lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         MIN(mdt->mdt_max_ea_size,
 602                                                             XATTR_SIZE_MAX);
 603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 604                                                         info->mti_big_aclsize);
 605                                         if (info->mti_big_acl == NULL) {
 606                                                 info->mti_big_aclsize = 0;
 607                                                 CERROR("%s: unable to grow "
 608                                                        DFID" ACL buffer\n",
 609                                                        mdt_obd_name(mdt),
 610                                                        PFID(mdt_object_fid(o)));
 611                                                 RETURN(-ENOMEM);
 612                                         }
 613                                 }
 614
 615                                 CDEBUG(D_INODE, "%s: grow the "DFID
 616                                        " ACL buffer to size %d\n",
 617                                        mdt_obd_name(mdt),
 618                                        PFID(mdt_object_fid(o)),
 619                                        info->mti_big_aclsize);
 620
 621                                 buf->lb_buf = info->mti_big_acl;
 622                                 buf->lb_len = info->mti_big_aclsize;
 623
 624                                 goto again;
 625                         }
 626
 627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 629                 }
 630         } else {
 631                 int client;
 632                 int server;
 633                 int acl_buflen;
 634                 int lmm_buflen = 0;
 635                 int lmmsize = 0;
 636
 637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638                 if (acl_buflen >= rc)
 639                         goto map;
 640
 641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 642                 client = ptlrpc_req_get_repsize(pill->rc_req);
 643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 646                                                           RCL_SERVER);
 647                         lmmsize = repbody->mbo_eadatasize;
 648                 }
 649
 650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 652                                "is not big enough with the ACL size %d (%d)\n",
 653                                mdt_obd_name(mdt), client, rc,
 654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 655                         repbody->mbo_aclsize = 0;
 656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 657                         RETURN(-ERANGE);
 658                 }
 659
 660 map:
 661                 if (buf->lb_buf == info->mti_big_acl)
 662                         info->mti_big_acl_used = 1;
 663
 664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 665                                      rc, NODEMAP_FS_TO_CLIENT);
 666                 /* if all ACLs mapped out, rc is still >= 0 */
 667                 if (rc < 0) {
 668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 670                                PFID(mdt_object_fid(o)), rc);
 671                         repbody->mbo_aclsize = 0;
 672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 673                 } else {
 674                         repbody->mbo_aclsize = rc;
 675                         repbody->mbo_valid |= OBD_MD_FLACL;
 676                         rc = 0;
 677                 }
 678         }
 679
 680         RETURN(rc);
 681 }
 682 #endif
 683
 684 /* XXX Look into layout in MDT layer. */
 685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 686 {
 687         struct lov_comp_md_v1   *comp_v1;
 688         struct lov_mds_md       *v1;
 689         int                      i;
 690
 691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 693
 694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 696                                 comp_v1->lcm_entries[i].lcme_offset);
 697                         /* We don't support partial release for now */
 698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 699                                 return false;
 700                 }
 701                 return true;
 702         } else {
 703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 704                         true : false;
 705         }
 706 }
 707
 708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 709                         const struct lu_attr *attr, const struct lu_fid *fid)
 710 {
 711         struct md_attr *ma = &info->mti_attr;
 712         struct obd_export *exp = info->mti_exp;
 713         struct lu_nodemap *nodemap = NULL;
 714
 715         LASSERT(ma->ma_valid & MA_INODE);
 716
 717         if (attr->la_valid & LA_ATIME) {
 718                 b->mbo_atime = attr->la_atime;
 719                 b->mbo_valid |= OBD_MD_FLATIME;
 720         }
 721         if (attr->la_valid & LA_MTIME) {
 722                 b->mbo_mtime = attr->la_mtime;
 723                 b->mbo_valid |= OBD_MD_FLMTIME;
 724         }
 725         if (attr->la_valid & LA_CTIME) {
 726                 b->mbo_ctime = attr->la_ctime;
 727                 b->mbo_valid |= OBD_MD_FLCTIME;
 728         }
 729         if (attr->la_valid & LA_FLAGS) {
 730                 b->mbo_flags = attr->la_flags;
 731                 b->mbo_valid |= OBD_MD_FLFLAGS;
 732         }
 733         if (attr->la_valid & LA_NLINK) {
 734                 b->mbo_nlink = attr->la_nlink;
 735                 b->mbo_valid |= OBD_MD_FLNLINK;
 736         }
 737         if (attr->la_valid & (LA_UID|LA_GID)) {
 738                 nodemap = nodemap_get_from_exp(exp);
 739                 if (IS_ERR(nodemap))
 740                         goto out;
 741         }
 742         if (attr->la_valid & LA_UID) {
 743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 744                                             NODEMAP_FS_TO_CLIENT,
 745                                             attr->la_uid);
 746                 b->mbo_valid |= OBD_MD_FLUID;
 747         }
 748         if (attr->la_valid & LA_GID) {
 749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 750                                             NODEMAP_FS_TO_CLIENT,
 751                                             attr->la_gid);
 752                 b->mbo_valid |= OBD_MD_FLGID;
 753         }
 754
 755         if (attr->la_valid & LA_PROJID) {
 756                 /* TODO, nodemap for project id */
 757                 b->mbo_projid = attr->la_projid;
 758                 b->mbo_valid |= OBD_MD_FLPROJID;
 759         }
 760
 761         b->mbo_mode = attr->la_mode;
 762         if (attr->la_valid & LA_MODE)
 763                 b->mbo_valid |= OBD_MD_FLMODE;
 764         if (attr->la_valid & LA_TYPE)
 765                 b->mbo_valid |= OBD_MD_FLTYPE;
 766
 767         if (fid != NULL) {
 768                 b->mbo_fid1 = *fid;
 769                 b->mbo_valid |= OBD_MD_FLID;
 770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 772         }
 773
 774         if (!(attr->la_valid & LA_TYPE))
 775                 return;
 776
 777         b->mbo_rdev   = attr->la_rdev;
 778         b->mbo_size   = attr->la_size;
 779         b->mbo_blocks = attr->la_blocks;
 780
 781         if (!S_ISREG(attr->la_mode)) {
 782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 784                 /* means no objects are allocated on osts. */
 785                 LASSERT(!(ma->ma_valid & MA_LOV));
 786                 /* just ignore blocks occupied by extend attributes on MDS */
 787                 b->mbo_blocks = 0;
 788                 /* if no object is allocated on osts, the size on mds is valid.
 789                  * b=22272 */
 790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 793                         /* A released file stores its size on MDS. */
 794                         /* But return 1 block for released file, unless tools
 795                          * like tar will consider it fully sparse. (LU-3864)
 796                          */
 797                         if (unlikely(b->mbo_size == 0))
 798                                 b->mbo_blocks = 0;
 799                         else
 800                                 b->mbo_blocks = 1;
 801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 802                 } else if (info->mti_som_valid) { /* som is valid */
 803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 804                 }
 805         }
 806
 807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 809                        PFID(fid), (unsigned long long)b->mbo_size);
 810
 811 out:
 812         if (!IS_ERR_OR_NULL(nodemap))
 813                 nodemap_putref(nodemap);
 814 }
 815
 816 static inline int mdt_body_has_lov(const struct lu_attr *la,
 817                                    const struct mdt_body *body)
 818 {
 819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 821 }
 822
 823 void mdt_client_compatibility(struct mdt_thread_info *info)
 824 {
 825         struct mdt_body       *body;
 826         struct ptlrpc_request *req = mdt_info_req(info);
 827         struct obd_export     *exp = req->rq_export;
 828         struct md_attr        *ma = &info->mti_attr;
 829         struct lu_attr        *la = &ma->ma_attr;
 830         ENTRY;
 831
 832         if (exp_connect_layout(exp))
 833                 /* the client can deal with 16-bit lmm_stripe_count */
 834                 RETURN_EXIT;
 835
 836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 837
 838         if (!mdt_body_has_lov(la, body))
 839                 RETURN_EXIT;
 840
 841         /* now we have a reply with a lov for a client not compatible with the
 842          * layout lock so we have to clean the layout generation number */
 843         if (S_ISREG(la->la_mode))
 844                 ma->ma_lmm->lmm_layout_gen = 0;
 845         EXIT;
 846 }
 847
 848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 849                                    struct mdt_object *o)
 850 {
 851         const struct lu_env *env = info->mti_env;
 852         int rc, rc2;
 853
 854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 855                           XATTR_NAME_LOV);
 856
 857         if (rc == -ENODATA)
 858                 rc = 0;
 859
 860         if (rc < 0)
 861                 goto out;
 862
 863         /* Is it a directory? Let's check for the LMV as well */
 864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 866                                    XATTR_NAME_LMV);
 867
 868                 if (rc2 == -ENODATA)
 869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 870                                            &LU_BUF_NULL,
 871                                            XATTR_NAME_DEFAULT_LMV);
 872
 873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 874                         rc = rc2;
 875         }
 876
 877 out:
 878         return rc;
 879 }
 880
 881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 882                       const char *name)
 883 {
 884         const struct lu_env *env = info->mti_env;
 885         int rc;
 886         ENTRY;
 887
 888         LASSERT(info->mti_big_lmm_used == 0);
 889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 890         if (rc < 0)
 891                 RETURN(rc);
 892
 893         /* big_lmm may need to be grown */
 894         if (info->mti_big_lmmsize < rc) {
 895                 int size = size_roundup_power2(rc);
 896
 897                 if (info->mti_big_lmmsize > 0) {
 898                         /* free old buffer */
 899                         LASSERT(info->mti_big_lmm);
 900                         OBD_FREE_LARGE(info->mti_big_lmm,
 901                                        info->mti_big_lmmsize);
 902                         info->mti_big_lmm = NULL;
 903                         info->mti_big_lmmsize = 0;
 904                 }
 905
 906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 907                 if (info->mti_big_lmm == NULL)
 908                         RETURN(-ENOMEM);
 909                 info->mti_big_lmmsize = size;
 910         }
 911         LASSERT(info->mti_big_lmmsize >= rc);
 912
 913         info->mti_buf.lb_buf = info->mti_big_lmm;
 914         info->mti_buf.lb_len = info->mti_big_lmmsize;
 915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 916
 917         RETURN(rc);
 918 }
 919
 920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 921                    struct md_attr *ma, const char *name)
 922 {
 923         struct md_object *next = mdt_object_child(o);
 924         struct lu_buf    *buf = &info->mti_buf;
 925         int rc;
 926
 927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 928                 buf->lb_buf = ma->ma_lmm;
 929                 buf->lb_len = ma->ma_lmm_size;
 930                 LASSERT(!(ma->ma_valid & MA_LOV));
 931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 932                 buf->lb_buf = ma->ma_lmv;
 933                 buf->lb_len = ma->ma_lmv_size;
 934                 LASSERT(!(ma->ma_valid & MA_LMV));
 935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_lmv;
 937                 buf->lb_len = ma->ma_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 939         } else {
 940                 return -EINVAL;
 941         }
 942
 943         LASSERT(buf->lb_buf);
 944
 945         rc = mo_xattr_get(info->mti_env, next, buf, name);
 946         if (rc > 0) {
 947
 948 got:
 949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 950                         if (info->mti_big_lmm_used)
 951                                 ma->ma_lmm = info->mti_big_lmm;
 952
 953                         /* NOT return LOV EA with hole to old client. */
 954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 955                                      LOV_PATTERN_F_HOLE) &&
 956                             !(exp_connect_flags(info->mti_exp) &
 957                               OBD_CONNECT_LFSCK)) {
 958                                 return -EIO;
 959                         } else {
 960                                 ma->ma_lmm_size = rc;
 961                                 ma->ma_valid |= MA_LOV;
 962                         }
 963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 964                         if (info->mti_big_lmm_used)
 965                                 ma->ma_lmv = info->mti_big_lmm;
 966
 967                         ma->ma_lmv_size = rc;
 968                         ma->ma_valid |= MA_LMV;
 969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 970                         ma->ma_lmv_size = rc;
 971                         ma->ma_valid |= MA_LMV_DEF;
 972                 }
 973
 974                 /* Update mdt_max_mdsize so all clients will be aware that */
 975                 if (info->mti_mdt->mdt_max_mdsize < rc)
 976                         info->mti_mdt->mdt_max_mdsize = rc;
 977
 978                 rc = 0;
 979         } else if (rc == -ENODATA) {
 980                 /* no LOV EA */
 981                 rc = 0;
 982         } else if (rc == -ERANGE) {
 983                 /* Default LMV has fixed size, so it must be able to fit
 984                  * in the original buffer */
 985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 986                         return rc;
 987                 rc = mdt_big_xattr_get(info, o, name);
 988                 if (rc > 0) {
 989                         info->mti_big_lmm_used = 1;
 990                         goto got;
 991                 }
 992         }
 993
 994         return rc;
 995 }
 996
 997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 998                       struct lu_fid *pfid)
 999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* If it is dir object and client require MEA, then we got MEA */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1194                 /* Assumption: MDT_MD size is enough for lmv size. */
1195                 ma->ma_lmv = buffer->lb_buf;
1196                 ma->ma_lmv_size = buffer->lb_len;
1197                 ma->ma_need = MA_INODE;
1198                 if (ma->ma_lmv_size > 0) {
1199                         if (reqbody->mbo_valid & OBD_MD_MEA)
1200                                 ma->ma_need |= MA_LMV;
1201                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1202                                 ma->ma_need |= MA_LMV_DEF;
1203                 }
1204         } else {
1205                 ma->ma_lmm = buffer->lb_buf;
1206                 ma->ma_lmm_size = buffer->lb_len;
1207                 ma->ma_need = MA_INODE | MA_HSM;
1208                 if (ma->ma_lmm_size > 0)
1209                         ma->ma_need |= MA_LOV;
1210         }
1211
1212         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1213             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1214             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1215                 /* get default stripe info for this dir. */
1216                 ma->ma_need |= MA_LOV_DEF;
1217         }
1218         ma->ma_need |= ma_need;
1219
1220         rc = mdt_attr_get_complex(info, o, ma);
1221         if (unlikely(rc)) {
1222                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1223                        "%s: getattr error for "DFID": rc = %d\n",
1224                        mdt_obd_name(info->mti_mdt),
1225                        PFID(mdt_object_fid(o)), rc);
1226                 RETURN(rc);
1227         }
1228
1229         /* if file is released, check if a restore is running */
1230         if (ma->ma_valid & MA_HSM) {
1231                 repbody->mbo_valid |= OBD_MD_TSTATE;
1232                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1233                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1234                         repbody->mbo_t_state = MS_RESTORE;
1235         }
1236
1237         if (likely(ma->ma_valid & MA_INODE))
1238                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1239         else
1240                 RETURN(-EFAULT);
1241
1242         if (mdt_body_has_lov(la, reqbody)) {
1243                 if (ma->ma_valid & MA_LOV) {
1244                         LASSERT(ma->ma_lmm_size);
1245                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1246                         if (S_ISDIR(la->la_mode))
1247                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1248                         else
1249                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1250                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1251                 }
1252                 if (ma->ma_valid & MA_LMV) {
1253                         /* Return -ENOTSUPP for old client */
1254                         if (!mdt_is_striped_client(req->rq_export))
1255                                 RETURN(-ENOTSUPP);
1256
1257                         LASSERT(S_ISDIR(la->la_mode));
1258                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1259                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1260                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1261                 }
1262                 if (ma->ma_valid & MA_LMV_DEF) {
1263                         /* Return -ENOTSUPP for old client */
1264                         if (!mdt_is_striped_client(req->rq_export))
1265                                 RETURN(-ENOTSUPP);
1266                         LASSERT(S_ISDIR(la->la_mode));
1267                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1268                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1269                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1270                                                OBD_MD_DEFAULT_MEA);
1271                 }
1272         } else if (S_ISLNK(la->la_mode) &&
1273                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1274                 buffer->lb_buf = ma->ma_lmm;
1275                 /* eadatasize from client includes NULL-terminator, so
1276                  * there is no need to read it */
1277                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1278                 rc = mo_readlink(env, next, buffer);
1279                 if (unlikely(rc <= 0)) {
1280                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1281                                mdt_obd_name(info->mti_mdt),
1282                                PFID(mdt_object_fid(o)), rc);
1283                         rc = -EFAULT;
1284                 } else {
1285                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1286
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1288                                 rc -= 2;
1289                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1290                         /* we need to report back size with NULL-terminator
1291                          * because client expects that */
1292                         repbody->mbo_eadatasize = rc + 1;
1293                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1294                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1295                                        "on "DFID ", expected %d\n",
1296                                        mdt_obd_name(info->mti_mdt),
1297                                        rc, PFID(mdt_object_fid(o)),
1298                                        reqbody->mbo_eadatasize - 1);
1299                         /* NULL terminate */
1300                         ((char *)ma->ma_lmm)[rc] = 0;
1301
1302                         /* If the total CDEBUG() size is larger than a page, it
1303                          * will print a warning to the console, avoid this by
1304                          * printing just the last part of the symlink. */
1305                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1306                                print_limit < rc ? "..." : "", print_limit,
1307                                (char *)ma->ma_lmm + rc - print_limit, rc);
1308                         rc = 0;
1309                 }
1310         }
1311
1312         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1313                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1314                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1315                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1316                        repbody->mbo_max_mdsize);
1317         }
1318
1319 #ifdef CONFIG_FS_POSIX_ACL
1320         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1321                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1322                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1323                 if (IS_ERR(nodemap))
1324                         RETURN(PTR_ERR(nodemap));
1325
1326                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1327                 nodemap_putref(nodemap);
1328         }
1329 #endif
1330
1331 out:
1332         if (rc == 0)
1333                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1334
1335         RETURN(rc);
1336 }
1337
1338 static int mdt_getattr(struct tgt_session_info *tsi)
1339 {
1340         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1341         struct mdt_object       *obj = info->mti_object;
1342         struct req_capsule      *pill = info->mti_pill;
1343         struct mdt_body         *reqbody;
1344         struct mdt_body         *repbody;
1345         int rc, rc2;
1346         ENTRY;
1347
1348         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1349         LASSERT(reqbody);
1350         LASSERT(obj != NULL);
1351         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1352
1353         /* Special case for Data-on-MDT files to get data version */
1354         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1355                 rc = mdt_data_version_get(tsi);
1356                 GOTO(out, rc);
1357         }
1358
1359         /* Unlike intent case where we need to pre-fill out buffers early on
1360          * in intent policy for ldlm reasons, here we can have a much better
1361          * guess at EA size by just reading it from disk.
1362          * Exceptions are readdir and (missing) directory striping */
1363         /* Readlink */
1364         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1365                 /* No easy way to know how long is the symlink, but it cannot
1366                  * be more than PATH_MAX, so we allocate +1 */
1367                 rc = PATH_MAX + 1;
1368         /* A special case for fs ROOT: getattr there might fetch
1369          * default EA for entire fs, not just for this dir!
1370          */
1371         } else if (lu_fid_eq(mdt_object_fid(obj),
1372                              &info->mti_mdt->mdt_md_root_fid) &&
1373                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1374                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1375                                                                  MDS_GETATTR)) {
1376                 /* Should the default strping be bigger, mdt_fix_reply
1377                  * will reallocate */
1378                 rc = DEF_REP_MD_SIZE;
1379         } else {
1380                 /* Read the actual EA size from disk */
1381                 rc = mdt_attr_get_eabuf_size(info, obj);
1382         }
1383
1384         if (rc < 0)
1385                 GOTO(out, rc = err_serious(rc));
1386
1387         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1388
1389         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1390          * by default. If the target object has more ACL entries, then
1391          * enlarge the buffer when necessary. */
1392         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1393                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1394
1395         rc = req_capsule_server_pack(pill);
1396         if (unlikely(rc != 0))
1397                 GOTO(out, rc = err_serious(rc));
1398
1399         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1400         LASSERT(repbody != NULL);
1401         repbody->mbo_eadatasize = 0;
1402         repbody->mbo_aclsize = 0;
1403
1404         rc = mdt_check_ucred(info);
1405         if (unlikely(rc))
1406                 GOTO(out_shrink, rc);
1407
1408         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1409
1410         rc = mdt_getattr_internal(info, obj, 0);
1411         EXIT;
1412 out_shrink:
1413         mdt_client_compatibility(info);
1414         rc2 = mdt_fix_reply(info);
1415         if (rc == 0)
1416                 rc = rc2;
1417 out:
1418         mdt_thread_info_fini(info);
1419         return rc;
1420 }
1421
1422 /**
1423  * Handler of layout intent RPC requiring the layout modification
1424  *
1425  * \param[in] info      thread environment
1426  * \param[in] obj       object
1427  * \param[in] layout    layout change descriptor
1428  *
1429  * \retval 0    on success
1430  * \retval < 0  error code
1431  */
1432 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1433                       struct md_layout_change *layout)
1434 {
1435         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1436         int rc;
1437         ENTRY;
1438
1439         if (!mdt_object_exists(obj))
1440                 GOTO(out, rc = -ENOENT);
1441
1442         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1443                 GOTO(out, rc = -EINVAL);
1444
1445         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1446                            MAY_WRITE);
1447         if (rc)
1448                 GOTO(out, rc);
1449
1450         /* take layout lock to prepare layout change */
1451         mdt_lock_reg_init(lh, LCK_EX);
1452         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1453         if (rc)
1454                 GOTO(out, rc);
1455
1456         mutex_lock(&obj->mot_som_mutex);
1457         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1458         mutex_unlock(&obj->mot_som_mutex);
1459         mdt_object_unlock(info, obj, lh, 1);
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 /**
1465  * Exchange MOF_LOV_CREATED flags between two objects after a
1466  * layout swap. No assumption is made on whether o1 or o2 have
1467  * created objects or not.
1468  *
1469  * \param[in,out] o1    First swap layout object
1470  * \param[in,out] o2    Second swap layout object
1471  */
1472 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1473 {
1474         unsigned int o1_lov_created = o1->mot_lov_created;
1475
1476         mutex_lock(&o1->mot_lov_mutex);
1477         mutex_lock(&o2->mot_lov_mutex);
1478
1479         o1->mot_lov_created = o2->mot_lov_created;
1480         o2->mot_lov_created = o1_lov_created;
1481
1482         mutex_unlock(&o2->mot_lov_mutex);
1483         mutex_unlock(&o1->mot_lov_mutex);
1484 }
1485
1486 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1487 {
1488         struct mdt_thread_info  *info;
1489         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1490         struct obd_export       *exp = req->rq_export;
1491         struct mdt_object       *o1, *o2, *o;
1492         struct mdt_lock_handle  *lh1, *lh2;
1493         struct mdc_swap_layouts *msl;
1494         int                      rc;
1495         ENTRY;
1496
1497         /* client does not support layout lock, so layout swaping
1498          * is disabled.
1499          * FIXME: there is a problem for old clients which don't support
1500          * layout lock yet. If those clients have already opened the file
1501          * they won't be notified at all so that old layout may still be
1502          * used to do IO. This can be fixed after file release is landed by
1503          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1504         if (!exp_connect_layout(exp))
1505                 RETURN(-EOPNOTSUPP);
1506
1507         info = tsi2mdt_info(tsi);
1508
1509         if (info->mti_dlm_req != NULL)
1510                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1511
1512         o1 = info->mti_object;
1513         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1514                                 &info->mti_body->mbo_fid2);
1515         if (IS_ERR(o))
1516                 GOTO(out, rc = PTR_ERR(o));
1517
1518         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1519                 GOTO(put, rc = -ENOENT);
1520
1521         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1522         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1523                 GOTO(put, rc);
1524
1525         if (rc < 0)
1526                 swap(o1, o2);
1527
1528         /* permission check. Make sure the calling process having permission
1529          * to write both files. */
1530         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1531                            MAY_WRITE);
1532         if (rc < 0)
1533                 GOTO(put, rc);
1534
1535         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1536                            MAY_WRITE);
1537         if (rc < 0)
1538                 GOTO(put, rc);
1539
1540         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1541         if (msl == NULL)
1542                 GOTO(put, rc = -EPROTO);
1543
1544         lh1 = &info->mti_lh[MDT_LH_NEW];
1545         mdt_lock_reg_init(lh1, LCK_EX);
1546         lh2 = &info->mti_lh[MDT_LH_OLD];
1547         mdt_lock_reg_init(lh2, LCK_EX);
1548
1549         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1550                              MDS_INODELOCK_XATTR);
1551         if (rc < 0)
1552                 GOTO(put, rc);
1553
1554         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1555                              MDS_INODELOCK_XATTR);
1556         if (rc < 0)
1557                 GOTO(unlock1, rc);
1558
1559         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1560                              mdt_object_child(o2), msl->msl_flags);
1561         if (rc < 0)
1562                 GOTO(unlock2, rc);
1563
1564         mdt_swap_lov_flag(o1, o2);
1565
1566 unlock2:
1567         mdt_object_unlock(info, o2, lh2, rc);
1568 unlock1:
1569         mdt_object_unlock(info, o1, lh1, rc);
1570 put:
1571         mdt_object_put(info->mti_env, o);
1572 out:
1573         mdt_thread_info_fini(info);
1574         RETURN(rc);
1575 }
1576
1577 static int mdt_raw_lookup(struct mdt_thread_info *info,
1578                           struct mdt_object *parent,
1579                           const struct lu_name *lname,
1580                           struct ldlm_reply *ldlm_rep)
1581 {
1582         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1583         int              rc;
1584         ENTRY;
1585
1586         LASSERT(!info->mti_cross_ref);
1587
1588         /* Only got the fid of this obj by name */
1589         fid_zero(child_fid);
1590         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1591                         lname, child_fid, &info->mti_spec);
1592         if (rc == 0) {
1593                 struct mdt_body *repbody;
1594
1595                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1596                 repbody->mbo_fid1 = *child_fid;
1597                 repbody->mbo_valid = OBD_MD_FLID;
1598                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1599         } else if (rc == -ENOENT) {
1600                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1601         }
1602
1603         RETURN(rc);
1604 }
1605
1606 /*
1607  * UPDATE lock should be taken against parent, and be released before exit;
1608  * child_bits lock should be taken against child, and be returned back:
1609  *            (1)normal request should release the child lock;
1610  *            (2)intent request will grant the lock to client.
1611  */
1612 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1613                                  struct mdt_lock_handle *lhc,
1614                                  __u64 child_bits,
1615                                  struct ldlm_reply *ldlm_rep)
1616 {
1617         struct ptlrpc_request  *req = mdt_info_req(info);
1618         struct mdt_body        *reqbody = NULL;
1619         struct mdt_object      *parent = info->mti_object;
1620         struct mdt_object      *child;
1621         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1622         struct lu_name         *lname = NULL;
1623         struct mdt_lock_handle *lhp = NULL;
1624         struct ldlm_lock       *lock;
1625         __u64 try_bits = 0;
1626         bool is_resent;
1627         int ma_need = 0;
1628         int rc;
1629
1630         ENTRY;
1631
1632         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1633         LASSERT(ergo(is_resent,
1634                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1635
1636         if (parent == NULL)
1637                 RETURN(-ENOENT);
1638
1639         if (info->mti_cross_ref) {
1640                 /* Only getattr on the child. Parent is on another node. */
1641                 mdt_set_disposition(info, ldlm_rep,
1642                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1643                 child = parent;
1644                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1645                        "ldlm_rep = %p\n",
1646                        PFID(mdt_object_fid(child)), ldlm_rep);
1647
1648                 rc = mdt_check_resent_lock(info, child, lhc);
1649                 if (rc < 0) {
1650                         RETURN(rc);
1651                 } else if (rc > 0) {
1652                         mdt_lock_handle_init(lhc);
1653                         mdt_lock_reg_init(lhc, LCK_PR);
1654
1655                         /*
1656                          * Object's name entry is on another MDS, it will
1657                          * request PERM lock only because LOOKUP lock is owned
1658                          * by the MDS where name entry resides.
1659                          *
1660                          * TODO: it should try layout lock too. - Jinshan
1661                          */
1662                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1663                                         MDS_INODELOCK_LAYOUT);
1664                         child_bits |= MDS_INODELOCK_PERM;
1665
1666                         rc = mdt_object_lock(info, child, lhc, child_bits);
1667                         if (rc < 0)
1668                                 RETURN(rc);
1669                 }
1670
1671                 /* Finally, we can get attr for child. */
1672                 if (!mdt_object_exists(child)) {
1673                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1674                                         &child->mot_obj,
1675                                         "remote object doesn't exist.");
1676                         mdt_object_unlock(info, child, lhc, 1);
1677                         RETURN(-ENOENT);
1678                 }
1679
1680                 rc = mdt_getattr_internal(info, child, 0);
1681                 if (unlikely(rc != 0))
1682                         mdt_object_unlock(info, child, lhc, 1);
1683
1684                 RETURN(rc);
1685         }
1686
1687         lname = &info->mti_name;
1688         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1689
1690         if (lu_name_is_valid(lname)) {
1691                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1692                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1693                        PNAME(lname), ldlm_rep);
1694         } else {
1695                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1696                 if (unlikely(reqbody == NULL))
1697                         RETURN(err_serious(-EPROTO));
1698
1699                 *child_fid = reqbody->mbo_fid2;
1700
1701                 if (unlikely(!fid_is_sane(child_fid)))
1702                         RETURN(err_serious(-EINVAL));
1703
1704                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1705                        "ldlm_rep = %p\n",
1706                        PFID(mdt_object_fid(parent)),
1707                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1708         }
1709
1710         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1711
1712         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1713                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1714                                 &parent->mot_obj,
1715                                 "Parent doesn't exist!");
1716                 RETURN(-ESTALE);
1717         }
1718
1719         if (mdt_object_remote(parent)) {
1720                 CERROR("%s: parent "DFID" is on remote target\n",
1721                        mdt_obd_name(info->mti_mdt),
1722                        PFID(mdt_object_fid(parent)));
1723                 RETURN(-EIO);
1724         }
1725
1726         if (lu_name_is_valid(lname)) {
1727                 /* Always allow to lookup ".." */
1728                 if (unlikely(lname->ln_namelen == 2 &&
1729                              lname->ln_name[0] == '.' &&
1730                              lname->ln_name[1] == '.'))
1731                         info->mti_spec.sp_permitted = 1;
1732
1733                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1734                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1735
1736                         RETURN(rc);
1737                 }
1738
1739                 /* step 1: lock parent only if parent is a directory */
1740                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1741                         lhp = &info->mti_lh[MDT_LH_PARENT];
1742                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1743                         rc = mdt_object_lock(info, parent, lhp,
1744                                              MDS_INODELOCK_UPDATE);
1745                         if (unlikely(rc != 0))
1746                                 RETURN(rc);
1747                 }
1748
1749                 /* step 2: lookup child's fid by name */
1750                 fid_zero(child_fid);
1751                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1752                                 child_fid, &info->mti_spec);
1753                 if (rc == -ENOENT)
1754                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1755
1756                 if (rc != 0)
1757                         GOTO(out_parent, rc);
1758         }
1759
1760         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1761
1762         /*
1763          *step 3: find the child object by fid & lock it.
1764          *        regardless if it is local or remote.
1765          *
1766          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1767          *      set parent dir fid the same as child fid in getattr by fid case
1768          *      we should not lu_object_find() the object again, could lead
1769          *      to hung if there is a concurrent unlink destroyed the object.
1770          */
1771         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1772                 mdt_object_get(info->mti_env, parent);
1773                 child = parent;
1774         } else {
1775                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1776                                         child_fid);
1777         }
1778
1779         if (unlikely(IS_ERR(child)))
1780                 GOTO(out_parent, rc = PTR_ERR(child));
1781
1782         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1783         if (!mdt_object_exists(child)) {
1784                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1785                                 &child->mot_obj,
1786                                 "Object doesn't exist!");
1787                 GOTO(out_child, rc = -ENOENT);
1788         }
1789
1790         rc = mdt_check_resent_lock(info, child, lhc);
1791         if (rc < 0) {
1792                 GOTO(out_child, rc);
1793         } else if (rc > 0) {
1794                 mdt_lock_handle_init(lhc);
1795                 mdt_lock_reg_init(lhc, LCK_PR);
1796
1797                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1798                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1799                         struct md_attr *ma = &info->mti_attr;
1800
1801                         ma->ma_valid = 0;
1802                         ma->ma_need = MA_INODE;
1803                         rc = mdt_attr_get_complex(info, child, ma);
1804                         if (unlikely(rc != 0))
1805                                 GOTO(out_child, rc);
1806
1807                         /* If the file has not been changed for some time, we
1808                          * return not only a LOOKUP lock, but also an UPDATE
1809                          * lock and this might save us RPC on later STAT. For
1810                          * directories, it also let negative dentry cache start
1811                          * working for this dir. */
1812                         if (ma->ma_valid & MA_INODE &&
1813                             ma->ma_attr.la_valid & LA_CTIME &&
1814                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1815                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1816                                 child_bits |= MDS_INODELOCK_UPDATE;
1817                 }
1818
1819                 /* layout lock must be granted in a best-effort way
1820                  * for IT operations */
1821                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1822                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1823                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1824                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1825                             exp_connect_layout(info->mti_exp)) {
1826                                 /* try to grant layout lock for regular file. */
1827                                 try_bits = MDS_INODELOCK_LAYOUT;
1828                         }
1829                         /* Acquire DOM lock in advance for data-on-mdt file */
1830                         if (child != parent)
1831                                 try_bits |= MDS_INODELOCK_DOM;
1832                 }
1833
1834                 if (try_bits != 0) {
1835                         /* try layout lock, it may fail to be granted due to
1836                          * contention at LOOKUP or UPDATE */
1837                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1838                                                  try_bits, false);
1839                         if (child_bits & MDS_INODELOCK_LAYOUT)
1840                                 ma_need |= MA_LOV;
1841                 } else {
1842                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1843                          * client will enqueue the lock to the remote MDT */
1844                         if (mdt_object_remote(child))
1845                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1846                         rc = mdt_object_lock(info, child, lhc, child_bits);
1847                 }
1848                 if (unlikely(rc != 0))
1849                         GOTO(out_child, rc);
1850         }
1851
1852         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1853
1854         /* finally, we can get attr for child. */
1855         rc = mdt_getattr_internal(info, child, ma_need);
1856         if (unlikely(rc != 0)) {
1857                 mdt_object_unlock(info, child, lhc, 1);
1858         } else if (lock) {
1859                 /* Debugging code. */
1860                 LDLM_DEBUG(lock, "Returning lock to client");
1861                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1862                                          &lock->l_resource->lr_name),
1863                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1864                          PLDLMRES(lock->l_resource),
1865                          PFID(mdt_object_fid(child)));
1866
1867                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1868                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1869                     child != parent) {
1870                         LDLM_LOCK_PUT(lock);
1871                         mdt_object_put(info->mti_env, child);
1872                         /* NB: call the mdt_pack_size2body always after
1873                          * mdt_object_put(), that is why this special
1874                          * exit path is used. */
1875                         rc = mdt_pack_size2body(info, child_fid,
1876                                                 &lhc->mlh_reg_lh);
1877                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1878                                 /* DOM lock was taken in advance but this is
1879                                  * not DoM file. Drop the lock. */
1880                                 lock_res_and_lock(lock);
1881                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1882                                 unlock_res_and_lock(lock);
1883                         }
1884
1885                         GOTO(out_parent, rc = 0);
1886                 }
1887         }
1888         if (lock)
1889                 LDLM_LOCK_PUT(lock);
1890
1891         EXIT;
1892 out_child:
1893         mdt_object_put(info->mti_env, child);
1894 out_parent:
1895         if (lhp)
1896                 mdt_object_unlock(info, parent, lhp, 1);
1897         return rc;
1898 }
1899
1900 /* normal handler: should release the child lock */
1901 static int mdt_getattr_name(struct tgt_session_info *tsi)
1902 {
1903         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1904         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1905         struct mdt_body        *reqbody;
1906         struct mdt_body        *repbody;
1907         int rc, rc2;
1908         ENTRY;
1909
1910         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1911         LASSERT(reqbody != NULL);
1912         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1913         LASSERT(repbody != NULL);
1914
1915         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1916         repbody->mbo_eadatasize = 0;
1917         repbody->mbo_aclsize = 0;
1918
1919         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1920         if (unlikely(rc))
1921                 GOTO(out_shrink, rc);
1922
1923         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1924         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1925                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1926                 lhc->mlh_reg_lh.cookie = 0;
1927         }
1928         mdt_exit_ucred(info);
1929         EXIT;
1930 out_shrink:
1931         mdt_client_compatibility(info);
1932         rc2 = mdt_fix_reply(info);
1933         if (rc == 0)
1934                 rc = rc2;
1935         mdt_thread_info_fini(info);
1936         return rc;
1937 }
1938
1939 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1940                          void *karg, void __user *uarg);
1941
1942 static int mdt_set_info(struct tgt_session_info *tsi)
1943 {
1944         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1945         char                    *key;
1946         void                    *val;
1947         int                      keylen, vallen, rc = 0;
1948
1949         ENTRY;
1950
1951         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1952         if (key == NULL) {
1953                 DEBUG_REQ(D_HA, req, "no set_info key");
1954                 RETURN(err_serious(-EFAULT));
1955         }
1956
1957         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1958                                       RCL_CLIENT);
1959
1960         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1961         if (val == NULL) {
1962                 DEBUG_REQ(D_HA, req, "no set_info val");
1963                 RETURN(err_serious(-EFAULT));
1964         }
1965
1966         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1967                                       RCL_CLIENT);
1968
1969         /* Swab any part of val you need to here */
1970         if (KEY_IS(KEY_READ_ONLY)) {
1971                 spin_lock(&req->rq_export->exp_lock);
1972                 if (*(__u32 *)val)
1973                         *exp_connect_flags_ptr(req->rq_export) |=
1974                                 OBD_CONNECT_RDONLY;
1975                 else
1976                         *exp_connect_flags_ptr(req->rq_export) &=
1977                                 ~OBD_CONNECT_RDONLY;
1978                 spin_unlock(&req->rq_export->exp_lock);
1979         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1980                 struct changelog_setinfo *cs = val;
1981
1982                 if (vallen != sizeof(*cs)) {
1983                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1984                                tgt_name(tsi->tsi_tgt), vallen);
1985                         RETURN(-EINVAL);
1986                 }
1987                 if (ptlrpc_req_need_swab(req)) {
1988                         __swab64s(&cs->cs_recno);
1989                         __swab32s(&cs->cs_id);
1990                 }
1991
1992                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1993                                    vallen, val, NULL);
1994         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1995                 if (vallen > 0)
1996                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1997         } else {
1998                 RETURN(-EINVAL);
1999         }
2000         RETURN(rc);
2001 }
2002
2003 static int mdt_readpage(struct tgt_session_info *tsi)
2004 {
2005         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2006         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2007         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2008         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2009         struct mdt_body         *repbody;
2010         int                      rc;
2011         int                      i;
2012
2013         ENTRY;
2014
2015         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2016                 RETURN(err_serious(-ENOMEM));
2017
2018         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2019         if (repbody == NULL || reqbody == NULL)
2020                 RETURN(err_serious(-EFAULT));
2021
2022         /*
2023          * prepare @rdpg before calling lower layers and transfer itself. Here
2024          * reqbody->size contains offset of where to start to read and
2025          * reqbody->nlink contains number bytes to read.
2026          */
2027         rdpg->rp_hash = reqbody->mbo_size;
2028         if (rdpg->rp_hash != reqbody->mbo_size) {
2029                 CERROR("Invalid hash: %#llx != %#llx\n",
2030                        rdpg->rp_hash, reqbody->mbo_size);
2031                 RETURN(-EFAULT);
2032         }
2033
2034         rdpg->rp_attrs = reqbody->mbo_mode;
2035         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2036                 rdpg->rp_attrs |= LUDA_64BITHASH;
2037         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2038                                 exp_max_brw_size(tsi->tsi_exp));
2039         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2040                           PAGE_SHIFT;
2041         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2042         if (rdpg->rp_pages == NULL)
2043                 RETURN(-ENOMEM);
2044
2045         for (i = 0; i < rdpg->rp_npages; ++i) {
2046                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2047                 if (rdpg->rp_pages[i] == NULL)
2048                         GOTO(free_rdpg, rc = -ENOMEM);
2049         }
2050
2051         /* call lower layers to fill allocated pages with directory data */
2052         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2053         if (rc < 0)
2054                 GOTO(free_rdpg, rc);
2055
2056         /* send pages to client */
2057         rc = tgt_sendpage(tsi, rdpg, rc);
2058
2059         EXIT;
2060 free_rdpg:
2061
2062         for (i = 0; i < rdpg->rp_npages; i++)
2063                 if (rdpg->rp_pages[i] != NULL)
2064                         __free_page(rdpg->rp_pages[i]);
2065         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2066
2067         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2068                 RETURN(0);
2069
2070         return rc;
2071 }
2072
2073 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2074 {
2075         struct lu_ucred *uc = mdt_ucred_check(info);
2076         struct lu_attr *attr = &info->mti_attr.ma_attr;
2077
2078         if (uc == NULL)
2079                 return -EINVAL;
2080
2081         if (op != REINT_SETATTR) {
2082                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2083                         attr->la_uid = uc->uc_fsuid;
2084                 /* for S_ISGID, inherit gid from his parent, such work will be
2085                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2086                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2087                         attr->la_gid = uc->uc_fsgid;
2088         }
2089
2090         return 0;
2091 }
2092
2093 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2094 {
2095         return op == REINT_OPEN &&
2096              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2097 }
2098
2099 static int mdt_reint_internal(struct mdt_thread_info *info,
2100                               struct mdt_lock_handle *lhc,
2101                               __u32 op)
2102 {
2103         struct req_capsule      *pill = info->mti_pill;
2104         struct mdt_body         *repbody;
2105         int                      rc = 0, rc2;
2106
2107         ENTRY;
2108
2109         rc = mdt_reint_unpack(info, op);
2110         if (rc != 0) {
2111                 CERROR("Can't unpack reint, rc %d\n", rc);
2112                 RETURN(err_serious(rc));
2113         }
2114
2115
2116         /* check if the file system is set to readonly. O_RDONLY open
2117          * is still allowed even the file system is set to readonly mode */
2118         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2119                 RETURN(err_serious(-EROFS));
2120
2121         /* for replay (no_create) lmm is not needed, client has it already */
2122         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2123                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2124                                      DEF_REP_MD_SIZE);
2125
2126         /* llog cookies are always 0, the field is kept for compatibility */
2127         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2128                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2129
2130         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2131          * by default. If the target object has more ACL entries, then
2132          * enlarge the buffer when necessary. */
2133         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2134                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2135                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2136
2137         rc = req_capsule_server_pack(pill);
2138         if (rc != 0) {
2139                 CERROR("Can't pack response, rc %d\n", rc);
2140                 RETURN(err_serious(rc));
2141         }
2142
2143         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2144                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2145                 LASSERT(repbody);
2146                 repbody->mbo_eadatasize = 0;
2147                 repbody->mbo_aclsize = 0;
2148         }
2149
2150         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2151
2152         /* for replay no cookkie / lmm need, because client have this already */
2153         if (info->mti_spec.no_create)
2154                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2155                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2156
2157         rc = mdt_init_ucred_reint(info);
2158         if (rc)
2159                 GOTO(out_shrink, rc);
2160
2161         rc = mdt_fix_attr_ucred(info, op);
2162         if (rc != 0)
2163                 GOTO(out_ucred, rc = err_serious(rc));
2164
2165         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2166         if (rc < 0) {
2167                 GOTO(out_ucred, rc);
2168         } else if (rc == 1) {
2169                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2170                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2171                 GOTO(out_ucred, rc);
2172         }
2173         rc = mdt_reint_rec(info, lhc);
2174         EXIT;
2175 out_ucred:
2176         mdt_exit_ucred(info);
2177 out_shrink:
2178         mdt_client_compatibility(info);
2179
2180         rc2 = mdt_fix_reply(info);
2181         if (rc == 0)
2182                 rc = rc2;
2183
2184         /*
2185          * Data-on-MDT optimization - read data along with OPEN and return it
2186          * in reply. Do that only if we have both DOM and LAYOUT locks.
2187          */
2188         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2189             info->mti_attr.ma_lmm != NULL &&
2190             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2191                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2192                                           &lhc->mlh_reg_lh);
2193         }
2194
2195         return rc;
2196 }
2197
2198 static long mdt_reint_opcode(struct ptlrpc_request *req,
2199                              const struct req_format **fmt)
2200 {
2201         struct mdt_device       *mdt;
2202         struct mdt_rec_reint    *rec;
2203         long                     opc;
2204
2205         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2206         if (rec != NULL) {
2207                 opc = rec->rr_opcode;
2208                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2209                 if (opc < REINT_MAX && fmt[opc] != NULL)
2210                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2211                 else {
2212                         mdt = mdt_exp2dev(req->rq_export);
2213                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2214                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2215                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2216                         opc = err_serious(-EFAULT);
2217                 }
2218         } else {
2219                 opc = err_serious(-EFAULT);
2220         }
2221         return opc;
2222 }
2223
2224 static int mdt_reint(struct tgt_session_info *tsi)
2225 {
2226         long opc;
2227         int  rc;
2228         static const struct req_format *reint_fmts[REINT_MAX] = {
2229                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2230                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2231                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2232                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2233                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2234                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2235                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2236                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2237                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2238                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2239         };
2240
2241         ENTRY;
2242
2243         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2244         if (opc >= 0) {
2245                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2246                 /*
2247                  * No lock possible here from client to pass it to reint code
2248                  * path.
2249                  */
2250                 rc = mdt_reint_internal(info, NULL, opc);
2251                 mdt_thread_info_fini(info);
2252         } else {
2253                 rc = opc;
2254         }
2255
2256         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2257         RETURN(rc);
2258 }
2259
2260 /* this should sync the whole device */
2261 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2262 {
2263         struct dt_device *dt = mdt->mdt_bottom;
2264         int rc;
2265         ENTRY;
2266
2267         rc = dt->dd_ops->dt_sync(env, dt);
2268         RETURN(rc);
2269 }
2270
2271 /* this should sync this object */
2272 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2273                            struct mdt_object *mo)
2274 {
2275         int rc;
2276
2277         ENTRY;
2278
2279         if (!mdt_object_exists(mo)) {
2280                 CWARN("%s: non existing object "DFID": rc = %d\n",
2281                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2282                       -ESTALE);
2283                 RETURN(-ESTALE);
2284         }
2285
2286         rc = mo_object_sync(env, mdt_object_child(mo));
2287
2288         RETURN(rc);
2289 }
2290
2291 static int mdt_sync(struct tgt_session_info *tsi)
2292 {
2293         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2294         struct req_capsule      *pill = tsi->tsi_pill;
2295         struct mdt_body         *body;
2296         int                      rc;
2297
2298         ENTRY;
2299
2300         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2301                 RETURN(err_serious(-ENOMEM));
2302
2303         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2304                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2305         } else {
2306                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2307
2308                 /* sync an object */
2309                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2310                                      info->mti_object);
2311                 if (rc == 0) {
2312                         const struct lu_fid *fid;
2313                         struct lu_attr *la = &info->mti_attr.ma_attr;
2314
2315                         info->mti_attr.ma_need = MA_INODE;
2316                         info->mti_attr.ma_valid = 0;
2317                         rc = mdt_attr_get_complex(info, info->mti_object,
2318                                                   &info->mti_attr);
2319                         if (rc == 0) {
2320                                 body = req_capsule_server_get(pill,
2321                                                               &RMF_MDT_BODY);
2322                                 fid = mdt_object_fid(info->mti_object);
2323                                 mdt_pack_attr2body(info, body, la, fid);
2324                         }
2325                 }
2326                 mdt_thread_info_fini(info);
2327         }
2328         if (rc == 0)
2329                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2330
2331         RETURN(rc);
2332 }
2333
2334 static int mdt_data_sync(struct tgt_session_info *tsi)
2335 {
2336         struct mdt_thread_info *info;
2337         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2338         struct ost_body *body = tsi->tsi_ost_body;
2339         struct ost_body *repbody;
2340         struct mdt_object *mo = NULL;
2341         struct md_attr *ma;
2342         int rc = 0;
2343
2344         ENTRY;
2345
2346         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2347
2348         /* if no fid is specified then do nothing,
2349          * device sync is done via MDS_SYNC */
2350         if (fid_is_zero(&tsi->tsi_fid))
2351                 RETURN(0);
2352
2353         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2354         if (IS_ERR(mo))
2355                 RETURN(PTR_ERR(mo));
2356
2357         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2358         if (rc)
2359                 GOTO(put, rc);
2360
2361         repbody->oa.o_oi = body->oa.o_oi;
2362         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2363
2364         info = tsi2mdt_info(tsi);
2365         ma = &info->mti_attr;
2366         ma->ma_need = MA_INODE;
2367         ma->ma_valid = 0;
2368         rc = mdt_attr_get_complex(info, mo, ma);
2369         if (rc == 0)
2370                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2371         else
2372                 rc = 0;
2373         mdt_thread_info_fini(info);
2374
2375         EXIT;
2376 put:
2377         if (mo != NULL)
2378                 mdt_object_put(tsi->tsi_env, mo);
2379         return rc;
2380 }
2381
2382 /*
2383  * Handle quota control requests to consult current usage/limit, but also
2384  * to configure quota enforcement
2385  */
2386 static int mdt_quotactl(struct tgt_session_info *tsi)
2387 {
2388         struct obd_export       *exp  = tsi->tsi_exp;
2389         struct req_capsule      *pill = tsi->tsi_pill;
2390         struct obd_quotactl     *oqctl, *repoqc;
2391         int                      id, rc;
2392         struct mdt_device       *mdt = mdt_exp2dev(exp);
2393         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2394         struct lu_nodemap       *nodemap;
2395         ENTRY;
2396
2397         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2398         if (oqctl == NULL)
2399                 RETURN(err_serious(-EPROTO));
2400
2401         rc = req_capsule_server_pack(pill);
2402         if (rc)
2403                 RETURN(err_serious(rc));
2404
2405         nodemap = nodemap_get_from_exp(exp);
2406         if (IS_ERR(nodemap))
2407                 RETURN(PTR_ERR(nodemap));
2408
2409         switch (oqctl->qc_cmd) {
2410                 /* master quotactl */
2411         case Q_SETINFO:
2412         case Q_SETQUOTA:
2413         case LUSTRE_Q_SETDEFAULT:
2414                 if (!nodemap_can_setquota(nodemap))
2415                         GOTO(out_nodemap, rc = -EPERM);
2416         case Q_GETINFO:
2417         case Q_GETQUOTA:
2418         case LUSTRE_Q_GETDEFAULT:
2419                 if (qmt == NULL)
2420                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2421                 /* slave quotactl */
2422         case Q_GETOINFO:
2423         case Q_GETOQUOTA:
2424                 break;
2425         default:
2426                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2427                 GOTO(out_nodemap, rc = -EFAULT);
2428         }
2429
2430         id = oqctl->qc_id;
2431         switch (oqctl->qc_type) {
2432         case USRQUOTA:
2433                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2434                                     NODEMAP_CLIENT_TO_FS, id);
2435                 break;
2436         case GRPQUOTA:
2437                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2438                                     NODEMAP_CLIENT_TO_FS, id);
2439                 break;
2440         case PRJQUOTA:
2441                 /* todo: check/map project id */
2442                 id = oqctl->qc_id;
2443                 break;
2444         default:
2445                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2446         }
2447         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2448         if (repoqc == NULL)
2449                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2450
2451         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2452                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2453
2454         if (oqctl->qc_id != id)
2455                 swap(oqctl->qc_id, id);
2456
2457         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2458                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2459                         RETURN(-EINPROGRESS);
2460         }
2461
2462         switch (oqctl->qc_cmd) {
2463
2464         case Q_GETINFO:
2465         case Q_SETINFO:
2466         case Q_SETQUOTA:
2467         case Q_GETQUOTA:
2468         case LUSTRE_Q_SETDEFAULT:
2469         case LUSTRE_Q_GETDEFAULT:
2470                 /* forward quotactl request to QMT */
2471                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2472                 break;
2473
2474         case Q_GETOINFO:
2475         case Q_GETOQUOTA:
2476                 /* slave quotactl */
2477                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2478                                    oqctl);
2479                 break;
2480
2481         default:
2482                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2483                 GOTO(out_nodemap, rc = -EFAULT);
2484         }
2485
2486         if (oqctl->qc_id != id)
2487                 swap(oqctl->qc_id, id);
2488
2489         *repoqc = *oqctl;
2490
2491         EXIT;
2492
2493 out_nodemap:
2494         nodemap_putref(nodemap);
2495
2496         return rc;
2497 }
2498
2499 /** clone llog ctxt from child (mdd)
2500  * This allows remote llog (replicator) access.
2501  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2502  * context was originally set up, or we can handle them directly.
2503  * I choose the latter, but that means I need any llog
2504  * contexts set up by child to be accessable by the mdt.  So we clone the
2505  * context into our context list here.
2506  */
2507 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2508                                int idx)
2509 {
2510         struct md_device  *next = mdt->mdt_child;
2511         struct llog_ctxt *ctxt;
2512         int rc;
2513
2514         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2515                 return 0;
2516
2517         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2518         if (rc || ctxt == NULL) {
2519                 return 0;
2520         }
2521
2522         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2523         if (rc)
2524                 CERROR("Can't set mdt ctxt %d\n", rc);
2525
2526         return rc;
2527 }
2528
2529 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2530                                  struct mdt_device *mdt, int idx)
2531 {
2532         struct llog_ctxt *ctxt;
2533
2534         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2535         if (ctxt == NULL)
2536                 return 0;
2537         /* Put once for the get we just did, and once for the clone */
2538         llog_ctxt_put(ctxt);
2539         llog_ctxt_put(ctxt);
2540         return 0;
2541 }
2542
2543 /*
2544  * sec context handlers
2545  */
2546 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2547 {
2548         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2549
2550         return 0;
2551 }
2552
2553 /*
2554  * quota request handlers
2555  */
2556 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2557 {
2558         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2559         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2560         int                      rc;
2561         ENTRY;
2562
2563         if (qmt == NULL)
2564                 RETURN(err_serious(-EOPNOTSUPP));
2565
2566         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2567         RETURN(rc);
2568 }
2569
2570 struct mdt_object *mdt_object_new(const struct lu_env *env,
2571                                   struct mdt_device *d,
2572                                   const struct lu_fid *f)
2573 {
2574         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2575         struct lu_object *o;
2576         struct mdt_object *m;
2577         ENTRY;
2578
2579         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2580         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2581         if (unlikely(IS_ERR(o)))
2582                 m = (struct mdt_object *)o;
2583         else
2584                 m = mdt_obj(o);
2585         RETURN(m);
2586 }
2587
2588 struct mdt_object *mdt_object_find(const struct lu_env *env,
2589                                    struct mdt_device *d,
2590                                    const struct lu_fid *f)
2591 {
2592         struct lu_object *o;
2593         struct mdt_object *m;
2594         ENTRY;
2595
2596         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2597         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2598         if (unlikely(IS_ERR(o)))
2599                 m = (struct mdt_object *)o;
2600         else
2601                 m = mdt_obj(o);
2602
2603         RETURN(m);
2604 }
2605
2606 /**
2607  * Asyncronous commit for mdt device.
2608  *
2609  * Pass asynchonous commit call down the MDS stack.
2610  *
2611  * \param env environment
2612  * \param mdt the mdt device
2613  */
2614 static void mdt_device_commit_async(const struct lu_env *env,
2615                                     struct mdt_device *mdt)
2616 {
2617         struct dt_device *dt = mdt->mdt_bottom;
2618         int rc;
2619         ENTRY;
2620
2621         rc = dt->dd_ops->dt_commit_async(env, dt);
2622         if (unlikely(rc != 0))
2623                 CWARN("%s: async commit start failed: rc = %d\n",
2624                       mdt_obd_name(mdt), rc);
2625         atomic_inc(&mdt->mdt_async_commit_count);
2626         EXIT;
2627 }
2628
2629 /**
2630  * Mark the lock as "synchonous".
2631  *
2632  * Mark the lock to deffer transaction commit to the unlock time.
2633  *
2634  * \param lock the lock to mark as "synchonous"
2635  *
2636  * \see mdt_is_lock_sync
2637  * \see mdt_save_lock
2638  */
2639 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2640 {
2641         lock->l_ast_data = (void*)1;
2642 }
2643
2644 /**
2645  * Check whehter the lock "synchonous" or not.
2646  *
2647  * \param lock the lock to check
2648  * \retval 1 the lock is "synchonous"
2649  * \retval 0 the lock isn't "synchronous"
2650  *
2651  * \see mdt_set_lock_sync
2652  * \see mdt_save_lock
2653  */
2654 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2655 {
2656         return lock->l_ast_data != NULL;
2657 }
2658
2659 /**
2660  * Blocking AST for mdt locks.
2661  *
2662  * Starts transaction commit if in case of COS lock conflict or
2663  * deffers such a commit to the mdt_save_lock.
2664  *
2665  * \param lock the lock which blocks a request or cancelling lock
2666  * \param desc unused
2667  * \param data unused
2668  * \param flag indicates whether this cancelling or blocking callback
2669  * \retval 0
2670  * \see ldlm_blocking_ast_nocheck
2671  */
2672 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2673                      void *data, int flag)
2674 {
2675         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2676         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2677         struct ldlm_cb_set_arg *arg = data;
2678         bool commit_async = false;
2679         int rc;
2680         ENTRY;
2681
2682         if (flag == LDLM_CB_CANCELING)
2683                 RETURN(0);
2684
2685         lock_res_and_lock(lock);
2686         if (lock->l_blocking_ast != mdt_blocking_ast) {
2687                 unlock_res_and_lock(lock);
2688                 RETURN(0);
2689         }
2690
2691         /* A blocking ast may be sent from ldlm_lock_decref_internal
2692          * when the last reference to a local lock was released and
2693          * during blocking event from ldlm_work_bl_ast_lock().
2694          * The 'data' parameter is l_ast_data in the first case and
2695          * callback arguments in the second one. Distinguish them by that.
2696          */
2697         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2698                 goto skip_cos_checks;
2699
2700         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2701                 if (mdt_cos_is_enabled(mdt)) {
2702                         if (!arg->bl_desc->bl_same_client)
2703                                 mdt_set_lock_sync(lock);
2704                 } else if (mdt_slc_is_enabled(mdt) &&
2705                            arg->bl_desc->bl_cos_incompat) {
2706                         mdt_set_lock_sync(lock);
2707                         /*
2708                          * we may do extra commit here, but there is a small
2709                          * window to miss a commit: lock was unlocked (saved),
2710                          * then a conflict lock queued and we come here, but
2711                          * REP-ACK not received, so lock was not converted to
2712                          * COS mode yet.
2713                          * Fortunately this window is quite small, so the
2714                          * extra commit should be rare (not to say distributed
2715                          * operation is rare too).
2716                          */
2717                         commit_async = true;
2718                 }
2719         } else if (lock->l_req_mode == LCK_COS) {
2720                 commit_async = true;
2721         }
2722
2723 skip_cos_checks:
2724         rc = ldlm_blocking_ast_nocheck(lock);
2725
2726         if (commit_async) {
2727                 struct lu_env env;
2728
2729                 rc = lu_env_init(&env, LCT_LOCAL);
2730                 if (unlikely(rc != 0))
2731                         CWARN("%s: lu_env initialization failed, cannot "
2732                               "start asynchronous commit: rc = %d\n",
2733                               obd->obd_name, rc);
2734                 else
2735                         mdt_device_commit_async(&env, mdt);
2736                 lu_env_fini(&env);
2737         }
2738         RETURN(rc);
2739 }
2740
2741 /*
2742  * Blocking AST for cross-MDT lock
2743  *
2744  * Discard lock from uncommitted_slc_locks and cancel it.
2745  *
2746  * \param lock  the lock which blocks a request or cancelling lock
2747  * \param desc  unused
2748  * \param data  unused
2749  * \param flag  indicates whether this cancelling or blocking callback
2750  * \retval      0 on success
2751  * \retval      negative number on error
2752  */
2753 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2754                             void *data, int flag)
2755 {
2756         int rc = 0;
2757         ENTRY;
2758
2759         switch (flag) {
2760         case LDLM_CB_BLOCKING: {
2761                 struct lustre_handle lockh;
2762
2763                 ldlm_lock2handle(lock, &lockh);
2764                 rc = ldlm_cli_cancel(&lockh,
2765                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2766                 if (rc < 0) {
2767                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2768                         RETURN(rc);
2769                 }
2770                 break;
2771         }
2772         case LDLM_CB_CANCELING: {
2773                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2774                 struct mdt_device *mdt =
2775                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2776
2777                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2778
2779                 /* discard slc lock here so that it can be cleaned anytime,
2780                  * especially for cleanup_resource() */
2781                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2782
2783                 /* once we cache lock, l_ast_data is set to mdt_object */
2784                 if (lock->l_ast_data != NULL) {
2785                         struct mdt_object *mo = lock->l_ast_data;
2786                         struct lu_env env;
2787
2788                         rc = lu_env_init(&env, LCT_MD_THREAD);
2789                         if (unlikely(rc != 0)) {
2790                                 CWARN("%s: lu_env initialization failed, object"
2791                                       "%p "DFID" is leaked!\n",
2792                                       obd->obd_name, mo,
2793                                       PFID(mdt_object_fid(mo)));
2794                                 RETURN(rc);
2795                         }
2796
2797                         if (lock->l_policy_data.l_inodebits.bits &
2798                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2799                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2800                                 mo->mot_cache_attr = 0;
2801                         }
2802                         mdt_object_put(&env, mo);
2803                         lu_env_fini(&env);
2804                 }
2805                 break;
2806         }
2807         default:
2808                 LBUG();
2809         }
2810
2811         RETURN(rc);
2812 }
2813
2814 int mdt_check_resent_lock(struct mdt_thread_info *info,
2815                           struct mdt_object *mo,
2816                           struct mdt_lock_handle *lhc)
2817 {
2818         /* the lock might already be gotten in ldlm_handle_enqueue() */
2819         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2820                 struct ptlrpc_request *req = mdt_info_req(info);
2821                 struct ldlm_lock      *lock;
2822
2823                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2824                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2825                 if (lock == NULL) {
2826                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2827                          * a resend case, however, it could be already destroyed
2828                          * due to client eviction or a raced cancel RPC. */
2829                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2830                                           lhc->mlh_reg_lh.cookie);
2831                         RETURN(-ESTALE);
2832                 }
2833
2834                 if (!fid_res_name_eq(mdt_object_fid(mo),
2835                                      &lock->l_resource->lr_name)) {
2836                         CWARN("%s: Although resent, but still not "
2837                               "get child lock:"DFID"\n",
2838                               info->mti_exp->exp_obd->obd_name,
2839                               PFID(mdt_object_fid(mo)));
2840                         LDLM_LOCK_PUT(lock);
2841                         RETURN(-EPROTO);
2842                 }
2843                 LDLM_LOCK_PUT(lock);
2844                 return 0;
2845         }
2846         return 1;
2847 }
2848
2849 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
2850 {
2851         mdt_object_get(NULL, lock->l_ast_data);
2852 }
2853
2854 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2855                                struct mdt_object *o, const struct lu_fid *fid,
2856                                struct lustre_handle *lh, enum ldlm_mode mode,
2857                                __u64 *ibits, __u64 trybits, bool cache)
2858 {
2859         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
2860         union ldlm_policy_data *policy = &mti->mti_policy;
2861         struct ldlm_res_id *res_id = &mti->mti_res_id;
2862         int rc = 0;
2863         ENTRY;
2864
2865         LASSERT(mdt_object_remote(o));
2866
2867         fid_build_reg_res_name(fid, res_id);
2868
2869         memset(einfo, 0, sizeof(*einfo));
2870         einfo->ei_type = LDLM_IBITS;
2871         einfo->ei_mode = mode;
2872         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2873         einfo->ei_cb_cp = ldlm_completion_ast;
2874         einfo->ei_enq_slave = 0;
2875         einfo->ei_res_id = res_id;
2876
2877         if (cache) {
2878                 /*
2879                  * if we cache lock, couple lock with mdt_object, so that object
2880                  * can be easily found in lock ASTs.
2881                  */
2882                 einfo->ei_cbdata = o;
2883                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
2884         }
2885
2886         memset(policy, 0, sizeof(*policy));
2887         policy->l_inodebits.bits = *ibits;
2888         policy->l_inodebits.try_bits = trybits;
2889
2890         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2891                             policy);
2892
2893         /* Return successfully acquired bits to a caller */
2894         if (rc == 0) {
2895                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2896
2897                 LASSERT(lock);
2898                 *ibits = lock->l_policy_data.l_inodebits.bits;
2899                 LDLM_LOCK_PUT(lock);
2900         }
2901         RETURN(rc);
2902 }
2903
2904 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2905                            const struct lu_fid *fid, struct lustre_handle *lh,
2906                            enum ldlm_mode mode, __u64 ibits, bool cache)
2907 {
2908         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2909                                           cache);
2910 }
2911
2912 static int mdt_object_local_lock(struct mdt_thread_info *info,
2913                                  struct mdt_object *o,
2914                                  struct mdt_lock_handle *lh, __u64 *ibits,
2915                                  __u64 trybits, bool cos_incompat)
2916 {
2917         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2918         union ldlm_policy_data *policy = &info->mti_policy;
2919         struct ldlm_res_id *res_id = &info->mti_res_id;
2920         __u64 dlmflags = 0, *cookie = NULL;
2921         int rc;
2922         ENTRY;
2923
2924         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2925         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2926         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2927         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2928
2929         if (cos_incompat) {
2930                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2931                         lh->mlh_reg_mode == LCK_EX);
2932                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2933         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2934                 dlmflags |= LDLM_FL_COS_ENABLED;
2935         }
2936
2937         /* Only enqueue LOOKUP lock for remote object */
2938         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
2939
2940         if (lh->mlh_type == MDT_PDO_LOCK) {
2941                 /* check for exists after object is locked */
2942                 if (mdt_object_exists(o) == 0) {
2943                         /* Non-existent object shouldn't have PDO lock */
2944                         RETURN(-ESTALE);
2945                 } else {
2946                         /* Non-dir object shouldn't have PDO lock */
2947                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2948                                 RETURN(-ENOTDIR);
2949                 }
2950         }
2951
2952         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2953         dlmflags |= LDLM_FL_ATOMIC_CB;
2954
2955         if (info->mti_exp)
2956                 cookie = &info->mti_exp->exp_handle.h_cookie;
2957
2958         /*
2959          * Take PDO lock on whole directory and build correct @res_id for lock
2960          * on part of directory.
2961          */
2962         if (lh->mlh_pdo_hash != 0) {
2963                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2964                 mdt_lock_pdo_mode(info, o, lh);
2965                 if (lh->mlh_pdo_mode != LCK_NL) {
2966                         /*
2967                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2968                          * is never going to be sent to client and we do not
2969                          * want it slowed down due to possible cancels.
2970                          */
2971                         policy->l_inodebits.bits =
2972                                 *ibits & MDS_INODELOCK_UPDATE;
2973                         policy->l_inodebits.try_bits =
2974                                 trybits & MDS_INODELOCK_UPDATE;
2975                         /* at least one of them should be set */
2976                         LASSERT(policy->l_inodebits.bits |
2977                                 policy->l_inodebits.try_bits);
2978                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
2979                                           lh->mlh_pdo_mode, policy, res_id,
2980                                           dlmflags, cookie);
2981                         if (unlikely(rc != 0))
2982                                 GOTO(out_unlock, rc);
2983                 }
2984
2985                 /*
2986                  * Finish res_id initializing by name hash marking part of
2987                  * directory which is taking modification.
2988                  */
2989                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2990         }
2991
2992         policy->l_inodebits.bits = *ibits;
2993         policy->l_inodebits.try_bits = trybits;
2994
2995         /*
2996          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2997          * going to be sent to client. If it is - mdt_intent_policy() path will
2998          * fix it up and turn FL_LOCAL flag off.
2999          */
3000         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3001                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3002                           cookie);
3003 out_unlock:
3004         if (rc != 0)
3005                 mdt_object_unlock(info, o, lh, 1);
3006         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3007                    lh->mlh_pdo_hash != 0 &&
3008                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3009                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3010
3011         /* Return successfully acquired bits to a caller */
3012         if (rc == 0) {
3013                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3014
3015                 LASSERT(lock);
3016                 *ibits = lock->l_policy_data.l_inodebits.bits;
3017                 LDLM_LOCK_PUT(lock);
3018         }
3019         RETURN(rc);
3020 }
3021
3022 static int
3023 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3024                          struct mdt_lock_handle *lh, __u64 *ibits,
3025                          __u64 trybits, bool cos_incompat)
3026 {
3027         struct mdt_lock_handle *local_lh = NULL;
3028         int rc;
3029         ENTRY;
3030
3031         if (!mdt_object_remote(o)) {
3032                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3033                                            cos_incompat);
3034                 RETURN(rc);
3035         }
3036
3037         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3038         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3039                     MDS_INODELOCK_XATTR);
3040
3041         /* Only enqueue LOOKUP lock for remote object */
3042         if (*ibits & MDS_INODELOCK_LOOKUP) {
3043                 __u64 local = MDS_INODELOCK_LOOKUP;
3044
3045                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3046                                            cos_incompat);
3047                 if (rc != ELDLM_OK)
3048                         RETURN(rc);
3049
3050                 local_lh = lh;
3051         }
3052
3053         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3054                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3055                  * enqueue RPC can only request 1 lock, to avoid extra
3056                  * RPC, so it will instead enqueue EX lock for remote
3057                  * object anyway XXX*/
3058                 if (lh->mlh_type == MDT_PDO_LOCK &&
3059                     lh->mlh_pdo_hash != 0) {
3060                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3061                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3062                                PFID(mdt_object_fid(o)));
3063                         lh->mlh_pdo_hash = 0;
3064                         lh->mlh_rreg_mode = LCK_EX;
3065                         lh->mlh_type = MDT_REG_LOCK;
3066                 }
3067
3068                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3069                                                 &lh->mlh_rreg_lh,
3070                                                 lh->mlh_rreg_mode,
3071                                                 ibits, trybits, false);
3072                 if (rc != ELDLM_OK) {
3073                         if (local_lh != NULL)
3074                                 mdt_object_unlock(info, o, local_lh, rc);
3075                         RETURN(rc);
3076                 }
3077         }
3078
3079         /* other components like LFSCK can use lockless access
3080          * and populate cache, so we better invalidate it */
3081         mo_invalidate(info->mti_env, mdt_object_child(o));
3082
3083         RETURN(0);
3084 }
3085
3086 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3087                     struct mdt_lock_handle *lh, __u64 ibits)
3088 {
3089         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3090 }
3091
3092 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3093                           struct mdt_lock_handle *lh, __u64 ibits,
3094                           bool cos_incompat)
3095 {
3096         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3097         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3098                                         cos_incompat);
3099 }
3100
3101 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3102                         struct mdt_lock_handle *lh, __u64 *ibits,
3103                         __u64 trybits, bool cos_incompat)
3104 {
3105         bool trylock_only = *ibits == 0;
3106         int rc;
3107
3108         LASSERT(!(*ibits & trybits));
3109         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3110                                       cos_incompat);
3111         if (rc && trylock_only) { /* clear error for try ibits lock only */
3112                 LASSERT(*ibits == 0);
3113                 rc = 0;
3114         }
3115         return rc;
3116 }
3117
3118 /**
3119  * Save a lock within request object.
3120  *
3121  * Keep the lock referenced until whether client ACK or transaction
3122  * commit happens or release the lock immediately depending on input
3123  * parameters. If COS is ON, a write lock is converted to COS lock
3124  * before saving.
3125  *
3126  * \param info thead info object
3127  * \param h lock handle
3128  * \param mode lock mode
3129  * \param decref force immediate lock releasing
3130  */
3131 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3132                    enum ldlm_mode mode, int decref)
3133 {
3134         ENTRY;
3135
3136         if (lustre_handle_is_used(h)) {
3137                 if (decref || !info->mti_has_trans ||
3138                     !(mode & (LCK_PW | LCK_EX))) {
3139                         mdt_fid_unlock(h, mode);
3140                 } else {
3141                         struct mdt_device *mdt = info->mti_mdt;
3142                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3143                         struct ptlrpc_request *req = mdt_info_req(info);
3144                         bool cos = mdt_cos_is_enabled(mdt);
3145                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3146
3147                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3148                                  h->cookie);
3149
3150                         /* there is no request if mdt_object_unlock() is called
3151                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3152                         if (likely(req != NULL)) {
3153                                 LDLM_DEBUG(lock, "save lock request %p reply "
3154                                         "state %p transno %lld\n", req,
3155                                         req->rq_reply_state, req->rq_transno);
3156                                 if (cos) {
3157                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3158                                         mode = LCK_COS;
3159                                 }
3160                                 if (req->rq_export->exp_disconnected)
3161                                         mdt_fid_unlock(h, mode);
3162                                 else
3163                                         ptlrpc_save_lock(req, h, mode, cos,
3164                                                          convert_lock);
3165                         } else {
3166                                 mdt_fid_unlock(h, mode);
3167                         }
3168                         if (mdt_is_lock_sync(lock)) {
3169                                 CDEBUG(D_HA, "found sync-lock,"
3170                                        " async commit started\n");
3171                                 mdt_device_commit_async(info->mti_env,
3172                                                         mdt);
3173                         }
3174                         LDLM_LOCK_PUT(lock);
3175                 }
3176                 h->cookie = 0ull;
3177         }
3178
3179         EXIT;
3180 }
3181
3182 /**
3183  * Save cross-MDT lock in uncommitted_slc_locks
3184  *
3185  * Keep the lock referenced until transaction commit happens or release the lock
3186  * immediately depending on input parameters.
3187  *
3188  * \param info thead info object
3189  * \param h lock handle
3190  * \param mode lock mode
3191  * \param decref force immediate lock releasing
3192  */
3193 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3194                                  struct mdt_object *o, struct lustre_handle *h,
3195                                  enum ldlm_mode mode, int decref)
3196 {
3197         ENTRY;
3198
3199         if (lustre_handle_is_used(h)) {
3200                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3201
3202                 if (o != NULL &&
3203                     (lock->l_policy_data.l_inodebits.bits &
3204                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3205                         mo_invalidate(info->mti_env, mdt_object_child(o));
3206
3207                 if (decref || !info->mti_has_trans ||
3208                     !(mode & (LCK_PW | LCK_EX))) {
3209                         ldlm_lock_decref_and_cancel(h, mode);
3210                         LDLM_LOCK_PUT(lock);
3211                 } else {
3212                         struct ptlrpc_request *req = mdt_info_req(info);
3213
3214                         LASSERT(req != NULL);
3215                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3216                                           req->rq_transno);
3217                         ldlm_lock_decref(h, mode);
3218                 }
3219                 h->cookie = 0ull;
3220         }
3221
3222         EXIT;
3223 }
3224
3225 /**
3226  * Unlock mdt object.
3227  *
3228  * Immeditely release the regular lock and the PDO lock or save the
3229  * lock in request and keep them referenced until client ACK or
3230  * transaction commit.
3231  *
3232  * \param info thread info object
3233  * \param o mdt object
3234  * \param lh mdt lock handle referencing regular and PDO locks
3235  * \param decref force immediate lock releasing
3236  *
3237  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3238  */
3239 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3240                        struct mdt_lock_handle *lh, int decref)
3241 {
3242         ENTRY;
3243
3244         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3245         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3246         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3247                              decref);
3248
3249         EXIT;
3250 }
3251
3252 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3253                                         const struct lu_fid *f,
3254                                         struct mdt_lock_handle *lh,
3255                                         __u64 ibits)
3256 {
3257         struct mdt_object *o;
3258
3259         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3260         if (!IS_ERR(o)) {
3261                 int rc;
3262
3263                 rc = mdt_object_lock(info, o, lh, ibits);
3264                 if (rc != 0) {
3265                         mdt_object_put(info->mti_env, o);
3266                         o = ERR_PTR(rc);
3267                 }
3268         }
3269         return o;
3270 }
3271
3272 void mdt_object_unlock_put(struct mdt_thread_info * info,
3273                            struct mdt_object * o,
3274                            struct mdt_lock_handle *lh,
3275                            int decref)
3276 {
3277         mdt_object_unlock(info, o, lh, decref);
3278         mdt_object_put(info->mti_env, o);
3279 }
3280
3281 /*
3282  * Generic code handling requests that have struct mdt_body passed in:
3283  *
3284  *  - extract mdt_body from request and save it in @info, if present;
3285  *
3286  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3287  *  @info;
3288  *
3289  *  - if HABEO_CORPUS flag is set for this request type check whether object
3290  *  actually exists on storage (lu_object_exists()).
3291  *
3292  */
3293 static int mdt_body_unpack(struct mdt_thread_info *info,
3294                            enum tgt_handler_flags flags)
3295 {
3296         const struct mdt_body    *body;
3297         struct mdt_object        *obj;
3298         const struct lu_env      *env;
3299         struct req_capsule       *pill;
3300         int                       rc;
3301         ENTRY;
3302
3303         env = info->mti_env;
3304         pill = info->mti_pill;
3305
3306         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3307         if (body == NULL)
3308                 RETURN(-EFAULT);
3309
3310         if (!(body->mbo_valid & OBD_MD_FLID))
3311                 RETURN(0);
3312
3313         if (!fid_is_sane(&body->mbo_fid1)) {
3314                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3315                 RETURN(-EINVAL);
3316         }
3317
3318         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3319         if (!IS_ERR(obj)) {
3320                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3321                         mdt_object_put(env, obj);
3322                         rc = -ENOENT;
3323                 } else {
3324                         info->mti_object = obj;
3325                         rc = 0;
3326                 }
3327         } else
3328                 rc = PTR_ERR(obj);
3329
3330         RETURN(rc);
3331 }
3332
3333 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3334                                    enum tgt_handler_flags flags)
3335 {
3336         struct req_capsule *pill = info->mti_pill;
3337         int rc;
3338         ENTRY;
3339
3340         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3341                 rc = mdt_body_unpack(info, flags);
3342         else
3343                 rc = 0;
3344
3345         if (rc == 0 && (flags & HABEO_REFERO)) {
3346                 /* Pack reply. */
3347                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3348                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3349                                              DEF_REP_MD_SIZE);
3350                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3351                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3352                                              RCL_SERVER, 0);
3353
3354                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3355                  * by default. If the target object has more ACL entries, then
3356                  * enlarge the buffer when necessary. */
3357                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3358                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3359                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3360
3361                 rc = req_capsule_server_pack(pill);
3362         }
3363         RETURN(rc);
3364 }
3365
3366 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3367 {
3368         lh->mlh_type = MDT_NUL_LOCK;
3369         lh->mlh_reg_lh.cookie = 0ull;
3370         lh->mlh_reg_mode = LCK_MINMODE;
3371         lh->mlh_pdo_lh.cookie = 0ull;
3372         lh->mlh_pdo_mode = LCK_MINMODE;
3373         lh->mlh_rreg_lh.cookie = 0ull;
3374         lh->mlh_rreg_mode = LCK_MINMODE;
3375 }
3376
3377 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3378 {
3379         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3380         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3381 }
3382
3383 /*
3384  * Initialize fields of struct mdt_thread_info. Other fields are left in
3385  * uninitialized state, because it's too expensive to zero out whole
3386  * mdt_thread_info (> 1K) on each request arrival.
3387  */
3388 void mdt_thread_info_init(struct ptlrpc_request *req,
3389                           struct mdt_thread_info *info)
3390 {
3391         int i;
3392
3393         info->mti_pill = &req->rq_pill;
3394
3395         /* lock handle */
3396         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3397                 mdt_lock_handle_init(&info->mti_lh[i]);
3398
3399         /* mdt device: it can be NULL while CONNECT */
3400         if (req->rq_export) {
3401                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3402                 info->mti_exp = req->rq_export;
3403         } else
3404                 info->mti_mdt = NULL;
3405         info->mti_env = req->rq_svc_thread->t_env;
3406         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3407
3408         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3409         info->mti_big_buf = LU_BUF_NULL;
3410         info->mti_body = NULL;
3411         info->mti_object = NULL;
3412         info->mti_dlm_req = NULL;
3413         info->mti_has_trans = 0;
3414         info->mti_cross_ref = 0;
3415         info->mti_opdata = 0;
3416         info->mti_big_lmm_used = 0;
3417         info->mti_big_acl_used = 0;
3418         info->mti_som_valid = 0;
3419
3420         info->mti_spec.no_create = 0;
3421         info->mti_spec.sp_rm_entry = 0;
3422         info->mti_spec.sp_permitted = 0;
3423         info->mti_spec.sp_migrate_close = 0;
3424
3425         info->mti_spec.u.sp_ea.eadata = NULL;
3426         info->mti_spec.u.sp_ea.eadatalen = 0;
3427 }
3428
3429 void mdt_thread_info_fini(struct mdt_thread_info *info)
3430 {
3431         int i;
3432
3433         if (info->mti_object != NULL) {
3434                 mdt_object_put(info->mti_env, info->mti_object);
3435                 info->mti_object = NULL;
3436         }
3437
3438         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3439                 mdt_lock_handle_fini(&info->mti_lh[i]);
3440         info->mti_env = NULL;
3441         info->mti_pill = NULL;
3442         info->mti_exp = NULL;
3443
3444         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3445                 lu_buf_free(&info->mti_big_buf);
3446 }
3447
3448 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3449 {
3450         struct mdt_thread_info  *mti;
3451
3452         mti = mdt_th_info(tsi->tsi_env);
3453         LASSERT(mti != NULL);
3454
3455         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3456         if (tsi->tsi_corpus != NULL) {
3457                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3458                 lu_object_get(tsi->tsi_corpus);
3459         }
3460         mti->mti_body = tsi->tsi_mdt_body;
3461         mti->mti_dlm_req = tsi->tsi_dlm_req;
3462
3463         return mti;
3464 }
3465
3466 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3467 {
3468         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3469             cfs_fail_val ==
3470             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3471                 set_current_state(TASK_UNINTERRUPTIBLE);
3472                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3473         }
3474
3475         return tgt_connect(tsi);
3476 }
3477
3478 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3479                               struct mdt_thread_info *info,
3480                               struct ldlm_lock **lockp, __u64 flags)
3481 {
3482         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3483                                    lockp, flags);
3484 }
3485 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3486                           struct mdt_thread_info *info,
3487                           struct ldlm_lock **lockp, __u64 flags)
3488 {
3489         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3490                                lockp, flags);
3491 }
3492
3493 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3494                             struct ldlm_lock **lockp,
3495                             struct mdt_lock_handle *lh,
3496                             __u64 flags, int result)
3497 {
3498         struct ptlrpc_request  *req = mdt_info_req(info);
3499         struct ldlm_lock       *lock = *lockp;
3500         struct ldlm_lock       *new_lock;
3501
3502         /* If possible resent found a lock, @lh is set to its handle */
3503         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3504
3505         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3506                 lh->mlh_reg_lh.cookie = 0;
3507                 RETURN(0);
3508         }
3509
3510         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3511                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3512                  * a resend case, however, it could be already destroyed
3513                  * due to client eviction or a raced cancel RPC. */
3514                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3515                                   lh->mlh_reg_lh.cookie);
3516                 lh->mlh_reg_lh.cookie = 0;
3517                 RETURN(-ESTALE);
3518         }
3519
3520         LASSERTF(new_lock != NULL,
3521                  "lockh %#llx flags %#llx : rc = %d\n",
3522                  lh->mlh_reg_lh.cookie, flags, result);
3523
3524         /*
3525          * If we've already given this lock to a client once, then we should
3526          * have no readers or writers.  Otherwise, we should have one reader
3527          * _or_ writer ref (which will be zeroed below) before returning the
3528          * lock to a client.
3529          */
3530         if (new_lock->l_export == req->rq_export) {
3531                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3532         } else {
3533                 LASSERT(new_lock->l_export == NULL);
3534                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3535         }
3536
3537         *lockp = new_lock;
3538
3539         if (new_lock->l_export == req->rq_export) {
3540                 /*
3541                  * Already gave this to the client, which means that we
3542                  * reconstructed a reply.
3543                  */
3544                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3545                         MSG_RESENT);
3546
3547                 LDLM_LOCK_RELEASE(new_lock);
3548                 lh->mlh_reg_lh.cookie = 0;
3549                 RETURN(ELDLM_LOCK_REPLACED);
3550         }
3551
3552         /*
3553          * Fixup the lock to be given to the client.
3554          */
3555         lock_res_and_lock(new_lock);
3556         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3557          * possible blocking AST. */
3558         while (new_lock->l_readers > 0) {
3559                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3560                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3561                 new_lock->l_readers--;
3562         }
3563         while (new_lock->l_writers > 0) {
3564                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3565                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3566                 new_lock->l_writers--;
3567         }
3568
3569         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3570         new_lock->l_blocking_ast = lock->l_blocking_ast;
3571         new_lock->l_completion_ast = lock->l_completion_ast;
3572         if (ldlm_has_dom(new_lock))
3573                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3574         new_lock->l_remote_handle = lock->l_remote_handle;
3575         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3576
3577         unlock_res_and_lock(new_lock);
3578
3579         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3580                      &new_lock->l_remote_handle,
3581                      &new_lock->l_exp_hash);
3582
3583         LDLM_LOCK_RELEASE(new_lock);
3584         lh->mlh_reg_lh.cookie = 0;
3585
3586         RETURN(ELDLM_LOCK_REPLACED);
3587 }
3588
3589 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3590                              struct ldlm_lock *new_lock,
3591                              struct mdt_lock_handle *lh, __u64 flags)
3592 {
3593         struct ptlrpc_request  *req = mdt_info_req(info);
3594         struct ldlm_request    *dlmreq;
3595
3596         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3597                 return;
3598
3599         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3600
3601         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3602          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3603          * initialized. */
3604         if (flags & LDLM_FL_RESENT) {
3605                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3606                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3607
3608                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3609                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3610                           lh->mlh_reg_lh.cookie);
3611                 return;
3612         }
3613
3614         /*
3615          * If the xid matches, then we know this is a resent request, and allow
3616          * it. (It's probably an OPEN, for which we don't send a lock.
3617          */
3618         if (req_can_reconstruct(req, NULL))
3619                 return;
3620
3621         /*
3622          * This remote handle isn't enqueued, so we never received or processed
3623          * this request.  Clear MSG_RESENT, because it can be handled like any
3624          * normal request now.
3625          */
3626         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3627
3628         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3629                   dlmreq->lock_handle[0].cookie);
3630 }
3631
3632 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3633                                struct mdt_thread_info *info,
3634                                struct ldlm_lock **lockp,
3635                                __u64 flags)
3636 {
3637         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3638         struct ldlm_reply      *ldlm_rep = NULL;
3639         int rc;
3640         ENTRY;
3641
3642         /*
3643          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3644          * (for the resend case) or a new lock. Below we will use it to
3645          * replace the original lock.
3646          */
3647         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3648         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3649                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3650                 rc = mdt_object_lock(info, info->mti_object, lhc,
3651                                      MDS_INODELOCK_XATTR);
3652                 if (rc)
3653                         return rc;
3654         }
3655
3656         rc = mdt_getxattr(info);
3657
3658         if (mdt_info_req(info)->rq_repmsg != NULL)
3659                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3660
3661         if (ldlm_rep == NULL ||
3662             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3663                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3664                 RETURN(err_serious(-EFAULT));
3665         }
3666
3667         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3668
3669         /* This is left for interop instead of adding a new interop flag.
3670          * LU-7433 */
3671 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3672         if (ldlm_rep->lock_policy_res2) {
3673                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3674                 RETURN(ELDLM_LOCK_ABORTED);
3675         }
3676 #endif
3677
3678         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3679         RETURN(rc);
3680 }
3681
3682 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3683                               struct mdt_thread_info *info,
3684                               struct ldlm_lock **lockp,
3685                               __u64 flags)
3686 {
3687         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3688         __u64                   child_bits;
3689         struct ldlm_reply      *ldlm_rep;
3690         struct mdt_body        *reqbody;
3691         struct mdt_body        *repbody;
3692         int                     rc, rc2;
3693         ENTRY;
3694
3695         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3696         LASSERT(reqbody);
3697
3698         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3699         LASSERT(repbody);
3700
3701         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3702         repbody->mbo_eadatasize = 0;
3703         repbody->mbo_aclsize = 0;
3704
3705         switch (it_opc) {
3706         case IT_LOOKUP:
3707                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3708                 break;
3709         case IT_GETATTR:
3710                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3711                              MDS_INODELOCK_PERM;
3712                 break;
3713         default:
3714                 CERROR("%s: unsupported intent %#x\n",
3715                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
3716                 GOTO(out_shrink, rc = -EINVAL);
3717         }
3718
3719         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3720         if (rc)
3721                 GOTO(out_shrink, rc);
3722
3723         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3724         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3725
3726         /* Get lock from request for possible resent case. */
3727         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3728
3729         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3730         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3731
3732         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3733                 ldlm_rep->lock_policy_res2 = 0;
3734         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3735             ldlm_rep->lock_policy_res2) {
3736                 lhc->mlh_reg_lh.cookie = 0ull;
3737                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3738         }
3739
3740         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3741         EXIT;
3742 out_ucred:
3743         mdt_exit_ucred(info);
3744 out_shrink:
3745         mdt_client_compatibility(info);
3746         rc2 = mdt_fix_reply(info);
3747         if (rc == 0)
3748                 rc = rc2;
3749         return rc;
3750 }
3751
3752 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
3753                              struct mdt_thread_info *info,
3754                              struct ldlm_lock **lockp,
3755                              __u64 flags)
3756 {
3757         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3758         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3759         struct layout_intent *intent;
3760         struct lu_fid *fid = &info->mti_tmp_fid2;
3761         struct mdt_object *obj = NULL;
3762         int layout_size = 0;
3763         int rc = 0;
3764         ENTRY;
3765
3766         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3767
3768         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3769         if (intent == NULL)
3770                 RETURN(-EPROTO);
3771
3772         CDEBUG(D_INFO, DFID "got layout change request from client: "
3773                "opc:%u flags:%#x extent "DEXT"\n",
3774                PFID(fid), intent->li_opc, intent->li_flags,
3775                PEXT(&intent->li_extent));
3776
3777         switch (intent->li_opc) {
3778         case LAYOUT_INTENT_TRUNC:
3779         case LAYOUT_INTENT_WRITE:
3780                 layout.mlc_opc = MD_LAYOUT_WRITE;
3781                 layout.mlc_intent = intent;
3782                 break;
3783         case LAYOUT_INTENT_ACCESS:
3784                 break;
3785         case LAYOUT_INTENT_READ:
3786         case LAYOUT_INTENT_GLIMPSE:
3787         case LAYOUT_INTENT_RELEASE:
3788         case LAYOUT_INTENT_RESTORE:
3789                 CERROR("%s: Unsupported layout intent opc %d\n",
3790                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3791                 rc = -ENOTSUPP;
3792                 break;
3793         default:
3794                 CERROR("%s: Unknown layout intent opc %d\n",
3795                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3796                 rc = -EINVAL;
3797                 break;
3798         }
3799         if (rc < 0)
3800                 RETURN(rc);
3801
3802         /* Get lock from request for possible resent case. */
3803         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3804
3805         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3806         if (IS_ERR(obj))
3807                 GOTO(out, rc = PTR_ERR(obj));
3808
3809
3810         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3811                 /* if layout is going to be changed don't use the current EA
3812                  * size but the maximum one. That buffer will be shrinked
3813                  * to the actual size in req_capsule_shrink() before reply.
3814                  */
3815                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
3816                         layout_size = info->mti_mdt->mdt_max_mdsize;
3817                 } else {
3818                         layout_size = mdt_attr_get_eabuf_size(info, obj);
3819                         if (layout_size < 0)
3820                                 GOTO(out_obj, rc = layout_size);
3821
3822                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
3823                                 info->mti_mdt->mdt_max_mdsize = layout_size;
3824                 }
3825                 CDEBUG(D_INFO, "%s: layout_size %d\n",
3826                        mdt_obd_name(info->mti_mdt), layout_size);
3827         }
3828
3829         /*
3830          * set reply buffer size, so that ldlm_handle_enqueue0()->
3831          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3832          */
3833         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3834         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3835                              layout_size);
3836         rc = req_capsule_server_pack(info->mti_pill);
3837         if (rc)
3838                 GOTO(out_obj, rc);
3839
3840
3841         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3842                 struct lu_buf *buf = &layout.mlc_buf;
3843
3844                 /**
3845                  * mdt_layout_change is a reint operation, when the request
3846                  * is resent, layout write shouldn't reprocess it again.
3847                  */
3848                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3849                 if (rc)
3850                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3851
3852                 /**
3853                  * There is another resent case: the client's job has been
3854                  * done by another client, referring lod_declare_layout_change
3855                  * -EALREADY case, and it became a operation w/o transaction,
3856                  * so we should not do the layout change, otherwise
3857                  * mdt_layout_change() will try to cancel the granted server
3858                  * CR lock whose remote counterpart is still in hold on the
3859                  * client, and a deadlock ensues.
3860                  */
3861                 rc = mdt_check_resent_lock(info, obj, lhc);
3862                 if (rc <= 0)
3863                         GOTO(out_obj, rc);
3864
3865                 buf->lb_buf = NULL;
3866                 buf->lb_len = 0;
3867                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3868                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3869                                         &RMF_EADATA);
3870                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3871                                         &RMF_EADATA, RCL_CLIENT);
3872                         /*
3873                          * If it's a replay of layout write intent RPC, the
3874                          * client has saved the extended lovea when
3875                          * it get reply then.
3876                          */
3877                         if (buf->lb_len > 0)
3878                                 mdt_fix_lov_magic(info, buf->lb_buf);
3879                 }
3880                 /*
3881                  * Instantiate some layout components, if @buf contains
3882                  * lovea, then it's a replay of the layout intent write
3883                  * RPC.
3884                  */
3885                 rc = mdt_layout_change(info, obj, &layout);
3886                 if (rc)
3887                         GOTO(out_obj, rc);
3888         }
3889 out_obj:
3890         mdt_object_put(info->mti_env, obj);
3891
3892         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3893                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3894
3895 out:
3896         lhc->mlh_reg_lh.cookie = 0;
3897
3898         RETURN(rc);
3899 }
3900
3901 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
3902                            struct mdt_thread_info *info,
3903                            struct ldlm_lock **lockp,
3904                            __u64 flags)
3905 {
3906         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3907         struct ldlm_reply      *rep = NULL;
3908         long                    opc;
3909         int                     rc;
3910
3911         static const struct req_format *intent_fmts[REINT_MAX] = {
3912                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3913                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3914         };
3915
3916         ENTRY;
3917
3918         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3919         if (opc < 0)
3920                 RETURN(opc);
3921
3922         /* Get lock from request for possible resent case. */
3923         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3924
3925         rc = mdt_reint_internal(info, lhc, opc);
3926
3927         /* Check whether the reply has been packed successfully. */
3928         if (mdt_info_req(info)->rq_repmsg != NULL)
3929                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3930         if (rep == NULL)
3931                 RETURN(err_serious(-EFAULT));
3932
3933         /* MDC expects this in any case */
3934         if (rc != 0)
3935                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3936
3937         /* the open lock or the lock for cross-ref object should be
3938          * returned to the client */
3939         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3940             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3941                 rep->lock_policy_res2 = 0;
3942                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3943                 RETURN(rc);
3944         }
3945
3946         rep->lock_policy_res2 = clear_serious(rc);
3947
3948         if (rep->lock_policy_res2 == -ENOENT &&
3949             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3950             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3951                 rep->lock_policy_res2 = 0;
3952
3953         lhc->mlh_reg_lh.cookie = 0ull;
3954         if (rc == -ENOTCONN || rc == -ENODEV ||
3955             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3956                 /*
3957                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3958                  * will be returned by rq_status, and client at ptlrpc layer
3959                  * will detect this, then disconnect, reconnect the import
3960                  * immediately, instead of impacting the following the rpc.
3961                  */
3962                 RETURN(rc);
3963         }
3964         /*
3965          * For other cases, the error will be returned by intent, and client
3966          * will retrieve the result from intent.
3967          */
3968         RETURN(ELDLM_LOCK_ABORTED);
3969 }
3970
3971 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
3972                           struct mdt_thread_info *info,
3973                           struct ldlm_lock **lockp,
3974                           u64 flags /* LDLM_FL_* */)
3975 {
3976         struct req_capsule *pill = info->mti_pill;
3977         struct ptlrpc_request *req = mdt_info_req(info);
3978         const struct req_format *it_format;
3979         int (*it_handler)(enum ldlm_intent_flags,
3980                           struct mdt_thread_info *,
3981                           struct ldlm_lock **,
3982                           u64);
3983         enum tgt_handler_flags it_handler_flags = 0;
3984         struct ldlm_reply *rep;
3985         int rc;
3986         ENTRY;
3987
3988         switch (it_opc) {
3989         case IT_OPEN:
3990         case IT_OPEN|IT_CREAT:
3991                 /*
3992                  * OCREAT is not a MUTABOR request since the file may
3993                  * already exist. We do the extra check of
3994                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
3995                  * really need to create the object.
3996                  */
3997                 it_format = &RQF_LDLM_INTENT;
3998                 it_handler = &mdt_intent_open;
3999                 break;
4000         case IT_GETATTR:
4001         case IT_LOOKUP:
4002                 it_format = &RQF_LDLM_INTENT_GETATTR;
4003                 it_handler = &mdt_intent_getattr;
4004                 it_handler_flags = HABEO_REFERO;
4005                 break;
4006         case IT_GETXATTR:
4007                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4008                 it_handler = &mdt_intent_getxattr;
4009                 it_handler_flags = HABEO_CORPUS;
4010                 break;
4011         case IT_LAYOUT:
4012                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4013                 it_handler = &mdt_intent_layout;
4014                 break;
4015         case IT_GLIMPSE:
4016                 it_format = &RQF_LDLM_INTENT;
4017                 it_handler = &mdt_intent_glimpse;
4018                 break;
4019         case IT_BRW:
4020                 it_format = &RQF_LDLM_INTENT;
4021                 it_handler = &mdt_intent_brw;
4022                 break;
4023         case IT_QUOTA_DQACQ:
4024         case IT_QUOTA_CONN: {
4025                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4026
4027                 if (qmt == NULL)
4028                         RETURN(-EOPNOTSUPP);
4029
4030                 if (mdt_rdonly(req->rq_export))
4031                         RETURN(-EROFS);
4032
4033                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4034                 /* pass the request to quota master */
4035                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4036                                                  mdt_info_req(info), lockp,
4037                                                  flags);
4038                 RETURN(rc);
4039         }
4040         default:
4041                 CERROR("%s: unknown intent code %#x\n",
4042                        mdt_obd_name(info->mti_mdt), it_opc);
4043                 RETURN(-EPROTO);
4044         }
4045
4046         req_capsule_extend(pill, it_format);
4047
4048         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4049         if (rc < 0)
4050                 RETURN(rc);
4051
4052         if (it_handler_flags & MUTABOR && mdt_rdonly(req->rq_export))
4053                 RETURN(-EROFS);
4054
4055         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4056
4057         /* execute policy */
4058         rc = (*it_handler)(it_opc, info, lockp, flags);
4059
4060         /* Check whether the reply has been packed successfully. */
4061         if (req->rq_repmsg != NULL) {
4062                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4063                 rep->lock_policy_res2 =
4064                         ptlrpc_status_hton(rep->lock_policy_res2);
4065         }
4066
4067         RETURN(rc);
4068 }
4069
4070 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4071                                     enum ldlm_intent_flags it_opc)
4072 {
4073         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4074
4075         /* update stats when IT code is known */
4076         if (srv_stats != NULL)
4077                 lprocfs_counter_incr(srv_stats,
4078                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4079                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4080 }
4081
4082 static int mdt_intent_policy(const struct lu_env *env,
4083                              struct ldlm_namespace *ns,
4084                              struct ldlm_lock **lockp,
4085                              void *req_cookie,
4086                              enum ldlm_mode mode,
4087                              __u64 flags, void *data)
4088 {
4089         struct tgt_session_info *tsi;
4090         struct mdt_thread_info  *info;
4091         struct ptlrpc_request   *req  =  req_cookie;
4092         struct ldlm_intent      *it;
4093         struct req_capsule      *pill;
4094         const struct ldlm_lock_desc *ldesc;
4095         int rc;
4096
4097         ENTRY;
4098
4099         LASSERT(req != NULL);
4100
4101         tsi = tgt_ses_info(env);
4102
4103         info = tsi2mdt_info(tsi);
4104         LASSERT(info != NULL);
4105         pill = info->mti_pill;
4106         LASSERT(pill->rc_req == req);
4107         ldesc = &info->mti_dlm_req->lock_desc;
4108
4109         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4110                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4111                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4112                 if (it != NULL) {
4113                         mdt_ptlrpc_stats_update(req, it->opc);
4114                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4115                         if (rc == 0)
4116                                 rc = ELDLM_OK;
4117
4118                         /* Lock without inodebits makes no sense and will oops
4119                          * later in ldlm. Let's check it now to see if we have
4120                          * ibits corrupted somewhere in mdt_intent_opc().
4121                          * The case for client miss to set ibits has been
4122                          * processed by others. */
4123                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4124                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4125                 } else {
4126                         rc = err_serious(-EFAULT);
4127                 }
4128         } else {
4129                 /* No intent was provided */
4130                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4131                 rc = req_capsule_server_pack(pill);
4132                 if (rc)
4133                         rc = err_serious(rc);
4134         }
4135         mdt_thread_info_fini(info);
4136         RETURN(rc);
4137 }
4138
4139 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4140 {
4141         struct seq_server_site  *ss = mdt_seq_site(mdt);
4142
4143         if (ss->ss_node_id == 0)
4144                 return;
4145
4146         if (ss->ss_client_seq != NULL) {
4147                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4148                 ss->ss_client_seq->lcs_exp = NULL;
4149         }
4150
4151         if (ss->ss_server_fld != NULL) {
4152                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4153                 ss->ss_server_fld->lsf_control_exp = NULL;
4154         }
4155 }
4156
4157 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4158 {
4159         struct seq_server_site *ss = mdt_seq_site(mdt);
4160
4161         if (ss == NULL)
4162                 return;
4163
4164         if (ss->ss_server_seq != NULL)
4165                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4166 }
4167
4168 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4169 {
4170         mdt_seq_fini_cli(mdt);
4171         mdt_deregister_seq_exp(mdt);
4172
4173         return seq_site_fini(env, mdt_seq_site(mdt));
4174 }
4175
4176 /**
4177  * It will retrieve its FLDB entries from MDT0, and it only happens
4178  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4179  * and it needs to refresh FLDB from the MDT0.
4180  **/
4181 static int mdt_register_lwp_callback(void *data)
4182 {
4183         struct lu_env           env;
4184         struct mdt_device       *mdt = data;
4185         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4186         int                     rc;
4187         ENTRY;
4188
4189         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4190
4191         rc = lu_env_init(&env, LCT_MD_THREAD);
4192         if (rc < 0) {
4193                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4194                 RETURN(rc);
4195         }
4196
4197         /* Allocate new sequence now to avoid creating local transaction
4198          * in the normal transaction process */
4199         rc = seq_server_check_and_alloc_super(&env,
4200                                               mdt_seq_site(mdt)->ss_server_seq);
4201         if (rc < 0)
4202                 GOTO(out, rc);
4203
4204         if (fld->lsf_new) {
4205                 rc = fld_update_from_controller(&env, fld);
4206                 if (rc != 0) {
4207                         CERROR("%s: cannot update controller: rc = %d\n",
4208                                mdt_obd_name(mdt), rc);
4209                         GOTO(out, rc);
4210                 }
4211         }
4212 out:
4213         lu_env_fini(&env);
4214         RETURN(rc);
4215 }
4216
4217 static int mdt_register_seq_exp(struct mdt_device *mdt)
4218 {
4219         struct seq_server_site  *ss = mdt_seq_site(mdt);
4220         char                    *lwp_name = NULL;
4221         int                     rc;
4222
4223         if (ss->ss_node_id == 0)
4224                 return 0;
4225
4226         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4227         if (lwp_name == NULL)
4228                 GOTO(out_free, rc = -ENOMEM);
4229
4230         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4231         if (rc != 0)
4232                 GOTO(out_free, rc);
4233
4234         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4235                                       NULL, NULL);
4236         if (rc != 0)
4237                 GOTO(out_free, rc);
4238
4239         rc = lustre_register_lwp_item(lwp_name,
4240                                       &ss->ss_server_fld->lsf_control_exp,
4241                                       mdt_register_lwp_callback, mdt);
4242         if (rc != 0) {
4243                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4244                 ss->ss_client_seq->lcs_exp = NULL;
4245                 GOTO(out_free, rc);
4246         }
4247 out_free:
4248         if (lwp_name != NULL)
4249                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4250
4251         return rc;
4252 }
4253
4254 /*
4255  * Init client sequence manager which is used by local MDS to talk to sequence
4256  * controller on remote node.
4257  */
4258 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4259 {
4260         struct seq_server_site  *ss = mdt_seq_site(mdt);
4261         int                     rc;
4262         char                    *prefix;
4263         ENTRY;
4264
4265         /* check if this is adding the first MDC and controller is not yet
4266          * initialized. */
4267         OBD_ALLOC_PTR(ss->ss_client_seq);
4268         if (ss->ss_client_seq == NULL)
4269                 RETURN(-ENOMEM);
4270
4271         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4272         if (prefix == NULL) {
4273                 OBD_FREE_PTR(ss->ss_client_seq);
4274                 ss->ss_client_seq = NULL;
4275                 RETURN(-ENOMEM);
4276         }
4277
4278         /* Note: seq_client_fini will be called in seq_site_fini */
4279         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4280         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4281                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4282                                                             NULL);
4283         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4284         if (rc != 0) {
4285                 OBD_FREE_PTR(ss->ss_client_seq);
4286                 ss->ss_client_seq = NULL;
4287                 RETURN(rc);
4288         }
4289
4290         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4291
4292         RETURN(rc);
4293 }
4294
4295 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4296 {
4297         struct seq_server_site  *ss;
4298         int                     rc;
4299         ENTRY;
4300
4301         ss = mdt_seq_site(mdt);
4302         /* init sequence controller server(MDT0) */
4303         if (ss->ss_node_id == 0) {
4304                 OBD_ALLOC_PTR(ss->ss_control_seq);
4305                 if (ss->ss_control_seq == NULL)
4306                         RETURN(-ENOMEM);
4307
4308                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4309                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4310                                      ss);
4311                 if (rc)
4312                         GOTO(out_seq_fini, rc);
4313         }
4314
4315         /* Init normal sequence server */
4316         OBD_ALLOC_PTR(ss->ss_server_seq);
4317         if (ss->ss_server_seq == NULL)
4318                 GOTO(out_seq_fini, rc = -ENOMEM);
4319
4320         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4321                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4322         if (rc)
4323                 GOTO(out_seq_fini, rc);
4324
4325         /* init seq client for seq server to talk to seq controller(MDT0) */
4326         rc = mdt_seq_init_cli(env, mdt);
4327         if (rc != 0)
4328                 GOTO(out_seq_fini, rc);
4329
4330         if (ss->ss_node_id != 0)
4331                 /* register controller export through lwp */
4332                 rc = mdt_register_seq_exp(mdt);
4333
4334         EXIT;
4335 out_seq_fini:
4336         if (rc)
4337                 mdt_seq_fini(env, mdt);
4338
4339         return rc;
4340 }
4341
4342 /*
4343  * FLD wrappers
4344  */
4345 static int mdt_fld_fini(const struct lu_env *env,
4346                         struct mdt_device *m)
4347 {
4348         struct seq_server_site *ss = mdt_seq_site(m);
4349         ENTRY;
4350
4351         if (ss && ss->ss_server_fld) {
4352                 fld_server_fini(env, ss->ss_server_fld);
4353                 OBD_FREE_PTR(ss->ss_server_fld);
4354                 ss->ss_server_fld = NULL;
4355         }
4356
4357         RETURN(0);
4358 }
4359
4360 static int mdt_fld_init(const struct lu_env *env,
4361                         const char *uuid,
4362                         struct mdt_device *m)
4363 {
4364         struct seq_server_site *ss;
4365         int rc;
4366         ENTRY;
4367
4368         ss = mdt_seq_site(m);
4369
4370         OBD_ALLOC_PTR(ss->ss_server_fld);
4371         if (ss->ss_server_fld == NULL)
4372                 RETURN(rc = -ENOMEM);
4373
4374         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4375                              LU_SEQ_RANGE_MDT);
4376         if (rc) {
4377                 OBD_FREE_PTR(ss->ss_server_fld);
4378                 ss->ss_server_fld = NULL;
4379                 RETURN(rc);
4380         }
4381
4382         RETURN(0);
4383 }
4384
4385 static void mdt_stack_pre_fini(const struct lu_env *env,
4386                            struct mdt_device *m, struct lu_device *top)
4387 {
4388         struct lustre_cfg_bufs  *bufs;
4389         struct lustre_cfg       *lcfg;
4390         struct mdt_thread_info  *info;
4391         ENTRY;
4392
4393         LASSERT(top);
4394
4395         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4396         LASSERT(info != NULL);
4397
4398         bufs = &info->mti_u.bufs;
4399
4400         LASSERT(m->mdt_child_exp);
4401         LASSERT(m->mdt_child_exp->exp_obd);
4402
4403         /* process cleanup, pass mdt obd name to get obd umount flags */
4404         /* XXX: this is needed because all layers are referenced by
4405          * objects (some of them are pinned by osd, for example *
4406          * the proper solution should be a model where object used
4407          * by osd only doesn't have mdt/mdd slices -bzzz */
4408         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4409         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4410         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4411         if (!lcfg)
4412                 RETURN_EXIT;
4413         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4414
4415         top->ld_ops->ldo_process_config(env, top, lcfg);
4416         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4417         EXIT;
4418 }
4419
4420 static void mdt_stack_fini(const struct lu_env *env,
4421                            struct mdt_device *m, struct lu_device *top)
4422 {
4423         struct obd_device       *obd = mdt2obd_dev(m);
4424         struct lustre_cfg_bufs  *bufs;
4425         struct lustre_cfg       *lcfg;
4426         struct mdt_thread_info  *info;
4427         char                     flags[3] = "";
4428         ENTRY;
4429
4430         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4431         LASSERT(info != NULL);
4432
4433         lu_dev_del_linkage(top->ld_site, top);
4434
4435         lu_site_purge(env, top->ld_site, -1);
4436
4437         bufs = &info->mti_u.bufs;
4438         /* process cleanup, pass mdt obd name to get obd umount flags */
4439         /* another purpose is to let all layers to release their objects */
4440         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4441         if (obd->obd_force)
4442                 strcat(flags, "F");
4443         if (obd->obd_fail)
4444                 strcat(flags, "A");
4445         lustre_cfg_bufs_set_string(bufs, 1, flags);
4446         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4447         if (!lcfg)
4448                 RETURN_EXIT;
4449         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4450
4451         LASSERT(top);
4452         top->ld_ops->ldo_process_config(env, top, lcfg);
4453         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4454
4455         lu_site_purge(env, top->ld_site, -1);
4456
4457         m->mdt_child = NULL;
4458         m->mdt_bottom = NULL;
4459
4460         obd_disconnect(m->mdt_child_exp);
4461         m->mdt_child_exp = NULL;
4462
4463         obd_disconnect(m->mdt_bottom_exp);
4464         m->mdt_child_exp = NULL;
4465 }
4466
4467 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4468                                const char *next, struct obd_export **exp)
4469 {
4470         struct obd_connect_data *data = NULL;
4471         struct obd_device       *obd;
4472         int                      rc;
4473         ENTRY;
4474
4475         OBD_ALLOC_PTR(data);
4476         if (data == NULL)
4477                 GOTO(out, rc = -ENOMEM);
4478
4479         obd = class_name2obd(next);
4480         if (obd == NULL) {
4481                 CERROR("%s: can't locate next device: %s\n",
4482                        mdt_obd_name(m), next);
4483                 GOTO(out, rc = -ENOTCONN);
4484         }
4485
4486         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4487         data->ocd_version = LUSTRE_VERSION_CODE;
4488
4489         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4490         if (rc) {
4491                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4492                        mdt_obd_name(m), next, rc);
4493                 GOTO(out, rc);
4494         }
4495
4496 out:
4497         if (data)
4498                 OBD_FREE_PTR(data);
4499         RETURN(rc);
4500 }
4501
4502 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4503                           struct lustre_cfg *cfg)
4504 {
4505         char                   *dev = lustre_cfg_string(cfg, 0);
4506         int                     rc, name_size, uuid_size;
4507         char                   *name, *uuid, *p;
4508         struct lustre_cfg_bufs *bufs;
4509         struct lustre_cfg      *lcfg;
4510         struct obd_device      *obd;
4511         struct lustre_profile  *lprof;
4512         struct lu_site         *site;
4513         ENTRY;
4514
4515         /* in 1.8 we had the only device in the stack - MDS.
4516          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4517          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4518          * to generate names and setup MDT, MDD. MDT will be using
4519          * generated name to connect to MDD. for MDD the next device
4520          * will be LOD with name taken from so called "profile" which
4521          * is generated by mount_option line
4522          *
4523          * 1.8 MGS generates config. commands like this:
4524          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4525          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4526          * 2.0 MGS generates config. commands like this:
4527          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4528          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4529          *                    3:lustre-MDT0000-mdtlov  4:f
4530          *
4531          * we generate MDD name from MDT one, just replacing T with D
4532          *
4533          * after all the preparations, the logical equivalent will be
4534          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4535          *                    3:lustre-MDT0000-mdtlov  4:f
4536          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4537          *                    3:lustre-MDD0000  4:f
4538          *
4539          *  notice we build the stack from down to top: MDD first, then MDT */
4540
4541         name_size = MAX_OBD_NAME;
4542         uuid_size = MAX_OBD_NAME;
4543
4544         OBD_ALLOC(name, name_size);
4545         OBD_ALLOC(uuid, uuid_size);
4546         if (name == NULL || uuid == NULL)
4547                 GOTO(cleanup_mem, rc = -ENOMEM);
4548
4549         OBD_ALLOC_PTR(bufs);
4550         if (!bufs)
4551                 GOTO(cleanup_mem, rc = -ENOMEM);
4552
4553         strcpy(name, dev);
4554         p = strstr(name, "-MDT");
4555         if (p == NULL)
4556                 GOTO(free_bufs, rc = -ENOMEM);
4557         p[3] = 'D';
4558
4559         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4560
4561         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4562         if (lprof == NULL || lprof->lp_dt == NULL) {
4563                 CERROR("can't find the profile: %s\n",
4564                        lustre_cfg_string(cfg, 0));
4565                 GOTO(free_bufs, rc = -EINVAL);
4566         }
4567
4568         lustre_cfg_bufs_reset(bufs, name);
4569         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4570         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4571         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4572
4573         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4574         if (!lcfg)
4575                 GOTO(put_profile, rc = -ENOMEM);
4576         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4577
4578         rc = class_attach(lcfg);
4579         if (rc)
4580                 GOTO(lcfg_cleanup, rc);
4581
4582         obd = class_name2obd(name);
4583         if (!obd) {
4584                 CERROR("Can not find obd %s (%s in config)\n",
4585                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4586                 GOTO(lcfg_cleanup, rc = -EINVAL);
4587         }
4588
4589         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4590
4591         lustre_cfg_bufs_reset(bufs, name);
4592         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4593         lustre_cfg_bufs_set_string(bufs, 2, dev);
4594         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4595
4596         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4597         if (!lcfg)
4598                 GOTO(class_detach, rc = -ENOMEM);
4599         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4600
4601         rc = class_setup(obd, lcfg);
4602         if (rc)
4603                 GOTO(class_detach, rc);
4604
4605         /* connect to MDD we just setup */
4606         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4607         if (rc)
4608                 GOTO(class_detach, rc);
4609
4610         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4611         LASSERT(site);
4612         LASSERT(mdt_lu_site(mdt) == NULL);
4613         mdt->mdt_lu_dev.ld_site = site;
4614         site->ls_top_dev = &mdt->mdt_lu_dev;
4615         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4616
4617         /* now connect to bottom OSD */
4618         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4619         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4620         if (rc)
4621                 GOTO(class_detach, rc);
4622         mdt->mdt_bottom =
4623                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4624
4625         rc = lu_env_refill((struct lu_env *)env);
4626         if (rc != 0)
4627                 CERROR("Failure to refill session: '%d'\n", rc);
4628
4629         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4630
4631         EXIT;
4632 class_detach:
4633         if (rc)
4634                 class_detach(obd, lcfg);
4635 lcfg_cleanup:
4636         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4637 put_profile:
4638         class_put_profile(lprof);
4639 free_bufs:
4640         OBD_FREE_PTR(bufs);
4641 cleanup_mem:
4642         if (name)
4643                 OBD_FREE(name, name_size);
4644         if (uuid)
4645                 OBD_FREE(uuid, uuid_size);
4646         RETURN(rc);
4647 }
4648
4649 /* setup quota master target on MDT0 */
4650 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4651                           struct lustre_cfg *cfg)
4652 {
4653         struct obd_device       *obd;
4654         char                    *dev = lustre_cfg_string(cfg, 0);
4655         char                    *qmtname, *uuid, *p;
4656         struct lustre_cfg_bufs  *bufs;
4657         struct lustre_cfg       *lcfg;
4658         struct lustre_profile   *lprof;
4659         struct obd_connect_data *data;
4660         int                      rc;
4661         ENTRY;
4662
4663         LASSERT(mdt->mdt_qmt_exp == NULL);
4664         LASSERT(mdt->mdt_qmt_dev == NULL);
4665
4666         /* quota master is on MDT0 only for now */
4667         if (mdt->mdt_seq_site.ss_node_id != 0)
4668                 RETURN(0);
4669
4670         /* MGS generates config commands which look as follows:
4671          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4672          *                    3:lustre-MDT0000-mdtlov  4:f
4673          *
4674          * We generate the QMT name from the MDT one, just replacing MD with QM
4675          * after all the preparations, the logical equivalent will be:
4676          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4677          *                    3:lustre-MDT0000-osd  4:f */
4678         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4679         OBD_ALLOC(uuid, UUID_MAX);
4680         OBD_ALLOC_PTR(bufs);
4681         OBD_ALLOC_PTR(data);
4682         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4683                 GOTO(cleanup_mem, rc = -ENOMEM);
4684
4685         strcpy(qmtname, dev);
4686         p = strstr(qmtname, "-MDT");
4687         if (p == NULL)
4688                 GOTO(cleanup_mem, rc = -ENOMEM);
4689         /* replace MD with QM */
4690         p[1] = 'Q';
4691         p[2] = 'M';
4692
4693         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4694
4695         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4696         if (lprof == NULL || lprof->lp_dt == NULL) {
4697                 CERROR("can't find profile for %s\n",
4698                        lustre_cfg_string(cfg, 0));
4699                 GOTO(cleanup_mem, rc = -EINVAL);
4700         }
4701
4702         lustre_cfg_bufs_reset(bufs, qmtname);
4703         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4704         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4705         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4706
4707         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4708         if (!lcfg)
4709                 GOTO(put_profile, rc = -ENOMEM);
4710         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4711
4712         rc = class_attach(lcfg);
4713         if (rc)
4714                 GOTO(lcfg_cleanup, rc);
4715
4716         obd = class_name2obd(qmtname);
4717         if (!obd) {
4718                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4719                        lustre_cfg_string(cfg, 0));
4720                 GOTO(lcfg_cleanup, rc = -EINVAL);
4721         }
4722
4723         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4724
4725         lustre_cfg_bufs_reset(bufs, qmtname);
4726         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4727         lustre_cfg_bufs_set_string(bufs, 2, dev);
4728
4729         /* for quota, the next device should be the OSD device */
4730         lustre_cfg_bufs_set_string(bufs, 3,
4731                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4732
4733         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4734         if (!lcfg)
4735                 GOTO(class_detach, rc = -ENOMEM);
4736         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4737
4738         rc = class_setup(obd, lcfg);
4739         if (rc)
4740                 GOTO(class_detach, rc);
4741
4742         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4743
4744         /* configure local quota objects */
4745         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4746                                                    &mdt->mdt_lu_dev,
4747                                                    mdt->mdt_qmt_dev);
4748         if (rc)
4749                 GOTO(class_cleanup, rc);
4750
4751         /* connect to quota master target */
4752         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4753         data->ocd_version = LUSTRE_VERSION_CODE;
4754         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4755                          data, NULL);
4756         if (rc) {
4757                 CERROR("cannot connect to quota master device %s (%d)\n",
4758                        qmtname, rc);
4759                 GOTO(class_cleanup, rc);
4760         }
4761
4762         EXIT;
4763 class_cleanup:
4764         if (rc) {
4765                 class_manual_cleanup(obd);
4766                 mdt->mdt_qmt_dev = NULL;
4767         }
4768 class_detach:
4769         if (rc)
4770                 class_detach(obd, lcfg);
4771 lcfg_cleanup:
4772         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4773 put_profile:
4774         class_put_profile(lprof);
4775 cleanup_mem:
4776         if (bufs)
4777                 OBD_FREE_PTR(bufs);
4778         if (qmtname)
4779                 OBD_FREE(qmtname, MAX_OBD_NAME);
4780         if (uuid)
4781                 OBD_FREE(uuid, UUID_MAX);
4782         if (data)
4783                 OBD_FREE_PTR(data);
4784         return rc;
4785 }
4786
4787 /* Shutdown quota master target associated with mdt */
4788 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4789 {
4790         ENTRY;
4791
4792         if (mdt->mdt_qmt_exp == NULL)
4793                 RETURN_EXIT;
4794         LASSERT(mdt->mdt_qmt_dev != NULL);
4795
4796         /* the qmt automatically shuts down when the mdt disconnects */
4797         obd_disconnect(mdt->mdt_qmt_exp);
4798         mdt->mdt_qmt_exp = NULL;
4799         mdt->mdt_qmt_dev = NULL;
4800         EXIT;
4801 }
4802
4803 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4804  * for now. This will be removed along with converting rest of MDT code
4805  * to use tgt_session_info */
4806 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4807 {
4808         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4809         int                      rc;
4810
4811         rc = mdt_getxattr(info);
4812
4813         mdt_thread_info_fini(info);
4814         return rc;
4815 }
4816
4817 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4818 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4819 #define OST_BRW_READ    OST_READ
4820 #define OST_BRW_WRITE   OST_WRITE
4821
4822 static struct tgt_handler mdt_tgt_handlers[] = {
4823 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4824                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4825                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4826 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4827                 0,                      MDS_DISCONNECT, tgt_disconnect,
4828                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4829 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4830                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4831                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4832 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4833 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4834 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4835 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4836                                                         mdt_getattr_name),
4837 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4838 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4839 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4840 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4841 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4842 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4843 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4844 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4845                                                         mdt_hsm_progress),
4846 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4847                                                         mdt_hsm_ct_register),
4848 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4849                                                         mdt_hsm_ct_unregister),
4850 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4851                                                         mdt_hsm_state_get),
4852 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4853                                                         mdt_hsm_state_set),
4854 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4855 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4856                                                         mdt_hsm_request),
4857 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4858             MDS_SWAP_LAYOUTS,
4859             mdt_swap_layouts),
4860 };
4861
4862 static struct tgt_handler mdt_io_ops[] = {
4863 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ, tgt_brw_read,
4864                                                         mdt_hp_brw),
4865 TGT_OST_HDL_HP(HABEO_CORPUS | MUTABOR,   OST_BRW_WRITE, tgt_brw_write,
4866                                                         mdt_hp_brw),
4867 TGT_OST_HDL_HP(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4868                                          OST_PUNCH,     mdt_punch_hdl,
4869                                                         mdt_hp_punch),
4870 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4871 };
4872
4873 static struct tgt_handler mdt_sec_ctx_ops[] = {
4874 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4875 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4876 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4877 };
4878
4879 static struct tgt_handler mdt_quota_ops[] = {
4880 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4881 };
4882
4883 static struct tgt_opc_slice mdt_common_slice[] = {
4884         {
4885                 .tos_opc_start  = MDS_FIRST_OPC,
4886                 .tos_opc_end    = MDS_LAST_OPC,
4887                 .tos_hs         = mdt_tgt_handlers
4888         },
4889         {
4890                 .tos_opc_start  = OBD_FIRST_OPC,
4891                 .tos_opc_end    = OBD_LAST_OPC,
4892                 .tos_hs         = tgt_obd_handlers
4893         },
4894         {
4895                 .tos_opc_start  = LDLM_FIRST_OPC,
4896                 .tos_opc_end    = LDLM_LAST_OPC,
4897                 .tos_hs         = tgt_dlm_handlers
4898         },
4899         {
4900                 .tos_opc_start  = SEC_FIRST_OPC,
4901                 .tos_opc_end    = SEC_LAST_OPC,
4902                 .tos_hs         = mdt_sec_ctx_ops
4903         },
4904         {
4905                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4906                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4907                 .tos_hs         = tgt_out_handlers
4908         },
4909         {
4910                 .tos_opc_start  = FLD_FIRST_OPC,
4911                 .tos_opc_end    = FLD_LAST_OPC,
4912                 .tos_hs         = fld_handlers
4913         },
4914         {
4915                 .tos_opc_start  = SEQ_FIRST_OPC,
4916                 .tos_opc_end    = SEQ_LAST_OPC,
4917                 .tos_hs         = seq_handlers
4918         },
4919         {
4920                 .tos_opc_start  = QUOTA_DQACQ,
4921                 .tos_opc_end    = QUOTA_LAST_OPC,
4922                 .tos_hs         = mdt_quota_ops
4923         },
4924         {
4925                 .tos_opc_start  = LLOG_FIRST_OPC,
4926                 .tos_opc_end    = LLOG_LAST_OPC,
4927                 .tos_hs         = tgt_llog_handlers
4928         },
4929         {
4930                 .tos_opc_start  = LFSCK_FIRST_OPC,
4931                 .tos_opc_end    = LFSCK_LAST_OPC,
4932                 .tos_hs         = tgt_lfsck_handlers
4933         },
4934         {
4935                 .tos_opc_start  = OST_FIRST_OPC,
4936                 .tos_opc_end    = OST_LAST_OPC,
4937                 .tos_hs         = mdt_io_ops
4938         },
4939         {
4940                 .tos_hs         = NULL
4941         }
4942 };
4943
4944 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4945 {
4946         struct md_device *next = m->mdt_child;
4947         struct lu_device *d = &m->mdt_lu_dev;
4948         struct obd_device *obd = mdt2obd_dev(m);
4949         struct lfsck_stop stop;
4950
4951         ENTRY;
4952         stop.ls_status = LS_PAUSED;
4953         stop.ls_flags = 0;
4954         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4955
4956         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4957         ping_evictor_stop();
4958
4959         /* Remove the HSM /proc entry so the coordinator cannot be
4960          * restarted by a user while it's shutting down. */
4961         hsm_cdt_procfs_fini(m);
4962         mdt_hsm_cdt_stop(m);
4963
4964         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4965         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4966
4967         if (m->mdt_namespace != NULL)
4968                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4969                                           d->ld_obd->obd_force);
4970
4971         obd_exports_barrier(obd);
4972         obd_zombie_barrier();
4973
4974         mdt_quota_fini(env, m);
4975
4976         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4977
4978         /* Calling the cleanup functions in the same order as in the mdt_init0
4979          * error path
4980          */
4981         mdt_procfs_fini(m);
4982
4983         target_recovery_fini(obd);
4984         upcall_cache_cleanup(m->mdt_identity_cache);
4985         m->mdt_identity_cache = NULL;
4986
4987         mdt_fs_cleanup(env, m);
4988
4989         tgt_fini(env, &m->mdt_lut);
4990
4991         mdt_hsm_cdt_fini(m);
4992
4993         if (m->mdt_los != NULL) {
4994                 local_oid_storage_fini(env, m->mdt_los);
4995                 m->mdt_los = NULL;
4996         }
4997
4998         if (m->mdt_namespace != NULL) {
4999                 ldlm_namespace_free_post(m->mdt_namespace);
5000                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5001         }
5002
5003         if (m->mdt_md_root != NULL) {
5004                 mdt_object_put(env, m->mdt_md_root);
5005                 m->mdt_md_root = NULL;
5006         }
5007
5008         mdt_seq_fini(env, m);
5009
5010         mdt_fld_fini(env, m);
5011
5012         /*
5013          * Finish the stack
5014          */
5015         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5016
5017         LASSERT(atomic_read(&d->ld_ref) == 0);
5018
5019         server_put_mount(mdt_obd_name(m), true);
5020
5021         EXIT;
5022 }
5023
5024 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5025
5026 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5027                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5028 {
5029         const struct dt_device_param *dt_conf;
5030         struct mdt_thread_info *info;
5031         struct obd_device *obd;
5032         const char *dev = lustre_cfg_string(cfg, 0);
5033         const char *num = lustre_cfg_string(cfg, 2);
5034         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5035         struct lustre_mount_info *lmi = NULL;
5036         struct lustre_sb_info *lsi;
5037         struct lu_site *s;
5038         struct seq_server_site *ss_site;
5039         const char *identity_upcall = "NONE";
5040         struct md_device *next;
5041         struct lu_fid fid;
5042         int rc;
5043         long node_id;
5044         mntopt_t mntopts;
5045         ENTRY;
5046
5047         lu_device_init(&m->mdt_lu_dev, ldt);
5048         /*
5049          * Environment (env) might be missing mdt_thread_key values at that
5050          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5051          * mode.
5052          *
5053          * Usually device allocation path doesn't use module key values, but
5054          * mdt has to do a lot of work here, so allocate key value.
5055          */
5056         rc = lu_env_refill((struct lu_env *)env);
5057         if (rc != 0)
5058                 RETURN(rc);
5059
5060         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5061         LASSERT(info != NULL);
5062
5063         obd = class_name2obd(dev);
5064         LASSERT(obd != NULL);
5065
5066         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5067         m->mdt_opts.mo_evict_tgt_nids = 1;
5068         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5069
5070         lmi = server_get_mount(dev);
5071         if (lmi == NULL) {
5072                 CERROR("Cannot get mount info for %s!\n", dev);
5073                 RETURN(-EFAULT);
5074         } else {
5075                 lsi = s2lsi(lmi->lmi_sb);
5076                 /* CMD is supported only in IAM mode */
5077                 LASSERT(num);
5078                 node_id = simple_strtol(num, NULL, 10);
5079                 obd->u.obt.obt_magic = OBT_MAGIC;
5080                 if (lsi->lsi_lmd != NULL &&
5081                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5082                         m->mdt_skip_lfsck = 1;
5083         }
5084
5085         /* DoM files get IO lock at open by default */
5086         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5087         /* DoM files are read at open and data is packed in the reply */
5088         m->mdt_opts.mo_dom_read_open = 1;
5089
5090         m->mdt_squash.rsi_uid = 0;
5091         m->mdt_squash.rsi_gid = 0;
5092         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5093         init_rwsem(&m->mdt_squash.rsi_sem);
5094         spin_lock_init(&m->mdt_lock);
5095         m->mdt_enable_remote_dir = 1;
5096         m->mdt_enable_striped_dir = 1;
5097         m->mdt_enable_dir_migration = 1;
5098         m->mdt_enable_remote_dir_gid = 0;
5099
5100         atomic_set(&m->mdt_mds_mds_conns, 0);
5101         atomic_set(&m->mdt_async_commit_count, 0);
5102
5103         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5104         m->mdt_lu_dev.ld_obd = obd;
5105         /* Set this lu_device to obd for error handling purposes. */
5106         obd->obd_lu_dev = &m->mdt_lu_dev;
5107
5108         /* init the stack */
5109         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5110         if (rc) {
5111                 CERROR("%s: Can't init device stack, rc %d\n",
5112                        mdt_obd_name(m), rc);
5113                 GOTO(err_lmi, rc);
5114         }
5115
5116         s = mdt_lu_site(m);
5117         ss_site = mdt_seq_site(m);
5118         s->ld_seq_site = ss_site;
5119         ss_site->ss_lu = s;
5120
5121         /* set server index */
5122         ss_site->ss_node_id = node_id;
5123
5124         /* failover is the default
5125          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5126          * assumed whose ss_node_id == 0 XXX
5127          * */
5128         obd->obd_replayable = 1;
5129         /* No connection accepted until configurations will finish */
5130         obd->obd_no_conn = 1;
5131
5132         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5133                 char *str = lustre_cfg_string(cfg, 4);
5134                 if (strchr(str, 'n')) {
5135                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5136                         obd->obd_replayable = 0;
5137                 }
5138         }
5139
5140         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5141         if (rc)
5142                 GOTO(err_fini_stack, rc);
5143
5144         rc = mdt_seq_init(env, m);
5145         if (rc)
5146                 GOTO(err_fini_fld, rc);
5147
5148         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5149                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5150         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5151                                               LDLM_NAMESPACE_SERVER,
5152                                               LDLM_NAMESPACE_GREEDY,
5153                                               LDLM_NS_TYPE_MDT);
5154         if (m->mdt_namespace == NULL)
5155                 GOTO(err_fini_seq, rc = -ENOMEM);
5156
5157         m->mdt_namespace->ns_lvbp = m;
5158         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5159
5160         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5161         /* set obd_namespace for compatibility with old code */
5162         obd->obd_namespace = m->mdt_namespace;
5163
5164         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5165                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5166                       OBD_FAIL_MDS_ALL_REPLY_NET);
5167         if (rc)
5168                 GOTO(err_free_ns, rc);
5169
5170         /* Amount of available space excluded from granting and reserved
5171          * for metadata. It is in percentage and 50% is default value. */
5172         tgd->tgd_reserved_pcnt = 50;
5173
5174         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5175                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5176         else
5177                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5178
5179         rc = mdt_fs_setup(env, m, obd, lsi);
5180         if (rc)
5181                 GOTO(err_tgt, rc);
5182
5183         fid.f_seq = FID_SEQ_LOCAL_NAME;
5184         fid.f_oid = 1;
5185         fid.f_ver = 0;
5186         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5187         if (rc != 0)
5188                 GOTO(err_fs_cleanup, rc);
5189
5190         rc = mdt_hsm_cdt_init(m);
5191         if (rc != 0) {
5192                 CERROR("%s: error initializing coordinator, rc %d\n",
5193                        mdt_obd_name(m), rc);
5194                 GOTO(err_los_fini, rc);
5195         }
5196
5197         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5198
5199         next = m->mdt_child;
5200         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5201
5202         mntopts = dt_conf->ddp_mntopts;
5203
5204         if (mntopts & MNTOPT_USERXATTR)
5205                 m->mdt_opts.mo_user_xattr = 1;
5206         else
5207                 m->mdt_opts.mo_user_xattr = 0;
5208
5209         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5210
5211         if (mntopts & MNTOPT_ACL)
5212                 m->mdt_opts.mo_acl = 1;
5213         else
5214                 m->mdt_opts.mo_acl = 0;
5215
5216         /* XXX: to support suppgid for ACL, we enable identity_upcall
5217          * by default, otherwise, maybe got unexpected -EACCESS. */
5218         if (m->mdt_opts.mo_acl)
5219                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5220
5221         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5222                                                 identity_upcall,
5223                                                 &mdt_identity_upcall_cache_ops);
5224         if (IS_ERR(m->mdt_identity_cache)) {
5225                 rc = PTR_ERR(m->mdt_identity_cache);
5226                 m->mdt_identity_cache = NULL;
5227                 GOTO(err_free_hsm, rc);
5228         }
5229
5230         rc = mdt_procfs_init(m, dev);
5231         if (rc) {
5232                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5233                 GOTO(err_recovery, rc);
5234         }
5235
5236         rc = mdt_quota_init(env, m, cfg);
5237         if (rc)
5238                 GOTO(err_procfs, rc);
5239
5240         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5241         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5242                            "mdt_ldlm_client", m->mdt_ldlm_client);
5243
5244         ping_evictor_start();
5245
5246         /* recovery will be started upon mdt_prepare()
5247          * when the whole stack is complete and ready
5248          * to serve the requests */
5249
5250         /* Reduce the initial timeout on an MDS because it doesn't need such
5251          * a long timeout as an OST does. Adaptive timeouts will adjust this
5252          * value appropriately. */
5253         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5254                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5255
5256         RETURN(0);
5257 err_procfs:
5258         mdt_procfs_fini(m);
5259 err_recovery:
5260         target_recovery_fini(obd);
5261         upcall_cache_cleanup(m->mdt_identity_cache);
5262         m->mdt_identity_cache = NULL;
5263 err_free_hsm:
5264         mdt_hsm_cdt_fini(m);
5265 err_los_fini:
5266         local_oid_storage_fini(env, m->mdt_los);
5267         m->mdt_los = NULL;
5268 err_fs_cleanup:
5269         mdt_fs_cleanup(env, m);
5270 err_tgt:
5271         tgt_fini(env, &m->mdt_lut);
5272 err_free_ns:
5273         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5274         obd->obd_namespace = m->mdt_namespace = NULL;
5275 err_fini_seq:
5276         mdt_seq_fini(env, m);
5277 err_fini_fld:
5278         mdt_fld_fini(env, m);
5279 err_fini_stack:
5280         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5281 err_lmi:
5282         if (lmi)
5283                 server_put_mount(dev, true);
5284         return(rc);
5285 }
5286
5287 /* For interoperability, the left element is old parameter, the right one
5288  * is the new version of the parameter, if some parameter is deprecated,
5289  * the new version should be set as NULL. */
5290 static struct cfg_interop_param mdt_interop_param[] = {
5291         { "mdt.group_upcall",   NULL },
5292         { "mdt.quota_type",     NULL },
5293         { "mdd.quota_type",     NULL },
5294         { "mdt.som",            NULL },
5295         { "mdt.rootsquash",     "mdt.root_squash" },
5296         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5297         { NULL }
5298 };
5299
5300 /* used by MGS to process specific configurations */
5301 static int mdt_process_config(const struct lu_env *env,
5302                               struct lu_device *d, struct lustre_cfg *cfg)
5303 {
5304         struct mdt_device *m = mdt_dev(d);
5305         struct md_device *md_next = m->mdt_child;
5306         struct lu_device *next = md2lu_dev(md_next);
5307         int rc;
5308         ENTRY;
5309
5310         switch (cfg->lcfg_command) {
5311         case LCFG_PARAM: {
5312                 struct obd_device          *obd = d->ld_obd;
5313
5314                 /* For interoperability */
5315                 struct cfg_interop_param   *ptr = NULL;
5316                 struct lustre_cfg          *old_cfg = NULL;
5317                 char                       *param = NULL;
5318
5319                 param = lustre_cfg_string(cfg, 1);
5320                 if (param == NULL) {
5321                         CERROR("param is empty\n");
5322                         rc = -EINVAL;
5323                         break;
5324                 }
5325
5326                 ptr = class_find_old_param(param, mdt_interop_param);
5327                 if (ptr != NULL) {
5328                         if (ptr->new_param == NULL) {
5329                                 rc = 0;
5330                                 CWARN("For interoperability, skip this %s."
5331                                       " It is obsolete.\n", ptr->old_param);
5332                                 break;
5333                         }
5334
5335                         CWARN("Found old param %s, changed it to %s.\n",
5336                               ptr->old_param, ptr->new_param);
5337
5338                         old_cfg = cfg;
5339                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5340                         if (IS_ERR(cfg)) {
5341                                 rc = PTR_ERR(cfg);
5342                                 break;
5343                         }
5344                 }
5345
5346                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5347                                               cfg, obd);
5348                 if (rc > 0 || rc == -ENOSYS) {
5349                         /* is it an HSM var ? */
5350                         rc = class_process_proc_param(PARAM_HSM,
5351                                                       hsm_cdt_get_proc_vars(),
5352                                                       cfg, obd);
5353                         if (rc > 0 || rc == -ENOSYS)
5354                                 /* we don't understand; pass it on */
5355                                 rc = next->ld_ops->ldo_process_config(env, next,
5356                                                                       cfg);
5357                 }
5358
5359                 if (old_cfg)
5360                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5361                                                      cfg->lcfg_buflens));
5362                 break;
5363         }
5364         default:
5365                 /* others are passed further */
5366                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5367                 break;
5368         }
5369         RETURN(rc);
5370 }
5371
5372 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5373                                           const struct lu_object_header *hdr,
5374                                           struct lu_device *d)
5375 {
5376         struct mdt_object *mo;
5377
5378         ENTRY;
5379
5380         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5381         if (mo != NULL) {
5382                 struct lu_object *o;
5383                 struct lu_object_header *h;
5384
5385                 o = &mo->mot_obj;
5386                 h = &mo->mot_header;
5387                 lu_object_header_init(h);
5388                 lu_object_init(o, h, d);
5389                 lu_object_add_top(h, o);
5390                 o->lo_ops = &mdt_obj_ops;
5391                 spin_lock_init(&mo->mot_write_lock);
5392                 mutex_init(&mo->mot_som_mutex);
5393                 mutex_init(&mo->mot_lov_mutex);
5394                 init_rwsem(&mo->mot_dom_sem);
5395                 init_rwsem(&mo->mot_open_sem);
5396                 atomic_set(&mo->mot_open_count, 0);
5397                 RETURN(o);
5398         }
5399         RETURN(NULL);
5400 }
5401
5402 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5403                            const struct lu_object_conf *unused)
5404 {
5405         struct mdt_device *d = mdt_dev(o->lo_dev);
5406         struct lu_device  *under;
5407         struct lu_object  *below;
5408         int                rc = 0;
5409         ENTRY;
5410
5411         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5412                PFID(lu_object_fid(o)));
5413
5414         under = &d->mdt_child->md_lu_dev;
5415         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5416         if (below != NULL) {
5417                 lu_object_add(o, below);
5418         } else
5419                 rc = -ENOMEM;
5420
5421         RETURN(rc);
5422 }
5423
5424 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5425 {
5426         struct mdt_object *mo = mdt_obj(o);
5427         struct lu_object_header *h;
5428         ENTRY;
5429
5430         h = o->lo_header;
5431         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5432                PFID(lu_object_fid(o)));
5433
5434         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5435         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5436
5437         lu_object_fini(o);
5438         lu_object_header_fini(h);
5439         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5440
5441         EXIT;
5442 }
5443
5444 static int mdt_object_print(const struct lu_env *env, void *cookie,
5445                             lu_printer_t p, const struct lu_object *o)
5446 {
5447         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5448
5449         return (*p)(env, cookie,
5450                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5451                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5452                     mdto->mot_cache_attr ? "cache_attr" : "",
5453                     mdto->mot_write_count);
5454 }
5455
5456 static int mdt_prepare(const struct lu_env *env,
5457                 struct lu_device *pdev,
5458                 struct lu_device *cdev)
5459 {
5460         struct mdt_device *mdt = mdt_dev(cdev);
5461         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5462         struct obd_device *obd = cdev->ld_obd;
5463         int rc;
5464
5465         ENTRY;
5466
5467         LASSERT(obd);
5468
5469         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5470         if (rc)
5471                 RETURN(rc);
5472
5473         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5474         if (rc)
5475                 RETURN(rc);
5476
5477         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5478         if (rc)
5479                 RETURN(rc);
5480
5481         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5482         /* The LFSCK instance is registered just now, so it must be there when
5483          * register the namespace to such instance. */
5484         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5485
5486         if (mdt->mdt_seq_site.ss_node_id == 0) {
5487                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5488                                                          &mdt->mdt_md_root_fid);
5489                 if (rc)
5490                         RETURN(rc);
5491         }
5492
5493         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5494
5495         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5496         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5497         LASSERT(obd->obd_no_conn);
5498         spin_lock(&obd->obd_dev_lock);
5499         obd->obd_no_conn = 0;
5500         spin_unlock(&obd->obd_dev_lock);
5501
5502         if (obd->obd_recovering == 0)
5503                 mdt_postrecov(env, mdt);
5504
5505         RETURN(rc);
5506 }
5507
5508 const struct lu_device_operations mdt_lu_ops = {
5509         .ldo_object_alloc   = mdt_object_alloc,
5510         .ldo_process_config = mdt_process_config,
5511         .ldo_prepare        = mdt_prepare,
5512 };
5513
5514 static const struct lu_object_operations mdt_obj_ops = {
5515         .loo_object_init    = mdt_object_init,
5516         .loo_object_free    = mdt_object_free,
5517         .loo_object_print   = mdt_object_print
5518 };
5519
5520 static int mdt_obd_set_info_async(const struct lu_env *env,
5521                                   struct obd_export *exp,
5522                                   __u32 keylen, void *key,
5523                                   __u32 vallen, void *val,
5524                                   struct ptlrpc_request_set *set)
5525 {
5526         int rc;
5527
5528         ENTRY;
5529
5530         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5531                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5532                 RETURN(rc);
5533         }
5534
5535         RETURN(0);
5536 }
5537
5538 /**
5539  * Match client and server connection feature flags.
5540  *
5541  * Compute the compatibility flags for a connection request based on
5542  * features mutually supported by client and server.
5543  *
5544  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5545  * must not be updated here, otherwise a partially initialized value may
5546  * be exposed. After the connection request is successfully processed,
5547  * the top-level MDT connect request handler atomically updates the export
5548  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5549  * reply. \see mdt_connect().
5550  *
5551  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5552  * full struct obd_connect_data. So care must be taken when accessing fields
5553  * that are not present in struct obd_connect_data_v1. See LU-16.
5554  *
5555  * \param exp   the obd_export associated with this client/target pair
5556  * \param mdt   the target device for the connection
5557  * \param data  stores data for this connect request
5558  *
5559  * \retval 0       success
5560  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5561  * \retval -EBADE  client and server feature requirements are incompatible
5562  */
5563 static int mdt_connect_internal(const struct lu_env *env,
5564                                 struct obd_export *exp,
5565                                 struct mdt_device *mdt,
5566                                 struct obd_connect_data *data, bool reconnect)
5567 {
5568         const char *obd_name = mdt_obd_name(mdt);
5569         LASSERT(data != NULL);
5570
5571         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5572
5573         if (mdt->mdt_bottom->dd_rdonly &&
5574             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5575             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5576                 RETURN(-EACCES);
5577
5578         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5579                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5580
5581         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5582
5583         if (!mdt->mdt_opts.mo_acl)
5584                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5585
5586         if (!mdt->mdt_opts.mo_user_xattr)
5587                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5588
5589         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5590                 data->ocd_brw_size = min(data->ocd_brw_size,
5591                                          mdt->mdt_brw_size);
5592                 if (data->ocd_brw_size == 0) {
5593                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5594                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5595                                "ocd_brw_size unexpectedly zero, network data "
5596                                "corruption? Refusing to connect this client\n",
5597                                obd_name, exp->exp_client_uuid.uuid,
5598                                exp, data->ocd_connect_flags, data->ocd_version,
5599                                data->ocd_grant, data->ocd_index);
5600                         return -EPROTO;
5601                 }
5602         }
5603
5604         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5605                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5606
5607                 /* client is reporting its page size, for future use */
5608                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5609                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5610                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5611                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5612                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5613                 /* ocd_grant_tax_kb is in 1K byte blocks */
5614                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5615                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5616         }
5617
5618         /* Save connect_data we have so far because tgt_grant_connect()
5619          * uses it to calculate grant, and we want to save the client
5620          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5621         exp->exp_connect_data = *data;
5622         if (OCD_HAS_FLAG(data, GRANT))
5623                 tgt_grant_connect(env, exp, data, !reconnect);
5624
5625         if (OCD_HAS_FLAG(data, MAXBYTES))
5626                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5627
5628         /* NB: Disregard the rule against updating
5629          * exp_connect_data.ocd_connect_flags in this case, since
5630          * tgt_client_new() needs to know if this is a lightweight
5631          * connection, and it is safe to expose this flag before
5632          * connection processing completes. */
5633         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5634                 spin_lock(&exp->exp_lock);
5635                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5636                 spin_unlock(&exp->exp_lock);
5637         }
5638
5639         data->ocd_version = LUSTRE_VERSION_CODE;
5640
5641         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5642                 CWARN("%s: MDS requires FID support, but client not\n",
5643                       obd_name);
5644                 return -EBADE;
5645         }
5646
5647         if (OCD_HAS_FLAG(data, PINGLESS)) {
5648                 if (ptlrpc_pinger_suppress_pings()) {
5649                         spin_lock(&exp->exp_obd->obd_dev_lock);
5650                         list_del_init(&exp->exp_obd_chain_timed);
5651                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5652                 } else {
5653                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5654                 }
5655         }
5656
5657         data->ocd_max_easize = mdt->mdt_max_ea_size;
5658
5659         /* NB: Disregard the rule against updating
5660          * exp_connect_data.ocd_connect_flags in this case, since
5661          * tgt_client_new() needs to know if this is client supports
5662          * multiple modify RPCs, and it is safe to expose this flag before
5663          * connection processing completes. */
5664         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5665                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5666                 spin_lock(&exp->exp_lock);
5667                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5668                 spin_unlock(&exp->exp_lock);
5669         }
5670
5671         if (OCD_HAS_FLAG(data, CKSUM)) {
5672                 __u32 cksum_types = data->ocd_cksum_types;
5673
5674                 /* The client set in ocd_cksum_types the checksum types it
5675                  * supports. We have to mask off the algorithms that we don't
5676                  * support */
5677                 data->ocd_cksum_types &=
5678                         obd_cksum_types_supported_server(obd_name);
5679
5680                 if (unlikely(data->ocd_cksum_types == 0)) {
5681                         CERROR("%s: Connect with checksum support but no "
5682                                "ocd_cksum_types is set\n",
5683                                exp->exp_obd->obd_name);
5684                         RETURN(-EPROTO);
5685                 }
5686
5687                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5688                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5689                        cksum_types, data->ocd_cksum_types);
5690         } else {
5691                 /* This client does not support OBD_CONNECT_CKSUM
5692                  * fall back to CRC32 */
5693                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5694                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5695                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5696         }
5697
5698         return 0;
5699 }
5700
5701 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5702                                    struct mdt_thread_info *info,
5703                                    struct mdt_file_data *mfd)
5704 {
5705         struct lu_context ses;
5706         int rc;
5707         ENTRY;
5708
5709         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5710         if (rc)
5711                 RETURN(rc);
5712
5713         env->le_ses = &ses;
5714         lu_context_enter(&ses);
5715
5716         mdt_ucred(info)->uc_valid = UCRED_OLD;
5717         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5718
5719         lu_context_exit(&ses);
5720         lu_context_fini(&ses);
5721         env->le_ses = NULL;
5722
5723         RETURN(rc);
5724 }
5725
5726 static int mdt_export_cleanup(struct obd_export *exp)
5727 {
5728         struct list_head         closing_list;
5729         struct mdt_export_data  *med = &exp->exp_mdt_data;
5730         struct obd_device       *obd = exp->exp_obd;
5731         struct mdt_device       *mdt;
5732         struct mdt_thread_info  *info;
5733         struct lu_env            env;
5734         struct mdt_file_data    *mfd, *n;
5735         int rc = 0;
5736         ENTRY;
5737
5738         INIT_LIST_HEAD(&closing_list);
5739         spin_lock(&med->med_open_lock);
5740         while (!list_empty(&med->med_open_head)) {
5741                 struct list_head *tmp = med->med_open_head.next;
5742                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5743
5744                 /* Remove mfd handle so it can't be found again.
5745                  * We are consuming the mfd_list reference here. */
5746                 class_handle_unhash(&mfd->mfd_open_handle);
5747                 list_move_tail(&mfd->mfd_list, &closing_list);
5748         }
5749         spin_unlock(&med->med_open_lock);
5750         mdt = mdt_dev(obd->obd_lu_dev);
5751         LASSERT(mdt != NULL);
5752
5753         rc = lu_env_init(&env, LCT_MD_THREAD);
5754         if (rc)
5755                 RETURN(rc);
5756
5757         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5758         LASSERT(info != NULL);
5759         memset(info, 0, sizeof *info);
5760         info->mti_env = &env;
5761         info->mti_mdt = mdt;
5762         info->mti_exp = exp;
5763
5764         if (!list_empty(&closing_list)) {
5765                 struct md_attr *ma = &info->mti_attr;
5766
5767                 /* Close any open files (which may also cause orphan
5768                  * unlinking). */
5769                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5770                         list_del_init(&mfd->mfd_list);
5771                         ma->ma_need = ma->ma_valid = 0;
5772
5773                         /* This file is being closed due to an eviction, it
5774                          * could have been modified and now dirty regarding to
5775                          * HSM archive, check this!
5776                          * The logic here is to mark a file dirty if there's a
5777                          * chance it was dirtied before the client was evicted,
5778                          * so that we don't have to wait for a release attempt
5779                          * before finding out the file was actually dirty and
5780                          * fail the release. Aggressively marking it dirty here
5781                          * will cause the policy engine to attempt to
5782                          * re-archive it; when rearchiving, we can compare the
5783                          * current version to the HSM data_version and make the
5784                          * archive request into a noop if it's not actually
5785                          * dirty.
5786                          */
5787                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
5788                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5789
5790                         /* Don't unlink orphan on failover umount, LU-184 */
5791                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5792                                 ma->ma_valid = MA_FLAGS;
5793                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5794                         }
5795                         mdt_mfd_close(info, mfd);
5796                 }
5797         }
5798         info->mti_mdt = NULL;
5799         /* cleanup client slot early */
5800         /* Do not erase record for recoverable client. */
5801         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5802                 tgt_client_del(&env, exp);
5803         lu_env_fini(&env);
5804
5805         RETURN(rc);
5806 }
5807
5808 static inline void mdt_enable_slc(struct mdt_device *mdt)
5809 {
5810         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5811                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5812 }
5813
5814 static inline void mdt_disable_slc(struct mdt_device *mdt)
5815 {
5816         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5817                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5818 }
5819
5820 static int mdt_obd_disconnect(struct obd_export *exp)
5821 {
5822         int rc;
5823
5824         ENTRY;
5825
5826         LASSERT(exp);
5827         class_export_get(exp);
5828
5829         if (!(exp->exp_flags & OBD_OPT_FORCE))
5830                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5831
5832         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5833             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5834                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5835
5836                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5837                         mdt_disable_slc(mdt);
5838         }
5839
5840         rc = server_disconnect_export(exp);
5841         if (rc != 0)
5842                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5843
5844         tgt_grant_discard(exp);
5845
5846         rc = mdt_export_cleanup(exp);
5847         nodemap_del_member(exp);
5848         class_export_put(exp);
5849         RETURN(rc);
5850 }
5851
5852 /* mds_connect copy */
5853 static int mdt_obd_connect(const struct lu_env *env,
5854                            struct obd_export **exp, struct obd_device *obd,
5855                            struct obd_uuid *cluuid,
5856                            struct obd_connect_data *data,
5857                            void *localdata)
5858 {
5859         struct obd_export       *lexp;
5860         struct lustre_handle    conn = { 0 };
5861         struct mdt_device       *mdt;
5862         int                      rc;
5863         lnet_nid_t              *client_nid = localdata;
5864         ENTRY;
5865
5866         LASSERT(env != NULL);
5867         LASSERT(data != NULL);
5868
5869         if (!exp || !obd || !cluuid)
5870                 RETURN(-EINVAL);
5871
5872         mdt = mdt_dev(obd->obd_lu_dev);
5873
5874         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5875             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5876                 atomic_inc(&mdt->mdt_mds_mds_conns);
5877                 mdt_enable_slc(mdt);
5878         }
5879
5880         /*
5881          * first, check whether the stack is ready to handle requests
5882          * XXX: probably not very appropriate method is used now
5883          *      at some point we should find a better one
5884          */
5885         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5886             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5887             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5888                 rc = obd_get_info(env, mdt->mdt_child_exp,
5889                                   sizeof(KEY_OSP_CONNECTED),
5890                                   KEY_OSP_CONNECTED, NULL, NULL);
5891                 if (rc)
5892                         RETURN(-EAGAIN);
5893                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5894         }
5895
5896         rc = class_connect(&conn, obd, cluuid);
5897         if (rc)
5898                 RETURN(rc);
5899
5900         lexp = class_conn2export(&conn);
5901         LASSERT(lexp != NULL);
5902
5903         rc = nodemap_add_member(*client_nid, lexp);
5904         if (rc != 0 && rc != -EEXIST)
5905                 GOTO(out, rc);
5906
5907         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5908         if (rc == 0) {
5909                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5910
5911                 LASSERT(lcd);
5912                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5913                 rc = tgt_client_new(env, lexp);
5914                 if (rc == 0)
5915                         mdt_export_stats_init(obd, lexp, localdata);
5916         }
5917 out:
5918         if (rc != 0) {
5919                 class_disconnect(lexp);
5920                 nodemap_del_member(lexp);
5921                 *exp = NULL;
5922         } else {
5923                 *exp = lexp;
5924                 /* Because we do not want this export to be evicted by pinger,
5925                  * let's not add this export to the timed chain list. */
5926                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5927                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5928                         list_del_init(&lexp->exp_obd_chain_timed);
5929                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5930                 }
5931         }
5932
5933         RETURN(rc);
5934 }
5935
5936 static int mdt_obd_reconnect(const struct lu_env *env,
5937                              struct obd_export *exp, struct obd_device *obd,
5938                              struct obd_uuid *cluuid,
5939                              struct obd_connect_data *data,
5940                              void *localdata)
5941 {
5942         lnet_nid_t             *client_nid = localdata;
5943         int                     rc;
5944         ENTRY;
5945
5946         if (exp == NULL || obd == NULL || cluuid == NULL)
5947                 RETURN(-EINVAL);
5948
5949         rc = nodemap_add_member(*client_nid, exp);
5950         if (rc != 0 && rc != -EEXIST)
5951                 RETURN(rc);
5952
5953         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5954                                   true);
5955         if (rc == 0)
5956                 mdt_export_stats_init(obd, exp, localdata);
5957         else
5958                 nodemap_del_member(exp);
5959
5960         RETURN(rc);
5961 }
5962
5963 /* FIXME: Can we avoid using these two interfaces? */
5964 static int mdt_init_export(struct obd_export *exp)
5965 {
5966         struct mdt_export_data *med = &exp->exp_mdt_data;
5967         int                     rc;
5968         ENTRY;
5969
5970         INIT_LIST_HEAD(&med->med_open_head);
5971         spin_lock_init(&med->med_open_lock);
5972         spin_lock(&exp->exp_lock);
5973         exp->exp_connecting = 1;
5974         spin_unlock(&exp->exp_lock);
5975
5976         /* self-export doesn't need client data and ldlm initialization */
5977         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5978                                      &exp->exp_client_uuid)))
5979                 RETURN(0);
5980
5981         rc = tgt_client_alloc(exp);
5982         if (rc)
5983                 GOTO(err, rc);
5984
5985         rc = ldlm_init_export(exp);
5986         if (rc)
5987                 GOTO(err_free, rc);
5988
5989         RETURN(rc);
5990
5991 err_free:
5992         tgt_client_free(exp);
5993 err:
5994         CERROR("%s: Failed to initialize export: rc = %d\n",
5995                exp->exp_obd->obd_name, rc);
5996         return rc;
5997 }
5998
5999 static int mdt_destroy_export(struct obd_export *exp)
6000 {
6001         ENTRY;
6002
6003         target_destroy_export(exp);
6004         /* destroy can be called from failed obd_setup, so
6005          * checking uuid is safer than obd_self_export */
6006         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6007                                      &exp->exp_client_uuid)))
6008                 RETURN(0);
6009
6010         ldlm_destroy_export(exp);
6011         tgt_client_free(exp);
6012
6013         LASSERT(list_empty(&exp->exp_outstanding_replies));
6014         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6015
6016         /*
6017          * discard grants once we're sure no more
6018          * interaction with the client is possible
6019          */
6020         tgt_grant_discard(exp);
6021         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6022                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6023
6024         if (!(exp->exp_flags & OBD_OPT_FORCE))
6025                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6026
6027         RETURN(0);
6028 }
6029
6030 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6031                    struct linkea_data *ldata)
6032 {
6033         int rc;
6034
6035         LASSERT(ldata->ld_buf->lb_buf != NULL);
6036
6037         if (!mdt_object_exists(mdt_obj))
6038                 return -ENODATA;
6039
6040         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6041                           ldata->ld_buf, XATTR_NAME_LINK);
6042         if (rc == -ERANGE) {
6043                 /* Buf was too small, figure out what we need. */
6044                 lu_buf_free(ldata->ld_buf);
6045                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6046                                   ldata->ld_buf, XATTR_NAME_LINK);
6047                 if (rc < 0)
6048                         return rc;
6049                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6050                 if (ldata->ld_buf->lb_buf == NULL)
6051                         return -ENOMEM;
6052                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6053                                   ldata->ld_buf, XATTR_NAME_LINK);
6054         }
6055         if (rc < 0)
6056                 return rc;
6057
6058         return linkea_init_with_rec(ldata);
6059 }
6060
6061 /**
6062  * Given an MDT object, try to look up the full path to the object.
6063  * Part of the MDT layer implementation of lfs fid2path.
6064  *
6065  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6066  * \param[in]     obj   Object to do path lookup of
6067  * \param[in,out] fp    User-provided struct to store path information
6068  * \param[in]     root_fid Root FID of current path should reach
6069  *
6070  * \retval 0 Lookup successful, path information stored in fp
6071  * \retval -EAGAIN Lookup failed, usually because object is being moved
6072  * \retval negative errno if there was a problem
6073  */
6074 static int mdt_path_current(struct mdt_thread_info *info,
6075                             struct mdt_object *obj,
6076                             struct getinfo_fid2path *fp,
6077                             struct lu_fid *root_fid)
6078 {
6079         struct mdt_device       *mdt = info->mti_mdt;
6080         struct mdt_object       *mdt_obj;
6081         struct link_ea_header   *leh;
6082         struct link_ea_entry    *lee;
6083         struct lu_name          *tmpname = &info->mti_name;
6084         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6085         struct lu_buf           *buf = &info->mti_big_buf;
6086         char                    *ptr;
6087         int                     reclen;
6088         struct linkea_data      ldata = { NULL };
6089         int                     rc = 0;
6090         bool                    first = true;
6091         ENTRY;
6092
6093         /* temp buffer for path element, the buffer will be finally freed
6094          * in mdt_thread_info_fini */
6095         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6096         if (buf->lb_buf == NULL)
6097                 RETURN(-ENOMEM);
6098
6099         ldata.ld_buf = buf;
6100         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6101         *ptr = 0;
6102         --ptr;
6103         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6104
6105         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6106                 struct lu_buf           lmv_buf;
6107
6108                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6109                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6110                         GOTO(out, rc = -ENOENT);
6111
6112                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6113                         mdt_obj = obj;
6114                         mdt_object_get(info->mti_env, mdt_obj);
6115                 } else {
6116                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6117                         if (IS_ERR(mdt_obj))
6118                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6119                 }
6120
6121                 if (!mdt_object_exists(mdt_obj)) {
6122                         mdt_object_put(info->mti_env, mdt_obj);
6123                         GOTO(out, rc = -ENOENT);
6124                 }
6125
6126                 if (mdt_object_remote(mdt_obj)) {
6127                         mdt_object_put(info->mti_env, mdt_obj);
6128                         GOTO(remote_out, rc = -EREMOTE);
6129                 }
6130
6131                 rc = mdt_links_read(info, mdt_obj, &ldata);
6132                 if (rc != 0) {
6133                         mdt_object_put(info->mti_env, mdt_obj);
6134                         GOTO(out, rc);
6135                 }
6136
6137                 leh = buf->lb_buf;
6138                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6139                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6140                 /* If set, use link #linkno for path lookup, otherwise use
6141                    link #0.  Only do this for the final path element. */
6142                 if (first && fp->gf_linkno < leh->leh_reccount) {
6143                         int count;
6144                         for (count = 0; count < fp->gf_linkno; count++) {
6145                                 lee = (struct link_ea_entry *)
6146                                      ((char *)lee + reclen);
6147                                 linkea_entry_unpack(lee, &reclen, tmpname,
6148                                                     tmpfid);
6149                         }
6150                         if (fp->gf_linkno < leh->leh_reccount - 1)
6151                                 /* indicate to user there are more links */
6152                                 fp->gf_linkno++;
6153                 }
6154
6155                 lmv_buf.lb_buf = info->mti_xattr_buf;
6156                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6157                 /* Check if it is slave stripes */
6158                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6159                                   &lmv_buf, XATTR_NAME_LMV);
6160                 mdt_object_put(info->mti_env, mdt_obj);
6161                 if (rc > 0) {
6162                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6163
6164                         /* For slave stripes, get its master */
6165                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6166                                 fp->gf_fid = *tmpfid;
6167                                 continue;
6168                         }
6169                 } else if (rc < 0 && rc != -ENODATA) {
6170                         GOTO(out, rc);
6171                 }
6172
6173                 rc = 0;
6174
6175                 /* Pack the name in the end of the buffer */
6176                 ptr -= tmpname->ln_namelen;
6177                 if (ptr - 1 <= fp->gf_u.gf_path)
6178                         GOTO(out, rc = -EOVERFLOW);
6179                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6180                 *(--ptr) = '/';
6181
6182                 /* keep the last resolved fid to the client, so the
6183                  * client will build the left path on another MDT for
6184                  * remote object */
6185                 fp->gf_fid = *tmpfid;
6186
6187                 first = false;
6188         }
6189
6190 remote_out:
6191         ptr++; /* skip leading / */
6192         memmove(fp->gf_u.gf_path, ptr,
6193                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6194
6195 out:
6196         RETURN(rc);
6197 }
6198
6199 /**
6200  * Given an MDT object, use mdt_path_current to get the path.
6201  * Essentially a wrapper to retry mdt_path_current a set number of times
6202  * if -EAGAIN is returned (usually because an object is being moved).
6203  *
6204  * Part of the MDT layer implementation of lfs fid2path.
6205  *
6206  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6207  * \param[in]     obj   Object to do path lookup of
6208  * \param[in,out] fp    User-provided struct for arguments and to store path
6209  *                      information
6210  *
6211  * \retval 0 Lookup successful, path information stored in fp
6212  * \retval negative errno if there was a problem
6213  */
6214 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6215                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6216 {
6217         struct mdt_device       *mdt = info->mti_mdt;
6218         int                     tries = 3;
6219         int                     rc = -EAGAIN;
6220         ENTRY;
6221
6222         if (fp->gf_pathlen < 3)
6223                 RETURN(-EOVERFLOW);
6224
6225         if (root_fid == NULL)
6226                 root_fid = &mdt->mdt_md_root_fid;
6227
6228         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6229                 fp->gf_u.gf_path[0] = '\0';
6230                 RETURN(0);
6231         }
6232
6233         /* Retry multiple times in case file is being moved */
6234         while (tries-- && rc == -EAGAIN)
6235                 rc = mdt_path_current(info, obj, fp, root_fid);
6236
6237         RETURN(rc);
6238 }
6239
6240 /**
6241  * Get the full path of the provided FID, as of changelog record recno.
6242  *
6243  * This checks sanity and looks up object for user provided FID
6244  * before calling the actual path lookup code.
6245  *
6246  * Part of the MDT layer implementation of lfs fid2path.
6247  *
6248  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6249  * \param[in,out] fp    User-provided struct for arguments and to store path
6250  *                      information
6251  *
6252  * \retval 0 Lookup successful, path information and recno stored in fp
6253  * \retval -ENOENT, object does not exist
6254  * \retval negative errno if there was a problem
6255  */
6256 static int mdt_fid2path(struct mdt_thread_info *info,
6257                         struct lu_fid *root_fid,
6258                         struct getinfo_fid2path *fp)
6259 {
6260         struct mdt_device *mdt = info->mti_mdt;
6261         struct mdt_object *obj;
6262         int    rc;
6263         ENTRY;
6264
6265         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6266                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6267
6268         if (!fid_is_sane(&fp->gf_fid))
6269                 RETURN(-EINVAL);
6270
6271         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6272                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6273                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6274                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6275                 RETURN(-EINVAL);
6276         }
6277
6278         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6279         if (IS_ERR(obj)) {
6280                 rc = PTR_ERR(obj);
6281                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6282                        PFID(&fp->gf_fid), rc);
6283                 RETURN(rc);
6284         }
6285
6286         if (mdt_object_remote(obj))
6287                 rc = -EREMOTE;
6288         else if (!mdt_object_exists(obj))
6289                 rc = -ENOENT;
6290         else
6291                 rc = 0;
6292
6293         if (rc < 0) {
6294                 mdt_object_put(info->mti_env, obj);
6295                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6296                        PFID(&fp->gf_fid), rc);
6297                 RETURN(rc);
6298         }
6299
6300         rc = mdt_path(info, obj, fp, root_fid);
6301
6302         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6303                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6304                fp->gf_recno, fp->gf_linkno);
6305
6306         mdt_object_put(info->mti_env, obj);
6307
6308         RETURN(rc);
6309 }
6310
6311 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6312                             void *val, int vallen)
6313 {
6314         struct getinfo_fid2path *fpout, *fpin;
6315         struct lu_fid *root_fid = NULL;
6316         int rc = 0;
6317
6318         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6319         fpout = val;
6320
6321         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6322                 lustre_swab_fid2path(fpin);
6323
6324         memcpy(fpout, fpin, sizeof(*fpin));
6325         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6326                 RETURN(-EINVAL);
6327
6328         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6329                       sizeof(struct lu_fid)) {
6330                 /* client sent its root FID, which is normally fileset FID */
6331                 root_fid = fpin->gf_u.gf_root_fid;
6332                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6333                         lustre_swab_lu_fid(root_fid);
6334
6335                 if (root_fid != NULL && !fid_is_sane(root_fid))
6336                         RETURN(-EINVAL);
6337         }
6338
6339         rc = mdt_fid2path(info, root_fid, fpout);
6340         RETURN(rc);
6341 }
6342
6343 int mdt_get_info(struct tgt_session_info *tsi)
6344 {
6345         char    *key;
6346         int      keylen;
6347         __u32   *vallen;
6348         void    *valout;
6349         int      rc;
6350
6351         ENTRY;
6352
6353         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6354         if (key == NULL) {
6355                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6356                 RETURN(err_serious(-EFAULT));
6357         }
6358         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6359                                       RCL_CLIENT);
6360
6361         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6362         if (vallen == NULL) {
6363                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6364                                 tgt_name(tsi->tsi_tgt));
6365                 RETURN(err_serious(-EFAULT));
6366         }
6367
6368         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6369                              *vallen);
6370         rc = req_capsule_server_pack(tsi->tsi_pill);
6371         if (rc)
6372                 RETURN(err_serious(rc));
6373
6374         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6375         if (valout == NULL) {
6376                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6377                                 tgt_name(tsi->tsi_tgt));
6378                 RETURN(err_serious(-EFAULT));
6379         }
6380
6381         if (KEY_IS(KEY_FID2PATH)) {
6382                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6383
6384                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6385                 mdt_thread_info_fini(info);
6386         } else {
6387                 rc = -EINVAL;
6388         }
6389         RETURN(rc);
6390 }
6391
6392 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6393 {
6394         struct obd_ioctl_data *data = karg;
6395         struct lu_fid *fid;
6396         __u64 version;
6397         struct mdt_object *obj;
6398         struct mdt_lock_handle  *lh;
6399         int rc;
6400         ENTRY;
6401
6402         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6403             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6404                 RETURN(-EINVAL);
6405
6406         fid = (struct lu_fid *)data->ioc_inlbuf1;
6407
6408         if (!fid_is_sane(fid))
6409                 RETURN(-EINVAL);
6410
6411         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6412
6413         lh = &mti->mti_lh[MDT_LH_PARENT];
6414         mdt_lock_reg_init(lh, LCK_CR);
6415
6416         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6417         if (IS_ERR(obj))
6418                 RETURN(PTR_ERR(obj));
6419
6420         if (mdt_object_remote(obj)) {
6421                 rc = -EREMOTE;
6422                 /**
6423                  * before calling version get the correct MDS should be
6424                  * fid, this is error to find remote object here
6425                  */
6426                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6427         } else if (!mdt_object_exists(obj)) {
6428                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6429                 rc = -ENOENT;
6430         } else {
6431                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6432                *(__u64 *)data->ioc_inlbuf2 = version;
6433                 rc = 0;
6434         }
6435         mdt_object_unlock_put(mti, obj, lh, 1);
6436         RETURN(rc);
6437 }
6438
6439 /* ioctls on obd dev */
6440 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6441                          void *karg, void __user *uarg)
6442 {
6443         struct lu_env      env;
6444         struct obd_device *obd = exp->exp_obd;
6445         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6446         struct dt_device  *dt = mdt->mdt_bottom;
6447         int rc;
6448
6449         ENTRY;
6450         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6451         rc = lu_env_init(&env, LCT_MD_THREAD);
6452         if (rc)
6453                 RETURN(rc);
6454
6455         switch (cmd) {
6456         case OBD_IOC_SYNC:
6457                 rc = mdt_device_sync(&env, mdt);
6458                 break;
6459         case OBD_IOC_SET_READONLY:
6460                 rc = dt_sync(&env, dt);
6461                 if (rc == 0)
6462                         rc = dt_ro(&env, dt);
6463                 break;
6464         case OBD_IOC_ABORT_RECOVERY:
6465                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6466                 obd->obd_abort_recovery = 1;
6467                 target_stop_recovery_thread(obd);
6468                 rc = 0;
6469                 break;
6470         case OBD_IOC_CHANGELOG_REG:
6471         case OBD_IOC_CHANGELOG_DEREG:
6472         case OBD_IOC_CHANGELOG_CLEAR:
6473                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6474                                                            mdt->mdt_child,
6475                                                            cmd, len, karg);
6476                 break;
6477         case OBD_IOC_START_LFSCK: {
6478                 struct md_device *next = mdt->mdt_child;
6479                 struct obd_ioctl_data *data = karg;
6480                 struct lfsck_start_param lsp;
6481
6482                 if (unlikely(data == NULL)) {
6483                         rc = -EINVAL;
6484                         break;
6485                 }
6486
6487                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6488                 lsp.lsp_index_valid = 0;
6489                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6490                 break;
6491         }
6492         case OBD_IOC_STOP_LFSCK: {
6493                 struct md_device        *next = mdt->mdt_child;
6494                 struct obd_ioctl_data   *data = karg;
6495                 struct lfsck_stop        stop;
6496
6497                 stop.ls_status = LS_STOPPED;
6498                 /* Old lfsck utils may pass NULL @stop. */
6499                 if (data->ioc_inlbuf1 == NULL)
6500                         stop.ls_flags = 0;
6501                 else
6502                         stop.ls_flags =
6503                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6504
6505                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6506                 break;
6507         }
6508         case OBD_IOC_QUERY_LFSCK: {
6509                 struct md_device        *next = mdt->mdt_child;
6510                 struct obd_ioctl_data   *data = karg;
6511
6512                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6513                                                  data->ioc_inlbuf1);
6514                 break;
6515         }
6516         case OBD_IOC_GET_OBJ_VERSION: {
6517                 struct mdt_thread_info *mti;
6518                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6519                 memset(mti, 0, sizeof *mti);
6520                 mti->mti_env = &env;
6521                 mti->mti_mdt = mdt;
6522                 mti->mti_exp = exp;
6523
6524                 rc = mdt_ioc_version_get(mti, karg);
6525                 break;
6526         }
6527         case OBD_IOC_CATLOGLIST: {
6528                 struct mdt_thread_info *mti;
6529
6530                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6531                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6532                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6533                                        &mti->mti_tmp_fid1);
6534                 break;
6535          }
6536         default:
6537                 rc = -EOPNOTSUPP;
6538                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6539                         mdt_obd_name(mdt), cmd, rc);
6540         }
6541
6542         lu_env_fini(&env);
6543         RETURN(rc);
6544 }
6545
6546 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6547 {
6548         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6549         int rc;
6550         ENTRY;
6551
6552         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6553                 struct lfsck_start_param lsp;
6554
6555                 lsp.lsp_start = NULL;
6556                 lsp.lsp_index_valid = 0;
6557                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6558                                                            OBD_IOC_START_LFSCK,
6559                                                            0, &lsp);
6560                 if (rc != 0 && rc != -EALREADY)
6561                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6562                               mdt_obd_name(mdt), rc);
6563         }
6564
6565         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6566         RETURN(rc);
6567 }
6568
6569 static int mdt_obd_postrecov(struct obd_device *obd)
6570 {
6571         struct lu_env env;
6572         int rc;
6573
6574         rc = lu_env_init(&env, LCT_MD_THREAD);
6575         if (rc)
6576                 RETURN(rc);
6577         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6578         lu_env_fini(&env);
6579         return rc;
6580 }
6581
6582 static struct obd_ops mdt_obd_device_ops = {
6583         .o_owner          = THIS_MODULE,
6584         .o_set_info_async = mdt_obd_set_info_async,
6585         .o_connect        = mdt_obd_connect,
6586         .o_reconnect      = mdt_obd_reconnect,
6587         .o_disconnect     = mdt_obd_disconnect,
6588         .o_init_export    = mdt_init_export,
6589         .o_destroy_export = mdt_destroy_export,
6590         .o_iocontrol      = mdt_iocontrol,
6591         .o_postrecov      = mdt_obd_postrecov,
6592         /* Data-on-MDT IO methods */
6593         .o_preprw         = mdt_obd_preprw,
6594         .o_commitrw       = mdt_obd_commitrw,
6595 };
6596
6597 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6598                                          struct lu_device *d)
6599 {
6600         struct mdt_device *m = mdt_dev(d);
6601         ENTRY;
6602
6603         mdt_fini(env, m);
6604         RETURN(NULL);
6605 }
6606
6607 static struct lu_device *mdt_device_free(const struct lu_env *env,
6608                                          struct lu_device *d)
6609 {
6610         struct mdt_device *m = mdt_dev(d);
6611         ENTRY;
6612
6613         lu_device_fini(&m->mdt_lu_dev);
6614         OBD_FREE_PTR(m);
6615
6616         RETURN(NULL);
6617 }
6618
6619 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6620                                           struct lu_device_type *t,
6621                                           struct lustre_cfg *cfg)
6622 {
6623         struct lu_device  *l;
6624         struct mdt_device *m;
6625
6626         OBD_ALLOC_PTR(m);
6627         if (m != NULL) {
6628                 int rc;
6629
6630                 l = &m->mdt_lu_dev;
6631                 rc = mdt_init0(env, m, t, cfg);
6632                 if (rc != 0) {
6633                         mdt_device_free(env, l);
6634                         l = ERR_PTR(rc);
6635                         return l;
6636                 }
6637         } else
6638                 l = ERR_PTR(-ENOMEM);
6639         return l;
6640 }
6641
6642 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6643 LU_KEY_INIT(mdt, struct mdt_thread_info);
6644
6645 static void mdt_key_fini(const struct lu_context *ctx,
6646                          struct lu_context_key *key, void* data)
6647 {
6648         struct mdt_thread_info *info = data;
6649
6650         if (info->mti_big_lmm) {
6651                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6652                 info->mti_big_lmm = NULL;
6653                 info->mti_big_lmmsize = 0;
6654         }
6655
6656         if (info->mti_big_acl) {
6657                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6658                 info->mti_big_acl = NULL;
6659                 info->mti_big_aclsize = 0;
6660         }
6661
6662         OBD_FREE_PTR(info);
6663 }
6664
6665 /* context key: mdt_thread_key */
6666 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6667
6668 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6669 {
6670         return lu_ucred(info->mti_env);
6671 }
6672
6673 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6674 {
6675         return lu_ucred_check(info->mti_env);
6676 }
6677
6678 /**
6679  * Enable/disable COS (Commit On Sharing).
6680  *
6681  * Set/Clear the COS flag in mdt options.
6682  *
6683  * \param mdt mdt device
6684  * \param val 0 disables COS, other values enable COS
6685  */
6686 void mdt_enable_cos(struct mdt_device *mdt, bool val)
6687 {
6688         struct lu_env env;
6689         int rc;
6690
6691         mdt->mdt_opts.mo_cos = val;
6692         rc = lu_env_init(&env, LCT_LOCAL);
6693         if (unlikely(rc != 0)) {
6694                 CWARN("%s: lu_env initialization failed, cannot "
6695                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6696                 return;
6697         }
6698         mdt_device_sync(&env, mdt);
6699         lu_env_fini(&env);
6700 }
6701
6702 /**
6703  * Check COS (Commit On Sharing) status.
6704  *
6705  * Return COS flag status.
6706  *
6707  * \param mdt mdt device
6708  */
6709 int mdt_cos_is_enabled(struct mdt_device *mdt)
6710 {
6711         return mdt->mdt_opts.mo_cos != 0;
6712 }
6713
6714 static struct lu_device_type_operations mdt_device_type_ops = {
6715         .ldto_device_alloc = mdt_device_alloc,
6716         .ldto_device_free  = mdt_device_free,
6717         .ldto_device_fini  = mdt_device_fini
6718 };
6719
6720 static struct lu_device_type mdt_device_type = {
6721         .ldt_tags     = LU_DEVICE_MD,
6722         .ldt_name     = LUSTRE_MDT_NAME,
6723         .ldt_ops      = &mdt_device_type_ops,
6724         .ldt_ctx_tags = LCT_MD_THREAD
6725 };
6726
6727 static int __init mdt_init(void)
6728 {
6729         int rc;
6730
6731         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6732                  FID_NOBRACE_LEN + 1);
6733         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6734                  FID_LEN + 1);
6735         rc = lu_kmem_init(mdt_caches);
6736         if (rc)
6737                 return rc;
6738
6739         rc = mds_mod_init();
6740         if (rc)
6741                 GOTO(lu_fini, rc);
6742
6743         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6744                                  LUSTRE_MDT_NAME, &mdt_device_type);
6745         if (rc)
6746                 GOTO(mds_fini, rc);
6747 lu_fini:
6748         if (rc)
6749                 lu_kmem_fini(mdt_caches);
6750 mds_fini:
6751         if (rc)
6752                 mds_mod_exit();
6753         return rc;
6754 }
6755
6756 static void __exit mdt_exit(void)
6757 {
6758         class_unregister_type(LUSTRE_MDT_NAME);
6759         mds_mod_exit();
6760         lu_kmem_fini(mdt_caches);
6761 }
6762
6763 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6764 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6765 MODULE_VERSION(LUSTRE_VERSION_STRING);
6766 MODULE_LICENSE("GPL");
6767
6768 module_init(mdt_init);
6769 module_exit(mdt_exit);