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 <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 /* For HSM restore handles */
 105 struct kmem_cache *mdt_hsm_cdt_kmem;
 106
 107 /* For HSM request handles */
 108 struct kmem_cache *mdt_hsm_car_kmem;
 109
 110 static struct lu_kmem_descr mdt_caches[] = {
 111         {
 112                 .ckd_cache = &mdt_object_kmem,
 113                 .ckd_name  = "mdt_obj",
 114                 .ckd_size  = sizeof(struct mdt_object)
 115         },
 116         {
 117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 118                 .ckd_name       = "mdt_cdt_restore_handle",
 119                 .ckd_size       = sizeof(struct cdt_restore_handle)
 120         },
 121         {
 122                 .ckd_cache      = &mdt_hsm_car_kmem,
 123                 .ckd_name       = "mdt_cdt_agent_req",
 124                 .ckd_size       = sizeof(struct cdt_agent_req)
 125         },
 126         {
 127                 .ckd_cache = NULL
 128         }
 129 };
 130
 131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 132 {
 133         if (!rep)
 134                 return 0;
 135         return rep->lock_policy_res1 & op_flag;
 136 }
 137
 138 void mdt_clear_disposition(struct mdt_thread_info *info,
 139                            struct ldlm_reply *rep, __u64 op_flag)
 140 {
 141         if (info) {
 142                 info->mti_opdata &= ~op_flag;
 143                 tgt_opdata_clear(info->mti_env, op_flag);
 144         }
 145         if (rep)
 146                 rep->lock_policy_res1 &= ~op_flag;
 147 }
 148
 149 void mdt_set_disposition(struct mdt_thread_info *info,
 150                          struct ldlm_reply *rep, __u64 op_flag)
 151 {
 152         if (info) {
 153                 info->mti_opdata |= op_flag;
 154                 tgt_opdata_set(info->mti_env, op_flag);
 155         }
 156         if (rep)
 157                 rep->lock_policy_res1 |= op_flag;
 158 }
 159
 160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 161 {
 162         lh->mlh_pdo_hash = 0;
 163         lh->mlh_reg_mode = lm;
 164         lh->mlh_rreg_mode = lm;
 165         lh->mlh_type = MDT_REG_LOCK;
 166 }
 167
 168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 169                        const struct lu_name *lname)
 170 {
 171         lh->mlh_reg_mode = lock_mode;
 172         lh->mlh_pdo_mode = LCK_MINMODE;
 173         lh->mlh_rreg_mode = lock_mode;
 174         lh->mlh_type = MDT_PDO_LOCK;
 175
 176         if (lu_name_is_valid(lname)) {
 177                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 178                                                      lname->ln_namelen);
 179                 /* XXX Workaround for LU-2856
 180                  *
 181                  * Zero is a valid return value of full_name_hash, but
 182                  * several users of mlh_pdo_hash assume a non-zero
 183                  * hash value. We therefore map zero onto an
 184                  * arbitrary, but consistent value (1) to avoid
 185                  * problems further down the road. */
 186                 if (unlikely(lh->mlh_pdo_hash == 0))
 187                         lh->mlh_pdo_hash = 1;
 188         } else {
 189                 lh->mlh_pdo_hash = 0;
 190         }
 191 }
 192
 193 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 194                               struct mdt_lock_handle *lh)
 195 {
 196         mdl_mode_t mode;
 197         ENTRY;
 198
 199         /*
 200          * Any dir access needs couple of locks:
 201          *
 202          * 1) on part of dir we gonna take lookup/modify;
 203          *
 204          * 2) on whole dir to protect it from concurrent splitting and/or to
 205          * flush client's cache for readdir().
 206          *
 207          * so, for a given mode and object this routine decides what lock mode
 208          * to use for lock #2:
 209          *
 210          * 1) if caller's gonna lookup in dir then we need to protect dir from
 211          * being splitted only - LCK_CR
 212          *
 213          * 2) if caller's gonna modify dir then we need to protect dir from
 214          * being splitted and to flush cache - LCK_CW
 215          *
 216          * 3) if caller's gonna modify dir and that dir seems ready for
 217          * splitting then we need to protect it from any type of access
 218          * (lookup/modify/split) - LCK_EX --bzzz
 219          */
 220
 221         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 222         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 223
 224         /*
 225          * Ask underlaying level its opinion about preferable PDO lock mode
 226          * having access type passed as regular lock mode:
 227          *
 228          * - MDL_MINMODE means that lower layer does not want to specify lock
 229          * mode;
 230          *
 231          * - MDL_NL means that no PDO lock should be taken. This is used in some
 232          * cases. Say, for non-splittable directories no need to use PDO locks
 233          * at all.
 234          */
 235         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 236                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 237
 238         if (mode != MDL_MINMODE) {
 239                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 240         } else {
 241                 /*
 242                  * Lower layer does not want to specify locking mode. We do it
 243                  * our selves. No special protection is needed, just flush
 244                  * client's cache on modification and allow concurrent
 245                  * mondification.
 246                  */
 247                 switch (lh->mlh_reg_mode) {
 248                 case LCK_EX:
 249                         lh->mlh_pdo_mode = LCK_EX;
 250                         break;
 251                 case LCK_PR:
 252                         lh->mlh_pdo_mode = LCK_CR;
 253                         break;
 254                 case LCK_PW:
 255                         lh->mlh_pdo_mode = LCK_CW;
 256                         break;
 257                 default:
 258                         CERROR("Not expected lock type (0x%x)\n",
 259                                (int)lh->mlh_reg_mode);
 260                         LBUG();
 261                 }
 262         }
 263
 264         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 265         EXIT;
 266 }
 267
 268 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 269                               struct lu_fid *fid)
 270 {
 271         struct mdt_device *mdt = info->mti_mdt;
 272         struct lu_name *lname = &info->mti_name;
 273         char *name = NULL;
 274         struct mdt_object *parent;
 275         u32 mode;
 276         int rc = 0;
 277
 278         LASSERT(!info->mti_cross_ref);
 279
 280         OBD_ALLOC(name, NAME_MAX + 1);
 281         if (name == NULL)
 282                 return -ENOMEM;
 283         lname->ln_name = name;
 284
 285         /*
 286          * We may want to allow this to mount a completely separate
 287          * fileset from the MDT in the future, but keeping it to
 288          * ROOT/ only for now avoid potential security issues.
 289          */
 290         *fid = mdt->mdt_md_root_fid;
 291
 292         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 293                 const char *s1 = fileset;
 294                 const char *s2;
 295
 296                 while (*++s1 == '/')
 297                         ;
 298                 s2 = s1;
 299                 while (*s2 != '/' && *s2 != '\0')
 300                         s2++;
 301
 302                 if (s2 == s1)
 303                         break;
 304
 305                 fileset = s2;
 306
 307                 lname->ln_namelen = s2 - s1;
 308                 if (lname->ln_namelen > NAME_MAX) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 /* reject .. as a path component */
 314                 if (lname->ln_namelen == 2 &&
 315                     strncmp(s1, "..", 2) == 0) {
 316                         rc = -EINVAL;
 317                         break;
 318                 }
 319
 320                 strncpy(name, s1, lname->ln_namelen);
 321                 name[lname->ln_namelen] = '\0';
 322
 323                 parent = mdt_object_find(info->mti_env, mdt, fid);
 324                 if (IS_ERR(parent)) {
 325                         rc = PTR_ERR(parent);
 326                         break;
 327                 }
 328                 /* Only got the fid of this obj by name */
 329                 fid_zero(fid);
 330                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 331                                 fid, &info->mti_spec);
 332                 mdt_object_put(info->mti_env, parent);
 333         }
 334         if (!rc) {
 335                 parent = mdt_object_find(info->mti_env, mdt, fid);
 336                 if (IS_ERR(parent))
 337                         rc = PTR_ERR(parent);
 338                 else {
 339                         mode = lu_object_attr(&parent->mot_obj);
 340                         mdt_object_put(info->mti_env, parent);
 341                         if (!S_ISDIR(mode))
 342                                 rc = -ENOTDIR;
 343                 }
 344         }
 345
 346         OBD_FREE(name, NAME_MAX + 1);
 347
 348         return rc;
 349 }
 350
 351 static int mdt_get_root(struct tgt_session_info *tsi)
 352 {
 353         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 354         struct mdt_device       *mdt = info->mti_mdt;
 355         struct mdt_body         *repbody;
 356         char                    *fileset = NULL, *buffer = NULL;
 357         int                      rc;
 358         struct obd_export       *exp = info->mti_exp;
 359         char                    *nodemap_fileset;
 360
 361         ENTRY;
 362
 363         rc = mdt_check_ucred(info);
 364         if (rc)
 365                 GOTO(out, rc = err_serious(rc));
 366
 367         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 368                 GOTO(out, rc = err_serious(-ENOMEM));
 369
 370         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 372                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 373                 if (fileset == NULL)
 374                         GOTO(out, rc = err_serious(-EFAULT));
 375         }
 376
 377         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 378         if (nodemap_fileset && nodemap_fileset[0]) {
 379                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 380                 if (fileset) {
 381                         /* consider fileset from client as a sub-fileset
 382                          * of the nodemap one */
 383                         OBD_ALLOC(buffer, PATH_MAX + 1);
 384                         if (buffer == NULL)
 385                                 GOTO(out, rc = err_serious(-ENOMEM));
 386                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 387                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 388                                 GOTO(out, rc = err_serious(-EINVAL));
 389                         fileset = buffer;
 390                 } else {
 391                         /* enforce fileset as specified in the nodemap */
 392                         fileset = nodemap_fileset;
 393                 }
 394         }
 395
 396         if (fileset) {
 397                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 398                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 399                 if (rc < 0)
 400                         GOTO(out, rc = err_serious(rc));
 401         } else {
 402                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 403         }
 404         repbody->mbo_valid |= OBD_MD_FLID;
 405
 406         EXIT;
 407 out:
 408         mdt_thread_info_fini(info);
 409         if (buffer)
 410                 OBD_FREE(buffer, PATH_MAX+1);
 411         return rc;
 412 }
 413
 414 static int mdt_statfs(struct tgt_session_info *tsi)
 415 {
 416         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 417         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 418         struct md_device                *next = info->mti_mdt->mdt_child;
 419         struct ptlrpc_service_part      *svcpt;
 420         struct obd_statfs               *osfs;
 421         int                             rc;
 422
 423         ENTRY;
 424
 425         svcpt = req->rq_rqbd->rqbd_svcpt;
 426
 427         /* This will trigger a watchdog timeout */
 428         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 429                          (MDT_SERVICE_WATCHDOG_FACTOR *
 430                           at_get(&svcpt->scp_at_estimate)) + 1);
 431
 432         rc = mdt_check_ucred(info);
 433         if (rc)
 434                 GOTO(out, rc = err_serious(rc));
 435
 436         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 437                 GOTO(out, rc = err_serious(-ENOMEM));
 438
 439         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 440         if (!osfs)
 441                 GOTO(out, rc = -EPROTO);
 442
 443         /** statfs information are cached in the mdt_device */
 444         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 445                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 446                 /** statfs data is too old, get up-to-date one */
 447                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 448                 if (rc)
 449                         GOTO(out, rc);
 450                 spin_lock(&info->mti_mdt->mdt_lock);
 451                 info->mti_mdt->mdt_osfs = *osfs;
 452                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 453                 spin_unlock(&info->mti_mdt->mdt_lock);
 454         } else {
 455                 /** use cached statfs data */
 456                 spin_lock(&info->mti_mdt->mdt_lock);
 457                 *osfs = info->mti_mdt->mdt_osfs;
 458                 spin_unlock(&info->mti_mdt->mdt_lock);
 459         }
 460
 461         if (rc == 0)
 462                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 463 out:
 464         mdt_thread_info_fini(info);
 465         RETURN(rc);
 466 }
 467
 468 #ifdef CONFIG_FS_POSIX_ACL
 469 /*
 470  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 471  *
 472  * \param       info    thread info object
 473  * \param       repbody reply to pack ACLs into
 474  * \param       o       mdt object of file to examine
 475  * \param       nodemap nodemap of client to reply to
 476  * \retval      0       success
 477  * \retval      -errno  error getting or parsing ACL from disk
 478  */
 479 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 480                       struct mdt_object *o, struct lu_nodemap *nodemap)
 481 {
 482         const struct lu_env     *env = info->mti_env;
 483         struct md_object        *next = mdt_object_child(o);
 484         struct lu_buf           *buf = &info->mti_buf;
 485         struct mdt_device       *mdt = info->mti_mdt;
 486         int rc;
 487
 488         ENTRY;
 489
 490         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 491         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 492                                            RCL_SERVER);
 493         if (buf->lb_len == 0)
 494                 RETURN(0);
 495
 496 again:
 497         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 498         if (rc < 0) {
 499                 if (rc == -ENODATA) {
 500                         repbody->mbo_aclsize = 0;
 501                         repbody->mbo_valid |= OBD_MD_FLACL;
 502                         rc = 0;
 503                 } else if (rc == -EOPNOTSUPP) {
 504                         rc = 0;
 505                 } else {
 506                         if (rc == -ERANGE &&
 507                             exp_connect_large_acl(info->mti_exp) &&
 508                             buf->lb_buf != info->mti_big_acl) {
 509                                 if (info->mti_big_acl == NULL) {
 510                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 511                                                         mdt->mdt_max_ea_size);
 512                                         if (info->mti_big_acl == NULL) {
 513                                                 CERROR("%s: unable to grow "
 514                                                        DFID" ACL buffer\n",
 515                                                        mdt_obd_name(mdt),
 516                                                        PFID(mdt_object_fid(o)));
 517                                                 RETURN(-ENOMEM);
 518                                         }
 519
 520                                         info->mti_big_aclsize =
 521                                                         mdt->mdt_max_ea_size;
 522                                 }
 523
 524                                 CDEBUG(D_INODE, "%s: grow the "DFID
 525                                        " ACL buffer to size %d\n",
 526                                        mdt_obd_name(mdt),
 527                                        PFID(mdt_object_fid(o)),
 528                                        mdt->mdt_max_ea_size);
 529
 530                                 buf->lb_buf = info->mti_big_acl;
 531                                 buf->lb_len = info->mti_big_aclsize;
 532
 533                                 goto again;
 534                         }
 535
 536                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 537                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 538                 }
 539         } else {
 540                 if (buf->lb_buf == info->mti_big_acl)
 541                         info->mti_big_acl_used = 1;
 542
 543                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 544                                      rc, NODEMAP_FS_TO_CLIENT);
 545                 /* if all ACLs mapped out, rc is still >= 0 */
 546                 if (rc < 0) {
 547                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 548                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 549                                PFID(mdt_object_fid(o)), rc);
 550                 } else {
 551                         repbody->mbo_aclsize = rc;
 552                         repbody->mbo_valid |= OBD_MD_FLACL;
 553                         rc = 0;
 554                 }
 555         }
 556
 557         RETURN(rc);
 558 }
 559 #endif
 560
 561 /* XXX Look into layout in MDT layer. */
 562 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 563 {
 564         struct lov_comp_md_v1   *comp_v1;
 565         struct lov_mds_md       *v1;
 566         int                      i;
 567
 568         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 569                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 570
 571                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 572                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 573                                 comp_v1->lcm_entries[i].lcme_offset);
 574                         /* We don't support partial release for now */
 575                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 576                                 return false;
 577                 }
 578                 return true;
 579         } else {
 580                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 581                         true : false;
 582         }
 583 }
 584
 585 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 586                         const struct lu_attr *attr, const struct lu_fid *fid)
 587 {
 588         struct md_attr *ma = &info->mti_attr;
 589         struct obd_export *exp = info->mti_exp;
 590         struct lu_nodemap *nodemap = NULL;
 591
 592         LASSERT(ma->ma_valid & MA_INODE);
 593
 594         if (attr->la_valid & LA_ATIME) {
 595                 b->mbo_atime = attr->la_atime;
 596                 b->mbo_valid |= OBD_MD_FLATIME;
 597         }
 598         if (attr->la_valid & LA_MTIME) {
 599                 b->mbo_mtime = attr->la_mtime;
 600                 b->mbo_valid |= OBD_MD_FLMTIME;
 601         }
 602         if (attr->la_valid & LA_CTIME) {
 603                 b->mbo_ctime = attr->la_ctime;
 604                 b->mbo_valid |= OBD_MD_FLCTIME;
 605         }
 606         if (attr->la_valid & LA_FLAGS) {
 607                 b->mbo_flags = attr->la_flags;
 608                 b->mbo_valid |= OBD_MD_FLFLAGS;
 609         }
 610         if (attr->la_valid & LA_NLINK) {
 611                 b->mbo_nlink = attr->la_nlink;
 612                 b->mbo_valid |= OBD_MD_FLNLINK;
 613         }
 614         if (attr->la_valid & (LA_UID|LA_GID)) {
 615                 nodemap = nodemap_get_from_exp(exp);
 616                 if (IS_ERR(nodemap))
 617                         goto out;
 618         }
 619         if (attr->la_valid & LA_UID) {
 620                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 621                                             NODEMAP_FS_TO_CLIENT,
 622                                             attr->la_uid);
 623                 b->mbo_valid |= OBD_MD_FLUID;
 624         }
 625         if (attr->la_valid & LA_GID) {
 626                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 627                                             NODEMAP_FS_TO_CLIENT,
 628                                             attr->la_gid);
 629                 b->mbo_valid |= OBD_MD_FLGID;
 630         }
 631
 632         if (attr->la_valid & LA_PROJID) {
 633                 /* TODO, nodemap for project id */
 634                 b->mbo_projid = attr->la_projid;
 635                 b->mbo_valid |= OBD_MD_FLPROJID;
 636         }
 637
 638         b->mbo_mode = attr->la_mode;
 639         if (attr->la_valid & LA_MODE)
 640                 b->mbo_valid |= OBD_MD_FLMODE;
 641         if (attr->la_valid & LA_TYPE)
 642                 b->mbo_valid |= OBD_MD_FLTYPE;
 643
 644         if (fid != NULL) {
 645                 b->mbo_fid1 = *fid;
 646                 b->mbo_valid |= OBD_MD_FLID;
 647                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 648                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 649         }
 650
 651         if (!(attr->la_valid & LA_TYPE))
 652                 return;
 653
 654         b->mbo_rdev   = attr->la_rdev;
 655         b->mbo_size   = attr->la_size;
 656         b->mbo_blocks = attr->la_blocks;
 657
 658         if (!S_ISREG(attr->la_mode)) {
 659                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 660         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 661                 /* means no objects are allocated on osts. */
 662                 LASSERT(!(ma->ma_valid & MA_LOV));
 663                 /* just ignore blocks occupied by extend attributes on MDS */
 664                 b->mbo_blocks = 0;
 665                 /* if no object is allocated on osts, the size on mds is valid.
 666                  * b=22272 */
 667                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 668         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 669                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 670                         /* A released file stores its size on MDS. */
 671                         /* But return 1 block for released file, unless tools
 672                          * like tar will consider it fully sparse. (LU-3864)
 673                          */
 674                         if (unlikely(b->mbo_size == 0))
 675                                 b->mbo_blocks = 0;
 676                         else
 677                                 b->mbo_blocks = 1;
 678                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 679                 } else if (lov_pattern(ma->ma_lmm->lmm_pattern) ==
 680                            LOV_PATTERN_MDT) {
 681                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 682                 }
 683         }
 684
 685         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 686                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 687                        PFID(fid), (unsigned long long)b->mbo_size);
 688
 689 out:
 690         if (!IS_ERR_OR_NULL(nodemap))
 691                 nodemap_putref(nodemap);
 692 }
 693
 694 static inline int mdt_body_has_lov(const struct lu_attr *la,
 695                                    const struct mdt_body *body)
 696 {
 697         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 698                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 699 }
 700
 701 void mdt_client_compatibility(struct mdt_thread_info *info)
 702 {
 703         struct mdt_body       *body;
 704         struct ptlrpc_request *req = mdt_info_req(info);
 705         struct obd_export     *exp = req->rq_export;
 706         struct md_attr        *ma = &info->mti_attr;
 707         struct lu_attr        *la = &ma->ma_attr;
 708         ENTRY;
 709
 710         if (exp_connect_layout(exp))
 711                 /* the client can deal with 16-bit lmm_stripe_count */
 712                 RETURN_EXIT;
 713
 714         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 715
 716         if (!mdt_body_has_lov(la, body))
 717                 RETURN_EXIT;
 718
 719         /* now we have a reply with a lov for a client not compatible with the
 720          * layout lock so we have to clean the layout generation number */
 721         if (S_ISREG(la->la_mode))
 722                 ma->ma_lmm->lmm_layout_gen = 0;
 723         EXIT;
 724 }
 725
 726 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 727                                    struct mdt_object *o)
 728 {
 729         const struct lu_env *env = info->mti_env;
 730         int rc, rc2;
 731
 732         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 733                           XATTR_NAME_LOV);
 734
 735         if (rc == -ENODATA)
 736                 rc = 0;
 737
 738         if (rc < 0)
 739                 goto out;
 740
 741         /* Is it a directory? Let's check for the LMV as well */
 742         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 743                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 744                                    XATTR_NAME_LMV);
 745
 746                 if (rc2 == -ENODATA)
 747                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 748                                            &LU_BUF_NULL,
 749                                            XATTR_NAME_DEFAULT_LMV);
 750
 751                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 752                         rc = rc2;
 753         }
 754
 755 out:
 756         return rc;
 757 }
 758
 759 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 760                       const char *name)
 761 {
 762         const struct lu_env *env = info->mti_env;
 763         int rc;
 764         ENTRY;
 765
 766         LASSERT(info->mti_big_lmm_used == 0);
 767         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 768         if (rc < 0)
 769                 RETURN(rc);
 770
 771         /* big_lmm may need to be grown */
 772         if (info->mti_big_lmmsize < rc) {
 773                 int size = size_roundup_power2(rc);
 774
 775                 if (info->mti_big_lmmsize > 0) {
 776                         /* free old buffer */
 777                         LASSERT(info->mti_big_lmm);
 778                         OBD_FREE_LARGE(info->mti_big_lmm,
 779                                        info->mti_big_lmmsize);
 780                         info->mti_big_lmm = NULL;
 781                         info->mti_big_lmmsize = 0;
 782                 }
 783
 784                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 785                 if (info->mti_big_lmm == NULL)
 786                         RETURN(-ENOMEM);
 787                 info->mti_big_lmmsize = size;
 788         }
 789         LASSERT(info->mti_big_lmmsize >= rc);
 790
 791         info->mti_buf.lb_buf = info->mti_big_lmm;
 792         info->mti_buf.lb_len = info->mti_big_lmmsize;
 793         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 794
 795         RETURN(rc);
 796 }
 797
 798 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 799                    struct md_attr *ma, const char *name)
 800 {
 801         struct md_object *next = mdt_object_child(o);
 802         struct lu_buf    *buf = &info->mti_buf;
 803         int rc;
 804
 805         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 806                 buf->lb_buf = ma->ma_lmm;
 807                 buf->lb_len = ma->ma_lmm_size;
 808                 LASSERT(!(ma->ma_valid & MA_LOV));
 809         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 810                 buf->lb_buf = ma->ma_lmv;
 811                 buf->lb_len = ma->ma_lmv_size;
 812                 LASSERT(!(ma->ma_valid & MA_LMV));
 813         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 814                 buf->lb_buf = ma->ma_lmv;
 815                 buf->lb_len = ma->ma_lmv_size;
 816                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 817         } else {
 818                 return -EINVAL;
 819         }
 820
 821         rc = mo_xattr_get(info->mti_env, next, buf, name);
 822         if (rc > 0) {
 823
 824 got:
 825                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 826                         if (info->mti_big_lmm_used)
 827                                 ma->ma_lmm = info->mti_big_lmm;
 828
 829                         /* NOT return LOV EA with hole to old client. */
 830                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 831                                      LOV_PATTERN_F_HOLE) &&
 832                             !(exp_connect_flags(info->mti_exp) &
 833                               OBD_CONNECT_LFSCK)) {
 834                                 return -EIO;
 835                         } else {
 836                                 ma->ma_lmm_size = rc;
 837                                 ma->ma_valid |= MA_LOV;
 838                         }
 839                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 840                         if (info->mti_big_lmm_used)
 841                                 ma->ma_lmv = info->mti_big_lmm;
 842
 843                         ma->ma_lmv_size = rc;
 844                         ma->ma_valid |= MA_LMV;
 845                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 846                         ma->ma_lmv_size = rc;
 847                         ma->ma_valid |= MA_LMV_DEF;
 848                 }
 849
 850                 /* Update mdt_max_mdsize so all clients will be aware that */
 851                 if (info->mti_mdt->mdt_max_mdsize < rc)
 852                         info->mti_mdt->mdt_max_mdsize = rc;
 853
 854                 rc = 0;
 855         } else if (rc == -ENODATA) {
 856                 /* no LOV EA */
 857                 rc = 0;
 858         } else if (rc == -ERANGE) {
 859                 /* Default LMV has fixed size, so it must be able to fit
 860                  * in the original buffer */
 861                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 862                         return rc;
 863                 rc = mdt_big_xattr_get(info, o, name);
 864                 if (rc > 0) {
 865                         info->mti_big_lmm_used = 1;
 866                         goto got;
 867                 }
 868         }
 869
 870         return rc;
 871 }
 872
 873 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 874                              struct mdt_object *o, struct lu_fid *pfid)
 875 {
 876         struct lu_buf           *buf = &info->mti_buf;
 877         struct link_ea_header   *leh;
 878         struct link_ea_entry    *lee;
 879         int                      rc;
 880         ENTRY;
 881
 882         buf->lb_buf = info->mti_big_lmm;
 883         buf->lb_len = info->mti_big_lmmsize;
 884         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 885                           buf, XATTR_NAME_LINK);
 886         /* ignore errors, MA_PFID won't be set and it is
 887          * up to the caller to treat this as an error */
 888         if (rc == -ERANGE || buf->lb_len == 0) {
 889                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 890                 buf->lb_buf = info->mti_big_lmm;
 891                 buf->lb_len = info->mti_big_lmmsize;
 892         }
 893
 894         if (rc < 0)
 895                 RETURN(rc);
 896         if (rc < sizeof(*leh)) {
 897                 CERROR("short LinkEA on "DFID": rc = %d\n",
 898                        PFID(mdt_object_fid(o)), rc);
 899                 RETURN(-ENODATA);
 900         }
 901
 902         leh = (struct link_ea_header *) buf->lb_buf;
 903         lee = (struct link_ea_entry *)(leh + 1);
 904         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 905                 leh->leh_magic = LINK_EA_MAGIC;
 906                 leh->leh_reccount = __swab32(leh->leh_reccount);
 907                 leh->leh_len = __swab64(leh->leh_len);
 908         }
 909         if (leh->leh_magic != LINK_EA_MAGIC)
 910                 RETURN(-EINVAL);
 911         if (leh->leh_reccount == 0)
 912                 RETURN(-ENODATA);
 913
 914         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 915         fid_be_to_cpu(pfid, pfid);
 916
 917         RETURN(0);
 918 }
 919
 920 int mdt_attr_get_complex(struct mdt_thread_info *info,
 921                          struct mdt_object *o, struct md_attr *ma)
 922 {
 923         const struct lu_env *env = info->mti_env;
 924         struct md_object    *next = mdt_object_child(o);
 925         struct lu_buf       *buf = &info->mti_buf;
 926         int                  need = ma->ma_need;
 927         int                  rc = 0, rc2;
 928         u32                  mode;
 929         ENTRY;
 930
 931         ma->ma_valid = 0;
 932
 933         if (mdt_object_exists(o) == 0)
 934                 GOTO(out, rc = -ENOENT);
 935         mode = lu_object_attr(&next->mo_lu);
 936
 937         if (need & MA_INODE) {
 938                 ma->ma_need = MA_INODE;
 939                 rc = mo_attr_get(env, next, ma);
 940                 if (rc)
 941                         GOTO(out, rc);
 942                 ma->ma_valid |= MA_INODE;
 943         }
 944
 945         if (need & MA_PFID) {
 946                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 947                 if (rc == 0)
 948                         ma->ma_valid |= MA_PFID;
 949                 /* ignore this error, parent fid is not mandatory */
 950                 rc = 0;
 951         }
 952
 953         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 954                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 955                 if (rc)
 956                         GOTO(out, rc);
 957         }
 958
 959         if (need & MA_LMV && S_ISDIR(mode)) {
 960                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 961                 if (rc != 0)
 962                         GOTO(out, rc);
 963         }
 964
 965         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 966                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 967                 if (rc != 0)
 968                         GOTO(out, rc);
 969         }
 970
 971         if (need & MA_HSM && S_ISREG(mode)) {
 972                 buf->lb_buf = info->mti_xattr_buf;
 973                 buf->lb_len = sizeof(info->mti_xattr_buf);
 974                 CLASSERT(sizeof(struct hsm_attrs) <=
 975                          sizeof(info->mti_xattr_buf));
 976                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 977                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 978                 if (rc2 == 0)
 979                         ma->ma_valid |= MA_HSM;
 980                 else if (rc2 < 0 && rc2 != -ENODATA)
 981                         GOTO(out, rc = rc2);
 982         }
 983
 984 #ifdef CONFIG_FS_POSIX_ACL
 985         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 986                 buf->lb_buf = ma->ma_acl;
 987                 buf->lb_len = ma->ma_acl_size;
 988                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 989                 if (rc2 > 0) {
 990                         ma->ma_acl_size = rc2;
 991                         ma->ma_valid |= MA_ACL_DEF;
 992                 } else if (rc2 == -ENODATA) {
 993                         /* no ACLs */
 994                         ma->ma_acl_size = 0;
 995                 } else
 996                         GOTO(out, rc = rc2);
 997         }
 998 #endif
 999 out:
1000         ma->ma_need = need;
1001         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1002                rc, ma->ma_valid, ma->ma_lmm);
1003         RETURN(rc);
1004 }
1005
1006 static int mdt_getattr_internal(struct mdt_thread_info *info,
1007                                 struct mdt_object *o, int ma_need)
1008 {
1009         struct md_object        *next = mdt_object_child(o);
1010         const struct mdt_body   *reqbody = info->mti_body;
1011         struct ptlrpc_request   *req = mdt_info_req(info);
1012         struct md_attr          *ma = &info->mti_attr;
1013         struct lu_attr          *la = &ma->ma_attr;
1014         struct req_capsule      *pill = info->mti_pill;
1015         const struct lu_env     *env = info->mti_env;
1016         struct mdt_body         *repbody;
1017         struct lu_buf           *buffer = &info->mti_buf;
1018         struct obd_export       *exp = info->mti_exp;
1019         int                      rc;
1020         ENTRY;
1021
1022         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1023                 RETURN(err_serious(-ENOMEM));
1024
1025         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1026
1027         ma->ma_valid = 0;
1028
1029         if (mdt_object_remote(o)) {
1030                 /* This object is located on remote node.*/
1031                 /* Return -ENOTSUPP for old client */
1032                 if (!mdt_is_dne_client(req->rq_export))
1033                         GOTO(out, rc = -ENOTSUPP);
1034
1035                 repbody->mbo_fid1 = *mdt_object_fid(o);
1036                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1037                 GOTO(out, rc = 0);
1038         }
1039
1040         if (reqbody->mbo_eadatasize > 0) {
1041                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1042                 if (buffer->lb_buf == NULL)
1043                         GOTO(out, rc = -EPROTO);
1044                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1045                                                       RCL_SERVER);
1046         } else {
1047                 buffer->lb_buf = NULL;
1048                 buffer->lb_len = 0;
1049                 ma_need &= ~(MA_LOV | MA_LMV);
1050                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1051                        mdt_obd_name(info->mti_mdt),
1052                        req->rq_export->exp_client_uuid.uuid);
1053         }
1054
1055         /* If it is dir object and client require MEA, then we got MEA */
1056         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1057             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1058                 /* Assumption: MDT_MD size is enough for lmv size. */
1059                 ma->ma_lmv = buffer->lb_buf;
1060                 ma->ma_lmv_size = buffer->lb_len;
1061                 ma->ma_need = MA_INODE;
1062                 if (ma->ma_lmv_size > 0) {
1063                         if (reqbody->mbo_valid & OBD_MD_MEA)
1064                                 ma->ma_need |= MA_LMV;
1065                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1066                                 ma->ma_need |= MA_LMV_DEF;
1067                 }
1068         } else {
1069                 ma->ma_lmm = buffer->lb_buf;
1070                 ma->ma_lmm_size = buffer->lb_len;
1071                 ma->ma_need = MA_INODE | MA_HSM;
1072                 if (ma->ma_lmm_size > 0)
1073                         ma->ma_need |= MA_LOV;
1074         }
1075
1076         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1077             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1078             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1079                 /* get default stripe info for this dir. */
1080                 ma->ma_need |= MA_LOV_DEF;
1081         }
1082         ma->ma_need |= ma_need;
1083
1084         rc = mdt_attr_get_complex(info, o, ma);
1085         if (unlikely(rc)) {
1086                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1087                        "%s: getattr error for "DFID": rc = %d\n",
1088                        mdt_obd_name(info->mti_mdt),
1089                        PFID(mdt_object_fid(o)), rc);
1090                 RETURN(rc);
1091         }
1092
1093         /* if file is released, check if a restore is running */
1094         if (ma->ma_valid & MA_HSM) {
1095                 repbody->mbo_valid |= OBD_MD_TSTATE;
1096                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1097                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1098                         repbody->mbo_t_state = MS_RESTORE;
1099         }
1100
1101         if (likely(ma->ma_valid & MA_INODE))
1102                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1103         else
1104                 RETURN(-EFAULT);
1105
1106         if (mdt_body_has_lov(la, reqbody)) {
1107                 if (ma->ma_valid & MA_LOV) {
1108                         LASSERT(ma->ma_lmm_size);
1109                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1110                         if (S_ISDIR(la->la_mode))
1111                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1112                         else
1113                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1114                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1115                 }
1116                 if (ma->ma_valid & MA_LMV) {
1117                         /* Return -ENOTSUPP for old client */
1118                         if (!mdt_is_striped_client(req->rq_export))
1119                                 RETURN(-ENOTSUPP);
1120
1121                         LASSERT(S_ISDIR(la->la_mode));
1122                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1123                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1124                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1125                 }
1126                 if (ma->ma_valid & MA_LMV_DEF) {
1127                         /* Return -ENOTSUPP for old client */
1128                         if (!mdt_is_striped_client(req->rq_export))
1129                                 RETURN(-ENOTSUPP);
1130                         LASSERT(S_ISDIR(la->la_mode));
1131                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1132                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1133                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1134                                                OBD_MD_DEFAULT_MEA);
1135                 }
1136         } else if (S_ISLNK(la->la_mode) &&
1137                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1138                 buffer->lb_buf = ma->ma_lmm;
1139                 /* eadatasize from client includes NULL-terminator, so
1140                  * there is no need to read it */
1141                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1142                 rc = mo_readlink(env, next, buffer);
1143                 if (unlikely(rc <= 0)) {
1144                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1145                                mdt_obd_name(info->mti_mdt),
1146                                PFID(mdt_object_fid(o)), rc);
1147                         rc = -EFAULT;
1148                 } else {
1149                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1150
1151                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1152                                 rc -= 2;
1153                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1154                         /* we need to report back size with NULL-terminator
1155                          * because client expects that */
1156                         repbody->mbo_eadatasize = rc + 1;
1157                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1158                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1159                                        "on "DFID ", expected %d\n",
1160                                        mdt_obd_name(info->mti_mdt),
1161                                        rc, PFID(mdt_object_fid(o)),
1162                                        reqbody->mbo_eadatasize - 1);
1163                         /* NULL terminate */
1164                         ((char *)ma->ma_lmm)[rc] = 0;
1165
1166                         /* If the total CDEBUG() size is larger than a page, it
1167                          * will print a warning to the console, avoid this by
1168                          * printing just the last part of the symlink. */
1169                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1170                                print_limit < rc ? "..." : "", print_limit,
1171                                (char *)ma->ma_lmm + rc - print_limit, rc);
1172                         rc = 0;
1173                 }
1174         }
1175
1176         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1177                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1178                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1179                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1180                        repbody->mbo_max_mdsize);
1181         }
1182
1183 #ifdef CONFIG_FS_POSIX_ACL
1184         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1185                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1186                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1187                 if (IS_ERR(nodemap))
1188                         RETURN(PTR_ERR(nodemap));
1189
1190                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1191                 nodemap_putref(nodemap);
1192         }
1193 #endif
1194
1195 out:
1196         if (rc == 0)
1197                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1198
1199         RETURN(rc);
1200 }
1201
1202 static int mdt_getattr(struct tgt_session_info *tsi)
1203 {
1204         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1205         struct mdt_object       *obj = info->mti_object;
1206         struct req_capsule      *pill = info->mti_pill;
1207         struct mdt_body         *reqbody;
1208         struct mdt_body         *repbody;
1209         int rc, rc2;
1210         ENTRY;
1211
1212         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1213         LASSERT(reqbody);
1214         LASSERT(obj != NULL);
1215         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1216
1217         /* Unlike intent case where we need to pre-fill out buffers early on
1218          * in intent policy for ldlm reasons, here we can have a much better
1219          * guess at EA size by just reading it from disk.
1220          * Exceptions are readdir and (missing) directory striping */
1221         /* Readlink */
1222         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1223                 /* No easy way to know how long is the symlink, but it cannot
1224                  * be more than PATH_MAX, so we allocate +1 */
1225                 rc = PATH_MAX + 1;
1226
1227         /* A special case for fs ROOT: getattr there might fetch
1228          * default EA for entire fs, not just for this dir!
1229          */
1230         } else if (lu_fid_eq(mdt_object_fid(obj),
1231                              &info->mti_mdt->mdt_md_root_fid) &&
1232                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1233                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1234                                                                  MDS_GETATTR)) {
1235                 /* Should the default strping be bigger, mdt_fix_reply
1236                  * will reallocate */
1237                 rc = DEF_REP_MD_SIZE;
1238         } else {
1239                 /* Read the actual EA size from disk */
1240                 rc = mdt_attr_get_eabuf_size(info, obj);
1241         }
1242
1243         if (rc < 0)
1244                 GOTO(out, rc = err_serious(rc));
1245
1246         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1247
1248         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1249          * by default. If the target object has more ACL entries, then
1250          * enlarge the buffer when necessary. */
1251         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1252                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1253
1254         rc = req_capsule_server_pack(pill);
1255         if (unlikely(rc != 0))
1256                 GOTO(out, rc = err_serious(rc));
1257
1258         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1259         LASSERT(repbody != NULL);
1260         repbody->mbo_eadatasize = 0;
1261         repbody->mbo_aclsize = 0;
1262
1263         rc = mdt_check_ucred(info);
1264         if (unlikely(rc))
1265                 GOTO(out_shrink, rc);
1266
1267         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1268
1269         rc = mdt_getattr_internal(info, obj, 0);
1270         EXIT;
1271 out_shrink:
1272         mdt_client_compatibility(info);
1273         rc2 = mdt_fix_reply(info);
1274         if (rc == 0)
1275                 rc = rc2;
1276 out:
1277         mdt_thread_info_fini(info);
1278         return rc;
1279 }
1280
1281 /**
1282  * Handler of layout intent RPC requiring the layout modification
1283  *
1284  * \param[in] info      thread environment
1285  * \param[in] obj       object
1286  * \param[in] layout    layout intent
1287  * \param[in] buf       buffer containing client's lovea, could be empty
1288  *
1289  * \retval 0    on success
1290  * \retval < 0  error code
1291  */
1292 static int mdt_layout_change(struct mdt_thread_info *info,
1293                              struct mdt_object *obj,
1294                              struct layout_intent *layout,
1295                              const struct lu_buf *buf)
1296 {
1297         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1298         int rc;
1299         ENTRY;
1300
1301         CDEBUG(D_INFO, "got layout change request from client: "
1302                "opc:%u flags:%#x extent[%#llx,%#llx)\n",
1303                layout->li_opc, layout->li_flags,
1304                layout->li_start, layout->li_end);
1305         if (layout->li_start >= layout->li_end) {
1306                 CERROR("Recieved an invalid layout change range [%llu, %llu) "
1307                        "for "DFID"\n", layout->li_start, layout->li_end,
1308                        PFID(mdt_object_fid(obj)));
1309                 RETURN(-EINVAL);
1310         }
1311
1312         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1313                 GOTO(out, rc = -EINVAL);
1314
1315         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1316                            MAY_WRITE);
1317         if (rc)
1318                 GOTO(out, rc);
1319
1320         /* take layout lock to prepare layout change */
1321         mdt_lock_reg_init(lh, LCK_EX);
1322         rc = mdt_object_lock(info, obj, lh,
1323                              MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1324         if (rc)
1325                 GOTO(out, rc);
1326
1327         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout,
1328                               buf);
1329
1330         mdt_object_unlock(info, obj, lh, 1);
1331 out:
1332         RETURN(rc);
1333 }
1334
1335 /**
1336  * Exchange MOF_LOV_CREATED flags between two objects after a
1337  * layout swap. No assumption is made on whether o1 or o2 have
1338  * created objects or not.
1339  *
1340  * \param[in,out] o1    First swap layout object
1341  * \param[in,out] o2    Second swap layout object
1342  */
1343 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1344 {
1345         unsigned int o1_lov_created = o1->mot_lov_created;
1346
1347         mutex_lock(&o1->mot_lov_mutex);
1348         mutex_lock(&o2->mot_lov_mutex);
1349
1350         o1->mot_lov_created = o2->mot_lov_created;
1351         o2->mot_lov_created = o1_lov_created;
1352
1353         mutex_unlock(&o2->mot_lov_mutex);
1354         mutex_unlock(&o1->mot_lov_mutex);
1355 }
1356
1357 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1358 {
1359         struct mdt_thread_info  *info;
1360         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1361         struct obd_export       *exp = req->rq_export;
1362         struct mdt_object       *o1, *o2, *o;
1363         struct mdt_lock_handle  *lh1, *lh2;
1364         struct mdc_swap_layouts *msl;
1365         int                      rc;
1366         ENTRY;
1367
1368         /* client does not support layout lock, so layout swaping
1369          * is disabled.
1370          * FIXME: there is a problem for old clients which don't support
1371          * layout lock yet. If those clients have already opened the file
1372          * they won't be notified at all so that old layout may still be
1373          * used to do IO. This can be fixed after file release is landed by
1374          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1375         if (!exp_connect_layout(exp))
1376                 RETURN(-EOPNOTSUPP);
1377
1378         info = tsi2mdt_info(tsi);
1379
1380         if (info->mti_dlm_req != NULL)
1381                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1382
1383         o1 = info->mti_object;
1384         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1385                                 &info->mti_body->mbo_fid2);
1386         if (IS_ERR(o))
1387                 GOTO(out, rc = PTR_ERR(o));
1388
1389         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1390                 GOTO(put, rc = -ENOENT);
1391
1392         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1393         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1394                 GOTO(put, rc);
1395
1396         if (rc < 0)
1397                 swap(o1, o2);
1398
1399         /* permission check. Make sure the calling process having permission
1400          * to write both files. */
1401         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1402                                 MAY_WRITE);
1403         if (rc < 0)
1404                 GOTO(put, rc);
1405
1406         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1407                                 MAY_WRITE);
1408         if (rc < 0)
1409                 GOTO(put, rc);
1410
1411         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1412         if (msl == NULL)
1413                 GOTO(put, rc = -EPROTO);
1414
1415         lh1 = &info->mti_lh[MDT_LH_NEW];
1416         mdt_lock_reg_init(lh1, LCK_EX);
1417         lh2 = &info->mti_lh[MDT_LH_OLD];
1418         mdt_lock_reg_init(lh2, LCK_EX);
1419
1420         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1421                              MDS_INODELOCK_XATTR);
1422         if (rc < 0)
1423                 GOTO(put, rc);
1424
1425         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1426                              MDS_INODELOCK_XATTR);
1427         if (rc < 0)
1428                 GOTO(unlock1, rc);
1429
1430         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1431                              mdt_object_child(o2), msl->msl_flags);
1432         if (rc < 0)
1433                 GOTO(unlock2, rc);
1434
1435         mdt_swap_lov_flag(o1, o2);
1436
1437 unlock2:
1438         mdt_object_unlock(info, o2, lh2, rc);
1439 unlock1:
1440         mdt_object_unlock(info, o1, lh1, rc);
1441 put:
1442         mdt_object_put(info->mti_env, o);
1443 out:
1444         mdt_thread_info_fini(info);
1445         RETURN(rc);
1446 }
1447
1448 static int mdt_raw_lookup(struct mdt_thread_info *info,
1449                           struct mdt_object *parent,
1450                           const struct lu_name *lname,
1451                           struct ldlm_reply *ldlm_rep)
1452 {
1453         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1454         int              rc;
1455         ENTRY;
1456
1457         LASSERT(!info->mti_cross_ref);
1458
1459         /* Only got the fid of this obj by name */
1460         fid_zero(child_fid);
1461         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1462                         lname, child_fid, &info->mti_spec);
1463         if (rc == 0) {
1464                 struct mdt_body *repbody;
1465
1466                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1467                 repbody->mbo_fid1 = *child_fid;
1468                 repbody->mbo_valid = OBD_MD_FLID;
1469                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1470         } else if (rc == -ENOENT) {
1471                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1472         }
1473
1474         RETURN(rc);
1475 }
1476
1477 /*
1478  * UPDATE lock should be taken against parent, and be released before exit;
1479  * child_bits lock should be taken against child, and be returned back:
1480  *            (1)normal request should release the child lock;
1481  *            (2)intent request will grant the lock to client.
1482  */
1483 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1484                                  struct mdt_lock_handle *lhc,
1485                                  __u64 child_bits,
1486                                  struct ldlm_reply *ldlm_rep)
1487 {
1488         struct ptlrpc_request  *req = mdt_info_req(info);
1489         struct mdt_body        *reqbody = NULL;
1490         struct mdt_object      *parent = info->mti_object;
1491         struct mdt_object      *child;
1492         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1493         struct lu_name         *lname = NULL;
1494         struct mdt_lock_handle *lhp = NULL;
1495         struct ldlm_lock       *lock;
1496         __u64 try_bits = 0;
1497         bool is_resent;
1498         int ma_need = 0;
1499         int rc;
1500
1501         ENTRY;
1502
1503         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1504         LASSERT(ergo(is_resent,
1505                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1506
1507         if (parent == NULL)
1508                 RETURN(-ENOENT);
1509
1510         if (info->mti_cross_ref) {
1511                 /* Only getattr on the child. Parent is on another node. */
1512                 mdt_set_disposition(info, ldlm_rep,
1513                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1514                 child = parent;
1515                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1516                        "ldlm_rep = %p\n",
1517                        PFID(mdt_object_fid(child)), ldlm_rep);
1518
1519                 rc = mdt_check_resent_lock(info, child, lhc);
1520                 if (rc < 0) {
1521                         RETURN(rc);
1522                 } else if (rc > 0) {
1523                         mdt_lock_handle_init(lhc);
1524                         mdt_lock_reg_init(lhc, LCK_PR);
1525
1526                         /*
1527                          * Object's name is on another MDS, no lookup or layout
1528                          * lock is needed here but update lock is.
1529                          */
1530                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1531                                         MDS_INODELOCK_LAYOUT);
1532                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1533
1534                         rc = mdt_object_lock(info, child, lhc, child_bits);
1535                         if (rc < 0)
1536                                 RETURN(rc);
1537                 }
1538
1539                 /* Finally, we can get attr for child. */
1540                 if (!mdt_object_exists(child)) {
1541                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1542                                         &child->mot_obj,
1543                                         "remote object doesn't exist.");
1544                         mdt_object_unlock(info, child, lhc, 1);
1545                         RETURN(-ENOENT);
1546                 }
1547
1548                 rc = mdt_getattr_internal(info, child, 0);
1549                 if (unlikely(rc != 0))
1550                         mdt_object_unlock(info, child, lhc, 1);
1551
1552                 RETURN(rc);
1553         }
1554
1555         lname = &info->mti_name;
1556         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1557
1558         if (lu_name_is_valid(lname)) {
1559                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1560                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1561                        PNAME(lname), ldlm_rep);
1562         } else {
1563                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1564                 if (unlikely(reqbody == NULL))
1565                         RETURN(err_serious(-EPROTO));
1566
1567                 *child_fid = reqbody->mbo_fid2;
1568
1569                 if (unlikely(!fid_is_sane(child_fid)))
1570                         RETURN(err_serious(-EINVAL));
1571
1572                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1573                        "ldlm_rep = %p\n",
1574                        PFID(mdt_object_fid(parent)),
1575                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1576         }
1577
1578         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1579
1580         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1581                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1582                                 &parent->mot_obj,
1583                                 "Parent doesn't exist!");
1584                 RETURN(-ESTALE);
1585         }
1586
1587         if (mdt_object_remote(parent)) {
1588                 CERROR("%s: parent "DFID" is on remote target\n",
1589                        mdt_obd_name(info->mti_mdt),
1590                        PFID(mdt_object_fid(parent)));
1591                 RETURN(-EIO);
1592         }
1593
1594         if (lu_name_is_valid(lname)) {
1595                 /* Always allow to lookup ".." */
1596                 if (unlikely(lname->ln_namelen == 2 &&
1597                              lname->ln_name[0] == '.' &&
1598                              lname->ln_name[1] == '.'))
1599                         info->mti_spec.sp_permitted = 1;
1600
1601                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1602                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1603
1604                         RETURN(rc);
1605                 }
1606
1607                 /* step 1: lock parent only if parent is a directory */
1608                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1609                         lhp = &info->mti_lh[MDT_LH_PARENT];
1610                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1611                         rc = mdt_object_lock(info, parent, lhp,
1612                                              MDS_INODELOCK_UPDATE);
1613                         if (unlikely(rc != 0))
1614                                 RETURN(rc);
1615                 }
1616
1617                 /* step 2: lookup child's fid by name */
1618                 fid_zero(child_fid);
1619                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1620                                 child_fid, &info->mti_spec);
1621                 if (rc == -ENOENT)
1622                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1623
1624                 if (rc != 0)
1625                         GOTO(out_parent, rc);
1626         }
1627
1628         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1629
1630         /*
1631          *step 3: find the child object by fid & lock it.
1632          *        regardless if it is local or remote.
1633          *
1634          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1635          *      set parent dir fid the same as child fid in getattr by fid case
1636          *      we should not lu_object_find() the object again, could lead
1637          *      to hung if there is a concurrent unlink destroyed the object.
1638          */
1639         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1640                 mdt_object_get(info->mti_env, parent);
1641                 child = parent;
1642         } else {
1643                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1644                                         child_fid);
1645         }
1646
1647         if (unlikely(IS_ERR(child)))
1648                 GOTO(out_parent, rc = PTR_ERR(child));
1649
1650         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1651         if (!mdt_object_exists(child)) {
1652                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1653                                 &child->mot_obj,
1654                                 "Object doesn't exist!");
1655                 GOTO(out_child, rc = -ENOENT);
1656         }
1657
1658         rc = mdt_check_resent_lock(info, child, lhc);
1659         if (rc < 0) {
1660                 GOTO(out_child, rc);
1661         } else if (rc > 0) {
1662                 mdt_lock_handle_init(lhc);
1663                 mdt_lock_reg_init(lhc, LCK_PR);
1664
1665                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1666                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1667                         struct md_attr *ma = &info->mti_attr;
1668
1669                         ma->ma_valid = 0;
1670                         ma->ma_need = MA_INODE;
1671                         rc = mdt_attr_get_complex(info, child, ma);
1672                         if (unlikely(rc != 0))
1673                                 GOTO(out_child, rc);
1674
1675                         /* If the file has not been changed for some time, we
1676                          * return not only a LOOKUP lock, but also an UPDATE
1677                          * lock and this might save us RPC on later STAT. For
1678                          * directories, it also let negative dentry cache start
1679                          * working for this dir. */
1680                         if (ma->ma_valid & MA_INODE &&
1681                             ma->ma_attr.la_valid & LA_CTIME &&
1682                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1683                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1684                                 child_bits |= MDS_INODELOCK_UPDATE;
1685                 }
1686
1687                 /* layout lock must be granted in a best-effort way
1688                  * for IT operations */
1689                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1690                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1691                     exp_connect_layout(info->mti_exp) &&
1692                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1693                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1694                         /* try to grant layout lock for regular file. */
1695                         try_bits = MDS_INODELOCK_LAYOUT;
1696                 }
1697
1698                 if (try_bits != 0) {
1699                         /* try layout lock, it may fail to be granted due to
1700                          * contention at LOOKUP or UPDATE */
1701                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1702                                                  try_bits, false);
1703                         if (child_bits & MDS_INODELOCK_LAYOUT)
1704                                 ma_need |= MA_LOV;
1705                 } else {
1706                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1707                          * client will enqueue the lock to the remote MDT */
1708                         if (mdt_object_remote(child))
1709                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1710                         rc = mdt_object_lock(info, child, lhc, child_bits);
1711                 }
1712                 if (unlikely(rc != 0))
1713                         GOTO(out_child, rc);
1714         }
1715
1716         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1717
1718         /* finally, we can get attr for child. */
1719         rc = mdt_getattr_internal(info, child, ma_need);
1720         if (unlikely(rc != 0)) {
1721                 mdt_object_unlock(info, child, lhc, 1);
1722         } else if (lock) {
1723                 /* Debugging code. */
1724                 LDLM_DEBUG(lock, "Returning lock to client");
1725                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1726                                          &lock->l_resource->lr_name),
1727                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1728                          PLDLMRES(lock->l_resource),
1729                          PFID(mdt_object_fid(child)));
1730         }
1731         if (lock)
1732                 LDLM_LOCK_PUT(lock);
1733
1734         EXIT;
1735 out_child:
1736         mdt_object_put(info->mti_env, child);
1737 out_parent:
1738         if (lhp)
1739                 mdt_object_unlock(info, parent, lhp, 1);
1740         return rc;
1741 }
1742
1743 /* normal handler: should release the child lock */
1744 static int mdt_getattr_name(struct tgt_session_info *tsi)
1745 {
1746         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1747         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1748         struct mdt_body        *reqbody;
1749         struct mdt_body        *repbody;
1750         int rc, rc2;
1751         ENTRY;
1752
1753         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1754         LASSERT(reqbody != NULL);
1755         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1756         LASSERT(repbody != NULL);
1757
1758         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1759         repbody->mbo_eadatasize = 0;
1760         repbody->mbo_aclsize = 0;
1761
1762         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1763         if (unlikely(rc))
1764                 GOTO(out_shrink, rc);
1765
1766         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1767         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1768                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1769                 lhc->mlh_reg_lh.cookie = 0;
1770         }
1771         mdt_exit_ucred(info);
1772         EXIT;
1773 out_shrink:
1774         mdt_client_compatibility(info);
1775         rc2 = mdt_fix_reply(info);
1776         if (rc == 0)
1777                 rc = rc2;
1778         mdt_thread_info_fini(info);
1779         return rc;
1780 }
1781
1782 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1783                          void *karg, void __user *uarg);
1784
1785 static int mdt_set_info(struct tgt_session_info *tsi)
1786 {
1787         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1788         char                    *key;
1789         void                    *val;
1790         int                      keylen, vallen, rc = 0;
1791
1792         ENTRY;
1793
1794         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1795         if (key == NULL) {
1796                 DEBUG_REQ(D_HA, req, "no set_info key");
1797                 RETURN(err_serious(-EFAULT));
1798         }
1799
1800         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1801                                       RCL_CLIENT);
1802
1803         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1804         if (val == NULL) {
1805                 DEBUG_REQ(D_HA, req, "no set_info val");
1806                 RETURN(err_serious(-EFAULT));
1807         }
1808
1809         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1810                                       RCL_CLIENT);
1811
1812         /* Swab any part of val you need to here */
1813         if (KEY_IS(KEY_READ_ONLY)) {
1814                 spin_lock(&req->rq_export->exp_lock);
1815                 if (*(__u32 *)val)
1816                         *exp_connect_flags_ptr(req->rq_export) |=
1817                                 OBD_CONNECT_RDONLY;
1818                 else
1819                         *exp_connect_flags_ptr(req->rq_export) &=
1820                                 ~OBD_CONNECT_RDONLY;
1821                 spin_unlock(&req->rq_export->exp_lock);
1822         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1823                 struct changelog_setinfo *cs = val;
1824
1825                 if (vallen != sizeof(*cs)) {
1826                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1827                                tgt_name(tsi->tsi_tgt), vallen);
1828                         RETURN(-EINVAL);
1829                 }
1830                 if (ptlrpc_req_need_swab(req)) {
1831                         __swab64s(&cs->cs_recno);
1832                         __swab32s(&cs->cs_id);
1833                 }
1834
1835                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1836                                    vallen, val, NULL);
1837         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1838                 if (vallen > 0)
1839                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1840         } else {
1841                 RETURN(-EINVAL);
1842         }
1843         RETURN(rc);
1844 }
1845
1846 static int mdt_readpage(struct tgt_session_info *tsi)
1847 {
1848         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1849         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1850         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1851         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1852         struct mdt_body         *repbody;
1853         int                      rc;
1854         int                      i;
1855
1856         ENTRY;
1857
1858         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1859                 RETURN(err_serious(-ENOMEM));
1860
1861         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1862         if (repbody == NULL || reqbody == NULL)
1863                 RETURN(err_serious(-EFAULT));
1864
1865         /*
1866          * prepare @rdpg before calling lower layers and transfer itself. Here
1867          * reqbody->size contains offset of where to start to read and
1868          * reqbody->nlink contains number bytes to read.
1869          */
1870         rdpg->rp_hash = reqbody->mbo_size;
1871         if (rdpg->rp_hash != reqbody->mbo_size) {
1872                 CERROR("Invalid hash: %#llx != %#llx\n",
1873                        rdpg->rp_hash, reqbody->mbo_size);
1874                 RETURN(-EFAULT);
1875         }
1876
1877         rdpg->rp_attrs = reqbody->mbo_mode;
1878         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1879                 rdpg->rp_attrs |= LUDA_64BITHASH;
1880         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1881                                 exp_max_brw_size(tsi->tsi_exp));
1882         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1883                           PAGE_SHIFT;
1884         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1885         if (rdpg->rp_pages == NULL)
1886                 RETURN(-ENOMEM);
1887
1888         for (i = 0; i < rdpg->rp_npages; ++i) {
1889                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1890                 if (rdpg->rp_pages[i] == NULL)
1891                         GOTO(free_rdpg, rc = -ENOMEM);
1892         }
1893
1894         /* call lower layers to fill allocated pages with directory data */
1895         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1896         if (rc < 0)
1897                 GOTO(free_rdpg, rc);
1898
1899         /* send pages to client */
1900         rc = tgt_sendpage(tsi, rdpg, rc);
1901
1902         EXIT;
1903 free_rdpg:
1904
1905         for (i = 0; i < rdpg->rp_npages; i++)
1906                 if (rdpg->rp_pages[i] != NULL)
1907                         __free_page(rdpg->rp_pages[i]);
1908         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1909
1910         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1911                 RETURN(0);
1912
1913         return rc;
1914 }
1915
1916 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1917 {
1918         struct lu_ucred *uc = mdt_ucred_check(info);
1919         struct lu_attr *attr = &info->mti_attr.ma_attr;
1920
1921         if (uc == NULL)
1922                 return -EINVAL;
1923
1924         if (op != REINT_SETATTR) {
1925                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1926                         attr->la_uid = uc->uc_fsuid;
1927                 /* for S_ISGID, inherit gid from his parent, such work will be
1928                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1929                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1930                         attr->la_gid = uc->uc_fsgid;
1931         }
1932
1933         return 0;
1934 }
1935
1936 static int mdt_reint_internal(struct mdt_thread_info *info,
1937                               struct mdt_lock_handle *lhc,
1938                               __u32 op)
1939 {
1940         struct req_capsule      *pill = info->mti_pill;
1941         struct mdt_body         *repbody;
1942         int                      rc = 0, rc2;
1943
1944         ENTRY;
1945
1946         rc = mdt_reint_unpack(info, op);
1947         if (rc != 0) {
1948                 CERROR("Can't unpack reint, rc %d\n", rc);
1949                 RETURN(err_serious(rc));
1950         }
1951
1952         /* for replay (no_create) lmm is not needed, client has it already */
1953         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1954                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1955                                      DEF_REP_MD_SIZE);
1956
1957         /* llog cookies are always 0, the field is kept for compatibility */
1958         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1959                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1960
1961         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1962          * by default. If the target object has more ACL entries, then
1963          * enlarge the buffer when necessary. */
1964         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
1965                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1966                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1967
1968         rc = req_capsule_server_pack(pill);
1969         if (rc != 0) {
1970                 CERROR("Can't pack response, rc %d\n", rc);
1971                 RETURN(err_serious(rc));
1972         }
1973
1974         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1975                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1976                 LASSERT(repbody);
1977                 repbody->mbo_eadatasize = 0;
1978                 repbody->mbo_aclsize = 0;
1979         }
1980
1981         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1982
1983         /* for replay no cookkie / lmm need, because client have this already */
1984         if (info->mti_spec.no_create)
1985                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1986                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1987
1988         rc = mdt_init_ucred_reint(info);
1989         if (rc)
1990                 GOTO(out_shrink, rc);
1991
1992         rc = mdt_fix_attr_ucred(info, op);
1993         if (rc != 0)
1994                 GOTO(out_ucred, rc = err_serious(rc));
1995
1996         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1997         if (rc < 0) {
1998                 GOTO(out_ucred, rc);
1999         } else if (rc == 1) {
2000                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2001                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2002                 GOTO(out_ucred, rc);
2003         }
2004         rc = mdt_reint_rec(info, lhc);
2005         EXIT;
2006 out_ucred:
2007         mdt_exit_ucred(info);
2008 out_shrink:
2009         mdt_client_compatibility(info);
2010         rc2 = mdt_fix_reply(info);
2011         if (rc == 0)
2012                 rc = rc2;
2013         return rc;
2014 }
2015
2016 static long mdt_reint_opcode(struct ptlrpc_request *req,
2017                              const struct req_format **fmt)
2018 {
2019         struct mdt_device       *mdt;
2020         struct mdt_rec_reint    *rec;
2021         long                     opc;
2022
2023         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2024         if (rec != NULL) {
2025                 opc = rec->rr_opcode;
2026                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2027                 if (opc < REINT_MAX && fmt[opc] != NULL)
2028                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2029                 else {
2030                         mdt = mdt_exp2dev(req->rq_export);
2031                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2032                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2033                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2034                         opc = err_serious(-EFAULT);
2035                 }
2036         } else {
2037                 opc = err_serious(-EFAULT);
2038         }
2039         return opc;
2040 }
2041
2042 static int mdt_reint(struct tgt_session_info *tsi)
2043 {
2044         long opc;
2045         int  rc;
2046         static const struct req_format *reint_fmts[REINT_MAX] = {
2047                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2048                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2049                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2050                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2051                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2052                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2053                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2054                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2055                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2056         };
2057
2058         ENTRY;
2059
2060         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2061         if (opc >= 0) {
2062                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2063                 /*
2064                  * No lock possible here from client to pass it to reint code
2065                  * path.
2066                  */
2067                 rc = mdt_reint_internal(info, NULL, opc);
2068                 mdt_thread_info_fini(info);
2069         } else {
2070                 rc = opc;
2071         }
2072
2073         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2074         RETURN(rc);
2075 }
2076
2077 /* this should sync the whole device */
2078 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2079 {
2080         struct dt_device *dt = mdt->mdt_bottom;
2081         int rc;
2082         ENTRY;
2083
2084         rc = dt->dd_ops->dt_sync(env, dt);
2085         RETURN(rc);
2086 }
2087
2088 /* this should sync this object */
2089 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2090                            struct mdt_object *mo)
2091 {
2092         int rc;
2093
2094         ENTRY;
2095
2096         if (!mdt_object_exists(mo)) {
2097                 CWARN("%s: non existing object "DFID": rc = %d\n",
2098                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2099                       -ESTALE);
2100                 RETURN(-ESTALE);
2101         }
2102
2103         rc = mo_object_sync(env, mdt_object_child(mo));
2104
2105         RETURN(rc);
2106 }
2107
2108 static int mdt_sync(struct tgt_session_info *tsi)
2109 {
2110         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2111         struct req_capsule      *pill = tsi->tsi_pill;
2112         struct mdt_body         *body;
2113         int                      rc;
2114
2115         ENTRY;
2116
2117         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2118                 RETURN(err_serious(-ENOMEM));
2119
2120         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2121                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2122         } else {
2123                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2124
2125                 /* sync an object */
2126                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2127                                      info->mti_object);
2128                 if (rc == 0) {
2129                         const struct lu_fid *fid;
2130                         struct lu_attr *la = &info->mti_attr.ma_attr;
2131
2132                         info->mti_attr.ma_need = MA_INODE;
2133                         info->mti_attr.ma_valid = 0;
2134                         rc = mdt_attr_get_complex(info, info->mti_object,
2135                                                   &info->mti_attr);
2136                         if (rc == 0) {
2137                                 body = req_capsule_server_get(pill,
2138                                                               &RMF_MDT_BODY);
2139                                 fid = mdt_object_fid(info->mti_object);
2140                                 mdt_pack_attr2body(info, body, la, fid);
2141                         }
2142                 }
2143                 mdt_thread_info_fini(info);
2144         }
2145         if (rc == 0)
2146                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2147
2148         RETURN(rc);
2149 }
2150
2151 static int mdt_data_sync(struct tgt_session_info *tsi)
2152 {
2153         struct mdt_thread_info *info;
2154         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2155         struct ost_body *body = tsi->tsi_ost_body;
2156         struct ost_body *repbody;
2157         struct mdt_object *mo = NULL;
2158         struct md_attr *ma;
2159         int rc = 0;
2160
2161         ENTRY;
2162
2163         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2164
2165         /* if no fid is specified then do nothing,
2166          * device sync is done via MDS_SYNC */
2167         if (fid_is_zero(&tsi->tsi_fid))
2168                 RETURN(0);
2169
2170         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2171         if (IS_ERR(mo))
2172                 RETURN(PTR_ERR(mo));
2173
2174         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2175         if (rc)
2176                 GOTO(put, rc);
2177
2178         repbody->oa.o_oi = body->oa.o_oi;
2179         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2180
2181         info = tsi2mdt_info(tsi);
2182         ma = &info->mti_attr;
2183         ma->ma_need = MA_INODE;
2184         ma->ma_valid = 0;
2185         rc = mdt_attr_get_complex(info, mo, ma);
2186         if (rc == 0)
2187                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2188         else
2189                 rc = 0;
2190         mdt_thread_info_fini(info);
2191
2192         EXIT;
2193 put:
2194         if (mo != NULL)
2195                 mdt_object_put(tsi->tsi_env, mo);
2196         return rc;
2197 }
2198
2199 /*
2200  * Handle quota control requests to consult current usage/limit, but also
2201  * to configure quota enforcement
2202  */
2203 static int mdt_quotactl(struct tgt_session_info *tsi)
2204 {
2205         struct obd_export       *exp  = tsi->tsi_exp;
2206         struct req_capsule      *pill = tsi->tsi_pill;
2207         struct obd_quotactl     *oqctl, *repoqc;
2208         int                      id, rc;
2209         struct mdt_device       *mdt = mdt_exp2dev(exp);
2210         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2211         struct lu_nodemap       *nodemap;
2212         ENTRY;
2213
2214         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2215         if (oqctl == NULL)
2216                 RETURN(err_serious(-EPROTO));
2217
2218         rc = req_capsule_server_pack(pill);
2219         if (rc)
2220                 RETURN(err_serious(rc));
2221
2222         nodemap = nodemap_get_from_exp(exp);
2223         if (IS_ERR(nodemap))
2224                 RETURN(PTR_ERR(nodemap));
2225
2226         switch (oqctl->qc_cmd) {
2227                 /* master quotactl */
2228         case Q_SETINFO:
2229         case Q_SETQUOTA:
2230                 if (!nodemap_can_setquota(nodemap))
2231                         GOTO(out_nodemap, rc = -EPERM);
2232         case Q_GETINFO:
2233         case Q_GETQUOTA:
2234                 if (qmt == NULL)
2235                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2236                 /* slave quotactl */
2237         case Q_GETOINFO:
2238         case Q_GETOQUOTA:
2239                 break;
2240         default:
2241                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2242                 GOTO(out_nodemap, rc = -EFAULT);
2243         }
2244
2245         id = oqctl->qc_id;
2246         switch (oqctl->qc_type) {
2247         case USRQUOTA:
2248                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2249                                     NODEMAP_CLIENT_TO_FS, id);
2250                 break;
2251         case GRPQUOTA:
2252                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2253                                     NODEMAP_CLIENT_TO_FS, id);
2254                 break;
2255         case PRJQUOTA:
2256                 /* todo: check/map project id */
2257                 id = oqctl->qc_id;
2258                 break;
2259         default:
2260                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2261         }
2262         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2263         if (repoqc == NULL)
2264                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2265
2266         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2267                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2268
2269         if (oqctl->qc_id != id)
2270                 swap(oqctl->qc_id, id);
2271
2272         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2273                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2274                         RETURN(-EINPROGRESS);
2275         }
2276
2277         switch (oqctl->qc_cmd) {
2278
2279         case Q_GETINFO:
2280         case Q_SETINFO:
2281         case Q_SETQUOTA:
2282         case Q_GETQUOTA:
2283                 /* forward quotactl request to QMT */
2284                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2285                 break;
2286
2287         case Q_GETOINFO:
2288         case Q_GETOQUOTA:
2289                 /* slave quotactl */
2290                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2291                                    oqctl);
2292                 break;
2293
2294         default:
2295                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2296                 GOTO(out_nodemap, rc = -EFAULT);
2297         }
2298
2299         if (oqctl->qc_id != id)
2300                 swap(oqctl->qc_id, id);
2301
2302         *repoqc = *oqctl;
2303
2304         EXIT;
2305
2306 out_nodemap:
2307         nodemap_putref(nodemap);
2308
2309         return rc;
2310 }
2311
2312 /** clone llog ctxt from child (mdd)
2313  * This allows remote llog (replicator) access.
2314  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2315  * context was originally set up, or we can handle them directly.
2316  * I choose the latter, but that means I need any llog
2317  * contexts set up by child to be accessable by the mdt.  So we clone the
2318  * context into our context list here.
2319  */
2320 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2321                                int idx)
2322 {
2323         struct md_device  *next = mdt->mdt_child;
2324         struct llog_ctxt *ctxt;
2325         int rc;
2326
2327         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2328                 return 0;
2329
2330         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2331         if (rc || ctxt == NULL) {
2332                 return 0;
2333         }
2334
2335         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2336         if (rc)
2337                 CERROR("Can't set mdt ctxt %d\n", rc);
2338
2339         return rc;
2340 }
2341
2342 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2343                                  struct mdt_device *mdt, int idx)
2344 {
2345         struct llog_ctxt *ctxt;
2346
2347         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2348         if (ctxt == NULL)
2349                 return 0;
2350         /* Put once for the get we just did, and once for the clone */
2351         llog_ctxt_put(ctxt);
2352         llog_ctxt_put(ctxt);
2353         return 0;
2354 }
2355
2356 /*
2357  * sec context handlers
2358  */
2359 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2360 {
2361         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2362
2363         return 0;
2364 }
2365
2366 /*
2367  * quota request handlers
2368  */
2369 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2370 {
2371         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2372         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2373         int                      rc;
2374         ENTRY;
2375
2376         if (qmt == NULL)
2377                 RETURN(err_serious(-EOPNOTSUPP));
2378
2379         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2380         RETURN(rc);
2381 }
2382
2383 struct mdt_object *mdt_object_new(const struct lu_env *env,
2384                                   struct mdt_device *d,
2385                                   const struct lu_fid *f)
2386 {
2387         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2388         struct lu_object *o;
2389         struct mdt_object *m;
2390         ENTRY;
2391
2392         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2393         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2394         if (unlikely(IS_ERR(o)))
2395                 m = (struct mdt_object *)o;
2396         else
2397                 m = mdt_obj(o);
2398         RETURN(m);
2399 }
2400
2401 struct mdt_object *mdt_object_find(const struct lu_env *env,
2402                                    struct mdt_device *d,
2403                                    const struct lu_fid *f)
2404 {
2405         struct lu_object *o;
2406         struct mdt_object *m;
2407         ENTRY;
2408
2409         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2410         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2411         if (unlikely(IS_ERR(o)))
2412                 m = (struct mdt_object *)o;
2413         else
2414                 m = mdt_obj(o);
2415
2416         RETURN(m);
2417 }
2418
2419 /**
2420  * Asyncronous commit for mdt device.
2421  *
2422  * Pass asynchonous commit call down the MDS stack.
2423  *
2424  * \param env environment
2425  * \param mdt the mdt device
2426  */
2427 static void mdt_device_commit_async(const struct lu_env *env,
2428                                     struct mdt_device *mdt)
2429 {
2430         struct dt_device *dt = mdt->mdt_bottom;
2431         int rc;
2432         ENTRY;
2433
2434         rc = dt->dd_ops->dt_commit_async(env, dt);
2435         if (unlikely(rc != 0))
2436                 CWARN("%s: async commit start failed: rc = %d\n",
2437                       mdt_obd_name(mdt), rc);
2438         atomic_inc(&mdt->mdt_async_commit_count);
2439         EXIT;
2440 }
2441
2442 /**
2443  * Mark the lock as "synchonous".
2444  *
2445  * Mark the lock to deffer transaction commit to the unlock time.
2446  *
2447  * \param lock the lock to mark as "synchonous"
2448  *
2449  * \see mdt_is_lock_sync
2450  * \see mdt_save_lock
2451  */
2452 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2453 {
2454         lock->l_ast_data = (void*)1;
2455 }
2456
2457 /**
2458  * Check whehter the lock "synchonous" or not.
2459  *
2460  * \param lock the lock to check
2461  * \retval 1 the lock is "synchonous"
2462  * \retval 0 the lock isn't "synchronous"
2463  *
2464  * \see mdt_set_lock_sync
2465  * \see mdt_save_lock
2466  */
2467 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2468 {
2469         return lock->l_ast_data != NULL;
2470 }
2471
2472 /**
2473  * Blocking AST for mdt locks.
2474  *
2475  * Starts transaction commit if in case of COS lock conflict or
2476  * deffers such a commit to the mdt_save_lock.
2477  *
2478  * \param lock the lock which blocks a request or cancelling lock
2479  * \param desc unused
2480  * \param data unused
2481  * \param flag indicates whether this cancelling or blocking callback
2482  * \retval 0
2483  * \see ldlm_blocking_ast_nocheck
2484  */
2485 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2486                      void *data, int flag)
2487 {
2488         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2489         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2490         bool commit_async = false;
2491         int rc;
2492         ENTRY;
2493
2494         if (flag == LDLM_CB_CANCELING)
2495                 RETURN(0);
2496
2497         lock_res_and_lock(lock);
2498         if (lock->l_blocking_ast != mdt_blocking_ast) {
2499                 unlock_res_and_lock(lock);
2500                 RETURN(0);
2501         }
2502         /* There is no lock conflict if l_blocking_lock == NULL,
2503          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2504          * when the last reference to a local lock was released */
2505         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2506             lock->l_blocking_lock != NULL) {
2507                 if (mdt_cos_is_enabled(mdt)) {
2508                         if (lock->l_client_cookie !=
2509                             lock->l_blocking_lock->l_client_cookie)
2510                                 mdt_set_lock_sync(lock);
2511                 } else if (mdt_slc_is_enabled(mdt) &&
2512                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2513                         mdt_set_lock_sync(lock);
2514                         /*
2515                          * we may do extra commit here, but there is a small
2516                          * window to miss a commit: lock was unlocked (saved),
2517                          * then a conflict lock queued and we come here, but
2518                          * REP-ACK not received, so lock was not converted to
2519                          * COS mode yet.
2520                          * Fortunately this window is quite small, so the
2521                          * extra commit should be rare (not to say distributed
2522                          * operation is rare too).
2523                          */
2524                         commit_async = true;
2525                 }
2526         } else if (lock->l_req_mode == LCK_COS &&
2527                    lock->l_blocking_lock != NULL) {
2528                 commit_async = true;
2529         }
2530
2531         rc = ldlm_blocking_ast_nocheck(lock);
2532
2533         if (commit_async) {
2534                 struct lu_env env;
2535
2536                 rc = lu_env_init(&env, LCT_LOCAL);
2537                 if (unlikely(rc != 0))
2538                         CWARN("%s: lu_env initialization failed, cannot "
2539                               "start asynchronous commit: rc = %d\n",
2540                               obd->obd_name, rc);
2541                 else
2542                         mdt_device_commit_async(&env, mdt);
2543                 lu_env_fini(&env);
2544         }
2545         RETURN(rc);
2546 }
2547
2548 /*
2549  * Blocking AST for cross-MDT lock
2550  *
2551  * Discard lock from uncommitted_slc_locks and cancel it.
2552  *
2553  * \param lock  the lock which blocks a request or cancelling lock
2554  * \param desc  unused
2555  * \param data  unused
2556  * \param flag  indicates whether this cancelling or blocking callback
2557  * \retval      0 on success
2558  * \retval      negative number on error
2559  */
2560 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2561                             void *data, int flag)
2562 {
2563         int rc = 0;
2564         ENTRY;
2565
2566         switch (flag) {
2567         case LDLM_CB_BLOCKING: {
2568                 struct lustre_handle lockh;
2569
2570                 ldlm_lock2handle(lock, &lockh);
2571                 rc = ldlm_cli_cancel(&lockh,
2572                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2573                 if (rc < 0) {
2574                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2575                         RETURN(rc);
2576                 }
2577                 break;
2578         }
2579         case LDLM_CB_CANCELING: {
2580                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2581                 struct mdt_device *mdt =
2582                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2583
2584                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2585
2586                 /* discard slc lock here so that it can be cleaned anytime,
2587                  * especially for cleanup_resource() */
2588                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2589
2590                 /* once we cache lock, l_ast_data is set to mdt_object */
2591                 if (lock->l_ast_data != NULL) {
2592                         struct mdt_object *mo = lock->l_ast_data;
2593                         struct lu_env env;
2594
2595                         rc = lu_env_init(&env, LCT_MD_THREAD);
2596                         if (unlikely(rc != 0)) {
2597                                 CWARN("%s: lu_env initialization failed, object"
2598                                       "%p "DFID" is leaked!\n",
2599                                       obd->obd_name, mo,
2600                                       PFID(mdt_object_fid(mo)));
2601                                 RETURN(rc);
2602                         }
2603
2604                         if (lock->l_policy_data.l_inodebits.bits &
2605                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2606                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2607                                 mo->mot_cache_attr = 0;
2608                         }
2609                         mdt_object_put(&env, mo);
2610                         lu_env_fini(&env);
2611                 }
2612                 break;
2613         }
2614         default:
2615                 LBUG();
2616         }
2617
2618         RETURN(rc);
2619 }
2620
2621 int mdt_check_resent_lock(struct mdt_thread_info *info,
2622                           struct mdt_object *mo,
2623                           struct mdt_lock_handle *lhc)
2624 {
2625         /* the lock might already be gotten in ldlm_handle_enqueue() */
2626         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2627                 struct ptlrpc_request *req = mdt_info_req(info);
2628                 struct ldlm_lock      *lock;
2629
2630                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2631                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2632                 if (lock == NULL) {
2633                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2634                          * a resend case, however, it could be already destroyed
2635                          * due to client eviction or a raced cancel RPC. */
2636                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2637                                           lhc->mlh_reg_lh.cookie);
2638                         RETURN(-ESTALE);
2639                 }
2640
2641                 if (!fid_res_name_eq(mdt_object_fid(mo),
2642                                      &lock->l_resource->lr_name)) {
2643                         CWARN("%s: Although resent, but still not "
2644                               "get child lock:"DFID"\n",
2645                               info->mti_exp->exp_obd->obd_name,
2646                               PFID(mdt_object_fid(mo)));
2647                         LDLM_LOCK_PUT(lock);
2648                         RETURN(-EPROTO);
2649                 }
2650                 LDLM_LOCK_PUT(lock);
2651                 return 0;
2652         }
2653         return 1;
2654 }
2655
2656 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2657                                struct mdt_object *o, const struct lu_fid *fid,
2658                                struct lustre_handle *lh, enum ldlm_mode mode,
2659                                __u64 *ibits, __u64 trybits, bool cache)
2660 {
2661         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2662         union ldlm_policy_data *policy = &mti->mti_policy;
2663         struct ldlm_res_id *res_id = &mti->mti_res_id;
2664         int rc = 0;
2665         ENTRY;
2666
2667         LASSERT(mdt_object_remote(o));
2668
2669         fid_build_reg_res_name(fid, res_id);
2670
2671         memset(einfo, 0, sizeof(*einfo));
2672         einfo->ei_type = LDLM_IBITS;
2673         einfo->ei_mode = mode;
2674         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2675         einfo->ei_cb_cp = ldlm_completion_ast;
2676         einfo->ei_enq_slave = 0;
2677         einfo->ei_res_id = res_id;
2678
2679         if (cache) {
2680                 /*
2681                  * if we cache lock, couple lock with mdt_object, so that object
2682                  * can be easily found in lock ASTs.
2683                  */
2684                 mdt_object_get(mti->mti_env, o);
2685                 einfo->ei_cbdata = o;
2686         }
2687
2688
2689         memset(policy, 0, sizeof(*policy));
2690         policy->l_inodebits.bits = *ibits;
2691         policy->l_inodebits.try_bits = trybits;
2692
2693         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2694                             policy);
2695         if (rc < 0 && cache) {
2696                 mdt_object_put(mti->mti_env, o);
2697                 einfo->ei_cbdata = NULL;
2698         }
2699
2700         /* Return successfully acquired bits to a caller */
2701         if (rc == 0) {
2702                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2703
2704                 LASSERT(lock);
2705                 *ibits = lock->l_policy_data.l_inodebits.bits;
2706                 LDLM_LOCK_PUT(lock);
2707         }
2708         RETURN(rc);
2709 }
2710
2711 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2712                            const struct lu_fid *fid, struct lustre_handle *lh,
2713                            enum ldlm_mode mode, __u64 ibits, bool cache)
2714 {
2715         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2716                                           cache);
2717 }
2718
2719 static int mdt_object_local_lock(struct mdt_thread_info *info,
2720                                  struct mdt_object *o,
2721                                  struct mdt_lock_handle *lh, __u64 *ibits,
2722                                  __u64 trybits, bool cos_incompat)
2723 {
2724         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2725         union ldlm_policy_data *policy = &info->mti_policy;
2726         struct ldlm_res_id *res_id = &info->mti_res_id;
2727         __u64 dlmflags = 0;
2728         int rc;
2729         ENTRY;
2730
2731         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2732         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2733         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2734         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2735
2736         if (cos_incompat) {
2737                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2738                         lh->mlh_reg_mode == LCK_EX);
2739                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2740         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2741                 dlmflags |= LDLM_FL_COS_ENABLED;
2742         }
2743
2744         /* Only enqueue LOOKUP lock for remote object */
2745         if (mdt_object_remote(o)) {
2746                 LASSERT(*ibits == MDS_INODELOCK_LOOKUP);
2747         }
2748
2749         if (lh->mlh_type == MDT_PDO_LOCK) {
2750                 /* check for exists after object is locked */
2751                 if (mdt_object_exists(o) == 0) {
2752                         /* Non-existent object shouldn't have PDO lock */
2753                         RETURN(-ESTALE);
2754                 } else {
2755                         /* Non-dir object shouldn't have PDO lock */
2756                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2757                                 RETURN(-ENOTDIR);
2758                 }
2759         }
2760
2761
2762         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2763         dlmflags |= LDLM_FL_ATOMIC_CB;
2764
2765         /*
2766          * Take PDO lock on whole directory and build correct @res_id for lock
2767          * on part of directory.
2768          */
2769         if (lh->mlh_pdo_hash != 0) {
2770                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2771                 mdt_lock_pdo_mode(info, o, lh);
2772                 if (lh->mlh_pdo_mode != LCK_NL) {
2773                         /*
2774                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2775                          * is never going to be sent to client and we do not
2776                          * want it slowed down due to possible cancels.
2777                          */
2778                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2779                         policy->l_inodebits.try_bits = 0;
2780                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2781                                           policy, res_id, dlmflags,
2782                                           info->mti_exp == NULL ? NULL :
2783                                           &info->mti_exp->exp_handle.h_cookie);
2784                         if (unlikely(rc != 0))
2785                                 GOTO(out_unlock, rc);
2786                 }
2787
2788                 /*
2789                  * Finish res_id initializing by name hash marking part of
2790                  * directory which is taking modification.
2791                  */
2792                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2793         }
2794
2795         policy->l_inodebits.bits = *ibits;
2796         policy->l_inodebits.try_bits = trybits;
2797
2798         /*
2799          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2800          * going to be sent to client. If it is - mdt_intent_policy() path will
2801          * fix it up and turn FL_LOCAL flag off.
2802          */
2803         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2804                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2805                           info->mti_exp == NULL ? NULL :
2806                           &info->mti_exp->exp_handle.h_cookie);
2807 out_unlock:
2808         if (rc != 0)
2809                 mdt_object_unlock(info, o, lh, 1);
2810         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2811                    lh->mlh_pdo_hash != 0 &&
2812                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2813                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2814
2815         /* Return successfully acquired bits to a caller */
2816         if (rc == 0) {
2817                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2818
2819                 LASSERT(lock);
2820                 *ibits = lock->l_policy_data.l_inodebits.bits;
2821                 LDLM_LOCK_PUT(lock);
2822         }
2823         RETURN(rc);
2824 }
2825
2826 static int
2827 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2828                          struct mdt_lock_handle *lh, __u64 *ibits,
2829                          __u64 trybits, bool cos_incompat)
2830 {
2831         struct mdt_lock_handle *local_lh = NULL;
2832         int rc;
2833         ENTRY;
2834
2835         if (!mdt_object_remote(o)) {
2836                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
2837                                            cos_incompat);
2838                 RETURN(rc);
2839         }
2840
2841         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2842         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2843                     MDS_INODELOCK_XATTR);
2844
2845         /* Only enqueue LOOKUP lock for remote object */
2846         if (*ibits & MDS_INODELOCK_LOOKUP) {
2847                 __u64 local = MDS_INODELOCK_LOOKUP;
2848
2849                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
2850                                            cos_incompat);
2851                 if (rc != ELDLM_OK)
2852                         RETURN(rc);
2853
2854                 local_lh = lh;
2855         }
2856
2857         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
2858                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2859                  * enqueue RPC can only request 1 lock, to avoid extra
2860                  * RPC, so it will instead enqueue EX lock for remote
2861                  * object anyway XXX*/
2862                 if (lh->mlh_type == MDT_PDO_LOCK &&
2863                     lh->mlh_pdo_hash != 0) {
2864                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2865                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2866                                PFID(mdt_object_fid(o)));
2867                         lh->mlh_pdo_hash = 0;
2868                         lh->mlh_rreg_mode = LCK_EX;
2869                         lh->mlh_type = MDT_REG_LOCK;
2870                 }
2871
2872                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
2873                                                 &lh->mlh_rreg_lh,
2874                                                 lh->mlh_rreg_mode,
2875                                                 ibits, trybits, false);
2876                 if (rc != ELDLM_OK) {
2877                         if (local_lh != NULL)
2878                                 mdt_object_unlock(info, o, local_lh, rc);
2879                         RETURN(rc);
2880                 }
2881         }
2882
2883         RETURN(0);
2884 }
2885
2886 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2887                     struct mdt_lock_handle *lh, __u64 ibits)
2888 {
2889         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
2890 }
2891
2892 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2893                           struct mdt_lock_handle *lh, __u64 ibits,
2894                           bool cos_incompat)
2895 {
2896         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2897         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
2898                                         cos_incompat);
2899 }
2900
2901 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2902                         struct mdt_lock_handle *lh, __u64 *ibits,
2903                         __u64 trybits, bool cos_incompat)
2904 {
2905         return mdt_object_lock_internal(info, o, lh, ibits, trybits,
2906                                         cos_incompat);
2907 }
2908
2909 /**
2910  * Save a lock within request object.
2911  *
2912  * Keep the lock referenced until whether client ACK or transaction
2913  * commit happens or release the lock immediately depending on input
2914  * parameters. If COS is ON, a write lock is converted to COS lock
2915  * before saving.
2916  *
2917  * \param info thead info object
2918  * \param h lock handle
2919  * \param mode lock mode
2920  * \param decref force immediate lock releasing
2921  */
2922 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2923                    enum ldlm_mode mode, int decref)
2924 {
2925         ENTRY;
2926
2927         if (lustre_handle_is_used(h)) {
2928                 if (decref || !info->mti_has_trans ||
2929                     !(mode & (LCK_PW | LCK_EX))) {
2930                         mdt_fid_unlock(h, mode);
2931                 } else {
2932                         struct mdt_device *mdt = info->mti_mdt;
2933                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2934                         struct ptlrpc_request *req = mdt_info_req(info);
2935                         bool cos = mdt_cos_is_enabled(mdt);
2936                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
2937
2938                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2939                                  h->cookie);
2940
2941                         /* there is no request if mdt_object_unlock() is called
2942                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2943                         if (likely(req != NULL)) {
2944                                 LDLM_DEBUG(lock, "save lock request %p reply "
2945                                         "state %p transno %lld\n", req,
2946                                         req->rq_reply_state, req->rq_transno);
2947                                 if (cos) {
2948                                         ldlm_lock_downgrade(lock, LCK_COS);
2949                                         mode = LCK_COS;
2950                                 }
2951                                 if (req->rq_export->exp_disconnected)
2952                                         mdt_fid_unlock(h, mode);
2953                                 else
2954                                         ptlrpc_save_lock(req, h, mode, cos,
2955                                                          convert_lock);
2956                         } else {
2957                                 mdt_fid_unlock(h, mode);
2958                         }
2959                         if (mdt_is_lock_sync(lock)) {
2960                                 CDEBUG(D_HA, "found sync-lock,"
2961                                        " async commit started\n");
2962                                 mdt_device_commit_async(info->mti_env,
2963                                                         mdt);
2964                         }
2965                         LDLM_LOCK_PUT(lock);
2966                 }
2967                 h->cookie = 0ull;
2968         }
2969
2970         EXIT;
2971 }
2972
2973 /**
2974  * Save cross-MDT lock in uncommitted_slc_locks
2975  *
2976  * Keep the lock referenced until transaction commit happens or release the lock
2977  * immediately depending on input parameters.
2978  *
2979  * \param info thead info object
2980  * \param h lock handle
2981  * \param mode lock mode
2982  * \param decref force immediate lock releasing
2983  */
2984 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2985                                  struct mdt_object *o, struct lustre_handle *h,
2986                                  enum ldlm_mode mode, int decref)
2987 {
2988         ENTRY;
2989
2990         if (lustre_handle_is_used(h)) {
2991                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2992
2993                 if (o != NULL &&
2994                     (lock->l_policy_data.l_inodebits.bits &
2995                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2996                         mo_invalidate(info->mti_env, mdt_object_child(o));
2997
2998                 if (decref || !info->mti_has_trans ||
2999                     !(mode & (LCK_PW | LCK_EX))) {
3000                         ldlm_lock_decref_and_cancel(h, mode);
3001                         LDLM_LOCK_PUT(lock);
3002                 } else {
3003                         struct ptlrpc_request *req = mdt_info_req(info);
3004
3005                         LASSERT(req != NULL);
3006                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3007                                           req->rq_transno);
3008                         ldlm_lock_decref(h, mode);
3009                 }
3010                 h->cookie = 0ull;
3011         }
3012
3013         EXIT;
3014 }
3015
3016 /**
3017  * Unlock mdt object.
3018  *
3019  * Immeditely release the regular lock and the PDO lock or save the
3020  * lock in request and keep them referenced until client ACK or
3021  * transaction commit.
3022  *
3023  * \param info thread info object
3024  * \param o mdt object
3025  * \param lh mdt lock handle referencing regular and PDO locks
3026  * \param decref force immediate lock releasing
3027  *
3028  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3029  */
3030 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3031                        struct mdt_lock_handle *lh, int decref)
3032 {
3033         ENTRY;
3034
3035         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3036         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3037         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3038                              decref);
3039
3040         EXIT;
3041 }
3042
3043 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3044                                         const struct lu_fid *f,
3045                                         struct mdt_lock_handle *lh,
3046                                         __u64 ibits)
3047 {
3048         struct mdt_object *o;
3049
3050         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3051         if (!IS_ERR(o)) {
3052                 int rc;
3053
3054                 rc = mdt_object_lock(info, o, lh, ibits);
3055                 if (rc != 0) {
3056                         mdt_object_put(info->mti_env, o);
3057                         o = ERR_PTR(rc);
3058                 }
3059         }
3060         return o;
3061 }
3062
3063 void mdt_object_unlock_put(struct mdt_thread_info * info,
3064                            struct mdt_object * o,
3065                            struct mdt_lock_handle *lh,
3066                            int decref)
3067 {
3068         mdt_object_unlock(info, o, lh, decref);
3069         mdt_object_put(info->mti_env, o);
3070 }
3071
3072 /*
3073  * Generic code handling requests that have struct mdt_body passed in:
3074  *
3075  *  - extract mdt_body from request and save it in @info, if present;
3076  *
3077  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3078  *  @info;
3079  *
3080  *  - if HABEO_CORPUS flag is set for this request type check whether object
3081  *  actually exists on storage (lu_object_exists()).
3082  *
3083  */
3084 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
3085 {
3086         const struct mdt_body    *body;
3087         struct mdt_object        *obj;
3088         const struct lu_env      *env;
3089         struct req_capsule       *pill;
3090         int                       rc;
3091         ENTRY;
3092
3093         env = info->mti_env;
3094         pill = info->mti_pill;
3095
3096         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3097         if (body == NULL)
3098                 RETURN(-EFAULT);
3099
3100         if (!(body->mbo_valid & OBD_MD_FLID))
3101                 RETURN(0);
3102
3103         if (!fid_is_sane(&body->mbo_fid1)) {
3104                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3105                 RETURN(-EINVAL);
3106         }
3107
3108         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3109         if (!IS_ERR(obj)) {
3110                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3111                         mdt_object_put(env, obj);
3112                         rc = -ENOENT;
3113                 } else {
3114                         info->mti_object = obj;
3115                         rc = 0;
3116                 }
3117         } else
3118                 rc = PTR_ERR(obj);
3119
3120         RETURN(rc);
3121 }
3122
3123 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
3124 {
3125         struct req_capsule *pill = info->mti_pill;
3126         int rc;
3127         ENTRY;
3128
3129         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3130                 rc = mdt_body_unpack(info, flags);
3131         else
3132                 rc = 0;
3133
3134         if (rc == 0 && (flags & HABEO_REFERO)) {
3135                 /* Pack reply. */
3136                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3137                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3138                                              DEF_REP_MD_SIZE);
3139                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3140                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3141                                              RCL_SERVER, 0);
3142
3143                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3144                  * by default. If the target object has more ACL entries, then
3145                  * enlarge the buffer when necessary. */
3146                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3147                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3148                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3149
3150                 rc = req_capsule_server_pack(pill);
3151         }
3152         RETURN(rc);
3153 }
3154
3155 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3156 {
3157         lh->mlh_type = MDT_NUL_LOCK;
3158         lh->mlh_reg_lh.cookie = 0ull;
3159         lh->mlh_reg_mode = LCK_MINMODE;
3160         lh->mlh_pdo_lh.cookie = 0ull;
3161         lh->mlh_pdo_mode = LCK_MINMODE;
3162         lh->mlh_rreg_lh.cookie = 0ull;
3163         lh->mlh_rreg_mode = LCK_MINMODE;
3164 }
3165
3166 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3167 {
3168         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3169         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3170 }
3171
3172 /*
3173  * Initialize fields of struct mdt_thread_info. Other fields are left in
3174  * uninitialized state, because it's too expensive to zero out whole
3175  * mdt_thread_info (> 1K) on each request arrival.
3176  */
3177 void mdt_thread_info_init(struct ptlrpc_request *req,
3178                           struct mdt_thread_info *info)
3179 {
3180         int i;
3181
3182         info->mti_pill = &req->rq_pill;
3183
3184         /* lock handle */
3185         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3186                 mdt_lock_handle_init(&info->mti_lh[i]);
3187
3188         /* mdt device: it can be NULL while CONNECT */
3189         if (req->rq_export) {
3190                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3191                 info->mti_exp = req->rq_export;
3192         } else
3193                 info->mti_mdt = NULL;
3194         info->mti_env = req->rq_svc_thread->t_env;
3195         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3196
3197         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3198         info->mti_big_buf = LU_BUF_NULL;
3199         info->mti_body = NULL;
3200         info->mti_object = NULL;
3201         info->mti_dlm_req = NULL;
3202         info->mti_has_trans = 0;
3203         info->mti_cross_ref = 0;
3204         info->mti_opdata = 0;
3205         info->mti_big_lmm_used = 0;
3206         info->mti_big_acl_used = 0;
3207
3208         info->mti_spec.no_create = 0;
3209         info->mti_spec.sp_rm_entry = 0;
3210         info->mti_spec.sp_permitted = 0;
3211         info->mti_spec.sp_migrate_close = 0;
3212
3213         info->mti_spec.u.sp_ea.eadata = NULL;
3214         info->mti_spec.u.sp_ea.eadatalen = 0;
3215 }
3216
3217 void mdt_thread_info_fini(struct mdt_thread_info *info)
3218 {
3219         int i;
3220
3221         if (info->mti_object != NULL) {
3222                 mdt_object_put(info->mti_env, info->mti_object);
3223                 info->mti_object = NULL;
3224         }
3225
3226         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3227                 mdt_lock_handle_fini(&info->mti_lh[i]);
3228         info->mti_env = NULL;
3229         info->mti_pill = NULL;
3230         info->mti_exp = NULL;
3231
3232         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3233                 lu_buf_free(&info->mti_big_buf);
3234 }
3235
3236 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3237 {
3238         struct mdt_thread_info  *mti;
3239
3240         mti = mdt_th_info(tsi->tsi_env);
3241         LASSERT(mti != NULL);
3242
3243         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3244         if (tsi->tsi_corpus != NULL) {
3245                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3246                 lu_object_get(tsi->tsi_corpus);
3247         }
3248         mti->mti_body = tsi->tsi_mdt_body;
3249         mti->mti_dlm_req = tsi->tsi_dlm_req;
3250
3251         return mti;
3252 }
3253
3254 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3255 {
3256         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3257             cfs_fail_val ==
3258             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3259                 set_current_state(TASK_UNINTERRUPTIBLE);
3260                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3261         }
3262
3263         return tgt_connect(tsi);
3264 }
3265
3266 enum mdt_it_code {
3267         MDT_IT_OPEN,
3268         MDT_IT_OCREAT,
3269         MDT_IT_CREATE,
3270         MDT_IT_GETATTR,
3271         MDT_IT_READDIR,
3272         MDT_IT_LOOKUP,
3273         MDT_IT_UNLINK,
3274         MDT_IT_TRUNC,
3275         MDT_IT_GETXATTR,
3276         MDT_IT_LAYOUT,
3277         MDT_IT_QUOTA,
3278         MDT_IT_NR
3279 };
3280
3281 static int mdt_intent_getattr(enum mdt_it_code opcode,
3282                               struct mdt_thread_info *info,
3283                               struct ldlm_lock **,
3284                               __u64);
3285
3286 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3287                                 struct mdt_thread_info *info,
3288                                 struct ldlm_lock **lockp,
3289                                 __u64 flags);
3290
3291 static int mdt_intent_layout(enum mdt_it_code opcode,
3292                              struct mdt_thread_info *info,
3293                              struct ldlm_lock **,
3294                              __u64);
3295 static int mdt_intent_reint(enum mdt_it_code opcode,
3296                             struct mdt_thread_info *info,
3297                             struct ldlm_lock **,
3298                             __u64);
3299
3300 static struct mdt_it_flavor {
3301         const struct req_format *it_fmt;
3302         __u32                    it_flags;
3303         int                    (*it_act)(enum mdt_it_code ,
3304                                          struct mdt_thread_info *,
3305                                          struct ldlm_lock **,
3306                                          __u64);
3307         long                     it_reint;
3308 } mdt_it_flavor[] = {
3309         [MDT_IT_OPEN]     = {
3310                 .it_fmt   = &RQF_LDLM_INTENT,
3311                 /*.it_flags = HABEO_REFERO,*/
3312                 .it_flags = 0,
3313                 .it_act   = mdt_intent_reint,
3314                 .it_reint = REINT_OPEN
3315         },
3316         [MDT_IT_OCREAT]   = {
3317                 .it_fmt   = &RQF_LDLM_INTENT,
3318                 /*
3319                  * OCREAT is not a MUTABOR request as if the file
3320                  * already exists.
3321                  * We do the extra check of OBD_CONNECT_RDONLY in
3322                  * mdt_reint_open() when we really need to create
3323                  * the object.
3324                  */
3325                 .it_flags = 0,
3326                 .it_act   = mdt_intent_reint,
3327                 .it_reint = REINT_OPEN
3328         },
3329         [MDT_IT_CREATE]   = {
3330                 .it_fmt   = &RQF_LDLM_INTENT,
3331                 .it_flags = MUTABOR,
3332                 .it_act   = mdt_intent_reint,
3333                 .it_reint = REINT_CREATE
3334         },
3335         [MDT_IT_GETATTR]  = {
3336                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3337                 .it_flags = HABEO_REFERO,
3338                 .it_act   = mdt_intent_getattr
3339         },
3340         [MDT_IT_READDIR]  = {
3341                 .it_fmt   = NULL,
3342                 .it_flags = 0,
3343                 .it_act   = NULL
3344         },
3345         [MDT_IT_LOOKUP]   = {
3346                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3347                 .it_flags = HABEO_REFERO,
3348                 .it_act   = mdt_intent_getattr
3349         },
3350         [MDT_IT_UNLINK]   = {
3351                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3352                 .it_flags = MUTABOR,
3353                 .it_act   = NULL,
3354                 .it_reint = REINT_UNLINK
3355         },
3356         [MDT_IT_TRUNC]    = {
3357                 .it_fmt   = NULL,
3358                 .it_flags = MUTABOR,
3359                 .it_act   = NULL
3360         },
3361         [MDT_IT_GETXATTR] = {
3362                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3363                 .it_flags = HABEO_CORPUS,
3364                 .it_act   = mdt_intent_getxattr
3365         },
3366         [MDT_IT_LAYOUT] = {
3367                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3368                 .it_flags = 0,
3369                 .it_act   = mdt_intent_layout
3370         }
3371 };
3372
3373 static int
3374 mdt_intent_lock_replace(struct mdt_thread_info *info,
3375                         struct ldlm_lock **lockp,
3376                         struct mdt_lock_handle *lh,
3377                         __u64 flags, int result)
3378 {
3379         struct ptlrpc_request  *req = mdt_info_req(info);
3380         struct ldlm_lock       *lock = *lockp;
3381         struct ldlm_lock       *new_lock;
3382
3383         /* If possible resent found a lock, @lh is set to its handle */
3384         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3385
3386         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3387                 lh->mlh_reg_lh.cookie = 0;
3388                 RETURN(0);
3389         }
3390
3391         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3392                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3393                  * a resend case, however, it could be already destroyed
3394                  * due to client eviction or a raced cancel RPC. */
3395                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3396                                   lh->mlh_reg_lh.cookie);
3397                 lh->mlh_reg_lh.cookie = 0;
3398                 RETURN(-ESTALE);
3399         }
3400
3401         LASSERTF(new_lock != NULL,
3402                  "lockh %#llx flags %#llx : rc = %d\n",
3403                  lh->mlh_reg_lh.cookie, flags, result);
3404
3405         /*
3406          * If we've already given this lock to a client once, then we should
3407          * have no readers or writers.  Otherwise, we should have one reader
3408          * _or_ writer ref (which will be zeroed below) before returning the
3409          * lock to a client.
3410          */
3411         if (new_lock->l_export == req->rq_export) {
3412                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3413         } else {
3414                 LASSERT(new_lock->l_export == NULL);
3415                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3416         }
3417
3418         *lockp = new_lock;
3419
3420         if (new_lock->l_export == req->rq_export) {
3421                 /*
3422                  * Already gave this to the client, which means that we
3423                  * reconstructed a reply.
3424                  */
3425                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3426                         MSG_RESENT);
3427
3428                 LDLM_LOCK_RELEASE(new_lock);
3429                 lh->mlh_reg_lh.cookie = 0;
3430                 RETURN(ELDLM_LOCK_REPLACED);
3431         }
3432
3433         /*
3434          * Fixup the lock to be given to the client.
3435          */
3436         lock_res_and_lock(new_lock);
3437         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3438          * possible blocking AST. */
3439         while (new_lock->l_readers > 0) {
3440                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3441                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3442                 new_lock->l_readers--;
3443         }
3444         while (new_lock->l_writers > 0) {
3445                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3446                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3447                 new_lock->l_writers--;
3448         }
3449
3450         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3451         new_lock->l_blocking_ast = lock->l_blocking_ast;
3452         new_lock->l_completion_ast = lock->l_completion_ast;
3453         new_lock->l_remote_handle = lock->l_remote_handle;
3454         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3455
3456         unlock_res_and_lock(new_lock);
3457
3458         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3459                      &new_lock->l_remote_handle,
3460                      &new_lock->l_exp_hash);
3461
3462         LDLM_LOCK_RELEASE(new_lock);
3463         lh->mlh_reg_lh.cookie = 0;
3464
3465         RETURN(ELDLM_LOCK_REPLACED);
3466 }
3467
3468 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3469                                     struct ldlm_lock *new_lock,
3470                                     struct mdt_lock_handle *lh,
3471                                     __u64 flags)
3472 {
3473         struct ptlrpc_request  *req = mdt_info_req(info);
3474         struct ldlm_request    *dlmreq;
3475
3476         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3477                 return;
3478
3479         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3480
3481         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3482          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3483          * initialized. */
3484         if (flags & LDLM_FL_RESENT) {
3485                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3486                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3487
3488                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3489                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3490                           lh->mlh_reg_lh.cookie);
3491                 return;
3492         }
3493
3494         /*
3495          * If the xid matches, then we know this is a resent request, and allow
3496          * it. (It's probably an OPEN, for which we don't send a lock.
3497          */
3498         if (req_can_reconstruct(req, NULL))
3499                 return;
3500
3501         /*
3502          * This remote handle isn't enqueued, so we never received or processed
3503          * this request.  Clear MSG_RESENT, because it can be handled like any
3504          * normal request now.
3505          */
3506         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3507
3508         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3509                   dlmreq->lock_handle[0].cookie);
3510 }
3511
3512 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3513                                 struct mdt_thread_info *info,
3514                                 struct ldlm_lock **lockp,
3515                                 __u64 flags)
3516 {
3517         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3518         struct ldlm_reply      *ldlm_rep = NULL;
3519         int rc;
3520         ENTRY;
3521
3522         /*
3523          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3524          * (for the resend case) or a new lock. Below we will use it to
3525          * replace the original lock.
3526          */
3527         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3528         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3529                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3530                 rc = mdt_object_lock(info, info->mti_object, lhc,
3531                                      MDS_INODELOCK_XATTR);
3532                 if (rc)
3533                         return rc;
3534         }
3535
3536         rc = mdt_getxattr(info);
3537
3538         if (mdt_info_req(info)->rq_repmsg != NULL)
3539                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3540
3541         if (ldlm_rep == NULL ||
3542             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3543                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3544                 RETURN(err_serious(-EFAULT));
3545         }
3546
3547         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3548
3549         /* This is left for interop instead of adding a new interop flag.
3550          * LU-7433 */
3551 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3552         if (ldlm_rep->lock_policy_res2) {
3553                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3554                 RETURN(ELDLM_LOCK_ABORTED);
3555         }
3556 #endif
3557
3558         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3559         RETURN(rc);
3560 }
3561
3562 static int mdt_intent_getattr(enum mdt_it_code opcode,
3563                               struct mdt_thread_info *info,
3564                               struct ldlm_lock **lockp,
3565                               __u64 flags)
3566 {
3567         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3568         __u64                   child_bits;
3569         struct ldlm_reply      *ldlm_rep;
3570         struct mdt_body        *reqbody;
3571         struct mdt_body        *repbody;
3572         int                     rc, rc2;
3573         ENTRY;
3574
3575         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3576         LASSERT(reqbody);
3577
3578         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3579         LASSERT(repbody);
3580
3581         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3582         repbody->mbo_eadatasize = 0;
3583         repbody->mbo_aclsize = 0;
3584
3585         switch (opcode) {
3586         case MDT_IT_LOOKUP:
3587                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3588                 break;
3589         case MDT_IT_GETATTR:
3590                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3591                              MDS_INODELOCK_PERM;
3592                 break;
3593         default:
3594                 CERROR("Unsupported intent (%d)\n", opcode);
3595                 GOTO(out_shrink, rc = -EINVAL);
3596         }
3597
3598         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3599         if (rc)
3600                 GOTO(out_shrink, rc);
3601
3602         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3603         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3604
3605         /* Get lock from request for possible resent case. */
3606         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3607
3608         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3609         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3610
3611         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3612                 ldlm_rep->lock_policy_res2 = 0;
3613         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3614             ldlm_rep->lock_policy_res2) {
3615                 lhc->mlh_reg_lh.cookie = 0ull;
3616                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3617         }
3618
3619         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3620         EXIT;
3621 out_ucred:
3622         mdt_exit_ucred(info);
3623 out_shrink:
3624         mdt_client_compatibility(info);
3625         rc2 = mdt_fix_reply(info);
3626         if (rc == 0)
3627                 rc = rc2;
3628         return rc;
3629 }
3630
3631 static int mdt_intent_layout(enum mdt_it_code opcode,
3632                              struct mdt_thread_info *info,
3633                              struct ldlm_lock **lockp,
3634                              __u64 flags)
3635 {
3636         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3637         struct layout_intent *layout;
3638         struct lu_fid *fid;
3639         struct mdt_object *obj = NULL;
3640         bool layout_change = false;
3641         int layout_size = 0;
3642         int rc = 0;
3643         ENTRY;
3644
3645         if (opcode != MDT_IT_LAYOUT) {
3646                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3647                         opcode);
3648                 RETURN(-EINVAL);
3649         }
3650
3651         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3652         if (layout == NULL)
3653                 RETURN(-EPROTO);
3654
3655         switch (layout->li_opc) {
3656         case LAYOUT_INTENT_TRUNC:
3657         case LAYOUT_INTENT_WRITE:
3658                 layout_change = true;
3659                 break;
3660         case LAYOUT_INTENT_ACCESS:
3661                 break;
3662         case LAYOUT_INTENT_READ:
3663         case LAYOUT_INTENT_GLIMPSE:
3664         case LAYOUT_INTENT_RELEASE:
3665         case LAYOUT_INTENT_RESTORE:
3666                 CERROR("%s: Unsupported layout intent opc %d\n",
3667                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3668                 rc = -ENOTSUPP;
3669                 break;
3670         default:
3671                 CERROR("%s: Unknown layout intent opc %d\n",
3672                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3673                 rc = -EINVAL;
3674                 break;
3675         }
3676         if (rc < 0)
3677                 RETURN(rc);
3678
3679         fid = &info->mti_tmp_fid2;
3680         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3681
3682         /* Get lock from request for possible resent case. */
3683         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3684
3685         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3686         if (IS_ERR(obj))
3687                 GOTO(out, rc = PTR_ERR(obj));
3688
3689         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3690                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3691                 if (layout_size < 0)
3692                         GOTO(out_obj, rc = layout_size);
3693
3694                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3695                         info->mti_mdt->mdt_max_mdsize = layout_size;
3696         }
3697
3698         /*
3699          * set reply buffer size, so that ldlm_handle_enqueue0()->
3700          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3701          */
3702         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3703         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3704                              layout_size);
3705         rc = req_capsule_server_pack(info->mti_pill);
3706         if (rc)
3707                 GOTO(out_obj, rc);
3708
3709
3710         if (layout_change) {
3711                 struct lu_buf *buf = &info->mti_buf;
3712
3713                 /**
3714                  * mdt_layout_change is a reint operation, when the request
3715                  * is resent, layout write shouldn't reprocess it again.
3716                  */
3717                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3718                 if (rc)
3719                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3720
3721                 /**
3722                  * There is another resent case: the client's job has been
3723                  * done by another client, referring lod_declare_layout_change
3724                  * -EALREADY case, and it became a operation w/o transaction,
3725                  * so we should not do the layout change, otherwise
3726                  * mdt_layout_change() will try to cancel the granted server
3727                  * CR lock whose remote counterpart is still in hold on the
3728                  * client, and a deadlock ensues.
3729                  */
3730                 rc = mdt_check_resent_lock(info, obj, lhc);
3731                 if (rc <= 0)
3732                         GOTO(out_obj, rc);
3733
3734                 buf->lb_buf = NULL;
3735                 buf->lb_len = 0;
3736                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3737                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3738                                         &RMF_EADATA);
3739                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3740                                         &RMF_EADATA, RCL_CLIENT);
3741                         /*
3742                          * If it's a replay of layout write intent RPC, the
3743                          * client has saved the extended lovea when
3744                          * it get reply then.
3745                          */
3746                         if (buf->lb_len > 0)
3747                                 mdt_fix_lov_magic(info, buf->lb_buf);
3748                 }
3749
3750                 /*
3751                  * Instantiate some layout components, if @buf contains
3752                  * lovea, then it's a replay of the layout intent write
3753                  * RPC.
3754                  */
3755                 rc = mdt_layout_change(info, obj, layout, buf);
3756                 if (rc)
3757                         GOTO(out_obj, rc);
3758         }
3759 out_obj:
3760         mdt_object_put(info->mti_env, obj);
3761
3762         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3763                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3764
3765 out:
3766         lhc->mlh_reg_lh.cookie = 0;
3767
3768         return rc;
3769 }
3770
3771 static int mdt_intent_reint(enum mdt_it_code opcode,
3772                             struct mdt_thread_info *info,
3773                             struct ldlm_lock **lockp,
3774                             __u64 flags)
3775 {
3776         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3777         struct ldlm_reply      *rep = NULL;
3778         long                    opc;
3779         int                     rc;
3780
3781         static const struct req_format *intent_fmts[REINT_MAX] = {
3782                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3783                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3784         };
3785
3786         ENTRY;
3787
3788         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3789         if (opc < 0)
3790                 RETURN(opc);
3791
3792         if (mdt_it_flavor[opcode].it_reint != opc) {
3793                 CERROR("Reint code %ld doesn't match intent: %d\n",
3794                        opc, opcode);
3795                 RETURN(err_serious(-EPROTO));
3796         }
3797
3798         /* Get lock from request for possible resent case. */
3799         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3800
3801         rc = mdt_reint_internal(info, lhc, opc);
3802
3803         /* Check whether the reply has been packed successfully. */
3804         if (mdt_info_req(info)->rq_repmsg != NULL)
3805                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3806         if (rep == NULL)
3807                 RETURN(err_serious(-EFAULT));
3808
3809         /* MDC expects this in any case */
3810         if (rc != 0)
3811                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3812
3813         /* the open lock or the lock for cross-ref object should be
3814          * returned to the client */
3815         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3816             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3817                 rep->lock_policy_res2 = 0;
3818                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3819                 RETURN(rc);
3820         }
3821
3822         rep->lock_policy_res2 = clear_serious(rc);
3823
3824         if (rep->lock_policy_res2 == -ENOENT &&
3825             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3826             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3827                 rep->lock_policy_res2 = 0;
3828
3829         lhc->mlh_reg_lh.cookie = 0ull;
3830         if (rc == -ENOTCONN || rc == -ENODEV ||
3831             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3832                 /*
3833                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3834                  * will be returned by rq_status, and client at ptlrpc layer
3835                  * will detect this, then disconnect, reconnect the import
3836                  * immediately, instead of impacting the following the rpc.
3837                  */
3838                 RETURN(rc);
3839         }
3840         /*
3841          * For other cases, the error will be returned by intent, and client
3842          * will retrieve the result from intent.
3843          */
3844         RETURN(ELDLM_LOCK_ABORTED);
3845 }
3846
3847 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3848 {
3849         int rc;
3850
3851         switch (itcode) {
3852         case IT_OPEN:
3853                 rc = MDT_IT_OPEN;
3854                 break;
3855         case IT_OPEN|IT_CREAT:
3856                 rc = MDT_IT_OCREAT;
3857                 break;
3858         case IT_CREAT:
3859                 rc = MDT_IT_CREATE;
3860                 break;
3861         case IT_READDIR:
3862                 rc = MDT_IT_READDIR;
3863                 break;
3864         case IT_GETATTR:
3865                 rc = MDT_IT_GETATTR;
3866                 break;
3867         case IT_LOOKUP:
3868                 rc = MDT_IT_LOOKUP;
3869                 break;
3870         case IT_UNLINK:
3871                 rc = MDT_IT_UNLINK;
3872                 break;
3873         case IT_TRUNC:
3874                 rc = MDT_IT_TRUNC;
3875                 break;
3876         case IT_GETXATTR:
3877                 rc = MDT_IT_GETXATTR;
3878                 break;
3879         case IT_LAYOUT:
3880                 rc = MDT_IT_LAYOUT;
3881                 break;
3882         case IT_QUOTA_DQACQ:
3883         case IT_QUOTA_CONN:
3884                 rc = MDT_IT_QUOTA;
3885                 break;
3886         default:
3887                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3888                 rc = -EINVAL;
3889                 break;
3890         }
3891         return rc;
3892 }
3893
3894 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3895                           struct mdt_thread_info *info,
3896                           struct ldlm_lock **lockp, __u64 flags)
3897 {
3898         struct req_capsule      *pill = info->mti_pill;
3899         struct ptlrpc_request   *req = mdt_info_req(info);
3900         struct mdt_it_flavor    *flv;
3901         int opc;
3902         int rc;
3903         ENTRY;
3904
3905         opc = mdt_intent_code(itopc);
3906         if (opc < 0)
3907                 RETURN(-EINVAL);
3908
3909         if (opc == MDT_IT_QUOTA) {
3910                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3911
3912                 if (qmt == NULL)
3913                         RETURN(-EOPNOTSUPP);
3914
3915                 if (mdt_rdonly(req->rq_export))
3916                         RETURN(-EROFS);
3917
3918                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3919                 /* pass the request to quota master */
3920                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3921                                                  mdt_info_req(info), lockp,
3922                                                  flags);
3923                 RETURN(rc);
3924         }
3925
3926         flv = &mdt_it_flavor[opc];
3927         if (flv->it_fmt != NULL)
3928                 req_capsule_extend(pill, flv->it_fmt);
3929
3930         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3931         if (rc < 0)
3932                 RETURN(rc);
3933
3934         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3935                 RETURN(-EROFS);
3936
3937         if (flv->it_act != NULL) {
3938                 struct ldlm_reply *rep;
3939
3940                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3941
3942                 /* execute policy */
3943                 rc = flv->it_act(opc, info, lockp, flags);
3944
3945                 /* Check whether the reply has been packed successfully. */
3946                 if (req->rq_repmsg != NULL) {
3947                         rep = req_capsule_server_get(info->mti_pill,
3948                                                      &RMF_DLM_REP);
3949                         rep->lock_policy_res2 =
3950                                 ptlrpc_status_hton(rep->lock_policy_res2);
3951                 }
3952         }
3953
3954         RETURN(rc);
3955 }
3956
3957 static int mdt_intent_policy(struct ldlm_namespace *ns,
3958                              struct ldlm_lock **lockp, void *req_cookie,
3959                              enum ldlm_mode mode, __u64 flags, void *data)
3960 {
3961         struct tgt_session_info *tsi;
3962         struct mdt_thread_info  *info;
3963         struct ptlrpc_request   *req  =  req_cookie;
3964         struct ldlm_intent      *it;
3965         struct req_capsule      *pill;
3966         int rc;
3967
3968         ENTRY;
3969
3970         LASSERT(req != NULL);
3971
3972         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3973
3974         info = tsi2mdt_info(tsi);
3975         LASSERT(info != NULL);
3976         pill = info->mti_pill;
3977         LASSERT(pill->rc_req == req);
3978
3979         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3980                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3981                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3982                 if (it != NULL) {
3983                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3984                         if (rc == 0)
3985                                 rc = ELDLM_OK;
3986
3987                         /* Lock without inodebits makes no sense and will oops
3988                          * later in ldlm. Let's check it now to see if we have
3989                          * ibits corrupted somewhere in mdt_intent_opc().
3990                          * The case for client miss to set ibits has been
3991                          * processed by others. */
3992                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3993                                         lr_type == LDLM_IBITS,
3994                                      info->mti_dlm_req->lock_desc.\
3995                                         l_policy_data.l_inodebits.bits != 0));
3996                 } else
3997                         rc = err_serious(-EFAULT);
3998         } else {
3999                 /* No intent was provided */
4000                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
4001                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4002                 rc = req_capsule_server_pack(pill);
4003                 if (rc)
4004                         rc = err_serious(rc);
4005         }
4006         mdt_thread_info_fini(info);
4007         RETURN(rc);
4008 }
4009
4010 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4011 {
4012         struct seq_server_site  *ss = mdt_seq_site(mdt);
4013
4014         if (ss->ss_node_id == 0)
4015                 return;
4016
4017         if (ss->ss_client_seq != NULL) {
4018                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4019                 ss->ss_client_seq->lcs_exp = NULL;
4020         }
4021
4022         if (ss->ss_server_fld != NULL) {
4023                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4024                 ss->ss_server_fld->lsf_control_exp = NULL;
4025         }
4026 }
4027
4028 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4029 {
4030         struct seq_server_site *ss = mdt_seq_site(mdt);
4031
4032         if (ss == NULL)
4033                 return;
4034
4035         if (ss->ss_server_seq != NULL)
4036                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4037 }
4038
4039 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4040 {
4041         mdt_seq_fini_cli(mdt);
4042         mdt_deregister_seq_exp(mdt);
4043
4044         return seq_site_fini(env, mdt_seq_site(mdt));
4045 }
4046
4047 /**
4048  * It will retrieve its FLDB entries from MDT0, and it only happens
4049  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4050  * and it needs to refresh FLDB from the MDT0.
4051  **/
4052 static int mdt_register_lwp_callback(void *data)
4053 {
4054         struct lu_env           env;
4055         struct mdt_device       *mdt = data;
4056         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4057         int                     rc;
4058         ENTRY;
4059
4060         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4061
4062         rc = lu_env_init(&env, LCT_MD_THREAD);
4063         if (rc < 0) {
4064                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4065                 RETURN(rc);
4066         }
4067
4068         /* Allocate new sequence now to avoid creating local transaction
4069          * in the normal transaction process */
4070         rc = seq_server_check_and_alloc_super(&env,
4071                                               mdt_seq_site(mdt)->ss_server_seq);
4072         if (rc < 0)
4073                 GOTO(out, rc);
4074
4075         if (fld->lsf_new) {
4076                 rc = fld_update_from_controller(&env, fld);
4077                 if (rc != 0) {
4078                         CERROR("%s: cannot update controller: rc = %d\n",
4079                                mdt_obd_name(mdt), rc);
4080                         GOTO(out, rc);
4081                 }
4082         }
4083 out:
4084         lu_env_fini(&env);
4085         RETURN(rc);
4086 }
4087
4088 static int mdt_register_seq_exp(struct mdt_device *mdt)
4089 {
4090         struct seq_server_site  *ss = mdt_seq_site(mdt);
4091         char                    *lwp_name = NULL;
4092         int                     rc;
4093
4094         if (ss->ss_node_id == 0)
4095                 return 0;
4096
4097         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4098         if (lwp_name == NULL)
4099                 GOTO(out_free, rc = -ENOMEM);
4100
4101         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4102         if (rc != 0)
4103                 GOTO(out_free, rc);
4104
4105         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4106                                       NULL, NULL);
4107         if (rc != 0)
4108                 GOTO(out_free, rc);
4109
4110         rc = lustre_register_lwp_item(lwp_name,
4111                                       &ss->ss_server_fld->lsf_control_exp,
4112                                       mdt_register_lwp_callback, mdt);
4113         if (rc != 0) {
4114                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4115                 ss->ss_client_seq->lcs_exp = NULL;
4116                 GOTO(out_free, rc);
4117         }
4118 out_free:
4119         if (lwp_name != NULL)
4120                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4121
4122         return rc;
4123 }
4124
4125 /*
4126  * Init client sequence manager which is used by local MDS to talk to sequence
4127  * controller on remote node.
4128  */
4129 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4130 {
4131         struct seq_server_site  *ss = mdt_seq_site(mdt);
4132         int                     rc;
4133         char                    *prefix;
4134         ENTRY;
4135
4136         /* check if this is adding the first MDC and controller is not yet
4137          * initialized. */
4138         OBD_ALLOC_PTR(ss->ss_client_seq);
4139         if (ss->ss_client_seq == NULL)
4140                 RETURN(-ENOMEM);
4141
4142         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4143         if (prefix == NULL) {
4144                 OBD_FREE_PTR(ss->ss_client_seq);
4145                 ss->ss_client_seq = NULL;
4146                 RETURN(-ENOMEM);
4147         }
4148
4149         /* Note: seq_client_fini will be called in seq_site_fini */
4150         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4151         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4152                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4153                                                             NULL);
4154         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4155         if (rc != 0) {
4156                 OBD_FREE_PTR(ss->ss_client_seq);
4157                 ss->ss_client_seq = NULL;
4158                 RETURN(rc);
4159         }
4160
4161         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4162
4163         RETURN(rc);
4164 }
4165
4166 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4167 {
4168         struct seq_server_site  *ss;
4169         int                     rc;
4170         ENTRY;
4171
4172         ss = mdt_seq_site(mdt);
4173         /* init sequence controller server(MDT0) */
4174         if (ss->ss_node_id == 0) {
4175                 OBD_ALLOC_PTR(ss->ss_control_seq);
4176                 if (ss->ss_control_seq == NULL)
4177                         RETURN(-ENOMEM);
4178
4179                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4180                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4181                                      ss);
4182                 if (rc)
4183                         GOTO(out_seq_fini, rc);
4184         }
4185
4186         /* Init normal sequence server */
4187         OBD_ALLOC_PTR(ss->ss_server_seq);
4188         if (ss->ss_server_seq == NULL)
4189                 GOTO(out_seq_fini, rc = -ENOMEM);
4190
4191         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4192                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4193         if (rc)
4194                 GOTO(out_seq_fini, rc);
4195
4196         /* init seq client for seq server to talk to seq controller(MDT0) */
4197         rc = mdt_seq_init_cli(env, mdt);
4198         if (rc != 0)
4199                 GOTO(out_seq_fini, rc);
4200
4201         if (ss->ss_node_id != 0)
4202                 /* register controller export through lwp */
4203                 rc = mdt_register_seq_exp(mdt);
4204
4205         EXIT;
4206 out_seq_fini:
4207         if (rc)
4208                 mdt_seq_fini(env, mdt);
4209
4210         return rc;
4211 }
4212
4213 /*
4214  * FLD wrappers
4215  */
4216 static int mdt_fld_fini(const struct lu_env *env,
4217                         struct mdt_device *m)
4218 {
4219         struct seq_server_site *ss = mdt_seq_site(m);
4220         ENTRY;
4221
4222         if (ss && ss->ss_server_fld) {
4223                 fld_server_fini(env, ss->ss_server_fld);
4224                 OBD_FREE_PTR(ss->ss_server_fld);
4225                 ss->ss_server_fld = NULL;
4226         }
4227
4228         RETURN(0);
4229 }
4230
4231 static int mdt_fld_init(const struct lu_env *env,
4232                         const char *uuid,
4233                         struct mdt_device *m)
4234 {
4235         struct seq_server_site *ss;
4236         int rc;
4237         ENTRY;
4238
4239         ss = mdt_seq_site(m);
4240
4241         OBD_ALLOC_PTR(ss->ss_server_fld);
4242         if (ss->ss_server_fld == NULL)
4243                 RETURN(rc = -ENOMEM);
4244
4245         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4246                              LU_SEQ_RANGE_MDT);
4247         if (rc) {
4248                 OBD_FREE_PTR(ss->ss_server_fld);
4249                 ss->ss_server_fld = NULL;
4250                 RETURN(rc);
4251         }
4252
4253         RETURN(0);
4254 }
4255
4256 static void mdt_stack_pre_fini(const struct lu_env *env,
4257                            struct mdt_device *m, struct lu_device *top)
4258 {
4259         struct lustre_cfg_bufs  *bufs;
4260         struct lustre_cfg       *lcfg;
4261         struct mdt_thread_info  *info;
4262         ENTRY;
4263
4264         LASSERT(top);
4265
4266         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4267         LASSERT(info != NULL);
4268
4269         bufs = &info->mti_u.bufs;
4270
4271         LASSERT(m->mdt_child_exp);
4272         LASSERT(m->mdt_child_exp->exp_obd);
4273
4274         /* process cleanup, pass mdt obd name to get obd umount flags */
4275         /* XXX: this is needed because all layers are referenced by
4276          * objects (some of them are pinned by osd, for example *
4277          * the proper solution should be a model where object used
4278          * by osd only doesn't have mdt/mdd slices -bzzz */
4279         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4280         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4281         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4282         if (!lcfg)
4283                 RETURN_EXIT;
4284         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4285
4286         top->ld_ops->ldo_process_config(env, top, lcfg);
4287         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4288         EXIT;
4289 }
4290
4291 static void mdt_stack_fini(const struct lu_env *env,
4292                            struct mdt_device *m, struct lu_device *top)
4293 {
4294         struct obd_device       *obd = mdt2obd_dev(m);
4295         struct lustre_cfg_bufs  *bufs;
4296         struct lustre_cfg       *lcfg;
4297         struct mdt_thread_info  *info;
4298         char                     flags[3] = "";
4299         ENTRY;
4300
4301         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4302         LASSERT(info != NULL);
4303
4304         lu_dev_del_linkage(top->ld_site, top);
4305
4306         lu_site_purge(env, top->ld_site, -1);
4307
4308         bufs = &info->mti_u.bufs;
4309         /* process cleanup, pass mdt obd name to get obd umount flags */
4310         /* another purpose is to let all layers to release their objects */
4311         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4312         if (obd->obd_force)
4313                 strcat(flags, "F");
4314         if (obd->obd_fail)
4315                 strcat(flags, "A");
4316         lustre_cfg_bufs_set_string(bufs, 1, flags);
4317         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4318         if (!lcfg)
4319                 RETURN_EXIT;
4320         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4321
4322         LASSERT(top);
4323         top->ld_ops->ldo_process_config(env, top, lcfg);
4324         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4325
4326         lu_site_purge(env, top->ld_site, -1);
4327
4328         m->mdt_child = NULL;
4329         m->mdt_bottom = NULL;
4330
4331         obd_disconnect(m->mdt_child_exp);
4332         m->mdt_child_exp = NULL;
4333
4334         obd_disconnect(m->mdt_bottom_exp);
4335         m->mdt_child_exp = NULL;
4336 }
4337
4338 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4339                                const char *next, struct obd_export **exp)
4340 {
4341         struct obd_connect_data *data = NULL;
4342         struct obd_device       *obd;
4343         int                      rc;
4344         ENTRY;
4345
4346         OBD_ALLOC_PTR(data);
4347         if (data == NULL)
4348                 GOTO(out, rc = -ENOMEM);
4349
4350         obd = class_name2obd(next);
4351         if (obd == NULL) {
4352                 CERROR("%s: can't locate next device: %s\n",
4353                        mdt_obd_name(m), next);
4354                 GOTO(out, rc = -ENOTCONN);
4355         }
4356
4357         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4358         data->ocd_version = LUSTRE_VERSION_CODE;
4359
4360         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4361         if (rc) {
4362                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4363                        mdt_obd_name(m), next, rc);
4364                 GOTO(out, rc);
4365         }
4366
4367 out:
4368         if (data)
4369                 OBD_FREE_PTR(data);
4370         RETURN(rc);
4371 }
4372
4373 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4374                           struct lustre_cfg *cfg)
4375 {
4376         char                   *dev = lustre_cfg_string(cfg, 0);
4377         int                     rc, name_size, uuid_size;
4378         char                   *name, *uuid, *p;
4379         struct lustre_cfg_bufs *bufs;
4380         struct lustre_cfg      *lcfg;
4381         struct obd_device      *obd;
4382         struct lustre_profile  *lprof;
4383         struct lu_site         *site;
4384         ENTRY;
4385
4386         /* in 1.8 we had the only device in the stack - MDS.
4387          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4388          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4389          * to generate names and setup MDT, MDD. MDT will be using
4390          * generated name to connect to MDD. for MDD the next device
4391          * will be LOD with name taken from so called "profile" which
4392          * is generated by mount_option line
4393          *
4394          * 1.8 MGS generates config. commands like this:
4395          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4396          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4397          * 2.0 MGS generates config. commands like this:
4398          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4399          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4400          *                    3:lustre-MDT0000-mdtlov  4:f
4401          *
4402          * we generate MDD name from MDT one, just replacing T with D
4403          *
4404          * after all the preparations, the logical equivalent will be
4405          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4406          *                    3:lustre-MDT0000-mdtlov  4:f
4407          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4408          *                    3:lustre-MDD0000  4:f
4409          *
4410          *  notice we build the stack from down to top: MDD first, then MDT */
4411
4412         name_size = MAX_OBD_NAME;
4413         uuid_size = MAX_OBD_NAME;
4414
4415         OBD_ALLOC(name, name_size);
4416         OBD_ALLOC(uuid, uuid_size);
4417         if (name == NULL || uuid == NULL)
4418                 GOTO(cleanup_mem, rc = -ENOMEM);
4419
4420         OBD_ALLOC_PTR(bufs);
4421         if (!bufs)
4422                 GOTO(cleanup_mem, rc = -ENOMEM);
4423
4424         strcpy(name, dev);
4425         p = strstr(name, "-MDT");
4426         if (p == NULL)
4427                 GOTO(free_bufs, rc = -ENOMEM);
4428         p[3] = 'D';
4429
4430         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4431
4432         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4433         if (lprof == NULL || lprof->lp_dt == NULL) {
4434                 CERROR("can't find the profile: %s\n",
4435                        lustre_cfg_string(cfg, 0));
4436                 GOTO(free_bufs, rc = -EINVAL);
4437         }
4438
4439         lustre_cfg_bufs_reset(bufs, name);
4440         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4441         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4442         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4443
4444         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4445         if (!lcfg)
4446                 GOTO(put_profile, rc = -ENOMEM);
4447         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4448
4449         rc = class_attach(lcfg);
4450         if (rc)
4451                 GOTO(lcfg_cleanup, rc);
4452
4453         obd = class_name2obd(name);
4454         if (!obd) {
4455                 CERROR("Can not find obd %s (%s in config)\n",
4456                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4457                 GOTO(lcfg_cleanup, rc = -EINVAL);
4458         }
4459
4460         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4461
4462         lustre_cfg_bufs_reset(bufs, name);
4463         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4464         lustre_cfg_bufs_set_string(bufs, 2, dev);
4465         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4466
4467         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4468         if (!lcfg)
4469                 GOTO(class_detach, rc = -ENOMEM);
4470         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4471
4472         rc = class_setup(obd, lcfg);
4473         if (rc)
4474                 GOTO(class_detach, rc);
4475
4476         /* connect to MDD we just setup */
4477         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4478         if (rc)
4479                 GOTO(class_detach, rc);
4480
4481         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4482         LASSERT(site);
4483         LASSERT(mdt_lu_site(mdt) == NULL);
4484         mdt->mdt_lu_dev.ld_site = site;
4485         site->ls_top_dev = &mdt->mdt_lu_dev;
4486         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4487
4488         /* now connect to bottom OSD */
4489         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4490         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4491         if (rc)
4492                 GOTO(class_detach, rc);
4493         mdt->mdt_bottom =
4494                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4495
4496         rc = lu_env_refill((struct lu_env *)env);
4497         if (rc != 0)
4498                 CERROR("Failure to refill session: '%d'\n", rc);
4499
4500         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4501
4502         EXIT;
4503 class_detach:
4504         if (rc)
4505                 class_detach(obd, lcfg);
4506 lcfg_cleanup:
4507         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4508 put_profile:
4509         class_put_profile(lprof);
4510 free_bufs:
4511         OBD_FREE_PTR(bufs);
4512 cleanup_mem:
4513         if (name)
4514                 OBD_FREE(name, name_size);
4515         if (uuid)
4516                 OBD_FREE(uuid, uuid_size);
4517         RETURN(rc);
4518 }
4519
4520 /* setup quota master target on MDT0 */
4521 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4522                           struct lustre_cfg *cfg)
4523 {
4524         struct obd_device       *obd;
4525         char                    *dev = lustre_cfg_string(cfg, 0);
4526         char                    *qmtname, *uuid, *p;
4527         struct lustre_cfg_bufs  *bufs;
4528         struct lustre_cfg       *lcfg;
4529         struct lustre_profile   *lprof;
4530         struct obd_connect_data *data;
4531         int                      rc;
4532         ENTRY;
4533
4534         LASSERT(mdt->mdt_qmt_exp == NULL);
4535         LASSERT(mdt->mdt_qmt_dev == NULL);
4536
4537         /* quota master is on MDT0 only for now */
4538         if (mdt->mdt_seq_site.ss_node_id != 0)
4539                 RETURN(0);
4540
4541         /* MGS generates config commands which look as follows:
4542          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4543          *                    3:lustre-MDT0000-mdtlov  4:f
4544          *
4545          * We generate the QMT name from the MDT one, just replacing MD with QM
4546          * after all the preparations, the logical equivalent will be:
4547          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4548          *                    3:lustre-MDT0000-osd  4:f */
4549         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4550         OBD_ALLOC(uuid, UUID_MAX);
4551         OBD_ALLOC_PTR(bufs);
4552         OBD_ALLOC_PTR(data);
4553         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4554                 GOTO(cleanup_mem, rc = -ENOMEM);
4555
4556         strcpy(qmtname, dev);
4557         p = strstr(qmtname, "-MDT");
4558         if (p == NULL)
4559                 GOTO(cleanup_mem, rc = -ENOMEM);
4560         /* replace MD with QM */
4561         p[1] = 'Q';
4562         p[2] = 'M';
4563
4564         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4565
4566         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4567         if (lprof == NULL || lprof->lp_dt == NULL) {
4568                 CERROR("can't find profile for %s\n",
4569                        lustre_cfg_string(cfg, 0));
4570                 GOTO(cleanup_mem, rc = -EINVAL);
4571         }
4572
4573         lustre_cfg_bufs_reset(bufs, qmtname);
4574         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4575         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4576         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4577
4578         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4579         if (!lcfg)
4580                 GOTO(put_profile, rc = -ENOMEM);
4581         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4582
4583         rc = class_attach(lcfg);
4584         if (rc)
4585                 GOTO(lcfg_cleanup, rc);
4586
4587         obd = class_name2obd(qmtname);
4588         if (!obd) {
4589                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4590                        lustre_cfg_string(cfg, 0));
4591                 GOTO(lcfg_cleanup, rc = -EINVAL);
4592         }
4593
4594         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4595
4596         lustre_cfg_bufs_reset(bufs, qmtname);
4597         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4598         lustre_cfg_bufs_set_string(bufs, 2, dev);
4599
4600         /* for quota, the next device should be the OSD device */
4601         lustre_cfg_bufs_set_string(bufs, 3,
4602                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4603
4604         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4605         if (!lcfg)
4606                 GOTO(class_detach, rc = -ENOMEM);
4607         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4608
4609         rc = class_setup(obd, lcfg);
4610         if (rc)
4611                 GOTO(class_detach, rc);
4612
4613         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4614
4615         /* configure local quota objects */
4616         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4617                                                    &mdt->mdt_lu_dev,
4618                                                    mdt->mdt_qmt_dev);
4619         if (rc)
4620                 GOTO(class_cleanup, rc);
4621
4622         /* connect to quota master target */
4623         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4624         data->ocd_version = LUSTRE_VERSION_CODE;
4625         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4626                          data, NULL);
4627         if (rc) {
4628                 CERROR("cannot connect to quota master device %s (%d)\n",
4629                        qmtname, rc);
4630                 GOTO(class_cleanup, rc);
4631         }
4632
4633         EXIT;
4634 class_cleanup:
4635         if (rc) {
4636                 class_manual_cleanup(obd);
4637                 mdt->mdt_qmt_dev = NULL;
4638         }
4639 class_detach:
4640         if (rc)
4641                 class_detach(obd, lcfg);
4642 lcfg_cleanup:
4643         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4644 put_profile:
4645         class_put_profile(lprof);
4646 cleanup_mem:
4647         if (bufs)
4648                 OBD_FREE_PTR(bufs);
4649         if (qmtname)
4650                 OBD_FREE(qmtname, MAX_OBD_NAME);
4651         if (uuid)
4652                 OBD_FREE(uuid, UUID_MAX);
4653         if (data)
4654                 OBD_FREE_PTR(data);
4655         return rc;
4656 }
4657
4658 /* Shutdown quota master target associated with mdt */
4659 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4660 {
4661         ENTRY;
4662
4663         if (mdt->mdt_qmt_exp == NULL)
4664                 RETURN_EXIT;
4665         LASSERT(mdt->mdt_qmt_dev != NULL);
4666
4667         /* the qmt automatically shuts down when the mdt disconnects */
4668         obd_disconnect(mdt->mdt_qmt_exp);
4669         mdt->mdt_qmt_exp = NULL;
4670         mdt->mdt_qmt_dev = NULL;
4671         EXIT;
4672 }
4673
4674 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4675  * for now. This will be removed along with converting rest of MDT code
4676  * to use tgt_session_info */
4677 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4678 {
4679         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4680         int                      rc;
4681
4682         rc = mdt_getxattr(info);
4683
4684         mdt_thread_info_fini(info);
4685         return rc;
4686 }
4687
4688 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4689 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4690 #define OST_BRW_READ    OST_READ
4691 #define OST_BRW_WRITE   OST_WRITE
4692
4693 static struct tgt_handler mdt_tgt_handlers[] = {
4694 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4695                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4696                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4697 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4698                 0,                      MDS_DISCONNECT, tgt_disconnect,
4699                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4700 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4701                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4702                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4703 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4704 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4705 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4706 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4707                                                         mdt_getattr_name),
4708 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4709 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4710 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4711 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4712 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4713 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4714 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4715 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4716                                                         mdt_hsm_progress),
4717 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4718                                                         mdt_hsm_ct_register),
4719 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4720                                                         mdt_hsm_ct_unregister),
4721 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4722                                                         mdt_hsm_state_get),
4723 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4724                                                         mdt_hsm_state_set),
4725 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4726 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4727                                                         mdt_hsm_request),
4728 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4729             MDS_SWAP_LAYOUTS,
4730             mdt_swap_layouts),
4731 };
4732
4733 static struct tgt_handler mdt_io_ops[] = {
4734 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ,  tgt_brw_read),
4735 TGT_OST_HDL(HABEO_CORPUS | MUTABOR,      OST_BRW_WRITE, tgt_brw_write),
4736 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4737                                          OST_PUNCH,     mdt_punch_hdl),
4738 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4739 };
4740
4741 static struct tgt_handler mdt_sec_ctx_ops[] = {
4742 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4743 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4744 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4745 };
4746
4747 static struct tgt_handler mdt_quota_ops[] = {
4748 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4749 };
4750
4751 static struct tgt_opc_slice mdt_common_slice[] = {
4752         {
4753                 .tos_opc_start  = MDS_FIRST_OPC,
4754                 .tos_opc_end    = MDS_LAST_OPC,
4755                 .tos_hs         = mdt_tgt_handlers
4756         },
4757         {
4758                 .tos_opc_start  = OBD_FIRST_OPC,
4759                 .tos_opc_end    = OBD_LAST_OPC,
4760                 .tos_hs         = tgt_obd_handlers
4761         },
4762         {
4763                 .tos_opc_start  = LDLM_FIRST_OPC,
4764                 .tos_opc_end    = LDLM_LAST_OPC,
4765                 .tos_hs         = tgt_dlm_handlers
4766         },
4767         {
4768                 .tos_opc_start  = SEC_FIRST_OPC,
4769                 .tos_opc_end    = SEC_LAST_OPC,
4770                 .tos_hs         = mdt_sec_ctx_ops
4771         },
4772         {
4773                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4774                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4775                 .tos_hs         = tgt_out_handlers
4776         },
4777         {
4778                 .tos_opc_start  = FLD_FIRST_OPC,
4779                 .tos_opc_end    = FLD_LAST_OPC,
4780                 .tos_hs         = fld_handlers
4781         },
4782         {
4783                 .tos_opc_start  = SEQ_FIRST_OPC,
4784                 .tos_opc_end    = SEQ_LAST_OPC,
4785                 .tos_hs         = seq_handlers
4786         },
4787         {
4788                 .tos_opc_start  = QUOTA_DQACQ,
4789                 .tos_opc_end    = QUOTA_LAST_OPC,
4790                 .tos_hs         = mdt_quota_ops
4791         },
4792         {
4793                 .tos_opc_start  = LLOG_FIRST_OPC,
4794                 .tos_opc_end    = LLOG_LAST_OPC,
4795                 .tos_hs         = tgt_llog_handlers
4796         },
4797         {
4798                 .tos_opc_start  = LFSCK_FIRST_OPC,
4799                 .tos_opc_end    = LFSCK_LAST_OPC,
4800                 .tos_hs         = tgt_lfsck_handlers
4801         },
4802         {
4803                 .tos_opc_start  = OST_FIRST_OPC,
4804                 .tos_opc_end    = OST_LAST_OPC,
4805                 .tos_hs         = mdt_io_ops
4806         },
4807         {
4808                 .tos_hs         = NULL
4809         }
4810 };
4811
4812 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4813 {
4814         struct md_device *next = m->mdt_child;
4815         struct lu_device *d = &m->mdt_lu_dev;
4816         struct obd_device *obd = mdt2obd_dev(m);
4817         struct lfsck_stop stop;
4818
4819         ENTRY;
4820         stop.ls_status = LS_PAUSED;
4821         stop.ls_flags = 0;
4822         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4823
4824         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4825         ping_evictor_stop();
4826
4827         /* Remove the HSM /proc entry so the coordinator cannot be
4828          * restarted by a user while it's shutting down. */
4829         hsm_cdt_procfs_fini(m);
4830         mdt_hsm_cdt_stop(m);
4831
4832         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4833         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4834
4835         if (m->mdt_namespace != NULL)
4836                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4837                                           d->ld_obd->obd_force);
4838
4839         obd_exports_barrier(obd);
4840         obd_zombie_barrier();
4841
4842         mdt_quota_fini(env, m);
4843
4844         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4845
4846         /* Calling the cleanup functions in the same order as in the mdt_init0
4847          * error path
4848          */
4849         mdt_procfs_fini(m);
4850
4851         target_recovery_fini(obd);
4852         upcall_cache_cleanup(m->mdt_identity_cache);
4853         m->mdt_identity_cache = NULL;
4854
4855         mdt_fs_cleanup(env, m);
4856
4857         tgt_fini(env, &m->mdt_lut);
4858
4859         mdt_hsm_cdt_fini(m);
4860
4861         if (m->mdt_namespace != NULL) {
4862                 ldlm_namespace_free_post(m->mdt_namespace);
4863                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4864         }
4865
4866         if (m->mdt_md_root != NULL) {
4867                 mdt_object_put(env, m->mdt_md_root);
4868                 m->mdt_md_root = NULL;
4869         }
4870
4871         mdt_seq_fini(env, m);
4872
4873         mdt_fld_fini(env, m);
4874
4875         /*
4876          * Finish the stack
4877          */
4878         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4879
4880         LASSERT(atomic_read(&d->ld_ref) == 0);
4881
4882         server_put_mount(mdt_obd_name(m), true);
4883
4884         EXIT;
4885 }
4886
4887 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4888
4889 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4890                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4891 {
4892         struct mdt_thread_info    *info;
4893         struct obd_device         *obd;
4894         const char                *dev = lustre_cfg_string(cfg, 0);
4895         const char                *num = lustre_cfg_string(cfg, 2);
4896         struct lustre_mount_info  *lmi = NULL;
4897         struct lustre_sb_info     *lsi;
4898         struct lu_site            *s;
4899         struct seq_server_site    *ss_site;
4900         const char                *identity_upcall = "NONE";
4901         struct md_device          *next;
4902         int                        rc;
4903         long                       node_id;
4904         mntopt_t                   mntopts;
4905         ENTRY;
4906
4907         lu_device_init(&m->mdt_lu_dev, ldt);
4908         /*
4909          * Environment (env) might be missing mdt_thread_key values at that
4910          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4911          * mode.
4912          *
4913          * Usually device allocation path doesn't use module key values, but
4914          * mdt has to do a lot of work here, so allocate key value.
4915          */
4916         rc = lu_env_refill((struct lu_env *)env);
4917         if (rc != 0)
4918                 RETURN(rc);
4919
4920         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4921         LASSERT(info != NULL);
4922
4923         obd = class_name2obd(dev);
4924         LASSERT(obd != NULL);
4925
4926         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4927         m->mdt_opts.mo_evict_tgt_nids = 1;
4928         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4929
4930         lmi = server_get_mount(dev);
4931         if (lmi == NULL) {
4932                 CERROR("Cannot get mount info for %s!\n", dev);
4933                 RETURN(-EFAULT);
4934         } else {
4935                 lsi = s2lsi(lmi->lmi_sb);
4936                 /* CMD is supported only in IAM mode */
4937                 LASSERT(num);
4938                 node_id = simple_strtol(num, NULL, 10);
4939                 obd->u.obt.obt_magic = OBT_MAGIC;
4940                 if (lsi->lsi_lmd != NULL &&
4941                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4942                         m->mdt_skip_lfsck = 1;
4943         }
4944
4945         m->mdt_squash.rsi_uid = 0;
4946         m->mdt_squash.rsi_gid = 0;
4947         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4948         init_rwsem(&m->mdt_squash.rsi_sem);
4949         spin_lock_init(&m->mdt_lock);
4950         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4951         m->mdt_enable_remote_dir = 0;
4952         m->mdt_enable_remote_dir_gid = 0;
4953
4954         atomic_set(&m->mdt_mds_mds_conns, 0);
4955         atomic_set(&m->mdt_async_commit_count, 0);
4956
4957         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4958         m->mdt_lu_dev.ld_obd = obd;
4959         /* Set this lu_device to obd for error handling purposes. */
4960         obd->obd_lu_dev = &m->mdt_lu_dev;
4961
4962         /* init the stack */
4963         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4964         if (rc) {
4965                 CERROR("%s: Can't init device stack, rc %d\n",
4966                        mdt_obd_name(m), rc);
4967                 GOTO(err_lmi, rc);
4968         }
4969
4970         s = mdt_lu_site(m);
4971         ss_site = mdt_seq_site(m);
4972         s->ld_seq_site = ss_site;
4973         ss_site->ss_lu = s;
4974
4975         /* set server index */
4976         ss_site->ss_node_id = node_id;
4977
4978         /* failover is the default
4979          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4980          * assumed whose ss_node_id == 0 XXX
4981          * */
4982         obd->obd_replayable = 1;
4983         /* No connection accepted until configurations will finish */
4984         obd->obd_no_conn = 1;
4985
4986         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4987                 char *str = lustre_cfg_string(cfg, 4);
4988                 if (strchr(str, 'n')) {
4989                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4990                         obd->obd_replayable = 0;
4991                 }
4992         }
4993
4994         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4995         if (rc)
4996                 GOTO(err_fini_stack, rc);
4997
4998         rc = mdt_seq_init(env, m);
4999         if (rc)
5000                 GOTO(err_fini_fld, rc);
5001
5002         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5003                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5004         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5005                                               LDLM_NAMESPACE_SERVER,
5006                                               LDLM_NAMESPACE_GREEDY,
5007                                               LDLM_NS_TYPE_MDT);
5008         if (m->mdt_namespace == NULL)
5009                 GOTO(err_fini_seq, rc = -ENOMEM);
5010
5011         m->mdt_namespace->ns_lvbp = m;
5012         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5013
5014         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5015         /* set obd_namespace for compatibility with old code */
5016         obd->obd_namespace = m->mdt_namespace;
5017
5018         rc = mdt_hsm_cdt_init(m);
5019         if (rc != 0) {
5020                 CERROR("%s: error initializing coordinator, rc %d\n",
5021                        mdt_obd_name(m), rc);
5022                 GOTO(err_free_ns, rc);
5023         }
5024
5025         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5026                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5027                       OBD_FAIL_MDS_ALL_REPLY_NET);
5028         if (rc)
5029                 GOTO(err_free_hsm, rc);
5030
5031         rc = mdt_fs_setup(env, m, obd, lsi);
5032         if (rc)
5033                 GOTO(err_tgt, rc);
5034
5035         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5036
5037         next = m->mdt_child;
5038         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
5039                                          &mntopts);
5040         if (rc)
5041                 GOTO(err_fs_cleanup, rc);
5042
5043         if (mntopts & MNTOPT_USERXATTR)
5044                 m->mdt_opts.mo_user_xattr = 1;
5045         else
5046                 m->mdt_opts.mo_user_xattr = 0;
5047
5048         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
5049         if (rc)
5050                 GOTO(err_fs_cleanup, rc);
5051
5052         if (mntopts & MNTOPT_ACL)
5053                 m->mdt_opts.mo_acl = 1;
5054         else
5055                 m->mdt_opts.mo_acl = 0;
5056
5057         /* XXX: to support suppgid for ACL, we enable identity_upcall
5058          * by default, otherwise, maybe got unexpected -EACCESS. */
5059         if (m->mdt_opts.mo_acl)
5060                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5061
5062         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5063                                                 identity_upcall,
5064                                                 &mdt_identity_upcall_cache_ops);
5065         if (IS_ERR(m->mdt_identity_cache)) {
5066                 rc = PTR_ERR(m->mdt_identity_cache);
5067                 m->mdt_identity_cache = NULL;
5068                 GOTO(err_fs_cleanup, rc);
5069         }
5070
5071         rc = mdt_procfs_init(m, dev);
5072         if (rc) {
5073                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5074                 GOTO(err_recovery, rc);
5075         }
5076
5077         rc = mdt_quota_init(env, m, cfg);
5078         if (rc)
5079                 GOTO(err_procfs, rc);
5080
5081         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5082         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5083                            "mdt_ldlm_client", m->mdt_ldlm_client);
5084
5085         ping_evictor_start();
5086
5087         /* recovery will be started upon mdt_prepare()
5088          * when the whole stack is complete and ready
5089          * to serve the requests */
5090
5091         /* Reduce the initial timeout on an MDS because it doesn't need such
5092          * a long timeout as an OST does. Adaptive timeouts will adjust this
5093          * value appropriately. */
5094         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5095                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5096
5097         RETURN(0);
5098 err_procfs:
5099         mdt_procfs_fini(m);
5100 err_recovery:
5101         target_recovery_fini(obd);
5102         upcall_cache_cleanup(m->mdt_identity_cache);
5103         m->mdt_identity_cache = NULL;
5104 err_fs_cleanup:
5105         mdt_fs_cleanup(env, m);
5106 err_tgt:
5107         tgt_fini(env, &m->mdt_lut);
5108 err_free_hsm:
5109         mdt_hsm_cdt_fini(m);
5110 err_free_ns:
5111         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5112         obd->obd_namespace = m->mdt_namespace = NULL;
5113 err_fini_seq:
5114         mdt_seq_fini(env, m);
5115 err_fini_fld:
5116         mdt_fld_fini(env, m);
5117 err_fini_stack:
5118         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5119 err_lmi:
5120         if (lmi)
5121                 server_put_mount(dev, true);
5122         return(rc);
5123 }
5124
5125 /* For interoperability, the left element is old parameter, the right one
5126  * is the new version of the parameter, if some parameter is deprecated,
5127  * the new version should be set as NULL. */
5128 static struct cfg_interop_param mdt_interop_param[] = {
5129         { "mdt.group_upcall",   NULL },
5130         { "mdt.quota_type",     NULL },
5131         { "mdd.quota_type",     NULL },
5132         { "mdt.som",            NULL },
5133         { "mdt.rootsquash",     "mdt.root_squash" },
5134         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5135         { NULL }
5136 };
5137
5138 /* used by MGS to process specific configurations */
5139 static int mdt_process_config(const struct lu_env *env,
5140                               struct lu_device *d, struct lustre_cfg *cfg)
5141 {
5142         struct mdt_device *m = mdt_dev(d);
5143         struct md_device *md_next = m->mdt_child;
5144         struct lu_device *next = md2lu_dev(md_next);
5145         int rc;
5146         ENTRY;
5147
5148         switch (cfg->lcfg_command) {
5149         case LCFG_PARAM: {
5150                 struct obd_device          *obd = d->ld_obd;
5151
5152                 /* For interoperability */
5153                 struct cfg_interop_param   *ptr = NULL;
5154                 struct lustre_cfg          *old_cfg = NULL;
5155                 char                       *param = NULL;
5156
5157                 param = lustre_cfg_string(cfg, 1);
5158                 if (param == NULL) {
5159                         CERROR("param is empty\n");
5160                         rc = -EINVAL;
5161                         break;
5162                 }
5163
5164                 ptr = class_find_old_param(param, mdt_interop_param);
5165                 if (ptr != NULL) {
5166                         if (ptr->new_param == NULL) {
5167                                 rc = 0;
5168                                 CWARN("For interoperability, skip this %s."
5169                                       " It is obsolete.\n", ptr->old_param);
5170                                 break;
5171                         }
5172
5173                         CWARN("Found old param %s, changed it to %s.\n",
5174                               ptr->old_param, ptr->new_param);
5175
5176                         old_cfg = cfg;
5177                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5178                         if (IS_ERR(cfg)) {
5179                                 rc = PTR_ERR(cfg);
5180                                 break;
5181                         }
5182                 }
5183
5184                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5185                                               cfg, obd);
5186                 if (rc > 0 || rc == -ENOSYS) {
5187                         /* is it an HSM var ? */
5188                         rc = class_process_proc_param(PARAM_HSM,
5189                                                       hsm_cdt_get_proc_vars(),
5190                                                       cfg, obd);
5191                         if (rc > 0 || rc == -ENOSYS)
5192                                 /* we don't understand; pass it on */
5193                                 rc = next->ld_ops->ldo_process_config(env, next,
5194                                                                       cfg);
5195                 }
5196
5197                 if (old_cfg)
5198                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5199                                                      cfg->lcfg_buflens));
5200                 break;
5201         }
5202         default:
5203                 /* others are passed further */
5204                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5205                 break;
5206         }
5207         RETURN(rc);
5208 }
5209
5210 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5211                                           const struct lu_object_header *hdr,
5212                                           struct lu_device *d)
5213 {
5214         struct mdt_object *mo;
5215
5216         ENTRY;
5217
5218         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5219         if (mo != NULL) {
5220                 struct lu_object *o;
5221                 struct lu_object_header *h;
5222
5223                 o = &mo->mot_obj;
5224                 h = &mo->mot_header;
5225                 lu_object_header_init(h);
5226                 lu_object_init(o, h, d);
5227                 lu_object_add_top(h, o);
5228                 o->lo_ops = &mdt_obj_ops;
5229                 spin_lock_init(&mo->mot_write_lock);
5230                 mutex_init(&mo->mot_lov_mutex);
5231                 init_rwsem(&mo->mot_dom_sem);
5232                 init_rwsem(&mo->mot_open_sem);
5233                 atomic_set(&mo->mot_open_count, 0);
5234                 RETURN(o);
5235         }
5236         RETURN(NULL);
5237 }
5238
5239 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5240                            const struct lu_object_conf *unused)
5241 {
5242         struct mdt_device *d = mdt_dev(o->lo_dev);
5243         struct lu_device  *under;
5244         struct lu_object  *below;
5245         int                rc = 0;
5246         ENTRY;
5247
5248         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5249                PFID(lu_object_fid(o)));
5250
5251         under = &d->mdt_child->md_lu_dev;
5252         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5253         if (below != NULL) {
5254                 lu_object_add(o, below);
5255         } else
5256                 rc = -ENOMEM;
5257
5258         RETURN(rc);
5259 }
5260
5261 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5262 {
5263         struct mdt_object *mo = mdt_obj(o);
5264         struct lu_object_header *h;
5265         ENTRY;
5266
5267         h = o->lo_header;
5268         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5269                PFID(lu_object_fid(o)));
5270
5271         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5272         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5273
5274         lu_object_fini(o);
5275         lu_object_header_fini(h);
5276         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5277
5278         EXIT;
5279 }
5280
5281 static int mdt_object_print(const struct lu_env *env, void *cookie,
5282                             lu_printer_t p, const struct lu_object *o)
5283 {
5284         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5285
5286         return (*p)(env, cookie,
5287                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5288                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5289                     mdto->mot_cache_attr ? "cache_attr" : "",
5290                     mdto->mot_write_count);
5291 }
5292
5293 static int mdt_prepare(const struct lu_env *env,
5294                 struct lu_device *pdev,
5295                 struct lu_device *cdev)
5296 {
5297         struct mdt_device *mdt = mdt_dev(cdev);
5298         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5299         struct obd_device *obd = cdev->ld_obd;
5300         int rc;
5301
5302         ENTRY;
5303
5304         LASSERT(obd);
5305
5306         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5307         if (rc)
5308                 RETURN(rc);
5309
5310         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5311         if (rc)
5312                 RETURN(rc);
5313
5314         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5315         if (rc)
5316                 RETURN(rc);
5317
5318         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5319         /* The LFSCK instance is registered just now, so it must be there when
5320          * register the namespace to such instance. */
5321         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5322
5323         if (mdt->mdt_seq_site.ss_node_id == 0) {
5324                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5325                                                          &mdt->mdt_md_root_fid);
5326                 if (rc)
5327                         RETURN(rc);
5328         }
5329
5330         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5331
5332         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5333         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5334         LASSERT(obd->obd_no_conn);
5335         spin_lock(&obd->obd_dev_lock);
5336         obd->obd_no_conn = 0;
5337         spin_unlock(&obd->obd_dev_lock);
5338
5339         if (obd->obd_recovering == 0)
5340                 mdt_postrecov(env, mdt);
5341
5342         RETURN(rc);
5343 }
5344
5345 const struct lu_device_operations mdt_lu_ops = {
5346         .ldo_object_alloc   = mdt_object_alloc,
5347         .ldo_process_config = mdt_process_config,
5348         .ldo_prepare        = mdt_prepare,
5349 };
5350
5351 static const struct lu_object_operations mdt_obj_ops = {
5352         .loo_object_init    = mdt_object_init,
5353         .loo_object_free    = mdt_object_free,
5354         .loo_object_print   = mdt_object_print
5355 };
5356
5357 static int mdt_obd_set_info_async(const struct lu_env *env,
5358                                   struct obd_export *exp,
5359                                   __u32 keylen, void *key,
5360                                   __u32 vallen, void *val,
5361                                   struct ptlrpc_request_set *set)
5362 {
5363         int rc;
5364
5365         ENTRY;
5366
5367         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5368                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5369                 RETURN(rc);
5370         }
5371
5372         RETURN(0);
5373 }
5374
5375 /**
5376  * Match client and server connection feature flags.
5377  *
5378  * Compute the compatibility flags for a connection request based on
5379  * features mutually supported by client and server.
5380  *
5381  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5382  * must not be updated here, otherwise a partially initialized value may
5383  * be exposed. After the connection request is successfully processed,
5384  * the top-level MDT connect request handler atomically updates the export
5385  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5386  * reply. \see mdt_connect().
5387  *
5388  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5389  * full struct obd_connect_data. So care must be taken when accessing fields
5390  * that are not present in struct obd_connect_data_v1. See LU-16.
5391  *
5392  * \param exp   the obd_export associated with this client/target pair
5393  * \param mdt   the target device for the connection
5394  * \param data  stores data for this connect request
5395  *
5396  * \retval 0       success
5397  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5398  * \retval -EBADE  client and server feature requirements are incompatible
5399  */
5400 static int mdt_connect_internal(const struct lu_env *env,
5401                                 struct obd_export *exp,
5402                                 struct mdt_device *mdt,
5403                                 struct obd_connect_data *data, bool reconnect)
5404 {
5405         LASSERT(data != NULL);
5406
5407         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5408
5409         if (mdt->mdt_bottom->dd_rdonly &&
5410             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5411             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5412                 RETURN(-EACCES);
5413
5414         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5415                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5416
5417         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5418
5419         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5420             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5421                 CWARN("%s: client %s does not support ibits lock, either "
5422                       "very old or an invalid client: flags %#llx\n",
5423                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5424                       data->ocd_connect_flags);
5425                 return -EBADE;
5426         }
5427
5428         if (!mdt->mdt_opts.mo_acl)
5429                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5430
5431         if (!mdt->mdt_opts.mo_user_xattr)
5432                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5433
5434         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5435                 data->ocd_brw_size = min(data->ocd_brw_size, MD_MAX_BRW_SIZE);
5436                 if (data->ocd_brw_size == 0) {
5437                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5438                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5439                                "ocd_brw_size unexpectedly zero, network data "
5440                                "corruption? Refusing to connect this client\n",
5441                                mdt_obd_name(mdt),
5442                                exp->exp_client_uuid.uuid,
5443                                exp, data->ocd_connect_flags, data->ocd_version,
5444                                data->ocd_grant, data->ocd_index);
5445                         return -EPROTO;
5446                 }
5447         }
5448
5449         if (OCD_HAS_FLAG(data, GRANT))
5450                 data->ocd_grant = mdt_grant_connect(env, exp, data->ocd_grant,
5451                                                     !reconnect);
5452
5453         /* NB: Disregard the rule against updating
5454          * exp_connect_data.ocd_connect_flags in this case, since
5455          * tgt_client_new() needs to know if this is a lightweight
5456          * connection, and it is safe to expose this flag before
5457          * connection processing completes. */
5458         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5459                 spin_lock(&exp->exp_lock);
5460                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5461                 spin_unlock(&exp->exp_lock);
5462         }
5463
5464         data->ocd_version = LUSTRE_VERSION_CODE;
5465
5466         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5467                 CWARN("%s: MDS requires FID support, but client not\n",
5468                       mdt_obd_name(mdt));
5469                 return -EBADE;
5470         }
5471
5472         if (OCD_HAS_FLAG(data, PINGLESS)) {
5473                 if (ptlrpc_pinger_suppress_pings()) {
5474                         spin_lock(&exp->exp_obd->obd_dev_lock);
5475                         list_del_init(&exp->exp_obd_chain_timed);
5476                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5477                 } else {
5478                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5479                 }
5480         }
5481
5482         data->ocd_max_easize = mdt->mdt_max_ea_size;
5483
5484         /* NB: Disregard the rule against updating
5485          * exp_connect_data.ocd_connect_flags in this case, since
5486          * tgt_client_new() needs to know if this is client supports
5487          * multiple modify RPCs, and it is safe to expose this flag before
5488          * connection processing completes. */
5489         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5490                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5491                 spin_lock(&exp->exp_lock);
5492                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5493                 spin_unlock(&exp->exp_lock);
5494         }
5495
5496         if (OCD_HAS_FLAG(data, CKSUM)) {
5497                 __u32 cksum_types = data->ocd_cksum_types;
5498
5499                 /* The client set in ocd_cksum_types the checksum types it
5500                  * supports. We have to mask off the algorithms that we don't
5501                  * support */
5502                 data->ocd_cksum_types &= cksum_types_supported_server();
5503
5504                 if (unlikely(data->ocd_cksum_types == 0)) {
5505                         CERROR("%s: Connect with checksum support but no "
5506                                "ocd_cksum_types is set\n",
5507                                exp->exp_obd->obd_name);
5508                         RETURN(-EPROTO);
5509                 }
5510
5511                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5512                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5513                        cksum_types, data->ocd_cksum_types);
5514         } else {
5515                 /* This client does not support OBD_CONNECT_CKSUM
5516                  * fall back to CRC32 */
5517                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5518                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5519                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5520         }
5521
5522         return 0;
5523 }
5524
5525 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5526                                    struct mdt_thread_info *info,
5527                                    struct mdt_file_data *mfd)
5528 {
5529         struct lu_context ses;
5530         int rc;
5531         ENTRY;
5532
5533         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5534         if (rc)
5535                 RETURN(rc);
5536
5537         env->le_ses = &ses;
5538         lu_context_enter(&ses);
5539
5540         mdt_ucred(info)->uc_valid = UCRED_OLD;
5541         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5542
5543         lu_context_exit(&ses);
5544         lu_context_fini(&ses);
5545         env->le_ses = NULL;
5546
5547         RETURN(rc);
5548 }
5549
5550 static int mdt_export_cleanup(struct obd_export *exp)
5551 {
5552         struct list_head         closing_list;
5553         struct mdt_export_data  *med = &exp->exp_mdt_data;
5554         struct obd_device       *obd = exp->exp_obd;
5555         struct mdt_device       *mdt;
5556         struct mdt_thread_info  *info;
5557         struct lu_env            env;
5558         struct mdt_file_data    *mfd, *n;
5559         int rc = 0;
5560         ENTRY;
5561
5562         INIT_LIST_HEAD(&closing_list);
5563         spin_lock(&med->med_open_lock);
5564         while (!list_empty(&med->med_open_head)) {
5565                 struct list_head *tmp = med->med_open_head.next;
5566                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5567
5568                 /* Remove mfd handle so it can't be found again.
5569                  * We are consuming the mfd_list reference here. */
5570                 class_handle_unhash(&mfd->mfd_handle);
5571                 list_move_tail(&mfd->mfd_list, &closing_list);
5572         }
5573         spin_unlock(&med->med_open_lock);
5574         mdt = mdt_dev(obd->obd_lu_dev);
5575         LASSERT(mdt != NULL);
5576
5577         rc = lu_env_init(&env, LCT_MD_THREAD);
5578         if (rc)
5579                 RETURN(rc);
5580
5581         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5582         LASSERT(info != NULL);
5583         memset(info, 0, sizeof *info);
5584         info->mti_env = &env;
5585         info->mti_mdt = mdt;
5586         info->mti_exp = exp;
5587
5588         if (!list_empty(&closing_list)) {
5589                 struct md_attr *ma = &info->mti_attr;
5590
5591                 /* Close any open files (which may also cause orphan
5592                  * unlinking). */
5593                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5594                         list_del_init(&mfd->mfd_list);
5595                         ma->ma_need = ma->ma_valid = 0;
5596
5597                         /* This file is being closed due to an eviction, it
5598                          * could have been modified and now dirty regarding to
5599                          * HSM archive, check this!
5600                          * The logic here is to mark a file dirty if there's a
5601                          * chance it was dirtied before the client was evicted,
5602                          * so that we don't have to wait for a release attempt
5603                          * before finding out the file was actually dirty and
5604                          * fail the release. Aggressively marking it dirty here
5605                          * will cause the policy engine to attempt to
5606                          * re-archive it; when rearchiving, we can compare the
5607                          * current version to the HSM data_version and make the
5608                          * archive request into a noop if it's not actually
5609                          * dirty.
5610                          */
5611                         if (mfd->mfd_mode & FMODE_WRITE)
5612                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5613
5614                         /* Don't unlink orphan on failover umount, LU-184 */
5615                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5616                                 ma->ma_valid = MA_FLAGS;
5617                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5618                         }
5619                         mdt_mfd_close(info, mfd);
5620                 }
5621         }
5622         info->mti_mdt = NULL;
5623         /* cleanup client slot early */
5624         /* Do not erase record for recoverable client. */
5625         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5626                 tgt_client_del(&env, exp);
5627         lu_env_fini(&env);
5628
5629         RETURN(rc);
5630 }
5631
5632 static inline void mdt_enable_slc(struct mdt_device *mdt)
5633 {
5634         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5635                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5636 }
5637
5638 static inline void mdt_disable_slc(struct mdt_device *mdt)
5639 {
5640         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5641                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5642 }
5643
5644 static int mdt_obd_disconnect(struct obd_export *exp)
5645 {
5646         int rc;
5647         ENTRY;
5648
5649         LASSERT(exp);
5650         class_export_get(exp);
5651
5652         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5653             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5654                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5655
5656                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5657                         mdt_disable_slc(mdt);
5658         }
5659
5660         rc = server_disconnect_export(exp);
5661         if (rc != 0)
5662                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5663
5664         rc = mdt_export_cleanup(exp);
5665         nodemap_del_member(exp);
5666         class_export_put(exp);
5667         RETURN(rc);
5668 }
5669
5670 /* mds_connect copy */
5671 static int mdt_obd_connect(const struct lu_env *env,
5672                            struct obd_export **exp, struct obd_device *obd,
5673                            struct obd_uuid *cluuid,
5674                            struct obd_connect_data *data,
5675                            void *localdata)
5676 {
5677         struct obd_export       *lexp;
5678         struct lustre_handle    conn = { 0 };
5679         struct mdt_device       *mdt;
5680         int                      rc;
5681         lnet_nid_t              *client_nid = localdata;
5682         ENTRY;
5683
5684         LASSERT(env != NULL);
5685         LASSERT(data != NULL);
5686
5687         if (!exp || !obd || !cluuid)
5688                 RETURN(-EINVAL);
5689
5690         mdt = mdt_dev(obd->obd_lu_dev);
5691
5692         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5693             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5694                 atomic_inc(&mdt->mdt_mds_mds_conns);
5695                 mdt_enable_slc(mdt);
5696         }
5697
5698         /*
5699          * first, check whether the stack is ready to handle requests
5700          * XXX: probably not very appropriate method is used now
5701          *      at some point we should find a better one
5702          */
5703         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5704             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5705             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5706                 rc = obd_get_info(env, mdt->mdt_child_exp,
5707                                   sizeof(KEY_OSP_CONNECTED),
5708                                   KEY_OSP_CONNECTED, NULL, NULL);
5709                 if (rc)
5710                         RETURN(-EAGAIN);
5711                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5712         }
5713
5714         rc = class_connect(&conn, obd, cluuid);
5715         if (rc)
5716                 RETURN(rc);
5717
5718         lexp = class_conn2export(&conn);
5719         LASSERT(lexp != NULL);
5720
5721         rc = nodemap_add_member(*client_nid, lexp);
5722         if (rc != 0 && rc != -EEXIST)
5723                 GOTO(out, rc);
5724
5725         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5726         if (rc == 0) {
5727                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5728
5729                 LASSERT(lcd);
5730                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5731                 rc = tgt_client_new(env, lexp);
5732                 if (rc == 0)
5733                         mdt_export_stats_init(obd, lexp, localdata);
5734         }
5735 out:
5736         if (rc != 0) {
5737                 class_disconnect(lexp);
5738                 nodemap_del_member(lexp);
5739                 *exp = NULL;
5740         } else {
5741                 *exp = lexp;
5742                 /* Because we do not want this export to be evicted by pinger,
5743                  * let's not add this export to the timed chain list. */
5744                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5745                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5746                         list_del_init(&lexp->exp_obd_chain_timed);
5747                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5748                 }
5749         }
5750
5751         RETURN(rc);
5752 }
5753
5754 static int mdt_obd_reconnect(const struct lu_env *env,
5755                              struct obd_export *exp, struct obd_device *obd,
5756                              struct obd_uuid *cluuid,
5757                              struct obd_connect_data *data,
5758                              void *localdata)
5759 {
5760         lnet_nid_t             *client_nid = localdata;
5761         int                     rc;
5762         ENTRY;
5763
5764         if (exp == NULL || obd == NULL || cluuid == NULL)
5765                 RETURN(-EINVAL);
5766
5767         rc = nodemap_add_member(*client_nid, exp);
5768         if (rc != 0 && rc != -EEXIST)
5769                 RETURN(rc);
5770
5771         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5772                                   true);
5773         if (rc == 0)
5774                 mdt_export_stats_init(obd, exp, localdata);
5775         else
5776                 nodemap_del_member(exp);
5777
5778         RETURN(rc);
5779 }
5780
5781 /* FIXME: Can we avoid using these two interfaces? */
5782 static int mdt_init_export(struct obd_export *exp)
5783 {
5784         struct mdt_export_data *med = &exp->exp_mdt_data;
5785         int                     rc;
5786         ENTRY;
5787
5788         INIT_LIST_HEAD(&med->med_open_head);
5789         spin_lock_init(&med->med_open_lock);
5790         spin_lock(&exp->exp_lock);
5791         exp->exp_connecting = 1;
5792         spin_unlock(&exp->exp_lock);
5793
5794         /* self-export doesn't need client data and ldlm initialization */
5795         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5796                                      &exp->exp_client_uuid)))
5797                 RETURN(0);
5798
5799         rc = tgt_client_alloc(exp);
5800         if (rc)
5801                 GOTO(err, rc);
5802
5803         rc = ldlm_init_export(exp);
5804         if (rc)
5805                 GOTO(err_free, rc);
5806
5807         RETURN(rc);
5808
5809 err_free:
5810         tgt_client_free(exp);
5811 err:
5812         CERROR("%s: Failed to initialize export: rc = %d\n",
5813                exp->exp_obd->obd_name, rc);
5814         return rc;
5815 }
5816
5817 static int mdt_destroy_export(struct obd_export *exp)
5818 {
5819         ENTRY;
5820
5821         target_destroy_export(exp);
5822         /* destroy can be called from failed obd_setup, so
5823          * checking uuid is safer than obd_self_export */
5824         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5825                                      &exp->exp_client_uuid)))
5826                 RETURN(0);
5827
5828         ldlm_destroy_export(exp);
5829         tgt_client_free(exp);
5830
5831         LASSERT(list_empty(&exp->exp_outstanding_replies));
5832         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5833
5834         RETURN(0);
5835 }
5836
5837 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5838                    struct linkea_data *ldata)
5839 {
5840         int rc;
5841
5842         LASSERT(ldata->ld_buf->lb_buf != NULL);
5843
5844         if (!mdt_object_exists(mdt_obj))
5845                 return -ENODATA;
5846
5847         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5848                           ldata->ld_buf, XATTR_NAME_LINK);
5849         if (rc == -ERANGE) {
5850                 /* Buf was too small, figure out what we need. */
5851                 lu_buf_free(ldata->ld_buf);
5852                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5853                                   ldata->ld_buf, XATTR_NAME_LINK);
5854                 if (rc < 0)
5855                         return rc;
5856                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5857                 if (ldata->ld_buf->lb_buf == NULL)
5858                         return -ENOMEM;
5859                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5860                                   ldata->ld_buf, XATTR_NAME_LINK);
5861         }
5862         if (rc < 0)
5863                 return rc;
5864
5865         return linkea_init_with_rec(ldata);
5866 }
5867
5868 /**
5869  * Given an MDT object, try to look up the full path to the object.
5870  * Part of the MDT layer implementation of lfs fid2path.
5871  *
5872  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5873  * \param[in]     obj   Object to do path lookup of
5874  * \param[in,out] fp    User-provided struct to store path information
5875  * \param[in]     root_fid Root FID of current path should reach
5876  *
5877  * \retval 0 Lookup successful, path information stored in fp
5878  * \retval -EAGAIN Lookup failed, usually because object is being moved
5879  * \retval negative errno if there was a problem
5880  */
5881 static int mdt_path_current(struct mdt_thread_info *info,
5882                             struct mdt_object *obj,
5883                             struct getinfo_fid2path *fp,
5884                             struct lu_fid *root_fid)
5885 {
5886         struct mdt_device       *mdt = info->mti_mdt;
5887         struct mdt_object       *mdt_obj;
5888         struct link_ea_header   *leh;
5889         struct link_ea_entry    *lee;
5890         struct lu_name          *tmpname = &info->mti_name;
5891         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5892         struct lu_buf           *buf = &info->mti_big_buf;
5893         char                    *ptr;
5894         int                     reclen;
5895         struct linkea_data      ldata = { NULL };
5896         int                     rc = 0;
5897         bool                    first = true;
5898         ENTRY;
5899
5900         /* temp buffer for path element, the buffer will be finally freed
5901          * in mdt_thread_info_fini */
5902         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5903         if (buf->lb_buf == NULL)
5904                 RETURN(-ENOMEM);
5905
5906         ldata.ld_buf = buf;
5907         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5908         *ptr = 0;
5909         --ptr;
5910         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5911
5912         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5913                 struct lu_buf           lmv_buf;
5914
5915                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5916                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5917                         GOTO(out, rc = -ENOENT);
5918
5919                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5920                         mdt_obj = obj;
5921                         mdt_object_get(info->mti_env, mdt_obj);
5922                 } else {
5923                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5924                         if (IS_ERR(mdt_obj))
5925                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5926                 }
5927
5928                 if (!mdt_object_exists(mdt_obj)) {
5929                         mdt_object_put(info->mti_env, mdt_obj);
5930                         GOTO(out, rc = -ENOENT);
5931                 }
5932
5933                 if (mdt_object_remote(mdt_obj)) {
5934                         mdt_object_put(info->mti_env, mdt_obj);
5935                         GOTO(remote_out, rc = -EREMOTE);
5936                 }
5937
5938                 rc = mdt_links_read(info, mdt_obj, &ldata);
5939                 if (rc != 0) {
5940                         mdt_object_put(info->mti_env, mdt_obj);
5941                         GOTO(out, rc);
5942                 }
5943
5944                 leh = buf->lb_buf;
5945                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5946                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5947                 /* If set, use link #linkno for path lookup, otherwise use
5948                    link #0.  Only do this for the final path element. */
5949                 if (first && fp->gf_linkno < leh->leh_reccount) {
5950                         int count;
5951                         for (count = 0; count < fp->gf_linkno; count++) {
5952                                 lee = (struct link_ea_entry *)
5953                                      ((char *)lee + reclen);
5954                                 linkea_entry_unpack(lee, &reclen, tmpname,
5955                                                     tmpfid);
5956                         }
5957                         if (fp->gf_linkno < leh->leh_reccount - 1)
5958                                 /* indicate to user there are more links */
5959                                 fp->gf_linkno++;
5960                 }
5961
5962                 lmv_buf.lb_buf = info->mti_xattr_buf;
5963                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5964                 /* Check if it is slave stripes */
5965                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5966                                   &lmv_buf, XATTR_NAME_LMV);
5967                 mdt_object_put(info->mti_env, mdt_obj);
5968                 if (rc > 0) {
5969                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5970
5971                         /* For slave stripes, get its master */
5972                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5973                                 fp->gf_fid = *tmpfid;
5974                                 continue;
5975                         }
5976                 } else if (rc < 0 && rc != -ENODATA) {
5977                         GOTO(out, rc);
5978                 }
5979
5980                 rc = 0;
5981
5982                 /* Pack the name in the end of the buffer */
5983                 ptr -= tmpname->ln_namelen;
5984                 if (ptr - 1 <= fp->gf_u.gf_path)
5985                         GOTO(out, rc = -EOVERFLOW);
5986                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5987                 *(--ptr) = '/';
5988
5989                 /* keep the last resolved fid to the client, so the
5990                  * client will build the left path on another MDT for
5991                  * remote object */
5992                 fp->gf_fid = *tmpfid;
5993
5994                 first = false;
5995         }
5996
5997 remote_out:
5998         ptr++; /* skip leading / */
5999         memmove(fp->gf_u.gf_path, ptr,
6000                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6001
6002 out:
6003         RETURN(rc);
6004 }
6005
6006 /**
6007  * Given an MDT object, use mdt_path_current to get the path.
6008  * Essentially a wrapper to retry mdt_path_current a set number of times
6009  * if -EAGAIN is returned (usually because an object is being moved).
6010  *
6011  * Part of the MDT layer implementation of lfs fid2path.
6012  *
6013  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6014  * \param[in]     obj   Object to do path lookup of
6015  * \param[in,out] fp    User-provided struct for arguments and to store path
6016  *                      information
6017  *
6018  * \retval 0 Lookup successful, path information stored in fp
6019  * \retval negative errno if there was a problem
6020  */
6021 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6022                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6023 {
6024         struct mdt_device       *mdt = info->mti_mdt;
6025         int                     tries = 3;
6026         int                     rc = -EAGAIN;
6027         ENTRY;
6028
6029         if (fp->gf_pathlen < 3)
6030                 RETURN(-EOVERFLOW);
6031
6032         if (root_fid == NULL)
6033                 root_fid = &mdt->mdt_md_root_fid;
6034
6035         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6036                 fp->gf_u.gf_path[0] = '\0';
6037                 RETURN(0);
6038         }
6039
6040         /* Retry multiple times in case file is being moved */
6041         while (tries-- && rc == -EAGAIN)
6042                 rc = mdt_path_current(info, obj, fp, root_fid);
6043
6044         RETURN(rc);
6045 }
6046
6047 /**
6048  * Get the full path of the provided FID, as of changelog record recno.
6049  *
6050  * This checks sanity and looks up object for user provided FID
6051  * before calling the actual path lookup code.
6052  *
6053  * Part of the MDT layer implementation of lfs fid2path.
6054  *
6055  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6056  * \param[in,out] fp    User-provided struct for arguments and to store path
6057  *                      information
6058  *
6059  * \retval 0 Lookup successful, path information and recno stored in fp
6060  * \retval -ENOENT, object does not exist
6061  * \retval negative errno if there was a problem
6062  */
6063 static int mdt_fid2path(struct mdt_thread_info *info,
6064                         struct lu_fid *root_fid,
6065                         struct getinfo_fid2path *fp)
6066 {
6067         struct mdt_device *mdt = info->mti_mdt;
6068         struct mdt_object *obj;
6069         int    rc;
6070         ENTRY;
6071
6072         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6073                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6074
6075         if (!fid_is_sane(&fp->gf_fid))
6076                 RETURN(-EINVAL);
6077
6078         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6079                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6080                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6081                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6082                 RETURN(-EINVAL);
6083         }
6084
6085         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6086         if (IS_ERR(obj)) {
6087                 rc = PTR_ERR(obj);
6088                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6089                        PFID(&fp->gf_fid), rc);
6090                 RETURN(rc);
6091         }
6092
6093         if (mdt_object_remote(obj))
6094                 rc = -EREMOTE;
6095         else if (!mdt_object_exists(obj))
6096                 rc = -ENOENT;
6097         else
6098                 rc = 0;
6099
6100         if (rc < 0) {
6101                 mdt_object_put(info->mti_env, obj);
6102                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6103                        PFID(&fp->gf_fid), rc);
6104                 RETURN(rc);
6105         }
6106
6107         rc = mdt_path(info, obj, fp, root_fid);
6108
6109         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6110                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6111                fp->gf_recno, fp->gf_linkno);
6112
6113         mdt_object_put(info->mti_env, obj);
6114
6115         RETURN(rc);
6116 }
6117
6118 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6119                             void *val, int vallen)
6120 {
6121         struct getinfo_fid2path *fpout, *fpin;
6122         struct lu_fid *root_fid = NULL;
6123         int rc = 0;
6124
6125         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6126         fpout = val;
6127
6128         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6129                 lustre_swab_fid2path(fpin);
6130
6131         memcpy(fpout, fpin, sizeof(*fpin));
6132         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6133                 RETURN(-EINVAL);
6134
6135         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6136                       sizeof(struct lu_fid)) {
6137                 /* client sent its root FID, which is normally fileset FID */
6138                 root_fid = fpin->gf_u.gf_root_fid;
6139                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6140                         lustre_swab_lu_fid(root_fid);
6141
6142                 if (root_fid != NULL && !fid_is_sane(root_fid))
6143                         RETURN(-EINVAL);
6144         }
6145
6146         rc = mdt_fid2path(info, root_fid, fpout);
6147         RETURN(rc);
6148 }
6149
6150 int mdt_get_info(struct tgt_session_info *tsi)
6151 {
6152         char    *key;
6153         int      keylen;
6154         __u32   *vallen;
6155         void    *valout;
6156         int      rc;
6157
6158         ENTRY;
6159
6160         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6161         if (key == NULL) {
6162                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6163                 RETURN(err_serious(-EFAULT));
6164         }
6165         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6166                                       RCL_CLIENT);
6167
6168         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6169         if (vallen == NULL) {
6170                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6171                                 tgt_name(tsi->tsi_tgt));
6172                 RETURN(err_serious(-EFAULT));
6173         }
6174
6175         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6176                              *vallen);
6177         rc = req_capsule_server_pack(tsi->tsi_pill);
6178         if (rc)
6179                 RETURN(err_serious(rc));
6180
6181         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6182         if (valout == NULL) {
6183                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6184                                 tgt_name(tsi->tsi_tgt));
6185                 RETURN(err_serious(-EFAULT));
6186         }
6187
6188         if (KEY_IS(KEY_FID2PATH)) {
6189                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6190
6191                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6192                 mdt_thread_info_fini(info);
6193         } else {
6194                 rc = -EINVAL;
6195         }
6196         RETURN(rc);
6197 }
6198
6199 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6200 {
6201         struct obd_ioctl_data *data = karg;
6202         struct lu_fid *fid;
6203         __u64 version;
6204         struct mdt_object *obj;
6205         struct mdt_lock_handle  *lh;
6206         int rc;
6207         ENTRY;
6208
6209         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6210             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6211                 RETURN(-EINVAL);
6212
6213         fid = (struct lu_fid *)data->ioc_inlbuf1;
6214
6215         if (!fid_is_sane(fid))
6216                 RETURN(-EINVAL);
6217
6218         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6219
6220         lh = &mti->mti_lh[MDT_LH_PARENT];
6221         mdt_lock_reg_init(lh, LCK_CR);
6222
6223         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6224         if (IS_ERR(obj))
6225                 RETURN(PTR_ERR(obj));
6226
6227         if (mdt_object_remote(obj)) {
6228                 rc = -EREMOTE;
6229                 /**
6230                  * before calling version get the correct MDS should be
6231                  * fid, this is error to find remote object here
6232                  */
6233                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6234         } else if (!mdt_object_exists(obj)) {
6235                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6236                 rc = -ENOENT;
6237         } else {
6238                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6239                *(__u64 *)data->ioc_inlbuf2 = version;
6240                 rc = 0;
6241         }
6242         mdt_object_unlock_put(mti, obj, lh, 1);
6243         RETURN(rc);
6244 }
6245
6246 /* ioctls on obd dev */
6247 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6248                          void *karg, void __user *uarg)
6249 {
6250         struct lu_env      env;
6251         struct obd_device *obd = exp->exp_obd;
6252         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6253         struct dt_device  *dt = mdt->mdt_bottom;
6254         int rc;
6255
6256         ENTRY;
6257         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6258         rc = lu_env_init(&env, LCT_MD_THREAD);
6259         if (rc)
6260                 RETURN(rc);
6261
6262         switch (cmd) {
6263         case OBD_IOC_SYNC:
6264                 rc = mdt_device_sync(&env, mdt);
6265                 break;
6266         case OBD_IOC_SET_READONLY:
6267                 rc = dt_sync(&env, dt);
6268                 if (rc == 0)
6269                         rc = dt_ro(&env, dt);
6270                 break;
6271         case OBD_IOC_ABORT_RECOVERY:
6272                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6273                 obd->obd_abort_recovery = 1;
6274                 target_stop_recovery_thread(obd);
6275                 rc = 0;
6276                 break;
6277         case OBD_IOC_CHANGELOG_REG:
6278         case OBD_IOC_CHANGELOG_DEREG:
6279         case OBD_IOC_CHANGELOG_CLEAR:
6280                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6281                                                            mdt->mdt_child,
6282                                                            cmd, len, karg);
6283                 break;
6284         case OBD_IOC_START_LFSCK: {
6285                 struct md_device *next = mdt->mdt_child;
6286                 struct obd_ioctl_data *data = karg;
6287                 struct lfsck_start_param lsp;
6288
6289                 if (unlikely(data == NULL)) {
6290                         rc = -EINVAL;
6291                         break;
6292                 }
6293
6294                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6295                 lsp.lsp_index_valid = 0;
6296                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6297                 break;
6298         }
6299         case OBD_IOC_STOP_LFSCK: {
6300                 struct md_device        *next = mdt->mdt_child;
6301                 struct obd_ioctl_data   *data = karg;
6302                 struct lfsck_stop        stop;
6303
6304                 stop.ls_status = LS_STOPPED;
6305                 /* Old lfsck utils may pass NULL @stop. */
6306                 if (data->ioc_inlbuf1 == NULL)
6307                         stop.ls_flags = 0;
6308                 else
6309                         stop.ls_flags =
6310                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6311
6312                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6313                 break;
6314         }
6315         case OBD_IOC_QUERY_LFSCK: {
6316                 struct md_device        *next = mdt->mdt_child;
6317                 struct obd_ioctl_data   *data = karg;
6318
6319                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6320                                                  data->ioc_inlbuf1);
6321                 break;
6322         }
6323         case OBD_IOC_GET_OBJ_VERSION: {
6324                 struct mdt_thread_info *mti;
6325                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6326                 memset(mti, 0, sizeof *mti);
6327                 mti->mti_env = &env;
6328                 mti->mti_mdt = mdt;
6329                 mti->mti_exp = exp;
6330
6331                 rc = mdt_ioc_version_get(mti, karg);
6332                 break;
6333         }
6334         case OBD_IOC_CATLOGLIST: {
6335                 struct mdt_thread_info *mti;
6336
6337                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6338                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6339                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6340                                        &mti->mti_tmp_fid1);
6341                 break;
6342          }
6343         default:
6344                 rc = -EOPNOTSUPP;
6345                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6346                         mdt_obd_name(mdt), cmd, rc);
6347         }
6348
6349         lu_env_fini(&env);
6350         RETURN(rc);
6351 }
6352
6353 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6354 {
6355         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6356         int rc;
6357         ENTRY;
6358
6359         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6360                 struct lfsck_start_param lsp;
6361
6362                 lsp.lsp_start = NULL;
6363                 lsp.lsp_index_valid = 0;
6364                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6365                                                            OBD_IOC_START_LFSCK,
6366                                                            0, &lsp);
6367                 if (rc != 0 && rc != -EALREADY)
6368                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6369                               mdt_obd_name(mdt), rc);
6370         }
6371
6372         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6373         RETURN(rc);
6374 }
6375
6376 static int mdt_obd_postrecov(struct obd_device *obd)
6377 {
6378         struct lu_env env;
6379         int rc;
6380
6381         rc = lu_env_init(&env, LCT_MD_THREAD);
6382         if (rc)
6383                 RETURN(rc);
6384         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6385         lu_env_fini(&env);
6386         return rc;
6387 }
6388
6389 static struct obd_ops mdt_obd_device_ops = {
6390         .o_owner          = THIS_MODULE,
6391         .o_set_info_async = mdt_obd_set_info_async,
6392         .o_connect        = mdt_obd_connect,
6393         .o_reconnect      = mdt_obd_reconnect,
6394         .o_disconnect     = mdt_obd_disconnect,
6395         .o_init_export    = mdt_init_export,
6396         .o_destroy_export = mdt_destroy_export,
6397         .o_iocontrol      = mdt_iocontrol,
6398         .o_postrecov      = mdt_obd_postrecov,
6399         /* Data-on-MDT IO methods */
6400         .o_preprw         = mdt_obd_preprw,
6401         .o_commitrw       = mdt_obd_commitrw,
6402 };
6403
6404 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6405                                          struct lu_device *d)
6406 {
6407         struct mdt_device *m = mdt_dev(d);
6408         ENTRY;
6409
6410         mdt_fini(env, m);
6411         RETURN(NULL);
6412 }
6413
6414 static struct lu_device *mdt_device_free(const struct lu_env *env,
6415                                          struct lu_device *d)
6416 {
6417         struct mdt_device *m = mdt_dev(d);
6418         ENTRY;
6419
6420         lu_device_fini(&m->mdt_lu_dev);
6421         OBD_FREE_PTR(m);
6422
6423         RETURN(NULL);
6424 }
6425
6426 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6427                                           struct lu_device_type *t,
6428                                           struct lustre_cfg *cfg)
6429 {
6430         struct lu_device  *l;
6431         struct mdt_device *m;
6432
6433         OBD_ALLOC_PTR(m);
6434         if (m != NULL) {
6435                 int rc;
6436
6437                 l = &m->mdt_lu_dev;
6438                 rc = mdt_init0(env, m, t, cfg);
6439                 if (rc != 0) {
6440                         mdt_device_free(env, l);
6441                         l = ERR_PTR(rc);
6442                         return l;
6443                 }
6444         } else
6445                 l = ERR_PTR(-ENOMEM);
6446         return l;
6447 }
6448
6449 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6450 LU_KEY_INIT(mdt, struct mdt_thread_info);
6451
6452 static void mdt_key_fini(const struct lu_context *ctx,
6453                          struct lu_context_key *key, void* data)
6454 {
6455         struct mdt_thread_info *info = data;
6456
6457         if (info->mti_big_lmm) {
6458                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6459                 info->mti_big_lmm = NULL;
6460                 info->mti_big_lmmsize = 0;
6461         }
6462
6463         if (info->mti_big_acl) {
6464                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6465                 info->mti_big_acl = NULL;
6466                 info->mti_big_aclsize = 0;
6467         }
6468
6469         OBD_FREE_PTR(info);
6470 }
6471
6472 /* context key: mdt_thread_key */
6473 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6474
6475 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6476 {
6477         return lu_ucred(info->mti_env);
6478 }
6479
6480 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6481 {
6482         return lu_ucred_check(info->mti_env);
6483 }
6484
6485 /**
6486  * Enable/disable COS (Commit On Sharing).
6487  *
6488  * Set/Clear the COS flag in mdt options.
6489  *
6490  * \param mdt mdt device
6491  * \param val 0 disables COS, other values enable COS
6492  */
6493 void mdt_enable_cos(struct mdt_device *mdt, int val)
6494 {
6495         struct lu_env env;
6496         int rc;
6497
6498         mdt->mdt_opts.mo_cos = !!val;
6499         rc = lu_env_init(&env, LCT_LOCAL);
6500         if (unlikely(rc != 0)) {
6501                 CWARN("%s: lu_env initialization failed, cannot "
6502                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6503                 return;
6504         }
6505         mdt_device_sync(&env, mdt);
6506         lu_env_fini(&env);
6507 }
6508
6509 /**
6510  * Check COS (Commit On Sharing) status.
6511  *
6512  * Return COS flag status.
6513  *
6514  * \param mdt mdt device
6515  */
6516 int mdt_cos_is_enabled(struct mdt_device *mdt)
6517 {
6518         return mdt->mdt_opts.mo_cos != 0;
6519 }
6520
6521 static struct lu_device_type_operations mdt_device_type_ops = {
6522         .ldto_device_alloc = mdt_device_alloc,
6523         .ldto_device_free  = mdt_device_free,
6524         .ldto_device_fini  = mdt_device_fini
6525 };
6526
6527 static struct lu_device_type mdt_device_type = {
6528         .ldt_tags     = LU_DEVICE_MD,
6529         .ldt_name     = LUSTRE_MDT_NAME,
6530         .ldt_ops      = &mdt_device_type_ops,
6531         .ldt_ctx_tags = LCT_MD_THREAD
6532 };
6533
6534 static int __init mdt_init(void)
6535 {
6536         int rc;
6537
6538         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6539                  FID_NOBRACE_LEN + 1);
6540         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6541                  FID_LEN + 1);
6542         rc = lu_kmem_init(mdt_caches);
6543         if (rc)
6544                 return rc;
6545
6546         rc = mds_mod_init();
6547         if (rc)
6548                 GOTO(lu_fini, rc);
6549
6550         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6551                                  LUSTRE_MDT_NAME, &mdt_device_type);
6552         if (rc)
6553                 GOTO(mds_fini, rc);
6554 lu_fini:
6555         if (rc)
6556                 lu_kmem_fini(mdt_caches);
6557 mds_fini:
6558         if (rc)
6559                 mds_mod_exit();
6560         return rc;
6561 }
6562
6563 static void __exit mdt_exit(void)
6564 {
6565         class_unregister_type(LUSTRE_MDT_NAME);
6566         mds_mod_exit();
6567         lu_kmem_fini(mdt_caches);
6568 }
6569
6570 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6571 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6572 MODULE_VERSION(LUSTRE_VERSION_STRING);
6573 MODULE_LICENSE("GPL");
6574
6575 module_init(mdt_init);
6576 module_exit(mdt_exit);