec != NULL) {
2515                 opc = rec->rr_opcode;
2516                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2517                 if (opc < REINT_MAX && fmt[opc] != NULL)
2518                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2519                 else {
2520                         mdt = mdt_exp2dev(req->rq_export);
2521                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2522                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2523                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2524                         opc = err_serious(-EFAULT);
2525                 }
2526         } else {
2527                 opc = err_serious(-EFAULT);
2528         }
2529         return opc;
2530 }
2531
2532 static int mdt_reint(struct tgt_session_info *tsi)
2533 {
2534         long opc;
2535         int  rc;
2536         static const struct req_format *reint_fmts[REINT_MAX] = {
2537                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2538                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2539                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2540                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2541                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2542                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2543                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2544                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2545                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2546                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2547         };
2548
2549         ENTRY;
2550
2551         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2552         if (opc >= 0) {
2553                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2554                 /*
2555                  * No lock possible here from client to pass it to reint code
2556                  * path.
2557                  */
2558                 rc = mdt_reint_internal(info, NULL, opc);
2559                 mdt_thread_info_fini(info);
2560         } else {
2561                 rc = opc;
2562         }
2563
2564         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2565         RETURN(rc);
2566 }
2567
2568 /* this should sync the whole device */
2569 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2570 {
2571         struct dt_device *dt = mdt->mdt_bottom;
2572         int rc;
2573         ENTRY;
2574
2575         rc = dt->dd_ops->dt_sync(env, dt);
2576         RETURN(rc);
2577 }
2578
2579 /* this should sync this object */
2580 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2581                            struct mdt_object *mo)
2582 {
2583         int rc = 0;
2584
2585         ENTRY;
2586
2587         if (!mdt_object_exists(mo)) {
2588                 CWARN("%s: non existing object "DFID": rc = %d\n",
2589                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2590                       -ESTALE);
2591                 RETURN(-ESTALE);
2592         }
2593
2594         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2595                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2596                 dt_obj_version_t version;
2597
2598                 version = dt_version_get(env, mdt_obj2dt(mo));
2599                 if (version > tgt->lut_obd->obd_last_committed)
2600                         rc = mo_object_sync(env, mdt_object_child(mo));
2601         } else {
2602                 rc = mo_object_sync(env, mdt_object_child(mo));
2603         }
2604
2605         RETURN(rc);
2606 }
2607
2608 static int mdt_sync(struct tgt_session_info *tsi)
2609 {
2610         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2611         struct req_capsule      *pill = tsi->tsi_pill;
2612         struct mdt_body         *body;
2613         int                      rc;
2614
2615         ENTRY;
2616
2617         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2618                 RETURN(err_serious(-ENOMEM));
2619
2620         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2621                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2622         } else {
2623                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2624
2625                 if (unlikely(info->mti_object == NULL))
2626                         RETURN(-EPROTO);
2627
2628                 /* sync an object */
2629                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2630                                      info->mti_object);
2631                 if (rc == 0) {
2632                         const struct lu_fid *fid;
2633                         struct lu_attr *la = &info->mti_attr.ma_attr;
2634
2635                         info->mti_attr.ma_need = MA_INODE;
2636                         info->mti_attr.ma_valid = 0;
2637                         rc = mdt_attr_get_complex(info, info->mti_object,
2638                                                   &info->mti_attr);
2639                         if (rc == 0) {
2640                                 body = req_capsule_server_get(pill,
2641                                                               &RMF_MDT_BODY);
2642                                 fid = mdt_object_fid(info->mti_object);
2643                                 mdt_pack_attr2body(info, body, la, fid);
2644                         }
2645                 }
2646                 mdt_thread_info_fini(info);
2647         }
2648         if (rc == 0)
2649                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2650
2651         RETURN(rc);
2652 }
2653
2654 static int mdt_data_sync(struct tgt_session_info *tsi)
2655 {
2656         struct mdt_thread_info *info;
2657         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2658         struct ost_body *body = tsi->tsi_ost_body;
2659         struct ost_body *repbody;
2660         struct mdt_object *mo = NULL;
2661         struct md_attr *ma;
2662         int rc = 0;
2663
2664         ENTRY;
2665
2666         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2667
2668         /* if no fid is specified then do nothing,
2669          * device sync is done via MDS_SYNC */
2670         if (fid_is_zero(&tsi->tsi_fid))
2671                 RETURN(0);
2672
2673         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2674         if (IS_ERR(mo))
2675                 RETURN(PTR_ERR(mo));
2676
2677         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2678         if (rc)
2679                 GOTO(put, rc);
2680
2681         repbody->oa.o_oi = body->oa.o_oi;
2682         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2683
2684         info = tsi2mdt_info(tsi);
2685         ma = &info->mti_attr;
2686         ma->ma_need = MA_INODE;
2687         ma->ma_valid = 0;
2688         rc = mdt_attr_get_complex(info, mo, ma);
2689         if (rc == 0)
2690                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2691         else
2692                 rc = 0;
2693         mdt_thread_info_fini(info);
2694
2695         EXIT;
2696 put:
2697         if (mo != NULL)
2698                 mdt_object_put(tsi->tsi_env, mo);
2699         return rc;
2700 }
2701
2702 /*
2703  * Handle quota control requests to consult current usage/limit, but also
2704  * to configure quota enforcement
2705  */
2706 static int mdt_quotactl(struct tgt_session_info *tsi)
2707 {
2708         struct obd_export       *exp  = tsi->tsi_exp;
2709         struct req_capsule      *pill = tsi->tsi_pill;
2710         struct obd_quotactl     *oqctl, *repoqc;
2711         int                      id, rc;
2712         struct mdt_device       *mdt = mdt_exp2dev(exp);
2713         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2714         struct lu_nodemap       *nodemap;
2715         ENTRY;
2716
2717         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2718         if (oqctl == NULL)
2719                 RETURN(err_serious(-EPROTO));
2720
2721         rc = req_capsule_server_pack(pill);
2722         if (rc)
2723                 RETURN(err_serious(rc));
2724
2725         nodemap = nodemap_get_from_exp(exp);
2726         if (IS_ERR(nodemap))
2727                 RETURN(PTR_ERR(nodemap));
2728
2729         switch (oqctl->qc_cmd) {
2730                 /* master quotactl */
2731         case Q_SETINFO:
2732         case Q_SETQUOTA:
2733         case LUSTRE_Q_SETDEFAULT:
2734                 if (!nodemap_can_setquota(nodemap))
2735                         GOTO(out_nodemap, rc = -EPERM);
2736                 /* fallthrough */
2737         case Q_GETINFO:
2738         case Q_GETQUOTA:
2739         case LUSTRE_Q_GETDEFAULT:
2740                 if (qmt == NULL)
2741                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2742                 /* slave quotactl */
2743                 /* fallthrough */
2744         case Q_GETOINFO:
2745         case Q_GETOQUOTA:
2746                 break;
2747         default:
2748                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2749                 GOTO(out_nodemap, rc = -EFAULT);
2750         }
2751
2752         id = oqctl->qc_id;
2753         switch (oqctl->qc_type) {
2754         case USRQUOTA:
2755                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2756                                     NODEMAP_CLIENT_TO_FS, id);
2757                 break;
2758         case GRPQUOTA:
2759                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2760                                     NODEMAP_CLIENT_TO_FS, id);
2761                 break;
2762         case PRJQUOTA:
2763                 /* todo: check/map project id */
2764                 id = oqctl->qc_id;
2765                 break;
2766         default:
2767                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2768         }
2769         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2770         if (repoqc == NULL)
2771                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2772
2773         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2774                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2775
2776         if (oqctl->qc_id != id)
2777                 swap(oqctl->qc_id, id);
2778
2779         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2780                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2781                         RETURN(-EINPROGRESS);
2782         }
2783
2784         switch (oqctl->qc_cmd) {
2785
2786         case Q_GETINFO:
2787         case Q_SETINFO:
2788         case Q_SETQUOTA:
2789         case Q_GETQUOTA:
2790         case LUSTRE_Q_SETDEFAULT:
2791         case LUSTRE_Q_GETDEFAULT:
2792                 /* forward quotactl request to QMT */
2793                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2794                 break;
2795
2796         case Q_GETOINFO:
2797         case Q_GETOQUOTA:
2798                 /* slave quotactl */
2799                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2800                                    oqctl);
2801                 break;
2802
2803         default:
2804                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2805                 GOTO(out_nodemap, rc = -EFAULT);
2806         }
2807
2808         if (oqctl->qc_id != id)
2809                 swap(oqctl->qc_id, id);
2810
2811         *repoqc = *oqctl;
2812
2813         EXIT;
2814
2815 out_nodemap:
2816         nodemap_putref(nodemap);
2817
2818         return rc;
2819 }
2820
2821 /** clone llog ctxt from child (mdd)
2822  * This allows remote llog (replicator) access.
2823  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2824  * context was originally set up, or we can handle them directly.
2825  * I choose the latter, but that means I need any llog
2826  * contexts set up by child to be accessable by the mdt.  So we clone the
2827  * context into our context list here.
2828  */
2829 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2830                                int idx)
2831 {
2832         struct md_device  *next = mdt->mdt_child;
2833         struct llog_ctxt *ctxt;
2834         int rc;
2835
2836         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2837                 return 0;
2838
2839         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2840         if (rc || ctxt == NULL) {
2841                 return 0;
2842         }
2843
2844         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2845         if (rc)
2846                 CERROR("Can't set mdt ctxt %d\n", rc);
2847
2848         return rc;
2849 }
2850
2851 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2852                                  struct mdt_device *mdt, int idx)
2853 {
2854         struct llog_ctxt *ctxt;
2855
2856         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2857         if (ctxt == NULL)
2858                 return 0;
2859         /* Put once for the get we just did, and once for the clone */
2860         llog_ctxt_put(ctxt);
2861         llog_ctxt_put(ctxt);
2862         return 0;
2863 }
2864
2865 /*
2866  * sec context handlers
2867  */
2868 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2869 {
2870         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2871
2872         return 0;
2873 }
2874
2875 /*
2876  * quota request handlers
2877  */
2878 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2879 {
2880         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2881         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2882         int                      rc;
2883         ENTRY;
2884
2885         if (qmt == NULL)
2886                 RETURN(err_serious(-EOPNOTSUPP));
2887
2888         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2889         RETURN(rc);
2890 }
2891
2892 struct mdt_object *mdt_object_new(const struct lu_env *env,
2893                                   struct mdt_device *d,
2894                                   const struct lu_fid *f)
2895 {
2896         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2897         struct lu_object *o;
2898         struct mdt_object *m;
2899         ENTRY;
2900
2901         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2902         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2903         if (unlikely(IS_ERR(o)))
2904                 m = (struct mdt_object *)o;
2905         else
2906                 m = mdt_obj(o);
2907         RETURN(m);
2908 }
2909
2910 struct mdt_object *mdt_object_find(const struct lu_env *env,
2911                                    struct mdt_device *d,
2912                                    const struct lu_fid *f)
2913 {
2914         struct lu_object *o;
2915         struct mdt_object *m;
2916         ENTRY;
2917
2918         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2919         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2920         if (unlikely(IS_ERR(o)))
2921                 m = (struct mdt_object *)o;
2922         else
2923                 m = mdt_obj(o);
2924
2925         RETURN(m);
2926 }
2927
2928 /**
2929  * Asyncronous commit for mdt device.
2930  *
2931  * Pass asynchonous commit call down the MDS stack.
2932  *
2933  * \param env environment
2934  * \param mdt the mdt device
2935  */
2936 static void mdt_device_commit_async(const struct lu_env *env,
2937                                     struct mdt_device *mdt)
2938 {
2939         struct dt_device *dt = mdt->mdt_bottom;
2940         int rc;
2941         ENTRY;
2942
2943         rc = dt->dd_ops->dt_commit_async(env, dt);
2944         if (unlikely(rc != 0))
2945                 CWARN("%s: async commit start failed: rc = %d\n",
2946                       mdt_obd_name(mdt), rc);
2947         atomic_inc(&mdt->mdt_async_commit_count);
2948         EXIT;
2949 }
2950
2951 /**
2952  * Mark the lock as "synchonous".
2953  *
2954  * Mark the lock to deffer transaction commit to the unlock time.
2955  *
2956  * \param lock the lock to mark as "synchonous"
2957  *
2958  * \see mdt_is_lock_sync
2959  * \see mdt_save_lock
2960  */
2961 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2962 {
2963         lock->l_ast_data = (void*)1;
2964 }
2965
2966 /**
2967  * Check whehter the lock "synchonous" or not.
2968  *
2969  * \param lock the lock to check
2970  * \retval 1 the lock is "synchonous"
2971  * \retval 0 the lock isn't "synchronous"
2972  *
2973  * \see mdt_set_lock_sync
2974  * \see mdt_save_lock
2975  */
2976 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2977 {
2978         return lock->l_ast_data != NULL;
2979 }
2980
2981 /**
2982  * Blocking AST for mdt locks.
2983  *
2984  * Starts transaction commit if in case of COS lock conflict or
2985  * deffers such a commit to the mdt_save_lock.
2986  *
2987  * \param lock the lock which blocks a request or cancelling lock
2988  * \param desc unused
2989  * \param data unused
2990  * \param flag indicates whether this cancelling or blocking callback
2991  * \retval 0
2992  * \see ldlm_blocking_ast_nocheck
2993  */
2994 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2995                      void *data, int flag)
2996 {
2997         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2998         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2999         struct ldlm_cb_set_arg *arg = data;
3000         bool commit_async = false;
3001         int rc;
3002         ENTRY;
3003
3004         if (flag == LDLM_CB_CANCELING)
3005                 RETURN(0);
3006
3007         lock_res_and_lock(lock);
3008         if (lock->l_blocking_ast != mdt_blocking_ast) {
3009                 unlock_res_and_lock(lock);
3010                 RETURN(0);
3011         }
3012
3013         /* A blocking ast may be sent from ldlm_lock_decref_internal
3014          * when the last reference to a local lock was released and
3015          * during blocking event from ldlm_work_bl_ast_lock().
3016          * The 'data' parameter is l_ast_data in the first case and
3017          * callback arguments in the second one. Distinguish them by that.
3018          */
3019         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3020                 goto skip_cos_checks;
3021
3022         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3023                 if (mdt_cos_is_enabled(mdt)) {
3024                         if (!arg->bl_desc->bl_same_client)
3025                                 mdt_set_lock_sync(lock);
3026                 } else if (mdt_slc_is_enabled(mdt) &&
3027                            arg->bl_desc->bl_cos_incompat) {
3028                         mdt_set_lock_sync(lock);
3029                         /*
3030                          * we may do extra commit here, but there is a small
3031                          * window to miss a commit: lock was unlocked (saved),
3032                          * then a conflict lock queued and we come here, but
3033                          * REP-ACK not received, so lock was not converted to
3034                          * COS mode yet.
3035                          * Fortunately this window is quite small, so the
3036                          * extra commit should be rare (not to say distributed
3037                          * operation is rare too).
3038                          */
3039                         commit_async = true;
3040                 }
3041         } else if (lock->l_req_mode == LCK_COS) {
3042                 commit_async = true;
3043         }
3044
3045 skip_cos_checks:
3046         rc = ldlm_blocking_ast_nocheck(lock);
3047
3048         if (commit_async) {
3049                 struct lu_env env;
3050
3051                 rc = lu_env_init(&env, LCT_LOCAL);
3052                 if (unlikely(rc != 0))
3053                         CWARN("%s: lu_env initialization failed, cannot "
3054                               "start asynchronous commit: rc = %d\n",
3055                               obd->obd_name, rc);
3056                 else
3057                         mdt_device_commit_async(&env, mdt);
3058                 lu_env_fini(&env);
3059         }
3060         RETURN(rc);
3061 }
3062
3063 /*
3064  * Blocking AST for cross-MDT lock
3065  *
3066  * Discard lock from uncommitted_slc_locks and cancel it.
3067  *
3068  * \param lock  the lock which blocks a request or cancelling lock
3069  * \param desc  unused
3070  * \param data  unused
3071  * \param flag  indicates whether this cancelling or blocking callback
3072  * \retval      0 on success
3073  * \retval      negative number on error
3074  */
3075 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3076                             void *data, int flag)
3077 {
3078         int rc = 0;
3079         ENTRY;
3080
3081         switch (flag) {
3082         case LDLM_CB_BLOCKING: {
3083                 struct lustre_handle lockh;
3084
3085                 ldlm_lock2handle(lock, &lockh);
3086                 rc = ldlm_cli_cancel(&lockh,
3087                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3088                 if (rc < 0) {
3089                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3090                         RETURN(rc);
3091                 }
3092                 break;
3093         }
3094         case LDLM_CB_CANCELING: {
3095                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3096                 struct mdt_device *mdt =
3097                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3098
3099                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3100
3101                 /* discard slc lock here so that it can be cleaned anytime,
3102                  * especially for cleanup_resource() */
3103                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3104
3105                 /* once we cache lock, l_ast_data is set to mdt_object */
3106                 if (lock->l_ast_data != NULL) {
3107                         struct mdt_object *mo = lock->l_ast_data;
3108                         struct lu_env env;
3109
3110                         rc = lu_env_init(&env, LCT_MD_THREAD);
3111                         if (unlikely(rc != 0)) {
3112                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3113                                       obd->obd_name, mo,
3114                                       PFID(mdt_object_fid(mo)), rc);
3115                                 RETURN(rc);
3116                         }
3117
3118                         if (lock->l_policy_data.l_inodebits.bits &
3119                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3120                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3121                                 mo->mot_cache_attr = 0;
3122                         }
3123                         mdt_object_put(&env, mo);
3124                         lu_env_fini(&env);
3125                 }
3126                 break;
3127         }
3128         default:
3129                 LBUG();
3130         }
3131
3132         RETURN(rc);
3133 }
3134
3135 int mdt_check_resent_lock(struct mdt_thread_info *info,
3136                           struct mdt_object *mo,
3137                           struct mdt_lock_handle *lhc)
3138 {
3139         /* the lock might already be gotten in ldlm_handle_enqueue() */
3140         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3141                 struct ptlrpc_request *req = mdt_info_req(info);
3142                 struct ldlm_lock      *lock;
3143
3144                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3145                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3146                 if (lock == NULL) {
3147                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3148                          * a resend case, however, it could be already destroyed
3149                          * due to client eviction or a raced cancel RPC. */
3150                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3151                                           lhc->mlh_reg_lh.cookie);
3152                         RETURN(-ESTALE);
3153                 }
3154
3155                 if (!fid_res_name_eq(mdt_object_fid(mo),
3156                                      &lock->l_resource->lr_name)) {
3157                         CWARN("%s: Although resent, but still not "
3158                               "get child lock:"DFID"\n",
3159                               info->mti_exp->exp_obd->obd_name,
3160                               PFID(mdt_object_fid(mo)));
3161                         LDLM_LOCK_PUT(lock);
3162                         RETURN(-EPROTO);
3163                 }
3164                 LDLM_LOCK_PUT(lock);
3165                 return 0;
3166         }
3167         return 1;
3168 }
3169
3170 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3171 {
3172         mdt_object_get(NULL, lock->l_ast_data);
3173 }
3174
3175 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3176                                struct mdt_object *o, const struct lu_fid *fid,
3177                                struct lustre_handle *lh, enum ldlm_mode mode,
3178                                __u64 *ibits, __u64 trybits, bool cache)
3179 {
3180         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3181         union ldlm_policy_data *policy = &mti->mti_policy;
3182         struct ldlm_res_id *res_id = &mti->mti_res_id;
3183         int rc = 0;
3184         ENTRY;
3185
3186         LASSERT(mdt_object_remote(o));
3187
3188         fid_build_reg_res_name(fid, res_id);
3189
3190         memset(einfo, 0, sizeof(*einfo));
3191         einfo->ei_type = LDLM_IBITS;
3192         einfo->ei_mode = mode;
3193         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3194         einfo->ei_cb_cp = ldlm_completion_ast;
3195         einfo->ei_enq_slave = 0;
3196         einfo->ei_res_id = res_id;
3197
3198         if (cache) {
3199                 /*
3200                  * if we cache lock, couple lock with mdt_object, so that object
3201                  * can be easily found in lock ASTs.
3202                  */
3203                 einfo->ei_cbdata = o;
3204                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3205         }
3206
3207         memset(policy, 0, sizeof(*policy));
3208         policy->l_inodebits.bits = *ibits;
3209         policy->l_inodebits.try_bits = trybits;
3210
3211         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3212                             policy);
3213
3214         /* Return successfully acquired bits to a caller */
3215         if (rc == 0) {
3216                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3217
3218                 LASSERT(lock);
3219                 *ibits = lock->l_policy_data.l_inodebits.bits;
3220                 LDLM_LOCK_PUT(lock);
3221         }
3222         RETURN(rc);
3223 }
3224
3225 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3226                            const struct lu_fid *fid, struct lustre_handle *lh,
3227                            enum ldlm_mode mode, __u64 ibits, bool cache)
3228 {
3229         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3230                                           cache);
3231 }
3232
3233 static int mdt_object_local_lock(struct mdt_thread_info *info,
3234                                  struct mdt_object *o,
3235                                  struct mdt_lock_handle *lh, __u64 *ibits,
3236                                  __u64 trybits, bool cos_incompat)
3237 {
3238         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3239         union ldlm_policy_data *policy = &info->mti_policy;
3240         struct ldlm_res_id *res_id = &info->mti_res_id;
3241         __u64 dlmflags = 0, *cookie = NULL;
3242         int rc;
3243         ENTRY;
3244
3245         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3246         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3247         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3248         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3249
3250         if (cos_incompat) {
3251                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3252                         lh->mlh_reg_mode == LCK_EX);
3253                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3254         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3255                 dlmflags |= LDLM_FL_COS_ENABLED;
3256         }
3257
3258         /* Only enqueue LOOKUP lock for remote object */
3259         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3260
3261         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3262         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3263             *ibits == MDS_INODELOCK_OPEN)
3264                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3265
3266         if (lh->mlh_type == MDT_PDO_LOCK) {
3267                 /* check for exists after object is locked */
3268                 if (mdt_object_exists(o) == 0) {
3269                         /* Non-existent object shouldn't have PDO lock */
3270                         RETURN(-ESTALE);
3271                 } else {
3272                         /* Non-dir object shouldn't have PDO lock */
3273                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3274                                 RETURN(-ENOTDIR);
3275                 }
3276         }
3277
3278         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3279         dlmflags |= LDLM_FL_ATOMIC_CB;
3280
3281         if (info->mti_exp)
3282                 cookie = &info->mti_exp->exp_handle.h_cookie;
3283
3284         /*
3285          * Take PDO lock on whole directory and build correct @res_id for lock
3286          * on part of directory.
3287          */
3288         if (lh->mlh_pdo_hash != 0) {
3289                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3290                 mdt_lock_pdo_mode(info, o, lh);
3291                 if (lh->mlh_pdo_mode != LCK_NL) {
3292                         /*
3293                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3294                          * is never going to be sent to client and we do not
3295                          * want it slowed down due to possible cancels.
3296                          */
3297                         policy->l_inodebits.bits =
3298                                 *ibits & MDS_INODELOCK_UPDATE;
3299                         policy->l_inodebits.try_bits =
3300                                 trybits & MDS_INODELOCK_UPDATE;
3301                         /* at least one of them should be set */
3302                         LASSERT(policy->l_inodebits.bits |
3303                                 policy->l_inodebits.try_bits);
3304                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3305                                           lh->mlh_pdo_mode, policy, res_id,
3306                                           dlmflags, cookie);
3307                         if (unlikely(rc != 0))
3308                                 GOTO(out_unlock, rc);
3309                 }
3310
3311                 /*
3312                  * Finish res_id initializing by name hash marking part of
3313                  * directory which is taking modification.
3314                  */
3315                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3316         }
3317
3318         policy->l_inodebits.bits = *ibits;
3319         policy->l_inodebits.try_bits = trybits;
3320
3321         /*
3322          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3323          * going to be sent to client. If it is - mdt_intent_policy() path will
3324          * fix it up and turn FL_LOCAL flag off.
3325          */
3326         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3327                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3328                           cookie);
3329 out_unlock:
3330         if (rc != 0)
3331                 mdt_object_unlock(info, o, lh, 1);
3332         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3333                    lh->mlh_pdo_hash != 0 &&
3334                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3335                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3336
3337         /* Return successfully acquired bits to a caller */
3338         if (rc == 0) {
3339                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3340
3341                 LASSERT(lock);
3342                 *ibits = lock->l_policy_data.l_inodebits.bits;
3343                 LDLM_LOCK_PUT(lock);
3344         }
3345         RETURN(rc);
3346 }
3347
3348 static int
3349 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3350                          struct mdt_lock_handle *lh, __u64 *ibits,
3351                          __u64 trybits, bool cos_incompat)
3352 {
3353         struct mdt_lock_handle *local_lh = NULL;
3354         int rc;
3355         ENTRY;
3356
3357         if (!mdt_object_remote(o)) {
3358                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3359                                            cos_incompat);
3360                 RETURN(rc);
3361         }
3362
3363         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3364         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3365                     MDS_INODELOCK_XATTR);
3366
3367         /* Only enqueue LOOKUP lock for remote object */
3368         if (*ibits & MDS_INODELOCK_LOOKUP) {
3369                 __u64 local = MDS_INODELOCK_LOOKUP;
3370
3371                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3372                                            cos_incompat);
3373                 if (rc != ELDLM_OK)
3374                         RETURN(rc);
3375
3376                 local_lh = lh;
3377         }
3378
3379         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3380                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3381                  * enqueue RPC can only request 1 lock, to avoid extra
3382                  * RPC, so it will instead enqueue EX lock for remote
3383                  * object anyway XXX*/
3384                 if (lh->mlh_type == MDT_PDO_LOCK &&
3385                     lh->mlh_pdo_hash != 0) {
3386                         CDEBUG(D_INFO,
3387                                "%s: "DFID" convert PDO lock to EX lock.\n",
3388                                mdt_obd_name(info->mti_mdt),
3389                                PFID(mdt_object_fid(o)));
3390                         lh->mlh_pdo_hash = 0;
3391                         lh->mlh_rreg_mode = LCK_EX;
3392                         lh->mlh_type = MDT_REG_LOCK;
3393                 }
3394
3395                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3396                                                 &lh->mlh_rreg_lh,
3397                                                 lh->mlh_rreg_mode,
3398                                                 ibits, trybits, false);
3399                 if (rc != ELDLM_OK) {
3400                         if (local_lh != NULL)
3401                                 mdt_object_unlock(info, o, local_lh, rc);
3402                         RETURN(rc);
3403                 }
3404         }
3405
3406         /* other components like LFSCK can use lockless access
3407          * and populate cache, so we better invalidate it */
3408         mo_invalidate(info->mti_env, mdt_object_child(o));
3409
3410         RETURN(0);
3411 }
3412
3413 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3414                     struct mdt_lock_handle *lh, __u64 ibits)
3415 {
3416         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3417 }
3418
3419 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3420                           struct mdt_lock_handle *lh, __u64 ibits,
3421                           bool cos_incompat)
3422 {
3423         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3424         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3425                                         cos_incompat);
3426 }
3427
3428 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3429                         struct mdt_lock_handle *lh, __u64 *ibits,
3430                         __u64 trybits, bool cos_incompat)
3431 {
3432         bool trylock_only = *ibits == 0;
3433         int rc;
3434
3435         LASSERT(!(*ibits & trybits));
3436         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3437                                       cos_incompat);
3438         if (rc && trylock_only) { /* clear error for try ibits lock only */
3439                 LASSERT(*ibits == 0);
3440                 rc = 0;
3441         }
3442         return rc;
3443 }
3444
3445 /**
3446  * Save a lock within request object.
3447  *
3448  * Keep the lock referenced until whether client ACK or transaction
3449  * commit happens or release the lock immediately depending on input
3450  * parameters. If COS is ON, a write lock is converted to COS lock
3451  * before saving.
3452  *
3453  * \param info thead info object
3454  * \param h lock handle
3455  * \param mode lock mode
3456  * \param decref force immediate lock releasing
3457  */
3458 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3459                    enum ldlm_mode mode, int decref)
3460 {
3461         ENTRY;
3462
3463         if (lustre_handle_is_used(h)) {
3464                 if (decref || !info->mti_has_trans ||
3465                     !(mode & (LCK_PW | LCK_EX))) {
3466                         mdt_fid_unlock(h, mode);
3467                 } else {
3468                         struct mdt_device *mdt = info->mti_mdt;
3469                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3470                         struct ptlrpc_request *req = mdt_info_req(info);
3471                         bool cos = mdt_cos_is_enabled(mdt);
3472                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3473
3474                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3475                                  h->cookie);
3476
3477                         /* there is no request if mdt_object_unlock() is called
3478                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3479                         if (likely(req != NULL)) {
3480                                 LDLM_DEBUG(lock, "save lock request %p reply "
3481                                         "state %p transno %lld\n", req,
3482                                         req->rq_reply_state, req->rq_transno);
3483                                 if (cos) {
3484                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3485                                         mode = LCK_COS;
3486                                 }
3487                                 if (req->rq_export->exp_disconnected)
3488                                         mdt_fid_unlock(h, mode);
3489                                 else
3490                                         ptlrpc_save_lock(req, h, mode, cos,
3491                                                          convert_lock);
3492                         } else {
3493                                 mdt_fid_unlock(h, mode);
3494                         }
3495                         if (mdt_is_lock_sync(lock)) {
3496                                 CDEBUG(D_HA, "found sync-lock,"
3497                                        " async commit started\n");
3498                                 mdt_device_commit_async(info->mti_env,
3499                                                         mdt);
3500                         }
3501                         LDLM_LOCK_PUT(lock);
3502                 }
3503                 h->cookie = 0ull;
3504         }
3505
3506         EXIT;
3507 }
3508
3509 /**
3510  * Save cross-MDT lock in uncommitted_slc_locks
3511  *
3512  * Keep the lock referenced until transaction commit happens or release the lock
3513  * immediately depending on input parameters.
3514  *
3515  * \param info thead info object
3516  * \param h lock handle
3517  * \param mode lock mode
3518  * \param decref force immediate lock releasing
3519  */
3520 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3521                                  struct mdt_object *o, struct lustre_handle *h,
3522                                  enum ldlm_mode mode, int decref)
3523 {
3524         ENTRY;
3525
3526         if (lustre_handle_is_used(h)) {
3527                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3528
3529                 if (o != NULL &&
3530                     (lock->l_policy_data.l_inodebits.bits &
3531                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3532                         mo_invalidate(info->mti_env, mdt_object_child(o));
3533
3534                 if (decref || !info->mti_has_trans ||
3535                     !(mode & (LCK_PW | LCK_EX))) {
3536                         ldlm_lock_decref_and_cancel(h, mode);
3537                         LDLM_LOCK_PUT(lock);
3538                 } else {
3539                         struct ptlrpc_request *req = mdt_info_req(info);
3540
3541                         LASSERT(req != NULL);
3542                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3543                                           req->rq_transno);
3544                         ldlm_lock_decref(h, mode);
3545                 }
3546                 h->cookie = 0ull;
3547         }
3548
3549         EXIT;
3550 }
3551
3552 /**
3553  * Unlock mdt object.
3554  *
3555  * Immeditely release the regular lock and the PDO lock or save the
3556  * lock in request and keep them referenced until client ACK or
3557  * transaction commit.
3558  *
3559  * \param info thread info object
3560  * \param o mdt object
3561  * \param lh mdt lock handle referencing regular and PDO locks
3562  * \param decref force immediate lock releasing
3563  *
3564  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3565  */
3566 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3567                        struct mdt_lock_handle *lh, int decref)
3568 {
3569         ENTRY;
3570
3571         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3572         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3573         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3574                              decref);
3575
3576         EXIT;
3577 }
3578
3579 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3580                                         const struct lu_fid *f,
3581                                         struct mdt_lock_handle *lh,
3582                                         __u64 ibits)
3583 {
3584         struct mdt_object *o;
3585
3586         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3587         if (!IS_ERR(o)) {
3588                 int rc;
3589
3590                 rc = mdt_object_lock(info, o, lh, ibits);
3591                 if (rc != 0) {
3592                         mdt_object_put(info->mti_env, o);
3593                         o = ERR_PTR(rc);
3594                 }
3595         }
3596         return o;
3597 }
3598
3599 void mdt_object_unlock_put(struct mdt_thread_info * info,
3600                            struct mdt_object * o,
3601                            struct mdt_lock_handle *lh,
3602                            int decref)
3603 {
3604         mdt_object_unlock(info, o, lh, decref);
3605         mdt_object_put(info->mti_env, o);
3606 }
3607
3608 /*
3609  * Generic code handling requests that have struct mdt_body passed in:
3610  *
3611  *  - extract mdt_body from request and save it in @info, if present;
3612  *
3613  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3614  *  @info;
3615  *
3616  *  - if HAS_BODY flag is set for this request type check whether object
3617  *  actually exists on storage (lu_object_exists()).
3618  *
3619  */
3620 static int mdt_body_unpack(struct mdt_thread_info *info,
3621                            enum tgt_handler_flags flags)
3622 {
3623         const struct mdt_body    *body;
3624         struct mdt_object        *obj;
3625         const struct lu_env      *env;
3626         struct req_capsule       *pill;
3627         int                       rc;
3628         ENTRY;
3629
3630         env = info->mti_env;
3631         pill = info->mti_pill;
3632
3633         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3634         if (body == NULL)
3635                 RETURN(-EFAULT);
3636
3637         if (!(body->mbo_valid & OBD_MD_FLID))
3638                 RETURN(0);
3639
3640         if (!fid_is_sane(&body->mbo_fid1)) {
3641                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3642                 RETURN(-EINVAL);
3643         }
3644
3645         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3646         if (!IS_ERR(obj)) {
3647                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3648                         mdt_object_put(env, obj);
3649                         rc = -ENOENT;
3650                 } else {
3651                         info->mti_object = obj;
3652                         rc = 0;
3653                 }
3654         } else
3655                 rc = PTR_ERR(obj);
3656
3657         RETURN(rc);
3658 }
3659
3660 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3661                                    enum tgt_handler_flags flags)
3662 {
3663         struct req_capsule *pill = info->mti_pill;
3664         int rc;
3665
3666         ENTRY;
3667
3668         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3669                 rc = mdt_body_unpack(info, flags);
3670         else
3671                 rc = 0;
3672
3673         if (rc == 0 && (flags & HAS_REPLY)) {
3674                 /* Pack reply. */
3675                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3676                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3677                                              DEF_REP_MD_SIZE);
3678                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3679                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3680                                              RCL_SERVER, 0);
3681
3682                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3683                  * by default. If the target object has more ACL entries, then
3684                  * enlarge the buffer when necessary. */
3685                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3686                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3687                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3688
3689                 mdt_preset_secctx_size(info);
3690
3691                 rc = req_capsule_server_pack(pill);
3692                 if (rc)
3693                         CWARN("%s: cannot pack response: rc = %d\n",
3694                                       mdt_obd_name(info->mti_mdt), rc);
3695         }
3696         RETURN(rc);
3697 }
3698
3699 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3700 {
3701         lh->mlh_type = MDT_NUL_LOCK;
3702         lh->mlh_reg_lh.cookie = 0ull;
3703         lh->mlh_reg_mode = LCK_MINMODE;
3704         lh->mlh_pdo_lh.cookie = 0ull;
3705         lh->mlh_pdo_mode = LCK_MINMODE;
3706         lh->mlh_rreg_lh.cookie = 0ull;
3707         lh->mlh_rreg_mode = LCK_MINMODE;
3708 }
3709
3710 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3711 {
3712         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3713         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3714 }
3715
3716 /*
3717  * Initialize fields of struct mdt_thread_info. Other fields are left in
3718  * uninitialized state, because it's too expensive to zero out whole
3719  * mdt_thread_info (> 1K) on each request arrival.
3720  */
3721 void mdt_thread_info_init(struct ptlrpc_request *req,
3722                           struct mdt_thread_info *info)
3723 {
3724         int i;
3725
3726         info->mti_pill = &req->rq_pill;
3727
3728         /* lock handle */
3729         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3730                 mdt_lock_handle_init(&info->mti_lh[i]);
3731
3732         /* mdt device: it can be NULL while CONNECT */
3733         if (req->rq_export) {
3734                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3735                 info->mti_exp = req->rq_export;
3736         } else
3737                 info->mti_mdt = NULL;
3738         info->mti_env = req->rq_svc_thread->t_env;
3739         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3740
3741         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3742         info->mti_big_buf = LU_BUF_NULL;
3743         info->mti_body = NULL;
3744         info->mti_object = NULL;
3745         info->mti_dlm_req = NULL;
3746         info->mti_has_trans = 0;
3747         info->mti_cross_ref = 0;
3748         info->mti_opdata = 0;
3749         info->mti_big_lmm_used = 0;
3750         info->mti_big_acl_used = 0;
3751         info->mti_som_valid = 0;
3752
3753         info->mti_spec.no_create = 0;
3754         info->mti_spec.sp_rm_entry = 0;
3755         info->mti_spec.sp_permitted = 0;
3756
3757         info->mti_spec.u.sp_ea.eadata = NULL;
3758         info->mti_spec.u.sp_ea.eadatalen = 0;
3759 }
3760
3761 void mdt_thread_info_fini(struct mdt_thread_info *info)
3762 {
3763         int i;
3764
3765         if (info->mti_object != NULL) {
3766                 mdt_object_put(info->mti_env, info->mti_object);
3767                 info->mti_object = NULL;
3768         }
3769
3770         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3771                 mdt_lock_handle_fini(&info->mti_lh[i]);
3772         info->mti_env = NULL;
3773         info->mti_pill = NULL;
3774         info->mti_exp = NULL;
3775         info->mti_mdt = NULL;
3776
3777         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3778                 lu_buf_free(&info->mti_big_buf);
3779 }
3780
3781 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3782 {
3783         struct mdt_thread_info  *mti;
3784
3785         mti = mdt_th_info(tsi->tsi_env);
3786         LASSERT(mti != NULL);
3787
3788         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3789         if (tsi->tsi_corpus != NULL) {
3790                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3791                 lu_object_get(tsi->tsi_corpus);
3792         }
3793         mti->mti_body = tsi->tsi_mdt_body;
3794         mti->mti_dlm_req = tsi->tsi_dlm_req;
3795
3796         return mti;
3797 }
3798
3799 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3800 {
3801         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3802             cfs_fail_val ==
3803             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
3804                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
3805
3806         return tgt_connect(tsi);
3807 }
3808
3809 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3810                               struct mdt_thread_info *info,
3811                               struct ldlm_lock **lockp, __u64 flags)
3812 {
3813         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3814                                    lockp, flags);
3815 }
3816 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3817                           struct mdt_thread_info *info,
3818                           struct ldlm_lock **lockp, __u64 flags)
3819 {
3820         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3821                                lockp, flags);
3822 }
3823
3824 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3825                             struct ldlm_lock **lockp,
3826                             struct mdt_lock_handle *lh,
3827                             __u64 flags, int result)
3828 {
3829         struct ptlrpc_request  *req = mdt_info_req(info);
3830         struct ldlm_lock       *lock = *lockp;
3831         struct ldlm_lock       *new_lock;
3832
3833         /* If possible resent found a lock, @lh is set to its handle */
3834         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3835
3836         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3837                 lh->mlh_reg_lh.cookie = 0;
3838                 RETURN(0);
3839         }
3840
3841         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3842                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3843                  * a resend case, however, it could be already destroyed
3844                  * due to client eviction or a raced cancel RPC. */
3845                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3846                                   lh->mlh_reg_lh.cookie);
3847                 lh->mlh_reg_lh.cookie = 0;
3848                 RETURN(-ESTALE);
3849         }
3850
3851         LASSERTF(new_lock != NULL,
3852                  "lockh %#llx flags %#llx : rc = %d\n",
3853                  lh->mlh_reg_lh.cookie, flags, result);
3854
3855         /*
3856          * If we've already given this lock to a client once, then we should
3857          * have no readers or writers.  Otherwise, we should have one reader
3858          * _or_ writer ref (which will be zeroed below) before returning the
3859          * lock to a client.
3860          */
3861         if (new_lock->l_export == req->rq_export) {
3862                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3863         } else {
3864                 LASSERT(new_lock->l_export == NULL);
3865                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3866         }
3867
3868         *lockp = new_lock;
3869
3870         if (new_lock->l_export == req->rq_export) {
3871                 /*
3872                  * Already gave this to the client, which means that we
3873                  * reconstructed a reply.
3874                  */
3875                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3876                         MSG_RESENT);
3877
3878                 LDLM_LOCK_RELEASE(new_lock);
3879                 lh->mlh_reg_lh.cookie = 0;
3880                 RETURN(ELDLM_LOCK_REPLACED);
3881         }
3882
3883         /*
3884          * Fixup the lock to be given to the client.
3885          */
3886         lock_res_and_lock(new_lock);
3887         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3888          * possible blocking AST. */
3889         while (new_lock->l_readers > 0) {
3890                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3891                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3892                 new_lock->l_readers--;
3893         }
3894         while (new_lock->l_writers > 0) {
3895                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3896                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3897                 new_lock->l_writers--;
3898         }
3899
3900         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3901         new_lock->l_blocking_ast = lock->l_blocking_ast;
3902         new_lock->l_completion_ast = lock->l_completion_ast;
3903         if (ldlm_has_dom(new_lock))
3904                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3905         new_lock->l_remote_handle = lock->l_remote_handle;
3906         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3907
3908         unlock_res_and_lock(new_lock);
3909
3910         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3911                      &new_lock->l_remote_handle,
3912                      &new_lock->l_exp_hash);
3913
3914         LDLM_LOCK_RELEASE(new_lock);
3915         lh->mlh_reg_lh.cookie = 0;
3916
3917         RETURN(ELDLM_LOCK_REPLACED);
3918 }
3919
3920 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3921                              struct ldlm_lock *new_lock,
3922                              struct mdt_lock_handle *lh, __u64 flags)
3923 {
3924         struct ptlrpc_request  *req = mdt_info_req(info);
3925         struct ldlm_request    *dlmreq;
3926
3927         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3928                 return;
3929
3930         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3931
3932         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3933          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3934          * initialized. */
3935         if (flags & LDLM_FL_RESENT) {
3936                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3937                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3938
3939                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3940                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3941                           lh->mlh_reg_lh.cookie);
3942                 return;
3943         }
3944
3945         /*
3946          * If the xid matches, then we know this is a resent request, and allow
3947          * it. (It's probably an OPEN, for which we don't send a lock.
3948          */
3949         if (req_can_reconstruct(req, NULL) != 0)
3950                 return;
3951
3952         /*
3953          * This remote handle isn't enqueued, so we never received or processed
3954          * this request.  Clear MSG_RESENT, because it can be handled like any
3955          * normal request now.
3956          */
3957         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3958
3959         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3960                   dlmreq->lock_handle[0].cookie);
3961 }
3962
3963 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3964                                struct mdt_thread_info *info,
3965                                struct ldlm_lock **lockp,
3966                                __u64 flags)
3967 {
3968         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3969         struct ldlm_reply      *ldlm_rep = NULL;
3970         int rc;
3971         ENTRY;
3972
3973         /*
3974          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3975          * (for the resend case) or a new lock. Below we will use it to
3976          * replace the original lock.
3977          */
3978         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3979         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3980                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3981                 rc = mdt_object_lock(info, info->mti_object, lhc,
3982                                      MDS_INODELOCK_XATTR);
3983                 if (rc)
3984                         return rc;
3985         }
3986
3987         rc = mdt_getxattr(info);
3988
3989         if (mdt_info_req(info)->rq_repmsg != NULL)
3990                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3991
3992         if (ldlm_rep == NULL ||
3993             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3994                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3995                 if (is_serious(rc))
3996                         RETURN(rc);
3997                 else
3998                         RETURN(err_serious(-EFAULT));
3999         }
4000
4001         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4002
4003         /* This is left for interop instead of adding a new interop flag.
4004          * LU-7433 */
4005 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4006         if (ldlm_rep->lock_policy_res2) {
4007                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4008                 RETURN(ELDLM_LOCK_ABORTED);
4009         }
4010 #endif
4011
4012         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4013         RETURN(rc);
4014 }
4015
4016 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4017                               struct mdt_thread_info *info,
4018                               struct ldlm_lock **lockp,
4019                               __u64 flags)
4020 {
4021         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4022         __u64                   child_bits;
4023         struct ldlm_reply      *ldlm_rep;
4024         struct mdt_body        *reqbody;
4025         struct mdt_body        *repbody;
4026         int                     rc, rc2;
4027         ENTRY;
4028
4029         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4030         LASSERT(reqbody);
4031
4032         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4033         LASSERT(repbody);
4034
4035         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4036         repbody->mbo_eadatasize = 0;
4037         repbody->mbo_aclsize = 0;
4038
4039         switch (it_opc) {
4040         case IT_LOOKUP:
4041                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4042                 break;
4043         case IT_GETATTR:
4044                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4045                              MDS_INODELOCK_PERM;
4046                 break;
4047         default:
4048                 CERROR("%s: unsupported intent %#x\n",
4049                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4050                 GOTO(out_shrink, rc = -EINVAL);
4051         }
4052
4053         rc = mdt_init_ucred_intent_getattr(info, reqbody);
4054         if (rc)
4055                 GOTO(out_shrink, rc);
4056
4057         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4058         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4059
4060         /* Get lock from request for possible resent case. */
4061         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4062
4063         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4064         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4065
4066         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4067                 ldlm_rep->lock_policy_res2 = 0;
4068         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4069             ldlm_rep->lock_policy_res2) {
4070                 lhc->mlh_reg_lh.cookie = 0ull;
4071                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4072         }
4073
4074         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4075         EXIT;
4076 out_ucred:
4077         mdt_exit_ucred(info);
4078 out_shrink:
4079         mdt_client_compatibility(info);
4080         rc2 = mdt_fix_reply(info);
4081         if (rc == 0)
4082                 rc = rc2;
4083         return rc;
4084 }
4085
4086 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4087                              struct mdt_thread_info *info,
4088                              struct ldlm_lock **lockp,
4089                              __u64 flags)
4090 {
4091         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4092         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4093         struct layout_intent *intent;
4094         struct ldlm_reply *ldlm_rep;
4095         struct lu_fid *fid = &info->mti_tmp_fid2;
4096         struct mdt_object *obj = NULL;
4097         int layout_size = 0;
4098         struct lu_buf *buf = &layout.mlc_buf;
4099         int rc = 0;
4100
4101         ENTRY;
4102
4103         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4104
4105         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4106         if (intent == NULL)
4107                 RETURN(-EPROTO);
4108
4109         CDEBUG(D_INFO, DFID "got layout change request from client: "
4110                "opc:%u flags:%#x extent "DEXT"\n",
4111                PFID(fid), intent->li_opc, intent->li_flags,
4112                PEXT(&intent->li_extent));
4113
4114         switch (intent->li_opc) {
4115         case LAYOUT_INTENT_TRUNC:
4116         case LAYOUT_INTENT_WRITE:
4117                 layout.mlc_opc = MD_LAYOUT_WRITE;
4118                 layout.mlc_intent = intent;
4119                 break;
4120         case LAYOUT_INTENT_ACCESS:
4121                 break;
4122         case LAYOUT_INTENT_READ:
4123         case LAYOUT_INTENT_GLIMPSE:
4124         case LAYOUT_INTENT_RELEASE:
4125         case LAYOUT_INTENT_RESTORE:
4126                 CERROR("%s: Unsupported layout intent opc %d\n",
4127                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4128                 RETURN(-ENOTSUPP);
4129         default:
4130                 CERROR("%s: Unknown layout intent opc %d\n",
4131                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4132                 RETURN(-EINVAL);
4133         }
4134
4135         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4136         if (IS_ERR(obj))
4137                 RETURN(PTR_ERR(obj));
4138
4139         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4140                 /* if layout is going to be changed don't use the current EA
4141                  * size but the maximum one. That buffer will be shrinked
4142                  * to the actual size in req_capsule_shrink() before reply.
4143                  */
4144                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4145                         layout_size = info->mti_mdt->mdt_max_mdsize;
4146                 } else {
4147                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4148                         if (layout_size < 0)
4149                                 GOTO(out, rc = layout_size);
4150
4151                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4152                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4153                 }
4154                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4155                        mdt_obd_name(info->mti_mdt), layout_size);
4156         }
4157
4158         /*
4159          * set reply buffer size, so that ldlm_handle_enqueue0()->
4160          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4161          */
4162         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4163                              layout_size);
4164         rc = req_capsule_server_pack(info->mti_pill);
4165         if (rc)
4166                 GOTO(out, rc);
4167
4168         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4169         if (!ldlm_rep)
4170                 GOTO(out, rc = -EPROTO);
4171
4172         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4173
4174         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4175         if (layout.mlc_opc == MD_LAYOUT_NOP)
4176                 GOTO(out, rc = 0);
4177
4178         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4179         if (rc < 0)
4180                 GOTO(out, rc);
4181         if (rc == 1) {
4182                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4183                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4184                 GOTO(out, rc);
4185         }
4186
4187         buf->lb_buf = NULL;
4188         buf->lb_len = 0;
4189         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4190                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4191                                                      &RMF_EADATA);
4192                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4193                                                      &RMF_EADATA, RCL_CLIENT);
4194                 /*
4195                  * If it's a replay of layout write intent RPC, the client has
4196                  * saved the extended lovea when it get reply then.
4197                  */
4198                 if (buf->lb_len > 0)
4199                         mdt_fix_lov_magic(info, buf->lb_buf);
4200         }
4201
4202         /* Get lock from request for possible resent case. */
4203         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4204         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4205
4206         /*
4207          * Instantiate some layout components, if @buf contains lovea, then it's
4208          * a replay of the layout intent write RPC.
4209          */
4210         rc = mdt_layout_change(info, obj, lhc, &layout);
4211         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4212
4213         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4214                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4215                 if (rc == ELDLM_LOCK_REPLACED &&
4216                     (*lockp)->l_granted_mode == LCK_EX)
4217                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4218         }
4219
4220         EXIT;
4221 out:
4222         mdt_object_put(info->mti_env, obj);
4223         return rc;
4224 }
4225
4226 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4227                            struct mdt_thread_info *info,
4228                            struct ldlm_lock **lockp,
4229                            __u64 flags)
4230 {
4231         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4232         struct ldlm_reply      *rep = NULL;
4233         long                    opc;
4234         int                     rc;
4235
4236         static const struct req_format *intent_fmts[REINT_MAX] = {
4237                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4238                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4239         };
4240
4241         ENTRY;
4242
4243         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4244         if (opc < 0)
4245                 RETURN(opc);
4246
4247         /* Get lock from request for possible resent case. */
4248         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4249
4250         rc = mdt_reint_internal(info, lhc, opc);
4251
4252         /* Check whether the reply has been packed successfully. */
4253         if (mdt_info_req(info)->rq_repmsg != NULL)
4254                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4255         if (rep == NULL) {
4256                 if (is_serious(rc))
4257                         RETURN(rc);
4258                 else
4259                         RETURN(err_serious(-EFAULT));
4260         }
4261
4262         /* MDC expects this in any case */
4263         if (rc != 0)
4264                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4265
4266         /* the open lock or the lock for cross-ref object should be
4267          * returned to the client */
4268         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4269             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4270                 rep->lock_policy_res2 = 0;
4271                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4272                 RETURN(rc);
4273         }
4274
4275         rep->lock_policy_res2 = clear_serious(rc);
4276
4277         if (rep->lock_policy_res2 == -ENOENT &&
4278             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4279             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4280                 rep->lock_policy_res2 = 0;
4281
4282         lhc->mlh_reg_lh.cookie = 0ull;
4283         if (rc == -ENOTCONN || rc == -ENODEV ||
4284             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4285                 /*
4286                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4287                  * will be returned by rq_status, and client at ptlrpc layer
4288                  * will detect this, then disconnect, reconnect the import
4289                  * immediately, instead of impacting the following the rpc.
4290                  */
4291                 RETURN(rc);
4292         }
4293         /*
4294          * For other cases, the error will be returned by intent, and client
4295          * will retrieve the result from intent.
4296          */
4297         RETURN(ELDLM_LOCK_ABORTED);
4298 }
4299
4300 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4301                           struct mdt_thread_info *info,
4302                           struct ldlm_lock **lockp,
4303                           u64 flags /* LDLM_FL_* */)
4304 {
4305         struct req_capsule *pill = info->mti_pill;
4306         struct ptlrpc_request *req = mdt_info_req(info);
4307         const struct req_format *it_format;
4308         int (*it_handler)(enum ldlm_intent_flags,
4309                           struct mdt_thread_info *,
4310                           struct ldlm_lock **,
4311                           u64);
4312         enum tgt_handler_flags it_handler_flags = 0;
4313         struct ldlm_reply *rep;
4314         bool check_mdt_object = false;
4315         int rc;
4316         ENTRY;
4317
4318         switch (it_opc) {
4319         case IT_OPEN:
4320         case IT_OPEN|IT_CREAT:
4321                 /*
4322                  * OCREAT is not a IS_MUTABLE request since the file may
4323                  * already exist. We do the extra check of
4324                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4325                  * really need to create the object.
4326                  */
4327                 it_format = &RQF_LDLM_INTENT;
4328                 it_handler = &mdt_intent_open;
4329                 break;
4330         case IT_GETATTR:
4331                 check_mdt_object = true;
4332                 /* fallthrough */
4333         case IT_LOOKUP:
4334                 it_format = &RQF_LDLM_INTENT_GETATTR;
4335                 it_handler = &mdt_intent_getattr;
4336                 it_handler_flags = HAS_REPLY;
4337                 break;
4338         case IT_GETXATTR:
4339                 check_mdt_object = true;
4340                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4341                 it_handler = &mdt_intent_getxattr;
4342                 it_handler_flags = HAS_BODY;
4343                 break;
4344         case IT_LAYOUT:
4345                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4346                 it_handler = &mdt_intent_layout;
4347                 break;
4348         case IT_GLIMPSE:
4349                 it_format = &RQF_LDLM_INTENT;
4350                 it_handler = &mdt_intent_glimpse;
4351                 break;
4352         case IT_BRW:
4353                 it_format = &RQF_LDLM_INTENT;
4354                 it_handler = &mdt_intent_brw;
4355                 break;
4356         case IT_QUOTA_DQACQ:
4357         case IT_QUOTA_CONN: {
4358                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4359
4360                 if (qmt == NULL)
4361                         RETURN(-EOPNOTSUPP);
4362
4363                 if (mdt_rdonly(req->rq_export))
4364                         RETURN(-EROFS);
4365
4366                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4367                 /* pass the request to quota master */
4368                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4369                                                  mdt_info_req(info), lockp,
4370                                                  flags);
4371                 RETURN(rc);
4372         }
4373         default:
4374                 CERROR("%s: unknown intent code %#x\n",
4375                        mdt_obd_name(info->mti_mdt), it_opc);
4376                 RETURN(-EPROTO);
4377         }
4378
4379         req_capsule_extend(pill, it_format);
4380
4381         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4382         if (rc < 0)
4383                 RETURN(rc);
4384
4385         if (unlikely(info->mti_object == NULL && check_mdt_object))
4386                 RETURN(-EPROTO);
4387
4388         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4389                 RETURN(-EROFS);
4390
4391         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4392
4393         /* execute policy */
4394         rc = (*it_handler)(it_opc, info, lockp, flags);
4395
4396         /* Check whether the reply has been packed successfully. */
4397         if (req->rq_repmsg != NULL) {
4398                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4399                 rep->lock_policy_res2 =
4400                         ptlrpc_status_hton(rep->lock_policy_res2);
4401         }
4402
4403         RETURN(rc);
4404 }
4405
4406 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4407                                     enum ldlm_intent_flags it_opc)
4408 {
4409         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4410
4411         /* update stats when IT code is known */
4412         if (srv_stats != NULL)
4413                 lprocfs_counter_incr(srv_stats,
4414                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4415                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4416 }
4417
4418 static int mdt_intent_policy(const struct lu_env *env,
4419                              struct ldlm_namespace *ns,
4420                              struct ldlm_lock **lockp,
4421                              void *req_cookie,
4422                              enum ldlm_mode mode,
4423                              __u64 flags, void *data)
4424 {
4425         struct tgt_session_info *tsi;
4426         struct mdt_thread_info  *info;
4427         struct ptlrpc_request   *req  =  req_cookie;
4428         struct ldlm_intent      *it;
4429         struct req_capsule      *pill;
4430         const struct ldlm_lock_desc *ldesc;
4431         int rc;
4432
4433         ENTRY;
4434
4435         LASSERT(req != NULL);
4436
4437         tsi = tgt_ses_info(env);
4438
4439         info = tsi2mdt_info(tsi);
4440         LASSERT(info != NULL);
4441         pill = info->mti_pill;
4442         LASSERT(pill->rc_req == req);
4443         ldesc = &info->mti_dlm_req->lock_desc;
4444
4445         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4446                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4447                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4448                 if (it != NULL) {
4449                         mdt_ptlrpc_stats_update(req, it->opc);
4450                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4451                         if (rc == 0)
4452                                 rc = ELDLM_OK;
4453
4454                         /* Lock without inodebits makes no sense and will oops
4455                          * later in ldlm. Let's check it now to see if we have
4456                          * ibits corrupted somewhere in mdt_intent_opc().
4457                          * The case for client miss to set ibits has been
4458                          * processed by others. */
4459                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4460                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4461                 } else {
4462                         rc = err_serious(-EFAULT);
4463                 }
4464         } else {
4465                 /* No intent was provided */
4466                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4467                 rc = req_capsule_server_pack(pill);
4468                 if (rc)
4469                         rc = err_serious(rc);
4470         }
4471         mdt_thread_info_fini(info);
4472         RETURN(rc);
4473 }
4474
4475 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4476 {
4477         struct seq_server_site  *ss = mdt_seq_site(mdt);
4478
4479         if (ss->ss_node_id == 0)
4480                 return;
4481
4482         if (ss->ss_client_seq != NULL) {
4483                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4484                 ss->ss_client_seq->lcs_exp = NULL;
4485         }
4486
4487         if (ss->ss_server_fld != NULL) {
4488                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4489                 ss->ss_server_fld->lsf_control_exp = NULL;
4490         }
4491 }
4492
4493 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4494 {
4495         struct seq_server_site *ss = mdt_seq_site(mdt);
4496
4497         if (ss == NULL)
4498                 return;
4499
4500         if (ss->ss_server_seq != NULL)
4501                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4502 }
4503
4504 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4505 {
4506         mdt_seq_fini_cli(mdt);
4507         mdt_deregister_seq_exp(mdt);
4508
4509         return seq_site_fini(env, mdt_seq_site(mdt));
4510 }
4511
4512 /**
4513  * It will retrieve its FLDB entries from MDT0, and it only happens
4514  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4515  * and it needs to refresh FLDB from the MDT0.
4516  **/
4517 static int mdt_register_lwp_callback(void *data)
4518 {
4519         struct lu_env           env;
4520         struct mdt_device       *mdt = data;
4521         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4522         int                     rc;
4523         ENTRY;
4524
4525         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4526
4527         rc = lu_env_init(&env, LCT_MD_THREAD);
4528         if (rc < 0) {
4529                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4530                 RETURN(rc);
4531         }
4532
4533         /* Allocate new sequence now to avoid creating local transaction
4534          * in the normal transaction process */
4535         rc = seq_server_check_and_alloc_super(&env,
4536                                               mdt_seq_site(mdt)->ss_server_seq);
4537         if (rc < 0)
4538                 GOTO(out, rc);
4539
4540         if (fld->lsf_new) {
4541                 rc = fld_update_from_controller(&env, fld);
4542                 if (rc != 0) {
4543                         CERROR("%s: cannot update controller: rc = %d\n",
4544                                mdt_obd_name(mdt), rc);
4545                         GOTO(out, rc);
4546                 }
4547         }
4548 out:
4549         lu_env_fini(&env);
4550         RETURN(rc);
4551 }
4552
4553 static int mdt_register_seq_exp(struct mdt_device *mdt)
4554 {
4555         struct seq_server_site  *ss = mdt_seq_site(mdt);
4556         char                    *lwp_name = NULL;
4557         int                     rc;
4558
4559         if (ss->ss_node_id == 0)
4560                 return 0;
4561
4562         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4563         if (lwp_name == NULL)
4564                 GOTO(out_free, rc = -ENOMEM);
4565
4566         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4567         if (rc != 0)
4568                 GOTO(out_free, rc);
4569
4570         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4571                                       NULL, NULL);
4572         if (rc != 0)
4573                 GOTO(out_free, rc);
4574
4575         rc = lustre_register_lwp_item(lwp_name,
4576                                       &ss->ss_server_fld->lsf_control_exp,
4577                                       mdt_register_lwp_callback, mdt);
4578         if (rc != 0) {
4579                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4580                 ss->ss_client_seq->lcs_exp = NULL;
4581                 GOTO(out_free, rc);
4582         }
4583 out_free:
4584         if (lwp_name != NULL)
4585                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4586
4587         return rc;
4588 }
4589
4590 /*
4591  * Init client sequence manager which is used by local MDS to talk to sequence
4592  * controller on remote node.
4593  */
4594 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4595 {
4596         struct seq_server_site *ss = mdt_seq_site(mdt);
4597         char *prefix;
4598         ENTRY;
4599
4600         /* check if this is adding the first MDC and controller is not yet
4601          * initialized. */
4602         OBD_ALLOC_PTR(ss->ss_client_seq);
4603         if (ss->ss_client_seq == NULL)
4604                 RETURN(-ENOMEM);
4605
4606         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4607         if (prefix == NULL) {
4608                 OBD_FREE_PTR(ss->ss_client_seq);
4609                 ss->ss_client_seq = NULL;
4610                 RETURN(-ENOMEM);
4611         }
4612
4613         /* Note: seq_client_fini will be called in seq_site_fini */
4614         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4615         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4616                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4617                                                             NULL);
4618         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4619
4620         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
4621 }
4622
4623 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4624 {
4625         struct seq_server_site  *ss;
4626         int                     rc;
4627         ENTRY;
4628
4629         ss = mdt_seq_site(mdt);
4630         /* init sequence controller server(MDT0) */
4631         if (ss->ss_node_id == 0) {
4632                 OBD_ALLOC_PTR(ss->ss_control_seq);
4633                 if (ss->ss_control_seq == NULL)
4634                         RETURN(-ENOMEM);
4635
4636                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4637                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4638                                      ss);
4639                 if (rc)
4640                         GOTO(out_seq_fini, rc);
4641         }
4642
4643         /* Init normal sequence server */
4644         OBD_ALLOC_PTR(ss->ss_server_seq);
4645         if (ss->ss_server_seq == NULL)
4646                 GOTO(out_seq_fini, rc = -ENOMEM);
4647
4648         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4649                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4650         if (rc)
4651                 GOTO(out_seq_fini, rc);
4652
4653         /* init seq client for seq server to talk to seq controller(MDT0) */
4654         rc = mdt_seq_init_cli(env, mdt);
4655         if (rc != 0)
4656                 GOTO(out_seq_fini, rc);
4657
4658         if (ss->ss_node_id != 0)
4659                 /* register controller export through lwp */
4660                 rc = mdt_register_seq_exp(mdt);
4661
4662         EXIT;
4663 out_seq_fini:
4664         if (rc)
4665                 mdt_seq_fini(env, mdt);
4666
4667         return rc;
4668 }
4669
4670 /*
4671  * FLD wrappers
4672  */
4673 static int mdt_fld_fini(const struct lu_env *env,
4674                         struct mdt_device *m)
4675 {
4676         struct seq_server_site *ss = mdt_seq_site(m);
4677         ENTRY;
4678
4679         if (ss && ss->ss_server_fld) {
4680                 fld_server_fini(env, ss->ss_server_fld);
4681                 OBD_FREE_PTR(ss->ss_server_fld);
4682                 ss->ss_server_fld = NULL;
4683         }
4684
4685         RETURN(0);
4686 }
4687
4688 static int mdt_fld_init(const struct lu_env *env,
4689                         const char *uuid,
4690                         struct mdt_device *m)
4691 {
4692         struct seq_server_site *ss;
4693         int rc;
4694         ENTRY;
4695
4696         ss = mdt_seq_site(m);
4697
4698         OBD_ALLOC_PTR(ss->ss_server_fld);
4699         if (ss->ss_server_fld == NULL)
4700                 RETURN(rc = -ENOMEM);
4701
4702         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4703                              LU_SEQ_RANGE_MDT);
4704         if (rc) {
4705                 OBD_FREE_PTR(ss->ss_server_fld);
4706                 ss->ss_server_fld = NULL;
4707                 RETURN(rc);
4708         }
4709
4710         RETURN(0);
4711 }
4712
4713 static void mdt_stack_pre_fini(const struct lu_env *env,
4714                            struct mdt_device *m, struct lu_device *top)
4715 {
4716         struct lustre_cfg_bufs  *bufs;
4717         struct lustre_cfg       *lcfg;
4718         struct mdt_thread_info  *info;
4719         ENTRY;
4720
4721         LASSERT(top);
4722
4723         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4724         LASSERT(info != NULL);
4725
4726         bufs = &info->mti_u.bufs;
4727
4728         LASSERT(m->mdt_child_exp);
4729         LASSERT(m->mdt_child_exp->exp_obd);
4730
4731         /* process cleanup, pass mdt obd name to get obd umount flags */
4732         /* XXX: this is needed because all layers are referenced by
4733          * objects (some of them are pinned by osd, for example *
4734          * the proper solution should be a model where object used
4735          * by osd only doesn't have mdt/mdd slices -bzzz */
4736         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4737         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4738         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4739         if (!lcfg)
4740                 RETURN_EXIT;
4741         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4742
4743         top->ld_ops->ldo_process_config(env, top, lcfg);
4744         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4745         EXIT;
4746 }
4747
4748 static void mdt_stack_fini(const struct lu_env *env,
4749                            struct mdt_device *m, struct lu_device *top)
4750 {
4751         struct obd_device       *obd = mdt2obd_dev(m);
4752         struct lustre_cfg_bufs  *bufs;
4753         struct lustre_cfg       *lcfg;
4754         struct mdt_thread_info  *info;
4755         char                     flags[3] = "";
4756         ENTRY;
4757
4758         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4759         LASSERT(info != NULL);
4760
4761         lu_dev_del_linkage(top->ld_site, top);
4762
4763         lu_site_purge(env, top->ld_site, -1);
4764
4765         bufs = &info->mti_u.bufs;
4766         /* process cleanup, pass mdt obd name to get obd umount flags */
4767         /* another purpose is to let all layers to release their objects */
4768         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4769         if (obd->obd_force)
4770                 strcat(flags, "F");
4771         if (obd->obd_fail)
4772                 strcat(flags, "A");
4773         lustre_cfg_bufs_set_string(bufs, 1, flags);
4774         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4775         if (!lcfg)
4776                 RETURN_EXIT;
4777         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4778
4779         LASSERT(top);
4780         top->ld_ops->ldo_process_config(env, top, lcfg);
4781         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4782
4783         lu_site_purge(env, top->ld_site, -1);
4784
4785         m->mdt_child = NULL;
4786         m->mdt_bottom = NULL;
4787
4788         obd_disconnect(m->mdt_child_exp);
4789         m->mdt_child_exp = NULL;
4790
4791         obd_disconnect(m->mdt_bottom_exp);
4792         m->mdt_child_exp = NULL;
4793 }
4794
4795 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4796                                const char *next, struct obd_export **exp)
4797 {
4798         struct obd_connect_data *data = NULL;
4799         struct obd_device       *obd;
4800         int                      rc;
4801         ENTRY;
4802
4803         OBD_ALLOC_PTR(data);
4804         if (data == NULL)
4805                 GOTO(out, rc = -ENOMEM);
4806
4807         obd = class_name2obd(next);
4808         if (obd == NULL) {
4809                 CERROR("%s: can't locate next device: %s\n",
4810                        mdt_obd_name(m), next);
4811                 GOTO(out, rc = -ENOTCONN);
4812         }
4813
4814         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4815         data->ocd_version = LUSTRE_VERSION_CODE;
4816
4817         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4818         if (rc) {
4819                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4820                        mdt_obd_name(m), next, rc);
4821                 GOTO(out, rc);
4822         }
4823
4824 out:
4825         if (data)
4826                 OBD_FREE_PTR(data);
4827         RETURN(rc);
4828 }
4829
4830 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4831                           struct lustre_cfg *cfg)
4832 {
4833         char                   *dev = lustre_cfg_string(cfg, 0);
4834         int                     rc, name_size, uuid_size;
4835         char                   *name, *uuid, *p;
4836         struct lustre_cfg_bufs *bufs;
4837         struct lustre_cfg      *lcfg;
4838         struct obd_device      *obd;
4839         struct lustre_profile  *lprof;
4840         struct lu_site         *site;
4841         ENTRY;
4842
4843         /* in 1.8 we had the only device in the stack - MDS.
4844          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4845          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4846          * to generate names and setup MDT, MDD. MDT will be using
4847          * generated name to connect to MDD. for MDD the next device
4848          * will be LOD with name taken from so called "profile" which
4849          * is generated by mount_option line
4850          *
4851          * 1.8 MGS generates config. commands like this:
4852          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4853          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4854          * 2.0 MGS generates config. commands like this:
4855          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4856          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4857          *                    3:lustre-MDT0000-mdtlov  4:f
4858          *
4859          * we generate MDD name from MDT one, just replacing T with D
4860          *
4861          * after all the preparations, the logical equivalent will be
4862          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4863          *                    3:lustre-MDT0000-mdtlov  4:f
4864          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4865          *                    3:lustre-MDD0000  4:f
4866          *
4867          *  notice we build the stack from down to top: MDD first, then MDT */
4868
4869         name_size = MAX_OBD_NAME;
4870         uuid_size = MAX_OBD_NAME;
4871
4872         OBD_ALLOC(name, name_size);
4873         OBD_ALLOC(uuid, uuid_size);
4874         if (name == NULL || uuid == NULL)
4875                 GOTO(cleanup_mem, rc = -ENOMEM);
4876
4877         OBD_ALLOC_PTR(bufs);
4878         if (!bufs)
4879                 GOTO(cleanup_mem, rc = -ENOMEM);
4880
4881         strcpy(name, dev);
4882         p = strstr(name, "-MDT");
4883         if (p == NULL)
4884                 GOTO(free_bufs, rc = -ENOMEM);
4885         p[3] = 'D';
4886
4887         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4888
4889         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4890         if (lprof == NULL || lprof->lp_dt == NULL) {
4891                 CERROR("can't find the profile: %s\n",
4892                        lustre_cfg_string(cfg, 0));
4893                 GOTO(free_bufs, rc = -EINVAL);
4894         }
4895
4896         lustre_cfg_bufs_reset(bufs, name);
4897         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4898         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4899         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4900
4901         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4902         if (!lcfg)
4903                 GOTO(put_profile, rc = -ENOMEM);
4904         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4905
4906         rc = class_attach(lcfg);
4907         if (rc)
4908                 GOTO(lcfg_cleanup, rc);
4909
4910         obd = class_name2obd(name);
4911         if (!obd) {
4912                 CERROR("Can not find obd %s (%s in config)\n",
4913                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4914                 GOTO(lcfg_cleanup, rc = -EINVAL);
4915         }
4916
4917         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4918
4919         lustre_cfg_bufs_reset(bufs, name);
4920         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4921         lustre_cfg_bufs_set_string(bufs, 2, dev);
4922         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4923
4924         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4925         if (!lcfg)
4926                 GOTO(class_detach, rc = -ENOMEM);
4927         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4928
4929         rc = class_setup(obd, lcfg);
4930         if (rc)
4931                 GOTO(class_detach, rc);
4932
4933         /* connect to MDD we just setup */
4934         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4935         if (rc)
4936                 GOTO(class_detach, rc);
4937
4938         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4939         LASSERT(site);
4940         LASSERT(mdt_lu_site(mdt) == NULL);
4941         mdt->mdt_lu_dev.ld_site = site;
4942         site->ls_top_dev = &mdt->mdt_lu_dev;
4943         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4944
4945         /* now connect to bottom OSD */
4946         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4947         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4948         if (rc)
4949                 GOTO(class_detach, rc);
4950         mdt->mdt_bottom =
4951                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4952
4953         rc = lu_env_refill((struct lu_env *)env);
4954         if (rc != 0)
4955                 CERROR("Failure to refill session: '%d'\n", rc);
4956
4957         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4958
4959         EXIT;
4960 class_detach:
4961         if (rc)
4962                 class_detach(obd, lcfg);
4963 lcfg_cleanup:
4964         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4965 put_profile:
4966         class_put_profile(lprof);
4967 free_bufs:
4968         OBD_FREE_PTR(bufs);
4969 cleanup_mem:
4970         if (name)
4971                 OBD_FREE(name, name_size);
4972         if (uuid)
4973                 OBD_FREE(uuid, uuid_size);
4974         RETURN(rc);
4975 }
4976
4977 /* setup quota master target on MDT0 */
4978 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4979                           struct lustre_cfg *cfg)
4980 {
4981         struct obd_device       *obd;
4982         char                    *dev = lustre_cfg_string(cfg, 0);
4983         char                    *qmtname, *uuid, *p;
4984         struct lustre_cfg_bufs  *bufs;
4985         struct lustre_cfg       *lcfg;
4986         struct lustre_profile   *lprof;
4987         struct obd_connect_data *data;
4988         int                      rc;
4989         ENTRY;
4990
4991         LASSERT(mdt->mdt_qmt_exp == NULL);
4992         LASSERT(mdt->mdt_qmt_dev == NULL);
4993
4994         /* quota master is on MDT0 only for now */
4995         if (mdt->mdt_seq_site.ss_node_id != 0)
4996                 RETURN(0);
4997
4998         /* MGS generates config commands which look as follows:
4999          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5000          *                    3:lustre-MDT0000-mdtlov  4:f
5001          *
5002          * We generate the QMT name from the MDT one, just replacing MD with QM
5003          * after all the preparations, the logical equivalent will be:
5004          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5005          *                    3:lustre-MDT0000-osd  4:f */
5006         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5007         OBD_ALLOC(uuid, UUID_MAX);
5008         OBD_ALLOC_PTR(bufs);
5009         OBD_ALLOC_PTR(data);
5010         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5011                 GOTO(cleanup_mem, rc = -ENOMEM);
5012
5013         strcpy(qmtname, dev);
5014         p = strstr(qmtname, "-MDT");
5015         if (p == NULL)
5016                 GOTO(cleanup_mem, rc = -ENOMEM);
5017         /* replace MD with QM */
5018         p[1] = 'Q';
5019         p[2] = 'M';
5020
5021         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5022
5023         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5024         if (lprof == NULL || lprof->lp_dt == NULL) {
5025                 CERROR("can't find profile for %s\n",
5026                        lustre_cfg_string(cfg, 0));
5027                 GOTO(cleanup_mem, rc = -EINVAL);
5028         }
5029
5030         lustre_cfg_bufs_reset(bufs, qmtname);
5031         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5032         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5033         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5034
5035         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5036         if (!lcfg)
5037                 GOTO(put_profile, rc = -ENOMEM);
5038         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5039
5040         rc = class_attach(lcfg);
5041         if (rc)
5042                 GOTO(lcfg_cleanup, rc);
5043
5044         obd = class_name2obd(qmtname);
5045         if (!obd) {
5046                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5047                        lustre_cfg_string(cfg, 0));
5048                 GOTO(lcfg_cleanup, rc = -EINVAL);
5049         }
5050
5051         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5052
5053         lustre_cfg_bufs_reset(bufs, qmtname);
5054         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5055         lustre_cfg_bufs_set_string(bufs, 2, dev);
5056
5057         /* for quota, the next device should be the OSD device */
5058         lustre_cfg_bufs_set_string(bufs, 3,
5059                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5060
5061         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5062         if (!lcfg)
5063                 GOTO(class_detach, rc = -ENOMEM);
5064         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5065
5066         rc = class_setup(obd, lcfg);
5067         if (rc)
5068                 GOTO(class_detach, rc);
5069
5070         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5071
5072         /* configure local quota objects */
5073         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5074                 rc = -EBADF;
5075         else
5076                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5077                                                            &mdt->mdt_lu_dev,
5078                                                            mdt->mdt_qmt_dev);
5079         if (rc)
5080                 GOTO(class_cleanup, rc);
5081
5082         /* connect to quota master target */
5083         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5084         data->ocd_version = LUSTRE_VERSION_CODE;
5085         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5086                          data, NULL);
5087         if (rc) {
5088                 CERROR("cannot connect to quota master device %s (%d)\n",
5089                        qmtname, rc);
5090                 GOTO(class_cleanup, rc);
5091         }
5092
5093         EXIT;
5094 class_cleanup:
5095         if (rc) {
5096                 class_manual_cleanup(obd);
5097                 mdt->mdt_qmt_dev = NULL;
5098                 GOTO(lcfg_cleanup, rc);
5099         }
5100 class_detach:
5101         if (rc)
5102                 class_detach(obd, lcfg);
5103 lcfg_cleanup:
5104         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5105 put_profile:
5106         class_put_profile(lprof);
5107 cleanup_mem:
5108         if (bufs)
5109                 OBD_FREE_PTR(bufs);
5110         if (qmtname)
5111                 OBD_FREE(qmtname, MAX_OBD_NAME);
5112         if (uuid)
5113                 OBD_FREE(uuid, UUID_MAX);
5114         if (data)
5115                 OBD_FREE_PTR(data);
5116         return rc;
5117 }
5118
5119 /* Shutdown quota master target associated with mdt */
5120 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5121 {
5122         ENTRY;
5123
5124         if (mdt->mdt_qmt_exp == NULL)
5125                 RETURN_EXIT;
5126         LASSERT(mdt->mdt_qmt_dev != NULL);
5127
5128         /* the qmt automatically shuts down when the mdt disconnects */
5129         obd_disconnect(mdt->mdt_qmt_exp);
5130         mdt->mdt_qmt_exp = NULL;
5131         mdt->mdt_qmt_dev = NULL;
5132         EXIT;
5133 }
5134
5135 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5136  * for now. This will be removed along with converting rest of MDT code
5137  * to use tgt_session_info */
5138 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5139 {
5140         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5141         int                      rc;
5142
5143         if (unlikely(info->mti_object == NULL))
5144                 return -EPROTO;
5145
5146         rc = mdt_getxattr(info);
5147
5148         mdt_thread_info_fini(info);
5149         return rc;
5150 }
5151
5152 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5153 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5154 #define OST_BRW_READ    OST_READ
5155 #define OST_BRW_WRITE   OST_WRITE
5156
5157 static struct tgt_handler mdt_tgt_handlers[] = {
5158 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5159                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5160                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5161 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5162                 0,                      MDS_DISCONNECT, tgt_disconnect,
5163                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5164 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5165                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5166                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
5167 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5168 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5169 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5170 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5171                                                         mdt_getattr_name),
5172 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5173 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5174 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5175 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5176 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5177 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5178 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5179 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5180                                                         mdt_hsm_progress),
5181 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5182                                                         mdt_hsm_ct_register),
5183 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5184                                                         mdt_hsm_ct_unregister),
5185 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5186                                                         mdt_hsm_state_get),
5187 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5188                                                         mdt_hsm_state_set),
5189 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5190 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5191                                                         mdt_hsm_request),
5192 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5193             MDS_SWAP_LAYOUTS,
5194             mdt_swap_layouts),
5195 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5196 };
5197
5198 static struct tgt_handler mdt_io_ops[] = {
5199 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5200                                                         mdt_hp_brw),
5201 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5202                                                         mdt_hp_brw),
5203 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5204                                          OST_PUNCH,     mdt_punch_hdl,
5205                                                         mdt_hp_punch),
5206 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5207 };
5208
5209 static struct tgt_handler mdt_sec_ctx_ops[] = {
5210 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5211 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5212 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5213 };
5214
5215 static struct tgt_handler mdt_quota_ops[] = {
5216 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5217 };
5218
5219 static struct tgt_opc_slice mdt_common_slice[] = {
5220         {
5221                 .tos_opc_start  = MDS_FIRST_OPC,
5222                 .tos_opc_end    = MDS_LAST_OPC,
5223                 .tos_hs         = mdt_tgt_handlers
5224         },
5225         {
5226                 .tos_opc_start  = OBD_FIRST_OPC,
5227                 .tos_opc_end    = OBD_LAST_OPC,
5228                 .tos_hs         = tgt_obd_handlers
5229         },
5230         {
5231                 .tos_opc_start  = LDLM_FIRST_OPC,
5232                 .tos_opc_end    = LDLM_LAST_OPC,
5233                 .tos_hs         = tgt_dlm_handlers
5234         },
5235         {
5236                 .tos_opc_start  = SEC_FIRST_OPC,
5237                 .tos_opc_end    = SEC_LAST_OPC,
5238                 .tos_hs         = mdt_sec_ctx_ops
5239         },
5240         {
5241                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5242                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5243                 .tos_hs         = tgt_out_handlers
5244         },
5245         {
5246                 .tos_opc_start  = FLD_FIRST_OPC,
5247                 .tos_opc_end    = FLD_LAST_OPC,
5248                 .tos_hs         = fld_handlers
5249         },
5250         {
5251                 .tos_opc_start  = SEQ_FIRST_OPC,
5252                 .tos_opc_end    = SEQ_LAST_OPC,
5253                 .tos_hs         = seq_handlers
5254         },
5255         {
5256                 .tos_opc_start  = QUOTA_DQACQ,
5257                 .tos_opc_end    = QUOTA_LAST_OPC,
5258                 .tos_hs         = mdt_quota_ops
5259         },
5260         {
5261                 .tos_opc_start  = LLOG_FIRST_OPC,
5262                 .tos_opc_end    = LLOG_LAST_OPC,
5263                 .tos_hs         = tgt_llog_handlers
5264         },
5265         {
5266                 .tos_opc_start  = LFSCK_FIRST_OPC,
5267                 .tos_opc_end    = LFSCK_LAST_OPC,
5268                 .tos_hs         = tgt_lfsck_handlers
5269         },
5270         {
5271                 .tos_opc_start  = OST_FIRST_OPC,
5272                 .tos_opc_end    = OST_LAST_OPC,
5273                 .tos_hs         = mdt_io_ops
5274         },
5275         {
5276                 .tos_hs         = NULL
5277         }
5278 };
5279
5280 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5281 {
5282         struct md_device *next = m->mdt_child;
5283         struct lu_device *d = &m->mdt_lu_dev;
5284         struct obd_device *obd = mdt2obd_dev(m);
5285         struct lfsck_stop stop;
5286
5287         ENTRY;
5288         stop.ls_status = LS_PAUSED;
5289         stop.ls_flags = 0;
5290         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5291
5292         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5293         ping_evictor_stop();
5294
5295         /* Remove the HSM /proc entry so the coordinator cannot be
5296          * restarted by a user while it's shutting down.
5297          */
5298         mdt_hsm_cdt_stop(m);
5299
5300         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5301         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5302
5303         if (m->mdt_namespace != NULL)
5304                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5305                                           d->ld_obd->obd_force);
5306
5307         obd_exports_barrier(obd);
5308         obd_zombie_barrier();
5309
5310         mdt_quota_fini(env, m);
5311
5312         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5313
5314         /* Calling the cleanup functions in the same order as in the mdt_init0
5315          * error path
5316          */
5317         mdt_tunables_fini(m);
5318
5319         target_recovery_fini(obd);
5320         upcall_cache_cleanup(m->mdt_identity_cache);
5321         m->mdt_identity_cache = NULL;
5322
5323         mdt_fs_cleanup(env, m);
5324
5325         tgt_fini(env, &m->mdt_lut);
5326
5327         mdt_hsm_cdt_fini(m);
5328
5329         if (m->mdt_los != NULL) {
5330                 local_oid_storage_fini(env, m->mdt_los);
5331                 m->mdt_los = NULL;
5332         }
5333
5334         if (m->mdt_namespace != NULL) {
5335                 ldlm_namespace_free_post(m->mdt_namespace);
5336                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5337         }
5338
5339         if (m->mdt_md_root != NULL) {
5340                 mdt_object_put(env, m->mdt_md_root);
5341                 m->mdt_md_root = NULL;
5342         }
5343
5344         mdt_seq_fini(env, m);
5345
5346         mdt_fld_fini(env, m);
5347
5348         /*
5349          * Finish the stack
5350          */
5351         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5352
5353         LASSERT(atomic_read(&d->ld_ref) == 0);
5354
5355         server_put_mount(mdt_obd_name(m), true);
5356
5357         EXIT;
5358 }
5359
5360 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5361
5362 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5363                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5364 {
5365         const struct dt_device_param *dt_conf;
5366         struct mdt_thread_info *info;
5367         struct obd_device *obd;
5368         const char *dev = lustre_cfg_string(cfg, 0);
5369         const char *num = lustre_cfg_string(cfg, 2);
5370         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5371         struct lustre_mount_info *lmi = NULL;
5372         struct lustre_sb_info *lsi;
5373         struct lu_site *s;
5374         struct seq_server_site *ss_site;
5375         const char *identity_upcall = "NONE";
5376         struct md_device *next;
5377         struct lu_fid fid;
5378         int rc;
5379         long node_id;
5380         mntopt_t mntopts;
5381         ENTRY;
5382
5383         lu_device_init(&m->mdt_lu_dev, ldt);
5384         /*
5385          * Environment (env) might be missing mdt_thread_key values at that
5386          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5387          * mode.
5388          *
5389          * Usually device allocation path doesn't use module key values, but
5390          * mdt has to do a lot of work here, so allocate key value.
5391          */
5392         rc = lu_env_refill((struct lu_env *)env);
5393         if (rc != 0)
5394                 RETURN(rc);
5395
5396         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5397         LASSERT(info != NULL);
5398
5399         obd = class_name2obd(dev);
5400         LASSERT(obd != NULL);
5401
5402         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5403         m->mdt_opts.mo_evict_tgt_nids = 1;
5404         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5405
5406         lmi = server_get_mount(dev);
5407         if (lmi == NULL) {
5408                 CERROR("Cannot get mount info for %s!\n", dev);
5409                 RETURN(-EFAULT);
5410         } else {
5411                 lsi = s2lsi(lmi->lmi_sb);
5412                 LASSERT(lsi->lsi_lmd);
5413                 /* CMD is supported only in IAM mode */
5414                 LASSERT(num);
5415                 node_id = simple_strtol(num, NULL, 10);
5416                 obd->u.obt.obt_magic = OBT_MAGIC;
5417                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5418                         m->mdt_skip_lfsck = 1;
5419         }
5420
5421         /* DoM files get IO lock at open by default */
5422         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5423         /* DoM files are read at open and data is packed in the reply */
5424         m->mdt_opts.mo_dom_read_open = 1;
5425
5426         m->mdt_squash.rsi_uid = 0;
5427         m->mdt_squash.rsi_gid = 0;
5428         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5429         spin_lock_init(&m->mdt_squash.rsi_lock);
5430         spin_lock_init(&m->mdt_lock);
5431         m->mdt_enable_remote_dir = 1;
5432         m->mdt_enable_striped_dir = 1;
5433         m->mdt_enable_dir_migration = 1;
5434         m->mdt_enable_remote_dir_gid = 0;
5435         m->mdt_enable_chprojid_gid = 0;
5436         m->mdt_enable_remote_rename = 1;
5437
5438         atomic_set(&m->mdt_mds_mds_conns, 0);
5439         atomic_set(&m->mdt_async_commit_count, 0);
5440
5441         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5442         m->mdt_lu_dev.ld_obd = obd;
5443         /* Set this lu_device to obd for error handling purposes. */
5444         obd->obd_lu_dev = &m->mdt_lu_dev;
5445
5446         /* init the stack */
5447         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5448         if (rc) {
5449                 CERROR("%s: Can't init device stack, rc %d\n",
5450                        mdt_obd_name(m), rc);
5451                 GOTO(err_lmi, rc);
5452         }
5453
5454         s = mdt_lu_site(m);
5455         ss_site = mdt_seq_site(m);
5456         s->ld_seq_site = ss_site;
5457         ss_site->ss_lu = s;
5458
5459         /* set server index */
5460         ss_site->ss_node_id = node_id;
5461
5462         /* failover is the default
5463          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5464          * assumed whose ss_node_id == 0 XXX
5465          * */
5466         obd->obd_replayable = 1;
5467         /* No connection accepted until configurations will finish */
5468         obd->obd_no_conn = 1;
5469
5470         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5471                 char *str = lustre_cfg_string(cfg, 4);
5472                 if (strchr(str, 'n')) {
5473                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5474                         obd->obd_replayable = 0;
5475                 }
5476         }
5477
5478         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5479         if (rc)
5480                 GOTO(err_fini_stack, rc);
5481
5482         rc = mdt_seq_init(env, m);
5483         if (rc)
5484                 GOTO(err_fini_fld, rc);
5485
5486         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5487                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5488         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5489                                               LDLM_NAMESPACE_SERVER,
5490                                               LDLM_NAMESPACE_GREEDY,
5491                                               LDLM_NS_TYPE_MDT);
5492         if (m->mdt_namespace == NULL)
5493                 GOTO(err_fini_seq, rc = -ENOMEM);
5494
5495         m->mdt_namespace->ns_lvbp = m;
5496         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5497
5498         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5499         /* set obd_namespace for compatibility with old code */
5500         obd->obd_namespace = m->mdt_namespace;
5501
5502         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5503                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5504                       OBD_FAIL_MDS_ALL_REPLY_NET);
5505         if (rc)
5506                 GOTO(err_free_ns, rc);
5507
5508         /* Amount of available space excluded from granting and reserved
5509          * for metadata. It is in percentage and 50% is default value. */
5510         tgd->tgd_reserved_pcnt = 50;
5511
5512         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5513                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5514         else
5515                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5516
5517         rc = mdt_fs_setup(env, m, obd, lsi);
5518         if (rc)
5519                 GOTO(err_tgt, rc);
5520
5521         fid.f_seq = FID_SEQ_LOCAL_NAME;
5522         fid.f_oid = 1;
5523         fid.f_ver = 0;
5524         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5525         if (rc != 0)
5526                 GOTO(err_fs_cleanup, rc);
5527
5528         rc = mdt_hsm_cdt_init(m);
5529         if (rc != 0) {
5530                 CERROR("%s: error initializing coordinator, rc %d\n",
5531                        mdt_obd_name(m), rc);
5532                 GOTO(err_los_fini, rc);
5533         }
5534
5535         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5536
5537         next = m->mdt_child;
5538         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5539
5540         mntopts = dt_conf->ddp_mntopts;
5541
5542         if (mntopts & MNTOPT_USERXATTR)
5543                 m->mdt_opts.mo_user_xattr = 1;
5544         else
5545                 m->mdt_opts.mo_user_xattr = 0;
5546
5547         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5548
5549         if (mntopts & MNTOPT_ACL)
5550                 m->mdt_opts.mo_acl = 1;
5551         else
5552                 m->mdt_opts.mo_acl = 0;
5553
5554         /* XXX: to support suppgid for ACL, we enable identity_upcall
5555          * by default, otherwise, maybe got unexpected -EACCESS. */
5556         if (m->mdt_opts.mo_acl)
5557                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5558
5559         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5560                                                 identity_upcall,
5561                                                 &mdt_identity_upcall_cache_ops);
5562         if (IS_ERR(m->mdt_identity_cache)) {
5563                 rc = PTR_ERR(m->mdt_identity_cache);
5564                 m->mdt_identity_cache = NULL;
5565                 GOTO(err_free_hsm, rc);
5566         }
5567
5568         rc = mdt_tunables_init(m, dev);
5569         if (rc) {
5570                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5571                 GOTO(err_recovery, rc);
5572         }
5573
5574         rc = mdt_quota_init(env, m, cfg);
5575         if (rc)
5576                 GOTO(err_procfs, rc);
5577
5578         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5579         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5580                            "mdt_ldlm_client", m->mdt_ldlm_client);
5581
5582         ping_evictor_start();
5583
5584         /* recovery will be started upon mdt_prepare()
5585          * when the whole stack is complete and ready
5586          * to serve the requests */
5587
5588         /* Reduce the initial timeout on an MDS because it doesn't need such
5589          * a long timeout as an OST does. Adaptive timeouts will adjust this
5590          * value appropriately. */
5591         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5592                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5593
5594         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
5595                 m->mdt_lut.lut_local_recovery = 1;
5596
5597         RETURN(0);
5598 err_procfs:
5599         mdt_tunables_fini(m);
5600 err_recovery:
5601         upcall_cache_cleanup(m->mdt_identity_cache);
5602         m->mdt_identity_cache = NULL;
5603 err_free_hsm:
5604         mdt_hsm_cdt_fini(m);
5605 err_los_fini:
5606         local_oid_storage_fini(env, m->mdt_los);
5607         m->mdt_los = NULL;
5608 err_fs_cleanup:
5609         mdt_fs_cleanup(env, m);
5610 err_tgt:
5611         /* keep recoverable clients */
5612         obd->obd_fail = 1;
5613         target_recovery_fini(obd);
5614         obd_exports_barrier(obd);
5615         obd_zombie_barrier();
5616         tgt_fini(env, &m->mdt_lut);
5617 err_free_ns:
5618         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5619         obd->obd_namespace = m->mdt_namespace = NULL;
5620 err_fini_seq:
5621         mdt_seq_fini(env, m);
5622 err_fini_fld:
5623         mdt_fld_fini(env, m);
5624 err_fini_stack:
5625         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5626 err_lmi:
5627         if (lmi)
5628                 server_put_mount(dev, true);
5629         return(rc);
5630 }
5631
5632 /* For interoperability, the left element is old parameter, the right one
5633  * is the new version of the parameter, if some parameter is deprecated,
5634  * the new version should be set as NULL. */
5635 static struct cfg_interop_param mdt_interop_param[] = {
5636         { "mdt.group_upcall",   NULL },
5637         { "mdt.quota_type",     NULL },
5638         { "mdd.quota_type",     NULL },
5639         { "mdt.som",            NULL },
5640         { "mdt.rootsquash",     "mdt.root_squash" },
5641         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5642         { NULL }
5643 };
5644
5645 /* used by MGS to process specific configurations */
5646 static int mdt_process_config(const struct lu_env *env,
5647                               struct lu_device *d, struct lustre_cfg *cfg)
5648 {
5649         struct mdt_device *m = mdt_dev(d);
5650         struct md_device *md_next = m->mdt_child;
5651         struct lu_device *next = md2lu_dev(md_next);
5652         int rc;
5653         ENTRY;
5654
5655         switch (cfg->lcfg_command) {
5656         case LCFG_PARAM: {
5657                 struct obd_device *obd = d->ld_obd;
5658                 /* For interoperability */
5659                 struct cfg_interop_param *ptr = NULL;
5660                 struct lustre_cfg *old_cfg = NULL;
5661                 char *param = NULL;
5662                 ssize_t count;
5663
5664                 param = lustre_cfg_string(cfg, 1);
5665                 if (param == NULL) {
5666                         CERROR("param is empty\n");
5667                         rc = -EINVAL;
5668                         break;
5669                 }
5670
5671                 ptr = class_find_old_param(param, mdt_interop_param);
5672                 if (ptr != NULL) {
5673                         if (ptr->new_param == NULL) {
5674                                 rc = 0;
5675                                 CWARN("For interoperability, skip this %s."
5676                                       " It is obsolete.\n", ptr->old_param);
5677                                 break;
5678                         }
5679
5680                         CWARN("Found old param %s, changed it to %s.\n",
5681                               ptr->old_param, ptr->new_param);
5682
5683                         old_cfg = cfg;
5684                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5685                         if (IS_ERR(cfg)) {
5686                                 rc = PTR_ERR(cfg);
5687                                 break;
5688                         }
5689                 }
5690
5691                 count = class_modify_config(cfg, PARAM_MDT,
5692                                             &obd->obd_kset.kobj);
5693                 if (count < 0) {
5694                         struct coordinator *cdt = &m->mdt_coordinator;
5695
5696                         /* is it an HSM var ? */
5697                         count = class_modify_config(cfg, PARAM_HSM,
5698                                                     &cdt->cdt_hsm_kobj);
5699                         if (count < 0)
5700                                 /* we don't understand; pass it on */
5701                                 rc = next->ld_ops->ldo_process_config(env, next,
5702                                                                       cfg);
5703                         else
5704                                 rc = count > 0 ? 0 : count;
5705                 } else {
5706                         rc = count > 0 ? 0 : count;
5707                 }
5708
5709                 if (old_cfg)
5710                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5711                                                      cfg->lcfg_buflens));
5712                 break;
5713         }
5714         default:
5715                 /* others are passed further */
5716                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5717                 break;
5718         }
5719         RETURN(rc);
5720 }
5721
5722 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5723                                           const struct lu_object_header *hdr,
5724                                           struct lu_device *d)
5725 {
5726         struct mdt_object *mo;
5727
5728         ENTRY;
5729
5730         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5731         if (mo != NULL) {
5732                 struct lu_object *o;
5733                 struct lu_object_header *h;
5734
5735                 o = &mo->mot_obj;
5736                 h = &mo->mot_header;
5737                 lu_object_header_init(h);
5738                 lu_object_init(o, h, d);
5739                 lu_object_add_top(h, o);
5740                 o->lo_ops = &mdt_obj_ops;
5741                 spin_lock_init(&mo->mot_write_lock);
5742                 mutex_init(&mo->mot_som_mutex);
5743                 mutex_init(&mo->mot_lov_mutex);
5744                 init_rwsem(&mo->mot_dom_sem);
5745                 init_rwsem(&mo->mot_open_sem);
5746                 atomic_set(&mo->mot_open_count, 0);
5747                 RETURN(o);
5748         }
5749         RETURN(NULL);
5750 }
5751
5752 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5753                            const struct lu_object_conf *unused)
5754 {
5755         struct mdt_device *d = mdt_dev(o->lo_dev);
5756         struct lu_device  *under;
5757         struct lu_object  *below;
5758         int                rc = 0;
5759         ENTRY;
5760
5761         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5762                PFID(lu_object_fid(o)));
5763
5764         under = &d->mdt_child->md_lu_dev;
5765         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5766         if (below != NULL) {
5767                 lu_object_add(o, below);
5768         } else
5769                 rc = -ENOMEM;
5770
5771         RETURN(rc);
5772 }
5773
5774 static void mdt_object_free_rcu(struct rcu_head *head)
5775 {
5776         struct mdt_object *mo = container_of(head, struct mdt_object,
5777                                              mot_header.loh_rcu);
5778
5779         kmem_cache_free(mdt_object_kmem, mo);
5780 }
5781
5782 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5783 {
5784         struct mdt_object *mo = mdt_obj(o);
5785         struct lu_object_header *h;
5786         ENTRY;
5787
5788         h = o->lo_header;
5789         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5790                PFID(lu_object_fid(o)));
5791
5792         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5793         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5794
5795         lu_object_fini(o);
5796         lu_object_header_fini(h);
5797         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
5798         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
5799
5800         EXIT;
5801 }
5802
5803 static int mdt_object_print(const struct lu_env *env, void *cookie,
5804                             lu_printer_t p, const struct lu_object *o)
5805 {
5806         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5807
5808         return (*p)(env, cookie,
5809                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5810                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5811                     mdto->mot_cache_attr ? "cache_attr" : "",
5812                     mdto->mot_write_count);
5813 }
5814
5815 static int mdt_prepare(const struct lu_env *env,
5816                 struct lu_device *pdev,
5817                 struct lu_device *cdev)
5818 {
5819         struct mdt_device *mdt = mdt_dev(cdev);
5820         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5821         struct obd_device *obd = cdev->ld_obd;
5822         int rc;
5823
5824         ENTRY;
5825
5826         LASSERT(obd);
5827
5828         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5829         if (rc)
5830                 RETURN(rc);
5831
5832         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5833         if (rc)
5834                 RETURN(rc);
5835
5836         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5837         if (rc)
5838                 RETURN(rc);
5839
5840         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5841         /* The LFSCK instance is registered just now, so it must be there when
5842          * register the namespace to such instance. */
5843         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5844
5845         if (mdt->mdt_seq_site.ss_node_id == 0) {
5846                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5847                                                          &mdt->mdt_md_root_fid);
5848                 if (rc)
5849                         RETURN(rc);
5850         }
5851
5852         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5853
5854         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5855         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5856         LASSERT(obd->obd_no_conn);
5857         spin_lock(&obd->obd_dev_lock);
5858         obd->obd_no_conn = 0;
5859         spin_unlock(&obd->obd_dev_lock);
5860
5861         if (obd->obd_recovering == 0)
5862                 mdt_postrecov(env, mdt);
5863
5864         RETURN(rc);
5865 }
5866
5867 const struct lu_device_operations mdt_lu_ops = {
5868         .ldo_object_alloc   = mdt_object_alloc,
5869         .ldo_process_config = mdt_process_config,
5870         .ldo_prepare        = mdt_prepare,
5871 };
5872
5873 static const struct lu_object_operations mdt_obj_ops = {
5874         .loo_object_init    = mdt_object_init,
5875         .loo_object_free    = mdt_object_free,
5876         .loo_object_print   = mdt_object_print
5877 };
5878
5879 static int mdt_obd_set_info_async(const struct lu_env *env,
5880                                   struct obd_export *exp,
5881                                   __u32 keylen, void *key,
5882                                   __u32 vallen, void *val,
5883                                   struct ptlrpc_request_set *set)
5884 {
5885         int rc;
5886
5887         ENTRY;
5888
5889         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5890                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5891                 RETURN(rc);
5892         }
5893
5894         RETURN(0);
5895 }
5896
5897 /**
5898  * Match client and server connection feature flags.
5899  *
5900  * Compute the compatibility flags for a connection request based on
5901  * features mutually supported by client and server.
5902  *
5903  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5904  * must not be updated here, otherwise a partially initialized value may
5905  * be exposed. After the connection request is successfully processed,
5906  * the top-level MDT connect request handler atomically updates the export
5907  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5908  * reply. \see mdt_connect().
5909  *
5910  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5911  * full struct obd_connect_data. So care must be taken when accessing fields
5912  * that are not present in struct obd_connect_data_v1. See LU-16.
5913  *
5914  * \param exp   the obd_export associated with this client/target pair
5915  * \param mdt   the target device for the connection
5916  * \param data  stores data for this connect request
5917  *
5918  * \retval 0       success
5919  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5920  * \retval -EBADE  client and server feature requirements are incompatible
5921  */
5922 static int mdt_connect_internal(const struct lu_env *env,
5923                                 struct obd_export *exp,
5924                                 struct mdt_device *mdt,
5925                                 struct obd_connect_data *data, bool reconnect)
5926 {
5927         const char *obd_name = mdt_obd_name(mdt);
5928         LASSERT(data != NULL);
5929
5930         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5931
5932         if (mdt->mdt_bottom->dd_rdonly &&
5933             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5934             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5935                 RETURN(-EACCES);
5936
5937         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5938                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5939
5940         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5941
5942         if (!mdt->mdt_opts.mo_acl)
5943                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5944
5945         if (!mdt->mdt_opts.mo_user_xattr)
5946                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5947
5948         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5949                 data->ocd_brw_size = min(data->ocd_brw_size,
5950                                          mdt->mdt_brw_size);
5951                 if (data->ocd_brw_size == 0) {
5952                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5953                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5954                                "ocd_brw_size unexpectedly zero, network data "
5955                                "corruption? Refusing to connect this client\n",
5956                                obd_name, exp->exp_client_uuid.uuid,
5957                                exp, data->ocd_connect_flags, data->ocd_version,
5958                                data->ocd_grant, data->ocd_index);
5959                         return -EPROTO;
5960                 }
5961         }
5962
5963         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5964                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5965
5966                 /* client is reporting its page size, for future use */
5967                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5968                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5969                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5970                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5971                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5972                 /* ocd_grant_tax_kb is in 1K byte blocks */
5973                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5974                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5975         }
5976
5977         /* Save connect_data we have so far because tgt_grant_connect()
5978          * uses it to calculate grant, and we want to save the client
5979          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5980         exp->exp_connect_data = *data;
5981         if (OCD_HAS_FLAG(data, GRANT))
5982                 tgt_grant_connect(env, exp, data, !reconnect);
5983
5984         if (OCD_HAS_FLAG(data, MAXBYTES))
5985                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5986
5987         /* NB: Disregard the rule against updating
5988          * exp_connect_data.ocd_connect_flags in this case, since
5989          * tgt_client_new() needs to know if this is a lightweight
5990          * connection, and it is safe to expose this flag before
5991          * connection processing completes. */
5992         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5993                 spin_lock(&exp->exp_lock);
5994                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5995                 spin_unlock(&exp->exp_lock);
5996         }
5997
5998         data->ocd_version = LUSTRE_VERSION_CODE;
5999
6000         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6001                 CWARN("%s: MDS requires FID support, but client not\n",
6002                       obd_name);
6003                 return -EBADE;
6004         }
6005
6006         if (OCD_HAS_FLAG(data, PINGLESS)) {
6007                 if (ptlrpc_pinger_suppress_pings()) {
6008                         spin_lock(&exp->exp_obd->obd_dev_lock);
6009                         list_del_init(&exp->exp_obd_chain_timed);
6010                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6011                 } else {
6012                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6013                 }
6014         }
6015
6016         data->ocd_max_easize = mdt->mdt_max_ea_size;
6017
6018         /* NB: Disregard the rule against updating
6019          * exp_connect_data.ocd_connect_flags in this case, since
6020          * tgt_client_new() needs to know if this is client supports
6021          * multiple modify RPCs, and it is safe to expose this flag before
6022          * connection processing completes. */
6023         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6024                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6025                 spin_lock(&exp->exp_lock);
6026                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6027                 spin_unlock(&exp->exp_lock);
6028         }
6029
6030         if (OCD_HAS_FLAG(data, CKSUM)) {
6031                 __u32 cksum_types = data->ocd_cksum_types;
6032
6033                 /* The client set in ocd_cksum_types the checksum types it
6034                  * supports. We have to mask off the algorithms that we don't
6035                  * support */
6036                 data->ocd_cksum_types &=
6037                         obd_cksum_types_supported_server(obd_name);
6038
6039                 if (unlikely(data->ocd_cksum_types == 0)) {
6040                         CERROR("%s: Connect with checksum support but no "
6041                                "ocd_cksum_types is set\n",
6042                                exp->exp_obd->obd_name);
6043                         RETURN(-EPROTO);
6044                 }
6045
6046                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6047                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6048                        cksum_types, data->ocd_cksum_types);
6049         } else {
6050                 /* This client does not support OBD_CONNECT_CKSUM
6051                  * fall back to CRC32 */
6052                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6053                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6054                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6055         }
6056
6057         return 0;
6058 }
6059
6060 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6061                                    struct mdt_thread_info *info,
6062                                    struct mdt_file_data *mfd)
6063 {
6064         struct lu_context ses;
6065         int rc;
6066         ENTRY;
6067
6068         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6069         if (rc)
6070                 RETURN(rc);
6071
6072         env->le_ses = &ses;
6073         lu_context_enter(&ses);
6074
6075         mdt_ucred(info)->uc_valid = UCRED_OLD;
6076         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6077
6078         lu_context_exit(&ses);
6079         lu_context_fini(&ses);
6080         env->le_ses = NULL;
6081
6082         RETURN(rc);
6083 }
6084
6085 static int mdt_export_cleanup(struct obd_export *exp)
6086 {
6087         LIST_HEAD(closing_list);
6088         struct mdt_export_data  *med = &exp->exp_mdt_data;
6089         struct obd_device       *obd = exp->exp_obd;
6090         struct mdt_device       *mdt;
6091         struct mdt_thread_info  *info;
6092         struct lu_env            env;
6093         struct mdt_file_data    *mfd, *n;
6094         int rc = 0;
6095         ENTRY;
6096
6097         spin_lock(&med->med_open_lock);
6098         while (!list_empty(&med->med_open_head)) {
6099                 struct list_head *tmp = med->med_open_head.next;
6100                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6101
6102                 /* Remove mfd handle so it can't be found again.
6103                  * We are consuming the mfd_list reference here. */
6104                 class_handle_unhash(&mfd->mfd_open_handle);
6105                 list_move_tail(&mfd->mfd_list, &closing_list);
6106         }
6107         spin_unlock(&med->med_open_lock);
6108         mdt = mdt_dev(obd->obd_lu_dev);
6109         LASSERT(mdt != NULL);
6110
6111         rc = lu_env_init(&env, LCT_MD_THREAD);
6112         if (rc)
6113                 RETURN(rc);
6114
6115         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6116         LASSERT(info != NULL);
6117         memset(info, 0, sizeof *info);
6118         info->mti_env = &env;
6119         info->mti_mdt = mdt;
6120         info->mti_exp = exp;
6121
6122         if (!list_empty(&closing_list)) {
6123                 struct md_attr *ma = &info->mti_attr;
6124
6125                 /* Close any open files (which may also cause orphan
6126                  * unlinking). */
6127                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6128                         list_del_init(&mfd->mfd_list);
6129                         ma->ma_need = ma->ma_valid = 0;
6130
6131                         /* This file is being closed due to an eviction, it
6132                          * could have been modified and now dirty regarding to
6133                          * HSM archive, check this!
6134                          * The logic here is to mark a file dirty if there's a
6135                          * chance it was dirtied before the client was evicted,
6136                          * so that we don't have to wait for a release attempt
6137                          * before finding out the file was actually dirty and
6138                          * fail the release. Aggressively marking it dirty here
6139                          * will cause the policy engine to attempt to
6140                          * re-archive it; when rearchiving, we can compare the
6141                          * current version to the HSM data_version and make the
6142                          * archive request into a noop if it's not actually
6143                          * dirty.
6144                          */
6145                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6146                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6147
6148                         /* Don't unlink orphan on failover umount, LU-184 */
6149                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6150                                 ma->ma_valid = MA_FLAGS;
6151                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6152                         }
6153                         ma->ma_valid |= MA_FORCE_LOG;
6154                         mdt_mfd_close(info, mfd);
6155                 }
6156         }
6157         info->mti_mdt = NULL;
6158         /* cleanup client slot early */
6159         /* Do not erase record for recoverable client. */
6160         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6161                 tgt_client_del(&env, exp);
6162         lu_env_fini(&env);
6163
6164         RETURN(rc);
6165 }
6166
6167 static inline void mdt_enable_slc(struct mdt_device *mdt)
6168 {
6169         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6170                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6171 }
6172
6173 static inline void mdt_disable_slc(struct mdt_device *mdt)
6174 {
6175         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6176                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6177 }
6178
6179 static int mdt_obd_disconnect(struct obd_export *exp)
6180 {
6181         int rc;
6182
6183         ENTRY;
6184
6185         LASSERT(exp);
6186         class_export_get(exp);
6187
6188         if (!(exp->exp_flags & OBD_OPT_FORCE))
6189                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6190
6191         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6192             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6193                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6194
6195                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6196                         mdt_disable_slc(mdt);
6197         }
6198
6199         rc = server_disconnect_export(exp);
6200         if (rc != 0)
6201                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6202
6203         tgt_grant_discard(exp);
6204
6205         rc = mdt_export_cleanup(exp);
6206         nodemap_del_member(exp);
6207         class_export_put(exp);
6208         RETURN(rc);
6209 }
6210
6211 /* mds_connect copy */
6212 static int mdt_obd_connect(const struct lu_env *env,
6213                            struct obd_export **exp, struct obd_device *obd,
6214                            struct obd_uuid *cluuid,
6215                            struct obd_connect_data *data,
6216                            void *localdata)
6217 {
6218         struct obd_export       *lexp;
6219         struct lustre_handle    conn = { 0 };
6220         struct mdt_device       *mdt;
6221         int                      rc;
6222         lnet_nid_t              *client_nid = localdata;
6223         ENTRY;
6224
6225         LASSERT(env != NULL);
6226         LASSERT(data != NULL);
6227
6228         if (!exp || !obd || !cluuid)
6229                 RETURN(-EINVAL);
6230
6231         mdt = mdt_dev(obd->obd_lu_dev);
6232
6233         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6234             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6235                 atomic_inc(&mdt->mdt_mds_mds_conns);
6236                 mdt_enable_slc(mdt);
6237         }
6238
6239         /*
6240          * first, check whether the stack is ready to handle requests
6241          * XXX: probably not very appropriate method is used now
6242          *      at some point we should find a better one
6243          */
6244         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6245             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6246             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6247                 rc = obd_get_info(env, mdt->mdt_child_exp,
6248                                   sizeof(KEY_OSP_CONNECTED),
6249                                   KEY_OSP_CONNECTED, NULL, NULL);
6250                 if (rc)
6251                         RETURN(-EAGAIN);
6252                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6253         }
6254
6255         rc = class_connect(&conn, obd, cluuid);
6256         if (rc)
6257                 RETURN(rc);
6258
6259         lexp = class_conn2export(&conn);
6260         LASSERT(lexp != NULL);
6261
6262         rc = nodemap_add_member(*client_nid, lexp);
6263         if (rc != 0 && rc != -EEXIST)
6264                 GOTO(out, rc);
6265
6266         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6267         if (rc == 0) {
6268                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6269
6270                 LASSERT(lcd);
6271                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6272                 rc = tgt_client_new(env, lexp);
6273                 if (rc == 0)
6274                         mdt_export_stats_init(obd, lexp, localdata);
6275         }
6276 out:
6277         if (rc != 0) {
6278                 class_disconnect(lexp);
6279                 nodemap_del_member(lexp);
6280                 *exp = NULL;
6281         } else {
6282                 *exp = lexp;
6283                 /* Because we do not want this export to be evicted by pinger,
6284                  * let's not add this export to the timed chain list. */
6285                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6286                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6287                         list_del_init(&lexp->exp_obd_chain_timed);
6288                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6289                 }
6290         }
6291
6292         RETURN(rc);
6293 }
6294
6295 static int mdt_obd_reconnect(const struct lu_env *env,
6296                              struct obd_export *exp, struct obd_device *obd,
6297                              struct obd_uuid *cluuid,
6298                              struct obd_connect_data *data,
6299                              void *localdata)
6300 {
6301         lnet_nid_t             *client_nid = localdata;
6302         int                     rc;
6303         ENTRY;
6304
6305         if (exp == NULL || obd == NULL || cluuid == NULL)
6306                 RETURN(-EINVAL);
6307
6308         rc = nodemap_add_member(*client_nid, exp);
6309         if (rc != 0 && rc != -EEXIST)
6310                 RETURN(rc);
6311
6312         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6313                                   true);
6314         if (rc == 0)
6315                 mdt_export_stats_init(obd, exp, localdata);
6316         else
6317                 nodemap_del_member(exp);
6318
6319         RETURN(rc);
6320 }
6321
6322 /* FIXME: Can we avoid using these two interfaces? */
6323 static int mdt_init_export(struct obd_export *exp)
6324 {
6325         struct mdt_export_data *med = &exp->exp_mdt_data;
6326         int                     rc;
6327         ENTRY;
6328
6329         INIT_LIST_HEAD(&med->med_open_head);
6330         spin_lock_init(&med->med_open_lock);
6331         spin_lock(&exp->exp_lock);
6332         exp->exp_connecting = 1;
6333         spin_unlock(&exp->exp_lock);
6334
6335         OBD_ALLOC(exp->exp_used_slots,
6336                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6337         if (exp->exp_used_slots == NULL)
6338                 RETURN(-ENOMEM);
6339
6340         /* self-export doesn't need client data and ldlm initialization */
6341         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6342                                      &exp->exp_client_uuid)))
6343                 RETURN(0);
6344
6345         rc = tgt_client_alloc(exp);
6346         if (rc)
6347                 GOTO(err, rc);
6348
6349         rc = ldlm_init_export(exp);
6350         if (rc)
6351                 GOTO(err_free, rc);
6352
6353         RETURN(rc);
6354
6355 err_free:
6356         tgt_client_free(exp);
6357 err:
6358         OBD_FREE(exp->exp_used_slots,
6359                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6360         exp->exp_used_slots = NULL;
6361
6362         CERROR("%s: Failed to initialize export: rc = %d\n",
6363                exp->exp_obd->obd_name, rc);
6364         return rc;
6365 }
6366
6367 static int mdt_destroy_export(struct obd_export *exp)
6368 {
6369         ENTRY;
6370
6371         target_destroy_export(exp);
6372         if (exp->exp_used_slots)
6373                 OBD_FREE(exp->exp_used_slots,
6374                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6375
6376         /* destroy can be called from failed obd_setup, so
6377          * checking uuid is safer than obd_self_export */
6378         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6379                                      &exp->exp_client_uuid)))
6380                 RETURN(0);
6381
6382         ldlm_destroy_export(exp);
6383         tgt_client_free(exp);
6384
6385         LASSERT(list_empty(&exp->exp_outstanding_replies));
6386         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6387
6388         /*
6389          * discard grants once we're sure no more
6390          * interaction with the client is possible
6391          */
6392         tgt_grant_discard(exp);
6393         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6394                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6395
6396         if (!(exp->exp_flags & OBD_OPT_FORCE))
6397                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6398
6399         RETURN(0);
6400 }
6401
6402 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6403                    struct linkea_data *ldata)
6404 {
6405         int rc;
6406
6407         LASSERT(ldata->ld_buf->lb_buf != NULL);
6408
6409         if (!mdt_object_exists(mdt_obj))
6410                 return -ENODATA;
6411
6412         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6413                           ldata->ld_buf, XATTR_NAME_LINK);
6414         if (rc == -ERANGE) {
6415                 /* Buf was too small, figure out what we need. */
6416                 lu_buf_free(ldata->ld_buf);
6417                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6418                                   ldata->ld_buf, XATTR_NAME_LINK);
6419                 if (rc < 0)
6420                         return rc;
6421                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6422                 if (ldata->ld_buf->lb_buf == NULL)
6423                         return -ENOMEM;
6424                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6425                                   ldata->ld_buf, XATTR_NAME_LINK);
6426         }
6427         if (rc < 0)
6428                 return rc;
6429
6430         return linkea_init_with_rec(ldata);
6431 }
6432
6433 /**
6434  * Given an MDT object, try to look up the full path to the object.
6435  * Part of the MDT layer implementation of lfs fid2path.
6436  *
6437  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6438  * \param[in]     obj   Object to do path lookup of
6439  * \param[in,out] fp    User-provided struct to store path information
6440  * \param[in]     root_fid Root FID of current path should reach
6441  *
6442  * \retval 0 Lookup successful, path information stored in fp
6443  * \retval -EAGAIN Lookup failed, usually because object is being moved
6444  * \retval negative errno if there was a problem
6445  */
6446 static int mdt_path_current(struct mdt_thread_info *info,
6447                             struct mdt_object *obj,
6448                             struct getinfo_fid2path *fp,
6449                             struct lu_fid *root_fid)
6450 {
6451         struct mdt_device       *mdt = info->mti_mdt;
6452         struct mdt_object       *mdt_obj;
6453         struct link_ea_header   *leh;
6454         struct link_ea_entry    *lee;
6455         struct lu_name          *tmpname = &info->mti_name;
6456         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6457         struct lu_buf           *buf = &info->mti_big_buf;
6458         char                    *ptr;
6459         int                     reclen;
6460         struct linkea_data      ldata = { NULL };
6461         int                     rc = 0;
6462         bool                    first = true;
6463         ENTRY;
6464
6465         /* temp buffer for path element, the buffer will be finally freed
6466          * in mdt_thread_info_fini */
6467         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6468         if (buf->lb_buf == NULL)
6469                 RETURN(-ENOMEM);
6470
6471         ldata.ld_buf = buf;
6472         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6473         *ptr = 0;
6474         --ptr;
6475         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6476
6477         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6478                 struct lu_buf           lmv_buf;
6479
6480                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6481                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6482                         GOTO(out, rc = -ENOENT);
6483
6484                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6485                         mdt_obj = obj;
6486                         mdt_object_get(info->mti_env, mdt_obj);
6487                 } else {
6488                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6489                         if (IS_ERR(mdt_obj))
6490                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6491                 }
6492
6493                 if (!mdt_object_exists(mdt_obj)) {
6494                         mdt_object_put(info->mti_env, mdt_obj);
6495                         GOTO(out, rc = -ENOENT);
6496                 }
6497
6498                 if (mdt_object_remote(mdt_obj)) {
6499                         mdt_object_put(info->mti_env, mdt_obj);
6500                         GOTO(remote_out, rc = -EREMOTE);
6501                 }
6502
6503                 rc = mdt_links_read(info, mdt_obj, &ldata);
6504                 if (rc != 0) {
6505                         mdt_object_put(info->mti_env, mdt_obj);
6506                         GOTO(out, rc);
6507                 }
6508
6509                 leh = buf->lb_buf;
6510                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6511                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6512                 /* If set, use link #linkno for path lookup, otherwise use
6513                    link #0.  Only do this for the final path element. */
6514                 if (first && fp->gf_linkno < leh->leh_reccount) {
6515                         int count;
6516                         for (count = 0; count < fp->gf_linkno; count++) {
6517                                 lee = (struct link_ea_entry *)
6518                                      ((char *)lee + reclen);
6519                                 linkea_entry_unpack(lee, &reclen, tmpname,
6520                                                     tmpfid);
6521                         }
6522                         if (fp->gf_linkno < leh->leh_reccount - 1)
6523                                 /* indicate to user there are more links */
6524                                 fp->gf_linkno++;
6525                 }
6526
6527                 lmv_buf.lb_buf = info->mti_xattr_buf;
6528                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6529                 /* Check if it is slave stripes */
6530                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6531                                   &lmv_buf, XATTR_NAME_LMV);
6532                 mdt_object_put(info->mti_env, mdt_obj);
6533                 if (rc > 0) {
6534                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6535
6536                         /* For slave stripes, get its master */
6537                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6538                                 fp->gf_fid = *tmpfid;
6539                                 continue;
6540                         }
6541                 } else if (rc < 0 && rc != -ENODATA) {
6542                         GOTO(out, rc);
6543                 }
6544
6545                 rc = 0;
6546
6547                 /* Pack the name in the end of the buffer */
6548                 ptr -= tmpname->ln_namelen;
6549                 if (ptr - 1 <= fp->gf_u.gf_path)
6550                         GOTO(out, rc = -EOVERFLOW);
6551                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6552                 *(--ptr) = '/';
6553
6554                 /* keep the last resolved fid to the client, so the
6555                  * client will build the left path on another MDT for
6556                  * remote object */
6557                 fp->gf_fid = *tmpfid;
6558
6559                 first = false;
6560         }
6561
6562 remote_out:
6563         ptr++; /* skip leading / */
6564         memmove(fp->gf_u.gf_path, ptr,
6565                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6566
6567 out:
6568         RETURN(rc);
6569 }
6570
6571 /**
6572  * Given an MDT object, use mdt_path_current to get the path.
6573  * Essentially a wrapper to retry mdt_path_current a set number of times
6574  * if -EAGAIN is returned (usually because an object is being moved).
6575  *
6576  * Part of the MDT layer implementation of lfs fid2path.
6577  *
6578  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6579  * \param[in]     obj   Object to do path lookup of
6580  * \param[in,out] fp    User-provided struct for arguments and to store path
6581  *                      information
6582  *
6583  * \retval 0 Lookup successful, path information stored in fp
6584  * \retval negative errno if there was a problem
6585  */
6586 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6587                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6588 {
6589         struct mdt_device       *mdt = info->mti_mdt;
6590         int                     tries = 3;
6591         int                     rc = -EAGAIN;
6592         ENTRY;
6593
6594         if (fp->gf_pathlen < 3)
6595                 RETURN(-EOVERFLOW);
6596
6597         if (root_fid == NULL)
6598                 root_fid = &mdt->mdt_md_root_fid;
6599
6600         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6601                 fp->gf_u.gf_path[0] = '\0';
6602                 RETURN(0);
6603         }
6604
6605         /* Retry multiple times in case file is being moved */
6606         while (tries-- && rc == -EAGAIN)
6607                 rc = mdt_path_current(info, obj, fp, root_fid);
6608
6609         RETURN(rc);
6610 }
6611
6612 /**
6613  * Get the full path of the provided FID, as of changelog record recno.
6614  *
6615  * This checks sanity and looks up object for user provided FID
6616  * before calling the actual path lookup code.
6617  *
6618  * Part of the MDT layer implementation of lfs fid2path.
6619  *
6620  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6621  * \param[in,out] fp    User-provided struct for arguments and to store path
6622  *                      information
6623  *
6624  * \retval 0 Lookup successful, path information and recno stored in fp
6625  * \retval -ENOENT, object does not exist
6626  * \retval negative errno if there was a problem
6627  */
6628 static int mdt_fid2path(struct mdt_thread_info *info,
6629                         struct lu_fid *root_fid,
6630                         struct getinfo_fid2path *fp)
6631 {
6632         struct mdt_device *mdt = info->mti_mdt;
6633         struct mdt_object *obj;
6634         int    rc;
6635         ENTRY;
6636
6637         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6638                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6639
6640         if (!fid_is_sane(&fp->gf_fid))
6641                 RETURN(-EINVAL);
6642
6643         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6644                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6645                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6646                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6647                 RETURN(-EINVAL);
6648         }
6649
6650         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6651         if (IS_ERR(obj)) {
6652                 rc = PTR_ERR(obj);
6653                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6654                        PFID(&fp->gf_fid), rc);
6655                 RETURN(rc);
6656         }
6657
6658         if (mdt_object_remote(obj))
6659                 rc = -EREMOTE;
6660         else if (!mdt_object_exists(obj))
6661                 rc = -ENOENT;
6662         else
6663                 rc = 0;
6664
6665         if (rc < 0) {
6666                 mdt_object_put(info->mti_env, obj);
6667                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6668                        PFID(&fp->gf_fid), rc);
6669                 RETURN(rc);
6670         }
6671
6672         rc = mdt_path(info, obj, fp, root_fid);
6673
6674         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6675                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6676                fp->gf_recno, fp->gf_linkno);
6677
6678         mdt_object_put(info->mti_env, obj);
6679
6680         RETURN(rc);
6681 }
6682
6683 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6684                             void *val, int vallen)
6685 {
6686         struct getinfo_fid2path *fpout, *fpin;
6687         struct lu_fid *root_fid = NULL;
6688         int rc = 0;
6689
6690         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6691         fpout = val;
6692
6693         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6694                 lustre_swab_fid2path(fpin);
6695
6696         memcpy(fpout, fpin, sizeof(*fpin));
6697         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6698                 RETURN(-EINVAL);
6699
6700         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6701                       sizeof(struct lu_fid)) {
6702                 /* client sent its root FID, which is normally fileset FID */
6703                 root_fid = fpin->gf_u.gf_root_fid;
6704                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6705                         lustre_swab_lu_fid(root_fid);
6706
6707                 if (root_fid != NULL && !fid_is_sane(root_fid))
6708                         RETURN(-EINVAL);
6709         }
6710
6711         rc = mdt_fid2path(info, root_fid, fpout);
6712         RETURN(rc);
6713 }
6714
6715 int mdt_get_info(struct tgt_session_info *tsi)
6716 {
6717         char    *key;
6718         int      keylen;
6719         __u32   *vallen;
6720         void    *valout;
6721         int      rc;
6722
6723         ENTRY;
6724
6725         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6726         if (key == NULL) {
6727                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6728                 RETURN(err_serious(-EFAULT));
6729         }
6730         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6731                                       RCL_CLIENT);
6732
6733         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6734         if (vallen == NULL) {
6735                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6736                                 tgt_name(tsi->tsi_tgt));
6737                 RETURN(err_serious(-EFAULT));
6738         }
6739
6740         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6741                              *vallen);
6742         rc = req_capsule_server_pack(tsi->tsi_pill);
6743         if (rc)
6744                 RETURN(err_serious(rc));
6745
6746         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6747         if (valout == NULL) {
6748                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6749                                 tgt_name(tsi->tsi_tgt));
6750                 RETURN(err_serious(-EFAULT));
6751         }
6752
6753         if (KEY_IS(KEY_FID2PATH)) {
6754                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6755
6756                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6757                 mdt_thread_info_fini(info);
6758         } else {
6759                 rc = -EINVAL;
6760         }
6761         RETURN(rc);
6762 }
6763
6764 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6765 {
6766         struct obd_ioctl_data *data = karg;
6767         struct lu_fid *fid;
6768         __u64 version;
6769         struct mdt_object *obj;
6770         struct mdt_lock_handle  *lh;
6771         int rc;
6772         ENTRY;
6773
6774         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6775             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6776                 RETURN(-EINVAL);
6777
6778         fid = (struct lu_fid *)data->ioc_inlbuf1;
6779
6780         if (!fid_is_sane(fid))
6781                 RETURN(-EINVAL);
6782
6783         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6784
6785         lh = &mti->mti_lh[MDT_LH_PARENT];
6786         mdt_lock_reg_init(lh, LCK_CR);
6787
6788         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6789         if (IS_ERR(obj))
6790                 RETURN(PTR_ERR(obj));
6791
6792         if (mdt_object_remote(obj)) {
6793                 rc = -EREMOTE;
6794                 /**
6795                  * before calling version get the correct MDS should be
6796                  * fid, this is error to find remote object here
6797                  */
6798                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6799         } else if (!mdt_object_exists(obj)) {
6800                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6801                 rc = -ENOENT;
6802         } else {
6803                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6804                *(__u64 *)data->ioc_inlbuf2 = version;
6805                 rc = 0;
6806         }
6807         mdt_object_unlock_put(mti, obj, lh, 1);
6808         RETURN(rc);
6809 }
6810
6811 /* ioctls on obd dev */
6812 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6813                          void *karg, void __user *uarg)
6814 {
6815         struct lu_env      env;
6816         struct obd_device *obd = exp->exp_obd;
6817         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6818         struct dt_device  *dt = mdt->mdt_bottom;
6819         int rc;
6820
6821         ENTRY;
6822         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6823         rc = lu_env_init(&env, LCT_MD_THREAD);
6824         if (rc)
6825                 RETURN(rc);
6826
6827         switch (cmd) {
6828         case OBD_IOC_SYNC:
6829                 rc = mdt_device_sync(&env, mdt);
6830                 break;
6831         case OBD_IOC_SET_READONLY:
6832                 rc = dt_sync(&env, dt);
6833                 if (rc == 0)
6834                         rc = dt_ro(&env, dt);
6835                 break;
6836         case OBD_IOC_ABORT_RECOVERY:
6837                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6838                 obd->obd_abort_recovery = 1;
6839                 target_stop_recovery_thread(obd);
6840                 rc = 0;
6841                 break;
6842         case OBD_IOC_CHANGELOG_REG:
6843         case OBD_IOC_CHANGELOG_DEREG:
6844         case OBD_IOC_CHANGELOG_CLEAR:
6845                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6846                                                            mdt->mdt_child,
6847                                                            cmd, len, karg);
6848                 break;
6849         case OBD_IOC_START_LFSCK: {
6850                 struct md_device *next = mdt->mdt_child;
6851                 struct obd_ioctl_data *data = karg;
6852                 struct lfsck_start_param lsp;
6853
6854                 if (unlikely(data == NULL)) {
6855                         rc = -EINVAL;
6856                         break;
6857                 }
6858
6859                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6860                 lsp.lsp_index_valid = 0;
6861                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6862                 break;
6863         }
6864         case OBD_IOC_STOP_LFSCK: {
6865                 struct md_device        *next = mdt->mdt_child;
6866                 struct obd_ioctl_data   *data = karg;
6867                 struct lfsck_stop        stop;
6868
6869                 stop.ls_status = LS_STOPPED;
6870                 /* Old lfsck utils may pass NULL @stop. */
6871                 if (data->ioc_inlbuf1 == NULL)
6872                         stop.ls_flags = 0;
6873                 else
6874                         stop.ls_flags =
6875                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6876
6877                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6878                 break;
6879         }
6880         case OBD_IOC_QUERY_LFSCK: {
6881                 struct md_device        *next = mdt->mdt_child;
6882                 struct obd_ioctl_data   *data = karg;
6883
6884                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6885                                                  data->ioc_inlbuf1);
6886                 break;
6887         }
6888         case OBD_IOC_GET_OBJ_VERSION: {
6889                 struct mdt_thread_info *mti;
6890                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6891                 memset(mti, 0, sizeof *mti);
6892                 mti->mti_env = &env;
6893                 mti->mti_mdt = mdt;
6894                 mti->mti_exp = exp;
6895
6896                 rc = mdt_ioc_version_get(mti, karg);
6897                 break;
6898         }
6899         case OBD_IOC_CATLOGLIST: {
6900                 struct mdt_thread_info *mti;
6901
6902                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6903                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6904                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6905                                        &mti->mti_tmp_fid1);
6906                 break;
6907          }
6908         default:
6909                 rc = -EOPNOTSUPP;
6910                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6911                         mdt_obd_name(mdt), cmd, rc);
6912         }
6913
6914         lu_env_fini(&env);
6915         RETURN(rc);
6916 }
6917
6918 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6919 {
6920         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6921         int rc;
6922         ENTRY;
6923
6924         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6925                 struct lfsck_start_param lsp;
6926
6927                 lsp.lsp_start = NULL;
6928                 lsp.lsp_index_valid = 0;
6929                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6930                                                            OBD_IOC_START_LFSCK,
6931                                                            0, &lsp);
6932                 if (rc != 0 && rc != -EALREADY)
6933                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6934                               mdt_obd_name(mdt), rc);
6935         }
6936
6937         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6938         RETURN(rc);
6939 }
6940
6941 static int mdt_obd_postrecov(struct obd_device *obd)
6942 {
6943         struct lu_env env;
6944         int rc;
6945
6946         rc = lu_env_init(&env, LCT_MD_THREAD);
6947         if (rc)
6948                 RETURN(rc);
6949         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6950         lu_env_fini(&env);
6951         return rc;
6952 }
6953
6954 static const struct obd_ops mdt_obd_device_ops = {
6955         .o_owner          = THIS_MODULE,
6956         .o_set_info_async = mdt_obd_set_info_async,
6957         .o_connect        = mdt_obd_connect,
6958         .o_reconnect      = mdt_obd_reconnect,
6959         .o_disconnect     = mdt_obd_disconnect,
6960         .o_init_export    = mdt_init_export,
6961         .o_destroy_export = mdt_destroy_export,
6962         .o_iocontrol      = mdt_iocontrol,
6963         .o_postrecov      = mdt_obd_postrecov,
6964         /* Data-on-MDT IO methods */
6965         .o_preprw         = mdt_obd_preprw,
6966         .o_commitrw       = mdt_obd_commitrw,
6967 };
6968
6969 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6970                                          struct lu_device *d)
6971 {
6972         struct mdt_device *m = mdt_dev(d);
6973         ENTRY;
6974
6975         mdt_fini(env, m);
6976         RETURN(NULL);
6977 }
6978
6979 static struct lu_device *mdt_device_free(const struct lu_env *env,
6980                                          struct lu_device *d)
6981 {
6982         struct mdt_device *m = mdt_dev(d);
6983         ENTRY;
6984
6985         lu_device_fini(&m->mdt_lu_dev);
6986         OBD_FREE_PTR(m);
6987
6988         RETURN(NULL);
6989 }
6990
6991 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6992                                           struct lu_device_type *t,
6993                                           struct lustre_cfg *cfg)
6994 {
6995         struct lu_device  *l;
6996         struct mdt_device *m;
6997
6998         OBD_ALLOC_PTR(m);
6999         if (m != NULL) {
7000                 int rc;
7001
7002                 l = &m->mdt_lu_dev;
7003                 rc = mdt_init0(env, m, t, cfg);
7004                 if (rc != 0) {
7005                         mdt_device_free(env, l);
7006                         l = ERR_PTR(rc);
7007                         return l;
7008                 }
7009         } else
7010                 l = ERR_PTR(-ENOMEM);
7011         return l;
7012 }
7013
7014 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7015 LU_KEY_INIT(mdt, struct mdt_thread_info);
7016
7017 static void mdt_key_fini(const struct lu_context *ctx,
7018                          struct lu_context_key *key, void* data)
7019 {
7020         struct mdt_thread_info *info = data;
7021
7022         if (info->mti_big_lmm) {
7023                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7024                 info->mti_big_lmm = NULL;
7025                 info->mti_big_lmmsize = 0;
7026         }
7027
7028         if (info->mti_big_acl) {
7029                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7030                 info->mti_big_acl = NULL;
7031                 info->mti_big_aclsize = 0;
7032         }
7033
7034         OBD_FREE_PTR(info);
7035 }
7036
7037 /* context key: mdt_thread_key */
7038 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7039
7040 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7041 {
7042         return lu_ucred(info->mti_env);
7043 }
7044
7045 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7046 {
7047         return lu_ucred_check(info->mti_env);
7048 }
7049
7050 /**
7051  * Enable/disable COS (Commit On Sharing).
7052  *
7053  * Set/Clear the COS flag in mdt options.
7054  *
7055  * \param mdt mdt device
7056  * \param val 0 disables COS, other values enable COS
7057  */
7058 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7059 {
7060         struct lu_env env;
7061         int rc;
7062
7063         mdt->mdt_opts.mo_cos = val;
7064         rc = lu_env_init(&env, LCT_LOCAL);
7065         if (unlikely(rc != 0)) {
7066                 CWARN("%s: lu_env initialization failed, cannot "
7067                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7068                 return;
7069         }
7070         mdt_device_sync(&env, mdt);
7071         lu_env_fini(&env);
7072 }
7073
7074 /**
7075  * Check COS (Commit On Sharing) status.
7076  *
7077  * Return COS flag status.
7078  *
7079  * \param mdt mdt device
7080  */
7081 int mdt_cos_is_enabled(struct mdt_device *mdt)
7082 {
7083         return mdt->mdt_opts.mo_cos != 0;
7084 }
7085
7086 static struct lu_device_type_operations mdt_device_type_ops = {
7087         .ldto_device_alloc = mdt_device_alloc,
7088         .ldto_device_free  = mdt_device_free,
7089         .ldto_device_fini  = mdt_device_fini
7090 };
7091
7092 static struct lu_device_type mdt_device_type = {
7093         .ldt_tags     = LU_DEVICE_MD,
7094         .ldt_name     = LUSTRE_MDT_NAME,
7095         .ldt_ops      = &mdt_device_type_ops,
7096         .ldt_ctx_tags = LCT_MD_THREAD
7097 };
7098
7099 static int __init mdt_init(void)
7100 {
7101         int rc;
7102
7103         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7104                      FID_NOBRACE_LEN + 1);
7105         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7106                      FID_LEN + 1);
7107         rc = lu_kmem_init(mdt_caches);
7108         if (rc)
7109                 return rc;
7110
7111         rc = mds_mod_init();
7112         if (rc)
7113                 GOTO(lu_fini, rc);
7114
7115         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
7116                                  LUSTRE_MDT_NAME, &mdt_device_type);
7117         if (rc)
7118                 GOTO(mds_fini, rc);
7119 lu_fini:
7120         if (rc)
7121                 lu_kmem_fini(mdt_caches);
7122 mds_fini:
7123         if (rc)
7124                 mds_mod_exit();
7125         return rc;
7126 }
7127
7128 static void __exit mdt_exit(void)
7129 {
7130         class_unregister_type(LUSTRE_MDT_NAME);
7131         mds_mod_exit();
7132         lu_kmem_fini(mdt_caches);
7133 }
7134
7135 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7136 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7137 MODULE_VERSION(LUSTRE_VERSION_STRING);
7138 MODULE_LICENSE("GPL");
7139
7140 module_init(mdt_init);
7141 module_exit(mdt_exit);