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