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