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