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                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2277                         RETURN(-EACCES);
2278                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2279                                    vallen, val, NULL);
2280         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2281                 if (vallen > 0)
2282                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2283         } else {
2284                 RETURN(-EINVAL);
2285         }
2286         RETURN(rc);
2287 }
2288
2289 static int mdt_readpage(struct tgt_session_info *tsi)
2290 {
2291         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2292         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2293         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2294         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2295         struct mdt_body         *repbody;
2296         int                      rc;
2297         int                      i;
2298
2299         ENTRY;
2300
2301         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2302                 RETURN(err_serious(-ENOMEM));
2303
2304         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2305         if (repbody == NULL || reqbody == NULL)
2306                 RETURN(err_serious(-EFAULT));
2307
2308         /*
2309          * prepare @rdpg before calling lower layers and transfer itself. Here
2310          * reqbody->size contains offset of where to start to read and
2311          * reqbody->nlink contains number bytes to read.
2312          */
2313         rdpg->rp_hash = reqbody->mbo_size;
2314         if (rdpg->rp_hash != reqbody->mbo_size) {
2315                 CERROR("Invalid hash: %#llx != %#llx\n",
2316                        rdpg->rp_hash, reqbody->mbo_size);
2317                 RETURN(-EFAULT);
2318         }
2319
2320         rdpg->rp_attrs = reqbody->mbo_mode;
2321         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2322                 rdpg->rp_attrs |= LUDA_64BITHASH;
2323         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2324                                 exp_max_brw_size(tsi->tsi_exp));
2325         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2326                           PAGE_SHIFT;
2327         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2328         if (rdpg->rp_pages == NULL)
2329                 RETURN(-ENOMEM);
2330
2331         for (i = 0; i < rdpg->rp_npages; ++i) {
2332                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2333                 if (rdpg->rp_pages[i] == NULL)
2334                         GOTO(free_rdpg, rc = -ENOMEM);
2335         }
2336
2337         /* call lower layers to fill allocated pages with directory data */
2338         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2339         if (rc < 0)
2340                 GOTO(free_rdpg, rc);
2341
2342         /* send pages to client */
2343         rc = tgt_sendpage(tsi, rdpg, rc);
2344
2345         EXIT;
2346 free_rdpg:
2347
2348         for (i = 0; i < rdpg->rp_npages; i++)
2349                 if (rdpg->rp_pages[i] != NULL)
2350                         __free_page(rdpg->rp_pages[i]);
2351         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2352
2353         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2354                 RETURN(0);
2355
2356         return rc;
2357 }
2358
2359 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2360 {
2361         struct lu_ucred *uc = mdt_ucred_check(info);
2362         struct lu_attr *attr = &info->mti_attr.ma_attr;
2363
2364         if (uc == NULL)
2365                 return -EINVAL;
2366
2367         if (op != REINT_SETATTR) {
2368                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2369                         attr->la_uid = uc->uc_fsuid;
2370                 /* for S_ISGID, inherit gid from his parent, such work will be
2371                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2372                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2373                         attr->la_gid = uc->uc_fsgid;
2374         }
2375
2376         return 0;
2377 }
2378
2379 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2380 {
2381         return op == REINT_OPEN &&
2382              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2383 }
2384
2385 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2386 {
2387         struct req_capsule *pill = info->mti_pill;
2388
2389         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2390                                   RCL_SERVER) &&
2391             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2392                                   RCL_CLIENT)) {
2393                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2394                                          RCL_CLIENT) != 0) {
2395                         /* pre-set size in server part with max size */
2396                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2397                                              RCL_SERVER,
2398                                              info->mti_mdt->mdt_max_ea_size);
2399                 } else {
2400                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2401                                              RCL_SERVER, 0);
2402                 }
2403         }
2404
2405 }
2406
2407 static int mdt_reint_internal(struct mdt_thread_info *info,
2408                               struct mdt_lock_handle *lhc,
2409                               __u32 op)
2410 {
2411         struct req_capsule      *pill = info->mti_pill;
2412         struct mdt_body         *repbody;
2413         int                      rc = 0, rc2;
2414
2415         ENTRY;
2416
2417         rc = mdt_reint_unpack(info, op);
2418         if (rc != 0) {
2419                 CERROR("Can't unpack reint, rc %d\n", rc);
2420                 RETURN(err_serious(rc));
2421         }
2422
2423
2424         /* check if the file system is set to readonly. O_RDONLY open
2425          * is still allowed even the file system is set to readonly mode */
2426         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2427                 RETURN(err_serious(-EROFS));
2428
2429         /* for replay (no_create) lmm is not needed, client has it already */
2430         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2431                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2432                                      DEF_REP_MD_SIZE);
2433
2434         /* llog cookies are always 0, the field is kept for compatibility */
2435         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2436                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2437
2438         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2439          * by default. If the target object has more ACL entries, then
2440          * enlarge the buffer when necessary. */
2441         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2442                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2443                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2444
2445         mdt_preset_secctx_size(info);
2446
2447         rc = req_capsule_server_pack(pill);
2448         if (rc != 0) {
2449                 CERROR("Can't pack response, rc %d\n", rc);
2450                 RETURN(err_serious(rc));
2451         }
2452
2453         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2454                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2455                 LASSERT(repbody);
2456                 repbody->mbo_eadatasize = 0;
2457                 repbody->mbo_aclsize = 0;
2458         }
2459
2460         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2461
2462         /* for replay no cookkie / lmm need, because client have this already */
2463         if (info->mti_spec.no_create)
2464                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2465                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2466
2467         rc = mdt_init_ucred_reint(info);
2468         if (rc)
2469                 GOTO(out_shrink, rc);
2470
2471         rc = mdt_fix_attr_ucred(info, op);
2472         if (rc != 0)
2473                 GOTO(out_ucred, rc = err_serious(rc));
2474
2475         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2476         if (rc < 0) {
2477                 GOTO(out_ucred, rc);
2478         } else if (rc == 1) {
2479                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2480                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2481                 GOTO(out_ucred, rc);
2482         }
2483         rc = mdt_reint_rec(info, lhc);
2484         EXIT;
2485 out_ucred:
2486         mdt_exit_ucred(info);
2487 out_shrink:
2488         mdt_client_compatibility(info);
2489
2490         rc2 = mdt_fix_reply(info);
2491         if (rc == 0)
2492                 rc = rc2;
2493
2494         /*
2495          * Data-on-MDT optimization - read data along with OPEN and return it
2496          * in reply. Do that only if we have both DOM and LAYOUT locks.
2497          */
2498         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2499             info->mti_attr.ma_lmm != NULL &&
2500             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2501                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2502                                           &lhc->mlh_reg_lh);
2503         }
2504
2505         return rc;
2506 }
2507
2508 static long mdt_reint_opcode(struct ptlrpc_request *req,
2509                              const struct req_format **fmt)
2510 {
2511         struct mdt_device       *mdt;
2512         struct mdt_rec_reint    *rec;
2513         long                     opc;
2514
2515         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2516         if (rec != NULL) {
2517                 opc = rec->rr_opcode;
2518                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2519                 if (opc < REINT_MAX && fmt[opc] != NULL)
2520                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2521                 else {
2522                         mdt = mdt_exp2dev(req->rq_export);
2523                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2524                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2525                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2526                         opc = err_serious(-EFAULT);
2527                 }
2528         } else {
2529                 opc = err_serious(-EFAULT);
2530         }
2531         return opc;
2532 }
2533
2534 static int mdt_reint(struct tgt_session_info *tsi)
2535 {
2536         long opc;
2537         int  rc;
2538         static const struct req_format *reint_fmts[REINT_MAX] = {
2539                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2540                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2541                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2542                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2543                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2544                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2545                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2546                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2547                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2548                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2549         };
2550
2551         ENTRY;
2552
2553         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2554         if (opc >= 0) {
2555                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2556                 /*
2557                  * No lock possible here from client to pass it to reint code
2558                  * path.
2559                  */
2560                 rc = mdt_reint_internal(info, NULL, opc);
2561                 mdt_thread_info_fini(info);
2562         } else {
2563                 rc = opc;
2564         }
2565
2566         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2567         RETURN(rc);
2568 }
2569
2570 /* this should sync the whole device */
2571 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2572 {
2573         struct dt_device *dt = mdt->mdt_bottom;
2574         int rc;
2575         ENTRY;
2576
2577         rc = dt->dd_ops->dt_sync(env, dt);
2578         RETURN(rc);
2579 }
2580
2581 /* this should sync this object */
2582 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2583                            struct mdt_object *mo)
2584 {
2585         int rc = 0;
2586
2587         ENTRY;
2588
2589         if (!mdt_object_exists(mo)) {
2590                 CWARN("%s: non existing object "DFID": rc = %d\n",
2591                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2592                       -ESTALE);
2593                 RETURN(-ESTALE);
2594         }
2595
2596         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2597                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2598                 dt_obj_version_t version;
2599
2600                 version = dt_version_get(env, mdt_obj2dt(mo));
2601                 if (version > tgt->lut_obd->obd_last_committed)
2602                         rc = mo_object_sync(env, mdt_object_child(mo));
2603         } else {
2604                 rc = mo_object_sync(env, mdt_object_child(mo));
2605         }
2606
2607         RETURN(rc);
2608 }
2609
2610 static int mdt_sync(struct tgt_session_info *tsi)
2611 {
2612         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2613         struct req_capsule      *pill = tsi->tsi_pill;
2614         struct mdt_body         *body;
2615         int                      rc;
2616
2617         ENTRY;
2618
2619         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2620                 RETURN(err_serious(-ENOMEM));
2621
2622         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2623                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2624         } else {
2625                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2626
2627                 if (unlikely(info->mti_object == NULL))
2628                         RETURN(-EPROTO);
2629
2630                 /* sync an object */
2631                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2632                                      info->mti_object);
2633                 if (rc == 0) {
2634                         const struct lu_fid *fid;
2635                         struct lu_attr *la = &info->mti_attr.ma_attr;
2636
2637                         info->mti_attr.ma_need = MA_INODE;
2638                         info->mti_attr.ma_valid = 0;
2639                         rc = mdt_attr_get_complex(info, info->mti_object,
2640                                                   &info->mti_attr);
2641                         if (rc == 0) {
2642                                 body = req_capsule_server_get(pill,
2643                                                               &RMF_MDT_BODY);
2644                                 fid = mdt_object_fid(info->mti_object);
2645                                 mdt_pack_attr2body(info, body, la, fid);
2646                         }
2647                 }
2648                 mdt_thread_info_fini(info);
2649         }
2650         if (rc == 0)
2651                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2652
2653         RETURN(rc);
2654 }
2655
2656 static int mdt_data_sync(struct tgt_session_info *tsi)
2657 {
2658         struct mdt_thread_info *info;
2659         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2660         struct ost_body *body = tsi->tsi_ost_body;
2661         struct ost_body *repbody;
2662         struct mdt_object *mo = NULL;
2663         struct md_attr *ma;
2664         int rc = 0;
2665
2666         ENTRY;
2667
2668         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2669
2670         /* if no fid is specified then do nothing,
2671          * device sync is done via MDS_SYNC */
2672         if (fid_is_zero(&tsi->tsi_fid))
2673                 RETURN(0);
2674
2675         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2676         if (IS_ERR(mo))
2677                 RETURN(PTR_ERR(mo));
2678
2679         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2680         if (rc)
2681                 GOTO(put, rc);
2682
2683         repbody->oa.o_oi = body->oa.o_oi;
2684         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2685
2686         info = tsi2mdt_info(tsi);
2687         ma = &info->mti_attr;
2688         ma->ma_need = MA_INODE;
2689         ma->ma_valid = 0;
2690         rc = mdt_attr_get_complex(info, mo, ma);
2691         if (rc == 0)
2692                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2693         else
2694                 rc = 0;
2695         mdt_thread_info_fini(info);
2696
2697         EXIT;
2698 put:
2699         if (mo != NULL)
2700                 mdt_object_put(tsi->tsi_env, mo);
2701         return rc;
2702 }
2703
2704 /*
2705  * Handle quota control requests to consult current usage/limit, but also
2706  * to configure quota enforcement
2707  */
2708 static int mdt_quotactl(struct tgt_session_info *tsi)
2709 {
2710         struct obd_export *exp  = tsi->tsi_exp;
2711         struct req_capsule *pill = tsi->tsi_pill;
2712         struct obd_quotactl *oqctl, *repoqc;
2713         int id, rc;
2714         struct mdt_device *mdt = mdt_exp2dev(exp);
2715         struct lu_device *qmt = mdt->mdt_qmt_dev;
2716         struct lu_nodemap *nodemap;
2717         ENTRY;
2718
2719         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2720         if (!oqctl)
2721                 RETURN(err_serious(-EPROTO));
2722
2723         rc = req_capsule_server_pack(pill);
2724         if (rc)
2725                 RETURN(err_serious(rc));
2726
2727         nodemap = nodemap_get_from_exp(exp);
2728         if (IS_ERR(nodemap))
2729                 RETURN(PTR_ERR(nodemap));
2730
2731         switch (oqctl->qc_cmd) {
2732                 /* master quotactl */
2733         case Q_SETINFO:
2734         case Q_SETQUOTA:
2735         case LUSTRE_Q_SETDEFAULT:
2736         case LUSTRE_Q_SETQUOTAPOOL:
2737         case LUSTRE_Q_SETINFOPOOL:
2738                 if (!nodemap_can_setquota(nodemap))
2739                         GOTO(out_nodemap, rc = -EPERM);
2740                 /* fallthrough */
2741         case Q_GETINFO:
2742         case Q_GETQUOTA:
2743         case LUSTRE_Q_GETDEFAULT:
2744         case LUSTRE_Q_GETQUOTAPOOL:
2745         case LUSTRE_Q_GETINFOPOOL:
2746                 if (qmt == NULL)
2747                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2748                 /* slave quotactl */
2749                 /* fallthrough */
2750         case Q_GETOINFO:
2751         case Q_GETOQUOTA:
2752                 break;
2753         default:
2754                 rc = -EFAULT;
2755                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
2756                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
2757                 GOTO(out_nodemap, rc);
2758         }
2759
2760         id = oqctl->qc_id;
2761         switch (oqctl->qc_type) {
2762         case USRQUOTA:
2763                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2764                                     NODEMAP_CLIENT_TO_FS, id);
2765                 break;
2766         case GRPQUOTA:
2767                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2768                                     NODEMAP_CLIENT_TO_FS, id);
2769                 break;
2770         case PRJQUOTA:
2771                 /* todo: check/map project id */
2772                 id = oqctl->qc_id;
2773                 break;
2774         default:
2775                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2776         }
2777         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2778         if (repoqc == NULL)
2779                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2780
2781         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2782                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2783
2784         if (oqctl->qc_id != id)
2785                 swap(oqctl->qc_id, id);
2786
2787         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2788                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2789                         RETURN(-EINPROGRESS);
2790         }
2791
2792         switch (oqctl->qc_cmd) {
2793
2794         case Q_GETINFO:
2795         case Q_SETINFO:
2796         case Q_SETQUOTA:
2797         case Q_GETQUOTA:
2798         case LUSTRE_Q_SETDEFAULT:
2799         case LUSTRE_Q_GETDEFAULT:
2800         case LUSTRE_Q_SETQUOTAPOOL:
2801         case LUSTRE_Q_GETQUOTAPOOL:
2802         case LUSTRE_Q_SETINFOPOOL:
2803         case LUSTRE_Q_GETINFOPOOL:
2804                 /* forward quotactl request to QMT */
2805                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2806                 break;
2807
2808         case Q_GETOINFO:
2809         case Q_GETOQUOTA:
2810                 /* slave quotactl */
2811                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2812                                    oqctl);
2813                 break;
2814
2815         default:
2816                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2817                 GOTO(out_nodemap, rc = -EFAULT);
2818         }
2819
2820         if (oqctl->qc_id != id)
2821                 swap(oqctl->qc_id, id);
2822
2823         QCTL_COPY(repoqc, oqctl);
2824         EXIT;
2825
2826 out_nodemap:
2827         nodemap_putref(nodemap);
2828
2829         return rc;
2830 }
2831
2832 /** clone llog ctxt from child (mdd)
2833  * This allows remote llog (replicator) access.
2834  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2835  * context was originally set up, or we can handle them directly.
2836  * I choose the latter, but that means I need any llog
2837  * contexts set up by child to be accessable by the mdt.  So we clone the
2838  * context into our context list here.
2839  */
2840 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2841                                int idx)
2842 {
2843         struct md_device  *next = mdt->mdt_child;
2844         struct llog_ctxt *ctxt;
2845         int rc;
2846
2847         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2848                 return 0;
2849
2850         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2851         if (rc || ctxt == NULL) {
2852                 return 0;
2853         }
2854
2855         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2856         if (rc)
2857                 CERROR("Can't set mdt ctxt %d\n", rc);
2858
2859         return rc;
2860 }
2861
2862 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2863                                  struct mdt_device *mdt, int idx)
2864 {
2865         struct llog_ctxt *ctxt;
2866
2867         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2868         if (ctxt == NULL)
2869                 return 0;
2870         /* Put once for the get we just did, and once for the clone */
2871         llog_ctxt_put(ctxt);
2872         llog_ctxt_put(ctxt);
2873         return 0;
2874 }
2875
2876 /*
2877  * sec context handlers
2878  */
2879 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2880 {
2881         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2882
2883         return 0;
2884 }
2885
2886 /*
2887  * quota request handlers
2888  */
2889 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2890 {
2891         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2892         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2893         int                      rc;
2894         ENTRY;
2895
2896         if (qmt == NULL)
2897                 RETURN(err_serious(-EOPNOTSUPP));
2898
2899         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2900         RETURN(rc);
2901 }
2902
2903 struct mdt_object *mdt_object_new(const struct lu_env *env,
2904                                   struct mdt_device *d,
2905                                   const struct lu_fid *f)
2906 {
2907         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2908         struct lu_object *o;
2909         struct mdt_object *m;
2910         ENTRY;
2911
2912         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2913         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2914         if (unlikely(IS_ERR(o)))
2915                 m = (struct mdt_object *)o;
2916         else
2917                 m = mdt_obj(o);
2918         RETURN(m);
2919 }
2920
2921 struct mdt_object *mdt_object_find(const struct lu_env *env,
2922                                    struct mdt_device *d,
2923                                    const struct lu_fid *f)
2924 {
2925         struct lu_object *o;
2926         struct mdt_object *m;
2927         ENTRY;
2928
2929         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2930         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2931         if (unlikely(IS_ERR(o)))
2932                 m = (struct mdt_object *)o;
2933         else
2934                 m = mdt_obj(o);
2935
2936         RETURN(m);
2937 }
2938
2939 /**
2940  * Asyncronous commit for mdt device.
2941  *
2942  * Pass asynchonous commit call down the MDS stack.
2943  *
2944  * \param env environment
2945  * \param mdt the mdt device
2946  */
2947 static void mdt_device_commit_async(const struct lu_env *env,
2948                                     struct mdt_device *mdt)
2949 {
2950         struct dt_device *dt = mdt->mdt_bottom;
2951         int rc;
2952         ENTRY;
2953
2954         rc = dt->dd_ops->dt_commit_async(env, dt);
2955         if (unlikely(rc != 0))
2956                 CWARN("%s: async commit start failed: rc = %d\n",
2957                       mdt_obd_name(mdt), rc);
2958         atomic_inc(&mdt->mdt_async_commit_count);
2959         EXIT;
2960 }
2961
2962 /**
2963  * Mark the lock as "synchonous".
2964  *
2965  * Mark the lock to deffer transaction commit to the unlock time.
2966  *
2967  * \param lock the lock to mark as "synchonous"
2968  *
2969  * \see mdt_is_lock_sync
2970  * \see mdt_save_lock
2971  */
2972 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2973 {
2974         lock->l_ast_data = (void*)1;
2975 }
2976
2977 /**
2978  * Check whehter the lock "synchonous" or not.
2979  *
2980  * \param lock the lock to check
2981  * \retval 1 the lock is "synchonous"
2982  * \retval 0 the lock isn't "synchronous"
2983  *
2984  * \see mdt_set_lock_sync
2985  * \see mdt_save_lock
2986  */
2987 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2988 {
2989         return lock->l_ast_data != NULL;
2990 }
2991
2992 /**
2993  * Blocking AST for mdt locks.
2994  *
2995  * Starts transaction commit if in case of COS lock conflict or
2996  * deffers such a commit to the mdt_save_lock.
2997  *
2998  * \param lock the lock which blocks a request or cancelling lock
2999  * \param desc unused
3000  * \param data unused
3001  * \param flag indicates whether this cancelling or blocking callback
3002  * \retval 0
3003  * \see ldlm_blocking_ast_nocheck
3004  */
3005 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3006                      void *data, int flag)
3007 {
3008         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3009         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3010         struct ldlm_cb_set_arg *arg = data;
3011         bool commit_async = false;
3012         int rc;
3013         ENTRY;
3014
3015         if (flag == LDLM_CB_CANCELING)
3016                 RETURN(0);
3017
3018         lock_res_and_lock(lock);
3019         if (lock->l_blocking_ast != mdt_blocking_ast) {
3020                 unlock_res_and_lock(lock);
3021                 RETURN(0);
3022         }
3023
3024         /* A blocking ast may be sent from ldlm_lock_decref_internal
3025          * when the last reference to a local lock was released and
3026          * during blocking event from ldlm_work_bl_ast_lock().
3027          * The 'data' parameter is l_ast_data in the first case and
3028          * callback arguments in the second one. Distinguish them by that.
3029          */
3030         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3031                 goto skip_cos_checks;
3032
3033         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3034                 if (mdt_cos_is_enabled(mdt)) {
3035                         if (!arg->bl_desc->bl_same_client)
3036                                 mdt_set_lock_sync(lock);
3037                 } else if (mdt_slc_is_enabled(mdt) &&
3038                            arg->bl_desc->bl_cos_incompat) {
3039                         mdt_set_lock_sync(lock);
3040                         /*
3041                          * we may do extra commit here, but there is a small
3042                          * window to miss a commit: lock was unlocked (saved),
3043                          * then a conflict lock queued and we come here, but
3044                          * REP-ACK not received, so lock was not converted to
3045                          * COS mode yet.
3046                          * Fortunately this window is quite small, so the
3047                          * extra commit should be rare (not to say distributed
3048                          * operation is rare too).
3049                          */
3050                         commit_async = true;
3051                 }
3052         } else if (lock->l_req_mode == LCK_COS) {
3053                 commit_async = true;
3054         }
3055
3056 skip_cos_checks:
3057         rc = ldlm_blocking_ast_nocheck(lock);
3058
3059         if (commit_async) {
3060                 struct lu_env env;
3061
3062                 rc = lu_env_init(&env, LCT_LOCAL);
3063                 if (unlikely(rc != 0))
3064                         CWARN("%s: lu_env initialization failed, cannot "
3065                               "start asynchronous commit: rc = %d\n",
3066                               obd->obd_name, rc);
3067                 else
3068                         mdt_device_commit_async(&env, mdt);
3069                 lu_env_fini(&env);
3070         }
3071         RETURN(rc);
3072 }
3073
3074 /*
3075  * Blocking AST for cross-MDT lock
3076  *
3077  * Discard lock from uncommitted_slc_locks and cancel it.
3078  *
3079  * \param lock  the lock which blocks a request or cancelling lock
3080  * \param desc  unused
3081  * \param data  unused
3082  * \param flag  indicates whether this cancelling or blocking callback
3083  * \retval      0 on success
3084  * \retval      negative number on error
3085  */
3086 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3087                             void *data, int flag)
3088 {
3089         int rc = 0;
3090         ENTRY;
3091
3092         switch (flag) {
3093         case LDLM_CB_BLOCKING: {
3094                 struct lustre_handle lockh;
3095
3096                 ldlm_lock2handle(lock, &lockh);
3097                 rc = ldlm_cli_cancel(&lockh,
3098                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3099                 if (rc < 0) {
3100                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3101                         RETURN(rc);
3102                 }
3103                 break;
3104         }
3105         case LDLM_CB_CANCELING: {
3106                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3107                 struct mdt_device *mdt =
3108                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3109
3110                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3111
3112                 /* discard slc lock here so that it can be cleaned anytime,
3113                  * especially for cleanup_resource() */
3114                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3115
3116                 /* once we cache lock, l_ast_data is set to mdt_object */
3117                 if (lock->l_ast_data != NULL) {
3118                         struct mdt_object *mo = lock->l_ast_data;
3119                         struct lu_env env;
3120
3121                         rc = lu_env_init(&env, LCT_MD_THREAD);
3122                         if (unlikely(rc != 0)) {
3123                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3124                                       obd->obd_name, mo,
3125                                       PFID(mdt_object_fid(mo)), rc);
3126                                 RETURN(rc);
3127                         }
3128
3129                         if (lock->l_policy_data.l_inodebits.bits &
3130                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3131                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3132                                 mo->mot_cache_attr = 0;
3133                         }
3134                         mdt_object_put(&env, mo);
3135                         lu_env_fini(&env);
3136                 }
3137                 break;
3138         }
3139         default:
3140                 LBUG();
3141         }
3142
3143         RETURN(rc);
3144 }
3145
3146 int mdt_check_resent_lock(struct mdt_thread_info *info,
3147                           struct mdt_object *mo,
3148                           struct mdt_lock_handle *lhc)
3149 {
3150         /* the lock might already be gotten in ldlm_handle_enqueue() */
3151         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3152                 struct ptlrpc_request *req = mdt_info_req(info);
3153                 struct ldlm_lock      *lock;
3154
3155                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3156                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3157                 if (lock == NULL) {
3158                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3159                          * a resend case, however, it could be already destroyed
3160                          * due to client eviction or a raced cancel RPC. */
3161                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3162                                           lhc->mlh_reg_lh.cookie);
3163                         RETURN(-ESTALE);
3164                 }
3165
3166                 if (!fid_res_name_eq(mdt_object_fid(mo),
3167                                      &lock->l_resource->lr_name)) {
3168                         CWARN("%s: Although resent, but still not "
3169                               "get child lock:"DFID"\n",
3170                               info->mti_exp->exp_obd->obd_name,
3171                               PFID(mdt_object_fid(mo)));
3172                         LDLM_LOCK_PUT(lock);
3173                         RETURN(-EPROTO);
3174                 }
3175                 LDLM_LOCK_PUT(lock);
3176                 return 0;
3177         }
3178         return 1;
3179 }
3180
3181 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3182 {
3183         mdt_object_get(NULL, lock->l_ast_data);
3184 }
3185
3186 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3187                                struct mdt_object *o, const struct lu_fid *fid,
3188                                struct lustre_handle *lh, enum ldlm_mode mode,
3189                                __u64 *ibits, __u64 trybits, bool cache)
3190 {
3191         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3192         union ldlm_policy_data *policy = &mti->mti_policy;
3193         struct ldlm_res_id *res_id = &mti->mti_res_id;
3194         int rc = 0;
3195         ENTRY;
3196
3197         LASSERT(mdt_object_remote(o));
3198
3199         fid_build_reg_res_name(fid, res_id);
3200
3201         memset(einfo, 0, sizeof(*einfo));
3202         einfo->ei_type = LDLM_IBITS;
3203         einfo->ei_mode = mode;
3204         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3205         einfo->ei_cb_cp = ldlm_completion_ast;
3206         einfo->ei_enq_slave = 0;
3207         einfo->ei_res_id = res_id;
3208
3209         if (cache) {
3210                 /*
3211                  * if we cache lock, couple lock with mdt_object, so that object
3212                  * can be easily found in lock ASTs.
3213                  */
3214                 einfo->ei_cbdata = o;
3215                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3216         }
3217
3218         memset(policy, 0, sizeof(*policy));
3219         policy->l_inodebits.bits = *ibits;
3220         policy->l_inodebits.try_bits = trybits;
3221
3222         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3223                             policy);
3224
3225         /* Return successfully acquired bits to a caller */
3226         if (rc == 0) {
3227                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3228
3229                 LASSERT(lock);
3230                 *ibits = lock->l_policy_data.l_inodebits.bits;
3231                 LDLM_LOCK_PUT(lock);
3232         }
3233         RETURN(rc);
3234 }
3235
3236 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3237                            const struct lu_fid *fid, struct lustre_handle *lh,
3238                            enum ldlm_mode mode, __u64 ibits, bool cache)
3239 {
3240         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3241                                           cache);
3242 }
3243
3244 static int mdt_object_local_lock(struct mdt_thread_info *info,
3245                                  struct mdt_object *o,
3246                                  struct mdt_lock_handle *lh, __u64 *ibits,
3247                                  __u64 trybits, bool cos_incompat)
3248 {
3249         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3250         union ldlm_policy_data *policy = &info->mti_policy;
3251         struct ldlm_res_id *res_id = &info->mti_res_id;
3252         __u64 dlmflags = 0, *cookie = NULL;
3253         int rc;
3254         ENTRY;
3255
3256         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3257         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3258         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3259         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3260
3261         if (cos_incompat) {
3262                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3263                         lh->mlh_reg_mode == LCK_EX);
3264                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3265         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3266                 dlmflags |= LDLM_FL_COS_ENABLED;
3267         }
3268
3269         /* Only enqueue LOOKUP lock for remote object */
3270         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3271
3272         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3273         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3274             *ibits == MDS_INODELOCK_OPEN)
3275                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3276
3277         if (lh->mlh_type == MDT_PDO_LOCK) {
3278                 /* check for exists after object is locked */
3279                 if (mdt_object_exists(o) == 0) {
3280                         /* Non-existent object shouldn't have PDO lock */
3281                         RETURN(-ESTALE);
3282                 } else {
3283                         /* Non-dir object shouldn't have PDO lock */
3284                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3285                                 RETURN(-ENOTDIR);
3286                 }
3287         }
3288
3289         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3290         dlmflags |= LDLM_FL_ATOMIC_CB;
3291
3292         if (info->mti_exp)
3293                 cookie = &info->mti_exp->exp_handle.h_cookie;
3294
3295         /*
3296          * Take PDO lock on whole directory and build correct @res_id for lock
3297          * on part of directory.
3298          */
3299         if (lh->mlh_pdo_hash != 0) {
3300                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3301                 mdt_lock_pdo_mode(info, o, lh);
3302                 if (lh->mlh_pdo_mode != LCK_NL) {
3303                         /*
3304                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3305                          * is never going to be sent to client and we do not
3306                          * want it slowed down due to possible cancels.
3307                          */
3308                         policy->l_inodebits.bits =
3309                                 *ibits & MDS_INODELOCK_UPDATE;
3310                         policy->l_inodebits.try_bits =
3311                                 trybits & MDS_INODELOCK_UPDATE;
3312                         /* at least one of them should be set */
3313                         LASSERT(policy->l_inodebits.bits |
3314                                 policy->l_inodebits.try_bits);
3315                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3316                                           lh->mlh_pdo_mode, policy, res_id,
3317                                           dlmflags, cookie);
3318                         if (unlikely(rc != 0))
3319                                 GOTO(out_unlock, rc);
3320                 }
3321
3322                 /*
3323                  * Finish res_id initializing by name hash marking part of
3324                  * directory which is taking modification.
3325                  */
3326                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3327         }
3328
3329         policy->l_inodebits.bits = *ibits;
3330         policy->l_inodebits.try_bits = trybits;
3331
3332         /*
3333          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3334          * going to be sent to client. If it is - mdt_intent_policy() path will
3335          * fix it up and turn FL_LOCAL flag off.
3336          */
3337         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3338                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3339                           cookie);
3340 out_unlock:
3341         if (rc != 0)
3342                 mdt_object_unlock(info, o, lh, 1);
3343         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3344                    lh->mlh_pdo_hash != 0 &&
3345                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3346                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3347
3348         /* Return successfully acquired bits to a caller */
3349         if (rc == 0) {
3350                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3351
3352                 LASSERT(lock);
3353                 *ibits = lock->l_policy_data.l_inodebits.bits;
3354                 LDLM_LOCK_PUT(lock);
3355         }
3356         RETURN(rc);
3357 }
3358
3359 static int
3360 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3361                          struct mdt_lock_handle *lh, __u64 *ibits,
3362                          __u64 trybits, bool cos_incompat)
3363 {
3364         struct mdt_lock_handle *local_lh = NULL;
3365         int rc;
3366         ENTRY;
3367
3368         if (!mdt_object_remote(o)) {
3369                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3370                                            cos_incompat);
3371                 RETURN(rc);
3372         }
3373
3374         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3375         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3376                     MDS_INODELOCK_XATTR);
3377
3378         /* Only enqueue LOOKUP lock for remote object */
3379         if (*ibits & MDS_INODELOCK_LOOKUP) {
3380                 __u64 local = MDS_INODELOCK_LOOKUP;
3381
3382                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3383                                            cos_incompat);
3384                 if (rc != ELDLM_OK)
3385                         RETURN(rc);
3386
3387                 local_lh = lh;
3388         }
3389
3390         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3391                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3392                  * enqueue RPC can only request 1 lock, to avoid extra
3393                  * RPC, so it will instead enqueue EX lock for remote
3394                  * object anyway XXX*/
3395                 if (lh->mlh_type == MDT_PDO_LOCK &&
3396                     lh->mlh_pdo_hash != 0) {
3397                         CDEBUG(D_INFO,
3398                                "%s: "DFID" convert PDO lock to EX lock.\n",
3399                                mdt_obd_name(info->mti_mdt),
3400                                PFID(mdt_object_fid(o)));
3401                         lh->mlh_pdo_hash = 0;
3402                         lh->mlh_rreg_mode = LCK_EX;
3403                         lh->mlh_type = MDT_REG_LOCK;
3404                 }
3405
3406                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3407                                                 &lh->mlh_rreg_lh,
3408                                                 lh->mlh_rreg_mode,
3409                                                 ibits, trybits, false);
3410                 if (rc != ELDLM_OK) {
3411                         if (local_lh != NULL)
3412                                 mdt_object_unlock(info, o, local_lh, rc);
3413                         RETURN(rc);
3414                 }
3415         }
3416
3417         /* other components like LFSCK can use lockless access
3418          * and populate cache, so we better invalidate it */
3419         mo_invalidate(info->mti_env, mdt_object_child(o));
3420
3421         RETURN(0);
3422 }
3423
3424 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3425                     struct mdt_lock_handle *lh, __u64 ibits)
3426 {
3427         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3428 }
3429
3430 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3431                           struct mdt_lock_handle *lh, __u64 ibits,
3432                           bool cos_incompat)
3433 {
3434         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3435         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3436                                         cos_incompat);
3437 }
3438
3439 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3440                         struct mdt_lock_handle *lh, __u64 *ibits,
3441                         __u64 trybits, bool cos_incompat)
3442 {
3443         bool trylock_only = *ibits == 0;
3444         int rc;
3445
3446         LASSERT(!(*ibits & trybits));
3447         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3448                                       cos_incompat);
3449         if (rc && trylock_only) { /* clear error for try ibits lock only */
3450                 LASSERT(*ibits == 0);
3451                 rc = 0;
3452         }
3453         return rc;
3454 }
3455
3456 /**
3457  * Save a lock within request object.
3458  *
3459  * Keep the lock referenced until whether client ACK or transaction
3460  * commit happens or release the lock immediately depending on input
3461  * parameters. If COS is ON, a write lock is converted to COS lock
3462  * before saving.
3463  *
3464  * \param info thead info object
3465  * \param h lock handle
3466  * \param mode lock mode
3467  * \param decref force immediate lock releasing
3468  */
3469 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3470                    enum ldlm_mode mode, int decref)
3471 {
3472         ENTRY;
3473
3474         if (lustre_handle_is_used(h)) {
3475                 if (decref || !info->mti_has_trans ||
3476                     !(mode & (LCK_PW | LCK_EX))) {
3477                         mdt_fid_unlock(h, mode);
3478                 } else {
3479                         struct mdt_device *mdt = info->mti_mdt;
3480                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3481                         struct ptlrpc_request *req = mdt_info_req(info);
3482                         bool cos = mdt_cos_is_enabled(mdt);
3483                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3484
3485                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3486                                  h->cookie);
3487
3488                         /* there is no request if mdt_object_unlock() is called
3489                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3490                         if (likely(req != NULL)) {
3491                                 LDLM_DEBUG(lock, "save lock request %p reply "
3492                                         "state %p transno %lld\n", req,
3493                                         req->rq_reply_state, req->rq_transno);
3494                                 if (cos) {
3495                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3496                                         mode = LCK_COS;
3497                                 }
3498                                 if (req->rq_export->exp_disconnected)
3499                                         mdt_fid_unlock(h, mode);
3500                                 else
3501                                         ptlrpc_save_lock(req, h, mode, cos,
3502                                                          convert_lock);
3503                         } else {
3504                                 mdt_fid_unlock(h, mode);
3505                         }
3506                         if (mdt_is_lock_sync(lock)) {
3507                                 CDEBUG(D_HA, "found sync-lock,"
3508                                        " async commit started\n");
3509                                 mdt_device_commit_async(info->mti_env,
3510                                                         mdt);
3511                         }
3512                         LDLM_LOCK_PUT(lock);
3513                 }
3514                 h->cookie = 0ull;
3515         }
3516
3517         EXIT;
3518 }
3519
3520 /**
3521  * Save cross-MDT lock in uncommitted_slc_locks
3522  *
3523  * Keep the lock referenced until transaction commit happens or release the lock
3524  * immediately depending on input parameters.
3525  *
3526  * \param info thead info object
3527  * \param h lock handle
3528  * \param mode lock mode
3529  * \param decref force immediate lock releasing
3530  */
3531 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3532                                  struct mdt_object *o, struct lustre_handle *h,
3533                                  enum ldlm_mode mode, int decref)
3534 {
3535         ENTRY;
3536
3537         if (lustre_handle_is_used(h)) {
3538                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3539                 struct ptlrpc_request *req = mdt_info_req(info);
3540
3541                 if (o != NULL &&
3542                     (lock->l_policy_data.l_inodebits.bits &
3543                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3544                         mo_invalidate(info->mti_env, mdt_object_child(o));
3545
3546                 if (decref || !info->mti_has_trans || !req ||
3547                     !(mode & (LCK_PW | LCK_EX))) {
3548                         ldlm_lock_decref_and_cancel(h, mode);
3549                         LDLM_LOCK_PUT(lock);
3550                 } else {
3551                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3552                                           req->rq_transno);
3553                         ldlm_lock_decref(h, mode);
3554                 }
3555                 h->cookie = 0ull;
3556         }
3557
3558         EXIT;
3559 }
3560
3561 /**
3562  * Unlock mdt object.
3563  *
3564  * Immeditely release the regular lock and the PDO lock or save the
3565  * lock in request and keep them referenced until client ACK or
3566  * transaction commit.
3567  *
3568  * \param info thread info object
3569  * \param o mdt object
3570  * \param lh mdt lock handle referencing regular and PDO locks
3571  * \param decref force immediate lock releasing
3572  *
3573  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3574  */
3575 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3576                        struct mdt_lock_handle *lh, int decref)
3577 {
3578         ENTRY;
3579
3580         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3581         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3582         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3583                              decref);
3584
3585         EXIT;
3586 }
3587
3588 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3589                                         const struct lu_fid *f,
3590                                         struct mdt_lock_handle *lh,
3591                                         __u64 ibits)
3592 {
3593         struct mdt_object *o;
3594
3595         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3596         if (!IS_ERR(o)) {
3597                 int rc;
3598
3599                 rc = mdt_object_lock(info, o, lh, ibits);
3600                 if (rc != 0) {
3601                         mdt_object_put(info->mti_env, o);
3602                         o = ERR_PTR(rc);
3603                 }
3604         }
3605         return o;
3606 }
3607
3608 void mdt_object_unlock_put(struct mdt_thread_info * info,
3609                            struct mdt_object * o,
3610                            struct mdt_lock_handle *lh,
3611                            int decref)
3612 {
3613         mdt_object_unlock(info, o, lh, decref);
3614         mdt_object_put(info->mti_env, o);
3615 }
3616
3617 /*
3618  * Generic code handling requests that have struct mdt_body passed in:
3619  *
3620  *  - extract mdt_body from request and save it in @info, if present;
3621  *
3622  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3623  *  @info;
3624  *
3625  *  - if HAS_BODY flag is set for this request type check whether object
3626  *  actually exists on storage (lu_object_exists()).
3627  *
3628  */
3629 static int mdt_body_unpack(struct mdt_thread_info *info,
3630                            enum tgt_handler_flags flags)
3631 {
3632         const struct mdt_body    *body;
3633         struct mdt_object        *obj;
3634         const struct lu_env      *env;
3635         struct req_capsule       *pill;
3636         int                       rc;
3637         ENTRY;
3638
3639         env = info->mti_env;
3640         pill = info->mti_pill;
3641
3642         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3643         if (body == NULL)
3644                 RETURN(-EFAULT);
3645
3646         if (!(body->mbo_valid & OBD_MD_FLID))
3647                 RETURN(0);
3648
3649         if (!fid_is_sane(&body->mbo_fid1)) {
3650                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3651                 RETURN(-EINVAL);
3652         }
3653
3654         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3655         if (!IS_ERR(obj)) {
3656                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3657                         mdt_object_put(env, obj);
3658                         rc = -ENOENT;
3659                 } else {
3660                         info->mti_object = obj;
3661                         rc = 0;
3662                 }
3663         } else
3664                 rc = PTR_ERR(obj);
3665
3666         RETURN(rc);
3667 }
3668
3669 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3670                                    enum tgt_handler_flags flags)
3671 {
3672         struct req_capsule *pill = info->mti_pill;
3673         int rc;
3674
3675         ENTRY;
3676
3677         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3678                 rc = mdt_body_unpack(info, flags);
3679         else
3680                 rc = 0;
3681
3682         if (rc == 0 && (flags & HAS_REPLY)) {
3683                 /* Pack reply. */
3684                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3685                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3686                                              DEF_REP_MD_SIZE);
3687                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3688                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3689                                              RCL_SERVER, 0);
3690
3691                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3692                  * by default. If the target object has more ACL entries, then
3693                  * enlarge the buffer when necessary. */
3694                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3695                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3696                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3697
3698                 mdt_preset_secctx_size(info);
3699
3700                 rc = req_capsule_server_pack(pill);
3701                 if (rc)
3702                         CWARN("%s: cannot pack response: rc = %d\n",
3703                                       mdt_obd_name(info->mti_mdt), rc);
3704         }
3705         RETURN(rc);
3706 }
3707
3708 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3709 {
3710         lh->mlh_type = MDT_NUL_LOCK;
3711         lh->mlh_reg_lh.cookie = 0ull;
3712         lh->mlh_reg_mode = LCK_MINMODE;
3713         lh->mlh_pdo_lh.cookie = 0ull;
3714         lh->mlh_pdo_mode = LCK_MINMODE;
3715         lh->mlh_rreg_lh.cookie = 0ull;
3716         lh->mlh_rreg_mode = LCK_MINMODE;
3717 }
3718
3719 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3720 {
3721         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3722         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3723 }
3724
3725 /*
3726  * Initialize fields of struct mdt_thread_info. Other fields are left in
3727  * uninitialized state, because it's too expensive to zero out whole
3728  * mdt_thread_info (> 1K) on each request arrival.
3729  */
3730 void mdt_thread_info_init(struct ptlrpc_request *req,
3731                           struct mdt_thread_info *info)
3732 {
3733         int i;
3734
3735         info->mti_pill = &req->rq_pill;
3736
3737         /* lock handle */
3738         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3739                 mdt_lock_handle_init(&info->mti_lh[i]);
3740
3741         /* mdt device: it can be NULL while CONNECT */
3742         if (req->rq_export) {
3743                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3744                 info->mti_exp = req->rq_export;
3745         } else
3746                 info->mti_mdt = NULL;
3747         info->mti_env = req->rq_svc_thread->t_env;
3748         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3749
3750         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3751         info->mti_big_buf = LU_BUF_NULL;
3752         info->mti_body = NULL;
3753         info->mti_object = NULL;
3754         info->mti_dlm_req = NULL;
3755         info->mti_has_trans = 0;
3756         info->mti_cross_ref = 0;
3757         info->mti_opdata = 0;
3758         info->mti_big_lmm_used = 0;
3759         info->mti_big_acl_used = 0;
3760         info->mti_som_valid = 0;
3761
3762         info->mti_spec.no_create = 0;
3763         info->mti_spec.sp_rm_entry = 0;
3764         info->mti_spec.sp_permitted = 0;
3765
3766         info->mti_spec.u.sp_ea.eadata = NULL;
3767         info->mti_spec.u.sp_ea.eadatalen = 0;
3768 }
3769
3770 void mdt_thread_info_fini(struct mdt_thread_info *info)
3771 {
3772         int i;
3773
3774         if (info->mti_object != NULL) {
3775                 mdt_object_put(info->mti_env, info->mti_object);
3776                 info->mti_object = NULL;
3777         }
3778
3779         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3780                 mdt_lock_handle_fini(&info->mti_lh[i]);
3781         info->mti_env = NULL;
3782         info->mti_pill = NULL;
3783         info->mti_exp = NULL;
3784         info->mti_mdt = NULL;
3785
3786         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3787                 lu_buf_free(&info->mti_big_buf);
3788 }
3789
3790 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3791 {
3792         struct mdt_thread_info  *mti;
3793
3794         mti = mdt_th_info(tsi->tsi_env);
3795         LASSERT(mti != NULL);
3796
3797         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3798         if (tsi->tsi_corpus != NULL) {
3799                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3800                 lu_object_get(tsi->tsi_corpus);
3801         }
3802         mti->mti_body = tsi->tsi_mdt_body;
3803         mti->mti_dlm_req = tsi->tsi_dlm_req;
3804
3805         return mti;
3806 }
3807
3808 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3809 {
3810         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3811             cfs_fail_val ==
3812             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
3813                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
3814
3815         return tgt_connect(tsi);
3816 }
3817
3818 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3819                               struct mdt_thread_info *info,
3820                               struct ldlm_lock **lockp, __u64 flags)
3821 {
3822         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3823                                    lockp, flags);
3824 }
3825 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3826                           struct mdt_thread_info *info,
3827                           struct ldlm_lock **lockp, __u64 flags)
3828 {
3829         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3830                                lockp, flags);
3831 }
3832
3833 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3834                             struct ldlm_lock **lockp,
3835                             struct mdt_lock_handle *lh,
3836                             __u64 flags, int result)
3837 {
3838         struct ptlrpc_request  *req = mdt_info_req(info);
3839         struct ldlm_lock       *lock = *lockp;
3840         struct ldlm_lock       *new_lock;
3841
3842         /* If possible resent found a lock, @lh is set to its handle */
3843         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3844
3845         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3846                 lh->mlh_reg_lh.cookie = 0;
3847                 RETURN(0);
3848         }
3849
3850         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3851                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3852                  * a resend case, however, it could be already destroyed
3853                  * due to client eviction or a raced cancel RPC. */
3854                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3855                                   lh->mlh_reg_lh.cookie);
3856                 lh->mlh_reg_lh.cookie = 0;
3857                 RETURN(-ESTALE);
3858         }
3859
3860         LASSERTF(new_lock != NULL,
3861                  "lockh %#llx flags %#llx : rc = %d\n",
3862                  lh->mlh_reg_lh.cookie, flags, result);
3863
3864         /*
3865          * If we've already given this lock to a client once, then we should
3866          * have no readers or writers.  Otherwise, we should have one reader
3867          * _or_ writer ref (which will be zeroed below) before returning the
3868          * lock to a client.
3869          */
3870         if (new_lock->l_export == req->rq_export) {
3871                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3872         } else {
3873                 LASSERT(new_lock->l_export == NULL);
3874                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3875         }
3876
3877         *lockp = new_lock;
3878
3879         if (new_lock->l_export == req->rq_export) {
3880                 /*
3881                  * Already gave this to the client, which means that we
3882                  * reconstructed a reply.
3883                  */
3884                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3885                         MSG_RESENT);
3886
3887                 LDLM_LOCK_RELEASE(new_lock);
3888                 lh->mlh_reg_lh.cookie = 0;
3889                 RETURN(ELDLM_LOCK_REPLACED);
3890         }
3891
3892         /*
3893          * Fixup the lock to be given to the client.
3894          */
3895         lock_res_and_lock(new_lock);
3896         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3897          * possible blocking AST. */
3898         while (new_lock->l_readers > 0) {
3899                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3900                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3901                 new_lock->l_readers--;
3902         }
3903         while (new_lock->l_writers > 0) {
3904                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3905                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3906                 new_lock->l_writers--;
3907         }
3908
3909         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3910         new_lock->l_blocking_ast = lock->l_blocking_ast;
3911         new_lock->l_completion_ast = lock->l_completion_ast;
3912         if (ldlm_has_dom(new_lock))
3913                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3914         new_lock->l_remote_handle = lock->l_remote_handle;
3915         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3916
3917         unlock_res_and_lock(new_lock);
3918
3919         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3920                      &new_lock->l_remote_handle,
3921                      &new_lock->l_exp_hash);
3922
3923         LDLM_LOCK_RELEASE(new_lock);
3924         lh->mlh_reg_lh.cookie = 0;
3925
3926         RETURN(ELDLM_LOCK_REPLACED);
3927 }
3928
3929 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3930                              struct ldlm_lock *new_lock,
3931                              struct mdt_lock_handle *lh, __u64 flags)
3932 {
3933         struct ptlrpc_request  *req = mdt_info_req(info);
3934         struct ldlm_request    *dlmreq;
3935
3936         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3937                 return;
3938
3939         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3940
3941         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3942          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3943          * initialized. */
3944         if (flags & LDLM_FL_RESENT) {
3945                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3946                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3947
3948                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3949                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3950                           lh->mlh_reg_lh.cookie);
3951                 return;
3952         }
3953
3954         /*
3955          * If the xid matches, then we know this is a resent request, and allow
3956          * it. (It's probably an OPEN, for which we don't send a lock.
3957          */
3958         if (req_can_reconstruct(req, NULL) != 0)
3959                 return;
3960
3961         /*
3962          * This remote handle isn't enqueued, so we never received or processed
3963          * this request.  Clear MSG_RESENT, because it can be handled like any
3964          * normal request now.
3965          */
3966         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3967
3968         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3969                   dlmreq->lock_handle[0].cookie);
3970 }
3971
3972 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3973                                struct mdt_thread_info *info,
3974                                struct ldlm_lock **lockp,
3975                                __u64 flags)
3976 {
3977         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3978         struct ldlm_reply      *ldlm_rep = NULL;
3979         int rc;
3980         ENTRY;
3981
3982         /*
3983          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3984          * (for the resend case) or a new lock. Below we will use it to
3985          * replace the original lock.
3986          */
3987         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3988         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3989                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3990                 rc = mdt_object_lock(info, info->mti_object, lhc,
3991                                      MDS_INODELOCK_XATTR);
3992                 if (rc)
3993                         return rc;
3994         }
3995
3996         rc = mdt_getxattr(info);
3997
3998         if (mdt_info_req(info)->rq_repmsg != NULL)
3999                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4000
4001         if (ldlm_rep == NULL ||
4002             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4003                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4004                 if (is_serious(rc))
4005                         RETURN(rc);
4006                 else
4007                         RETURN(err_serious(-EFAULT));
4008         }
4009
4010         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4011
4012         /* This is left for interop instead of adding a new interop flag.
4013          * LU-7433 */
4014 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4015         if (ldlm_rep->lock_policy_res2) {
4016                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4017                 RETURN(ELDLM_LOCK_ABORTED);
4018         }
4019 #endif
4020
4021         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4022         RETURN(rc);
4023 }
4024
4025 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4026                               struct mdt_thread_info *info,
4027                               struct ldlm_lock **lockp,
4028                               __u64 flags)
4029 {
4030         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4031         __u64                   child_bits;
4032         struct ldlm_reply      *ldlm_rep;
4033         struct mdt_body        *reqbody;
4034         struct mdt_body        *repbody;
4035         int                     rc, rc2;
4036         ENTRY;
4037
4038         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4039         LASSERT(reqbody);
4040
4041         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4042         LASSERT(repbody);
4043
4044         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4045         repbody->mbo_eadatasize = 0;
4046         repbody->mbo_aclsize = 0;
4047
4048         switch (it_opc) {
4049         case IT_LOOKUP:
4050                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4051                 break;
4052         case IT_GETATTR:
4053                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4054                              MDS_INODELOCK_PERM;
4055                 break;
4056         default:
4057                 CERROR("%s: unsupported intent %#x\n",
4058                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4059                 GOTO(out_shrink, rc = -EINVAL);
4060         }
4061
4062         rc = mdt_init_ucred_intent_getattr(info, reqbody);
4063         if (rc)
4064                 GOTO(out_shrink, rc);
4065
4066         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4067         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4068
4069         /* Get lock from request for possible resent case. */
4070         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4071
4072         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4073         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4074
4075         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4076                 ldlm_rep->lock_policy_res2 = 0;
4077         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4078             ldlm_rep->lock_policy_res2) {
4079                 lhc->mlh_reg_lh.cookie = 0ull;
4080                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4081         }
4082
4083         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4084         EXIT;
4085 out_ucred:
4086         mdt_exit_ucred(info);
4087 out_shrink:
4088         mdt_client_compatibility(info);
4089         rc2 = mdt_fix_reply(info);
4090         if (rc == 0)
4091                 rc = rc2;
4092         return rc;
4093 }
4094
4095 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4096                              struct mdt_thread_info *info,
4097                              struct ldlm_lock **lockp,
4098                              __u64 flags)
4099 {
4100         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4101         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4102         struct layout_intent *intent;
4103         struct ldlm_reply *ldlm_rep;
4104         struct lu_fid *fid = &info->mti_tmp_fid2;
4105         struct mdt_object *obj = NULL;
4106         int layout_size = 0;
4107         struct lu_buf *buf = &layout.mlc_buf;
4108         int rc = 0;
4109
4110         ENTRY;
4111
4112         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4113
4114         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4115         if (intent == NULL)
4116                 RETURN(-EPROTO);
4117
4118         CDEBUG(D_INFO, DFID "got layout change request from client: "
4119                "opc:%u flags:%#x extent "DEXT"\n",
4120                PFID(fid), intent->li_opc, intent->li_flags,
4121                PEXT(&intent->li_extent));
4122
4123         switch (intent->li_opc) {
4124         case LAYOUT_INTENT_TRUNC:
4125         case LAYOUT_INTENT_WRITE:
4126                 layout.mlc_opc = MD_LAYOUT_WRITE;
4127                 layout.mlc_intent = intent;
4128                 break;
4129         case LAYOUT_INTENT_ACCESS:
4130                 break;
4131         case LAYOUT_INTENT_READ:
4132         case LAYOUT_INTENT_GLIMPSE:
4133         case LAYOUT_INTENT_RELEASE:
4134         case LAYOUT_INTENT_RESTORE:
4135                 CERROR("%s: Unsupported layout intent opc %d\n",
4136                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4137                 RETURN(-ENOTSUPP);
4138         default:
4139                 CERROR("%s: Unknown layout intent opc %d\n",
4140                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4141                 RETURN(-EINVAL);
4142         }
4143
4144         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4145         if (IS_ERR(obj))
4146                 RETURN(PTR_ERR(obj));
4147
4148         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4149                 /* if layout is going to be changed don't use the current EA
4150                  * size but the maximum one. That buffer will be shrinked
4151                  * to the actual size in req_capsule_shrink() before reply.
4152                  */
4153                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4154                         layout_size = info->mti_mdt->mdt_max_mdsize;
4155                 } else {
4156                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4157                         if (layout_size < 0)
4158                                 GOTO(out, rc = layout_size);
4159
4160                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4161                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4162                 }
4163                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4164                        mdt_obd_name(info->mti_mdt), layout_size);
4165         }
4166
4167         /*
4168          * set reply buffer size, so that ldlm_handle_enqueue0()->
4169          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4170          */
4171         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4172                              layout_size);
4173         rc = req_capsule_server_pack(info->mti_pill);
4174         if (rc)
4175                 GOTO(out, rc);
4176
4177         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4178         if (!ldlm_rep)
4179                 GOTO(out, rc = -EPROTO);
4180
4181         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4182
4183         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4184         if (layout.mlc_opc == MD_LAYOUT_NOP)
4185                 GOTO(out, rc = 0);
4186
4187         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4188         if (rc < 0)
4189                 GOTO(out, rc);
4190         if (rc == 1) {
4191                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4192                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4193                 GOTO(out, rc);
4194         }
4195
4196         buf->lb_buf = NULL;
4197         buf->lb_len = 0;
4198         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4199                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4200                                                      &RMF_EADATA);
4201                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4202                                                      &RMF_EADATA, RCL_CLIENT);
4203                 /*
4204                  * If it's a replay of layout write intent RPC, the client has
4205                  * saved the extended lovea when it get reply then.
4206                  */
4207                 if (buf->lb_len > 0)
4208                         mdt_fix_lov_magic(info, buf->lb_buf);
4209         }
4210
4211         /* Get lock from request for possible resent case. */
4212         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4213         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4214
4215         /*
4216          * Instantiate some layout components, if @buf contains lovea, then it's
4217          * a replay of the layout intent write RPC.
4218          */
4219         rc = mdt_layout_change(info, obj, lhc, &layout);
4220         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4221
4222         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4223                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4224                 if (rc == ELDLM_LOCK_REPLACED &&
4225                     (*lockp)->l_granted_mode == LCK_EX)
4226                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4227         }
4228
4229         EXIT;
4230 out:
4231         mdt_object_put(info->mti_env, obj);
4232         return rc;
4233 }
4234
4235 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4236                            struct mdt_thread_info *info,
4237                            struct ldlm_lock **lockp,
4238                            __u64 flags)
4239 {
4240         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4241         struct ldlm_reply      *rep = NULL;
4242         long                    opc;
4243         int                     rc;
4244
4245         static const struct req_format *intent_fmts[REINT_MAX] = {
4246                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4247                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4248         };
4249
4250         ENTRY;
4251
4252         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4253         if (opc < 0)
4254                 RETURN(opc);
4255
4256         /* Get lock from request for possible resent case. */
4257         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4258
4259         rc = mdt_reint_internal(info, lhc, opc);
4260
4261         /* Check whether the reply has been packed successfully. */
4262         if (mdt_info_req(info)->rq_repmsg != NULL)
4263                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4264         if (rep == NULL) {
4265                 if (is_serious(rc))
4266                         RETURN(rc);
4267                 else
4268                         RETURN(err_serious(-EFAULT));
4269         }
4270
4271         /* MDC expects this in any case */
4272         if (rc != 0)
4273                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4274
4275         /* the open lock or the lock for cross-ref object should be
4276          * returned to the client */
4277         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4278             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4279                 rep->lock_policy_res2 = 0;
4280                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4281                 RETURN(rc);
4282         }
4283
4284         rep->lock_policy_res2 = clear_serious(rc);
4285
4286         if (rep->lock_policy_res2 == -ENOENT &&
4287             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4288             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4289                 rep->lock_policy_res2 = 0;
4290
4291         lhc->mlh_reg_lh.cookie = 0ull;
4292         if (rc == -ENOTCONN || rc == -ENODEV ||
4293             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4294                 /*
4295                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4296                  * will be returned by rq_status, and client at ptlrpc layer
4297                  * will detect this, then disconnect, reconnect the import
4298                  * immediately, instead of impacting the following the rpc.
4299                  */
4300                 RETURN(rc);
4301         }
4302         /*
4303          * For other cases, the error will be returned by intent, and client
4304          * will retrieve the result from intent.
4305          */
4306         RETURN(ELDLM_LOCK_ABORTED);
4307 }
4308
4309 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4310                           struct mdt_thread_info *info,
4311                           struct ldlm_lock **lockp,
4312                           u64 flags /* LDLM_FL_* */)
4313 {
4314         struct req_capsule *pill = info->mti_pill;
4315         struct ptlrpc_request *req = mdt_info_req(info);
4316         const struct req_format *it_format;
4317         int (*it_handler)(enum ldlm_intent_flags,
4318                           struct mdt_thread_info *,
4319                           struct ldlm_lock **,
4320                           u64);
4321         enum tgt_handler_flags it_handler_flags = 0;
4322         struct ldlm_reply *rep;
4323         bool check_mdt_object = false;
4324         int rc;
4325         ENTRY;
4326
4327         switch (it_opc) {
4328         case IT_OPEN:
4329         case IT_OPEN|IT_CREAT:
4330                 /*
4331                  * OCREAT is not a IS_MUTABLE request since the file may
4332                  * already exist. We do the extra check of
4333                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4334                  * really need to create the object.
4335                  */
4336                 it_format = &RQF_LDLM_INTENT;
4337                 it_handler = &mdt_intent_open;
4338                 break;
4339         case IT_GETATTR:
4340                 check_mdt_object = true;
4341                 /* fallthrough */
4342         case IT_LOOKUP:
4343                 it_format = &RQF_LDLM_INTENT_GETATTR;
4344                 it_handler = &mdt_intent_getattr;
4345                 it_handler_flags = HAS_REPLY;
4346                 break;
4347         case IT_GETXATTR:
4348                 check_mdt_object = true;
4349                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4350                 it_handler = &mdt_intent_getxattr;
4351                 it_handler_flags = HAS_BODY;
4352                 break;
4353         case IT_LAYOUT:
4354                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4355                 it_handler = &mdt_intent_layout;
4356                 break;
4357         case IT_GLIMPSE:
4358                 it_format = &RQF_LDLM_INTENT;
4359                 it_handler = &mdt_intent_glimpse;
4360                 break;
4361         case IT_BRW:
4362                 it_format = &RQF_LDLM_INTENT;
4363                 it_handler = &mdt_intent_brw;
4364                 break;
4365         case IT_QUOTA_DQACQ:
4366         case IT_QUOTA_CONN: {
4367                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4368
4369                 if (qmt == NULL)
4370                         RETURN(-EOPNOTSUPP);
4371
4372                 if (mdt_rdonly(req->rq_export))
4373                         RETURN(-EROFS);
4374
4375                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4376                 /* pass the request to quota master */
4377                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4378                                                  mdt_info_req(info), lockp,
4379                                                  flags);
4380                 RETURN(rc);
4381         }
4382         default:
4383                 CERROR("%s: unknown intent code %#x\n",
4384                        mdt_obd_name(info->mti_mdt), it_opc);
4385                 RETURN(-EPROTO);
4386         }
4387
4388         req_capsule_extend(pill, it_format);
4389
4390         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4391         if (rc < 0)
4392                 RETURN(rc);
4393
4394         if (unlikely(info->mti_object == NULL && check_mdt_object))
4395                 RETURN(-EPROTO);
4396
4397         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4398                 RETURN(-EROFS);
4399
4400         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4401
4402         /* execute policy */
4403         rc = (*it_handler)(it_opc, info, lockp, flags);
4404
4405         /* Check whether the reply has been packed successfully. */
4406         if (req->rq_repmsg != NULL) {
4407                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4408                 rep->lock_policy_res2 =
4409                         ptlrpc_status_hton(rep->lock_policy_res2);
4410         }
4411
4412         RETURN(rc);
4413 }
4414
4415 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4416                                     enum ldlm_intent_flags it_opc)
4417 {
4418         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4419
4420         /* update stats when IT code is known */
4421         if (srv_stats != NULL)
4422                 lprocfs_counter_incr(srv_stats,
4423                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4424                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4425 }
4426
4427 static int mdt_intent_policy(const struct lu_env *env,
4428                              struct ldlm_namespace *ns,
4429                              struct ldlm_lock **lockp,
4430                              void *req_cookie,
4431                              enum ldlm_mode mode,
4432                              __u64 flags, void *data)
4433 {
4434         struct tgt_session_info *tsi;
4435         struct mdt_thread_info  *info;
4436         struct ptlrpc_request   *req  =  req_cookie;
4437         struct ldlm_intent      *it;
4438         struct req_capsule      *pill;
4439         const struct ldlm_lock_desc *ldesc;
4440         int rc;
4441
4442         ENTRY;
4443
4444         LASSERT(req != NULL);
4445
4446         tsi = tgt_ses_info(env);
4447
4448         info = tsi2mdt_info(tsi);
4449         LASSERT(info != NULL);
4450         pill = info->mti_pill;
4451         LASSERT(pill->rc_req == req);
4452         ldesc = &info->mti_dlm_req->lock_desc;
4453
4454         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4455                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4456                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4457                 if (it != NULL) {
4458                         mdt_ptlrpc_stats_update(req, it->opc);
4459                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4460                         if (rc == 0)
4461                                 rc = ELDLM_OK;
4462
4463                         /* Lock without inodebits makes no sense and will oops
4464                          * later in ldlm. Let's check it now to see if we have
4465                          * ibits corrupted somewhere in mdt_intent_opc().
4466                          * The case for client miss to set ibits has been
4467                          * processed by others. */
4468                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4469                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4470                 } else {
4471                         rc = err_serious(-EFAULT);
4472                 }
4473         } else {
4474                 /* No intent was provided */
4475                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4476                 rc = req_capsule_server_pack(pill);
4477                 if (rc)
4478                         rc = err_serious(rc);
4479         }
4480         mdt_thread_info_fini(info);
4481         RETURN(rc);
4482 }
4483
4484 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4485 {
4486         struct seq_server_site  *ss = mdt_seq_site(mdt);
4487
4488         if (ss->ss_node_id == 0)
4489                 return;
4490
4491         if (ss->ss_client_seq != NULL) {
4492                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4493                 ss->ss_client_seq->lcs_exp = NULL;
4494         }
4495
4496         if (ss->ss_server_fld != NULL) {
4497                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4498                 ss->ss_server_fld->lsf_control_exp = NULL;
4499         }
4500 }
4501
4502 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4503 {
4504         struct seq_server_site *ss = mdt_seq_site(mdt);
4505
4506         if (ss == NULL)
4507                 return;
4508
4509         if (ss->ss_server_seq != NULL)
4510                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4511 }
4512
4513 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4514 {
4515         mdt_seq_fini_cli(mdt);
4516         mdt_deregister_seq_exp(mdt);
4517
4518         return seq_site_fini(env, mdt_seq_site(mdt));
4519 }
4520
4521 /**
4522  * It will retrieve its FLDB entries from MDT0, and it only happens
4523  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4524  * and it needs to refresh FLDB from the MDT0.
4525  **/
4526 static int mdt_register_lwp_callback(void *data)
4527 {
4528         struct lu_env           env;
4529         struct mdt_device       *mdt = data;
4530         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4531         int                     rc;
4532         ENTRY;
4533
4534         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4535
4536         rc = lu_env_init(&env, LCT_MD_THREAD);
4537         if (rc < 0) {
4538                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4539                 RETURN(rc);
4540         }
4541
4542         /* Allocate new sequence now to avoid creating local transaction
4543          * in the normal transaction process */
4544         rc = seq_server_check_and_alloc_super(&env,
4545                                               mdt_seq_site(mdt)->ss_server_seq);
4546         if (rc < 0)
4547                 GOTO(out, rc);
4548
4549         if (fld->lsf_new) {
4550                 rc = fld_update_from_controller(&env, fld);
4551                 if (rc != 0) {
4552                         CERROR("%s: cannot update controller: rc = %d\n",
4553                                mdt_obd_name(mdt), rc);
4554                         GOTO(out, rc);
4555                 }
4556         }
4557 out:
4558         lu_env_fini(&env);
4559         RETURN(rc);
4560 }
4561
4562 static int mdt_register_seq_exp(struct mdt_device *mdt)
4563 {
4564         struct seq_server_site  *ss = mdt_seq_site(mdt);
4565         char                    *lwp_name = NULL;
4566         int                     rc;
4567
4568         if (ss->ss_node_id == 0)
4569                 return 0;
4570
4571         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4572         if (lwp_name == NULL)
4573                 GOTO(out_free, rc = -ENOMEM);
4574
4575         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4576         if (rc != 0)
4577                 GOTO(out_free, rc);
4578
4579         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4580                                       NULL, NULL);
4581         if (rc != 0)
4582                 GOTO(out_free, rc);
4583
4584         rc = lustre_register_lwp_item(lwp_name,
4585                                       &ss->ss_server_fld->lsf_control_exp,
4586                                       mdt_register_lwp_callback, mdt);
4587         if (rc != 0) {
4588                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4589                 ss->ss_client_seq->lcs_exp = NULL;
4590                 GOTO(out_free, rc);
4591         }
4592 out_free:
4593         if (lwp_name != NULL)
4594                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4595
4596         return rc;
4597 }
4598
4599 /*
4600  * Init client sequence manager which is used by local MDS to talk to sequence
4601  * controller on remote node.
4602  */
4603 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4604 {
4605         struct seq_server_site *ss = mdt_seq_site(mdt);
4606         char *prefix;
4607         ENTRY;
4608
4609         /* check if this is adding the first MDC and controller is not yet
4610          * initialized. */
4611         OBD_ALLOC_PTR(ss->ss_client_seq);
4612         if (ss->ss_client_seq == NULL)
4613                 RETURN(-ENOMEM);
4614
4615         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4616         if (prefix == NULL) {
4617                 OBD_FREE_PTR(ss->ss_client_seq);
4618                 ss->ss_client_seq = NULL;
4619                 RETURN(-ENOMEM);
4620         }
4621
4622         /* Note: seq_client_fini will be called in seq_site_fini */
4623         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4624         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4625                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4626                                                             NULL);
4627         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4628
4629         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
4630 }
4631
4632 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4633 {
4634         struct seq_server_site  *ss;
4635         int                     rc;
4636         ENTRY;
4637
4638         ss = mdt_seq_site(mdt);
4639         /* init sequence controller server(MDT0) */
4640         if (ss->ss_node_id == 0) {
4641                 OBD_ALLOC_PTR(ss->ss_control_seq);
4642                 if (ss->ss_control_seq == NULL)
4643                         RETURN(-ENOMEM);
4644
4645                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4646                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4647                                      ss);
4648                 if (rc)
4649                         GOTO(out_seq_fini, rc);
4650         }
4651
4652         /* Init normal sequence server */
4653         OBD_ALLOC_PTR(ss->ss_server_seq);
4654         if (ss->ss_server_seq == NULL)
4655                 GOTO(out_seq_fini, rc = -ENOMEM);
4656
4657         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4658                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4659         if (rc)
4660                 GOTO(out_seq_fini, rc);
4661
4662         /* init seq client for seq server to talk to seq controller(MDT0) */
4663         rc = mdt_seq_init_cli(env, mdt);
4664         if (rc != 0)
4665                 GOTO(out_seq_fini, rc);
4666
4667         if (ss->ss_node_id != 0)
4668                 /* register controller export through lwp */
4669                 rc = mdt_register_seq_exp(mdt);
4670
4671         EXIT;
4672 out_seq_fini:
4673         if (rc)
4674                 mdt_seq_fini(env, mdt);
4675
4676         return rc;
4677 }
4678
4679 /*
4680  * FLD wrappers
4681  */
4682 static int mdt_fld_fini(const struct lu_env *env,
4683                         struct mdt_device *m)
4684 {
4685         struct seq_server_site *ss = mdt_seq_site(m);
4686         ENTRY;
4687
4688         if (ss && ss->ss_server_fld) {
4689                 fld_server_fini(env, ss->ss_server_fld);
4690                 OBD_FREE_PTR(ss->ss_server_fld);
4691                 ss->ss_server_fld = NULL;
4692         }
4693
4694         RETURN(0);
4695 }
4696
4697 static int mdt_fld_init(const struct lu_env *env,
4698                         const char *uuid,
4699                         struct mdt_device *m)
4700 {
4701         struct seq_server_site *ss;
4702         int rc;
4703         ENTRY;
4704
4705         ss = mdt_seq_site(m);
4706
4707         OBD_ALLOC_PTR(ss->ss_server_fld);
4708         if (ss->ss_server_fld == NULL)
4709                 RETURN(rc = -ENOMEM);
4710
4711         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4712                              LU_SEQ_RANGE_MDT);
4713         if (rc) {
4714                 OBD_FREE_PTR(ss->ss_server_fld);
4715                 ss->ss_server_fld = NULL;
4716                 RETURN(rc);
4717         }
4718
4719         RETURN(0);
4720 }
4721
4722 static void mdt_stack_pre_fini(const struct lu_env *env,
4723                            struct mdt_device *m, struct lu_device *top)
4724 {
4725         struct lustre_cfg_bufs  *bufs;
4726         struct lustre_cfg       *lcfg;
4727         struct mdt_thread_info  *info;
4728         ENTRY;
4729
4730         LASSERT(top);
4731
4732         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4733         LASSERT(info != NULL);
4734
4735         bufs = &info->mti_u.bufs;
4736
4737         LASSERT(m->mdt_child_exp);
4738         LASSERT(m->mdt_child_exp->exp_obd);
4739
4740         /* process cleanup, pass mdt obd name to get obd umount flags */
4741         /* XXX: this is needed because all layers are referenced by
4742          * objects (some of them are pinned by osd, for example *
4743          * the proper solution should be a model where object used
4744          * by osd only doesn't have mdt/mdd slices -bzzz */
4745         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4746         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4747         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4748         if (!lcfg)
4749                 RETURN_EXIT;
4750         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4751
4752         top->ld_ops->ldo_process_config(env, top, lcfg);
4753         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4754         EXIT;
4755 }
4756
4757 static void mdt_stack_fini(const struct lu_env *env,
4758                            struct mdt_device *m, struct lu_device *top)
4759 {
4760         struct obd_device       *obd = mdt2obd_dev(m);
4761         struct lustre_cfg_bufs  *bufs;
4762         struct lustre_cfg       *lcfg;
4763         struct mdt_thread_info  *info;
4764         char                     flags[3] = "";
4765         ENTRY;
4766
4767         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4768         LASSERT(info != NULL);
4769
4770         lu_dev_del_linkage(top->ld_site, top);
4771
4772         lu_site_purge(env, top->ld_site, -1);
4773
4774         bufs = &info->mti_u.bufs;
4775         /* process cleanup, pass mdt obd name to get obd umount flags */
4776         /* another purpose is to let all layers to release their objects */
4777         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4778         if (obd->obd_force)
4779                 strcat(flags, "F");
4780         if (obd->obd_fail)
4781                 strcat(flags, "A");
4782         lustre_cfg_bufs_set_string(bufs, 1, flags);
4783         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4784         if (!lcfg)
4785                 RETURN_EXIT;
4786         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4787
4788         LASSERT(top);
4789         top->ld_ops->ldo_process_config(env, top, lcfg);
4790         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4791
4792         lu_site_purge(env, top->ld_site, -1);
4793
4794         m->mdt_child = NULL;
4795         m->mdt_bottom = NULL;
4796
4797         obd_disconnect(m->mdt_child_exp);
4798         m->mdt_child_exp = NULL;
4799
4800         obd_disconnect(m->mdt_bottom_exp);
4801         m->mdt_child_exp = NULL;
4802 }
4803
4804 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4805                                const char *next, struct obd_export **exp)
4806 {
4807         struct obd_connect_data *data = NULL;
4808         struct obd_device       *obd;
4809         int                      rc;
4810         ENTRY;
4811
4812         OBD_ALLOC_PTR(data);
4813         if (data == NULL)
4814                 GOTO(out, rc = -ENOMEM);
4815
4816         obd = class_name2obd(next);
4817         if (obd == NULL) {
4818                 CERROR("%s: can't locate next device: %s\n",
4819                        mdt_obd_name(m), next);
4820                 GOTO(out, rc = -ENOTCONN);
4821         }
4822
4823         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4824         data->ocd_version = LUSTRE_VERSION_CODE;
4825
4826         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4827         if (rc) {
4828                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4829                        mdt_obd_name(m), next, rc);
4830                 GOTO(out, rc);
4831         }
4832
4833 out:
4834         if (data)
4835                 OBD_FREE_PTR(data);
4836         RETURN(rc);
4837 }
4838
4839 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4840                           struct lustre_cfg *cfg)
4841 {
4842         char                   *dev = lustre_cfg_string(cfg, 0);
4843         int                     rc, name_size, uuid_size;
4844         char                   *name, *uuid, *p;
4845         struct lustre_cfg_bufs *bufs;
4846         struct lustre_cfg      *lcfg;
4847         struct obd_device      *obd;
4848         struct lustre_profile  *lprof;
4849         struct lu_site         *site;
4850         ENTRY;
4851
4852         /* in 1.8 we had the only device in the stack - MDS.
4853          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4854          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4855          * to generate names and setup MDT, MDD. MDT will be using
4856          * generated name to connect to MDD. for MDD the next device
4857          * will be LOD with name taken from so called "profile" which
4858          * is generated by mount_option line
4859          *
4860          * 1.8 MGS generates config. commands like this:
4861          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4862          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4863          * 2.0 MGS generates config. commands like this:
4864          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4865          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4866          *                    3:lustre-MDT0000-mdtlov  4:f
4867          *
4868          * we generate MDD name from MDT one, just replacing T with D
4869          *
4870          * after all the preparations, the logical equivalent will be
4871          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4872          *                    3:lustre-MDT0000-mdtlov  4:f
4873          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4874          *                    3:lustre-MDD0000  4:f
4875          *
4876          *  notice we build the stack from down to top: MDD first, then MDT */
4877
4878         name_size = MAX_OBD_NAME;
4879         uuid_size = MAX_OBD_NAME;
4880
4881         OBD_ALLOC(name, name_size);
4882         OBD_ALLOC(uuid, uuid_size);
4883         if (name == NULL || uuid == NULL)
4884                 GOTO(cleanup_mem, rc = -ENOMEM);
4885
4886         OBD_ALLOC_PTR(bufs);
4887         if (!bufs)
4888                 GOTO(cleanup_mem, rc = -ENOMEM);
4889
4890         strcpy(name, dev);
4891         p = strstr(name, "-MDT");
4892         if (p == NULL)
4893                 GOTO(free_bufs, rc = -ENOMEM);
4894         p[3] = 'D';
4895
4896         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4897
4898         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4899         if (lprof == NULL || lprof->lp_dt == NULL) {
4900                 CERROR("can't find the profile: %s\n",
4901                        lustre_cfg_string(cfg, 0));
4902                 GOTO(free_bufs, rc = -EINVAL);
4903         }
4904
4905         lustre_cfg_bufs_reset(bufs, name);
4906         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4907         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4908         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4909
4910         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4911         if (!lcfg)
4912                 GOTO(put_profile, rc = -ENOMEM);
4913         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4914
4915         rc = class_attach(lcfg);
4916         if (rc)
4917                 GOTO(lcfg_cleanup, rc);
4918
4919         obd = class_name2obd(name);
4920         if (!obd) {
4921                 CERROR("Can not find obd %s (%s in config)\n",
4922                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4923                 GOTO(lcfg_cleanup, rc = -EINVAL);
4924         }
4925
4926         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4927
4928         lustre_cfg_bufs_reset(bufs, name);
4929         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4930         lustre_cfg_bufs_set_string(bufs, 2, dev);
4931         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4932
4933         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4934         if (!lcfg)
4935                 GOTO(class_detach, rc = -ENOMEM);
4936         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4937
4938         rc = class_setup(obd, lcfg);
4939         if (rc)
4940                 GOTO(class_detach, rc);
4941
4942         /* connect to MDD we just setup */
4943         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4944         if (rc)
4945                 GOTO(class_detach, rc);
4946
4947         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4948         LASSERT(site);
4949         LASSERT(mdt_lu_site(mdt) == NULL);
4950         mdt->mdt_lu_dev.ld_site = site;
4951         site->ls_top_dev = &mdt->mdt_lu_dev;
4952         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4953
4954         /* now connect to bottom OSD */
4955         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4956         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4957         if (rc)
4958                 GOTO(class_detach, rc);
4959         mdt->mdt_bottom =
4960                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4961
4962         rc = lu_env_refill((struct lu_env *)env);
4963         if (rc != 0)
4964                 CERROR("Failure to refill session: '%d'\n", rc);
4965
4966         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4967
4968         EXIT;
4969 class_detach:
4970         if (rc)
4971                 class_detach(obd, lcfg);
4972 lcfg_cleanup:
4973         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4974 put_profile:
4975         class_put_profile(lprof);
4976 free_bufs:
4977         OBD_FREE_PTR(bufs);
4978 cleanup_mem:
4979         if (name)
4980                 OBD_FREE(name, name_size);
4981         if (uuid)
4982                 OBD_FREE(uuid, uuid_size);
4983         RETURN(rc);
4984 }
4985
4986 /* setup quota master target on MDT0 */
4987 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4988                           struct lustre_cfg *cfg)
4989 {
4990         struct obd_device       *obd;
4991         char                    *dev = lustre_cfg_string(cfg, 0);
4992         char                    *qmtname, *uuid, *p;
4993         struct lustre_cfg_bufs  *bufs;
4994         struct lustre_cfg       *lcfg;
4995         struct lustre_profile   *lprof;
4996         struct obd_connect_data *data;
4997         int                      rc;
4998         ENTRY;
4999
5000         LASSERT(mdt->mdt_qmt_exp == NULL);
5001         LASSERT(mdt->mdt_qmt_dev == NULL);
5002
5003         /* quota master is on MDT0 only for now */
5004         if (mdt->mdt_seq_site.ss_node_id != 0)
5005                 RETURN(0);
5006
5007         /* MGS generates config commands which look as follows:
5008          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5009          *                    3:lustre-MDT0000-mdtlov  4:f
5010          *
5011          * We generate the QMT name from the MDT one, just replacing MD with QM
5012          * after all the preparations, the logical equivalent will be:
5013          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5014          *                    3:lustre-MDT0000-osd  4:f */
5015         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5016         OBD_ALLOC(uuid, UUID_MAX);
5017         OBD_ALLOC_PTR(bufs);
5018         OBD_ALLOC_PTR(data);
5019         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5020                 GOTO(cleanup_mem, rc = -ENOMEM);
5021
5022         strcpy(qmtname, dev);
5023         p = strstr(qmtname, "-MDT");
5024         if (p == NULL)
5025                 GOTO(cleanup_mem, rc = -ENOMEM);
5026         /* replace MD with QM */
5027         p[1] = 'Q';
5028         p[2] = 'M';
5029
5030         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5031
5032         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5033         if (lprof == NULL || lprof->lp_dt == NULL) {
5034                 CERROR("can't find profile for %s\n",
5035                        lustre_cfg_string(cfg, 0));
5036                 GOTO(cleanup_mem, rc = -EINVAL);
5037         }
5038
5039         lustre_cfg_bufs_reset(bufs, qmtname);
5040         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5041         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5042         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5043
5044         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5045         if (!lcfg)
5046                 GOTO(put_profile, rc = -ENOMEM);
5047         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5048
5049         rc = class_attach(lcfg);
5050         if (rc)
5051                 GOTO(lcfg_cleanup, rc);
5052
5053         obd = class_name2obd(qmtname);
5054         if (!obd) {
5055                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5056                        lustre_cfg_string(cfg, 0));
5057                 GOTO(lcfg_cleanup, rc = -EINVAL);
5058         }
5059
5060         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5061
5062         lustre_cfg_bufs_reset(bufs, qmtname);
5063         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5064         lustre_cfg_bufs_set_string(bufs, 2, dev);
5065
5066         /* for quota, the next device should be the OSD device */
5067         lustre_cfg_bufs_set_string(bufs, 3,
5068                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5069
5070         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5071         if (!lcfg)
5072                 GOTO(class_detach, rc = -ENOMEM);
5073         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5074
5075         rc = class_setup(obd, lcfg);
5076         if (rc)
5077                 GOTO(class_detach, rc);
5078
5079         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5080
5081         /* configure local quota objects */
5082         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5083                 rc = -EBADF;
5084         else
5085                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5086                                                            &mdt->mdt_lu_dev,
5087                                                            mdt->mdt_qmt_dev);
5088         if (rc)
5089                 GOTO(class_cleanup, rc);
5090
5091         /* connect to quota master target */
5092         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5093         data->ocd_version = LUSTRE_VERSION_CODE;
5094         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5095                          data, NULL);
5096         if (rc) {
5097                 CERROR("cannot connect to quota master device %s (%d)\n",
5098                        qmtname, rc);
5099                 GOTO(class_cleanup, rc);
5100         }
5101
5102         EXIT;
5103 class_cleanup:
5104         if (rc) {
5105                 class_manual_cleanup(obd);
5106                 mdt->mdt_qmt_dev = NULL;
5107                 GOTO(lcfg_cleanup, rc);
5108         }
5109 class_detach:
5110         if (rc)
5111                 class_detach(obd, lcfg);
5112 lcfg_cleanup:
5113         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5114 put_profile:
5115         class_put_profile(lprof);
5116 cleanup_mem:
5117         if (bufs)
5118                 OBD_FREE_PTR(bufs);
5119         if (qmtname)
5120                 OBD_FREE(qmtname, MAX_OBD_NAME);
5121         if (uuid)
5122                 OBD_FREE(uuid, UUID_MAX);
5123         if (data)
5124                 OBD_FREE_PTR(data);
5125         return rc;
5126 }
5127
5128 /* Shutdown quota master target associated with mdt */
5129 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5130 {
5131         ENTRY;
5132
5133         if (mdt->mdt_qmt_exp == NULL)
5134                 RETURN_EXIT;
5135         LASSERT(mdt->mdt_qmt_dev != NULL);
5136
5137         /* the qmt automatically shuts down when the mdt disconnects */
5138         obd_disconnect(mdt->mdt_qmt_exp);
5139         mdt->mdt_qmt_exp = NULL;
5140         mdt->mdt_qmt_dev = NULL;
5141         EXIT;
5142 }
5143
5144 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5145  * for now. This will be removed along with converting rest of MDT code
5146  * to use tgt_session_info */
5147 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5148 {
5149         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5150         int                      rc;
5151
5152         if (unlikely(info->mti_object == NULL))
5153                 return -EPROTO;
5154
5155         rc = mdt_getxattr(info);
5156
5157         mdt_thread_info_fini(info);
5158         return rc;
5159 }
5160
5161 static int mdt_llog_open(struct tgt_session_info *tsi)
5162 {
5163         ENTRY;
5164
5165         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5166                 RETURN(err_serious(-EACCES));
5167
5168         RETURN(tgt_llog_open(tsi));
5169 }
5170
5171 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5172 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5173 #define OST_BRW_READ    OST_READ
5174 #define OST_BRW_WRITE   OST_WRITE
5175
5176 static struct tgt_handler mdt_tgt_handlers[] = {
5177 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5178                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5179                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5180 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5181                 0,                      MDS_DISCONNECT, tgt_disconnect,
5182                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5183 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5184                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5185                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5186 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5187 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5188 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5189 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5190                                                         mdt_getattr_name),
5191 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5192 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5193 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5194 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5195 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5196 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5197 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5198 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5199                                                         mdt_hsm_progress),
5200 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5201                                                         mdt_hsm_ct_register),
5202 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5203                                                         mdt_hsm_ct_unregister),
5204 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5205                                                         mdt_hsm_state_get),
5206 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5207                                                         mdt_hsm_state_set),
5208 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5209 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5210                                                         mdt_hsm_request),
5211 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5212             MDS_SWAP_LAYOUTS,
5213             mdt_swap_layouts),
5214 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5215 };
5216
5217 static struct tgt_handler mdt_io_ops[] = {
5218 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5219                                                         mdt_hp_brw),
5220 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5221                                                         mdt_hp_brw),
5222 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5223                                          OST_PUNCH,     mdt_punch_hdl,
5224                                                         mdt_hp_punch),
5225 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5226 };
5227
5228 static struct tgt_handler mdt_sec_ctx_ops[] = {
5229 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5230 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5231 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5232 };
5233
5234 static struct tgt_handler mdt_quota_ops[] = {
5235 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5236 };
5237
5238 static struct tgt_handler mdt_llog_handlers[] = {
5239         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5240         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5241         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5242         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5243 };
5244
5245 static struct tgt_opc_slice mdt_common_slice[] = {
5246         {
5247                 .tos_opc_start  = MDS_FIRST_OPC,
5248                 .tos_opc_end    = MDS_LAST_OPC,
5249                 .tos_hs         = mdt_tgt_handlers
5250         },
5251         {
5252                 .tos_opc_start  = OBD_FIRST_OPC,
5253                 .tos_opc_end    = OBD_LAST_OPC,
5254                 .tos_hs         = tgt_obd_handlers
5255         },
5256         {
5257                 .tos_opc_start  = LDLM_FIRST_OPC,
5258                 .tos_opc_end    = LDLM_LAST_OPC,
5259                 .tos_hs         = tgt_dlm_handlers
5260         },
5261         {
5262                 .tos_opc_start  = SEC_FIRST_OPC,
5263                 .tos_opc_end    = SEC_LAST_OPC,
5264                 .tos_hs         = mdt_sec_ctx_ops
5265         },
5266         {
5267                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5268                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5269                 .tos_hs         = tgt_out_handlers
5270         },
5271         {
5272                 .tos_opc_start  = FLD_FIRST_OPC,
5273                 .tos_opc_end    = FLD_LAST_OPC,
5274                 .tos_hs         = fld_handlers
5275         },
5276         {
5277                 .tos_opc_start  = SEQ_FIRST_OPC,
5278                 .tos_opc_end    = SEQ_LAST_OPC,
5279                 .tos_hs         = seq_handlers
5280         },
5281         {
5282                 .tos_opc_start  = QUOTA_DQACQ,
5283                 .tos_opc_end    = QUOTA_LAST_OPC,
5284                 .tos_hs         = mdt_quota_ops
5285         },
5286         {
5287                 .tos_opc_start  = LLOG_FIRST_OPC,
5288                 .tos_opc_end    = LLOG_LAST_OPC,
5289                 .tos_hs         = mdt_llog_handlers
5290         },
5291         {
5292                 .tos_opc_start  = LFSCK_FIRST_OPC,
5293                 .tos_opc_end    = LFSCK_LAST_OPC,
5294                 .tos_hs         = tgt_lfsck_handlers
5295         },
5296         {
5297                 .tos_opc_start  = OST_FIRST_OPC,
5298                 .tos_opc_end    = OST_LAST_OPC,
5299                 .tos_hs         = mdt_io_ops
5300         },
5301         {
5302                 .tos_hs         = NULL
5303         }
5304 };
5305
5306 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5307 {
5308         struct md_device *next = m->mdt_child;
5309         struct lu_device *d = &m->mdt_lu_dev;
5310         struct obd_device *obd = mdt2obd_dev(m);
5311         struct lfsck_stop stop;
5312
5313         ENTRY;
5314         stop.ls_status = LS_PAUSED;
5315         stop.ls_flags = 0;
5316         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5317
5318         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5319         ping_evictor_stop();
5320
5321         /* Remove the HSM /proc entry so the coordinator cannot be
5322          * restarted by a user while it's shutting down.
5323          */
5324         mdt_hsm_cdt_stop(m);
5325
5326         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5327         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5328
5329         if (m->mdt_namespace != NULL)
5330                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5331                                           d->ld_obd->obd_force);
5332
5333         obd_exports_barrier(obd);
5334         obd_zombie_barrier();
5335
5336         mdt_quota_fini(env, m);
5337
5338         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5339
5340         /* Calling the cleanup functions in the same order as in the mdt_init0
5341          * error path
5342          */
5343         mdt_tunables_fini(m);
5344
5345         target_recovery_fini(obd);
5346         upcall_cache_cleanup(m->mdt_identity_cache);
5347         m->mdt_identity_cache = NULL;
5348
5349         mdt_fs_cleanup(env, m);
5350
5351         tgt_fini(env, &m->mdt_lut);
5352
5353         mdt_hsm_cdt_fini(m);
5354
5355         if (m->mdt_los != NULL) {
5356                 local_oid_storage_fini(env, m->mdt_los);
5357                 m->mdt_los = NULL;
5358         }
5359
5360         if (m->mdt_namespace != NULL) {
5361                 ldlm_namespace_free_post(m->mdt_namespace);
5362                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5363         }
5364
5365         if (m->mdt_md_root != NULL) {
5366                 mdt_object_put(env, m->mdt_md_root);
5367                 m->mdt_md_root = NULL;
5368         }
5369
5370         mdt_seq_fini(env, m);
5371
5372         mdt_fld_fini(env, m);
5373
5374         /*
5375          * Finish the stack
5376          */
5377         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5378
5379         LASSERT(atomic_read(&d->ld_ref) == 0);
5380
5381         server_put_mount(mdt_obd_name(m), true);
5382
5383         EXIT;
5384 }
5385
5386 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5387
5388 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5389                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5390 {
5391         const struct dt_device_param *dt_conf;
5392         struct mdt_thread_info *info;
5393         struct obd_device *obd;
5394         const char *dev = lustre_cfg_string(cfg, 0);
5395         const char *num = lustre_cfg_string(cfg, 2);
5396         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5397         struct lustre_mount_info *lmi = NULL;
5398         struct lustre_sb_info *lsi;
5399         struct lu_site *s;
5400         struct seq_server_site *ss_site;
5401         const char *identity_upcall = "NONE";
5402         struct md_device *next;
5403         struct lu_fid fid;
5404         int rc;
5405         long node_id;
5406         mntopt_t mntopts;
5407         ENTRY;
5408
5409         lu_device_init(&m->mdt_lu_dev, ldt);
5410         /*
5411          * Environment (env) might be missing mdt_thread_key values at that
5412          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5413          * mode.
5414          *
5415          * Usually device allocation path doesn't use module key values, but
5416          * mdt has to do a lot of work here, so allocate key value.
5417          */
5418         rc = lu_env_refill((struct lu_env *)env);
5419         if (rc != 0)
5420                 RETURN(rc);
5421
5422         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5423         LASSERT(info != NULL);
5424
5425         obd = class_name2obd(dev);
5426         LASSERT(obd != NULL);
5427
5428         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5429         m->mdt_opts.mo_evict_tgt_nids = 1;
5430         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5431
5432         lmi = server_get_mount(dev);
5433         if (lmi == NULL) {
5434                 CERROR("Cannot get mount info for %s!\n", dev);
5435                 RETURN(-EFAULT);
5436         } else {
5437                 lsi = s2lsi(lmi->lmi_sb);
5438                 LASSERT(lsi->lsi_lmd);
5439                 /* CMD is supported only in IAM mode */
5440                 LASSERT(num);
5441                 node_id = simple_strtol(num, NULL, 10);
5442                 obd->u.obt.obt_magic = OBT_MAGIC;
5443                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5444                         m->mdt_skip_lfsck = 1;
5445         }
5446
5447         /* DoM files get IO lock at open by default */
5448         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5449         /* DoM files are read at open and data is packed in the reply */
5450         m->mdt_opts.mo_dom_read_open = 1;
5451
5452         m->mdt_squash.rsi_uid = 0;
5453         m->mdt_squash.rsi_gid = 0;
5454         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5455         spin_lock_init(&m->mdt_squash.rsi_lock);
5456         spin_lock_init(&m->mdt_lock);
5457         m->mdt_enable_remote_dir = 1;
5458         m->mdt_enable_striped_dir = 1;
5459         m->mdt_enable_dir_migration = 1;
5460         m->mdt_enable_remote_dir_gid = 0;
5461         m->mdt_enable_chprojid_gid = 0;
5462         m->mdt_enable_remote_rename = 1;
5463
5464         atomic_set(&m->mdt_mds_mds_conns, 0);
5465         atomic_set(&m->mdt_async_commit_count, 0);
5466
5467         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5468         m->mdt_lu_dev.ld_obd = obd;
5469         /* Set this lu_device to obd for error handling purposes. */
5470         obd->obd_lu_dev = &m->mdt_lu_dev;
5471
5472         /* init the stack */
5473         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5474         if (rc) {
5475                 CERROR("%s: Can't init device stack, rc %d\n",
5476                        mdt_obd_name(m), rc);
5477                 GOTO(err_lmi, rc);
5478         }
5479
5480         s = mdt_lu_site(m);
5481         ss_site = mdt_seq_site(m);
5482         s->ld_seq_site = ss_site;
5483         ss_site->ss_lu = s;
5484
5485         /* set server index */
5486         ss_site->ss_node_id = node_id;
5487
5488         /* failover is the default
5489          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5490          * assumed whose ss_node_id == 0 XXX
5491          * */
5492         obd->obd_replayable = 1;
5493         /* No connection accepted until configurations will finish */
5494         obd->obd_no_conn = 1;
5495
5496         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5497                 char *str = lustre_cfg_string(cfg, 4);
5498                 if (strchr(str, 'n')) {
5499                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5500                         obd->obd_replayable = 0;
5501                 }
5502         }
5503
5504         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5505         if (rc)
5506                 GOTO(err_fini_stack, rc);
5507
5508         rc = mdt_seq_init(env, m);
5509         if (rc)
5510                 GOTO(err_fini_fld, rc);
5511
5512         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5513                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5514         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5515                                               LDLM_NAMESPACE_SERVER,
5516                                               LDLM_NAMESPACE_GREEDY,
5517                                               LDLM_NS_TYPE_MDT);
5518         if (m->mdt_namespace == NULL)
5519                 GOTO(err_fini_seq, rc = -ENOMEM);
5520
5521         m->mdt_namespace->ns_lvbp = m;
5522         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5523
5524         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5525         /* set obd_namespace for compatibility with old code */
5526         obd->obd_namespace = m->mdt_namespace;
5527
5528         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5529                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5530                       OBD_FAIL_MDS_ALL_REPLY_NET);
5531         if (rc)
5532                 GOTO(err_free_ns, rc);
5533
5534         /* Amount of available space excluded from granting and reserved
5535          * for metadata. It is a percentage of the total MDT size. */
5536         tgd->tgd_reserved_pcnt = 10;
5537
5538         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5539                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5540         else
5541                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5542
5543         rc = mdt_fs_setup(env, m, obd, lsi);
5544         if (rc)
5545                 GOTO(err_tgt, rc);
5546
5547         fid.f_seq = FID_SEQ_LOCAL_NAME;
5548         fid.f_oid = 1;
5549         fid.f_ver = 0;
5550         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5551         if (rc != 0)
5552                 GOTO(err_fs_cleanup, rc);
5553
5554         rc = mdt_hsm_cdt_init(m);
5555         if (rc != 0) {
5556                 CERROR("%s: error initializing coordinator, rc %d\n",
5557                        mdt_obd_name(m), rc);
5558                 GOTO(err_los_fini, rc);
5559         }
5560
5561         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5562
5563         next = m->mdt_child;
5564         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5565
5566         mntopts = dt_conf->ddp_mntopts;
5567
5568         if (mntopts & MNTOPT_USERXATTR)
5569                 m->mdt_opts.mo_user_xattr = 1;
5570         else
5571                 m->mdt_opts.mo_user_xattr = 0;
5572
5573         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5574
5575         if (mntopts & MNTOPT_ACL)
5576                 m->mdt_opts.mo_acl = 1;
5577         else
5578                 m->mdt_opts.mo_acl = 0;
5579
5580         /* XXX: to support suppgid for ACL, we enable identity_upcall
5581          * by default, otherwise, maybe got unexpected -EACCESS. */
5582         if (m->mdt_opts.mo_acl)
5583                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5584
5585         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5586                                                 identity_upcall,
5587                                                 &mdt_identity_upcall_cache_ops);
5588         if (IS_ERR(m->mdt_identity_cache)) {
5589                 rc = PTR_ERR(m->mdt_identity_cache);
5590                 m->mdt_identity_cache = NULL;
5591                 GOTO(err_free_hsm, rc);
5592         }
5593
5594         rc = mdt_tunables_init(m, dev);
5595         if (rc) {
5596                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5597                 GOTO(err_recovery, rc);
5598         }
5599
5600         rc = mdt_quota_init(env, m, cfg);
5601         if (rc)
5602                 GOTO(err_procfs, rc);
5603
5604         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5605         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5606                            "mdt_ldlm_client", m->mdt_ldlm_client);
5607
5608         ping_evictor_start();
5609
5610         /* recovery will be started upon mdt_prepare()
5611          * when the whole stack is complete and ready
5612          * to serve the requests */
5613
5614         /* Reduce the initial timeout on an MDS because it doesn't need such
5615          * a long timeout as an OST does. Adaptive timeouts will adjust this
5616          * value appropriately. */
5617         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5618                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5619
5620         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
5621                 m->mdt_lut.lut_local_recovery = 1;
5622
5623         RETURN(0);
5624 err_procfs:
5625         mdt_tunables_fini(m);
5626 err_recovery:
5627         upcall_cache_cleanup(m->mdt_identity_cache);
5628         m->mdt_identity_cache = NULL;
5629 err_free_hsm:
5630         mdt_hsm_cdt_fini(m);
5631 err_los_fini:
5632         local_oid_storage_fini(env, m->mdt_los);
5633         m->mdt_los = NULL;
5634 err_fs_cleanup:
5635         mdt_fs_cleanup(env, m);
5636 err_tgt:
5637         /* keep recoverable clients */
5638         obd->obd_fail = 1;
5639         target_recovery_fini(obd);
5640         obd_exports_barrier(obd);
5641         obd_zombie_barrier();
5642         tgt_fini(env, &m->mdt_lut);
5643 err_free_ns:
5644         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5645         obd->obd_namespace = m->mdt_namespace = NULL;
5646 err_fini_seq:
5647         mdt_seq_fini(env, m);
5648 err_fini_fld:
5649         mdt_fld_fini(env, m);
5650 err_fini_stack:
5651         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5652 err_lmi:
5653         if (lmi)
5654                 server_put_mount(dev, true);
5655         return(rc);
5656 }
5657
5658 /* For interoperability, the left element is old parameter, the right one
5659  * is the new version of the parameter, if some parameter is deprecated,
5660  * the new version should be set as NULL. */
5661 static struct cfg_interop_param mdt_interop_param[] = {
5662         { "mdt.group_upcall",   NULL },
5663         { "mdt.quota_type",     NULL },
5664         { "mdd.quota_type",     NULL },
5665         { "mdt.som",            NULL },
5666         { "mdt.rootsquash",     "mdt.root_squash" },
5667         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5668         { NULL }
5669 };
5670
5671 /* used by MGS to process specific configurations */
5672 static int mdt_process_config(const struct lu_env *env,
5673                               struct lu_device *d, struct lustre_cfg *cfg)
5674 {
5675         struct mdt_device *m = mdt_dev(d);
5676         struct md_device *md_next = m->mdt_child;
5677         struct lu_device *next = md2lu_dev(md_next);
5678         int rc;
5679         ENTRY;
5680
5681         switch (cfg->lcfg_command) {
5682         case LCFG_PARAM: {
5683                 struct obd_device *obd = d->ld_obd;
5684                 /* For interoperability */
5685                 struct cfg_interop_param *ptr = NULL;
5686                 struct lustre_cfg *old_cfg = NULL;
5687                 char *param = NULL;
5688                 ssize_t count;
5689
5690                 param = lustre_cfg_string(cfg, 1);
5691                 if (param == NULL) {
5692                         CERROR("param is empty\n");
5693                         rc = -EINVAL;
5694                         break;
5695                 }
5696
5697                 ptr = class_find_old_param(param, mdt_interop_param);
5698                 if (ptr != NULL) {
5699                         if (ptr->new_param == NULL) {
5700                                 rc = 0;
5701                                 CWARN("For interoperability, skip this %s."
5702                                       " It is obsolete.\n", ptr->old_param);
5703                                 break;
5704                         }
5705
5706                         CWARN("Found old param %s, changed it to %s.\n",
5707                               ptr->old_param, ptr->new_param);
5708
5709                         old_cfg = cfg;
5710                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5711                         if (IS_ERR(cfg)) {
5712                                 rc = PTR_ERR(cfg);
5713                                 break;
5714                         }
5715                 }
5716
5717                 count = class_modify_config(cfg, PARAM_MDT,
5718                                             &obd->obd_kset.kobj);
5719                 if (count < 0) {
5720                         struct coordinator *cdt = &m->mdt_coordinator;
5721
5722                         /* is it an HSM var ? */
5723                         count = class_modify_config(cfg, PARAM_HSM,
5724                                                     &cdt->cdt_hsm_kobj);
5725                         if (count < 0)
5726                                 /* we don't understand; pass it on */
5727                                 rc = next->ld_ops->ldo_process_config(env, next,
5728                                                                       cfg);
5729                         else
5730                                 rc = count > 0 ? 0 : count;
5731                 } else {
5732                         rc = count > 0 ? 0 : count;
5733                 }
5734
5735                 if (old_cfg)
5736                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5737                                                      cfg->lcfg_buflens));
5738                 break;
5739         }
5740         default:
5741                 /* others are passed further */
5742                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5743                 break;
5744         }
5745         RETURN(rc);
5746 }
5747
5748 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5749                                           const struct lu_object_header *hdr,
5750                                           struct lu_device *d)
5751 {
5752         struct mdt_object *mo;
5753
5754         ENTRY;
5755
5756         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5757         if (mo != NULL) {
5758                 struct lu_object *o;
5759                 struct lu_object_header *h;
5760
5761                 o = &mo->mot_obj;
5762                 h = &mo->mot_header;
5763                 lu_object_header_init(h);
5764                 lu_object_init(o, h, d);
5765                 lu_object_add_top(h, o);
5766                 o->lo_ops = &mdt_obj_ops;
5767                 spin_lock_init(&mo->mot_write_lock);
5768                 mutex_init(&mo->mot_som_mutex);
5769                 mutex_init(&mo->mot_lov_mutex);
5770                 init_rwsem(&mo->mot_dom_sem);
5771                 init_rwsem(&mo->mot_open_sem);
5772                 atomic_set(&mo->mot_open_count, 0);
5773                 RETURN(o);
5774         }
5775         RETURN(NULL);
5776 }
5777
5778 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5779                            const struct lu_object_conf *unused)
5780 {
5781         struct mdt_device *d = mdt_dev(o->lo_dev);
5782         struct lu_device  *under;
5783         struct lu_object  *below;
5784         int                rc = 0;
5785         ENTRY;
5786
5787         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5788                PFID(lu_object_fid(o)));
5789
5790         under = &d->mdt_child->md_lu_dev;
5791         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5792         if (below != NULL) {
5793                 lu_object_add(o, below);
5794         } else
5795                 rc = -ENOMEM;
5796
5797         RETURN(rc);
5798 }
5799
5800 static void mdt_object_free_rcu(struct rcu_head *head)
5801 {
5802         struct mdt_object *mo = container_of(head, struct mdt_object,
5803                                              mot_header.loh_rcu);
5804
5805         kmem_cache_free(mdt_object_kmem, mo);
5806 }
5807
5808 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5809 {
5810         struct mdt_object *mo = mdt_obj(o);
5811         struct lu_object_header *h;
5812         ENTRY;
5813
5814         h = o->lo_header;
5815         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5816                PFID(lu_object_fid(o)));
5817
5818         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5819         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5820
5821         lu_object_fini(o);
5822         lu_object_header_fini(h);
5823         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
5824         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
5825
5826         EXIT;
5827 }
5828
5829 static int mdt_object_print(const struct lu_env *env, void *cookie,
5830                             lu_printer_t p, const struct lu_object *o)
5831 {
5832         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5833
5834         return (*p)(env, cookie,
5835                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5836                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5837                     mdto->mot_cache_attr ? "cache_attr" : "",
5838                     mdto->mot_write_count);
5839 }
5840
5841 static int mdt_prepare(const struct lu_env *env,
5842                 struct lu_device *pdev,
5843                 struct lu_device *cdev)
5844 {
5845         struct mdt_device *mdt = mdt_dev(cdev);
5846         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5847         struct obd_device *obd = cdev->ld_obd;
5848         int rc;
5849
5850         ENTRY;
5851
5852         LASSERT(obd);
5853
5854         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5855         if (rc)
5856                 RETURN(rc);
5857
5858         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5859         if (rc)
5860                 RETURN(rc);
5861
5862         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5863         if (rc)
5864                 RETURN(rc);
5865
5866         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5867         /* The LFSCK instance is registered just now, so it must be there when
5868          * register the namespace to such instance. */
5869         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5870
5871         if (mdt->mdt_seq_site.ss_node_id == 0) {
5872                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5873                                                          &mdt->mdt_md_root_fid);
5874                 if (rc)
5875                         RETURN(rc);
5876         }
5877
5878         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5879
5880         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5881         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5882         LASSERT(obd->obd_no_conn);
5883         spin_lock(&obd->obd_dev_lock);
5884         obd->obd_no_conn = 0;
5885         spin_unlock(&obd->obd_dev_lock);
5886
5887         if (obd->obd_recovering == 0)
5888                 mdt_postrecov(env, mdt);
5889
5890         RETURN(rc);
5891 }
5892
5893 const struct lu_device_operations mdt_lu_ops = {
5894         .ldo_object_alloc   = mdt_object_alloc,
5895         .ldo_process_config = mdt_process_config,
5896         .ldo_prepare        = mdt_prepare,
5897 };
5898
5899 static const struct lu_object_operations mdt_obj_ops = {
5900         .loo_object_init    = mdt_object_init,
5901         .loo_object_free    = mdt_object_free,
5902         .loo_object_print   = mdt_object_print
5903 };
5904
5905 static int mdt_obd_set_info_async(const struct lu_env *env,
5906                                   struct obd_export *exp,
5907                                   __u32 keylen, void *key,
5908                                   __u32 vallen, void *val,
5909                                   struct ptlrpc_request_set *set)
5910 {
5911         int rc;
5912
5913         ENTRY;
5914
5915         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5916                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5917                 RETURN(rc);
5918         }
5919
5920         RETURN(0);
5921 }
5922
5923 /**
5924  * Match client and server connection feature flags.
5925  *
5926  * Compute the compatibility flags for a connection request based on
5927  * features mutually supported by client and server.
5928  *
5929  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5930  * must not be updated here, otherwise a partially initialized value may
5931  * be exposed. After the connection request is successfully processed,
5932  * the top-level MDT connect request handler atomically updates the export
5933  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5934  * reply. \see mdt_connect().
5935  *
5936  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5937  * full struct obd_connect_data. So care must be taken when accessing fields
5938  * that are not present in struct obd_connect_data_v1. See LU-16.
5939  *
5940  * \param exp   the obd_export associated with this client/target pair
5941  * \param mdt   the target device for the connection
5942  * \param data  stores data for this connect request
5943  *
5944  * \retval 0       success
5945  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5946  * \retval -EBADE  client and server feature requirements are incompatible
5947  */
5948 static int mdt_connect_internal(const struct lu_env *env,
5949                                 struct obd_export *exp,
5950                                 struct mdt_device *mdt,
5951                                 struct obd_connect_data *data, bool reconnect)
5952 {
5953         const char *obd_name = mdt_obd_name(mdt);
5954         LASSERT(data != NULL);
5955
5956         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5957
5958         if (mdt->mdt_bottom->dd_rdonly &&
5959             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5960             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5961                 RETURN(-EACCES);
5962
5963         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5964                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5965
5966         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5967
5968         if (!mdt->mdt_opts.mo_acl)
5969                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5970
5971         if (!mdt->mdt_opts.mo_user_xattr)
5972                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5973
5974         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5975                 data->ocd_brw_size = min(data->ocd_brw_size,
5976                                          mdt->mdt_brw_size);
5977                 if (data->ocd_brw_size == 0) {
5978                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5979                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5980                                "ocd_brw_size unexpectedly zero, network data "
5981                                "corruption? Refusing to connect this client\n",
5982                                obd_name, exp->exp_client_uuid.uuid,
5983                                exp, data->ocd_connect_flags, data->ocd_version,
5984                                data->ocd_grant, data->ocd_index);
5985                         return -EPROTO;
5986                 }
5987         }
5988
5989         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5990                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5991
5992                 /* client is reporting its page size, for future use */
5993                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5994                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5995                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5996                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5997                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5998                 /* ocd_grant_tax_kb is in 1K byte blocks */
5999                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6000                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6001         }
6002
6003         /* Save connect_data we have so far because tgt_grant_connect()
6004          * uses it to calculate grant, and we want to save the client
6005          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6006         exp->exp_connect_data = *data;
6007         if (OCD_HAS_FLAG(data, GRANT))
6008                 tgt_grant_connect(env, exp, data, !reconnect);
6009
6010         if (OCD_HAS_FLAG(data, MAXBYTES))
6011                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6012
6013         /* NB: Disregard the rule against updating
6014          * exp_connect_data.ocd_connect_flags in this case, since
6015          * tgt_client_new() needs to know if this is a lightweight
6016          * connection, and it is safe to expose this flag before
6017          * connection processing completes. */
6018         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6019                 spin_lock(&exp->exp_lock);
6020                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6021                 spin_unlock(&exp->exp_lock);
6022         }
6023
6024         data->ocd_version = LUSTRE_VERSION_CODE;
6025
6026         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6027                 CWARN("%s: MDS requires FID support, but client not\n",
6028                       obd_name);
6029                 return -EBADE;
6030         }
6031
6032         if (OCD_HAS_FLAG(data, PINGLESS)) {
6033                 if (ptlrpc_pinger_suppress_pings()) {
6034                         spin_lock(&exp->exp_obd->obd_dev_lock);
6035                         list_del_init(&exp->exp_obd_chain_timed);
6036                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6037                 } else {
6038                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6039                 }
6040         }
6041
6042         data->ocd_max_easize = mdt->mdt_max_ea_size;
6043
6044         /* NB: Disregard the rule against updating
6045          * exp_connect_data.ocd_connect_flags in this case, since
6046          * tgt_client_new() needs to know if this is client supports
6047          * multiple modify RPCs, and it is safe to expose this flag before
6048          * connection processing completes. */
6049         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6050                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6051                 spin_lock(&exp->exp_lock);
6052                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6053                 spin_unlock(&exp->exp_lock);
6054         }
6055
6056         if (OCD_HAS_FLAG(data, CKSUM)) {
6057                 __u32 cksum_types = data->ocd_cksum_types;
6058
6059                 /* The client set in ocd_cksum_types the checksum types it
6060                  * supports. We have to mask off the algorithms that we don't
6061                  * support */
6062                 data->ocd_cksum_types &=
6063                         obd_cksum_types_supported_server(obd_name);
6064
6065                 if (unlikely(data->ocd_cksum_types == 0)) {
6066                         CERROR("%s: Connect with checksum support but no "
6067                                "ocd_cksum_types is set\n",
6068                                exp->exp_obd->obd_name);
6069                         RETURN(-EPROTO);
6070                 }
6071
6072                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6073                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6074                        cksum_types, data->ocd_cksum_types);
6075         } else {
6076                 /* This client does not support OBD_CONNECT_CKSUM
6077                  * fall back to CRC32 */
6078                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6079                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6080                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6081         }
6082
6083         return 0;
6084 }
6085
6086 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6087                                    struct mdt_thread_info *info,
6088                                    struct mdt_file_data *mfd)
6089 {
6090         struct lu_context ses;
6091         int rc;
6092         ENTRY;
6093
6094         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6095         if (rc)
6096                 RETURN(rc);
6097
6098         env->le_ses = &ses;
6099         lu_context_enter(&ses);
6100
6101         mdt_ucred(info)->uc_valid = UCRED_OLD;
6102         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6103
6104         lu_context_exit(&ses);
6105         lu_context_fini(&ses);
6106         env->le_ses = NULL;
6107
6108         RETURN(rc);
6109 }
6110
6111 static int mdt_export_cleanup(struct obd_export *exp)
6112 {
6113         LIST_HEAD(closing_list);
6114         struct mdt_export_data  *med = &exp->exp_mdt_data;
6115         struct obd_device       *obd = exp->exp_obd;
6116         struct mdt_device       *mdt;
6117         struct mdt_thread_info  *info;
6118         struct lu_env            env;
6119         struct mdt_file_data    *mfd, *n;
6120         int rc = 0;
6121         ENTRY;
6122
6123         spin_lock(&med->med_open_lock);
6124         while (!list_empty(&med->med_open_head)) {
6125                 struct list_head *tmp = med->med_open_head.next;
6126                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6127
6128                 /* Remove mfd handle so it can't be found again.
6129                  * We are consuming the mfd_list reference here. */
6130                 class_handle_unhash(&mfd->mfd_open_handle);
6131                 list_move_tail(&mfd->mfd_list, &closing_list);
6132         }
6133         spin_unlock(&med->med_open_lock);
6134         mdt = mdt_dev(obd->obd_lu_dev);
6135         LASSERT(mdt != NULL);
6136
6137         rc = lu_env_init(&env, LCT_MD_THREAD);
6138         if (rc)
6139                 RETURN(rc);
6140
6141         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6142         LASSERT(info != NULL);
6143         memset(info, 0, sizeof *info);
6144         info->mti_env = &env;
6145         info->mti_mdt = mdt;
6146         info->mti_exp = exp;
6147
6148         if (!list_empty(&closing_list)) {
6149                 struct md_attr *ma = &info->mti_attr;
6150
6151                 /* Close any open files (which may also cause orphan
6152                  * unlinking). */
6153                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6154                         list_del_init(&mfd->mfd_list);
6155                         ma->ma_need = ma->ma_valid = 0;
6156
6157                         /* This file is being closed due to an eviction, it
6158                          * could have been modified and now dirty regarding to
6159                          * HSM archive, check this!
6160                          * The logic here is to mark a file dirty if there's a
6161                          * chance it was dirtied before the client was evicted,
6162                          * so that we don't have to wait for a release attempt
6163                          * before finding out the file was actually dirty and
6164                          * fail the release. Aggressively marking it dirty here
6165                          * will cause the policy engine to attempt to
6166                          * re-archive it; when rearchiving, we can compare the
6167                          * current version to the HSM data_version and make the
6168                          * archive request into a noop if it's not actually
6169                          * dirty.
6170                          */
6171                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6172                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6173
6174                         /* Don't unlink orphan on failover umount, LU-184 */
6175                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6176                                 ma->ma_valid = MA_FLAGS;
6177                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6178                         }
6179                         ma->ma_valid |= MA_FORCE_LOG;
6180                         mdt_mfd_close(info, mfd);
6181                 }
6182         }
6183         info->mti_mdt = NULL;
6184         /* cleanup client slot early */
6185         /* Do not erase record for recoverable client. */
6186         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6187                 tgt_client_del(&env, exp);
6188         lu_env_fini(&env);
6189
6190         RETURN(rc);
6191 }
6192
6193 static inline void mdt_enable_slc(struct mdt_device *mdt)
6194 {
6195         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6196                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6197 }
6198
6199 static inline void mdt_disable_slc(struct mdt_device *mdt)
6200 {
6201         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6202                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6203 }
6204
6205 static int mdt_obd_disconnect(struct obd_export *exp)
6206 {
6207         int rc;
6208
6209         ENTRY;
6210
6211         LASSERT(exp);
6212         class_export_get(exp);
6213
6214         if (!(exp->exp_flags & OBD_OPT_FORCE))
6215                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6216
6217         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6218             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6219                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6220
6221                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6222                         mdt_disable_slc(mdt);
6223         }
6224
6225         rc = server_disconnect_export(exp);
6226         if (rc != 0)
6227                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6228
6229         tgt_grant_discard(exp);
6230
6231         rc = mdt_export_cleanup(exp);
6232         nodemap_del_member(exp);
6233         class_export_put(exp);
6234         RETURN(rc);
6235 }
6236
6237 /* mds_connect copy */
6238 static int mdt_obd_connect(const struct lu_env *env,
6239                            struct obd_export **exp, struct obd_device *obd,
6240                            struct obd_uuid *cluuid,
6241                            struct obd_connect_data *data,
6242                            void *localdata)
6243 {
6244         struct obd_export       *lexp;
6245         struct lustre_handle    conn = { 0 };
6246         struct mdt_device       *mdt;
6247         int                      rc;
6248         lnet_nid_t              *client_nid = localdata;
6249         ENTRY;
6250
6251         LASSERT(env != NULL);
6252         LASSERT(data != NULL);
6253
6254         if (!exp || !obd || !cluuid)
6255                 RETURN(-EINVAL);
6256
6257         mdt = mdt_dev(obd->obd_lu_dev);
6258
6259         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6260             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6261                 atomic_inc(&mdt->mdt_mds_mds_conns);
6262                 mdt_enable_slc(mdt);
6263         }
6264
6265         /*
6266          * first, check whether the stack is ready to handle requests
6267          * XXX: probably not very appropriate method is used now
6268          *      at some point we should find a better one
6269          */
6270         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6271             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6272             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6273                 rc = obd_get_info(env, mdt->mdt_child_exp,
6274                                   sizeof(KEY_OSP_CONNECTED),
6275                                   KEY_OSP_CONNECTED, NULL, NULL);
6276                 if (rc)
6277                         RETURN(-EAGAIN);
6278                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6279         }
6280
6281         rc = class_connect(&conn, obd, cluuid);
6282         if (rc)
6283                 RETURN(rc);
6284
6285         lexp = class_conn2export(&conn);
6286         LASSERT(lexp != NULL);
6287
6288         rc = nodemap_add_member(*client_nid, lexp);
6289         if (rc != 0 && rc != -EEXIST)
6290                 GOTO(out, rc);
6291
6292         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6293         if (rc == 0) {
6294                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6295
6296                 LASSERT(lcd);
6297                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6298                 rc = tgt_client_new(env, lexp);
6299                 if (rc == 0)
6300                         mdt_export_stats_init(obd, lexp, localdata);
6301         }
6302 out:
6303         if (rc != 0) {
6304                 class_disconnect(lexp);
6305                 nodemap_del_member(lexp);
6306                 *exp = NULL;
6307         } else {
6308                 *exp = lexp;
6309                 /* Because we do not want this export to be evicted by pinger,
6310                  * let's not add this export to the timed chain list. */
6311                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6312                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6313                         list_del_init(&lexp->exp_obd_chain_timed);
6314                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6315                 }
6316         }
6317
6318         RETURN(rc);
6319 }
6320
6321 static int mdt_obd_reconnect(const struct lu_env *env,
6322                              struct obd_export *exp, struct obd_device *obd,
6323                              struct obd_uuid *cluuid,
6324                              struct obd_connect_data *data,
6325                              void *localdata)
6326 {
6327         lnet_nid_t             *client_nid = localdata;
6328         int                     rc;
6329         ENTRY;
6330
6331         if (exp == NULL || obd == NULL || cluuid == NULL)
6332                 RETURN(-EINVAL);
6333
6334         rc = nodemap_add_member(*client_nid, exp);
6335         if (rc != 0 && rc != -EEXIST)
6336                 RETURN(rc);
6337
6338         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6339                                   true);
6340         if (rc == 0)
6341                 mdt_export_stats_init(obd, exp, localdata);
6342         else
6343                 nodemap_del_member(exp);
6344
6345         RETURN(rc);
6346 }
6347
6348 /* FIXME: Can we avoid using these two interfaces? */
6349 static int mdt_init_export(struct obd_export *exp)
6350 {
6351         struct mdt_export_data *med = &exp->exp_mdt_data;
6352         int                     rc;
6353         ENTRY;
6354
6355         INIT_LIST_HEAD(&med->med_open_head);
6356         spin_lock_init(&med->med_open_lock);
6357         spin_lock(&exp->exp_lock);
6358         exp->exp_connecting = 1;
6359         spin_unlock(&exp->exp_lock);
6360
6361         OBD_ALLOC(exp->exp_used_slots,
6362                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6363         if (exp->exp_used_slots == NULL)
6364                 RETURN(-ENOMEM);
6365
6366         /* self-export doesn't need client data and ldlm initialization */
6367         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6368                                      &exp->exp_client_uuid)))
6369                 RETURN(0);
6370
6371         rc = tgt_client_alloc(exp);
6372         if (rc)
6373                 GOTO(err, rc);
6374
6375         rc = ldlm_init_export(exp);
6376         if (rc)
6377                 GOTO(err_free, rc);
6378
6379         RETURN(rc);
6380
6381 err_free:
6382         tgt_client_free(exp);
6383 err:
6384         OBD_FREE(exp->exp_used_slots,
6385                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6386         exp->exp_used_slots = NULL;
6387
6388         CERROR("%s: Failed to initialize export: rc = %d\n",
6389                exp->exp_obd->obd_name, rc);
6390         return rc;
6391 }
6392
6393 static int mdt_destroy_export(struct obd_export *exp)
6394 {
6395         ENTRY;
6396
6397         target_destroy_export(exp);
6398         if (exp->exp_used_slots)
6399                 OBD_FREE(exp->exp_used_slots,
6400                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6401
6402         /* destroy can be called from failed obd_setup, so
6403          * checking uuid is safer than obd_self_export */
6404         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6405                                      &exp->exp_client_uuid)))
6406                 RETURN(0);
6407
6408         ldlm_destroy_export(exp);
6409         tgt_client_free(exp);
6410
6411         LASSERT(list_empty(&exp->exp_outstanding_replies));
6412         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6413
6414         /*
6415          * discard grants once we're sure no more
6416          * interaction with the client is possible
6417          */
6418         tgt_grant_discard(exp);
6419         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6420                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6421
6422         if (!(exp->exp_flags & OBD_OPT_FORCE))
6423                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6424
6425         RETURN(0);
6426 }
6427
6428 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6429                    struct linkea_data *ldata)
6430 {
6431         int rc;
6432
6433         LASSERT(ldata->ld_buf->lb_buf != NULL);
6434
6435         if (!mdt_object_exists(mdt_obj))
6436                 return -ENODATA;
6437
6438         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6439                           ldata->ld_buf, XATTR_NAME_LINK);
6440         if (rc == -ERANGE) {
6441                 /* Buf was too small, figure out what we need. */
6442                 lu_buf_free(ldata->ld_buf);
6443                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6444                                   ldata->ld_buf, XATTR_NAME_LINK);
6445                 if (rc < 0)
6446                         return rc;
6447                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6448                 if (ldata->ld_buf->lb_buf == NULL)
6449                         return -ENOMEM;
6450                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6451                                   ldata->ld_buf, XATTR_NAME_LINK);
6452         }
6453         if (rc < 0)
6454                 return rc;
6455
6456         return linkea_init_with_rec(ldata);
6457 }
6458
6459 /**
6460  * Given an MDT object, try to look up the full path to the object.
6461  * Part of the MDT layer implementation of lfs fid2path.
6462  *
6463  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6464  * \param[in]     obj   Object to do path lookup of
6465  * \param[in,out] fp    User-provided struct to store path information
6466  * \param[in]     root_fid Root FID of current path should reach
6467  *
6468  * \retval 0 Lookup successful, path information stored in fp
6469  * \retval -EAGAIN Lookup failed, usually because object is being moved
6470  * \retval negative errno if there was a problem
6471  */
6472 static int mdt_path_current(struct mdt_thread_info *info,
6473                             struct mdt_object *obj,
6474                             struct getinfo_fid2path *fp,
6475                             struct lu_fid *root_fid)
6476 {
6477         struct mdt_device       *mdt = info->mti_mdt;
6478         struct mdt_object       *mdt_obj;
6479         struct link_ea_header   *leh;
6480         struct link_ea_entry    *lee;
6481         struct lu_name          *tmpname = &info->mti_name;
6482         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6483         struct lu_buf           *buf = &info->mti_big_buf;
6484         char                    *ptr;
6485         int                     reclen;
6486         struct linkea_data      ldata = { NULL };
6487         int                     rc = 0;
6488         bool                    first = true;
6489         ENTRY;
6490
6491         /* temp buffer for path element, the buffer will be finally freed
6492          * in mdt_thread_info_fini */
6493         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6494         if (buf->lb_buf == NULL)
6495                 RETURN(-ENOMEM);
6496
6497         ldata.ld_buf = buf;
6498         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6499         *ptr = 0;
6500         --ptr;
6501         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6502
6503         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6504                 struct lu_buf           lmv_buf;
6505
6506                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6507                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6508                         GOTO(out, rc = -ENOENT);
6509
6510                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6511                         mdt_obj = obj;
6512                         mdt_object_get(info->mti_env, mdt_obj);
6513                 } else {
6514                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6515                         if (IS_ERR(mdt_obj))
6516                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6517                 }
6518
6519                 if (!mdt_object_exists(mdt_obj)) {
6520                         mdt_object_put(info->mti_env, mdt_obj);
6521                         GOTO(out, rc = -ENOENT);
6522                 }
6523
6524                 if (mdt_object_remote(mdt_obj)) {
6525                         mdt_object_put(info->mti_env, mdt_obj);
6526                         GOTO(remote_out, rc = -EREMOTE);
6527                 }
6528
6529                 rc = mdt_links_read(info, mdt_obj, &ldata);
6530                 if (rc != 0) {
6531                         mdt_object_put(info->mti_env, mdt_obj);
6532                         GOTO(out, rc);
6533                 }
6534
6535                 leh = buf->lb_buf;
6536                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6537                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6538                 /* If set, use link #linkno for path lookup, otherwise use
6539                    link #0.  Only do this for the final path element. */
6540                 if (first && fp->gf_linkno < leh->leh_reccount) {
6541                         int count;
6542                         for (count = 0; count < fp->gf_linkno; count++) {
6543                                 lee = (struct link_ea_entry *)
6544                                      ((char *)lee + reclen);
6545                                 linkea_entry_unpack(lee, &reclen, tmpname,
6546                                                     tmpfid);
6547                         }
6548                         if (fp->gf_linkno < leh->leh_reccount - 1)
6549                                 /* indicate to user there are more links */
6550                                 fp->gf_linkno++;
6551                 }
6552
6553                 lmv_buf.lb_buf = info->mti_xattr_buf;
6554                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6555                 /* Check if it is slave stripes */
6556                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6557                                   &lmv_buf, XATTR_NAME_LMV);
6558                 mdt_object_put(info->mti_env, mdt_obj);
6559                 if (rc > 0) {
6560                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6561
6562                         /* For slave stripes, get its master */
6563                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6564                                 fp->gf_fid = *tmpfid;
6565                                 continue;
6566                         }
6567                 } else if (rc < 0 && rc != -ENODATA) {
6568                         GOTO(out, rc);
6569                 }
6570
6571                 rc = 0;
6572
6573                 /* Pack the name in the end of the buffer */
6574                 ptr -= tmpname->ln_namelen;
6575                 if (ptr - 1 <= fp->gf_u.gf_path)
6576                         GOTO(out, rc = -EOVERFLOW);
6577                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6578                 *(--ptr) = '/';
6579
6580                 /* keep the last resolved fid to the client, so the
6581                  * client will build the left path on another MDT for
6582                  * remote object */
6583                 fp->gf_fid = *tmpfid;
6584
6585                 first = false;
6586         }
6587
6588 remote_out:
6589         ptr++; /* skip leading / */
6590         memmove(fp->gf_u.gf_path, ptr,
6591                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6592
6593 out:
6594         RETURN(rc);
6595 }
6596
6597 /**
6598  * Given an MDT object, use mdt_path_current to get the path.
6599  * Essentially a wrapper to retry mdt_path_current a set number of times
6600  * if -EAGAIN is returned (usually because an object is being moved).
6601  *
6602  * Part of the MDT layer implementation of lfs fid2path.
6603  *
6604  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6605  * \param[in]     obj   Object to do path lookup of
6606  * \param[in,out] fp    User-provided struct for arguments and to store path
6607  *                      information
6608  *
6609  * \retval 0 Lookup successful, path information stored in fp
6610  * \retval negative errno if there was a problem
6611  */
6612 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6613                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6614 {
6615         struct mdt_device       *mdt = info->mti_mdt;
6616         int                     tries = 3;
6617         int                     rc = -EAGAIN;
6618         ENTRY;
6619
6620         if (fp->gf_pathlen < 3)
6621                 RETURN(-EOVERFLOW);
6622
6623         if (root_fid == NULL)
6624                 root_fid = &mdt->mdt_md_root_fid;
6625
6626         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6627                 fp->gf_u.gf_path[0] = '\0';
6628                 RETURN(0);
6629         }
6630
6631         /* Retry multiple times in case file is being moved */
6632         while (tries-- && rc == -EAGAIN)
6633                 rc = mdt_path_current(info, obj, fp, root_fid);
6634
6635         RETURN(rc);
6636 }
6637
6638 /**
6639  * Get the full path of the provided FID, as of changelog record recno.
6640  *
6641  * This checks sanity and looks up object for user provided FID
6642  * before calling the actual path lookup code.
6643  *
6644  * Part of the MDT layer implementation of lfs fid2path.
6645  *
6646  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6647  * \param[in,out] fp    User-provided struct for arguments and to store path
6648  *                      information
6649  *
6650  * \retval 0 Lookup successful, path information and recno stored in fp
6651  * \retval -ENOENT, object does not exist
6652  * \retval negative errno if there was a problem
6653  */
6654 static int mdt_fid2path(struct mdt_thread_info *info,
6655                         struct lu_fid *root_fid,
6656                         struct getinfo_fid2path *fp)
6657 {
6658         struct mdt_device *mdt = info->mti_mdt;
6659         struct mdt_object *obj;
6660         int    rc;
6661         ENTRY;
6662
6663         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6664                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6665
6666         if (!fid_is_sane(&fp->gf_fid))
6667                 RETURN(-EINVAL);
6668
6669         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6670                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6671                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6672                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6673                 RETURN(-EINVAL);
6674         }
6675
6676         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6677         if (IS_ERR(obj)) {
6678                 rc = PTR_ERR(obj);
6679                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6680                        PFID(&fp->gf_fid), rc);
6681                 RETURN(rc);
6682         }
6683
6684         if (mdt_object_remote(obj))
6685                 rc = -EREMOTE;
6686         else if (!mdt_object_exists(obj))
6687                 rc = -ENOENT;
6688         else
6689                 rc = 0;
6690
6691         if (rc < 0) {
6692                 mdt_object_put(info->mti_env, obj);
6693                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6694                        PFID(&fp->gf_fid), rc);
6695                 RETURN(rc);
6696         }
6697
6698         rc = mdt_path(info, obj, fp, root_fid);
6699
6700         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6701                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6702                fp->gf_recno, fp->gf_linkno);
6703
6704         mdt_object_put(info->mti_env, obj);
6705
6706         RETURN(rc);
6707 }
6708
6709 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6710                             void *val, int vallen)
6711 {
6712         struct getinfo_fid2path *fpout, *fpin;
6713         struct lu_fid *root_fid = NULL;
6714         int rc = 0;
6715
6716         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6717         fpout = val;
6718
6719         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6720                 lustre_swab_fid2path(fpin);
6721
6722         memcpy(fpout, fpin, sizeof(*fpin));
6723         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6724                 RETURN(-EINVAL);
6725
6726         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6727                       sizeof(struct lu_fid)) {
6728                 /* client sent its root FID, which is normally fileset FID */
6729                 root_fid = fpin->gf_u.gf_root_fid;
6730                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6731                         lustre_swab_lu_fid(root_fid);
6732
6733                 if (root_fid != NULL && !fid_is_sane(root_fid))
6734                         RETURN(-EINVAL);
6735         }
6736
6737         rc = mdt_fid2path(info, root_fid, fpout);
6738         RETURN(rc);
6739 }
6740
6741 int mdt_get_info(struct tgt_session_info *tsi)
6742 {
6743         char    *key;
6744         int      keylen;
6745         __u32   *vallen;
6746         void    *valout;
6747         int      rc;
6748
6749         ENTRY;
6750
6751         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6752         if (key == NULL) {
6753                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6754                 RETURN(err_serious(-EFAULT));
6755         }
6756         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6757                                       RCL_CLIENT);
6758
6759         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6760         if (vallen == NULL) {
6761                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6762                                 tgt_name(tsi->tsi_tgt));
6763                 RETURN(err_serious(-EFAULT));
6764         }
6765
6766         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6767                              *vallen);
6768         rc = req_capsule_server_pack(tsi->tsi_pill);
6769         if (rc)
6770                 RETURN(err_serious(rc));
6771
6772         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6773         if (valout == NULL) {
6774                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6775                                 tgt_name(tsi->tsi_tgt));
6776                 RETURN(err_serious(-EFAULT));
6777         }
6778
6779         if (KEY_IS(KEY_FID2PATH)) {
6780                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6781
6782                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6783                 mdt_thread_info_fini(info);
6784         } else {
6785                 rc = -EINVAL;
6786         }
6787         RETURN(rc);
6788 }
6789
6790 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6791 {
6792         struct obd_ioctl_data *data = karg;
6793         struct lu_fid *fid;
6794         __u64 version;
6795         struct mdt_object *obj;
6796         struct mdt_lock_handle  *lh;
6797         int rc;
6798         ENTRY;
6799
6800         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6801             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6802                 RETURN(-EINVAL);
6803
6804         fid = (struct lu_fid *)data->ioc_inlbuf1;
6805
6806         if (!fid_is_sane(fid))
6807                 RETURN(-EINVAL);
6808
6809         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6810
6811         lh = &mti->mti_lh[MDT_LH_PARENT];
6812         mdt_lock_reg_init(lh, LCK_CR);
6813
6814         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6815         if (IS_ERR(obj))
6816                 RETURN(PTR_ERR(obj));
6817
6818         if (mdt_object_remote(obj)) {
6819                 rc = -EREMOTE;
6820                 /**
6821                  * before calling version get the correct MDS should be
6822                  * fid, this is error to find remote object here
6823                  */
6824                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6825         } else if (!mdt_object_exists(obj)) {
6826                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6827                 rc = -ENOENT;
6828         } else {
6829                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6830                *(__u64 *)data->ioc_inlbuf2 = version;
6831                 rc = 0;
6832         }
6833         mdt_object_unlock_put(mti, obj, lh, 1);
6834         RETURN(rc);
6835 }
6836
6837 /* ioctls on obd dev */
6838 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6839                          void *karg, void __user *uarg)
6840 {
6841         struct lu_env      env;
6842         struct obd_device *obd = exp->exp_obd;
6843         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6844         struct dt_device  *dt = mdt->mdt_bottom;
6845         int rc;
6846
6847         ENTRY;
6848         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6849         rc = lu_env_init(&env, LCT_MD_THREAD);
6850         if (rc)
6851                 RETURN(rc);
6852
6853         switch (cmd) {
6854         case OBD_IOC_SYNC:
6855                 rc = mdt_device_sync(&env, mdt);
6856                 break;
6857         case OBD_IOC_SET_READONLY:
6858                 rc = dt_sync(&env, dt);
6859                 if (rc == 0)
6860                         rc = dt_ro(&env, dt);
6861                 break;
6862         case OBD_IOC_ABORT_RECOVERY:
6863                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6864                 obd->obd_abort_recovery = 1;
6865                 target_stop_recovery_thread(obd);
6866                 rc = 0;
6867                 break;
6868         case OBD_IOC_CHANGELOG_REG:
6869         case OBD_IOC_CHANGELOG_DEREG:
6870         case OBD_IOC_CHANGELOG_CLEAR:
6871                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6872                                                            mdt->mdt_child,
6873                                                            cmd, len, karg);
6874                 break;
6875         case OBD_IOC_START_LFSCK: {
6876                 struct md_device *next = mdt->mdt_child;
6877                 struct obd_ioctl_data *data = karg;
6878                 struct lfsck_start_param lsp;
6879
6880                 if (unlikely(data == NULL)) {
6881                         rc = -EINVAL;
6882                         break;
6883                 }
6884
6885                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6886                 lsp.lsp_index_valid = 0;
6887                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6888                 break;
6889         }
6890         case OBD_IOC_STOP_LFSCK: {
6891                 struct md_device        *next = mdt->mdt_child;
6892                 struct obd_ioctl_data   *data = karg;
6893                 struct lfsck_stop        stop;
6894
6895                 stop.ls_status = LS_STOPPED;
6896                 /* Old lfsck utils may pass NULL @stop. */
6897                 if (data->ioc_inlbuf1 == NULL)
6898                         stop.ls_flags = 0;
6899                 else
6900                         stop.ls_flags =
6901                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6902
6903                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6904                 break;
6905         }
6906         case OBD_IOC_QUERY_LFSCK: {
6907                 struct md_device        *next = mdt->mdt_child;
6908                 struct obd_ioctl_data   *data = karg;
6909
6910                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6911                                                  data->ioc_inlbuf1);
6912                 break;
6913         }
6914         case OBD_IOC_GET_OBJ_VERSION: {
6915                 struct mdt_thread_info *mti;
6916                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6917                 memset(mti, 0, sizeof *mti);
6918                 mti->mti_env = &env;
6919                 mti->mti_mdt = mdt;
6920                 mti->mti_exp = exp;
6921
6922                 rc = mdt_ioc_version_get(mti, karg);
6923                 break;
6924         }
6925         case OBD_IOC_CATLOGLIST: {
6926                 struct mdt_thread_info *mti;
6927
6928                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6929                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6930                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6931                                        &mti->mti_tmp_fid1);
6932                 break;
6933          }
6934         default:
6935                 rc = -EOPNOTSUPP;
6936                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6937                         mdt_obd_name(mdt), cmd, rc);
6938         }
6939
6940         lu_env_fini(&env);
6941         RETURN(rc);
6942 }
6943
6944 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6945 {
6946         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6947         int rc;
6948         ENTRY;
6949
6950         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6951                 struct lfsck_start_param lsp;
6952
6953                 lsp.lsp_start = NULL;
6954                 lsp.lsp_index_valid = 0;
6955                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6956                                                            OBD_IOC_START_LFSCK,
6957                                                            0, &lsp);
6958                 if (rc != 0 && rc != -EALREADY)
6959                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6960                               mdt_obd_name(mdt), rc);
6961         }
6962
6963         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6964         RETURN(rc);
6965 }
6966
6967 static int mdt_obd_postrecov(struct obd_device *obd)
6968 {
6969         struct lu_env env;
6970         int rc;
6971
6972         rc = lu_env_init(&env, LCT_MD_THREAD);
6973         if (rc)
6974                 RETURN(rc);
6975         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6976         lu_env_fini(&env);
6977         return rc;
6978 }
6979
6980 static const struct obd_ops mdt_obd_device_ops = {
6981         .o_owner          = THIS_MODULE,
6982         .o_set_info_async = mdt_obd_set_info_async,
6983         .o_connect        = mdt_obd_connect,
6984         .o_reconnect      = mdt_obd_reconnect,
6985         .o_disconnect     = mdt_obd_disconnect,
6986         .o_init_export    = mdt_init_export,
6987         .o_destroy_export = mdt_destroy_export,
6988         .o_iocontrol      = mdt_iocontrol,
6989         .o_postrecov      = mdt_obd_postrecov,
6990         /* Data-on-MDT IO methods */
6991         .o_preprw         = mdt_obd_preprw,
6992         .o_commitrw       = mdt_obd_commitrw,
6993 };
6994
6995 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6996                                          struct lu_device *d)
6997 {
6998         struct mdt_device *m = mdt_dev(d);
6999         ENTRY;
7000
7001         mdt_fini(env, m);
7002         RETURN(NULL);
7003 }
7004
7005 static struct lu_device *mdt_device_free(const struct lu_env *env,
7006                                          struct lu_device *d)
7007 {
7008         struct mdt_device *m = mdt_dev(d);
7009         ENTRY;
7010
7011         lu_device_fini(&m->mdt_lu_dev);
7012         OBD_FREE_PTR(m);
7013
7014         RETURN(NULL);
7015 }
7016
7017 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7018                                           struct lu_device_type *t,
7019                                           struct lustre_cfg *cfg)
7020 {
7021         struct lu_device  *l;
7022         struct mdt_device *m;
7023
7024         OBD_ALLOC_PTR(m);
7025         if (m != NULL) {
7026                 int rc;
7027
7028                 l = &m->mdt_lu_dev;
7029                 rc = mdt_init0(env, m, t, cfg);
7030                 if (rc != 0) {
7031                         mdt_device_free(env, l);
7032                         l = ERR_PTR(rc);
7033                         return l;
7034                 }
7035         } else
7036                 l = ERR_PTR(-ENOMEM);
7037         return l;
7038 }
7039
7040 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7041 LU_KEY_INIT(mdt, struct mdt_thread_info);
7042
7043 static void mdt_key_fini(const struct lu_context *ctx,
7044                          struct lu_context_key *key, void* data)
7045 {
7046         struct mdt_thread_info *info = data;
7047
7048         if (info->mti_big_lmm) {
7049                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7050                 info->mti_big_lmm = NULL;
7051                 info->mti_big_lmmsize = 0;
7052         }
7053
7054         if (info->mti_big_acl) {
7055                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7056                 info->mti_big_acl = NULL;
7057                 info->mti_big_aclsize = 0;
7058         }
7059
7060         OBD_FREE_PTR(info);
7061 }
7062
7063 /* context key: mdt_thread_key */
7064 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7065
7066 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7067 {
7068         return lu_ucred(info->mti_env);
7069 }
7070
7071 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7072 {
7073         return lu_ucred_check(info->mti_env);
7074 }
7075
7076 /**
7077  * Enable/disable COS (Commit On Sharing).
7078  *
7079  * Set/Clear the COS flag in mdt options.
7080  *
7081  * \param mdt mdt device
7082  * \param val 0 disables COS, other values enable COS
7083  */
7084 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7085 {
7086         struct lu_env env;
7087         int rc;
7088
7089         mdt->mdt_opts.mo_cos = val;
7090         rc = lu_env_init(&env, LCT_LOCAL);
7091         if (unlikely(rc != 0)) {
7092                 CWARN("%s: lu_env initialization failed, cannot "
7093                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7094                 return;
7095         }
7096         mdt_device_sync(&env, mdt);
7097         lu_env_fini(&env);
7098 }
7099
7100 /**
7101  * Check COS (Commit On Sharing) status.
7102  *
7103  * Return COS flag status.
7104  *
7105  * \param mdt mdt device
7106  */
7107 int mdt_cos_is_enabled(struct mdt_device *mdt)
7108 {
7109         return mdt->mdt_opts.mo_cos != 0;
7110 }
7111
7112 static struct lu_device_type_operations mdt_device_type_ops = {
7113         .ldto_device_alloc = mdt_device_alloc,
7114         .ldto_device_free  = mdt_device_free,
7115         .ldto_device_fini  = mdt_device_fini
7116 };
7117
7118 static struct lu_device_type mdt_device_type = {
7119         .ldt_tags     = LU_DEVICE_MD,
7120         .ldt_name     = LUSTRE_MDT_NAME,
7121         .ldt_ops      = &mdt_device_type_ops,
7122         .ldt_ctx_tags = LCT_MD_THREAD
7123 };
7124
7125 static int __init mdt_init(void)
7126 {
7127         int rc;
7128
7129         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7130                      FID_NOBRACE_LEN + 1);
7131         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7132                      FID_LEN + 1);
7133         rc = lu_kmem_init(mdt_caches);
7134         if (rc)
7135                 return rc;
7136
7137         rc = mds_mod_init();
7138         if (rc)
7139                 GOTO(lu_fini, rc);
7140
7141         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
7142                                  LUSTRE_MDT_NAME, &mdt_device_type);
7143         if (rc)
7144                 GOTO(mds_fini, rc);
7145 lu_fini:
7146         if (rc)
7147                 lu_kmem_fini(mdt_caches);
7148 mds_fini:
7149         if (rc)
7150                 mds_mod_exit();
7151         return rc;
7152 }
7153
7154 static void __exit mdt_exit(void)
7155 {
7156         class_unregister_type(LUSTRE_MDT_NAME);
7157         mds_mod_exit();
7158         lu_kmem_fini(mdt_caches);
7159 }
7160
7161 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7162 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7163 MODULE_VERSION(LUSTRE_VERSION_STRING);
7164 MODULE_LICENSE("GPL");
7165
7166 module_init(mdt_init);
7167 module_exit(mdt_exit);