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, 2016, 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 <lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_net.h>
  57 #include <lustre_nodemap.h>
  58 #include <lustre_mds.h>
  59 #include <lustre_param.h>
  60 #include <lustre_quota.h>
  61 #include <lustre_swab.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 module_param(max_mod_rpcs_per_client, uint, 0644);
  73 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 270                               struct lu_fid *fid)
 271 {
 272         struct mdt_device *mdt = info->mti_mdt;
 273         struct lu_name *lname = &info->mti_name;
 274         char *name = NULL;
 275         struct mdt_object *parent;
 276         u32 mode;
 277         int rc = 0;
 278
 279         LASSERT(!info->mti_cross_ref);
 280
 281         OBD_ALLOC(name, NAME_MAX + 1);
 282         if (name == NULL)
 283                 return -ENOMEM;
 284         lname->ln_name = name;
 285
 286         /*
 287          * We may want to allow this to mount a completely separate
 288          * fileset from the MDT in the future, but keeping it to
 289          * ROOT/ only for now avoid potential security issues.
 290          */
 291         *fid = mdt->mdt_md_root_fid;
 292
 293         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 294                 const char *s1 = fileset;
 295                 const char *s2;
 296
 297                 while (*++s1 == '/')
 298                         ;
 299                 s2 = s1;
 300                 while (*s2 != '/' && *s2 != '\0')
 301                         s2++;
 302
 303                 if (s2 == s1)
 304                         break;
 305
 306                 fileset = s2;
 307
 308                 lname->ln_namelen = s2 - s1;
 309                 if (lname->ln_namelen > NAME_MAX) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 /* reject .. as a path component */
 315                 if (lname->ln_namelen == 2 &&
 316                     strncmp(s1, "..", 2) == 0) {
 317                         rc = -EINVAL;
 318                         break;
 319                 }
 320
 321                 strncpy(name, s1, lname->ln_namelen);
 322                 name[lname->ln_namelen] = '\0';
 323
 324                 parent = mdt_object_find(info->mti_env, mdt, fid);
 325                 if (IS_ERR(parent)) {
 326                         rc = PTR_ERR(parent);
 327                         break;
 328                 }
 329                 /* Only got the fid of this obj by name */
 330                 fid_zero(fid);
 331                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 332                                 fid, &info->mti_spec);
 333                 mdt_object_put(info->mti_env, parent);
 334         }
 335         if (!rc) {
 336                 parent = mdt_object_find(info->mti_env, mdt, fid);
 337                 if (IS_ERR(parent))
 338                         rc = PTR_ERR(parent);
 339                 else {
 340                         mode = lu_object_attr(&parent->mot_obj);
 341                         mdt_object_put(info->mti_env, parent);
 342                         if (!S_ISDIR(mode))
 343                                 rc = -ENOTDIR;
 344                 }
 345         }
 346
 347         OBD_FREE(name, NAME_MAX + 1);
 348
 349         return rc;
 350 }
 351
 352 static int mdt_get_root(struct tgt_session_info *tsi)
 353 {
 354         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 355         struct mdt_device       *mdt = info->mti_mdt;
 356         struct mdt_body         *repbody;
 357         char                    *fileset = NULL, *buffer = NULL;
 358         int                      rc;
 359         struct obd_export       *exp = info->mti_exp;
 360         char                    *nodemap_fileset;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376         }
 377
 378         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 379         if (nodemap_fileset && nodemap_fileset[0]) {
 380                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 381                 if (fileset) {
 382                         /* consider fileset from client as a sub-fileset
 383                          * of the nodemap one */
 384                         OBD_ALLOC(buffer, PATH_MAX + 1);
 385                         if (buffer == NULL)
 386                                 GOTO(out, rc = err_serious(-ENOMEM));
 387                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 388                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 389                                 GOTO(out, rc = err_serious(-EINVAL));
 390                         fileset = buffer;
 391                 } else {
 392                         /* enforce fileset as specified in the nodemap */
 393                         fileset = nodemap_fileset;
 394                 }
 395         }
 396
 397         if (fileset) {
 398                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 399                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 400                 if (rc < 0)
 401                         GOTO(out, rc = err_serious(rc));
 402         } else {
 403                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 404         }
 405         repbody->mbo_valid |= OBD_MD_FLID;
 406
 407         EXIT;
 408 out:
 409         mdt_thread_info_fini(info);
 410         if (buffer)
 411                 OBD_FREE(buffer, PATH_MAX+1);
 412         return rc;
 413 }
 414
 415 static int mdt_statfs(struct tgt_session_info *tsi)
 416 {
 417         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 418         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 419         struct md_device                *next = info->mti_mdt->mdt_child;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         /** statfs information are cached in the mdt_device */
 445         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 446                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 447                 /** statfs data is too old, get up-to-date one */
 448                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 449                 if (rc)
 450                         GOTO(out, rc);
 451                 spin_lock(&info->mti_mdt->mdt_lock);
 452                 info->mti_mdt->mdt_osfs = *osfs;
 453                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 454                 spin_unlock(&info->mti_mdt->mdt_lock);
 455         } else {
 456                 /** use cached statfs data */
 457                 spin_lock(&info->mti_mdt->mdt_lock);
 458                 *osfs = info->mti_mdt->mdt_osfs;
 459                 spin_unlock(&info->mti_mdt->mdt_lock);
 460         }
 461
 462         if (rc == 0)
 463                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 464 out:
 465         mdt_thread_info_fini(info);
 466         RETURN(rc);
 467 }
 468
 469 #ifdef CONFIG_FS_POSIX_ACL
 470 /*
 471  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 472  *
 473  * \param       info    thread info object
 474  * \param       repbody reply to pack ACLs into
 475  * \param       o       mdt object of file to examine
 476  * \param       nodemap nodemap of client to reply to
 477  * \retval      0       success
 478  * \retval      -errno  error getting or parsing ACL from disk
 479  */
 480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 481                       struct mdt_object *o, struct lu_nodemap *nodemap)
 482 {
 483         const struct lu_env     *env = info->mti_env;
 484         struct md_object        *next = mdt_object_child(o);
 485         struct lu_buf           *buf = &info->mti_buf;
 486         int rc;
 487
 488         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 489         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 490                                            RCL_SERVER);
 491         if (buf->lb_len == 0)
 492                 return 0;
 493
 494         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 495         if (rc < 0) {
 496                 if (rc == -ENODATA) {
 497                         repbody->mbo_aclsize = 0;
 498                         repbody->mbo_valid |= OBD_MD_FLACL;
 499                         rc = 0;
 500                 } else if (rc == -EOPNOTSUPP) {
 501                         rc = 0;
 502                 } else {
 503                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 504                                mdt_obd_name(info->mti_mdt),
 505                                PFID(mdt_object_fid(o)), rc);
 506                 }
 507         } else {
 508                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 509                                      rc, NODEMAP_FS_TO_CLIENT);
 510                 /* if all ACLs mapped out, rc is still >= 0 */
 511                 if (rc < 0) {
 512                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 513                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 514                                PFID(mdt_object_fid(o)), rc);
 515                 } else {
 516                         repbody->mbo_aclsize = rc;
 517                         repbody->mbo_valid |= OBD_MD_FLACL;
 518                         rc = 0;
 519                 }
 520         }
 521         return rc;
 522 }
 523 #endif
 524
 525 /* XXX Look into layout in MDT layer. */
 526 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 527 {
 528         struct lov_comp_md_v1   *comp_v1;
 529         struct lov_mds_md       *v1;
 530         int                      i;
 531
 532         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 533                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 534
 535                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 536                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 537                                 comp_v1->lcm_entries[i].lcme_offset);
 538                         /* We don't support partial release for now */
 539                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 540                                 return false;
 541                 }
 542                 return true;
 543         } else {
 544                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 545                         true : false;
 546         }
 547 }
 548
 549 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 550                         const struct lu_attr *attr, const struct lu_fid *fid)
 551 {
 552         struct md_attr *ma = &info->mti_attr;
 553         struct obd_export *exp = info->mti_exp;
 554         struct lu_nodemap *nodemap = NULL;
 555
 556         LASSERT(ma->ma_valid & MA_INODE);
 557
 558         if (attr->la_valid & LA_ATIME) {
 559                 b->mbo_atime = attr->la_atime;
 560                 b->mbo_valid |= OBD_MD_FLATIME;
 561         }
 562         if (attr->la_valid & LA_MTIME) {
 563                 b->mbo_mtime = attr->la_mtime;
 564                 b->mbo_valid |= OBD_MD_FLMTIME;
 565         }
 566         if (attr->la_valid & LA_CTIME) {
 567                 b->mbo_ctime = attr->la_ctime;
 568                 b->mbo_valid |= OBD_MD_FLCTIME;
 569         }
 570         if (attr->la_valid & LA_FLAGS) {
 571                 b->mbo_flags = attr->la_flags;
 572                 b->mbo_valid |= OBD_MD_FLFLAGS;
 573         }
 574         if (attr->la_valid & LA_NLINK) {
 575                 b->mbo_nlink = attr->la_nlink;
 576                 b->mbo_valid |= OBD_MD_FLNLINK;
 577         }
 578         if (attr->la_valid & (LA_UID|LA_GID)) {
 579                 nodemap = nodemap_get_from_exp(exp);
 580                 if (IS_ERR(nodemap))
 581                         goto out;
 582         }
 583         if (attr->la_valid & LA_UID) {
 584                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 585                                             NODEMAP_FS_TO_CLIENT,
 586                                             attr->la_uid);
 587                 b->mbo_valid |= OBD_MD_FLUID;
 588         }
 589         if (attr->la_valid & LA_GID) {
 590                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 591                                             NODEMAP_FS_TO_CLIENT,
 592                                             attr->la_gid);
 593                 b->mbo_valid |= OBD_MD_FLGID;
 594         }
 595
 596         b->mbo_mode = attr->la_mode;
 597         if (attr->la_valid & LA_MODE)
 598                 b->mbo_valid |= OBD_MD_FLMODE;
 599         if (attr->la_valid & LA_TYPE)
 600                 b->mbo_valid |= OBD_MD_FLTYPE;
 601
 602         if (fid != NULL) {
 603                 b->mbo_fid1 = *fid;
 604                 b->mbo_valid |= OBD_MD_FLID;
 605                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 606                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 607         }
 608
 609         if (!(attr->la_valid & LA_TYPE))
 610                 return;
 611
 612         b->mbo_rdev   = attr->la_rdev;
 613         b->mbo_size   = attr->la_size;
 614         b->mbo_blocks = attr->la_blocks;
 615
 616         if (!S_ISREG(attr->la_mode)) {
 617                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 618         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 619                 /* means no objects are allocated on osts. */
 620                 LASSERT(!(ma->ma_valid & MA_LOV));
 621                 /* just ignore blocks occupied by extend attributes on MDS */
 622                 b->mbo_blocks = 0;
 623                 /* if no object is allocated on osts, the size on mds is valid.
 624                  * b=22272 */
 625                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 626         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 627                    mdt_hsm_is_released(ma->ma_lmm)) {
 628                 /* A released file stores its size on MDS. */
 629                 /* But return 1 block for released file, unless tools like tar
 630                  * will consider it fully sparse. (LU-3864)
 631                  */
 632                 if (unlikely(b->mbo_size == 0))
 633                         b->mbo_blocks = 0;
 634                 else
 635                         b->mbo_blocks = 1;
 636                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 637         }
 638
 639         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 640                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 641                        PFID(fid), (unsigned long long)b->mbo_size);
 642
 643 out:
 644         if (!IS_ERR_OR_NULL(nodemap))
 645                 nodemap_putref(nodemap);
 646 }
 647
 648 static inline int mdt_body_has_lov(const struct lu_attr *la,
 649                                    const struct mdt_body *body)
 650 {
 651         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 652                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 653 }
 654
 655 void mdt_client_compatibility(struct mdt_thread_info *info)
 656 {
 657         struct mdt_body       *body;
 658         struct ptlrpc_request *req = mdt_info_req(info);
 659         struct obd_export     *exp = req->rq_export;
 660         struct md_attr        *ma = &info->mti_attr;
 661         struct lu_attr        *la = &ma->ma_attr;
 662         ENTRY;
 663
 664         if (exp_connect_layout(exp))
 665                 /* the client can deal with 16-bit lmm_stripe_count */
 666                 RETURN_EXIT;
 667
 668         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 669
 670         if (!mdt_body_has_lov(la, body))
 671                 RETURN_EXIT;
 672
 673         /* now we have a reply with a lov for a client not compatible with the
 674          * layout lock so we have to clean the layout generation number */
 675         if (S_ISREG(la->la_mode))
 676                 ma->ma_lmm->lmm_layout_gen = 0;
 677         EXIT;
 678 }
 679
 680 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 681                                    struct mdt_object *o)
 682 {
 683         const struct lu_env *env = info->mti_env;
 684         int rc, rc2;
 685
 686         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 687                           XATTR_NAME_LOV);
 688
 689         if (rc == -ENODATA)
 690                 rc = 0;
 691
 692         if (rc < 0)
 693                 goto out;
 694
 695         /* Is it a directory? Let's check for the LMV as well */
 696         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 697                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 698                                    XATTR_NAME_LMV);
 699
 700                 if (rc2 == -ENODATA)
 701                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 702                                            &LU_BUF_NULL,
 703                                            XATTR_NAME_DEFAULT_LMV);
 704
 705                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 706                         rc = rc2;
 707         }
 708
 709 out:
 710         return rc;
 711 }
 712
 713 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 714                       const char *name)
 715 {
 716         const struct lu_env *env = info->mti_env;
 717         int rc;
 718         ENTRY;
 719
 720         LASSERT(info->mti_big_lmm_used == 0);
 721         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 722         if (rc < 0)
 723                 RETURN(rc);
 724
 725         /* big_lmm may need to be grown */
 726         if (info->mti_big_lmmsize < rc) {
 727                 int size = size_roundup_power2(rc);
 728
 729                 if (info->mti_big_lmmsize > 0) {
 730                         /* free old buffer */
 731                         LASSERT(info->mti_big_lmm);
 732                         OBD_FREE_LARGE(info->mti_big_lmm,
 733                                        info->mti_big_lmmsize);
 734                         info->mti_big_lmm = NULL;
 735                         info->mti_big_lmmsize = 0;
 736                 }
 737
 738                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 739                 if (info->mti_big_lmm == NULL)
 740                         RETURN(-ENOMEM);
 741                 info->mti_big_lmmsize = size;
 742         }
 743         LASSERT(info->mti_big_lmmsize >= rc);
 744
 745         info->mti_buf.lb_buf = info->mti_big_lmm;
 746         info->mti_buf.lb_len = info->mti_big_lmmsize;
 747         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 748
 749         RETURN(rc);
 750 }
 751
 752 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 753                    struct md_attr *ma, const char *name)
 754 {
 755         struct md_object *next = mdt_object_child(o);
 756         struct lu_buf    *buf = &info->mti_buf;
 757         int rc;
 758
 759         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 760                 buf->lb_buf = ma->ma_lmm;
 761                 buf->lb_len = ma->ma_lmm_size;
 762                 LASSERT(!(ma->ma_valid & MA_LOV));
 763         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 764                 buf->lb_buf = ma->ma_lmv;
 765                 buf->lb_len = ma->ma_lmv_size;
 766                 LASSERT(!(ma->ma_valid & MA_LMV));
 767         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 768                 buf->lb_buf = ma->ma_lmv;
 769                 buf->lb_len = ma->ma_lmv_size;
 770                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 771         } else {
 772                 return -EINVAL;
 773         }
 774
 775         rc = mo_xattr_get(info->mti_env, next, buf, name);
 776         if (rc > 0) {
 777
 778 got:
 779                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 780                         if (info->mti_big_lmm_used)
 781                                 ma->ma_lmm = info->mti_big_lmm;
 782
 783                         /* NOT return LOV EA with hole to old client. */
 784                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 785                                      LOV_PATTERN_F_HOLE) &&
 786                             !(exp_connect_flags(info->mti_exp) &
 787                               OBD_CONNECT_LFSCK)) {
 788                                 return -EIO;
 789                         } else {
 790                                 ma->ma_lmm_size = rc;
 791                                 ma->ma_valid |= MA_LOV;
 792                         }
 793                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 794                         if (info->mti_big_lmm_used)
 795                                 ma->ma_lmv = info->mti_big_lmm;
 796
 797                         ma->ma_lmv_size = rc;
 798                         ma->ma_valid |= MA_LMV;
 799                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 800                         ma->ma_lmv_size = rc;
 801                         ma->ma_valid |= MA_LMV_DEF;
 802                 }
 803
 804                 /* Update mdt_max_mdsize so all clients will be aware that */
 805                 if (info->mti_mdt->mdt_max_mdsize < rc)
 806                         info->mti_mdt->mdt_max_mdsize = rc;
 807
 808                 rc = 0;
 809         } else if (rc == -ENODATA) {
 810                 /* no LOV EA */
 811                 rc = 0;
 812         } else if (rc == -ERANGE) {
 813                 /* Default LMV has fixed size, so it must be able to fit
 814                  * in the original buffer */
 815                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 816                         return rc;
 817                 rc = mdt_big_xattr_get(info, o, name);
 818                 if (rc > 0) {
 819                         info->mti_big_lmm_used = 1;
 820                         goto got;
 821                 }
 822         }
 823
 824         return rc;
 825 }
 826
 827 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 828                              struct mdt_object *o, struct lu_fid *pfid)
 829 {
 830         struct lu_buf           *buf = &info->mti_buf;
 831         struct link_ea_header   *leh;
 832         struct link_ea_entry    *lee;
 833         int                      rc;
 834         ENTRY;
 835
 836         buf->lb_buf = info->mti_big_lmm;
 837         buf->lb_len = info->mti_big_lmmsize;
 838         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 839                           buf, XATTR_NAME_LINK);
 840         /* ignore errors, MA_PFID won't be set and it is
 841          * up to the caller to treat this as an error */
 842         if (rc == -ERANGE || buf->lb_len == 0) {
 843                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 844                 buf->lb_buf = info->mti_big_lmm;
 845                 buf->lb_len = info->mti_big_lmmsize;
 846         }
 847
 848         if (rc < 0)
 849                 RETURN(rc);
 850         if (rc < sizeof(*leh)) {
 851                 CERROR("short LinkEA on "DFID": rc = %d\n",
 852                        PFID(mdt_object_fid(o)), rc);
 853                 RETURN(-ENODATA);
 854         }
 855
 856         leh = (struct link_ea_header *) buf->lb_buf;
 857         lee = (struct link_ea_entry *)(leh + 1);
 858         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 859                 leh->leh_magic = LINK_EA_MAGIC;
 860                 leh->leh_reccount = __swab32(leh->leh_reccount);
 861                 leh->leh_len = __swab64(leh->leh_len);
 862         }
 863         if (leh->leh_magic != LINK_EA_MAGIC)
 864                 RETURN(-EINVAL);
 865         if (leh->leh_reccount == 0)
 866                 RETURN(-ENODATA);
 867
 868         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 869         fid_be_to_cpu(pfid, pfid);
 870
 871         RETURN(0);
 872 }
 873
 874 int mdt_attr_get_complex(struct mdt_thread_info *info,
 875                          struct mdt_object *o, struct md_attr *ma)
 876 {
 877         const struct lu_env *env = info->mti_env;
 878         struct md_object    *next = mdt_object_child(o);
 879         struct lu_buf       *buf = &info->mti_buf;
 880         int                  need = ma->ma_need;
 881         int                  rc = 0, rc2;
 882         u32                  mode;
 883         ENTRY;
 884
 885         ma->ma_valid = 0;
 886
 887         if (mdt_object_exists(o) == 0)
 888                 GOTO(out, rc = -ENOENT);
 889         mode = lu_object_attr(&next->mo_lu);
 890
 891         if (need & MA_INODE) {
 892                 ma->ma_need = MA_INODE;
 893                 rc = mo_attr_get(env, next, ma);
 894                 if (rc)
 895                         GOTO(out, rc);
 896                 ma->ma_valid |= MA_INODE;
 897         }
 898
 899         if (need & MA_PFID) {
 900                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 901                 if (rc == 0)
 902                         ma->ma_valid |= MA_PFID;
 903                 /* ignore this error, parent fid is not mandatory */
 904                 rc = 0;
 905         }
 906
 907         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 908                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 909                 if (rc)
 910                         GOTO(out, rc);
 911         }
 912
 913         if (need & MA_LMV && S_ISDIR(mode)) {
 914                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 915                 if (rc != 0)
 916                         GOTO(out, rc);
 917         }
 918
 919         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 920                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 921                 if (rc != 0)
 922                         GOTO(out, rc);
 923         }
 924
 925         if (need & MA_HSM && S_ISREG(mode)) {
 926                 buf->lb_buf = info->mti_xattr_buf;
 927                 buf->lb_len = sizeof(info->mti_xattr_buf);
 928                 CLASSERT(sizeof(struct hsm_attrs) <=
 929                          sizeof(info->mti_xattr_buf));
 930                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 931                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 932                 if (rc2 == 0)
 933                         ma->ma_valid |= MA_HSM;
 934                 else if (rc2 < 0 && rc2 != -ENODATA)
 935                         GOTO(out, rc = rc2);
 936         }
 937
 938 #ifdef CONFIG_FS_POSIX_ACL
 939         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 940                 buf->lb_buf = ma->ma_acl;
 941                 buf->lb_len = ma->ma_acl_size;
 942                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 943                 if (rc2 > 0) {
 944                         ma->ma_acl_size = rc2;
 945                         ma->ma_valid |= MA_ACL_DEF;
 946                 } else if (rc2 == -ENODATA) {
 947                         /* no ACLs */
 948                         ma->ma_acl_size = 0;
 949                 } else
 950                         GOTO(out, rc = rc2);
 951         }
 952 #endif
 953 out:
 954         ma->ma_need = need;
 955         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
 956                rc, ma->ma_valid, ma->ma_lmm);
 957         RETURN(rc);
 958 }
 959
 960 static int mdt_getattr_internal(struct mdt_thread_info *info,
 961                                 struct mdt_object *o, int ma_need)
 962 {
 963         struct md_object        *next = mdt_object_child(o);
 964         const struct mdt_body   *reqbody = info->mti_body;
 965         struct ptlrpc_request   *req = mdt_info_req(info);
 966         struct md_attr          *ma = &info->mti_attr;
 967         struct lu_attr          *la = &ma->ma_attr;
 968         struct req_capsule      *pill = info->mti_pill;
 969         const struct lu_env     *env = info->mti_env;
 970         struct mdt_body         *repbody;
 971         struct lu_buf           *buffer = &info->mti_buf;
 972         struct obd_export       *exp = info->mti_exp;
 973         int                      rc;
 974         ENTRY;
 975
 976         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 977                 RETURN(err_serious(-ENOMEM));
 978
 979         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 980
 981         ma->ma_valid = 0;
 982
 983         if (mdt_object_remote(o)) {
 984                 /* This object is located on remote node.*/
 985                 /* Return -ENOTSUPP for old client */
 986                 if (!mdt_is_dne_client(req->rq_export))
 987                         GOTO(out, rc = -ENOTSUPP);
 988
 989                 repbody->mbo_fid1 = *mdt_object_fid(o);
 990                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 991                 GOTO(out, rc = 0);
 992         }
 993
 994         if (reqbody->mbo_eadatasize > 0) {
 995                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 996                 if (buffer->lb_buf == NULL)
 997                         GOTO(out, rc = -EPROTO);
 998                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 999                                                       RCL_SERVER);
1000         } else {
1001                 buffer->lb_buf = NULL;
1002                 buffer->lb_len = 0;
1003                 ma_need &= ~(MA_LOV | MA_LMV);
1004                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1005                        mdt_obd_name(info->mti_mdt),
1006                        req->rq_export->exp_client_uuid.uuid);
1007         }
1008
1009         /* If it is dir object and client require MEA, then we got MEA */
1010         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1011             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1012                 /* Assumption: MDT_MD size is enough for lmv size. */
1013                 ma->ma_lmv = buffer->lb_buf;
1014                 ma->ma_lmv_size = buffer->lb_len;
1015                 ma->ma_need = MA_INODE;
1016                 if (ma->ma_lmv_size > 0) {
1017                         if (reqbody->mbo_valid & OBD_MD_MEA)
1018                                 ma->ma_need |= MA_LMV;
1019                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1020                                 ma->ma_need |= MA_LMV_DEF;
1021                 }
1022         } else {
1023                 ma->ma_lmm = buffer->lb_buf;
1024                 ma->ma_lmm_size = buffer->lb_len;
1025                 ma->ma_need = MA_INODE | MA_HSM;
1026                 if (ma->ma_lmm_size > 0)
1027                         ma->ma_need |= MA_LOV;
1028         }
1029
1030         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1031             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1032             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1033                 /* get default stripe info for this dir. */
1034                 ma->ma_need |= MA_LOV_DEF;
1035         }
1036         ma->ma_need |= ma_need;
1037
1038         rc = mdt_attr_get_complex(info, o, ma);
1039         if (unlikely(rc)) {
1040                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1041                        "%s: getattr error for "DFID": rc = %d\n",
1042                        mdt_obd_name(info->mti_mdt),
1043                        PFID(mdt_object_fid(o)), rc);
1044                 RETURN(rc);
1045         }
1046
1047         /* if file is released, check if a restore is running */
1048         if (ma->ma_valid & MA_HSM) {
1049                 repbody->mbo_valid |= OBD_MD_TSTATE;
1050                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1051                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1052                         repbody->mbo_t_state = MS_RESTORE;
1053         }
1054
1055         if (likely(ma->ma_valid & MA_INODE))
1056                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1057         else
1058                 RETURN(-EFAULT);
1059
1060         if (mdt_body_has_lov(la, reqbody)) {
1061                 if (ma->ma_valid & MA_LOV) {
1062                         LASSERT(ma->ma_lmm_size);
1063                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1064                         if (S_ISDIR(la->la_mode))
1065                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1066                         else
1067                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1068                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1069                 }
1070                 if (ma->ma_valid & MA_LMV) {
1071                         /* Return -ENOTSUPP for old client */
1072                         if (!mdt_is_striped_client(req->rq_export))
1073                                 RETURN(-ENOTSUPP);
1074
1075                         LASSERT(S_ISDIR(la->la_mode));
1076                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1077                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1078                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1079                 }
1080                 if (ma->ma_valid & MA_LMV_DEF) {
1081                         /* Return -ENOTSUPP for old client */
1082                         if (!mdt_is_striped_client(req->rq_export))
1083                                 RETURN(-ENOTSUPP);
1084                         LASSERT(S_ISDIR(la->la_mode));
1085                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1086                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1087                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1088                                                OBD_MD_DEFAULT_MEA);
1089                 }
1090         } else if (S_ISLNK(la->la_mode) &&
1091                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1092                 buffer->lb_buf = ma->ma_lmm;
1093                 /* eadatasize from client includes NULL-terminator, so
1094                  * there is no need to read it */
1095                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1096                 rc = mo_readlink(env, next, buffer);
1097                 if (unlikely(rc <= 0)) {
1098                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1099                                mdt_obd_name(info->mti_mdt),
1100                                PFID(mdt_object_fid(o)), rc);
1101                         rc = -EFAULT;
1102                 } else {
1103                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1104
1105                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1106                                 rc -= 2;
1107                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1108                         /* we need to report back size with NULL-terminator
1109                          * because client expects that */
1110                         repbody->mbo_eadatasize = rc + 1;
1111                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1112                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1113                                        "on "DFID ", expected %d\n",
1114                                        mdt_obd_name(info->mti_mdt),
1115                                        rc, PFID(mdt_object_fid(o)),
1116                                        reqbody->mbo_eadatasize - 1);
1117                         /* NULL terminate */
1118                         ((char *)ma->ma_lmm)[rc] = 0;
1119
1120                         /* If the total CDEBUG() size is larger than a page, it
1121                          * will print a warning to the console, avoid this by
1122                          * printing just the last part of the symlink. */
1123                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1124                                print_limit < rc ? "..." : "", print_limit,
1125                                (char *)ma->ma_lmm + rc - print_limit, rc);
1126                         rc = 0;
1127                 }
1128         }
1129
1130         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1131                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1132                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1133                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1134                        repbody->mbo_max_mdsize);
1135         }
1136
1137 #ifdef CONFIG_FS_POSIX_ACL
1138         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1139                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1140                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1141                 if (IS_ERR(nodemap))
1142                         RETURN(PTR_ERR(nodemap));
1143
1144                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1145                 nodemap_putref(nodemap);
1146         }
1147 #endif
1148
1149 out:
1150         if (rc == 0)
1151                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1152
1153         RETURN(rc);
1154 }
1155
1156 static int mdt_getattr(struct tgt_session_info *tsi)
1157 {
1158         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1159         struct mdt_object       *obj = info->mti_object;
1160         struct req_capsule      *pill = info->mti_pill;
1161         struct mdt_body         *reqbody;
1162         struct mdt_body         *repbody;
1163         int rc, rc2;
1164         ENTRY;
1165
1166         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1167         LASSERT(reqbody);
1168         LASSERT(obj != NULL);
1169         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1170
1171         /* Unlike intent case where we need to pre-fill out buffers early on
1172          * in intent policy for ldlm reasons, here we can have a much better
1173          * guess at EA size by just reading it from disk.
1174          * Exceptions are readdir and (missing) directory striping */
1175         /* Readlink */
1176         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1177                 /* No easy way to know how long is the symlink, but it cannot
1178                  * be more than PATH_MAX, so we allocate +1 */
1179                 rc = PATH_MAX + 1;
1180
1181         /* A special case for fs ROOT: getattr there might fetch
1182          * default EA for entire fs, not just for this dir!
1183          */
1184         } else if (lu_fid_eq(mdt_object_fid(obj),
1185                              &info->mti_mdt->mdt_md_root_fid) &&
1186                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1187                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1188                                                                  MDS_GETATTR)) {
1189                 /* Should the default strping be bigger, mdt_fix_reply
1190                  * will reallocate */
1191                 rc = DEF_REP_MD_SIZE;
1192         } else {
1193                 /* Read the actual EA size from disk */
1194                 rc = mdt_attr_get_eabuf_size(info, obj);
1195         }
1196
1197         if (rc < 0)
1198                 GOTO(out, rc = err_serious(rc));
1199
1200         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1201
1202         rc = req_capsule_server_pack(pill);
1203         if (unlikely(rc != 0))
1204                 GOTO(out, rc = err_serious(rc));
1205
1206         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1207         LASSERT(repbody != NULL);
1208         repbody->mbo_eadatasize = 0;
1209         repbody->mbo_aclsize = 0;
1210
1211         rc = mdt_check_ucred(info);
1212         if (unlikely(rc))
1213                 GOTO(out_shrink, rc);
1214
1215         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1216
1217         rc = mdt_getattr_internal(info, obj, 0);
1218         EXIT;
1219 out_shrink:
1220         mdt_client_compatibility(info);
1221         rc2 = mdt_fix_reply(info);
1222         if (rc == 0)
1223                 rc = rc2;
1224 out:
1225         mdt_thread_info_fini(info);
1226         return rc;
1227 }
1228
1229 /**
1230  * Handler of layout intent RPC requiring the layout modification
1231  *
1232  * \param info [in]     thread environment
1233  * \param obj [in]      object
1234  * \param layout [in]   layout intent
1235  *
1236  * \retval 0    on success
1237  * \retval < 0  error code
1238  */
1239 static int mdt_layout_change(struct mdt_thread_info *info,
1240                              struct mdt_object *obj,
1241                              struct layout_intent *layout)
1242 {
1243         /* XXX: to do */
1244         return 0;
1245 }
1246
1247 /**
1248  * Exchange MOF_LOV_CREATED flags between two objects after a
1249  * layout swap. No assumption is made on whether o1 or o2 have
1250  * created objects or not.
1251  *
1252  * \param[in,out] o1    First swap layout object
1253  * \param[in,out] o2    Second swap layout object
1254  */
1255 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1256 {
1257         unsigned int o1_lov_created = o1->mot_lov_created;
1258
1259         mutex_lock(&o1->mot_lov_mutex);
1260         mutex_lock(&o2->mot_lov_mutex);
1261
1262         o1->mot_lov_created = o2->mot_lov_created;
1263         o2->mot_lov_created = o1_lov_created;
1264
1265         mutex_unlock(&o2->mot_lov_mutex);
1266         mutex_unlock(&o1->mot_lov_mutex);
1267 }
1268
1269 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1270 {
1271         struct mdt_thread_info  *info;
1272         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1273         struct obd_export       *exp = req->rq_export;
1274         struct mdt_object       *o1, *o2, *o;
1275         struct mdt_lock_handle  *lh1, *lh2;
1276         struct mdc_swap_layouts *msl;
1277         int                      rc;
1278         ENTRY;
1279
1280         /* client does not support layout lock, so layout swaping
1281          * is disabled.
1282          * FIXME: there is a problem for old clients which don't support
1283          * layout lock yet. If those clients have already opened the file
1284          * they won't be notified at all so that old layout may still be
1285          * used to do IO. This can be fixed after file release is landed by
1286          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1287         if (!exp_connect_layout(exp))
1288                 RETURN(-EOPNOTSUPP);
1289
1290         info = tsi2mdt_info(tsi);
1291
1292         if (info->mti_dlm_req != NULL)
1293                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1294
1295         o1 = info->mti_object;
1296         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1297                                 &info->mti_body->mbo_fid2);
1298         if (IS_ERR(o))
1299                 GOTO(out, rc = PTR_ERR(o));
1300
1301         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1302                 GOTO(put, rc = -ENOENT);
1303
1304         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1305         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1306                 GOTO(put, rc);
1307
1308         if (rc < 0)
1309                 swap(o1, o2);
1310
1311         /* permission check. Make sure the calling process having permission
1312          * to write both files. */
1313         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1314                                 MAY_WRITE);
1315         if (rc < 0)
1316                 GOTO(put, rc);
1317
1318         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1319                                 MAY_WRITE);
1320         if (rc < 0)
1321                 GOTO(put, rc);
1322
1323         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1324         if (msl == NULL)
1325                 GOTO(put, rc = -EPROTO);
1326
1327         lh1 = &info->mti_lh[MDT_LH_NEW];
1328         mdt_lock_reg_init(lh1, LCK_EX);
1329         lh2 = &info->mti_lh[MDT_LH_OLD];
1330         mdt_lock_reg_init(lh2, LCK_EX);
1331
1332         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1333                              MDS_INODELOCK_XATTR);
1334         if (rc < 0)
1335                 GOTO(put, rc);
1336
1337         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1338                              MDS_INODELOCK_XATTR);
1339         if (rc < 0)
1340                 GOTO(unlock1, rc);
1341
1342         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1343                              mdt_object_child(o2), msl->msl_flags);
1344         if (rc < 0)
1345                 GOTO(unlock2, rc);
1346
1347         mdt_swap_lov_flag(o1, o2);
1348
1349 unlock2:
1350         mdt_object_unlock(info, o2, lh2, rc);
1351 unlock1:
1352         mdt_object_unlock(info, o1, lh1, rc);
1353 put:
1354         mdt_object_put(info->mti_env, o);
1355 out:
1356         mdt_thread_info_fini(info);
1357         RETURN(rc);
1358 }
1359
1360 static int mdt_raw_lookup(struct mdt_thread_info *info,
1361                           struct mdt_object *parent,
1362                           const struct lu_name *lname,
1363                           struct ldlm_reply *ldlm_rep)
1364 {
1365         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1366         int              rc;
1367         ENTRY;
1368
1369         LASSERT(!info->mti_cross_ref);
1370
1371         /* Only got the fid of this obj by name */
1372         fid_zero(child_fid);
1373         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1374                         lname, child_fid, &info->mti_spec);
1375         if (rc == 0) {
1376                 struct mdt_body *repbody;
1377
1378                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1379                 repbody->mbo_fid1 = *child_fid;
1380                 repbody->mbo_valid = OBD_MD_FLID;
1381                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1382         } else if (rc == -ENOENT) {
1383                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1384         }
1385
1386         RETURN(rc);
1387 }
1388
1389 /*
1390  * UPDATE lock should be taken against parent, and be released before exit;
1391  * child_bits lock should be taken against child, and be returned back:
1392  *            (1)normal request should release the child lock;
1393  *            (2)intent request will grant the lock to client.
1394  */
1395 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1396                                  struct mdt_lock_handle *lhc,
1397                                  __u64 child_bits,
1398                                  struct ldlm_reply *ldlm_rep)
1399 {
1400         struct ptlrpc_request  *req       = mdt_info_req(info);
1401         struct mdt_body        *reqbody   = NULL;
1402         struct mdt_object      *parent    = info->mti_object;
1403         struct mdt_object      *child;
1404         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1405         struct lu_name         *lname     = NULL;
1406         struct mdt_lock_handle *lhp       = NULL;
1407         struct ldlm_lock       *lock;
1408         bool                    is_resent;
1409         bool                    try_layout;
1410         int                     ma_need = 0;
1411         int                     rc;
1412         ENTRY;
1413
1414         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1415         LASSERT(ergo(is_resent,
1416                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1417
1418         if (parent == NULL)
1419                 RETURN(-ENOENT);
1420
1421         if (info->mti_cross_ref) {
1422                 /* Only getattr on the child. Parent is on another node. */
1423                 mdt_set_disposition(info, ldlm_rep,
1424                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1425                 child = parent;
1426                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1427                        "ldlm_rep = %p\n",
1428                        PFID(mdt_object_fid(child)), ldlm_rep);
1429
1430                 rc = mdt_check_resent_lock(info, child, lhc);
1431                 if (rc < 0) {
1432                         RETURN(rc);
1433                 } else if (rc > 0) {
1434                         mdt_lock_handle_init(lhc);
1435                         mdt_lock_reg_init(lhc, LCK_PR);
1436
1437                         /*
1438                          * Object's name is on another MDS, no lookup or layout
1439                          * lock is needed here but update lock is.
1440                          */
1441                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1442                                         MDS_INODELOCK_LAYOUT);
1443                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1444
1445                         rc = mdt_object_lock(info, child, lhc, child_bits);
1446                         if (rc < 0)
1447                                 RETURN(rc);
1448                 }
1449
1450                 /* Finally, we can get attr for child. */
1451                 if (!mdt_object_exists(child)) {
1452                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1453                                         &child->mot_obj,
1454                                         "remote object doesn't exist.");
1455                         mdt_object_unlock(info, child, lhc, 1);
1456                         RETURN(-ENOENT);
1457                 }
1458
1459                 rc = mdt_getattr_internal(info, child, 0);
1460                 if (unlikely(rc != 0))
1461                         mdt_object_unlock(info, child, lhc, 1);
1462
1463                 RETURN(rc);
1464         }
1465
1466         lname = &info->mti_name;
1467         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1468
1469         if (lu_name_is_valid(lname)) {
1470                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1471                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1472                        PNAME(lname), ldlm_rep);
1473         } else {
1474                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1475                 if (unlikely(reqbody == NULL))
1476                         RETURN(err_serious(-EPROTO));
1477
1478                 *child_fid = reqbody->mbo_fid2;
1479
1480                 if (unlikely(!fid_is_sane(child_fid)))
1481                         RETURN(err_serious(-EINVAL));
1482
1483                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1484                        "ldlm_rep = %p\n",
1485                        PFID(mdt_object_fid(parent)),
1486                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1487         }
1488
1489         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1490
1491         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1492                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1493                                 &parent->mot_obj,
1494                                 "Parent doesn't exist!");
1495                 RETURN(-ESTALE);
1496         }
1497
1498         if (mdt_object_remote(parent)) {
1499                 CERROR("%s: parent "DFID" is on remote target\n",
1500                        mdt_obd_name(info->mti_mdt),
1501                        PFID(mdt_object_fid(parent)));
1502                 RETURN(-EIO);
1503         }
1504
1505         if (lu_name_is_valid(lname)) {
1506                 /* Always allow to lookup ".." */
1507                 if (unlikely(lname->ln_namelen == 2 &&
1508                              lname->ln_name[0] == '.' &&
1509                              lname->ln_name[1] == '.'))
1510                         info->mti_spec.sp_permitted = 1;
1511
1512                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1513                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1514
1515                         RETURN(rc);
1516                 }
1517
1518                 /* step 1: lock parent only if parent is a directory */
1519                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1520                         lhp = &info->mti_lh[MDT_LH_PARENT];
1521                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1522                         rc = mdt_object_lock(info, parent, lhp,
1523                                              MDS_INODELOCK_UPDATE);
1524                         if (unlikely(rc != 0))
1525                                 RETURN(rc);
1526                 }
1527
1528                 /* step 2: lookup child's fid by name */
1529                 fid_zero(child_fid);
1530                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1531                                 child_fid, &info->mti_spec);
1532                 if (rc == -ENOENT)
1533                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1534
1535                 if (rc != 0)
1536                         GOTO(out_parent, rc);
1537         }
1538
1539         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1540
1541         /*
1542          *step 3: find the child object by fid & lock it.
1543          *        regardless if it is local or remote.
1544          *
1545          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1546          *      set parent dir fid the same as child fid in getattr by fid case
1547          *      we should not lu_object_find() the object again, could lead
1548          *      to hung if there is a concurrent unlink destroyed the object.
1549          */
1550         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1551                 mdt_object_get(info->mti_env, parent);
1552                 child = parent;
1553         } else {
1554                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1555                                         child_fid);
1556         }
1557
1558         if (unlikely(IS_ERR(child)))
1559                 GOTO(out_parent, rc = PTR_ERR(child));
1560
1561         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1562         if (!mdt_object_exists(child)) {
1563                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1564                                 &child->mot_obj,
1565                                 "Object doesn't exist!");
1566                 GOTO(out_child, rc = -ENOENT);
1567         }
1568
1569         rc = mdt_check_resent_lock(info, child, lhc);
1570         if (rc < 0) {
1571                 GOTO(out_child, rc);
1572         } else if (rc > 0) {
1573                 mdt_lock_handle_init(lhc);
1574                 mdt_lock_reg_init(lhc, LCK_PR);
1575                 try_layout = false;
1576
1577                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1578                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1579                         struct md_attr *ma = &info->mti_attr;
1580
1581                         ma->ma_valid = 0;
1582                         ma->ma_need = MA_INODE;
1583                         rc = mdt_attr_get_complex(info, child, ma);
1584                         if (unlikely(rc != 0))
1585                                 GOTO(out_child, rc);
1586
1587                         /* If the file has not been changed for some time, we
1588                          * return not only a LOOKUP lock, but also an UPDATE
1589                          * lock and this might save us RPC on later STAT. For
1590                          * directories, it also let negative dentry cache start
1591                          * working for this dir. */
1592                         if (ma->ma_valid & MA_INODE &&
1593                             ma->ma_attr.la_valid & LA_CTIME &&
1594                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1595                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1596                                 child_bits |= MDS_INODELOCK_UPDATE;
1597                 }
1598
1599                 /* layout lock must be granted in a best-effort way
1600                  * for IT operations */
1601                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1602                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1603                     exp_connect_layout(info->mti_exp) &&
1604                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1605                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1606                         /* try to grant layout lock for regular file. */
1607                         try_layout = true;
1608                 }
1609
1610                 rc = 0;
1611                 if (try_layout) {
1612                         child_bits |= MDS_INODELOCK_LAYOUT;
1613                         /* try layout lock, it may fail to be granted due to
1614                          * contention at LOOKUP or UPDATE */
1615                         if (!mdt_object_lock_try(info, child, lhc,
1616                                                  child_bits)) {
1617                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1618                                 LASSERT(child_bits != 0);
1619                                 rc = mdt_object_lock(info, child, lhc,
1620                                                      child_bits);
1621                         } else {
1622                                 ma_need |= MA_LOV;
1623                         }
1624                 } else {
1625                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1626                          * client will enqueue the lock to the remote MDT */
1627                         if (mdt_object_remote(child))
1628                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1629                         rc = mdt_object_lock(info, child, lhc, child_bits);
1630                 }
1631                 if (unlikely(rc != 0))
1632                         GOTO(out_child, rc);
1633         }
1634
1635         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1636
1637         /* finally, we can get attr for child. */
1638         rc = mdt_getattr_internal(info, child, ma_need);
1639         if (unlikely(rc != 0)) {
1640                 mdt_object_unlock(info, child, lhc, 1);
1641         } else if (lock) {
1642                 /* Debugging code. */
1643                 LDLM_DEBUG(lock, "Returning lock to client");
1644                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1645                                          &lock->l_resource->lr_name),
1646                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1647                          PLDLMRES(lock->l_resource),
1648                          PFID(mdt_object_fid(child)));
1649         }
1650         if (lock)
1651                 LDLM_LOCK_PUT(lock);
1652
1653         EXIT;
1654 out_child:
1655         mdt_object_put(info->mti_env, child);
1656 out_parent:
1657         if (lhp)
1658                 mdt_object_unlock(info, parent, lhp, 1);
1659         return rc;
1660 }
1661
1662 /* normal handler: should release the child lock */
1663 static int mdt_getattr_name(struct tgt_session_info *tsi)
1664 {
1665         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1666         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1667         struct mdt_body        *reqbody;
1668         struct mdt_body        *repbody;
1669         int rc, rc2;
1670         ENTRY;
1671
1672         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1673         LASSERT(reqbody != NULL);
1674         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1675         LASSERT(repbody != NULL);
1676
1677         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1678         repbody->mbo_eadatasize = 0;
1679         repbody->mbo_aclsize = 0;
1680
1681         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1682         if (unlikely(rc))
1683                 GOTO(out_shrink, rc);
1684
1685         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1686         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1687                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1688                 lhc->mlh_reg_lh.cookie = 0;
1689         }
1690         mdt_exit_ucred(info);
1691         EXIT;
1692 out_shrink:
1693         mdt_client_compatibility(info);
1694         rc2 = mdt_fix_reply(info);
1695         if (rc == 0)
1696                 rc = rc2;
1697         mdt_thread_info_fini(info);
1698         return rc;
1699 }
1700
1701 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1702                          void *karg, void __user *uarg);
1703
1704 static int mdt_set_info(struct tgt_session_info *tsi)
1705 {
1706         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1707         char                    *key;
1708         void                    *val;
1709         int                      keylen, vallen, rc = 0;
1710
1711         ENTRY;
1712
1713         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1714         if (key == NULL) {
1715                 DEBUG_REQ(D_HA, req, "no set_info key");
1716                 RETURN(err_serious(-EFAULT));
1717         }
1718
1719         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1720                                       RCL_CLIENT);
1721
1722         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1723         if (val == NULL) {
1724                 DEBUG_REQ(D_HA, req, "no set_info val");
1725                 RETURN(err_serious(-EFAULT));
1726         }
1727
1728         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1729                                       RCL_CLIENT);
1730
1731         /* Swab any part of val you need to here */
1732         if (KEY_IS(KEY_READ_ONLY)) {
1733                 spin_lock(&req->rq_export->exp_lock);
1734                 if (*(__u32 *)val)
1735                         *exp_connect_flags_ptr(req->rq_export) |=
1736                                 OBD_CONNECT_RDONLY;
1737                 else
1738                         *exp_connect_flags_ptr(req->rq_export) &=
1739                                 ~OBD_CONNECT_RDONLY;
1740                 spin_unlock(&req->rq_export->exp_lock);
1741         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1742                 struct changelog_setinfo *cs = val;
1743
1744                 if (vallen != sizeof(*cs)) {
1745                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1746                                tgt_name(tsi->tsi_tgt), vallen);
1747                         RETURN(-EINVAL);
1748                 }
1749                 if (ptlrpc_req_need_swab(req)) {
1750                         __swab64s(&cs->cs_recno);
1751                         __swab32s(&cs->cs_id);
1752                 }
1753
1754                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1755                                    vallen, val, NULL);
1756         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1757                 if (vallen > 0)
1758                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1759         } else {
1760                 RETURN(-EINVAL);
1761         }
1762         RETURN(rc);
1763 }
1764
1765 static int mdt_readpage(struct tgt_session_info *tsi)
1766 {
1767         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1768         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1769         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1770         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1771         struct mdt_body         *repbody;
1772         int                      rc;
1773         int                      i;
1774
1775         ENTRY;
1776
1777         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1778                 RETURN(err_serious(-ENOMEM));
1779
1780         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1781         if (repbody == NULL || reqbody == NULL)
1782                 RETURN(err_serious(-EFAULT));
1783
1784         /*
1785          * prepare @rdpg before calling lower layers and transfer itself. Here
1786          * reqbody->size contains offset of where to start to read and
1787          * reqbody->nlink contains number bytes to read.
1788          */
1789         rdpg->rp_hash = reqbody->mbo_size;
1790         if (rdpg->rp_hash != reqbody->mbo_size) {
1791                 CERROR("Invalid hash: %#llx != %#llx\n",
1792                        rdpg->rp_hash, reqbody->mbo_size);
1793                 RETURN(-EFAULT);
1794         }
1795
1796         rdpg->rp_attrs = reqbody->mbo_mode;
1797         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1798                 rdpg->rp_attrs |= LUDA_64BITHASH;
1799         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1800                                 exp_max_brw_size(tsi->tsi_exp));
1801         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1802                           PAGE_SHIFT;
1803         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1804         if (rdpg->rp_pages == NULL)
1805                 RETURN(-ENOMEM);
1806
1807         for (i = 0; i < rdpg->rp_npages; ++i) {
1808                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1809                 if (rdpg->rp_pages[i] == NULL)
1810                         GOTO(free_rdpg, rc = -ENOMEM);
1811         }
1812
1813         /* call lower layers to fill allocated pages with directory data */
1814         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1815         if (rc < 0)
1816                 GOTO(free_rdpg, rc);
1817
1818         /* send pages to client */
1819         rc = tgt_sendpage(tsi, rdpg, rc);
1820
1821         EXIT;
1822 free_rdpg:
1823
1824         for (i = 0; i < rdpg->rp_npages; i++)
1825                 if (rdpg->rp_pages[i] != NULL)
1826                         __free_page(rdpg->rp_pages[i]);
1827         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1828
1829         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1830                 RETURN(0);
1831
1832         return rc;
1833 }
1834
1835 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1836 {
1837         struct lu_ucred *uc = mdt_ucred_check(info);
1838         struct lu_attr *attr = &info->mti_attr.ma_attr;
1839
1840         if (uc == NULL)
1841                 return -EINVAL;
1842
1843         if (op != REINT_SETATTR) {
1844                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1845                         attr->la_uid = uc->uc_fsuid;
1846                 /* for S_ISGID, inherit gid from his parent, such work will be
1847                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1848                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1849                         attr->la_gid = uc->uc_fsgid;
1850         }
1851
1852         return 0;
1853 }
1854
1855 static int mdt_reint_internal(struct mdt_thread_info *info,
1856                               struct mdt_lock_handle *lhc,
1857                               __u32 op)
1858 {
1859         struct req_capsule      *pill = info->mti_pill;
1860         struct mdt_body         *repbody;
1861         int                      rc = 0, rc2;
1862
1863         ENTRY;
1864
1865         rc = mdt_reint_unpack(info, op);
1866         if (rc != 0) {
1867                 CERROR("Can't unpack reint, rc %d\n", rc);
1868                 RETURN(err_serious(rc));
1869         }
1870
1871         /* for replay (no_create) lmm is not needed, client has it already */
1872         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1873                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1874                                      DEF_REP_MD_SIZE);
1875
1876         /* llog cookies are always 0, the field is kept for compatibility */
1877         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1878                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1879
1880         rc = req_capsule_server_pack(pill);
1881         if (rc != 0) {
1882                 CERROR("Can't pack response, rc %d\n", rc);
1883                 RETURN(err_serious(rc));
1884         }
1885
1886         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1887                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1888                 LASSERT(repbody);
1889                 repbody->mbo_eadatasize = 0;
1890                 repbody->mbo_aclsize = 0;
1891         }
1892
1893         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1894
1895         /* for replay no cookkie / lmm need, because client have this already */
1896         if (info->mti_spec.no_create)
1897                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1898                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1899
1900         rc = mdt_init_ucred_reint(info);
1901         if (rc)
1902                 GOTO(out_shrink, rc);
1903
1904         rc = mdt_fix_attr_ucred(info, op);
1905         if (rc != 0)
1906                 GOTO(out_ucred, rc = err_serious(rc));
1907
1908         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1909         if (rc < 0) {
1910                 GOTO(out_ucred, rc);
1911         } else if (rc == 1) {
1912                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1913                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1914                 GOTO(out_ucred, rc);
1915         }
1916         rc = mdt_reint_rec(info, lhc);
1917         EXIT;
1918 out_ucred:
1919         mdt_exit_ucred(info);
1920 out_shrink:
1921         mdt_client_compatibility(info);
1922         rc2 = mdt_fix_reply(info);
1923         if (rc == 0)
1924                 rc = rc2;
1925         return rc;
1926 }
1927
1928 static long mdt_reint_opcode(struct ptlrpc_request *req,
1929                              const struct req_format **fmt)
1930 {
1931         struct mdt_device       *mdt;
1932         struct mdt_rec_reint    *rec;
1933         long                     opc;
1934
1935         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1936         if (rec != NULL) {
1937                 opc = rec->rr_opcode;
1938                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1939                 if (opc < REINT_MAX && fmt[opc] != NULL)
1940                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1941                 else {
1942                         mdt = mdt_exp2dev(req->rq_export);
1943                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1944                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1945                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1946                         opc = err_serious(-EFAULT);
1947                 }
1948         } else {
1949                 opc = err_serious(-EFAULT);
1950         }
1951         return opc;
1952 }
1953
1954 static int mdt_reint(struct tgt_session_info *tsi)
1955 {
1956         long opc;
1957         int  rc;
1958         static const struct req_format *reint_fmts[REINT_MAX] = {
1959                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1960                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1961                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1962                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1963                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1964                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1965                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1966                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1967                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1968         };
1969
1970         ENTRY;
1971
1972         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1973         if (opc >= 0) {
1974                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1975                 /*
1976                  * No lock possible here from client to pass it to reint code
1977                  * path.
1978                  */
1979                 rc = mdt_reint_internal(info, NULL, opc);
1980                 mdt_thread_info_fini(info);
1981         } else {
1982                 rc = opc;
1983         }
1984
1985         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1986         RETURN(rc);
1987 }
1988
1989 /* this should sync the whole device */
1990 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1991 {
1992         struct dt_device *dt = mdt->mdt_bottom;
1993         int rc;
1994         ENTRY;
1995
1996         rc = dt->dd_ops->dt_sync(env, dt);
1997         RETURN(rc);
1998 }
1999
2000 /* this should sync this object */
2001 static int mdt_object_sync(struct mdt_thread_info *info)
2002 {
2003         struct md_object *next;
2004         int rc;
2005         ENTRY;
2006
2007         if (!mdt_object_exists(info->mti_object)) {
2008                 CWARN("%s: non existing object "DFID": rc = %d\n",
2009                       mdt_obd_name(info->mti_mdt),
2010                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
2011                 RETURN(-ESTALE);
2012         }
2013         next = mdt_object_child(info->mti_object);
2014         rc = mo_object_sync(info->mti_env, next);
2015
2016         RETURN(rc);
2017 }
2018
2019 static int mdt_sync(struct tgt_session_info *tsi)
2020 {
2021         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2022         struct req_capsule      *pill = tsi->tsi_pill;
2023         struct mdt_body         *body;
2024         int                      rc;
2025
2026         ENTRY;
2027
2028         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2029                 RETURN(err_serious(-ENOMEM));
2030
2031         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2032                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2033         } else {
2034                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2035
2036                 /* sync an object */
2037                 rc = mdt_object_sync(info);
2038                 if (rc == 0) {
2039                         const struct lu_fid *fid;
2040                         struct lu_attr *la = &info->mti_attr.ma_attr;
2041
2042                         info->mti_attr.ma_need = MA_INODE;
2043                         info->mti_attr.ma_valid = 0;
2044                         rc = mdt_attr_get_complex(info, info->mti_object,
2045                                                   &info->mti_attr);
2046                         if (rc == 0) {
2047                                 body = req_capsule_server_get(pill,
2048                                                               &RMF_MDT_BODY);
2049                                 fid = mdt_object_fid(info->mti_object);
2050                                 mdt_pack_attr2body(info, body, la, fid);
2051                         }
2052                 }
2053                 mdt_thread_info_fini(info);
2054         }
2055         if (rc == 0)
2056                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2057
2058         RETURN(rc);
2059 }
2060
2061 /*
2062  * Handle quota control requests to consult current usage/limit, but also
2063  * to configure quota enforcement
2064  */
2065 static int mdt_quotactl(struct tgt_session_info *tsi)
2066 {
2067         struct obd_export       *exp  = tsi->tsi_exp;
2068         struct req_capsule      *pill = tsi->tsi_pill;
2069         struct obd_quotactl     *oqctl, *repoqc;
2070         int                      id, rc;
2071         struct mdt_device       *mdt = mdt_exp2dev(exp);
2072         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2073         struct lu_nodemap       *nodemap;
2074         ENTRY;
2075
2076         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2077         if (oqctl == NULL)
2078                 RETURN(err_serious(-EPROTO));
2079
2080         rc = req_capsule_server_pack(pill);
2081         if (rc)
2082                 RETURN(err_serious(rc));
2083
2084         nodemap = nodemap_get_from_exp(exp);
2085         if (IS_ERR(nodemap))
2086                 RETURN(PTR_ERR(nodemap));
2087
2088         switch (oqctl->qc_cmd) {
2089                 /* master quotactl */
2090         case Q_SETINFO:
2091         case Q_SETQUOTA:
2092                 if (!nodemap_can_setquota(nodemap))
2093                         GOTO(out_nodemap, rc = -EPERM);
2094         case Q_GETINFO:
2095         case Q_GETQUOTA:
2096                 if (qmt == NULL)
2097                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2098                 /* slave quotactl */
2099         case Q_GETOINFO:
2100         case Q_GETOQUOTA:
2101                 break;
2102         default:
2103                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2104                 GOTO(out_nodemap, rc = -EFAULT);
2105         }
2106
2107         id = oqctl->qc_id;
2108         if (oqctl->qc_type == USRQUOTA)
2109                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2110                                     NODEMAP_CLIENT_TO_FS, id);
2111         else if (oqctl->qc_type == GRPQUOTA)
2112                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2113                                     NODEMAP_CLIENT_TO_FS, id);
2114
2115         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2116         if (repoqc == NULL)
2117                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2118
2119         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2120                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2121
2122         if (oqctl->qc_id != id)
2123                 swap(oqctl->qc_id, id);
2124
2125         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2126                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2127                         RETURN(-EINPROGRESS);
2128         }
2129
2130         switch (oqctl->qc_cmd) {
2131
2132         case Q_GETINFO:
2133         case Q_SETINFO:
2134         case Q_SETQUOTA:
2135         case Q_GETQUOTA:
2136                 /* forward quotactl request to QMT */
2137                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2138                 break;
2139
2140         case Q_GETOINFO:
2141         case Q_GETOQUOTA:
2142                 /* slave quotactl */
2143                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2144                                    oqctl);
2145                 break;
2146
2147         default:
2148                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2149                 GOTO(out_nodemap, rc = -EFAULT);
2150         }
2151
2152         if (oqctl->qc_id != id)
2153                 swap(oqctl->qc_id, id);
2154
2155         *repoqc = *oqctl;
2156
2157         EXIT;
2158
2159 out_nodemap:
2160         nodemap_putref(nodemap);
2161
2162         return rc;
2163 }
2164
2165 /** clone llog ctxt from child (mdd)
2166  * This allows remote llog (replicator) access.
2167  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2168  * context was originally set up, or we can handle them directly.
2169  * I choose the latter, but that means I need any llog
2170  * contexts set up by child to be accessable by the mdt.  So we clone the
2171  * context into our context list here.
2172  */
2173 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2174                                int idx)
2175 {
2176         struct md_device  *next = mdt->mdt_child;
2177         struct llog_ctxt *ctxt;
2178         int rc;
2179
2180         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2181                 return 0;
2182
2183         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2184         if (rc || ctxt == NULL) {
2185                 return 0;
2186         }
2187
2188         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2189         if (rc)
2190                 CERROR("Can't set mdt ctxt %d\n", rc);
2191
2192         return rc;
2193 }
2194
2195 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2196                                  struct mdt_device *mdt, int idx)
2197 {
2198         struct llog_ctxt *ctxt;
2199
2200         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2201         if (ctxt == NULL)
2202                 return 0;
2203         /* Put once for the get we just did, and once for the clone */
2204         llog_ctxt_put(ctxt);
2205         llog_ctxt_put(ctxt);
2206         return 0;
2207 }
2208
2209 /*
2210  * sec context handlers
2211  */
2212 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2213 {
2214         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2215
2216         return 0;
2217 }
2218
2219 /*
2220  * quota request handlers
2221  */
2222 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2223 {
2224         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2225         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2226         int                      rc;
2227         ENTRY;
2228
2229         if (qmt == NULL)
2230                 RETURN(err_serious(-EOPNOTSUPP));
2231
2232         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2233         RETURN(rc);
2234 }
2235
2236 struct mdt_object *mdt_object_new(const struct lu_env *env,
2237                                   struct mdt_device *d,
2238                                   const struct lu_fid *f)
2239 {
2240         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2241         struct lu_object *o;
2242         struct mdt_object *m;
2243         ENTRY;
2244
2245         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2246         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2247         if (unlikely(IS_ERR(o)))
2248                 m = (struct mdt_object *)o;
2249         else
2250                 m = mdt_obj(o);
2251         RETURN(m);
2252 }
2253
2254 struct mdt_object *mdt_object_find(const struct lu_env *env,
2255                                    struct mdt_device *d,
2256                                    const struct lu_fid *f)
2257 {
2258         struct lu_object *o;
2259         struct mdt_object *m;
2260         ENTRY;
2261
2262         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2263         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2264         if (unlikely(IS_ERR(o)))
2265                 m = (struct mdt_object *)o;
2266         else
2267                 m = mdt_obj(o);
2268
2269         RETURN(m);
2270 }
2271
2272 /**
2273  * Asyncronous commit for mdt device.
2274  *
2275  * Pass asynchonous commit call down the MDS stack.
2276  *
2277  * \param env environment
2278  * \param mdt the mdt device
2279  */
2280 static void mdt_device_commit_async(const struct lu_env *env,
2281                                     struct mdt_device *mdt)
2282 {
2283         struct dt_device *dt = mdt->mdt_bottom;
2284         int rc;
2285         ENTRY;
2286
2287         rc = dt->dd_ops->dt_commit_async(env, dt);
2288         if (unlikely(rc != 0))
2289                 CWARN("%s: async commit start failed: rc = %d\n",
2290                       mdt_obd_name(mdt), rc);
2291         atomic_inc(&mdt->mdt_async_commit_count);
2292         EXIT;
2293 }
2294
2295 /**
2296  * Mark the lock as "synchonous".
2297  *
2298  * Mark the lock to deffer transaction commit to the unlock time.
2299  *
2300  * \param lock the lock to mark as "synchonous"
2301  *
2302  * \see mdt_is_lock_sync
2303  * \see mdt_save_lock
2304  */
2305 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2306 {
2307         lock->l_ast_data = (void*)1;
2308 }
2309
2310 /**
2311  * Check whehter the lock "synchonous" or not.
2312  *
2313  * \param lock the lock to check
2314  * \retval 1 the lock is "synchonous"
2315  * \retval 0 the lock isn't "synchronous"
2316  *
2317  * \see mdt_set_lock_sync
2318  * \see mdt_save_lock
2319  */
2320 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2321 {
2322         return lock->l_ast_data != NULL;
2323 }
2324
2325 /**
2326  * Blocking AST for mdt locks.
2327  *
2328  * Starts transaction commit if in case of COS lock conflict or
2329  * deffers such a commit to the mdt_save_lock.
2330  *
2331  * \param lock the lock which blocks a request or cancelling lock
2332  * \param desc unused
2333  * \param data unused
2334  * \param flag indicates whether this cancelling or blocking callback
2335  * \retval 0
2336  * \see ldlm_blocking_ast_nocheck
2337  */
2338 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2339                      void *data, int flag)
2340 {
2341         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2342         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2343         int rc;
2344         ENTRY;
2345
2346         if (flag == LDLM_CB_CANCELING)
2347                 RETURN(0);
2348
2349         lock_res_and_lock(lock);
2350         if (lock->l_blocking_ast != mdt_blocking_ast) {
2351                 unlock_res_and_lock(lock);
2352                 RETURN(0);
2353         }
2354         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2355             lock->l_blocking_lock != NULL) {
2356                 if (mdt_cos_is_enabled(mdt) &&
2357                     lock->l_client_cookie !=
2358                     lock->l_blocking_lock->l_client_cookie)
2359                         mdt_set_lock_sync(lock);
2360                 else if (mdt_slc_is_enabled(mdt) &&
2361                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2362                         mdt_set_lock_sync(lock);
2363         }
2364         rc = ldlm_blocking_ast_nocheck(lock);
2365
2366         /* There is no lock conflict if l_blocking_lock == NULL,
2367          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2368          * when the last reference to a local lock was released */
2369         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2370                 struct lu_env env;
2371
2372                 rc = lu_env_init(&env, LCT_LOCAL);
2373                 if (unlikely(rc != 0))
2374                         CWARN("%s: lu_env initialization failed, cannot "
2375                               "start asynchronous commit: rc = %d\n",
2376                               obd->obd_name, rc);
2377                 else
2378                         mdt_device_commit_async(&env, mdt);
2379                 lu_env_fini(&env);
2380         }
2381         RETURN(rc);
2382 }
2383
2384 /*
2385  * Blocking AST for cross-MDT lock
2386  *
2387  * Discard lock from uncommitted_slc_locks and cancel it.
2388  *
2389  * \param lock  the lock which blocks a request or cancelling lock
2390  * \param desc  unused
2391  * \param data  unused
2392  * \param flag  indicates whether this cancelling or blocking callback
2393  * \retval      0 on success
2394  * \retval      negative number on error
2395  */
2396 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2397                             void *data, int flag)
2398 {
2399         int rc = 0;
2400         ENTRY;
2401
2402         switch (flag) {
2403         case LDLM_CB_BLOCKING: {
2404                 struct lustre_handle lockh;
2405
2406                 ldlm_lock2handle(lock, &lockh);
2407                 rc = ldlm_cli_cancel(&lockh,
2408                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2409                 if (rc < 0) {
2410                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2411                         RETURN(rc);
2412                 }
2413                 break;
2414         }
2415         case LDLM_CB_CANCELING: {
2416                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2417                 struct mdt_device *mdt =
2418                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2419
2420                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2421
2422                 /* discard slc lock here so that it can be cleaned anytime,
2423                  * especially for cleanup_resource() */
2424                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2425
2426                 /* once we cache lock, l_ast_data is set to mdt_object */
2427                 if (lock->l_ast_data != NULL) {
2428                         struct mdt_object *mo = lock->l_ast_data;
2429                         struct lu_env env;
2430
2431                         rc = lu_env_init(&env, LCT_MD_THREAD);
2432                         if (unlikely(rc != 0)) {
2433                                 CWARN("%s: lu_env initialization failed, object"
2434                                       "%p "DFID" is leaked!\n",
2435                                       obd->obd_name, mo,
2436                                       PFID(mdt_object_fid(mo)));
2437                                 RETURN(rc);
2438                         }
2439
2440                         if (lock->l_policy_data.l_inodebits.bits &
2441                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2442                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2443                                 mo->mot_cache_attr = 0;
2444                         }
2445                         mdt_object_put(&env, mo);
2446                         lu_env_fini(&env);
2447                 }
2448                 break;
2449         }
2450         default:
2451                 LBUG();
2452         }
2453
2454         RETURN(rc);
2455 }
2456
2457 int mdt_check_resent_lock(struct mdt_thread_info *info,
2458                           struct mdt_object *mo,
2459                           struct mdt_lock_handle *lhc)
2460 {
2461         /* the lock might already be gotten in ldlm_handle_enqueue() */
2462         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2463                 struct ptlrpc_request *req = mdt_info_req(info);
2464                 struct ldlm_lock      *lock;
2465
2466                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2467                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2468                 if (lock == NULL) {
2469                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2470                          * a resend case, however, it could be already destroyed
2471                          * due to client eviction or a raced cancel RPC. */
2472                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2473                                           lhc->mlh_reg_lh.cookie);
2474                         RETURN(-ESTALE);
2475                 }
2476
2477                 if (!fid_res_name_eq(mdt_object_fid(mo),
2478                                      &lock->l_resource->lr_name)) {
2479                         CWARN("%s: Although resent, but still not "
2480                               "get child lock:"DFID"\n",
2481                               info->mti_exp->exp_obd->obd_name,
2482                               PFID(mdt_object_fid(mo)));
2483                         LDLM_LOCK_PUT(lock);
2484                         RETURN(-EPROTO);
2485                 }
2486                 LDLM_LOCK_PUT(lock);
2487                 return 0;
2488         }
2489         return 1;
2490 }
2491
2492 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2493                            const struct lu_fid *fid, struct lustre_handle *lh,
2494                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2495                            bool cache)
2496 {
2497         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2498         union ldlm_policy_data *policy = &mti->mti_policy;
2499         struct ldlm_res_id *res_id = &mti->mti_res_id;
2500         int rc = 0;
2501         ENTRY;
2502
2503         LASSERT(mdt_object_remote(o));
2504
2505         fid_build_reg_res_name(fid, res_id);
2506
2507         memset(einfo, 0, sizeof(*einfo));
2508         einfo->ei_type = LDLM_IBITS;
2509         einfo->ei_mode = mode;
2510         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2511         einfo->ei_cb_cp = ldlm_completion_ast;
2512         einfo->ei_enq_slave = 0;
2513         einfo->ei_res_id = res_id;
2514         if (nonblock)
2515                 einfo->ei_nonblock = 1;
2516         if (cache) {
2517                 /*
2518                  * if we cache lock, couple lock with mdt_object, so that object
2519                  * can be easily found in lock ASTs.
2520                  */
2521                 mdt_object_get(mti->mti_env, o);
2522                 einfo->ei_cbdata = o;
2523         }
2524
2525         memset(policy, 0, sizeof(*policy));
2526         policy->l_inodebits.bits = ibits;
2527
2528         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2529                             policy);
2530         if (rc < 0 && cache) {
2531                 mdt_object_put(mti->mti_env, o);
2532                 einfo->ei_cbdata = NULL;
2533         }
2534         RETURN(rc);
2535 }
2536
2537 static int mdt_object_local_lock(struct mdt_thread_info *info,
2538                                  struct mdt_object *o,
2539                                  struct mdt_lock_handle *lh, __u64 ibits,
2540                                  bool nonblock, bool cos_incompat)
2541 {
2542         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2543         union ldlm_policy_data *policy = &info->mti_policy;
2544         struct ldlm_res_id *res_id = &info->mti_res_id;
2545         __u64 dlmflags = 0;
2546         int rc;
2547         ENTRY;
2548
2549         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2550         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2551         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2552         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2553
2554         if (cos_incompat) {
2555                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2556                         lh->mlh_reg_mode == LCK_EX);
2557                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2558         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2559                 dlmflags |= LDLM_FL_COS_ENABLED;
2560         }
2561
2562         /* Only enqueue LOOKUP lock for remote object */
2563         if (mdt_object_remote(o))
2564                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2565
2566         if (lh->mlh_type == MDT_PDO_LOCK) {
2567                 /* check for exists after object is locked */
2568                 if (mdt_object_exists(o) == 0) {
2569                         /* Non-existent object shouldn't have PDO lock */
2570                         RETURN(-ESTALE);
2571                 } else {
2572                         /* Non-dir object shouldn't have PDO lock */
2573                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2574                                 RETURN(-ENOTDIR);
2575                 }
2576         }
2577
2578         memset(policy, 0, sizeof(*policy));
2579         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2580
2581         dlmflags |= LDLM_FL_ATOMIC_CB;
2582         if (nonblock)
2583                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2584
2585         /*
2586          * Take PDO lock on whole directory and build correct @res_id for lock
2587          * on part of directory.
2588          */
2589         if (lh->mlh_pdo_hash != 0) {
2590                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2591                 mdt_lock_pdo_mode(info, o, lh);
2592                 if (lh->mlh_pdo_mode != LCK_NL) {
2593                         /*
2594                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2595                          * is never going to be sent to client and we do not
2596                          * want it slowed down due to possible cancels.
2597                          */
2598                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2599                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2600                                           policy, res_id, dlmflags,
2601                                           info->mti_exp == NULL ? NULL :
2602                                           &info->mti_exp->exp_handle.h_cookie);
2603                         if (unlikely(rc != 0))
2604                                 GOTO(out_unlock, rc);
2605                 }
2606
2607                 /*
2608                  * Finish res_id initializing by name hash marking part of
2609                  * directory which is taking modification.
2610                  */
2611                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2612         }
2613
2614         policy->l_inodebits.bits = ibits;
2615
2616         /*
2617          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2618          * going to be sent to client. If it is - mdt_intent_policy() path will
2619          * fix it up and turn FL_LOCAL flag off.
2620          */
2621         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2622                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2623                           info->mti_exp == NULL ? NULL :
2624                           &info->mti_exp->exp_handle.h_cookie);
2625 out_unlock:
2626         if (rc != 0)
2627                 mdt_object_unlock(info, o, lh, 1);
2628         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2629                    lh->mlh_pdo_hash != 0 &&
2630                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2631                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2632
2633         RETURN(rc);
2634 }
2635
2636 static int
2637 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2638                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2639                          bool cos_incompat)
2640 {
2641         struct mdt_lock_handle *local_lh = NULL;
2642         int rc;
2643         ENTRY;
2644
2645         if (!mdt_object_remote(o)) {
2646                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2647                                            cos_incompat);
2648                 RETURN(rc);
2649         }
2650
2651         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2652         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2653                    MDS_INODELOCK_XATTR);
2654
2655         /* Only enqueue LOOKUP lock for remote object */
2656         if (ibits & MDS_INODELOCK_LOOKUP) {
2657                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2658                                            nonblock, cos_incompat);
2659                 if (rc != ELDLM_OK)
2660                         RETURN(rc);
2661
2662                 local_lh = lh;
2663         }
2664
2665         if (ibits & MDS_INODELOCK_UPDATE) {
2666                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2667                  * enqueue RPC can only request 1 lock, to avoid extra
2668                  * RPC, so it will instead enqueue EX lock for remote
2669                  * object anyway XXX*/
2670                 if (lh->mlh_type == MDT_PDO_LOCK &&
2671                     lh->mlh_pdo_hash != 0) {
2672                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2673                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2674                                PFID(mdt_object_fid(o)));
2675                         lh->mlh_pdo_hash = 0;
2676                         lh->mlh_rreg_mode = LCK_EX;
2677                         lh->mlh_type = MDT_REG_LOCK;
2678                 }
2679                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2680                                             &lh->mlh_rreg_lh,
2681                                             lh->mlh_rreg_mode,
2682                                             MDS_INODELOCK_UPDATE, nonblock,
2683                                             false);
2684                 if (rc != ELDLM_OK) {
2685                         if (local_lh != NULL)
2686                                 mdt_object_unlock(info, o, local_lh, rc);
2687                         RETURN(rc);
2688                 }
2689         }
2690
2691         RETURN(0);
2692 }
2693
2694 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2695                     struct mdt_lock_handle *lh, __u64 ibits)
2696 {
2697         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2698 }
2699
2700 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2701                           struct mdt_lock_handle *lh, __u64 ibits,
2702                           bool cos_incompat)
2703 {
2704         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2705         return mdt_object_lock_internal(info, o, lh, ibits, false,
2706                                         cos_incompat);
2707 }
2708
2709 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2710                         struct mdt_lock_handle *lh, __u64 ibits)
2711 {
2712         struct mdt_lock_handle tmp = *lh;
2713         int rc;
2714
2715         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2716         if (rc == 0)
2717                 *lh = tmp;
2718
2719         return rc == 0;
2720 }
2721
2722 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2723                               struct mdt_object *o, struct mdt_lock_handle *lh,
2724                               __u64 ibits, bool cos_incompat)
2725 {
2726         struct mdt_lock_handle tmp = *lh;
2727         int rc;
2728
2729         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2730         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2731         if (rc == 0)
2732                 *lh = tmp;
2733
2734         return rc == 0;
2735 }
2736
2737 /**
2738  * Save a lock within request object.
2739  *
2740  * Keep the lock referenced until whether client ACK or transaction
2741  * commit happens or release the lock immediately depending on input
2742  * parameters. If COS is ON, a write lock is converted to COS lock
2743  * before saving.
2744  *
2745  * \param info thead info object
2746  * \param h lock handle
2747  * \param mode lock mode
2748  * \param decref force immediate lock releasing
2749  */
2750 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2751                           enum ldlm_mode mode, int decref)
2752 {
2753         ENTRY;
2754
2755         if (lustre_handle_is_used(h)) {
2756                 if (decref || !info->mti_has_trans ||
2757                     !(mode & (LCK_PW | LCK_EX))) {
2758                         mdt_fid_unlock(h, mode);
2759                 } else {
2760                         struct mdt_device *mdt = info->mti_mdt;
2761                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2762                         struct ptlrpc_request *req = mdt_info_req(info);
2763                         int cos;
2764
2765                         cos = (mdt_cos_is_enabled(mdt) ||
2766                                mdt_slc_is_enabled(mdt));
2767
2768                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2769                                  h->cookie);
2770
2771                         /* there is no request if mdt_object_unlock() is called
2772                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2773                         if (likely(req != NULL)) {
2774                                 CDEBUG(D_HA, "request = %p reply state = %p"
2775                                        " transno = %lld\n", req,
2776                                        req->rq_reply_state, req->rq_transno);
2777                                 if (cos) {
2778                                         ldlm_lock_downgrade(lock, LCK_COS);
2779                                         mode = LCK_COS;
2780                                 }
2781                                 ptlrpc_save_lock(req, h, mode, cos);
2782                         } else {
2783                                 mdt_fid_unlock(h, mode);
2784                         }
2785                         if (mdt_is_lock_sync(lock)) {
2786                                 CDEBUG(D_HA, "found sync-lock,"
2787                                        " async commit started\n");
2788                                 mdt_device_commit_async(info->mti_env,
2789                                                         mdt);
2790                         }
2791                         LDLM_LOCK_PUT(lock);
2792                 }
2793                 h->cookie = 0ull;
2794         }
2795
2796         EXIT;
2797 }
2798
2799 /**
2800  * Save cross-MDT lock in uncommitted_slc_locks
2801  *
2802  * Keep the lock referenced until transaction commit happens or release the lock
2803  * immediately depending on input parameters.
2804  *
2805  * \param info thead info object
2806  * \param h lock handle
2807  * \param mode lock mode
2808  * \param decref force immediate lock releasing
2809  */
2810 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2811                                  struct mdt_object *o, struct lustre_handle *h,
2812                                  enum ldlm_mode mode, int decref)
2813 {
2814         ENTRY;
2815
2816         if (lustre_handle_is_used(h)) {
2817                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2818
2819                 if (o != NULL &&
2820                     (lock->l_policy_data.l_inodebits.bits &
2821                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2822                         mo_invalidate(info->mti_env, mdt_object_child(o));
2823
2824                 if (decref || !info->mti_has_trans ||
2825                     !(mode & (LCK_PW | LCK_EX))) {
2826                         ldlm_lock_decref_and_cancel(h, mode);
2827                         LDLM_LOCK_PUT(lock);
2828                 } else {
2829                         struct ptlrpc_request *req = mdt_info_req(info);
2830
2831                         LASSERT(req != NULL);
2832                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
2833                                           req->rq_transno);
2834                         ldlm_lock_decref(h, mode);
2835                 }
2836                 h->cookie = 0ull;
2837         }
2838
2839         EXIT;
2840 }
2841
2842 /**
2843  * Unlock mdt object.
2844  *
2845  * Immeditely release the regular lock and the PDO lock or save the
2846  * lock in request and keep them referenced until client ACK or
2847  * transaction commit.
2848  *
2849  * \param info thread info object
2850  * \param o mdt object
2851  * \param lh mdt lock handle referencing regular and PDO locks
2852  * \param decref force immediate lock releasing
2853  *
2854  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
2855  */
2856 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2857                        struct mdt_lock_handle *lh, int decref)
2858 {
2859         ENTRY;
2860
2861         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2862         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2863         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2864                              decref);
2865
2866         EXIT;
2867 }
2868
2869 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2870                                         const struct lu_fid *f,
2871                                         struct mdt_lock_handle *lh,
2872                                         __u64 ibits)
2873 {
2874         struct mdt_object *o;
2875
2876         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2877         if (!IS_ERR(o)) {
2878                 int rc;
2879
2880                 rc = mdt_object_lock(info, o, lh, ibits);
2881                 if (rc != 0) {
2882                         mdt_object_put(info->mti_env, o);
2883                         o = ERR_PTR(rc);
2884                 }
2885         }
2886         return o;
2887 }
2888
2889 void mdt_object_unlock_put(struct mdt_thread_info * info,
2890                            struct mdt_object * o,
2891                            struct mdt_lock_handle *lh,
2892                            int decref)
2893 {
2894         mdt_object_unlock(info, o, lh, decref);
2895         mdt_object_put(info->mti_env, o);
2896 }
2897
2898 /*
2899  * Generic code handling requests that have struct mdt_body passed in:
2900  *
2901  *  - extract mdt_body from request and save it in @info, if present;
2902  *
2903  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2904  *  @info;
2905  *
2906  *  - if HABEO_CORPUS flag is set for this request type check whether object
2907  *  actually exists on storage (lu_object_exists()).
2908  *
2909  */
2910 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2911 {
2912         const struct mdt_body    *body;
2913         struct mdt_object        *obj;
2914         const struct lu_env      *env;
2915         struct req_capsule       *pill;
2916         int                       rc;
2917         ENTRY;
2918
2919         env = info->mti_env;
2920         pill = info->mti_pill;
2921
2922         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2923         if (body == NULL)
2924                 RETURN(-EFAULT);
2925
2926         if (!(body->mbo_valid & OBD_MD_FLID))
2927                 RETURN(0);
2928
2929         if (!fid_is_sane(&body->mbo_fid1)) {
2930                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2931                 RETURN(-EINVAL);
2932         }
2933
2934         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2935         if (!IS_ERR(obj)) {
2936                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2937                         mdt_object_put(env, obj);
2938                         rc = -ENOENT;
2939                 } else {
2940                         info->mti_object = obj;
2941                         rc = 0;
2942                 }
2943         } else
2944                 rc = PTR_ERR(obj);
2945
2946         RETURN(rc);
2947 }
2948
2949 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2950 {
2951         struct req_capsule *pill = info->mti_pill;
2952         int rc;
2953         ENTRY;
2954
2955         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2956                 rc = mdt_body_unpack(info, flags);
2957         else
2958                 rc = 0;
2959
2960         if (rc == 0 && (flags & HABEO_REFERO)) {
2961                 /* Pack reply. */
2962                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2963                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2964                                              DEF_REP_MD_SIZE);
2965                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2966                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2967                                              RCL_SERVER, 0);
2968
2969                 rc = req_capsule_server_pack(pill);
2970         }
2971         RETURN(rc);
2972 }
2973
2974 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2975 {
2976         lh->mlh_type = MDT_NUL_LOCK;
2977         lh->mlh_reg_lh.cookie = 0ull;
2978         lh->mlh_reg_mode = LCK_MINMODE;
2979         lh->mlh_pdo_lh.cookie = 0ull;
2980         lh->mlh_pdo_mode = LCK_MINMODE;
2981         lh->mlh_rreg_lh.cookie = 0ull;
2982         lh->mlh_rreg_mode = LCK_MINMODE;
2983 }
2984
2985 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2986 {
2987         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2988         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2989 }
2990
2991 /*
2992  * Initialize fields of struct mdt_thread_info. Other fields are left in
2993  * uninitialized state, because it's too expensive to zero out whole
2994  * mdt_thread_info (> 1K) on each request arrival.
2995  */
2996 void mdt_thread_info_init(struct ptlrpc_request *req,
2997                           struct mdt_thread_info *info)
2998 {
2999         int i;
3000
3001         info->mti_pill = &req->rq_pill;
3002
3003         /* lock handle */
3004         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3005                 mdt_lock_handle_init(&info->mti_lh[i]);
3006
3007         /* mdt device: it can be NULL while CONNECT */
3008         if (req->rq_export) {
3009                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3010                 info->mti_exp = req->rq_export;
3011         } else
3012                 info->mti_mdt = NULL;
3013         info->mti_env = req->rq_svc_thread->t_env;
3014         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3015
3016         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3017         info->mti_big_buf = LU_BUF_NULL;
3018         info->mti_body = NULL;
3019         info->mti_object = NULL;
3020         info->mti_dlm_req = NULL;
3021         info->mti_has_trans = 0;
3022         info->mti_cross_ref = 0;
3023         info->mti_opdata = 0;
3024         info->mti_big_lmm_used = 0;
3025
3026         info->mti_spec.no_create = 0;
3027         info->mti_spec.sp_rm_entry = 0;
3028         info->mti_spec.sp_permitted = 0;
3029         info->mti_spec.sp_migrate_close = 0;
3030
3031         info->mti_spec.u.sp_ea.eadata = NULL;
3032         info->mti_spec.u.sp_ea.eadatalen = 0;
3033 }
3034
3035 void mdt_thread_info_fini(struct mdt_thread_info *info)
3036 {
3037         int i;
3038
3039         if (info->mti_object != NULL) {
3040                 mdt_object_put(info->mti_env, info->mti_object);
3041                 info->mti_object = NULL;
3042         }
3043
3044         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3045                 mdt_lock_handle_fini(&info->mti_lh[i]);
3046         info->mti_env = NULL;
3047         info->mti_pill = NULL;
3048         info->mti_exp = NULL;
3049
3050         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3051                 lu_buf_free(&info->mti_big_buf);
3052 }
3053
3054 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3055 {
3056         struct mdt_thread_info  *mti;
3057
3058         mti = mdt_th_info(tsi->tsi_env);
3059         LASSERT(mti != NULL);
3060
3061         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3062         if (tsi->tsi_corpus != NULL) {
3063                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3064                 lu_object_get(tsi->tsi_corpus);
3065         }
3066         mti->mti_body = tsi->tsi_mdt_body;
3067         mti->mti_dlm_req = tsi->tsi_dlm_req;
3068
3069         return mti;
3070 }
3071
3072 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3073 {
3074         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3075             cfs_fail_val ==
3076             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3077                 set_current_state(TASK_UNINTERRUPTIBLE);
3078                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3079         }
3080
3081         return tgt_connect(tsi);
3082 }
3083
3084 enum mdt_it_code {
3085         MDT_IT_OPEN,
3086         MDT_IT_OCREAT,
3087         MDT_IT_CREATE,
3088         MDT_IT_GETATTR,
3089         MDT_IT_READDIR,
3090         MDT_IT_LOOKUP,
3091         MDT_IT_UNLINK,
3092         MDT_IT_TRUNC,
3093         MDT_IT_GETXATTR,
3094         MDT_IT_LAYOUT,
3095         MDT_IT_QUOTA,
3096         MDT_IT_NR
3097 };
3098
3099 static int mdt_intent_getattr(enum mdt_it_code opcode,
3100                               struct mdt_thread_info *info,
3101                               struct ldlm_lock **,
3102                               __u64);
3103
3104 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3105                                 struct mdt_thread_info *info,
3106                                 struct ldlm_lock **lockp,
3107                                 __u64 flags);
3108
3109 static int mdt_intent_layout(enum mdt_it_code opcode,
3110                              struct mdt_thread_info *info,
3111                              struct ldlm_lock **,
3112                              __u64);
3113 static int mdt_intent_reint(enum mdt_it_code opcode,
3114                             struct mdt_thread_info *info,
3115                             struct ldlm_lock **,
3116                             __u64);
3117
3118 static struct mdt_it_flavor {
3119         const struct req_format *it_fmt;
3120         __u32                    it_flags;
3121         int                    (*it_act)(enum mdt_it_code ,
3122                                          struct mdt_thread_info *,
3123                                          struct ldlm_lock **,
3124                                          __u64);
3125         long                     it_reint;
3126 } mdt_it_flavor[] = {
3127         [MDT_IT_OPEN]     = {
3128                 .it_fmt   = &RQF_LDLM_INTENT,
3129                 /*.it_flags = HABEO_REFERO,*/
3130                 .it_flags = 0,
3131                 .it_act   = mdt_intent_reint,
3132                 .it_reint = REINT_OPEN
3133         },
3134         [MDT_IT_OCREAT]   = {
3135                 .it_fmt   = &RQF_LDLM_INTENT,
3136                 /*
3137                  * OCREAT is not a MUTABOR request as if the file
3138                  * already exists.
3139                  * We do the extra check of OBD_CONNECT_RDONLY in
3140                  * mdt_reint_open() when we really need to create
3141                  * the object.
3142                  */
3143                 .it_flags = 0,
3144                 .it_act   = mdt_intent_reint,
3145                 .it_reint = REINT_OPEN
3146         },
3147         [MDT_IT_CREATE]   = {
3148                 .it_fmt   = &RQF_LDLM_INTENT,
3149                 .it_flags = MUTABOR,
3150                 .it_act   = mdt_intent_reint,
3151                 .it_reint = REINT_CREATE
3152         },
3153         [MDT_IT_GETATTR]  = {
3154                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3155                 .it_flags = HABEO_REFERO,
3156                 .it_act   = mdt_intent_getattr
3157         },
3158         [MDT_IT_READDIR]  = {
3159                 .it_fmt   = NULL,
3160                 .it_flags = 0,
3161                 .it_act   = NULL
3162         },
3163         [MDT_IT_LOOKUP]   = {
3164                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3165                 .it_flags = HABEO_REFERO,
3166                 .it_act   = mdt_intent_getattr
3167         },
3168         [MDT_IT_UNLINK]   = {
3169                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3170                 .it_flags = MUTABOR,
3171                 .it_act   = NULL,
3172                 .it_reint = REINT_UNLINK
3173         },
3174         [MDT_IT_TRUNC]    = {
3175                 .it_fmt   = NULL,
3176                 .it_flags = MUTABOR,
3177                 .it_act   = NULL
3178         },
3179         [MDT_IT_GETXATTR] = {
3180                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3181                 .it_flags = HABEO_CORPUS,
3182                 .it_act   = mdt_intent_getxattr
3183         },
3184         [MDT_IT_LAYOUT] = {
3185                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3186                 .it_flags = 0,
3187                 .it_act   = mdt_intent_layout
3188         }
3189 };
3190
3191 static int
3192 mdt_intent_lock_replace(struct mdt_thread_info *info,
3193                         struct ldlm_lock **lockp,
3194                         struct mdt_lock_handle *lh,
3195                         __u64 flags, int result)
3196 {
3197         struct ptlrpc_request  *req = mdt_info_req(info);
3198         struct ldlm_lock       *lock = *lockp;
3199         struct ldlm_lock       *new_lock;
3200
3201         /* If possible resent found a lock, @lh is set to its handle */
3202         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3203
3204         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3205                 lh->mlh_reg_lh.cookie = 0;
3206                 RETURN(0);
3207         }
3208
3209         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3210                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3211                  * a resend case, however, it could be already destroyed
3212                  * due to client eviction or a raced cancel RPC. */
3213                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3214                                   lh->mlh_reg_lh.cookie);
3215                 lh->mlh_reg_lh.cookie = 0;
3216                 RETURN(-ESTALE);
3217         }
3218
3219         LASSERTF(new_lock != NULL,
3220                  "lockh %#llx flags %#llx : rc = %d\n",
3221                  lh->mlh_reg_lh.cookie, flags, result);
3222
3223         /*
3224          * If we've already given this lock to a client once, then we should
3225          * have no readers or writers.  Otherwise, we should have one reader
3226          * _or_ writer ref (which will be zeroed below) before returning the
3227          * lock to a client.
3228          */
3229         if (new_lock->l_export == req->rq_export) {
3230                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3231         } else {
3232                 LASSERT(new_lock->l_export == NULL);
3233                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3234         }
3235
3236         *lockp = new_lock;
3237
3238         if (new_lock->l_export == req->rq_export) {
3239                 /*
3240                  * Already gave this to the client, which means that we
3241                  * reconstructed a reply.
3242                  */
3243                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3244                         MSG_RESENT);
3245
3246                 LDLM_LOCK_RELEASE(new_lock);
3247                 lh->mlh_reg_lh.cookie = 0;
3248                 RETURN(ELDLM_LOCK_REPLACED);
3249         }
3250
3251         /*
3252          * Fixup the lock to be given to the client.
3253          */
3254         lock_res_and_lock(new_lock);
3255         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3256          * possible blocking AST. */
3257         while (new_lock->l_readers > 0) {
3258                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3259                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3260                 new_lock->l_readers--;
3261         }
3262         while (new_lock->l_writers > 0) {
3263                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3264                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3265                 new_lock->l_writers--;
3266         }
3267
3268         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3269         new_lock->l_blocking_ast = lock->l_blocking_ast;
3270         new_lock->l_completion_ast = lock->l_completion_ast;
3271         new_lock->l_remote_handle = lock->l_remote_handle;
3272         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3273
3274         unlock_res_and_lock(new_lock);
3275
3276         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3277                      &new_lock->l_remote_handle,
3278                      &new_lock->l_exp_hash);
3279
3280         LDLM_LOCK_RELEASE(new_lock);
3281         lh->mlh_reg_lh.cookie = 0;
3282
3283         RETURN(ELDLM_LOCK_REPLACED);
3284 }
3285
3286 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3287                                     struct ldlm_lock *new_lock,
3288                                     struct mdt_lock_handle *lh,
3289                                     __u64 flags)
3290 {
3291         struct ptlrpc_request  *req = mdt_info_req(info);
3292         struct ldlm_request    *dlmreq;
3293
3294         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3295                 return;
3296
3297         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3298
3299         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3300          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3301          * initialized. */
3302         if (flags & LDLM_FL_RESENT) {
3303                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3304                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3305
3306                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3307                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3308                           lh->mlh_reg_lh.cookie);
3309                 return;
3310         }
3311
3312         /*
3313          * If the xid matches, then we know this is a resent request, and allow
3314          * it. (It's probably an OPEN, for which we don't send a lock.
3315          */
3316         if (req_can_reconstruct(req, NULL))
3317                 return;
3318
3319         /*
3320          * This remote handle isn't enqueued, so we never received or processed
3321          * this request.  Clear MSG_RESENT, because it can be handled like any
3322          * normal request now.
3323          */
3324         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3325
3326         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3327                   dlmreq->lock_handle[0].cookie);
3328 }
3329
3330 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3331                                 struct mdt_thread_info *info,
3332                                 struct ldlm_lock **lockp,
3333                                 __u64 flags)
3334 {
3335         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3336         struct ldlm_reply      *ldlm_rep = NULL;
3337         int rc;
3338         ENTRY;
3339
3340         /*
3341          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3342          * (for the resend case) or a new lock. Below we will use it to
3343          * replace the original lock.
3344          */
3345         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3346         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3347                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3348                 rc = mdt_object_lock(info, info->mti_object, lhc,
3349                                      MDS_INODELOCK_XATTR);
3350                 if (rc)
3351                         return rc;
3352         }
3353
3354         rc = mdt_getxattr(info);
3355
3356         if (mdt_info_req(info)->rq_repmsg != NULL)
3357                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3358
3359         if (ldlm_rep == NULL ||
3360             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3361                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3362                 RETURN(err_serious(-EFAULT));
3363         }
3364
3365         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3366
3367         /* This is left for interop instead of adding a new interop flag.
3368          * LU-7433 */
3369 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3370         if (ldlm_rep->lock_policy_res2) {
3371                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3372                 RETURN(ELDLM_LOCK_ABORTED);
3373         }
3374 #endif
3375
3376         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3377         RETURN(rc);
3378 }
3379
3380 static int mdt_intent_getattr(enum mdt_it_code opcode,
3381                               struct mdt_thread_info *info,
3382                               struct ldlm_lock **lockp,
3383                               __u64 flags)
3384 {
3385         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3386         __u64                   child_bits;
3387         struct ldlm_reply      *ldlm_rep;
3388         struct mdt_body        *reqbody;
3389         struct mdt_body        *repbody;
3390         int                     rc, rc2;
3391         ENTRY;
3392
3393         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3394         LASSERT(reqbody);
3395
3396         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3397         LASSERT(repbody);
3398
3399         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3400         repbody->mbo_eadatasize = 0;
3401         repbody->mbo_aclsize = 0;
3402
3403         switch (opcode) {
3404         case MDT_IT_LOOKUP:
3405                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3406                 break;
3407         case MDT_IT_GETATTR:
3408                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3409                              MDS_INODELOCK_PERM;
3410                 break;
3411         default:
3412                 CERROR("Unsupported intent (%d)\n", opcode);
3413                 GOTO(out_shrink, rc = -EINVAL);
3414         }
3415
3416         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3417         if (rc)
3418                 GOTO(out_shrink, rc);
3419
3420         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3421         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3422
3423         /* Get lock from request for possible resent case. */
3424         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3425
3426         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3427         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3428
3429         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3430                 ldlm_rep->lock_policy_res2 = 0;
3431         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3432             ldlm_rep->lock_policy_res2) {
3433                 lhc->mlh_reg_lh.cookie = 0ull;
3434                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3435         }
3436
3437         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3438         EXIT;
3439 out_ucred:
3440         mdt_exit_ucred(info);
3441 out_shrink:
3442         mdt_client_compatibility(info);
3443         rc2 = mdt_fix_reply(info);
3444         if (rc == 0)
3445                 rc = rc2;
3446         return rc;
3447 }
3448
3449 static int mdt_intent_layout(enum mdt_it_code opcode,
3450                              struct mdt_thread_info *info,
3451                              struct ldlm_lock **lockp,
3452                              __u64 flags)
3453 {
3454         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3455         struct layout_intent *layout;
3456         struct lu_fid *fid;
3457         struct mdt_object *obj = NULL;
3458         bool layout_change = false;
3459         int layout_size = 0;
3460         int rc = 0;
3461         ENTRY;
3462
3463         if (opcode != MDT_IT_LAYOUT) {
3464                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3465                         opcode);
3466                 RETURN(-EINVAL);
3467         }
3468
3469         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3470         if (layout == NULL)
3471                 RETURN(-EPROTO);
3472
3473         switch (layout->li_opc) {
3474         case LAYOUT_INTENT_TRUNC:
3475         case LAYOUT_INTENT_WRITE:
3476                 layout_change = true;
3477                 break;
3478         case LAYOUT_INTENT_ACCESS:
3479                 break;
3480         case LAYOUT_INTENT_READ:
3481         case LAYOUT_INTENT_GLIMPSE:
3482         case LAYOUT_INTENT_RELEASE:
3483         case LAYOUT_INTENT_RESTORE:
3484                 CERROR("%s: Unsupported layout intent opc %d\n",
3485                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3486                 rc = -ENOTSUPP;
3487                 break;
3488         default:
3489                 CERROR("%s: Unknown layout intent opc %d\n",
3490                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3491                 rc = -EINVAL;
3492                 break;
3493         }
3494         if (rc < 0)
3495                 RETURN(rc);
3496
3497         fid = &info->mti_tmp_fid2;
3498         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3499
3500         /* Get lock from request for possible resent case. */
3501         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3502
3503         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3504         if (IS_ERR(obj))
3505                 GOTO(out, rc = PTR_ERR(obj));
3506
3507         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3508                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3509                 if (layout_size < 0)
3510                         GOTO(out_obj, rc = layout_size);
3511
3512                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3513                         info->mti_mdt->mdt_max_mdsize = layout_size;
3514         }
3515
3516         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3517         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3518                              layout_size);
3519         rc = req_capsule_server_pack(info->mti_pill);
3520         if (rc)
3521                 GOTO(out_obj, rc);
3522
3523         if (layout_change) {
3524                 rc = mdt_layout_change(info, obj, layout);
3525                 if (rc)
3526                         GOTO(out_obj, rc);
3527         }
3528 out_obj:
3529         mdt_object_put(info->mti_env, obj);
3530
3531         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3532                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3533
3534 out:
3535         lhc->mlh_reg_lh.cookie = 0;
3536
3537         return rc;
3538 }
3539
3540 static int mdt_intent_reint(enum mdt_it_code opcode,
3541                             struct mdt_thread_info *info,
3542                             struct ldlm_lock **lockp,
3543                             __u64 flags)
3544 {
3545         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3546         struct ldlm_reply      *rep = NULL;
3547         long                    opc;
3548         int                     rc;
3549
3550         static const struct req_format *intent_fmts[REINT_MAX] = {
3551                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3552                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3553         };
3554
3555         ENTRY;
3556
3557         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3558         if (opc < 0)
3559                 RETURN(opc);
3560
3561         if (mdt_it_flavor[opcode].it_reint != opc) {
3562                 CERROR("Reint code %ld doesn't match intent: %d\n",
3563                        opc, opcode);
3564                 RETURN(err_serious(-EPROTO));
3565         }
3566
3567         /* Get lock from request for possible resent case. */
3568         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3569
3570         rc = mdt_reint_internal(info, lhc, opc);
3571
3572         /* Check whether the reply has been packed successfully. */
3573         if (mdt_info_req(info)->rq_repmsg != NULL)
3574                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3575         if (rep == NULL)
3576                 RETURN(err_serious(-EFAULT));
3577
3578         /* MDC expects this in any case */
3579         if (rc != 0)
3580                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3581
3582         /* the open lock or the lock for cross-ref object should be
3583          * returned to the client */
3584         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3585             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3586                 rep->lock_policy_res2 = 0;
3587                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3588                 RETURN(rc);
3589         }
3590
3591         rep->lock_policy_res2 = clear_serious(rc);
3592
3593         if (rep->lock_policy_res2 == -ENOENT &&
3594             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3595             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3596                 rep->lock_policy_res2 = 0;
3597
3598         lhc->mlh_reg_lh.cookie = 0ull;
3599         if (rc == -ENOTCONN || rc == -ENODEV ||
3600             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3601                 /*
3602                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3603                  * will be returned by rq_status, and client at ptlrpc layer
3604                  * will detect this, then disconnect, reconnect the import
3605                  * immediately, instead of impacting the following the rpc.
3606                  */
3607                 RETURN(rc);
3608         }
3609         /*
3610          * For other cases, the error will be returned by intent, and client
3611          * will retrieve the result from intent.
3612          */
3613         RETURN(ELDLM_LOCK_ABORTED);
3614 }
3615
3616 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3617 {
3618         int rc;
3619
3620         switch (itcode) {
3621         case IT_OPEN:
3622                 rc = MDT_IT_OPEN;
3623                 break;
3624         case IT_OPEN|IT_CREAT:
3625                 rc = MDT_IT_OCREAT;
3626                 break;
3627         case IT_CREAT:
3628                 rc = MDT_IT_CREATE;
3629                 break;
3630         case IT_READDIR:
3631                 rc = MDT_IT_READDIR;
3632                 break;
3633         case IT_GETATTR:
3634                 rc = MDT_IT_GETATTR;
3635                 break;
3636         case IT_LOOKUP:
3637                 rc = MDT_IT_LOOKUP;
3638                 break;
3639         case IT_UNLINK:
3640                 rc = MDT_IT_UNLINK;
3641                 break;
3642         case IT_TRUNC:
3643                 rc = MDT_IT_TRUNC;
3644                 break;
3645         case IT_GETXATTR:
3646                 rc = MDT_IT_GETXATTR;
3647                 break;
3648         case IT_LAYOUT:
3649                 rc = MDT_IT_LAYOUT;
3650                 break;
3651         case IT_QUOTA_DQACQ:
3652         case IT_QUOTA_CONN:
3653                 rc = MDT_IT_QUOTA;
3654                 break;
3655         default:
3656                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3657                 rc = -EINVAL;
3658                 break;
3659         }
3660         return rc;
3661 }
3662
3663 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3664                           struct mdt_thread_info *info,
3665                           struct ldlm_lock **lockp, __u64 flags)
3666 {
3667         struct req_capsule      *pill = info->mti_pill;
3668         struct ptlrpc_request   *req = mdt_info_req(info);
3669         struct mdt_it_flavor    *flv;
3670         int opc;
3671         int rc;
3672         ENTRY;
3673
3674         opc = mdt_intent_code(itopc);
3675         if (opc < 0)
3676                 RETURN(-EINVAL);
3677
3678         if (opc == MDT_IT_QUOTA) {
3679                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3680
3681                 if (qmt == NULL)
3682                         RETURN(-EOPNOTSUPP);
3683
3684                 if (mdt_rdonly(req->rq_export))
3685                         RETURN(-EROFS);
3686
3687                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3688                 /* pass the request to quota master */
3689                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3690                                                  mdt_info_req(info), lockp,
3691                                                  flags);
3692                 RETURN(rc);
3693         }
3694
3695         flv = &mdt_it_flavor[opc];
3696         if (flv->it_fmt != NULL)
3697                 req_capsule_extend(pill, flv->it_fmt);
3698
3699         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3700         if (rc < 0)
3701                 RETURN(rc);
3702
3703         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3704                 RETURN(-EROFS);
3705
3706         if (flv->it_act != NULL) {
3707                 struct ldlm_reply *rep;
3708
3709                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3710
3711                 /* execute policy */
3712                 rc = flv->it_act(opc, info, lockp, flags);
3713
3714                 /* Check whether the reply has been packed successfully. */
3715                 if (req->rq_repmsg != NULL) {
3716                         rep = req_capsule_server_get(info->mti_pill,
3717                                                      &RMF_DLM_REP);
3718                         rep->lock_policy_res2 =
3719                                 ptlrpc_status_hton(rep->lock_policy_res2);
3720                 }
3721         }
3722
3723         RETURN(rc);
3724 }
3725
3726 static int mdt_intent_policy(struct ldlm_namespace *ns,
3727                              struct ldlm_lock **lockp, void *req_cookie,
3728                              enum ldlm_mode mode, __u64 flags, void *data)
3729 {
3730         struct tgt_session_info *tsi;
3731         struct mdt_thread_info  *info;
3732         struct ptlrpc_request   *req  =  req_cookie;
3733         struct ldlm_intent      *it;
3734         struct req_capsule      *pill;
3735         int rc;
3736
3737         ENTRY;
3738
3739         LASSERT(req != NULL);
3740
3741         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3742
3743         info = tsi2mdt_info(tsi);
3744         LASSERT(info != NULL);
3745         pill = info->mti_pill;
3746         LASSERT(pill->rc_req == req);
3747
3748         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3749                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3750                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3751                 if (it != NULL) {
3752                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3753                         if (rc == 0)
3754                                 rc = ELDLM_OK;
3755
3756                         /* Lock without inodebits makes no sense and will oops
3757                          * later in ldlm. Let's check it now to see if we have
3758                          * ibits corrupted somewhere in mdt_intent_opc().
3759                          * The case for client miss to set ibits has been
3760                          * processed by others. */
3761                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3762                                         lr_type == LDLM_IBITS,
3763                                      info->mti_dlm_req->lock_desc.\
3764                                         l_policy_data.l_inodebits.bits != 0));
3765                 } else
3766                         rc = err_serious(-EFAULT);
3767         } else {
3768                 /* No intent was provided */
3769                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3770                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3771                 rc = req_capsule_server_pack(pill);
3772                 if (rc)
3773                         rc = err_serious(rc);
3774         }
3775         mdt_thread_info_fini(info);
3776         RETURN(rc);
3777 }
3778
3779 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3780 {
3781         struct seq_server_site  *ss = mdt_seq_site(mdt);
3782
3783         if (ss->ss_node_id == 0)
3784                 return;
3785
3786         if (ss->ss_client_seq != NULL) {
3787                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3788                 ss->ss_client_seq->lcs_exp = NULL;
3789         }
3790
3791         if (ss->ss_server_fld != NULL) {
3792                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3793                 ss->ss_server_fld->lsf_control_exp = NULL;
3794         }
3795 }
3796
3797 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3798 {
3799         struct seq_server_site *ss = mdt_seq_site(mdt);
3800
3801         if (ss == NULL)
3802                 return;
3803
3804         if (ss->ss_server_seq != NULL)
3805                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3806 }
3807
3808 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3809 {
3810         mdt_seq_fini_cli(mdt);
3811         mdt_deregister_seq_exp(mdt);
3812
3813         return seq_site_fini(env, mdt_seq_site(mdt));
3814 }
3815
3816 /**
3817  * It will retrieve its FLDB entries from MDT0, and it only happens
3818  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3819  * and it needs to refresh FLDB from the MDT0.
3820  **/
3821 static int mdt_register_lwp_callback(void *data)
3822 {
3823         struct lu_env           env;
3824         struct mdt_device       *mdt = data;
3825         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3826         int                     rc;
3827         ENTRY;
3828
3829         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3830
3831         rc = lu_env_init(&env, LCT_MD_THREAD);
3832         if (rc < 0) {
3833                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3834                 RETURN(rc);
3835         }
3836
3837         /* Allocate new sequence now to avoid creating local transaction
3838          * in the normal transaction process */
3839         rc = seq_server_check_and_alloc_super(&env,
3840                                               mdt_seq_site(mdt)->ss_server_seq);
3841         if (rc < 0)
3842                 GOTO(out, rc);
3843
3844         if (fld->lsf_new) {
3845                 rc = fld_update_from_controller(&env, fld);
3846                 if (rc != 0) {
3847                         CERROR("%s: cannot update controller: rc = %d\n",
3848                                mdt_obd_name(mdt), rc);
3849                         GOTO(out, rc);
3850                 }
3851         }
3852 out:
3853         lu_env_fini(&env);
3854         RETURN(rc);
3855 }
3856
3857 static int mdt_register_seq_exp(struct mdt_device *mdt)
3858 {
3859         struct seq_server_site  *ss = mdt_seq_site(mdt);
3860         char                    *lwp_name = NULL;
3861         int                     rc;
3862
3863         if (ss->ss_node_id == 0)
3864                 return 0;
3865
3866         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3867         if (lwp_name == NULL)
3868                 GOTO(out_free, rc = -ENOMEM);
3869
3870         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3871         if (rc != 0)
3872                 GOTO(out_free, rc);
3873
3874         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3875                                       NULL, NULL);
3876         if (rc != 0)
3877                 GOTO(out_free, rc);
3878
3879         rc = lustre_register_lwp_item(lwp_name,
3880                                       &ss->ss_server_fld->lsf_control_exp,
3881                                       mdt_register_lwp_callback, mdt);
3882         if (rc != 0) {
3883                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3884                 ss->ss_client_seq->lcs_exp = NULL;
3885                 GOTO(out_free, rc);
3886         }
3887 out_free:
3888         if (lwp_name != NULL)
3889                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3890
3891         return rc;
3892 }
3893
3894 /*
3895  * Init client sequence manager which is used by local MDS to talk to sequence
3896  * controller on remote node.
3897  */
3898 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3899 {
3900         struct seq_server_site  *ss = mdt_seq_site(mdt);
3901         int                     rc;
3902         char                    *prefix;
3903         ENTRY;
3904
3905         /* check if this is adding the first MDC and controller is not yet
3906          * initialized. */
3907         OBD_ALLOC_PTR(ss->ss_client_seq);
3908         if (ss->ss_client_seq == NULL)
3909                 RETURN(-ENOMEM);
3910
3911         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3912         if (prefix == NULL) {
3913                 OBD_FREE_PTR(ss->ss_client_seq);
3914                 ss->ss_client_seq = NULL;
3915                 RETURN(-ENOMEM);
3916         }
3917
3918         /* Note: seq_client_fini will be called in seq_site_fini */
3919         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3920         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3921                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3922                                                             NULL);
3923         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3924         if (rc != 0) {
3925                 OBD_FREE_PTR(ss->ss_client_seq);
3926                 ss->ss_client_seq = NULL;
3927                 RETURN(rc);
3928         }
3929
3930         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3931
3932         RETURN(rc);
3933 }
3934
3935 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3936 {
3937         struct seq_server_site  *ss;
3938         int                     rc;
3939         ENTRY;
3940
3941         ss = mdt_seq_site(mdt);
3942         /* init sequence controller server(MDT0) */
3943         if (ss->ss_node_id == 0) {
3944                 OBD_ALLOC_PTR(ss->ss_control_seq);
3945                 if (ss->ss_control_seq == NULL)
3946                         RETURN(-ENOMEM);
3947
3948                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3949                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3950                                      ss);
3951                 if (rc)
3952                         GOTO(out_seq_fini, rc);
3953         }
3954
3955         /* Init normal sequence server */
3956         OBD_ALLOC_PTR(ss->ss_server_seq);
3957         if (ss->ss_server_seq == NULL)
3958                 GOTO(out_seq_fini, rc = -ENOMEM);
3959
3960         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3961                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3962         if (rc)
3963                 GOTO(out_seq_fini, rc);
3964
3965         /* init seq client for seq server to talk to seq controller(MDT0) */
3966         rc = mdt_seq_init_cli(env, mdt);
3967         if (rc != 0)
3968                 GOTO(out_seq_fini, rc);
3969
3970         if (ss->ss_node_id != 0)
3971                 /* register controller export through lwp */
3972                 rc = mdt_register_seq_exp(mdt);
3973
3974         EXIT;
3975 out_seq_fini:
3976         if (rc)
3977                 mdt_seq_fini(env, mdt);
3978
3979         return rc;
3980 }
3981
3982 /*
3983  * FLD wrappers
3984  */
3985 static int mdt_fld_fini(const struct lu_env *env,
3986                         struct mdt_device *m)
3987 {
3988         struct seq_server_site *ss = mdt_seq_site(m);
3989         ENTRY;
3990
3991         if (ss && ss->ss_server_fld) {
3992                 fld_server_fini(env, ss->ss_server_fld);
3993                 OBD_FREE_PTR(ss->ss_server_fld);
3994                 ss->ss_server_fld = NULL;
3995         }
3996
3997         RETURN(0);
3998 }
3999
4000 static int mdt_fld_init(const struct lu_env *env,
4001                         const char *uuid,
4002                         struct mdt_device *m)
4003 {
4004         struct seq_server_site *ss;
4005         int rc;
4006         ENTRY;
4007
4008         ss = mdt_seq_site(m);
4009
4010         OBD_ALLOC_PTR(ss->ss_server_fld);
4011         if (ss->ss_server_fld == NULL)
4012                 RETURN(rc = -ENOMEM);
4013
4014         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4015                              LU_SEQ_RANGE_MDT);
4016         if (rc) {
4017                 OBD_FREE_PTR(ss->ss_server_fld);
4018                 ss->ss_server_fld = NULL;
4019                 RETURN(rc);
4020         }
4021
4022         RETURN(0);
4023 }
4024
4025 static void mdt_stack_pre_fini(const struct lu_env *env,
4026                            struct mdt_device *m, struct lu_device *top)
4027 {
4028         struct lustre_cfg_bufs  *bufs;
4029         struct lustre_cfg       *lcfg;
4030         struct mdt_thread_info  *info;
4031         ENTRY;
4032
4033         LASSERT(top);
4034
4035         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4036         LASSERT(info != NULL);
4037
4038         bufs = &info->mti_u.bufs;
4039
4040         LASSERT(m->mdt_child_exp);
4041         LASSERT(m->mdt_child_exp->exp_obd);
4042
4043         /* process cleanup, pass mdt obd name to get obd umount flags */
4044         /* XXX: this is needed because all layers are referenced by
4045          * objects (some of them are pinned by osd, for example *
4046          * the proper solution should be a model where object used
4047          * by osd only doesn't have mdt/mdd slices -bzzz */
4048         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4049         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4050         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4051         if (lcfg == NULL)
4052                 RETURN_EXIT;
4053
4054         top->ld_ops->ldo_process_config(env, top, lcfg);
4055         lustre_cfg_free(lcfg);
4056         EXIT;
4057 }
4058
4059 static void mdt_stack_fini(const struct lu_env *env,
4060                            struct mdt_device *m, struct lu_device *top)
4061 {
4062         struct obd_device       *obd = mdt2obd_dev(m);
4063         struct lustre_cfg_bufs  *bufs;
4064         struct lustre_cfg       *lcfg;
4065         struct mdt_thread_info  *info;
4066         char                     flags[3] = "";
4067         ENTRY;
4068
4069         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4070         LASSERT(info != NULL);
4071
4072         lu_dev_del_linkage(top->ld_site, top);
4073
4074         lu_site_purge(env, top->ld_site, -1);
4075
4076         bufs = &info->mti_u.bufs;
4077         /* process cleanup, pass mdt obd name to get obd umount flags */
4078         /* another purpose is to let all layers to release their objects */
4079         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4080         if (obd->obd_force)
4081                 strcat(flags, "F");
4082         if (obd->obd_fail)
4083                 strcat(flags, "A");
4084         lustre_cfg_bufs_set_string(bufs, 1, flags);
4085         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4086         if (lcfg == NULL)
4087                 RETURN_EXIT;
4088
4089         LASSERT(top);
4090         top->ld_ops->ldo_process_config(env, top, lcfg);
4091         lustre_cfg_free(lcfg);
4092
4093         lu_site_purge(env, top->ld_site, -1);
4094
4095         m->mdt_child = NULL;
4096         m->mdt_bottom = NULL;
4097
4098         obd_disconnect(m->mdt_child_exp);
4099         m->mdt_child_exp = NULL;
4100
4101         obd_disconnect(m->mdt_bottom_exp);
4102         m->mdt_child_exp = NULL;
4103 }
4104
4105 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4106                                const char *next, struct obd_export **exp)
4107 {
4108         struct obd_connect_data *data = NULL;
4109         struct obd_device       *obd;
4110         int                      rc;
4111         ENTRY;
4112
4113         OBD_ALLOC_PTR(data);
4114         if (data == NULL)
4115                 GOTO(out, rc = -ENOMEM);
4116
4117         obd = class_name2obd(next);
4118         if (obd == NULL) {
4119                 CERROR("%s: can't locate next device: %s\n",
4120                        mdt_obd_name(m), next);
4121                 GOTO(out, rc = -ENOTCONN);
4122         }
4123
4124         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4125         data->ocd_version = LUSTRE_VERSION_CODE;
4126
4127         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4128         if (rc) {
4129                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4130                        mdt_obd_name(m), next, rc);
4131                 GOTO(out, rc);
4132         }
4133
4134 out:
4135         if (data)
4136                 OBD_FREE_PTR(data);
4137         RETURN(rc);
4138 }
4139
4140 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4141                           struct lustre_cfg *cfg)
4142 {
4143         char                   *dev = lustre_cfg_string(cfg, 0);
4144         int                     rc, name_size, uuid_size;
4145         char                   *name, *uuid, *p;
4146         struct lustre_cfg_bufs *bufs;
4147         struct lustre_cfg      *lcfg;
4148         struct obd_device      *obd;
4149         struct lustre_profile  *lprof;
4150         struct lu_site         *site;
4151         ENTRY;
4152
4153         /* in 1.8 we had the only device in the stack - MDS.
4154          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4155          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4156          * to generate names and setup MDT, MDD. MDT will be using
4157          * generated name to connect to MDD. for MDD the next device
4158          * will be LOD with name taken from so called "profile" which
4159          * is generated by mount_option line
4160          *
4161          * 1.8 MGS generates config. commands like this:
4162          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4163          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4164          * 2.0 MGS generates config. commands like this:
4165          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4166          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4167          *                    3:lustre-MDT0000-mdtlov  4:f
4168          *
4169          * we generate MDD name from MDT one, just replacing T with D
4170          *
4171          * after all the preparations, the logical equivalent will be
4172          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4173          *                    3:lustre-MDT0000-mdtlov  4:f
4174          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4175          *                    3:lustre-MDD0000  4:f
4176          *
4177          *  notice we build the stack from down to top: MDD first, then MDT */
4178
4179         name_size = MAX_OBD_NAME;
4180         uuid_size = MAX_OBD_NAME;
4181
4182         OBD_ALLOC(name, name_size);
4183         OBD_ALLOC(uuid, uuid_size);
4184         if (name == NULL || uuid == NULL)
4185                 GOTO(cleanup_mem, rc = -ENOMEM);
4186
4187         OBD_ALLOC_PTR(bufs);
4188         if (!bufs)
4189                 GOTO(cleanup_mem, rc = -ENOMEM);
4190
4191         strcpy(name, dev);
4192         p = strstr(name, "-MDT");
4193         if (p == NULL)
4194                 GOTO(free_bufs, rc = -ENOMEM);
4195         p[3] = 'D';
4196
4197         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4198
4199         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4200         if (lprof == NULL || lprof->lp_dt == NULL) {
4201                 CERROR("can't find the profile: %s\n",
4202                        lustre_cfg_string(cfg, 0));
4203                 GOTO(free_bufs, rc = -EINVAL);
4204         }
4205
4206         lustre_cfg_bufs_reset(bufs, name);
4207         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4208         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4209         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4210
4211         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4212         if (lcfg == NULL)
4213                 GOTO(put_profile, rc = -ENOMEM);
4214
4215         rc = class_attach(lcfg);
4216         if (rc)
4217                 GOTO(lcfg_cleanup, rc);
4218
4219         obd = class_name2obd(name);
4220         if (!obd) {
4221                 CERROR("Can not find obd %s (%s in config)\n",
4222                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4223                 GOTO(lcfg_cleanup, rc = -EINVAL);
4224         }
4225
4226         lustre_cfg_free(lcfg);
4227
4228         lustre_cfg_bufs_reset(bufs, name);
4229         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4230         lustre_cfg_bufs_set_string(bufs, 2, dev);
4231         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4232
4233         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4234         if (lcfg == NULL)
4235                 GOTO(class_detach, rc = -ENOMEM);
4236
4237         rc = class_setup(obd, lcfg);
4238         if (rc)
4239                 GOTO(class_detach, rc);
4240
4241         /* connect to MDD we just setup */
4242         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4243         if (rc)
4244                 GOTO(class_detach, rc);
4245
4246         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4247         LASSERT(site);
4248         LASSERT(mdt_lu_site(mdt) == NULL);
4249         mdt->mdt_lu_dev.ld_site = site;
4250         site->ls_top_dev = &mdt->mdt_lu_dev;
4251         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4252
4253         /* now connect to bottom OSD */
4254         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4255         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4256         if (rc)
4257                 GOTO(class_detach, rc);
4258         mdt->mdt_bottom =
4259                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4260
4261         rc = lu_env_refill((struct lu_env *)env);
4262         if (rc != 0)
4263                 CERROR("Failure to refill session: '%d'\n", rc);
4264
4265         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4266
4267         EXIT;
4268 class_detach:
4269         if (rc)
4270                 class_detach(obd, lcfg);
4271 lcfg_cleanup:
4272         lustre_cfg_free(lcfg);
4273 put_profile:
4274         class_put_profile(lprof);
4275 free_bufs:
4276         OBD_FREE_PTR(bufs);
4277 cleanup_mem:
4278         if (name)
4279                 OBD_FREE(name, name_size);
4280         if (uuid)
4281                 OBD_FREE(uuid, uuid_size);
4282         RETURN(rc);
4283 }
4284
4285 /* setup quota master target on MDT0 */
4286 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4287                           struct lustre_cfg *cfg)
4288 {
4289         struct obd_device       *obd;
4290         char                    *dev = lustre_cfg_string(cfg, 0);
4291         char                    *qmtname, *uuid, *p;
4292         struct lustre_cfg_bufs  *bufs;
4293         struct lustre_cfg       *lcfg;
4294         struct lustre_profile   *lprof;
4295         struct obd_connect_data *data;
4296         int                      rc;
4297         ENTRY;
4298
4299         LASSERT(mdt->mdt_qmt_exp == NULL);
4300         LASSERT(mdt->mdt_qmt_dev == NULL);
4301
4302         /* quota master is on MDT0 only for now */
4303         if (mdt->mdt_seq_site.ss_node_id != 0)
4304                 RETURN(0);
4305
4306         /* MGS generates config commands which look as follows:
4307          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4308          *                    3:lustre-MDT0000-mdtlov  4:f
4309          *
4310          * We generate the QMT name from the MDT one, just replacing MD with QM
4311          * after all the preparations, the logical equivalent will be:
4312          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4313          *                    3:lustre-MDT0000-osd  4:f */
4314         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4315         OBD_ALLOC(uuid, UUID_MAX);
4316         OBD_ALLOC_PTR(bufs);
4317         OBD_ALLOC_PTR(data);
4318         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4319                 GOTO(cleanup_mem, rc = -ENOMEM);
4320
4321         strcpy(qmtname, dev);
4322         p = strstr(qmtname, "-MDT");
4323         if (p == NULL)
4324                 GOTO(cleanup_mem, rc = -ENOMEM);
4325         /* replace MD with QM */
4326         p[1] = 'Q';
4327         p[2] = 'M';
4328
4329         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4330
4331         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4332         if (lprof == NULL || lprof->lp_dt == NULL) {
4333                 CERROR("can't find profile for %s\n",
4334                        lustre_cfg_string(cfg, 0));
4335                 GOTO(cleanup_mem, rc = -EINVAL);
4336         }
4337
4338         lustre_cfg_bufs_reset(bufs, qmtname);
4339         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4340         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4341         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4342
4343         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4344         if (lcfg == NULL)
4345                 GOTO(put_profile, rc = -ENOMEM);
4346
4347         rc = class_attach(lcfg);
4348         if (rc)
4349                 GOTO(lcfg_cleanup, rc);
4350
4351         obd = class_name2obd(qmtname);
4352         if (!obd) {
4353                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4354                        lustre_cfg_string(cfg, 0));
4355                 GOTO(lcfg_cleanup, rc = -EINVAL);
4356         }
4357
4358         lustre_cfg_free(lcfg);
4359
4360         lustre_cfg_bufs_reset(bufs, qmtname);
4361         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4362         lustre_cfg_bufs_set_string(bufs, 2, dev);
4363
4364         /* for quota, the next device should be the OSD device */
4365         lustre_cfg_bufs_set_string(bufs, 3,
4366                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4367
4368         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4369         if (lcfg == NULL)
4370                 GOTO(class_detach, rc = -ENOMEM);
4371
4372         rc = class_setup(obd, lcfg);
4373         if (rc)
4374                 GOTO(class_detach, rc);
4375
4376         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4377
4378         /* configure local quota objects */
4379         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4380                                                    &mdt->mdt_lu_dev,
4381                                                    mdt->mdt_qmt_dev);
4382         if (rc)
4383                 GOTO(class_cleanup, rc);
4384
4385         /* connect to quota master target */
4386         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4387         data->ocd_version = LUSTRE_VERSION_CODE;
4388         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4389                          data, NULL);
4390         if (rc) {
4391                 CERROR("cannot connect to quota master device %s (%d)\n",
4392                        qmtname, rc);
4393                 GOTO(class_cleanup, rc);
4394         }
4395
4396         EXIT;
4397 class_cleanup:
4398         if (rc) {
4399                 class_manual_cleanup(obd);
4400                 mdt->mdt_qmt_dev = NULL;
4401         }
4402 class_detach:
4403         if (rc)
4404                 class_detach(obd, lcfg);
4405 lcfg_cleanup:
4406         lustre_cfg_free(lcfg);
4407 put_profile:
4408         class_put_profile(lprof);
4409 cleanup_mem:
4410         if (bufs)
4411                 OBD_FREE_PTR(bufs);
4412         if (qmtname)
4413                 OBD_FREE(qmtname, MAX_OBD_NAME);
4414         if (uuid)
4415                 OBD_FREE(uuid, UUID_MAX);
4416         if (data)
4417                 OBD_FREE_PTR(data);
4418         return rc;
4419 }
4420
4421 /* Shutdown quota master target associated with mdt */
4422 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4423 {
4424         ENTRY;
4425
4426         if (mdt->mdt_qmt_exp == NULL)
4427                 RETURN_EXIT;
4428         LASSERT(mdt->mdt_qmt_dev != NULL);
4429
4430         /* the qmt automatically shuts down when the mdt disconnects */
4431         obd_disconnect(mdt->mdt_qmt_exp);
4432         mdt->mdt_qmt_exp = NULL;
4433         mdt->mdt_qmt_dev = NULL;
4434         EXIT;
4435 }
4436
4437 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4438  * for now. This will be removed along with converting rest of MDT code
4439  * to use tgt_session_info */
4440 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4441 {
4442         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4443         int                      rc;
4444
4445         rc = mdt_getxattr(info);
4446
4447         mdt_thread_info_fini(info);
4448         return rc;
4449 }
4450
4451 static struct tgt_handler mdt_tgt_handlers[] = {
4452 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4453                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4454                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4455 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4456                 0,                      MDS_DISCONNECT, tgt_disconnect,
4457                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4458 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4459                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4460                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4461 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4462 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4463 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4464 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4465                                                         mdt_getattr_name),
4466 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4467 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4468 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4469 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4470 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4471 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4472 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4473 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4474                                                         mdt_hsm_progress),
4475 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4476                                                         mdt_hsm_ct_register),
4477 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4478                                                         mdt_hsm_ct_unregister),
4479 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4480                                                         mdt_hsm_state_get),
4481 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4482                                                         mdt_hsm_state_set),
4483 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4484 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4485                                                         mdt_hsm_request),
4486 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4487             MDS_SWAP_LAYOUTS,
4488             mdt_swap_layouts),
4489 };
4490
4491 static struct tgt_handler mdt_sec_ctx_ops[] = {
4492 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4493 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4494 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4495 };
4496
4497 static struct tgt_handler mdt_quota_ops[] = {
4498 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4499 };
4500
4501 static struct tgt_opc_slice mdt_common_slice[] = {
4502         {
4503                 .tos_opc_start  = MDS_FIRST_OPC,
4504                 .tos_opc_end    = MDS_LAST_OPC,
4505                 .tos_hs         = mdt_tgt_handlers
4506         },
4507         {
4508                 .tos_opc_start  = OBD_FIRST_OPC,
4509                 .tos_opc_end    = OBD_LAST_OPC,
4510                 .tos_hs         = tgt_obd_handlers
4511         },
4512         {
4513                 .tos_opc_start  = LDLM_FIRST_OPC,
4514                 .tos_opc_end    = LDLM_LAST_OPC,
4515                 .tos_hs         = tgt_dlm_handlers
4516         },
4517         {
4518                 .tos_opc_start  = SEC_FIRST_OPC,
4519                 .tos_opc_end    = SEC_LAST_OPC,
4520                 .tos_hs         = mdt_sec_ctx_ops
4521         },
4522         {
4523                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4524                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4525                 .tos_hs         = tgt_out_handlers
4526         },
4527         {
4528                 .tos_opc_start  = FLD_FIRST_OPC,
4529                 .tos_opc_end    = FLD_LAST_OPC,
4530                 .tos_hs         = fld_handlers
4531         },
4532         {
4533                 .tos_opc_start  = SEQ_FIRST_OPC,
4534                 .tos_opc_end    = SEQ_LAST_OPC,
4535                 .tos_hs         = seq_handlers
4536         },
4537         {
4538                 .tos_opc_start  = QUOTA_DQACQ,
4539                 .tos_opc_end    = QUOTA_LAST_OPC,
4540                 .tos_hs         = mdt_quota_ops
4541         },
4542         {
4543                 .tos_opc_start  = LLOG_FIRST_OPC,
4544                 .tos_opc_end    = LLOG_LAST_OPC,
4545                 .tos_hs         = tgt_llog_handlers
4546         },
4547         {
4548                 .tos_opc_start  = LFSCK_FIRST_OPC,
4549                 .tos_opc_end    = LFSCK_LAST_OPC,
4550                 .tos_hs         = tgt_lfsck_handlers
4551         },
4552
4553         {
4554                 .tos_hs         = NULL
4555         }
4556 };
4557
4558 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4559 {
4560         struct md_device *next = m->mdt_child;
4561         struct lu_device *d = &m->mdt_lu_dev;
4562         struct obd_device *obd = mdt2obd_dev(m);
4563         struct lfsck_stop stop;
4564
4565         ENTRY;
4566         stop.ls_status = LS_PAUSED;
4567         stop.ls_flags = 0;
4568         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4569
4570         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4571         ping_evictor_stop();
4572
4573         if (m->mdt_opts.mo_coordinator)
4574                 mdt_hsm_cdt_stop(m);
4575
4576         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4577         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4578
4579         if (m->mdt_namespace != NULL)
4580                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4581                                           d->ld_obd->obd_force);
4582
4583         obd_exports_barrier(obd);
4584         obd_zombie_barrier();
4585
4586         mdt_quota_fini(env, m);
4587
4588         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4589
4590         /* Calling the cleanup functions in the same order as in the mdt_init0
4591          * error path
4592          */
4593         mdt_procfs_fini(m);
4594
4595         target_recovery_fini(obd);
4596         upcall_cache_cleanup(m->mdt_identity_cache);
4597         m->mdt_identity_cache = NULL;
4598
4599         mdt_fs_cleanup(env, m);
4600
4601         tgt_fini(env, &m->mdt_lut);
4602
4603         mdt_hsm_cdt_fini(m);
4604
4605         if (m->mdt_namespace != NULL) {
4606                 ldlm_namespace_free_post(m->mdt_namespace);
4607                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4608         }
4609
4610         if (m->mdt_md_root != NULL) {
4611                 mdt_object_put(env, m->mdt_md_root);
4612                 m->mdt_md_root = NULL;
4613         }
4614
4615         mdt_seq_fini(env, m);
4616
4617         mdt_fld_fini(env, m);
4618
4619         /*
4620          * Finish the stack
4621          */
4622         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4623
4624         LASSERT(atomic_read(&d->ld_ref) == 0);
4625
4626         server_put_mount(mdt_obd_name(m), true);
4627
4628         EXIT;
4629 }
4630
4631 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4632
4633 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4634                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4635 {
4636         struct mdt_thread_info    *info;
4637         struct obd_device         *obd;
4638         const char                *dev = lustre_cfg_string(cfg, 0);
4639         const char                *num = lustre_cfg_string(cfg, 2);
4640         struct lustre_mount_info  *lmi = NULL;
4641         struct lustre_sb_info     *lsi;
4642         struct lu_site            *s;
4643         struct seq_server_site    *ss_site;
4644         const char                *identity_upcall = "NONE";
4645         struct md_device          *next;
4646         int                        rc;
4647         long                       node_id;
4648         mntopt_t                   mntopts;
4649         ENTRY;
4650
4651         lu_device_init(&m->mdt_lu_dev, ldt);
4652         /*
4653          * Environment (env) might be missing mdt_thread_key values at that
4654          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4655          * mode.
4656          *
4657          * Usually device allocation path doesn't use module key values, but
4658          * mdt has to do a lot of work here, so allocate key value.
4659          */
4660         rc = lu_env_refill((struct lu_env *)env);
4661         if (rc != 0)
4662                 RETURN(rc);
4663
4664         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4665         LASSERT(info != NULL);
4666
4667         obd = class_name2obd(dev);
4668         LASSERT(obd != NULL);
4669
4670         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4671         m->mdt_opts.mo_evict_tgt_nids = 1;
4672         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4673
4674         /* default is coordinator off, it is started through conf_param
4675          * or /proc */
4676         m->mdt_opts.mo_coordinator = 0;
4677
4678         lmi = server_get_mount(dev);
4679         if (lmi == NULL) {
4680                 CERROR("Cannot get mount info for %s!\n", dev);
4681                 RETURN(-EFAULT);
4682         } else {
4683                 lsi = s2lsi(lmi->lmi_sb);
4684                 /* CMD is supported only in IAM mode */
4685                 LASSERT(num);
4686                 node_id = simple_strtol(num, NULL, 10);
4687                 obd->u.obt.obt_magic = OBT_MAGIC;
4688                 if (lsi->lsi_lmd != NULL &&
4689                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4690                         m->mdt_skip_lfsck = 1;
4691         }
4692
4693         m->mdt_squash.rsi_uid = 0;
4694         m->mdt_squash.rsi_gid = 0;
4695         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4696         init_rwsem(&m->mdt_squash.rsi_sem);
4697         spin_lock_init(&m->mdt_lock);
4698         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4699         m->mdt_enable_remote_dir = 0;
4700         m->mdt_enable_remote_dir_gid = 0;
4701
4702         atomic_set(&m->mdt_mds_mds_conns, 0);
4703         atomic_set(&m->mdt_async_commit_count, 0);
4704
4705         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4706         m->mdt_lu_dev.ld_obd = obd;
4707         /* Set this lu_device to obd for error handling purposes. */
4708         obd->obd_lu_dev = &m->mdt_lu_dev;
4709
4710         /* init the stack */
4711         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4712         if (rc) {
4713                 CERROR("%s: Can't init device stack, rc %d\n",
4714                        mdt_obd_name(m), rc);
4715                 GOTO(err_lmi, rc);
4716         }
4717
4718         s = mdt_lu_site(m);
4719         ss_site = mdt_seq_site(m);
4720         s->ld_seq_site = ss_site;
4721         ss_site->ss_lu = s;
4722
4723         /* set server index */
4724         ss_site->ss_node_id = node_id;
4725
4726         /* failover is the default
4727          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4728          * assumed whose ss_node_id == 0 XXX
4729          * */
4730         obd->obd_replayable = 1;
4731         /* No connection accepted until configurations will finish */
4732         obd->obd_no_conn = 1;
4733
4734         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4735                 char *str = lustre_cfg_string(cfg, 4);
4736                 if (strchr(str, 'n')) {
4737                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4738                         obd->obd_replayable = 0;
4739                 }
4740         }
4741
4742         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4743         if (rc)
4744                 GOTO(err_fini_stack, rc);
4745
4746         rc = mdt_seq_init(env, m);
4747         if (rc)
4748                 GOTO(err_fini_fld, rc);
4749
4750         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4751                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4752         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4753                                               LDLM_NAMESPACE_SERVER,
4754                                               LDLM_NAMESPACE_GREEDY,
4755                                               LDLM_NS_TYPE_MDT);
4756         if (m->mdt_namespace == NULL)
4757                 GOTO(err_fini_seq, rc = -ENOMEM);
4758
4759         m->mdt_namespace->ns_lvbp = m;
4760         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4761
4762         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4763         /* set obd_namespace for compatibility with old code */
4764         obd->obd_namespace = m->mdt_namespace;
4765
4766         rc = mdt_hsm_cdt_init(m);
4767         if (rc != 0) {
4768                 CERROR("%s: error initializing coordinator, rc %d\n",
4769                        mdt_obd_name(m), rc);
4770                 GOTO(err_free_ns, rc);
4771         }
4772
4773         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4774                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4775                       OBD_FAIL_MDS_ALL_REPLY_NET);
4776         if (rc)
4777                 GOTO(err_free_hsm, rc);
4778
4779         rc = mdt_fs_setup(env, m, obd, lsi);
4780         if (rc)
4781                 GOTO(err_tgt, rc);
4782
4783         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
4784
4785         next = m->mdt_child;
4786         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4787                                          &mntopts);
4788         if (rc)
4789                 GOTO(err_fs_cleanup, rc);
4790
4791         if (mntopts & MNTOPT_USERXATTR)
4792                 m->mdt_opts.mo_user_xattr = 1;
4793         else
4794                 m->mdt_opts.mo_user_xattr = 0;
4795
4796         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4797         if (rc)
4798                 GOTO(err_fs_cleanup, rc);
4799
4800         if (mntopts & MNTOPT_ACL)
4801                 m->mdt_opts.mo_acl = 1;
4802         else
4803                 m->mdt_opts.mo_acl = 0;
4804
4805         /* XXX: to support suppgid for ACL, we enable identity_upcall
4806          * by default, otherwise, maybe got unexpected -EACCESS. */
4807         if (m->mdt_opts.mo_acl)
4808                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4809
4810         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4811                                                 identity_upcall,
4812                                                 &mdt_identity_upcall_cache_ops);
4813         if (IS_ERR(m->mdt_identity_cache)) {
4814                 rc = PTR_ERR(m->mdt_identity_cache);
4815                 m->mdt_identity_cache = NULL;
4816                 GOTO(err_fs_cleanup, rc);
4817         }
4818
4819         rc = mdt_procfs_init(m, dev);
4820         if (rc) {
4821                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4822                 GOTO(err_recovery, rc);
4823         }
4824
4825         rc = mdt_quota_init(env, m, cfg);
4826         if (rc)
4827                 GOTO(err_procfs, rc);
4828
4829         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4830         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4831                            "mdt_ldlm_client", m->mdt_ldlm_client);
4832
4833         ping_evictor_start();
4834
4835         /* recovery will be started upon mdt_prepare()
4836          * when the whole stack is complete and ready
4837          * to serve the requests */
4838
4839         /* Reduce the initial timeout on an MDS because it doesn't need such
4840          * a long timeout as an OST does. Adaptive timeouts will adjust this
4841          * value appropriately. */
4842         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4843                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4844
4845         RETURN(0);
4846 err_procfs:
4847         mdt_procfs_fini(m);
4848 err_recovery:
4849         target_recovery_fini(obd);
4850         upcall_cache_cleanup(m->mdt_identity_cache);
4851         m->mdt_identity_cache = NULL;
4852 err_fs_cleanup:
4853         mdt_fs_cleanup(env, m);
4854 err_tgt:
4855         tgt_fini(env, &m->mdt_lut);
4856 err_free_hsm:
4857         mdt_hsm_cdt_fini(m);
4858 err_free_ns:
4859         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4860         obd->obd_namespace = m->mdt_namespace = NULL;
4861 err_fini_seq:
4862         mdt_seq_fini(env, m);
4863 err_fini_fld:
4864         mdt_fld_fini(env, m);
4865 err_fini_stack:
4866         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4867 err_lmi:
4868         if (lmi)
4869                 server_put_mount(dev, true);
4870         return(rc);
4871 }
4872
4873 /* For interoperability, the left element is old parameter, the right one
4874  * is the new version of the parameter, if some parameter is deprecated,
4875  * the new version should be set as NULL. */
4876 static struct cfg_interop_param mdt_interop_param[] = {
4877         { "mdt.group_upcall",   NULL },
4878         { "mdt.quota_type",     NULL },
4879         { "mdd.quota_type",     NULL },
4880         { "mdt.som",            NULL },
4881         { "mdt.rootsquash",     "mdt.root_squash" },
4882         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4883         { NULL }
4884 };
4885
4886 /* used by MGS to process specific configurations */
4887 static int mdt_process_config(const struct lu_env *env,
4888                               struct lu_device *d, struct lustre_cfg *cfg)
4889 {
4890         struct mdt_device *m = mdt_dev(d);
4891         struct md_device *md_next = m->mdt_child;
4892         struct lu_device *next = md2lu_dev(md_next);
4893         int rc;
4894         ENTRY;
4895
4896         switch (cfg->lcfg_command) {
4897         case LCFG_PARAM: {
4898                 struct obd_device          *obd = d->ld_obd;
4899
4900                 /* For interoperability */
4901                 struct cfg_interop_param   *ptr = NULL;
4902                 struct lustre_cfg          *old_cfg = NULL;
4903                 char                       *param = NULL;
4904
4905                 param = lustre_cfg_string(cfg, 1);
4906                 if (param == NULL) {
4907                         CERROR("param is empty\n");
4908                         rc = -EINVAL;
4909                         break;
4910                 }
4911
4912                 ptr = class_find_old_param(param, mdt_interop_param);
4913                 if (ptr != NULL) {
4914                         if (ptr->new_param == NULL) {
4915                                 rc = 0;
4916                                 CWARN("For interoperability, skip this %s."
4917                                       " It is obsolete.\n", ptr->old_param);
4918                                 break;
4919                         }
4920
4921                         CWARN("Found old param %s, changed it to %s.\n",
4922                               ptr->old_param, ptr->new_param);
4923
4924                         old_cfg = cfg;
4925                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4926                         if (IS_ERR(cfg)) {
4927                                 rc = PTR_ERR(cfg);
4928                                 break;
4929                         }
4930                 }
4931
4932                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4933                                               cfg, obd);
4934                 if (rc > 0 || rc == -ENOSYS) {
4935                         /* is it an HSM var ? */
4936                         rc = class_process_proc_param(PARAM_HSM,
4937                                                       hsm_cdt_get_proc_vars(),
4938                                                       cfg, obd);
4939                         if (rc > 0 || rc == -ENOSYS)
4940                                 /* we don't understand; pass it on */
4941                                 rc = next->ld_ops->ldo_process_config(env, next,
4942                                                                       cfg);
4943                 }
4944
4945                 if (old_cfg != NULL)
4946                         lustre_cfg_free(cfg);
4947
4948                 break;
4949         }
4950         default:
4951                 /* others are passed further */
4952                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4953                 break;
4954         }
4955         RETURN(rc);
4956 }
4957
4958 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4959                                           const struct lu_object_header *hdr,
4960                                           struct lu_device *d)
4961 {
4962         struct mdt_object *mo;
4963
4964         ENTRY;
4965
4966         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4967         if (mo != NULL) {
4968                 struct lu_object *o;
4969                 struct lu_object_header *h;
4970
4971                 o = &mo->mot_obj;
4972                 h = &mo->mot_header;
4973                 lu_object_header_init(h);
4974                 lu_object_init(o, h, d);
4975                 lu_object_add_top(h, o);
4976                 o->lo_ops = &mdt_obj_ops;
4977                 spin_lock_init(&mo->mot_write_lock);
4978                 mutex_init(&mo->mot_lov_mutex);
4979                 init_rwsem(&mo->mot_open_sem);
4980                 atomic_set(&mo->mot_open_count, 0);
4981                 RETURN(o);
4982         }
4983         RETURN(NULL);
4984 }
4985
4986 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4987                            const struct lu_object_conf *unused)
4988 {
4989         struct mdt_device *d = mdt_dev(o->lo_dev);
4990         struct lu_device  *under;
4991         struct lu_object  *below;
4992         int                rc = 0;
4993         ENTRY;
4994
4995         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4996                PFID(lu_object_fid(o)));
4997
4998         under = &d->mdt_child->md_lu_dev;
4999         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5000         if (below != NULL) {
5001                 lu_object_add(o, below);
5002         } else
5003                 rc = -ENOMEM;
5004
5005         RETURN(rc);
5006 }
5007
5008 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5009 {
5010         struct mdt_object *mo = mdt_obj(o);
5011         struct lu_object_header *h;
5012         ENTRY;
5013
5014         h = o->lo_header;
5015         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5016                PFID(lu_object_fid(o)));
5017
5018         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5019         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5020
5021         lu_object_fini(o);
5022         lu_object_header_fini(h);
5023         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5024
5025         EXIT;
5026 }
5027
5028 static int mdt_object_print(const struct lu_env *env, void *cookie,
5029                             lu_printer_t p, const struct lu_object *o)
5030 {
5031         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5032
5033         return (*p)(env, cookie,
5034                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5035                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5036                     mdto->mot_cache_attr ? "cache_attr" : "",
5037                     mdto->mot_write_count);
5038 }
5039
5040 static int mdt_prepare(const struct lu_env *env,
5041                 struct lu_device *pdev,
5042                 struct lu_device *cdev)
5043 {
5044         struct mdt_device *mdt = mdt_dev(cdev);
5045         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5046         struct obd_device *obd = cdev->ld_obd;
5047         int rc;
5048
5049         ENTRY;
5050
5051         LASSERT(obd);
5052
5053         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5054         if (rc)
5055                 RETURN(rc);
5056
5057         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5058         if (rc)
5059                 RETURN(rc);
5060
5061         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5062         if (rc)
5063                 RETURN(rc);
5064
5065         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5066         /* The LFSCK instance is registered just now, so it must be there when
5067          * register the namespace to such instance. */
5068         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5069
5070         if (mdt->mdt_seq_site.ss_node_id == 0) {
5071                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5072                                                          &mdt->mdt_md_root_fid);
5073                 if (rc)
5074                         RETURN(rc);
5075         }
5076
5077         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5078
5079         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5080         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5081         LASSERT(obd->obd_no_conn);
5082         spin_lock(&obd->obd_dev_lock);
5083         obd->obd_no_conn = 0;
5084         spin_unlock(&obd->obd_dev_lock);
5085
5086         if (obd->obd_recovering == 0)
5087                 mdt_postrecov(env, mdt);
5088
5089         RETURN(rc);
5090 }
5091
5092 const struct lu_device_operations mdt_lu_ops = {
5093         .ldo_object_alloc   = mdt_object_alloc,
5094         .ldo_process_config = mdt_process_config,
5095         .ldo_prepare        = mdt_prepare,
5096 };
5097
5098 static const struct lu_object_operations mdt_obj_ops = {
5099         .loo_object_init    = mdt_object_init,
5100         .loo_object_free    = mdt_object_free,
5101         .loo_object_print   = mdt_object_print
5102 };
5103
5104 static int mdt_obd_set_info_async(const struct lu_env *env,
5105                                   struct obd_export *exp,
5106                                   __u32 keylen, void *key,
5107                                   __u32 vallen, void *val,
5108                                   struct ptlrpc_request_set *set)
5109 {
5110         int rc;
5111
5112         ENTRY;
5113
5114         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5115                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5116                 RETURN(rc);
5117         }
5118
5119         RETURN(0);
5120 }
5121
5122 /**
5123  * Match client and server connection feature flags.
5124  *
5125  * Compute the compatibility flags for a connection request based on
5126  * features mutually supported by client and server.
5127  *
5128  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5129  * must not be updated here, otherwise a partially initialized value may
5130  * be exposed. After the connection request is successfully processed,
5131  * the top-level MDT connect request handler atomically updates the export
5132  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5133  * reply. \see mdt_connect().
5134  *
5135  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5136  * full struct obd_connect_data. So care must be taken when accessing fields
5137  * that are not present in struct obd_connect_data_v1. See LU-16.
5138  *
5139  * \param exp   the obd_export associated with this client/target pair
5140  * \param mdt   the target device for the connection
5141  * \param data  stores data for this connect request
5142  *
5143  * \retval 0       success
5144  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5145  * \retval -EBADE  client and server feature requirements are incompatible
5146  */
5147 static int mdt_connect_internal(struct obd_export *exp,
5148                                 struct mdt_device *mdt,
5149                                 struct obd_connect_data *data)
5150 {
5151         LASSERT(data != NULL);
5152
5153         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5154
5155         if (mdt->mdt_bottom->dd_rdonly &&
5156             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5157             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5158                 RETURN(-EACCES);
5159
5160         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5161                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5162
5163         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5164
5165         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5166             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5167                 CWARN("%s: client %s does not support ibits lock, either "
5168                       "very old or an invalid client: flags %#llx\n",
5169                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5170                       data->ocd_connect_flags);
5171                 return -EBADE;
5172         }
5173
5174         if (!mdt->mdt_opts.mo_acl)
5175                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5176
5177         if (!mdt->mdt_opts.mo_user_xattr)
5178                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5179
5180         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5181                 data->ocd_brw_size = min(data->ocd_brw_size,
5182                                          (__u32)MD_MAX_BRW_SIZE);
5183                 if (data->ocd_brw_size == 0) {
5184                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
5185                                " ocd_version: %x ocd_grant: %d "
5186                                "ocd_index: %u ocd_brw_size is "
5187                                "unexpectedly zero, network data "
5188                                "corruption? Refusing connection of this"
5189                                " client\n",
5190                                mdt_obd_name(mdt),
5191                                exp->exp_client_uuid.uuid,
5192                                exp, data->ocd_connect_flags, data->ocd_version,
5193                                data->ocd_grant, data->ocd_index);
5194                         return -EPROTO;
5195                 }
5196         }
5197
5198         /* NB: Disregard the rule against updating
5199          * exp_connect_data.ocd_connect_flags in this case, since
5200          * tgt_client_new() needs to know if this is a lightweight
5201          * connection, and it is safe to expose this flag before
5202          * connection processing completes. */
5203         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5204                 spin_lock(&exp->exp_lock);
5205                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5206                 spin_unlock(&exp->exp_lock);
5207         }
5208
5209         data->ocd_version = LUSTRE_VERSION_CODE;
5210
5211         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5212                 CWARN("%s: MDS requires FID support, but client not\n",
5213                       mdt_obd_name(mdt));
5214                 return -EBADE;
5215         }
5216
5217         if (OCD_HAS_FLAG(data, PINGLESS)) {
5218                 if (ptlrpc_pinger_suppress_pings()) {
5219                         spin_lock(&exp->exp_obd->obd_dev_lock);
5220                         list_del_init(&exp->exp_obd_chain_timed);
5221                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5222                 } else {
5223                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5224                 }
5225         }
5226
5227         data->ocd_max_easize = mdt->mdt_max_ea_size;
5228
5229         /* NB: Disregard the rule against updating
5230          * exp_connect_data.ocd_connect_flags in this case, since
5231          * tgt_client_new() needs to know if this is client supports
5232          * multiple modify RPCs, and it is safe to expose this flag before
5233          * connection processing completes. */
5234         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5235                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5236                 spin_lock(&exp->exp_lock);
5237                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5238                 spin_unlock(&exp->exp_lock);
5239         }
5240
5241         return 0;
5242 }
5243
5244 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5245                                    struct mdt_thread_info *info,
5246                                    struct mdt_file_data *mfd)
5247 {
5248         struct lu_context ses;
5249         int rc;
5250         ENTRY;
5251
5252         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5253         if (rc)
5254                 RETURN(rc);
5255
5256         env->le_ses = &ses;
5257         lu_context_enter(&ses);
5258
5259         mdt_ucred(info)->uc_valid = UCRED_OLD;
5260         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5261
5262         lu_context_exit(&ses);
5263         lu_context_fini(&ses);
5264         env->le_ses = NULL;
5265
5266         RETURN(rc);
5267 }
5268
5269 static int mdt_export_cleanup(struct obd_export *exp)
5270 {
5271         struct list_head         closing_list;
5272         struct mdt_export_data  *med = &exp->exp_mdt_data;
5273         struct obd_device       *obd = exp->exp_obd;
5274         struct mdt_device       *mdt;
5275         struct mdt_thread_info  *info;
5276         struct lu_env            env;
5277         struct mdt_file_data    *mfd, *n;
5278         int rc = 0;
5279         ENTRY;
5280
5281         INIT_LIST_HEAD(&closing_list);
5282         spin_lock(&med->med_open_lock);
5283         while (!list_empty(&med->med_open_head)) {
5284                 struct list_head *tmp = med->med_open_head.next;
5285                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5286
5287                 /* Remove mfd handle so it can't be found again.
5288                  * We are consuming the mfd_list reference here. */
5289                 class_handle_unhash(&mfd->mfd_handle);
5290                 list_move_tail(&mfd->mfd_list, &closing_list);
5291         }
5292         spin_unlock(&med->med_open_lock);
5293         mdt = mdt_dev(obd->obd_lu_dev);
5294         LASSERT(mdt != NULL);
5295
5296         rc = lu_env_init(&env, LCT_MD_THREAD);
5297         if (rc)
5298                 RETURN(rc);
5299
5300         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5301         LASSERT(info != NULL);
5302         memset(info, 0, sizeof *info);
5303         info->mti_env = &env;
5304         info->mti_mdt = mdt;
5305         info->mti_exp = exp;
5306
5307         if (!list_empty(&closing_list)) {
5308                 struct md_attr *ma = &info->mti_attr;
5309
5310                 /* Close any open files (which may also cause orphan
5311                  * unlinking). */
5312                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5313                         list_del_init(&mfd->mfd_list);
5314                         ma->ma_need = ma->ma_valid = 0;
5315
5316                         /* This file is being closed due to an eviction, it
5317                          * could have been modified and now dirty regarding to
5318                          * HSM archive, check this!
5319                          * The logic here is to mark a file dirty if there's a
5320                          * chance it was dirtied before the client was evicted,
5321                          * so that we don't have to wait for a release attempt
5322                          * before finding out the file was actually dirty and
5323                          * fail the release. Aggressively marking it dirty here
5324                          * will cause the policy engine to attempt to
5325                          * re-archive it; when rearchiving, we can compare the
5326                          * current version to the HSM data_version and make the
5327                          * archive request into a noop if it's not actually
5328                          * dirty.
5329                          */
5330                         if (mfd->mfd_mode & FMODE_WRITE)
5331                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5332
5333                         /* Don't unlink orphan on failover umount, LU-184 */
5334                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5335                                 ma->ma_valid = MA_FLAGS;
5336                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5337                         }
5338                         mdt_mfd_close(info, mfd);
5339                 }
5340         }
5341         info->mti_mdt = NULL;
5342         /* cleanup client slot early */
5343         /* Do not erase record for recoverable client. */
5344         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5345                 tgt_client_del(&env, exp);
5346         lu_env_fini(&env);
5347
5348         RETURN(rc);
5349 }
5350
5351 static inline void mdt_enable_slc(struct mdt_device *mdt)
5352 {
5353         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5354                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5355 }
5356
5357 static inline void mdt_disable_slc(struct mdt_device *mdt)
5358 {
5359         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5360                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5361 }
5362
5363 static int mdt_obd_disconnect(struct obd_export *exp)
5364 {
5365         int rc;
5366         ENTRY;
5367
5368         LASSERT(exp);
5369         class_export_get(exp);
5370
5371         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5372             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5373                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5374
5375                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5376                         mdt_disable_slc(mdt);
5377         }
5378
5379         rc = server_disconnect_export(exp);
5380         if (rc != 0)
5381                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5382
5383         rc = mdt_export_cleanup(exp);
5384         nodemap_del_member(exp);
5385         class_export_put(exp);
5386         RETURN(rc);
5387 }
5388
5389 /* mds_connect copy */
5390 static int mdt_obd_connect(const struct lu_env *env,
5391                            struct obd_export **exp, struct obd_device *obd,
5392                            struct obd_uuid *cluuid,
5393                            struct obd_connect_data *data,
5394                            void *localdata)
5395 {
5396         struct obd_export       *lexp;
5397         struct lustre_handle    conn = { 0 };
5398         struct mdt_device       *mdt;
5399         int                      rc;
5400         lnet_nid_t              *client_nid = localdata;
5401         ENTRY;
5402
5403         LASSERT(env != NULL);
5404         LASSERT(data != NULL);
5405
5406         if (!exp || !obd || !cluuid)
5407                 RETURN(-EINVAL);
5408
5409         mdt = mdt_dev(obd->obd_lu_dev);
5410
5411         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5412             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5413                 atomic_inc(&mdt->mdt_mds_mds_conns);
5414                 mdt_enable_slc(mdt);
5415         }
5416
5417         /*
5418          * first, check whether the stack is ready to handle requests
5419          * XXX: probably not very appropriate method is used now
5420          *      at some point we should find a better one
5421          */
5422         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5423             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5424             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5425                 rc = obd_get_info(env, mdt->mdt_child_exp,
5426                                   sizeof(KEY_OSP_CONNECTED),
5427                                   KEY_OSP_CONNECTED, NULL, NULL);
5428                 if (rc)
5429                         RETURN(-EAGAIN);
5430                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5431         }
5432
5433         rc = class_connect(&conn, obd, cluuid);
5434         if (rc)
5435                 RETURN(rc);
5436
5437         lexp = class_conn2export(&conn);
5438         LASSERT(lexp != NULL);
5439
5440         rc = nodemap_add_member(*client_nid, lexp);
5441         if (rc != 0 && rc != -EEXIST)
5442                 GOTO(out, rc);
5443
5444         rc = mdt_connect_internal(lexp, mdt, data);
5445         if (rc == 0) {
5446                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5447
5448                 LASSERT(lcd);
5449                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5450                 rc = tgt_client_new(env, lexp);
5451                 if (rc == 0)
5452                         mdt_export_stats_init(obd, lexp, localdata);
5453         }
5454 out:
5455         if (rc != 0) {
5456                 class_disconnect(lexp);
5457                 nodemap_del_member(lexp);
5458                 *exp = NULL;
5459         } else {
5460                 *exp = lexp;
5461                 /* Because we do not want this export to be evicted by pinger,
5462                  * let's not add this export to the timed chain list. */
5463                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5464                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5465                         list_del_init(&lexp->exp_obd_chain_timed);
5466                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5467                 }
5468         }
5469
5470         RETURN(rc);
5471 }
5472
5473 static int mdt_obd_reconnect(const struct lu_env *env,
5474                              struct obd_export *exp, struct obd_device *obd,
5475                              struct obd_uuid *cluuid,
5476                              struct obd_connect_data *data,
5477                              void *localdata)
5478 {
5479         lnet_nid_t             *client_nid = localdata;
5480         int                     rc;
5481         ENTRY;
5482
5483         if (exp == NULL || obd == NULL || cluuid == NULL)
5484                 RETURN(-EINVAL);
5485
5486         rc = nodemap_add_member(*client_nid, exp);
5487         if (rc != 0 && rc != -EEXIST)
5488                 RETURN(rc);
5489
5490         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5491         if (rc == 0)
5492                 mdt_export_stats_init(obd, exp, localdata);
5493         else
5494                 nodemap_del_member(exp);
5495
5496         RETURN(rc);
5497 }
5498
5499 /* FIXME: Can we avoid using these two interfaces? */
5500 static int mdt_init_export(struct obd_export *exp)
5501 {
5502         struct mdt_export_data *med = &exp->exp_mdt_data;
5503         int                     rc;
5504         ENTRY;
5505
5506         INIT_LIST_HEAD(&med->med_open_head);
5507         spin_lock_init(&med->med_open_lock);
5508         spin_lock(&exp->exp_lock);
5509         exp->exp_connecting = 1;
5510         spin_unlock(&exp->exp_lock);
5511
5512         /* self-export doesn't need client data and ldlm initialization */
5513         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5514                                      &exp->exp_client_uuid)))
5515                 RETURN(0);
5516
5517         rc = tgt_client_alloc(exp);
5518         if (rc)
5519                 GOTO(err, rc);
5520
5521         rc = ldlm_init_export(exp);
5522         if (rc)
5523                 GOTO(err_free, rc);
5524
5525         RETURN(rc);
5526
5527 err_free:
5528         tgt_client_free(exp);
5529 err:
5530         CERROR("%s: Failed to initialize export: rc = %d\n",
5531                exp->exp_obd->obd_name, rc);
5532         return rc;
5533 }
5534
5535 static int mdt_destroy_export(struct obd_export *exp)
5536 {
5537         ENTRY;
5538
5539         target_destroy_export(exp);
5540         /* destroy can be called from failed obd_setup, so
5541          * checking uuid is safer than obd_self_export */
5542         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5543                                      &exp->exp_client_uuid)))
5544                 RETURN(0);
5545
5546         ldlm_destroy_export(exp);
5547         tgt_client_free(exp);
5548
5549         LASSERT(list_empty(&exp->exp_outstanding_replies));
5550         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5551
5552         RETURN(0);
5553 }
5554
5555 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5556                    struct linkea_data *ldata)
5557 {
5558         int rc;
5559
5560         LASSERT(ldata->ld_buf->lb_buf != NULL);
5561
5562         if (!mdt_object_exists(mdt_obj))
5563                 return -ENODATA;
5564
5565         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5566                           ldata->ld_buf, XATTR_NAME_LINK);
5567         if (rc == -ERANGE) {
5568                 /* Buf was too small, figure out what we need. */
5569                 lu_buf_free(ldata->ld_buf);
5570                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5571                                   ldata->ld_buf, XATTR_NAME_LINK);
5572                 if (rc < 0)
5573                         return rc;
5574                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5575                 if (ldata->ld_buf->lb_buf == NULL)
5576                         return -ENOMEM;
5577                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5578                                   ldata->ld_buf, XATTR_NAME_LINK);
5579         }
5580         if (rc < 0)
5581                 return rc;
5582
5583         return linkea_init_with_rec(ldata);
5584 }
5585
5586 /**
5587  * Given an MDT object, try to look up the full path to the object.
5588  * Part of the MDT layer implementation of lfs fid2path.
5589  *
5590  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5591  * \param[in]     obj   Object to do path lookup of
5592  * \param[in,out] fp    User-provided struct to store path information
5593  * \param[in]     root_fid Root FID of current path should reach
5594  *
5595  * \retval 0 Lookup successful, path information stored in fp
5596  * \retval -EAGAIN Lookup failed, usually because object is being moved
5597  * \retval negative errno if there was a problem
5598  */
5599 static int mdt_path_current(struct mdt_thread_info *info,
5600                             struct mdt_object *obj,
5601                             struct getinfo_fid2path *fp,
5602                             struct lu_fid *root_fid)
5603 {
5604         struct mdt_device       *mdt = info->mti_mdt;
5605         struct mdt_object       *mdt_obj;
5606         struct link_ea_header   *leh;
5607         struct link_ea_entry    *lee;
5608         struct lu_name          *tmpname = &info->mti_name;
5609         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5610         struct lu_buf           *buf = &info->mti_big_buf;
5611         char                    *ptr;
5612         int                     reclen;
5613         struct linkea_data      ldata = { NULL };
5614         int                     rc = 0;
5615         bool                    first = true;
5616         ENTRY;
5617
5618         /* temp buffer for path element, the buffer will be finally freed
5619          * in mdt_thread_info_fini */
5620         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5621         if (buf->lb_buf == NULL)
5622                 RETURN(-ENOMEM);
5623
5624         ldata.ld_buf = buf;
5625         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5626         *ptr = 0;
5627         --ptr;
5628         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5629
5630         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5631                 struct lu_buf           lmv_buf;
5632
5633                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5634                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5635                         GOTO(out, rc = -ENOENT);
5636
5637                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5638                         mdt_obj = obj;
5639                         mdt_object_get(info->mti_env, mdt_obj);
5640                 } else {
5641                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5642                         if (IS_ERR(mdt_obj))
5643                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5644                 }
5645
5646                 if (!mdt_object_exists(mdt_obj)) {
5647                         mdt_object_put(info->mti_env, mdt_obj);
5648                         GOTO(out, rc = -ENOENT);
5649                 }
5650
5651                 if (mdt_object_remote(mdt_obj)) {
5652                         mdt_object_put(info->mti_env, mdt_obj);
5653                         GOTO(remote_out, rc = -EREMOTE);
5654                 }
5655
5656                 rc = mdt_links_read(info, mdt_obj, &ldata);
5657                 if (rc != 0) {
5658                         mdt_object_put(info->mti_env, mdt_obj);
5659                         GOTO(out, rc);
5660                 }
5661
5662                 leh = buf->lb_buf;
5663                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5664                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5665                 /* If set, use link #linkno for path lookup, otherwise use
5666                    link #0.  Only do this for the final path element. */
5667                 if (first && fp->gf_linkno < leh->leh_reccount) {
5668                         int count;
5669                         for (count = 0; count < fp->gf_linkno; count++) {
5670                                 lee = (struct link_ea_entry *)
5671                                      ((char *)lee + reclen);
5672                                 linkea_entry_unpack(lee, &reclen, tmpname,
5673                                                     tmpfid);
5674                         }
5675                         if (fp->gf_linkno < leh->leh_reccount - 1)
5676                                 /* indicate to user there are more links */
5677                                 fp->gf_linkno++;
5678                 }
5679
5680                 lmv_buf.lb_buf = info->mti_xattr_buf;
5681                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5682                 /* Check if it is slave stripes */
5683                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5684                                   &lmv_buf, XATTR_NAME_LMV);
5685                 mdt_object_put(info->mti_env, mdt_obj);
5686                 if (rc > 0) {
5687                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5688
5689                         /* For slave stripes, get its master */
5690                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5691                                 fp->gf_fid = *tmpfid;
5692                                 continue;
5693                         }
5694                 } else if (rc < 0 && rc != -ENODATA) {
5695                         GOTO(out, rc);
5696                 }
5697
5698                 rc = 0;
5699
5700                 /* Pack the name in the end of the buffer */
5701                 ptr -= tmpname->ln_namelen;
5702                 if (ptr - 1 <= fp->gf_u.gf_path)
5703                         GOTO(out, rc = -EOVERFLOW);
5704                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5705                 *(--ptr) = '/';
5706
5707                 /* keep the last resolved fid to the client, so the
5708                  * client will build the left path on another MDT for
5709                  * remote object */
5710                 fp->gf_fid = *tmpfid;
5711
5712                 first = false;
5713         }
5714
5715 remote_out:
5716         ptr++; /* skip leading / */
5717         memmove(fp->gf_u.gf_path, ptr,
5718                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5719
5720 out:
5721         RETURN(rc);
5722 }
5723
5724 /**
5725  * Given an MDT object, use mdt_path_current to get the path.
5726  * Essentially a wrapper to retry mdt_path_current a set number of times
5727  * if -EAGAIN is returned (usually because an object is being moved).
5728  *
5729  * Part of the MDT layer implementation of lfs fid2path.
5730  *
5731  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5732  * \param[in]     obj   Object to do path lookup of
5733  * \param[in,out] fp    User-provided struct for arguments and to store path
5734  *                      information
5735  *
5736  * \retval 0 Lookup successful, path information stored in fp
5737  * \retval negative errno if there was a problem
5738  */
5739 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5740                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5741 {
5742         struct mdt_device       *mdt = info->mti_mdt;
5743         int                     tries = 3;
5744         int                     rc = -EAGAIN;
5745         ENTRY;
5746
5747         if (fp->gf_pathlen < 3)
5748                 RETURN(-EOVERFLOW);
5749
5750         if (root_fid == NULL)
5751                 root_fid = &mdt->mdt_md_root_fid;
5752
5753         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5754                 fp->gf_u.gf_path[0] = '\0';
5755                 RETURN(0);
5756         }
5757
5758         /* Retry multiple times in case file is being moved */
5759         while (tries-- && rc == -EAGAIN)
5760                 rc = mdt_path_current(info, obj, fp, root_fid);
5761
5762         RETURN(rc);
5763 }
5764
5765 /**
5766  * Get the full path of the provided FID, as of changelog record recno.
5767  *
5768  * This checks sanity and looks up object for user provided FID
5769  * before calling the actual path lookup code.
5770  *
5771  * Part of the MDT layer implementation of lfs fid2path.
5772  *
5773  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5774  * \param[in,out] fp    User-provided struct for arguments and to store path
5775  *                      information
5776  *
5777  * \retval 0 Lookup successful, path information and recno stored in fp
5778  * \retval -ENOENT, object does not exist
5779  * \retval negative errno if there was a problem
5780  */
5781 static int mdt_fid2path(struct mdt_thread_info *info,
5782                         struct lu_fid *root_fid,
5783                         struct getinfo_fid2path *fp)
5784 {
5785         struct mdt_device *mdt = info->mti_mdt;
5786         struct mdt_object *obj;
5787         int    rc;
5788         ENTRY;
5789
5790         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
5791                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5792
5793         if (!fid_is_sane(&fp->gf_fid))
5794                 RETURN(-EINVAL);
5795
5796         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5797                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
5798                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
5799                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5800                 RETURN(-EINVAL);
5801         }
5802
5803         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5804         if (IS_ERR(obj)) {
5805                 rc = PTR_ERR(obj);
5806                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5807                        PFID(&fp->gf_fid), rc);
5808                 RETURN(rc);
5809         }
5810
5811         if (mdt_object_remote(obj))
5812                 rc = -EREMOTE;
5813         else if (!mdt_object_exists(obj))
5814                 rc = -ENOENT;
5815         else
5816                 rc = 0;
5817
5818         if (rc < 0) {
5819                 mdt_object_put(info->mti_env, obj);
5820                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5821                        PFID(&fp->gf_fid), rc);
5822                 RETURN(rc);
5823         }
5824
5825         rc = mdt_path(info, obj, fp, root_fid);
5826
5827         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
5828                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5829                fp->gf_recno, fp->gf_linkno);
5830
5831         mdt_object_put(info->mti_env, obj);
5832
5833         RETURN(rc);
5834 }
5835
5836 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5837                             void *val, int vallen)
5838 {
5839         struct getinfo_fid2path *fpout, *fpin;
5840         struct lu_fid *root_fid = NULL;
5841         int rc = 0;
5842
5843         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5844         fpout = val;
5845
5846         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5847                 lustre_swab_fid2path(fpin);
5848
5849         memcpy(fpout, fpin, sizeof(*fpin));
5850         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5851                 RETURN(-EINVAL);
5852
5853         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5854                       sizeof(struct lu_fid)) {
5855                 /* client sent its root FID, which is normally fileset FID */
5856                 root_fid = fpin->gf_u.gf_root_fid;
5857                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5858                         lustre_swab_lu_fid(root_fid);
5859
5860                 if (root_fid != NULL && !fid_is_sane(root_fid))
5861                         RETURN(-EINVAL);
5862         }
5863
5864         rc = mdt_fid2path(info, root_fid, fpout);
5865         RETURN(rc);
5866 }
5867
5868 int mdt_get_info(struct tgt_session_info *tsi)
5869 {
5870         char    *key;
5871         int      keylen;
5872         __u32   *vallen;
5873         void    *valout;
5874         int      rc;
5875
5876         ENTRY;
5877
5878         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5879         if (key == NULL) {
5880                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5881                 RETURN(err_serious(-EFAULT));
5882         }
5883         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5884                                       RCL_CLIENT);
5885
5886         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5887         if (vallen == NULL) {
5888                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5889                                 tgt_name(tsi->tsi_tgt));
5890                 RETURN(err_serious(-EFAULT));
5891         }
5892
5893         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5894                              *vallen);
5895         rc = req_capsule_server_pack(tsi->tsi_pill);
5896         if (rc)
5897                 RETURN(err_serious(rc));
5898
5899         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5900         if (valout == NULL) {
5901                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5902                                 tgt_name(tsi->tsi_tgt));
5903                 RETURN(err_serious(-EFAULT));
5904         }
5905
5906         if (KEY_IS(KEY_FID2PATH)) {
5907                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5908
5909                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5910                 mdt_thread_info_fini(info);
5911         } else {
5912                 rc = -EINVAL;
5913         }
5914         RETURN(rc);
5915 }
5916
5917 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5918 {
5919         struct obd_ioctl_data *data = karg;
5920         struct lu_fid *fid;
5921         __u64 version;
5922         struct mdt_object *obj;
5923         struct mdt_lock_handle  *lh;
5924         int rc;
5925         ENTRY;
5926
5927         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5928             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5929                 RETURN(-EINVAL);
5930
5931         fid = (struct lu_fid *)data->ioc_inlbuf1;
5932
5933         if (!fid_is_sane(fid))
5934                 RETURN(-EINVAL);
5935
5936         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5937
5938         lh = &mti->mti_lh[MDT_LH_PARENT];
5939         mdt_lock_reg_init(lh, LCK_CR);
5940
5941         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5942         if (IS_ERR(obj))
5943                 RETURN(PTR_ERR(obj));
5944
5945         if (mdt_object_remote(obj)) {
5946                 rc = -EREMOTE;
5947                 /**
5948                  * before calling version get the correct MDS should be
5949                  * fid, this is error to find remote object here
5950                  */
5951                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5952         } else if (!mdt_object_exists(obj)) {
5953                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5954                 rc = -ENOENT;
5955         } else {
5956                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5957                *(__u64 *)data->ioc_inlbuf2 = version;
5958                 rc = 0;
5959         }
5960         mdt_object_unlock_put(mti, obj, lh, 1);
5961         RETURN(rc);
5962 }
5963
5964 /* ioctls on obd dev */
5965 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5966                          void *karg, void __user *uarg)
5967 {
5968         struct lu_env      env;
5969         struct obd_device *obd = exp->exp_obd;
5970         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5971         struct dt_device  *dt = mdt->mdt_bottom;
5972         int rc;
5973
5974         ENTRY;
5975         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5976         rc = lu_env_init(&env, LCT_MD_THREAD);
5977         if (rc)
5978                 RETURN(rc);
5979
5980         switch (cmd) {
5981         case OBD_IOC_SYNC:
5982                 rc = mdt_device_sync(&env, mdt);
5983                 break;
5984         case OBD_IOC_SET_READONLY:
5985                 rc = dt_sync(&env, dt);
5986                 if (rc == 0)
5987                         rc = dt_ro(&env, dt);
5988                 break;
5989         case OBD_IOC_ABORT_RECOVERY:
5990                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5991                 obd->obd_abort_recovery = 1;
5992                 target_stop_recovery_thread(obd);
5993                 rc = 0;
5994                 break;
5995         case OBD_IOC_CHANGELOG_REG:
5996         case OBD_IOC_CHANGELOG_DEREG:
5997         case OBD_IOC_CHANGELOG_CLEAR:
5998                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
5999                                                            mdt->mdt_child,
6000                                                            cmd, len, karg);
6001                 break;
6002         case OBD_IOC_START_LFSCK: {
6003                 struct md_device *next = mdt->mdt_child;
6004                 struct obd_ioctl_data *data = karg;
6005                 struct lfsck_start_param lsp;
6006
6007                 if (unlikely(data == NULL)) {
6008                         rc = -EINVAL;
6009                         break;
6010                 }
6011
6012                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6013                 lsp.lsp_index_valid = 0;
6014                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6015                 break;
6016         }
6017         case OBD_IOC_STOP_LFSCK: {
6018                 struct md_device        *next = mdt->mdt_child;
6019                 struct obd_ioctl_data   *data = karg;
6020                 struct lfsck_stop        stop;
6021
6022                 stop.ls_status = LS_STOPPED;
6023                 /* Old lfsck utils may pass NULL @stop. */
6024                 if (data->ioc_inlbuf1 == NULL)
6025                         stop.ls_flags = 0;
6026                 else
6027                         stop.ls_flags =
6028                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6029
6030                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6031                 break;
6032         }
6033         case OBD_IOC_QUERY_LFSCK: {
6034                 struct md_device        *next = mdt->mdt_child;
6035                 struct obd_ioctl_data   *data = karg;
6036
6037                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6038                                                  data->ioc_inlbuf1);
6039                 break;
6040         }
6041         case OBD_IOC_GET_OBJ_VERSION: {
6042                 struct mdt_thread_info *mti;
6043                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6044                 memset(mti, 0, sizeof *mti);
6045                 mti->mti_env = &env;
6046                 mti->mti_mdt = mdt;
6047                 mti->mti_exp = exp;
6048
6049                 rc = mdt_ioc_version_get(mti, karg);
6050                 break;
6051         }
6052         case OBD_IOC_CATLOGLIST: {
6053                 struct mdt_thread_info *mti;
6054
6055                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6056                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6057                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6058                                        &mti->mti_tmp_fid1);
6059                 break;
6060          }
6061         default:
6062                 rc = -EOPNOTSUPP;
6063                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6064                         mdt_obd_name(mdt), cmd, rc);
6065         }
6066
6067         lu_env_fini(&env);
6068         RETURN(rc);
6069 }
6070
6071 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6072 {
6073         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6074         int rc;
6075         ENTRY;
6076
6077         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6078                 struct lfsck_start_param lsp;
6079
6080                 lsp.lsp_start = NULL;
6081                 lsp.lsp_index_valid = 0;
6082                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6083                                                            OBD_IOC_START_LFSCK,
6084                                                            0, &lsp);
6085                 if (rc != 0 && rc != -EALREADY)
6086                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6087                               mdt_obd_name(mdt), rc);
6088         }
6089
6090         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6091         RETURN(rc);
6092 }
6093
6094 static int mdt_obd_postrecov(struct obd_device *obd)
6095 {
6096         struct lu_env env;
6097         int rc;
6098
6099         rc = lu_env_init(&env, LCT_MD_THREAD);
6100         if (rc)
6101                 RETURN(rc);
6102         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6103         lu_env_fini(&env);
6104         return rc;
6105 }
6106
6107 static struct obd_ops mdt_obd_device_ops = {
6108         .o_owner          = THIS_MODULE,
6109         .o_set_info_async = mdt_obd_set_info_async,
6110         .o_connect        = mdt_obd_connect,
6111         .o_reconnect      = mdt_obd_reconnect,
6112         .o_disconnect     = mdt_obd_disconnect,
6113         .o_init_export    = mdt_init_export,
6114         .o_destroy_export = mdt_destroy_export,
6115         .o_iocontrol      = mdt_iocontrol,
6116         .o_postrecov      = mdt_obd_postrecov,
6117 };
6118
6119 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6120                                          struct lu_device *d)
6121 {
6122         struct mdt_device *m = mdt_dev(d);
6123         ENTRY;
6124
6125         mdt_fini(env, m);
6126         RETURN(NULL);
6127 }
6128
6129 static struct lu_device *mdt_device_free(const struct lu_env *env,
6130                                          struct lu_device *d)
6131 {
6132         struct mdt_device *m = mdt_dev(d);
6133         ENTRY;
6134
6135         lu_device_fini(&m->mdt_lu_dev);
6136         OBD_FREE_PTR(m);
6137
6138         RETURN(NULL);
6139 }
6140
6141 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6142                                           struct lu_device_type *t,
6143                                           struct lustre_cfg *cfg)
6144 {
6145         struct lu_device  *l;
6146         struct mdt_device *m;
6147
6148         OBD_ALLOC_PTR(m);
6149         if (m != NULL) {
6150                 int rc;
6151
6152                 l = &m->mdt_lu_dev;
6153                 rc = mdt_init0(env, m, t, cfg);
6154                 if (rc != 0) {
6155                         mdt_device_free(env, l);
6156                         l = ERR_PTR(rc);
6157                         return l;
6158                 }
6159         } else
6160                 l = ERR_PTR(-ENOMEM);
6161         return l;
6162 }
6163
6164 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6165 LU_KEY_INIT(mdt, struct mdt_thread_info);
6166
6167 static void mdt_key_fini(const struct lu_context *ctx,
6168                          struct lu_context_key *key, void* data)
6169 {
6170         struct mdt_thread_info *info = data;
6171
6172         if (info->mti_big_lmm) {
6173                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6174                 info->mti_big_lmm = NULL;
6175                 info->mti_big_lmmsize = 0;
6176         }
6177         OBD_FREE_PTR(info);
6178 }
6179
6180 /* context key: mdt_thread_key */
6181 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6182
6183 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6184 {
6185         return lu_ucred(info->mti_env);
6186 }
6187
6188 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6189 {
6190         return lu_ucred_check(info->mti_env);
6191 }
6192
6193 /**
6194  * Enable/disable COS (Commit On Sharing).
6195  *
6196  * Set/Clear the COS flag in mdt options.
6197  *
6198  * \param mdt mdt device
6199  * \param val 0 disables COS, other values enable COS
6200  */
6201 void mdt_enable_cos(struct mdt_device *mdt, int val)
6202 {
6203         struct lu_env env;
6204         int rc;
6205
6206         mdt->mdt_opts.mo_cos = !!val;
6207         rc = lu_env_init(&env, LCT_LOCAL);
6208         if (unlikely(rc != 0)) {
6209                 CWARN("%s: lu_env initialization failed, cannot "
6210                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6211                 return;
6212         }
6213         mdt_device_sync(&env, mdt);
6214         lu_env_fini(&env);
6215 }
6216
6217 /**
6218  * Check COS (Commit On Sharing) status.
6219  *
6220  * Return COS flag status.
6221  *
6222  * \param mdt mdt device
6223  */
6224 int mdt_cos_is_enabled(struct mdt_device *mdt)
6225 {
6226         return mdt->mdt_opts.mo_cos != 0;
6227 }
6228
6229 static struct lu_device_type_operations mdt_device_type_ops = {
6230         .ldto_device_alloc = mdt_device_alloc,
6231         .ldto_device_free  = mdt_device_free,
6232         .ldto_device_fini  = mdt_device_fini
6233 };
6234
6235 static struct lu_device_type mdt_device_type = {
6236         .ldt_tags     = LU_DEVICE_MD,
6237         .ldt_name     = LUSTRE_MDT_NAME,
6238         .ldt_ops      = &mdt_device_type_ops,
6239         .ldt_ctx_tags = LCT_MD_THREAD
6240 };
6241
6242 static int __init mdt_init(void)
6243 {
6244         int rc;
6245
6246         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6247                  FID_NOBRACE_LEN + 1);
6248         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6249                  FID_LEN + 1);
6250         rc = lu_kmem_init(mdt_caches);
6251         if (rc)
6252                 return rc;
6253
6254         rc = mds_mod_init();
6255         if (rc)
6256                 GOTO(lu_fini, rc);
6257
6258         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6259                                  LUSTRE_MDT_NAME, &mdt_device_type);
6260         if (rc)
6261                 GOTO(mds_fini, rc);
6262 lu_fini:
6263         if (rc)
6264                 lu_kmem_fini(mdt_caches);
6265 mds_fini:
6266         if (rc)
6267                 mds_mod_exit();
6268         return rc;
6269 }
6270
6271 static void __exit mdt_exit(void)
6272 {
6273         class_unregister_type(LUSTRE_MDT_NAME);
6274         mds_mod_exit();
6275         lu_kmem_fini(mdt_caches);
6276 }
6277
6278 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6279 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6280 MODULE_VERSION(LUSTRE_VERSION_STRING);
6281 MODULE_LICENSE("GPL");
6282
6283 module_init(mdt_init);
6284 module_exit(mdt_exit);