lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         MIN(mdt->mdt_max_ea_size,
 602                                                             XATTR_SIZE_MAX);
 603                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 604                                                         info->mti_big_aclsize);
 605                                         if (info->mti_big_acl == NULL) {
 606                                                 info->mti_big_aclsize = 0;
 607                                                 CERROR("%s: unable to grow "
 608                                                        DFID" ACL buffer\n",
 609                                                        mdt_obd_name(mdt),
 610                                                        PFID(mdt_object_fid(o)));
 611                                                 RETURN(-ENOMEM);
 612                                         }
 613                                 }
 614
 615                                 CDEBUG(D_INODE, "%s: grow the "DFID
 616                                        " ACL buffer to size %d\n",
 617                                        mdt_obd_name(mdt),
 618                                        PFID(mdt_object_fid(o)),
 619                                        info->mti_big_aclsize);
 620
 621                                 buf->lb_buf = info->mti_big_acl;
 622                                 buf->lb_len = info->mti_big_aclsize;
 623
 624                                 goto again;
 625                         }
 626
 627                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 628                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 629                 }
 630         } else {
 631                 int client;
 632                 int server;
 633                 int acl_buflen;
 634                 int lmm_buflen = 0;
 635                 int lmmsize = 0;
 636
 637                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638                 if (acl_buflen >= rc)
 639                         goto map;
 640
 641                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 642                 client = ptlrpc_req_get_repsize(pill->rc_req);
 643                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 644                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 645                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 646                                                           RCL_SERVER);
 647                         lmmsize = repbody->mbo_eadatasize;
 648                 }
 649
 650                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 651                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 652                                "is not big enough with the ACL size %d (%d)\n",
 653                                mdt_obd_name(mdt), client, rc,
 654                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 655                         repbody->mbo_aclsize = 0;
 656                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 657                         RETURN(-ERANGE);
 658                 }
 659
 660 map:
 661                 if (buf->lb_buf == info->mti_big_acl)
 662                         info->mti_big_acl_used = 1;
 663
 664                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 665                                      rc, NODEMAP_FS_TO_CLIENT);
 666                 /* if all ACLs mapped out, rc is still >= 0 */
 667                 if (rc < 0) {
 668                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 669                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 670                                PFID(mdt_object_fid(o)), rc);
 671                         repbody->mbo_aclsize = 0;
 672                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 673                 } else {
 674                         repbody->mbo_aclsize = rc;
 675                         repbody->mbo_valid |= OBD_MD_FLACL;
 676                         rc = 0;
 677                 }
 678         }
 679
 680         RETURN(rc);
 681 }
 682 #endif
 683
 684 /* XXX Look into layout in MDT layer. */
 685 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 686 {
 687         struct lov_comp_md_v1   *comp_v1;
 688         struct lov_mds_md       *v1;
 689         int                      i;
 690
 691         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 692                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 693
 694                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 695                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 696                                 comp_v1->lcm_entries[i].lcme_offset);
 697                         /* We don't support partial release for now */
 698                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 699                                 return false;
 700                 }
 701                 return true;
 702         } else {
 703                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 704                         true : false;
 705         }
 706 }
 707
 708 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 709                         const struct lu_attr *attr, const struct lu_fid *fid)
 710 {
 711         struct md_attr *ma = &info->mti_attr;
 712         struct obd_export *exp = info->mti_exp;
 713         struct lu_nodemap *nodemap = NULL;
 714
 715         LASSERT(ma->ma_valid & MA_INODE);
 716
 717         if (attr->la_valid & LA_ATIME) {
 718                 b->mbo_atime = attr->la_atime;
 719                 b->mbo_valid |= OBD_MD_FLATIME;
 720         }
 721         if (attr->la_valid & LA_MTIME) {
 722                 b->mbo_mtime = attr->la_mtime;
 723                 b->mbo_valid |= OBD_MD_FLMTIME;
 724         }
 725         if (attr->la_valid & LA_CTIME) {
 726                 b->mbo_ctime = attr->la_ctime;
 727                 b->mbo_valid |= OBD_MD_FLCTIME;
 728         }
 729         if (attr->la_valid & LA_FLAGS) {
 730                 b->mbo_flags = attr->la_flags;
 731                 b->mbo_valid |= OBD_MD_FLFLAGS;
 732         }
 733         if (attr->la_valid & LA_NLINK) {
 734                 b->mbo_nlink = attr->la_nlink;
 735                 b->mbo_valid |= OBD_MD_FLNLINK;
 736         }
 737         if (attr->la_valid & (LA_UID|LA_GID)) {
 738                 nodemap = nodemap_get_from_exp(exp);
 739                 if (IS_ERR(nodemap))
 740                         goto out;
 741         }
 742         if (attr->la_valid & LA_UID) {
 743                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 744                                             NODEMAP_FS_TO_CLIENT,
 745                                             attr->la_uid);
 746                 b->mbo_valid |= OBD_MD_FLUID;
 747         }
 748         if (attr->la_valid & LA_GID) {
 749                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 750                                             NODEMAP_FS_TO_CLIENT,
 751                                             attr->la_gid);
 752                 b->mbo_valid |= OBD_MD_FLGID;
 753         }
 754
 755         if (attr->la_valid & LA_PROJID) {
 756                 /* TODO, nodemap for project id */
 757                 b->mbo_projid = attr->la_projid;
 758                 b->mbo_valid |= OBD_MD_FLPROJID;
 759         }
 760
 761         b->mbo_mode = attr->la_mode;
 762         if (attr->la_valid & LA_MODE)
 763                 b->mbo_valid |= OBD_MD_FLMODE;
 764         if (attr->la_valid & LA_TYPE)
 765                 b->mbo_valid |= OBD_MD_FLTYPE;
 766
 767         if (fid != NULL) {
 768                 b->mbo_fid1 = *fid;
 769                 b->mbo_valid |= OBD_MD_FLID;
 770                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 771                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 772         }
 773
 774         if (!(attr->la_valid & LA_TYPE))
 775                 return;
 776
 777         b->mbo_rdev   = attr->la_rdev;
 778         b->mbo_size   = attr->la_size;
 779         b->mbo_blocks = attr->la_blocks;
 780
 781         if (!S_ISREG(attr->la_mode)) {
 782                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 783         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 784                 /* means no objects are allocated on osts. */
 785                 LASSERT(!(ma->ma_valid & MA_LOV));
 786                 /* just ignore blocks occupied by extend attributes on MDS */
 787                 b->mbo_blocks = 0;
 788                 /* if no object is allocated on osts, the size on mds is valid.
 789                  * b=22272 */
 790                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 791         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 792                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 793                         /* A released file stores its size on MDS. */
 794                         /* But return 1 block for released file, unless tools
 795                          * like tar will consider it fully sparse. (LU-3864)
 796                          */
 797                         if (unlikely(b->mbo_size == 0))
 798                                 b->mbo_blocks = 0;
 799                         else
 800                                 b->mbo_blocks = 1;
 801                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 802                 } else if (info->mti_som_valid) { /* som is valid */
 803                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 804                 }
 805         }
 806
 807         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 808                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 809                        PFID(fid), (unsigned long long)b->mbo_size);
 810
 811 out:
 812         if (!IS_ERR_OR_NULL(nodemap))
 813                 nodemap_putref(nodemap);
 814 }
 815
 816 static inline int mdt_body_has_lov(const struct lu_attr *la,
 817                                    const struct mdt_body *body)
 818 {
 819         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 820                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 821 }
 822
 823 void mdt_client_compatibility(struct mdt_thread_info *info)
 824 {
 825         struct mdt_body       *body;
 826         struct ptlrpc_request *req = mdt_info_req(info);
 827         struct obd_export     *exp = req->rq_export;
 828         struct md_attr        *ma = &info->mti_attr;
 829         struct lu_attr        *la = &ma->ma_attr;
 830         ENTRY;
 831
 832         if (exp_connect_layout(exp))
 833                 /* the client can deal with 16-bit lmm_stripe_count */
 834                 RETURN_EXIT;
 835
 836         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 837
 838         if (!mdt_body_has_lov(la, body))
 839                 RETURN_EXIT;
 840
 841         /* now we have a reply with a lov for a client not compatible with the
 842          * layout lock so we have to clean the layout generation number */
 843         if (S_ISREG(la->la_mode))
 844                 ma->ma_lmm->lmm_layout_gen = 0;
 845         EXIT;
 846 }
 847
 848 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 849                                    struct mdt_object *o)
 850 {
 851         const struct lu_env *env = info->mti_env;
 852         int rc, rc2;
 853
 854         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 855                           XATTR_NAME_LOV);
 856
 857         if (rc == -ENODATA)
 858                 rc = 0;
 859
 860         if (rc < 0)
 861                 goto out;
 862
 863         /* Is it a directory? Let's check for the LMV as well */
 864         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 865                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 866                                    XATTR_NAME_LMV);
 867
 868                 if (rc2 == -ENODATA)
 869                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 870                                            &LU_BUF_NULL,
 871                                            XATTR_NAME_DEFAULT_LMV);
 872
 873                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 874                         rc = rc2;
 875         }
 876
 877 out:
 878         return rc;
 879 }
 880
 881 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 882                       const char *name)
 883 {
 884         const struct lu_env *env = info->mti_env;
 885         int rc;
 886         ENTRY;
 887
 888         LASSERT(info->mti_big_lmm_used == 0);
 889         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 890         if (rc < 0)
 891                 RETURN(rc);
 892
 893         /* big_lmm may need to be grown */
 894         if (info->mti_big_lmmsize < rc) {
 895                 int size = size_roundup_power2(rc);
 896
 897                 if (info->mti_big_lmmsize > 0) {
 898                         /* free old buffer */
 899                         LASSERT(info->mti_big_lmm);
 900                         OBD_FREE_LARGE(info->mti_big_lmm,
 901                                        info->mti_big_lmmsize);
 902                         info->mti_big_lmm = NULL;
 903                         info->mti_big_lmmsize = 0;
 904                 }
 905
 906                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 907                 if (info->mti_big_lmm == NULL)
 908                         RETURN(-ENOMEM);
 909                 info->mti_big_lmmsize = size;
 910         }
 911         LASSERT(info->mti_big_lmmsize >= rc);
 912
 913         info->mti_buf.lb_buf = info->mti_big_lmm;
 914         info->mti_buf.lb_len = info->mti_big_lmmsize;
 915         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 916
 917         RETURN(rc);
 918 }
 919
 920 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 921                    struct md_attr *ma, const char *name)
 922 {
 923         struct md_object *next = mdt_object_child(o);
 924         struct lu_buf    *buf = &info->mti_buf;
 925         int rc;
 926
 927         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 928                 buf->lb_buf = ma->ma_lmm;
 929                 buf->lb_len = ma->ma_lmm_size;
 930                 LASSERT(!(ma->ma_valid & MA_LOV));
 931         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 932                 buf->lb_buf = ma->ma_lmv;
 933                 buf->lb_len = ma->ma_lmv_size;
 934                 LASSERT(!(ma->ma_valid & MA_LMV));
 935         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 936                 buf->lb_buf = ma->ma_default_lmv;
 937                 buf->lb_len = ma->ma_default_lmv_size;
 938                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 939         } else {
 940                 return -EINVAL;
 941         }
 942
 943         LASSERT(buf->lb_buf);
 944
 945         rc = mo_xattr_get(info->mti_env, next, buf, name);
 946         if (rc > 0) {
 947
 948 got:
 949                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 950                         if (info->mti_big_lmm_used)
 951                                 ma->ma_lmm = info->mti_big_lmm;
 952
 953                         /* NOT return LOV EA with hole to old client. */
 954                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 955                                      LOV_PATTERN_F_HOLE) &&
 956                             !(exp_connect_flags(info->mti_exp) &
 957                               OBD_CONNECT_LFSCK)) {
 958                                 return -EIO;
 959                         } else {
 960                                 ma->ma_lmm_size = rc;
 961                                 ma->ma_valid |= MA_LOV;
 962                         }
 963                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 964                         if (info->mti_big_lmm_used)
 965                                 ma->ma_lmv = info->mti_big_lmm;
 966
 967                         ma->ma_lmv_size = rc;
 968                         ma->ma_valid |= MA_LMV;
 969                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 970                         ma->ma_default_lmv_size = rc;
 971                         ma->ma_valid |= MA_LMV_DEF;
 972                 }
 973
 974                 /* Update mdt_max_mdsize so all clients will be aware that */
 975                 if (info->mti_mdt->mdt_max_mdsize < rc)
 976                         info->mti_mdt->mdt_max_mdsize = rc;
 977
 978                 rc = 0;
 979         } else if (rc == -ENODATA) {
 980                 /* no LOV EA */
 981                 rc = 0;
 982         } else if (rc == -ERANGE) {
 983                 /* Default LMV has fixed size, so it must be able to fit
 984                  * in the original buffer */
 985                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 986                         return rc;
 987                 rc = mdt_big_xattr_get(info, o, name);
 988                 if (rc > 0) {
 989                         info->mti_big_lmm_used = 1;
 990                         goto got;
 991                 }
 992         }
 993
 994         return rc;
 995 }
 996
 997 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
 998                       struct lu_fid *pfid)
 999 {
1000         struct lu_buf           *buf = &info->mti_buf;
1001         struct link_ea_header   *leh;
1002         struct link_ea_entry    *lee;
1003         int                      rc;
1004         ENTRY;
1005
1006         buf->lb_buf = info->mti_big_lmm;
1007         buf->lb_len = info->mti_big_lmmsize;
1008         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1009                           buf, XATTR_NAME_LINK);
1010         /* ignore errors, MA_PFID won't be set and it is
1011          * up to the caller to treat this as an error */
1012         if (rc == -ERANGE || buf->lb_len == 0) {
1013                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1014                 buf->lb_buf = info->mti_big_lmm;
1015                 buf->lb_len = info->mti_big_lmmsize;
1016         }
1017
1018         if (rc < 0)
1019                 RETURN(rc);
1020         if (rc < sizeof(*leh)) {
1021                 CERROR("short LinkEA on "DFID": rc = %d\n",
1022                        PFID(mdt_object_fid(o)), rc);
1023                 RETURN(-ENODATA);
1024         }
1025
1026         leh = (struct link_ea_header *) buf->lb_buf;
1027         lee = (struct link_ea_entry *)(leh + 1);
1028         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1029                 leh->leh_magic = LINK_EA_MAGIC;
1030                 leh->leh_reccount = __swab32(leh->leh_reccount);
1031                 leh->leh_len = __swab64(leh->leh_len);
1032         }
1033         if (leh->leh_magic != LINK_EA_MAGIC)
1034                 RETURN(-EINVAL);
1035         if (leh->leh_reccount == 0)
1036                 RETURN(-ENODATA);
1037
1038         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1039         fid_be_to_cpu(pfid, pfid);
1040
1041         RETURN(0);
1042 }
1043
1044 int mdt_attr_get_complex(struct mdt_thread_info *info,
1045                          struct mdt_object *o, struct md_attr *ma)
1046 {
1047         const struct lu_env *env = info->mti_env;
1048         struct md_object    *next = mdt_object_child(o);
1049         struct lu_buf       *buf = &info->mti_buf;
1050         int                  need = ma->ma_need;
1051         int                  rc = 0, rc2;
1052         u32                  mode;
1053         ENTRY;
1054
1055         ma->ma_valid = 0;
1056
1057         if (mdt_object_exists(o) == 0)
1058                 GOTO(out, rc = -ENOENT);
1059         mode = lu_object_attr(&next->mo_lu);
1060
1061         if (need & MA_INODE) {
1062                 ma->ma_need = MA_INODE;
1063                 rc = mo_attr_get(env, next, ma);
1064                 if (rc)
1065                         GOTO(out, rc);
1066
1067                 if (S_ISREG(mode))
1068                         (void) mdt_get_som(info, o, ma);
1069                 ma->ma_valid |= MA_INODE;
1070         }
1071
1072         if (need & MA_PFID) {
1073                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1074                 if (rc == 0)
1075                         ma->ma_valid |= MA_PFID;
1076                 /* ignore this error, parent fid is not mandatory */
1077                 rc = 0;
1078         }
1079
1080         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1081                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1082                 if (rc)
1083                         GOTO(out, rc);
1084         }
1085
1086         if (need & MA_LMV && S_ISDIR(mode)) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1088                 if (rc != 0)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         /*
1099          * In the handle of MA_INODE, we may already get the SOM attr.
1100          */
1101         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1102                 rc = mdt_get_som(info, o, ma);
1103                 if (rc != 0)
1104                         GOTO(out, rc);
1105         }
1106
1107         if (need & MA_HSM && S_ISREG(mode)) {
1108                 buf->lb_buf = info->mti_xattr_buf;
1109                 buf->lb_len = sizeof(info->mti_xattr_buf);
1110                 CLASSERT(sizeof(struct hsm_attrs) <=
1111                          sizeof(info->mti_xattr_buf));
1112                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1113                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1114                 if (rc2 == 0)
1115                         ma->ma_valid |= MA_HSM;
1116                 else if (rc2 < 0 && rc2 != -ENODATA)
1117                         GOTO(out, rc = rc2);
1118         }
1119
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1122                 buf->lb_buf = ma->ma_acl;
1123                 buf->lb_len = ma->ma_acl_size;
1124                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1125                 if (rc2 > 0) {
1126                         ma->ma_acl_size = rc2;
1127                         ma->ma_valid |= MA_ACL_DEF;
1128                 } else if (rc2 == -ENODATA) {
1129                         /* no ACLs */
1130                         ma->ma_acl_size = 0;
1131                 } else
1132                         GOTO(out, rc = rc2);
1133         }
1134 #endif
1135 out:
1136         ma->ma_need = need;
1137         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1138                rc, ma->ma_valid, ma->ma_lmm);
1139         RETURN(rc);
1140 }
1141
1142 static int mdt_getattr_internal(struct mdt_thread_info *info,
1143                                 struct mdt_object *o, int ma_need)
1144 {
1145         struct md_object        *next = mdt_object_child(o);
1146         const struct mdt_body   *reqbody = info->mti_body;
1147         struct ptlrpc_request   *req = mdt_info_req(info);
1148         struct md_attr          *ma = &info->mti_attr;
1149         struct lu_attr          *la = &ma->ma_attr;
1150         struct req_capsule      *pill = info->mti_pill;
1151         const struct lu_env     *env = info->mti_env;
1152         struct mdt_body         *repbody;
1153         struct lu_buf           *buffer = &info->mti_buf;
1154         struct obd_export       *exp = info->mti_exp;
1155         int                      rc;
1156         ENTRY;
1157
1158         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1159                 RETURN(err_serious(-ENOMEM));
1160
1161         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162
1163         ma->ma_valid = 0;
1164
1165         if (mdt_object_remote(o)) {
1166                 /* This object is located on remote node.*/
1167                 /* Return -ENOTSUPP for old client */
1168                 if (!mdt_is_dne_client(req->rq_export))
1169                         GOTO(out, rc = -ENOTSUPP);
1170
1171                 repbody->mbo_fid1 = *mdt_object_fid(o);
1172                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1173                 GOTO(out, rc = 0);
1174         }
1175
1176         if (reqbody->mbo_eadatasize > 0) {
1177                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1178                 if (buffer->lb_buf == NULL)
1179                         GOTO(out, rc = -EPROTO);
1180                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1181                                                       RCL_SERVER);
1182         } else {
1183                 buffer->lb_buf = NULL;
1184                 buffer->lb_len = 0;
1185                 ma_need &= ~(MA_LOV | MA_LMV);
1186                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1187                        mdt_obd_name(info->mti_mdt),
1188                        req->rq_export->exp_client_uuid.uuid);
1189         }
1190
1191         /* from 2.12.58 intent_getattr pack default LMV in reply */
1192         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1193             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1194                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1195             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1196                                   RCL_SERVER)) {
1197                 ma->ma_lmv = buffer->lb_buf;
1198                 ma->ma_lmv_size = buffer->lb_len;
1199                 ma->ma_default_lmv = req_capsule_server_get(pill,
1200                                                 &RMF_DEFAULT_MDT_MD);
1201                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1202                                                 &RMF_DEFAULT_MDT_MD,
1203                                                 RCL_SERVER);
1204                 ma->ma_need = MA_INODE;
1205                 if (ma->ma_lmv_size > 0)
1206                         ma->ma_need |= MA_LMV;
1207                 if (ma->ma_default_lmv_size > 0)
1208                         ma->ma_need |= MA_LMV_DEF;
1209         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1210                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1211                 /* If it is dir and client require MEA, then we got MEA */
1212                 /* Assumption: MDT_MD size is enough for lmv size. */
1213                 ma->ma_lmv = buffer->lb_buf;
1214                 ma->ma_lmv_size = buffer->lb_len;
1215                 ma->ma_need = MA_INODE;
1216                 if (ma->ma_lmv_size > 0) {
1217                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1218                                 ma->ma_need |= MA_LMV;
1219                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1220                                 ma->ma_need |= MA_LMV_DEF;
1221                                 ma->ma_default_lmv = buffer->lb_buf;
1222                                 ma->ma_lmv = NULL;
1223                                 ma->ma_default_lmv_size = buffer->lb_len;
1224                                 ma->ma_lmv_size = 0;
1225                         }
1226                 }
1227         } else {
1228                 ma->ma_lmm = buffer->lb_buf;
1229                 ma->ma_lmm_size = buffer->lb_len;
1230                 ma->ma_need = MA_INODE | MA_HSM;
1231                 if (ma->ma_lmm_size > 0) {
1232                         ma->ma_need |= MA_LOV;
1233                         /* Older clients may crash if they getattr overstriped
1234                          * files
1235                          */
1236                         if (!exp_connect_overstriping(exp) &&
1237                             mdt_lmm_is_overstriping(ma->ma_lmm))
1238                                 RETURN(-EOPNOTSUPP);
1239                 }
1240         }
1241
1242         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1243             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1244             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1245                 /* get default stripe info for this dir. */
1246                 ma->ma_need |= MA_LOV_DEF;
1247         }
1248         ma->ma_need |= ma_need;
1249
1250         rc = mdt_attr_get_complex(info, o, ma);
1251         if (unlikely(rc)) {
1252                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1253                        "%s: getattr error for "DFID": rc = %d\n",
1254                        mdt_obd_name(info->mti_mdt),
1255                        PFID(mdt_object_fid(o)), rc);
1256                 RETURN(rc);
1257         }
1258
1259         /* if file is released, check if a restore is running */
1260         if (ma->ma_valid & MA_HSM) {
1261                 repbody->mbo_valid |= OBD_MD_TSTATE;
1262                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1263                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1264                         repbody->mbo_t_state = MS_RESTORE;
1265         }
1266
1267         if (likely(ma->ma_valid & MA_INODE))
1268                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1269         else
1270                 RETURN(-EFAULT);
1271
1272         if (mdt_body_has_lov(la, reqbody)) {
1273                 if (ma->ma_valid & MA_LOV) {
1274                         LASSERT(ma->ma_lmm_size);
1275                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1276                         if (S_ISDIR(la->la_mode))
1277                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1278                         else
1279                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1280                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1281                 }
1282                 if (ma->ma_valid & MA_LMV) {
1283                         /* Return -ENOTSUPP for old client */
1284                         if (!mdt_is_striped_client(req->rq_export))
1285                                 RETURN(-ENOTSUPP);
1286
1287                         LASSERT(S_ISDIR(la->la_mode));
1288                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1289                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1290                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1291                 }
1292                 if (ma->ma_valid & MA_LMV_DEF) {
1293                         /* Return -ENOTSUPP for old client */
1294                         if (!mdt_is_striped_client(req->rq_export))
1295                                 RETURN(-ENOTSUPP);
1296                         LASSERT(S_ISDIR(la->la_mode));
1297                         /*
1298                          * when ll_dir_getstripe() gets default LMV, it
1299                          * checks mbo_eadatasize.
1300                          */
1301                         if (!(ma->ma_valid & MA_LMV))
1302                                 repbody->mbo_eadatasize =
1303                                         ma->ma_default_lmv_size;
1304                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1305                                                OBD_MD_DEFAULT_MEA);
1306                 }
1307         } else if (S_ISLNK(la->la_mode) &&
1308                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1309                 buffer->lb_buf = ma->ma_lmm;
1310                 /* eadatasize from client includes NULL-terminator, so
1311                  * there is no need to read it */
1312                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1313                 rc = mo_readlink(env, next, buffer);
1314                 if (unlikely(rc <= 0)) {
1315                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1316                                mdt_obd_name(info->mti_mdt),
1317                                PFID(mdt_object_fid(o)), rc);
1318                         rc = -EFAULT;
1319                 } else {
1320                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1321
1322                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1323                                 rc -= 2;
1324                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1325                         /* we need to report back size with NULL-terminator
1326                          * because client expects that */
1327                         repbody->mbo_eadatasize = rc + 1;
1328                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1329                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1330                                        "on "DFID ", expected %d\n",
1331                                        mdt_obd_name(info->mti_mdt),
1332                                        rc, PFID(mdt_object_fid(o)),
1333                                        reqbody->mbo_eadatasize - 1);
1334                         /* NULL terminate */
1335                         ((char *)ma->ma_lmm)[rc] = 0;
1336
1337                         /* If the total CDEBUG() size is larger than a page, it
1338                          * will print a warning to the console, avoid this by
1339                          * printing just the last part of the symlink. */
1340                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1341                                print_limit < rc ? "..." : "", print_limit,
1342                                (char *)ma->ma_lmm + rc - print_limit, rc);
1343                         rc = 0;
1344                 }
1345         }
1346
1347         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1348                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1349                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1350                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1351                        repbody->mbo_max_mdsize);
1352         }
1353
1354 #ifdef CONFIG_FS_POSIX_ACL
1355         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1356                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1357                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1358                 if (IS_ERR(nodemap))
1359                         RETURN(PTR_ERR(nodemap));
1360
1361                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1362                 nodemap_putref(nodemap);
1363         }
1364 #endif
1365
1366 out:
1367         if (rc == 0)
1368                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1369
1370         RETURN(rc);
1371 }
1372
1373 static int mdt_getattr(struct tgt_session_info *tsi)
1374 {
1375         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1376         struct mdt_object       *obj = info->mti_object;
1377         struct req_capsule      *pill = info->mti_pill;
1378         struct mdt_body         *reqbody;
1379         struct mdt_body         *repbody;
1380         int rc, rc2;
1381         ENTRY;
1382
1383         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1384         LASSERT(reqbody);
1385         LASSERT(obj != NULL);
1386         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1387
1388         /* Special case for Data-on-MDT files to get data version */
1389         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1390                 rc = mdt_data_version_get(tsi);
1391                 GOTO(out, rc);
1392         }
1393
1394         /* Unlike intent case where we need to pre-fill out buffers early on
1395          * in intent policy for ldlm reasons, here we can have a much better
1396          * guess at EA size by just reading it from disk.
1397          * Exceptions are readdir and (missing) directory striping */
1398         /* Readlink */
1399         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1400                 /* No easy way to know how long is the symlink, but it cannot
1401                  * be more than PATH_MAX, so we allocate +1 */
1402                 rc = PATH_MAX + 1;
1403         /* A special case for fs ROOT: getattr there might fetch
1404          * default EA for entire fs, not just for this dir!
1405          */
1406         } else if (lu_fid_eq(mdt_object_fid(obj),
1407                              &info->mti_mdt->mdt_md_root_fid) &&
1408                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1409                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1410                                                                  MDS_GETATTR)) {
1411                 /* Should the default strping be bigger, mdt_fix_reply
1412                  * will reallocate */
1413                 rc = DEF_REP_MD_SIZE;
1414         } else {
1415                 /* Read the actual EA size from disk */
1416                 rc = mdt_attr_get_eabuf_size(info, obj);
1417         }
1418
1419         if (rc < 0)
1420                 GOTO(out, rc = err_serious(rc));
1421
1422         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1423
1424         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1425          * by default. If the target object has more ACL entries, then
1426          * enlarge the buffer when necessary. */
1427         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1428                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1429
1430         rc = req_capsule_server_pack(pill);
1431         if (unlikely(rc != 0))
1432                 GOTO(out, rc = err_serious(rc));
1433
1434         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1435         LASSERT(repbody != NULL);
1436         repbody->mbo_eadatasize = 0;
1437         repbody->mbo_aclsize = 0;
1438
1439         rc = mdt_check_ucred(info);
1440         if (unlikely(rc))
1441                 GOTO(out_shrink, rc);
1442
1443         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1444
1445         rc = mdt_getattr_internal(info, obj, 0);
1446         EXIT;
1447 out_shrink:
1448         mdt_client_compatibility(info);
1449         rc2 = mdt_fix_reply(info);
1450         if (rc == 0)
1451                 rc = rc2;
1452 out:
1453         mdt_thread_info_fini(info);
1454         return rc;
1455 }
1456
1457 /**
1458  * Handler of layout intent RPC requiring the layout modification
1459  *
1460  * \param[in] info      thread environment
1461  * \param[in] obj       object
1462  * \param[in] layout    layout change descriptor
1463  *
1464  * \retval 0    on success
1465  * \retval < 0  error code
1466  */
1467 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1468                       struct md_layout_change *layout)
1469 {
1470         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1471         int rc;
1472         ENTRY;
1473
1474         if (!mdt_object_exists(obj))
1475                 GOTO(out, rc = -ENOENT);
1476
1477         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1478                 GOTO(out, rc = -EINVAL);
1479
1480         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1481                            MAY_WRITE);
1482         if (rc)
1483                 GOTO(out, rc);
1484
1485         /* take layout lock to prepare layout change */
1486         mdt_lock_reg_init(lh, LCK_EX);
1487         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1488         if (rc)
1489                 GOTO(out, rc);
1490
1491         mutex_lock(&obj->mot_som_mutex);
1492         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1493         mutex_unlock(&obj->mot_som_mutex);
1494         mdt_object_unlock(info, obj, lh, 1);
1495 out:
1496         RETURN(rc);
1497 }
1498
1499 /**
1500  * Exchange MOF_LOV_CREATED flags between two objects after a
1501  * layout swap. No assumption is made on whether o1 or o2 have
1502  * created objects or not.
1503  *
1504  * \param[in,out] o1    First swap layout object
1505  * \param[in,out] o2    Second swap layout object
1506  */
1507 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1508 {
1509         unsigned int o1_lov_created = o1->mot_lov_created;
1510
1511         mutex_lock(&o1->mot_lov_mutex);
1512         mutex_lock(&o2->mot_lov_mutex);
1513
1514         o1->mot_lov_created = o2->mot_lov_created;
1515         o2->mot_lov_created = o1_lov_created;
1516
1517         mutex_unlock(&o2->mot_lov_mutex);
1518         mutex_unlock(&o1->mot_lov_mutex);
1519 }
1520
1521 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1522 {
1523         struct mdt_thread_info  *info;
1524         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1525         struct obd_export       *exp = req->rq_export;
1526         struct mdt_object       *o1, *o2, *o;
1527         struct mdt_lock_handle  *lh1, *lh2;
1528         struct mdc_swap_layouts *msl;
1529         int                      rc;
1530         ENTRY;
1531
1532         /* client does not support layout lock, so layout swaping
1533          * is disabled.
1534          * FIXME: there is a problem for old clients which don't support
1535          * layout lock yet. If those clients have already opened the file
1536          * they won't be notified at all so that old layout may still be
1537          * used to do IO. This can be fixed after file release is landed by
1538          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1539         if (!exp_connect_layout(exp))
1540                 RETURN(-EOPNOTSUPP);
1541
1542         info = tsi2mdt_info(tsi);
1543
1544         if (info->mti_dlm_req != NULL)
1545                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1546
1547         o1 = info->mti_object;
1548         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1549                                 &info->mti_body->mbo_fid2);
1550         if (IS_ERR(o))
1551                 GOTO(out, rc = PTR_ERR(o));
1552
1553         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1554                 GOTO(put, rc = -ENOENT);
1555
1556         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1557         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1558                 GOTO(put, rc);
1559
1560         if (rc < 0)
1561                 swap(o1, o2);
1562
1563         /* permission check. Make sure the calling process having permission
1564          * to write both files. */
1565         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1566                            MAY_WRITE);
1567         if (rc < 0)
1568                 GOTO(put, rc);
1569
1570         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1571                            MAY_WRITE);
1572         if (rc < 0)
1573                 GOTO(put, rc);
1574
1575         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1576         if (msl == NULL)
1577                 GOTO(put, rc = -EPROTO);
1578
1579         lh1 = &info->mti_lh[MDT_LH_NEW];
1580         mdt_lock_reg_init(lh1, LCK_EX);
1581         lh2 = &info->mti_lh[MDT_LH_OLD];
1582         mdt_lock_reg_init(lh2, LCK_EX);
1583
1584         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1585                              MDS_INODELOCK_XATTR);
1586         if (rc < 0)
1587                 GOTO(put, rc);
1588
1589         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1590                              MDS_INODELOCK_XATTR);
1591         if (rc < 0)
1592                 GOTO(unlock1, rc);
1593
1594         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1595                              mdt_object_child(o2), msl->msl_flags);
1596         if (rc < 0)
1597                 GOTO(unlock2, rc);
1598
1599         mdt_swap_lov_flag(o1, o2);
1600
1601 unlock2:
1602         mdt_object_unlock(info, o2, lh2, rc);
1603 unlock1:
1604         mdt_object_unlock(info, o1, lh1, rc);
1605 put:
1606         mdt_object_put(info->mti_env, o);
1607 out:
1608         mdt_thread_info_fini(info);
1609         RETURN(rc);
1610 }
1611
1612 static int mdt_raw_lookup(struct mdt_thread_info *info,
1613                           struct mdt_object *parent,
1614                           const struct lu_name *lname,
1615                           struct ldlm_reply *ldlm_rep)
1616 {
1617         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1618         int              rc;
1619         ENTRY;
1620
1621         LASSERT(!info->mti_cross_ref);
1622
1623         /* Only got the fid of this obj by name */
1624         fid_zero(child_fid);
1625         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1626                         lname, child_fid, &info->mti_spec);
1627         if (rc == 0) {
1628                 struct mdt_body *repbody;
1629
1630                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1631                 repbody->mbo_fid1 = *child_fid;
1632                 repbody->mbo_valid = OBD_MD_FLID;
1633                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1634         } else if (rc == -ENOENT) {
1635                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1636         }
1637
1638         RETURN(rc);
1639 }
1640
1641 /*
1642  * UPDATE lock should be taken against parent, and be released before exit;
1643  * child_bits lock should be taken against child, and be returned back:
1644  *            (1)normal request should release the child lock;
1645  *            (2)intent request will grant the lock to client.
1646  */
1647 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1648                                  struct mdt_lock_handle *lhc,
1649                                  __u64 child_bits,
1650                                  struct ldlm_reply *ldlm_rep)
1651 {
1652         struct ptlrpc_request  *req = mdt_info_req(info);
1653         struct mdt_body        *reqbody = NULL;
1654         struct mdt_object      *parent = info->mti_object;
1655         struct mdt_object      *child;
1656         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1657         struct lu_name         *lname = NULL;
1658         struct mdt_lock_handle *lhp = NULL;
1659         struct ldlm_lock       *lock;
1660         struct req_capsule *pill = info->mti_pill;
1661         __u64 try_bits = 0;
1662         bool is_resent;
1663         int ma_need = 0;
1664         int rc;
1665
1666         ENTRY;
1667
1668         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1669         LASSERT(ergo(is_resent,
1670                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1671
1672         if (parent == NULL)
1673                 RETURN(-ENOENT);
1674
1675         if (info->mti_cross_ref) {
1676                 /* Only getattr on the child. Parent is on another node. */
1677                 mdt_set_disposition(info, ldlm_rep,
1678                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1679                 child = parent;
1680                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1681                        "ldlm_rep = %p\n",
1682                        PFID(mdt_object_fid(child)), ldlm_rep);
1683
1684                 rc = mdt_check_resent_lock(info, child, lhc);
1685                 if (rc < 0) {
1686                         RETURN(rc);
1687                 } else if (rc > 0) {
1688                         mdt_lock_handle_init(lhc);
1689                         mdt_lock_reg_init(lhc, LCK_PR);
1690
1691                         /*
1692                          * Object's name entry is on another MDS, it will
1693                          * request PERM lock only because LOOKUP lock is owned
1694                          * by the MDS where name entry resides.
1695                          *
1696                          * TODO: it should try layout lock too. - Jinshan
1697                          */
1698                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1699                                         MDS_INODELOCK_LAYOUT);
1700                         child_bits |= MDS_INODELOCK_PERM;
1701
1702                         rc = mdt_object_lock(info, child, lhc, child_bits);
1703                         if (rc < 0)
1704                                 RETURN(rc);
1705                 }
1706
1707                 /* Finally, we can get attr for child. */
1708                 if (!mdt_object_exists(child)) {
1709                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1710                                         &child->mot_obj,
1711                                         "remote object doesn't exist.");
1712                         mdt_object_unlock(info, child, lhc, 1);
1713                         RETURN(-ENOENT);
1714                 }
1715
1716                 rc = mdt_getattr_internal(info, child, 0);
1717                 if (unlikely(rc != 0))
1718                         mdt_object_unlock(info, child, lhc, 1);
1719
1720                 mdt_pack_secctx_in_reply(info, child);
1721
1722                 RETURN(rc);
1723         }
1724
1725         lname = &info->mti_name;
1726         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1727
1728         if (lu_name_is_valid(lname)) {
1729                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1730                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1731                        PNAME(lname), ldlm_rep);
1732         } else {
1733                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1734                 if (unlikely(reqbody == NULL))
1735                         RETURN(err_serious(-EPROTO));
1736
1737                 *child_fid = reqbody->mbo_fid2;
1738
1739                 if (unlikely(!fid_is_sane(child_fid)))
1740                         RETURN(err_serious(-EINVAL));
1741
1742                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1743                        "ldlm_rep = %p\n",
1744                        PFID(mdt_object_fid(parent)),
1745                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1746         }
1747
1748         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1749
1750         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1751                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1752                                 &parent->mot_obj,
1753                                 "Parent doesn't exist!");
1754                 RETURN(-ESTALE);
1755         }
1756
1757         if (mdt_object_remote(parent)) {
1758                 CERROR("%s: parent "DFID" is on remote target\n",
1759                        mdt_obd_name(info->mti_mdt),
1760                        PFID(mdt_object_fid(parent)));
1761                 RETURN(-EIO);
1762         }
1763
1764         if (lu_name_is_valid(lname)) {
1765                 /* Always allow to lookup ".." */
1766                 if (unlikely(lname->ln_namelen == 2 &&
1767                              lname->ln_name[0] == '.' &&
1768                              lname->ln_name[1] == '.'))
1769                         info->mti_spec.sp_permitted = 1;
1770
1771                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1772                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1773
1774                         RETURN(rc);
1775                 }
1776
1777                 /* step 1: lock parent only if parent is a directory */
1778                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1779                         lhp = &info->mti_lh[MDT_LH_PARENT];
1780                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1781                         rc = mdt_object_lock(info, parent, lhp,
1782                                              MDS_INODELOCK_UPDATE);
1783                         if (unlikely(rc != 0))
1784                                 RETURN(rc);
1785                 }
1786
1787                 /* step 2: lookup child's fid by name */
1788                 fid_zero(child_fid);
1789                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1790                                 child_fid, &info->mti_spec);
1791                 if (rc == -ENOENT)
1792                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1793
1794                 if (rc != 0)
1795                         GOTO(out_parent, rc);
1796         }
1797
1798         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1799
1800         /*
1801          *step 3: find the child object by fid & lock it.
1802          *        regardless if it is local or remote.
1803          *
1804          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1805          *      set parent dir fid the same as child fid in getattr by fid case
1806          *      we should not lu_object_find() the object again, could lead
1807          *      to hung if there is a concurrent unlink destroyed the object.
1808          */
1809         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1810                 mdt_object_get(info->mti_env, parent);
1811                 child = parent;
1812         } else {
1813                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1814                                         child_fid);
1815         }
1816
1817         if (unlikely(IS_ERR(child)))
1818                 GOTO(out_parent, rc = PTR_ERR(child));
1819
1820         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1821         if (!mdt_object_exists(child)) {
1822                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1823                                 &child->mot_obj,
1824                                 "Object doesn't exist!");
1825                 GOTO(out_child, rc = -ENOENT);
1826         }
1827
1828         rc = mdt_check_resent_lock(info, child, lhc);
1829         if (rc < 0) {
1830                 GOTO(out_child, rc);
1831         } else if (rc > 0) {
1832                 mdt_lock_handle_init(lhc);
1833                 mdt_lock_reg_init(lhc, LCK_PR);
1834
1835                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1836                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1837                         struct md_attr *ma = &info->mti_attr;
1838
1839                         ma->ma_valid = 0;
1840                         ma->ma_need = MA_INODE;
1841                         rc = mdt_attr_get_complex(info, child, ma);
1842                         if (unlikely(rc != 0))
1843                                 GOTO(out_child, rc);
1844
1845                         /* If the file has not been changed for some time, we
1846                          * return not only a LOOKUP lock, but also an UPDATE
1847                          * lock and this might save us RPC on later STAT. For
1848                          * directories, it also let negative dentry cache start
1849                          * working for this dir. */
1850                         if (ma->ma_valid & MA_INODE &&
1851                             ma->ma_attr.la_valid & LA_CTIME &&
1852                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1853                                 ma->ma_attr.la_ctime < ktime_get_real_seconds())
1854                                 child_bits |= MDS_INODELOCK_UPDATE;
1855                 }
1856
1857                 /* layout lock must be granted in a best-effort way
1858                  * for IT operations */
1859                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1860                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1861                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1862                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1863                             exp_connect_layout(info->mti_exp)) {
1864                                 /* try to grant layout lock for regular file. */
1865                                 try_bits = MDS_INODELOCK_LAYOUT;
1866                         }
1867                         /* Acquire DOM lock in advance for data-on-mdt file */
1868                         if (child != parent)
1869                                 try_bits |= MDS_INODELOCK_DOM;
1870                 }
1871
1872                 if (try_bits != 0) {
1873                         /* try layout lock, it may fail to be granted due to
1874                          * contention at LOOKUP or UPDATE */
1875                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1876                                                  try_bits, false);
1877                         if (child_bits & MDS_INODELOCK_LAYOUT)
1878                                 ma_need |= MA_LOV;
1879                 } else {
1880                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1881                          * client will enqueue the lock to the remote MDT */
1882                         if (mdt_object_remote(child))
1883                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1884                         rc = mdt_object_lock(info, child, lhc, child_bits);
1885                 }
1886                 if (unlikely(rc != 0))
1887                         GOTO(out_child, rc);
1888         }
1889
1890         /* finally, we can get attr for child. */
1891         rc = mdt_getattr_internal(info, child, ma_need);
1892         if (unlikely(rc != 0)) {
1893                 mdt_object_unlock(info, child, lhc, 1);
1894                 GOTO(out_child, rc);
1895         }
1896
1897         mdt_pack_secctx_in_reply(info, child);
1898
1899         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1900         if (lock) {
1901                 /* Debugging code. */
1902                 LDLM_DEBUG(lock, "Returning lock to client");
1903                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1904                                          &lock->l_resource->lr_name),
1905                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1906                          PLDLMRES(lock->l_resource),
1907                          PFID(mdt_object_fid(child)));
1908
1909                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1910                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1911                     child != parent) {
1912                         mdt_object_put(info->mti_env, child);
1913                         rc = mdt_pack_size2body(info, child_fid,
1914                                                 &lhc->mlh_reg_lh);
1915                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1916                                 /* DOM lock was taken in advance but this is
1917                                  * not DoM file. Drop the lock.
1918                                  */
1919                                 lock_res_and_lock(lock);
1920                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1921                                 unlock_res_and_lock(lock);
1922                         }
1923                         LDLM_LOCK_PUT(lock);
1924                         GOTO(out_parent, rc = 0);
1925                 }
1926                 LDLM_LOCK_PUT(lock);
1927         }
1928
1929         EXIT;
1930 out_child:
1931         mdt_object_put(info->mti_env, child);
1932 out_parent:
1933         if (lhp)
1934                 mdt_object_unlock(info, parent, lhp, 1);
1935         return rc;
1936 }
1937
1938 /* normal handler: should release the child lock */
1939 static int mdt_getattr_name(struct tgt_session_info *tsi)
1940 {
1941         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1942         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1943         struct mdt_body        *reqbody;
1944         struct mdt_body        *repbody;
1945         int rc, rc2;
1946         ENTRY;
1947
1948         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1949         LASSERT(reqbody != NULL);
1950         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1951         LASSERT(repbody != NULL);
1952
1953         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1954         repbody->mbo_eadatasize = 0;
1955         repbody->mbo_aclsize = 0;
1956
1957         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1958         if (unlikely(rc))
1959                 GOTO(out_shrink, rc);
1960
1961         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1962         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1963                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1964                 lhc->mlh_reg_lh.cookie = 0;
1965         }
1966         mdt_exit_ucred(info);
1967         EXIT;
1968 out_shrink:
1969         mdt_client_compatibility(info);
1970         rc2 = mdt_fix_reply(info);
1971         if (rc == 0)
1972                 rc = rc2;
1973         mdt_thread_info_fini(info);
1974         return rc;
1975 }
1976
1977 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
1978                             const struct lu_fid *pfid,
1979                             const struct lu_name *name,
1980                             struct mdt_object *obj, s64 ctime)
1981 {
1982         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1983         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
1984         struct mdt_device *mdt = info->mti_mdt;
1985         struct md_attr *ma = &info->mti_attr;
1986         struct mdt_lock_handle *parent_lh;
1987         struct mdt_lock_handle *child_lh;
1988         struct mdt_object *pobj;
1989         bool cos_incompat = false;
1990         int rc;
1991         ENTRY;
1992
1993         pobj = mdt_object_find(info->mti_env, mdt, pfid);
1994         if (IS_ERR(pobj))
1995                 GOTO(out, rc = PTR_ERR(pobj));
1996
1997         parent_lh = &info->mti_lh[MDT_LH_PARENT];
1998         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
1999         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2000         if (rc != 0)
2001                 GOTO(put_parent, rc);
2002
2003         if (mdt_object_remote(pobj))
2004                 cos_incompat = true;
2005
2006         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2007                         name, child_fid, &info->mti_spec);
2008         if (rc != 0)
2009                 GOTO(unlock_parent, rc);
2010
2011         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2012                 GOTO(unlock_parent, rc = -EREMCHG);
2013
2014         child_lh = &info->mti_lh[MDT_LH_CHILD];
2015         mdt_lock_reg_init(child_lh, LCK_EX);
2016         rc = mdt_reint_striped_lock(info, obj, child_lh,
2017                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2018                                     einfo, cos_incompat);
2019         if (rc != 0)
2020                 GOTO(unlock_parent, rc);
2021
2022         if (atomic_read(&obj->mot_open_count)) {
2023                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2024                        PFID(mdt_object_fid(obj)));
2025                 GOTO(unlock_child, rc = -EBUSY);
2026         }
2027
2028         ma->ma_need = 0;
2029         ma->ma_valid = MA_INODE;
2030         ma->ma_attr.la_valid = LA_CTIME;
2031         ma->ma_attr.la_ctime = ctime;
2032
2033         mutex_lock(&obj->mot_lov_mutex);
2034
2035         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2036                         mdt_object_child(obj), name, ma, 0);
2037
2038         mutex_unlock(&obj->mot_lov_mutex);
2039
2040 unlock_child:
2041         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2042 unlock_parent:
2043         mdt_object_unlock(info, pobj, parent_lh, 1);
2044 put_parent:
2045         mdt_object_put(info->mti_env, pobj);
2046 out:
2047         RETURN(rc);
2048 }
2049
2050 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2051                                         struct mdt_object *obj)
2052 {
2053         struct lu_ucred *uc = lu_ucred(info->mti_env);
2054         struct md_attr *ma = &info->mti_attr;
2055         struct lu_attr *la = &ma->ma_attr;
2056         int rc = 0;
2057         ENTRY;
2058
2059         ma->ma_need = MA_INODE;
2060         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2061         if (rc)
2062                 GOTO(out, rc);
2063
2064         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2065                         rc = -EACCES;
2066
2067         if (md_capable(uc, CFS_CAP_DAC_OVERRIDE))
2068                 RETURN(0);
2069         if (uc->uc_fsuid == la->la_uid) {
2070                 if ((la->la_mode & S_IWUSR) == 0)
2071                         rc = -EACCES;
2072         } else if (uc->uc_fsgid == la->la_gid) {
2073                 if ((la->la_mode & S_IWGRP) == 0)
2074                         rc = -EACCES;
2075         } else if ((la->la_mode & S_IWOTH) == 0) {
2076                         rc = -EACCES;
2077         }
2078
2079 out:
2080         RETURN(rc);
2081 }
2082
2083 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2084                          s64 ctime)
2085 {
2086         struct mdt_device *mdt = info->mti_mdt;
2087         struct mdt_object *obj = NULL;
2088         struct linkea_data ldata = { NULL };
2089         struct lu_buf *buf = &info->mti_big_buf;
2090         struct lu_name *name = &info->mti_name;
2091         struct lu_fid *pfid = &info->mti_tmp_fid1;
2092         struct link_ea_header *leh;
2093         struct link_ea_entry *lee;
2094         int reclen, count, rc = 0;
2095         ENTRY;
2096
2097         if (!fid_is_sane(fid))
2098                 GOTO(out, rc = -EINVAL);
2099
2100         if (!fid_is_namespace_visible(fid))
2101                 GOTO(out, rc = -EINVAL);
2102
2103         obj = mdt_object_find(info->mti_env, mdt, fid);
2104         if (IS_ERR(obj))
2105                 GOTO(out, rc = PTR_ERR(obj));
2106
2107         if (mdt_object_remote(obj))
2108                 GOTO(out, rc = -EREMOTE);
2109         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2110                 GOTO(out, rc = -ENOENT);
2111
2112         rc = mdt_rmfid_check_permission(info, obj);
2113         if (rc)
2114                 GOTO(out, rc);
2115
2116         /* take LinkEA */
2117         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2118         if (!buf->lb_buf)
2119                 GOTO(out, rc = -ENOMEM);
2120
2121         ldata.ld_buf = buf;
2122         rc = mdt_links_read(info, obj, &ldata);
2123         if (rc)
2124                 GOTO(out, rc);
2125
2126         leh = buf->lb_buf;
2127         lee = (struct link_ea_entry *)(leh + 1);
2128         for (count = 0; count < leh->leh_reccount; count++) {
2129                 /* remove every hardlink */
2130                 linkea_entry_unpack(lee, &reclen, name, pfid);
2131                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2132                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2133                 if (rc)
2134                         break;
2135         }
2136
2137 out:
2138         if (obj && !IS_ERR(obj))
2139                 mdt_object_put(info->mti_env, obj);
2140         if (info->mti_big_buf.lb_buf)
2141                 lu_buf_free(&info->mti_big_buf);
2142
2143         RETURN(rc);
2144 }
2145
2146 static int mdt_rmfid(struct tgt_session_info *tsi)
2147 {
2148         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2149         struct mdt_body *reqbody;
2150         struct lu_fid *fids, *rfids;
2151         int bufsize, rc;
2152         __u32 *rcs;
2153         int i, nr;
2154         ENTRY;
2155
2156         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2157         if (reqbody == NULL)
2158                 RETURN(-EPROTO);
2159         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2160                                        RCL_CLIENT);
2161         nr = bufsize / sizeof(struct lu_fid);
2162         if (nr * sizeof(struct lu_fid) != bufsize)
2163                 RETURN(-EINVAL);
2164         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2165                              RCL_SERVER, nr * sizeof(__u32));
2166         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2167                              RCL_SERVER, nr * sizeof(struct lu_fid));
2168         rc = req_capsule_server_pack(tsi->tsi_pill);
2169         if (rc)
2170                 GOTO(out, rc = err_serious(rc));
2171         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2172         if (fids == NULL)
2173                 RETURN(-EPROTO);
2174         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2175         LASSERT(rcs);
2176         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2177         LASSERT(rfids);
2178
2179         mdt_init_ucred(mti, reqbody);
2180         for (i = 0; i < nr; i++) {
2181                 rfids[i] = fids[i];
2182                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2183         }
2184         mdt_exit_ucred(mti);
2185
2186 out:
2187         RETURN(rc);
2188 }
2189
2190 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2191                          void *karg, void __user *uarg);
2192
2193 static int mdt_set_info(struct tgt_session_info *tsi)
2194 {
2195         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2196         char                    *key;
2197         void                    *val;
2198         int                      keylen, vallen, rc = 0;
2199
2200         ENTRY;
2201
2202         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2203         if (key == NULL) {
2204                 DEBUG_REQ(D_HA, req, "no set_info key");
2205                 RETURN(err_serious(-EFAULT));
2206         }
2207
2208         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2209                                       RCL_CLIENT);
2210
2211         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2212         if (val == NULL) {
2213                 DEBUG_REQ(D_HA, req, "no set_info val");
2214                 RETURN(err_serious(-EFAULT));
2215         }
2216
2217         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2218                                       RCL_CLIENT);
2219
2220         /* Swab any part of val you need to here */
2221         if (KEY_IS(KEY_READ_ONLY)) {
2222                 spin_lock(&req->rq_export->exp_lock);
2223                 if (*(__u32 *)val)
2224                         *exp_connect_flags_ptr(req->rq_export) |=
2225                                 OBD_CONNECT_RDONLY;
2226                 else
2227                         *exp_connect_flags_ptr(req->rq_export) &=
2228                                 ~OBD_CONNECT_RDONLY;
2229                 spin_unlock(&req->rq_export->exp_lock);
2230         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2231                 struct changelog_setinfo *cs = val;
2232
2233                 if (vallen != sizeof(*cs)) {
2234                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2235                                tgt_name(tsi->tsi_tgt), vallen);
2236                         RETURN(-EINVAL);
2237                 }
2238                 if (ptlrpc_req_need_swab(req)) {
2239                         __swab64s(&cs->cs_recno);
2240                         __swab32s(&cs->cs_id);
2241                 }
2242
2243                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2244                                    vallen, val, NULL);
2245         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2246                 if (vallen > 0)
2247                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2248         } else {
2249                 RETURN(-EINVAL);
2250         }
2251         RETURN(rc);
2252 }
2253
2254 static int mdt_readpage(struct tgt_session_info *tsi)
2255 {
2256         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2257         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2258         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2259         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2260         struct mdt_body         *repbody;
2261         int                      rc;
2262         int                      i;
2263
2264         ENTRY;
2265
2266         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2267                 RETURN(err_serious(-ENOMEM));
2268
2269         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2270         if (repbody == NULL || reqbody == NULL)
2271                 RETURN(err_serious(-EFAULT));
2272
2273         /*
2274          * prepare @rdpg before calling lower layers and transfer itself. Here
2275          * reqbody->size contains offset of where to start to read and
2276          * reqbody->nlink contains number bytes to read.
2277          */
2278         rdpg->rp_hash = reqbody->mbo_size;
2279         if (rdpg->rp_hash != reqbody->mbo_size) {
2280                 CERROR("Invalid hash: %#llx != %#llx\n",
2281                        rdpg->rp_hash, reqbody->mbo_size);
2282                 RETURN(-EFAULT);
2283         }
2284
2285         rdpg->rp_attrs = reqbody->mbo_mode;
2286         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2287                 rdpg->rp_attrs |= LUDA_64BITHASH;
2288         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2289                                 exp_max_brw_size(tsi->tsi_exp));
2290         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2291                           PAGE_SHIFT;
2292         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2293         if (rdpg->rp_pages == NULL)
2294                 RETURN(-ENOMEM);
2295
2296         for (i = 0; i < rdpg->rp_npages; ++i) {
2297                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2298                 if (rdpg->rp_pages[i] == NULL)
2299                         GOTO(free_rdpg, rc = -ENOMEM);
2300         }
2301
2302         /* call lower layers to fill allocated pages with directory data */
2303         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2304         if (rc < 0)
2305                 GOTO(free_rdpg, rc);
2306
2307         /* send pages to client */
2308         rc = tgt_sendpage(tsi, rdpg, rc);
2309
2310         EXIT;
2311 free_rdpg:
2312
2313         for (i = 0; i < rdpg->rp_npages; i++)
2314                 if (rdpg->rp_pages[i] != NULL)
2315                         __free_page(rdpg->rp_pages[i]);
2316         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2317
2318         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2319                 RETURN(0);
2320
2321         return rc;
2322 }
2323
2324 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2325 {
2326         struct lu_ucred *uc = mdt_ucred_check(info);
2327         struct lu_attr *attr = &info->mti_attr.ma_attr;
2328
2329         if (uc == NULL)
2330                 return -EINVAL;
2331
2332         if (op != REINT_SETATTR) {
2333                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2334                         attr->la_uid = uc->uc_fsuid;
2335                 /* for S_ISGID, inherit gid from his parent, such work will be
2336                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2337                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2338                         attr->la_gid = uc->uc_fsgid;
2339         }
2340
2341         return 0;
2342 }
2343
2344 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2345 {
2346         return op == REINT_OPEN &&
2347              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2348 }
2349
2350 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2351 {
2352         struct req_capsule *pill = info->mti_pill;
2353
2354         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2355                                   RCL_SERVER) &&
2356             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2357                                   RCL_CLIENT)) {
2358                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2359                                          RCL_CLIENT) != 0) {
2360                         /* pre-set size in server part with max size */
2361                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2362                                              RCL_SERVER,
2363                                              info->mti_mdt->mdt_max_ea_size);
2364                 } else {
2365                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2366                                              RCL_SERVER, 0);
2367                 }
2368         }
2369
2370 }
2371
2372 static int mdt_reint_internal(struct mdt_thread_info *info,
2373                               struct mdt_lock_handle *lhc,
2374                               __u32 op)
2375 {
2376         struct req_capsule      *pill = info->mti_pill;
2377         struct mdt_body         *repbody;
2378         int                      rc = 0, rc2;
2379
2380         ENTRY;
2381
2382         rc = mdt_reint_unpack(info, op);
2383         if (rc != 0) {
2384                 CERROR("Can't unpack reint, rc %d\n", rc);
2385                 RETURN(err_serious(rc));
2386         }
2387
2388
2389         /* check if the file system is set to readonly. O_RDONLY open
2390          * is still allowed even the file system is set to readonly mode */
2391         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2392                 RETURN(err_serious(-EROFS));
2393
2394         /* for replay (no_create) lmm is not needed, client has it already */
2395         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2396                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2397                                      DEF_REP_MD_SIZE);
2398
2399         /* llog cookies are always 0, the field is kept for compatibility */
2400         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2401                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2402
2403         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2404          * by default. If the target object has more ACL entries, then
2405          * enlarge the buffer when necessary. */
2406         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2407                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2408                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2409
2410         mdt_preset_secctx_size(info);
2411
2412         rc = req_capsule_server_pack(pill);
2413         if (rc != 0) {
2414                 CERROR("Can't pack response, rc %d\n", rc);
2415                 RETURN(err_serious(rc));
2416         }
2417
2418         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2419                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2420                 LASSERT(repbody);
2421                 repbody->mbo_eadatasize = 0;
2422                 repbody->mbo_aclsize = 0;
2423         }
2424
2425         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2426
2427         /* for replay no cookkie / lmm need, because client have this already */
2428         if (info->mti_spec.no_create)
2429                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2430                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2431
2432         rc = mdt_init_ucred_reint(info);
2433         if (rc)
2434                 GOTO(out_shrink, rc);
2435
2436         rc = mdt_fix_attr_ucred(info, op);
2437         if (rc != 0)
2438                 GOTO(out_ucred, rc = err_serious(rc));
2439
2440         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2441         if (rc < 0) {
2442                 GOTO(out_ucred, rc);
2443         } else if (rc == 1) {
2444                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2445                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2446                 GOTO(out_ucred, rc);
2447         }
2448         rc = mdt_reint_rec(info, lhc);
2449         EXIT;
2450 out_ucred:
2451         mdt_exit_ucred(info);
2452 out_shrink:
2453         mdt_client_compatibility(info);
2454
2455         rc2 = mdt_fix_reply(info);
2456         if (rc == 0)
2457                 rc = rc2;
2458
2459         /*
2460          * Data-on-MDT optimization - read data along with OPEN and return it
2461          * in reply. Do that only if we have both DOM and LAYOUT locks.
2462          */
2463         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2464             info->mti_attr.ma_lmm != NULL &&
2465             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2466                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2467                                           &lhc->mlh_reg_lh);
2468         }
2469
2470         return rc;
2471 }
2472
2473 static long mdt_reint_opcode(struct ptlrpc_request *req,
2474                              const struct req_format **fmt)
2475 {
2476         struct mdt_device       *mdt;
2477         struct mdt_rec_reint    *rec;
2478         long                     opc;
2479
2480         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2481         if (rec != NULL) {
2482                 opc = rec->rr_opcode;
2483                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2484                 if (opc < REINT_MAX && fmt[opc] != NULL)
2485                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2486                 else {
2487                         mdt = mdt_exp2dev(req->rq_export);
2488                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2489                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2490                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2491                         opc = err_serious(-EFAULT);
2492                 }
2493         } else {
2494                 opc = err_serious(-EFAULT);
2495         }
2496         return opc;
2497 }
2498
2499 static int mdt_reint(struct tgt_session_info *tsi)
2500 {
2501         long opc;
2502         int  rc;
2503         static const struct req_format *reint_fmts[REINT_MAX] = {
2504                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2505                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2506                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2507                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2508                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2509                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2510                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2511                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2512                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2513                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2514         };
2515
2516         ENTRY;
2517
2518         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2519         if (opc >= 0) {
2520                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2521                 /*
2522                  * No lock possible here from client to pass it to reint code
2523                  * path.
2524                  */
2525                 rc = mdt_reint_internal(info, NULL, opc);
2526                 mdt_thread_info_fini(info);
2527         } else {
2528                 rc = opc;
2529         }
2530
2531         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2532         RETURN(rc);
2533 }
2534
2535 /* this should sync the whole device */
2536 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2537 {
2538         struct dt_device *dt = mdt->mdt_bottom;
2539         int rc;
2540         ENTRY;
2541
2542         rc = dt->dd_ops->dt_sync(env, dt);
2543         RETURN(rc);
2544 }
2545
2546 /* this should sync this object */
2547 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2548                            struct mdt_object *mo)
2549 {
2550         int rc = 0;
2551
2552         ENTRY;
2553
2554         if (!mdt_object_exists(mo)) {
2555                 CWARN("%s: non existing object "DFID": rc = %d\n",
2556                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2557                       -ESTALE);
2558                 RETURN(-ESTALE);
2559         }
2560
2561         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2562                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2563                 dt_obj_version_t version;
2564
2565                 version = dt_version_get(env, mdt_obj2dt(mo));
2566                 if (version > tgt->lut_obd->obd_last_committed)
2567                         rc = mo_object_sync(env, mdt_object_child(mo));
2568         } else {
2569                 rc = mo_object_sync(env, mdt_object_child(mo));
2570         }
2571
2572         RETURN(rc);
2573 }
2574
2575 static int mdt_sync(struct tgt_session_info *tsi)
2576 {
2577         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2578         struct req_capsule      *pill = tsi->tsi_pill;
2579         struct mdt_body         *body;
2580         int                      rc;
2581
2582         ENTRY;
2583
2584         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2585                 RETURN(err_serious(-ENOMEM));
2586
2587         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2588                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2589         } else {
2590                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2591
2592                 /* sync an object */
2593                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2594                                      info->mti_object);
2595                 if (rc == 0) {
2596                         const struct lu_fid *fid;
2597                         struct lu_attr *la = &info->mti_attr.ma_attr;
2598
2599                         info->mti_attr.ma_need = MA_INODE;
2600                         info->mti_attr.ma_valid = 0;
2601                         rc = mdt_attr_get_complex(info, info->mti_object,
2602                                                   &info->mti_attr);
2603                         if (rc == 0) {
2604                                 body = req_capsule_server_get(pill,
2605                                                               &RMF_MDT_BODY);
2606                                 fid = mdt_object_fid(info->mti_object);
2607                                 mdt_pack_attr2body(info, body, la, fid);
2608                         }
2609                 }
2610                 mdt_thread_info_fini(info);
2611         }
2612         if (rc == 0)
2613                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2614
2615         RETURN(rc);
2616 }
2617
2618 static int mdt_data_sync(struct tgt_session_info *tsi)
2619 {
2620         struct mdt_thread_info *info;
2621         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2622         struct ost_body *body = tsi->tsi_ost_body;
2623         struct ost_body *repbody;
2624         struct mdt_object *mo = NULL;
2625         struct md_attr *ma;
2626         int rc = 0;
2627
2628         ENTRY;
2629
2630         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2631
2632         /* if no fid is specified then do nothing,
2633          * device sync is done via MDS_SYNC */
2634         if (fid_is_zero(&tsi->tsi_fid))
2635                 RETURN(0);
2636
2637         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2638         if (IS_ERR(mo))
2639                 RETURN(PTR_ERR(mo));
2640
2641         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2642         if (rc)
2643                 GOTO(put, rc);
2644
2645         repbody->oa.o_oi = body->oa.o_oi;
2646         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2647
2648         info = tsi2mdt_info(tsi);
2649         ma = &info->mti_attr;
2650         ma->ma_need = MA_INODE;
2651         ma->ma_valid = 0;
2652         rc = mdt_attr_get_complex(info, mo, ma);
2653         if (rc == 0)
2654                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2655         else
2656                 rc = 0;
2657         mdt_thread_info_fini(info);
2658
2659         EXIT;
2660 put:
2661         if (mo != NULL)
2662                 mdt_object_put(tsi->tsi_env, mo);
2663         return rc;
2664 }
2665
2666 /*
2667  * Handle quota control requests to consult current usage/limit, but also
2668  * to configure quota enforcement
2669  */
2670 static int mdt_quotactl(struct tgt_session_info *tsi)
2671 {
2672         struct obd_export       *exp  = tsi->tsi_exp;
2673         struct req_capsule      *pill = tsi->tsi_pill;
2674         struct obd_quotactl     *oqctl, *repoqc;
2675         int                      id, rc;
2676         struct mdt_device       *mdt = mdt_exp2dev(exp);
2677         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2678         struct lu_nodemap       *nodemap;
2679         ENTRY;
2680
2681         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2682         if (oqctl == NULL)
2683                 RETURN(err_serious(-EPROTO));
2684
2685         rc = req_capsule_server_pack(pill);
2686         if (rc)
2687                 RETURN(err_serious(rc));
2688
2689         nodemap = nodemap_get_from_exp(exp);
2690         if (IS_ERR(nodemap))
2691                 RETURN(PTR_ERR(nodemap));
2692
2693         switch (oqctl->qc_cmd) {
2694                 /* master quotactl */
2695         case Q_SETINFO:
2696         case Q_SETQUOTA:
2697         case LUSTRE_Q_SETDEFAULT:
2698                 if (!nodemap_can_setquota(nodemap))
2699                         GOTO(out_nodemap, rc = -EPERM);
2700         case Q_GETINFO:
2701         case Q_GETQUOTA:
2702         case LUSTRE_Q_GETDEFAULT:
2703                 if (qmt == NULL)
2704                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2705                 /* slave quotactl */
2706         case Q_GETOINFO:
2707         case Q_GETOQUOTA:
2708                 break;
2709         default:
2710                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2711                 GOTO(out_nodemap, rc = -EFAULT);
2712         }
2713
2714         id = oqctl->qc_id;
2715         switch (oqctl->qc_type) {
2716         case USRQUOTA:
2717                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2718                                     NODEMAP_CLIENT_TO_FS, id);
2719                 break;
2720         case GRPQUOTA:
2721                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2722                                     NODEMAP_CLIENT_TO_FS, id);
2723                 break;
2724         case PRJQUOTA:
2725                 /* todo: check/map project id */
2726                 id = oqctl->qc_id;
2727                 break;
2728         default:
2729                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2730         }
2731         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2732         if (repoqc == NULL)
2733                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2734
2735         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2736                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2737
2738         if (oqctl->qc_id != id)
2739                 swap(oqctl->qc_id, id);
2740
2741         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2742                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2743                         RETURN(-EINPROGRESS);
2744         }
2745
2746         switch (oqctl->qc_cmd) {
2747
2748         case Q_GETINFO:
2749         case Q_SETINFO:
2750         case Q_SETQUOTA:
2751         case Q_GETQUOTA:
2752         case LUSTRE_Q_SETDEFAULT:
2753         case LUSTRE_Q_GETDEFAULT:
2754                 /* forward quotactl request to QMT */
2755                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2756                 break;
2757
2758         case Q_GETOINFO:
2759         case Q_GETOQUOTA:
2760                 /* slave quotactl */
2761                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2762                                    oqctl);
2763                 break;
2764
2765         default:
2766                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2767                 GOTO(out_nodemap, rc = -EFAULT);
2768         }
2769
2770         if (oqctl->qc_id != id)
2771                 swap(oqctl->qc_id, id);
2772
2773         *repoqc = *oqctl;
2774
2775         EXIT;
2776
2777 out_nodemap:
2778         nodemap_putref(nodemap);
2779
2780         return rc;
2781 }
2782
2783 /** clone llog ctxt from child (mdd)
2784  * This allows remote llog (replicator) access.
2785  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2786  * context was originally set up, or we can handle them directly.
2787  * I choose the latter, but that means I need any llog
2788  * contexts set up by child to be accessable by the mdt.  So we clone the
2789  * context into our context list here.
2790  */
2791 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2792                                int idx)
2793 {
2794         struct md_device  *next = mdt->mdt_child;
2795         struct llog_ctxt *ctxt;
2796         int rc;
2797
2798         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2799                 return 0;
2800
2801         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2802         if (rc || ctxt == NULL) {
2803                 return 0;
2804         }
2805
2806         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2807         if (rc)
2808                 CERROR("Can't set mdt ctxt %d\n", rc);
2809
2810         return rc;
2811 }
2812
2813 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2814                                  struct mdt_device *mdt, int idx)
2815 {
2816         struct llog_ctxt *ctxt;
2817
2818         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2819         if (ctxt == NULL)
2820                 return 0;
2821         /* Put once for the get we just did, and once for the clone */
2822         llog_ctxt_put(ctxt);
2823         llog_ctxt_put(ctxt);
2824         return 0;
2825 }
2826
2827 /*
2828  * sec context handlers
2829  */
2830 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2831 {
2832         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2833
2834         return 0;
2835 }
2836
2837 /*
2838  * quota request handlers
2839  */
2840 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2841 {
2842         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2843         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2844         int                      rc;
2845         ENTRY;
2846
2847         if (qmt == NULL)
2848                 RETURN(err_serious(-EOPNOTSUPP));
2849
2850         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2851         RETURN(rc);
2852 }
2853
2854 struct mdt_object *mdt_object_new(const struct lu_env *env,
2855                                   struct mdt_device *d,
2856                                   const struct lu_fid *f)
2857 {
2858         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2859         struct lu_object *o;
2860         struct mdt_object *m;
2861         ENTRY;
2862
2863         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2864         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2865         if (unlikely(IS_ERR(o)))
2866                 m = (struct mdt_object *)o;
2867         else
2868                 m = mdt_obj(o);
2869         RETURN(m);
2870 }
2871
2872 struct mdt_object *mdt_object_find(const struct lu_env *env,
2873                                    struct mdt_device *d,
2874                                    const struct lu_fid *f)
2875 {
2876         struct lu_object *o;
2877         struct mdt_object *m;
2878         ENTRY;
2879
2880         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2881         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2882         if (unlikely(IS_ERR(o)))
2883                 m = (struct mdt_object *)o;
2884         else
2885                 m = mdt_obj(o);
2886
2887         RETURN(m);
2888 }
2889
2890 /**
2891  * Asyncronous commit for mdt device.
2892  *
2893  * Pass asynchonous commit call down the MDS stack.
2894  *
2895  * \param env environment
2896  * \param mdt the mdt device
2897  */
2898 static void mdt_device_commit_async(const struct lu_env *env,
2899                                     struct mdt_device *mdt)
2900 {
2901         struct dt_device *dt = mdt->mdt_bottom;
2902         int rc;
2903         ENTRY;
2904
2905         rc = dt->dd_ops->dt_commit_async(env, dt);
2906         if (unlikely(rc != 0))
2907                 CWARN("%s: async commit start failed: rc = %d\n",
2908                       mdt_obd_name(mdt), rc);
2909         atomic_inc(&mdt->mdt_async_commit_count);
2910         EXIT;
2911 }
2912
2913 /**
2914  * Mark the lock as "synchonous".
2915  *
2916  * Mark the lock to deffer transaction commit to the unlock time.
2917  *
2918  * \param lock the lock to mark as "synchonous"
2919  *
2920  * \see mdt_is_lock_sync
2921  * \see mdt_save_lock
2922  */
2923 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2924 {
2925         lock->l_ast_data = (void*)1;
2926 }
2927
2928 /**
2929  * Check whehter the lock "synchonous" or not.
2930  *
2931  * \param lock the lock to check
2932  * \retval 1 the lock is "synchonous"
2933  * \retval 0 the lock isn't "synchronous"
2934  *
2935  * \see mdt_set_lock_sync
2936  * \see mdt_save_lock
2937  */
2938 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2939 {
2940         return lock->l_ast_data != NULL;
2941 }
2942
2943 /**
2944  * Blocking AST for mdt locks.
2945  *
2946  * Starts transaction commit if in case of COS lock conflict or
2947  * deffers such a commit to the mdt_save_lock.
2948  *
2949  * \param lock the lock which blocks a request or cancelling lock
2950  * \param desc unused
2951  * \param data unused
2952  * \param flag indicates whether this cancelling or blocking callback
2953  * \retval 0
2954  * \see ldlm_blocking_ast_nocheck
2955  */
2956 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2957                      void *data, int flag)
2958 {
2959         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2960         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2961         struct ldlm_cb_set_arg *arg = data;
2962         bool commit_async = false;
2963         int rc;
2964         ENTRY;
2965
2966         if (flag == LDLM_CB_CANCELING)
2967                 RETURN(0);
2968
2969         lock_res_and_lock(lock);
2970         if (lock->l_blocking_ast != mdt_blocking_ast) {
2971                 unlock_res_and_lock(lock);
2972                 RETURN(0);
2973         }
2974
2975         /* A blocking ast may be sent from ldlm_lock_decref_internal
2976          * when the last reference to a local lock was released and
2977          * during blocking event from ldlm_work_bl_ast_lock().
2978          * The 'data' parameter is l_ast_data in the first case and
2979          * callback arguments in the second one. Distinguish them by that.
2980          */
2981         if (!data || data == lock->l_ast_data || !arg->bl_desc)
2982                 goto skip_cos_checks;
2983
2984         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
2985                 if (mdt_cos_is_enabled(mdt)) {
2986                         if (!arg->bl_desc->bl_same_client)
2987                                 mdt_set_lock_sync(lock);
2988                 } else if (mdt_slc_is_enabled(mdt) &&
2989                            arg->bl_desc->bl_cos_incompat) {
2990                         mdt_set_lock_sync(lock);
2991                         /*
2992                          * we may do extra commit here, but there is a small
2993                          * window to miss a commit: lock was unlocked (saved),
2994                          * then a conflict lock queued and we come here, but
2995                          * REP-ACK not received, so lock was not converted to
2996                          * COS mode yet.
2997                          * Fortunately this window is quite small, so the
2998                          * extra commit should be rare (not to say distributed
2999                          * operation is rare too).
3000                          */
3001                         commit_async = true;
3002                 }
3003         } else if (lock->l_req_mode == LCK_COS) {
3004                 commit_async = true;
3005         }
3006
3007 skip_cos_checks:
3008         rc = ldlm_blocking_ast_nocheck(lock);
3009
3010         if (commit_async) {
3011                 struct lu_env env;
3012
3013                 rc = lu_env_init(&env, LCT_LOCAL);
3014                 if (unlikely(rc != 0))
3015                         CWARN("%s: lu_env initialization failed, cannot "
3016                               "start asynchronous commit: rc = %d\n",
3017                               obd->obd_name, rc);
3018                 else
3019                         mdt_device_commit_async(&env, mdt);
3020                 lu_env_fini(&env);
3021         }
3022         RETURN(rc);
3023 }
3024
3025 /*
3026  * Blocking AST for cross-MDT lock
3027  *
3028  * Discard lock from uncommitted_slc_locks and cancel it.
3029  *
3030  * \param lock  the lock which blocks a request or cancelling lock
3031  * \param desc  unused
3032  * \param data  unused
3033  * \param flag  indicates whether this cancelling or blocking callback
3034  * \retval      0 on success
3035  * \retval      negative number on error
3036  */
3037 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3038                             void *data, int flag)
3039 {
3040         int rc = 0;
3041         ENTRY;
3042
3043         switch (flag) {
3044         case LDLM_CB_BLOCKING: {
3045                 struct lustre_handle lockh;
3046
3047                 ldlm_lock2handle(lock, &lockh);
3048                 rc = ldlm_cli_cancel(&lockh,
3049                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3050                 if (rc < 0) {
3051                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3052                         RETURN(rc);
3053                 }
3054                 break;
3055         }
3056         case LDLM_CB_CANCELING: {
3057                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3058                 struct mdt_device *mdt =
3059                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3060
3061                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3062
3063                 /* discard slc lock here so that it can be cleaned anytime,
3064                  * especially for cleanup_resource() */
3065                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3066
3067                 /* once we cache lock, l_ast_data is set to mdt_object */
3068                 if (lock->l_ast_data != NULL) {
3069                         struct mdt_object *mo = lock->l_ast_data;
3070                         struct lu_env env;
3071
3072                         rc = lu_env_init(&env, LCT_MD_THREAD);
3073                         if (unlikely(rc != 0)) {
3074                                 CWARN("%s: lu_env initialization failed, object"
3075                                       "%p "DFID" is leaked!\n",
3076                                       obd->obd_name, mo,
3077                                       PFID(mdt_object_fid(mo)));
3078                                 RETURN(rc);
3079                         }
3080
3081                         if (lock->l_policy_data.l_inodebits.bits &
3082                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3083                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3084                                 mo->mot_cache_attr = 0;
3085                         }
3086                         mdt_object_put(&env, mo);
3087                         lu_env_fini(&env);
3088                 }
3089                 break;
3090         }
3091         default:
3092                 LBUG();
3093         }
3094
3095         RETURN(rc);
3096 }
3097
3098 int mdt_check_resent_lock(struct mdt_thread_info *info,
3099                           struct mdt_object *mo,
3100                           struct mdt_lock_handle *lhc)
3101 {
3102         /* the lock might already be gotten in ldlm_handle_enqueue() */
3103         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3104                 struct ptlrpc_request *req = mdt_info_req(info);
3105                 struct ldlm_lock      *lock;
3106
3107                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3108                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3109                 if (lock == NULL) {
3110                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3111                          * a resend case, however, it could be already destroyed
3112                          * due to client eviction or a raced cancel RPC. */
3113                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3114                                           lhc->mlh_reg_lh.cookie);
3115                         RETURN(-ESTALE);
3116                 }
3117
3118                 if (!fid_res_name_eq(mdt_object_fid(mo),
3119                                      &lock->l_resource->lr_name)) {
3120                         CWARN("%s: Although resent, but still not "
3121                               "get child lock:"DFID"\n",
3122                               info->mti_exp->exp_obd->obd_name,
3123                               PFID(mdt_object_fid(mo)));
3124                         LDLM_LOCK_PUT(lock);
3125                         RETURN(-EPROTO);
3126                 }
3127                 LDLM_LOCK_PUT(lock);
3128                 return 0;
3129         }
3130         return 1;
3131 }
3132
3133 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3134 {
3135         mdt_object_get(NULL, lock->l_ast_data);
3136 }
3137
3138 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3139                                struct mdt_object *o, const struct lu_fid *fid,
3140                                struct lustre_handle *lh, enum ldlm_mode mode,
3141                                __u64 *ibits, __u64 trybits, bool cache)
3142 {
3143         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3144         union ldlm_policy_data *policy = &mti->mti_policy;
3145         struct ldlm_res_id *res_id = &mti->mti_res_id;
3146         int rc = 0;
3147         ENTRY;
3148
3149         LASSERT(mdt_object_remote(o));
3150
3151         fid_build_reg_res_name(fid, res_id);
3152
3153         memset(einfo, 0, sizeof(*einfo));
3154         einfo->ei_type = LDLM_IBITS;
3155         einfo->ei_mode = mode;
3156         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3157         einfo->ei_cb_cp = ldlm_completion_ast;
3158         einfo->ei_enq_slave = 0;
3159         einfo->ei_res_id = res_id;
3160
3161         if (cache) {
3162                 /*
3163                  * if we cache lock, couple lock with mdt_object, so that object
3164                  * can be easily found in lock ASTs.
3165                  */
3166                 einfo->ei_cbdata = o;
3167                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3168         }
3169
3170         memset(policy, 0, sizeof(*policy));
3171         policy->l_inodebits.bits = *ibits;
3172         policy->l_inodebits.try_bits = trybits;
3173
3174         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3175                             policy);
3176
3177         /* Return successfully acquired bits to a caller */
3178         if (rc == 0) {
3179                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3180
3181                 LASSERT(lock);
3182                 *ibits = lock->l_policy_data.l_inodebits.bits;
3183                 LDLM_LOCK_PUT(lock);
3184         }
3185         RETURN(rc);
3186 }
3187
3188 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3189                            const struct lu_fid *fid, struct lustre_handle *lh,
3190                            enum ldlm_mode mode, __u64 ibits, bool cache)
3191 {
3192         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3193                                           cache);
3194 }
3195
3196 static int mdt_object_local_lock(struct mdt_thread_info *info,
3197                                  struct mdt_object *o,
3198                                  struct mdt_lock_handle *lh, __u64 *ibits,
3199                                  __u64 trybits, bool cos_incompat)
3200 {
3201         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3202         union ldlm_policy_data *policy = &info->mti_policy;
3203         struct ldlm_res_id *res_id = &info->mti_res_id;
3204         __u64 dlmflags = 0, *cookie = NULL;
3205         int rc;
3206         ENTRY;
3207
3208         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3209         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3210         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3211         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3212
3213         if (cos_incompat) {
3214                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3215                         lh->mlh_reg_mode == LCK_EX);
3216                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3217         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3218                 dlmflags |= LDLM_FL_COS_ENABLED;
3219         }
3220
3221         /* Only enqueue LOOKUP lock for remote object */
3222         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3223
3224         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3225         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3226             *ibits == MDS_INODELOCK_OPEN)
3227                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3228
3229         if (lh->mlh_type == MDT_PDO_LOCK) {
3230                 /* check for exists after object is locked */
3231                 if (mdt_object_exists(o) == 0) {
3232                         /* Non-existent object shouldn't have PDO lock */
3233                         RETURN(-ESTALE);
3234                 } else {
3235                         /* Non-dir object shouldn't have PDO lock */
3236                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3237                                 RETURN(-ENOTDIR);
3238                 }
3239         }
3240
3241         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3242         dlmflags |= LDLM_FL_ATOMIC_CB;
3243
3244         if (info->mti_exp)
3245                 cookie = &info->mti_exp->exp_handle.h_cookie;
3246
3247         /*
3248          * Take PDO lock on whole directory and build correct @res_id for lock
3249          * on part of directory.
3250          */
3251         if (lh->mlh_pdo_hash != 0) {
3252                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3253                 mdt_lock_pdo_mode(info, o, lh);
3254                 if (lh->mlh_pdo_mode != LCK_NL) {
3255                         /*
3256                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3257                          * is never going to be sent to client and we do not
3258                          * want it slowed down due to possible cancels.
3259                          */
3260                         policy->l_inodebits.bits =
3261                                 *ibits & MDS_INODELOCK_UPDATE;
3262                         policy->l_inodebits.try_bits =
3263                                 trybits & MDS_INODELOCK_UPDATE;
3264                         /* at least one of them should be set */
3265                         LASSERT(policy->l_inodebits.bits |
3266                                 policy->l_inodebits.try_bits);
3267                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3268                                           lh->mlh_pdo_mode, policy, res_id,
3269                                           dlmflags, cookie);
3270                         if (unlikely(rc != 0))
3271                                 GOTO(out_unlock, rc);
3272                 }
3273
3274                 /*
3275                  * Finish res_id initializing by name hash marking part of
3276                  * directory which is taking modification.
3277                  */
3278                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3279         }
3280
3281         policy->l_inodebits.bits = *ibits;
3282         policy->l_inodebits.try_bits = trybits;
3283
3284         /*
3285          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3286          * going to be sent to client. If it is - mdt_intent_policy() path will
3287          * fix it up and turn FL_LOCAL flag off.
3288          */
3289         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3290                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3291                           cookie);
3292 out_unlock:
3293         if (rc != 0)
3294                 mdt_object_unlock(info, o, lh, 1);
3295         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3296                    lh->mlh_pdo_hash != 0 &&
3297                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3298                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3299
3300         /* Return successfully acquired bits to a caller */
3301         if (rc == 0) {
3302                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3303
3304                 LASSERT(lock);
3305                 *ibits = lock->l_policy_data.l_inodebits.bits;
3306                 LDLM_LOCK_PUT(lock);
3307         }
3308         RETURN(rc);
3309 }
3310
3311 static int
3312 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3313                          struct mdt_lock_handle *lh, __u64 *ibits,
3314                          __u64 trybits, bool cos_incompat)
3315 {
3316         struct mdt_lock_handle *local_lh = NULL;
3317         int rc;
3318         ENTRY;
3319
3320         if (!mdt_object_remote(o)) {
3321                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3322                                            cos_incompat);
3323                 RETURN(rc);
3324         }
3325
3326         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3327         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3328                     MDS_INODELOCK_XATTR);
3329
3330         /* Only enqueue LOOKUP lock for remote object */
3331         if (*ibits & MDS_INODELOCK_LOOKUP) {
3332                 __u64 local = MDS_INODELOCK_LOOKUP;
3333
3334                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3335                                            cos_incompat);
3336                 if (rc != ELDLM_OK)
3337                         RETURN(rc);
3338
3339                 local_lh = lh;
3340         }
3341
3342         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3343                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3344                  * enqueue RPC can only request 1 lock, to avoid extra
3345                  * RPC, so it will instead enqueue EX lock for remote
3346                  * object anyway XXX*/
3347                 if (lh->mlh_type == MDT_PDO_LOCK &&
3348                     lh->mlh_pdo_hash != 0) {
3349                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
3350                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
3351                                PFID(mdt_object_fid(o)));
3352                         lh->mlh_pdo_hash = 0;
3353                         lh->mlh_rreg_mode = LCK_EX;
3354                         lh->mlh_type = MDT_REG_LOCK;
3355                 }
3356
3357                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3358                                                 &lh->mlh_rreg_lh,
3359                                                 lh->mlh_rreg_mode,
3360                                                 ibits, trybits, false);
3361                 if (rc != ELDLM_OK) {
3362                         if (local_lh != NULL)
3363                                 mdt_object_unlock(info, o, local_lh, rc);
3364                         RETURN(rc);
3365                 }
3366         }
3367
3368         /* other components like LFSCK can use lockless access
3369          * and populate cache, so we better invalidate it */
3370         mo_invalidate(info->mti_env, mdt_object_child(o));
3371
3372         RETURN(0);
3373 }
3374
3375 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3376                     struct mdt_lock_handle *lh, __u64 ibits)
3377 {
3378         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3379 }
3380
3381 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3382                           struct mdt_lock_handle *lh, __u64 ibits,
3383                           bool cos_incompat)
3384 {
3385         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3386         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3387                                         cos_incompat);
3388 }
3389
3390 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3391                         struct mdt_lock_handle *lh, __u64 *ibits,
3392                         __u64 trybits, bool cos_incompat)
3393 {
3394         bool trylock_only = *ibits == 0;
3395         int rc;
3396
3397         LASSERT(!(*ibits & trybits));
3398         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3399                                       cos_incompat);
3400         if (rc && trylock_only) { /* clear error for try ibits lock only */
3401                 LASSERT(*ibits == 0);
3402                 rc = 0;
3403         }
3404         return rc;
3405 }
3406
3407 /**
3408  * Save a lock within request object.
3409  *
3410  * Keep the lock referenced until whether client ACK or transaction
3411  * commit happens or release the lock immediately depending on input
3412  * parameters. If COS is ON, a write lock is converted to COS lock
3413  * before saving.
3414  *
3415  * \param info thead info object
3416  * \param h lock handle
3417  * \param mode lock mode
3418  * \param decref force immediate lock releasing
3419  */
3420 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3421                    enum ldlm_mode mode, int decref)
3422 {
3423         ENTRY;
3424
3425         if (lustre_handle_is_used(h)) {
3426                 if (decref || !info->mti_has_trans ||
3427                     !(mode & (LCK_PW | LCK_EX))) {
3428                         mdt_fid_unlock(h, mode);
3429                 } else {
3430                         struct mdt_device *mdt = info->mti_mdt;
3431                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3432                         struct ptlrpc_request *req = mdt_info_req(info);
3433                         bool cos = mdt_cos_is_enabled(mdt);
3434                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3435
3436                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3437                                  h->cookie);
3438
3439                         /* there is no request if mdt_object_unlock() is called
3440                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3441                         if (likely(req != NULL)) {
3442                                 LDLM_DEBUG(lock, "save lock request %p reply "
3443                                         "state %p transno %lld\n", req,
3444                                         req->rq_reply_state, req->rq_transno);
3445                                 if (cos) {
3446                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3447                                         mode = LCK_COS;
3448                                 }
3449                                 if (req->rq_export->exp_disconnected)
3450                                         mdt_fid_unlock(h, mode);
3451                                 else
3452                                         ptlrpc_save_lock(req, h, mode, cos,
3453                                                          convert_lock);
3454                         } else {
3455                                 mdt_fid_unlock(h, mode);
3456                         }
3457                         if (mdt_is_lock_sync(lock)) {
3458                                 CDEBUG(D_HA, "found sync-lock,"
3459                                        " async commit started\n");
3460                                 mdt_device_commit_async(info->mti_env,
3461                                                         mdt);
3462                         }
3463                         LDLM_LOCK_PUT(lock);
3464                 }
3465                 h->cookie = 0ull;
3466         }
3467
3468         EXIT;
3469 }
3470
3471 /**
3472  * Save cross-MDT lock in uncommitted_slc_locks
3473  *
3474  * Keep the lock referenced until transaction commit happens or release the lock
3475  * immediately depending on input parameters.
3476  *
3477  * \param info thead info object
3478  * \param h lock handle
3479  * \param mode lock mode
3480  * \param decref force immediate lock releasing
3481  */
3482 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3483                                  struct mdt_object *o, struct lustre_handle *h,
3484                                  enum ldlm_mode mode, int decref)
3485 {
3486         ENTRY;
3487
3488         if (lustre_handle_is_used(h)) {
3489                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3490
3491                 if (o != NULL &&
3492                     (lock->l_policy_data.l_inodebits.bits &
3493                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3494                         mo_invalidate(info->mti_env, mdt_object_child(o));
3495
3496                 if (decref || !info->mti_has_trans ||
3497                     !(mode & (LCK_PW | LCK_EX))) {
3498                         ldlm_lock_decref_and_cancel(h, mode);
3499                         LDLM_LOCK_PUT(lock);
3500                 } else {
3501                         struct ptlrpc_request *req = mdt_info_req(info);
3502
3503                         LASSERT(req != NULL);
3504                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3505                                           req->rq_transno);
3506                         ldlm_lock_decref(h, mode);
3507                 }
3508                 h->cookie = 0ull;
3509         }
3510
3511         EXIT;
3512 }
3513
3514 /**
3515  * Unlock mdt object.
3516  *
3517  * Immeditely release the regular lock and the PDO lock or save the
3518  * lock in request and keep them referenced until client ACK or
3519  * transaction commit.
3520  *
3521  * \param info thread info object
3522  * \param o mdt object
3523  * \param lh mdt lock handle referencing regular and PDO locks
3524  * \param decref force immediate lock releasing
3525  *
3526  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3527  */
3528 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3529                        struct mdt_lock_handle *lh, int decref)
3530 {
3531         ENTRY;
3532
3533         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3534         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3535         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3536                              decref);
3537
3538         EXIT;
3539 }
3540
3541 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3542                                         const struct lu_fid *f,
3543                                         struct mdt_lock_handle *lh,
3544                                         __u64 ibits)
3545 {
3546         struct mdt_object *o;
3547
3548         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3549         if (!IS_ERR(o)) {
3550                 int rc;
3551
3552                 rc = mdt_object_lock(info, o, lh, ibits);
3553                 if (rc != 0) {
3554                         mdt_object_put(info->mti_env, o);
3555                         o = ERR_PTR(rc);
3556                 }
3557         }
3558         return o;
3559 }
3560
3561 void mdt_object_unlock_put(struct mdt_thread_info * info,
3562                            struct mdt_object * o,
3563                            struct mdt_lock_handle *lh,
3564                            int decref)
3565 {
3566         mdt_object_unlock(info, o, lh, decref);
3567         mdt_object_put(info->mti_env, o);
3568 }
3569
3570 /*
3571  * Generic code handling requests that have struct mdt_body passed in:
3572  *
3573  *  - extract mdt_body from request and save it in @info, if present;
3574  *
3575  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3576  *  @info;
3577  *
3578  *  - if HAS_BODY flag is set for this request type check whether object
3579  *  actually exists on storage (lu_object_exists()).
3580  *
3581  */
3582 static int mdt_body_unpack(struct mdt_thread_info *info,
3583                            enum tgt_handler_flags flags)
3584 {
3585         const struct mdt_body    *body;
3586         struct mdt_object        *obj;
3587         const struct lu_env      *env;
3588         struct req_capsule       *pill;
3589         int                       rc;
3590         ENTRY;
3591
3592         env = info->mti_env;
3593         pill = info->mti_pill;
3594
3595         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3596         if (body == NULL)
3597                 RETURN(-EFAULT);
3598
3599         if (!(body->mbo_valid & OBD_MD_FLID))
3600                 RETURN(0);
3601
3602         if (!fid_is_sane(&body->mbo_fid1)) {
3603                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3604                 RETURN(-EINVAL);
3605         }
3606
3607         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3608         if (!IS_ERR(obj)) {
3609                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3610                         mdt_object_put(env, obj);
3611                         rc = -ENOENT;
3612                 } else {
3613                         info->mti_object = obj;
3614                         rc = 0;
3615                 }
3616         } else
3617                 rc = PTR_ERR(obj);
3618
3619         RETURN(rc);
3620 }
3621
3622 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3623                                    enum tgt_handler_flags flags)
3624 {
3625         struct req_capsule *pill = info->mti_pill;
3626         int rc;
3627
3628         ENTRY;
3629
3630         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3631                 rc = mdt_body_unpack(info, flags);
3632         else
3633                 rc = 0;
3634
3635         if (rc == 0 && (flags & HAS_REPLY)) {
3636                 /* Pack reply. */
3637                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3638                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3639                                              DEF_REP_MD_SIZE);
3640                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3641                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3642                                              RCL_SERVER, 0);
3643
3644                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3645                  * by default. If the target object has more ACL entries, then
3646                  * enlarge the buffer when necessary. */
3647                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3648                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3649                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3650
3651                 mdt_preset_secctx_size(info);
3652
3653                 rc = req_capsule_server_pack(pill);
3654                 if (rc)
3655                         CWARN("%s: cannot pack response: rc = %d\n",
3656                                       mdt_obd_name(info->mti_mdt), rc);
3657         }
3658         RETURN(rc);
3659 }
3660
3661 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3662 {
3663         lh->mlh_type = MDT_NUL_LOCK;
3664         lh->mlh_reg_lh.cookie = 0ull;
3665         lh->mlh_reg_mode = LCK_MINMODE;
3666         lh->mlh_pdo_lh.cookie = 0ull;
3667         lh->mlh_pdo_mode = LCK_MINMODE;
3668         lh->mlh_rreg_lh.cookie = 0ull;
3669         lh->mlh_rreg_mode = LCK_MINMODE;
3670 }
3671
3672 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3673 {
3674         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3675         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3676 }
3677
3678 /*
3679  * Initialize fields of struct mdt_thread_info. Other fields are left in
3680  * uninitialized state, because it's too expensive to zero out whole
3681  * mdt_thread_info (> 1K) on each request arrival.
3682  */
3683 void mdt_thread_info_init(struct ptlrpc_request *req,
3684                           struct mdt_thread_info *info)
3685 {
3686         int i;
3687
3688         info->mti_pill = &req->rq_pill;
3689
3690         /* lock handle */
3691         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3692                 mdt_lock_handle_init(&info->mti_lh[i]);
3693
3694         /* mdt device: it can be NULL while CONNECT */
3695         if (req->rq_export) {
3696                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3697                 info->mti_exp = req->rq_export;
3698         } else
3699                 info->mti_mdt = NULL;
3700         info->mti_env = req->rq_svc_thread->t_env;
3701         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3702
3703         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3704         info->mti_big_buf = LU_BUF_NULL;
3705         info->mti_body = NULL;
3706         info->mti_object = NULL;
3707         info->mti_dlm_req = NULL;
3708         info->mti_has_trans = 0;
3709         info->mti_cross_ref = 0;
3710         info->mti_opdata = 0;
3711         info->mti_big_lmm_used = 0;
3712         info->mti_big_acl_used = 0;
3713         info->mti_som_valid = 0;
3714
3715         info->mti_spec.no_create = 0;
3716         info->mti_spec.sp_rm_entry = 0;
3717         info->mti_spec.sp_permitted = 0;
3718         info->mti_spec.sp_migrate_close = 0;
3719
3720         info->mti_spec.u.sp_ea.eadata = NULL;
3721         info->mti_spec.u.sp_ea.eadatalen = 0;
3722 }
3723
3724 void mdt_thread_info_fini(struct mdt_thread_info *info)
3725 {
3726         int i;
3727
3728         if (info->mti_object != NULL) {
3729                 mdt_object_put(info->mti_env, info->mti_object);
3730                 info->mti_object = NULL;
3731         }
3732
3733         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3734                 mdt_lock_handle_fini(&info->mti_lh[i]);
3735         info->mti_env = NULL;
3736         info->mti_pill = NULL;
3737         info->mti_exp = NULL;
3738
3739         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3740                 lu_buf_free(&info->mti_big_buf);
3741 }
3742
3743 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3744 {
3745         struct mdt_thread_info  *mti;
3746
3747         mti = mdt_th_info(tsi->tsi_env);
3748         LASSERT(mti != NULL);
3749
3750         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3751         if (tsi->tsi_corpus != NULL) {
3752                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3753                 lu_object_get(tsi->tsi_corpus);
3754         }
3755         mti->mti_body = tsi->tsi_mdt_body;
3756         mti->mti_dlm_req = tsi->tsi_dlm_req;
3757
3758         return mti;
3759 }
3760
3761 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3762 {
3763         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3764             cfs_fail_val ==
3765             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3766                 set_current_state(TASK_UNINTERRUPTIBLE);
3767                 schedule_timeout(cfs_time_seconds(3));
3768         }
3769
3770         return tgt_connect(tsi);
3771 }
3772
3773 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3774                               struct mdt_thread_info *info,
3775                               struct ldlm_lock **lockp, __u64 flags)
3776 {
3777         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3778                                    lockp, flags);
3779 }
3780 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3781                           struct mdt_thread_info *info,
3782                           struct ldlm_lock **lockp, __u64 flags)
3783 {
3784         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3785                                lockp, flags);
3786 }
3787
3788 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3789                             struct ldlm_lock **lockp,
3790                             struct mdt_lock_handle *lh,
3791                             __u64 flags, int result)
3792 {
3793         struct ptlrpc_request  *req = mdt_info_req(info);
3794         struct ldlm_lock       *lock = *lockp;
3795         struct ldlm_lock       *new_lock;
3796
3797         /* If possible resent found a lock, @lh is set to its handle */
3798         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3799
3800         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3801                 lh->mlh_reg_lh.cookie = 0;
3802                 RETURN(0);
3803         }
3804
3805         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3806                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3807                  * a resend case, however, it could be already destroyed
3808                  * due to client eviction or a raced cancel RPC. */
3809                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3810                                   lh->mlh_reg_lh.cookie);
3811                 lh->mlh_reg_lh.cookie = 0;
3812                 RETURN(-ESTALE);
3813         }
3814
3815         LASSERTF(new_lock != NULL,
3816                  "lockh %#llx flags %#llx : rc = %d\n",
3817                  lh->mlh_reg_lh.cookie, flags, result);
3818
3819         /*
3820          * If we've already given this lock to a client once, then we should
3821          * have no readers or writers.  Otherwise, we should have one reader
3822          * _or_ writer ref (which will be zeroed below) before returning the
3823          * lock to a client.
3824          */
3825         if (new_lock->l_export == req->rq_export) {
3826                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3827         } else {
3828                 LASSERT(new_lock->l_export == NULL);
3829                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3830         }
3831
3832         *lockp = new_lock;
3833
3834         if (new_lock->l_export == req->rq_export) {
3835                 /*
3836                  * Already gave this to the client, which means that we
3837                  * reconstructed a reply.
3838                  */
3839                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3840                         MSG_RESENT);
3841
3842                 LDLM_LOCK_RELEASE(new_lock);
3843                 lh->mlh_reg_lh.cookie = 0;
3844                 RETURN(ELDLM_LOCK_REPLACED);
3845         }
3846
3847         /*
3848          * Fixup the lock to be given to the client.
3849          */
3850         lock_res_and_lock(new_lock);
3851         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3852          * possible blocking AST. */
3853         while (new_lock->l_readers > 0) {
3854                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3855                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3856                 new_lock->l_readers--;
3857         }
3858         while (new_lock->l_writers > 0) {
3859                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3860                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3861                 new_lock->l_writers--;
3862         }
3863
3864         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3865         new_lock->l_blocking_ast = lock->l_blocking_ast;
3866         new_lock->l_completion_ast = lock->l_completion_ast;
3867         if (ldlm_has_dom(new_lock))
3868                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3869         new_lock->l_remote_handle = lock->l_remote_handle;
3870         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3871
3872         unlock_res_and_lock(new_lock);
3873
3874         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3875                      &new_lock->l_remote_handle,
3876                      &new_lock->l_exp_hash);
3877
3878         LDLM_LOCK_RELEASE(new_lock);
3879         lh->mlh_reg_lh.cookie = 0;
3880
3881         RETURN(ELDLM_LOCK_REPLACED);
3882 }
3883
3884 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3885                              struct ldlm_lock *new_lock,
3886                              struct mdt_lock_handle *lh, __u64 flags)
3887 {
3888         struct ptlrpc_request  *req = mdt_info_req(info);
3889         struct ldlm_request    *dlmreq;
3890
3891         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3892                 return;
3893
3894         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3895
3896         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3897          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3898          * initialized. */
3899         if (flags & LDLM_FL_RESENT) {
3900                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3901                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3902
3903                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3904                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3905                           lh->mlh_reg_lh.cookie);
3906                 return;
3907         }
3908
3909         /*
3910          * If the xid matches, then we know this is a resent request, and allow
3911          * it. (It's probably an OPEN, for which we don't send a lock.
3912          */
3913         if (req_can_reconstruct(req, NULL))
3914                 return;
3915
3916         /*
3917          * This remote handle isn't enqueued, so we never received or processed
3918          * this request.  Clear MSG_RESENT, because it can be handled like any
3919          * normal request now.
3920          */
3921         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3922
3923         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3924                   dlmreq->lock_handle[0].cookie);
3925 }
3926
3927 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3928                                struct mdt_thread_info *info,
3929                                struct ldlm_lock **lockp,
3930                                __u64 flags)
3931 {
3932         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3933         struct ldlm_reply      *ldlm_rep = NULL;
3934         int rc;
3935         ENTRY;
3936
3937         /*
3938          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3939          * (for the resend case) or a new lock. Below we will use it to
3940          * replace the original lock.
3941          */
3942         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3943         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3944                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3945                 rc = mdt_object_lock(info, info->mti_object, lhc,
3946                                      MDS_INODELOCK_XATTR);
3947                 if (rc)
3948                         return rc;
3949         }
3950
3951         rc = mdt_getxattr(info);
3952
3953         if (mdt_info_req(info)->rq_repmsg != NULL)
3954                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3955
3956         if (ldlm_rep == NULL ||
3957             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3958                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3959                 if (is_serious(rc))
3960                         RETURN(rc);
3961                 else
3962                         RETURN(err_serious(-EFAULT));
3963         }
3964
3965         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3966
3967         /* This is left for interop instead of adding a new interop flag.
3968          * LU-7433 */
3969 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3970         if (ldlm_rep->lock_policy_res2) {
3971                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3972                 RETURN(ELDLM_LOCK_ABORTED);
3973         }
3974 #endif
3975
3976         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3977         RETURN(rc);
3978 }
3979
3980 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
3981                               struct mdt_thread_info *info,
3982                               struct ldlm_lock **lockp,
3983                               __u64 flags)
3984 {
3985         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3986         __u64                   child_bits;
3987         struct ldlm_reply      *ldlm_rep;
3988         struct mdt_body        *reqbody;
3989         struct mdt_body        *repbody;
3990         int                     rc, rc2;
3991         ENTRY;
3992
3993         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3994         LASSERT(reqbody);
3995
3996         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3997         LASSERT(repbody);
3998
3999         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4000         repbody->mbo_eadatasize = 0;
4001         repbody->mbo_aclsize = 0;
4002
4003         switch (it_opc) {
4004         case IT_LOOKUP:
4005                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4006                 break;
4007         case IT_GETATTR:
4008                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4009                              MDS_INODELOCK_PERM;
4010                 break;
4011         default:
4012                 CERROR("%s: unsupported intent %#x\n",
4013                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4014                 GOTO(out_shrink, rc = -EINVAL);
4015         }
4016
4017         rc = mdt_init_ucred_intent_getattr(info, reqbody);
4018         if (rc)
4019                 GOTO(out_shrink, rc);
4020
4021         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4022         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4023
4024         /* Get lock from request for possible resent case. */
4025         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4026
4027         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4028         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4029
4030         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4031                 ldlm_rep->lock_policy_res2 = 0;
4032         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4033             ldlm_rep->lock_policy_res2) {
4034                 lhc->mlh_reg_lh.cookie = 0ull;
4035                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4036         }
4037
4038         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4039         EXIT;
4040 out_ucred:
4041         mdt_exit_ucred(info);
4042 out_shrink:
4043         mdt_client_compatibility(info);
4044         rc2 = mdt_fix_reply(info);
4045         if (rc == 0)
4046                 rc = rc2;
4047         return rc;
4048 }
4049
4050 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4051                              struct mdt_thread_info *info,
4052                              struct ldlm_lock **lockp,
4053                              __u64 flags)
4054 {
4055         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
4056         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4057         struct layout_intent *intent;
4058         struct lu_fid *fid = &info->mti_tmp_fid2;
4059         struct mdt_object *obj = NULL;
4060         int layout_size = 0;
4061         int rc = 0;
4062         ENTRY;
4063
4064         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4065
4066         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4067         if (intent == NULL)
4068                 RETURN(-EPROTO);
4069
4070         CDEBUG(D_INFO, DFID "got layout change request from client: "
4071                "opc:%u flags:%#x extent "DEXT"\n",
4072                PFID(fid), intent->li_opc, intent->li_flags,
4073                PEXT(&intent->li_extent));
4074
4075         switch (intent->li_opc) {
4076         case LAYOUT_INTENT_TRUNC:
4077         case LAYOUT_INTENT_WRITE:
4078                 layout.mlc_opc = MD_LAYOUT_WRITE;
4079                 layout.mlc_intent = intent;
4080                 break;
4081         case LAYOUT_INTENT_ACCESS:
4082                 break;
4083         case LAYOUT_INTENT_READ:
4084         case LAYOUT_INTENT_GLIMPSE:
4085         case LAYOUT_INTENT_RELEASE:
4086         case LAYOUT_INTENT_RESTORE:
4087                 CERROR("%s: Unsupported layout intent opc %d\n",
4088                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4089                 rc = -ENOTSUPP;
4090                 break;
4091         default:
4092                 CERROR("%s: Unknown layout intent opc %d\n",
4093                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4094                 rc = -EINVAL;
4095                 break;
4096         }
4097         if (rc < 0)
4098                 RETURN(rc);
4099
4100         /* Get lock from request for possible resent case. */
4101         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4102
4103         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4104         if (IS_ERR(obj))
4105                 GOTO(out, rc = PTR_ERR(obj));
4106
4107
4108         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4109                 /* if layout is going to be changed don't use the current EA
4110                  * size but the maximum one. That buffer will be shrinked
4111                  * to the actual size in req_capsule_shrink() before reply.
4112                  */
4113                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4114                         layout_size = info->mti_mdt->mdt_max_mdsize;
4115                 } else {
4116                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4117                         if (layout_size < 0)
4118                                 GOTO(out_obj, rc = layout_size);
4119
4120                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4121                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4122                 }
4123                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4124                        mdt_obd_name(info->mti_mdt), layout_size);
4125         }
4126
4127         /*
4128          * set reply buffer size, so that ldlm_handle_enqueue0()->
4129          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4130          */
4131         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4132         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4133                              layout_size);
4134         rc = req_capsule_server_pack(info->mti_pill);
4135         if (rc)
4136                 GOTO(out_obj, rc);
4137
4138
4139         if (layout.mlc_opc != MD_LAYOUT_NOP) {
4140                 struct lu_buf *buf = &layout.mlc_buf;
4141
4142                 /**
4143                  * mdt_layout_change is a reint operation, when the request
4144                  * is resent, layout write shouldn't reprocess it again.
4145                  */
4146                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4147                 if (rc)
4148                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
4149
4150                 /**
4151                  * There is another resent case: the client's job has been
4152                  * done by another client, referring lod_declare_layout_change
4153                  * -EALREADY case, and it became a operation w/o transaction,
4154                  * so we should not do the layout change, otherwise
4155                  * mdt_layout_change() will try to cancel the granted server
4156                  * CR lock whose remote counterpart is still in hold on the
4157                  * client, and a deadlock ensues.
4158                  */
4159                 rc = mdt_check_resent_lock(info, obj, lhc);
4160                 if (rc <= 0)
4161                         GOTO(out_obj, rc);
4162
4163                 buf->lb_buf = NULL;
4164                 buf->lb_len = 0;
4165                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
4166                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
4167                                         &RMF_EADATA);
4168                         buf->lb_len = req_capsule_get_size(info->mti_pill,
4169                                         &RMF_EADATA, RCL_CLIENT);
4170                         /*
4171                          * If it's a replay of layout write intent RPC, the
4172                          * client has saved the extended lovea when
4173                          * it get reply then.
4174                          */
4175                         if (buf->lb_len > 0)
4176                                 mdt_fix_lov_magic(info, buf->lb_buf);
4177                 }
4178                 /*
4179                  * Instantiate some layout components, if @buf contains
4180                  * lovea, then it's a replay of the layout intent write
4181                  * RPC.
4182                  */
4183                 rc = mdt_layout_change(info, obj, &layout);
4184                 if (rc)
4185                         GOTO(out_obj, rc);
4186         }
4187 out_obj:
4188         mdt_object_put(info->mti_env, obj);
4189
4190         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
4191                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4192
4193 out:
4194         lhc->mlh_reg_lh.cookie = 0;
4195
4196         RETURN(rc);
4197 }
4198
4199 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4200                            struct mdt_thread_info *info,
4201                            struct ldlm_lock **lockp,
4202                            __u64 flags)
4203 {
4204         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4205         struct ldlm_reply      *rep = NULL;
4206         long                    opc;
4207         int                     rc;
4208
4209         static const struct req_format *intent_fmts[REINT_MAX] = {
4210                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4211                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4212         };
4213
4214         ENTRY;
4215
4216         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4217         if (opc < 0)
4218                 RETURN(opc);
4219
4220         /* Get lock from request for possible resent case. */
4221         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4222
4223         rc = mdt_reint_internal(info, lhc, opc);
4224
4225         /* Check whether the reply has been packed successfully. */
4226         if (mdt_info_req(info)->rq_repmsg != NULL)
4227                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4228         if (rep == NULL) {
4229                 if (is_serious(rc))
4230                         RETURN(rc);
4231                 else
4232                         RETURN(err_serious(-EFAULT));
4233         }
4234
4235         /* MDC expects this in any case */
4236         if (rc != 0)
4237                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4238
4239         /* the open lock or the lock for cross-ref object should be
4240          * returned to the client */
4241         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4242             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4243                 rep->lock_policy_res2 = 0;
4244                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4245                 RETURN(rc);
4246         }
4247
4248         rep->lock_policy_res2 = clear_serious(rc);
4249
4250         if (rep->lock_policy_res2 == -ENOENT &&
4251             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4252             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4253                 rep->lock_policy_res2 = 0;
4254
4255         lhc->mlh_reg_lh.cookie = 0ull;
4256         if (rc == -ENOTCONN || rc == -ENODEV ||
4257             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4258                 /*
4259                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4260                  * will be returned by rq_status, and client at ptlrpc layer
4261                  * will detect this, then disconnect, reconnect the import
4262                  * immediately, instead of impacting the following the rpc.
4263                  */
4264                 RETURN(rc);
4265         }
4266         /*
4267          * For other cases, the error will be returned by intent, and client
4268          * will retrieve the result from intent.
4269          */
4270         RETURN(ELDLM_LOCK_ABORTED);
4271 }
4272
4273 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4274                           struct mdt_thread_info *info,
4275                           struct ldlm_lock **lockp,
4276                           u64 flags /* LDLM_FL_* */)
4277 {
4278         struct req_capsule *pill = info->mti_pill;
4279         struct ptlrpc_request *req = mdt_info_req(info);
4280         const struct req_format *it_format;
4281         int (*it_handler)(enum ldlm_intent_flags,
4282                           struct mdt_thread_info *,
4283                           struct ldlm_lock **,
4284                           u64);
4285         enum tgt_handler_flags it_handler_flags = 0;
4286         struct ldlm_reply *rep;
4287         int rc;
4288         ENTRY;
4289
4290         switch (it_opc) {
4291         case IT_OPEN:
4292         case IT_OPEN|IT_CREAT:
4293                 /*
4294                  * OCREAT is not a IS_MUTABLE request since the file may
4295                  * already exist. We do the extra check of
4296                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4297                  * really need to create the object.
4298                  */
4299                 it_format = &RQF_LDLM_INTENT;
4300                 it_handler = &mdt_intent_open;
4301                 break;
4302         case IT_GETATTR:
4303         case IT_LOOKUP:
4304                 it_format = &RQF_LDLM_INTENT_GETATTR;
4305                 it_handler = &mdt_intent_getattr;
4306                 it_handler_flags = HAS_REPLY;
4307                 break;
4308         case IT_GETXATTR:
4309                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4310                 it_handler = &mdt_intent_getxattr;
4311                 it_handler_flags = HAS_BODY;
4312                 break;
4313         case IT_LAYOUT:
4314                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4315                 it_handler = &mdt_intent_layout;
4316                 break;
4317         case IT_GLIMPSE:
4318                 it_format = &RQF_LDLM_INTENT;
4319                 it_handler = &mdt_intent_glimpse;
4320                 break;
4321         case IT_BRW:
4322                 it_format = &RQF_LDLM_INTENT;
4323                 it_handler = &mdt_intent_brw;
4324                 break;
4325         case IT_QUOTA_DQACQ:
4326         case IT_QUOTA_CONN: {
4327                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4328
4329                 if (qmt == NULL)
4330                         RETURN(-EOPNOTSUPP);
4331
4332                 if (mdt_rdonly(req->rq_export))
4333                         RETURN(-EROFS);
4334
4335                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4336                 /* pass the request to quota master */
4337                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4338                                                  mdt_info_req(info), lockp,
4339                                                  flags);
4340                 RETURN(rc);
4341         }
4342         default:
4343                 CERROR("%s: unknown intent code %#x\n",
4344                        mdt_obd_name(info->mti_mdt), it_opc);
4345                 RETURN(-EPROTO);
4346         }
4347
4348         req_capsule_extend(pill, it_format);
4349
4350         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4351         if (rc < 0)
4352                 RETURN(rc);
4353
4354         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4355                 RETURN(-EROFS);
4356
4357         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4358
4359         /* execute policy */
4360         rc = (*it_handler)(it_opc, info, lockp, flags);
4361
4362         /* Check whether the reply has been packed successfully. */
4363         if (req->rq_repmsg != NULL) {
4364                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4365                 rep->lock_policy_res2 =
4366                         ptlrpc_status_hton(rep->lock_policy_res2);
4367         }
4368
4369         RETURN(rc);
4370 }
4371
4372 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4373                                     enum ldlm_intent_flags it_opc)
4374 {
4375         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4376
4377         /* update stats when IT code is known */
4378         if (srv_stats != NULL)
4379                 lprocfs_counter_incr(srv_stats,
4380                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4381                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4382 }
4383
4384 static int mdt_intent_policy(const struct lu_env *env,
4385                              struct ldlm_namespace *ns,
4386                              struct ldlm_lock **lockp,
4387                              void *req_cookie,
4388                              enum ldlm_mode mode,
4389                              __u64 flags, void *data)
4390 {
4391         struct tgt_session_info *tsi;
4392         struct mdt_thread_info  *info;
4393         struct ptlrpc_request   *req  =  req_cookie;
4394         struct ldlm_intent      *it;
4395         struct req_capsule      *pill;
4396         const struct ldlm_lock_desc *ldesc;
4397         int rc;
4398
4399         ENTRY;
4400
4401         LASSERT(req != NULL);
4402
4403         tsi = tgt_ses_info(env);
4404
4405         info = tsi2mdt_info(tsi);
4406         LASSERT(info != NULL);
4407         pill = info->mti_pill;
4408         LASSERT(pill->rc_req == req);
4409         ldesc = &info->mti_dlm_req->lock_desc;
4410
4411         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4412                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4413                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4414                 if (it != NULL) {
4415                         mdt_ptlrpc_stats_update(req, it->opc);
4416                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4417                         if (rc == 0)
4418                                 rc = ELDLM_OK;
4419
4420                         /* Lock without inodebits makes no sense and will oops
4421                          * later in ldlm. Let's check it now to see if we have
4422                          * ibits corrupted somewhere in mdt_intent_opc().
4423                          * The case for client miss to set ibits has been
4424                          * processed by others. */
4425                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4426                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4427                 } else {
4428                         rc = err_serious(-EFAULT);
4429                 }
4430         } else {
4431                 /* No intent was provided */
4432                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4433                 rc = req_capsule_server_pack(pill);
4434                 if (rc)
4435                         rc = err_serious(rc);
4436         }
4437         mdt_thread_info_fini(info);
4438         RETURN(rc);
4439 }
4440
4441 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4442 {
4443         struct seq_server_site  *ss = mdt_seq_site(mdt);
4444
4445         if (ss->ss_node_id == 0)
4446                 return;
4447
4448         if (ss->ss_client_seq != NULL) {
4449                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4450                 ss->ss_client_seq->lcs_exp = NULL;
4451         }
4452
4453         if (ss->ss_server_fld != NULL) {
4454                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4455                 ss->ss_server_fld->lsf_control_exp = NULL;
4456         }
4457 }
4458
4459 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4460 {
4461         struct seq_server_site *ss = mdt_seq_site(mdt);
4462
4463         if (ss == NULL)
4464                 return;
4465
4466         if (ss->ss_server_seq != NULL)
4467                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4468 }
4469
4470 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4471 {
4472         mdt_seq_fini_cli(mdt);
4473         mdt_deregister_seq_exp(mdt);
4474
4475         return seq_site_fini(env, mdt_seq_site(mdt));
4476 }
4477
4478 /**
4479  * It will retrieve its FLDB entries from MDT0, and it only happens
4480  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4481  * and it needs to refresh FLDB from the MDT0.
4482  **/
4483 static int mdt_register_lwp_callback(void *data)
4484 {
4485         struct lu_env           env;
4486         struct mdt_device       *mdt = data;
4487         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4488         int                     rc;
4489         ENTRY;
4490
4491         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4492
4493         rc = lu_env_init(&env, LCT_MD_THREAD);
4494         if (rc < 0) {
4495                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4496                 RETURN(rc);
4497         }
4498
4499         /* Allocate new sequence now to avoid creating local transaction
4500          * in the normal transaction process */
4501         rc = seq_server_check_and_alloc_super(&env,
4502                                               mdt_seq_site(mdt)->ss_server_seq);
4503         if (rc < 0)
4504                 GOTO(out, rc);
4505
4506         if (fld->lsf_new) {
4507                 rc = fld_update_from_controller(&env, fld);
4508                 if (rc != 0) {
4509                         CERROR("%s: cannot update controller: rc = %d\n",
4510                                mdt_obd_name(mdt), rc);
4511                         GOTO(out, rc);
4512                 }
4513         }
4514 out:
4515         lu_env_fini(&env);
4516         RETURN(rc);
4517 }
4518
4519 static int mdt_register_seq_exp(struct mdt_device *mdt)
4520 {
4521         struct seq_server_site  *ss = mdt_seq_site(mdt);
4522         char                    *lwp_name = NULL;
4523         int                     rc;
4524
4525         if (ss->ss_node_id == 0)
4526                 return 0;
4527
4528         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4529         if (lwp_name == NULL)
4530                 GOTO(out_free, rc = -ENOMEM);
4531
4532         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4533         if (rc != 0)
4534                 GOTO(out_free, rc);
4535
4536         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4537                                       NULL, NULL);
4538         if (rc != 0)
4539                 GOTO(out_free, rc);
4540
4541         rc = lustre_register_lwp_item(lwp_name,
4542                                       &ss->ss_server_fld->lsf_control_exp,
4543                                       mdt_register_lwp_callback, mdt);
4544         if (rc != 0) {
4545                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4546                 ss->ss_client_seq->lcs_exp = NULL;
4547                 GOTO(out_free, rc);
4548         }
4549 out_free:
4550         if (lwp_name != NULL)
4551                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4552
4553         return rc;
4554 }
4555
4556 /*
4557  * Init client sequence manager which is used by local MDS to talk to sequence
4558  * controller on remote node.
4559  */
4560 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4561 {
4562         struct seq_server_site  *ss = mdt_seq_site(mdt);
4563         int                     rc;
4564         char                    *prefix;
4565         ENTRY;
4566
4567         /* check if this is adding the first MDC and controller is not yet
4568          * initialized. */
4569         OBD_ALLOC_PTR(ss->ss_client_seq);
4570         if (ss->ss_client_seq == NULL)
4571                 RETURN(-ENOMEM);
4572
4573         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4574         if (prefix == NULL) {
4575                 OBD_FREE_PTR(ss->ss_client_seq);
4576                 ss->ss_client_seq = NULL;
4577                 RETURN(-ENOMEM);
4578         }
4579
4580         /* Note: seq_client_fini will be called in seq_site_fini */
4581         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4582         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4583                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4584                                                             NULL);
4585         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4586         if (rc != 0) {
4587                 OBD_FREE_PTR(ss->ss_client_seq);
4588                 ss->ss_client_seq = NULL;
4589                 RETURN(rc);
4590         }
4591
4592         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4593
4594         RETURN(rc);
4595 }
4596
4597 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4598 {
4599         struct seq_server_site  *ss;
4600         int                     rc;
4601         ENTRY;
4602
4603         ss = mdt_seq_site(mdt);
4604         /* init sequence controller server(MDT0) */
4605         if (ss->ss_node_id == 0) {
4606                 OBD_ALLOC_PTR(ss->ss_control_seq);
4607                 if (ss->ss_control_seq == NULL)
4608                         RETURN(-ENOMEM);
4609
4610                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4611                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4612                                      ss);
4613                 if (rc)
4614                         GOTO(out_seq_fini, rc);
4615         }
4616
4617         /* Init normal sequence server */
4618         OBD_ALLOC_PTR(ss->ss_server_seq);
4619         if (ss->ss_server_seq == NULL)
4620                 GOTO(out_seq_fini, rc = -ENOMEM);
4621
4622         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4623                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4624         if (rc)
4625                 GOTO(out_seq_fini, rc);
4626
4627         /* init seq client for seq server to talk to seq controller(MDT0) */
4628         rc = mdt_seq_init_cli(env, mdt);
4629         if (rc != 0)
4630                 GOTO(out_seq_fini, rc);
4631
4632         if (ss->ss_node_id != 0)
4633                 /* register controller export through lwp */
4634                 rc = mdt_register_seq_exp(mdt);
4635
4636         EXIT;
4637 out_seq_fini:
4638         if (rc)
4639                 mdt_seq_fini(env, mdt);
4640
4641         return rc;
4642 }
4643
4644 /*
4645  * FLD wrappers
4646  */
4647 static int mdt_fld_fini(const struct lu_env *env,
4648                         struct mdt_device *m)
4649 {
4650         struct seq_server_site *ss = mdt_seq_site(m);
4651         ENTRY;
4652
4653         if (ss && ss->ss_server_fld) {
4654                 fld_server_fini(env, ss->ss_server_fld);
4655                 OBD_FREE_PTR(ss->ss_server_fld);
4656                 ss->ss_server_fld = NULL;
4657         }
4658
4659         RETURN(0);
4660 }
4661
4662 static int mdt_fld_init(const struct lu_env *env,
4663                         const char *uuid,
4664                         struct mdt_device *m)
4665 {
4666         struct seq_server_site *ss;
4667         int rc;
4668         ENTRY;
4669
4670         ss = mdt_seq_site(m);
4671
4672         OBD_ALLOC_PTR(ss->ss_server_fld);
4673         if (ss->ss_server_fld == NULL)
4674                 RETURN(rc = -ENOMEM);
4675
4676         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4677                              LU_SEQ_RANGE_MDT);
4678         if (rc) {
4679                 OBD_FREE_PTR(ss->ss_server_fld);
4680                 ss->ss_server_fld = NULL;
4681                 RETURN(rc);
4682         }
4683
4684         RETURN(0);
4685 }
4686
4687 static void mdt_stack_pre_fini(const struct lu_env *env,
4688                            struct mdt_device *m, struct lu_device *top)
4689 {
4690         struct lustre_cfg_bufs  *bufs;
4691         struct lustre_cfg       *lcfg;
4692         struct mdt_thread_info  *info;
4693         ENTRY;
4694
4695         LASSERT(top);
4696
4697         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4698         LASSERT(info != NULL);
4699
4700         bufs = &info->mti_u.bufs;
4701
4702         LASSERT(m->mdt_child_exp);
4703         LASSERT(m->mdt_child_exp->exp_obd);
4704
4705         /* process cleanup, pass mdt obd name to get obd umount flags */
4706         /* XXX: this is needed because all layers are referenced by
4707          * objects (some of them are pinned by osd, for example *
4708          * the proper solution should be a model where object used
4709          * by osd only doesn't have mdt/mdd slices -bzzz */
4710         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4711         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4712         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4713         if (!lcfg)
4714                 RETURN_EXIT;
4715         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4716
4717         top->ld_ops->ldo_process_config(env, top, lcfg);
4718         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4719         EXIT;
4720 }
4721
4722 static void mdt_stack_fini(const struct lu_env *env,
4723                            struct mdt_device *m, struct lu_device *top)
4724 {
4725         struct obd_device       *obd = mdt2obd_dev(m);
4726         struct lustre_cfg_bufs  *bufs;
4727         struct lustre_cfg       *lcfg;
4728         struct mdt_thread_info  *info;
4729         char                     flags[3] = "";
4730         ENTRY;
4731
4732         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4733         LASSERT(info != NULL);
4734
4735         lu_dev_del_linkage(top->ld_site, top);
4736
4737         lu_site_purge(env, top->ld_site, -1);
4738
4739         bufs = &info->mti_u.bufs;
4740         /* process cleanup, pass mdt obd name to get obd umount flags */
4741         /* another purpose is to let all layers to release their objects */
4742         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4743         if (obd->obd_force)
4744                 strcat(flags, "F");
4745         if (obd->obd_fail)
4746                 strcat(flags, "A");
4747         lustre_cfg_bufs_set_string(bufs, 1, flags);
4748         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4749         if (!lcfg)
4750                 RETURN_EXIT;
4751         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4752
4753         LASSERT(top);
4754         top->ld_ops->ldo_process_config(env, top, lcfg);
4755         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4756
4757         lu_site_purge(env, top->ld_site, -1);
4758
4759         m->mdt_child = NULL;
4760         m->mdt_bottom = NULL;
4761
4762         obd_disconnect(m->mdt_child_exp);
4763         m->mdt_child_exp = NULL;
4764
4765         obd_disconnect(m->mdt_bottom_exp);
4766         m->mdt_child_exp = NULL;
4767 }
4768
4769 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4770                                const char *next, struct obd_export **exp)
4771 {
4772         struct obd_connect_data *data = NULL;
4773         struct obd_device       *obd;
4774         int                      rc;
4775         ENTRY;
4776
4777         OBD_ALLOC_PTR(data);
4778         if (data == NULL)
4779                 GOTO(out, rc = -ENOMEM);
4780
4781         obd = class_name2obd(next);
4782         if (obd == NULL) {
4783                 CERROR("%s: can't locate next device: %s\n",
4784                        mdt_obd_name(m), next);
4785                 GOTO(out, rc = -ENOTCONN);
4786         }
4787
4788         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4789         data->ocd_version = LUSTRE_VERSION_CODE;
4790
4791         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4792         if (rc) {
4793                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4794                        mdt_obd_name(m), next, rc);
4795                 GOTO(out, rc);
4796         }
4797
4798 out:
4799         if (data)
4800                 OBD_FREE_PTR(data);
4801         RETURN(rc);
4802 }
4803
4804 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4805                           struct lustre_cfg *cfg)
4806 {
4807         char                   *dev = lustre_cfg_string(cfg, 0);
4808         int                     rc, name_size, uuid_size;
4809         char                   *name, *uuid, *p;
4810         struct lustre_cfg_bufs *bufs;
4811         struct lustre_cfg      *lcfg;
4812         struct obd_device      *obd;
4813         struct lustre_profile  *lprof;
4814         struct lu_site         *site;
4815         ENTRY;
4816
4817         /* in 1.8 we had the only device in the stack - MDS.
4818          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4819          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4820          * to generate names and setup MDT, MDD. MDT will be using
4821          * generated name to connect to MDD. for MDD the next device
4822          * will be LOD with name taken from so called "profile" which
4823          * is generated by mount_option line
4824          *
4825          * 1.8 MGS generates config. commands like this:
4826          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4827          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4828          * 2.0 MGS generates config. commands like this:
4829          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4830          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4831          *                    3:lustre-MDT0000-mdtlov  4:f
4832          *
4833          * we generate MDD name from MDT one, just replacing T with D
4834          *
4835          * after all the preparations, the logical equivalent will be
4836          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4837          *                    3:lustre-MDT0000-mdtlov  4:f
4838          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4839          *                    3:lustre-MDD0000  4:f
4840          *
4841          *  notice we build the stack from down to top: MDD first, then MDT */
4842
4843         name_size = MAX_OBD_NAME;
4844         uuid_size = MAX_OBD_NAME;
4845
4846         OBD_ALLOC(name, name_size);
4847         OBD_ALLOC(uuid, uuid_size);
4848         if (name == NULL || uuid == NULL)
4849                 GOTO(cleanup_mem, rc = -ENOMEM);
4850
4851         OBD_ALLOC_PTR(bufs);
4852         if (!bufs)
4853                 GOTO(cleanup_mem, rc = -ENOMEM);
4854
4855         strcpy(name, dev);
4856         p = strstr(name, "-MDT");
4857         if (p == NULL)
4858                 GOTO(free_bufs, rc = -ENOMEM);
4859         p[3] = 'D';
4860
4861         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4862
4863         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4864         if (lprof == NULL || lprof->lp_dt == NULL) {
4865                 CERROR("can't find the profile: %s\n",
4866                        lustre_cfg_string(cfg, 0));
4867                 GOTO(free_bufs, rc = -EINVAL);
4868         }
4869
4870         lustre_cfg_bufs_reset(bufs, name);
4871         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4872         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4873         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4874
4875         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4876         if (!lcfg)
4877                 GOTO(put_profile, rc = -ENOMEM);
4878         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4879
4880         rc = class_attach(lcfg);
4881         if (rc)
4882                 GOTO(lcfg_cleanup, rc);
4883
4884         obd = class_name2obd(name);
4885         if (!obd) {
4886                 CERROR("Can not find obd %s (%s in config)\n",
4887                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4888                 GOTO(lcfg_cleanup, rc = -EINVAL);
4889         }
4890
4891         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4892
4893         lustre_cfg_bufs_reset(bufs, name);
4894         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4895         lustre_cfg_bufs_set_string(bufs, 2, dev);
4896         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4897
4898         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4899         if (!lcfg)
4900                 GOTO(class_detach, rc = -ENOMEM);
4901         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4902
4903         rc = class_setup(obd, lcfg);
4904         if (rc)
4905                 GOTO(class_detach, rc);
4906
4907         /* connect to MDD we just setup */
4908         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4909         if (rc)
4910                 GOTO(class_detach, rc);
4911
4912         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4913         LASSERT(site);
4914         LASSERT(mdt_lu_site(mdt) == NULL);
4915         mdt->mdt_lu_dev.ld_site = site;
4916         site->ls_top_dev = &mdt->mdt_lu_dev;
4917         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4918
4919         /* now connect to bottom OSD */
4920         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4921         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4922         if (rc)
4923                 GOTO(class_detach, rc);
4924         mdt->mdt_bottom =
4925                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4926
4927         rc = lu_env_refill((struct lu_env *)env);
4928         if (rc != 0)
4929                 CERROR("Failure to refill session: '%d'\n", rc);
4930
4931         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4932
4933         EXIT;
4934 class_detach:
4935         if (rc)
4936                 class_detach(obd, lcfg);
4937 lcfg_cleanup:
4938         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4939 put_profile:
4940         class_put_profile(lprof);
4941 free_bufs:
4942         OBD_FREE_PTR(bufs);
4943 cleanup_mem:
4944         if (name)
4945                 OBD_FREE(name, name_size);
4946         if (uuid)
4947                 OBD_FREE(uuid, uuid_size);
4948         RETURN(rc);
4949 }
4950
4951 /* setup quota master target on MDT0 */
4952 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4953                           struct lustre_cfg *cfg)
4954 {
4955         struct obd_device       *obd;
4956         char                    *dev = lustre_cfg_string(cfg, 0);
4957         char                    *qmtname, *uuid, *p;
4958         struct lustre_cfg_bufs  *bufs;
4959         struct lustre_cfg       *lcfg;
4960         struct lustre_profile   *lprof;
4961         struct obd_connect_data *data;
4962         int                      rc;
4963         ENTRY;
4964
4965         LASSERT(mdt->mdt_qmt_exp == NULL);
4966         LASSERT(mdt->mdt_qmt_dev == NULL);
4967
4968         /* quota master is on MDT0 only for now */
4969         if (mdt->mdt_seq_site.ss_node_id != 0)
4970                 RETURN(0);
4971
4972         /* MGS generates config commands which look as follows:
4973          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4974          *                    3:lustre-MDT0000-mdtlov  4:f
4975          *
4976          * We generate the QMT name from the MDT one, just replacing MD with QM
4977          * after all the preparations, the logical equivalent will be:
4978          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4979          *                    3:lustre-MDT0000-osd  4:f */
4980         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4981         OBD_ALLOC(uuid, UUID_MAX);
4982         OBD_ALLOC_PTR(bufs);
4983         OBD_ALLOC_PTR(data);
4984         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4985                 GOTO(cleanup_mem, rc = -ENOMEM);
4986
4987         strcpy(qmtname, dev);
4988         p = strstr(qmtname, "-MDT");
4989         if (p == NULL)
4990                 GOTO(cleanup_mem, rc = -ENOMEM);
4991         /* replace MD with QM */
4992         p[1] = 'Q';
4993         p[2] = 'M';
4994
4995         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4996
4997         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4998         if (lprof == NULL || lprof->lp_dt == NULL) {
4999                 CERROR("can't find profile for %s\n",
5000                        lustre_cfg_string(cfg, 0));
5001                 GOTO(cleanup_mem, rc = -EINVAL);
5002         }
5003
5004         lustre_cfg_bufs_reset(bufs, qmtname);
5005         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5006         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5007         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5008
5009         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5010         if (!lcfg)
5011                 GOTO(put_profile, rc = -ENOMEM);
5012         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5013
5014         rc = class_attach(lcfg);
5015         if (rc)
5016                 GOTO(lcfg_cleanup, rc);
5017
5018         obd = class_name2obd(qmtname);
5019         if (!obd) {
5020                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5021                        lustre_cfg_string(cfg, 0));
5022                 GOTO(lcfg_cleanup, rc = -EINVAL);
5023         }
5024
5025         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5026
5027         lustre_cfg_bufs_reset(bufs, qmtname);
5028         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5029         lustre_cfg_bufs_set_string(bufs, 2, dev);
5030
5031         /* for quota, the next device should be the OSD device */
5032         lustre_cfg_bufs_set_string(bufs, 3,
5033                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5034
5035         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5036         if (!lcfg)
5037                 GOTO(class_detach, rc = -ENOMEM);
5038         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5039
5040         rc = class_setup(obd, lcfg);
5041         if (rc)
5042                 GOTO(class_detach, rc);
5043
5044         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5045
5046         /* configure local quota objects */
5047         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5048                 rc = -EBADF;
5049         else
5050                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5051                                                            &mdt->mdt_lu_dev,
5052                                                            mdt->mdt_qmt_dev);
5053         if (rc)
5054                 GOTO(class_cleanup, rc);
5055
5056         /* connect to quota master target */
5057         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5058         data->ocd_version = LUSTRE_VERSION_CODE;
5059         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5060                          data, NULL);
5061         if (rc) {
5062                 CERROR("cannot connect to quota master device %s (%d)\n",
5063                        qmtname, rc);
5064                 GOTO(class_cleanup, rc);
5065         }
5066
5067         EXIT;
5068 class_cleanup:
5069         if (rc) {
5070                 class_manual_cleanup(obd);
5071                 mdt->mdt_qmt_dev = NULL;
5072                 GOTO(lcfg_cleanup, rc);
5073         }
5074 class_detach:
5075         if (rc)
5076                 class_detach(obd, lcfg);
5077 lcfg_cleanup:
5078         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5079 put_profile:
5080         class_put_profile(lprof);
5081 cleanup_mem:
5082         if (bufs)
5083                 OBD_FREE_PTR(bufs);
5084         if (qmtname)
5085                 OBD_FREE(qmtname, MAX_OBD_NAME);
5086         if (uuid)
5087                 OBD_FREE(uuid, UUID_MAX);
5088         if (data)
5089                 OBD_FREE_PTR(data);
5090         return rc;
5091 }
5092
5093 /* Shutdown quota master target associated with mdt */
5094 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5095 {
5096         ENTRY;
5097
5098         if (mdt->mdt_qmt_exp == NULL)
5099                 RETURN_EXIT;
5100         LASSERT(mdt->mdt_qmt_dev != NULL);
5101
5102         /* the qmt automatically shuts down when the mdt disconnects */
5103         obd_disconnect(mdt->mdt_qmt_exp);
5104         mdt->mdt_qmt_exp = NULL;
5105         mdt->mdt_qmt_dev = NULL;
5106         EXIT;
5107 }
5108
5109 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5110  * for now. This will be removed along with converting rest of MDT code
5111  * to use tgt_session_info */
5112 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5113 {
5114         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5115         int                      rc;
5116
5117         rc = mdt_getxattr(info);
5118
5119         mdt_thread_info_fini(info);
5120         return rc;
5121 }
5122
5123 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5124 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5125 #define OST_BRW_READ    OST_READ
5126 #define OST_BRW_WRITE   OST_WRITE
5127
5128 static struct tgt_handler mdt_tgt_handlers[] = {
5129 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5130                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5131                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5132 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5133                 0,                      MDS_DISCONNECT, tgt_disconnect,
5134                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5135 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5136                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5137                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
5138 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5139 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5140 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5141 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5142                                                         mdt_getattr_name),
5143 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5144 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5145 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5146 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5147 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5148 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5149 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5150 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5151                                                         mdt_hsm_progress),
5152 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5153                                                         mdt_hsm_ct_register),
5154 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5155                                                         mdt_hsm_ct_unregister),
5156 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5157                                                         mdt_hsm_state_get),
5158 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5159                                                         mdt_hsm_state_set),
5160 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5161 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5162                                                         mdt_hsm_request),
5163 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5164             MDS_SWAP_LAYOUTS,
5165             mdt_swap_layouts),
5166 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5167 };
5168
5169 static struct tgt_handler mdt_io_ops[] = {
5170 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5171                                                         mdt_hp_brw),
5172 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5173                                                         mdt_hp_brw),
5174 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5175                                          OST_PUNCH,     mdt_punch_hdl,
5176                                                         mdt_hp_punch),
5177 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5178 };
5179
5180 static struct tgt_handler mdt_sec_ctx_ops[] = {
5181 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5182 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5183 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5184 };
5185
5186 static struct tgt_handler mdt_quota_ops[] = {
5187 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5188 };
5189
5190 static struct tgt_opc_slice mdt_common_slice[] = {
5191         {
5192                 .tos_opc_start  = MDS_FIRST_OPC,
5193                 .tos_opc_end    = MDS_LAST_OPC,
5194                 .tos_hs         = mdt_tgt_handlers
5195         },
5196         {
5197                 .tos_opc_start  = OBD_FIRST_OPC,
5198                 .tos_opc_end    = OBD_LAST_OPC,
5199                 .tos_hs         = tgt_obd_handlers
5200         },
5201         {
5202                 .tos_opc_start  = LDLM_FIRST_OPC,
5203                 .tos_opc_end    = LDLM_LAST_OPC,
5204                 .tos_hs         = tgt_dlm_handlers
5205         },
5206         {
5207                 .tos_opc_start  = SEC_FIRST_OPC,
5208                 .tos_opc_end    = SEC_LAST_OPC,
5209                 .tos_hs         = mdt_sec_ctx_ops
5210         },
5211         {
5212                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5213                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5214                 .tos_hs         = tgt_out_handlers
5215         },
5216         {
5217                 .tos_opc_start  = FLD_FIRST_OPC,
5218                 .tos_opc_end    = FLD_LAST_OPC,
5219                 .tos_hs         = fld_handlers
5220         },
5221         {
5222                 .tos_opc_start  = SEQ_FIRST_OPC,
5223                 .tos_opc_end    = SEQ_LAST_OPC,
5224                 .tos_hs         = seq_handlers
5225         },
5226         {
5227                 .tos_opc_start  = QUOTA_DQACQ,
5228                 .tos_opc_end    = QUOTA_LAST_OPC,
5229                 .tos_hs         = mdt_quota_ops
5230         },
5231         {
5232                 .tos_opc_start  = LLOG_FIRST_OPC,
5233                 .tos_opc_end    = LLOG_LAST_OPC,
5234                 .tos_hs         = tgt_llog_handlers
5235         },
5236         {
5237                 .tos_opc_start  = LFSCK_FIRST_OPC,
5238                 .tos_opc_end    = LFSCK_LAST_OPC,
5239                 .tos_hs         = tgt_lfsck_handlers
5240         },
5241         {
5242                 .tos_opc_start  = OST_FIRST_OPC,
5243                 .tos_opc_end    = OST_LAST_OPC,
5244                 .tos_hs         = mdt_io_ops
5245         },
5246         {
5247                 .tos_hs         = NULL
5248         }
5249 };
5250
5251 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5252 {
5253         struct md_device *next = m->mdt_child;
5254         struct lu_device *d = &m->mdt_lu_dev;
5255         struct obd_device *obd = mdt2obd_dev(m);
5256         struct lfsck_stop stop;
5257
5258         ENTRY;
5259         stop.ls_status = LS_PAUSED;
5260         stop.ls_flags = 0;
5261         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5262
5263         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5264         ping_evictor_stop();
5265
5266         /* Remove the HSM /proc entry so the coordinator cannot be
5267          * restarted by a user while it's shutting down.
5268          */
5269         mdt_hsm_cdt_stop(m);
5270
5271         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5272         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5273
5274         if (m->mdt_namespace != NULL)
5275                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5276                                           d->ld_obd->obd_force);
5277
5278         obd_exports_barrier(obd);
5279         obd_zombie_barrier();
5280
5281         mdt_quota_fini(env, m);
5282
5283         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5284
5285         /* Calling the cleanup functions in the same order as in the mdt_init0
5286          * error path
5287          */
5288         mdt_tunables_fini(m);
5289
5290         target_recovery_fini(obd);
5291         upcall_cache_cleanup(m->mdt_identity_cache);
5292         m->mdt_identity_cache = NULL;
5293
5294         mdt_fs_cleanup(env, m);
5295
5296         tgt_fini(env, &m->mdt_lut);
5297
5298         mdt_hsm_cdt_fini(m);
5299
5300         if (m->mdt_los != NULL) {
5301                 local_oid_storage_fini(env, m->mdt_los);
5302                 m->mdt_los = NULL;
5303         }
5304
5305         if (m->mdt_namespace != NULL) {
5306                 ldlm_namespace_free_post(m->mdt_namespace);
5307                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5308         }
5309
5310         if (m->mdt_md_root != NULL) {
5311                 mdt_object_put(env, m->mdt_md_root);
5312                 m->mdt_md_root = NULL;
5313         }
5314
5315         mdt_seq_fini(env, m);
5316
5317         mdt_fld_fini(env, m);
5318
5319         /*
5320          * Finish the stack
5321          */
5322         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5323
5324         LASSERT(atomic_read(&d->ld_ref) == 0);
5325
5326         server_put_mount(mdt_obd_name(m), true);
5327
5328         EXIT;
5329 }
5330
5331 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5332
5333 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5334                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5335 {
5336         const struct dt_device_param *dt_conf;
5337         struct mdt_thread_info *info;
5338         struct obd_device *obd;
5339         const char *dev = lustre_cfg_string(cfg, 0);
5340         const char *num = lustre_cfg_string(cfg, 2);
5341         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5342         struct lustre_mount_info *lmi = NULL;
5343         struct lustre_sb_info *lsi;
5344         struct lu_site *s;
5345         struct seq_server_site *ss_site;
5346         const char *identity_upcall = "NONE";
5347         struct md_device *next;
5348         struct lu_fid fid;
5349         int rc;
5350         long node_id;
5351         mntopt_t mntopts;
5352         ENTRY;
5353
5354         lu_device_init(&m->mdt_lu_dev, ldt);
5355         /*
5356          * Environment (env) might be missing mdt_thread_key values at that
5357          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5358          * mode.
5359          *
5360          * Usually device allocation path doesn't use module key values, but
5361          * mdt has to do a lot of work here, so allocate key value.
5362          */
5363         rc = lu_env_refill((struct lu_env *)env);
5364         if (rc != 0)
5365                 RETURN(rc);
5366
5367         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5368         LASSERT(info != NULL);
5369
5370         obd = class_name2obd(dev);
5371         LASSERT(obd != NULL);
5372
5373         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5374         m->mdt_opts.mo_evict_tgt_nids = 1;
5375         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5376
5377         lmi = server_get_mount(dev);
5378         if (lmi == NULL) {
5379                 CERROR("Cannot get mount info for %s!\n", dev);
5380                 RETURN(-EFAULT);
5381         } else {
5382                 lsi = s2lsi(lmi->lmi_sb);
5383                 /* CMD is supported only in IAM mode */
5384                 LASSERT(num);
5385                 node_id = simple_strtol(num, NULL, 10);
5386                 obd->u.obt.obt_magic = OBT_MAGIC;
5387                 if (lsi->lsi_lmd != NULL &&
5388                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5389                         m->mdt_skip_lfsck = 1;
5390         }
5391
5392         /* DoM files get IO lock at open by default */
5393         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5394         /* DoM files are read at open and data is packed in the reply */
5395         m->mdt_opts.mo_dom_read_open = 1;
5396
5397         m->mdt_squash.rsi_uid = 0;
5398         m->mdt_squash.rsi_gid = 0;
5399         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5400         spin_lock_init(&m->mdt_squash.rsi_lock);
5401         spin_lock_init(&m->mdt_lock);
5402         m->mdt_enable_remote_dir = 1;
5403         m->mdt_enable_striped_dir = 1;
5404         m->mdt_enable_dir_migration = 1;
5405         m->mdt_enable_remote_dir_gid = 0;
5406         m->mdt_enable_remote_rename = 1;
5407
5408         atomic_set(&m->mdt_mds_mds_conns, 0);
5409         atomic_set(&m->mdt_async_commit_count, 0);
5410
5411         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5412         m->mdt_lu_dev.ld_obd = obd;
5413         /* Set this lu_device to obd for error handling purposes. */
5414         obd->obd_lu_dev = &m->mdt_lu_dev;
5415
5416         /* init the stack */
5417         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5418         if (rc) {
5419                 CERROR("%s: Can't init device stack, rc %d\n",
5420                        mdt_obd_name(m), rc);
5421                 GOTO(err_lmi, rc);
5422         }
5423
5424         s = mdt_lu_site(m);
5425         ss_site = mdt_seq_site(m);
5426         s->ld_seq_site = ss_site;
5427         ss_site->ss_lu = s;
5428
5429         /* set server index */
5430         ss_site->ss_node_id = node_id;
5431
5432         /* failover is the default
5433          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5434          * assumed whose ss_node_id == 0 XXX
5435          * */
5436         obd->obd_replayable = 1;
5437         /* No connection accepted until configurations will finish */
5438         obd->obd_no_conn = 1;
5439
5440         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5441                 char *str = lustre_cfg_string(cfg, 4);
5442                 if (strchr(str, 'n')) {
5443                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5444                         obd->obd_replayable = 0;
5445                 }
5446         }
5447
5448         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5449         if (rc)
5450                 GOTO(err_fini_stack, rc);
5451
5452         rc = mdt_seq_init(env, m);
5453         if (rc)
5454                 GOTO(err_fini_fld, rc);
5455
5456         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5457                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5458         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5459                                               LDLM_NAMESPACE_SERVER,
5460                                               LDLM_NAMESPACE_GREEDY,
5461                                               LDLM_NS_TYPE_MDT);
5462         if (m->mdt_namespace == NULL)
5463                 GOTO(err_fini_seq, rc = -ENOMEM);
5464
5465         m->mdt_namespace->ns_lvbp = m;
5466         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5467
5468         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5469         /* set obd_namespace for compatibility with old code */
5470         obd->obd_namespace = m->mdt_namespace;
5471
5472         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5473                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5474                       OBD_FAIL_MDS_ALL_REPLY_NET);
5475         if (rc)
5476                 GOTO(err_free_ns, rc);
5477
5478         /* Amount of available space excluded from granting and reserved
5479          * for metadata. It is in percentage and 50% is default value. */
5480         tgd->tgd_reserved_pcnt = 50;
5481
5482         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5483                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5484         else
5485                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5486
5487         rc = mdt_fs_setup(env, m, obd, lsi);
5488         if (rc)
5489                 GOTO(err_tgt, rc);
5490
5491         fid.f_seq = FID_SEQ_LOCAL_NAME;
5492         fid.f_oid = 1;
5493         fid.f_ver = 0;
5494         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5495         if (rc != 0)
5496                 GOTO(err_fs_cleanup, rc);
5497
5498         rc = mdt_hsm_cdt_init(m);
5499         if (rc != 0) {
5500                 CERROR("%s: error initializing coordinator, rc %d\n",
5501                        mdt_obd_name(m), rc);
5502                 GOTO(err_los_fini, rc);
5503         }
5504
5505         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5506
5507         next = m->mdt_child;
5508         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5509
5510         mntopts = dt_conf->ddp_mntopts;
5511
5512         if (mntopts & MNTOPT_USERXATTR)
5513                 m->mdt_opts.mo_user_xattr = 1;
5514         else
5515                 m->mdt_opts.mo_user_xattr = 0;
5516
5517         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5518
5519         if (mntopts & MNTOPT_ACL)
5520                 m->mdt_opts.mo_acl = 1;
5521         else
5522                 m->mdt_opts.mo_acl = 0;
5523
5524         /* XXX: to support suppgid for ACL, we enable identity_upcall
5525          * by default, otherwise, maybe got unexpected -EACCESS. */
5526         if (m->mdt_opts.mo_acl)
5527                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5528
5529         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5530                                                 identity_upcall,
5531                                                 &mdt_identity_upcall_cache_ops);
5532         if (IS_ERR(m->mdt_identity_cache)) {
5533                 rc = PTR_ERR(m->mdt_identity_cache);
5534                 m->mdt_identity_cache = NULL;
5535                 GOTO(err_free_hsm, rc);
5536         }
5537
5538         rc = mdt_tunables_init(m, dev);
5539         if (rc) {
5540                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5541                 GOTO(err_recovery, rc);
5542         }
5543
5544         rc = mdt_quota_init(env, m, cfg);
5545         if (rc)
5546                 GOTO(err_procfs, rc);
5547
5548         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5549         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5550                            "mdt_ldlm_client", m->mdt_ldlm_client);
5551
5552         ping_evictor_start();
5553
5554         /* recovery will be started upon mdt_prepare()
5555          * when the whole stack is complete and ready
5556          * to serve the requests */
5557
5558         /* Reduce the initial timeout on an MDS because it doesn't need such
5559          * a long timeout as an OST does. Adaptive timeouts will adjust this
5560          * value appropriately. */
5561         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5562                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5563
5564         RETURN(0);
5565 err_procfs:
5566         mdt_tunables_fini(m);
5567 err_recovery:
5568         upcall_cache_cleanup(m->mdt_identity_cache);
5569         m->mdt_identity_cache = NULL;
5570 err_free_hsm:
5571         mdt_hsm_cdt_fini(m);
5572 err_los_fini:
5573         local_oid_storage_fini(env, m->mdt_los);
5574         m->mdt_los = NULL;
5575 err_fs_cleanup:
5576         mdt_fs_cleanup(env, m);
5577 err_tgt:
5578         /* keep recoverable clients */
5579         obd->obd_fail = 1;
5580         target_recovery_fini(obd);
5581         obd_exports_barrier(obd);
5582         obd_zombie_barrier();
5583         tgt_fini(env, &m->mdt_lut);
5584 err_free_ns:
5585         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5586         obd->obd_namespace = m->mdt_namespace = NULL;
5587 err_fini_seq:
5588         mdt_seq_fini(env, m);
5589 err_fini_fld:
5590         mdt_fld_fini(env, m);
5591 err_fini_stack:
5592         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5593 err_lmi:
5594         if (lmi)
5595                 server_put_mount(dev, true);
5596         return(rc);
5597 }
5598
5599 /* For interoperability, the left element is old parameter, the right one
5600  * is the new version of the parameter, if some parameter is deprecated,
5601  * the new version should be set as NULL. */
5602 static struct cfg_interop_param mdt_interop_param[] = {
5603         { "mdt.group_upcall",   NULL },
5604         { "mdt.quota_type",     NULL },
5605         { "mdd.quota_type",     NULL },
5606         { "mdt.som",            NULL },
5607         { "mdt.rootsquash",     "mdt.root_squash" },
5608         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5609         { NULL }
5610 };
5611
5612 /* used by MGS to process specific configurations */
5613 static int mdt_process_config(const struct lu_env *env,
5614                               struct lu_device *d, struct lustre_cfg *cfg)
5615 {
5616         struct mdt_device *m = mdt_dev(d);
5617         struct md_device *md_next = m->mdt_child;
5618         struct lu_device *next = md2lu_dev(md_next);
5619         int rc;
5620         ENTRY;
5621
5622         switch (cfg->lcfg_command) {
5623         case LCFG_PARAM: {
5624                 struct obd_device *obd = d->ld_obd;
5625                 /* For interoperability */
5626                 struct cfg_interop_param *ptr = NULL;
5627                 struct lustre_cfg *old_cfg = NULL;
5628                 char *param = NULL;
5629                 ssize_t count;
5630
5631                 param = lustre_cfg_string(cfg, 1);
5632                 if (param == NULL) {
5633                         CERROR("param is empty\n");
5634                         rc = -EINVAL;
5635                         break;
5636                 }
5637
5638                 ptr = class_find_old_param(param, mdt_interop_param);
5639                 if (ptr != NULL) {
5640                         if (ptr->new_param == NULL) {
5641                                 rc = 0;
5642                                 CWARN("For interoperability, skip this %s."
5643                                       " It is obsolete.\n", ptr->old_param);
5644                                 break;
5645                         }
5646
5647                         CWARN("Found old param %s, changed it to %s.\n",
5648                               ptr->old_param, ptr->new_param);
5649
5650                         old_cfg = cfg;
5651                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5652                         if (IS_ERR(cfg)) {
5653                                 rc = PTR_ERR(cfg);
5654                                 break;
5655                         }
5656                 }
5657
5658                 count = class_modify_config(cfg, PARAM_MDT,
5659                                             &obd->obd_kset.kobj);
5660                 if (count < 0) {
5661                         struct coordinator *cdt = &m->mdt_coordinator;
5662
5663                         /* is it an HSM var ? */
5664                         count = class_modify_config(cfg, PARAM_HSM,
5665                                                     &cdt->cdt_hsm_kobj);
5666                         if (count < 0)
5667                                 /* we don't understand; pass it on */
5668                                 rc = next->ld_ops->ldo_process_config(env, next,
5669                                                                       cfg);
5670                         else
5671                                 rc = count > 0 ? 0 : count;
5672                 } else {
5673                         rc = count > 0 ? 0 : count;
5674                 }
5675
5676                 if (old_cfg)
5677                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5678                                                      cfg->lcfg_buflens));
5679                 break;
5680         }
5681         default:
5682                 /* others are passed further */
5683                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5684                 break;
5685         }
5686         RETURN(rc);
5687 }
5688
5689 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5690                                           const struct lu_object_header *hdr,
5691                                           struct lu_device *d)
5692 {
5693         struct mdt_object *mo;
5694
5695         ENTRY;
5696
5697         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5698         if (mo != NULL) {
5699                 struct lu_object *o;
5700                 struct lu_object_header *h;
5701
5702                 o = &mo->mot_obj;
5703                 h = &mo->mot_header;
5704                 lu_object_header_init(h);
5705                 lu_object_init(o, h, d);
5706                 lu_object_add_top(h, o);
5707                 o->lo_ops = &mdt_obj_ops;
5708                 spin_lock_init(&mo->mot_write_lock);
5709                 mutex_init(&mo->mot_som_mutex);
5710                 mutex_init(&mo->mot_lov_mutex);
5711                 init_rwsem(&mo->mot_dom_sem);
5712                 init_rwsem(&mo->mot_open_sem);
5713                 atomic_set(&mo->mot_open_count, 0);
5714                 RETURN(o);
5715         }
5716         RETURN(NULL);
5717 }
5718
5719 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5720                            const struct lu_object_conf *unused)
5721 {
5722         struct mdt_device *d = mdt_dev(o->lo_dev);
5723         struct lu_device  *under;
5724         struct lu_object  *below;
5725         int                rc = 0;
5726         ENTRY;
5727
5728         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5729                PFID(lu_object_fid(o)));
5730
5731         under = &d->mdt_child->md_lu_dev;
5732         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5733         if (below != NULL) {
5734                 lu_object_add(o, below);
5735         } else
5736                 rc = -ENOMEM;
5737
5738         RETURN(rc);
5739 }
5740
5741 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5742 {
5743         struct mdt_object *mo = mdt_obj(o);
5744         struct lu_object_header *h;
5745         ENTRY;
5746
5747         h = o->lo_header;
5748         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5749                PFID(lu_object_fid(o)));
5750
5751         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5752         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5753
5754         lu_object_fini(o);
5755         lu_object_header_fini(h);
5756         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5757
5758         EXIT;
5759 }
5760
5761 static int mdt_object_print(const struct lu_env *env, void *cookie,
5762                             lu_printer_t p, const struct lu_object *o)
5763 {
5764         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5765
5766         return (*p)(env, cookie,
5767                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5768                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5769                     mdto->mot_cache_attr ? "cache_attr" : "",
5770                     mdto->mot_write_count);
5771 }
5772
5773 static int mdt_prepare(const struct lu_env *env,
5774                 struct lu_device *pdev,
5775                 struct lu_device *cdev)
5776 {
5777         struct mdt_device *mdt = mdt_dev(cdev);
5778         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5779         struct obd_device *obd = cdev->ld_obd;
5780         int rc;
5781
5782         ENTRY;
5783
5784         LASSERT(obd);
5785
5786         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5787         if (rc)
5788                 RETURN(rc);
5789
5790         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5791         if (rc)
5792                 RETURN(rc);
5793
5794         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5795         if (rc)
5796                 RETURN(rc);
5797
5798         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5799         /* The LFSCK instance is registered just now, so it must be there when
5800          * register the namespace to such instance. */
5801         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5802
5803         if (mdt->mdt_seq_site.ss_node_id == 0) {
5804                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5805                                                          &mdt->mdt_md_root_fid);
5806                 if (rc)
5807                         RETURN(rc);
5808         }
5809
5810         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5811
5812         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5813         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5814         LASSERT(obd->obd_no_conn);
5815         spin_lock(&obd->obd_dev_lock);
5816         obd->obd_no_conn = 0;
5817         spin_unlock(&obd->obd_dev_lock);
5818
5819         if (obd->obd_recovering == 0)
5820                 mdt_postrecov(env, mdt);
5821
5822         RETURN(rc);
5823 }
5824
5825 const struct lu_device_operations mdt_lu_ops = {
5826         .ldo_object_alloc   = mdt_object_alloc,
5827         .ldo_process_config = mdt_process_config,
5828         .ldo_prepare        = mdt_prepare,
5829 };
5830
5831 static const struct lu_object_operations mdt_obj_ops = {
5832         .loo_object_init    = mdt_object_init,
5833         .loo_object_free    = mdt_object_free,
5834         .loo_object_print   = mdt_object_print
5835 };
5836
5837 static int mdt_obd_set_info_async(const struct lu_env *env,
5838                                   struct obd_export *exp,
5839                                   __u32 keylen, void *key,
5840                                   __u32 vallen, void *val,
5841                                   struct ptlrpc_request_set *set)
5842 {
5843         int rc;
5844
5845         ENTRY;
5846
5847         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5848                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5849                 RETURN(rc);
5850         }
5851
5852         RETURN(0);
5853 }
5854
5855 /**
5856  * Match client and server connection feature flags.
5857  *
5858  * Compute the compatibility flags for a connection request based on
5859  * features mutually supported by client and server.
5860  *
5861  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5862  * must not be updated here, otherwise a partially initialized value may
5863  * be exposed. After the connection request is successfully processed,
5864  * the top-level MDT connect request handler atomically updates the export
5865  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5866  * reply. \see mdt_connect().
5867  *
5868  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5869  * full struct obd_connect_data. So care must be taken when accessing fields
5870  * that are not present in struct obd_connect_data_v1. See LU-16.
5871  *
5872  * \param exp   the obd_export associated with this client/target pair
5873  * \param mdt   the target device for the connection
5874  * \param data  stores data for this connect request
5875  *
5876  * \retval 0       success
5877  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5878  * \retval -EBADE  client and server feature requirements are incompatible
5879  */
5880 static int mdt_connect_internal(const struct lu_env *env,
5881                                 struct obd_export *exp,
5882                                 struct mdt_device *mdt,
5883                                 struct obd_connect_data *data, bool reconnect)
5884 {
5885         const char *obd_name = mdt_obd_name(mdt);
5886         LASSERT(data != NULL);
5887
5888         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5889
5890         if (mdt->mdt_bottom->dd_rdonly &&
5891             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5892             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5893                 RETURN(-EACCES);
5894
5895         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5896                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5897
5898         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5899
5900         if (!mdt->mdt_opts.mo_acl)
5901                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5902
5903         if (!mdt->mdt_opts.mo_user_xattr)
5904                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5905
5906         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5907                 data->ocd_brw_size = min(data->ocd_brw_size,
5908                                          mdt->mdt_brw_size);
5909                 if (data->ocd_brw_size == 0) {
5910                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5911                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5912                                "ocd_brw_size unexpectedly zero, network data "
5913                                "corruption? Refusing to connect this client\n",
5914                                obd_name, exp->exp_client_uuid.uuid,
5915                                exp, data->ocd_connect_flags, data->ocd_version,
5916                                data->ocd_grant, data->ocd_index);
5917                         return -EPROTO;
5918                 }
5919         }
5920
5921         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5922                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5923
5924                 /* client is reporting its page size, for future use */
5925                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5926                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5927                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5928                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5929                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5930                 /* ocd_grant_tax_kb is in 1K byte blocks */
5931                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5932                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5933         }
5934
5935         /* Save connect_data we have so far because tgt_grant_connect()
5936          * uses it to calculate grant, and we want to save the client
5937          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5938         exp->exp_connect_data = *data;
5939         if (OCD_HAS_FLAG(data, GRANT))
5940                 tgt_grant_connect(env, exp, data, !reconnect);
5941
5942         if (OCD_HAS_FLAG(data, MAXBYTES))
5943                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5944
5945         /* NB: Disregard the rule against updating
5946          * exp_connect_data.ocd_connect_flags in this case, since
5947          * tgt_client_new() needs to know if this is a lightweight
5948          * connection, and it is safe to expose this flag before
5949          * connection processing completes. */
5950         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5951                 spin_lock(&exp->exp_lock);
5952                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5953                 spin_unlock(&exp->exp_lock);
5954         }
5955
5956         data->ocd_version = LUSTRE_VERSION_CODE;
5957
5958         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5959                 CWARN("%s: MDS requires FID support, but client not\n",
5960                       obd_name);
5961                 return -EBADE;
5962         }
5963
5964         if (OCD_HAS_FLAG(data, PINGLESS)) {
5965                 if (ptlrpc_pinger_suppress_pings()) {
5966                         spin_lock(&exp->exp_obd->obd_dev_lock);
5967                         list_del_init(&exp->exp_obd_chain_timed);
5968                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5969                 } else {
5970                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5971                 }
5972         }
5973
5974         data->ocd_max_easize = mdt->mdt_max_ea_size;
5975
5976         /* NB: Disregard the rule against updating
5977          * exp_connect_data.ocd_connect_flags in this case, since
5978          * tgt_client_new() needs to know if this is client supports
5979          * multiple modify RPCs, and it is safe to expose this flag before
5980          * connection processing completes. */
5981         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5982                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5983                 spin_lock(&exp->exp_lock);
5984                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5985                 spin_unlock(&exp->exp_lock);
5986         }
5987
5988         if (OCD_HAS_FLAG(data, CKSUM)) {
5989                 __u32 cksum_types = data->ocd_cksum_types;
5990
5991                 /* The client set in ocd_cksum_types the checksum types it
5992                  * supports. We have to mask off the algorithms that we don't
5993                  * support */
5994                 data->ocd_cksum_types &=
5995                         obd_cksum_types_supported_server(obd_name);
5996
5997                 if (unlikely(data->ocd_cksum_types == 0)) {
5998                         CERROR("%s: Connect with checksum support but no "
5999                                "ocd_cksum_types is set\n",
6000                                exp->exp_obd->obd_name);
6001                         RETURN(-EPROTO);
6002                 }
6003
6004                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6005                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6006                        cksum_types, data->ocd_cksum_types);
6007         } else {
6008                 /* This client does not support OBD_CONNECT_CKSUM
6009                  * fall back to CRC32 */
6010                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6011                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6012                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6013         }
6014
6015         return 0;
6016 }
6017
6018 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6019                                    struct mdt_thread_info *info,
6020                                    struct mdt_file_data *mfd)
6021 {
6022         struct lu_context ses;
6023         int rc;
6024         ENTRY;
6025
6026         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6027         if (rc)
6028                 RETURN(rc);
6029
6030         env->le_ses = &ses;
6031         lu_context_enter(&ses);
6032
6033         mdt_ucred(info)->uc_valid = UCRED_OLD;
6034         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6035
6036         lu_context_exit(&ses);
6037         lu_context_fini(&ses);
6038         env->le_ses = NULL;
6039
6040         RETURN(rc);
6041 }
6042
6043 static int mdt_export_cleanup(struct obd_export *exp)
6044 {
6045         struct list_head         closing_list;
6046         struct mdt_export_data  *med = &exp->exp_mdt_data;
6047         struct obd_device       *obd = exp->exp_obd;
6048         struct mdt_device       *mdt;
6049         struct mdt_thread_info  *info;
6050         struct lu_env            env;
6051         struct mdt_file_data    *mfd, *n;
6052         int rc = 0;
6053         ENTRY;
6054
6055         INIT_LIST_HEAD(&closing_list);
6056         spin_lock(&med->med_open_lock);
6057         while (!list_empty(&med->med_open_head)) {
6058                 struct list_head *tmp = med->med_open_head.next;
6059                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6060
6061                 /* Remove mfd handle so it can't be found again.
6062                  * We are consuming the mfd_list reference here. */
6063                 class_handle_unhash(&mfd->mfd_open_handle);
6064                 list_move_tail(&mfd->mfd_list, &closing_list);
6065         }
6066         spin_unlock(&med->med_open_lock);
6067         mdt = mdt_dev(obd->obd_lu_dev);
6068         LASSERT(mdt != NULL);
6069
6070         rc = lu_env_init(&env, LCT_MD_THREAD);
6071         if (rc)
6072                 RETURN(rc);
6073
6074         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6075         LASSERT(info != NULL);
6076         memset(info, 0, sizeof *info);
6077         info->mti_env = &env;
6078         info->mti_mdt = mdt;
6079         info->mti_exp = exp;
6080
6081         if (!list_empty(&closing_list)) {
6082                 struct md_attr *ma = &info->mti_attr;
6083
6084                 /* Close any open files (which may also cause orphan
6085                  * unlinking). */
6086                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6087                         list_del_init(&mfd->mfd_list);
6088                         ma->ma_need = ma->ma_valid = 0;
6089
6090                         /* This file is being closed due to an eviction, it
6091                          * could have been modified and now dirty regarding to
6092                          * HSM archive, check this!
6093                          * The logic here is to mark a file dirty if there's a
6094                          * chance it was dirtied before the client was evicted,
6095                          * so that we don't have to wait for a release attempt
6096                          * before finding out the file was actually dirty and
6097                          * fail the release. Aggressively marking it dirty here
6098                          * will cause the policy engine to attempt to
6099                          * re-archive it; when rearchiving, we can compare the
6100                          * current version to the HSM data_version and make the
6101                          * archive request into a noop if it's not actually
6102                          * dirty.
6103                          */
6104                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6105                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6106
6107                         /* Don't unlink orphan on failover umount, LU-184 */
6108                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6109                                 ma->ma_valid = MA_FLAGS;
6110                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6111                         }
6112                         mdt_mfd_close(info, mfd);
6113                 }
6114         }
6115         info->mti_mdt = NULL;
6116         /* cleanup client slot early */
6117         /* Do not erase record for recoverable client. */
6118         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6119                 tgt_client_del(&env, exp);
6120         lu_env_fini(&env);
6121
6122         RETURN(rc);
6123 }
6124
6125 static inline void mdt_enable_slc(struct mdt_device *mdt)
6126 {
6127         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
6128                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
6129 }
6130
6131 static inline void mdt_disable_slc(struct mdt_device *mdt)
6132 {
6133         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
6134                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
6135 }
6136
6137 static int mdt_obd_disconnect(struct obd_export *exp)
6138 {
6139         int rc;
6140
6141         ENTRY;
6142
6143         LASSERT(exp);
6144         class_export_get(exp);
6145
6146         if (!(exp->exp_flags & OBD_OPT_FORCE))
6147                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6148
6149         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6150             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6151                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6152
6153                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6154                         mdt_disable_slc(mdt);
6155         }
6156
6157         rc = server_disconnect_export(exp);
6158         if (rc != 0)
6159                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6160
6161         tgt_grant_discard(exp);
6162
6163         rc = mdt_export_cleanup(exp);
6164         nodemap_del_member(exp);
6165         class_export_put(exp);
6166         RETURN(rc);
6167 }
6168
6169 /* mds_connect copy */
6170 static int mdt_obd_connect(const struct lu_env *env,
6171                            struct obd_export **exp, struct obd_device *obd,
6172                            struct obd_uuid *cluuid,
6173                            struct obd_connect_data *data,
6174                            void *localdata)
6175 {
6176         struct obd_export       *lexp;
6177         struct lustre_handle    conn = { 0 };
6178         struct mdt_device       *mdt;
6179         int                      rc;
6180         lnet_nid_t              *client_nid = localdata;
6181         ENTRY;
6182
6183         LASSERT(env != NULL);
6184         LASSERT(data != NULL);
6185
6186         if (!exp || !obd || !cluuid)
6187                 RETURN(-EINVAL);
6188
6189         mdt = mdt_dev(obd->obd_lu_dev);
6190
6191         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6192             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6193                 atomic_inc(&mdt->mdt_mds_mds_conns);
6194                 mdt_enable_slc(mdt);
6195         }
6196
6197         /*
6198          * first, check whether the stack is ready to handle requests
6199          * XXX: probably not very appropriate method is used now
6200          *      at some point we should find a better one
6201          */
6202         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6203             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6204             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6205                 rc = obd_get_info(env, mdt->mdt_child_exp,
6206                                   sizeof(KEY_OSP_CONNECTED),
6207                                   KEY_OSP_CONNECTED, NULL, NULL);
6208                 if (rc)
6209                         RETURN(-EAGAIN);
6210                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6211         }
6212
6213         rc = class_connect(&conn, obd, cluuid);
6214         if (rc)
6215                 RETURN(rc);
6216
6217         lexp = class_conn2export(&conn);
6218         LASSERT(lexp != NULL);
6219
6220         rc = nodemap_add_member(*client_nid, lexp);
6221         if (rc != 0 && rc != -EEXIST)
6222                 GOTO(out, rc);
6223
6224         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6225         if (rc == 0) {
6226                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6227
6228                 LASSERT(lcd);
6229                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6230                 rc = tgt_client_new(env, lexp);
6231                 if (rc == 0)
6232                         mdt_export_stats_init(obd, lexp, localdata);
6233         }
6234 out:
6235         if (rc != 0) {
6236                 class_disconnect(lexp);
6237                 nodemap_del_member(lexp);
6238                 *exp = NULL;
6239         } else {
6240                 *exp = lexp;
6241                 /* Because we do not want this export to be evicted by pinger,
6242                  * let's not add this export to the timed chain list. */
6243                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6244                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6245                         list_del_init(&lexp->exp_obd_chain_timed);
6246                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6247                 }
6248         }
6249
6250         RETURN(rc);
6251 }
6252
6253 static int mdt_obd_reconnect(const struct lu_env *env,
6254                              struct obd_export *exp, struct obd_device *obd,
6255                              struct obd_uuid *cluuid,
6256                              struct obd_connect_data *data,
6257                              void *localdata)
6258 {
6259         lnet_nid_t             *client_nid = localdata;
6260         int                     rc;
6261         ENTRY;
6262
6263         if (exp == NULL || obd == NULL || cluuid == NULL)
6264                 RETURN(-EINVAL);
6265
6266         rc = nodemap_add_member(*client_nid, exp);
6267         if (rc != 0 && rc != -EEXIST)
6268                 RETURN(rc);
6269
6270         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6271                                   true);
6272         if (rc == 0)
6273                 mdt_export_stats_init(obd, exp, localdata);
6274         else
6275                 nodemap_del_member(exp);
6276
6277         RETURN(rc);
6278 }
6279
6280 /* FIXME: Can we avoid using these two interfaces? */
6281 static int mdt_init_export(struct obd_export *exp)
6282 {
6283         struct mdt_export_data *med = &exp->exp_mdt_data;
6284         int                     rc;
6285         ENTRY;
6286
6287         INIT_LIST_HEAD(&med->med_open_head);
6288         spin_lock_init(&med->med_open_lock);
6289         spin_lock(&exp->exp_lock);
6290         exp->exp_connecting = 1;
6291         spin_unlock(&exp->exp_lock);
6292
6293         /* self-export doesn't need client data and ldlm initialization */
6294         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6295                                      &exp->exp_client_uuid)))
6296                 RETURN(0);
6297
6298         rc = tgt_client_alloc(exp);
6299         if (rc)
6300                 GOTO(err, rc);
6301
6302         rc = ldlm_init_export(exp);
6303         if (rc)
6304                 GOTO(err_free, rc);
6305
6306         RETURN(rc);
6307
6308 err_free:
6309         tgt_client_free(exp);
6310 err:
6311         CERROR("%s: Failed to initialize export: rc = %d\n",
6312                exp->exp_obd->obd_name, rc);
6313         return rc;
6314 }
6315
6316 static int mdt_destroy_export(struct obd_export *exp)
6317 {
6318         ENTRY;
6319
6320         target_destroy_export(exp);
6321         /* destroy can be called from failed obd_setup, so
6322          * checking uuid is safer than obd_self_export */
6323         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6324                                      &exp->exp_client_uuid)))
6325                 RETURN(0);
6326
6327         ldlm_destroy_export(exp);
6328         tgt_client_free(exp);
6329
6330         LASSERT(list_empty(&exp->exp_outstanding_replies));
6331         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6332
6333         /*
6334          * discard grants once we're sure no more
6335          * interaction with the client is possible
6336          */
6337         tgt_grant_discard(exp);
6338         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6339                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6340
6341         if (!(exp->exp_flags & OBD_OPT_FORCE))
6342                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6343
6344         RETURN(0);
6345 }
6346
6347 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6348                    struct linkea_data *ldata)
6349 {
6350         int rc;
6351
6352         LASSERT(ldata->ld_buf->lb_buf != NULL);
6353
6354         if (!mdt_object_exists(mdt_obj))
6355                 return -ENODATA;
6356
6357         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6358                           ldata->ld_buf, XATTR_NAME_LINK);
6359         if (rc == -ERANGE) {
6360                 /* Buf was too small, figure out what we need. */
6361                 lu_buf_free(ldata->ld_buf);
6362                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6363                                   ldata->ld_buf, XATTR_NAME_LINK);
6364                 if (rc < 0)
6365                         return rc;
6366                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6367                 if (ldata->ld_buf->lb_buf == NULL)
6368                         return -ENOMEM;
6369                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6370                                   ldata->ld_buf, XATTR_NAME_LINK);
6371         }
6372         if (rc < 0)
6373                 return rc;
6374
6375         return linkea_init_with_rec(ldata);
6376 }
6377
6378 /**
6379  * Given an MDT object, try to look up the full path to the object.
6380  * Part of the MDT layer implementation of lfs fid2path.
6381  *
6382  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6383  * \param[in]     obj   Object to do path lookup of
6384  * \param[in,out] fp    User-provided struct to store path information
6385  * \param[in]     root_fid Root FID of current path should reach
6386  *
6387  * \retval 0 Lookup successful, path information stored in fp
6388  * \retval -EAGAIN Lookup failed, usually because object is being moved
6389  * \retval negative errno if there was a problem
6390  */
6391 static int mdt_path_current(struct mdt_thread_info *info,
6392                             struct mdt_object *obj,
6393                             struct getinfo_fid2path *fp,
6394                             struct lu_fid *root_fid)
6395 {
6396         struct mdt_device       *mdt = info->mti_mdt;
6397         struct mdt_object       *mdt_obj;
6398         struct link_ea_header   *leh;
6399         struct link_ea_entry    *lee;
6400         struct lu_name          *tmpname = &info->mti_name;
6401         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6402         struct lu_buf           *buf = &info->mti_big_buf;
6403         char                    *ptr;
6404         int                     reclen;
6405         struct linkea_data      ldata = { NULL };
6406         int                     rc = 0;
6407         bool                    first = true;
6408         ENTRY;
6409
6410         /* temp buffer for path element, the buffer will be finally freed
6411          * in mdt_thread_info_fini */
6412         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6413         if (buf->lb_buf == NULL)
6414                 RETURN(-ENOMEM);
6415
6416         ldata.ld_buf = buf;
6417         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6418         *ptr = 0;
6419         --ptr;
6420         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6421
6422         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6423                 struct lu_buf           lmv_buf;
6424
6425                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6426                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6427                         GOTO(out, rc = -ENOENT);
6428
6429                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6430                         mdt_obj = obj;
6431                         mdt_object_get(info->mti_env, mdt_obj);
6432                 } else {
6433                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6434                         if (IS_ERR(mdt_obj))
6435                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6436                 }
6437
6438                 if (!mdt_object_exists(mdt_obj)) {
6439                         mdt_object_put(info->mti_env, mdt_obj);
6440                         GOTO(out, rc = -ENOENT);
6441                 }
6442
6443                 if (mdt_object_remote(mdt_obj)) {
6444                         mdt_object_put(info->mti_env, mdt_obj);
6445                         GOTO(remote_out, rc = -EREMOTE);
6446                 }
6447
6448                 rc = mdt_links_read(info, mdt_obj, &ldata);
6449                 if (rc != 0) {
6450                         mdt_object_put(info->mti_env, mdt_obj);
6451                         GOTO(out, rc);
6452                 }
6453
6454                 leh = buf->lb_buf;
6455                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6456                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6457                 /* If set, use link #linkno for path lookup, otherwise use
6458                    link #0.  Only do this for the final path element. */
6459                 if (first && fp->gf_linkno < leh->leh_reccount) {
6460                         int count;
6461                         for (count = 0; count < fp->gf_linkno; count++) {
6462                                 lee = (struct link_ea_entry *)
6463                                      ((char *)lee + reclen);
6464                                 linkea_entry_unpack(lee, &reclen, tmpname,
6465                                                     tmpfid);
6466                         }
6467                         if (fp->gf_linkno < leh->leh_reccount - 1)
6468                                 /* indicate to user there are more links */
6469                                 fp->gf_linkno++;
6470                 }
6471
6472                 lmv_buf.lb_buf = info->mti_xattr_buf;
6473                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6474                 /* Check if it is slave stripes */
6475                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6476                                   &lmv_buf, XATTR_NAME_LMV);
6477                 mdt_object_put(info->mti_env, mdt_obj);
6478                 if (rc > 0) {
6479                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6480
6481                         /* For slave stripes, get its master */
6482                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6483                                 fp->gf_fid = *tmpfid;
6484                                 continue;
6485                         }
6486                 } else if (rc < 0 && rc != -ENODATA) {
6487                         GOTO(out, rc);
6488                 }
6489
6490                 rc = 0;
6491
6492                 /* Pack the name in the end of the buffer */
6493                 ptr -= tmpname->ln_namelen;
6494                 if (ptr - 1 <= fp->gf_u.gf_path)
6495                         GOTO(out, rc = -EOVERFLOW);
6496                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6497                 *(--ptr) = '/';
6498
6499                 /* keep the last resolved fid to the client, so the
6500                  * client will build the left path on another MDT for
6501                  * remote object */
6502                 fp->gf_fid = *tmpfid;
6503
6504                 first = false;
6505         }
6506
6507 remote_out:
6508         ptr++; /* skip leading / */
6509         memmove(fp->gf_u.gf_path, ptr,
6510                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6511
6512 out:
6513         RETURN(rc);
6514 }
6515
6516 /**
6517  * Given an MDT object, use mdt_path_current to get the path.
6518  * Essentially a wrapper to retry mdt_path_current a set number of times
6519  * if -EAGAIN is returned (usually because an object is being moved).
6520  *
6521  * Part of the MDT layer implementation of lfs fid2path.
6522  *
6523  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6524  * \param[in]     obj   Object to do path lookup of
6525  * \param[in,out] fp    User-provided struct for arguments and to store path
6526  *                      information
6527  *
6528  * \retval 0 Lookup successful, path information stored in fp
6529  * \retval negative errno if there was a problem
6530  */
6531 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6532                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6533 {
6534         struct mdt_device       *mdt = info->mti_mdt;
6535         int                     tries = 3;
6536         int                     rc = -EAGAIN;
6537         ENTRY;
6538
6539         if (fp->gf_pathlen < 3)
6540                 RETURN(-EOVERFLOW);
6541
6542         if (root_fid == NULL)
6543                 root_fid = &mdt->mdt_md_root_fid;
6544
6545         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6546                 fp->gf_u.gf_path[0] = '\0';
6547                 RETURN(0);
6548         }
6549
6550         /* Retry multiple times in case file is being moved */
6551         while (tries-- && rc == -EAGAIN)
6552                 rc = mdt_path_current(info, obj, fp, root_fid);
6553
6554         RETURN(rc);
6555 }
6556
6557 /**
6558  * Get the full path of the provided FID, as of changelog record recno.
6559  *
6560  * This checks sanity and looks up object for user provided FID
6561  * before calling the actual path lookup code.
6562  *
6563  * Part of the MDT layer implementation of lfs fid2path.
6564  *
6565  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6566  * \param[in,out] fp    User-provided struct for arguments and to store path
6567  *                      information
6568  *
6569  * \retval 0 Lookup successful, path information and recno stored in fp
6570  * \retval -ENOENT, object does not exist
6571  * \retval negative errno if there was a problem
6572  */
6573 static int mdt_fid2path(struct mdt_thread_info *info,
6574                         struct lu_fid *root_fid,
6575                         struct getinfo_fid2path *fp)
6576 {
6577         struct mdt_device *mdt = info->mti_mdt;
6578         struct mdt_object *obj;
6579         int    rc;
6580         ENTRY;
6581
6582         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6583                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6584
6585         if (!fid_is_sane(&fp->gf_fid))
6586                 RETURN(-EINVAL);
6587
6588         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6589                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6590                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6591                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6592                 RETURN(-EINVAL);
6593         }
6594
6595         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6596         if (IS_ERR(obj)) {
6597                 rc = PTR_ERR(obj);
6598                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6599                        PFID(&fp->gf_fid), rc);
6600                 RETURN(rc);
6601         }
6602
6603         if (mdt_object_remote(obj))
6604                 rc = -EREMOTE;
6605         else if (!mdt_object_exists(obj))
6606                 rc = -ENOENT;
6607         else
6608                 rc = 0;
6609
6610         if (rc < 0) {
6611                 mdt_object_put(info->mti_env, obj);
6612                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6613                        PFID(&fp->gf_fid), rc);
6614                 RETURN(rc);
6615         }
6616
6617         rc = mdt_path(info, obj, fp, root_fid);
6618
6619         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6620                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6621                fp->gf_recno, fp->gf_linkno);
6622
6623         mdt_object_put(info->mti_env, obj);
6624
6625         RETURN(rc);
6626 }
6627
6628 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6629                             void *val, int vallen)
6630 {
6631         struct getinfo_fid2path *fpout, *fpin;
6632         struct lu_fid *root_fid = NULL;
6633         int rc = 0;
6634
6635         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6636         fpout = val;
6637
6638         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6639                 lustre_swab_fid2path(fpin);
6640
6641         memcpy(fpout, fpin, sizeof(*fpin));
6642         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6643                 RETURN(-EINVAL);
6644
6645         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6646                       sizeof(struct lu_fid)) {
6647                 /* client sent its root FID, which is normally fileset FID */
6648                 root_fid = fpin->gf_u.gf_root_fid;
6649                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6650                         lustre_swab_lu_fid(root_fid);
6651
6652                 if (root_fid != NULL && !fid_is_sane(root_fid))
6653                         RETURN(-EINVAL);
6654         }
6655
6656         rc = mdt_fid2path(info, root_fid, fpout);
6657         RETURN(rc);
6658 }
6659
6660 int mdt_get_info(struct tgt_session_info *tsi)
6661 {
6662         char    *key;
6663         int      keylen;
6664         __u32   *vallen;
6665         void    *valout;
6666         int      rc;
6667
6668         ENTRY;
6669
6670         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6671         if (key == NULL) {
6672                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6673                 RETURN(err_serious(-EFAULT));
6674         }
6675         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6676                                       RCL_CLIENT);
6677
6678         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6679         if (vallen == NULL) {
6680                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6681                                 tgt_name(tsi->tsi_tgt));
6682                 RETURN(err_serious(-EFAULT));
6683         }
6684
6685         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6686                              *vallen);
6687         rc = req_capsule_server_pack(tsi->tsi_pill);
6688         if (rc)
6689                 RETURN(err_serious(rc));
6690
6691         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6692         if (valout == NULL) {
6693                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6694                                 tgt_name(tsi->tsi_tgt));
6695                 RETURN(err_serious(-EFAULT));
6696         }
6697
6698         if (KEY_IS(KEY_FID2PATH)) {
6699                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6700
6701                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6702                 mdt_thread_info_fini(info);
6703         } else {
6704                 rc = -EINVAL;
6705         }
6706         RETURN(rc);
6707 }
6708
6709 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6710 {
6711         struct obd_ioctl_data *data = karg;
6712         struct lu_fid *fid;
6713         __u64 version;
6714         struct mdt_object *obj;
6715         struct mdt_lock_handle  *lh;
6716         int rc;
6717         ENTRY;
6718
6719         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6720             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6721                 RETURN(-EINVAL);
6722
6723         fid = (struct lu_fid *)data->ioc_inlbuf1;
6724
6725         if (!fid_is_sane(fid))
6726                 RETURN(-EINVAL);
6727
6728         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6729
6730         lh = &mti->mti_lh[MDT_LH_PARENT];
6731         mdt_lock_reg_init(lh, LCK_CR);
6732
6733         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6734         if (IS_ERR(obj))
6735                 RETURN(PTR_ERR(obj));
6736
6737         if (mdt_object_remote(obj)) {
6738                 rc = -EREMOTE;
6739                 /**
6740                  * before calling version get the correct MDS should be
6741                  * fid, this is error to find remote object here
6742                  */
6743                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6744         } else if (!mdt_object_exists(obj)) {
6745                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6746                 rc = -ENOENT;
6747         } else {
6748                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6749                *(__u64 *)data->ioc_inlbuf2 = version;
6750                 rc = 0;
6751         }
6752         mdt_object_unlock_put(mti, obj, lh, 1);
6753         RETURN(rc);
6754 }
6755
6756 /* ioctls on obd dev */
6757 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6758                          void *karg, void __user *uarg)
6759 {
6760         struct lu_env      env;
6761         struct obd_device *obd = exp->exp_obd;
6762         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6763         struct dt_device  *dt = mdt->mdt_bottom;
6764         int rc;
6765
6766         ENTRY;
6767         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6768         rc = lu_env_init(&env, LCT_MD_THREAD);
6769         if (rc)
6770                 RETURN(rc);
6771
6772         switch (cmd) {
6773         case OBD_IOC_SYNC:
6774                 rc = mdt_device_sync(&env, mdt);
6775                 break;
6776         case OBD_IOC_SET_READONLY:
6777                 rc = dt_sync(&env, dt);
6778                 if (rc == 0)
6779                         rc = dt_ro(&env, dt);
6780                 break;
6781         case OBD_IOC_ABORT_RECOVERY:
6782                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6783                 obd->obd_abort_recovery = 1;
6784                 target_stop_recovery_thread(obd);
6785                 rc = 0;
6786                 break;
6787         case OBD_IOC_CHANGELOG_REG:
6788         case OBD_IOC_CHANGELOG_DEREG:
6789         case OBD_IOC_CHANGELOG_CLEAR:
6790                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6791                                                            mdt->mdt_child,
6792                                                            cmd, len, karg);
6793                 break;
6794         case OBD_IOC_START_LFSCK: {
6795                 struct md_device *next = mdt->mdt_child;
6796                 struct obd_ioctl_data *data = karg;
6797                 struct lfsck_start_param lsp;
6798
6799                 if (unlikely(data == NULL)) {
6800                         rc = -EINVAL;
6801                         break;
6802                 }
6803
6804                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6805                 lsp.lsp_index_valid = 0;
6806                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6807                 break;
6808         }
6809         case OBD_IOC_STOP_LFSCK: {
6810                 struct md_device        *next = mdt->mdt_child;
6811                 struct obd_ioctl_data   *data = karg;
6812                 struct lfsck_stop        stop;
6813
6814                 stop.ls_status = LS_STOPPED;
6815                 /* Old lfsck utils may pass NULL @stop. */
6816                 if (data->ioc_inlbuf1 == NULL)
6817                         stop.ls_flags = 0;
6818                 else
6819                         stop.ls_flags =
6820                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6821
6822                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6823                 break;
6824         }
6825         case OBD_IOC_QUERY_LFSCK: {
6826                 struct md_device        *next = mdt->mdt_child;
6827                 struct obd_ioctl_data   *data = karg;
6828
6829                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6830                                                  data->ioc_inlbuf1);
6831                 break;
6832         }
6833         case OBD_IOC_GET_OBJ_VERSION: {
6834                 struct mdt_thread_info *mti;
6835                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6836                 memset(mti, 0, sizeof *mti);
6837                 mti->mti_env = &env;
6838                 mti->mti_mdt = mdt;
6839                 mti->mti_exp = exp;
6840
6841                 rc = mdt_ioc_version_get(mti, karg);
6842                 break;
6843         }
6844         case OBD_IOC_CATLOGLIST: {
6845                 struct mdt_thread_info *mti;
6846
6847                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6848                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6849                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6850                                        &mti->mti_tmp_fid1);
6851                 break;
6852          }
6853         default:
6854                 rc = -EOPNOTSUPP;
6855                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6856                         mdt_obd_name(mdt), cmd, rc);
6857         }
6858
6859         lu_env_fini(&env);
6860         RETURN(rc);
6861 }
6862
6863 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6864 {
6865         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6866         int rc;
6867         ENTRY;
6868
6869         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6870                 struct lfsck_start_param lsp;
6871
6872                 lsp.lsp_start = NULL;
6873                 lsp.lsp_index_valid = 0;
6874                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6875                                                            OBD_IOC_START_LFSCK,
6876                                                            0, &lsp);
6877                 if (rc != 0 && rc != -EALREADY)
6878                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6879                               mdt_obd_name(mdt), rc);
6880         }
6881
6882         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6883         RETURN(rc);
6884 }
6885
6886 static int mdt_obd_postrecov(struct obd_device *obd)
6887 {
6888         struct lu_env env;
6889         int rc;
6890
6891         rc = lu_env_init(&env, LCT_MD_THREAD);
6892         if (rc)
6893                 RETURN(rc);
6894         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6895         lu_env_fini(&env);
6896         return rc;
6897 }
6898
6899 static struct obd_ops mdt_obd_device_ops = {
6900         .o_owner          = THIS_MODULE,
6901         .o_set_info_async = mdt_obd_set_info_async,
6902         .o_connect        = mdt_obd_connect,
6903         .o_reconnect      = mdt_obd_reconnect,
6904         .o_disconnect     = mdt_obd_disconnect,
6905         .o_init_export    = mdt_init_export,
6906         .o_destroy_export = mdt_destroy_export,
6907         .o_iocontrol      = mdt_iocontrol,
6908         .o_postrecov      = mdt_obd_postrecov,
6909         /* Data-on-MDT IO methods */
6910         .o_preprw         = mdt_obd_preprw,
6911         .o_commitrw       = mdt_obd_commitrw,
6912 };
6913
6914 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6915                                          struct lu_device *d)
6916 {
6917         struct mdt_device *m = mdt_dev(d);
6918         ENTRY;
6919
6920         mdt_fini(env, m);
6921         RETURN(NULL);
6922 }
6923
6924 static struct lu_device *mdt_device_free(const struct lu_env *env,
6925                                          struct lu_device *d)
6926 {
6927         struct mdt_device *m = mdt_dev(d);
6928         ENTRY;
6929
6930         lu_device_fini(&m->mdt_lu_dev);
6931         OBD_FREE_PTR(m);
6932
6933         RETURN(NULL);
6934 }
6935
6936 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6937                                           struct lu_device_type *t,
6938                                           struct lustre_cfg *cfg)
6939 {
6940         struct lu_device  *l;
6941         struct mdt_device *m;
6942
6943         OBD_ALLOC_PTR(m);
6944         if (m != NULL) {
6945                 int rc;
6946
6947                 l = &m->mdt_lu_dev;
6948                 rc = mdt_init0(env, m, t, cfg);
6949                 if (rc != 0) {
6950                         mdt_device_free(env, l);
6951                         l = ERR_PTR(rc);
6952                         return l;
6953                 }
6954         } else
6955                 l = ERR_PTR(-ENOMEM);
6956         return l;
6957 }
6958
6959 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6960 LU_KEY_INIT(mdt, struct mdt_thread_info);
6961
6962 static void mdt_key_fini(const struct lu_context *ctx,
6963                          struct lu_context_key *key, void* data)
6964 {
6965         struct mdt_thread_info *info = data;
6966
6967         if (info->mti_big_lmm) {
6968                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6969                 info->mti_big_lmm = NULL;
6970                 info->mti_big_lmmsize = 0;
6971         }
6972
6973         if (info->mti_big_acl) {
6974                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6975                 info->mti_big_acl = NULL;
6976                 info->mti_big_aclsize = 0;
6977         }
6978
6979         OBD_FREE_PTR(info);
6980 }
6981
6982 /* context key: mdt_thread_key */
6983 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6984
6985 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6986 {
6987         return lu_ucred(info->mti_env);
6988 }
6989
6990 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6991 {
6992         return lu_ucred_check(info->mti_env);
6993 }
6994
6995 /**
6996  * Enable/disable COS (Commit On Sharing).
6997  *
6998  * Set/Clear the COS flag in mdt options.
6999  *
7000  * \param mdt mdt device
7001  * \param val 0 disables COS, other values enable COS
7002  */
7003 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7004 {
7005         struct lu_env env;
7006         int rc;
7007
7008         mdt->mdt_opts.mo_cos = val;
7009         rc = lu_env_init(&env, LCT_LOCAL);
7010         if (unlikely(rc != 0)) {
7011                 CWARN("%s: lu_env initialization failed, cannot "
7012                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7013                 return;
7014         }
7015         mdt_device_sync(&env, mdt);
7016         lu_env_fini(&env);
7017 }
7018
7019 /**
7020  * Check COS (Commit On Sharing) status.
7021  *
7022  * Return COS flag status.
7023  *
7024  * \param mdt mdt device
7025  */
7026 int mdt_cos_is_enabled(struct mdt_device *mdt)
7027 {
7028         return mdt->mdt_opts.mo_cos != 0;
7029 }
7030
7031 static struct lu_device_type_operations mdt_device_type_ops = {
7032         .ldto_device_alloc = mdt_device_alloc,
7033         .ldto_device_free  = mdt_device_free,
7034         .ldto_device_fini  = mdt_device_fini
7035 };
7036
7037 static struct lu_device_type mdt_device_type = {
7038         .ldt_tags     = LU_DEVICE_MD,
7039         .ldt_name     = LUSTRE_MDT_NAME,
7040         .ldt_ops      = &mdt_device_type_ops,
7041         .ldt_ctx_tags = LCT_MD_THREAD
7042 };
7043
7044 static int __init mdt_init(void)
7045 {
7046         int rc;
7047
7048         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
7049                  FID_NOBRACE_LEN + 1);
7050         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
7051                  FID_LEN + 1);
7052         rc = lu_kmem_init(mdt_caches);
7053         if (rc)
7054                 return rc;
7055
7056         rc = mds_mod_init();
7057         if (rc)
7058                 GOTO(lu_fini, rc);
7059
7060         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
7061                                  LUSTRE_MDT_NAME, &mdt_device_type);
7062         if (rc)
7063                 GOTO(mds_fini, rc);
7064 lu_fini:
7065         if (rc)
7066                 lu_kmem_fini(mdt_caches);
7067 mds_fini:
7068         if (rc)
7069                 mds_mod_exit();
7070         return rc;
7071 }
7072
7073 static void __exit mdt_exit(void)
7074 {
7075         class_unregister_type(LUSTRE_MDT_NAME);
7076         mds_mod_exit();
7077         lu_kmem_fini(mdt_caches);
7078 }
7079
7080 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7081 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7082 MODULE_VERSION(LUSTRE_VERSION_STRING);
7083 MODULE_LICENSE("GPL");
7084
7085 module_init(mdt_init);
7086 module_exit(mdt_exit);