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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52
  53 #include <dt_object.h>
  54 #include <lustre_acl.h>
  55 #include <lustre_export.h>
  56 #include <lustre_ioctl.h>
  57 #include <lustre_lfsck.h>
  58 #include <lustre_log.h>
  59 #include <lustre_net.h>
  60 #include <lustre_nodemap.h>
  61 #include <lustre_mds.h>
  62 #include <lustre_param.h>
  63 #include <lustre_quota.h>
  64 #include <lustre_swab.h>
  65 #include <obd.h>
  66 #include <obd_support.h>
  67
  68 #include <llog_swab.h>
  69
  70 #include "mdt_internal.h"
  71
  72
  73 static unsigned int max_mod_rpcs_per_client = 8;
  74 module_param(max_mod_rpcs_per_client, uint, 0644);
  75 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 enum ldlm_mode mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE]   = LCK_MINMODE,
  90         [MDL_EX]        = LCK_EX,
  91         [MDL_PW]        = LCK_PW,
  92         [MDL_PR]        = LCK_PR,
  93         [MDL_CW]        = LCK_CW,
  94         [MDL_CR]        = LCK_CR,
  95         [MDL_NL]        = LCK_NL,
  96         [MDL_GROUP]     = LCK_GROUP
  97 };
  98
  99 static struct mdt_device *mdt_dev(struct lu_device *d);
 100 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 101
 102 static const struct lu_object_operations mdt_obj_ops;
 103
 104 /* Slab for MDT object allocation */
 105 static struct kmem_cache *mdt_object_kmem;
 106
 107 /* For HSM restore handles */
 108 struct kmem_cache *mdt_hsm_cdt_kmem;
 109
 110 /* For HSM request handles */
 111 struct kmem_cache *mdt_hsm_car_kmem;
 112
 113 static struct lu_kmem_descr mdt_caches[] = {
 114         {
 115                 .ckd_cache = &mdt_object_kmem,
 116                 .ckd_name  = "mdt_obj",
 117                 .ckd_size  = sizeof(struct mdt_object)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 121                 .ckd_name       = "mdt_cdt_restore_handle",
 122                 .ckd_size       = sizeof(struct cdt_restore_handle)
 123         },
 124         {
 125                 .ckd_cache      = &mdt_hsm_car_kmem,
 126                 .ckd_name       = "mdt_cdt_agent_req",
 127                 .ckd_size       = sizeof(struct cdt_agent_req)
 128         },
 129         {
 130                 .ckd_cache = NULL
 131         }
 132 };
 133
 134 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 135 {
 136         if (!rep)
 137                 return 0;
 138         return rep->lock_policy_res1 & op_flag;
 139 }
 140
 141 void mdt_clear_disposition(struct mdt_thread_info *info,
 142                            struct ldlm_reply *rep, __u64 op_flag)
 143 {
 144         if (info) {
 145                 info->mti_opdata &= ~op_flag;
 146                 tgt_opdata_clear(info->mti_env, op_flag);
 147         }
 148         if (rep)
 149                 rep->lock_policy_res1 &= ~op_flag;
 150 }
 151
 152 void mdt_set_disposition(struct mdt_thread_info *info,
 153                          struct ldlm_reply *rep, __u64 op_flag)
 154 {
 155         if (info) {
 156                 info->mti_opdata |= op_flag;
 157                 tgt_opdata_set(info->mti_env, op_flag);
 158         }
 159         if (rep)
 160                 rep->lock_policy_res1 |= op_flag;
 161 }
 162
 163 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 164 {
 165         lh->mlh_pdo_hash = 0;
 166         lh->mlh_reg_mode = lm;
 167         lh->mlh_rreg_mode = lm;
 168         lh->mlh_type = MDT_REG_LOCK;
 169 }
 170
 171 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 172                        const struct lu_name *lname)
 173 {
 174         lh->mlh_reg_mode = lock_mode;
 175         lh->mlh_pdo_mode = LCK_MINMODE;
 176         lh->mlh_rreg_mode = lock_mode;
 177         lh->mlh_type = MDT_PDO_LOCK;
 178
 179         if (lu_name_is_valid(lname)) {
 180                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 181                                                   lname->ln_namelen);
 182                 /* XXX Workaround for LU-2856
 183                  *
 184                  * Zero is a valid return value of full_name_hash, but
 185                  * several users of mlh_pdo_hash assume a non-zero
 186                  * hash value. We therefore map zero onto an
 187                  * arbitrary, but consistent value (1) to avoid
 188                  * problems further down the road. */
 189                 if (unlikely(lh->mlh_pdo_hash == 0))
 190                         lh->mlh_pdo_hash = 1;
 191         } else {
 192                 lh->mlh_pdo_hash = 0;
 193         }
 194 }
 195
 196 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 197                               struct mdt_lock_handle *lh)
 198 {
 199         mdl_mode_t mode;
 200         ENTRY;
 201
 202         /*
 203          * Any dir access needs couple of locks:
 204          *
 205          * 1) on part of dir we gonna take lookup/modify;
 206          *
 207          * 2) on whole dir to protect it from concurrent splitting and/or to
 208          * flush client's cache for readdir().
 209          *
 210          * so, for a given mode and object this routine decides what lock mode
 211          * to use for lock #2:
 212          *
 213          * 1) if caller's gonna lookup in dir then we need to protect dir from
 214          * being splitted only - LCK_CR
 215          *
 216          * 2) if caller's gonna modify dir then we need to protect dir from
 217          * being splitted and to flush cache - LCK_CW
 218          *
 219          * 3) if caller's gonna modify dir and that dir seems ready for
 220          * splitting then we need to protect it from any type of access
 221          * (lookup/modify/split) - LCK_EX --bzzz
 222          */
 223
 224         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 225         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 226
 227         /*
 228          * Ask underlaying level its opinion about preferable PDO lock mode
 229          * having access type passed as regular lock mode:
 230          *
 231          * - MDL_MINMODE means that lower layer does not want to specify lock
 232          * mode;
 233          *
 234          * - MDL_NL means that no PDO lock should be taken. This is used in some
 235          * cases. Say, for non-splittable directories no need to use PDO locks
 236          * at all.
 237          */
 238         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 239                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 240
 241         if (mode != MDL_MINMODE) {
 242                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 243         } else {
 244                 /*
 245                  * Lower layer does not want to specify locking mode. We do it
 246                  * our selves. No special protection is needed, just flush
 247                  * client's cache on modification and allow concurrent
 248                  * mondification.
 249                  */
 250                 switch (lh->mlh_reg_mode) {
 251                 case LCK_EX:
 252                         lh->mlh_pdo_mode = LCK_EX;
 253                         break;
 254                 case LCK_PR:
 255                         lh->mlh_pdo_mode = LCK_CR;
 256                         break;
 257                 case LCK_PW:
 258                         lh->mlh_pdo_mode = LCK_CW;
 259                         break;
 260                 default:
 261                         CERROR("Not expected lock type (0x%x)\n",
 262                                (int)lh->mlh_reg_mode);
 263                         LBUG();
 264                 }
 265         }
 266
 267         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 268         EXIT;
 269 }
 270
 271 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 272                               struct lu_fid *fid)
 273 {
 274         struct mdt_device *mdt = info->mti_mdt;
 275         struct lu_name *lname = &info->mti_name;
 276         char *name = NULL;
 277         struct mdt_object *parent;
 278         u32 mode;
 279         int rc = 0;
 280
 281         LASSERT(!info->mti_cross_ref);
 282
 283         OBD_ALLOC(name, NAME_MAX + 1);
 284         if (name == NULL)
 285                 return -ENOMEM;
 286         lname->ln_name = name;
 287
 288         /*
 289          * We may want to allow this to mount a completely separate
 290          * fileset from the MDT in the future, but keeping it to
 291          * ROOT/ only for now avoid potential security issues.
 292          */
 293         *fid = mdt->mdt_md_root_fid;
 294
 295         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 296                 const char *s1 = fileset;
 297                 const char *s2;
 298
 299                 while (*++s1 == '/')
 300                         ;
 301                 s2 = s1;
 302                 while (*s2 != '/' && *s2 != '\0')
 303                         s2++;
 304
 305                 if (s2 == s1)
 306                         break;
 307
 308                 fileset = s2;
 309
 310                 lname->ln_namelen = s2 - s1;
 311                 if (lname->ln_namelen > NAME_MAX) {
 312                         rc = -EINVAL;
 313                         break;
 314                 }
 315
 316                 /* reject .. as a path component */
 317                 if (lname->ln_namelen == 2 &&
 318                     strncmp(s1, "..", 2) == 0) {
 319                         rc = -EINVAL;
 320                         break;
 321                 }
 322
 323                 strncpy(name, s1, lname->ln_namelen);
 324                 name[lname->ln_namelen] = '\0';
 325
 326                 parent = mdt_object_find(info->mti_env, mdt, fid);
 327                 if (IS_ERR(parent)) {
 328                         rc = PTR_ERR(parent);
 329                         break;
 330                 }
 331                 /* Only got the fid of this obj by name */
 332                 fid_zero(fid);
 333                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 334                                 fid, &info->mti_spec);
 335                 mdt_object_put(info->mti_env, parent);
 336         }
 337         if (!rc) {
 338                 parent = mdt_object_find(info->mti_env, mdt, fid);
 339                 if (IS_ERR(parent))
 340                         rc = PTR_ERR(parent);
 341                 else {
 342                         mode = lu_object_attr(&parent->mot_obj);
 343                         mdt_object_put(info->mti_env, parent);
 344                         if (!S_ISDIR(mode))
 345                                 rc = -ENOTDIR;
 346                 }
 347         }
 348
 349         OBD_FREE(name, NAME_MAX + 1);
 350
 351         return rc;
 352 }
 353
 354 static int mdt_get_root(struct tgt_session_info *tsi)
 355 {
 356         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 357         struct mdt_device       *mdt = info->mti_mdt;
 358         struct mdt_body         *repbody;
 359         char                    *fileset;
 360         int                      rc;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376
 377                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 378                 if (rc < 0)
 379                         GOTO(out, rc = err_serious(rc));
 380         } else {
 381                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 382         }
 383         repbody->mbo_valid |= OBD_MD_FLID;
 384
 385         EXIT;
 386 out:
 387         mdt_thread_info_fini(info);
 388         return rc;
 389 }
 390
 391 static int mdt_statfs(struct tgt_session_info *tsi)
 392 {
 393         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 394         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 395         struct md_device                *next = info->mti_mdt->mdt_child;
 396         struct ptlrpc_service_part      *svcpt;
 397         struct obd_statfs               *osfs;
 398         int                             rc;
 399
 400         ENTRY;
 401
 402         svcpt = req->rq_rqbd->rqbd_svcpt;
 403
 404         /* This will trigger a watchdog timeout */
 405         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 406                          (MDT_SERVICE_WATCHDOG_FACTOR *
 407                           at_get(&svcpt->scp_at_estimate)) + 1);
 408
 409         rc = mdt_check_ucred(info);
 410         if (rc)
 411                 GOTO(out, rc = err_serious(rc));
 412
 413         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 414                 GOTO(out, rc = err_serious(-ENOMEM));
 415
 416         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 417         if (!osfs)
 418                 GOTO(out, rc = -EPROTO);
 419
 420         /** statfs information are cached in the mdt_device */
 421         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 422                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 423                 /** statfs data is too old, get up-to-date one */
 424                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 425                 if (rc)
 426                         GOTO(out, rc);
 427                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 428                 info->mti_mdt->mdt_osfs = *osfs;
 429                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 430                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 431         } else {
 432                 /** use cached statfs data */
 433                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 434                 *osfs = info->mti_mdt->mdt_osfs;
 435                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 436         }
 437
 438         if (rc == 0)
 439                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 440 out:
 441         mdt_thread_info_fini(info);
 442         RETURN(rc);
 443 }
 444
 445 #ifdef CONFIG_FS_POSIX_ACL
 446 /*
 447  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 448  *
 449  * \param       info    thread info object
 450  * \param       repbody reply to pack ACLs into
 451  * \param       o       mdt object of file to examine
 452  * \param       nodemap nodemap of client to reply to
 453  * \retval      0       success
 454  * \retval      -errno  error getting or parsing ACL from disk
 455  */
 456 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 457                       struct mdt_object *o, struct lu_nodemap *nodemap)
 458 {
 459         const struct lu_env     *env = info->mti_env;
 460         struct md_object        *next = mdt_object_child(o);
 461         struct lu_buf           *buf = &info->mti_buf;
 462         int rc;
 463
 464         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 465         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 466                                            RCL_SERVER);
 467         if (buf->lb_len == 0)
 468                 return 0;
 469
 470         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 471         if (rc < 0) {
 472                 if (rc == -ENODATA) {
 473                         repbody->mbo_aclsize = 0;
 474                         repbody->mbo_valid |= OBD_MD_FLACL;
 475                         rc = 0;
 476                 } else if (rc == -EOPNOTSUPP) {
 477                         rc = 0;
 478                 } else {
 479                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 480                                mdt_obd_name(info->mti_mdt),
 481                                PFID(mdt_object_fid(o)), rc);
 482                 }
 483         } else {
 484                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 485                                      rc, NODEMAP_FS_TO_CLIENT);
 486                 /* if all ACLs mapped out, rc is still >= 0 */
 487                 if (rc < 0) {
 488                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 489                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 490                                PFID(mdt_object_fid(o)), rc);
 491                 } else {
 492                         repbody->mbo_aclsize = rc;
 493                         repbody->mbo_valid |= OBD_MD_FLACL;
 494                         rc = 0;
 495                 }
 496         }
 497         return rc;
 498 }
 499 #endif
 500
 501 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 502                         const struct lu_attr *attr, const struct lu_fid *fid)
 503 {
 504         struct md_attr          *ma = &info->mti_attr;
 505         struct obd_export       *exp = info->mti_exp;
 506         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 507
 508         LASSERT(ma->ma_valid & MA_INODE);
 509
 510         if (attr->la_valid & LA_ATIME) {
 511                 b->mbo_atime = attr->la_atime;
 512                 b->mbo_valid |= OBD_MD_FLATIME;
 513         }
 514         if (attr->la_valid & LA_MTIME) {
 515                 b->mbo_mtime = attr->la_mtime;
 516                 b->mbo_valid |= OBD_MD_FLMTIME;
 517         }
 518         if (attr->la_valid & LA_CTIME) {
 519                 b->mbo_ctime = attr->la_ctime;
 520                 b->mbo_valid |= OBD_MD_FLCTIME;
 521         }
 522         if (attr->la_valid & LA_FLAGS) {
 523                 b->mbo_flags = attr->la_flags;
 524                 b->mbo_valid |= OBD_MD_FLFLAGS;
 525         }
 526         if (attr->la_valid & LA_NLINK) {
 527                 b->mbo_nlink = attr->la_nlink;
 528                 b->mbo_valid |= OBD_MD_FLNLINK;
 529         }
 530         if (attr->la_valid & LA_UID) {
 531                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 532                                             NODEMAP_FS_TO_CLIENT,
 533                                             attr->la_uid);
 534                 b->mbo_valid |= OBD_MD_FLUID;
 535         }
 536         if (attr->la_valid & LA_GID) {
 537                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 538                                             NODEMAP_FS_TO_CLIENT,
 539                                             attr->la_gid);
 540                 b->mbo_valid |= OBD_MD_FLGID;
 541         }
 542         b->mbo_mode = attr->la_mode;
 543         if (attr->la_valid & LA_MODE)
 544                 b->mbo_valid |= OBD_MD_FLMODE;
 545         if (attr->la_valid & LA_TYPE)
 546                 b->mbo_valid |= OBD_MD_FLTYPE;
 547
 548         if (fid != NULL) {
 549                 b->mbo_fid1 = *fid;
 550                 b->mbo_valid |= OBD_MD_FLID;
 551                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
 552                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 553         }
 554
 555         if (info != NULL)
 556                 mdt_body_reverse_idmap(info, b);
 557
 558         if (!(attr->la_valid & LA_TYPE))
 559                 return;
 560
 561         b->mbo_rdev   = attr->la_rdev;
 562         b->mbo_size   = attr->la_size;
 563         b->mbo_blocks = attr->la_blocks;
 564
 565         if (!S_ISREG(attr->la_mode)) {
 566                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 567         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 568                 /* means no objects are allocated on osts. */
 569                 LASSERT(!(ma->ma_valid & MA_LOV));
 570                 /* just ignore blocks occupied by extend attributes on MDS */
 571                 b->mbo_blocks = 0;
 572                 /* if no object is allocated on osts, the size on mds is valid.
 573                  * b=22272 */
 574                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 575         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 576                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 577                 /* A released file stores its size on MDS. */
 578                 /* But return 1 block for released file, unless tools like tar
 579                  * will consider it fully sparse. (LU-3864)
 580                  */
 581                 if (unlikely(b->mbo_size == 0))
 582                         b->mbo_blocks = 0;
 583                 else
 584                         b->mbo_blocks = 1;
 585                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 586         }
 587
 588         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 589                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 590                        PFID(fid), (unsigned long long)b->mbo_size);
 591 }
 592
 593 static inline int mdt_body_has_lov(const struct lu_attr *la,
 594                                    const struct mdt_body *body)
 595 {
 596         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 597                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 598 }
 599
 600 void mdt_client_compatibility(struct mdt_thread_info *info)
 601 {
 602         struct mdt_body       *body;
 603         struct ptlrpc_request *req = mdt_info_req(info);
 604         struct obd_export     *exp = req->rq_export;
 605         struct md_attr        *ma = &info->mti_attr;
 606         struct lu_attr        *la = &ma->ma_attr;
 607         ENTRY;
 608
 609         if (exp_connect_layout(exp))
 610                 /* the client can deal with 16-bit lmm_stripe_count */
 611                 RETURN_EXIT;
 612
 613         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 614
 615         if (!mdt_body_has_lov(la, body))
 616                 RETURN_EXIT;
 617
 618         /* now we have a reply with a lov for a client not compatible with the
 619          * layout lock so we have to clean the layout generation number */
 620         if (S_ISREG(la->la_mode))
 621                 ma->ma_lmm->lmm_layout_gen = 0;
 622         EXIT;
 623 }
 624
 625 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 626                                    struct mdt_object *o)
 627 {
 628         const struct lu_env *env = info->mti_env;
 629         int rc, rc2;
 630
 631         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 632                           XATTR_NAME_LOV);
 633
 634         if (rc == -ENODATA)
 635                 rc = 0;
 636
 637         if (rc < 0)
 638                 goto out;
 639
 640         /* Is it a directory? Let's check for the LMV as well */
 641         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 642                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 643                                    XATTR_NAME_LMV);
 644
 645                 if (rc2 == -ENODATA)
 646                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 647                                            &LU_BUF_NULL,
 648                                            XATTR_NAME_DEFAULT_LMV);
 649
 650                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 651                         rc = rc2;
 652         }
 653
 654 out:
 655         return rc;
 656 }
 657
 658 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 659                       const char *name)
 660 {
 661         const struct lu_env *env = info->mti_env;
 662         int rc;
 663         ENTRY;
 664
 665         LASSERT(info->mti_big_lmm_used == 0);
 666         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 667         if (rc < 0)
 668                 RETURN(rc);
 669
 670         /* big_lmm may need to be grown */
 671         if (info->mti_big_lmmsize < rc) {
 672                 int size = size_roundup_power2(rc);
 673
 674                 if (info->mti_big_lmmsize > 0) {
 675                         /* free old buffer */
 676                         LASSERT(info->mti_big_lmm);
 677                         OBD_FREE_LARGE(info->mti_big_lmm,
 678                                        info->mti_big_lmmsize);
 679                         info->mti_big_lmm = NULL;
 680                         info->mti_big_lmmsize = 0;
 681                 }
 682
 683                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 684                 if (info->mti_big_lmm == NULL)
 685                         RETURN(-ENOMEM);
 686                 info->mti_big_lmmsize = size;
 687         }
 688         LASSERT(info->mti_big_lmmsize >= rc);
 689
 690         info->mti_buf.lb_buf = info->mti_big_lmm;
 691         info->mti_buf.lb_len = info->mti_big_lmmsize;
 692         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 693
 694         RETURN(rc);
 695 }
 696
 697 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 698                    struct md_attr *ma, const char *name)
 699 {
 700         struct md_object *next = mdt_object_child(o);
 701         struct lu_buf    *buf = &info->mti_buf;
 702         int rc;
 703
 704         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 705                 buf->lb_buf = ma->ma_lmm;
 706                 buf->lb_len = ma->ma_lmm_size;
 707                 LASSERT(!(ma->ma_valid & MA_LOV));
 708         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 709                 buf->lb_buf = ma->ma_lmv;
 710                 buf->lb_len = ma->ma_lmv_size;
 711                 LASSERT(!(ma->ma_valid & MA_LMV));
 712         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 713                 buf->lb_buf = ma->ma_lmv;
 714                 buf->lb_len = ma->ma_lmv_size;
 715                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 716         } else {
 717                 return -EINVAL;
 718         }
 719
 720         rc = mo_xattr_get(info->mti_env, next, buf, name);
 721         if (rc > 0) {
 722
 723 got:
 724                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 725                         if (info->mti_big_lmm_used)
 726                                 ma->ma_lmm = info->mti_big_lmm;
 727
 728                         /* NOT return LOV EA with hole to old client. */
 729                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 730                                      LOV_PATTERN_F_HOLE) &&
 731                             !(exp_connect_flags(info->mti_exp) &
 732                               OBD_CONNECT_LFSCK)) {
 733                                 return -EIO;
 734                         } else {
 735                                 ma->ma_lmm_size = rc;
 736                                 ma->ma_valid |= MA_LOV;
 737                         }
 738                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 739                         if (info->mti_big_lmm_used)
 740                                 ma->ma_lmv = info->mti_big_lmm;
 741
 742                         ma->ma_lmv_size = rc;
 743                         ma->ma_valid |= MA_LMV;
 744                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 745                         ma->ma_lmv_size = rc;
 746                         ma->ma_valid |= MA_LMV_DEF;
 747                 }
 748
 749                 /* Update mdt_max_mdsize so all clients will be aware that */
 750                 if (info->mti_mdt->mdt_max_mdsize < rc)
 751                         info->mti_mdt->mdt_max_mdsize = rc;
 752
 753                 rc = 0;
 754         } else if (rc == -ENODATA) {
 755                 /* no LOV EA */
 756                 rc = 0;
 757         } else if (rc == -ERANGE) {
 758                 /* Default LMV has fixed size, so it must be able to fit
 759                  * in the original buffer */
 760                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 761                         return rc;
 762                 rc = mdt_big_xattr_get(info, o, name);
 763                 if (rc > 0) {
 764                         info->mti_big_lmm_used = 1;
 765                         goto got;
 766                 }
 767         }
 768
 769         return rc;
 770 }
 771
 772 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 773                              struct mdt_object *o, struct lu_fid *pfid)
 774 {
 775         struct lu_buf           *buf = &info->mti_buf;
 776         struct link_ea_header   *leh;
 777         struct link_ea_entry    *lee;
 778         int                      rc;
 779         ENTRY;
 780
 781         buf->lb_buf = info->mti_big_lmm;
 782         buf->lb_len = info->mti_big_lmmsize;
 783         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 784                           buf, XATTR_NAME_LINK);
 785         /* ignore errors, MA_PFID won't be set and it is
 786          * up to the caller to treat this as an error */
 787         if (rc == -ERANGE || buf->lb_len == 0) {
 788                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 789                 buf->lb_buf = info->mti_big_lmm;
 790                 buf->lb_len = info->mti_big_lmmsize;
 791         }
 792
 793         if (rc < 0)
 794                 RETURN(rc);
 795         if (rc < sizeof(*leh)) {
 796                 CERROR("short LinkEA on "DFID": rc = %d\n",
 797                        PFID(mdt_object_fid(o)), rc);
 798                 RETURN(-ENODATA);
 799         }
 800
 801         leh = (struct link_ea_header *) buf->lb_buf;
 802         lee = (struct link_ea_entry *)(leh + 1);
 803         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 804                 leh->leh_magic = LINK_EA_MAGIC;
 805                 leh->leh_reccount = __swab32(leh->leh_reccount);
 806                 leh->leh_len = __swab64(leh->leh_len);
 807         }
 808         if (leh->leh_magic != LINK_EA_MAGIC)
 809                 RETURN(-EINVAL);
 810         if (leh->leh_reccount == 0)
 811                 RETURN(-ENODATA);
 812
 813         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 814         fid_be_to_cpu(pfid, pfid);
 815
 816         RETURN(0);
 817 }
 818
 819 int mdt_attr_get_complex(struct mdt_thread_info *info,
 820                          struct mdt_object *o, struct md_attr *ma)
 821 {
 822         const struct lu_env *env = info->mti_env;
 823         struct md_object    *next = mdt_object_child(o);
 824         struct lu_buf       *buf = &info->mti_buf;
 825         int                  need = ma->ma_need;
 826         int                  rc = 0, rc2;
 827         u32                  mode;
 828         ENTRY;
 829
 830         ma->ma_valid = 0;
 831
 832         if (mdt_object_exists(o) == 0)
 833                 GOTO(out, rc = -ENOENT);
 834         mode = lu_object_attr(&next->mo_lu);
 835
 836         if (need & MA_INODE) {
 837                 ma->ma_need = MA_INODE;
 838                 rc = mo_attr_get(env, next, ma);
 839                 if (rc)
 840                         GOTO(out, rc);
 841                 ma->ma_valid |= MA_INODE;
 842         }
 843
 844         if (need & MA_PFID) {
 845                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 846                 if (rc == 0)
 847                         ma->ma_valid |= MA_PFID;
 848                 /* ignore this error, parent fid is not mandatory */
 849                 rc = 0;
 850         }
 851
 852         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 853                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 854                 if (rc)
 855                         GOTO(out, rc);
 856         }
 857
 858         if (need & MA_LMV && S_ISDIR(mode)) {
 859                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 860                 if (rc != 0)
 861                         GOTO(out, rc);
 862         }
 863
 864         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 865                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 866                 if (rc != 0)
 867                         GOTO(out, rc);
 868         }
 869
 870         if (need & MA_HSM && S_ISREG(mode)) {
 871                 buf->lb_buf = info->mti_xattr_buf;
 872                 buf->lb_len = sizeof(info->mti_xattr_buf);
 873                 CLASSERT(sizeof(struct hsm_attrs) <=
 874                          sizeof(info->mti_xattr_buf));
 875                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 876                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 877                 if (rc2 == 0)
 878                         ma->ma_valid |= MA_HSM;
 879                 else if (rc2 < 0 && rc2 != -ENODATA)
 880                         GOTO(out, rc = rc2);
 881         }
 882
 883 #ifdef CONFIG_FS_POSIX_ACL
 884         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 885                 buf->lb_buf = ma->ma_acl;
 886                 buf->lb_len = ma->ma_acl_size;
 887                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 888                 if (rc2 > 0) {
 889                         ma->ma_acl_size = rc2;
 890                         ma->ma_valid |= MA_ACL_DEF;
 891                 } else if (rc2 == -ENODATA) {
 892                         /* no ACLs */
 893                         ma->ma_acl_size = 0;
 894                 } else
 895                         GOTO(out, rc = rc2);
 896         }
 897 #endif
 898 out:
 899         ma->ma_need = need;
 900         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 901                rc, ma->ma_valid, ma->ma_lmm);
 902         RETURN(rc);
 903 }
 904
 905 static int mdt_getattr_internal(struct mdt_thread_info *info,
 906                                 struct mdt_object *o, int ma_need)
 907 {
 908         struct md_object        *next = mdt_object_child(o);
 909         const struct mdt_body   *reqbody = info->mti_body;
 910         struct ptlrpc_request   *req = mdt_info_req(info);
 911         struct md_attr          *ma = &info->mti_attr;
 912         struct lu_attr          *la = &ma->ma_attr;
 913         struct req_capsule      *pill = info->mti_pill;
 914         const struct lu_env     *env = info->mti_env;
 915         struct mdt_body         *repbody;
 916         struct lu_buf           *buffer = &info->mti_buf;
 917         struct obd_export       *exp = info->mti_exp;
 918         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 919         int                      rc;
 920         int                      is_root;
 921         ENTRY;
 922
 923         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 924                 RETURN(err_serious(-ENOMEM));
 925
 926         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 927
 928         ma->ma_valid = 0;
 929
 930         if (mdt_object_remote(o)) {
 931                 /* This object is located on remote node.*/
 932                 /* Return -ENOTSUPP for old client */
 933                 if (!mdt_is_dne_client(req->rq_export))
 934                         GOTO(out, rc = -ENOTSUPP);
 935
 936                 repbody->mbo_fid1 = *mdt_object_fid(o);
 937                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 938                 GOTO(out, rc = 0);
 939         }
 940
 941         if (reqbody->mbo_eadatasize > 0) {
 942                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 943                 if (buffer->lb_buf == NULL)
 944                         GOTO(out, rc = -EPROTO);
 945                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 946                                                       RCL_SERVER);
 947         } else {
 948                 buffer->lb_buf = NULL;
 949                 buffer->lb_len = 0;
 950                 ma_need &= ~(MA_LOV | MA_LMV);
 951                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 952                        mdt_obd_name(info->mti_mdt),
 953                        req->rq_export->exp_client_uuid.uuid);
 954         }
 955
 956         /* If it is dir object and client require MEA, then we got MEA */
 957         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 958             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 959                 /* Assumption: MDT_MD size is enough for lmv size. */
 960                 ma->ma_lmv = buffer->lb_buf;
 961                 ma->ma_lmv_size = buffer->lb_len;
 962                 ma->ma_need = MA_INODE;
 963                 if (ma->ma_lmv_size > 0) {
 964                         if (reqbody->mbo_valid & OBD_MD_MEA)
 965                                 ma->ma_need |= MA_LMV;
 966                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 967                                 ma->ma_need |= MA_LMV_DEF;
 968                 }
 969         } else {
 970                 ma->ma_lmm = buffer->lb_buf;
 971                 ma->ma_lmm_size = buffer->lb_len;
 972                 ma->ma_need = MA_INODE | MA_HSM;
 973                 if (ma->ma_lmm_size > 0)
 974                         ma->ma_need |= MA_LOV;
 975         }
 976
 977         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 978             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 979             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 980                 /* get default stripe info for this dir. */
 981                 ma->ma_need |= MA_LOV_DEF;
 982         }
 983         ma->ma_need |= ma_need;
 984
 985         rc = mdt_attr_get_complex(info, o, ma);
 986         if (unlikely(rc)) {
 987                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 988                        mdt_obd_name(info->mti_mdt),
 989                        PFID(mdt_object_fid(o)), rc);
 990                 RETURN(rc);
 991         }
 992
 993         /* if file is released, check if a restore is running */
 994         if (ma->ma_valid & MA_HSM) {
 995                 repbody->mbo_valid |= OBD_MD_TSTATE;
 996                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
 997                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
 998                         repbody->mbo_t_state = MS_RESTORE;
 999         }
1000
1001         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
1002
1003         /* the Lustre protocol supposes to return default striping
1004          * on the user-visible root if explicitly requested */
1005         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
1006             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
1007                 struct lu_fid      rootfid;
1008                 struct mdt_object *root;
1009                 struct mdt_device *mdt = info->mti_mdt;
1010
1011                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1012                 if (rc)
1013                         RETURN(rc);
1014                 root = mdt_object_find(env, mdt, &rootfid);
1015                 if (IS_ERR(root))
1016                         RETURN(PTR_ERR(root));
1017                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1018                 mdt_object_put(info->mti_env, root);
1019                 if (unlikely(rc)) {
1020                         CERROR("%s: getattr error for "DFID": rc = %d\n",
1021                                mdt_obd_name(info->mti_mdt),
1022                                PFID(mdt_object_fid(o)), rc);
1023                         RETURN(rc);
1024                 }
1025         }
1026
1027         if (likely(ma->ma_valid & MA_INODE))
1028                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1029         else
1030                 RETURN(-EFAULT);
1031
1032         if (mdt_body_has_lov(la, reqbody)) {
1033                 if (ma->ma_valid & MA_LOV) {
1034                         LASSERT(ma->ma_lmm_size);
1035                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1036                         if (S_ISDIR(la->la_mode))
1037                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1038                         else
1039                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1040                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1041                 }
1042                 if (ma->ma_valid & MA_LMV) {
1043                         /* Return -ENOTSUPP for old client */
1044                         if (!mdt_is_striped_client(req->rq_export))
1045                                 RETURN(-ENOTSUPP);
1046
1047                         LASSERT(S_ISDIR(la->la_mode));
1048                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1049                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1050                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1051                 }
1052                 if (ma->ma_valid & MA_LMV_DEF) {
1053                         /* Return -ENOTSUPP for old client */
1054                         if (!mdt_is_striped_client(req->rq_export))
1055                                 RETURN(-ENOTSUPP);
1056                         LASSERT(S_ISDIR(la->la_mode));
1057                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1058                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1059                                                OBD_MD_DEFAULT_MEA);
1060                 }
1061         } else if (S_ISLNK(la->la_mode) &&
1062                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1063                 buffer->lb_buf = ma->ma_lmm;
1064                 /* eadatasize from client includes NULL-terminator, so
1065                  * there is no need to read it */
1066                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1067                 rc = mo_readlink(env, next, buffer);
1068                 if (unlikely(rc <= 0)) {
1069                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1070                                mdt_obd_name(info->mti_mdt),
1071                                PFID(mdt_object_fid(o)), rc);
1072                         rc = -EFAULT;
1073                 } else {
1074                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1075
1076                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1077                                 rc -= 2;
1078                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1079                         /* we need to report back size with NULL-terminator
1080                          * because client expects that */
1081                         repbody->mbo_eadatasize = rc + 1;
1082                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1083                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1084                                        "on "DFID ", expected %d\n",
1085                                        mdt_obd_name(info->mti_mdt),
1086                                        rc, PFID(mdt_object_fid(o)),
1087                                        reqbody->mbo_eadatasize - 1);
1088                         /* NULL terminate */
1089                         ((char *)ma->ma_lmm)[rc] = 0;
1090
1091                         /* If the total CDEBUG() size is larger than a page, it
1092                          * will print a warning to the console, avoid this by
1093                          * printing just the last part of the symlink. */
1094                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1095                                print_limit < rc ? "..." : "", print_limit,
1096                                (char *)ma->ma_lmm + rc - print_limit, rc);
1097                         rc = 0;
1098                 }
1099         }
1100
1101         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1102                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1103                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1104                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1105                        repbody->mbo_max_mdsize);
1106         }
1107
1108         if (exp_connect_rmtclient(info->mti_exp) &&
1109             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1110                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1111
1112                 /* mdt_getattr_lock only */
1113                 rc = mdt_pack_remote_perm(info, o, buf);
1114                 if (rc) {
1115                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1116                         repbody->mbo_aclsize = 0;
1117                         RETURN(rc);
1118                 } else {
1119                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1120                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1121                 }
1122         }
1123 #ifdef CONFIG_FS_POSIX_ACL
1124         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1125                  (reqbody->mbo_valid & OBD_MD_FLACL))
1126                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1127 #endif
1128
1129 out:
1130         if (rc == 0)
1131                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1132
1133         RETURN(rc);
1134 }
1135
1136 static int mdt_getattr(struct tgt_session_info *tsi)
1137 {
1138         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1139         struct mdt_object       *obj = info->mti_object;
1140         struct req_capsule      *pill = info->mti_pill;
1141         struct mdt_body         *reqbody;
1142         struct mdt_body         *repbody;
1143         int rc, rc2;
1144         ENTRY;
1145
1146         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1147         LASSERT(reqbody);
1148         LASSERT(obj != NULL);
1149         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1150
1151         /* Unlike intent case where we need to pre-fill out buffers early on
1152          * in intent policy for ldlm reasons, here we can have a much better
1153          * guess at EA size by just reading it from disk.
1154          * Exceptions are readdir and (missing) directory striping */
1155         /* Readlink */
1156         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1157                 /* No easy way to know how long is the symlink, but it cannot
1158                  * be more than PATH_MAX, so we allocate +1 */
1159                 rc = PATH_MAX + 1;
1160
1161         /* A special case for fs ROOT: getattr there might fetch
1162          * default EA for entire fs, not just for this dir!
1163          */
1164         } else if (lu_fid_eq(mdt_object_fid(obj),
1165                              &info->mti_mdt->mdt_md_root_fid) &&
1166                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1167                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1168                                                                  MDS_GETATTR)) {
1169                 /* Should the default strping be bigger, mdt_fix_reply
1170                  * will reallocate */
1171                 rc = DEF_REP_MD_SIZE;
1172         } else {
1173                 /* Read the actual EA size from disk */
1174                 rc = mdt_attr_get_eabuf_size(info, obj);
1175         }
1176
1177         if (rc < 0)
1178                 GOTO(out, rc = err_serious(rc));
1179
1180         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1181
1182         rc = req_capsule_server_pack(pill);
1183         if (unlikely(rc != 0))
1184                 GOTO(out, rc = err_serious(rc));
1185
1186         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1187         LASSERT(repbody != NULL);
1188         repbody->mbo_eadatasize = 0;
1189         repbody->mbo_aclsize = 0;
1190
1191         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1192                 rc = mdt_init_ucred(info, reqbody);
1193         else
1194                 rc = mdt_check_ucred(info);
1195         if (unlikely(rc))
1196                 GOTO(out_shrink, rc);
1197
1198         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1199
1200         rc = mdt_getattr_internal(info, obj, 0);
1201         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1202                 mdt_exit_ucred(info);
1203         EXIT;
1204 out_shrink:
1205         mdt_client_compatibility(info);
1206         rc2 = mdt_fix_reply(info);
1207         if (rc == 0)
1208                 rc = rc2;
1209 out:
1210         mdt_thread_info_fini(info);
1211         return rc;
1212 }
1213
1214 /**
1215  * Exchange MOF_LOV_CREATED flags between two objects after a
1216  * layout swap. No assumption is made on whether o1 or o2 have
1217  * created objects or not.
1218  *
1219  * \param[in,out] o1    First swap layout object
1220  * \param[in,out] o2    Second swap layout object
1221  */
1222 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1223 {
1224         __u64   o1_flags;
1225
1226         mutex_lock(&o1->mot_lov_mutex);
1227         mutex_lock(&o2->mot_lov_mutex);
1228
1229         o1_flags = o1->mot_flags;
1230         o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1231                         (o2->mot_flags & MOF_LOV_CREATED);
1232
1233         o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1234                         (o1_flags & MOF_LOV_CREATED);
1235
1236         mutex_unlock(&o2->mot_lov_mutex);
1237         mutex_unlock(&o1->mot_lov_mutex);
1238 }
1239
1240 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1241 {
1242         struct mdt_thread_info  *info;
1243         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1244         struct obd_export       *exp = req->rq_export;
1245         struct mdt_object       *o1, *o2, *o;
1246         struct mdt_lock_handle  *lh1, *lh2;
1247         struct mdc_swap_layouts *msl;
1248         int                      rc;
1249         ENTRY;
1250
1251         /* client does not support layout lock, so layout swaping
1252          * is disabled.
1253          * FIXME: there is a problem for old clients which don't support
1254          * layout lock yet. If those clients have already opened the file
1255          * they won't be notified at all so that old layout may still be
1256          * used to do IO. This can be fixed after file release is landed by
1257          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1258         if (!exp_connect_layout(exp))
1259                 RETURN(-EOPNOTSUPP);
1260
1261         info = tsi2mdt_info(tsi);
1262
1263         if (info->mti_dlm_req != NULL)
1264                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1265
1266         o1 = info->mti_object;
1267         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1268                                 &info->mti_body->mbo_fid2);
1269         if (IS_ERR(o))
1270                 GOTO(out, rc = PTR_ERR(o));
1271
1272         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1273                 GOTO(put, rc = -ENOENT);
1274
1275         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1276         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1277                 GOTO(put, rc);
1278
1279         if (rc < 0)
1280                 swap(o1, o2);
1281
1282         /* permission check. Make sure the calling process having permission
1283          * to write both files. */
1284         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1285                                 MAY_WRITE);
1286         if (rc < 0)
1287                 GOTO(put, rc);
1288
1289         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1290                                 MAY_WRITE);
1291         if (rc < 0)
1292                 GOTO(put, rc);
1293
1294         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1295         if (msl == NULL)
1296                 GOTO(put, rc = -EPROTO);
1297
1298         lh1 = &info->mti_lh[MDT_LH_NEW];
1299         mdt_lock_reg_init(lh1, LCK_EX);
1300         lh2 = &info->mti_lh[MDT_LH_OLD];
1301         mdt_lock_reg_init(lh2, LCK_EX);
1302
1303         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1304                              MDS_INODELOCK_XATTR);
1305         if (rc < 0)
1306                 GOTO(put, rc);
1307
1308         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1309                              MDS_INODELOCK_XATTR);
1310         if (rc < 0)
1311                 GOTO(unlock1, rc);
1312
1313         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1314                              mdt_object_child(o2), msl->msl_flags);
1315         if (rc < 0)
1316                 GOTO(unlock2, rc);
1317
1318         mdt_swap_lov_flag(o1, o2);
1319
1320 unlock2:
1321         mdt_object_unlock(info, o2, lh2, rc);
1322 unlock1:
1323         mdt_object_unlock(info, o1, lh1, rc);
1324 put:
1325         mdt_object_put(info->mti_env, o);
1326 out:
1327         mdt_thread_info_fini(info);
1328         RETURN(rc);
1329 }
1330
1331 static int mdt_raw_lookup(struct mdt_thread_info *info,
1332                           struct mdt_object *parent,
1333                           const struct lu_name *lname,
1334                           struct ldlm_reply *ldlm_rep)
1335 {
1336         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1337         int              rc;
1338         ENTRY;
1339
1340         LASSERT(!info->mti_cross_ref);
1341
1342         /* Only got the fid of this obj by name */
1343         fid_zero(child_fid);
1344         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1345                         lname, child_fid, &info->mti_spec);
1346         if (rc == 0) {
1347                 struct mdt_body *repbody;
1348
1349                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1350                 repbody->mbo_fid1 = *child_fid;
1351                 repbody->mbo_valid = OBD_MD_FLID;
1352                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1353         } else if (rc == -ENOENT) {
1354                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1355         }
1356
1357         RETURN(rc);
1358 }
1359
1360 /*
1361  * UPDATE lock should be taken against parent, and be released before exit;
1362  * child_bits lock should be taken against child, and be returned back:
1363  *            (1)normal request should release the child lock;
1364  *            (2)intent request will grant the lock to client.
1365  */
1366 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1367                                  struct mdt_lock_handle *lhc,
1368                                  __u64 child_bits,
1369                                  struct ldlm_reply *ldlm_rep)
1370 {
1371         struct ptlrpc_request  *req       = mdt_info_req(info);
1372         struct mdt_body        *reqbody   = NULL;
1373         struct mdt_object      *parent    = info->mti_object;
1374         struct mdt_object      *child;
1375         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1376         struct lu_name         *lname     = NULL;
1377         struct mdt_lock_handle *lhp       = NULL;
1378         struct ldlm_lock       *lock;
1379         bool                    is_resent;
1380         bool                    try_layout;
1381         int                     ma_need = 0;
1382         int                     rc;
1383         ENTRY;
1384
1385         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1386         LASSERT(ergo(is_resent,
1387                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1388
1389         if (parent == NULL)
1390                 RETURN(-ENOENT);
1391
1392         if (info->mti_cross_ref) {
1393                 /* Only getattr on the child. Parent is on another node. */
1394                 mdt_set_disposition(info, ldlm_rep,
1395                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1396                 child = parent;
1397                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1398                        "ldlm_rep = %p\n",
1399                        PFID(mdt_object_fid(child)), ldlm_rep);
1400
1401                 rc = mdt_check_resent_lock(info, child, lhc);
1402                 if (rc < 0) {
1403                         RETURN(rc);
1404                 } else if (rc > 0) {
1405                         mdt_lock_handle_init(lhc);
1406                         mdt_lock_reg_init(lhc, LCK_PR);
1407
1408                         /*
1409                          * Object's name is on another MDS, no lookup or layout
1410                          * lock is needed here but update lock is.
1411                          */
1412                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1413                                         MDS_INODELOCK_LAYOUT);
1414                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1415
1416                         rc = mdt_object_lock(info, child, lhc, child_bits);
1417                         if (rc < 0)
1418                                 RETURN(rc);
1419                 }
1420
1421                 /* Finally, we can get attr for child. */
1422                 if (!mdt_object_exists(child)) {
1423                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1424                                         &child->mot_obj,
1425                                         "remote object doesn't exist.");
1426                         mdt_object_unlock(info, child, lhc, 1);
1427                         RETURN(-ENOENT);
1428                 }
1429
1430                 rc = mdt_getattr_internal(info, child, 0);
1431                 if (unlikely(rc != 0))
1432                         mdt_object_unlock(info, child, lhc, 1);
1433
1434                 RETURN(rc);
1435         }
1436
1437         lname = &info->mti_name;
1438         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1439
1440         if (lu_name_is_valid(lname)) {
1441                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1442                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1443                        PNAME(lname), ldlm_rep);
1444         } else {
1445                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1446                 if (unlikely(reqbody == NULL))
1447                         RETURN(err_serious(-EPROTO));
1448
1449                 *child_fid = reqbody->mbo_fid2;
1450
1451                 if (unlikely(!fid_is_sane(child_fid)))
1452                         RETURN(err_serious(-EINVAL));
1453
1454                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1455                        "ldlm_rep = %p\n",
1456                        PFID(mdt_object_fid(parent)),
1457                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1458         }
1459
1460         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1461
1462         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1463                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1464                                 &parent->mot_obj,
1465                                 "Parent doesn't exist!");
1466                 RETURN(-ESTALE);
1467         }
1468
1469         if (mdt_object_remote(parent)) {
1470                 CERROR("%s: parent "DFID" is on remote target\n",
1471                        mdt_obd_name(info->mti_mdt),
1472                        PFID(mdt_object_fid(parent)));
1473                 RETURN(-EIO);
1474         }
1475
1476         if (lu_name_is_valid(lname)) {
1477                 /* Always allow to lookup ".." */
1478                 if (unlikely(lname->ln_namelen == 2 &&
1479                              lname->ln_name[0] == '.' &&
1480                              lname->ln_name[1] == '.'))
1481                         info->mti_spec.sp_permitted = 1;
1482
1483                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1484                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1485
1486                         RETURN(rc);
1487                 }
1488
1489                 /* step 1: lock parent only if parent is a directory */
1490                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1491                         lhp = &info->mti_lh[MDT_LH_PARENT];
1492                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1493                         rc = mdt_object_lock(info, parent, lhp,
1494                                              MDS_INODELOCK_UPDATE);
1495                         if (unlikely(rc != 0))
1496                                 RETURN(rc);
1497                 }
1498
1499                 /* step 2: lookup child's fid by name */
1500                 fid_zero(child_fid);
1501                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1502                                 child_fid, &info->mti_spec);
1503                 if (rc == -ENOENT)
1504                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1505
1506                 if (rc != 0)
1507                         GOTO(out_parent, rc);
1508         }
1509
1510         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1511
1512         /*
1513          *step 3: find the child object by fid & lock it.
1514          *        regardless if it is local or remote.
1515          *
1516          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1517          *      set parent dir fid the same as child fid in getattr by fid case
1518          *      we should not lu_object_find() the object again, could lead
1519          *      to hung if there is a concurrent unlink destroyed the object.
1520          */
1521         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1522                 mdt_object_get(info->mti_env, parent);
1523                 child = parent;
1524         } else {
1525                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1526                                         child_fid);
1527         }
1528
1529         if (unlikely(IS_ERR(child)))
1530                 GOTO(out_parent, rc = PTR_ERR(child));
1531
1532         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1533         if (!mdt_object_exists(child)) {
1534                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1535                                 &child->mot_obj,
1536                                 "Object doesn't exist!");
1537                 GOTO(out_child, rc = -ENOENT);
1538         }
1539
1540         rc = mdt_check_resent_lock(info, child, lhc);
1541         if (rc < 0) {
1542                 GOTO(out_child, rc);
1543         } else if (rc > 0) {
1544                 mdt_lock_handle_init(lhc);
1545                 mdt_lock_reg_init(lhc, LCK_PR);
1546                 try_layout = false;
1547
1548                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1549                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1550                         struct md_attr *ma = &info->mti_attr;
1551
1552                         ma->ma_valid = 0;
1553                         ma->ma_need = MA_INODE;
1554                         rc = mdt_attr_get_complex(info, child, ma);
1555                         if (unlikely(rc != 0))
1556                                 GOTO(out_child, rc);
1557
1558                         /* If the file has not been changed for some time, we
1559                          * return not only a LOOKUP lock, but also an UPDATE
1560                          * lock and this might save us RPC on later STAT. For
1561                          * directories, it also let negative dentry cache start
1562                          * working for this dir. */
1563                         if (ma->ma_valid & MA_INODE &&
1564                             ma->ma_attr.la_valid & LA_CTIME &&
1565                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1566                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1567                                 child_bits |= MDS_INODELOCK_UPDATE;
1568                 }
1569
1570                 /* layout lock must be granted in a best-effort way
1571                  * for IT operations */
1572                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1573                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1574                     exp_connect_layout(info->mti_exp) &&
1575                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1576                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1577                         /* try to grant layout lock for regular file. */
1578                         try_layout = true;
1579                 }
1580
1581                 rc = 0;
1582                 if (try_layout) {
1583                         child_bits |= MDS_INODELOCK_LAYOUT;
1584                         /* try layout lock, it may fail to be granted due to
1585                          * contention at LOOKUP or UPDATE */
1586                         if (!mdt_object_lock_try(info, child, lhc,
1587                                                  child_bits)) {
1588                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1589                                 LASSERT(child_bits != 0);
1590                                 rc = mdt_object_lock(info, child, lhc,
1591                                                      child_bits);
1592                         } else {
1593                                 ma_need |= MA_LOV;
1594                         }
1595                 } else {
1596                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1597                          * client will enqueue the lock to the remote MDT */
1598                         if (mdt_object_remote(child))
1599                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1600                         rc = mdt_object_lock(info, child, lhc, child_bits);
1601                 }
1602                 if (unlikely(rc != 0))
1603                         GOTO(out_child, rc);
1604         }
1605
1606         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1607
1608         /* finally, we can get attr for child. */
1609         rc = mdt_getattr_internal(info, child, ma_need);
1610         if (unlikely(rc != 0)) {
1611                 mdt_object_unlock(info, child, lhc, 1);
1612         } else if (lock) {
1613                 /* Debugging code. */
1614                 LDLM_DEBUG(lock, "Returning lock to client");
1615                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1616                                          &lock->l_resource->lr_name),
1617                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1618                          PLDLMRES(lock->l_resource),
1619                          PFID(mdt_object_fid(child)));
1620         }
1621         if (lock)
1622                 LDLM_LOCK_PUT(lock);
1623
1624         EXIT;
1625 out_child:
1626         mdt_object_put(info->mti_env, child);
1627 out_parent:
1628         if (lhp)
1629                 mdt_object_unlock(info, parent, lhp, 1);
1630         return rc;
1631 }
1632
1633 /* normal handler: should release the child lock */
1634 static int mdt_getattr_name(struct tgt_session_info *tsi)
1635 {
1636         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1637         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1638         struct mdt_body        *reqbody;
1639         struct mdt_body        *repbody;
1640         int rc, rc2;
1641         ENTRY;
1642
1643         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1644         LASSERT(reqbody != NULL);
1645         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1646         LASSERT(repbody != NULL);
1647
1648         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1649         repbody->mbo_eadatasize = 0;
1650         repbody->mbo_aclsize = 0;
1651
1652         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1653         if (unlikely(rc))
1654                 GOTO(out_shrink, rc);
1655
1656         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1657         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1658                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1659                 lhc->mlh_reg_lh.cookie = 0;
1660         }
1661         mdt_exit_ucred(info);
1662         EXIT;
1663 out_shrink:
1664         mdt_client_compatibility(info);
1665         rc2 = mdt_fix_reply(info);
1666         if (rc == 0)
1667                 rc = rc2;
1668         mdt_thread_info_fini(info);
1669         return rc;
1670 }
1671
1672 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1673                          void *karg, void __user *uarg);
1674
1675 static int mdt_set_info(struct tgt_session_info *tsi)
1676 {
1677         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1678         char                    *key;
1679         void                    *val;
1680         int                      keylen, vallen, rc = 0;
1681
1682         ENTRY;
1683
1684         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1685         if (key == NULL) {
1686                 DEBUG_REQ(D_HA, req, "no set_info key");
1687                 RETURN(err_serious(-EFAULT));
1688         }
1689
1690         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1691                                       RCL_CLIENT);
1692
1693         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1694         if (val == NULL) {
1695                 DEBUG_REQ(D_HA, req, "no set_info val");
1696                 RETURN(err_serious(-EFAULT));
1697         }
1698
1699         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1700                                       RCL_CLIENT);
1701
1702         /* Swab any part of val you need to here */
1703         if (KEY_IS(KEY_READ_ONLY)) {
1704                 spin_lock(&req->rq_export->exp_lock);
1705                 if (*(__u32 *)val)
1706                         *exp_connect_flags_ptr(req->rq_export) |=
1707                                 OBD_CONNECT_RDONLY;
1708                 else
1709                         *exp_connect_flags_ptr(req->rq_export) &=
1710                                 ~OBD_CONNECT_RDONLY;
1711                 spin_unlock(&req->rq_export->exp_lock);
1712         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1713                 struct changelog_setinfo *cs = val;
1714
1715                 if (vallen != sizeof(*cs)) {
1716                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1717                                tgt_name(tsi->tsi_tgt), vallen);
1718                         RETURN(-EINVAL);
1719                 }
1720                 if (ptlrpc_req_need_swab(req)) {
1721                         __swab64s(&cs->cs_recno);
1722                         __swab32s(&cs->cs_id);
1723                 }
1724
1725                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1726                                    vallen, val, NULL);
1727         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1728                 if (vallen > 0)
1729                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1730         } else {
1731                 RETURN(-EINVAL);
1732         }
1733         RETURN(rc);
1734 }
1735
1736 static int mdt_readpage(struct tgt_session_info *tsi)
1737 {
1738         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1739         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1740         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1741         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1742         struct mdt_body         *repbody;
1743         int                      rc;
1744         int                      i;
1745
1746         ENTRY;
1747
1748         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1749                 RETURN(err_serious(-ENOMEM));
1750
1751         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1752         if (repbody == NULL || reqbody == NULL)
1753                 RETURN(err_serious(-EFAULT));
1754
1755         /*
1756          * prepare @rdpg before calling lower layers and transfer itself. Here
1757          * reqbody->size contains offset of where to start to read and
1758          * reqbody->nlink contains number bytes to read.
1759          */
1760         rdpg->rp_hash = reqbody->mbo_size;
1761         if (rdpg->rp_hash != reqbody->mbo_size) {
1762                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1763                        rdpg->rp_hash, reqbody->mbo_size);
1764                 RETURN(-EFAULT);
1765         }
1766
1767         rdpg->rp_attrs = reqbody->mbo_mode;
1768         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1769                 rdpg->rp_attrs |= LUDA_64BITHASH;
1770         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1771                                 exp_max_brw_size(tsi->tsi_exp));
1772         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1773                           PAGE_CACHE_SHIFT;
1774         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1775         if (rdpg->rp_pages == NULL)
1776                 RETURN(-ENOMEM);
1777
1778         for (i = 0; i < rdpg->rp_npages; ++i) {
1779                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1780                 if (rdpg->rp_pages[i] == NULL)
1781                         GOTO(free_rdpg, rc = -ENOMEM);
1782         }
1783
1784         /* call lower layers to fill allocated pages with directory data */
1785         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1786         if (rc < 0)
1787                 GOTO(free_rdpg, rc);
1788
1789         /* send pages to client */
1790         rc = tgt_sendpage(tsi, rdpg, rc);
1791
1792         EXIT;
1793 free_rdpg:
1794
1795         for (i = 0; i < rdpg->rp_npages; i++)
1796                 if (rdpg->rp_pages[i] != NULL)
1797                         __free_page(rdpg->rp_pages[i]);
1798         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1799
1800         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1801                 RETURN(0);
1802
1803         return rc;
1804 }
1805
1806 static int mdt_reint_internal(struct mdt_thread_info *info,
1807                               struct mdt_lock_handle *lhc,
1808                               __u32 op)
1809 {
1810         struct req_capsule      *pill = info->mti_pill;
1811         struct mdt_body         *repbody;
1812         int                      rc = 0, rc2;
1813
1814         ENTRY;
1815
1816         rc = mdt_reint_unpack(info, op);
1817         if (rc != 0) {
1818                 CERROR("Can't unpack reint, rc %d\n", rc);
1819                 RETURN(err_serious(rc));
1820         }
1821
1822         /* for replay (no_create) lmm is not needed, client has it already */
1823         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1824                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1825                                      DEF_REP_MD_SIZE);
1826
1827         /* llog cookies are always 0, the field is kept for compatibility */
1828         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1829                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1830
1831         rc = req_capsule_server_pack(pill);
1832         if (rc != 0) {
1833                 CERROR("Can't pack response, rc %d\n", rc);
1834                 RETURN(err_serious(rc));
1835         }
1836
1837         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1838                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1839                 LASSERT(repbody);
1840                 repbody->mbo_eadatasize = 0;
1841                 repbody->mbo_aclsize = 0;
1842         }
1843
1844         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1845
1846         /* for replay no cookkie / lmm need, because client have this already */
1847         if (info->mti_spec.no_create)
1848                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1849                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1850
1851         rc = mdt_init_ucred_reint(info);
1852         if (rc)
1853                 GOTO(out_shrink, rc);
1854
1855         rc = mdt_fix_attr_ucred(info, op);
1856         if (rc != 0)
1857                 GOTO(out_ucred, rc = err_serious(rc));
1858
1859         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1860         if (rc < 0) {
1861                 GOTO(out_ucred, rc);
1862         } else if (rc == 1) {
1863                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1864                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1865                 GOTO(out_ucred, rc);
1866         }
1867         rc = mdt_reint_rec(info, lhc);
1868         EXIT;
1869 out_ucred:
1870         mdt_exit_ucred(info);
1871 out_shrink:
1872         mdt_client_compatibility(info);
1873         rc2 = mdt_fix_reply(info);
1874         if (rc == 0)
1875                 rc = rc2;
1876         return rc;
1877 }
1878
1879 static long mdt_reint_opcode(struct ptlrpc_request *req,
1880                              const struct req_format **fmt)
1881 {
1882         struct mdt_device       *mdt;
1883         struct mdt_rec_reint    *rec;
1884         long                     opc;
1885
1886         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1887         if (rec != NULL) {
1888                 opc = rec->rr_opcode;
1889                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1890                 if (opc < REINT_MAX && fmt[opc] != NULL)
1891                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1892                 else {
1893                         mdt = mdt_exp2dev(req->rq_export);
1894                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1895                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1896                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1897                         opc = err_serious(-EFAULT);
1898                 }
1899         } else {
1900                 opc = err_serious(-EFAULT);
1901         }
1902         return opc;
1903 }
1904
1905 static int mdt_reint(struct tgt_session_info *tsi)
1906 {
1907         long opc;
1908         int  rc;
1909         static const struct req_format *reint_fmts[REINT_MAX] = {
1910                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1911                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1912                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1913                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1914                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1915                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1916                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1917                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1918                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1919         };
1920
1921         ENTRY;
1922
1923         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1924         if (opc >= 0) {
1925                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1926                 /*
1927                  * No lock possible here from client to pass it to reint code
1928                  * path.
1929                  */
1930                 rc = mdt_reint_internal(info, NULL, opc);
1931                 mdt_thread_info_fini(info);
1932         } else {
1933                 rc = opc;
1934         }
1935
1936         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1937         RETURN(rc);
1938 }
1939
1940 /* this should sync the whole device */
1941 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1942 {
1943         struct dt_device *dt = mdt->mdt_bottom;
1944         int rc;
1945         ENTRY;
1946
1947         rc = dt->dd_ops->dt_sync(env, dt);
1948         RETURN(rc);
1949 }
1950
1951 /* this should sync this object */
1952 static int mdt_object_sync(struct mdt_thread_info *info)
1953 {
1954         struct md_object *next;
1955         int rc;
1956         ENTRY;
1957
1958         if (!mdt_object_exists(info->mti_object)) {
1959                 CWARN("%s: non existing object "DFID": rc = %d\n",
1960                       mdt_obd_name(info->mti_mdt),
1961                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1962                 RETURN(-ESTALE);
1963         }
1964         next = mdt_object_child(info->mti_object);
1965         rc = mo_object_sync(info->mti_env, next);
1966
1967         RETURN(rc);
1968 }
1969
1970 static int mdt_sync(struct tgt_session_info *tsi)
1971 {
1972         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1973         struct req_capsule      *pill = tsi->tsi_pill;
1974         struct mdt_body         *body;
1975         int                      rc;
1976
1977         ENTRY;
1978
1979         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1980                 RETURN(err_serious(-ENOMEM));
1981
1982         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1983                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1984         } else {
1985                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1986
1987                 /* sync an object */
1988                 rc = mdt_object_sync(info);
1989                 if (rc == 0) {
1990                         const struct lu_fid *fid;
1991                         struct lu_attr *la = &info->mti_attr.ma_attr;
1992
1993                         info->mti_attr.ma_need = MA_INODE;
1994                         info->mti_attr.ma_valid = 0;
1995                         rc = mdt_attr_get_complex(info, info->mti_object,
1996                                                   &info->mti_attr);
1997                         if (rc == 0) {
1998                                 body = req_capsule_server_get(pill,
1999                                                               &RMF_MDT_BODY);
2000                                 fid = mdt_object_fid(info->mti_object);
2001                                 mdt_pack_attr2body(info, body, la, fid);
2002                         }
2003                 }
2004                 mdt_thread_info_fini(info);
2005         }
2006         if (rc == 0)
2007                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2008
2009         RETURN(rc);
2010 }
2011
2012 /*
2013  * Handle quota control requests to consult current usage/limit, but also
2014  * to configure quota enforcement
2015  */
2016 static int mdt_quotactl(struct tgt_session_info *tsi)
2017 {
2018         struct obd_export       *exp  = tsi->tsi_exp;
2019         struct req_capsule      *pill = tsi->tsi_pill;
2020         struct obd_quotactl     *oqctl, *repoqc;
2021         int                      id, rc;
2022         struct mdt_device       *mdt = mdt_exp2dev(exp);
2023         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2024         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
2025         ENTRY;
2026
2027         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2028         if (oqctl == NULL)
2029                 RETURN(err_serious(-EPROTO));
2030
2031         rc = req_capsule_server_pack(pill);
2032         if (rc)
2033                 RETURN(err_serious(rc));
2034
2035         switch (oqctl->qc_cmd) {
2036                 /* master quotactl */
2037         case Q_SETINFO:
2038         case Q_SETQUOTA:
2039                 if (!nodemap_can_setquota(nodemap))
2040                         RETURN(-EPERM);
2041         case Q_GETINFO:
2042         case Q_GETQUOTA:
2043                 if (qmt == NULL)
2044                         RETURN(-EOPNOTSUPP);
2045                 /* slave quotactl */
2046         case Q_GETOINFO:
2047         case Q_GETOQUOTA:
2048                 break;
2049         default:
2050                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2051                 RETURN(-EFAULT);
2052         }
2053
2054         /* map uid/gid for remote client */
2055         id = oqctl->qc_id;
2056         if (exp_connect_rmtclient(exp)) {
2057                 struct lustre_idmap_table *idmap;
2058
2059                 idmap = exp->exp_mdt_data.med_idmap;
2060
2061                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2062                              oqctl->qc_cmd != Q_GETINFO))
2063                         RETURN(-EPERM);
2064
2065                 if (oqctl->qc_type == USRQUOTA)
2066                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2067                                                      oqctl->qc_id);
2068                 else if (oqctl->qc_type == GRPQUOTA)
2069                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2070                                                      oqctl->qc_id);
2071                 else
2072                         RETURN(-EINVAL);
2073
2074                 if (id == CFS_IDMAP_NOTFOUND) {
2075                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2076                         RETURN(-EACCES);
2077                 }
2078         }
2079
2080         if (oqctl->qc_type == USRQUOTA)
2081                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2082                                     NODEMAP_CLIENT_TO_FS, id);
2083         else if (oqctl->qc_type == GRPQUOTA)
2084                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2085                                     NODEMAP_CLIENT_TO_FS, id);
2086
2087         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2088         if (repoqc == NULL)
2089                 RETURN(err_serious(-EFAULT));
2090
2091         if (oqctl->qc_id != id)
2092                 swap(oqctl->qc_id, id);
2093
2094         switch (oqctl->qc_cmd) {
2095
2096         case Q_GETINFO:
2097         case Q_SETINFO:
2098         case Q_SETQUOTA:
2099         case Q_GETQUOTA:
2100                 /* forward quotactl request to QMT */
2101                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2102                 break;
2103
2104         case Q_GETOINFO:
2105         case Q_GETOQUOTA:
2106                 /* slave quotactl */
2107                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2108                                    oqctl);
2109                 break;
2110
2111         default:
2112                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2113                 RETURN(-EFAULT);
2114         }
2115
2116         if (oqctl->qc_id != id)
2117                 swap(oqctl->qc_id, id);
2118
2119         *repoqc = *oqctl;
2120         RETURN(rc);
2121 }
2122
2123 /** clone llog ctxt from child (mdd)
2124  * This allows remote llog (replicator) access.
2125  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2126  * context was originally set up, or we can handle them directly.
2127  * I choose the latter, but that means I need any llog
2128  * contexts set up by child to be accessable by the mdt.  So we clone the
2129  * context into our context list here.
2130  */
2131 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2132                                int idx)
2133 {
2134         struct md_device  *next = mdt->mdt_child;
2135         struct llog_ctxt *ctxt;
2136         int rc;
2137
2138         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2139                 return 0;
2140
2141         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2142         if (rc || ctxt == NULL) {
2143                 return 0;
2144         }
2145
2146         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2147         if (rc)
2148                 CERROR("Can't set mdt ctxt %d\n", rc);
2149
2150         return rc;
2151 }
2152
2153 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2154                                  struct mdt_device *mdt, int idx)
2155 {
2156         struct llog_ctxt *ctxt;
2157
2158         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2159         if (ctxt == NULL)
2160                 return 0;
2161         /* Put once for the get we just did, and once for the clone */
2162         llog_ctxt_put(ctxt);
2163         llog_ctxt_put(ctxt);
2164         return 0;
2165 }
2166
2167 /*
2168  * sec context handlers
2169  */
2170 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2171 {
2172         int rc;
2173
2174         rc = mdt_handle_idmap(tsi);
2175         if (unlikely(rc)) {
2176                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2177                 __u32                    opc;
2178
2179                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2180                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2181                         sptlrpc_svc_ctx_invalidate(req);
2182         }
2183
2184         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2185
2186         return rc;
2187 }
2188
2189 /*
2190  * quota request handlers
2191  */
2192 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2193 {
2194         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2195         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2196         int                      rc;
2197         ENTRY;
2198
2199         if (qmt == NULL)
2200                 RETURN(err_serious(-EOPNOTSUPP));
2201
2202         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2203         RETURN(rc);
2204 }
2205
2206 struct mdt_object *mdt_object_new(const struct lu_env *env,
2207                                   struct mdt_device *d,
2208                                   const struct lu_fid *f)
2209 {
2210         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2211         struct lu_object *o;
2212         struct mdt_object *m;
2213         ENTRY;
2214
2215         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2216         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2217         if (unlikely(IS_ERR(o)))
2218                 m = (struct mdt_object *)o;
2219         else
2220                 m = mdt_obj(o);
2221         RETURN(m);
2222 }
2223
2224 struct mdt_object *mdt_object_find(const struct lu_env *env,
2225                                    struct mdt_device *d,
2226                                    const struct lu_fid *f)
2227 {
2228         struct lu_object *o;
2229         struct mdt_object *m;
2230         ENTRY;
2231
2232         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2233         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2234         if (unlikely(IS_ERR(o)))
2235                 m = (struct mdt_object *)o;
2236         else
2237                 m = mdt_obj(o);
2238
2239         RETURN(m);
2240 }
2241
2242 /**
2243  * Asyncronous commit for mdt device.
2244  *
2245  * Pass asynchonous commit call down the MDS stack.
2246  *
2247  * \param env environment
2248  * \param mdt the mdt device
2249  */
2250 static void mdt_device_commit_async(const struct lu_env *env,
2251                                     struct mdt_device *mdt)
2252 {
2253         struct dt_device *dt = mdt->mdt_bottom;
2254         int rc;
2255         ENTRY;
2256
2257         rc = dt->dd_ops->dt_commit_async(env, dt);
2258         if (unlikely(rc != 0))
2259                 CWARN("%s: async commit start failed: rc = %d\n",
2260                       mdt_obd_name(mdt), rc);
2261         atomic_inc(&mdt->mdt_async_commit_count);
2262         EXIT;
2263 }
2264
2265 /**
2266  * Mark the lock as "synchonous".
2267  *
2268  * Mark the lock to deffer transaction commit to the unlock time.
2269  *
2270  * \param lock the lock to mark as "synchonous"
2271  *
2272  * \see mdt_is_lock_sync
2273  * \see mdt_save_lock
2274  */
2275 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2276 {
2277         lock->l_ast_data = (void*)1;
2278 }
2279
2280 /**
2281  * Check whehter the lock "synchonous" or not.
2282  *
2283  * \param lock the lock to check
2284  * \retval 1 the lock is "synchonous"
2285  * \retval 0 the lock isn't "synchronous"
2286  *
2287  * \see mdt_set_lock_sync
2288  * \see mdt_save_lock
2289  */
2290 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2291 {
2292         return lock->l_ast_data != NULL;
2293 }
2294
2295 /**
2296  * Blocking AST for mdt locks.
2297  *
2298  * Starts transaction commit if in case of COS lock conflict or
2299  * deffers such a commit to the mdt_save_lock.
2300  *
2301  * \param lock the lock which blocks a request or cancelling lock
2302  * \param desc unused
2303  * \param data unused
2304  * \param flag indicates whether this cancelling or blocking callback
2305  * \retval 0
2306  * \see ldlm_blocking_ast_nocheck
2307  */
2308 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2309                      void *data, int flag)
2310 {
2311         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2312         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2313         int rc;
2314         ENTRY;
2315
2316         if (flag == LDLM_CB_CANCELING)
2317                 RETURN(0);
2318
2319         lock_res_and_lock(lock);
2320         if (lock->l_blocking_ast != mdt_blocking_ast) {
2321                 unlock_res_and_lock(lock);
2322                 RETURN(0);
2323         }
2324         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2325             lock->l_blocking_lock != NULL) {
2326                 if (mdt_cos_is_enabled(mdt) &&
2327                     lock->l_client_cookie !=
2328                     lock->l_blocking_lock->l_client_cookie)
2329                         mdt_set_lock_sync(lock);
2330                 else if (mdt_slc_is_enabled(mdt) &&
2331                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2332                         mdt_set_lock_sync(lock);
2333         }
2334         rc = ldlm_blocking_ast_nocheck(lock);
2335
2336         /* There is no lock conflict if l_blocking_lock == NULL,
2337          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2338          * when the last reference to a local lock was released */
2339         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2340                 struct lu_env env;
2341
2342                 rc = lu_env_init(&env, LCT_LOCAL);
2343                 if (unlikely(rc != 0))
2344                         CWARN("%s: lu_env initialization failed, cannot "
2345                               "start asynchronous commit: rc = %d\n",
2346                               obd->obd_name, rc);
2347                 else
2348                         mdt_device_commit_async(&env, mdt);
2349                 lu_env_fini(&env);
2350         }
2351         RETURN(rc);
2352 }
2353
2354 /*
2355  * Blocking AST for cross-MDT lock
2356  *
2357  * Discard lock from uncommitted_slc_locks and cancel it.
2358  *
2359  * \param lock  the lock which blocks a request or cancelling lock
2360  * \param desc  unused
2361  * \param data  unused
2362  * \param flag  indicates whether this cancelling or blocking callback
2363  * \retval      0 on success
2364  * \retval      negative number on error
2365  */
2366 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2367                             void *data, int flag)
2368 {
2369         struct lustre_handle lockh;
2370         int               rc;
2371         ENTRY;
2372
2373         switch (flag) {
2374         case LDLM_CB_BLOCKING:
2375                 ldlm_lock2handle(lock, &lockh);
2376                 rc = ldlm_cli_cancel(&lockh,
2377                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2378                 if (rc < 0) {
2379                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2380                         RETURN(rc);
2381                 }
2382                 break;
2383         case LDLM_CB_CANCELING:
2384                 LDLM_DEBUG(lock, "Revoke remote lock");
2385                 /* discard slc lock here so that it can be cleaned anytime,
2386                  * especially for cleanup_resource() */
2387                 tgt_discard_slc_lock(lock);
2388                 break;
2389         default:
2390                 LBUG();
2391         }
2392
2393         RETURN(0);
2394 }
2395
2396 int mdt_check_resent_lock(struct mdt_thread_info *info,
2397                           struct mdt_object *mo,
2398                           struct mdt_lock_handle *lhc)
2399 {
2400         /* the lock might already be gotten in ldlm_handle_enqueue() */
2401         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2402                 struct ptlrpc_request *req = mdt_info_req(info);
2403                 struct ldlm_lock      *lock;
2404
2405                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2406                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2407                 if (lock == NULL) {
2408                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2409                          * a resend case, however, it could be already destroyed
2410                          * due to client eviction or a raced cancel RPC. */
2411                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64,
2412                                           lhc->mlh_reg_lh.cookie);
2413                         RETURN(-ESTALE);
2414                 }
2415
2416                 if (!fid_res_name_eq(mdt_object_fid(mo),
2417                                      &lock->l_resource->lr_name)) {
2418                         CWARN("%s: Although resent, but still not "
2419                               "get child lock:"DFID"\n",
2420                               info->mti_exp->exp_obd->obd_name,
2421                               PFID(mdt_object_fid(mo)));
2422                         LDLM_LOCK_PUT(lock);
2423                         RETURN(-EPROTO);
2424                 }
2425                 LDLM_LOCK_PUT(lock);
2426                 return 0;
2427         }
2428         return 1;
2429 }
2430
2431 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2432                            const struct lu_fid *fid, struct lustre_handle *lh,
2433                            enum ldlm_mode mode, __u64 ibits, bool nonblock)
2434 {
2435         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2436         union ldlm_policy_data *policy = &mti->mti_policy;
2437         struct ldlm_res_id *res_id = &mti->mti_res_id;
2438         int rc = 0;
2439         ENTRY;
2440
2441         LASSERT(mdt_object_remote(o));
2442
2443         fid_build_reg_res_name(fid, res_id);
2444
2445         memset(einfo, 0, sizeof(*einfo));
2446         einfo->ei_type = LDLM_IBITS;
2447         einfo->ei_mode = mode;
2448         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2449         einfo->ei_cb_cp = ldlm_completion_ast;
2450         einfo->ei_enq_slave = 0;
2451         einfo->ei_res_id = res_id;
2452         if (nonblock)
2453                 einfo->ei_nonblock = 1;
2454
2455         memset(policy, 0, sizeof(*policy));
2456         policy->l_inodebits.bits = ibits;
2457
2458         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2459                             policy);
2460         RETURN(rc);
2461 }
2462
2463 static int mdt_object_local_lock(struct mdt_thread_info *info,
2464                                  struct mdt_object *o,
2465                                  struct mdt_lock_handle *lh, __u64 ibits,
2466                                  bool nonblock, bool cos_incompat)
2467 {
2468         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2469         union ldlm_policy_data *policy = &info->mti_policy;
2470         struct ldlm_res_id *res_id = &info->mti_res_id;
2471         __u64 dlmflags = 0;
2472         int rc;
2473         ENTRY;
2474
2475         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2476         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2477         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2478         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2479
2480         if (cos_incompat) {
2481                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2482                         lh->mlh_reg_mode == LCK_EX);
2483                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2484         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2485                 dlmflags |= LDLM_FL_COS_ENABLED;
2486         }
2487
2488         /* Only enqueue LOOKUP lock for remote object */
2489         if (mdt_object_remote(o))
2490                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2491
2492         if (lh->mlh_type == MDT_PDO_LOCK) {
2493                 /* check for exists after object is locked */
2494                 if (mdt_object_exists(o) == 0) {
2495                         /* Non-existent object shouldn't have PDO lock */
2496                         RETURN(-ESTALE);
2497                 } else {
2498                         /* Non-dir object shouldn't have PDO lock */
2499                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2500                                 RETURN(-ENOTDIR);
2501                 }
2502         }
2503
2504         memset(policy, 0, sizeof(*policy));
2505         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2506
2507         dlmflags |= LDLM_FL_ATOMIC_CB;
2508         if (nonblock)
2509                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2510
2511         /*
2512          * Take PDO lock on whole directory and build correct @res_id for lock
2513          * on part of directory.
2514          */
2515         if (lh->mlh_pdo_hash != 0) {
2516                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2517                 mdt_lock_pdo_mode(info, o, lh);
2518                 if (lh->mlh_pdo_mode != LCK_NL) {
2519                         /*
2520                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2521                          * is never going to be sent to client and we do not
2522                          * want it slowed down due to possible cancels.
2523                          */
2524                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2525                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2526                                           policy, res_id, dlmflags,
2527                                           info->mti_exp == NULL ? NULL :
2528                                           &info->mti_exp->exp_handle.h_cookie);
2529                         if (unlikely(rc != 0))
2530                                 GOTO(out_unlock, rc);
2531                 }
2532
2533                 /*
2534                  * Finish res_id initializing by name hash marking part of
2535                  * directory which is taking modification.
2536                  */
2537                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2538         }
2539
2540         policy->l_inodebits.bits = ibits;
2541
2542         /*
2543          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2544          * going to be sent to client. If it is - mdt_intent_policy() path will
2545          * fix it up and turn FL_LOCAL flag off.
2546          */
2547         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2548                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2549                           info->mti_exp == NULL ? NULL :
2550                           &info->mti_exp->exp_handle.h_cookie);
2551 out_unlock:
2552         if (rc != 0)
2553                 mdt_object_unlock(info, o, lh, 1);
2554         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2555                    lh->mlh_pdo_hash != 0 &&
2556                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2557                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2558
2559         RETURN(rc);
2560 }
2561
2562 static int
2563 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2564                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2565                          bool cos_incompat)
2566 {
2567         struct mdt_lock_handle *local_lh = NULL;
2568         int rc;
2569         ENTRY;
2570
2571         if (!mdt_object_remote(o)) {
2572                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2573                                            cos_incompat);
2574                 RETURN(rc);
2575         }
2576
2577         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2578         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2579                    MDS_INODELOCK_XATTR);
2580
2581         /* Only enqueue LOOKUP lock for remote object */
2582         if (ibits & MDS_INODELOCK_LOOKUP) {
2583                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2584                                            nonblock, cos_incompat);
2585                 if (rc != ELDLM_OK)
2586                         RETURN(rc);
2587
2588                 local_lh = lh;
2589         }
2590
2591         if (ibits & MDS_INODELOCK_UPDATE) {
2592                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2593                  * enqueue RPC can only request 1 lock, to avoid extra
2594                  * RPC, so it will instead enqueue EX lock for remote
2595                  * object anyway XXX*/
2596                 if (lh->mlh_type == MDT_PDO_LOCK &&
2597                     lh->mlh_pdo_hash != 0) {
2598                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2599                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2600                                PFID(mdt_object_fid(o)));
2601                         lh->mlh_pdo_hash = 0;
2602                         lh->mlh_rreg_mode = LCK_EX;
2603                         lh->mlh_type = MDT_REG_LOCK;
2604                 }
2605                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2606                                             &lh->mlh_rreg_lh,
2607                                             lh->mlh_rreg_mode,
2608                                             MDS_INODELOCK_UPDATE, nonblock);
2609                 if (rc != ELDLM_OK) {
2610                         if (local_lh != NULL)
2611                                 mdt_object_unlock(info, o, local_lh, rc);
2612                         RETURN(rc);
2613                 }
2614         }
2615
2616         RETURN(0);
2617 }
2618
2619 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2620                     struct mdt_lock_handle *lh, __u64 ibits)
2621 {
2622         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2623 }
2624
2625 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2626                           struct mdt_lock_handle *lh, __u64 ibits,
2627                           bool cos_incompat)
2628 {
2629         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2630         return mdt_object_lock_internal(info, o, lh, ibits, false,
2631                                         cos_incompat);
2632 }
2633
2634 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2635                         struct mdt_lock_handle *lh, __u64 ibits)
2636 {
2637         struct mdt_lock_handle tmp = *lh;
2638         int rc;
2639
2640         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2641         if (rc == 0)
2642                 *lh = tmp;
2643
2644         return rc == 0;
2645 }
2646
2647 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2648                               struct mdt_object *o, struct mdt_lock_handle *lh,
2649                               __u64 ibits, bool cos_incompat)
2650 {
2651         struct mdt_lock_handle tmp = *lh;
2652         int rc;
2653
2654         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2655         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2656         if (rc == 0)
2657                 *lh = tmp;
2658
2659         return rc == 0;
2660 }
2661
2662 /**
2663  * Save a lock within request object.
2664  *
2665  * Keep the lock referenced until whether client ACK or transaction
2666  * commit happens or release the lock immediately depending on input
2667  * parameters. If COS is ON, a write lock is converted to COS lock
2668  * before saving.
2669  *
2670  * \param info thead info object
2671  * \param h lock handle
2672  * \param mode lock mode
2673  * \param decref force immediate lock releasing
2674  */
2675 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2676                           enum ldlm_mode mode, int decref)
2677 {
2678         ENTRY;
2679
2680         if (lustre_handle_is_used(h)) {
2681                 if (decref || !info->mti_has_trans ||
2682                     !(mode & (LCK_PW | LCK_EX))) {
2683                         mdt_fid_unlock(h, mode);
2684                 } else {
2685                         struct mdt_device *mdt = info->mti_mdt;
2686                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2687                         struct ptlrpc_request *req = mdt_info_req(info);
2688                         int cos;
2689
2690                         cos = (mdt_cos_is_enabled(mdt) ||
2691                                mdt_slc_is_enabled(mdt));
2692
2693                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2694                                  h->cookie);
2695
2696                         /* there is no request if mdt_object_unlock() is called
2697                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2698                         if (likely(req != NULL)) {
2699                                 CDEBUG(D_HA, "request = %p reply state = %p"
2700                                        " transno = "LPD64"\n", req,
2701                                        req->rq_reply_state, req->rq_transno);
2702                                 if (cos) {
2703                                         ldlm_lock_downgrade(lock, LCK_COS);
2704                                         mode = LCK_COS;
2705                                 }
2706                                 ptlrpc_save_lock(req, h, mode, cos);
2707                         } else {
2708                                 mdt_fid_unlock(h, mode);
2709                         }
2710                         if (mdt_is_lock_sync(lock)) {
2711                                 CDEBUG(D_HA, "found sync-lock,"
2712                                        " async commit started\n");
2713                                 mdt_device_commit_async(info->mti_env,
2714                                                         mdt);
2715                         }
2716                         LDLM_LOCK_PUT(lock);
2717                 }
2718                 h->cookie = 0ull;
2719         }
2720
2721         EXIT;
2722 }
2723
2724 /**
2725  * Save cross-MDT lock in uncommitted_slc_locks
2726  *
2727  * Keep the lock referenced until transaction commit happens or release the lock
2728  * immediately depending on input parameters.
2729  *
2730  * \param info thead info object
2731  * \param h lock handle
2732  * \param mode lock mode
2733  * \param decref force immediate lock releasing
2734  */
2735 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2736                                  struct lustre_handle *h, enum ldlm_mode mode,
2737                                  int decref)
2738 {
2739         ENTRY;
2740
2741         if (lustre_handle_is_used(h)) {
2742                 if (decref || !info->mti_has_trans ||
2743                     !(mode & (LCK_PW | LCK_EX))) {
2744                         ldlm_lock_decref_and_cancel(h, mode);
2745                 } else {
2746                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2747                         struct ptlrpc_request *req = mdt_info_req(info);
2748
2749                         LASSERT(req != NULL);
2750                         tgt_save_slc_lock(lock, req->rq_transno);
2751                         ldlm_lock_decref(h, mode);
2752                 }
2753                 h->cookie = 0ull;
2754         }
2755
2756         EXIT;
2757 }
2758
2759 /**
2760  * Unlock mdt object.
2761  *
2762  * Immeditely release the regular lock and the PDO lock or save the
2763  * lock in request and keep them referenced until client ACK or
2764  * transaction commit.
2765  *
2766  * \param info thread info object
2767  * \param o mdt object
2768  * \param lh mdt lock handle referencing regular and PDO locks
2769  * \param decref force immediate lock releasing
2770  */
2771 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2772                        struct mdt_lock_handle *lh, int decref)
2773 {
2774         ENTRY;
2775
2776         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2777         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2778         mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
2779
2780         EXIT;
2781 }
2782
2783 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2784                                         const struct lu_fid *f,
2785                                         struct mdt_lock_handle *lh,
2786                                         __u64 ibits)
2787 {
2788         struct mdt_object *o;
2789
2790         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2791         if (!IS_ERR(o)) {
2792                 int rc;
2793
2794                 rc = mdt_object_lock(info, o, lh, ibits);
2795                 if (rc != 0) {
2796                         mdt_object_put(info->mti_env, o);
2797                         o = ERR_PTR(rc);
2798                 }
2799         }
2800         return o;
2801 }
2802
2803 void mdt_object_unlock_put(struct mdt_thread_info * info,
2804                            struct mdt_object * o,
2805                            struct mdt_lock_handle *lh,
2806                            int decref)
2807 {
2808         mdt_object_unlock(info, o, lh, decref);
2809         mdt_object_put(info->mti_env, o);
2810 }
2811
2812 /*
2813  * Generic code handling requests that have struct mdt_body passed in:
2814  *
2815  *  - extract mdt_body from request and save it in @info, if present;
2816  *
2817  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2818  *  @info;
2819  *
2820  *  - if HABEO_CORPUS flag is set for this request type check whether object
2821  *  actually exists on storage (lu_object_exists()).
2822  *
2823  */
2824 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2825 {
2826         const struct mdt_body    *body;
2827         struct mdt_object        *obj;
2828         const struct lu_env      *env;
2829         struct req_capsule       *pill;
2830         int                       rc;
2831         ENTRY;
2832
2833         env = info->mti_env;
2834         pill = info->mti_pill;
2835
2836         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2837         if (body == NULL)
2838                 RETURN(-EFAULT);
2839
2840         if (!(body->mbo_valid & OBD_MD_FLID))
2841                 RETURN(0);
2842
2843         if (!fid_is_sane(&body->mbo_fid1)) {
2844                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2845                 RETURN(-EINVAL);
2846         }
2847
2848         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2849         if (!IS_ERR(obj)) {
2850                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2851                         mdt_object_put(env, obj);
2852                         rc = -ENOENT;
2853                 } else {
2854                         info->mti_object = obj;
2855                         rc = 0;
2856                 }
2857         } else
2858                 rc = PTR_ERR(obj);
2859
2860         RETURN(rc);
2861 }
2862
2863 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2864 {
2865         struct req_capsule *pill = info->mti_pill;
2866         int rc;
2867         ENTRY;
2868
2869         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2870                 rc = mdt_body_unpack(info, flags);
2871         else
2872                 rc = 0;
2873
2874         if (rc == 0 && (flags & HABEO_REFERO)) {
2875                 /* Pack reply. */
2876                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2877                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2878                                              DEF_REP_MD_SIZE);
2879                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2880                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2881                                              RCL_SERVER, 0);
2882
2883                 rc = req_capsule_server_pack(pill);
2884         }
2885         RETURN(rc);
2886 }
2887
2888 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2889 {
2890         lh->mlh_type = MDT_NUL_LOCK;
2891         lh->mlh_reg_lh.cookie = 0ull;
2892         lh->mlh_reg_mode = LCK_MINMODE;
2893         lh->mlh_pdo_lh.cookie = 0ull;
2894         lh->mlh_pdo_mode = LCK_MINMODE;
2895         lh->mlh_rreg_lh.cookie = 0ull;
2896         lh->mlh_rreg_mode = LCK_MINMODE;
2897 }
2898
2899 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2900 {
2901         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2902         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2903 }
2904
2905 /*
2906  * Initialize fields of struct mdt_thread_info. Other fields are left in
2907  * uninitialized state, because it's too expensive to zero out whole
2908  * mdt_thread_info (> 1K) on each request arrival.
2909  */
2910 void mdt_thread_info_init(struct ptlrpc_request *req,
2911                           struct mdt_thread_info *info)
2912 {
2913         int i;
2914
2915         info->mti_pill = &req->rq_pill;
2916
2917         /* lock handle */
2918         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2919                 mdt_lock_handle_init(&info->mti_lh[i]);
2920
2921         /* mdt device: it can be NULL while CONNECT */
2922         if (req->rq_export) {
2923                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2924                 info->mti_exp = req->rq_export;
2925         } else
2926                 info->mti_mdt = NULL;
2927         info->mti_env = req->rq_svc_thread->t_env;
2928         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2929
2930         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2931         info->mti_big_buf = LU_BUF_NULL;
2932         info->mti_body = NULL;
2933         info->mti_object = NULL;
2934         info->mti_dlm_req = NULL;
2935         info->mti_has_trans = 0;
2936         info->mti_cross_ref = 0;
2937         info->mti_opdata = 0;
2938         info->mti_big_lmm_used = 0;
2939
2940         info->mti_spec.no_create = 0;
2941         info->mti_spec.sp_rm_entry = 0;
2942         info->mti_spec.sp_permitted = 0;
2943         info->mti_spec.sp_migrate_close = 0;
2944
2945         info->mti_spec.u.sp_ea.eadata = NULL;
2946         info->mti_spec.u.sp_ea.eadatalen = 0;
2947 }
2948
2949 void mdt_thread_info_fini(struct mdt_thread_info *info)
2950 {
2951         int i;
2952
2953         if (info->mti_object != NULL) {
2954                 mdt_object_put(info->mti_env, info->mti_object);
2955                 info->mti_object = NULL;
2956         }
2957
2958         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2959                 mdt_lock_handle_fini(&info->mti_lh[i]);
2960         info->mti_env = NULL;
2961         info->mti_pill = NULL;
2962         info->mti_exp = NULL;
2963
2964         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2965                 lu_buf_free(&info->mti_big_buf);
2966 }
2967
2968 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2969 {
2970         struct mdt_thread_info  *mti;
2971
2972         mti = mdt_th_info(tsi->tsi_env);
2973         LASSERT(mti != NULL);
2974
2975         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2976         if (tsi->tsi_corpus != NULL) {
2977                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2978                 lu_object_get(tsi->tsi_corpus);
2979         }
2980         mti->mti_body = tsi->tsi_mdt_body;
2981         mti->mti_dlm_req = tsi->tsi_dlm_req;
2982
2983         return mti;
2984 }
2985
2986 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2987 {
2988         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2989         int                      rc;
2990
2991         ENTRY;
2992
2993         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2994             cfs_fail_val ==
2995             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2996                 set_current_state(TASK_UNINTERRUPTIBLE);
2997                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2998         }
2999
3000         rc = tgt_connect(tsi);
3001         if (rc != 0)
3002                 RETURN(rc);
3003
3004         rc = mdt_init_idmap(tsi);
3005         if (rc != 0)
3006                 GOTO(err, rc);
3007         RETURN(0);
3008 err:
3009         obd_disconnect(class_export_get(req->rq_export));
3010         return rc;
3011 }
3012
3013 enum mdt_it_code {
3014         MDT_IT_OPEN,
3015         MDT_IT_OCREAT,
3016         MDT_IT_CREATE,
3017         MDT_IT_GETATTR,
3018         MDT_IT_READDIR,
3019         MDT_IT_LOOKUP,
3020         MDT_IT_UNLINK,
3021         MDT_IT_TRUNC,
3022         MDT_IT_GETXATTR,
3023         MDT_IT_LAYOUT,
3024         MDT_IT_QUOTA,
3025         MDT_IT_NR
3026 };
3027
3028 static int mdt_intent_getattr(enum mdt_it_code opcode,
3029                               struct mdt_thread_info *info,
3030                               struct ldlm_lock **,
3031                               __u64);
3032
3033 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3034                                 struct mdt_thread_info *info,
3035                                 struct ldlm_lock **lockp,
3036                                 __u64 flags);
3037
3038 static int mdt_intent_layout(enum mdt_it_code opcode,
3039                              struct mdt_thread_info *info,
3040                              struct ldlm_lock **,
3041                              __u64);
3042 static int mdt_intent_reint(enum mdt_it_code opcode,
3043                             struct mdt_thread_info *info,
3044                             struct ldlm_lock **,
3045                             __u64);
3046
3047 static struct mdt_it_flavor {
3048         const struct req_format *it_fmt;
3049         __u32                    it_flags;
3050         int                    (*it_act)(enum mdt_it_code ,
3051                                          struct mdt_thread_info *,
3052                                          struct ldlm_lock **,
3053                                          __u64);
3054         long                     it_reint;
3055 } mdt_it_flavor[] = {
3056         [MDT_IT_OPEN]     = {
3057                 .it_fmt   = &RQF_LDLM_INTENT,
3058                 /*.it_flags = HABEO_REFERO,*/
3059                 .it_flags = 0,
3060                 .it_act   = mdt_intent_reint,
3061                 .it_reint = REINT_OPEN
3062         },
3063         [MDT_IT_OCREAT]   = {
3064                 .it_fmt   = &RQF_LDLM_INTENT,
3065                 /*
3066                  * OCREAT is not a MUTABOR request as if the file
3067                  * already exists.
3068                  * We do the extra check of OBD_CONNECT_RDONLY in
3069                  * mdt_reint_open() when we really need to create
3070                  * the object.
3071                  */
3072                 .it_flags = 0,
3073                 .it_act   = mdt_intent_reint,
3074                 .it_reint = REINT_OPEN
3075         },
3076         [MDT_IT_CREATE]   = {
3077                 .it_fmt   = &RQF_LDLM_INTENT,
3078                 .it_flags = MUTABOR,
3079                 .it_act   = mdt_intent_reint,
3080                 .it_reint = REINT_CREATE
3081         },
3082         [MDT_IT_GETATTR]  = {
3083                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3084                 .it_flags = HABEO_REFERO,
3085                 .it_act   = mdt_intent_getattr
3086         },
3087         [MDT_IT_READDIR]  = {
3088                 .it_fmt   = NULL,
3089                 .it_flags = 0,
3090                 .it_act   = NULL
3091         },
3092         [MDT_IT_LOOKUP]   = {
3093                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3094                 .it_flags = HABEO_REFERO,
3095                 .it_act   = mdt_intent_getattr
3096         },
3097         [MDT_IT_UNLINK]   = {
3098                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3099                 .it_flags = MUTABOR,
3100                 .it_act   = NULL,
3101                 .it_reint = REINT_UNLINK
3102         },
3103         [MDT_IT_TRUNC]    = {
3104                 .it_fmt   = NULL,
3105                 .it_flags = MUTABOR,
3106                 .it_act   = NULL
3107         },
3108         [MDT_IT_GETXATTR] = {
3109                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3110                 .it_flags = HABEO_CORPUS,
3111                 .it_act   = mdt_intent_getxattr
3112         },
3113         [MDT_IT_LAYOUT] = {
3114                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3115                 .it_flags = 0,
3116                 .it_act   = mdt_intent_layout
3117         }
3118 };
3119
3120 static int
3121 mdt_intent_lock_replace(struct mdt_thread_info *info,
3122                         struct ldlm_lock **lockp,
3123                         struct mdt_lock_handle *lh,
3124                         __u64 flags, int result)
3125 {
3126         struct ptlrpc_request  *req = mdt_info_req(info);
3127         struct ldlm_lock       *lock = *lockp;
3128         struct ldlm_lock       *new_lock;
3129
3130         /* If possible resent found a lock, @lh is set to its handle */
3131         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3132
3133         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3134                 lh->mlh_reg_lh.cookie = 0;
3135                 RETURN(0);
3136         }
3137
3138         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3139                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3140                  * a resend case, however, it could be already destroyed
3141                  * due to client eviction or a raced cancel RPC. */
3142                 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
3143                                   lh->mlh_reg_lh.cookie);
3144                 lh->mlh_reg_lh.cookie = 0;
3145                 RETURN(-ESTALE);
3146         }
3147
3148         LASSERTF(new_lock != NULL,
3149                  "lockh "LPX64" flags "LPX64" rc %d\n",
3150                  lh->mlh_reg_lh.cookie, flags, result);
3151
3152         /*
3153          * If we've already given this lock to a client once, then we should
3154          * have no readers or writers.  Otherwise, we should have one reader
3155          * _or_ writer ref (which will be zeroed below) before returning the
3156          * lock to a client.
3157          */
3158         if (new_lock->l_export == req->rq_export) {
3159                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3160         } else {
3161                 LASSERT(new_lock->l_export == NULL);
3162                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3163         }
3164
3165         *lockp = new_lock;
3166
3167         if (new_lock->l_export == req->rq_export) {
3168                 /*
3169                  * Already gave this to the client, which means that we
3170                  * reconstructed a reply.
3171                  */
3172                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3173                         MSG_RESENT);
3174
3175                 LDLM_LOCK_RELEASE(new_lock);
3176                 lh->mlh_reg_lh.cookie = 0;
3177                 RETURN(ELDLM_LOCK_REPLACED);
3178         }
3179
3180         /*
3181          * Fixup the lock to be given to the client.
3182          */
3183         lock_res_and_lock(new_lock);
3184         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3185          * possible blocking AST. */
3186         while (new_lock->l_readers > 0) {
3187                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3188                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3189                 new_lock->l_readers--;
3190         }
3191         while (new_lock->l_writers > 0) {
3192                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3193                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3194                 new_lock->l_writers--;
3195         }
3196
3197         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3198         new_lock->l_blocking_ast = lock->l_blocking_ast;
3199         new_lock->l_completion_ast = lock->l_completion_ast;
3200         new_lock->l_remote_handle = lock->l_remote_handle;
3201         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3202
3203         unlock_res_and_lock(new_lock);
3204
3205         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3206                      &new_lock->l_remote_handle,
3207                      &new_lock->l_exp_hash);
3208
3209         LDLM_LOCK_RELEASE(new_lock);
3210         lh->mlh_reg_lh.cookie = 0;
3211
3212         RETURN(ELDLM_LOCK_REPLACED);
3213 }
3214
3215 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3216                                     struct ldlm_lock *new_lock,
3217                                     struct mdt_lock_handle *lh,
3218                                     __u64 flags)
3219 {
3220         struct ptlrpc_request  *req = mdt_info_req(info);
3221         struct ldlm_request    *dlmreq;
3222
3223         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3224                 return;
3225
3226         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3227
3228         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3229          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3230          * initialized. */
3231         if (flags & LDLM_FL_RESENT) {
3232                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3233                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3234
3235                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3236                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3237                           lh->mlh_reg_lh.cookie);
3238                 return;
3239         }
3240
3241         /*
3242          * If the xid matches, then we know this is a resent request, and allow
3243          * it. (It's probably an OPEN, for which we don't send a lock.
3244          */
3245         if (req_can_reconstruct(req, NULL))
3246                 return;
3247
3248         /*
3249          * This remote handle isn't enqueued, so we never received or processed
3250          * this request.  Clear MSG_RESENT, because it can be handled like any
3251          * normal request now.
3252          */
3253         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3254
3255         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3256                   dlmreq->lock_handle[0].cookie);
3257 }
3258
3259 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3260                                 struct mdt_thread_info *info,
3261                                 struct ldlm_lock **lockp,
3262                                 __u64 flags)
3263 {
3264         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3265         struct ldlm_reply      *ldlm_rep = NULL;
3266         int rc, grc;
3267
3268         /*
3269          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3270          * (for the resend case) or a new lock. Below we will use it to
3271          * replace the original lock.
3272          */
3273         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3274         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3275                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3276                 rc = mdt_object_lock(info, info->mti_object, lhc,
3277                                      MDS_INODELOCK_XATTR);
3278                 if (rc)
3279                         return rc;
3280         }
3281
3282         grc = mdt_getxattr(info);
3283
3284         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, 0);
3285
3286         if (mdt_info_req(info)->rq_repmsg != NULL)
3287                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3288         if (ldlm_rep == NULL)
3289                 RETURN(err_serious(-EFAULT));
3290
3291         ldlm_rep->lock_policy_res2 = grc;
3292
3293         return rc;
3294 }
3295
3296 static int mdt_intent_getattr(enum mdt_it_code opcode,
3297                               struct mdt_thread_info *info,
3298                               struct ldlm_lock **lockp,
3299                               __u64 flags)
3300 {
3301         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3302         __u64                   child_bits;
3303         struct ldlm_reply      *ldlm_rep;
3304         struct mdt_body        *reqbody;
3305         struct mdt_body        *repbody;
3306         int                     rc, rc2;
3307         ENTRY;
3308
3309         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3310         LASSERT(reqbody);
3311
3312         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3313         LASSERT(repbody);
3314
3315         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3316         repbody->mbo_eadatasize = 0;
3317         repbody->mbo_aclsize = 0;
3318
3319         switch (opcode) {
3320         case MDT_IT_LOOKUP:
3321                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3322                 break;
3323         case MDT_IT_GETATTR:
3324                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3325                              MDS_INODELOCK_PERM;
3326                 break;
3327         default:
3328                 CERROR("Unsupported intent (%d)\n", opcode);
3329                 GOTO(out_shrink, rc = -EINVAL);
3330         }
3331
3332         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3333         if (rc)
3334                 GOTO(out_shrink, rc);
3335
3336         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3337         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3338
3339         /* Get lock from request for possible resent case. */
3340         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3341
3342         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3343         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3344
3345         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3346                 ldlm_rep->lock_policy_res2 = 0;
3347         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3348             ldlm_rep->lock_policy_res2) {
3349                 lhc->mlh_reg_lh.cookie = 0ull;
3350                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3351         }
3352
3353         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3354         EXIT;
3355 out_ucred:
3356         mdt_exit_ucred(info);
3357 out_shrink:
3358         mdt_client_compatibility(info);
3359         rc2 = mdt_fix_reply(info);
3360         if (rc == 0)
3361                 rc = rc2;
3362         return rc;
3363 }
3364
3365 static int mdt_intent_layout(enum mdt_it_code opcode,
3366                              struct mdt_thread_info *info,
3367                              struct ldlm_lock **lockp,
3368                              __u64 flags)
3369 {
3370         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3371         struct layout_intent *layout;
3372         struct lu_fid *fid;
3373         struct mdt_object *obj = NULL;
3374         int layout_size = 0;
3375         int rc = 0;
3376         ENTRY;
3377
3378         if (opcode != MDT_IT_LAYOUT) {
3379                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3380                         opcode);
3381                 RETURN(-EINVAL);
3382         }
3383
3384         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3385         if (layout == NULL)
3386                 RETURN(-EPROTO);
3387
3388         if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3389                 CERROR("%s: Unsupported layout intent opc %d\n",
3390                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3391                 RETURN(-EINVAL);
3392         }
3393
3394         fid = &info->mti_tmp_fid2;
3395         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3396
3397         /* Get lock from request for possible resent case. */
3398         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3399
3400         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3401         if (IS_ERR(obj))
3402                 GOTO(out, rc = PTR_ERR(obj));
3403
3404         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3405                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3406                 if (layout_size < 0)
3407                         GOTO(out_obj, rc = layout_size);
3408
3409                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3410                         info->mti_mdt->mdt_max_mdsize = layout_size;
3411         }
3412
3413         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3414         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3415                              layout_size);
3416         rc = req_capsule_server_pack(info->mti_pill);
3417         GOTO(out_obj, rc);
3418
3419 out_obj:
3420         mdt_object_put(info->mti_env, obj);
3421
3422         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3423                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3424
3425 out:
3426         lhc->mlh_reg_lh.cookie = 0;
3427
3428         return rc;
3429 }
3430
3431 static int mdt_intent_reint(enum mdt_it_code opcode,
3432                             struct mdt_thread_info *info,
3433                             struct ldlm_lock **lockp,
3434                             __u64 flags)
3435 {
3436         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3437         struct ldlm_reply      *rep = NULL;
3438         long                    opc;
3439         int                     rc;
3440
3441         static const struct req_format *intent_fmts[REINT_MAX] = {
3442                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3443                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3444         };
3445
3446         ENTRY;
3447
3448         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3449         if (opc < 0)
3450                 RETURN(opc);
3451
3452         if (mdt_it_flavor[opcode].it_reint != opc) {
3453                 CERROR("Reint code %ld doesn't match intent: %d\n",
3454                        opc, opcode);
3455                 RETURN(err_serious(-EPROTO));
3456         }
3457
3458         /* Get lock from request for possible resent case. */
3459         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3460
3461         rc = mdt_reint_internal(info, lhc, opc);
3462
3463         /* Check whether the reply has been packed successfully. */
3464         if (mdt_info_req(info)->rq_repmsg != NULL)
3465                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3466         if (rep == NULL)
3467                 RETURN(err_serious(-EFAULT));
3468
3469         /* MDC expects this in any case */
3470         if (rc != 0)
3471                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3472
3473         /* the open lock or the lock for cross-ref object should be
3474          * returned to the client */
3475         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3476             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3477                 rep->lock_policy_res2 = 0;
3478                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3479                 RETURN(rc);
3480         }
3481
3482         rep->lock_policy_res2 = clear_serious(rc);
3483
3484         if (rep->lock_policy_res2 == -ENOENT &&
3485             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3486             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3487                 rep->lock_policy_res2 = 0;
3488
3489         lhc->mlh_reg_lh.cookie = 0ull;
3490         if (rc == -ENOTCONN || rc == -ENODEV ||
3491             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3492                 /*
3493                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3494                  * will be returned by rq_status, and client at ptlrpc layer
3495                  * will detect this, then disconnect, reconnect the import
3496                  * immediately, instead of impacting the following the rpc.
3497                  */
3498                 RETURN(rc);
3499         }
3500         /*
3501          * For other cases, the error will be returned by intent, and client
3502          * will retrieve the result from intent.
3503          */
3504         RETURN(ELDLM_LOCK_ABORTED);
3505 }
3506
3507 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3508 {
3509         int rc;
3510
3511         switch (itcode) {
3512         case IT_OPEN:
3513                 rc = MDT_IT_OPEN;
3514                 break;
3515         case IT_OPEN|IT_CREAT:
3516                 rc = MDT_IT_OCREAT;
3517                 break;
3518         case IT_CREAT:
3519                 rc = MDT_IT_CREATE;
3520                 break;
3521         case IT_READDIR:
3522                 rc = MDT_IT_READDIR;
3523                 break;
3524         case IT_GETATTR:
3525                 rc = MDT_IT_GETATTR;
3526                 break;
3527         case IT_LOOKUP:
3528                 rc = MDT_IT_LOOKUP;
3529                 break;
3530         case IT_UNLINK:
3531                 rc = MDT_IT_UNLINK;
3532                 break;
3533         case IT_TRUNC:
3534                 rc = MDT_IT_TRUNC;
3535                 break;
3536         case IT_GETXATTR:
3537                 rc = MDT_IT_GETXATTR;
3538                 break;
3539         case IT_LAYOUT:
3540                 rc = MDT_IT_LAYOUT;
3541                 break;
3542         case IT_QUOTA_DQACQ:
3543         case IT_QUOTA_CONN:
3544                 rc = MDT_IT_QUOTA;
3545                 break;
3546         default:
3547                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3548                 rc = -EINVAL;
3549                 break;
3550         }
3551         return rc;
3552 }
3553
3554 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3555                           struct mdt_thread_info *info,
3556                           struct ldlm_lock **lockp, __u64 flags)
3557 {
3558         struct req_capsule      *pill = info->mti_pill;
3559         struct ptlrpc_request   *req = mdt_info_req(info);
3560         struct mdt_it_flavor    *flv;
3561         int opc;
3562         int rc;
3563         ENTRY;
3564
3565         opc = mdt_intent_code(itopc);
3566         if (opc < 0)
3567                 RETURN(-EINVAL);
3568
3569         if (opc == MDT_IT_QUOTA) {
3570                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3571
3572                 if (qmt == NULL)
3573                         RETURN(-EOPNOTSUPP);
3574
3575                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3576                 /* pass the request to quota master */
3577                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3578                                                  mdt_info_req(info), lockp,
3579                                                  flags);
3580                 RETURN(rc);
3581         }
3582
3583         flv = &mdt_it_flavor[opc];
3584         if (flv->it_fmt != NULL)
3585                 req_capsule_extend(pill, flv->it_fmt);
3586
3587         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3588         if (rc < 0)
3589                 RETURN(rc);
3590
3591         if (flv->it_flags & MUTABOR &&
3592             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3593                 RETURN(-EROFS);
3594
3595         if (flv->it_act != NULL) {
3596                 struct ldlm_reply *rep;
3597
3598                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3599
3600                 /* execute policy */
3601                 rc = flv->it_act(opc, info, lockp, flags);
3602
3603                 /* Check whether the reply has been packed successfully. */
3604                 if (req->rq_repmsg != NULL) {
3605                         rep = req_capsule_server_get(info->mti_pill,
3606                                                      &RMF_DLM_REP);
3607                         rep->lock_policy_res2 =
3608                                 ptlrpc_status_hton(rep->lock_policy_res2);
3609                 }
3610         }
3611
3612         RETURN(rc);
3613 }
3614
3615 static int mdt_intent_policy(struct ldlm_namespace *ns,
3616                              struct ldlm_lock **lockp, void *req_cookie,
3617                              enum ldlm_mode mode, __u64 flags, void *data)
3618 {
3619         struct tgt_session_info *tsi;
3620         struct mdt_thread_info  *info;
3621         struct ptlrpc_request   *req  =  req_cookie;
3622         struct ldlm_intent      *it;
3623         struct req_capsule      *pill;
3624         int rc;
3625
3626         ENTRY;
3627
3628         LASSERT(req != NULL);
3629
3630         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3631
3632         info = tsi2mdt_info(tsi);
3633         LASSERT(info != NULL);
3634         pill = info->mti_pill;
3635         LASSERT(pill->rc_req == req);
3636
3637         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3638                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3639                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3640                 if (it != NULL) {
3641                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3642                         if (rc == 0)
3643                                 rc = ELDLM_OK;
3644
3645                         /* Lock without inodebits makes no sense and will oops
3646                          * later in ldlm. Let's check it now to see if we have
3647                          * ibits corrupted somewhere in mdt_intent_opc().
3648                          * The case for client miss to set ibits has been
3649                          * processed by others. */
3650                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3651                                         lr_type == LDLM_IBITS,
3652                                      info->mti_dlm_req->lock_desc.\
3653                                         l_policy_data.l_inodebits.bits != 0));
3654                 } else
3655                         rc = err_serious(-EFAULT);
3656         } else {
3657                 /* No intent was provided */
3658                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3659                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3660                 rc = req_capsule_server_pack(pill);
3661                 if (rc)
3662                         rc = err_serious(rc);
3663         }
3664         mdt_thread_info_fini(info);
3665         RETURN(rc);
3666 }
3667
3668 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3669 {
3670         struct seq_server_site  *ss = mdt_seq_site(mdt);
3671
3672         if (ss->ss_node_id == 0)
3673                 return;
3674
3675         if (ss->ss_client_seq != NULL) {
3676                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3677                 ss->ss_client_seq->lcs_exp = NULL;
3678         }
3679
3680         if (ss->ss_server_fld != NULL) {
3681                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3682                 ss->ss_server_fld->lsf_control_exp = NULL;
3683         }
3684 }
3685
3686 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3687 {
3688         struct seq_server_site *ss = mdt_seq_site(mdt);
3689
3690         if (ss == NULL)
3691                 return;
3692
3693         if (ss->ss_server_seq != NULL)
3694                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3695 }
3696
3697 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3698 {
3699         mdt_seq_fini_cli(mdt);
3700         mdt_deregister_seq_exp(mdt);
3701
3702         return seq_site_fini(env, mdt_seq_site(mdt));
3703 }
3704
3705 /**
3706  * It will retrieve its FLDB entries from MDT0, and it only happens
3707  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3708  * and it needs to refresh FLDB from the MDT0.
3709  **/
3710 static int mdt_register_lwp_callback(void *data)
3711 {
3712         struct lu_env           env;
3713         struct mdt_device       *mdt = data;
3714         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3715         int                     rc;
3716         ENTRY;
3717
3718         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3719
3720         rc = lu_env_init(&env, LCT_MD_THREAD);
3721         if (rc < 0) {
3722                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3723                 RETURN(rc);
3724         }
3725
3726         /* Allocate new sequence now to avoid creating local transaction
3727          * in the normal transaction process */
3728         rc = seq_server_check_and_alloc_super(&env,
3729                                               mdt_seq_site(mdt)->ss_server_seq);
3730         if (rc < 0)
3731                 GOTO(out, rc);
3732
3733         if (fld->lsf_new) {
3734                 rc = fld_update_from_controller(&env, fld);
3735                 if (rc != 0) {
3736                         CERROR("%s: cannot update controller: rc = %d\n",
3737                                mdt_obd_name(mdt), rc);
3738                         GOTO(out, rc);
3739                 }
3740         }
3741 out:
3742         lu_env_fini(&env);
3743         RETURN(rc);
3744 }
3745
3746 static int mdt_register_seq_exp(struct mdt_device *mdt)
3747 {
3748         struct seq_server_site  *ss = mdt_seq_site(mdt);
3749         char                    *lwp_name = NULL;
3750         int                     rc;
3751
3752         if (ss->ss_node_id == 0)
3753                 return 0;
3754
3755         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3756         if (lwp_name == NULL)
3757                 GOTO(out_free, rc = -ENOMEM);
3758
3759         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3760         if (rc != 0)
3761                 GOTO(out_free, rc);
3762
3763         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3764                                       NULL, NULL);
3765         if (rc != 0)
3766                 GOTO(out_free, rc);
3767
3768         rc = lustre_register_lwp_item(lwp_name,
3769                                       &ss->ss_server_fld->lsf_control_exp,
3770                                       mdt_register_lwp_callback, mdt);
3771         if (rc != 0) {
3772                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3773                 ss->ss_client_seq->lcs_exp = NULL;
3774                 GOTO(out_free, rc);
3775         }
3776 out_free:
3777         if (lwp_name != NULL)
3778                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3779
3780         return rc;
3781 }
3782
3783 /*
3784  * Init client sequence manager which is used by local MDS to talk to sequence
3785  * controller on remote node.
3786  */
3787 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3788 {
3789         struct seq_server_site  *ss = mdt_seq_site(mdt);
3790         int                     rc;
3791         char                    *prefix;
3792         ENTRY;
3793
3794         /* check if this is adding the first MDC and controller is not yet
3795          * initialized. */
3796         OBD_ALLOC_PTR(ss->ss_client_seq);
3797         if (ss->ss_client_seq == NULL)
3798                 RETURN(-ENOMEM);
3799
3800         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3801         if (prefix == NULL) {
3802                 OBD_FREE_PTR(ss->ss_client_seq);
3803                 ss->ss_client_seq = NULL;
3804                 RETURN(-ENOMEM);
3805         }
3806
3807         /* Note: seq_client_fini will be called in seq_site_fini */
3808         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3809         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3810                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3811                                                             NULL);
3812         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3813         if (rc != 0) {
3814                 OBD_FREE_PTR(ss->ss_client_seq);
3815                 ss->ss_client_seq = NULL;
3816                 RETURN(rc);
3817         }
3818
3819         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3820
3821         RETURN(rc);
3822 }
3823
3824 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3825 {
3826         struct seq_server_site  *ss;
3827         int                     rc;
3828         ENTRY;
3829
3830         ss = mdt_seq_site(mdt);
3831         /* init sequence controller server(MDT0) */
3832         if (ss->ss_node_id == 0) {
3833                 OBD_ALLOC_PTR(ss->ss_control_seq);
3834                 if (ss->ss_control_seq == NULL)
3835                         RETURN(-ENOMEM);
3836
3837                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3838                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3839                                      ss);
3840                 if (rc)
3841                         GOTO(out_seq_fini, rc);
3842         }
3843
3844         /* Init normal sequence server */
3845         OBD_ALLOC_PTR(ss->ss_server_seq);
3846         if (ss->ss_server_seq == NULL)
3847                 GOTO(out_seq_fini, rc = -ENOMEM);
3848
3849         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3850                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3851         if (rc)
3852                 GOTO(out_seq_fini, rc);
3853
3854         /* init seq client for seq server to talk to seq controller(MDT0) */
3855         rc = mdt_seq_init_cli(env, mdt);
3856         if (rc != 0)
3857                 GOTO(out_seq_fini, rc);
3858
3859         if (ss->ss_node_id != 0)
3860                 /* register controller export through lwp */
3861                 rc = mdt_register_seq_exp(mdt);
3862
3863         EXIT;
3864 out_seq_fini:
3865         if (rc)
3866                 mdt_seq_fini(env, mdt);
3867
3868         return rc;
3869 }
3870
3871 /*
3872  * FLD wrappers
3873  */
3874 static int mdt_fld_fini(const struct lu_env *env,
3875                         struct mdt_device *m)
3876 {
3877         struct seq_server_site *ss = mdt_seq_site(m);
3878         ENTRY;
3879
3880         if (ss && ss->ss_server_fld) {
3881                 fld_server_fini(env, ss->ss_server_fld);
3882                 OBD_FREE_PTR(ss->ss_server_fld);
3883                 ss->ss_server_fld = NULL;
3884         }
3885
3886         RETURN(0);
3887 }
3888
3889 static int mdt_fld_init(const struct lu_env *env,
3890                         const char *uuid,
3891                         struct mdt_device *m)
3892 {
3893         struct seq_server_site *ss;
3894         int rc;
3895         ENTRY;
3896
3897         ss = mdt_seq_site(m);
3898
3899         OBD_ALLOC_PTR(ss->ss_server_fld);
3900         if (ss->ss_server_fld == NULL)
3901                 RETURN(rc = -ENOMEM);
3902
3903         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3904                              LU_SEQ_RANGE_MDT);
3905         if (rc) {
3906                 OBD_FREE_PTR(ss->ss_server_fld);
3907                 ss->ss_server_fld = NULL;
3908                 RETURN(rc);
3909         }
3910
3911         RETURN(0);
3912 }
3913
3914 static void mdt_stack_pre_fini(const struct lu_env *env,
3915                            struct mdt_device *m, struct lu_device *top)
3916 {
3917         struct lustre_cfg_bufs  *bufs;
3918         struct lustre_cfg       *lcfg;
3919         struct mdt_thread_info  *info;
3920         ENTRY;
3921
3922         LASSERT(top);
3923
3924         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3925         LASSERT(info != NULL);
3926
3927         bufs = &info->mti_u.bufs;
3928
3929         LASSERT(m->mdt_child_exp);
3930         LASSERT(m->mdt_child_exp->exp_obd);
3931
3932         /* process cleanup, pass mdt obd name to get obd umount flags */
3933         /* XXX: this is needed because all layers are referenced by
3934          * objects (some of them are pinned by osd, for example *
3935          * the proper solution should be a model where object used
3936          * by osd only doesn't have mdt/mdd slices -bzzz */
3937         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3938         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3939         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3940         if (lcfg == NULL)
3941                 RETURN_EXIT;
3942
3943         top->ld_ops->ldo_process_config(env, top, lcfg);
3944         lustre_cfg_free(lcfg);
3945         EXIT;
3946 }
3947
3948 static void mdt_stack_fini(const struct lu_env *env,
3949                            struct mdt_device *m, struct lu_device *top)
3950 {
3951         struct obd_device       *obd = mdt2obd_dev(m);
3952         struct lustre_cfg_bufs  *bufs;
3953         struct lustre_cfg       *lcfg;
3954         struct mdt_thread_info  *info;
3955         char                     flags[3] = "";
3956         ENTRY;
3957
3958         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3959         LASSERT(info != NULL);
3960
3961         lu_dev_del_linkage(top->ld_site, top);
3962
3963         lu_site_purge(env, top->ld_site, -1);
3964
3965         bufs = &info->mti_u.bufs;
3966         /* process cleanup, pass mdt obd name to get obd umount flags */
3967         /* another purpose is to let all layers to release their objects */
3968         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3969         if (obd->obd_force)
3970                 strcat(flags, "F");
3971         if (obd->obd_fail)
3972                 strcat(flags, "A");
3973         lustre_cfg_bufs_set_string(bufs, 1, flags);
3974         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3975         if (lcfg == NULL)
3976                 RETURN_EXIT;
3977
3978         LASSERT(top);
3979         top->ld_ops->ldo_process_config(env, top, lcfg);
3980         lustre_cfg_free(lcfg);
3981
3982         lu_site_purge(env, top->ld_site, -1);
3983
3984         m->mdt_child = NULL;
3985         m->mdt_bottom = NULL;
3986
3987         obd_disconnect(m->mdt_child_exp);
3988         m->mdt_child_exp = NULL;
3989
3990         obd_disconnect(m->mdt_bottom_exp);
3991         m->mdt_child_exp = NULL;
3992 }
3993
3994 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3995                                const char *next, struct obd_export **exp)
3996 {
3997         struct obd_connect_data *data = NULL;
3998         struct obd_device       *obd;
3999         int                      rc;
4000         ENTRY;
4001
4002         OBD_ALLOC_PTR(data);
4003         if (data == NULL)
4004                 GOTO(out, rc = -ENOMEM);
4005
4006         obd = class_name2obd(next);
4007         if (obd == NULL) {
4008                 CERROR("%s: can't locate next device: %s\n",
4009                        mdt_obd_name(m), next);
4010                 GOTO(out, rc = -ENOTCONN);
4011         }
4012
4013         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4014         data->ocd_version = LUSTRE_VERSION_CODE;
4015
4016         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4017         if (rc) {
4018                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4019                        mdt_obd_name(m), next, rc);
4020                 GOTO(out, rc);
4021         }
4022
4023 out:
4024         if (data)
4025                 OBD_FREE_PTR(data);
4026         RETURN(rc);
4027 }
4028
4029 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4030                           struct lustre_cfg *cfg)
4031 {
4032         char                   *dev = lustre_cfg_string(cfg, 0);
4033         int                     rc, name_size, uuid_size;
4034         char                   *name, *uuid, *p;
4035         struct lustre_cfg_bufs *bufs;
4036         struct lustre_cfg      *lcfg;
4037         struct obd_device      *obd;
4038         struct lustre_profile  *lprof;
4039         struct lu_site         *site;
4040         ENTRY;
4041
4042         /* in 1.8 we had the only device in the stack - MDS.
4043          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4044          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4045          * to generate names and setup MDT, MDD. MDT will be using
4046          * generated name to connect to MDD. for MDD the next device
4047          * will be LOD with name taken from so called "profile" which
4048          * is generated by mount_option line
4049          *
4050          * 1.8 MGS generates config. commands like this:
4051          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4052          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4053          * 2.0 MGS generates config. commands like this:
4054          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4055          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4056          *                    3:lustre-MDT0000-mdtlov  4:f
4057          *
4058          * we generate MDD name from MDT one, just replacing T with D
4059          *
4060          * after all the preparations, the logical equivalent will be
4061          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4062          *                    3:lustre-MDT0000-mdtlov  4:f
4063          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4064          *                    3:lustre-MDD0000  4:f
4065          *
4066          *  notice we build the stack from down to top: MDD first, then MDT */
4067
4068         name_size = MAX_OBD_NAME;
4069         uuid_size = MAX_OBD_NAME;
4070
4071         OBD_ALLOC(name, name_size);
4072         OBD_ALLOC(uuid, uuid_size);
4073         if (name == NULL || uuid == NULL)
4074                 GOTO(cleanup_mem, rc = -ENOMEM);
4075
4076         OBD_ALLOC_PTR(bufs);
4077         if (!bufs)
4078                 GOTO(cleanup_mem, rc = -ENOMEM);
4079
4080         strcpy(name, dev);
4081         p = strstr(name, "-MDT");
4082         if (p == NULL)
4083                 GOTO(free_bufs, rc = -ENOMEM);
4084         p[3] = 'D';
4085
4086         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4087
4088         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4089         if (lprof == NULL || lprof->lp_dt == NULL) {
4090                 CERROR("can't find the profile: %s\n",
4091                        lustre_cfg_string(cfg, 0));
4092                 GOTO(free_bufs, rc = -EINVAL);
4093         }
4094
4095         lustre_cfg_bufs_reset(bufs, name);
4096         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4097         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4098         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4099
4100         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4101         if (lcfg == NULL)
4102                 GOTO(put_profile, rc = -ENOMEM);
4103
4104         rc = class_attach(lcfg);
4105         if (rc)
4106                 GOTO(lcfg_cleanup, rc);
4107
4108         obd = class_name2obd(name);
4109         if (!obd) {
4110                 CERROR("Can not find obd %s (%s in config)\n",
4111                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4112                 GOTO(lcfg_cleanup, rc = -EINVAL);
4113         }
4114
4115         lustre_cfg_free(lcfg);
4116
4117         lustre_cfg_bufs_reset(bufs, name);
4118         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4119         lustre_cfg_bufs_set_string(bufs, 2, dev);
4120         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4121
4122         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4123         if (lcfg == NULL)
4124                 GOTO(class_detach, rc = -ENOMEM);
4125
4126         rc = class_setup(obd, lcfg);
4127         if (rc)
4128                 GOTO(class_detach, rc);
4129
4130         /* connect to MDD we just setup */
4131         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4132         if (rc)
4133                 GOTO(class_detach, rc);
4134
4135         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4136         LASSERT(site);
4137         LASSERT(mdt_lu_site(mdt) == NULL);
4138         mdt->mdt_lu_dev.ld_site = site;
4139         site->ls_top_dev = &mdt->mdt_lu_dev;
4140         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4141
4142         /* now connect to bottom OSD */
4143         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4144         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4145         if (rc)
4146                 GOTO(class_detach, rc);
4147         mdt->mdt_bottom =
4148                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4149
4150         rc = lu_env_refill((struct lu_env *)env);
4151         if (rc != 0)
4152                 CERROR("Failure to refill session: '%d'\n", rc);
4153
4154         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4155
4156         EXIT;
4157 class_detach:
4158         if (rc)
4159                 class_detach(obd, lcfg);
4160 lcfg_cleanup:
4161         lustre_cfg_free(lcfg);
4162 put_profile:
4163         class_put_profile(lprof);
4164 free_bufs:
4165         OBD_FREE_PTR(bufs);
4166 cleanup_mem:
4167         if (name)
4168                 OBD_FREE(name, name_size);
4169         if (uuid)
4170                 OBD_FREE(uuid, uuid_size);
4171         RETURN(rc);
4172 }
4173
4174 /* setup quota master target on MDT0 */
4175 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4176                           struct lustre_cfg *cfg)
4177 {
4178         struct obd_device       *obd;
4179         char                    *dev = lustre_cfg_string(cfg, 0);
4180         char                    *qmtname, *uuid, *p;
4181         struct lustre_cfg_bufs  *bufs;
4182         struct lustre_cfg       *lcfg;
4183         struct lustre_profile   *lprof;
4184         struct obd_connect_data *data;
4185         int                      rc;
4186         ENTRY;
4187
4188         LASSERT(mdt->mdt_qmt_exp == NULL);
4189         LASSERT(mdt->mdt_qmt_dev == NULL);
4190
4191         /* quota master is on MDT0 only for now */
4192         if (mdt->mdt_seq_site.ss_node_id != 0)
4193                 RETURN(0);
4194
4195         /* MGS generates config commands which look as follows:
4196          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4197          *                    3:lustre-MDT0000-mdtlov  4:f
4198          *
4199          * We generate the QMT name from the MDT one, just replacing MD with QM
4200          * after all the preparations, the logical equivalent will be:
4201          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4202          *                    3:lustre-MDT0000-osd  4:f */
4203         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4204         OBD_ALLOC(uuid, UUID_MAX);
4205         OBD_ALLOC_PTR(bufs);
4206         OBD_ALLOC_PTR(data);
4207         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4208                 GOTO(cleanup_mem, rc = -ENOMEM);
4209
4210         strcpy(qmtname, dev);
4211         p = strstr(qmtname, "-MDT");
4212         if (p == NULL)
4213                 GOTO(cleanup_mem, rc = -ENOMEM);
4214         /* replace MD with QM */
4215         p[1] = 'Q';
4216         p[2] = 'M';
4217
4218         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4219
4220         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4221         if (lprof == NULL || lprof->lp_dt == NULL) {
4222                 CERROR("can't find profile for %s\n",
4223                        lustre_cfg_string(cfg, 0));
4224                 GOTO(cleanup_mem, rc = -EINVAL);
4225         }
4226
4227         lustre_cfg_bufs_reset(bufs, qmtname);
4228         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4229         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4230         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4231
4232         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4233         if (lcfg == NULL)
4234                 GOTO(put_profile, rc = -ENOMEM);
4235
4236         rc = class_attach(lcfg);
4237         if (rc)
4238                 GOTO(lcfg_cleanup, rc);
4239
4240         obd = class_name2obd(qmtname);
4241         if (!obd) {
4242                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4243                        lustre_cfg_string(cfg, 0));
4244                 GOTO(lcfg_cleanup, rc = -EINVAL);
4245         }
4246
4247         lustre_cfg_free(lcfg);
4248
4249         lustre_cfg_bufs_reset(bufs, qmtname);
4250         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4251         lustre_cfg_bufs_set_string(bufs, 2, dev);
4252
4253         /* for quota, the next device should be the OSD device */
4254         lustre_cfg_bufs_set_string(bufs, 3,
4255                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4256
4257         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4258         if (lcfg == NULL)
4259                 GOTO(class_detach, rc = -ENOMEM);
4260
4261         rc = class_setup(obd, lcfg);
4262         if (rc)
4263                 GOTO(class_detach, rc);
4264
4265         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4266
4267         /* configure local quota objects */
4268         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4269                                                    &mdt->mdt_lu_dev,
4270                                                    mdt->mdt_qmt_dev);
4271         if (rc)
4272                 GOTO(class_cleanup, rc);
4273
4274         /* connect to quota master target */
4275         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4276         data->ocd_version = LUSTRE_VERSION_CODE;
4277         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4278                          data, NULL);
4279         if (rc) {
4280                 CERROR("cannot connect to quota master device %s (%d)\n",
4281                        qmtname, rc);
4282                 GOTO(class_cleanup, rc);
4283         }
4284
4285         EXIT;
4286 class_cleanup:
4287         if (rc) {
4288                 class_manual_cleanup(obd);
4289                 mdt->mdt_qmt_dev = NULL;
4290         }
4291 class_detach:
4292         if (rc)
4293                 class_detach(obd, lcfg);
4294 lcfg_cleanup:
4295         lustre_cfg_free(lcfg);
4296 put_profile:
4297         class_put_profile(lprof);
4298 cleanup_mem:
4299         if (bufs)
4300                 OBD_FREE_PTR(bufs);
4301         if (qmtname)
4302                 OBD_FREE(qmtname, MAX_OBD_NAME);
4303         if (uuid)
4304                 OBD_FREE(uuid, UUID_MAX);
4305         if (data)
4306                 OBD_FREE_PTR(data);
4307         return rc;
4308 }
4309
4310 /* Shutdown quota master target associated with mdt */
4311 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4312 {
4313         ENTRY;
4314
4315         if (mdt->mdt_qmt_exp == NULL)
4316                 RETURN_EXIT;
4317         LASSERT(mdt->mdt_qmt_dev != NULL);
4318
4319         /* the qmt automatically shuts down when the mdt disconnects */
4320         obd_disconnect(mdt->mdt_qmt_exp);
4321         mdt->mdt_qmt_exp = NULL;
4322         mdt->mdt_qmt_dev = NULL;
4323         EXIT;
4324 }
4325
4326 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4327  * for now. This will be removed along with converting rest of MDT code
4328  * to use tgt_session_info */
4329 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4330 {
4331         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4332         int                      rc;
4333
4334         rc = mdt_getxattr(info);
4335
4336         mdt_thread_info_fini(info);
4337         return rc;
4338 }
4339
4340 static struct tgt_handler mdt_tgt_handlers[] = {
4341 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4342                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4343                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4344 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4345                 0,                      MDS_DISCONNECT, tgt_disconnect,
4346                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4347 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4348                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4349                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4350 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4351 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4352 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4353 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4354                                                         mdt_getattr_name),
4355 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4356 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4357 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4358 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4359 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4360 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4361 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4362 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4363                                                         mdt_hsm_progress),
4364 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4365                                                         mdt_hsm_ct_register),
4366 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4367                                                         mdt_hsm_ct_unregister),
4368 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4369                                                         mdt_hsm_state_get),
4370 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4371                                                         mdt_hsm_state_set),
4372 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4373 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4374                                                         mdt_hsm_request),
4375 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4376             MDS_SWAP_LAYOUTS,
4377             mdt_swap_layouts),
4378 };
4379
4380 static struct tgt_handler mdt_sec_ctx_ops[] = {
4381 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4382 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4383 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4384 };
4385
4386 static struct tgt_handler mdt_quota_ops[] = {
4387 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4388 };
4389
4390 static struct tgt_opc_slice mdt_common_slice[] = {
4391         {
4392                 .tos_opc_start  = MDS_FIRST_OPC,
4393                 .tos_opc_end    = MDS_LAST_OPC,
4394                 .tos_hs         = mdt_tgt_handlers
4395         },
4396         {
4397                 .tos_opc_start  = OBD_FIRST_OPC,
4398                 .tos_opc_end    = OBD_LAST_OPC,
4399                 .tos_hs         = tgt_obd_handlers
4400         },
4401         {
4402                 .tos_opc_start  = LDLM_FIRST_OPC,
4403                 .tos_opc_end    = LDLM_LAST_OPC,
4404                 .tos_hs         = tgt_dlm_handlers
4405         },
4406         {
4407                 .tos_opc_start  = SEC_FIRST_OPC,
4408                 .tos_opc_end    = SEC_LAST_OPC,
4409                 .tos_hs         = mdt_sec_ctx_ops
4410         },
4411         {
4412                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4413                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4414                 .tos_hs         = tgt_out_handlers
4415         },
4416         {
4417                 .tos_opc_start  = FLD_FIRST_OPC,
4418                 .tos_opc_end    = FLD_LAST_OPC,
4419                 .tos_hs         = fld_handlers
4420         },
4421         {
4422                 .tos_opc_start  = SEQ_FIRST_OPC,
4423                 .tos_opc_end    = SEQ_LAST_OPC,
4424                 .tos_hs         = seq_handlers
4425         },
4426         {
4427                 .tos_opc_start  = QUOTA_DQACQ,
4428                 .tos_opc_end    = QUOTA_LAST_OPC,
4429                 .tos_hs         = mdt_quota_ops
4430         },
4431         {
4432                 .tos_opc_start  = LLOG_FIRST_OPC,
4433                 .tos_opc_end    = LLOG_LAST_OPC,
4434                 .tos_hs         = tgt_llog_handlers
4435         },
4436         {
4437                 .tos_opc_start  = LFSCK_FIRST_OPC,
4438                 .tos_opc_end    = LFSCK_LAST_OPC,
4439                 .tos_hs         = tgt_lfsck_handlers
4440         },
4441
4442         {
4443                 .tos_hs         = NULL
4444         }
4445 };
4446
4447 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4448 {
4449         struct md_device        *next = m->mdt_child;
4450         struct lu_device        *d    = &m->mdt_lu_dev;
4451         struct obd_device       *obd  = mdt2obd_dev(m);
4452         struct lfsck_stop        stop;
4453         ENTRY;
4454
4455         stop.ls_status = LS_PAUSED;
4456         stop.ls_flags = 0;
4457         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4458
4459         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4460         target_recovery_fini(obd);
4461         ping_evictor_stop();
4462
4463         if (m->mdt_opts.mo_coordinator)
4464                 mdt_hsm_cdt_stop(m);
4465
4466         mdt_hsm_cdt_fini(m);
4467
4468         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4469         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4470
4471         if (m->mdt_namespace != NULL)
4472                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4473                                           d->ld_obd->obd_force);
4474
4475         obd_exports_barrier(obd);
4476         obd_zombie_barrier();
4477
4478         mdt_procfs_fini(m);
4479
4480         tgt_fini(env, &m->mdt_lut);
4481         mdt_fs_cleanup(env, m);
4482         upcall_cache_cleanup(m->mdt_identity_cache);
4483         m->mdt_identity_cache = NULL;
4484
4485         if (m->mdt_namespace != NULL) {
4486                 ldlm_namespace_free_post(m->mdt_namespace);
4487                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4488         }
4489
4490         mdt_quota_fini(env, m);
4491
4492         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4493
4494         mdt_seq_fini(env, m);
4495         mdt_fld_fini(env, m);
4496
4497         /*
4498          * Finish the stack
4499          */
4500         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4501
4502         LASSERT(atomic_read(&d->ld_ref) == 0);
4503
4504         server_put_mount(mdt_obd_name(m), true);
4505
4506         EXIT;
4507 }
4508
4509 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4510
4511 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4512                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4513 {
4514         struct mdt_thread_info    *info;
4515         struct obd_device         *obd;
4516         const char                *dev = lustre_cfg_string(cfg, 0);
4517         const char                *num = lustre_cfg_string(cfg, 2);
4518         struct lustre_mount_info  *lmi = NULL;
4519         struct lustre_sb_info     *lsi;
4520         struct lu_site            *s;
4521         struct seq_server_site    *ss_site;
4522         const char                *identity_upcall = "NONE";
4523         struct md_device          *next;
4524         int                        rc;
4525         long                       node_id;
4526         mntopt_t                   mntopts;
4527         ENTRY;
4528
4529         lu_device_init(&m->mdt_lu_dev, ldt);
4530         /*
4531          * Environment (env) might be missing mdt_thread_key values at that
4532          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4533          * mode.
4534          *
4535          * Usually device allocation path doesn't use module key values, but
4536          * mdt has to do a lot of work here, so allocate key value.
4537          */
4538         rc = lu_env_refill((struct lu_env *)env);
4539         if (rc != 0)
4540                 RETURN(rc);
4541
4542         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4543         LASSERT(info != NULL);
4544
4545         obd = class_name2obd(dev);
4546         LASSERT(obd != NULL);
4547
4548         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4549         m->mdt_opts.mo_evict_tgt_nids = 1;
4550         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4551
4552         /* default is coordinator off, it is started through conf_param
4553          * or /proc */
4554         m->mdt_opts.mo_coordinator = 0;
4555
4556         lmi = server_get_mount(dev);
4557         if (lmi == NULL) {
4558                 CERROR("Cannot get mount info for %s!\n", dev);
4559                 RETURN(-EFAULT);
4560         } else {
4561                 lsi = s2lsi(lmi->lmi_sb);
4562                 /* CMD is supported only in IAM mode */
4563                 LASSERT(num);
4564                 node_id = simple_strtol(num, NULL, 10);
4565                 obd->u.obt.obt_magic = OBT_MAGIC;
4566                 if (lsi->lsi_lmd != NULL &&
4567                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4568                         m->mdt_skip_lfsck = 1;
4569         }
4570
4571         m->mdt_squash.rsi_uid = 0;
4572         m->mdt_squash.rsi_gid = 0;
4573         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4574         init_rwsem(&m->mdt_squash.rsi_sem);
4575         spin_lock_init(&m->mdt_osfs_lock);
4576         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4577         m->mdt_enable_remote_dir = 0;
4578         m->mdt_enable_remote_dir_gid = 0;
4579
4580         atomic_set(&m->mdt_mds_mds_conns, 0);
4581         atomic_set(&m->mdt_async_commit_count, 0);
4582
4583         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4584         m->mdt_lu_dev.ld_obd = obd;
4585         /* Set this lu_device to obd for error handling purposes. */
4586         obd->obd_lu_dev = &m->mdt_lu_dev;
4587
4588         /* init the stack */
4589         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4590         if (rc) {
4591                 CERROR("%s: Can't init device stack, rc %d\n",
4592                        mdt_obd_name(m), rc);
4593                 GOTO(err_lmi, rc);
4594         }
4595
4596         s = mdt_lu_site(m);
4597         ss_site = mdt_seq_site(m);
4598         s->ld_seq_site = ss_site;
4599         ss_site->ss_lu = s;
4600
4601         /* set server index */
4602         ss_site->ss_node_id = node_id;
4603
4604         /* failover is the default
4605          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4606          * assumed whose ss_node_id == 0 XXX
4607          * */
4608         obd->obd_replayable = 1;
4609         /* No connection accepted until configurations will finish */
4610         obd->obd_no_conn = 1;
4611
4612         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4613                 char *str = lustre_cfg_string(cfg, 4);
4614                 if (strchr(str, 'n')) {
4615                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4616                         obd->obd_replayable = 0;
4617                 }
4618         }
4619
4620         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4621         if (rc)
4622                 GOTO(err_fini_stack, rc);
4623
4624         rc = mdt_seq_init(env, m);
4625         if (rc)
4626                 GOTO(err_fini_fld, rc);
4627
4628         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4629                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4630         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4631                                               LDLM_NAMESPACE_SERVER,
4632                                               LDLM_NAMESPACE_GREEDY,
4633                                               LDLM_NS_TYPE_MDT);
4634         if (m->mdt_namespace == NULL)
4635                 GOTO(err_fini_seq, rc = -ENOMEM);
4636
4637         m->mdt_namespace->ns_lvbp = m;
4638         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4639
4640         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4641         /* set obd_namespace for compatibility with old code */
4642         obd->obd_namespace = m->mdt_namespace;
4643
4644         rc = mdt_hsm_cdt_init(m);
4645         if (rc != 0) {
4646                 CERROR("%s: error initializing coordinator, rc %d\n",
4647                        mdt_obd_name(m), rc);
4648                 GOTO(err_free_ns, rc);
4649         }
4650
4651         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4652                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4653                       OBD_FAIL_MDS_ALL_REPLY_NET);
4654         if (rc)
4655                 GOTO(err_free_hsm, rc);
4656
4657         rc = mdt_fs_setup(env, m, obd, lsi);
4658         if (rc)
4659                 GOTO(err_tgt, rc);
4660
4661         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4662
4663         next = m->mdt_child;
4664         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4665                                          &mntopts);
4666         if (rc)
4667                 GOTO(err_fs_cleanup, rc);
4668
4669         if (mntopts & MNTOPT_USERXATTR)
4670                 m->mdt_opts.mo_user_xattr = 1;
4671         else
4672                 m->mdt_opts.mo_user_xattr = 0;
4673
4674         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4675         if (rc)
4676                 GOTO(err_fs_cleanup, rc);
4677
4678         if (mntopts & MNTOPT_ACL)
4679                 m->mdt_opts.mo_acl = 1;
4680         else
4681                 m->mdt_opts.mo_acl = 0;
4682
4683         /* XXX: to support suppgid for ACL, we enable identity_upcall
4684          * by default, otherwise, maybe got unexpected -EACCESS. */
4685         if (m->mdt_opts.mo_acl)
4686                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4687
4688         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4689                                                 identity_upcall,
4690                                                 &mdt_identity_upcall_cache_ops);
4691         if (IS_ERR(m->mdt_identity_cache)) {
4692                 rc = PTR_ERR(m->mdt_identity_cache);
4693                 m->mdt_identity_cache = NULL;
4694                 GOTO(err_fs_cleanup, rc);
4695         }
4696
4697         rc = mdt_procfs_init(m, dev);
4698         if (rc) {
4699                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4700                 GOTO(err_recovery, rc);
4701         }
4702
4703         rc = mdt_quota_init(env, m, cfg);
4704         if (rc)
4705                 GOTO(err_procfs, rc);
4706
4707         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4708         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4709                            "mdt_ldlm_client", m->mdt_ldlm_client);
4710
4711         ping_evictor_start();
4712
4713         /* recovery will be started upon mdt_prepare()
4714          * when the whole stack is complete and ready
4715          * to serve the requests */
4716
4717         /* Reduce the initial timeout on an MDS because it doesn't need such
4718          * a long timeout as an OST does. Adaptive timeouts will adjust this
4719          * value appropriately. */
4720         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4721                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4722
4723         RETURN(0);
4724 err_procfs:
4725         mdt_procfs_fini(m);
4726 err_recovery:
4727         target_recovery_fini(obd);
4728         upcall_cache_cleanup(m->mdt_identity_cache);
4729         m->mdt_identity_cache = NULL;
4730 err_fs_cleanup:
4731         mdt_fs_cleanup(env, m);
4732 err_tgt:
4733         tgt_fini(env, &m->mdt_lut);
4734 err_free_hsm:
4735         mdt_hsm_cdt_fini(m);
4736 err_free_ns:
4737         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4738         obd->obd_namespace = m->mdt_namespace = NULL;
4739 err_fini_seq:
4740         mdt_seq_fini(env, m);
4741 err_fini_fld:
4742         mdt_fld_fini(env, m);
4743 err_fini_stack:
4744         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4745 err_lmi:
4746         if (lmi)
4747                 server_put_mount(dev, true);
4748         return(rc);
4749 }
4750
4751 /* For interoperability, the left element is old parameter, the right one
4752  * is the new version of the parameter, if some parameter is deprecated,
4753  * the new version should be set as NULL. */
4754 static struct cfg_interop_param mdt_interop_param[] = {
4755         { "mdt.group_upcall",   NULL },
4756         { "mdt.quota_type",     NULL },
4757         { "mdd.quota_type",     NULL },
4758         { "mdt.som",            NULL },
4759         { "mdt.rootsquash",     "mdt.root_squash" },
4760         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4761         { NULL }
4762 };
4763
4764 /* used by MGS to process specific configurations */
4765 static int mdt_process_config(const struct lu_env *env,
4766                               struct lu_device *d, struct lustre_cfg *cfg)
4767 {
4768         struct mdt_device *m = mdt_dev(d);
4769         struct md_device *md_next = m->mdt_child;
4770         struct lu_device *next = md2lu_dev(md_next);
4771         int rc;
4772         ENTRY;
4773
4774         switch (cfg->lcfg_command) {
4775         case LCFG_PARAM: {
4776                 struct obd_device          *obd = d->ld_obd;
4777
4778                 /* For interoperability */
4779                 struct cfg_interop_param   *ptr = NULL;
4780                 struct lustre_cfg          *old_cfg = NULL;
4781                 char                       *param = NULL;
4782
4783                 param = lustre_cfg_string(cfg, 1);
4784                 if (param == NULL) {
4785                         CERROR("param is empty\n");
4786                         rc = -EINVAL;
4787                         break;
4788                 }
4789
4790                 ptr = class_find_old_param(param, mdt_interop_param);
4791                 if (ptr != NULL) {
4792                         if (ptr->new_param == NULL) {
4793                                 rc = 0;
4794                                 CWARN("For interoperability, skip this %s."
4795                                       " It is obsolete.\n", ptr->old_param);
4796                                 break;
4797                         }
4798
4799                         CWARN("Found old param %s, changed it to %s.\n",
4800                               ptr->old_param, ptr->new_param);
4801
4802                         old_cfg = cfg;
4803                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4804                         if (IS_ERR(cfg)) {
4805                                 rc = PTR_ERR(cfg);
4806                                 break;
4807                         }
4808                 }
4809
4810                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4811                                               cfg, obd);
4812                 if (rc > 0 || rc == -ENOSYS) {
4813                         /* is it an HSM var ? */
4814                         rc = class_process_proc_param(PARAM_HSM,
4815                                                       hsm_cdt_get_proc_vars(),
4816                                                       cfg, obd);
4817                         if (rc > 0 || rc == -ENOSYS)
4818                                 /* we don't understand; pass it on */
4819                                 rc = next->ld_ops->ldo_process_config(env, next,
4820                                                                       cfg);
4821                 }
4822
4823                 if (old_cfg != NULL)
4824                         lustre_cfg_free(cfg);
4825
4826                 break;
4827         }
4828         default:
4829                 /* others are passed further */
4830                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4831                 break;
4832         }
4833         RETURN(rc);
4834 }
4835
4836 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4837                                           const struct lu_object_header *hdr,
4838                                           struct lu_device *d)
4839 {
4840         struct mdt_object *mo;
4841
4842         ENTRY;
4843
4844         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4845         if (mo != NULL) {
4846                 struct lu_object *o;
4847                 struct lu_object_header *h;
4848
4849                 o = &mo->mot_obj;
4850                 h = &mo->mot_header;
4851                 lu_object_header_init(h);
4852                 lu_object_init(o, h, d);
4853                 lu_object_add_top(h, o);
4854                 o->lo_ops = &mdt_obj_ops;
4855                 spin_lock_init(&mo->mot_write_lock);
4856                 mutex_init(&mo->mot_lov_mutex);
4857                 init_rwsem(&mo->mot_open_sem);
4858                 RETURN(o);
4859         }
4860         RETURN(NULL);
4861 }
4862
4863 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4864                            const struct lu_object_conf *unused)
4865 {
4866         struct mdt_device *d = mdt_dev(o->lo_dev);
4867         struct lu_device  *under;
4868         struct lu_object  *below;
4869         int                rc = 0;
4870         ENTRY;
4871
4872         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4873                PFID(lu_object_fid(o)));
4874
4875         under = &d->mdt_child->md_lu_dev;
4876         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4877         if (below != NULL) {
4878                 lu_object_add(o, below);
4879         } else
4880                 rc = -ENOMEM;
4881
4882         RETURN(rc);
4883 }
4884
4885 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4886 {
4887         struct mdt_object *mo = mdt_obj(o);
4888         struct lu_object_header *h;
4889         ENTRY;
4890
4891         h = o->lo_header;
4892         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4893                PFID(lu_object_fid(o)));
4894
4895         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4896         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4897
4898         lu_object_fini(o);
4899         lu_object_header_fini(h);
4900         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4901
4902         EXIT;
4903 }
4904
4905 static int mdt_object_print(const struct lu_env *env, void *cookie,
4906                             lu_printer_t p, const struct lu_object *o)
4907 {
4908         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4909
4910         return (*p)(env, cookie,
4911                     LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4912                     mdto, mdto->mot_flags, mdto->mot_write_count);
4913 }
4914
4915 static int mdt_prepare(const struct lu_env *env,
4916                 struct lu_device *pdev,
4917                 struct lu_device *cdev)
4918 {
4919         struct mdt_device *mdt = mdt_dev(cdev);
4920         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4921         struct obd_device *obd = cdev->ld_obd;
4922         int rc;
4923
4924         ENTRY;
4925
4926         LASSERT(obd);
4927
4928         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4929         if (rc)
4930                 RETURN(rc);
4931
4932         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4933         if (rc)
4934                 RETURN(rc);
4935
4936         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4937         if (rc)
4938                 RETURN(rc);
4939
4940         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4941         /* The LFSCK instance is registered just now, so it must be there when
4942          * register the namespace to such instance. */
4943         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4944
4945         if (mdt->mdt_seq_site.ss_node_id == 0) {
4946                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4947                                                          &mdt->mdt_md_root_fid);
4948                 if (rc)
4949                         RETURN(rc);
4950         }
4951
4952         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4953
4954         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4955         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4956         LASSERT(obd->obd_no_conn);
4957         spin_lock(&obd->obd_dev_lock);
4958         obd->obd_no_conn = 0;
4959         spin_unlock(&obd->obd_dev_lock);
4960
4961         if (obd->obd_recovering == 0)
4962                 mdt_postrecov(env, mdt);
4963
4964         RETURN(rc);
4965 }
4966
4967 const struct lu_device_operations mdt_lu_ops = {
4968         .ldo_object_alloc   = mdt_object_alloc,
4969         .ldo_process_config = mdt_process_config,
4970         .ldo_prepare        = mdt_prepare,
4971 };
4972
4973 static const struct lu_object_operations mdt_obj_ops = {
4974         .loo_object_init    = mdt_object_init,
4975         .loo_object_free    = mdt_object_free,
4976         .loo_object_print   = mdt_object_print
4977 };
4978
4979 static int mdt_obd_set_info_async(const struct lu_env *env,
4980                                   struct obd_export *exp,
4981                                   __u32 keylen, void *key,
4982                                   __u32 vallen, void *val,
4983                                   struct ptlrpc_request_set *set)
4984 {
4985         int rc;
4986
4987         ENTRY;
4988
4989         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4990                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4991                 RETURN(rc);
4992         }
4993
4994         RETURN(0);
4995 }
4996
4997 /**
4998  * Match client and server connection feature flags.
4999  *
5000  * Compute the compatibility flags for a connection request based on
5001  * features mutually supported by client and server.
5002  *
5003  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5004  * must not be updated here, otherwise a partially initialized value may
5005  * be exposed. After the connection request is successfully processed,
5006  * the top-level MDT connect request handler atomically updates the export
5007  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5008  * reply. \see mdt_connect().
5009  *
5010  * \param exp   the obd_export associated with this client/target pair
5011  * \param mdt   the target device for the connection
5012  * \param data  stores data for this connect request
5013  *
5014  * \retval 0       success
5015  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5016  * \retval -EBADE  client and server feature requirements are incompatible
5017  */
5018 static int mdt_connect_internal(struct obd_export *exp,
5019                                 struct mdt_device *mdt,
5020                                 struct obd_connect_data *data)
5021 {
5022         LASSERT(data != NULL);
5023
5024         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5025         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5026
5027         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5028             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5029                 CWARN("%s: client %s does not support ibits lock, either "
5030                       "very old or an invalid client: flags "LPX64"\n",
5031                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5032                       data->ocd_connect_flags);
5033                 return -EBADE;
5034         }
5035
5036         if (!mdt->mdt_opts.mo_acl)
5037                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5038
5039         if (!mdt->mdt_opts.mo_user_xattr)
5040                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5041
5042         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5043                 data->ocd_brw_size = min(data->ocd_brw_size,
5044                                          (__u32)MD_MAX_BRW_SIZE);
5045                 if (data->ocd_brw_size == 0) {
5046                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5047                                " ocd_version: %x ocd_grant: %d "
5048                                "ocd_index: %u ocd_brw_size is "
5049                                "unexpectedly zero, network data "
5050                                "corruption? Refusing connection of this"
5051                                " client\n",
5052                                mdt_obd_name(mdt),
5053                                exp->exp_client_uuid.uuid,
5054                                exp, data->ocd_connect_flags, data->ocd_version,
5055                                data->ocd_grant, data->ocd_index);
5056                         return -EPROTO;
5057                 }
5058         }
5059
5060         /* NB: Disregard the rule against updating
5061          * exp_connect_data.ocd_connect_flags in this case, since
5062          * tgt_client_new() needs to know if this is a lightweight
5063          * connection, and it is safe to expose this flag before
5064          * connection processing completes. */
5065         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5066                 spin_lock(&exp->exp_lock);
5067                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5068                 spin_unlock(&exp->exp_lock);
5069         }
5070
5071         data->ocd_version = LUSTRE_VERSION_CODE;
5072
5073         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5074                 CWARN("%s: MDS requires FID support, but client not\n",
5075                       mdt_obd_name(mdt));
5076                 return -EBADE;
5077         }
5078
5079         if (OCD_HAS_FLAG(data, PINGLESS)) {
5080                 if (ptlrpc_pinger_suppress_pings()) {
5081                         spin_lock(&exp->exp_obd->obd_dev_lock);
5082                         list_del_init(&exp->exp_obd_chain_timed);
5083                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5084                 } else {
5085                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5086                 }
5087         }
5088
5089         data->ocd_max_easize = mdt->mdt_max_ea_size;
5090
5091         /* NB: Disregard the rule against updating
5092          * exp_connect_data.ocd_connect_flags in this case, since
5093          * tgt_client_new() needs to know if this is client supports
5094          * multiple modify RPCs, and it is safe to expose this flag before
5095          * connection processing completes. */
5096         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5097                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5098                 spin_lock(&exp->exp_lock);
5099                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5100                 spin_unlock(&exp->exp_lock);
5101         }
5102
5103         return 0;
5104 }
5105
5106 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5107                                    struct mdt_thread_info *info,
5108                                    struct mdt_file_data *mfd)
5109 {
5110         struct lu_context ses;
5111         int rc;
5112         ENTRY;
5113
5114         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5115         if (rc)
5116                 RETURN(rc);
5117
5118         env->le_ses = &ses;
5119         lu_context_enter(&ses);
5120
5121         mdt_ucred(info)->uc_valid = UCRED_OLD;
5122         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5123
5124         lu_context_exit(&ses);
5125         lu_context_fini(&ses);
5126         env->le_ses = NULL;
5127
5128         RETURN(rc);
5129 }
5130
5131 static int mdt_export_cleanup(struct obd_export *exp)
5132 {
5133         struct list_head         closing_list;
5134         struct mdt_export_data  *med = &exp->exp_mdt_data;
5135         struct obd_device       *obd = exp->exp_obd;
5136         struct mdt_device       *mdt;
5137         struct mdt_thread_info  *info;
5138         struct lu_env            env;
5139         struct mdt_file_data    *mfd, *n;
5140         int rc = 0;
5141         ENTRY;
5142
5143         INIT_LIST_HEAD(&closing_list);
5144         spin_lock(&med->med_open_lock);
5145         while (!list_empty(&med->med_open_head)) {
5146                 struct list_head *tmp = med->med_open_head.next;
5147                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5148
5149                 /* Remove mfd handle so it can't be found again.
5150                  * We are consuming the mfd_list reference here. */
5151                 class_handle_unhash(&mfd->mfd_handle);
5152                 list_move_tail(&mfd->mfd_list, &closing_list);
5153         }
5154         spin_unlock(&med->med_open_lock);
5155         mdt = mdt_dev(obd->obd_lu_dev);
5156         LASSERT(mdt != NULL);
5157
5158         rc = lu_env_init(&env, LCT_MD_THREAD);
5159         if (rc)
5160                 RETURN(rc);
5161
5162         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5163         LASSERT(info != NULL);
5164         memset(info, 0, sizeof *info);
5165         info->mti_env = &env;
5166         info->mti_mdt = mdt;
5167         info->mti_exp = exp;
5168
5169         if (!list_empty(&closing_list)) {
5170                 struct md_attr *ma = &info->mti_attr;
5171
5172                 /* Close any open files (which may also cause orphan
5173                  * unlinking). */
5174                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5175                         list_del_init(&mfd->mfd_list);
5176                         ma->ma_need = ma->ma_valid = 0;
5177
5178                         /* This file is being closed due to an eviction, it
5179                          * could have been modified and now dirty regarding to
5180                          * HSM archive, check this!
5181                          * The logic here is to mark a file dirty if there's a
5182                          * chance it was dirtied before the client was evicted,
5183                          * so that we don't have to wait for a release attempt
5184                          * before finding out the file was actually dirty and
5185                          * fail the release. Aggressively marking it dirty here
5186                          * will cause the policy engine to attempt to
5187                          * re-archive it; when rearchiving, we can compare the
5188                          * current version to the HSM data_version and make the
5189                          * archive request into a noop if it's not actually
5190                          * dirty.
5191                          */
5192                         if (mfd->mfd_mode & FMODE_WRITE)
5193                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5194
5195                         /* Don't unlink orphan on failover umount, LU-184 */
5196                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5197                                 ma->ma_valid = MA_FLAGS;
5198                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5199                         }
5200                         mdt_mfd_close(info, mfd);
5201                 }
5202         }
5203         info->mti_mdt = NULL;
5204         /* cleanup client slot early */
5205         /* Do not erase record for recoverable client. */
5206         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5207                 tgt_client_del(&env, exp);
5208         lu_env_fini(&env);
5209
5210         RETURN(rc);
5211 }
5212
5213 static inline void mdt_enable_slc(struct mdt_device *mdt)
5214 {
5215         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5216                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5217 }
5218
5219 static inline void mdt_disable_slc(struct mdt_device *mdt)
5220 {
5221         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5222                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5223 }
5224
5225 static int mdt_obd_disconnect(struct obd_export *exp)
5226 {
5227         int rc;
5228         ENTRY;
5229
5230         LASSERT(exp);
5231         class_export_get(exp);
5232
5233         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5234             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5235                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5236
5237                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5238                         mdt_disable_slc(mdt);
5239         }
5240
5241         rc = server_disconnect_export(exp);
5242         if (rc != 0)
5243                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5244
5245         rc = mdt_export_cleanup(exp);
5246         nodemap_del_member(exp);
5247         class_export_put(exp);
5248         RETURN(rc);
5249 }
5250
5251 /* mds_connect copy */
5252 static int mdt_obd_connect(const struct lu_env *env,
5253                            struct obd_export **exp, struct obd_device *obd,
5254                            struct obd_uuid *cluuid,
5255                            struct obd_connect_data *data,
5256                            void *localdata)
5257 {
5258         struct obd_export       *lexp;
5259         struct lustre_handle    conn = { 0 };
5260         struct mdt_device       *mdt;
5261         int                      rc;
5262         lnet_nid_t              *client_nid = localdata;
5263         ENTRY;
5264
5265         LASSERT(env != NULL);
5266         if (!exp || !obd || !cluuid)
5267                 RETURN(-EINVAL);
5268
5269         mdt = mdt_dev(obd->obd_lu_dev);
5270
5271         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5272             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5273                 atomic_inc(&mdt->mdt_mds_mds_conns);
5274                 mdt_enable_slc(mdt);
5275         }
5276
5277         /*
5278          * first, check whether the stack is ready to handle requests
5279          * XXX: probably not very appropriate method is used now
5280          *      at some point we should find a better one
5281          */
5282         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5283             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5284             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5285                 rc = obd_get_info(env, mdt->mdt_child_exp,
5286                                   sizeof(KEY_OSP_CONNECTED),
5287                                   KEY_OSP_CONNECTED, NULL, NULL);
5288                 if (rc)
5289                         RETURN(-EAGAIN);
5290                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5291         }
5292
5293         rc = class_connect(&conn, obd, cluuid);
5294         if (rc)
5295                 RETURN(rc);
5296
5297         lexp = class_conn2export(&conn);
5298         LASSERT(lexp != NULL);
5299
5300         rc = nodemap_add_member(*client_nid, lexp);
5301         if (rc != 0 && rc != -EEXIST)
5302                 GOTO(out, rc);
5303
5304         rc = mdt_connect_internal(lexp, mdt, data);
5305         if (rc == 0) {
5306                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5307
5308                 LASSERT(lcd);
5309                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5310                 rc = tgt_client_new(env, lexp);
5311                 if (rc == 0)
5312                         mdt_export_stats_init(obd, lexp, localdata);
5313         }
5314 out:
5315         if (rc != 0) {
5316                 class_disconnect(lexp);
5317                 nodemap_del_member(lexp);
5318                 *exp = NULL;
5319         } else {
5320                 *exp = lexp;
5321                 /* Because we do not want this export to be evicted by pinger,
5322                  * let's not add this export to the timed chain list. */
5323                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5324                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5325                         list_del_init(&lexp->exp_obd_chain_timed);
5326                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5327                 }
5328         }
5329
5330         RETURN(rc);
5331 }
5332
5333 static int mdt_obd_reconnect(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         lnet_nid_t             *client_nid = localdata;
5340         int                     rc;
5341         ENTRY;
5342
5343         if (exp == NULL || obd == NULL || cluuid == NULL)
5344                 RETURN(-EINVAL);
5345
5346         rc = nodemap_add_member(*client_nid, exp);
5347         if (rc != 0 && rc != -EEXIST)
5348                 RETURN(rc);
5349
5350         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5351         if (rc == 0)
5352                 mdt_export_stats_init(obd, exp, localdata);
5353         else
5354                 nodemap_del_member(exp);
5355
5356         RETURN(rc);
5357 }
5358
5359 /* FIXME: Can we avoid using these two interfaces? */
5360 static int mdt_init_export(struct obd_export *exp)
5361 {
5362         struct mdt_export_data *med = &exp->exp_mdt_data;
5363         int                     rc;
5364         ENTRY;
5365
5366         INIT_LIST_HEAD(&med->med_open_head);
5367         spin_lock_init(&med->med_open_lock);
5368         mutex_init(&med->med_idmap_mutex);
5369         med->med_idmap = NULL;
5370         spin_lock(&exp->exp_lock);
5371         exp->exp_connecting = 1;
5372         spin_unlock(&exp->exp_lock);
5373
5374         /* self-export doesn't need client data and ldlm initialization */
5375         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5376                                      &exp->exp_client_uuid)))
5377                 RETURN(0);
5378
5379         rc = tgt_client_alloc(exp);
5380         if (rc)
5381                 GOTO(err, rc);
5382
5383         rc = ldlm_init_export(exp);
5384         if (rc)
5385                 GOTO(err_free, rc);
5386
5387         RETURN(rc);
5388
5389 err_free:
5390         tgt_client_free(exp);
5391 err:
5392         CERROR("%s: Failed to initialize export: rc = %d\n",
5393                exp->exp_obd->obd_name, rc);
5394         return rc;
5395 }
5396
5397 static int mdt_destroy_export(struct obd_export *exp)
5398 {
5399         ENTRY;
5400
5401         if (exp_connect_rmtclient(exp))
5402                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5403
5404         target_destroy_export(exp);
5405         /* destroy can be called from failed obd_setup, so
5406          * checking uuid is safer than obd_self_export */
5407         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5408                                      &exp->exp_client_uuid)))
5409                 RETURN(0);
5410
5411         ldlm_destroy_export(exp);
5412         tgt_client_free(exp);
5413
5414         LASSERT(list_empty(&exp->exp_outstanding_replies));
5415         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5416
5417         RETURN(0);
5418 }
5419
5420 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5421                    struct linkea_data *ldata)
5422 {
5423         int rc;
5424
5425         LASSERT(ldata->ld_buf->lb_buf != NULL);
5426
5427         if (!mdt_object_exists(mdt_obj))
5428                 return -ENODATA;
5429
5430         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5431                           ldata->ld_buf, XATTR_NAME_LINK);
5432         if (rc == -ERANGE) {
5433                 /* Buf was too small, figure out what we need. */
5434                 lu_buf_free(ldata->ld_buf);
5435                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5436                                   ldata->ld_buf, XATTR_NAME_LINK);
5437                 if (rc < 0)
5438                         return rc;
5439                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5440                 if (ldata->ld_buf->lb_buf == NULL)
5441                         return -ENOMEM;
5442                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5443                                   ldata->ld_buf, XATTR_NAME_LINK);
5444         }
5445         if (rc < 0)
5446                 return rc;
5447
5448         return linkea_init(ldata);
5449 }
5450
5451 /**
5452  * Given an MDT object, try to look up the full path to the object.
5453  * Part of the MDT layer implementation of lfs fid2path.
5454  *
5455  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5456  * \param[in]     obj   Object to do path lookup of
5457  * \param[in,out] fp    User-provided struct to store path information
5458  * \param[in]     root_fid Root FID of current path should reach
5459  *
5460  * \retval 0 Lookup successful, path information stored in fp
5461  * \retval -EAGAIN Lookup failed, usually because object is being moved
5462  * \retval negative errno if there was a problem
5463  */
5464 static int mdt_path_current(struct mdt_thread_info *info,
5465                             struct mdt_object *obj,
5466                             struct getinfo_fid2path *fp,
5467                             struct lu_fid *root_fid)
5468 {
5469         struct mdt_device       *mdt = info->mti_mdt;
5470         struct mdt_object       *mdt_obj;
5471         struct link_ea_header   *leh;
5472         struct link_ea_entry    *lee;
5473         struct lu_name          *tmpname = &info->mti_name;
5474         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5475         struct lu_buf           *buf = &info->mti_big_buf;
5476         char                    *ptr;
5477         int                     reclen;
5478         struct linkea_data      ldata = { NULL };
5479         int                     rc = 0;
5480         bool                    first = true;
5481         ENTRY;
5482
5483         /* temp buffer for path element, the buffer will be finally freed
5484          * in mdt_thread_info_fini */
5485         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5486         if (buf->lb_buf == NULL)
5487                 RETURN(-ENOMEM);
5488
5489         ldata.ld_buf = buf;
5490         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5491         *ptr = 0;
5492         --ptr;
5493         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5494
5495         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5496                 struct lu_buf           lmv_buf;
5497
5498                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5499                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5500                         GOTO(out, rc = -ENOENT);
5501
5502                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5503                 if (IS_ERR(mdt_obj))
5504                         GOTO(out, rc = PTR_ERR(mdt_obj));
5505
5506                 if (!mdt_object_exists(mdt_obj)) {
5507                         mdt_object_put(info->mti_env, mdt_obj);
5508                         GOTO(out, rc = -ENOENT);
5509                 }
5510
5511                 if (mdt_object_remote(mdt_obj)) {
5512                         mdt_object_put(info->mti_env, mdt_obj);
5513                         GOTO(remote_out, rc = -EREMOTE);
5514                 }
5515
5516                 rc = mdt_links_read(info, mdt_obj, &ldata);
5517                 if (rc != 0) {
5518                         mdt_object_put(info->mti_env, mdt_obj);
5519                         GOTO(out, rc);
5520                 }
5521
5522                 leh = buf->lb_buf;
5523                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5524                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5525                 /* If set, use link #linkno for path lookup, otherwise use
5526                    link #0.  Only do this for the final path element. */
5527                 if (first && fp->gf_linkno < leh->leh_reccount) {
5528                         int count;
5529                         for (count = 0; count < fp->gf_linkno; count++) {
5530                                 lee = (struct link_ea_entry *)
5531                                      ((char *)lee + reclen);
5532                                 linkea_entry_unpack(lee, &reclen, tmpname,
5533                                                     tmpfid);
5534                         }
5535                         if (fp->gf_linkno < leh->leh_reccount - 1)
5536                                 /* indicate to user there are more links */
5537                                 fp->gf_linkno++;
5538                 }
5539
5540                 lmv_buf.lb_buf = info->mti_xattr_buf;
5541                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5542                 /* Check if it is slave stripes */
5543                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5544                                   &lmv_buf, XATTR_NAME_LMV);
5545                 mdt_object_put(info->mti_env, mdt_obj);
5546                 if (rc > 0) {
5547                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5548
5549                         /* For slave stripes, get its master */
5550                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5551                                 fp->gf_fid = *tmpfid;
5552                                 continue;
5553                         }
5554                 } else if (rc < 0 && rc != -ENODATA) {
5555                         GOTO(out, rc);
5556                 }
5557
5558                 rc = 0;
5559
5560                 /* Pack the name in the end of the buffer */
5561                 ptr -= tmpname->ln_namelen;
5562                 if (ptr - 1 <= fp->gf_u.gf_path)
5563                         GOTO(out, rc = -EOVERFLOW);
5564                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5565                 *(--ptr) = '/';
5566
5567                 /* keep the last resolved fid to the client, so the
5568                  * client will build the left path on another MDT for
5569                  * remote object */
5570                 fp->gf_fid = *tmpfid;
5571
5572                 first = false;
5573         }
5574
5575 remote_out:
5576         ptr++; /* skip leading / */
5577         memmove(fp->gf_u.gf_path, ptr,
5578                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5579
5580 out:
5581         RETURN(rc);
5582 }
5583
5584 /**
5585  * Given an MDT object, use mdt_path_current to get the path.
5586  * Essentially a wrapper to retry mdt_path_current a set number of times
5587  * if -EAGAIN is returned (usually because an object is being moved).
5588  *
5589  * Part of the MDT layer implementation of lfs fid2path.
5590  *
5591  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5592  * \param[in]     obj   Object to do path lookup of
5593  * \param[in,out] fp    User-provided struct for arguments and to store path
5594  *                      information
5595  *
5596  * \retval 0 Lookup successful, path information stored in fp
5597  * \retval negative errno if there was a problem
5598  */
5599 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5600                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5601 {
5602         struct mdt_device       *mdt = info->mti_mdt;
5603         int                     tries = 3;
5604         int                     rc = -EAGAIN;
5605         ENTRY;
5606
5607         if (fp->gf_pathlen < 3)
5608                 RETURN(-EOVERFLOW);
5609
5610         if (root_fid == NULL)
5611                 root_fid = &mdt->mdt_md_root_fid;
5612
5613         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5614                 fp->gf_u.gf_path[0] = '\0';
5615                 RETURN(0);
5616         }
5617
5618         /* Retry multiple times in case file is being moved */
5619         while (tries-- && rc == -EAGAIN)
5620                 rc = mdt_path_current(info, obj, fp, root_fid);
5621
5622         RETURN(rc);
5623 }
5624
5625 /**
5626  * Get the full path of the provided FID, as of changelog record recno.
5627  *
5628  * This checks sanity and looks up object for user provided FID
5629  * before calling the actual path lookup code.
5630  *
5631  * Part of the MDT layer implementation of lfs fid2path.
5632  *
5633  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5634  * \param[in,out] fp    User-provided struct for arguments and to store path
5635  *                      information
5636  *
5637  * \retval 0 Lookup successful, path information and recno stored in fp
5638  * \retval -ENOENT, object does not exist
5639  * \retval negative errno if there was a problem
5640  */
5641 static int mdt_fid2path(struct mdt_thread_info *info,
5642                         struct lu_fid *root_fid,
5643                         struct getinfo_fid2path *fp)
5644 {
5645         struct mdt_device *mdt = info->mti_mdt;
5646         struct mdt_object *obj;
5647         int    rc;
5648         ENTRY;
5649
5650         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5651                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5652
5653         if (!fid_is_sane(&fp->gf_fid))
5654                 RETURN(-EINVAL);
5655
5656         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5657                 CWARN("%s: "DFID" is invalid, sequence should be "
5658                       ">= "LPX64"\n", mdt_obd_name(mdt),
5659                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5660                 RETURN(-EINVAL);
5661         }
5662
5663         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5664         if (IS_ERR(obj)) {
5665                 rc = PTR_ERR(obj);
5666                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5667                        PFID(&fp->gf_fid), rc);
5668                 RETURN(rc);
5669         }
5670
5671         if (mdt_object_remote(obj))
5672                 rc = -EREMOTE;
5673         else if (!mdt_object_exists(obj))
5674                 rc = -ENOENT;
5675         else
5676                 rc = 0;
5677
5678         if (rc < 0) {
5679                 mdt_object_put(info->mti_env, obj);
5680                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5681                        PFID(&fp->gf_fid), rc);
5682                 RETURN(rc);
5683         }
5684
5685         rc = mdt_path(info, obj, fp, root_fid);
5686
5687         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5688                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5689                fp->gf_recno, fp->gf_linkno);
5690
5691         mdt_object_put(info->mti_env, obj);
5692
5693         RETURN(rc);
5694 }
5695
5696 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5697                             void *val, int vallen)
5698 {
5699         struct getinfo_fid2path *fpout, *fpin;
5700         struct lu_fid *root_fid = NULL;
5701         int rc = 0;
5702
5703         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5704         fpout = val;
5705
5706         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5707                 lustre_swab_fid2path(fpin);
5708
5709         memcpy(fpout, fpin, sizeof(*fpin));
5710         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5711                 RETURN(-EINVAL);
5712
5713         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5714                       sizeof(struct lu_fid)) {
5715                 /* client sent its root FID, which is normally fileset FID */
5716                 root_fid = fpin->gf_u.gf_root_fid;
5717                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5718                         lustre_swab_lu_fid(root_fid);
5719
5720                 if (root_fid != NULL && !fid_is_sane(root_fid))
5721                         RETURN(-EINVAL);
5722         }
5723
5724         rc = mdt_fid2path(info, root_fid, fpout);
5725         RETURN(rc);
5726 }
5727
5728 int mdt_get_info(struct tgt_session_info *tsi)
5729 {
5730         char    *key;
5731         int      keylen;
5732         __u32   *vallen;
5733         void    *valout;
5734         int      rc;
5735
5736         ENTRY;
5737
5738         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5739         if (key == NULL) {
5740                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5741                 RETURN(err_serious(-EFAULT));
5742         }
5743         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5744                                       RCL_CLIENT);
5745
5746         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5747         if (vallen == NULL) {
5748                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5749                                 tgt_name(tsi->tsi_tgt));
5750                 RETURN(err_serious(-EFAULT));
5751         }
5752
5753         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5754                              *vallen);
5755         rc = req_capsule_server_pack(tsi->tsi_pill);
5756         if (rc)
5757                 RETURN(err_serious(rc));
5758
5759         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5760         if (valout == NULL) {
5761                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5762                                 tgt_name(tsi->tsi_tgt));
5763                 RETURN(err_serious(-EFAULT));
5764         }
5765
5766         if (KEY_IS(KEY_FID2PATH)) {
5767                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5768
5769                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5770                 mdt_thread_info_fini(info);
5771         } else {
5772                 rc = -EINVAL;
5773         }
5774         RETURN(rc);
5775 }
5776
5777 /* Pass the ioc down */
5778 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5779                          unsigned int cmd, int len, void *data)
5780 {
5781         struct lu_context ioctl_session;
5782         struct md_device *next = mdt->mdt_child;
5783         int rc;
5784         ENTRY;
5785
5786         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5787         if (rc)
5788                 RETURN(rc);
5789         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5790         lu_context_enter(&ioctl_session);
5791         env->le_ses = &ioctl_session;
5792
5793         LASSERT(next->md_ops->mdo_iocontrol);
5794         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5795
5796         lu_context_exit(&ioctl_session);
5797         lu_context_fini(&ioctl_session);
5798         RETURN(rc);
5799 }
5800
5801 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5802 {
5803         struct obd_ioctl_data *data = karg;
5804         struct lu_fid *fid;
5805         __u64 version;
5806         struct mdt_object *obj;
5807         struct mdt_lock_handle  *lh;
5808         int rc;
5809         ENTRY;
5810
5811         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5812             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5813                 RETURN(-EINVAL);
5814
5815         fid = (struct lu_fid *)data->ioc_inlbuf1;
5816
5817         if (!fid_is_sane(fid))
5818                 RETURN(-EINVAL);
5819
5820         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5821
5822         lh = &mti->mti_lh[MDT_LH_PARENT];
5823         mdt_lock_reg_init(lh, LCK_CR);
5824
5825         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5826         if (IS_ERR(obj))
5827                 RETURN(PTR_ERR(obj));
5828
5829         if (mdt_object_remote(obj)) {
5830                 rc = -EREMOTE;
5831                 /**
5832                  * before calling version get the correct MDS should be
5833                  * fid, this is error to find remote object here
5834                  */
5835                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5836         } else if (!mdt_object_exists(obj)) {
5837                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5838                 rc = -ENOENT;
5839         } else {
5840                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5841                *(__u64 *)data->ioc_inlbuf2 = version;
5842                 rc = 0;
5843         }
5844         mdt_object_unlock_put(mti, obj, lh, 1);
5845         RETURN(rc);
5846 }
5847
5848 /* ioctls on obd dev */
5849 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5850                          void *karg, void __user *uarg)
5851 {
5852         struct lu_env      env;
5853         struct obd_device *obd = exp->exp_obd;
5854         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5855         struct dt_device  *dt = mdt->mdt_bottom;
5856         int rc;
5857
5858         ENTRY;
5859         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5860         rc = lu_env_init(&env, LCT_MD_THREAD);
5861         if (rc)
5862                 RETURN(rc);
5863
5864         switch (cmd) {
5865         case OBD_IOC_SYNC:
5866                 rc = mdt_device_sync(&env, mdt);
5867                 break;
5868         case OBD_IOC_SET_READONLY:
5869                 rc = dt->dd_ops->dt_ro(&env, dt);
5870                 break;
5871         case OBD_IOC_ABORT_RECOVERY:
5872                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5873                 obd->obd_abort_recovery = 1;
5874                 target_stop_recovery_thread(obd);
5875                 rc = 0;
5876                 break;
5877         case OBD_IOC_CHANGELOG_REG:
5878         case OBD_IOC_CHANGELOG_DEREG:
5879         case OBD_IOC_CHANGELOG_CLEAR:
5880                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5881                 break;
5882         case OBD_IOC_START_LFSCK: {
5883                 struct md_device *next = mdt->mdt_child;
5884                 struct obd_ioctl_data *data = karg;
5885                 struct lfsck_start_param lsp;
5886
5887                 if (unlikely(data == NULL)) {
5888                         rc = -EINVAL;
5889                         break;
5890                 }
5891
5892                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5893                 lsp.lsp_index_valid = 0;
5894                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5895                 break;
5896         }
5897         case OBD_IOC_STOP_LFSCK: {
5898                 struct md_device        *next = mdt->mdt_child;
5899                 struct obd_ioctl_data   *data = karg;
5900                 struct lfsck_stop        stop;
5901
5902                 stop.ls_status = LS_STOPPED;
5903                 /* Old lfsck utils may pass NULL @stop. */
5904                 if (data->ioc_inlbuf1 == NULL)
5905                         stop.ls_flags = 0;
5906                 else
5907                         stop.ls_flags =
5908                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5909
5910                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5911                 break;
5912         }
5913         case OBD_IOC_QUERY_LFSCK: {
5914                 struct md_device        *next = mdt->mdt_child;
5915                 struct obd_ioctl_data   *data = karg;
5916
5917                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5918                                                  data->ioc_inlbuf1);
5919                 break;
5920         }
5921         case OBD_IOC_GET_OBJ_VERSION: {
5922                 struct mdt_thread_info *mti;
5923                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5924                 memset(mti, 0, sizeof *mti);
5925                 mti->mti_env = &env;
5926                 mti->mti_mdt = mdt;
5927                 mti->mti_exp = exp;
5928
5929                 rc = mdt_ioc_version_get(mti, karg);
5930                 break;
5931         }
5932         case OBD_IOC_CATLOGLIST: {
5933                 struct mdt_thread_info *mti;
5934
5935                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5936                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5937                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5938                                        &mti->mti_tmp_fid1);
5939                 break;
5940          }
5941         default:
5942                 rc = -EOPNOTSUPP;
5943                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5944                         mdt_obd_name(mdt), cmd, rc);
5945         }
5946
5947         lu_env_fini(&env);
5948         RETURN(rc);
5949 }
5950
5951 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5952 {
5953         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5954         int rc;
5955         ENTRY;
5956
5957         if (!mdt->mdt_skip_lfsck) {
5958                 struct lfsck_start_param lsp;
5959
5960                 lsp.lsp_start = NULL;
5961                 lsp.lsp_index_valid = 0;
5962                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5963                                                            OBD_IOC_START_LFSCK,
5964                                                            0, &lsp);
5965                 if (rc != 0 && rc != -EALREADY)
5966                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5967                               mdt_obd_name(mdt), rc);
5968         }
5969
5970         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5971         RETURN(rc);
5972 }
5973
5974 static int mdt_obd_postrecov(struct obd_device *obd)
5975 {
5976         struct lu_env env;
5977         int rc;
5978
5979         rc = lu_env_init(&env, LCT_MD_THREAD);
5980         if (rc)
5981                 RETURN(rc);
5982         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5983         lu_env_fini(&env);
5984         return rc;
5985 }
5986
5987 static struct obd_ops mdt_obd_device_ops = {
5988         .o_owner          = THIS_MODULE,
5989         .o_set_info_async = mdt_obd_set_info_async,
5990         .o_connect        = mdt_obd_connect,
5991         .o_reconnect      = mdt_obd_reconnect,
5992         .o_disconnect     = mdt_obd_disconnect,
5993         .o_init_export    = mdt_init_export,
5994         .o_destroy_export = mdt_destroy_export,
5995         .o_iocontrol      = mdt_iocontrol,
5996         .o_postrecov      = mdt_obd_postrecov,
5997 };
5998
5999 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6000                                          struct lu_device *d)
6001 {
6002         struct mdt_device *m = mdt_dev(d);
6003         ENTRY;
6004
6005         mdt_fini(env, m);
6006         RETURN(NULL);
6007 }
6008
6009 static struct lu_device *mdt_device_free(const struct lu_env *env,
6010                                          struct lu_device *d)
6011 {
6012         struct mdt_device *m = mdt_dev(d);
6013         ENTRY;
6014
6015         lu_device_fini(&m->mdt_lu_dev);
6016         OBD_FREE_PTR(m);
6017
6018         RETURN(NULL);
6019 }
6020
6021 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6022                                           struct lu_device_type *t,
6023                                           struct lustre_cfg *cfg)
6024 {
6025         struct lu_device  *l;
6026         struct mdt_device *m;
6027
6028         OBD_ALLOC_PTR(m);
6029         if (m != NULL) {
6030                 int rc;
6031
6032                 l = &m->mdt_lu_dev;
6033                 rc = mdt_init0(env, m, t, cfg);
6034                 if (rc != 0) {
6035                         mdt_device_free(env, l);
6036                         l = ERR_PTR(rc);
6037                         return l;
6038                 }
6039         } else
6040                 l = ERR_PTR(-ENOMEM);
6041         return l;
6042 }
6043
6044 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6045 LU_KEY_INIT(mdt, struct mdt_thread_info);
6046
6047 static void mdt_key_fini(const struct lu_context *ctx,
6048                          struct lu_context_key *key, void* data)
6049 {
6050         struct mdt_thread_info *info = data;
6051
6052         if (info->mti_big_lmm) {
6053                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6054                 info->mti_big_lmm = NULL;
6055                 info->mti_big_lmmsize = 0;
6056         }
6057         OBD_FREE_PTR(info);
6058 }
6059
6060 /* context key: mdt_thread_key */
6061 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6062
6063 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6064 {
6065         return lu_ucred(info->mti_env);
6066 }
6067
6068 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6069 {
6070         return lu_ucred_check(info->mti_env);
6071 }
6072
6073 /**
6074  * Enable/disable COS (Commit On Sharing).
6075  *
6076  * Set/Clear the COS flag in mdt options.
6077  *
6078  * \param mdt mdt device
6079  * \param val 0 disables COS, other values enable COS
6080  */
6081 void mdt_enable_cos(struct mdt_device *mdt, int val)
6082 {
6083         struct lu_env env;
6084         int rc;
6085
6086         mdt->mdt_opts.mo_cos = !!val;
6087         rc = lu_env_init(&env, LCT_LOCAL);
6088         if (unlikely(rc != 0)) {
6089                 CWARN("%s: lu_env initialization failed, cannot "
6090                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6091                 return;
6092         }
6093         mdt_device_sync(&env, mdt);
6094         lu_env_fini(&env);
6095 }
6096
6097 /**
6098  * Check COS (Commit On Sharing) status.
6099  *
6100  * Return COS flag status.
6101  *
6102  * \param mdt mdt device
6103  */
6104 int mdt_cos_is_enabled(struct mdt_device *mdt)
6105 {
6106         return mdt->mdt_opts.mo_cos != 0;
6107 }
6108
6109 static struct lu_device_type_operations mdt_device_type_ops = {
6110         .ldto_device_alloc = mdt_device_alloc,
6111         .ldto_device_free  = mdt_device_free,
6112         .ldto_device_fini  = mdt_device_fini
6113 };
6114
6115 static struct lu_device_type mdt_device_type = {
6116         .ldt_tags     = LU_DEVICE_MD,
6117         .ldt_name     = LUSTRE_MDT_NAME,
6118         .ldt_ops      = &mdt_device_type_ops,
6119         .ldt_ctx_tags = LCT_MD_THREAD
6120 };
6121
6122 static int __init mdt_init(void)
6123 {
6124         int rc;
6125
6126         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6127                  FID_NOBRACE_LEN + 1);
6128         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6129                  FID_LEN + 1);
6130         rc = lu_kmem_init(mdt_caches);
6131         if (rc)
6132                 return rc;
6133
6134         rc = mds_mod_init();
6135         if (rc)
6136                 GOTO(lu_fini, rc);
6137
6138         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6139                                  LUSTRE_MDT_NAME, &mdt_device_type);
6140         if (rc)
6141                 GOTO(mds_fini, rc);
6142 lu_fini:
6143         if (rc)
6144                 lu_kmem_fini(mdt_caches);
6145 mds_fini:
6146         if (rc)
6147                 mds_mod_exit();
6148         return rc;
6149 }
6150
6151 static void __exit mdt_exit(void)
6152 {
6153         class_unregister_type(LUSTRE_MDT_NAME);
6154         mds_mod_exit();
6155         lu_kmem_fini(mdt_caches);
6156 }
6157
6158 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6159 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6160 MODULE_VERSION(LUSTRE_VERSION_STRING);
6161 MODULE_LICENSE("GPL");
6162
6163 module_init(mdt_init);
6164 module_exit(mdt_exit);