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