lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_net.h>
  57 #include <lustre_nodemap.h>
  58 #include <lustre_mds.h>
  59 #include <lustre_param.h>
  60 #include <lustre_quota.h>
  61 #include <lustre_swab.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 module_param(max_mod_rpcs_per_client, uint, 0644);
  73 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 270                               struct lu_fid *fid)
 271 {
 272         struct mdt_device *mdt = info->mti_mdt;
 273         struct lu_name *lname = &info->mti_name;
 274         char *name = NULL;
 275         struct mdt_object *parent;
 276         u32 mode;
 277         int rc = 0;
 278
 279         LASSERT(!info->mti_cross_ref);
 280
 281         OBD_ALLOC(name, NAME_MAX + 1);
 282         if (name == NULL)
 283                 return -ENOMEM;
 284         lname->ln_name = name;
 285
 286         /*
 287          * We may want to allow this to mount a completely separate
 288          * fileset from the MDT in the future, but keeping it to
 289          * ROOT/ only for now avoid potential security issues.
 290          */
 291         *fid = mdt->mdt_md_root_fid;
 292
 293         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 294                 const char *s1 = fileset;
 295                 const char *s2;
 296
 297                 while (*++s1 == '/')
 298                         ;
 299                 s2 = s1;
 300                 while (*s2 != '/' && *s2 != '\0')
 301                         s2++;
 302
 303                 if (s2 == s1)
 304                         break;
 305
 306                 fileset = s2;
 307
 308                 lname->ln_namelen = s2 - s1;
 309                 if (lname->ln_namelen > NAME_MAX) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 /* reject .. as a path component */
 315                 if (lname->ln_namelen == 2 &&
 316                     strncmp(s1, "..", 2) == 0) {
 317                         rc = -EINVAL;
 318                         break;
 319                 }
 320
 321                 strncpy(name, s1, lname->ln_namelen);
 322                 name[lname->ln_namelen] = '\0';
 323
 324                 parent = mdt_object_find(info->mti_env, mdt, fid);
 325                 if (IS_ERR(parent)) {
 326                         rc = PTR_ERR(parent);
 327                         break;
 328                 }
 329                 /* Only got the fid of this obj by name */
 330                 fid_zero(fid);
 331                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 332                                 fid, &info->mti_spec);
 333                 mdt_object_put(info->mti_env, parent);
 334         }
 335         if (!rc) {
 336                 parent = mdt_object_find(info->mti_env, mdt, fid);
 337                 if (IS_ERR(parent))
 338                         rc = PTR_ERR(parent);
 339                 else {
 340                         mode = lu_object_attr(&parent->mot_obj);
 341                         mdt_object_put(info->mti_env, parent);
 342                         if (!S_ISDIR(mode))
 343                                 rc = -ENOTDIR;
 344                 }
 345         }
 346
 347         OBD_FREE(name, NAME_MAX + 1);
 348
 349         return rc;
 350 }
 351
 352 static int mdt_get_root(struct tgt_session_info *tsi)
 353 {
 354         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 355         struct mdt_device       *mdt = info->mti_mdt;
 356         struct mdt_body         *repbody;
 357         char                    *fileset = NULL, *buffer = NULL;
 358         int                      rc;
 359         struct obd_export       *exp = info->mti_exp;
 360         char                    *nodemap_fileset;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376         }
 377
 378         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 379         if (nodemap_fileset && nodemap_fileset[0]) {
 380                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 381                 if (fileset) {
 382                         /* consider fileset from client as a sub-fileset
 383                          * of the nodemap one */
 384                         OBD_ALLOC(buffer, PATH_MAX + 1);
 385                         if (buffer == NULL)
 386                                 GOTO(out, rc = err_serious(-ENOMEM));
 387                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 388                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 389                                 GOTO(out, rc = err_serious(-EINVAL));
 390                         fileset = buffer;
 391                 } else {
 392                         /* enforce fileset as specified in the nodemap */
 393                         fileset = nodemap_fileset;
 394                 }
 395         }
 396
 397         if (fileset) {
 398                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 399                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 400                 if (rc < 0)
 401                         GOTO(out, rc = err_serious(rc));
 402         } else {
 403                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 404         }
 405         repbody->mbo_valid |= OBD_MD_FLID;
 406
 407         EXIT;
 408 out:
 409         mdt_thread_info_fini(info);
 410         if (buffer)
 411                 OBD_FREE(buffer, PATH_MAX+1);
 412         return rc;
 413 }
 414
 415 static int mdt_statfs(struct tgt_session_info *tsi)
 416 {
 417         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 418         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 419         struct md_device                *next = info->mti_mdt->mdt_child;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         /** statfs information are cached in the mdt_device */
 445         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 446                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 447                 /** statfs data is too old, get up-to-date one */
 448                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 449                 if (rc)
 450                         GOTO(out, rc);
 451                 spin_lock(&info->mti_mdt->mdt_lock);
 452                 info->mti_mdt->mdt_osfs = *osfs;
 453                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 454                 spin_unlock(&info->mti_mdt->mdt_lock);
 455         } else {
 456                 /** use cached statfs data */
 457                 spin_lock(&info->mti_mdt->mdt_lock);
 458                 *osfs = info->mti_mdt->mdt_osfs;
 459                 spin_unlock(&info->mti_mdt->mdt_lock);
 460         }
 461
 462         if (rc == 0)
 463                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 464 out:
 465         mdt_thread_info_fini(info);
 466         RETURN(rc);
 467 }
 468
 469 #ifdef CONFIG_FS_POSIX_ACL
 470 /*
 471  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 472  *
 473  * \param       info    thread info object
 474  * \param       repbody reply to pack ACLs into
 475  * \param       o       mdt object of file to examine
 476  * \param       nodemap nodemap of client to reply to
 477  * \retval      0       success
 478  * \retval      -errno  error getting or parsing ACL from disk
 479  */
 480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 481                       struct mdt_object *o, struct lu_nodemap *nodemap)
 482 {
 483         const struct lu_env     *env = info->mti_env;
 484         struct md_object        *next = mdt_object_child(o);
 485         struct lu_buf           *buf = &info->mti_buf;
 486         int rc;
 487
 488         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 489         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 490                                            RCL_SERVER);
 491         if (buf->lb_len == 0)
 492                 return 0;
 493
 494         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 495         if (rc < 0) {
 496                 if (rc == -ENODATA) {
 497                         repbody->mbo_aclsize = 0;
 498                         repbody->mbo_valid |= OBD_MD_FLACL;
 499                         rc = 0;
 500                 } else if (rc == -EOPNOTSUPP) {
 501                         rc = 0;
 502                 } else {
 503                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 504                                mdt_obd_name(info->mti_mdt),
 505                                PFID(mdt_object_fid(o)), rc);
 506                 }
 507         } else {
 508                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 509                                      rc, NODEMAP_FS_TO_CLIENT);
 510                 /* if all ACLs mapped out, rc is still >= 0 */
 511                 if (rc < 0) {
 512                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 513                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 514                                PFID(mdt_object_fid(o)), rc);
 515                 } else {
 516                         repbody->mbo_aclsize = rc;
 517                         repbody->mbo_valid |= OBD_MD_FLACL;
 518                         rc = 0;
 519                 }
 520         }
 521         return rc;
 522 }
 523 #endif
 524
 525 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 526                         const struct lu_attr *attr, const struct lu_fid *fid)
 527 {
 528         struct md_attr *ma = &info->mti_attr;
 529         struct obd_export *exp = info->mti_exp;
 530         struct lu_nodemap *nodemap = NULL;
 531
 532         LASSERT(ma->ma_valid & MA_INODE);
 533
 534         if (attr->la_valid & LA_ATIME) {
 535                 b->mbo_atime = attr->la_atime;
 536                 b->mbo_valid |= OBD_MD_FLATIME;
 537         }
 538         if (attr->la_valid & LA_MTIME) {
 539                 b->mbo_mtime = attr->la_mtime;
 540                 b->mbo_valid |= OBD_MD_FLMTIME;
 541         }
 542         if (attr->la_valid & LA_CTIME) {
 543                 b->mbo_ctime = attr->la_ctime;
 544                 b->mbo_valid |= OBD_MD_FLCTIME;
 545         }
 546         if (attr->la_valid & LA_FLAGS) {
 547                 b->mbo_flags = attr->la_flags;
 548                 b->mbo_valid |= OBD_MD_FLFLAGS;
 549         }
 550         if (attr->la_valid & LA_NLINK) {
 551                 b->mbo_nlink = attr->la_nlink;
 552                 b->mbo_valid |= OBD_MD_FLNLINK;
 553         }
 554         if (attr->la_valid & (LA_UID|LA_GID)) {
 555                 nodemap = nodemap_get_from_exp(exp);
 556                 if (IS_ERR(nodemap))
 557                         goto out;
 558         }
 559         if (attr->la_valid & LA_UID) {
 560                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 561                                             NODEMAP_FS_TO_CLIENT,
 562                                             attr->la_uid);
 563                 b->mbo_valid |= OBD_MD_FLUID;
 564         }
 565         if (attr->la_valid & LA_GID) {
 566                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 567                                             NODEMAP_FS_TO_CLIENT,
 568                                             attr->la_gid);
 569                 b->mbo_valid |= OBD_MD_FLGID;
 570         }
 571
 572         b->mbo_mode = attr->la_mode;
 573         if (attr->la_valid & LA_MODE)
 574                 b->mbo_valid |= OBD_MD_FLMODE;
 575         if (attr->la_valid & LA_TYPE)
 576                 b->mbo_valid |= OBD_MD_FLTYPE;
 577
 578         if (fid != NULL) {
 579                 b->mbo_fid1 = *fid;
 580                 b->mbo_valid |= OBD_MD_FLID;
 581                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 582                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 583         }
 584
 585         if (!(attr->la_valid & LA_TYPE))
 586                 return;
 587
 588         b->mbo_rdev   = attr->la_rdev;
 589         b->mbo_size   = attr->la_size;
 590         b->mbo_blocks = attr->la_blocks;
 591
 592         if (!S_ISREG(attr->la_mode)) {
 593                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 594         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 595                 /* means no objects are allocated on osts. */
 596                 LASSERT(!(ma->ma_valid & MA_LOV));
 597                 /* just ignore blocks occupied by extend attributes on MDS */
 598                 b->mbo_blocks = 0;
 599                 /* if no object is allocated on osts, the size on mds is valid.
 600                  * b=22272 */
 601                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 602         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 603                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 604                 /* A released file stores its size on MDS. */
 605                 /* But return 1 block for released file, unless tools like tar
 606                  * will consider it fully sparse. (LU-3864)
 607                  */
 608                 if (unlikely(b->mbo_size == 0))
 609                         b->mbo_blocks = 0;
 610                 else
 611                         b->mbo_blocks = 1;
 612                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 613         }
 614
 615         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 616                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 617                        PFID(fid), (unsigned long long)b->mbo_size);
 618
 619 out:
 620         if (!IS_ERR_OR_NULL(nodemap))
 621                 nodemap_putref(nodemap);
 622 }
 623
 624 static inline int mdt_body_has_lov(const struct lu_attr *la,
 625                                    const struct mdt_body *body)
 626 {
 627         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 628                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 629 }
 630
 631 void mdt_client_compatibility(struct mdt_thread_info *info)
 632 {
 633         struct mdt_body       *body;
 634         struct ptlrpc_request *req = mdt_info_req(info);
 635         struct obd_export     *exp = req->rq_export;
 636         struct md_attr        *ma = &info->mti_attr;
 637         struct lu_attr        *la = &ma->ma_attr;
 638         ENTRY;
 639
 640         if (exp_connect_layout(exp))
 641                 /* the client can deal with 16-bit lmm_stripe_count */
 642                 RETURN_EXIT;
 643
 644         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 645
 646         if (!mdt_body_has_lov(la, body))
 647                 RETURN_EXIT;
 648
 649         /* now we have a reply with a lov for a client not compatible with the
 650          * layout lock so we have to clean the layout generation number */
 651         if (S_ISREG(la->la_mode))
 652                 ma->ma_lmm->lmm_layout_gen = 0;
 653         EXIT;
 654 }
 655
 656 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 657                                    struct mdt_object *o)
 658 {
 659         const struct lu_env *env = info->mti_env;
 660         int rc, rc2;
 661
 662         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 663                           XATTR_NAME_LOV);
 664
 665         if (rc == -ENODATA)
 666                 rc = 0;
 667
 668         if (rc < 0)
 669                 goto out;
 670
 671         /* Is it a directory? Let's check for the LMV as well */
 672         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 673                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 674                                    XATTR_NAME_LMV);
 675
 676                 if (rc2 == -ENODATA)
 677                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 678                                            &LU_BUF_NULL,
 679                                            XATTR_NAME_DEFAULT_LMV);
 680
 681                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 682                         rc = rc2;
 683         }
 684
 685 out:
 686         return rc;
 687 }
 688
 689 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 690                       const char *name)
 691 {
 692         const struct lu_env *env = info->mti_env;
 693         int rc;
 694         ENTRY;
 695
 696         LASSERT(info->mti_big_lmm_used == 0);
 697         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 698         if (rc < 0)
 699                 RETURN(rc);
 700
 701         /* big_lmm may need to be grown */
 702         if (info->mti_big_lmmsize < rc) {
 703                 int size = size_roundup_power2(rc);
 704
 705                 if (info->mti_big_lmmsize > 0) {
 706                         /* free old buffer */
 707                         LASSERT(info->mti_big_lmm);
 708                         OBD_FREE_LARGE(info->mti_big_lmm,
 709                                        info->mti_big_lmmsize);
 710                         info->mti_big_lmm = NULL;
 711                         info->mti_big_lmmsize = 0;
 712                 }
 713
 714                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 715                 if (info->mti_big_lmm == NULL)
 716                         RETURN(-ENOMEM);
 717                 info->mti_big_lmmsize = size;
 718         }
 719         LASSERT(info->mti_big_lmmsize >= rc);
 720
 721         info->mti_buf.lb_buf = info->mti_big_lmm;
 722         info->mti_buf.lb_len = info->mti_big_lmmsize;
 723         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 724
 725         RETURN(rc);
 726 }
 727
 728 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 729                    struct md_attr *ma, const char *name)
 730 {
 731         struct md_object *next = mdt_object_child(o);
 732         struct lu_buf    *buf = &info->mti_buf;
 733         int rc;
 734
 735         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 736                 buf->lb_buf = ma->ma_lmm;
 737                 buf->lb_len = ma->ma_lmm_size;
 738                 LASSERT(!(ma->ma_valid & MA_LOV));
 739         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 740                 buf->lb_buf = ma->ma_lmv;
 741                 buf->lb_len = ma->ma_lmv_size;
 742                 LASSERT(!(ma->ma_valid & MA_LMV));
 743         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 744                 buf->lb_buf = ma->ma_lmv;
 745                 buf->lb_len = ma->ma_lmv_size;
 746                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 747         } else {
 748                 return -EINVAL;
 749         }
 750
 751         rc = mo_xattr_get(info->mti_env, next, buf, name);
 752         if (rc > 0) {
 753
 754 got:
 755                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 756                         if (info->mti_big_lmm_used)
 757                                 ma->ma_lmm = info->mti_big_lmm;
 758
 759                         /* NOT return LOV EA with hole to old client. */
 760                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 761                                      LOV_PATTERN_F_HOLE) &&
 762                             !(exp_connect_flags(info->mti_exp) &
 763                               OBD_CONNECT_LFSCK)) {
 764                                 return -EIO;
 765                         } else {
 766                                 ma->ma_lmm_size = rc;
 767                                 ma->ma_valid |= MA_LOV;
 768                         }
 769                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 770                         if (info->mti_big_lmm_used)
 771                                 ma->ma_lmv = info->mti_big_lmm;
 772
 773                         ma->ma_lmv_size = rc;
 774                         ma->ma_valid |= MA_LMV;
 775                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 776                         ma->ma_lmv_size = rc;
 777                         ma->ma_valid |= MA_LMV_DEF;
 778                 }
 779
 780                 /* Update mdt_max_mdsize so all clients will be aware that */
 781                 if (info->mti_mdt->mdt_max_mdsize < rc)
 782                         info->mti_mdt->mdt_max_mdsize = rc;
 783
 784                 rc = 0;
 785         } else if (rc == -ENODATA) {
 786                 /* no LOV EA */
 787                 rc = 0;
 788         } else if (rc == -ERANGE) {
 789                 /* Default LMV has fixed size, so it must be able to fit
 790                  * in the original buffer */
 791                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 792                         return rc;
 793                 rc = mdt_big_xattr_get(info, o, name);
 794                 if (rc > 0) {
 795                         info->mti_big_lmm_used = 1;
 796                         goto got;
 797                 }
 798         }
 799
 800         return rc;
 801 }
 802
 803 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 804                              struct mdt_object *o, struct lu_fid *pfid)
 805 {
 806         struct lu_buf           *buf = &info->mti_buf;
 807         struct link_ea_header   *leh;
 808         struct link_ea_entry    *lee;
 809         int                      rc;
 810         ENTRY;
 811
 812         buf->lb_buf = info->mti_big_lmm;
 813         buf->lb_len = info->mti_big_lmmsize;
 814         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 815                           buf, XATTR_NAME_LINK);
 816         /* ignore errors, MA_PFID won't be set and it is
 817          * up to the caller to treat this as an error */
 818         if (rc == -ERANGE || buf->lb_len == 0) {
 819                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 820                 buf->lb_buf = info->mti_big_lmm;
 821                 buf->lb_len = info->mti_big_lmmsize;
 822         }
 823
 824         if (rc < 0)
 825                 RETURN(rc);
 826         if (rc < sizeof(*leh)) {
 827                 CERROR("short LinkEA on "DFID": rc = %d\n",
 828                        PFID(mdt_object_fid(o)), rc);
 829                 RETURN(-ENODATA);
 830         }
 831
 832         leh = (struct link_ea_header *) buf->lb_buf;
 833         lee = (struct link_ea_entry *)(leh + 1);
 834         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 835                 leh->leh_magic = LINK_EA_MAGIC;
 836                 leh->leh_reccount = __swab32(leh->leh_reccount);
 837                 leh->leh_len = __swab64(leh->leh_len);
 838         }
 839         if (leh->leh_magic != LINK_EA_MAGIC)
 840                 RETURN(-EINVAL);
 841         if (leh->leh_reccount == 0)
 842                 RETURN(-ENODATA);
 843
 844         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 845         fid_be_to_cpu(pfid, pfid);
 846
 847         RETURN(0);
 848 }
 849
 850 int mdt_attr_get_complex(struct mdt_thread_info *info,
 851                          struct mdt_object *o, struct md_attr *ma)
 852 {
 853         const struct lu_env *env = info->mti_env;
 854         struct md_object    *next = mdt_object_child(o);
 855         struct lu_buf       *buf = &info->mti_buf;
 856         int                  need = ma->ma_need;
 857         int                  rc = 0, rc2;
 858         u32                  mode;
 859         ENTRY;
 860
 861         ma->ma_valid = 0;
 862
 863         if (mdt_object_exists(o) == 0)
 864                 GOTO(out, rc = -ENOENT);
 865         mode = lu_object_attr(&next->mo_lu);
 866
 867         if (need & MA_INODE) {
 868                 ma->ma_need = MA_INODE;
 869                 rc = mo_attr_get(env, next, ma);
 870                 if (rc)
 871                         GOTO(out, rc);
 872                 ma->ma_valid |= MA_INODE;
 873         }
 874
 875         if (need & MA_PFID) {
 876                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 877                 if (rc == 0)
 878                         ma->ma_valid |= MA_PFID;
 879                 /* ignore this error, parent fid is not mandatory */
 880                 rc = 0;
 881         }
 882
 883         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 884                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 885                 if (rc)
 886                         GOTO(out, rc);
 887         }
 888
 889         if (need & MA_LMV && S_ISDIR(mode)) {
 890                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 891                 if (rc != 0)
 892                         GOTO(out, rc);
 893         }
 894
 895         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 896                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 897                 if (rc != 0)
 898                         GOTO(out, rc);
 899         }
 900
 901         if (need & MA_HSM && S_ISREG(mode)) {
 902                 buf->lb_buf = info->mti_xattr_buf;
 903                 buf->lb_len = sizeof(info->mti_xattr_buf);
 904                 CLASSERT(sizeof(struct hsm_attrs) <=
 905                          sizeof(info->mti_xattr_buf));
 906                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 907                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 908                 if (rc2 == 0)
 909                         ma->ma_valid |= MA_HSM;
 910                 else if (rc2 < 0 && rc2 != -ENODATA)
 911                         GOTO(out, rc = rc2);
 912         }
 913
 914 #ifdef CONFIG_FS_POSIX_ACL
 915         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 916                 buf->lb_buf = ma->ma_acl;
 917                 buf->lb_len = ma->ma_acl_size;
 918                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 919                 if (rc2 > 0) {
 920                         ma->ma_acl_size = rc2;
 921                         ma->ma_valid |= MA_ACL_DEF;
 922                 } else if (rc2 == -ENODATA) {
 923                         /* no ACLs */
 924                         ma->ma_acl_size = 0;
 925                 } else
 926                         GOTO(out, rc = rc2);
 927         }
 928 #endif
 929 out:
 930         ma->ma_need = need;
 931         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
 932                rc, ma->ma_valid, ma->ma_lmm);
 933         RETURN(rc);
 934 }
 935
 936 static int mdt_getattr_internal(struct mdt_thread_info *info,
 937                                 struct mdt_object *o, int ma_need)
 938 {
 939         struct md_object        *next = mdt_object_child(o);
 940         const struct mdt_body   *reqbody = info->mti_body;
 941         struct ptlrpc_request   *req = mdt_info_req(info);
 942         struct md_attr          *ma = &info->mti_attr;
 943         struct lu_attr          *la = &ma->ma_attr;
 944         struct req_capsule      *pill = info->mti_pill;
 945         const struct lu_env     *env = info->mti_env;
 946         struct mdt_body         *repbody;
 947         struct lu_buf           *buffer = &info->mti_buf;
 948         struct obd_export       *exp = info->mti_exp;
 949         int                      rc;
 950         int                      is_root;
 951         ENTRY;
 952
 953         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 954                 RETURN(err_serious(-ENOMEM));
 955
 956         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 957
 958         ma->ma_valid = 0;
 959
 960         if (mdt_object_remote(o)) {
 961                 /* This object is located on remote node.*/
 962                 /* Return -ENOTSUPP for old client */
 963                 if (!mdt_is_dne_client(req->rq_export))
 964                         GOTO(out, rc = -ENOTSUPP);
 965
 966                 repbody->mbo_fid1 = *mdt_object_fid(o);
 967                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 968                 GOTO(out, rc = 0);
 969         }
 970
 971         if (reqbody->mbo_eadatasize > 0) {
 972                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 973                 if (buffer->lb_buf == NULL)
 974                         GOTO(out, rc = -EPROTO);
 975                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 976                                                       RCL_SERVER);
 977         } else {
 978                 buffer->lb_buf = NULL;
 979                 buffer->lb_len = 0;
 980                 ma_need &= ~(MA_LOV | MA_LMV);
 981                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 982                        mdt_obd_name(info->mti_mdt),
 983                        req->rq_export->exp_client_uuid.uuid);
 984         }
 985
 986         /* If it is dir object and client require MEA, then we got MEA */
 987         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 988             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 989                 /* Assumption: MDT_MD size is enough for lmv size. */
 990                 ma->ma_lmv = buffer->lb_buf;
 991                 ma->ma_lmv_size = buffer->lb_len;
 992                 ma->ma_need = MA_INODE;
 993                 if (ma->ma_lmv_size > 0) {
 994                         if (reqbody->mbo_valid & OBD_MD_MEA)
 995                                 ma->ma_need |= MA_LMV;
 996                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 997                                 ma->ma_need |= MA_LMV_DEF;
 998                 }
 999         } else {
1000                 ma->ma_lmm = buffer->lb_buf;
1001                 ma->ma_lmm_size = buffer->lb_len;
1002                 ma->ma_need = MA_INODE | MA_HSM;
1003                 if (ma->ma_lmm_size > 0)
1004                         ma->ma_need |= MA_LOV;
1005         }
1006
1007         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1008             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1009             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1010                 /* get default stripe info for this dir. */
1011                 ma->ma_need |= MA_LOV_DEF;
1012         }
1013         ma->ma_need |= ma_need;
1014
1015         rc = mdt_attr_get_complex(info, o, ma);
1016         if (unlikely(rc)) {
1017                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1018                        "%s: getattr error for "DFID": rc = %d\n",
1019                        mdt_obd_name(info->mti_mdt),
1020                        PFID(mdt_object_fid(o)), rc);
1021                 RETURN(rc);
1022         }
1023
1024         /* if file is released, check if a restore is running */
1025         if (ma->ma_valid & MA_HSM) {
1026                 repbody->mbo_valid |= OBD_MD_TSTATE;
1027                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1028                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1029                         repbody->mbo_t_state = MS_RESTORE;
1030         }
1031
1032         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
1033
1034         /* the Lustre protocol supposes to return default striping
1035          * on the user-visible root if explicitly requested */
1036         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
1037             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
1038                 struct lu_fid      rootfid;
1039                 struct mdt_object *root;
1040                 struct mdt_device *mdt = info->mti_mdt;
1041
1042                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1043                 if (rc)
1044                         RETURN(rc);
1045                 root = mdt_object_find(env, mdt, &rootfid);
1046                 if (IS_ERR(root))
1047                         RETURN(PTR_ERR(root));
1048                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1049                 mdt_object_put(info->mti_env, root);
1050                 if (unlikely(rc)) {
1051                         CERROR("%s: getattr error for "DFID": rc = %d\n",
1052                                mdt_obd_name(info->mti_mdt),
1053                                PFID(mdt_object_fid(o)), rc);
1054                         RETURN(rc);
1055                 }
1056         }
1057
1058         if (likely(ma->ma_valid & MA_INODE))
1059                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1060         else
1061                 RETURN(-EFAULT);
1062
1063         if (mdt_body_has_lov(la, reqbody)) {
1064                 if (ma->ma_valid & MA_LOV) {
1065                         LASSERT(ma->ma_lmm_size);
1066                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1067                         if (S_ISDIR(la->la_mode))
1068                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1069                         else
1070                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1071                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1072                 }
1073                 if (ma->ma_valid & MA_LMV) {
1074                         /* Return -ENOTSUPP for old client */
1075                         if (!mdt_is_striped_client(req->rq_export))
1076                                 RETURN(-ENOTSUPP);
1077
1078                         LASSERT(S_ISDIR(la->la_mode));
1079                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1080                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1081                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1082                 }
1083                 if (ma->ma_valid & MA_LMV_DEF) {
1084                         /* Return -ENOTSUPP for old client */
1085                         if (!mdt_is_striped_client(req->rq_export))
1086                                 RETURN(-ENOTSUPP);
1087                         LASSERT(S_ISDIR(la->la_mode));
1088                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1089                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1090                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1091                                                OBD_MD_DEFAULT_MEA);
1092                 }
1093         } else if (S_ISLNK(la->la_mode) &&
1094                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1095                 buffer->lb_buf = ma->ma_lmm;
1096                 /* eadatasize from client includes NULL-terminator, so
1097                  * there is no need to read it */
1098                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1099                 rc = mo_readlink(env, next, buffer);
1100                 if (unlikely(rc <= 0)) {
1101                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1102                                mdt_obd_name(info->mti_mdt),
1103                                PFID(mdt_object_fid(o)), rc);
1104                         rc = -EFAULT;
1105                 } else {
1106                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1107
1108                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1109                                 rc -= 2;
1110                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1111                         /* we need to report back size with NULL-terminator
1112                          * because client expects that */
1113                         repbody->mbo_eadatasize = rc + 1;
1114                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1115                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1116                                        "on "DFID ", expected %d\n",
1117                                        mdt_obd_name(info->mti_mdt),
1118                                        rc, PFID(mdt_object_fid(o)),
1119                                        reqbody->mbo_eadatasize - 1);
1120                         /* NULL terminate */
1121                         ((char *)ma->ma_lmm)[rc] = 0;
1122
1123                         /* If the total CDEBUG() size is larger than a page, it
1124                          * will print a warning to the console, avoid this by
1125                          * printing just the last part of the symlink. */
1126                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1127                                print_limit < rc ? "..." : "", print_limit,
1128                                (char *)ma->ma_lmm + rc - print_limit, rc);
1129                         rc = 0;
1130                 }
1131         }
1132
1133         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1134                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1135                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1136                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1137                        repbody->mbo_max_mdsize);
1138         }
1139
1140 #ifdef CONFIG_FS_POSIX_ACL
1141         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1142                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1143                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1144                 if (IS_ERR(nodemap))
1145                         RETURN(PTR_ERR(nodemap));
1146
1147                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1148                 nodemap_putref(nodemap);
1149         }
1150 #endif
1151
1152 out:
1153         if (rc == 0)
1154                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1155
1156         RETURN(rc);
1157 }
1158
1159 static int mdt_getattr(struct tgt_session_info *tsi)
1160 {
1161         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1162         struct mdt_object       *obj = info->mti_object;
1163         struct req_capsule      *pill = info->mti_pill;
1164         struct mdt_body         *reqbody;
1165         struct mdt_body         *repbody;
1166         int rc, rc2;
1167         ENTRY;
1168
1169         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1170         LASSERT(reqbody);
1171         LASSERT(obj != NULL);
1172         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1173
1174         /* Unlike intent case where we need to pre-fill out buffers early on
1175          * in intent policy for ldlm reasons, here we can have a much better
1176          * guess at EA size by just reading it from disk.
1177          * Exceptions are readdir and (missing) directory striping */
1178         /* Readlink */
1179         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1180                 /* No easy way to know how long is the symlink, but it cannot
1181                  * be more than PATH_MAX, so we allocate +1 */
1182                 rc = PATH_MAX + 1;
1183
1184         /* A special case for fs ROOT: getattr there might fetch
1185          * default EA for entire fs, not just for this dir!
1186          */
1187         } else if (lu_fid_eq(mdt_object_fid(obj),
1188                              &info->mti_mdt->mdt_md_root_fid) &&
1189                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1190                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1191                                                                  MDS_GETATTR)) {
1192                 /* Should the default strping be bigger, mdt_fix_reply
1193                  * will reallocate */
1194                 rc = DEF_REP_MD_SIZE;
1195         } else {
1196                 /* Read the actual EA size from disk */
1197                 rc = mdt_attr_get_eabuf_size(info, obj);
1198         }
1199
1200         if (rc < 0)
1201                 GOTO(out, rc = err_serious(rc));
1202
1203         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1204
1205         rc = req_capsule_server_pack(pill);
1206         if (unlikely(rc != 0))
1207                 GOTO(out, rc = err_serious(rc));
1208
1209         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1210         LASSERT(repbody != NULL);
1211         repbody->mbo_eadatasize = 0;
1212         repbody->mbo_aclsize = 0;
1213
1214         rc = mdt_check_ucred(info);
1215         if (unlikely(rc))
1216                 GOTO(out_shrink, rc);
1217
1218         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1219
1220         rc = mdt_getattr_internal(info, obj, 0);
1221         EXIT;
1222 out_shrink:
1223         mdt_client_compatibility(info);
1224         rc2 = mdt_fix_reply(info);
1225         if (rc == 0)
1226                 rc = rc2;
1227 out:
1228         mdt_thread_info_fini(info);
1229         return rc;
1230 }
1231
1232 /**
1233  * Exchange MOF_LOV_CREATED flags between two objects after a
1234  * layout swap. No assumption is made on whether o1 or o2 have
1235  * created objects or not.
1236  *
1237  * \param[in,out] o1    First swap layout object
1238  * \param[in,out] o2    Second swap layout object
1239  */
1240 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1241 {
1242         unsigned int o1_lov_created = o1->mot_lov_created;
1243
1244         mutex_lock(&o1->mot_lov_mutex);
1245         mutex_lock(&o2->mot_lov_mutex);
1246
1247         o1->mot_lov_created = o2->mot_lov_created;
1248         o2->mot_lov_created = o1_lov_created;
1249
1250         mutex_unlock(&o2->mot_lov_mutex);
1251         mutex_unlock(&o1->mot_lov_mutex);
1252 }
1253
1254 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1255 {
1256         struct mdt_thread_info  *info;
1257         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1258         struct obd_export       *exp = req->rq_export;
1259         struct mdt_object       *o1, *o2, *o;
1260         struct mdt_lock_handle  *lh1, *lh2;
1261         struct mdc_swap_layouts *msl;
1262         int                      rc;
1263         ENTRY;
1264
1265         /* client does not support layout lock, so layout swaping
1266          * is disabled.
1267          * FIXME: there is a problem for old clients which don't support
1268          * layout lock yet. If those clients have already opened the file
1269          * they won't be notified at all so that old layout may still be
1270          * used to do IO. This can be fixed after file release is landed by
1271          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1272         if (!exp_connect_layout(exp))
1273                 RETURN(-EOPNOTSUPP);
1274
1275         info = tsi2mdt_info(tsi);
1276
1277         if (info->mti_dlm_req != NULL)
1278                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1279
1280         o1 = info->mti_object;
1281         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1282                                 &info->mti_body->mbo_fid2);
1283         if (IS_ERR(o))
1284                 GOTO(out, rc = PTR_ERR(o));
1285
1286         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1287                 GOTO(put, rc = -ENOENT);
1288
1289         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1290         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1291                 GOTO(put, rc);
1292
1293         if (rc < 0)
1294                 swap(o1, o2);
1295
1296         /* permission check. Make sure the calling process having permission
1297          * to write both files. */
1298         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1299                                 MAY_WRITE);
1300         if (rc < 0)
1301                 GOTO(put, rc);
1302
1303         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1304                                 MAY_WRITE);
1305         if (rc < 0)
1306                 GOTO(put, rc);
1307
1308         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1309         if (msl == NULL)
1310                 GOTO(put, rc = -EPROTO);
1311
1312         lh1 = &info->mti_lh[MDT_LH_NEW];
1313         mdt_lock_reg_init(lh1, LCK_EX);
1314         lh2 = &info->mti_lh[MDT_LH_OLD];
1315         mdt_lock_reg_init(lh2, LCK_EX);
1316
1317         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1318                              MDS_INODELOCK_XATTR);
1319         if (rc < 0)
1320                 GOTO(put, rc);
1321
1322         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1323                              MDS_INODELOCK_XATTR);
1324         if (rc < 0)
1325                 GOTO(unlock1, rc);
1326
1327         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1328                              mdt_object_child(o2), msl->msl_flags);
1329         if (rc < 0)
1330                 GOTO(unlock2, rc);
1331
1332         mdt_swap_lov_flag(o1, o2);
1333
1334 unlock2:
1335         mdt_object_unlock(info, o2, lh2, rc);
1336 unlock1:
1337         mdt_object_unlock(info, o1, lh1, rc);
1338 put:
1339         mdt_object_put(info->mti_env, o);
1340 out:
1341         mdt_thread_info_fini(info);
1342         RETURN(rc);
1343 }
1344
1345 static int mdt_raw_lookup(struct mdt_thread_info *info,
1346                           struct mdt_object *parent,
1347                           const struct lu_name *lname,
1348                           struct ldlm_reply *ldlm_rep)
1349 {
1350         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1351         int              rc;
1352         ENTRY;
1353
1354         LASSERT(!info->mti_cross_ref);
1355
1356         /* Only got the fid of this obj by name */
1357         fid_zero(child_fid);
1358         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1359                         lname, child_fid, &info->mti_spec);
1360         if (rc == 0) {
1361                 struct mdt_body *repbody;
1362
1363                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1364                 repbody->mbo_fid1 = *child_fid;
1365                 repbody->mbo_valid = OBD_MD_FLID;
1366                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1367         } else if (rc == -ENOENT) {
1368                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1369         }
1370
1371         RETURN(rc);
1372 }
1373
1374 /*
1375  * UPDATE lock should be taken against parent, and be released before exit;
1376  * child_bits lock should be taken against child, and be returned back:
1377  *            (1)normal request should release the child lock;
1378  *            (2)intent request will grant the lock to client.
1379  */
1380 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1381                                  struct mdt_lock_handle *lhc,
1382                                  __u64 child_bits,
1383                                  struct ldlm_reply *ldlm_rep)
1384 {
1385         struct ptlrpc_request  *req       = mdt_info_req(info);
1386         struct mdt_body        *reqbody   = NULL;
1387         struct mdt_object      *parent    = info->mti_object;
1388         struct mdt_object      *child;
1389         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1390         struct lu_name         *lname     = NULL;
1391         struct mdt_lock_handle *lhp       = NULL;
1392         struct ldlm_lock       *lock;
1393         bool                    is_resent;
1394         bool                    try_layout;
1395         int                     ma_need = 0;
1396         int                     rc;
1397         ENTRY;
1398
1399         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1400         LASSERT(ergo(is_resent,
1401                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1402
1403         if (parent == NULL)
1404                 RETURN(-ENOENT);
1405
1406         if (info->mti_cross_ref) {
1407                 /* Only getattr on the child. Parent is on another node. */
1408                 mdt_set_disposition(info, ldlm_rep,
1409                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1410                 child = parent;
1411                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1412                        "ldlm_rep = %p\n",
1413                        PFID(mdt_object_fid(child)), ldlm_rep);
1414
1415                 rc = mdt_check_resent_lock(info, child, lhc);
1416                 if (rc < 0) {
1417                         RETURN(rc);
1418                 } else if (rc > 0) {
1419                         mdt_lock_handle_init(lhc);
1420                         mdt_lock_reg_init(lhc, LCK_PR);
1421
1422                         /*
1423                          * Object's name is on another MDS, no lookup or layout
1424                          * lock is needed here but update lock is.
1425                          */
1426                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1427                                         MDS_INODELOCK_LAYOUT);
1428                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1429
1430                         rc = mdt_object_lock(info, child, lhc, child_bits);
1431                         if (rc < 0)
1432                                 RETURN(rc);
1433                 }
1434
1435                 /* Finally, we can get attr for child. */
1436                 if (!mdt_object_exists(child)) {
1437                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1438                                         &child->mot_obj,
1439                                         "remote object doesn't exist.");
1440                         mdt_object_unlock(info, child, lhc, 1);
1441                         RETURN(-ENOENT);
1442                 }
1443
1444                 rc = mdt_getattr_internal(info, child, 0);
1445                 if (unlikely(rc != 0))
1446                         mdt_object_unlock(info, child, lhc, 1);
1447
1448                 RETURN(rc);
1449         }
1450
1451         lname = &info->mti_name;
1452         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1453
1454         if (lu_name_is_valid(lname)) {
1455                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1456                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1457                        PNAME(lname), ldlm_rep);
1458         } else {
1459                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1460                 if (unlikely(reqbody == NULL))
1461                         RETURN(err_serious(-EPROTO));
1462
1463                 *child_fid = reqbody->mbo_fid2;
1464
1465                 if (unlikely(!fid_is_sane(child_fid)))
1466                         RETURN(err_serious(-EINVAL));
1467
1468                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1469                        "ldlm_rep = %p\n",
1470                        PFID(mdt_object_fid(parent)),
1471                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1472         }
1473
1474         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1475
1476         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1477                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1478                                 &parent->mot_obj,
1479                                 "Parent doesn't exist!");
1480                 RETURN(-ESTALE);
1481         }
1482
1483         if (mdt_object_remote(parent)) {
1484                 CERROR("%s: parent "DFID" is on remote target\n",
1485                        mdt_obd_name(info->mti_mdt),
1486                        PFID(mdt_object_fid(parent)));
1487                 RETURN(-EIO);
1488         }
1489
1490         if (lu_name_is_valid(lname)) {
1491                 /* Always allow to lookup ".." */
1492                 if (unlikely(lname->ln_namelen == 2 &&
1493                              lname->ln_name[0] == '.' &&
1494                              lname->ln_name[1] == '.'))
1495                         info->mti_spec.sp_permitted = 1;
1496
1497                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1498                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1499
1500                         RETURN(rc);
1501                 }
1502
1503                 /* step 1: lock parent only if parent is a directory */
1504                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1505                         lhp = &info->mti_lh[MDT_LH_PARENT];
1506                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1507                         rc = mdt_object_lock(info, parent, lhp,
1508                                              MDS_INODELOCK_UPDATE);
1509                         if (unlikely(rc != 0))
1510                                 RETURN(rc);
1511                 }
1512
1513                 /* step 2: lookup child's fid by name */
1514                 fid_zero(child_fid);
1515                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1516                                 child_fid, &info->mti_spec);
1517                 if (rc == -ENOENT)
1518                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1519
1520                 if (rc != 0)
1521                         GOTO(out_parent, rc);
1522         }
1523
1524         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1525
1526         /*
1527          *step 3: find the child object by fid & lock it.
1528          *        regardless if it is local or remote.
1529          *
1530          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1531          *      set parent dir fid the same as child fid in getattr by fid case
1532          *      we should not lu_object_find() the object again, could lead
1533          *      to hung if there is a concurrent unlink destroyed the object.
1534          */
1535         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1536                 mdt_object_get(info->mti_env, parent);
1537                 child = parent;
1538         } else {
1539                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1540                                         child_fid);
1541         }
1542
1543         if (unlikely(IS_ERR(child)))
1544                 GOTO(out_parent, rc = PTR_ERR(child));
1545
1546         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1547         if (!mdt_object_exists(child)) {
1548                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1549                                 &child->mot_obj,
1550                                 "Object doesn't exist!");
1551                 GOTO(out_child, rc = -ENOENT);
1552         }
1553
1554         rc = mdt_check_resent_lock(info, child, lhc);
1555         if (rc < 0) {
1556                 GOTO(out_child, rc);
1557         } else if (rc > 0) {
1558                 mdt_lock_handle_init(lhc);
1559                 mdt_lock_reg_init(lhc, LCK_PR);
1560                 try_layout = false;
1561
1562                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1563                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1564                         struct md_attr *ma = &info->mti_attr;
1565
1566                         ma->ma_valid = 0;
1567                         ma->ma_need = MA_INODE;
1568                         rc = mdt_attr_get_complex(info, child, ma);
1569                         if (unlikely(rc != 0))
1570                                 GOTO(out_child, rc);
1571
1572                         /* If the file has not been changed for some time, we
1573                          * return not only a LOOKUP lock, but also an UPDATE
1574                          * lock and this might save us RPC on later STAT. For
1575                          * directories, it also let negative dentry cache start
1576                          * working for this dir. */
1577                         if (ma->ma_valid & MA_INODE &&
1578                             ma->ma_attr.la_valid & LA_CTIME &&
1579                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1580                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1581                                 child_bits |= MDS_INODELOCK_UPDATE;
1582                 }
1583
1584                 /* layout lock must be granted in a best-effort way
1585                  * for IT operations */
1586                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1587                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1588                     exp_connect_layout(info->mti_exp) &&
1589                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1590                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1591                         /* try to grant layout lock for regular file. */
1592                         try_layout = true;
1593                 }
1594
1595                 rc = 0;
1596                 if (try_layout) {
1597                         child_bits |= MDS_INODELOCK_LAYOUT;
1598                         /* try layout lock, it may fail to be granted due to
1599                          * contention at LOOKUP or UPDATE */
1600                         if (!mdt_object_lock_try(info, child, lhc,
1601                                                  child_bits)) {
1602                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1603                                 LASSERT(child_bits != 0);
1604                                 rc = mdt_object_lock(info, child, lhc,
1605                                                      child_bits);
1606                         } else {
1607                                 ma_need |= MA_LOV;
1608                         }
1609                 } else {
1610                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1611                          * client will enqueue the lock to the remote MDT */
1612                         if (mdt_object_remote(child))
1613                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1614                         rc = mdt_object_lock(info, child, lhc, child_bits);
1615                 }
1616                 if (unlikely(rc != 0))
1617                         GOTO(out_child, rc);
1618         }
1619
1620         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1621
1622         /* finally, we can get attr for child. */
1623         rc = mdt_getattr_internal(info, child, ma_need);
1624         if (unlikely(rc != 0)) {
1625                 mdt_object_unlock(info, child, lhc, 1);
1626         } else if (lock) {
1627                 /* Debugging code. */
1628                 LDLM_DEBUG(lock, "Returning lock to client");
1629                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1630                                          &lock->l_resource->lr_name),
1631                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1632                          PLDLMRES(lock->l_resource),
1633                          PFID(mdt_object_fid(child)));
1634         }
1635         if (lock)
1636                 LDLM_LOCK_PUT(lock);
1637
1638         EXIT;
1639 out_child:
1640         mdt_object_put(info->mti_env, child);
1641 out_parent:
1642         if (lhp)
1643                 mdt_object_unlock(info, parent, lhp, 1);
1644         return rc;
1645 }
1646
1647 /* normal handler: should release the child lock */
1648 static int mdt_getattr_name(struct tgt_session_info *tsi)
1649 {
1650         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1651         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1652         struct mdt_body        *reqbody;
1653         struct mdt_body        *repbody;
1654         int rc, rc2;
1655         ENTRY;
1656
1657         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1658         LASSERT(reqbody != NULL);
1659         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1660         LASSERT(repbody != NULL);
1661
1662         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1663         repbody->mbo_eadatasize = 0;
1664         repbody->mbo_aclsize = 0;
1665
1666         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1667         if (unlikely(rc))
1668                 GOTO(out_shrink, rc);
1669
1670         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1671         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1672                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1673                 lhc->mlh_reg_lh.cookie = 0;
1674         }
1675         mdt_exit_ucred(info);
1676         EXIT;
1677 out_shrink:
1678         mdt_client_compatibility(info);
1679         rc2 = mdt_fix_reply(info);
1680         if (rc == 0)
1681                 rc = rc2;
1682         mdt_thread_info_fini(info);
1683         return rc;
1684 }
1685
1686 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1687                          void *karg, void __user *uarg);
1688
1689 static int mdt_set_info(struct tgt_session_info *tsi)
1690 {
1691         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1692         char                    *key;
1693         void                    *val;
1694         int                      keylen, vallen, rc = 0;
1695
1696         ENTRY;
1697
1698         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1699         if (key == NULL) {
1700                 DEBUG_REQ(D_HA, req, "no set_info key");
1701                 RETURN(err_serious(-EFAULT));
1702         }
1703
1704         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1705                                       RCL_CLIENT);
1706
1707         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1708         if (val == NULL) {
1709                 DEBUG_REQ(D_HA, req, "no set_info val");
1710                 RETURN(err_serious(-EFAULT));
1711         }
1712
1713         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1714                                       RCL_CLIENT);
1715
1716         /* Swab any part of val you need to here */
1717         if (KEY_IS(KEY_READ_ONLY)) {
1718                 spin_lock(&req->rq_export->exp_lock);
1719                 if (*(__u32 *)val)
1720                         *exp_connect_flags_ptr(req->rq_export) |=
1721                                 OBD_CONNECT_RDONLY;
1722                 else
1723                         *exp_connect_flags_ptr(req->rq_export) &=
1724                                 ~OBD_CONNECT_RDONLY;
1725                 spin_unlock(&req->rq_export->exp_lock);
1726         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1727                 struct changelog_setinfo *cs = val;
1728
1729                 if (vallen != sizeof(*cs)) {
1730                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1731                                tgt_name(tsi->tsi_tgt), vallen);
1732                         RETURN(-EINVAL);
1733                 }
1734                 if (ptlrpc_req_need_swab(req)) {
1735                         __swab64s(&cs->cs_recno);
1736                         __swab32s(&cs->cs_id);
1737                 }
1738
1739                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1740                                    vallen, val, NULL);
1741         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1742                 if (vallen > 0)
1743                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1744         } else {
1745                 RETURN(-EINVAL);
1746         }
1747         RETURN(rc);
1748 }
1749
1750 static int mdt_readpage(struct tgt_session_info *tsi)
1751 {
1752         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1753         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1754         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1755         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1756         struct mdt_body         *repbody;
1757         int                      rc;
1758         int                      i;
1759
1760         ENTRY;
1761
1762         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1763                 RETURN(err_serious(-ENOMEM));
1764
1765         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1766         if (repbody == NULL || reqbody == NULL)
1767                 RETURN(err_serious(-EFAULT));
1768
1769         /*
1770          * prepare @rdpg before calling lower layers and transfer itself. Here
1771          * reqbody->size contains offset of where to start to read and
1772          * reqbody->nlink contains number bytes to read.
1773          */
1774         rdpg->rp_hash = reqbody->mbo_size;
1775         if (rdpg->rp_hash != reqbody->mbo_size) {
1776                 CERROR("Invalid hash: %#llx != %#llx\n",
1777                        rdpg->rp_hash, reqbody->mbo_size);
1778                 RETURN(-EFAULT);
1779         }
1780
1781         rdpg->rp_attrs = reqbody->mbo_mode;
1782         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1783                 rdpg->rp_attrs |= LUDA_64BITHASH;
1784         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1785                                 exp_max_brw_size(tsi->tsi_exp));
1786         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1787                           PAGE_SHIFT;
1788         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1789         if (rdpg->rp_pages == NULL)
1790                 RETURN(-ENOMEM);
1791
1792         for (i = 0; i < rdpg->rp_npages; ++i) {
1793                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1794                 if (rdpg->rp_pages[i] == NULL)
1795                         GOTO(free_rdpg, rc = -ENOMEM);
1796         }
1797
1798         /* call lower layers to fill allocated pages with directory data */
1799         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1800         if (rc < 0)
1801                 GOTO(free_rdpg, rc);
1802
1803         /* send pages to client */
1804         rc = tgt_sendpage(tsi, rdpg, rc);
1805
1806         EXIT;
1807 free_rdpg:
1808
1809         for (i = 0; i < rdpg->rp_npages; i++)
1810                 if (rdpg->rp_pages[i] != NULL)
1811                         __free_page(rdpg->rp_pages[i]);
1812         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1813
1814         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1815                 RETURN(0);
1816
1817         return rc;
1818 }
1819
1820 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1821 {
1822         struct lu_ucred *uc = mdt_ucred_check(info);
1823         struct lu_attr *attr = &info->mti_attr.ma_attr;
1824
1825         if (uc == NULL)
1826                 return -EINVAL;
1827
1828         if (op != REINT_SETATTR) {
1829                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1830                         attr->la_uid = uc->uc_fsuid;
1831                 /* for S_ISGID, inherit gid from his parent, such work will be
1832                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1833                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1834                         attr->la_gid = uc->uc_fsgid;
1835         }
1836
1837         return 0;
1838 }
1839
1840 static int mdt_reint_internal(struct mdt_thread_info *info,
1841                               struct mdt_lock_handle *lhc,
1842                               __u32 op)
1843 {
1844         struct req_capsule      *pill = info->mti_pill;
1845         struct mdt_body         *repbody;
1846         int                      rc = 0, rc2;
1847
1848         ENTRY;
1849
1850         rc = mdt_reint_unpack(info, op);
1851         if (rc != 0) {
1852                 CERROR("Can't unpack reint, rc %d\n", rc);
1853                 RETURN(err_serious(rc));
1854         }
1855
1856         /* for replay (no_create) lmm is not needed, client has it already */
1857         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1858                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1859                                      DEF_REP_MD_SIZE);
1860
1861         /* llog cookies are always 0, the field is kept for compatibility */
1862         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1863                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1864
1865         rc = req_capsule_server_pack(pill);
1866         if (rc != 0) {
1867                 CERROR("Can't pack response, rc %d\n", rc);
1868                 RETURN(err_serious(rc));
1869         }
1870
1871         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1872                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1873                 LASSERT(repbody);
1874                 repbody->mbo_eadatasize = 0;
1875                 repbody->mbo_aclsize = 0;
1876         }
1877
1878         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1879
1880         /* for replay no cookkie / lmm need, because client have this already */
1881         if (info->mti_spec.no_create)
1882                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1883                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1884
1885         rc = mdt_init_ucred_reint(info);
1886         if (rc)
1887                 GOTO(out_shrink, rc);
1888
1889         rc = mdt_fix_attr_ucred(info, op);
1890         if (rc != 0)
1891                 GOTO(out_ucred, rc = err_serious(rc));
1892
1893         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1894         if (rc < 0) {
1895                 GOTO(out_ucred, rc);
1896         } else if (rc == 1) {
1897                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1898                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1899                 GOTO(out_ucred, rc);
1900         }
1901         rc = mdt_reint_rec(info, lhc);
1902         EXIT;
1903 out_ucred:
1904         mdt_exit_ucred(info);
1905 out_shrink:
1906         mdt_client_compatibility(info);
1907         rc2 = mdt_fix_reply(info);
1908         if (rc == 0)
1909                 rc = rc2;
1910         return rc;
1911 }
1912
1913 static long mdt_reint_opcode(struct ptlrpc_request *req,
1914                              const struct req_format **fmt)
1915 {
1916         struct mdt_device       *mdt;
1917         struct mdt_rec_reint    *rec;
1918         long                     opc;
1919
1920         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1921         if (rec != NULL) {
1922                 opc = rec->rr_opcode;
1923                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1924                 if (opc < REINT_MAX && fmt[opc] != NULL)
1925                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1926                 else {
1927                         mdt = mdt_exp2dev(req->rq_export);
1928                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1929                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1930                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1931                         opc = err_serious(-EFAULT);
1932                 }
1933         } else {
1934                 opc = err_serious(-EFAULT);
1935         }
1936         return opc;
1937 }
1938
1939 static int mdt_reint(struct tgt_session_info *tsi)
1940 {
1941         long opc;
1942         int  rc;
1943         static const struct req_format *reint_fmts[REINT_MAX] = {
1944                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1945                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1946                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1947                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1948                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1949                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1950                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1951                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1952                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1953         };
1954
1955         ENTRY;
1956
1957         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1958         if (opc >= 0) {
1959                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1960                 /*
1961                  * No lock possible here from client to pass it to reint code
1962                  * path.
1963                  */
1964                 rc = mdt_reint_internal(info, NULL, opc);
1965                 mdt_thread_info_fini(info);
1966         } else {
1967                 rc = opc;
1968         }
1969
1970         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1971         RETURN(rc);
1972 }
1973
1974 /* this should sync the whole device */
1975 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1976 {
1977         struct dt_device *dt = mdt->mdt_bottom;
1978         int rc;
1979         ENTRY;
1980
1981         rc = dt->dd_ops->dt_sync(env, dt);
1982         RETURN(rc);
1983 }
1984
1985 /* this should sync this object */
1986 static int mdt_object_sync(struct mdt_thread_info *info)
1987 {
1988         struct md_object *next;
1989         int rc;
1990         ENTRY;
1991
1992         if (!mdt_object_exists(info->mti_object)) {
1993                 CWARN("%s: non existing object "DFID": rc = %d\n",
1994                       mdt_obd_name(info->mti_mdt),
1995                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1996                 RETURN(-ESTALE);
1997         }
1998         next = mdt_object_child(info->mti_object);
1999         rc = mo_object_sync(info->mti_env, next);
2000
2001         RETURN(rc);
2002 }
2003
2004 static int mdt_sync(struct tgt_session_info *tsi)
2005 {
2006         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2007         struct req_capsule      *pill = tsi->tsi_pill;
2008         struct mdt_body         *body;
2009         int                      rc;
2010
2011         ENTRY;
2012
2013         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2014                 RETURN(err_serious(-ENOMEM));
2015
2016         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2017                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2018         } else {
2019                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2020
2021                 /* sync an object */
2022                 rc = mdt_object_sync(info);
2023                 if (rc == 0) {
2024                         const struct lu_fid *fid;
2025                         struct lu_attr *la = &info->mti_attr.ma_attr;
2026
2027                         info->mti_attr.ma_need = MA_INODE;
2028                         info->mti_attr.ma_valid = 0;
2029                         rc = mdt_attr_get_complex(info, info->mti_object,
2030                                                   &info->mti_attr);
2031                         if (rc == 0) {
2032                                 body = req_capsule_server_get(pill,
2033                                                               &RMF_MDT_BODY);
2034                                 fid = mdt_object_fid(info->mti_object);
2035                                 mdt_pack_attr2body(info, body, la, fid);
2036                         }
2037                 }
2038                 mdt_thread_info_fini(info);
2039         }
2040         if (rc == 0)
2041                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2042
2043         RETURN(rc);
2044 }
2045
2046 /*
2047  * Handle quota control requests to consult current usage/limit, but also
2048  * to configure quota enforcement
2049  */
2050 static int mdt_quotactl(struct tgt_session_info *tsi)
2051 {
2052         struct obd_export       *exp  = tsi->tsi_exp;
2053         struct req_capsule      *pill = tsi->tsi_pill;
2054         struct obd_quotactl     *oqctl, *repoqc;
2055         int                      id, rc;
2056         struct mdt_device       *mdt = mdt_exp2dev(exp);
2057         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2058         struct lu_nodemap       *nodemap;
2059         ENTRY;
2060
2061         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2062         if (oqctl == NULL)
2063                 RETURN(err_serious(-EPROTO));
2064
2065         rc = req_capsule_server_pack(pill);
2066         if (rc)
2067                 RETURN(err_serious(rc));
2068
2069         nodemap = nodemap_get_from_exp(exp);
2070         if (IS_ERR(nodemap))
2071                 RETURN(PTR_ERR(nodemap));
2072
2073         switch (oqctl->qc_cmd) {
2074                 /* master quotactl */
2075         case Q_SETINFO:
2076         case Q_SETQUOTA:
2077                 if (!nodemap_can_setquota(nodemap))
2078                         GOTO(out_nodemap, rc = -EPERM);
2079         case Q_GETINFO:
2080         case Q_GETQUOTA:
2081                 if (qmt == NULL)
2082                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2083                 /* slave quotactl */
2084         case Q_GETOINFO:
2085         case Q_GETOQUOTA:
2086                 break;
2087         default:
2088                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2089                 GOTO(out_nodemap, rc = -EFAULT);
2090         }
2091
2092         id = oqctl->qc_id;
2093         if (oqctl->qc_type == USRQUOTA)
2094                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2095                                     NODEMAP_CLIENT_TO_FS, id);
2096         else if (oqctl->qc_type == GRPQUOTA)
2097                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2098                                     NODEMAP_CLIENT_TO_FS, id);
2099
2100         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2101         if (repoqc == NULL)
2102                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2103
2104         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2105                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2106
2107         if (oqctl->qc_id != id)
2108                 swap(oqctl->qc_id, id);
2109
2110         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2111                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2112                         RETURN(-EINPROGRESS);
2113         }
2114
2115         switch (oqctl->qc_cmd) {
2116
2117         case Q_GETINFO:
2118         case Q_SETINFO:
2119         case Q_SETQUOTA:
2120         case Q_GETQUOTA:
2121                 /* forward quotactl request to QMT */
2122                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2123                 break;
2124
2125         case Q_GETOINFO:
2126         case Q_GETOQUOTA:
2127                 /* slave quotactl */
2128                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2129                                    oqctl);
2130                 break;
2131
2132         default:
2133                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2134                 GOTO(out_nodemap, rc = -EFAULT);
2135         }
2136
2137         if (oqctl->qc_id != id)
2138                 swap(oqctl->qc_id, id);
2139
2140         *repoqc = *oqctl;
2141
2142         EXIT;
2143
2144 out_nodemap:
2145         nodemap_putref(nodemap);
2146
2147         return rc;
2148 }
2149
2150 /** clone llog ctxt from child (mdd)
2151  * This allows remote llog (replicator) access.
2152  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2153  * context was originally set up, or we can handle them directly.
2154  * I choose the latter, but that means I need any llog
2155  * contexts set up by child to be accessable by the mdt.  So we clone the
2156  * context into our context list here.
2157  */
2158 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2159                                int idx)
2160 {
2161         struct md_device  *next = mdt->mdt_child;
2162         struct llog_ctxt *ctxt;
2163         int rc;
2164
2165         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2166                 return 0;
2167
2168         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2169         if (rc || ctxt == NULL) {
2170                 return 0;
2171         }
2172
2173         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2174         if (rc)
2175                 CERROR("Can't set mdt ctxt %d\n", rc);
2176
2177         return rc;
2178 }
2179
2180 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2181                                  struct mdt_device *mdt, int idx)
2182 {
2183         struct llog_ctxt *ctxt;
2184
2185         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2186         if (ctxt == NULL)
2187                 return 0;
2188         /* Put once for the get we just did, and once for the clone */
2189         llog_ctxt_put(ctxt);
2190         llog_ctxt_put(ctxt);
2191         return 0;
2192 }
2193
2194 /*
2195  * sec context handlers
2196  */
2197 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2198 {
2199         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2200
2201         return 0;
2202 }
2203
2204 /*
2205  * quota request handlers
2206  */
2207 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2208 {
2209         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2210         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2211         int                      rc;
2212         ENTRY;
2213
2214         if (qmt == NULL)
2215                 RETURN(err_serious(-EOPNOTSUPP));
2216
2217         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2218         RETURN(rc);
2219 }
2220
2221 struct mdt_object *mdt_object_new(const struct lu_env *env,
2222                                   struct mdt_device *d,
2223                                   const struct lu_fid *f)
2224 {
2225         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2226         struct lu_object *o;
2227         struct mdt_object *m;
2228         ENTRY;
2229
2230         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2231         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2232         if (unlikely(IS_ERR(o)))
2233                 m = (struct mdt_object *)o;
2234         else
2235                 m = mdt_obj(o);
2236         RETURN(m);
2237 }
2238
2239 struct mdt_object *mdt_object_find(const struct lu_env *env,
2240                                    struct mdt_device *d,
2241                                    const struct lu_fid *f)
2242 {
2243         struct lu_object *o;
2244         struct mdt_object *m;
2245         ENTRY;
2246
2247         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2248         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2249         if (unlikely(IS_ERR(o)))
2250                 m = (struct mdt_object *)o;
2251         else
2252                 m = mdt_obj(o);
2253
2254         RETURN(m);
2255 }
2256
2257 /**
2258  * Asyncronous commit for mdt device.
2259  *
2260  * Pass asynchonous commit call down the MDS stack.
2261  *
2262  * \param env environment
2263  * \param mdt the mdt device
2264  */
2265 static void mdt_device_commit_async(const struct lu_env *env,
2266                                     struct mdt_device *mdt)
2267 {
2268         struct dt_device *dt = mdt->mdt_bottom;
2269         int rc;
2270         ENTRY;
2271
2272         rc = dt->dd_ops->dt_commit_async(env, dt);
2273         if (unlikely(rc != 0))
2274                 CWARN("%s: async commit start failed: rc = %d\n",
2275                       mdt_obd_name(mdt), rc);
2276         atomic_inc(&mdt->mdt_async_commit_count);
2277         EXIT;
2278 }
2279
2280 /**
2281  * Mark the lock as "synchonous".
2282  *
2283  * Mark the lock to deffer transaction commit to the unlock time.
2284  *
2285  * \param lock the lock to mark as "synchonous"
2286  *
2287  * \see mdt_is_lock_sync
2288  * \see mdt_save_lock
2289  */
2290 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2291 {
2292         lock->l_ast_data = (void*)1;
2293 }
2294
2295 /**
2296  * Check whehter the lock "synchonous" or not.
2297  *
2298  * \param lock the lock to check
2299  * \retval 1 the lock is "synchonous"
2300  * \retval 0 the lock isn't "synchronous"
2301  *
2302  * \see mdt_set_lock_sync
2303  * \see mdt_save_lock
2304  */
2305 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2306 {
2307         return lock->l_ast_data != NULL;
2308 }
2309
2310 /**
2311  * Blocking AST for mdt locks.
2312  *
2313  * Starts transaction commit if in case of COS lock conflict or
2314  * deffers such a commit to the mdt_save_lock.
2315  *
2316  * \param lock the lock which blocks a request or cancelling lock
2317  * \param desc unused
2318  * \param data unused
2319  * \param flag indicates whether this cancelling or blocking callback
2320  * \retval 0
2321  * \see ldlm_blocking_ast_nocheck
2322  */
2323 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2324                      void *data, int flag)
2325 {
2326         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2327         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2328         int rc;
2329         ENTRY;
2330
2331         if (flag == LDLM_CB_CANCELING)
2332                 RETURN(0);
2333
2334         lock_res_and_lock(lock);
2335         if (lock->l_blocking_ast != mdt_blocking_ast) {
2336                 unlock_res_and_lock(lock);
2337                 RETURN(0);
2338         }
2339         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2340             lock->l_blocking_lock != NULL) {
2341                 if (mdt_cos_is_enabled(mdt) &&
2342                     lock->l_client_cookie !=
2343                     lock->l_blocking_lock->l_client_cookie)
2344                         mdt_set_lock_sync(lock);
2345                 else if (mdt_slc_is_enabled(mdt) &&
2346                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2347                         mdt_set_lock_sync(lock);
2348         }
2349         rc = ldlm_blocking_ast_nocheck(lock);
2350
2351         /* There is no lock conflict if l_blocking_lock == NULL,
2352          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2353          * when the last reference to a local lock was released */
2354         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2355                 struct lu_env env;
2356
2357                 rc = lu_env_init(&env, LCT_LOCAL);
2358                 if (unlikely(rc != 0))
2359                         CWARN("%s: lu_env initialization failed, cannot "
2360                               "start asynchronous commit: rc = %d\n",
2361                               obd->obd_name, rc);
2362                 else
2363                         mdt_device_commit_async(&env, mdt);
2364                 lu_env_fini(&env);
2365         }
2366         RETURN(rc);
2367 }
2368
2369 /*
2370  * Blocking AST for cross-MDT lock
2371  *
2372  * Discard lock from uncommitted_slc_locks and cancel it.
2373  *
2374  * \param lock  the lock which blocks a request or cancelling lock
2375  * \param desc  unused
2376  * \param data  unused
2377  * \param flag  indicates whether this cancelling or blocking callback
2378  * \retval      0 on success
2379  * \retval      negative number on error
2380  */
2381 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2382                             void *data, int flag)
2383 {
2384         int rc = 0;
2385         ENTRY;
2386
2387         switch (flag) {
2388         case LDLM_CB_BLOCKING: {
2389                 struct lustre_handle lockh;
2390
2391                 ldlm_lock2handle(lock, &lockh);
2392                 rc = ldlm_cli_cancel(&lockh,
2393                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2394                 if (rc < 0) {
2395                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2396                         RETURN(rc);
2397                 }
2398                 break;
2399         }
2400         case LDLM_CB_CANCELING: {
2401                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2402                 struct mdt_device *mdt =
2403                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2404
2405                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2406
2407                 /* discard slc lock here so that it can be cleaned anytime,
2408                  * especially for cleanup_resource() */
2409                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2410
2411                 /* once we cache lock, l_ast_data is set to mdt_object */
2412                 if (lock->l_ast_data != NULL) {
2413                         struct mdt_object *mo = lock->l_ast_data;
2414                         struct lu_env env;
2415
2416                         rc = lu_env_init(&env, LCT_MD_THREAD);
2417                         if (unlikely(rc != 0)) {
2418                                 CWARN("%s: lu_env initialization failed, object"
2419                                       "%p "DFID" is leaked!\n",
2420                                       obd->obd_name, mo,
2421                                       PFID(mdt_object_fid(mo)));
2422                                 RETURN(rc);
2423                         }
2424
2425                         if (lock->l_policy_data.l_inodebits.bits &
2426                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2427                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2428                                 mo->mot_cache_attr = 0;
2429                         }
2430                         mdt_object_put(&env, mo);
2431                         lu_env_fini(&env);
2432                 }
2433                 break;
2434         }
2435         default:
2436                 LBUG();
2437         }
2438
2439         RETURN(rc);
2440 }
2441
2442 int mdt_check_resent_lock(struct mdt_thread_info *info,
2443                           struct mdt_object *mo,
2444                           struct mdt_lock_handle *lhc)
2445 {
2446         /* the lock might already be gotten in ldlm_handle_enqueue() */
2447         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2448                 struct ptlrpc_request *req = mdt_info_req(info);
2449                 struct ldlm_lock      *lock;
2450
2451                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2452                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2453                 if (lock == NULL) {
2454                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2455                          * a resend case, however, it could be already destroyed
2456                          * due to client eviction or a raced cancel RPC. */
2457                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2458                                           lhc->mlh_reg_lh.cookie);
2459                         RETURN(-ESTALE);
2460                 }
2461
2462                 if (!fid_res_name_eq(mdt_object_fid(mo),
2463                                      &lock->l_resource->lr_name)) {
2464                         CWARN("%s: Although resent, but still not "
2465                               "get child lock:"DFID"\n",
2466                               info->mti_exp->exp_obd->obd_name,
2467                               PFID(mdt_object_fid(mo)));
2468                         LDLM_LOCK_PUT(lock);
2469                         RETURN(-EPROTO);
2470                 }
2471                 LDLM_LOCK_PUT(lock);
2472                 return 0;
2473         }
2474         return 1;
2475 }
2476
2477 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2478                            const struct lu_fid *fid, struct lustre_handle *lh,
2479                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2480                            bool cache)
2481 {
2482         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2483         union ldlm_policy_data *policy = &mti->mti_policy;
2484         struct ldlm_res_id *res_id = &mti->mti_res_id;
2485         int rc = 0;
2486         ENTRY;
2487
2488         LASSERT(mdt_object_remote(o));
2489
2490         fid_build_reg_res_name(fid, res_id);
2491
2492         memset(einfo, 0, sizeof(*einfo));
2493         einfo->ei_type = LDLM_IBITS;
2494         einfo->ei_mode = mode;
2495         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2496         einfo->ei_cb_cp = ldlm_completion_ast;
2497         einfo->ei_enq_slave = 0;
2498         einfo->ei_res_id = res_id;
2499         if (nonblock)
2500                 einfo->ei_nonblock = 1;
2501         if (cache) {
2502                 /*
2503                  * if we cache lock, couple lock with mdt_object, so that object
2504                  * can be easily found in lock ASTs.
2505                  */
2506                 mdt_object_get(mti->mti_env, o);
2507                 einfo->ei_cbdata = o;
2508         }
2509
2510         memset(policy, 0, sizeof(*policy));
2511         policy->l_inodebits.bits = ibits;
2512
2513         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2514                             policy);
2515         if (rc < 0 && cache) {
2516                 mdt_object_put(mti->mti_env, o);
2517                 einfo->ei_cbdata = NULL;
2518         }
2519         RETURN(rc);
2520 }
2521
2522 static int mdt_object_local_lock(struct mdt_thread_info *info,
2523                                  struct mdt_object *o,
2524                                  struct mdt_lock_handle *lh, __u64 ibits,
2525                                  bool nonblock, bool cos_incompat)
2526 {
2527         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2528         union ldlm_policy_data *policy = &info->mti_policy;
2529         struct ldlm_res_id *res_id = &info->mti_res_id;
2530         __u64 dlmflags = 0;
2531         int rc;
2532         ENTRY;
2533
2534         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2535         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2536         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2537         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2538
2539         if (cos_incompat) {
2540                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2541                         lh->mlh_reg_mode == LCK_EX);
2542                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2543         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2544                 dlmflags |= LDLM_FL_COS_ENABLED;
2545         }
2546
2547         /* Only enqueue LOOKUP lock for remote object */
2548         if (mdt_object_remote(o))
2549                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2550
2551         if (lh->mlh_type == MDT_PDO_LOCK) {
2552                 /* check for exists after object is locked */
2553                 if (mdt_object_exists(o) == 0) {
2554                         /* Non-existent object shouldn't have PDO lock */
2555                         RETURN(-ESTALE);
2556                 } else {
2557                         /* Non-dir object shouldn't have PDO lock */
2558                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2559                                 RETURN(-ENOTDIR);
2560                 }
2561         }
2562
2563         memset(policy, 0, sizeof(*policy));
2564         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2565
2566         dlmflags |= LDLM_FL_ATOMIC_CB;
2567         if (nonblock)
2568                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2569
2570         /*
2571          * Take PDO lock on whole directory and build correct @res_id for lock
2572          * on part of directory.
2573          */
2574         if (lh->mlh_pdo_hash != 0) {
2575                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2576                 mdt_lock_pdo_mode(info, o, lh);
2577                 if (lh->mlh_pdo_mode != LCK_NL) {
2578                         /*
2579                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2580                          * is never going to be sent to client and we do not
2581                          * want it slowed down due to possible cancels.
2582                          */
2583                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2584                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2585                                           policy, res_id, dlmflags,
2586                                           info->mti_exp == NULL ? NULL :
2587                                           &info->mti_exp->exp_handle.h_cookie);
2588                         if (unlikely(rc != 0))
2589                                 GOTO(out_unlock, rc);
2590                 }
2591
2592                 /*
2593                  * Finish res_id initializing by name hash marking part of
2594                  * directory which is taking modification.
2595                  */
2596                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2597         }
2598
2599         policy->l_inodebits.bits = ibits;
2600
2601         /*
2602          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2603          * going to be sent to client. If it is - mdt_intent_policy() path will
2604          * fix it up and turn FL_LOCAL flag off.
2605          */
2606         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2607                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2608                           info->mti_exp == NULL ? NULL :
2609                           &info->mti_exp->exp_handle.h_cookie);
2610 out_unlock:
2611         if (rc != 0)
2612                 mdt_object_unlock(info, o, lh, 1);
2613         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2614                    lh->mlh_pdo_hash != 0 &&
2615                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2616                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2617
2618         RETURN(rc);
2619 }
2620
2621 static int
2622 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2623                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2624                          bool cos_incompat)
2625 {
2626         struct mdt_lock_handle *local_lh = NULL;
2627         int rc;
2628         ENTRY;
2629
2630         if (!mdt_object_remote(o)) {
2631                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2632                                            cos_incompat);
2633                 RETURN(rc);
2634         }
2635
2636         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2637         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2638                    MDS_INODELOCK_XATTR);
2639
2640         /* Only enqueue LOOKUP lock for remote object */
2641         if (ibits & MDS_INODELOCK_LOOKUP) {
2642                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2643                                            nonblock, cos_incompat);
2644                 if (rc != ELDLM_OK)
2645                         RETURN(rc);
2646
2647                 local_lh = lh;
2648         }
2649
2650         if (ibits & MDS_INODELOCK_UPDATE) {
2651                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2652                  * enqueue RPC can only request 1 lock, to avoid extra
2653                  * RPC, so it will instead enqueue EX lock for remote
2654                  * object anyway XXX*/
2655                 if (lh->mlh_type == MDT_PDO_LOCK &&
2656                     lh->mlh_pdo_hash != 0) {
2657                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2658                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2659                                PFID(mdt_object_fid(o)));
2660                         lh->mlh_pdo_hash = 0;
2661                         lh->mlh_rreg_mode = LCK_EX;
2662                         lh->mlh_type = MDT_REG_LOCK;
2663                 }
2664                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2665                                             &lh->mlh_rreg_lh,
2666                                             lh->mlh_rreg_mode,
2667                                             MDS_INODELOCK_UPDATE, nonblock,
2668                                             false);
2669                 if (rc != ELDLM_OK) {
2670                         if (local_lh != NULL)
2671                                 mdt_object_unlock(info, o, local_lh, rc);
2672                         RETURN(rc);
2673                 }
2674         }
2675
2676         RETURN(0);
2677 }
2678
2679 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2680                     struct mdt_lock_handle *lh, __u64 ibits)
2681 {
2682         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2683 }
2684
2685 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2686                           struct mdt_lock_handle *lh, __u64 ibits,
2687                           bool cos_incompat)
2688 {
2689         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2690         return mdt_object_lock_internal(info, o, lh, ibits, false,
2691                                         cos_incompat);
2692 }
2693
2694 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2695                         struct mdt_lock_handle *lh, __u64 ibits)
2696 {
2697         struct mdt_lock_handle tmp = *lh;
2698         int rc;
2699
2700         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2701         if (rc == 0)
2702                 *lh = tmp;
2703
2704         return rc == 0;
2705 }
2706
2707 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2708                               struct mdt_object *o, struct mdt_lock_handle *lh,
2709                               __u64 ibits, bool cos_incompat)
2710 {
2711         struct mdt_lock_handle tmp = *lh;
2712         int rc;
2713
2714         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2715         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2716         if (rc == 0)
2717                 *lh = tmp;
2718
2719         return rc == 0;
2720 }
2721
2722 /**
2723  * Save a lock within request object.
2724  *
2725  * Keep the lock referenced until whether client ACK or transaction
2726  * commit happens or release the lock immediately depending on input
2727  * parameters. If COS is ON, a write lock is converted to COS lock
2728  * before saving.
2729  *
2730  * \param info thead info object
2731  * \param h lock handle
2732  * \param mode lock mode
2733  * \param decref force immediate lock releasing
2734  */
2735 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2736                           enum ldlm_mode mode, int decref)
2737 {
2738         ENTRY;
2739
2740         if (lustre_handle_is_used(h)) {
2741                 if (decref || !info->mti_has_trans ||
2742                     !(mode & (LCK_PW | LCK_EX))) {
2743                         mdt_fid_unlock(h, mode);
2744                 } else {
2745                         struct mdt_device *mdt = info->mti_mdt;
2746                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2747                         struct ptlrpc_request *req = mdt_info_req(info);
2748                         int cos;
2749
2750                         cos = (mdt_cos_is_enabled(mdt) ||
2751                                mdt_slc_is_enabled(mdt));
2752
2753                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2754                                  h->cookie);
2755
2756                         /* there is no request if mdt_object_unlock() is called
2757                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2758                         if (likely(req != NULL)) {
2759                                 CDEBUG(D_HA, "request = %p reply state = %p"
2760                                        " transno = %lld\n", req,
2761                                        req->rq_reply_state, req->rq_transno);
2762                                 if (cos) {
2763                                         ldlm_lock_downgrade(lock, LCK_COS);
2764                                         mode = LCK_COS;
2765                                 }
2766                                 ptlrpc_save_lock(req, h, mode, cos);
2767                         } else {
2768                                 mdt_fid_unlock(h, mode);
2769                         }
2770                         if (mdt_is_lock_sync(lock)) {
2771                                 CDEBUG(D_HA, "found sync-lock,"
2772                                        " async commit started\n");
2773                                 mdt_device_commit_async(info->mti_env,
2774                                                         mdt);
2775                         }
2776                         LDLM_LOCK_PUT(lock);
2777                 }
2778                 h->cookie = 0ull;
2779         }
2780
2781         EXIT;
2782 }
2783
2784 /**
2785  * Save cross-MDT lock in uncommitted_slc_locks
2786  *
2787  * Keep the lock referenced until transaction commit happens or release the lock
2788  * immediately depending on input parameters.
2789  *
2790  * \param info thead info object
2791  * \param h lock handle
2792  * \param mode lock mode
2793  * \param decref force immediate lock releasing
2794  */
2795 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2796                                  struct mdt_object *o, struct lustre_handle *h,
2797                                  enum ldlm_mode mode, int decref)
2798 {
2799         ENTRY;
2800
2801         if (lustre_handle_is_used(h)) {
2802                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2803
2804                 if (o != NULL &&
2805                     (lock->l_policy_data.l_inodebits.bits &
2806                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2807                         mo_invalidate(info->mti_env, mdt_object_child(o));
2808
2809                 if (decref || !info->mti_has_trans ||
2810                     !(mode & (LCK_PW | LCK_EX))) {
2811                         ldlm_lock_decref_and_cancel(h, mode);
2812                         LDLM_LOCK_PUT(lock);
2813                 } else {
2814                         struct ptlrpc_request *req = mdt_info_req(info);
2815
2816                         LASSERT(req != NULL);
2817                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
2818                                           req->rq_transno);
2819                         ldlm_lock_decref(h, mode);
2820                 }
2821                 h->cookie = 0ull;
2822         }
2823
2824         EXIT;
2825 }
2826
2827 /**
2828  * Unlock mdt object.
2829  *
2830  * Immeditely release the regular lock and the PDO lock or save the
2831  * lock in request and keep them referenced until client ACK or
2832  * transaction commit.
2833  *
2834  * \param info thread info object
2835  * \param o mdt object
2836  * \param lh mdt lock handle referencing regular and PDO locks
2837  * \param decref force immediate lock releasing
2838  *
2839  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
2840  */
2841 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2842                        struct mdt_lock_handle *lh, int decref)
2843 {
2844         ENTRY;
2845
2846         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2847         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2848         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2849                              decref);
2850
2851         EXIT;
2852 }
2853
2854 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2855                                         const struct lu_fid *f,
2856                                         struct mdt_lock_handle *lh,
2857                                         __u64 ibits)
2858 {
2859         struct mdt_object *o;
2860
2861         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2862         if (!IS_ERR(o)) {
2863                 int rc;
2864
2865                 rc = mdt_object_lock(info, o, lh, ibits);
2866                 if (rc != 0) {
2867                         mdt_object_put(info->mti_env, o);
2868                         o = ERR_PTR(rc);
2869                 }
2870         }
2871         return o;
2872 }
2873
2874 void mdt_object_unlock_put(struct mdt_thread_info * info,
2875                            struct mdt_object * o,
2876                            struct mdt_lock_handle *lh,
2877                            int decref)
2878 {
2879         mdt_object_unlock(info, o, lh, decref);
2880         mdt_object_put(info->mti_env, o);
2881 }
2882
2883 /*
2884  * Generic code handling requests that have struct mdt_body passed in:
2885  *
2886  *  - extract mdt_body from request and save it in @info, if present;
2887  *
2888  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2889  *  @info;
2890  *
2891  *  - if HABEO_CORPUS flag is set for this request type check whether object
2892  *  actually exists on storage (lu_object_exists()).
2893  *
2894  */
2895 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2896 {
2897         const struct mdt_body    *body;
2898         struct mdt_object        *obj;
2899         const struct lu_env      *env;
2900         struct req_capsule       *pill;
2901         int                       rc;
2902         ENTRY;
2903
2904         env = info->mti_env;
2905         pill = info->mti_pill;
2906
2907         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2908         if (body == NULL)
2909                 RETURN(-EFAULT);
2910
2911         if (!(body->mbo_valid & OBD_MD_FLID))
2912                 RETURN(0);
2913
2914         if (!fid_is_sane(&body->mbo_fid1)) {
2915                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2916                 RETURN(-EINVAL);
2917         }
2918
2919         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2920         if (!IS_ERR(obj)) {
2921                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2922                         mdt_object_put(env, obj);
2923                         rc = -ENOENT;
2924                 } else {
2925                         info->mti_object = obj;
2926                         rc = 0;
2927                 }
2928         } else
2929                 rc = PTR_ERR(obj);
2930
2931         RETURN(rc);
2932 }
2933
2934 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2935 {
2936         struct req_capsule *pill = info->mti_pill;
2937         int rc;
2938         ENTRY;
2939
2940         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2941                 rc = mdt_body_unpack(info, flags);
2942         else
2943                 rc = 0;
2944
2945         if (rc == 0 && (flags & HABEO_REFERO)) {
2946                 /* Pack reply. */
2947                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2948                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2949                                              DEF_REP_MD_SIZE);
2950                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2951                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2952                                              RCL_SERVER, 0);
2953
2954                 rc = req_capsule_server_pack(pill);
2955         }
2956         RETURN(rc);
2957 }
2958
2959 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2960 {
2961         lh->mlh_type = MDT_NUL_LOCK;
2962         lh->mlh_reg_lh.cookie = 0ull;
2963         lh->mlh_reg_mode = LCK_MINMODE;
2964         lh->mlh_pdo_lh.cookie = 0ull;
2965         lh->mlh_pdo_mode = LCK_MINMODE;
2966         lh->mlh_rreg_lh.cookie = 0ull;
2967         lh->mlh_rreg_mode = LCK_MINMODE;
2968 }
2969
2970 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2971 {
2972         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2973         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2974 }
2975
2976 /*
2977  * Initialize fields of struct mdt_thread_info. Other fields are left in
2978  * uninitialized state, because it's too expensive to zero out whole
2979  * mdt_thread_info (> 1K) on each request arrival.
2980  */
2981 void mdt_thread_info_init(struct ptlrpc_request *req,
2982                           struct mdt_thread_info *info)
2983 {
2984         int i;
2985
2986         info->mti_pill = &req->rq_pill;
2987
2988         /* lock handle */
2989         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2990                 mdt_lock_handle_init(&info->mti_lh[i]);
2991
2992         /* mdt device: it can be NULL while CONNECT */
2993         if (req->rq_export) {
2994                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2995                 info->mti_exp = req->rq_export;
2996         } else
2997                 info->mti_mdt = NULL;
2998         info->mti_env = req->rq_svc_thread->t_env;
2999         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3000
3001         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3002         info->mti_big_buf = LU_BUF_NULL;
3003         info->mti_body = NULL;
3004         info->mti_object = NULL;
3005         info->mti_dlm_req = NULL;
3006         info->mti_has_trans = 0;
3007         info->mti_cross_ref = 0;
3008         info->mti_opdata = 0;
3009         info->mti_big_lmm_used = 0;
3010
3011         info->mti_spec.no_create = 0;
3012         info->mti_spec.sp_rm_entry = 0;
3013         info->mti_spec.sp_permitted = 0;
3014         info->mti_spec.sp_migrate_close = 0;
3015
3016         info->mti_spec.u.sp_ea.eadata = NULL;
3017         info->mti_spec.u.sp_ea.eadatalen = 0;
3018 }
3019
3020 void mdt_thread_info_fini(struct mdt_thread_info *info)
3021 {
3022         int i;
3023
3024         if (info->mti_object != NULL) {
3025                 mdt_object_put(info->mti_env, info->mti_object);
3026                 info->mti_object = NULL;
3027         }
3028
3029         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3030                 mdt_lock_handle_fini(&info->mti_lh[i]);
3031         info->mti_env = NULL;
3032         info->mti_pill = NULL;
3033         info->mti_exp = NULL;
3034
3035         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3036                 lu_buf_free(&info->mti_big_buf);
3037 }
3038
3039 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3040 {
3041         struct mdt_thread_info  *mti;
3042
3043         mti = mdt_th_info(tsi->tsi_env);
3044         LASSERT(mti != NULL);
3045
3046         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3047         if (tsi->tsi_corpus != NULL) {
3048                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3049                 lu_object_get(tsi->tsi_corpus);
3050         }
3051         mti->mti_body = tsi->tsi_mdt_body;
3052         mti->mti_dlm_req = tsi->tsi_dlm_req;
3053
3054         return mti;
3055 }
3056
3057 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3058 {
3059         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3060             cfs_fail_val ==
3061             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3062                 set_current_state(TASK_UNINTERRUPTIBLE);
3063                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3064         }
3065
3066         return tgt_connect(tsi);
3067 }
3068
3069 enum mdt_it_code {
3070         MDT_IT_OPEN,
3071         MDT_IT_OCREAT,
3072         MDT_IT_CREATE,
3073         MDT_IT_GETATTR,
3074         MDT_IT_READDIR,
3075         MDT_IT_LOOKUP,
3076         MDT_IT_UNLINK,
3077         MDT_IT_TRUNC,
3078         MDT_IT_GETXATTR,
3079         MDT_IT_LAYOUT,
3080         MDT_IT_QUOTA,
3081         MDT_IT_NR
3082 };
3083
3084 static int mdt_intent_getattr(enum mdt_it_code opcode,
3085                               struct mdt_thread_info *info,
3086                               struct ldlm_lock **,
3087                               __u64);
3088
3089 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3090                                 struct mdt_thread_info *info,
3091                                 struct ldlm_lock **lockp,
3092                                 __u64 flags);
3093
3094 static int mdt_intent_layout(enum mdt_it_code opcode,
3095                              struct mdt_thread_info *info,
3096                              struct ldlm_lock **,
3097                              __u64);
3098 static int mdt_intent_reint(enum mdt_it_code opcode,
3099                             struct mdt_thread_info *info,
3100                             struct ldlm_lock **,
3101                             __u64);
3102
3103 static struct mdt_it_flavor {
3104         const struct req_format *it_fmt;
3105         __u32                    it_flags;
3106         int                    (*it_act)(enum mdt_it_code ,
3107                                          struct mdt_thread_info *,
3108                                          struct ldlm_lock **,
3109                                          __u64);
3110         long                     it_reint;
3111 } mdt_it_flavor[] = {
3112         [MDT_IT_OPEN]     = {
3113                 .it_fmt   = &RQF_LDLM_INTENT,
3114                 /*.it_flags = HABEO_REFERO,*/
3115                 .it_flags = 0,
3116                 .it_act   = mdt_intent_reint,
3117                 .it_reint = REINT_OPEN
3118         },
3119         [MDT_IT_OCREAT]   = {
3120                 .it_fmt   = &RQF_LDLM_INTENT,
3121                 /*
3122                  * OCREAT is not a MUTABOR request as if the file
3123                  * already exists.
3124                  * We do the extra check of OBD_CONNECT_RDONLY in
3125                  * mdt_reint_open() when we really need to create
3126                  * the object.
3127                  */
3128                 .it_flags = 0,
3129                 .it_act   = mdt_intent_reint,
3130                 .it_reint = REINT_OPEN
3131         },
3132         [MDT_IT_CREATE]   = {
3133                 .it_fmt   = &RQF_LDLM_INTENT,
3134                 .it_flags = MUTABOR,
3135                 .it_act   = mdt_intent_reint,
3136                 .it_reint = REINT_CREATE
3137         },
3138         [MDT_IT_GETATTR]  = {
3139                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3140                 .it_flags = HABEO_REFERO,
3141                 .it_act   = mdt_intent_getattr
3142         },
3143         [MDT_IT_READDIR]  = {
3144                 .it_fmt   = NULL,
3145                 .it_flags = 0,
3146                 .it_act   = NULL
3147         },
3148         [MDT_IT_LOOKUP]   = {
3149                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3150                 .it_flags = HABEO_REFERO,
3151                 .it_act   = mdt_intent_getattr
3152         },
3153         [MDT_IT_UNLINK]   = {
3154                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3155                 .it_flags = MUTABOR,
3156                 .it_act   = NULL,
3157                 .it_reint = REINT_UNLINK
3158         },
3159         [MDT_IT_TRUNC]    = {
3160                 .it_fmt   = NULL,
3161                 .it_flags = MUTABOR,
3162                 .it_act   = NULL
3163         },
3164         [MDT_IT_GETXATTR] = {
3165                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3166                 .it_flags = HABEO_CORPUS,
3167                 .it_act   = mdt_intent_getxattr
3168         },
3169         [MDT_IT_LAYOUT] = {
3170                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3171                 .it_flags = 0,
3172                 .it_act   = mdt_intent_layout
3173         }
3174 };
3175
3176 static int
3177 mdt_intent_lock_replace(struct mdt_thread_info *info,
3178                         struct ldlm_lock **lockp,
3179                         struct mdt_lock_handle *lh,
3180                         __u64 flags, int result)
3181 {
3182         struct ptlrpc_request  *req = mdt_info_req(info);
3183         struct ldlm_lock       *lock = *lockp;
3184         struct ldlm_lock       *new_lock;
3185
3186         /* If possible resent found a lock, @lh is set to its handle */
3187         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3188
3189         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3190                 lh->mlh_reg_lh.cookie = 0;
3191                 RETURN(0);
3192         }
3193
3194         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3195                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3196                  * a resend case, however, it could be already destroyed
3197                  * due to client eviction or a raced cancel RPC. */
3198                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3199                                   lh->mlh_reg_lh.cookie);
3200                 lh->mlh_reg_lh.cookie = 0;
3201                 RETURN(-ESTALE);
3202         }
3203
3204         LASSERTF(new_lock != NULL,
3205                  "lockh %#llx flags %#llx : rc = %d\n",
3206                  lh->mlh_reg_lh.cookie, flags, result);
3207
3208         /*
3209          * If we've already given this lock to a client once, then we should
3210          * have no readers or writers.  Otherwise, we should have one reader
3211          * _or_ writer ref (which will be zeroed below) before returning the
3212          * lock to a client.
3213          */
3214         if (new_lock->l_export == req->rq_export) {
3215                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3216         } else {
3217                 LASSERT(new_lock->l_export == NULL);
3218                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3219         }
3220
3221         *lockp = new_lock;
3222
3223         if (new_lock->l_export == req->rq_export) {
3224                 /*
3225                  * Already gave this to the client, which means that we
3226                  * reconstructed a reply.
3227                  */
3228                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3229                         MSG_RESENT);
3230
3231                 LDLM_LOCK_RELEASE(new_lock);
3232                 lh->mlh_reg_lh.cookie = 0;
3233                 RETURN(ELDLM_LOCK_REPLACED);
3234         }
3235
3236         /*
3237          * Fixup the lock to be given to the client.
3238          */
3239         lock_res_and_lock(new_lock);
3240         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3241          * possible blocking AST. */
3242         while (new_lock->l_readers > 0) {
3243                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3244                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3245                 new_lock->l_readers--;
3246         }
3247         while (new_lock->l_writers > 0) {
3248                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3249                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3250                 new_lock->l_writers--;
3251         }
3252
3253         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3254         new_lock->l_blocking_ast = lock->l_blocking_ast;
3255         new_lock->l_completion_ast = lock->l_completion_ast;
3256         new_lock->l_remote_handle = lock->l_remote_handle;
3257         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3258
3259         unlock_res_and_lock(new_lock);
3260
3261         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3262                      &new_lock->l_remote_handle,
3263                      &new_lock->l_exp_hash);
3264
3265         LDLM_LOCK_RELEASE(new_lock);
3266         lh->mlh_reg_lh.cookie = 0;
3267
3268         RETURN(ELDLM_LOCK_REPLACED);
3269 }
3270
3271 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3272                                     struct ldlm_lock *new_lock,
3273                                     struct mdt_lock_handle *lh,
3274                                     __u64 flags)
3275 {
3276         struct ptlrpc_request  *req = mdt_info_req(info);
3277         struct ldlm_request    *dlmreq;
3278
3279         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3280                 return;
3281
3282         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3283
3284         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3285          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3286          * initialized. */
3287         if (flags & LDLM_FL_RESENT) {
3288                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3289                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3290
3291                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3292                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3293                           lh->mlh_reg_lh.cookie);
3294                 return;
3295         }
3296
3297         /*
3298          * If the xid matches, then we know this is a resent request, and allow
3299          * it. (It's probably an OPEN, for which we don't send a lock.
3300          */
3301         if (req_can_reconstruct(req, NULL))
3302                 return;
3303
3304         /*
3305          * This remote handle isn't enqueued, so we never received or processed
3306          * this request.  Clear MSG_RESENT, because it can be handled like any
3307          * normal request now.
3308          */
3309         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3310
3311         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3312                   dlmreq->lock_handle[0].cookie);
3313 }
3314
3315 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3316                                 struct mdt_thread_info *info,
3317                                 struct ldlm_lock **lockp,
3318                                 __u64 flags)
3319 {
3320         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3321         struct ldlm_reply      *ldlm_rep = NULL;
3322         int rc;
3323         ENTRY;
3324
3325         /*
3326          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3327          * (for the resend case) or a new lock. Below we will use it to
3328          * replace the original lock.
3329          */
3330         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3331         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3332                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3333                 rc = mdt_object_lock(info, info->mti_object, lhc,
3334                                      MDS_INODELOCK_XATTR);
3335                 if (rc)
3336                         return rc;
3337         }
3338
3339         rc = mdt_getxattr(info);
3340
3341         if (mdt_info_req(info)->rq_repmsg != NULL)
3342                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3343
3344         if (ldlm_rep == NULL ||
3345             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3346                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3347                 RETURN(err_serious(-EFAULT));
3348         }
3349
3350         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3351
3352         /* This is left for interop instead of adding a new interop flag.
3353          * LU-7433 */
3354 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3355         if (ldlm_rep->lock_policy_res2) {
3356                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3357                 RETURN(ELDLM_LOCK_ABORTED);
3358         }
3359 #endif
3360
3361         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3362         RETURN(rc);
3363 }
3364
3365 static int mdt_intent_getattr(enum mdt_it_code opcode,
3366                               struct mdt_thread_info *info,
3367                               struct ldlm_lock **lockp,
3368                               __u64 flags)
3369 {
3370         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3371         __u64                   child_bits;
3372         struct ldlm_reply      *ldlm_rep;
3373         struct mdt_body        *reqbody;
3374         struct mdt_body        *repbody;
3375         int                     rc, rc2;
3376         ENTRY;
3377
3378         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3379         LASSERT(reqbody);
3380
3381         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3382         LASSERT(repbody);
3383
3384         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3385         repbody->mbo_eadatasize = 0;
3386         repbody->mbo_aclsize = 0;
3387
3388         switch (opcode) {
3389         case MDT_IT_LOOKUP:
3390                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3391                 break;
3392         case MDT_IT_GETATTR:
3393                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3394                              MDS_INODELOCK_PERM;
3395                 break;
3396         default:
3397                 CERROR("Unsupported intent (%d)\n", opcode);
3398                 GOTO(out_shrink, rc = -EINVAL);
3399         }
3400
3401         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3402         if (rc)
3403                 GOTO(out_shrink, rc);
3404
3405         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3406         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3407
3408         /* Get lock from request for possible resent case. */
3409         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3410
3411         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3412         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3413
3414         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3415                 ldlm_rep->lock_policy_res2 = 0;
3416         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3417             ldlm_rep->lock_policy_res2) {
3418                 lhc->mlh_reg_lh.cookie = 0ull;
3419                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3420         }
3421
3422         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3423         EXIT;
3424 out_ucred:
3425         mdt_exit_ucred(info);
3426 out_shrink:
3427         mdt_client_compatibility(info);
3428         rc2 = mdt_fix_reply(info);
3429         if (rc == 0)
3430                 rc = rc2;
3431         return rc;
3432 }
3433
3434 static int mdt_intent_layout(enum mdt_it_code opcode,
3435                              struct mdt_thread_info *info,
3436                              struct ldlm_lock **lockp,
3437                              __u64 flags)
3438 {
3439         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3440         struct layout_intent *layout;
3441         struct lu_fid *fid;
3442         struct mdt_object *obj = NULL;
3443         int layout_size = 0;
3444         int rc = 0;
3445         ENTRY;
3446
3447         if (opcode != MDT_IT_LAYOUT) {
3448                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3449                         opcode);
3450                 RETURN(-EINVAL);
3451         }
3452
3453         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3454         if (layout == NULL)
3455                 RETURN(-EPROTO);
3456
3457         if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3458                 CERROR("%s: Unsupported layout intent opc %d\n",
3459                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3460                 RETURN(-EINVAL);
3461         }
3462
3463         fid = &info->mti_tmp_fid2;
3464         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3465
3466         /* Get lock from request for possible resent case. */
3467         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3468
3469         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3470         if (IS_ERR(obj))
3471                 GOTO(out, rc = PTR_ERR(obj));
3472
3473         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3474                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3475                 if (layout_size < 0)
3476                         GOTO(out_obj, rc = layout_size);
3477
3478                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3479                         info->mti_mdt->mdt_max_mdsize = layout_size;
3480         }
3481
3482         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3483         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3484                              layout_size);
3485         rc = req_capsule_server_pack(info->mti_pill);
3486         GOTO(out_obj, rc);
3487
3488 out_obj:
3489         mdt_object_put(info->mti_env, obj);
3490
3491         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3492                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3493
3494 out:
3495         lhc->mlh_reg_lh.cookie = 0;
3496
3497         return rc;
3498 }
3499
3500 static int mdt_intent_reint(enum mdt_it_code opcode,
3501                             struct mdt_thread_info *info,
3502                             struct ldlm_lock **lockp,
3503                             __u64 flags)
3504 {
3505         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3506         struct ldlm_reply      *rep = NULL;
3507         long                    opc;
3508         int                     rc;
3509
3510         static const struct req_format *intent_fmts[REINT_MAX] = {
3511                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3512                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3513         };
3514
3515         ENTRY;
3516
3517         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3518         if (opc < 0)
3519                 RETURN(opc);
3520
3521         if (mdt_it_flavor[opcode].it_reint != opc) {
3522                 CERROR("Reint code %ld doesn't match intent: %d\n",
3523                        opc, opcode);
3524                 RETURN(err_serious(-EPROTO));
3525         }
3526
3527         /* Get lock from request for possible resent case. */
3528         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3529
3530         rc = mdt_reint_internal(info, lhc, opc);
3531
3532         /* Check whether the reply has been packed successfully. */
3533         if (mdt_info_req(info)->rq_repmsg != NULL)
3534                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3535         if (rep == NULL)
3536                 RETURN(err_serious(-EFAULT));
3537
3538         /* MDC expects this in any case */
3539         if (rc != 0)
3540                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3541
3542         /* the open lock or the lock for cross-ref object should be
3543          * returned to the client */
3544         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3545             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3546                 rep->lock_policy_res2 = 0;
3547                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3548                 RETURN(rc);
3549         }
3550
3551         rep->lock_policy_res2 = clear_serious(rc);
3552
3553         if (rep->lock_policy_res2 == -ENOENT &&
3554             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3555             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3556                 rep->lock_policy_res2 = 0;
3557
3558         lhc->mlh_reg_lh.cookie = 0ull;
3559         if (rc == -ENOTCONN || rc == -ENODEV ||
3560             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3561                 /*
3562                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3563                  * will be returned by rq_status, and client at ptlrpc layer
3564                  * will detect this, then disconnect, reconnect the import
3565                  * immediately, instead of impacting the following the rpc.
3566                  */
3567                 RETURN(rc);
3568         }
3569         /*
3570          * For other cases, the error will be returned by intent, and client
3571          * will retrieve the result from intent.
3572          */
3573         RETURN(ELDLM_LOCK_ABORTED);
3574 }
3575
3576 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3577 {
3578         int rc;
3579
3580         switch (itcode) {
3581         case IT_OPEN:
3582                 rc = MDT_IT_OPEN;
3583                 break;
3584         case IT_OPEN|IT_CREAT:
3585                 rc = MDT_IT_OCREAT;
3586                 break;
3587         case IT_CREAT:
3588                 rc = MDT_IT_CREATE;
3589                 break;
3590         case IT_READDIR:
3591                 rc = MDT_IT_READDIR;
3592                 break;
3593         case IT_GETATTR:
3594                 rc = MDT_IT_GETATTR;
3595                 break;
3596         case IT_LOOKUP:
3597                 rc = MDT_IT_LOOKUP;
3598                 break;
3599         case IT_UNLINK:
3600                 rc = MDT_IT_UNLINK;
3601                 break;
3602         case IT_TRUNC:
3603                 rc = MDT_IT_TRUNC;
3604                 break;
3605         case IT_GETXATTR:
3606                 rc = MDT_IT_GETXATTR;
3607                 break;
3608         case IT_LAYOUT:
3609                 rc = MDT_IT_LAYOUT;
3610                 break;
3611         case IT_QUOTA_DQACQ:
3612         case IT_QUOTA_CONN:
3613                 rc = MDT_IT_QUOTA;
3614                 break;
3615         default:
3616                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3617                 rc = -EINVAL;
3618                 break;
3619         }
3620         return rc;
3621 }
3622
3623 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3624                           struct mdt_thread_info *info,
3625                           struct ldlm_lock **lockp, __u64 flags)
3626 {
3627         struct req_capsule      *pill = info->mti_pill;
3628         struct ptlrpc_request   *req = mdt_info_req(info);
3629         struct mdt_it_flavor    *flv;
3630         int opc;
3631         int rc;
3632         ENTRY;
3633
3634         opc = mdt_intent_code(itopc);
3635         if (opc < 0)
3636                 RETURN(-EINVAL);
3637
3638         if (opc == MDT_IT_QUOTA) {
3639                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3640
3641                 if (qmt == NULL)
3642                         RETURN(-EOPNOTSUPP);
3643
3644                 if (mdt_rdonly(req->rq_export))
3645                         RETURN(-EROFS);
3646
3647                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3648                 /* pass the request to quota master */
3649                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3650                                                  mdt_info_req(info), lockp,
3651                                                  flags);
3652                 RETURN(rc);
3653         }
3654
3655         flv = &mdt_it_flavor[opc];
3656         if (flv->it_fmt != NULL)
3657                 req_capsule_extend(pill, flv->it_fmt);
3658
3659         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3660         if (rc < 0)
3661                 RETURN(rc);
3662
3663         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3664                 RETURN(-EROFS);
3665
3666         if (flv->it_act != NULL) {
3667                 struct ldlm_reply *rep;
3668
3669                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3670
3671                 /* execute policy */
3672                 rc = flv->it_act(opc, info, lockp, flags);
3673
3674                 /* Check whether the reply has been packed successfully. */
3675                 if (req->rq_repmsg != NULL) {
3676                         rep = req_capsule_server_get(info->mti_pill,
3677                                                      &RMF_DLM_REP);
3678                         rep->lock_policy_res2 =
3679                                 ptlrpc_status_hton(rep->lock_policy_res2);
3680                 }
3681         }
3682
3683         RETURN(rc);
3684 }
3685
3686 static int mdt_intent_policy(struct ldlm_namespace *ns,
3687                              struct ldlm_lock **lockp, void *req_cookie,
3688                              enum ldlm_mode mode, __u64 flags, void *data)
3689 {
3690         struct tgt_session_info *tsi;
3691         struct mdt_thread_info  *info;
3692         struct ptlrpc_request   *req  =  req_cookie;
3693         struct ldlm_intent      *it;
3694         struct req_capsule      *pill;
3695         int rc;
3696
3697         ENTRY;
3698
3699         LASSERT(req != NULL);
3700
3701         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3702
3703         info = tsi2mdt_info(tsi);
3704         LASSERT(info != NULL);
3705         pill = info->mti_pill;
3706         LASSERT(pill->rc_req == req);
3707
3708         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3709                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3710                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3711                 if (it != NULL) {
3712                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3713                         if (rc == 0)
3714                                 rc = ELDLM_OK;
3715
3716                         /* Lock without inodebits makes no sense and will oops
3717                          * later in ldlm. Let's check it now to see if we have
3718                          * ibits corrupted somewhere in mdt_intent_opc().
3719                          * The case for client miss to set ibits has been
3720                          * processed by others. */
3721                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3722                                         lr_type == LDLM_IBITS,
3723                                      info->mti_dlm_req->lock_desc.\
3724                                         l_policy_data.l_inodebits.bits != 0));
3725                 } else
3726                         rc = err_serious(-EFAULT);
3727         } else {
3728                 /* No intent was provided */
3729                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3730                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3731                 rc = req_capsule_server_pack(pill);
3732                 if (rc)
3733                         rc = err_serious(rc);
3734         }
3735         mdt_thread_info_fini(info);
3736         RETURN(rc);
3737 }
3738
3739 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3740 {
3741         struct seq_server_site  *ss = mdt_seq_site(mdt);
3742
3743         if (ss->ss_node_id == 0)
3744                 return;
3745
3746         if (ss->ss_client_seq != NULL) {
3747                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3748                 ss->ss_client_seq->lcs_exp = NULL;
3749         }
3750
3751         if (ss->ss_server_fld != NULL) {
3752                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3753                 ss->ss_server_fld->lsf_control_exp = NULL;
3754         }
3755 }
3756
3757 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3758 {
3759         struct seq_server_site *ss = mdt_seq_site(mdt);
3760
3761         if (ss == NULL)
3762                 return;
3763
3764         if (ss->ss_server_seq != NULL)
3765                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3766 }
3767
3768 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3769 {
3770         mdt_seq_fini_cli(mdt);
3771         mdt_deregister_seq_exp(mdt);
3772
3773         return seq_site_fini(env, mdt_seq_site(mdt));
3774 }
3775
3776 /**
3777  * It will retrieve its FLDB entries from MDT0, and it only happens
3778  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3779  * and it needs to refresh FLDB from the MDT0.
3780  **/
3781 static int mdt_register_lwp_callback(void *data)
3782 {
3783         struct lu_env           env;
3784         struct mdt_device       *mdt = data;
3785         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3786         int                     rc;
3787         ENTRY;
3788
3789         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3790
3791         rc = lu_env_init(&env, LCT_MD_THREAD);
3792         if (rc < 0) {
3793                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3794                 RETURN(rc);
3795         }
3796
3797         /* Allocate new sequence now to avoid creating local transaction
3798          * in the normal transaction process */
3799         rc = seq_server_check_and_alloc_super(&env,
3800                                               mdt_seq_site(mdt)->ss_server_seq);
3801         if (rc < 0)
3802                 GOTO(out, rc);
3803
3804         if (fld->lsf_new) {
3805                 rc = fld_update_from_controller(&env, fld);
3806                 if (rc != 0) {
3807                         CERROR("%s: cannot update controller: rc = %d\n",
3808                                mdt_obd_name(mdt), rc);
3809                         GOTO(out, rc);
3810                 }
3811         }
3812 out:
3813         lu_env_fini(&env);
3814         RETURN(rc);
3815 }
3816
3817 static int mdt_register_seq_exp(struct mdt_device *mdt)
3818 {
3819         struct seq_server_site  *ss = mdt_seq_site(mdt);
3820         char                    *lwp_name = NULL;
3821         int                     rc;
3822
3823         if (ss->ss_node_id == 0)
3824                 return 0;
3825
3826         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3827         if (lwp_name == NULL)
3828                 GOTO(out_free, rc = -ENOMEM);
3829
3830         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3831         if (rc != 0)
3832                 GOTO(out_free, rc);
3833
3834         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3835                                       NULL, NULL);
3836         if (rc != 0)
3837                 GOTO(out_free, rc);
3838
3839         rc = lustre_register_lwp_item(lwp_name,
3840                                       &ss->ss_server_fld->lsf_control_exp,
3841                                       mdt_register_lwp_callback, mdt);
3842         if (rc != 0) {
3843                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3844                 ss->ss_client_seq->lcs_exp = NULL;
3845                 GOTO(out_free, rc);
3846         }
3847 out_free:
3848         if (lwp_name != NULL)
3849                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3850
3851         return rc;
3852 }
3853
3854 /*
3855  * Init client sequence manager which is used by local MDS to talk to sequence
3856  * controller on remote node.
3857  */
3858 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3859 {
3860         struct seq_server_site  *ss = mdt_seq_site(mdt);
3861         int                     rc;
3862         char                    *prefix;
3863         ENTRY;
3864
3865         /* check if this is adding the first MDC and controller is not yet
3866          * initialized. */
3867         OBD_ALLOC_PTR(ss->ss_client_seq);
3868         if (ss->ss_client_seq == NULL)
3869                 RETURN(-ENOMEM);
3870
3871         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3872         if (prefix == NULL) {
3873                 OBD_FREE_PTR(ss->ss_client_seq);
3874                 ss->ss_client_seq = NULL;
3875                 RETURN(-ENOMEM);
3876         }
3877
3878         /* Note: seq_client_fini will be called in seq_site_fini */
3879         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3880         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3881                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3882                                                             NULL);
3883         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3884         if (rc != 0) {
3885                 OBD_FREE_PTR(ss->ss_client_seq);
3886                 ss->ss_client_seq = NULL;
3887                 RETURN(rc);
3888         }
3889
3890         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3891
3892         RETURN(rc);
3893 }
3894
3895 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3896 {
3897         struct seq_server_site  *ss;
3898         int                     rc;
3899         ENTRY;
3900
3901         ss = mdt_seq_site(mdt);
3902         /* init sequence controller server(MDT0) */
3903         if (ss->ss_node_id == 0) {
3904                 OBD_ALLOC_PTR(ss->ss_control_seq);
3905                 if (ss->ss_control_seq == NULL)
3906                         RETURN(-ENOMEM);
3907
3908                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3909                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3910                                      ss);
3911                 if (rc)
3912                         GOTO(out_seq_fini, rc);
3913         }
3914
3915         /* Init normal sequence server */
3916         OBD_ALLOC_PTR(ss->ss_server_seq);
3917         if (ss->ss_server_seq == NULL)
3918                 GOTO(out_seq_fini, rc = -ENOMEM);
3919
3920         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3921                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3922         if (rc)
3923                 GOTO(out_seq_fini, rc);
3924
3925         /* init seq client for seq server to talk to seq controller(MDT0) */
3926         rc = mdt_seq_init_cli(env, mdt);
3927         if (rc != 0)
3928                 GOTO(out_seq_fini, rc);
3929
3930         if (ss->ss_node_id != 0)
3931                 /* register controller export through lwp */
3932                 rc = mdt_register_seq_exp(mdt);
3933
3934         EXIT;
3935 out_seq_fini:
3936         if (rc)
3937                 mdt_seq_fini(env, mdt);
3938
3939         return rc;
3940 }
3941
3942 /*
3943  * FLD wrappers
3944  */
3945 static int mdt_fld_fini(const struct lu_env *env,
3946                         struct mdt_device *m)
3947 {
3948         struct seq_server_site *ss = mdt_seq_site(m);
3949         ENTRY;
3950
3951         if (ss && ss->ss_server_fld) {
3952                 fld_server_fini(env, ss->ss_server_fld);
3953                 OBD_FREE_PTR(ss->ss_server_fld);
3954                 ss->ss_server_fld = NULL;
3955         }
3956
3957         RETURN(0);
3958 }
3959
3960 static int mdt_fld_init(const struct lu_env *env,
3961                         const char *uuid,
3962                         struct mdt_device *m)
3963 {
3964         struct seq_server_site *ss;
3965         int rc;
3966         ENTRY;
3967
3968         ss = mdt_seq_site(m);
3969
3970         OBD_ALLOC_PTR(ss->ss_server_fld);
3971         if (ss->ss_server_fld == NULL)
3972                 RETURN(rc = -ENOMEM);
3973
3974         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3975                              LU_SEQ_RANGE_MDT);
3976         if (rc) {
3977                 OBD_FREE_PTR(ss->ss_server_fld);
3978                 ss->ss_server_fld = NULL;
3979                 RETURN(rc);
3980         }
3981
3982         RETURN(0);
3983 }
3984
3985 static void mdt_stack_pre_fini(const struct lu_env *env,
3986                            struct mdt_device *m, struct lu_device *top)
3987 {
3988         struct lustre_cfg_bufs  *bufs;
3989         struct lustre_cfg       *lcfg;
3990         struct mdt_thread_info  *info;
3991         ENTRY;
3992
3993         LASSERT(top);
3994
3995         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3996         LASSERT(info != NULL);
3997
3998         bufs = &info->mti_u.bufs;
3999
4000         LASSERT(m->mdt_child_exp);
4001         LASSERT(m->mdt_child_exp->exp_obd);
4002
4003         /* process cleanup, pass mdt obd name to get obd umount flags */
4004         /* XXX: this is needed because all layers are referenced by
4005          * objects (some of them are pinned by osd, for example *
4006          * the proper solution should be a model where object used
4007          * by osd only doesn't have mdt/mdd slices -bzzz */
4008         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4009         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4010         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4011         if (lcfg == NULL)
4012                 RETURN_EXIT;
4013
4014         top->ld_ops->ldo_process_config(env, top, lcfg);
4015         lustre_cfg_free(lcfg);
4016         EXIT;
4017 }
4018
4019 static void mdt_stack_fini(const struct lu_env *env,
4020                            struct mdt_device *m, struct lu_device *top)
4021 {
4022         struct obd_device       *obd = mdt2obd_dev(m);
4023         struct lustre_cfg_bufs  *bufs;
4024         struct lustre_cfg       *lcfg;
4025         struct mdt_thread_info  *info;
4026         char                     flags[3] = "";
4027         ENTRY;
4028
4029         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4030         LASSERT(info != NULL);
4031
4032         lu_dev_del_linkage(top->ld_site, top);
4033
4034         lu_site_purge(env, top->ld_site, -1);
4035
4036         bufs = &info->mti_u.bufs;
4037         /* process cleanup, pass mdt obd name to get obd umount flags */
4038         /* another purpose is to let all layers to release their objects */
4039         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4040         if (obd->obd_force)
4041                 strcat(flags, "F");
4042         if (obd->obd_fail)
4043                 strcat(flags, "A");
4044         lustre_cfg_bufs_set_string(bufs, 1, flags);
4045         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4046         if (lcfg == NULL)
4047                 RETURN_EXIT;
4048
4049         LASSERT(top);
4050         top->ld_ops->ldo_process_config(env, top, lcfg);
4051         lustre_cfg_free(lcfg);
4052
4053         lu_site_purge(env, top->ld_site, -1);
4054
4055         m->mdt_child = NULL;
4056         m->mdt_bottom = NULL;
4057
4058         obd_disconnect(m->mdt_child_exp);
4059         m->mdt_child_exp = NULL;
4060
4061         obd_disconnect(m->mdt_bottom_exp);
4062         m->mdt_child_exp = NULL;
4063 }
4064
4065 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4066                                const char *next, struct obd_export **exp)
4067 {
4068         struct obd_connect_data *data = NULL;
4069         struct obd_device       *obd;
4070         int                      rc;
4071         ENTRY;
4072
4073         OBD_ALLOC_PTR(data);
4074         if (data == NULL)
4075                 GOTO(out, rc = -ENOMEM);
4076
4077         obd = class_name2obd(next);
4078         if (obd == NULL) {
4079                 CERROR("%s: can't locate next device: %s\n",
4080                        mdt_obd_name(m), next);
4081                 GOTO(out, rc = -ENOTCONN);
4082         }
4083
4084         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4085         data->ocd_version = LUSTRE_VERSION_CODE;
4086
4087         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4088         if (rc) {
4089                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4090                        mdt_obd_name(m), next, rc);
4091                 GOTO(out, rc);
4092         }
4093
4094 out:
4095         if (data)
4096                 OBD_FREE_PTR(data);
4097         RETURN(rc);
4098 }
4099
4100 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4101                           struct lustre_cfg *cfg)
4102 {
4103         char                   *dev = lustre_cfg_string(cfg, 0);
4104         int                     rc, name_size, uuid_size;
4105         char                   *name, *uuid, *p;
4106         struct lustre_cfg_bufs *bufs;
4107         struct lustre_cfg      *lcfg;
4108         struct obd_device      *obd;
4109         struct lustre_profile  *lprof;
4110         struct lu_site         *site;
4111         ENTRY;
4112
4113         /* in 1.8 we had the only device in the stack - MDS.
4114          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4115          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4116          * to generate names and setup MDT, MDD. MDT will be using
4117          * generated name to connect to MDD. for MDD the next device
4118          * will be LOD with name taken from so called "profile" which
4119          * is generated by mount_option line
4120          *
4121          * 1.8 MGS generates config. commands like this:
4122          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4123          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4124          * 2.0 MGS generates config. commands like this:
4125          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4126          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4127          *                    3:lustre-MDT0000-mdtlov  4:f
4128          *
4129          * we generate MDD name from MDT one, just replacing T with D
4130          *
4131          * after all the preparations, the logical equivalent will be
4132          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4133          *                    3:lustre-MDT0000-mdtlov  4:f
4134          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4135          *                    3:lustre-MDD0000  4:f
4136          *
4137          *  notice we build the stack from down to top: MDD first, then MDT */
4138
4139         name_size = MAX_OBD_NAME;
4140         uuid_size = MAX_OBD_NAME;
4141
4142         OBD_ALLOC(name, name_size);
4143         OBD_ALLOC(uuid, uuid_size);
4144         if (name == NULL || uuid == NULL)
4145                 GOTO(cleanup_mem, rc = -ENOMEM);
4146
4147         OBD_ALLOC_PTR(bufs);
4148         if (!bufs)
4149                 GOTO(cleanup_mem, rc = -ENOMEM);
4150
4151         strcpy(name, dev);
4152         p = strstr(name, "-MDT");
4153         if (p == NULL)
4154                 GOTO(free_bufs, rc = -ENOMEM);
4155         p[3] = 'D';
4156
4157         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4158
4159         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4160         if (lprof == NULL || lprof->lp_dt == NULL) {
4161                 CERROR("can't find the profile: %s\n",
4162                        lustre_cfg_string(cfg, 0));
4163                 GOTO(free_bufs, rc = -EINVAL);
4164         }
4165
4166         lustre_cfg_bufs_reset(bufs, name);
4167         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4168         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4169         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4170
4171         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4172         if (lcfg == NULL)
4173                 GOTO(put_profile, rc = -ENOMEM);
4174
4175         rc = class_attach(lcfg);
4176         if (rc)
4177                 GOTO(lcfg_cleanup, rc);
4178
4179         obd = class_name2obd(name);
4180         if (!obd) {
4181                 CERROR("Can not find obd %s (%s in config)\n",
4182                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4183                 GOTO(lcfg_cleanup, rc = -EINVAL);
4184         }
4185
4186         lustre_cfg_free(lcfg);
4187
4188         lustre_cfg_bufs_reset(bufs, name);
4189         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4190         lustre_cfg_bufs_set_string(bufs, 2, dev);
4191         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4192
4193         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4194         if (lcfg == NULL)
4195                 GOTO(class_detach, rc = -ENOMEM);
4196
4197         rc = class_setup(obd, lcfg);
4198         if (rc)
4199                 GOTO(class_detach, rc);
4200
4201         /* connect to MDD we just setup */
4202         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4203         if (rc)
4204                 GOTO(class_detach, rc);
4205
4206         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4207         LASSERT(site);
4208         LASSERT(mdt_lu_site(mdt) == NULL);
4209         mdt->mdt_lu_dev.ld_site = site;
4210         site->ls_top_dev = &mdt->mdt_lu_dev;
4211         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4212
4213         /* now connect to bottom OSD */
4214         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4215         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4216         if (rc)
4217                 GOTO(class_detach, rc);
4218         mdt->mdt_bottom =
4219                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4220
4221         rc = lu_env_refill((struct lu_env *)env);
4222         if (rc != 0)
4223                 CERROR("Failure to refill session: '%d'\n", rc);
4224
4225         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4226
4227         EXIT;
4228 class_detach:
4229         if (rc)
4230                 class_detach(obd, lcfg);
4231 lcfg_cleanup:
4232         lustre_cfg_free(lcfg);
4233 put_profile:
4234         class_put_profile(lprof);
4235 free_bufs:
4236         OBD_FREE_PTR(bufs);
4237 cleanup_mem:
4238         if (name)
4239                 OBD_FREE(name, name_size);
4240         if (uuid)
4241                 OBD_FREE(uuid, uuid_size);
4242         RETURN(rc);
4243 }
4244
4245 /* setup quota master target on MDT0 */
4246 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4247                           struct lustre_cfg *cfg)
4248 {
4249         struct obd_device       *obd;
4250         char                    *dev = lustre_cfg_string(cfg, 0);
4251         char                    *qmtname, *uuid, *p;
4252         struct lustre_cfg_bufs  *bufs;
4253         struct lustre_cfg       *lcfg;
4254         struct lustre_profile   *lprof;
4255         struct obd_connect_data *data;
4256         int                      rc;
4257         ENTRY;
4258
4259         LASSERT(mdt->mdt_qmt_exp == NULL);
4260         LASSERT(mdt->mdt_qmt_dev == NULL);
4261
4262         /* quota master is on MDT0 only for now */
4263         if (mdt->mdt_seq_site.ss_node_id != 0)
4264                 RETURN(0);
4265
4266         /* MGS generates config commands which look as follows:
4267          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4268          *                    3:lustre-MDT0000-mdtlov  4:f
4269          *
4270          * We generate the QMT name from the MDT one, just replacing MD with QM
4271          * after all the preparations, the logical equivalent will be:
4272          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4273          *                    3:lustre-MDT0000-osd  4:f */
4274         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4275         OBD_ALLOC(uuid, UUID_MAX);
4276         OBD_ALLOC_PTR(bufs);
4277         OBD_ALLOC_PTR(data);
4278         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4279                 GOTO(cleanup_mem, rc = -ENOMEM);
4280
4281         strcpy(qmtname, dev);
4282         p = strstr(qmtname, "-MDT");
4283         if (p == NULL)
4284                 GOTO(cleanup_mem, rc = -ENOMEM);
4285         /* replace MD with QM */
4286         p[1] = 'Q';
4287         p[2] = 'M';
4288
4289         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4290
4291         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4292         if (lprof == NULL || lprof->lp_dt == NULL) {
4293                 CERROR("can't find profile for %s\n",
4294                        lustre_cfg_string(cfg, 0));
4295                 GOTO(cleanup_mem, rc = -EINVAL);
4296         }
4297
4298         lustre_cfg_bufs_reset(bufs, qmtname);
4299         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4300         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4301         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4302
4303         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4304         if (lcfg == NULL)
4305                 GOTO(put_profile, rc = -ENOMEM);
4306
4307         rc = class_attach(lcfg);
4308         if (rc)
4309                 GOTO(lcfg_cleanup, rc);
4310
4311         obd = class_name2obd(qmtname);
4312         if (!obd) {
4313                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4314                        lustre_cfg_string(cfg, 0));
4315                 GOTO(lcfg_cleanup, rc = -EINVAL);
4316         }
4317
4318         lustre_cfg_free(lcfg);
4319
4320         lustre_cfg_bufs_reset(bufs, qmtname);
4321         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4322         lustre_cfg_bufs_set_string(bufs, 2, dev);
4323
4324         /* for quota, the next device should be the OSD device */
4325         lustre_cfg_bufs_set_string(bufs, 3,
4326                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4327
4328         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4329         if (lcfg == NULL)
4330                 GOTO(class_detach, rc = -ENOMEM);
4331
4332         rc = class_setup(obd, lcfg);
4333         if (rc)
4334                 GOTO(class_detach, rc);
4335
4336         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4337
4338         /* configure local quota objects */
4339         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4340                                                    &mdt->mdt_lu_dev,
4341                                                    mdt->mdt_qmt_dev);
4342         if (rc)
4343                 GOTO(class_cleanup, rc);
4344
4345         /* connect to quota master target */
4346         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4347         data->ocd_version = LUSTRE_VERSION_CODE;
4348         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4349                          data, NULL);
4350         if (rc) {
4351                 CERROR("cannot connect to quota master device %s (%d)\n",
4352                        qmtname, rc);
4353                 GOTO(class_cleanup, rc);
4354         }
4355
4356         EXIT;
4357 class_cleanup:
4358         if (rc) {
4359                 class_manual_cleanup(obd);
4360                 mdt->mdt_qmt_dev = NULL;
4361         }
4362 class_detach:
4363         if (rc)
4364                 class_detach(obd, lcfg);
4365 lcfg_cleanup:
4366         lustre_cfg_free(lcfg);
4367 put_profile:
4368         class_put_profile(lprof);
4369 cleanup_mem:
4370         if (bufs)
4371                 OBD_FREE_PTR(bufs);
4372         if (qmtname)
4373                 OBD_FREE(qmtname, MAX_OBD_NAME);
4374         if (uuid)
4375                 OBD_FREE(uuid, UUID_MAX);
4376         if (data)
4377                 OBD_FREE_PTR(data);
4378         return rc;
4379 }
4380
4381 /* Shutdown quota master target associated with mdt */
4382 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4383 {
4384         ENTRY;
4385
4386         if (mdt->mdt_qmt_exp == NULL)
4387                 RETURN_EXIT;
4388         LASSERT(mdt->mdt_qmt_dev != NULL);
4389
4390         /* the qmt automatically shuts down when the mdt disconnects */
4391         obd_disconnect(mdt->mdt_qmt_exp);
4392         mdt->mdt_qmt_exp = NULL;
4393         mdt->mdt_qmt_dev = NULL;
4394         EXIT;
4395 }
4396
4397 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4398  * for now. This will be removed along with converting rest of MDT code
4399  * to use tgt_session_info */
4400 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4401 {
4402         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4403         int                      rc;
4404
4405         rc = mdt_getxattr(info);
4406
4407         mdt_thread_info_fini(info);
4408         return rc;
4409 }
4410
4411 static struct tgt_handler mdt_tgt_handlers[] = {
4412 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4413                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4414                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4415 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4416                 0,                      MDS_DISCONNECT, tgt_disconnect,
4417                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4418 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4419                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4420                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4421 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4422 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4423 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4424 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4425                                                         mdt_getattr_name),
4426 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4427 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4428 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4429 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4430 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4431 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4432 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4433 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4434                                                         mdt_hsm_progress),
4435 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4436                                                         mdt_hsm_ct_register),
4437 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4438                                                         mdt_hsm_ct_unregister),
4439 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4440                                                         mdt_hsm_state_get),
4441 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4442                                                         mdt_hsm_state_set),
4443 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4444 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4445                                                         mdt_hsm_request),
4446 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4447             MDS_SWAP_LAYOUTS,
4448             mdt_swap_layouts),
4449 };
4450
4451 static struct tgt_handler mdt_sec_ctx_ops[] = {
4452 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4453 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4454 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4455 };
4456
4457 static struct tgt_handler mdt_quota_ops[] = {
4458 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4459 };
4460
4461 static struct tgt_opc_slice mdt_common_slice[] = {
4462         {
4463                 .tos_opc_start  = MDS_FIRST_OPC,
4464                 .tos_opc_end    = MDS_LAST_OPC,
4465                 .tos_hs         = mdt_tgt_handlers
4466         },
4467         {
4468                 .tos_opc_start  = OBD_FIRST_OPC,
4469                 .tos_opc_end    = OBD_LAST_OPC,
4470                 .tos_hs         = tgt_obd_handlers
4471         },
4472         {
4473                 .tos_opc_start  = LDLM_FIRST_OPC,
4474                 .tos_opc_end    = LDLM_LAST_OPC,
4475                 .tos_hs         = tgt_dlm_handlers
4476         },
4477         {
4478                 .tos_opc_start  = SEC_FIRST_OPC,
4479                 .tos_opc_end    = SEC_LAST_OPC,
4480                 .tos_hs         = mdt_sec_ctx_ops
4481         },
4482         {
4483                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4484                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4485                 .tos_hs         = tgt_out_handlers
4486         },
4487         {
4488                 .tos_opc_start  = FLD_FIRST_OPC,
4489                 .tos_opc_end    = FLD_LAST_OPC,
4490                 .tos_hs         = fld_handlers
4491         },
4492         {
4493                 .tos_opc_start  = SEQ_FIRST_OPC,
4494                 .tos_opc_end    = SEQ_LAST_OPC,
4495                 .tos_hs         = seq_handlers
4496         },
4497         {
4498                 .tos_opc_start  = QUOTA_DQACQ,
4499                 .tos_opc_end    = QUOTA_LAST_OPC,
4500                 .tos_hs         = mdt_quota_ops
4501         },
4502         {
4503                 .tos_opc_start  = LLOG_FIRST_OPC,
4504                 .tos_opc_end    = LLOG_LAST_OPC,
4505                 .tos_hs         = tgt_llog_handlers
4506         },
4507         {
4508                 .tos_opc_start  = LFSCK_FIRST_OPC,
4509                 .tos_opc_end    = LFSCK_LAST_OPC,
4510                 .tos_hs         = tgt_lfsck_handlers
4511         },
4512
4513         {
4514                 .tos_hs         = NULL
4515         }
4516 };
4517
4518 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4519 {
4520         struct md_device *next = m->mdt_child;
4521         struct lu_device *d = &m->mdt_lu_dev;
4522         struct obd_device *obd = mdt2obd_dev(m);
4523         struct lfsck_stop stop;
4524
4525         ENTRY;
4526         stop.ls_status = LS_PAUSED;
4527         stop.ls_flags = 0;
4528         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4529
4530         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4531         ping_evictor_stop();
4532
4533         if (m->mdt_opts.mo_coordinator)
4534                 mdt_hsm_cdt_stop(m);
4535
4536         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4537         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4538
4539         if (m->mdt_namespace != NULL)
4540                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4541                                           d->ld_obd->obd_force);
4542
4543         obd_exports_barrier(obd);
4544         obd_zombie_barrier();
4545
4546         mdt_quota_fini(env, m);
4547
4548         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4549
4550         /* Calling the cleanup functions in the same order as in the mdt_init0
4551          * error path
4552          */
4553         mdt_procfs_fini(m);
4554
4555         target_recovery_fini(obd);
4556         upcall_cache_cleanup(m->mdt_identity_cache);
4557         m->mdt_identity_cache = NULL;
4558
4559         mdt_fs_cleanup(env, m);
4560
4561         tgt_fini(env, &m->mdt_lut);
4562
4563         mdt_hsm_cdt_fini(m);
4564
4565         if (m->mdt_namespace != NULL) {
4566                 ldlm_namespace_free_post(m->mdt_namespace);
4567                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4568         }
4569
4570         if (m->mdt_md_root != NULL) {
4571                 mdt_object_put(env, m->mdt_md_root);
4572                 m->mdt_md_root = NULL;
4573         }
4574
4575         mdt_seq_fini(env, m);
4576
4577         mdt_fld_fini(env, m);
4578
4579         /*
4580          * Finish the stack
4581          */
4582         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4583
4584         LASSERT(atomic_read(&d->ld_ref) == 0);
4585
4586         server_put_mount(mdt_obd_name(m), true);
4587
4588         EXIT;
4589 }
4590
4591 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4592
4593 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4594                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4595 {
4596         struct mdt_thread_info    *info;
4597         struct obd_device         *obd;
4598         const char                *dev = lustre_cfg_string(cfg, 0);
4599         const char                *num = lustre_cfg_string(cfg, 2);
4600         struct lustre_mount_info  *lmi = NULL;
4601         struct lustre_sb_info     *lsi;
4602         struct lu_site            *s;
4603         struct seq_server_site    *ss_site;
4604         const char                *identity_upcall = "NONE";
4605         struct md_device          *next;
4606         int                        rc;
4607         long                       node_id;
4608         mntopt_t                   mntopts;
4609         ENTRY;
4610
4611         lu_device_init(&m->mdt_lu_dev, ldt);
4612         /*
4613          * Environment (env) might be missing mdt_thread_key values at that
4614          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4615          * mode.
4616          *
4617          * Usually device allocation path doesn't use module key values, but
4618          * mdt has to do a lot of work here, so allocate key value.
4619          */
4620         rc = lu_env_refill((struct lu_env *)env);
4621         if (rc != 0)
4622                 RETURN(rc);
4623
4624         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4625         LASSERT(info != NULL);
4626
4627         obd = class_name2obd(dev);
4628         LASSERT(obd != NULL);
4629
4630         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4631         m->mdt_opts.mo_evict_tgt_nids = 1;
4632         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4633
4634         /* default is coordinator off, it is started through conf_param
4635          * or /proc */
4636         m->mdt_opts.mo_coordinator = 0;
4637
4638         lmi = server_get_mount(dev);
4639         if (lmi == NULL) {
4640                 CERROR("Cannot get mount info for %s!\n", dev);
4641                 RETURN(-EFAULT);
4642         } else {
4643                 lsi = s2lsi(lmi->lmi_sb);
4644                 /* CMD is supported only in IAM mode */
4645                 LASSERT(num);
4646                 node_id = simple_strtol(num, NULL, 10);
4647                 obd->u.obt.obt_magic = OBT_MAGIC;
4648                 if (lsi->lsi_lmd != NULL &&
4649                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4650                         m->mdt_skip_lfsck = 1;
4651         }
4652
4653         m->mdt_squash.rsi_uid = 0;
4654         m->mdt_squash.rsi_gid = 0;
4655         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4656         init_rwsem(&m->mdt_squash.rsi_sem);
4657         spin_lock_init(&m->mdt_lock);
4658         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4659         m->mdt_enable_remote_dir = 0;
4660         m->mdt_enable_remote_dir_gid = 0;
4661
4662         atomic_set(&m->mdt_mds_mds_conns, 0);
4663         atomic_set(&m->mdt_async_commit_count, 0);
4664
4665         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4666         m->mdt_lu_dev.ld_obd = obd;
4667         /* Set this lu_device to obd for error handling purposes. */
4668         obd->obd_lu_dev = &m->mdt_lu_dev;
4669
4670         /* init the stack */
4671         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4672         if (rc) {
4673                 CERROR("%s: Can't init device stack, rc %d\n",
4674                        mdt_obd_name(m), rc);
4675                 GOTO(err_lmi, rc);
4676         }
4677
4678         s = mdt_lu_site(m);
4679         ss_site = mdt_seq_site(m);
4680         s->ld_seq_site = ss_site;
4681         ss_site->ss_lu = s;
4682
4683         /* set server index */
4684         ss_site->ss_node_id = node_id;
4685
4686         /* failover is the default
4687          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4688          * assumed whose ss_node_id == 0 XXX
4689          * */
4690         obd->obd_replayable = 1;
4691         /* No connection accepted until configurations will finish */
4692         obd->obd_no_conn = 1;
4693
4694         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4695                 char *str = lustre_cfg_string(cfg, 4);
4696                 if (strchr(str, 'n')) {
4697                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4698                         obd->obd_replayable = 0;
4699                 }
4700         }
4701
4702         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4703         if (rc)
4704                 GOTO(err_fini_stack, rc);
4705
4706         rc = mdt_seq_init(env, m);
4707         if (rc)
4708                 GOTO(err_fini_fld, rc);
4709
4710         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4711                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4712         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4713                                               LDLM_NAMESPACE_SERVER,
4714                                               LDLM_NAMESPACE_GREEDY,
4715                                               LDLM_NS_TYPE_MDT);
4716         if (m->mdt_namespace == NULL)
4717                 GOTO(err_fini_seq, rc = -ENOMEM);
4718
4719         m->mdt_namespace->ns_lvbp = m;
4720         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4721
4722         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4723         /* set obd_namespace for compatibility with old code */
4724         obd->obd_namespace = m->mdt_namespace;
4725
4726         rc = mdt_hsm_cdt_init(m);
4727         if (rc != 0) {
4728                 CERROR("%s: error initializing coordinator, rc %d\n",
4729                        mdt_obd_name(m), rc);
4730                 GOTO(err_free_ns, rc);
4731         }
4732
4733         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4734                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4735                       OBD_FAIL_MDS_ALL_REPLY_NET);
4736         if (rc)
4737                 GOTO(err_free_hsm, rc);
4738
4739         rc = mdt_fs_setup(env, m, obd, lsi);
4740         if (rc)
4741                 GOTO(err_tgt, rc);
4742
4743         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
4744
4745         next = m->mdt_child;
4746         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4747                                          &mntopts);
4748         if (rc)
4749                 GOTO(err_fs_cleanup, rc);
4750
4751         if (mntopts & MNTOPT_USERXATTR)
4752                 m->mdt_opts.mo_user_xattr = 1;
4753         else
4754                 m->mdt_opts.mo_user_xattr = 0;
4755
4756         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4757         if (rc)
4758                 GOTO(err_fs_cleanup, rc);
4759
4760         if (mntopts & MNTOPT_ACL)
4761                 m->mdt_opts.mo_acl = 1;
4762         else
4763                 m->mdt_opts.mo_acl = 0;
4764
4765         /* XXX: to support suppgid for ACL, we enable identity_upcall
4766          * by default, otherwise, maybe got unexpected -EACCESS. */
4767         if (m->mdt_opts.mo_acl)
4768                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4769
4770         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4771                                                 identity_upcall,
4772                                                 &mdt_identity_upcall_cache_ops);
4773         if (IS_ERR(m->mdt_identity_cache)) {
4774                 rc = PTR_ERR(m->mdt_identity_cache);
4775                 m->mdt_identity_cache = NULL;
4776                 GOTO(err_fs_cleanup, rc);
4777         }
4778
4779         rc = mdt_procfs_init(m, dev);
4780         if (rc) {
4781                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4782                 GOTO(err_recovery, rc);
4783         }
4784
4785         rc = mdt_quota_init(env, m, cfg);
4786         if (rc)
4787                 GOTO(err_procfs, rc);
4788
4789         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4790         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4791                            "mdt_ldlm_client", m->mdt_ldlm_client);
4792
4793         ping_evictor_start();
4794
4795         /* recovery will be started upon mdt_prepare()
4796          * when the whole stack is complete and ready
4797          * to serve the requests */
4798
4799         /* Reduce the initial timeout on an MDS because it doesn't need such
4800          * a long timeout as an OST does. Adaptive timeouts will adjust this
4801          * value appropriately. */
4802         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4803                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4804
4805         RETURN(0);
4806 err_procfs:
4807         mdt_procfs_fini(m);
4808 err_recovery:
4809         target_recovery_fini(obd);
4810         upcall_cache_cleanup(m->mdt_identity_cache);
4811         m->mdt_identity_cache = NULL;
4812 err_fs_cleanup:
4813         mdt_fs_cleanup(env, m);
4814 err_tgt:
4815         tgt_fini(env, &m->mdt_lut);
4816 err_free_hsm:
4817         mdt_hsm_cdt_fini(m);
4818 err_free_ns:
4819         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4820         obd->obd_namespace = m->mdt_namespace = NULL;
4821 err_fini_seq:
4822         mdt_seq_fini(env, m);
4823 err_fini_fld:
4824         mdt_fld_fini(env, m);
4825 err_fini_stack:
4826         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4827 err_lmi:
4828         if (lmi)
4829                 server_put_mount(dev, true);
4830         return(rc);
4831 }
4832
4833 /* For interoperability, the left element is old parameter, the right one
4834  * is the new version of the parameter, if some parameter is deprecated,
4835  * the new version should be set as NULL. */
4836 static struct cfg_interop_param mdt_interop_param[] = {
4837         { "mdt.group_upcall",   NULL },
4838         { "mdt.quota_type",     NULL },
4839         { "mdd.quota_type",     NULL },
4840         { "mdt.som",            NULL },
4841         { "mdt.rootsquash",     "mdt.root_squash" },
4842         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4843         { NULL }
4844 };
4845
4846 /* used by MGS to process specific configurations */
4847 static int mdt_process_config(const struct lu_env *env,
4848                               struct lu_device *d, struct lustre_cfg *cfg)
4849 {
4850         struct mdt_device *m = mdt_dev(d);
4851         struct md_device *md_next = m->mdt_child;
4852         struct lu_device *next = md2lu_dev(md_next);
4853         int rc;
4854         ENTRY;
4855
4856         switch (cfg->lcfg_command) {
4857         case LCFG_PARAM: {
4858                 struct obd_device          *obd = d->ld_obd;
4859
4860                 /* For interoperability */
4861                 struct cfg_interop_param   *ptr = NULL;
4862                 struct lustre_cfg          *old_cfg = NULL;
4863                 char                       *param = NULL;
4864
4865                 param = lustre_cfg_string(cfg, 1);
4866                 if (param == NULL) {
4867                         CERROR("param is empty\n");
4868                         rc = -EINVAL;
4869                         break;
4870                 }
4871
4872                 ptr = class_find_old_param(param, mdt_interop_param);
4873                 if (ptr != NULL) {
4874                         if (ptr->new_param == NULL) {
4875                                 rc = 0;
4876                                 CWARN("For interoperability, skip this %s."
4877                                       " It is obsolete.\n", ptr->old_param);
4878                                 break;
4879                         }
4880
4881                         CWARN("Found old param %s, changed it to %s.\n",
4882                               ptr->old_param, ptr->new_param);
4883
4884                         old_cfg = cfg;
4885                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4886                         if (IS_ERR(cfg)) {
4887                                 rc = PTR_ERR(cfg);
4888                                 break;
4889                         }
4890                 }
4891
4892                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4893                                               cfg, obd);
4894                 if (rc > 0 || rc == -ENOSYS) {
4895                         /* is it an HSM var ? */
4896                         rc = class_process_proc_param(PARAM_HSM,
4897                                                       hsm_cdt_get_proc_vars(),
4898                                                       cfg, obd);
4899                         if (rc > 0 || rc == -ENOSYS)
4900                                 /* we don't understand; pass it on */
4901                                 rc = next->ld_ops->ldo_process_config(env, next,
4902                                                                       cfg);
4903                 }
4904
4905                 if (old_cfg != NULL)
4906                         lustre_cfg_free(cfg);
4907
4908                 break;
4909         }
4910         default:
4911                 /* others are passed further */
4912                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4913                 break;
4914         }
4915         RETURN(rc);
4916 }
4917
4918 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4919                                           const struct lu_object_header *hdr,
4920                                           struct lu_device *d)
4921 {
4922         struct mdt_object *mo;
4923
4924         ENTRY;
4925
4926         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4927         if (mo != NULL) {
4928                 struct lu_object *o;
4929                 struct lu_object_header *h;
4930
4931                 o = &mo->mot_obj;
4932                 h = &mo->mot_header;
4933                 lu_object_header_init(h);
4934                 lu_object_init(o, h, d);
4935                 lu_object_add_top(h, o);
4936                 o->lo_ops = &mdt_obj_ops;
4937                 spin_lock_init(&mo->mot_write_lock);
4938                 mutex_init(&mo->mot_lov_mutex);
4939                 init_rwsem(&mo->mot_open_sem);
4940                 atomic_set(&mo->mot_open_count, 0);
4941                 RETURN(o);
4942         }
4943         RETURN(NULL);
4944 }
4945
4946 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4947                            const struct lu_object_conf *unused)
4948 {
4949         struct mdt_device *d = mdt_dev(o->lo_dev);
4950         struct lu_device  *under;
4951         struct lu_object  *below;
4952         int                rc = 0;
4953         ENTRY;
4954
4955         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4956                PFID(lu_object_fid(o)));
4957
4958         under = &d->mdt_child->md_lu_dev;
4959         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4960         if (below != NULL) {
4961                 lu_object_add(o, below);
4962         } else
4963                 rc = -ENOMEM;
4964
4965         RETURN(rc);
4966 }
4967
4968 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4969 {
4970         struct mdt_object *mo = mdt_obj(o);
4971         struct lu_object_header *h;
4972         ENTRY;
4973
4974         h = o->lo_header;
4975         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4976                PFID(lu_object_fid(o)));
4977
4978         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4979         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4980
4981         lu_object_fini(o);
4982         lu_object_header_fini(h);
4983         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4984
4985         EXIT;
4986 }
4987
4988 static int mdt_object_print(const struct lu_env *env, void *cookie,
4989                             lu_printer_t p, const struct lu_object *o)
4990 {
4991         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4992
4993         return (*p)(env, cookie,
4994                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
4995                     mdto, mdto->mot_lov_created ? "lov_created" : "",
4996                     mdto->mot_cache_attr ? "cache_attr" : "",
4997                     mdto->mot_write_count);
4998 }
4999
5000 static int mdt_prepare(const struct lu_env *env,
5001                 struct lu_device *pdev,
5002                 struct lu_device *cdev)
5003 {
5004         struct mdt_device *mdt = mdt_dev(cdev);
5005         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5006         struct obd_device *obd = cdev->ld_obd;
5007         int rc;
5008
5009         ENTRY;
5010
5011         LASSERT(obd);
5012
5013         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5014         if (rc)
5015                 RETURN(rc);
5016
5017         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5018         if (rc)
5019                 RETURN(rc);
5020
5021         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5022         if (rc)
5023                 RETURN(rc);
5024
5025         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5026         /* The LFSCK instance is registered just now, so it must be there when
5027          * register the namespace to such instance. */
5028         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5029
5030         if (mdt->mdt_seq_site.ss_node_id == 0) {
5031                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5032                                                          &mdt->mdt_md_root_fid);
5033                 if (rc)
5034                         RETURN(rc);
5035         }
5036
5037         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5038
5039         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5040         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5041         LASSERT(obd->obd_no_conn);
5042         spin_lock(&obd->obd_dev_lock);
5043         obd->obd_no_conn = 0;
5044         spin_unlock(&obd->obd_dev_lock);
5045
5046         if (obd->obd_recovering == 0)
5047                 mdt_postrecov(env, mdt);
5048
5049         RETURN(rc);
5050 }
5051
5052 const struct lu_device_operations mdt_lu_ops = {
5053         .ldo_object_alloc   = mdt_object_alloc,
5054         .ldo_process_config = mdt_process_config,
5055         .ldo_prepare        = mdt_prepare,
5056 };
5057
5058 static const struct lu_object_operations mdt_obj_ops = {
5059         .loo_object_init    = mdt_object_init,
5060         .loo_object_free    = mdt_object_free,
5061         .loo_object_print   = mdt_object_print
5062 };
5063
5064 static int mdt_obd_set_info_async(const struct lu_env *env,
5065                                   struct obd_export *exp,
5066                                   __u32 keylen, void *key,
5067                                   __u32 vallen, void *val,
5068                                   struct ptlrpc_request_set *set)
5069 {
5070         int rc;
5071
5072         ENTRY;
5073
5074         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5075                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5076                 RETURN(rc);
5077         }
5078
5079         RETURN(0);
5080 }
5081
5082 /**
5083  * Match client and server connection feature flags.
5084  *
5085  * Compute the compatibility flags for a connection request based on
5086  * features mutually supported by client and server.
5087  *
5088  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5089  * must not be updated here, otherwise a partially initialized value may
5090  * be exposed. After the connection request is successfully processed,
5091  * the top-level MDT connect request handler atomically updates the export
5092  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5093  * reply. \see mdt_connect().
5094  *
5095  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5096  * full struct obd_connect_data. So care must be taken when accessing fields
5097  * that are not present in struct obd_connect_data_v1. See LU-16.
5098  *
5099  * \param exp   the obd_export associated with this client/target pair
5100  * \param mdt   the target device for the connection
5101  * \param data  stores data for this connect request
5102  *
5103  * \retval 0       success
5104  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5105  * \retval -EBADE  client and server feature requirements are incompatible
5106  */
5107 static int mdt_connect_internal(struct obd_export *exp,
5108                                 struct mdt_device *mdt,
5109                                 struct obd_connect_data *data)
5110 {
5111         LASSERT(data != NULL);
5112
5113         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5114
5115         if (mdt->mdt_bottom->dd_rdonly &&
5116             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5117             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5118                 RETURN(-EACCES);
5119
5120         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5121                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5122
5123         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5124
5125         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5126             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5127                 CWARN("%s: client %s does not support ibits lock, either "
5128                       "very old or an invalid client: flags %#llx\n",
5129                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5130                       data->ocd_connect_flags);
5131                 return -EBADE;
5132         }
5133
5134         if (!mdt->mdt_opts.mo_acl)
5135                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5136
5137         if (!mdt->mdt_opts.mo_user_xattr)
5138                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5139
5140         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5141                 data->ocd_brw_size = min(data->ocd_brw_size,
5142                                          (__u32)MD_MAX_BRW_SIZE);
5143                 if (data->ocd_brw_size == 0) {
5144                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
5145                                " ocd_version: %x ocd_grant: %d "
5146                                "ocd_index: %u ocd_brw_size is "
5147                                "unexpectedly zero, network data "
5148                                "corruption? Refusing connection of this"
5149                                " client\n",
5150                                mdt_obd_name(mdt),
5151                                exp->exp_client_uuid.uuid,
5152                                exp, data->ocd_connect_flags, data->ocd_version,
5153                                data->ocd_grant, data->ocd_index);
5154                         return -EPROTO;
5155                 }
5156         }
5157
5158         /* NB: Disregard the rule against updating
5159          * exp_connect_data.ocd_connect_flags in this case, since
5160          * tgt_client_new() needs to know if this is a lightweight
5161          * connection, and it is safe to expose this flag before
5162          * connection processing completes. */
5163         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5164                 spin_lock(&exp->exp_lock);
5165                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5166                 spin_unlock(&exp->exp_lock);
5167         }
5168
5169         data->ocd_version = LUSTRE_VERSION_CODE;
5170
5171         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5172                 CWARN("%s: MDS requires FID support, but client not\n",
5173                       mdt_obd_name(mdt));
5174                 return -EBADE;
5175         }
5176
5177         if (OCD_HAS_FLAG(data, PINGLESS)) {
5178                 if (ptlrpc_pinger_suppress_pings()) {
5179                         spin_lock(&exp->exp_obd->obd_dev_lock);
5180                         list_del_init(&exp->exp_obd_chain_timed);
5181                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5182                 } else {
5183                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5184                 }
5185         }
5186
5187         data->ocd_max_easize = mdt->mdt_max_ea_size;
5188
5189         /* NB: Disregard the rule against updating
5190          * exp_connect_data.ocd_connect_flags in this case, since
5191          * tgt_client_new() needs to know if this is client supports
5192          * multiple modify RPCs, and it is safe to expose this flag before
5193          * connection processing completes. */
5194         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5195                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5196                 spin_lock(&exp->exp_lock);
5197                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5198                 spin_unlock(&exp->exp_lock);
5199         }
5200
5201         return 0;
5202 }
5203
5204 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5205                                    struct mdt_thread_info *info,
5206                                    struct mdt_file_data *mfd)
5207 {
5208         struct lu_context ses;
5209         int rc;
5210         ENTRY;
5211
5212         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5213         if (rc)
5214                 RETURN(rc);
5215
5216         env->le_ses = &ses;
5217         lu_context_enter(&ses);
5218
5219         mdt_ucred(info)->uc_valid = UCRED_OLD;
5220         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5221
5222         lu_context_exit(&ses);
5223         lu_context_fini(&ses);
5224         env->le_ses = NULL;
5225
5226         RETURN(rc);
5227 }
5228
5229 static int mdt_export_cleanup(struct obd_export *exp)
5230 {
5231         struct list_head         closing_list;
5232         struct mdt_export_data  *med = &exp->exp_mdt_data;
5233         struct obd_device       *obd = exp->exp_obd;
5234         struct mdt_device       *mdt;
5235         struct mdt_thread_info  *info;
5236         struct lu_env            env;
5237         struct mdt_file_data    *mfd, *n;
5238         int rc = 0;
5239         ENTRY;
5240
5241         INIT_LIST_HEAD(&closing_list);
5242         spin_lock(&med->med_open_lock);
5243         while (!list_empty(&med->med_open_head)) {
5244                 struct list_head *tmp = med->med_open_head.next;
5245                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5246
5247                 /* Remove mfd handle so it can't be found again.
5248                  * We are consuming the mfd_list reference here. */
5249                 class_handle_unhash(&mfd->mfd_handle);
5250                 list_move_tail(&mfd->mfd_list, &closing_list);
5251         }
5252         spin_unlock(&med->med_open_lock);
5253         mdt = mdt_dev(obd->obd_lu_dev);
5254         LASSERT(mdt != NULL);
5255
5256         rc = lu_env_init(&env, LCT_MD_THREAD);
5257         if (rc)
5258                 RETURN(rc);
5259
5260         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5261         LASSERT(info != NULL);
5262         memset(info, 0, sizeof *info);
5263         info->mti_env = &env;
5264         info->mti_mdt = mdt;
5265         info->mti_exp = exp;
5266
5267         if (!list_empty(&closing_list)) {
5268                 struct md_attr *ma = &info->mti_attr;
5269
5270                 /* Close any open files (which may also cause orphan
5271                  * unlinking). */
5272                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5273                         list_del_init(&mfd->mfd_list);
5274                         ma->ma_need = ma->ma_valid = 0;
5275
5276                         /* This file is being closed due to an eviction, it
5277                          * could have been modified and now dirty regarding to
5278                          * HSM archive, check this!
5279                          * The logic here is to mark a file dirty if there's a
5280                          * chance it was dirtied before the client was evicted,
5281                          * so that we don't have to wait for a release attempt
5282                          * before finding out the file was actually dirty and
5283                          * fail the release. Aggressively marking it dirty here
5284                          * will cause the policy engine to attempt to
5285                          * re-archive it; when rearchiving, we can compare the
5286                          * current version to the HSM data_version and make the
5287                          * archive request into a noop if it's not actually
5288                          * dirty.
5289                          */
5290                         if (mfd->mfd_mode & FMODE_WRITE)
5291                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5292
5293                         /* Don't unlink orphan on failover umount, LU-184 */
5294                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5295                                 ma->ma_valid = MA_FLAGS;
5296                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5297                         }
5298                         mdt_mfd_close(info, mfd);
5299                 }
5300         }
5301         info->mti_mdt = NULL;
5302         /* cleanup client slot early */
5303         /* Do not erase record for recoverable client. */
5304         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5305                 tgt_client_del(&env, exp);
5306         lu_env_fini(&env);
5307
5308         RETURN(rc);
5309 }
5310
5311 static inline void mdt_enable_slc(struct mdt_device *mdt)
5312 {
5313         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5314                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5315 }
5316
5317 static inline void mdt_disable_slc(struct mdt_device *mdt)
5318 {
5319         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5320                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5321 }
5322
5323 static int mdt_obd_disconnect(struct obd_export *exp)
5324 {
5325         int rc;
5326         ENTRY;
5327
5328         LASSERT(exp);
5329         class_export_get(exp);
5330
5331         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5332             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5333                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5334
5335                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5336                         mdt_disable_slc(mdt);
5337         }
5338
5339         rc = server_disconnect_export(exp);
5340         if (rc != 0)
5341                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5342
5343         rc = mdt_export_cleanup(exp);
5344         nodemap_del_member(exp);
5345         class_export_put(exp);
5346         RETURN(rc);
5347 }
5348
5349 /* mds_connect copy */
5350 static int mdt_obd_connect(const struct lu_env *env,
5351                            struct obd_export **exp, struct obd_device *obd,
5352                            struct obd_uuid *cluuid,
5353                            struct obd_connect_data *data,
5354                            void *localdata)
5355 {
5356         struct obd_export       *lexp;
5357         struct lustre_handle    conn = { 0 };
5358         struct mdt_device       *mdt;
5359         int                      rc;
5360         lnet_nid_t              *client_nid = localdata;
5361         ENTRY;
5362
5363         LASSERT(env != NULL);
5364         LASSERT(data != NULL);
5365
5366         if (!exp || !obd || !cluuid)
5367                 RETURN(-EINVAL);
5368
5369         mdt = mdt_dev(obd->obd_lu_dev);
5370
5371         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5372             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5373                 atomic_inc(&mdt->mdt_mds_mds_conns);
5374                 mdt_enable_slc(mdt);
5375         }
5376
5377         /*
5378          * first, check whether the stack is ready to handle requests
5379          * XXX: probably not very appropriate method is used now
5380          *      at some point we should find a better one
5381          */
5382         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5383             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5384             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5385                 rc = obd_get_info(env, mdt->mdt_child_exp,
5386                                   sizeof(KEY_OSP_CONNECTED),
5387                                   KEY_OSP_CONNECTED, NULL, NULL);
5388                 if (rc)
5389                         RETURN(-EAGAIN);
5390                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5391         }
5392
5393         rc = class_connect(&conn, obd, cluuid);
5394         if (rc)
5395                 RETURN(rc);
5396
5397         lexp = class_conn2export(&conn);
5398         LASSERT(lexp != NULL);
5399
5400         rc = nodemap_add_member(*client_nid, lexp);
5401         if (rc != 0 && rc != -EEXIST)
5402                 GOTO(out, rc);
5403
5404         rc = mdt_connect_internal(lexp, mdt, data);
5405         if (rc == 0) {
5406                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5407
5408                 LASSERT(lcd);
5409                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5410                 rc = tgt_client_new(env, lexp);
5411                 if (rc == 0)
5412                         mdt_export_stats_init(obd, lexp, localdata);
5413         }
5414 out:
5415         if (rc != 0) {
5416                 class_disconnect(lexp);
5417                 nodemap_del_member(lexp);
5418                 *exp = NULL;
5419         } else {
5420                 *exp = lexp;
5421                 /* Because we do not want this export to be evicted by pinger,
5422                  * let's not add this export to the timed chain list. */
5423                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5424                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5425                         list_del_init(&lexp->exp_obd_chain_timed);
5426                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5427                 }
5428         }
5429
5430         RETURN(rc);
5431 }
5432
5433 static int mdt_obd_reconnect(const struct lu_env *env,
5434                              struct obd_export *exp, struct obd_device *obd,
5435                              struct obd_uuid *cluuid,
5436                              struct obd_connect_data *data,
5437                              void *localdata)
5438 {
5439         lnet_nid_t             *client_nid = localdata;
5440         int                     rc;
5441         ENTRY;
5442
5443         if (exp == NULL || obd == NULL || cluuid == NULL)
5444                 RETURN(-EINVAL);
5445
5446         rc = nodemap_add_member(*client_nid, exp);
5447         if (rc != 0 && rc != -EEXIST)
5448                 RETURN(rc);
5449
5450         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5451         if (rc == 0)
5452                 mdt_export_stats_init(obd, exp, localdata);
5453         else
5454                 nodemap_del_member(exp);
5455
5456         RETURN(rc);
5457 }
5458
5459 /* FIXME: Can we avoid using these two interfaces? */
5460 static int mdt_init_export(struct obd_export *exp)
5461 {
5462         struct mdt_export_data *med = &exp->exp_mdt_data;
5463         int                     rc;
5464         ENTRY;
5465
5466         INIT_LIST_HEAD(&med->med_open_head);
5467         spin_lock_init(&med->med_open_lock);
5468         spin_lock(&exp->exp_lock);
5469         exp->exp_connecting = 1;
5470         spin_unlock(&exp->exp_lock);
5471
5472         /* self-export doesn't need client data and ldlm initialization */
5473         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5474                                      &exp->exp_client_uuid)))
5475                 RETURN(0);
5476
5477         rc = tgt_client_alloc(exp);
5478         if (rc)
5479                 GOTO(err, rc);
5480
5481         rc = ldlm_init_export(exp);
5482         if (rc)
5483                 GOTO(err_free, rc);
5484
5485         RETURN(rc);
5486
5487 err_free:
5488         tgt_client_free(exp);
5489 err:
5490         CERROR("%s: Failed to initialize export: rc = %d\n",
5491                exp->exp_obd->obd_name, rc);
5492         return rc;
5493 }
5494
5495 static int mdt_destroy_export(struct obd_export *exp)
5496 {
5497         ENTRY;
5498
5499         target_destroy_export(exp);
5500         /* destroy can be called from failed obd_setup, so
5501          * checking uuid is safer than obd_self_export */
5502         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5503                                      &exp->exp_client_uuid)))
5504                 RETURN(0);
5505
5506         ldlm_destroy_export(exp);
5507         tgt_client_free(exp);
5508
5509         LASSERT(list_empty(&exp->exp_outstanding_replies));
5510         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5511
5512         RETURN(0);
5513 }
5514
5515 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5516                    struct linkea_data *ldata)
5517 {
5518         int rc;
5519
5520         LASSERT(ldata->ld_buf->lb_buf != NULL);
5521
5522         if (!mdt_object_exists(mdt_obj))
5523                 return -ENODATA;
5524
5525         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5526                           ldata->ld_buf, XATTR_NAME_LINK);
5527         if (rc == -ERANGE) {
5528                 /* Buf was too small, figure out what we need. */
5529                 lu_buf_free(ldata->ld_buf);
5530                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5531                                   ldata->ld_buf, XATTR_NAME_LINK);
5532                 if (rc < 0)
5533                         return rc;
5534                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5535                 if (ldata->ld_buf->lb_buf == NULL)
5536                         return -ENOMEM;
5537                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5538                                   ldata->ld_buf, XATTR_NAME_LINK);
5539         }
5540         if (rc < 0)
5541                 return rc;
5542
5543         return linkea_init_with_rec(ldata);
5544 }
5545
5546 /**
5547  * Given an MDT object, try to look up the full path to the object.
5548  * Part of the MDT layer implementation of lfs fid2path.
5549  *
5550  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5551  * \param[in]     obj   Object to do path lookup of
5552  * \param[in,out] fp    User-provided struct to store path information
5553  * \param[in]     root_fid Root FID of current path should reach
5554  *
5555  * \retval 0 Lookup successful, path information stored in fp
5556  * \retval -EAGAIN Lookup failed, usually because object is being moved
5557  * \retval negative errno if there was a problem
5558  */
5559 static int mdt_path_current(struct mdt_thread_info *info,
5560                             struct mdt_object *obj,
5561                             struct getinfo_fid2path *fp,
5562                             struct lu_fid *root_fid)
5563 {
5564         struct mdt_device       *mdt = info->mti_mdt;
5565         struct mdt_object       *mdt_obj;
5566         struct link_ea_header   *leh;
5567         struct link_ea_entry    *lee;
5568         struct lu_name          *tmpname = &info->mti_name;
5569         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5570         struct lu_buf           *buf = &info->mti_big_buf;
5571         char                    *ptr;
5572         int                     reclen;
5573         struct linkea_data      ldata = { NULL };
5574         int                     rc = 0;
5575         bool                    first = true;
5576         ENTRY;
5577
5578         /* temp buffer for path element, the buffer will be finally freed
5579          * in mdt_thread_info_fini */
5580         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5581         if (buf->lb_buf == NULL)
5582                 RETURN(-ENOMEM);
5583
5584         ldata.ld_buf = buf;
5585         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5586         *ptr = 0;
5587         --ptr;
5588         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5589
5590         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5591                 struct lu_buf           lmv_buf;
5592
5593                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5594                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5595                         GOTO(out, rc = -ENOENT);
5596
5597                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5598                         mdt_obj = obj;
5599                         mdt_object_get(info->mti_env, mdt_obj);
5600                 } else {
5601                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5602                         if (IS_ERR(mdt_obj))
5603                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5604                 }
5605
5606                 if (!mdt_object_exists(mdt_obj)) {
5607                         mdt_object_put(info->mti_env, mdt_obj);
5608                         GOTO(out, rc = -ENOENT);
5609                 }
5610
5611                 if (mdt_object_remote(mdt_obj)) {
5612                         mdt_object_put(info->mti_env, mdt_obj);
5613                         GOTO(remote_out, rc = -EREMOTE);
5614                 }
5615
5616                 rc = mdt_links_read(info, mdt_obj, &ldata);
5617                 if (rc != 0) {
5618                         mdt_object_put(info->mti_env, mdt_obj);
5619                         GOTO(out, rc);
5620                 }
5621
5622                 leh = buf->lb_buf;
5623                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5624                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5625                 /* If set, use link #linkno for path lookup, otherwise use
5626                    link #0.  Only do this for the final path element. */
5627                 if (first && fp->gf_linkno < leh->leh_reccount) {
5628                         int count;
5629                         for (count = 0; count < fp->gf_linkno; count++) {
5630                                 lee = (struct link_ea_entry *)
5631                                      ((char *)lee + reclen);
5632                                 linkea_entry_unpack(lee, &reclen, tmpname,
5633                                                     tmpfid);
5634                         }
5635                         if (fp->gf_linkno < leh->leh_reccount - 1)
5636                                 /* indicate to user there are more links */
5637                                 fp->gf_linkno++;
5638                 }
5639
5640                 lmv_buf.lb_buf = info->mti_xattr_buf;
5641                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5642                 /* Check if it is slave stripes */
5643                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5644                                   &lmv_buf, XATTR_NAME_LMV);
5645                 mdt_object_put(info->mti_env, mdt_obj);
5646                 if (rc > 0) {
5647                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5648
5649                         /* For slave stripes, get its master */
5650                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5651                                 fp->gf_fid = *tmpfid;
5652                                 continue;
5653                         }
5654                 } else if (rc < 0 && rc != -ENODATA) {
5655                         GOTO(out, rc);
5656                 }
5657
5658                 rc = 0;
5659
5660                 /* Pack the name in the end of the buffer */
5661                 ptr -= tmpname->ln_namelen;
5662                 if (ptr - 1 <= fp->gf_u.gf_path)
5663                         GOTO(out, rc = -EOVERFLOW);
5664                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5665                 *(--ptr) = '/';
5666
5667                 /* keep the last resolved fid to the client, so the
5668                  * client will build the left path on another MDT for
5669                  * remote object */
5670                 fp->gf_fid = *tmpfid;
5671
5672                 first = false;
5673         }
5674
5675 remote_out:
5676         ptr++; /* skip leading / */
5677         memmove(fp->gf_u.gf_path, ptr,
5678                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5679
5680 out:
5681         RETURN(rc);
5682 }
5683
5684 /**
5685  * Given an MDT object, use mdt_path_current to get the path.
5686  * Essentially a wrapper to retry mdt_path_current a set number of times
5687  * if -EAGAIN is returned (usually because an object is being moved).
5688  *
5689  * Part of the MDT layer implementation of lfs fid2path.
5690  *
5691  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5692  * \param[in]     obj   Object to do path lookup of
5693  * \param[in,out] fp    User-provided struct for arguments and to store path
5694  *                      information
5695  *
5696  * \retval 0 Lookup successful, path information stored in fp
5697  * \retval negative errno if there was a problem
5698  */
5699 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5700                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5701 {
5702         struct mdt_device       *mdt = info->mti_mdt;
5703         int                     tries = 3;
5704         int                     rc = -EAGAIN;
5705         ENTRY;
5706
5707         if (fp->gf_pathlen < 3)
5708                 RETURN(-EOVERFLOW);
5709
5710         if (root_fid == NULL)
5711                 root_fid = &mdt->mdt_md_root_fid;
5712
5713         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5714                 fp->gf_u.gf_path[0] = '\0';
5715                 RETURN(0);
5716         }
5717
5718         /* Retry multiple times in case file is being moved */
5719         while (tries-- && rc == -EAGAIN)
5720                 rc = mdt_path_current(info, obj, fp, root_fid);
5721
5722         RETURN(rc);
5723 }
5724
5725 /**
5726  * Get the full path of the provided FID, as of changelog record recno.
5727  *
5728  * This checks sanity and looks up object for user provided FID
5729  * before calling the actual path lookup code.
5730  *
5731  * Part of the MDT layer implementation of lfs fid2path.
5732  *
5733  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5734  * \param[in,out] fp    User-provided struct for arguments and to store path
5735  *                      information
5736  *
5737  * \retval 0 Lookup successful, path information and recno stored in fp
5738  * \retval -ENOENT, object does not exist
5739  * \retval negative errno if there was a problem
5740  */
5741 static int mdt_fid2path(struct mdt_thread_info *info,
5742                         struct lu_fid *root_fid,
5743                         struct getinfo_fid2path *fp)
5744 {
5745         struct mdt_device *mdt = info->mti_mdt;
5746         struct mdt_object *obj;
5747         int    rc;
5748         ENTRY;
5749
5750         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
5751                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5752
5753         if (!fid_is_sane(&fp->gf_fid))
5754                 RETURN(-EINVAL);
5755
5756         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5757                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
5758                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
5759                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5760                 RETURN(-EINVAL);
5761         }
5762
5763         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5764         if (IS_ERR(obj)) {
5765                 rc = PTR_ERR(obj);
5766                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5767                        PFID(&fp->gf_fid), rc);
5768                 RETURN(rc);
5769         }
5770
5771         if (mdt_object_remote(obj))
5772                 rc = -EREMOTE;
5773         else if (!mdt_object_exists(obj))
5774                 rc = -ENOENT;
5775         else
5776                 rc = 0;
5777
5778         if (rc < 0) {
5779                 mdt_object_put(info->mti_env, obj);
5780                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5781                        PFID(&fp->gf_fid), rc);
5782                 RETURN(rc);
5783         }
5784
5785         rc = mdt_path(info, obj, fp, root_fid);
5786
5787         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
5788                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5789                fp->gf_recno, fp->gf_linkno);
5790
5791         mdt_object_put(info->mti_env, obj);
5792
5793         RETURN(rc);
5794 }
5795
5796 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5797                             void *val, int vallen)
5798 {
5799         struct getinfo_fid2path *fpout, *fpin;
5800         struct lu_fid *root_fid = NULL;
5801         int rc = 0;
5802
5803         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5804         fpout = val;
5805
5806         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5807                 lustre_swab_fid2path(fpin);
5808
5809         memcpy(fpout, fpin, sizeof(*fpin));
5810         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5811                 RETURN(-EINVAL);
5812
5813         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5814                       sizeof(struct lu_fid)) {
5815                 /* client sent its root FID, which is normally fileset FID */
5816                 root_fid = fpin->gf_u.gf_root_fid;
5817                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5818                         lustre_swab_lu_fid(root_fid);
5819
5820                 if (root_fid != NULL && !fid_is_sane(root_fid))
5821                         RETURN(-EINVAL);
5822         }
5823
5824         rc = mdt_fid2path(info, root_fid, fpout);
5825         RETURN(rc);
5826 }
5827
5828 int mdt_get_info(struct tgt_session_info *tsi)
5829 {
5830         char    *key;
5831         int      keylen;
5832         __u32   *vallen;
5833         void    *valout;
5834         int      rc;
5835
5836         ENTRY;
5837
5838         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5839         if (key == NULL) {
5840                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5841                 RETURN(err_serious(-EFAULT));
5842         }
5843         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5844                                       RCL_CLIENT);
5845
5846         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5847         if (vallen == NULL) {
5848                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5849                                 tgt_name(tsi->tsi_tgt));
5850                 RETURN(err_serious(-EFAULT));
5851         }
5852
5853         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5854                              *vallen);
5855         rc = req_capsule_server_pack(tsi->tsi_pill);
5856         if (rc)
5857                 RETURN(err_serious(rc));
5858
5859         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5860         if (valout == NULL) {
5861                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5862                                 tgt_name(tsi->tsi_tgt));
5863                 RETURN(err_serious(-EFAULT));
5864         }
5865
5866         if (KEY_IS(KEY_FID2PATH)) {
5867                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5868
5869                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5870                 mdt_thread_info_fini(info);
5871         } else {
5872                 rc = -EINVAL;
5873         }
5874         RETURN(rc);
5875 }
5876
5877 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5878 {
5879         struct obd_ioctl_data *data = karg;
5880         struct lu_fid *fid;
5881         __u64 version;
5882         struct mdt_object *obj;
5883         struct mdt_lock_handle  *lh;
5884         int rc;
5885         ENTRY;
5886
5887         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5888             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5889                 RETURN(-EINVAL);
5890
5891         fid = (struct lu_fid *)data->ioc_inlbuf1;
5892
5893         if (!fid_is_sane(fid))
5894                 RETURN(-EINVAL);
5895
5896         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5897
5898         lh = &mti->mti_lh[MDT_LH_PARENT];
5899         mdt_lock_reg_init(lh, LCK_CR);
5900
5901         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5902         if (IS_ERR(obj))
5903                 RETURN(PTR_ERR(obj));
5904
5905         if (mdt_object_remote(obj)) {
5906                 rc = -EREMOTE;
5907                 /**
5908                  * before calling version get the correct MDS should be
5909                  * fid, this is error to find remote object here
5910                  */
5911                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5912         } else if (!mdt_object_exists(obj)) {
5913                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5914                 rc = -ENOENT;
5915         } else {
5916                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5917                *(__u64 *)data->ioc_inlbuf2 = version;
5918                 rc = 0;
5919         }
5920         mdt_object_unlock_put(mti, obj, lh, 1);
5921         RETURN(rc);
5922 }
5923
5924 /* ioctls on obd dev */
5925 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5926                          void *karg, void __user *uarg)
5927 {
5928         struct lu_env      env;
5929         struct obd_device *obd = exp->exp_obd;
5930         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5931         struct dt_device  *dt = mdt->mdt_bottom;
5932         int rc;
5933
5934         ENTRY;
5935         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5936         rc = lu_env_init(&env, LCT_MD_THREAD);
5937         if (rc)
5938                 RETURN(rc);
5939
5940         switch (cmd) {
5941         case OBD_IOC_SYNC:
5942                 rc = mdt_device_sync(&env, mdt);
5943                 break;
5944         case OBD_IOC_SET_READONLY:
5945                 rc = dt_sync(&env, dt);
5946                 if (rc == 0)
5947                         rc = dt_ro(&env, dt);
5948                 break;
5949         case OBD_IOC_ABORT_RECOVERY:
5950                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5951                 obd->obd_abort_recovery = 1;
5952                 target_stop_recovery_thread(obd);
5953                 rc = 0;
5954                 break;
5955         case OBD_IOC_CHANGELOG_REG:
5956         case OBD_IOC_CHANGELOG_DEREG:
5957         case OBD_IOC_CHANGELOG_CLEAR:
5958                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
5959                                                            mdt->mdt_child,
5960                                                            cmd, len, karg);
5961                 break;
5962         case OBD_IOC_START_LFSCK: {
5963                 struct md_device *next = mdt->mdt_child;
5964                 struct obd_ioctl_data *data = karg;
5965                 struct lfsck_start_param lsp;
5966
5967                 if (unlikely(data == NULL)) {
5968                         rc = -EINVAL;
5969                         break;
5970                 }
5971
5972                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5973                 lsp.lsp_index_valid = 0;
5974                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5975                 break;
5976         }
5977         case OBD_IOC_STOP_LFSCK: {
5978                 struct md_device        *next = mdt->mdt_child;
5979                 struct obd_ioctl_data   *data = karg;
5980                 struct lfsck_stop        stop;
5981
5982                 stop.ls_status = LS_STOPPED;
5983                 /* Old lfsck utils may pass NULL @stop. */
5984                 if (data->ioc_inlbuf1 == NULL)
5985                         stop.ls_flags = 0;
5986                 else
5987                         stop.ls_flags =
5988                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5989
5990                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5991                 break;
5992         }
5993         case OBD_IOC_QUERY_LFSCK: {
5994                 struct md_device        *next = mdt->mdt_child;
5995                 struct obd_ioctl_data   *data = karg;
5996
5997                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5998                                                  data->ioc_inlbuf1);
5999                 break;
6000         }
6001         case OBD_IOC_GET_OBJ_VERSION: {
6002                 struct mdt_thread_info *mti;
6003                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6004                 memset(mti, 0, sizeof *mti);
6005                 mti->mti_env = &env;
6006                 mti->mti_mdt = mdt;
6007                 mti->mti_exp = exp;
6008
6009                 rc = mdt_ioc_version_get(mti, karg);
6010                 break;
6011         }
6012         case OBD_IOC_CATLOGLIST: {
6013                 struct mdt_thread_info *mti;
6014
6015                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6016                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6017                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6018                                        &mti->mti_tmp_fid1);
6019                 break;
6020          }
6021         default:
6022                 rc = -EOPNOTSUPP;
6023                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6024                         mdt_obd_name(mdt), cmd, rc);
6025         }
6026
6027         lu_env_fini(&env);
6028         RETURN(rc);
6029 }
6030
6031 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6032 {
6033         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6034         int rc;
6035         ENTRY;
6036
6037         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6038                 struct lfsck_start_param lsp;
6039
6040                 lsp.lsp_start = NULL;
6041                 lsp.lsp_index_valid = 0;
6042                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6043                                                            OBD_IOC_START_LFSCK,
6044                                                            0, &lsp);
6045                 if (rc != 0 && rc != -EALREADY)
6046                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6047                               mdt_obd_name(mdt), rc);
6048         }
6049
6050         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6051         RETURN(rc);
6052 }
6053
6054 static int mdt_obd_postrecov(struct obd_device *obd)
6055 {
6056         struct lu_env env;
6057         int rc;
6058
6059         rc = lu_env_init(&env, LCT_MD_THREAD);
6060         if (rc)
6061                 RETURN(rc);
6062         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6063         lu_env_fini(&env);
6064         return rc;
6065 }
6066
6067 static struct obd_ops mdt_obd_device_ops = {
6068         .o_owner          = THIS_MODULE,
6069         .o_set_info_async = mdt_obd_set_info_async,
6070         .o_connect        = mdt_obd_connect,
6071         .o_reconnect      = mdt_obd_reconnect,
6072         .o_disconnect     = mdt_obd_disconnect,
6073         .o_init_export    = mdt_init_export,
6074         .o_destroy_export = mdt_destroy_export,
6075         .o_iocontrol      = mdt_iocontrol,
6076         .o_postrecov      = mdt_obd_postrecov,
6077 };
6078
6079 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6080                                          struct lu_device *d)
6081 {
6082         struct mdt_device *m = mdt_dev(d);
6083         ENTRY;
6084
6085         mdt_fini(env, m);
6086         RETURN(NULL);
6087 }
6088
6089 static struct lu_device *mdt_device_free(const struct lu_env *env,
6090                                          struct lu_device *d)
6091 {
6092         struct mdt_device *m = mdt_dev(d);
6093         ENTRY;
6094
6095         lu_device_fini(&m->mdt_lu_dev);
6096         OBD_FREE_PTR(m);
6097
6098         RETURN(NULL);
6099 }
6100
6101 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6102                                           struct lu_device_type *t,
6103                                           struct lustre_cfg *cfg)
6104 {
6105         struct lu_device  *l;
6106         struct mdt_device *m;
6107
6108         OBD_ALLOC_PTR(m);
6109         if (m != NULL) {
6110                 int rc;
6111
6112                 l = &m->mdt_lu_dev;
6113                 rc = mdt_init0(env, m, t, cfg);
6114                 if (rc != 0) {
6115                         mdt_device_free(env, l);
6116                         l = ERR_PTR(rc);
6117                         return l;
6118                 }
6119         } else
6120                 l = ERR_PTR(-ENOMEM);
6121         return l;
6122 }
6123
6124 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6125 LU_KEY_INIT(mdt, struct mdt_thread_info);
6126
6127 static void mdt_key_fini(const struct lu_context *ctx,
6128                          struct lu_context_key *key, void* data)
6129 {
6130         struct mdt_thread_info *info = data;
6131
6132         if (info->mti_big_lmm) {
6133                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6134                 info->mti_big_lmm = NULL;
6135                 info->mti_big_lmmsize = 0;
6136         }
6137         OBD_FREE_PTR(info);
6138 }
6139
6140 /* context key: mdt_thread_key */
6141 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6142
6143 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6144 {
6145         return lu_ucred(info->mti_env);
6146 }
6147
6148 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6149 {
6150         return lu_ucred_check(info->mti_env);
6151 }
6152
6153 /**
6154  * Enable/disable COS (Commit On Sharing).
6155  *
6156  * Set/Clear the COS flag in mdt options.
6157  *
6158  * \param mdt mdt device
6159  * \param val 0 disables COS, other values enable COS
6160  */
6161 void mdt_enable_cos(struct mdt_device *mdt, int val)
6162 {
6163         struct lu_env env;
6164         int rc;
6165
6166         mdt->mdt_opts.mo_cos = !!val;
6167         rc = lu_env_init(&env, LCT_LOCAL);
6168         if (unlikely(rc != 0)) {
6169                 CWARN("%s: lu_env initialization failed, cannot "
6170                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6171                 return;
6172         }
6173         mdt_device_sync(&env, mdt);
6174         lu_env_fini(&env);
6175 }
6176
6177 /**
6178  * Check COS (Commit On Sharing) status.
6179  *
6180  * Return COS flag status.
6181  *
6182  * \param mdt mdt device
6183  */
6184 int mdt_cos_is_enabled(struct mdt_device *mdt)
6185 {
6186         return mdt->mdt_opts.mo_cos != 0;
6187 }
6188
6189 static struct lu_device_type_operations mdt_device_type_ops = {
6190         .ldto_device_alloc = mdt_device_alloc,
6191         .ldto_device_free  = mdt_device_free,
6192         .ldto_device_fini  = mdt_device_fini
6193 };
6194
6195 static struct lu_device_type mdt_device_type = {
6196         .ldt_tags     = LU_DEVICE_MD,
6197         .ldt_name     = LUSTRE_MDT_NAME,
6198         .ldt_ops      = &mdt_device_type_ops,
6199         .ldt_ctx_tags = LCT_MD_THREAD
6200 };
6201
6202 static int __init mdt_init(void)
6203 {
6204         int rc;
6205
6206         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6207                  FID_NOBRACE_LEN + 1);
6208         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6209                  FID_LEN + 1);
6210         rc = lu_kmem_init(mdt_caches);
6211         if (rc)
6212                 return rc;
6213
6214         rc = mds_mod_init();
6215         if (rc)
6216                 GOTO(lu_fini, rc);
6217
6218         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6219                                  LUSTRE_MDT_NAME, &mdt_device_type);
6220         if (rc)
6221                 GOTO(mds_fini, rc);
6222 lu_fini:
6223         if (rc)
6224                 lu_kmem_fini(mdt_caches);
6225 mds_fini:
6226         if (rc)
6227                 mds_mod_exit();
6228         return rc;
6229 }
6230
6231 static void __exit mdt_exit(void)
6232 {
6233         class_unregister_type(LUSTRE_MDT_NAME);
6234         mds_mod_exit();
6235         lu_kmem_fini(mdt_caches);
6236 }
6237
6238 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6239 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6240 MODULE_VERSION(LUSTRE_VERSION_STRING);
6241 MODULE_LICENSE("GPL");
6242
6243 module_init(mdt_init);
6244 module_exit(mdt_exit);