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, 2013, 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  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 /* struct ptlrpc_request */
  57 #include <lustre_net.h>
  58 /* struct obd_export */
  59 #include <lustre_export.h>
  60 /* struct obd_device */
  61 #include <obd.h>
  62 /* lu2dt_dev() */
  63 #include <dt_object.h>
  64 #include <lustre_mds.h>
  65 #include <lustre_log.h>
  66 #include "mdt_internal.h"
  67 #include <lustre_acl.h>
  68 #include <lustre_param.h>
  69 #include <lustre_quota.h>
  70 #include <lustre_linkea.h>
  71 #include <lustre_lfsck.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return (rep->lock_policy_res1 & flag);
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, int flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~flag;
 142                 tgt_opdata_clear(info->mti_env, flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, int flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= flag;
 153                 tgt_opdata_set(info->mti_env, flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 176                                                   lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 int mdt_getstatus(struct tgt_session_info *tsi)
 267 {
 268         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 269         struct mdt_device       *mdt = info->mti_mdt;
 270         struct mdt_body         *repbody;
 271         int                      rc;
 272
 273         ENTRY;
 274
 275         rc = mdt_check_ucred(info);
 276         if (rc)
 277                 GOTO(out, rc = err_serious(rc));
 278
 279         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 280                 GOTO(out, rc = err_serious(-ENOMEM));
 281
 282         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 283         repbody->fid1 = mdt->mdt_md_root_fid;
 284         repbody->valid |= OBD_MD_FLID;
 285
 286         if (tsi->tsi_tgt->lut_mds_capa &&
 287             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 288                 struct mdt_object       *root;
 289                 struct lustre_capa      *capa;
 290
 291                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 292                 if (IS_ERR(root))
 293                         GOTO(out, rc = PTR_ERR(root));
 294
 295                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 296                 LASSERT(capa);
 297                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 298                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 299                                  0);
 300                 mdt_object_put(info->mti_env, root);
 301                 if (rc == 0)
 302                         repbody->valid |= OBD_MD_FLMDSCAPA;
 303         }
 304         EXIT;
 305 out:
 306         mdt_thread_info_fini(info);
 307         return rc;
 308 }
 309
 310 int mdt_statfs(struct tgt_session_info *tsi)
 311 {
 312         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 313         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 314         struct md_device                *next = info->mti_mdt->mdt_child;
 315         struct ptlrpc_service_part      *svcpt;
 316         struct obd_statfs               *osfs;
 317         int                             rc;
 318
 319         ENTRY;
 320
 321         svcpt = req->rq_rqbd->rqbd_svcpt;
 322
 323         /* This will trigger a watchdog timeout */
 324         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 325                          (MDT_SERVICE_WATCHDOG_FACTOR *
 326                           at_get(&svcpt->scp_at_estimate)) + 1);
 327
 328         rc = mdt_check_ucred(info);
 329         if (rc)
 330                 GOTO(out, rc = err_serious(rc));
 331
 332         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 333                 GOTO(out, rc = err_serious(-ENOMEM));
 334
 335         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 336         if (!osfs)
 337                 GOTO(out, rc = -EPROTO);
 338
 339         /** statfs information are cached in the mdt_device */
 340         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 341                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 342                 /** statfs data is too old, get up-to-date one */
 343                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 344                 if (rc)
 345                         GOTO(out, rc);
 346                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 347                 info->mti_mdt->mdt_osfs = *osfs;
 348                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 349                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 350         } else {
 351                 /** use cached statfs data */
 352                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 353                 *osfs = info->mti_mdt->mdt_osfs;
 354                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 355         }
 356
 357         if (rc == 0)
 358                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 359 out:
 360         mdt_thread_info_fini(info);
 361         RETURN(rc);
 362 }
 363
 364 /**
 365  * Pack SOM attributes into the reply.
 366  * Call under a DLM UPDATE lock.
 367  */
 368 static void mdt_pack_size2body(struct mdt_thread_info *info,
 369                                struct mdt_object *mo)
 370 {
 371         struct mdt_body *b;
 372         struct md_attr *ma = &info->mti_attr;
 373
 374         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 375         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 376
 377         /* Check if Size-on-MDS is supported, if this is a regular file,
 378          * if SOM is enabled on the object and if SOM cache exists and valid.
 379          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 380         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 381             !S_ISREG(ma->ma_attr.la_mode) ||
 382             !mdt_object_is_som_enabled(mo) ||
 383             !(ma->ma_valid & MA_SOM))
 384                 return;
 385
 386         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 387         b->size = ma->ma_som->msd_size;
 388         b->blocks = ma->ma_som->msd_blocks;
 389 }
 390
 391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 392                         const struct lu_attr *attr, const struct lu_fid *fid)
 393 {
 394         struct md_attr *ma = &info->mti_attr;
 395
 396         LASSERT(ma->ma_valid & MA_INODE);
 397
 398         b->atime      = attr->la_atime;
 399         b->mtime      = attr->la_mtime;
 400         b->ctime      = attr->la_ctime;
 401         b->mode       = attr->la_mode;
 402         b->size       = attr->la_size;
 403         b->blocks     = attr->la_blocks;
 404         b->uid        = attr->la_uid;
 405         b->gid        = attr->la_gid;
 406         b->flags      = attr->la_flags;
 407         b->nlink      = attr->la_nlink;
 408         b->rdev       = attr->la_rdev;
 409
 410         /*XXX should pack the reply body according to lu_valid*/
 411         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 412                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 413                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 414                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 415
 416         if (!S_ISREG(attr->la_mode)) {
 417                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 418         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 419                 /* means no objects are allocated on osts. */
 420                 LASSERT(!(ma->ma_valid & MA_LOV));
 421                 /* just ignore blocks occupied by extend attributes on MDS */
 422                 b->blocks = 0;
 423                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 424                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 425         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 426                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 427                 /* A released file stores its size on MDS. */
 428                 /* But return 1 block for released file, unless tools like tar
 429                  * will consider it fully sparse. (LU-3864)
 430                  */
 431                 if (unlikely(b->size == 0))
 432                         b->blocks = 0;
 433                 else
 434                         b->blocks = 1;
 435                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 436         }
 437
 438         if (fid) {
 439                 b->fid1 = *fid;
 440                 b->valid |= OBD_MD_FLID;
 441                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 442                                 PFID(fid), b->nlink, b->mode, b->size);
 443         }
 444
 445         if (info)
 446                 mdt_body_reverse_idmap(info, b);
 447
 448         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
 449                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 450                        PFID(fid), (unsigned long long)b->size);
 451 }
 452
 453 static inline int mdt_body_has_lov(const struct lu_attr *la,
 454                                    const struct mdt_body *body)
 455 {
 456         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 457                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 458 }
 459
 460 void mdt_client_compatibility(struct mdt_thread_info *info)
 461 {
 462         struct mdt_body       *body;
 463         struct ptlrpc_request *req = mdt_info_req(info);
 464         struct obd_export     *exp = req->rq_export;
 465         struct md_attr        *ma = &info->mti_attr;
 466         struct lu_attr        *la = &ma->ma_attr;
 467         ENTRY;
 468
 469         if (exp_connect_layout(exp))
 470                 /* the client can deal with 16-bit lmm_stripe_count */
 471                 RETURN_EXIT;
 472
 473         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 474
 475         if (!mdt_body_has_lov(la, body))
 476                 RETURN_EXIT;
 477
 478         /* now we have a reply with a lov for a client not compatible with the
 479          * layout lock so we have to clean the layout generation number */
 480         if (S_ISREG(la->la_mode))
 481                 ma->ma_lmm->lmm_layout_gen = 0;
 482         EXIT;
 483 }
 484
 485 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 486                              char *name)
 487 {
 488         const struct lu_env *env = info->mti_env;
 489         int rc;
 490         ENTRY;
 491
 492         LASSERT(info->mti_big_lmm_used == 0);
 493         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 494         if (rc < 0)
 495                 RETURN(rc);
 496
 497         /* big_lmm may need to be grown */
 498         if (info->mti_big_lmmsize < rc) {
 499                 int size = size_roundup_power2(rc);
 500
 501                 if (info->mti_big_lmmsize > 0) {
 502                         /* free old buffer */
 503                         LASSERT(info->mti_big_lmm);
 504                         OBD_FREE_LARGE(info->mti_big_lmm,
 505                                        info->mti_big_lmmsize);
 506                         info->mti_big_lmm = NULL;
 507                         info->mti_big_lmmsize = 0;
 508                 }
 509
 510                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 511                 if (info->mti_big_lmm == NULL)
 512                         RETURN(-ENOMEM);
 513                 info->mti_big_lmmsize = size;
 514         }
 515         LASSERT(info->mti_big_lmmsize >= rc);
 516
 517         info->mti_buf.lb_buf = info->mti_big_lmm;
 518         info->mti_buf.lb_len = info->mti_big_lmmsize;
 519         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 520
 521         RETURN(rc);
 522 }
 523
 524 int mdt_attr_get_lov(struct mdt_thread_info *info,
 525                      struct mdt_object *o, struct md_attr *ma)
 526 {
 527         struct md_object *next = mdt_object_child(o);
 528         struct lu_buf    *buf = &info->mti_buf;
 529         int rc;
 530
 531         buf->lb_buf = ma->ma_lmm;
 532         buf->lb_len = ma->ma_lmm_size;
 533         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 534
 535         if (rc > 0) {
 536                 ma->ma_lmm_size = rc;
 537                 ma->ma_valid |= MA_LOV;
 538                 rc = 0;
 539         } else if (rc == -ENODATA) {
 540                 /* no LOV EA */
 541                 rc = 0;
 542         } else if (rc == -ERANGE) {
 543                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 544                 if (rc > 0) {
 545                         info->mti_big_lmm_used = 1;
 546                         ma->ma_valid |= MA_LOV;
 547                         ma->ma_lmm = info->mti_big_lmm;
 548                         ma->ma_lmm_size = rc;
 549                         /* update mdt_max_mdsize so all clients
 550                          * will be aware about that */
 551                         if (info->mti_mdt->mdt_max_mdsize < rc)
 552                                 info->mti_mdt->mdt_max_mdsize = rc;
 553                         rc = 0;
 554                 }
 555         }
 556
 557         return rc;
 558 }
 559
 560 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 561                       struct mdt_object *o, struct lu_fid *pfid)
 562 {
 563         struct lu_buf           *buf = &info->mti_buf;
 564         struct link_ea_header   *leh;
 565         struct link_ea_entry    *lee;
 566         int                      rc;
 567         ENTRY;
 568
 569         buf->lb_buf = info->mti_big_lmm;
 570         buf->lb_len = info->mti_big_lmmsize;
 571         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 572                           buf, XATTR_NAME_LINK);
 573         /* ignore errors, MA_PFID won't be set and it is
 574          * up to the caller to treat this as an error */
 575         if (rc == -ERANGE || buf->lb_len == 0) {
 576                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 577                 buf->lb_buf = info->mti_big_lmm;
 578                 buf->lb_len = info->mti_big_lmmsize;
 579         }
 580
 581         if (rc < 0)
 582                 RETURN(rc);
 583         if (rc < sizeof(*leh)) {
 584                 CERROR("short LinkEA on "DFID": rc = %d\n",
 585                        PFID(mdt_object_fid(o)), rc);
 586                 RETURN(-ENODATA);
 587         }
 588
 589         leh = (struct link_ea_header *) buf->lb_buf;
 590         lee = (struct link_ea_entry *)(leh + 1);
 591         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 592                 leh->leh_magic = LINK_EA_MAGIC;
 593                 leh->leh_reccount = __swab32(leh->leh_reccount);
 594                 leh->leh_len = __swab64(leh->leh_len);
 595         }
 596         if (leh->leh_magic != LINK_EA_MAGIC)
 597                 RETURN(-EINVAL);
 598         if (leh->leh_reccount == 0)
 599                 RETURN(-ENODATA);
 600
 601         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 602         fid_be_to_cpu(pfid, pfid);
 603
 604         RETURN(0);
 605 }
 606
 607 int mdt_attr_get_complex(struct mdt_thread_info *info,
 608                          struct mdt_object *o, struct md_attr *ma)
 609 {
 610         const struct lu_env *env = info->mti_env;
 611         struct md_object    *next = mdt_object_child(o);
 612         struct lu_buf       *buf = &info->mti_buf;
 613         u32                  mode = lu_object_attr(&next->mo_lu);
 614         int                  need = ma->ma_need;
 615         int                  rc = 0, rc2;
 616         ENTRY;
 617
 618         ma->ma_valid = 0;
 619
 620         if (need & MA_INODE) {
 621                 ma->ma_need = MA_INODE;
 622                 rc = mo_attr_get(env, next, ma);
 623                 if (rc)
 624                         GOTO(out, rc);
 625                 ma->ma_valid |= MA_INODE;
 626         }
 627
 628         if (need & MA_PFID) {
 629                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 630                 if (rc == 0)
 631                         ma->ma_valid |= MA_PFID;
 632                 /* ignore this error, parent fid is not mandatory */
 633                 rc = 0;
 634         }
 635
 636         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 637                 rc = mdt_attr_get_lov(info, o, ma);
 638                 if (rc)
 639                         GOTO(out, rc);
 640         }
 641
 642         if (need & MA_LMV && S_ISDIR(mode)) {
 643                 buf->lb_buf = ma->ma_lmv;
 644                 buf->lb_len = ma->ma_lmv_size;
 645                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 646                 if (rc2 > 0) {
 647                         ma->ma_lmv_size = rc2;
 648                         ma->ma_valid |= MA_LMV;
 649                 } else if (rc2 == -ENODATA) {
 650                         /* no LMV EA */
 651                         ma->ma_lmv_size = 0;
 652                 } else
 653                         GOTO(out, rc = rc2);
 654         }
 655
 656         if (need & MA_SOM && S_ISREG(mode)) {
 657                 buf->lb_buf = info->mti_xattr_buf;
 658                 buf->lb_len = sizeof(info->mti_xattr_buf);
 659                 CLASSERT(sizeof(struct som_attrs) <=
 660                          sizeof(info->mti_xattr_buf));
 661                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 662                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 663                 if (rc2 == 0)
 664                         ma->ma_valid |= MA_SOM;
 665                 else if (rc2 < 0 && rc2 != -ENODATA)
 666                         GOTO(out, rc = rc2);
 667         }
 668
 669         if (need & MA_HSM && S_ISREG(mode)) {
 670                 buf->lb_buf = info->mti_xattr_buf;
 671                 buf->lb_len = sizeof(info->mti_xattr_buf);
 672                 CLASSERT(sizeof(struct hsm_attrs) <=
 673                          sizeof(info->mti_xattr_buf));
 674                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 675                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 676                 if (rc2 == 0)
 677                         ma->ma_valid |= MA_HSM;
 678                 else if (rc2 < 0 && rc2 != -ENODATA)
 679                         GOTO(out, rc = rc2);
 680         }
 681
 682 #ifdef CONFIG_FS_POSIX_ACL
 683         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 684                 buf->lb_buf = ma->ma_acl;
 685                 buf->lb_len = ma->ma_acl_size;
 686                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 687                 if (rc2 > 0) {
 688                         ma->ma_acl_size = rc2;
 689                         ma->ma_valid |= MA_ACL_DEF;
 690                 } else if (rc2 == -ENODATA) {
 691                         /* no ACLs */
 692                         ma->ma_acl_size = 0;
 693                 } else
 694                         GOTO(out, rc = rc2);
 695         }
 696 #endif
 697 out:
 698         ma->ma_need = need;
 699         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 700                rc, ma->ma_valid, ma->ma_lmm);
 701         RETURN(rc);
 702 }
 703
 704 static int mdt_getattr_internal(struct mdt_thread_info *info,
 705                                 struct mdt_object *o, int ma_need)
 706 {
 707         struct md_object        *next = mdt_object_child(o);
 708         const struct mdt_body   *reqbody = info->mti_body;
 709         struct ptlrpc_request   *req = mdt_info_req(info);
 710         struct md_attr          *ma = &info->mti_attr;
 711         struct lu_attr          *la = &ma->ma_attr;
 712         struct req_capsule      *pill = info->mti_pill;
 713         const struct lu_env     *env = info->mti_env;
 714         struct mdt_body         *repbody;
 715         struct lu_buf           *buffer = &info->mti_buf;
 716         int                     rc;
 717         int                     is_root;
 718         ENTRY;
 719
 720         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 721                 RETURN(err_serious(-ENOMEM));
 722
 723         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 724
 725         ma->ma_valid = 0;
 726
 727         if (mdt_object_remote(o)) {
 728                 /* This object is located on remote node.*/
 729                 /* Return -EIO for old client */
 730                 if (!mdt_is_dne_client(req->rq_export))
 731                         GOTO(out, rc = -EIO);
 732
 733                 repbody->fid1 = *mdt_object_fid(o);
 734                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 735                 GOTO(out, rc = 0);
 736         }
 737
 738         buffer->lb_len = reqbody->eadatasize;
 739         if (buffer->lb_len > 0) {
 740                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 741                 if (buffer->lb_buf == NULL)
 742                         GOTO(out, rc = -EPROTO);
 743         } else {
 744                 buffer->lb_buf = NULL;
 745                 ma_need &= ~(MA_LOV | MA_LMV);
 746                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 747                        mdt_obd_name(info->mti_mdt),
 748                        req->rq_export->exp_client_uuid.uuid);
 749         }
 750
 751         /* If it is dir object and client require MEA, then we got MEA */
 752         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 753             reqbody->valid & OBD_MD_MEA) {
 754                 /* Assumption: MDT_MD size is enough for lmv size. */
 755                 ma->ma_lmv = buffer->lb_buf;
 756                 ma->ma_lmv_size = buffer->lb_len;
 757                 ma->ma_need = MA_INODE;
 758                 if (ma->ma_lmv_size > 0)
 759                         ma->ma_need |= MA_LMV;
 760         } else {
 761                 ma->ma_lmm = buffer->lb_buf;
 762                 ma->ma_lmm_size = buffer->lb_len;
 763                 ma->ma_need = MA_INODE | MA_HSM;
 764                 if (ma->ma_lmm_size > 0)
 765                         ma->ma_need |= MA_LOV;
 766         }
 767
 768         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 769             reqbody->valid & OBD_MD_FLDIREA  &&
 770             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 771                 /* get default stripe info for this dir. */
 772                 ma->ma_need |= MA_LOV_DEF;
 773         }
 774         ma->ma_need |= ma_need;
 775         if (ma->ma_need & MA_SOM)
 776                 ma->ma_som = &info->mti_u.som.data;
 777
 778         rc = mdt_attr_get_complex(info, o, ma);
 779         if (unlikely(rc)) {
 780                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 781                        mdt_obd_name(info->mti_mdt),
 782                        PFID(mdt_object_fid(o)), rc);
 783                 RETURN(rc);
 784         }
 785
 786         /* if file is released, check if a restore is running */
 787         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 788             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 789                 repbody->t_state = MS_RESTORE;
 790                 repbody->valid |= OBD_MD_TSTATE;
 791         }
 792
 793         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 794
 795         /* the Lustre protocol supposes to return default striping
 796          * on the user-visible root if explicitly requested */
 797         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 798             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 799                 struct lu_fid      rootfid;
 800                 struct mdt_object *root;
 801                 struct mdt_device *mdt = info->mti_mdt;
 802
 803                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 804                 if (rc)
 805                         RETURN(rc);
 806                 root = mdt_object_find(env, mdt, &rootfid);
 807                 if (IS_ERR(root))
 808                         RETURN(PTR_ERR(root));
 809                 rc = mdt_attr_get_lov(info, root, ma);
 810                 mdt_object_put(info->mti_env, root);
 811                 if (unlikely(rc)) {
 812                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 813                                mdt_obd_name(info->mti_mdt),
 814                                PFID(mdt_object_fid(o)), rc);
 815                         RETURN(rc);
 816                 }
 817         }
 818
 819         if (likely(ma->ma_valid & MA_INODE))
 820                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 821         else
 822                 RETURN(-EFAULT);
 823
 824         if (mdt_body_has_lov(la, reqbody)) {
 825                 if (ma->ma_valid & MA_LOV) {
 826                         LASSERT(ma->ma_lmm_size);
 827                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 828                         repbody->eadatasize = ma->ma_lmm_size;
 829                         if (S_ISDIR(la->la_mode))
 830                                 repbody->valid |= OBD_MD_FLDIREA;
 831                         else
 832                                 repbody->valid |= OBD_MD_FLEASIZE;
 833                 }
 834                 if (ma->ma_valid & MA_LMV) {
 835                         LASSERT(S_ISDIR(la->la_mode));
 836                         repbody->eadatasize = ma->ma_lmv_size;
 837                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 838                 }
 839         } else if (S_ISLNK(la->la_mode) &&
 840                    reqbody->valid & OBD_MD_LINKNAME) {
 841                 buffer->lb_buf = ma->ma_lmm;
 842                 /* eadatasize from client includes NULL-terminator, so
 843                  * there is no need to read it */
 844                 buffer->lb_len = reqbody->eadatasize - 1;
 845                 rc = mo_readlink(env, next, buffer);
 846                 if (unlikely(rc <= 0)) {
 847                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 848                                mdt_obd_name(info->mti_mdt),
 849                                PFID(mdt_object_fid(o)), rc);
 850                         rc = -EFAULT;
 851                 } else {
 852                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 853
 854                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 855                                 rc -= 2;
 856                         repbody->valid |= OBD_MD_LINKNAME;
 857                         /* we need to report back size with NULL-terminator
 858                          * because client expects that */
 859                         repbody->eadatasize = rc + 1;
 860                         if (repbody->eadatasize != reqbody->eadatasize)
 861                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 862                                        "on "DFID ", expected %d\n",
 863                                        mdt_obd_name(info->mti_mdt),
 864                                        rc, PFID(mdt_object_fid(o)),
 865                                        reqbody->eadatasize - 1);
 866                         /* NULL terminate */
 867                         ((char *)ma->ma_lmm)[rc] = 0;
 868
 869                         /* If the total CDEBUG() size is larger than a page, it
 870                          * will print a warning to the console, avoid this by
 871                          * printing just the last part of the symlink. */
 872                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 873                                print_limit < rc ? "..." : "", print_limit,
 874                                (char *)ma->ma_lmm + rc - print_limit, rc);
 875                         rc = 0;
 876                 }
 877         }
 878
 879         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 880                 repbody->max_cookiesize = 0;
 881                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 882                 repbody->valid |= OBD_MD_FLMODEASIZE;
 883                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 884                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 885                        repbody->max_cookiesize);
 886         }
 887
 888         if (exp_connect_rmtclient(info->mti_exp) &&
 889             reqbody->valid & OBD_MD_FLRMTPERM) {
 890                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 891
 892                 /* mdt_getattr_lock only */
 893                 rc = mdt_pack_remote_perm(info, o, buf);
 894                 if (rc) {
 895                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 896                         repbody->aclsize = 0;
 897                         RETURN(rc);
 898                 } else {
 899                         repbody->valid |= OBD_MD_FLRMTPERM;
 900                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 901                 }
 902         }
 903 #ifdef CONFIG_FS_POSIX_ACL
 904         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 905                  (reqbody->valid & OBD_MD_FLACL)) {
 906                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 907                 buffer->lb_len = req_capsule_get_size(pill,
 908                                                       &RMF_ACL, RCL_SERVER);
 909                 if (buffer->lb_len > 0) {
 910                         rc = mo_xattr_get(env, next, buffer,
 911                                           XATTR_NAME_ACL_ACCESS);
 912                         if (rc < 0) {
 913                                 if (rc == -ENODATA) {
 914                                         repbody->aclsize = 0;
 915                                         repbody->valid |= OBD_MD_FLACL;
 916                                         rc = 0;
 917                                 } else if (rc == -EOPNOTSUPP) {
 918                                         rc = 0;
 919                                 } else {
 920                                         CERROR("%s: unable to read "DFID
 921                                                " ACL: rc = %d\n",
 922                                                mdt_obd_name(info->mti_mdt),
 923                                                PFID(mdt_object_fid(o)), rc);
 924                                 }
 925                         } else {
 926                                 repbody->aclsize = rc;
 927                                 repbody->valid |= OBD_MD_FLACL;
 928                                 rc = 0;
 929                         }
 930                 }
 931         }
 932 #endif
 933
 934         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 935             info->mti_mdt->mdt_lut.lut_mds_capa &&
 936             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 937                 struct lustre_capa *capa;
 938
 939                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 940                 LASSERT(capa);
 941                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 942                 rc = mo_capa_get(env, next, capa, 0);
 943                 if (rc)
 944                         RETURN(rc);
 945                 repbody->valid |= OBD_MD_FLMDSCAPA;
 946         }
 947
 948 out:
 949         if (rc == 0)
 950                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 951
 952         RETURN(rc);
 953 }
 954
 955 static int mdt_renew_capa(struct mdt_thread_info *info)
 956 {
 957         struct mdt_object  *obj = info->mti_object;
 958         struct mdt_body    *body;
 959         struct lustre_capa *capa, *c;
 960         int rc;
 961         ENTRY;
 962
 963         /* if object doesn't exist, or server has disabled capability,
 964          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 965          * flag not set.
 966          */
 967         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
 968             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 969                 RETURN(0);
 970
 971         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 972         LASSERT(body != NULL);
 973
 974         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 975         LASSERT(c);
 976
 977         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 978         LASSERT(capa);
 979
 980         *capa = *c;
 981         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 982         if (rc == 0)
 983                 body->valid |= OBD_MD_FLOSSCAPA;
 984         RETURN(rc);
 985 }
 986
 987 int mdt_getattr(struct tgt_session_info *tsi)
 988 {
 989         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 990         struct mdt_object       *obj = info->mti_object;
 991         struct req_capsule      *pill = info->mti_pill;
 992         struct mdt_body         *reqbody;
 993         struct mdt_body         *repbody;
 994         mode_t                   mode;
 995         int rc, rc2;
 996         ENTRY;
 997
 998         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 999         LASSERT(reqbody);
1000
1001         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
1002                 rc = req_capsule_server_pack(pill);
1003                 if (unlikely(rc))
1004                         RETURN(err_serious(rc));
1005                 rc = mdt_renew_capa(info);
1006                 GOTO(out_shrink, rc);
1007         }
1008
1009         LASSERT(obj != NULL);
1010         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1011
1012         mode = lu_object_attr(&obj->mot_obj);
1013
1014         /* old clients may not report needed easize, use max value then */
1015         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1016                              reqbody->eadatasize == 0 ?
1017                              info->mti_mdt->mdt_max_mdsize :
1018                              reqbody->eadatasize);
1019
1020         rc = req_capsule_server_pack(pill);
1021         if (unlikely(rc != 0))
1022                 GOTO(out, rc = err_serious(rc));
1023
1024         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1025         LASSERT(repbody != NULL);
1026         repbody->eadatasize = 0;
1027         repbody->aclsize = 0;
1028
1029         if (reqbody->valid & OBD_MD_FLRMTPERM)
1030                 rc = mdt_init_ucred(info, reqbody);
1031         else
1032                 rc = mdt_check_ucred(info);
1033         if (unlikely(rc))
1034                 GOTO(out_shrink, rc);
1035
1036         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1037
1038         /*
1039          * Don't check capability at all, because rename might getattr for
1040          * remote obj, and at that time no capability is available.
1041          */
1042         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1043         rc = mdt_getattr_internal(info, obj, 0);
1044         if (reqbody->valid & OBD_MD_FLRMTPERM)
1045                 mdt_exit_ucred(info);
1046         EXIT;
1047 out_shrink:
1048         mdt_client_compatibility(info);
1049         rc2 = mdt_fix_reply(info);
1050         if (rc == 0)
1051                 rc = rc2;
1052 out:
1053         mdt_thread_info_fini(info);
1054         return rc;
1055 }
1056
1057 int mdt_is_subdir(struct tgt_session_info *tsi)
1058 {
1059         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1060         struct mdt_object     *o = info->mti_object;
1061         struct req_capsule    *pill = info->mti_pill;
1062         const struct mdt_body *body = info->mti_body;
1063         struct mdt_body       *repbody;
1064         int                    rc;
1065         ENTRY;
1066
1067         LASSERT(o != NULL);
1068
1069         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1070
1071         /*
1072          * We save last checked parent fid to @repbody->fid1 for remote
1073          * directory case.
1074          */
1075         LASSERT(fid_is_sane(&body->fid2));
1076         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1077         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1078                            &body->fid2, &repbody->fid1);
1079         if (rc == 0 || rc == -EREMOTE)
1080                 repbody->valid |= OBD_MD_FLID;
1081
1082         mdt_thread_info_fini(info);
1083         RETURN(rc);
1084 }
1085
1086 int mdt_swap_layouts(struct tgt_session_info *tsi)
1087 {
1088         struct mdt_thread_info  *info;
1089         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1090         struct obd_export       *exp = req->rq_export;
1091         struct mdt_object       *o1, *o2, *o;
1092         struct mdt_lock_handle  *lh1, *lh2;
1093         struct mdc_swap_layouts *msl;
1094         int                      rc;
1095         ENTRY;
1096
1097         /* client does not support layout lock, so layout swaping
1098          * is disabled.
1099          * FIXME: there is a problem for old clients which don't support
1100          * layout lock yet. If those clients have already opened the file
1101          * they won't be notified at all so that old layout may still be
1102          * used to do IO. This can be fixed after file release is landed by
1103          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1104         if (!exp_connect_layout(exp))
1105                 RETURN(-EOPNOTSUPP);
1106
1107         info = tsi2mdt_info(tsi);
1108         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1109                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1110                                  req_capsule_client_get(info->mti_pill,
1111                                                         &RMF_CAPA1));
1112
1113         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1114                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1115                                  req_capsule_client_get(info->mti_pill,
1116                                                         &RMF_CAPA2));
1117
1118         o1 = info->mti_object;
1119         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1120                                 &info->mti_body->fid2);
1121         if (IS_ERR(o))
1122                 GOTO(out, rc = PTR_ERR(o));
1123
1124         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1125                 GOTO(put, rc = -ENOENT);
1126
1127         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1128         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1129                 GOTO(put, rc);
1130
1131         if (rc < 0)
1132                 swap(o1, o2);
1133
1134         /* permission check. Make sure the calling process having permission
1135          * to write both files. */
1136         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1137                                 MAY_WRITE);
1138         if (rc < 0)
1139                 GOTO(put, rc);
1140
1141         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1142                                 MAY_WRITE);
1143         if (rc < 0)
1144                 GOTO(put, rc);
1145
1146         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1147         if (msl == NULL)
1148                 GOTO(put, rc = -EPROTO);
1149
1150         lh1 = &info->mti_lh[MDT_LH_NEW];
1151         mdt_lock_reg_init(lh1, LCK_EX);
1152         lh2 = &info->mti_lh[MDT_LH_OLD];
1153         mdt_lock_reg_init(lh2, LCK_EX);
1154
1155         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1156                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1157         if (rc < 0)
1158                 GOTO(put, rc);
1159
1160         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1161                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1162         if (rc < 0)
1163                 GOTO(unlock1, rc);
1164
1165         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1166                              mdt_object_child(o2), msl->msl_flags);
1167         GOTO(unlock2, rc);
1168 unlock2:
1169         mdt_object_unlock(info, o2, lh2, rc);
1170 unlock1:
1171         mdt_object_unlock(info, o1, lh1, rc);
1172 put:
1173         mdt_object_put(info->mti_env, o);
1174 out:
1175         mdt_thread_info_fini(info);
1176         RETURN(rc);
1177 }
1178
1179 static int mdt_raw_lookup(struct mdt_thread_info *info,
1180                           struct mdt_object *parent,
1181                           const struct lu_name *lname,
1182                           struct ldlm_reply *ldlm_rep)
1183 {
1184         struct md_object *next = mdt_object_child(info->mti_object);
1185         const struct mdt_body *reqbody = info->mti_body;
1186         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1187         struct mdt_body *repbody;
1188         int rc;
1189         ENTRY;
1190
1191         if (reqbody->valid != OBD_MD_FLID)
1192                 RETURN(0);
1193
1194         LASSERT(!info->mti_cross_ref);
1195
1196         /* Only got the fid of this obj by name */
1197         fid_zero(child_fid);
1198         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1199                         &info->mti_spec);
1200 #if 0
1201         /* XXX is raw_lookup possible as intent operation? */
1202         if (rc != 0) {
1203                 if (rc == -ENOENT)
1204                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1205                 RETURN(rc);
1206         } else
1207                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1208
1209         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1210 #endif
1211         if (rc == 0) {
1212                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1213                 repbody->fid1 = *child_fid;
1214                 repbody->valid = OBD_MD_FLID;
1215         }
1216         RETURN(1);
1217 }
1218
1219 /*
1220  * UPDATE lock should be taken against parent, and be release before exit;
1221  * child_bits lock should be taken against child, and be returned back:
1222  *            (1)normal request should release the child lock;
1223  *            (2)intent request will grant the lock to client.
1224  */
1225 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1226                                  struct mdt_lock_handle *lhc,
1227                                  __u64 child_bits,
1228                                  struct ldlm_reply *ldlm_rep)
1229 {
1230         struct ptlrpc_request  *req       = mdt_info_req(info);
1231         struct mdt_body        *reqbody   = NULL;
1232         struct mdt_object      *parent    = info->mti_object;
1233         struct mdt_object      *child;
1234         struct md_object       *next      = mdt_object_child(parent);
1235         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1236         struct lu_name         *lname     = NULL;
1237         struct mdt_lock_handle *lhp       = NULL;
1238         struct ldlm_lock       *lock;
1239         struct ldlm_res_id     *res_id;
1240         int                     is_resent;
1241         int                     ma_need = 0;
1242         int                     rc;
1243
1244         ENTRY;
1245
1246         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1247         LASSERT(ergo(is_resent,
1248                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1249
1250         LASSERT(parent != NULL);
1251
1252         lname = &info->mti_name;
1253         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1254
1255         if (!info->mti_cross_ref) {
1256                 /*
1257                  * XXX: Check for anonymous name is for getattr by fid
1258                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1259                  * that is the name must contain at least one character and
1260                  * the terminating '\0'.
1261                  */
1262                 if (!lu_name_is_valid(lname)) {
1263                         reqbody = req_capsule_client_get(info->mti_pill,
1264                                                          &RMF_MDT_BODY);
1265                         if (unlikely(reqbody == NULL))
1266                                 RETURN(err_serious(-EFAULT));
1267
1268                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1269                                 RETURN(err_serious(-EINVAL));
1270
1271                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1272                                "ldlm_rep = %p\n",
1273                                PFID(mdt_object_fid(parent)),
1274                                PFID(&reqbody->fid2), ldlm_rep);
1275                 } else {
1276                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1277                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1278                                PNAME(lname), ldlm_rep);
1279                 }
1280         }
1281
1282         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1283
1284         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1285                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1286                                 &parent->mot_obj,
1287                                 "Parent doesn't exist!\n");
1288                 RETURN(-ESTALE);
1289         } else if (!info->mti_cross_ref) {
1290                 LASSERTF(!mdt_object_remote(parent),
1291                          "Parent "DFID" is on remote server\n",
1292                          PFID(mdt_object_fid(parent)));
1293         }
1294
1295         if (lu_name_is_valid(lname)) {
1296                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1297                 if (rc != 0) {
1298                         if (rc > 0)
1299                                 rc = 0;
1300                         RETURN(rc);
1301                 }
1302         }
1303
1304         if (info->mti_cross_ref) {
1305                 /* Only getattr on the child. Parent is on another node. */
1306                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1307                 child = parent;
1308                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1309                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1310
1311                 if (is_resent) {
1312                         /* Do not take lock for resent case. */
1313                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1314                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1315                                  lhc->mlh_reg_lh.cookie);
1316                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1317                                                 &lock->l_resource->lr_name));
1318                         LDLM_LOCK_PUT(lock);
1319                         rc = 0;
1320                 } else {
1321                         mdt_lock_handle_init(lhc);
1322                         mdt_lock_reg_init(lhc, LCK_PR);
1323
1324                         /*
1325                          * Object's name is on another MDS, no lookup or layout
1326                          * lock is needed here but update lock is.
1327                          */
1328                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1329                                         MDS_INODELOCK_LAYOUT);
1330                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1331
1332                         rc = mdt_object_lock(info, child, lhc, child_bits,
1333                                              MDT_LOCAL_LOCK);
1334                 }
1335                 if (rc == 0) {
1336                         /* Finally, we can get attr for child. */
1337                         if (!mdt_object_exists(child)) {
1338                                 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1339                                                 &child->mot_obj,
1340                                              "remote object doesn't exist.\n");
1341                                 mdt_object_unlock(info, child, lhc, 1);
1342                                 RETURN(-ENOENT);
1343                         }
1344
1345                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1346                                          BYPASS_CAPA);
1347                         rc = mdt_getattr_internal(info, child, 0);
1348                         if (unlikely(rc != 0))
1349                                 mdt_object_unlock(info, child, lhc, 1);
1350                 }
1351                 RETURN(rc);
1352         }
1353
1354         if (lu_name_is_valid(lname)) {
1355                 /* step 1: lock parent only if parent is a directory */
1356                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1357                         lhp = &info->mti_lh[MDT_LH_PARENT];
1358                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1359                         rc = mdt_object_lock(info, parent, lhp,
1360                                              MDS_INODELOCK_UPDATE,
1361                                              MDT_LOCAL_LOCK);
1362                         if (unlikely(rc != 0))
1363                                 RETURN(rc);
1364                 }
1365
1366                 /* step 2: lookup child's fid by name */
1367                 fid_zero(child_fid);
1368                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1369                                 &info->mti_spec);
1370
1371                 if (rc != 0) {
1372                         if (rc == -ENOENT)
1373                                 mdt_set_disposition(info, ldlm_rep,
1374                                                     DISP_LOOKUP_NEG);
1375                         GOTO(out_parent, rc);
1376                 } else
1377                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1378         } else {
1379                 *child_fid = reqbody->fid2;
1380                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1381         }
1382
1383         /*
1384          *step 3: find the child object by fid & lock it.
1385          *        regardless if it is local or remote.
1386          *
1387          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1388          *      set parent dir fid the same as child fid in getattr by fid case
1389          *      we should not lu_object_find() the object again, could lead
1390          *      to hung if there is a concurrent unlink destroyed the object.
1391          */
1392         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1393                 mdt_object_get(info->mti_env, parent);
1394                 child = parent;
1395         } else {
1396                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1397                                         child_fid);
1398         }
1399
1400         if (unlikely(IS_ERR(child)))
1401                 GOTO(out_parent, rc = PTR_ERR(child));
1402         if (is_resent) {
1403                 /* Do not take lock for resent case. */
1404                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1405                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1406                          lhc->mlh_reg_lh.cookie);
1407
1408                 res_id = &lock->l_resource->lr_name;
1409                 if (!fid_res_name_eq(mdt_object_fid(child),
1410                                      &lock->l_resource->lr_name)) {
1411                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1412                                                  &lock->l_resource->lr_name),
1413                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1414                                  PLDLMRES(lock->l_resource),
1415                                  PFID(mdt_object_fid(parent)));
1416                         CWARN("Although resent, but still not get child lock"
1417                               "parent:"DFID" child:"DFID"\n",
1418                               PFID(mdt_object_fid(parent)),
1419                               PFID(mdt_object_fid(child)));
1420                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1421                         LDLM_LOCK_PUT(lock);
1422                         GOTO(relock, 0);
1423                 }
1424                 LDLM_LOCK_PUT(lock);
1425                 rc = 0;
1426         } else {
1427                 bool try_layout = false;
1428
1429 relock:
1430                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1431                 mdt_lock_handle_init(lhc);
1432                 mdt_lock_reg_init(lhc, LCK_PR);
1433
1434                 if (!mdt_object_exists(child)) {
1435                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1436                                         &child->mot_obj,
1437                                         "Object doesn't exist!\n");
1438                         GOTO(out_child, rc = -ENOENT);
1439                 }
1440
1441                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1442                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1443                         struct md_attr *ma = &info->mti_attr;
1444
1445                         ma->ma_valid = 0;
1446                         ma->ma_need = MA_INODE;
1447                         rc = mdt_attr_get_complex(info, child, ma);
1448                         if (unlikely(rc != 0))
1449                                 GOTO(out_child, rc);
1450
1451                         /* If the file has not been changed for some time, we
1452                          * return not only a LOOKUP lock, but also an UPDATE
1453                          * lock and this might save us RPC on later STAT. For
1454                          * directories, it also let negative dentry cache start
1455                          * working for this dir. */
1456                         if (ma->ma_valid & MA_INODE &&
1457                             ma->ma_attr.la_valid & LA_CTIME &&
1458                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1459                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1460                                 child_bits |= MDS_INODELOCK_UPDATE;
1461                 }
1462
1463                 /* layout lock must be granted in a best-effort way
1464                  * for IT operations */
1465                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1466                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1467                     exp_connect_layout(info->mti_exp) &&
1468                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1469                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1470                         /* try to grant layout lock for regular file. */
1471                         try_layout = true;
1472                 }
1473
1474                 rc = 0;
1475                 if (try_layout) {
1476                         child_bits |= MDS_INODELOCK_LAYOUT;
1477                         /* try layout lock, it may fail to be granted due to
1478                          * contention at LOOKUP or UPDATE */
1479                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1480                                                  MDT_CROSS_LOCK)) {
1481                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1482                                 LASSERT(child_bits != 0);
1483                                 rc = mdt_object_lock(info, child, lhc,
1484                                                 child_bits, MDT_CROSS_LOCK);
1485                         } else {
1486                                 ma_need |= MA_LOV;
1487                         }
1488                 } else {
1489                         rc = mdt_object_lock(info, child, lhc, child_bits,
1490                                                 MDT_CROSS_LOCK);
1491                 }
1492                 if (unlikely(rc != 0))
1493                         GOTO(out_child, rc);
1494         }
1495
1496         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1497         /* Get MA_SOM attributes if update lock is given. */
1498         if (lock &&
1499             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1500             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1501                 ma_need |= MA_SOM;
1502
1503         /* finally, we can get attr for child. */
1504         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1505         rc = mdt_getattr_internal(info, child, ma_need);
1506         if (unlikely(rc != 0)) {
1507                 mdt_object_unlock(info, child, lhc, 1);
1508         } else if (lock) {
1509                 /* Debugging code. */
1510                 res_id = &lock->l_resource->lr_name;
1511                 LDLM_DEBUG(lock, "Returning lock to client");
1512                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1513                                          &lock->l_resource->lr_name),
1514                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1515                          PLDLMRES(lock->l_resource),
1516                          PFID(mdt_object_fid(child)));
1517                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1518                         mdt_pack_size2body(info, child);
1519         }
1520         if (lock)
1521                 LDLM_LOCK_PUT(lock);
1522
1523         EXIT;
1524 out_child:
1525         mdt_object_put(info->mti_env, child);
1526 out_parent:
1527         if (lhp)
1528                 mdt_object_unlock(info, parent, lhp, 1);
1529         return rc;
1530 }
1531
1532 /* normal handler: should release the child lock */
1533 int mdt_getattr_name(struct tgt_session_info *tsi)
1534 {
1535         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1536         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1537         struct mdt_body        *reqbody;
1538         struct mdt_body        *repbody;
1539         int rc, rc2;
1540         ENTRY;
1541
1542         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1543         LASSERT(reqbody != NULL);
1544         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1545         LASSERT(repbody != NULL);
1546
1547         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1548         repbody->eadatasize = 0;
1549         repbody->aclsize = 0;
1550
1551         rc = mdt_init_ucred(info, reqbody);
1552         if (unlikely(rc))
1553                 GOTO(out_shrink, rc);
1554
1555         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1556         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1557                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1558                 lhc->mlh_reg_lh.cookie = 0;
1559         }
1560         mdt_exit_ucred(info);
1561         EXIT;
1562 out_shrink:
1563         mdt_client_compatibility(info);
1564         rc2 = mdt_fix_reply(info);
1565         if (rc == 0)
1566                 rc = rc2;
1567         mdt_thread_info_fini(info);
1568         return rc;
1569 }
1570
1571 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1572                          void *karg, void *uarg);
1573
1574 int mdt_set_info(struct tgt_session_info *tsi)
1575 {
1576         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1577         char                    *key;
1578         void                    *val;
1579         int                      keylen, vallen, rc = 0;
1580
1581         ENTRY;
1582
1583         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1584         if (key == NULL) {
1585                 DEBUG_REQ(D_HA, req, "no set_info key");
1586                 RETURN(err_serious(-EFAULT));
1587         }
1588
1589         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1590                                       RCL_CLIENT);
1591
1592         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1593         if (val == NULL) {
1594                 DEBUG_REQ(D_HA, req, "no set_info val");
1595                 RETURN(err_serious(-EFAULT));
1596         }
1597
1598         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1599                                       RCL_CLIENT);
1600
1601         /* Swab any part of val you need to here */
1602         if (KEY_IS(KEY_READ_ONLY)) {
1603                 spin_lock(&req->rq_export->exp_lock);
1604                 if (*(__u32 *)val)
1605                         *exp_connect_flags_ptr(req->rq_export) |=
1606                                 OBD_CONNECT_RDONLY;
1607                 else
1608                         *exp_connect_flags_ptr(req->rq_export) &=
1609                                 ~OBD_CONNECT_RDONLY;
1610                 spin_unlock(&req->rq_export->exp_lock);
1611         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1612                 struct changelog_setinfo *cs = val;
1613
1614                 if (vallen != sizeof(*cs)) {
1615                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1616                                tgt_name(tsi->tsi_tgt), vallen);
1617                         RETURN(-EINVAL);
1618                 }
1619                 if (ptlrpc_req_need_swab(req)) {
1620                         __swab64s(&cs->cs_recno);
1621                         __swab32s(&cs->cs_id);
1622                 }
1623
1624                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1625                                    vallen, val, NULL);
1626         } else {
1627                 RETURN(-EINVAL);
1628         }
1629         RETURN(rc);
1630 }
1631
1632 int mdt_readpage(struct tgt_session_info *tsi)
1633 {
1634         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1635         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1636         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1637         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1638         struct mdt_body         *repbody;
1639         int                      rc;
1640         int                      i;
1641
1642         ENTRY;
1643
1644         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1645                 RETURN(err_serious(-ENOMEM));
1646
1647         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1648         if (repbody == NULL || reqbody == NULL)
1649                 RETURN(err_serious(-EFAULT));
1650
1651         /*
1652          * prepare @rdpg before calling lower layers and transfer itself. Here
1653          * reqbody->size contains offset of where to start to read and
1654          * reqbody->nlink contains number bytes to read.
1655          */
1656         rdpg->rp_hash = reqbody->size;
1657         if (rdpg->rp_hash != reqbody->size) {
1658                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1659                        rdpg->rp_hash, reqbody->size);
1660                 RETURN(-EFAULT);
1661         }
1662
1663         rdpg->rp_attrs = reqbody->mode;
1664         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1665                 rdpg->rp_attrs |= LUDA_64BITHASH;
1666         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1667                                 exp_max_brw_size(tsi->tsi_exp));
1668         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1669                           PAGE_CACHE_SHIFT;
1670         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1671         if (rdpg->rp_pages == NULL)
1672                 RETURN(-ENOMEM);
1673
1674         for (i = 0; i < rdpg->rp_npages; ++i) {
1675                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1676                 if (rdpg->rp_pages[i] == NULL)
1677                         GOTO(free_rdpg, rc = -ENOMEM);
1678         }
1679
1680         /* call lower layers to fill allocated pages with directory data */
1681         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1682         if (rc < 0)
1683                 GOTO(free_rdpg, rc);
1684
1685         /* send pages to client */
1686         rc = tgt_sendpage(tsi, rdpg, rc);
1687
1688         EXIT;
1689 free_rdpg:
1690
1691         for (i = 0; i < rdpg->rp_npages; i++)
1692                 if (rdpg->rp_pages[i] != NULL)
1693                         __free_page(rdpg->rp_pages[i]);
1694         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1695
1696         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1697                 RETURN(0);
1698
1699         return rc;
1700 }
1701
1702 static int mdt_reint_internal(struct mdt_thread_info *info,
1703                               struct mdt_lock_handle *lhc,
1704                               __u32 op)
1705 {
1706         struct req_capsule      *pill = info->mti_pill;
1707         struct mdt_body         *repbody;
1708         int                      rc = 0, rc2;
1709
1710         ENTRY;
1711
1712         rc = mdt_reint_unpack(info, op);
1713         if (rc != 0) {
1714                 CERROR("Can't unpack reint, rc %d\n", rc);
1715                 RETURN(err_serious(rc));
1716         }
1717
1718         /* for replay (no_create) lmm is not needed, client has it already */
1719         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1720                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1721                                      info->mti_rr.rr_eadatalen);
1722
1723         /* llog cookies are always 0, the field is kept for compatibility */
1724         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1725                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1726
1727         rc = req_capsule_server_pack(pill);
1728         if (rc != 0) {
1729                 CERROR("Can't pack response, rc %d\n", rc);
1730                 RETURN(err_serious(rc));
1731         }
1732
1733         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1734                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1735                 LASSERT(repbody);
1736                 repbody->eadatasize = 0;
1737                 repbody->aclsize = 0;
1738         }
1739
1740         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1741
1742         /* for replay no cookkie / lmm need, because client have this already */
1743         if (info->mti_spec.no_create)
1744                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1745                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1746
1747         rc = mdt_init_ucred_reint(info);
1748         if (rc)
1749                 GOTO(out_shrink, rc);
1750
1751         rc = mdt_fix_attr_ucred(info, op);
1752         if (rc != 0)
1753                 GOTO(out_ucred, rc = err_serious(rc));
1754
1755         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1756                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1757                 GOTO(out_ucred, rc);
1758         }
1759         rc = mdt_reint_rec(info, lhc);
1760         EXIT;
1761 out_ucred:
1762         mdt_exit_ucred(info);
1763 out_shrink:
1764         mdt_client_compatibility(info);
1765         rc2 = mdt_fix_reply(info);
1766         if (rc == 0)
1767                 rc = rc2;
1768         return rc;
1769 }
1770
1771 static long mdt_reint_opcode(struct ptlrpc_request *req,
1772                              const struct req_format **fmt)
1773 {
1774         struct mdt_device       *mdt;
1775         struct mdt_rec_reint    *rec;
1776         long                     opc;
1777
1778         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1779         if (rec != NULL) {
1780                 opc = rec->rr_opcode;
1781                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1782                 if (opc < REINT_MAX && fmt[opc] != NULL)
1783                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1784                 else {
1785                         mdt = mdt_exp2dev(req->rq_export);
1786                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1787                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1788                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1789                         opc = err_serious(-EFAULT);
1790                 }
1791         } else {
1792                 opc = err_serious(-EFAULT);
1793         }
1794         return opc;
1795 }
1796
1797 int mdt_reint(struct tgt_session_info *tsi)
1798 {
1799         long opc;
1800         int  rc;
1801         static const struct req_format *reint_fmts[REINT_MAX] = {
1802                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1803                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1804                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1805                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1806                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1807                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1808                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1809                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK
1810         };
1811
1812         ENTRY;
1813
1814         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1815         if (opc >= 0) {
1816                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1817                 /*
1818                  * No lock possible here from client to pass it to reint code
1819                  * path.
1820                  */
1821                 rc = mdt_reint_internal(info, NULL, opc);
1822                 mdt_thread_info_fini(info);
1823         } else {
1824                 rc = opc;
1825         }
1826
1827         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1828         RETURN(rc);
1829 }
1830
1831 /* this should sync the whole device */
1832 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1833 {
1834         struct dt_device *dt = mdt->mdt_bottom;
1835         int rc;
1836         ENTRY;
1837
1838         rc = dt->dd_ops->dt_sync(env, dt);
1839         RETURN(rc);
1840 }
1841
1842 /* this should sync this object */
1843 static int mdt_object_sync(struct mdt_thread_info *info)
1844 {
1845         struct md_object *next;
1846         int rc;
1847         ENTRY;
1848
1849         if (!mdt_object_exists(info->mti_object)) {
1850                 CWARN("Non existing object  "DFID"!\n",
1851                       PFID(mdt_object_fid(info->mti_object)));
1852                 RETURN(-ESTALE);
1853         }
1854         next = mdt_object_child(info->mti_object);
1855         rc = mo_object_sync(info->mti_env, next);
1856
1857         RETURN(rc);
1858 }
1859
1860 int mdt_sync(struct tgt_session_info *tsi)
1861 {
1862         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1863         struct req_capsule      *pill = tsi->tsi_pill;
1864         struct mdt_body         *body;
1865         int                      rc;
1866
1867         ENTRY;
1868
1869         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1870                 RETURN(err_serious(-ENOMEM));
1871
1872         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1873                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1874         } else {
1875                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1876
1877                 /* sync an object */
1878                 rc = mdt_object_sync(info);
1879                 if (rc == 0) {
1880                         const struct lu_fid *fid;
1881                         struct lu_attr *la = &info->mti_attr.ma_attr;
1882
1883                         info->mti_attr.ma_need = MA_INODE;
1884                         info->mti_attr.ma_valid = 0;
1885                         rc = mdt_attr_get_complex(info, info->mti_object,
1886                                                   &info->mti_attr);
1887                         if (rc == 0) {
1888                                 body = req_capsule_server_get(pill,
1889                                                               &RMF_MDT_BODY);
1890                                 fid = mdt_object_fid(info->mti_object);
1891                                 mdt_pack_attr2body(info, body, la, fid);
1892                         }
1893                 }
1894                 mdt_thread_info_fini(info);
1895         }
1896         if (rc == 0)
1897                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1898
1899         RETURN(rc);
1900 }
1901
1902 /*
1903  * Handle quota control requests to consult current usage/limit, but also
1904  * to configure quota enforcement
1905  */
1906 int mdt_quotactl(struct tgt_session_info *tsi)
1907 {
1908         struct obd_export       *exp  = tsi->tsi_exp;
1909         struct req_capsule      *pill = tsi->tsi_pill;
1910         struct obd_quotactl     *oqctl, *repoqc;
1911         int                      id, rc;
1912         struct mdt_device       *mdt = mdt_exp2dev(exp);
1913         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1914         ENTRY;
1915
1916         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1917         if (oqctl == NULL)
1918                 RETURN(err_serious(-EPROTO));
1919
1920         rc = req_capsule_server_pack(pill);
1921         if (rc)
1922                 RETURN(err_serious(rc));
1923
1924         switch (oqctl->qc_cmd) {
1925         case Q_QUOTACHECK:
1926         case LUSTRE_Q_INVALIDATE:
1927         case LUSTRE_Q_FINVALIDATE:
1928         case Q_QUOTAON:
1929         case Q_QUOTAOFF:
1930         case Q_INITQUOTA:
1931                 /* deprecated, not used any more */
1932                 RETURN(-EOPNOTSUPP);
1933                 /* master quotactl */
1934         case Q_GETINFO:
1935         case Q_SETINFO:
1936         case Q_SETQUOTA:
1937         case Q_GETQUOTA:
1938                 if (qmt == NULL)
1939                         RETURN(-EOPNOTSUPP);
1940                 /* slave quotactl */
1941         case Q_GETOINFO:
1942         case Q_GETOQUOTA:
1943                 break;
1944         default:
1945                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1946                 RETURN(-EFAULT);
1947         }
1948
1949         /* map uid/gid for remote client */
1950         id = oqctl->qc_id;
1951         if (exp_connect_rmtclient(exp)) {
1952                 struct lustre_idmap_table *idmap;
1953
1954                 idmap = exp->exp_mdt_data.med_idmap;
1955
1956                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1957                              oqctl->qc_cmd != Q_GETINFO))
1958                         RETURN(-EPERM);
1959
1960                 if (oqctl->qc_type == USRQUOTA)
1961                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1962                                                      oqctl->qc_id);
1963                 else if (oqctl->qc_type == GRPQUOTA)
1964                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1965                                                      oqctl->qc_id);
1966                 else
1967                         RETURN(-EINVAL);
1968
1969                 if (id == CFS_IDMAP_NOTFOUND) {
1970                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1971                         RETURN(-EACCES);
1972                 }
1973         }
1974
1975         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1976         if (repoqc == NULL)
1977                 RETURN(err_serious(-EFAULT));
1978
1979         if (oqctl->qc_id != id)
1980                 swap(oqctl->qc_id, id);
1981
1982         switch (oqctl->qc_cmd) {
1983
1984         case Q_GETINFO:
1985         case Q_SETINFO:
1986         case Q_SETQUOTA:
1987         case Q_GETQUOTA:
1988                 /* forward quotactl request to QMT */
1989                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
1990                 break;
1991
1992         case Q_GETOINFO:
1993         case Q_GETOQUOTA:
1994                 /* slave quotactl */
1995                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1996                                    oqctl);
1997                 break;
1998
1999         default:
2000                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2001                 RETURN(-EFAULT);
2002         }
2003
2004         if (oqctl->qc_id != id)
2005                 swap(oqctl->qc_id, id);
2006
2007         *repoqc = *oqctl;
2008         RETURN(rc);
2009 }
2010
2011 /** clone llog ctxt from child (mdd)
2012  * This allows remote llog (replicator) access.
2013  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2014  * context was originally set up, or we can handle them directly.
2015  * I choose the latter, but that means I need any llog
2016  * contexts set up by child to be accessable by the mdt.  So we clone the
2017  * context into our context list here.
2018  */
2019 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2020                                int idx)
2021 {
2022         struct md_device  *next = mdt->mdt_child;
2023         struct llog_ctxt *ctxt;
2024         int rc;
2025
2026         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2027                 return 0;
2028
2029         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2030         if (rc || ctxt == NULL) {
2031                 return 0;
2032         }
2033
2034         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2035         if (rc)
2036                 CERROR("Can't set mdt ctxt %d\n", rc);
2037
2038         return rc;
2039 }
2040
2041 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2042                                  struct mdt_device *mdt, int idx)
2043 {
2044         struct llog_ctxt *ctxt;
2045
2046         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2047         if (ctxt == NULL)
2048                 return 0;
2049         /* Put once for the get we just did, and once for the clone */
2050         llog_ctxt_put(ctxt);
2051         llog_ctxt_put(ctxt);
2052         return 0;
2053 }
2054
2055 /*
2056  * sec context handlers
2057  */
2058 int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2059 {
2060         int rc;
2061
2062         rc = mdt_handle_idmap(tsi);
2063         if (unlikely(rc)) {
2064                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2065                 __u32                    opc;
2066
2067                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2068                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2069                         sptlrpc_svc_ctx_invalidate(req);
2070         }
2071
2072         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2073
2074         return rc;
2075 }
2076
2077 /*
2078  * quota request handlers
2079  */
2080 int mdt_quota_dqacq(struct tgt_session_info *tsi)
2081 {
2082         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2083         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2084         int                      rc;
2085         ENTRY;
2086
2087         if (qmt == NULL)
2088                 RETURN(err_serious(-EOPNOTSUPP));
2089
2090         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2091         RETURN(rc);
2092 }
2093
2094 struct mdt_object *mdt_object_new(const struct lu_env *env,
2095                                   struct mdt_device *d,
2096                                   const struct lu_fid *f)
2097 {
2098         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2099         struct lu_object *o;
2100         struct mdt_object *m;
2101         ENTRY;
2102
2103         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2104         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2105         if (unlikely(IS_ERR(o)))
2106                 m = (struct mdt_object *)o;
2107         else
2108                 m = mdt_obj(o);
2109         RETURN(m);
2110 }
2111
2112 struct mdt_object *mdt_object_find(const struct lu_env *env,
2113                                    struct mdt_device *d,
2114                                    const struct lu_fid *f)
2115 {
2116         struct lu_object *o;
2117         struct mdt_object *m;
2118         ENTRY;
2119
2120         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2121         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2122         if (unlikely(IS_ERR(o)))
2123                 m = (struct mdt_object *)o;
2124         else
2125                 m = mdt_obj(o);
2126
2127         RETURN(m);
2128 }
2129
2130 /**
2131  * Asyncronous commit for mdt device.
2132  *
2133  * Pass asynchonous commit call down the MDS stack.
2134  *
2135  * \param env environment
2136  * \param mdt the mdt device
2137  */
2138 static void mdt_device_commit_async(const struct lu_env *env,
2139                                     struct mdt_device *mdt)
2140 {
2141         struct dt_device *dt = mdt->mdt_bottom;
2142         int rc;
2143
2144         rc = dt->dd_ops->dt_commit_async(env, dt);
2145         if (unlikely(rc != 0))
2146                 CWARN("async commit start failed with rc = %d", rc);
2147 }
2148
2149 /**
2150  * Mark the lock as "synchonous".
2151  *
2152  * Mark the lock to deffer transaction commit to the unlock time.
2153  *
2154  * \param lock the lock to mark as "synchonous"
2155  *
2156  * \see mdt_is_lock_sync
2157  * \see mdt_save_lock
2158  */
2159 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2160 {
2161         lock->l_ast_data = (void*)1;
2162 }
2163
2164 /**
2165  * Check whehter the lock "synchonous" or not.
2166  *
2167  * \param lock the lock to check
2168  * \retval 1 the lock is "synchonous"
2169  * \retval 0 the lock isn't "synchronous"
2170  *
2171  * \see mdt_set_lock_sync
2172  * \see mdt_save_lock
2173  */
2174 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2175 {
2176         return lock->l_ast_data != NULL;
2177 }
2178
2179 /**
2180  * Blocking AST for mdt locks.
2181  *
2182  * Starts transaction commit if in case of COS lock conflict or
2183  * deffers such a commit to the mdt_save_lock.
2184  *
2185  * \param lock the lock which blocks a request or cancelling lock
2186  * \param desc unused
2187  * \param data unused
2188  * \param flag indicates whether this cancelling or blocking callback
2189  * \retval 0
2190  * \see ldlm_blocking_ast_nocheck
2191  */
2192 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2193                      void *data, int flag)
2194 {
2195         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2196         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2197         int rc;
2198         ENTRY;
2199
2200         if (flag == LDLM_CB_CANCELING)
2201                 RETURN(0);
2202         lock_res_and_lock(lock);
2203         if (lock->l_blocking_ast != mdt_blocking_ast) {
2204                 unlock_res_and_lock(lock);
2205                 RETURN(0);
2206         }
2207         if (mdt_cos_is_enabled(mdt) &&
2208             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2209             lock->l_blocking_lock != NULL &&
2210             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2211                 mdt_set_lock_sync(lock);
2212         }
2213         rc = ldlm_blocking_ast_nocheck(lock);
2214
2215         /* There is no lock conflict if l_blocking_lock == NULL,
2216          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2217          * when the last reference to a local lock was released */
2218         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2219                 struct lu_env env;
2220
2221                 rc = lu_env_init(&env, LCT_LOCAL);
2222                 if (unlikely(rc != 0))
2223                         CWARN("lu_env initialization failed with rc = %d,"
2224                               "cannot start asynchronous commit\n", rc);
2225                 else
2226                         mdt_device_commit_async(&env, mdt);
2227                 lu_env_fini(&env);
2228         }
2229         RETURN(rc);
2230 }
2231
2232 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2233                         void *data, int flag)
2234 {
2235         struct lustre_handle lockh;
2236         int               rc;
2237
2238         switch (flag) {
2239         case LDLM_CB_BLOCKING:
2240                 ldlm_lock2handle(lock, &lockh);
2241                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2242                 if (rc < 0) {
2243                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2244                         RETURN(rc);
2245                 }
2246                 break;
2247         case LDLM_CB_CANCELING:
2248                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2249                 break;
2250         default:
2251                 LBUG();
2252         }
2253         RETURN(0);
2254 }
2255
2256 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2257                            struct mdt_object *o, struct lustre_handle *lh,
2258                            ldlm_mode_t mode, __u64 ibits)
2259 {
2260         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2261         ldlm_policy_data_t *policy = &mti->mti_policy;
2262         int rc = 0;
2263         ENTRY;
2264
2265         LASSERT(mdt_object_remote(o));
2266
2267         LASSERT(ibits & MDS_INODELOCK_UPDATE);
2268
2269         memset(einfo, 0, sizeof(*einfo));
2270         einfo->ei_type = LDLM_IBITS;
2271         einfo->ei_mode = mode;
2272         einfo->ei_cb_bl = mdt_md_blocking_ast;
2273         einfo->ei_cb_cp = ldlm_completion_ast;
2274
2275         memset(policy, 0, sizeof(*policy));
2276         policy->l_inodebits.bits = ibits;
2277
2278         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2279                             policy);
2280         RETURN(rc);
2281 }
2282
2283 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2284                             struct mdt_lock_handle *lh, __u64 ibits,
2285                             bool nonblock, int locality)
2286 {
2287         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2288         ldlm_policy_data_t *policy = &info->mti_policy;
2289         struct ldlm_res_id *res_id = &info->mti_res_id;
2290         __u64 dlmflags;
2291         int rc;
2292         ENTRY;
2293
2294         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2295         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2296         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2297         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2298
2299         if (mdt_object_remote(o)) {
2300                 if (locality == MDT_CROSS_LOCK) {
2301                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2302                                    MDS_INODELOCK_LAYOUT);
2303                         ibits |= MDS_INODELOCK_LOOKUP;
2304                 } else {
2305                         LASSERTF(!(ibits &
2306                                  (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2307                                   MDS_INODELOCK_LAYOUT)),
2308                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2309                                 mdt_obd_name(info->mti_mdt), ibits,
2310                                 PFID(mdt_object_fid(o)));
2311                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2312                 }
2313                 /* No PDO lock on remote object */
2314                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2315         }
2316
2317         if (lh->mlh_type == MDT_PDO_LOCK) {
2318                 /* check for exists after object is locked */
2319                 if (mdt_object_exists(o) == 0) {
2320                         /* Non-existent object shouldn't have PDO lock */
2321                         RETURN(-ESTALE);
2322                 } else {
2323                         /* Non-dir object shouldn't have PDO lock */
2324                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2325                                 RETURN(-ENOTDIR);
2326                 }
2327         }
2328
2329         memset(policy, 0, sizeof(*policy));
2330         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2331
2332         dlmflags = LDLM_FL_ATOMIC_CB;
2333         if (nonblock)
2334                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2335
2336         /*
2337          * Take PDO lock on whole directory and build correct @res_id for lock
2338          * on part of directory.
2339          */
2340         if (lh->mlh_pdo_hash != 0) {
2341                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2342                 mdt_lock_pdo_mode(info, o, lh);
2343                 if (lh->mlh_pdo_mode != LCK_NL) {
2344                         /*
2345                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2346                          * is never going to be sent to client and we do not
2347                          * want it slowed down due to possible cancels.
2348                          */
2349                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2350                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2351                                           policy, res_id, dlmflags,
2352                                           &info->mti_exp->exp_handle.h_cookie);
2353                         if (unlikely(rc))
2354                                 RETURN(rc);
2355                 }
2356
2357                 /*
2358                  * Finish res_id initializing by name hash marking part of
2359                  * directory which is taking modification.
2360                  */
2361                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2362         }
2363
2364         policy->l_inodebits.bits = ibits;
2365
2366         /*
2367          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2368          * going to be sent to client. If it is - mdt_intent_policy() path will
2369          * fix it up and turn FL_LOCAL flag off.
2370          */
2371         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2372                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2373                           &info->mti_exp->exp_handle.h_cookie);
2374         if (rc)
2375                 mdt_object_unlock(info, o, lh, 1);
2376         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2377                  lh->mlh_pdo_hash != 0 &&
2378                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2379                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2380         }
2381
2382         RETURN(rc);
2383 }
2384
2385 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2386                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2387 {
2388         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2389 }
2390
2391 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2392                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2393 {
2394         struct mdt_lock_handle tmp = *lh;
2395         int rc;
2396
2397         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2398         if (rc == 0)
2399                 *lh = tmp;
2400
2401         return rc == 0;
2402 }
2403
2404 /**
2405  * Save a lock within request object.
2406  *
2407  * Keep the lock referenced until whether client ACK or transaction
2408  * commit happens or release the lock immediately depending on input
2409  * parameters. If COS is ON, a write lock is converted to COS lock
2410  * before saving.
2411  *
2412  * \param info thead info object
2413  * \param h lock handle
2414  * \param mode lock mode
2415  * \param decref force immediate lock releasing
2416  */
2417 static
2418 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2419                    ldlm_mode_t mode, int decref)
2420 {
2421         ENTRY;
2422
2423         if (lustre_handle_is_used(h)) {
2424                 if (decref || !info->mti_has_trans ||
2425                     !(mode & (LCK_PW | LCK_EX))){
2426                         mdt_fid_unlock(h, mode);
2427                 } else {
2428                         struct mdt_device *mdt = info->mti_mdt;
2429                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2430                         struct ptlrpc_request *req = mdt_info_req(info);
2431                         int no_ack = 0;
2432
2433                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2434                                  h->cookie);
2435                         /* there is no request if mdt_object_unlock() is called
2436                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2437                         if (likely(req != NULL)) {
2438                                 CDEBUG(D_HA, "request = %p reply state = %p"
2439                                        " transno = "LPD64"\n", req,
2440                                        req->rq_reply_state, req->rq_transno);
2441                                 if (mdt_cos_is_enabled(mdt)) {
2442                                         no_ack = 1;
2443                                         ldlm_lock_downgrade(lock, LCK_COS);
2444                                         mode = LCK_COS;
2445                                 }
2446                                 ptlrpc_save_lock(req, h, mode, no_ack);
2447                         } else {
2448                                 ldlm_lock_decref(h, mode);
2449                         }
2450                         if (mdt_is_lock_sync(lock)) {
2451                                 CDEBUG(D_HA, "found sync-lock,"
2452                                        " async commit started\n");
2453                                 mdt_device_commit_async(info->mti_env,
2454                                                         mdt);
2455                         }
2456                         LDLM_LOCK_PUT(lock);
2457                 }
2458                 h->cookie = 0ull;
2459         }
2460
2461         EXIT;
2462 }
2463
2464 /**
2465  * Unlock mdt object.
2466  *
2467  * Immeditely release the regular lock and the PDO lock or save the
2468  * lock in reqeuest and keep them referenced until client ACK or
2469  * transaction commit.
2470  *
2471  * \param info thread info object
2472  * \param o mdt object
2473  * \param lh mdt lock handle referencing regular and PDO locks
2474  * \param decref force immediate lock releasing
2475  */
2476 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2477                        struct mdt_lock_handle *lh, int decref)
2478 {
2479         ENTRY;
2480
2481         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2482         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2483
2484         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2485                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2486
2487         EXIT;
2488 }
2489
2490 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2491                                         const struct lu_fid *f,
2492                                         struct mdt_lock_handle *lh,
2493                                         __u64 ibits)
2494 {
2495         struct mdt_object *o;
2496
2497         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2498         if (!IS_ERR(o)) {
2499                 int rc;
2500
2501                 rc = mdt_object_lock(info, o, lh, ibits,
2502                                      MDT_LOCAL_LOCK);
2503                 if (rc != 0) {
2504                         mdt_object_put(info->mti_env, o);
2505                         o = ERR_PTR(rc);
2506                 }
2507         }
2508         return o;
2509 }
2510
2511 void mdt_object_unlock_put(struct mdt_thread_info * info,
2512                            struct mdt_object * o,
2513                            struct mdt_lock_handle *lh,
2514                            int decref)
2515 {
2516         mdt_object_unlock(info, o, lh, decref);
2517         mdt_object_put(info->mti_env, o);
2518 }
2519
2520 /*
2521  * Generic code handling requests that have struct mdt_body passed in:
2522  *
2523  *  - extract mdt_body from request and save it in @info, if present;
2524  *
2525  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2526  *  @info;
2527  *
2528  *  - if HABEO_CORPUS flag is set for this request type check whether object
2529  *  actually exists on storage (lu_object_exists()).
2530  *
2531  */
2532 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2533 {
2534         const struct mdt_body    *body;
2535         struct mdt_object        *obj;
2536         const struct lu_env      *env;
2537         struct req_capsule       *pill;
2538         int                       rc;
2539         ENTRY;
2540
2541         env = info->mti_env;
2542         pill = info->mti_pill;
2543
2544         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2545         if (body == NULL)
2546                 RETURN(-EFAULT);
2547
2548         if (!(body->valid & OBD_MD_FLID))
2549                 RETURN(0);
2550
2551         if (!fid_is_sane(&body->fid1)) {
2552                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2553                 RETURN(-EINVAL);
2554         }
2555
2556         /*
2557          * Do not get size or any capa fields before we check that request
2558          * contains capa actually. There are some requests which do not, for
2559          * instance MDS_IS_SUBDIR.
2560          */
2561         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2562             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2563                 mdt_set_capainfo(info, 0, &body->fid1,
2564                                  req_capsule_client_get(pill, &RMF_CAPA1));
2565
2566         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2567         if (!IS_ERR(obj)) {
2568                 if ((flags & HABEO_CORPUS) &&
2569                     !mdt_object_exists(obj)) {
2570                         mdt_object_put(env, obj);
2571                         /* for capability renew ENOENT will be handled in
2572                          * mdt_renew_capa */
2573                         if (body->valid & OBD_MD_FLOSSCAPA)
2574                                 rc = 0;
2575                         else
2576                                 rc = -ENOENT;
2577                 } else {
2578                         info->mti_object = obj;
2579                         rc = 0;
2580                 }
2581         } else
2582                 rc = PTR_ERR(obj);
2583
2584         RETURN(rc);
2585 }
2586
2587 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2588 {
2589         struct req_capsule *pill = info->mti_pill;
2590         int rc;
2591         ENTRY;
2592
2593         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2594                 rc = mdt_body_unpack(info, flags);
2595         else
2596                 rc = 0;
2597
2598         if (rc == 0 && (flags & HABEO_REFERO)) {
2599                 /* Pack reply. */
2600                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2601                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2602                                              info->mti_body->eadatasize);
2603                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2604                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2605                                              RCL_SERVER, 0);
2606
2607                 rc = req_capsule_server_pack(pill);
2608         }
2609         RETURN(rc);
2610 }
2611
2612 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2613 {
2614         struct md_device *next = m->mdt_child;
2615
2616         return next->md_ops->mdo_init_capa_ctxt(env, next,
2617                                                 m->mdt_lut.lut_mds_capa,
2618                                                 m->mdt_capa_timeout,
2619                                                 m->mdt_capa_alg,
2620                                                 m->mdt_capa_keys);
2621 }
2622
2623 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2624 {
2625         lh->mlh_type = MDT_NUL_LOCK;
2626         lh->mlh_reg_lh.cookie = 0ull;
2627         lh->mlh_reg_mode = LCK_MINMODE;
2628         lh->mlh_pdo_lh.cookie = 0ull;
2629         lh->mlh_pdo_mode = LCK_MINMODE;
2630         lh->mlh_rreg_lh.cookie = 0ull;
2631         lh->mlh_rreg_mode = LCK_MINMODE;
2632 }
2633
2634 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2635 {
2636         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2637         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2638 }
2639
2640 /*
2641  * Initialize fields of struct mdt_thread_info. Other fields are left in
2642  * uninitialized state, because it's too expensive to zero out whole
2643  * mdt_thread_info (> 1K) on each request arrival.
2644  */
2645 void mdt_thread_info_init(struct ptlrpc_request *req,
2646                           struct mdt_thread_info *info)
2647 {
2648         int i;
2649
2650         info->mti_pill = &req->rq_pill;
2651
2652         /* lock handle */
2653         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2654                 mdt_lock_handle_init(&info->mti_lh[i]);
2655
2656         /* mdt device: it can be NULL while CONNECT */
2657         if (req->rq_export) {
2658                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2659                 info->mti_exp = req->rq_export;
2660         } else
2661                 info->mti_mdt = NULL;
2662         info->mti_env = req->rq_svc_thread->t_env;
2663         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2664
2665         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2666         info->mti_big_buf = LU_BUF_NULL;
2667         info->mti_body = NULL;
2668         info->mti_object = NULL;
2669         info->mti_dlm_req = NULL;
2670         info->mti_has_trans = 0;
2671         info->mti_cross_ref = 0;
2672         info->mti_opdata = 0;
2673         info->mti_big_lmm_used = 0;
2674
2675         info->mti_spec.no_create = 0;
2676         info->mti_spec.sp_rm_entry = 0;
2677 }
2678
2679 void mdt_thread_info_fini(struct mdt_thread_info *info)
2680 {
2681         int i;
2682
2683         if (info->mti_object != NULL) {
2684                 mdt_object_put(info->mti_env, info->mti_object);
2685                 info->mti_object = NULL;
2686         }
2687
2688         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2689                 mdt_lock_handle_fini(&info->mti_lh[i]);
2690         info->mti_env = NULL;
2691         info->mti_pill = NULL;
2692         info->mti_exp = NULL;
2693
2694         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2695                 lu_buf_free(&info->mti_big_buf);
2696 }
2697
2698 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2699 {
2700         struct mdt_thread_info  *mti;
2701         struct lustre_capa      *lc;
2702
2703         mti = mdt_th_info(tsi->tsi_env);
2704         LASSERT(mti != NULL);
2705
2706         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2707         if (tsi->tsi_corpus != NULL) {
2708                 struct req_capsule *pill = tsi->tsi_pill;
2709
2710                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2711                 lu_object_get(tsi->tsi_corpus);
2712
2713                 /*
2714                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2715                  * due to mdt_set_capainfo().
2716                  */
2717                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2718                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2719                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2720                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2721                 }
2722         }
2723         mti->mti_body = tsi->tsi_mdt_body;
2724         mti->mti_dlm_req = tsi->tsi_dlm_req;
2725
2726         return mti;
2727 }
2728
2729 int mdt_tgt_connect(struct tgt_session_info *tsi)
2730 {
2731         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2732         int                      rc;
2733
2734         ENTRY;
2735
2736         rc = tgt_connect(tsi);
2737         if (rc != 0)
2738                 RETURN(rc);
2739
2740         rc = mdt_init_idmap(tsi);
2741         if (rc != 0)
2742                 GOTO(err, rc);
2743         RETURN(0);
2744 err:
2745         obd_disconnect(class_export_get(req->rq_export));
2746         return rc;
2747 }
2748
2749 enum mdt_it_code {
2750         MDT_IT_OPEN,
2751         MDT_IT_OCREAT,
2752         MDT_IT_CREATE,
2753         MDT_IT_GETATTR,
2754         MDT_IT_READDIR,
2755         MDT_IT_LOOKUP,
2756         MDT_IT_UNLINK,
2757         MDT_IT_TRUNC,
2758         MDT_IT_GETXATTR,
2759         MDT_IT_LAYOUT,
2760         MDT_IT_QUOTA,
2761         MDT_IT_NR
2762 };
2763
2764 static int mdt_intent_getattr(enum mdt_it_code opcode,
2765                               struct mdt_thread_info *info,
2766                               struct ldlm_lock **,
2767                               __u64);
2768
2769 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2770                                 struct mdt_thread_info *info,
2771                                 struct ldlm_lock **lockp,
2772                                 __u64 flags);
2773
2774 static int mdt_intent_layout(enum mdt_it_code opcode,
2775                              struct mdt_thread_info *info,
2776                              struct ldlm_lock **,
2777                              __u64);
2778 static int mdt_intent_reint(enum mdt_it_code opcode,
2779                             struct mdt_thread_info *info,
2780                             struct ldlm_lock **,
2781                             __u64);
2782
2783 static struct mdt_it_flavor {
2784         const struct req_format *it_fmt;
2785         __u32                    it_flags;
2786         int                    (*it_act)(enum mdt_it_code ,
2787                                          struct mdt_thread_info *,
2788                                          struct ldlm_lock **,
2789                                          __u64);
2790         long                     it_reint;
2791 } mdt_it_flavor[] = {
2792         [MDT_IT_OPEN]     = {
2793                 .it_fmt   = &RQF_LDLM_INTENT,
2794                 /*.it_flags = HABEO_REFERO,*/
2795                 .it_flags = 0,
2796                 .it_act   = mdt_intent_reint,
2797                 .it_reint = REINT_OPEN
2798         },
2799         [MDT_IT_OCREAT]   = {
2800                 .it_fmt   = &RQF_LDLM_INTENT,
2801                 /*
2802                  * OCREAT is not a MUTABOR request as if the file
2803                  * already exists.
2804                  * We do the extra check of OBD_CONNECT_RDONLY in
2805                  * mdt_reint_open() when we really need to create
2806                  * the object.
2807                  */
2808                 .it_flags = 0,
2809                 .it_act   = mdt_intent_reint,
2810                 .it_reint = REINT_OPEN
2811         },
2812         [MDT_IT_CREATE]   = {
2813                 .it_fmt   = &RQF_LDLM_INTENT,
2814                 .it_flags = MUTABOR,
2815                 .it_act   = mdt_intent_reint,
2816                 .it_reint = REINT_CREATE
2817         },
2818         [MDT_IT_GETATTR]  = {
2819                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2820                 .it_flags = HABEO_REFERO,
2821                 .it_act   = mdt_intent_getattr
2822         },
2823         [MDT_IT_READDIR]  = {
2824                 .it_fmt   = NULL,
2825                 .it_flags = 0,
2826                 .it_act   = NULL
2827         },
2828         [MDT_IT_LOOKUP]   = {
2829                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2830                 .it_flags = HABEO_REFERO,
2831                 .it_act   = mdt_intent_getattr
2832         },
2833         [MDT_IT_UNLINK]   = {
2834                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2835                 .it_flags = MUTABOR,
2836                 .it_act   = NULL,
2837                 .it_reint = REINT_UNLINK
2838         },
2839         [MDT_IT_TRUNC]    = {
2840                 .it_fmt   = NULL,
2841                 .it_flags = MUTABOR,
2842                 .it_act   = NULL
2843         },
2844         [MDT_IT_GETXATTR] = {
2845                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2846                 .it_flags = HABEO_CORPUS,
2847                 .it_act   = mdt_intent_getxattr
2848         },
2849         [MDT_IT_LAYOUT] = {
2850                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2851                 .it_flags = 0,
2852                 .it_act   = mdt_intent_layout
2853         }
2854 };
2855
2856 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2857                             struct ldlm_lock **lockp,
2858                             struct ldlm_lock *new_lock,
2859                             struct mdt_lock_handle *lh,
2860                             __u64 flags)
2861 {
2862         struct ptlrpc_request  *req = mdt_info_req(info);
2863         struct ldlm_lock       *lock = *lockp;
2864
2865         /*
2866          * Get new lock only for cases when possible resent did not find any
2867          * lock.
2868          */
2869         if (new_lock == NULL)
2870                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2871
2872         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2873                 lh->mlh_reg_lh.cookie = 0;
2874                 RETURN(0);
2875         }
2876
2877         LASSERTF(new_lock != NULL,
2878                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2879
2880         /*
2881          * If we've already given this lock to a client once, then we should
2882          * have no readers or writers.  Otherwise, we should have one reader
2883          * _or_ writer ref (which will be zeroed below) before returning the
2884          * lock to a client.
2885          */
2886         if (new_lock->l_export == req->rq_export) {
2887                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2888         } else {
2889                 LASSERT(new_lock->l_export == NULL);
2890                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2891         }
2892
2893         *lockp = new_lock;
2894
2895         if (new_lock->l_export == req->rq_export) {
2896                 /*
2897                  * Already gave this to the client, which means that we
2898                  * reconstructed a reply.
2899                  */
2900                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2901                         MSG_RESENT);
2902                 lh->mlh_reg_lh.cookie = 0;
2903                 RETURN(ELDLM_LOCK_REPLACED);
2904         }
2905
2906         /*
2907          * Fixup the lock to be given to the client.
2908          */
2909         lock_res_and_lock(new_lock);
2910         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2911          * possible blocking AST. */
2912         while (new_lock->l_readers > 0) {
2913                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2914                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2915                 new_lock->l_readers--;
2916         }
2917         while (new_lock->l_writers > 0) {
2918                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2919                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2920                 new_lock->l_writers--;
2921         }
2922
2923         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2924         new_lock->l_blocking_ast = lock->l_blocking_ast;
2925         new_lock->l_completion_ast = lock->l_completion_ast;
2926         new_lock->l_remote_handle = lock->l_remote_handle;
2927         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2928
2929         unlock_res_and_lock(new_lock);
2930
2931         cfs_hash_add(new_lock->l_export->exp_lock_hash,
2932                      &new_lock->l_remote_handle,
2933                      &new_lock->l_exp_hash);
2934
2935         LDLM_LOCK_RELEASE(new_lock);
2936         lh->mlh_reg_lh.cookie = 0;
2937
2938         RETURN(ELDLM_LOCK_REPLACED);
2939 }
2940
2941 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2942                                     struct ldlm_lock *new_lock,
2943                                     struct ldlm_lock **old_lock,
2944                                     struct mdt_lock_handle *lh,
2945                                     enum mdt_it_code opcode)
2946 {
2947         struct ptlrpc_request  *req = mdt_info_req(info);
2948         struct obd_export      *exp = req->rq_export;
2949         struct lustre_handle    remote_hdl;
2950         struct ldlm_request    *dlmreq;
2951         struct ldlm_lock       *lock;
2952
2953         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2954                 return;
2955
2956         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2957         remote_hdl = dlmreq->lock_handle[0];
2958         /* If the client does not require open lock, it does not need to
2959          * search lock in exp_lock_hash, since the server thread will
2960          * make sure the lock will be released, and the resend request
2961          * can always re-enqueue the lock */
2962         if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
2963             info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
2964                 /* In the function below, .hs_keycmp resolves to
2965                  * ldlm_export_lock_keycmp() */
2966                 /* coverity[overrun-buffer-val] */
2967                 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
2968                 if (lock) {
2969                         lock_res_and_lock(lock);
2970                         if (lock != new_lock) {
2971                                 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2972                                 lh->mlh_reg_mode = lock->l_granted_mode;
2973
2974                                 LDLM_DEBUG(lock, "Restoring lock cookie");
2975                                 DEBUG_REQ(D_DLMTRACE, req,
2976                                           "restoring lock cookie "LPX64,
2977                                           lh->mlh_reg_lh.cookie);
2978                                 if (old_lock)
2979                                         *old_lock = LDLM_LOCK_GET(lock);
2980                                 cfs_hash_put(exp->exp_lock_hash,
2981                                              &lock->l_exp_hash);
2982                                 unlock_res_and_lock(lock);
2983                                 return;
2984                         }
2985                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
2986                         unlock_res_and_lock(lock);
2987                 }
2988         }
2989         /*
2990          * If the xid matches, then we know this is a resent request, and allow
2991          * it. (It's probably an OPEN, for which we don't send a lock.
2992          */
2993         if (req_xid_is_last(req))
2994                 return;
2995
2996         /*
2997          * This remote handle isn't enqueued, so we never received or processed
2998          * this request.  Clear MSG_RESENT, because it can be handled like any
2999          * normal request now.
3000          */
3001         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3002
3003         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3004                   remote_hdl.cookie);
3005 }
3006
3007 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3008                                 struct mdt_thread_info *info,
3009                                 struct ldlm_lock **lockp,
3010                                 __u64 flags)
3011 {
3012         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3013         struct ldlm_reply      *ldlm_rep = NULL;
3014         int rc, grc;
3015
3016         /*
3017          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3018          * (for the resend case) or a new lock. Below we will use it to
3019          * replace the original lock.
3020          */
3021         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3022         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3023                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3024                 rc = mdt_object_lock(info, info->mti_object, lhc,
3025                                         MDS_INODELOCK_XATTR,
3026                                         MDT_LOCAL_LOCK);
3027                 if (rc)
3028                         return rc;
3029         }
3030
3031         grc = mdt_getxattr(info);
3032
3033         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3034
3035         if (mdt_info_req(info)->rq_repmsg != NULL)
3036                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3037         if (ldlm_rep == NULL)
3038                 RETURN(err_serious(-EFAULT));
3039
3040         ldlm_rep->lock_policy_res2 = grc;
3041
3042         return rc;
3043 }
3044
3045 static int mdt_intent_getattr(enum mdt_it_code opcode,
3046                               struct mdt_thread_info *info,
3047                               struct ldlm_lock **lockp,
3048                               __u64 flags)
3049 {
3050         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3051         struct ldlm_lock       *new_lock = NULL;
3052         __u64                   child_bits;
3053         struct ldlm_reply      *ldlm_rep;
3054         struct ptlrpc_request  *req;
3055         struct mdt_body        *reqbody;
3056         struct mdt_body        *repbody;
3057         int                     rc, rc2;
3058         ENTRY;
3059
3060         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3061         LASSERT(reqbody);
3062
3063         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3064         LASSERT(repbody);
3065
3066         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3067         repbody->eadatasize = 0;
3068         repbody->aclsize = 0;
3069
3070         switch (opcode) {
3071         case MDT_IT_LOOKUP:
3072                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3073                 break;
3074         case MDT_IT_GETATTR:
3075                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3076                              MDS_INODELOCK_PERM;
3077                 break;
3078         default:
3079                 CERROR("Unsupported intent (%d)\n", opcode);
3080                 GOTO(out_shrink, rc = -EINVAL);
3081         }
3082
3083         rc = mdt_init_ucred(info, reqbody);
3084         if (rc)
3085                 GOTO(out_shrink, rc);
3086
3087         req = info->mti_pill->rc_req;
3088         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3089         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3090
3091         /* Get lock from request for possible resent case. */
3092         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3093
3094         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3095         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3096
3097         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3098                 ldlm_rep->lock_policy_res2 = 0;
3099         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3100             ldlm_rep->lock_policy_res2) {
3101                 lhc->mlh_reg_lh.cookie = 0ull;
3102                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3103         }
3104
3105         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3106         EXIT;
3107 out_ucred:
3108         mdt_exit_ucred(info);
3109 out_shrink:
3110         mdt_client_compatibility(info);
3111         rc2 = mdt_fix_reply(info);
3112         if (rc == 0)
3113                 rc = rc2;
3114         return rc;
3115 }
3116
3117 static int mdt_intent_layout(enum mdt_it_code opcode,
3118                              struct mdt_thread_info *info,
3119                              struct ldlm_lock **lockp,
3120                              __u64 flags)
3121 {
3122         struct layout_intent *layout;
3123         struct lu_fid *fid;
3124         struct mdt_object *obj = NULL;
3125         struct md_object *child = NULL;
3126         int rc;
3127         ENTRY;
3128
3129         if (opcode != MDT_IT_LAYOUT) {
3130                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3131                         opcode);
3132                 RETURN(-EINVAL);
3133         }
3134
3135         fid = &info->mti_tmp_fid2;
3136         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3137
3138         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3139         if (IS_ERR(obj))
3140                 RETURN(PTR_ERR(obj));
3141
3142         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3143                 child = mdt_object_child(obj);
3144
3145                 /* get the length of lsm */
3146                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3147                                   XATTR_NAME_LOV);
3148
3149                 if (rc > info->mti_mdt->mdt_max_mdsize)
3150                         info->mti_mdt->mdt_max_mdsize = rc;
3151         }
3152
3153         mdt_object_put(info->mti_env, obj);
3154
3155         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3156         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3157                         ldlm_lvbo_size(*lockp));
3158         rc = req_capsule_server_pack(info->mti_pill);
3159         if (rc != 0)
3160                 RETURN(-EINVAL);
3161
3162         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3163         LASSERT(layout != NULL);
3164         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3165                 /* return to normal ldlm handling */
3166                 RETURN(0);
3167
3168         CERROR("%s: Unsupported layout intent (%d)\n",
3169                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3170         RETURN(-EINVAL);
3171 }
3172
3173 static int mdt_intent_reint(enum mdt_it_code opcode,
3174                             struct mdt_thread_info *info,
3175                             struct ldlm_lock **lockp,
3176                             __u64 flags)
3177 {
3178         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3179         struct ldlm_reply      *rep = NULL;
3180         long                    opc;
3181         int                     rc;
3182
3183         static const struct req_format *intent_fmts[REINT_MAX] = {
3184                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3185                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3186         };
3187
3188         ENTRY;
3189
3190         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3191         if (opc < 0)
3192                 RETURN(opc);
3193
3194         if (mdt_it_flavor[opcode].it_reint != opc) {
3195                 CERROR("Reint code %ld doesn't match intent: %d\n",
3196                        opc, opcode);
3197                 RETURN(err_serious(-EPROTO));
3198         }
3199
3200         /* Get lock from request for possible resent case. */
3201         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3202
3203         rc = mdt_reint_internal(info, lhc, opc);
3204
3205         /* Check whether the reply has been packed successfully. */
3206         if (mdt_info_req(info)->rq_repmsg != NULL)
3207                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3208         if (rep == NULL)
3209                 RETURN(err_serious(-EFAULT));
3210
3211         /* MDC expects this in any case */
3212         if (rc != 0)
3213                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3214
3215         /* the open lock or the lock for cross-ref object should be
3216          * returned to the client */
3217         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3218                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3219                 rep->lock_policy_res2 = 0;
3220                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3221                 RETURN(rc);
3222         }
3223
3224         rep->lock_policy_res2 = clear_serious(rc);
3225
3226         if (rep->lock_policy_res2 == -ENOENT &&
3227             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3228                 rep->lock_policy_res2 = 0;
3229
3230         if (rc == -ENOTCONN || rc == -ENODEV ||
3231             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3232                 /*
3233                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3234                  * will be returned by rq_status, and client at ptlrpc layer
3235                  * will detect this, then disconnect, reconnect the import
3236                  * immediately, instead of impacting the following the rpc.
3237                  */
3238                 lhc->mlh_reg_lh.cookie = 0ull;
3239                 RETURN(rc);
3240         } else {
3241                 /*
3242                  * For other cases, the error will be returned by intent.
3243                  * and client will retrieve the result from intent.
3244                  */
3245                  /*
3246                   * FIXME: when open lock is finished, that should be
3247                   * checked here.
3248                   */
3249                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3250                         LASSERTF(rc == 0, "Error occurred but lock handle "
3251                                  "is still in use, rc = %d\n", rc);
3252                         rep->lock_policy_res2 = 0;
3253                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3254                         RETURN(rc);
3255                 } else {
3256                         lhc->mlh_reg_lh.cookie = 0ull;
3257                         RETURN(ELDLM_LOCK_ABORTED);
3258                 }
3259         }
3260 }
3261
3262 static int mdt_intent_code(long itcode)
3263 {
3264         int rc;
3265
3266         switch(itcode) {
3267         case IT_OPEN:
3268                 rc = MDT_IT_OPEN;
3269                 break;
3270         case IT_OPEN|IT_CREAT:
3271                 rc = MDT_IT_OCREAT;
3272                 break;
3273         case IT_CREAT:
3274                 rc = MDT_IT_CREATE;
3275                 break;
3276         case IT_READDIR:
3277                 rc = MDT_IT_READDIR;
3278                 break;
3279         case IT_GETATTR:
3280                 rc = MDT_IT_GETATTR;
3281                 break;
3282         case IT_LOOKUP:
3283                 rc = MDT_IT_LOOKUP;
3284                 break;
3285         case IT_UNLINK:
3286                 rc = MDT_IT_UNLINK;
3287                 break;
3288         case IT_TRUNC:
3289                 rc = MDT_IT_TRUNC;
3290                 break;
3291         case IT_GETXATTR:
3292                 rc = MDT_IT_GETXATTR;
3293                 break;
3294         case IT_LAYOUT:
3295                 rc = MDT_IT_LAYOUT;
3296                 break;
3297         case IT_QUOTA_DQACQ:
3298         case IT_QUOTA_CONN:
3299                 rc = MDT_IT_QUOTA;
3300                 break;
3301         default:
3302                 CERROR("Unknown intent opcode: %ld\n", itcode);
3303                 rc = -EINVAL;
3304                 break;
3305         }
3306         return rc;
3307 }
3308
3309 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3310                           struct ldlm_lock **lockp, __u64 flags)
3311 {
3312         struct req_capsule   *pill;
3313         struct mdt_it_flavor *flv;
3314         int opc;
3315         int rc;
3316         ENTRY;
3317
3318         opc = mdt_intent_code(itopc);
3319         if (opc < 0)
3320                 RETURN(-EINVAL);
3321
3322         pill = info->mti_pill;
3323
3324         if (opc == MDT_IT_QUOTA) {
3325                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3326
3327                 if (qmt == NULL)
3328                         RETURN(-EOPNOTSUPP);
3329
3330                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3331                 /* pass the request to quota master */
3332                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3333                                                  mdt_info_req(info), lockp,
3334                                                  flags);
3335                 RETURN(rc);
3336         }
3337
3338         flv  = &mdt_it_flavor[opc];
3339         if (flv->it_fmt != NULL)
3340                 req_capsule_extend(pill, flv->it_fmt);
3341
3342         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3343         if (rc == 0) {
3344                 struct ptlrpc_request *req = mdt_info_req(info);
3345                 if (flv->it_flags & MUTABOR &&
3346                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3347                         RETURN(-EROFS);
3348         }
3349         if (rc == 0 && flv->it_act != NULL) {
3350                 struct ldlm_reply *rep;
3351
3352                 /* execute policy */
3353                 rc = flv->it_act(opc, info, lockp, flags);
3354
3355                 /* Check whether the reply has been packed successfully. */
3356                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3357                         rep = req_capsule_server_get(info->mti_pill,
3358                                                      &RMF_DLM_REP);
3359                         rep->lock_policy_res2 =
3360                                 ptlrpc_status_hton(rep->lock_policy_res2);
3361                 }
3362         } else {
3363                 rc = -EPROTO;
3364         }
3365         RETURN(rc);
3366 }
3367
3368 static int mdt_intent_policy(struct ldlm_namespace *ns,
3369                              struct ldlm_lock **lockp, void *req_cookie,
3370                              ldlm_mode_t mode, __u64 flags, void *data)
3371 {
3372         struct tgt_session_info *tsi;
3373         struct mdt_thread_info  *info;
3374         struct ptlrpc_request   *req  =  req_cookie;
3375         struct ldlm_intent      *it;
3376         struct req_capsule      *pill;
3377         int rc;
3378
3379         ENTRY;
3380
3381         LASSERT(req != NULL);
3382
3383         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3384
3385         info = tsi2mdt_info(tsi);
3386         LASSERT(info != NULL);
3387         pill = info->mti_pill;
3388         LASSERT(pill->rc_req == req);
3389
3390         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3391                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3392                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3393                 if (it != NULL) {
3394                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3395                         if (rc == 0)
3396                                 rc = ELDLM_OK;
3397
3398                         /* Lock without inodebits makes no sense and will oops
3399                          * later in ldlm. Let's check it now to see if we have
3400                          * ibits corrupted somewhere in mdt_intent_opc().
3401                          * The case for client miss to set ibits has been
3402                          * processed by others. */
3403                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3404                                         lr_type == LDLM_IBITS,
3405                                      info->mti_dlm_req->lock_desc.\
3406                                         l_policy_data.l_inodebits.bits != 0));
3407                 } else
3408                         rc = err_serious(-EFAULT);
3409         } else {
3410                 /* No intent was provided */
3411                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3412                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3413                 rc = req_capsule_server_pack(pill);
3414                 if (rc)
3415                         rc = err_serious(rc);
3416         }
3417         mdt_thread_info_fini(info);
3418         RETURN(rc);
3419 }
3420
3421 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3422 {
3423         struct seq_server_site  *ss = mdt_seq_site(mdt);
3424
3425         if (ss->ss_node_id == 0)
3426                 return;
3427
3428         if (ss->ss_client_seq != NULL) {
3429                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3430                 ss->ss_client_seq->lcs_exp = NULL;
3431         }
3432
3433         if (ss->ss_server_fld != NULL) {
3434                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3435                 ss->ss_server_fld->lsf_control_exp = NULL;
3436         }
3437 }
3438
3439 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3440 {
3441         struct seq_server_site *ss = mdt_seq_site(mdt);
3442
3443         if (ss == NULL)
3444                 return;
3445
3446         if (ss->ss_server_seq == NULL)
3447                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3448
3449         return;
3450 }
3451
3452 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3453 {
3454         mdt_seq_fini_cli(mdt);
3455         mdt_deregister_seq_exp(mdt);
3456
3457         return seq_site_fini(env, mdt_seq_site(mdt));
3458 }
3459
3460 /**
3461  * It will retrieve its FLDB entries from MDT0, and it only happens
3462  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3463  * and it needs to refresh FLDB from the MDT0.
3464  **/
3465 static int mdt_register_lwp_callback(void *data)
3466 {
3467         struct lu_env           env;
3468         struct mdt_device       *mdt = data;
3469         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3470         int                     rc;
3471         ENTRY;
3472
3473         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3474
3475         if (!likely(fld->lsf_new))
3476                 RETURN(0);
3477
3478         rc = lu_env_init(&env, LCT_MD_THREAD);
3479         if (rc) {
3480                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3481                 RETURN(rc);
3482         }
3483
3484         rc = fld_update_from_controller(&env, fld);
3485         if (rc != 0) {
3486                 CERROR("%s: cannot update controller: rc = %d\n",
3487                        mdt_obd_name(mdt), rc);
3488                 GOTO(out, rc);
3489         }
3490 out:
3491         lu_env_fini(&env);
3492         RETURN(rc);
3493 }
3494
3495 static int mdt_register_seq_exp(struct mdt_device *mdt)
3496 {
3497         struct seq_server_site  *ss = mdt_seq_site(mdt);
3498         char                    *lwp_name = NULL;
3499         int                     rc;
3500
3501         if (ss->ss_node_id == 0)
3502                 return 0;
3503
3504         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3505         if (lwp_name == NULL)
3506                 GOTO(out_free, rc = -ENOMEM);
3507
3508         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3509         if (rc != 0)
3510                 GOTO(out_free, rc);
3511
3512         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3513                                       NULL, NULL);
3514         if (rc != 0)
3515                 GOTO(out_free, rc);
3516
3517         rc = lustre_register_lwp_item(lwp_name,
3518                                       &ss->ss_server_fld->lsf_control_exp,
3519                                       mdt_register_lwp_callback, mdt);
3520         if (rc != 0) {
3521                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3522                 ss->ss_client_seq->lcs_exp = NULL;
3523                 GOTO(out_free, rc);
3524         }
3525 out_free:
3526         if (lwp_name != NULL)
3527                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3528
3529         return rc;
3530 }
3531
3532 /*
3533  * Init client sequence manager which is used by local MDS to talk to sequence
3534  * controller on remote node.
3535  */
3536 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3537 {
3538         struct seq_server_site  *ss = mdt_seq_site(mdt);
3539         int                     rc;
3540         char                    *prefix;
3541         ENTRY;
3542
3543         /* check if this is adding the first MDC and controller is not yet
3544          * initialized. */
3545         OBD_ALLOC_PTR(ss->ss_client_seq);
3546         if (ss->ss_client_seq == NULL)
3547                 RETURN(-ENOMEM);
3548
3549         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3550         if (prefix == NULL) {
3551                 OBD_FREE_PTR(ss->ss_client_seq);
3552                 ss->ss_client_seq = NULL;
3553                 RETURN(-ENOMEM);
3554         }
3555
3556         /* Note: seq_client_fini will be called in seq_site_fini */
3557         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3558         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3559                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3560                                                             NULL);
3561         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3562         if (rc != 0) {
3563                 OBD_FREE_PTR(ss->ss_client_seq);
3564                 ss->ss_client_seq = NULL;
3565                 RETURN(rc);
3566         }
3567
3568         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3569
3570         RETURN(rc);
3571 }
3572
3573 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3574 {
3575         struct seq_server_site  *ss;
3576         int                     rc;
3577         ENTRY;
3578
3579         ss = mdt_seq_site(mdt);
3580         /* init sequence controller server(MDT0) */
3581         if (ss->ss_node_id == 0) {
3582                 OBD_ALLOC_PTR(ss->ss_control_seq);
3583                 if (ss->ss_control_seq == NULL)
3584                         RETURN(-ENOMEM);
3585
3586                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3587                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3588                                      ss);
3589                 if (rc)
3590                         GOTO(out_seq_fini, rc);
3591         }
3592
3593         /* Init normal sequence server */
3594         OBD_ALLOC_PTR(ss->ss_server_seq);
3595         if (ss->ss_server_seq == NULL)
3596                 GOTO(out_seq_fini, rc = -ENOMEM);
3597
3598         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3599                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3600         if (rc)
3601                 GOTO(out_seq_fini, rc);
3602
3603         /* init seq client for seq server to talk to seq controller(MDT0) */
3604         rc = mdt_seq_init_cli(env, mdt);
3605         if (rc != 0)
3606                 GOTO(out_seq_fini, rc);
3607
3608         if (ss->ss_node_id != 0)
3609                 /* register controler export through lwp */
3610                 rc = mdt_register_seq_exp(mdt);
3611
3612         EXIT;
3613 out_seq_fini:
3614         if (rc)
3615                 mdt_seq_fini(env, mdt);
3616
3617         return rc;
3618 }
3619
3620 /*
3621  * FLD wrappers
3622  */
3623 static int mdt_fld_fini(const struct lu_env *env,
3624                         struct mdt_device *m)
3625 {
3626         struct seq_server_site *ss = mdt_seq_site(m);
3627         ENTRY;
3628
3629         if (ss && ss->ss_server_fld) {
3630                 fld_server_fini(env, ss->ss_server_fld);
3631                 OBD_FREE_PTR(ss->ss_server_fld);
3632                 ss->ss_server_fld = NULL;
3633         }
3634
3635         RETURN(0);
3636 }
3637
3638 static int mdt_fld_init(const struct lu_env *env,
3639                         const char *uuid,
3640                         struct mdt_device *m)
3641 {
3642         struct seq_server_site *ss;
3643         int rc;
3644         ENTRY;
3645
3646         ss = mdt_seq_site(m);
3647
3648         OBD_ALLOC_PTR(ss->ss_server_fld);
3649         if (ss->ss_server_fld == NULL)
3650                 RETURN(rc = -ENOMEM);
3651
3652         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3653                              LU_SEQ_RANGE_MDT);
3654         if (rc) {
3655                 OBD_FREE_PTR(ss->ss_server_fld);
3656                 ss->ss_server_fld = NULL;
3657                 RETURN(rc);
3658         }
3659
3660         RETURN(0);
3661 }
3662
3663 static void mdt_stack_pre_fini(const struct lu_env *env,
3664                            struct mdt_device *m, struct lu_device *top)
3665 {
3666         struct obd_device       *obd;
3667         struct lustre_cfg_bufs  *bufs;
3668         struct lustre_cfg       *lcfg;
3669         struct mdt_thread_info  *info;
3670         ENTRY;
3671
3672         LASSERT(top);
3673
3674         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3675         LASSERT(info != NULL);
3676
3677         bufs = &info->mti_u.bufs;
3678
3679         LASSERT(m->mdt_child_exp);
3680         LASSERT(m->mdt_child_exp->exp_obd);
3681         obd = m->mdt_child_exp->exp_obd;
3682
3683         /* process cleanup, pass mdt obd name to get obd umount flags */
3684         /* XXX: this is needed because all layers are referenced by
3685          * objects (some of them are pinned by osd, for example *
3686          * the proper solution should be a model where object used
3687          * by osd only doesn't have mdt/mdd slices -bzzz */
3688         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3689         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3690         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3691         if (!lcfg) {
3692                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3693                 return;
3694         }
3695         top->ld_ops->ldo_process_config(env, top, lcfg);
3696         lustre_cfg_free(lcfg);
3697         EXIT;
3698 }
3699
3700 static void mdt_stack_fini(const struct lu_env *env,
3701                            struct mdt_device *m, struct lu_device *top)
3702 {
3703         struct obd_device       *obd = mdt2obd_dev(m);
3704         struct lustre_cfg_bufs  *bufs;
3705         struct lustre_cfg       *lcfg;
3706         struct mdt_thread_info  *info;
3707         char                     flags[3] = "";
3708         ENTRY;
3709
3710         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3711         LASSERT(info != NULL);
3712
3713         lu_dev_del_linkage(top->ld_site, top);
3714
3715         lu_site_purge(env, top->ld_site, -1);
3716
3717         bufs = &info->mti_u.bufs;
3718         /* process cleanup, pass mdt obd name to get obd umount flags */
3719         /* another purpose is to let all layers to release their objects */
3720         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3721         if (obd->obd_force)
3722                 strcat(flags, "F");
3723         if (obd->obd_fail)
3724                 strcat(flags, "A");
3725         lustre_cfg_bufs_set_string(bufs, 1, flags);
3726         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3727         if (!lcfg) {
3728                 CERROR("Cannot alloc lcfg!\n");
3729                 return;
3730         }
3731         LASSERT(top);
3732         top->ld_ops->ldo_process_config(env, top, lcfg);
3733         lustre_cfg_free(lcfg);
3734
3735         lu_site_purge(env, top->ld_site, -1);
3736
3737         m->mdt_child = NULL;
3738         m->mdt_bottom = NULL;
3739
3740         obd_disconnect(m->mdt_child_exp);
3741         m->mdt_child_exp = NULL;
3742
3743         obd_disconnect(m->mdt_bottom_exp);
3744         m->mdt_child_exp = NULL;
3745 }
3746
3747 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3748                                const char *next, struct obd_export **exp)
3749 {
3750         struct obd_connect_data *data = NULL;
3751         struct obd_device       *obd;
3752         int                      rc;
3753         ENTRY;
3754
3755         OBD_ALLOC_PTR(data);
3756         if (data == NULL)
3757                 GOTO(out, rc = -ENOMEM);
3758
3759         obd = class_name2obd(next);
3760         if (obd == NULL) {
3761                 CERROR("%s: can't locate next device: %s\n",
3762                        mdt_obd_name(m), next);
3763                 GOTO(out, rc = -ENOTCONN);
3764         }
3765
3766         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3767         data->ocd_version = LUSTRE_VERSION_CODE;
3768
3769         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3770         if (rc) {
3771                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3772                        mdt_obd_name(m), next, rc);
3773                 GOTO(out, rc);
3774         }
3775
3776 out:
3777         if (data)
3778                 OBD_FREE_PTR(data);
3779         RETURN(rc);
3780 }
3781
3782 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3783                           struct lustre_cfg *cfg)
3784 {
3785         char                   *dev = lustre_cfg_string(cfg, 0);
3786         int                     rc, name_size, uuid_size;
3787         char                   *name, *uuid, *p;
3788         struct lustre_cfg_bufs *bufs;
3789         struct lustre_cfg      *lcfg;
3790         struct obd_device      *obd;
3791         struct lustre_profile  *lprof;
3792         struct lu_site         *site;
3793         ENTRY;
3794
3795         /* in 1.8 we had the only device in the stack - MDS.
3796          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3797          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3798          * to generate names and setup MDT, MDD. MDT will be using
3799          * generated name to connect to MDD. for MDD the next device
3800          * will be LOD with name taken from so called "profile" which
3801          * is generated by mount_option line
3802          *
3803          * 1.8 MGS generates config. commands like this:
3804          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3805          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3806          * 2.0 MGS generates config. commands like this:
3807          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3808          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3809          *                    3:lustre-MDT0000-mdtlov  4:f
3810          *
3811          * we generate MDD name from MDT one, just replacing T with D
3812          *
3813          * after all the preparations, the logical equivalent will be
3814          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3815          *                    3:lustre-MDT0000-mdtlov  4:f
3816          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3817          *                    3:lustre-MDD0000  4:f
3818          *
3819          *  notice we build the stack from down to top: MDD first, then MDT */
3820
3821         name_size = MAX_OBD_NAME;
3822         uuid_size = MAX_OBD_NAME;
3823
3824         OBD_ALLOC(name, name_size);
3825         OBD_ALLOC(uuid, uuid_size);
3826         if (name == NULL || uuid == NULL)
3827                 GOTO(cleanup_mem, rc = -ENOMEM);
3828
3829         OBD_ALLOC_PTR(bufs);
3830         if (!bufs)
3831                 GOTO(cleanup_mem, rc = -ENOMEM);
3832
3833         strcpy(name, dev);
3834         p = strstr(name, "-MDT");
3835         if (p == NULL)
3836                 GOTO(free_bufs, rc = -ENOMEM);
3837         p[3] = 'D';
3838
3839         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3840
3841         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3842         if (lprof == NULL || lprof->lp_dt == NULL) {
3843                 CERROR("can't find the profile: %s\n",
3844                        lustre_cfg_string(cfg, 0));
3845                 GOTO(free_bufs, rc = -EINVAL);
3846         }
3847
3848         lustre_cfg_bufs_reset(bufs, name);
3849         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3850         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3851         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3852
3853         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3854         if (!lcfg)
3855                 GOTO(free_bufs, rc = -ENOMEM);
3856
3857         rc = class_attach(lcfg);
3858         if (rc)
3859                 GOTO(lcfg_cleanup, rc);
3860
3861         obd = class_name2obd(name);
3862         if (!obd) {
3863                 CERROR("Can not find obd %s (%s in config)\n",
3864                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3865                 GOTO(class_detach, rc = -EINVAL);
3866         }
3867
3868         lustre_cfg_free(lcfg);
3869
3870         lustre_cfg_bufs_reset(bufs, name);
3871         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3872         lustre_cfg_bufs_set_string(bufs, 2, dev);
3873         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3874
3875         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3876
3877         rc = class_setup(obd, lcfg);
3878         if (rc)
3879                 GOTO(class_detach, rc);
3880
3881         /* connect to MDD we just setup */
3882         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3883         if (rc)
3884                 GOTO(class_detach, rc);
3885
3886         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3887         LASSERT(site);
3888         LASSERT(mdt_lu_site(mdt) == NULL);
3889         mdt->mdt_lu_dev.ld_site = site;
3890         site->ls_top_dev = &mdt->mdt_lu_dev;
3891         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3892
3893
3894         /* now connect to bottom OSD */
3895         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3896         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3897         if (rc)
3898                 GOTO(class_detach, rc);
3899         mdt->mdt_bottom =
3900                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3901
3902
3903         rc = lu_env_refill((struct lu_env *)env);
3904         if (rc != 0)
3905                 CERROR("Failure to refill session: '%d'\n", rc);
3906
3907         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3908
3909         EXIT;
3910 class_detach:
3911         if (rc)
3912                 class_detach(obd, lcfg);
3913 lcfg_cleanup:
3914         lustre_cfg_free(lcfg);
3915 free_bufs:
3916         OBD_FREE_PTR(bufs);
3917 cleanup_mem:
3918         if (name)
3919                 OBD_FREE(name, name_size);
3920         if (uuid)
3921                 OBD_FREE(uuid, uuid_size);
3922         RETURN(rc);
3923 }
3924
3925 /* setup quota master target on MDT0 */
3926 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3927                           struct lustre_cfg *cfg)
3928 {
3929         struct obd_device       *obd;
3930         char                    *dev = lustre_cfg_string(cfg, 0);
3931         char                    *qmtname, *uuid, *p;
3932         struct lustre_cfg_bufs  *bufs;
3933         struct lustre_cfg       *lcfg;
3934         struct lustre_profile   *lprof;
3935         struct obd_connect_data *data;
3936         int                      rc;
3937         ENTRY;
3938
3939         LASSERT(mdt->mdt_qmt_exp == NULL);
3940         LASSERT(mdt->mdt_qmt_dev == NULL);
3941
3942         /* quota master is on MDT0 only for now */
3943         if (mdt->mdt_seq_site.ss_node_id != 0)
3944                 RETURN(0);
3945
3946         /* MGS generates config commands which look as follows:
3947          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3948          *                    3:lustre-MDT0000-mdtlov  4:f
3949          *
3950          * We generate the QMT name from the MDT one, just replacing MD with QM
3951          * after all the preparations, the logical equivalent will be:
3952          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
3953          *                    3:lustre-MDT0000-osd  4:f */
3954         OBD_ALLOC(qmtname, MAX_OBD_NAME);
3955         OBD_ALLOC(uuid, UUID_MAX);
3956         OBD_ALLOC_PTR(bufs);
3957         OBD_ALLOC_PTR(data);
3958         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
3959                 GOTO(cleanup_mem, rc = -ENOMEM);
3960
3961         strcpy(qmtname, dev);
3962         p = strstr(qmtname, "-MDT");
3963         if (p == NULL)
3964                 GOTO(cleanup_mem, rc = -ENOMEM);
3965         /* replace MD with QM */
3966         p[1] = 'Q';
3967         p[2] = 'M';
3968
3969         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
3970
3971         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3972         if (lprof == NULL || lprof->lp_dt == NULL) {
3973                 CERROR("can't find profile for %s\n",
3974                        lustre_cfg_string(cfg, 0));
3975                 GOTO(cleanup_mem, rc = -EINVAL);
3976         }
3977
3978         lustre_cfg_bufs_reset(bufs, qmtname);
3979         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
3980         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3981         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3982
3983         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3984         if (!lcfg)
3985                 GOTO(cleanup_mem, rc = -ENOMEM);
3986
3987         rc = class_attach(lcfg);
3988         if (rc)
3989                 GOTO(lcfg_cleanup, rc);
3990
3991         obd = class_name2obd(qmtname);
3992         if (!obd) {
3993                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
3994                        lustre_cfg_string(cfg, 0));
3995                 GOTO(class_detach, rc = -EINVAL);
3996         }
3997
3998         lustre_cfg_free(lcfg);
3999
4000         lustre_cfg_bufs_reset(bufs, qmtname);
4001         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4002         lustre_cfg_bufs_set_string(bufs, 2, dev);
4003
4004         /* for quota, the next device should be the OSD device */
4005         lustre_cfg_bufs_set_string(bufs, 3,
4006                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4007
4008         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4009
4010         rc = class_setup(obd, lcfg);
4011         if (rc)
4012                 GOTO(class_detach, rc);
4013
4014         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4015
4016         /* configure local quota objects */
4017         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4018                                                    &mdt->mdt_lu_dev,
4019                                                    mdt->mdt_qmt_dev);
4020         if (rc)
4021                 GOTO(class_cleanup, rc);
4022
4023         /* connect to quota master target */
4024         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4025         data->ocd_version = LUSTRE_VERSION_CODE;
4026         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4027                          data, NULL);
4028         if (rc) {
4029                 CERROR("cannot connect to quota master device %s (%d)\n",
4030                        qmtname, rc);
4031                 GOTO(class_cleanup, rc);
4032         }
4033
4034         EXIT;
4035 class_cleanup:
4036         if (rc) {
4037                 class_manual_cleanup(obd);
4038                 mdt->mdt_qmt_dev = NULL;
4039         }
4040 class_detach:
4041         if (rc)
4042                 class_detach(obd, lcfg);
4043 lcfg_cleanup:
4044         lustre_cfg_free(lcfg);
4045 cleanup_mem:
4046         if (bufs)
4047                 OBD_FREE_PTR(bufs);
4048         if (qmtname)
4049                 OBD_FREE(qmtname, MAX_OBD_NAME);
4050         if (uuid)
4051                 OBD_FREE(uuid, UUID_MAX);
4052         if (data)
4053                 OBD_FREE_PTR(data);
4054         return rc;
4055 }
4056
4057 /* Shutdown quota master target associated with mdt */
4058 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4059 {
4060         ENTRY;
4061
4062         if (mdt->mdt_qmt_exp == NULL)
4063                 RETURN_EXIT;
4064         LASSERT(mdt->mdt_qmt_dev != NULL);
4065
4066         /* the qmt automatically shuts down when the mdt disconnects */
4067         obd_disconnect(mdt->mdt_qmt_exp);
4068         mdt->mdt_qmt_exp = NULL;
4069         mdt->mdt_qmt_dev = NULL;
4070         EXIT;
4071 }
4072
4073 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4074  * for now. This will be removed along with converting rest of MDT code
4075  * to use tgt_session_info */
4076 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4077 {
4078         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4079         int                      rc;
4080
4081         rc = mdt_getxattr(info);
4082
4083         mdt_thread_info_fini(info);
4084         return rc;
4085 }
4086
4087 static struct tgt_handler mdt_tgt_handlers[] = {
4088 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4089                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4090                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4091 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4092                 0,                      MDS_DISCONNECT, tgt_disconnect,
4093                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4094 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4095                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4096                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4097 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4098 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4099 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4100 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4101                                                         mdt_getattr_name),
4102 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4103 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4104 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4105 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4106 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4107                                                         mdt_done_writing),
4108 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4109 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4110 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR,  mdt_is_subdir),
4111 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4112 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4113                                                         mdt_hsm_progress),
4114 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4115                                                         mdt_hsm_ct_register),
4116 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4117                                                         mdt_hsm_ct_unregister),
4118 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4119                                                         mdt_hsm_state_get),
4120 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4121                                                         mdt_hsm_state_set),
4122 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4123 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4124                                                         mdt_hsm_request),
4125 TGT_MDT_HDL(HABEO_CORPUS|HABEO_REFERO | MUTABOR, MDS_SWAP_LAYOUTS,
4126                                                         mdt_swap_layouts)
4127 };
4128
4129 static struct tgt_handler mdt_sec_ctx_ops[] = {
4130 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4131 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4132 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4133 };
4134
4135 static struct tgt_handler mdt_quota_ops[] = {
4136 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4137 };
4138
4139 static struct tgt_opc_slice mdt_common_slice[] = {
4140         {
4141                 .tos_opc_start  = MDS_FIRST_OPC,
4142                 .tos_opc_end    = MDS_LAST_OPC,
4143                 .tos_hs         = mdt_tgt_handlers
4144         },
4145         {
4146                 .tos_opc_start  = OBD_FIRST_OPC,
4147                 .tos_opc_end    = OBD_LAST_OPC,
4148                 .tos_hs         = tgt_obd_handlers
4149         },
4150         {
4151                 .tos_opc_start  = LDLM_FIRST_OPC,
4152                 .tos_opc_end    = LDLM_LAST_OPC,
4153                 .tos_hs         = tgt_dlm_handlers
4154         },
4155         {
4156                 .tos_opc_start  = SEC_FIRST_OPC,
4157                 .tos_opc_end    = SEC_LAST_OPC,
4158                 .tos_hs         = mdt_sec_ctx_ops
4159         },
4160         {
4161                 .tos_opc_start  = UPDATE_OBJ,
4162                 .tos_opc_end    = UPDATE_LAST_OPC,
4163                 .tos_hs         = tgt_out_handlers
4164         },
4165         {
4166                 .tos_opc_start  = FLD_FIRST_OPC,
4167                 .tos_opc_end    = FLD_LAST_OPC,
4168                 .tos_hs         = fld_handlers
4169         },
4170         {
4171                 .tos_opc_start  = SEQ_FIRST_OPC,
4172                 .tos_opc_end    = SEQ_LAST_OPC,
4173                 .tos_hs         = seq_handlers
4174         },
4175         {
4176                 .tos_opc_start  = QUOTA_DQACQ,
4177                 .tos_opc_end    = QUOTA_LAST_OPC,
4178                 .tos_hs         = mdt_quota_ops
4179         },
4180         {
4181                 .tos_opc_start  = LLOG_FIRST_OPC,
4182                 .tos_opc_end    = LLOG_LAST_OPC,
4183                 .tos_hs         = tgt_llog_handlers
4184         },
4185         {
4186                 .tos_opc_start  = LFSCK_FIRST_OPC,
4187                 .tos_opc_end    = LFSCK_LAST_OPC,
4188                 .tos_hs         = tgt_lfsck_handlers
4189         },
4190
4191         {
4192                 .tos_hs         = NULL
4193         }
4194 };
4195
4196 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4197 {
4198         struct md_device        *next = m->mdt_child;
4199         struct lu_device        *d    = &m->mdt_lu_dev;
4200         struct obd_device       *obd  = mdt2obd_dev(m);
4201         struct lfsck_stop        stop;
4202         ENTRY;
4203
4204         target_recovery_fini(obd);
4205         ping_evictor_stop();
4206         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4207
4208         if (m->mdt_opts.mo_coordinator)
4209                 mdt_hsm_cdt_stop(m);
4210
4211         mdt_hsm_cdt_fini(m);
4212
4213         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4214         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4215         obd_exports_barrier(obd);
4216         obd_zombie_barrier();
4217
4218         mdt_procfs_fini(m);
4219
4220         tgt_fini(env, &m->mdt_lut);
4221         mdt_fs_cleanup(env, m);
4222         upcall_cache_cleanup(m->mdt_identity_cache);
4223         m->mdt_identity_cache = NULL;
4224
4225         if (m->mdt_namespace != NULL) {
4226                 ldlm_namespace_free(m->mdt_namespace, NULL,
4227                                     d->ld_obd->obd_force);
4228                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4229         }
4230
4231         mdt_quota_fini(env, m);
4232
4233         cfs_free_nidlist(&m->mdt_nosquash_nids);
4234         if (m->mdt_nosquash_str) {
4235                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4236                 m->mdt_nosquash_str = NULL;
4237                 m->mdt_nosquash_strlen = 0;
4238         }
4239
4240         stop.ls_status = LS_PAUSED;
4241         stop.ls_flags = 0;
4242         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4243
4244         mdt_seq_fini(env, m);
4245         mdt_fld_fini(env, m);
4246
4247         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4248         cfs_timer_disarm(&m->mdt_ck_timer);
4249         mdt_ck_thread_stop(m);
4250
4251         /*
4252          * Finish the stack
4253          */
4254         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4255
4256         LASSERT(atomic_read(&d->ld_ref) == 0);
4257
4258         server_put_mount(mdt_obd_name(m));
4259
4260         EXIT;
4261 }
4262
4263 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4264
4265 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4266                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4267 {
4268         struct mdt_thread_info    *info;
4269         struct obd_device         *obd;
4270         const char                *dev = lustre_cfg_string(cfg, 0);
4271         const char                *num = lustre_cfg_string(cfg, 2);
4272         struct lustre_mount_info  *lmi = NULL;
4273         struct lustre_sb_info     *lsi;
4274         struct lu_site            *s;
4275         struct seq_server_site    *ss_site;
4276         const char                *identity_upcall = "NONE";
4277         struct md_device          *next;
4278         int                        rc;
4279         long                       node_id;
4280         mntopt_t                   mntopts;
4281         ENTRY;
4282
4283         lu_device_init(&m->mdt_lu_dev, ldt);
4284         /*
4285          * Environment (env) might be missing mdt_thread_key values at that
4286          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4287          * mode.
4288          *
4289          * Usually device allocation path doesn't use module key values, but
4290          * mdt has to do a lot of work here, so allocate key value.
4291          */
4292         rc = lu_env_refill((struct lu_env *)env);
4293         if (rc != 0)
4294                 RETURN(rc);
4295
4296         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4297         LASSERT(info != NULL);
4298
4299         obd = class_name2obd(dev);
4300         LASSERT(obd != NULL);
4301
4302         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4303
4304         m->mdt_som_conf = 0;
4305
4306         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4307
4308         /* default is coordinator off, it is started through conf_param
4309          * or /proc */
4310         m->mdt_opts.mo_coordinator = 0;
4311
4312         lmi = server_get_mount(dev);
4313         if (lmi == NULL) {
4314                 CERROR("Cannot get mount info for %s!\n", dev);
4315                 RETURN(-EFAULT);
4316         } else {
4317                 lsi = s2lsi(lmi->lmi_sb);
4318                 /* CMD is supported only in IAM mode */
4319                 LASSERT(num);
4320                 node_id = simple_strtol(num, NULL, 10);
4321                 obd->u.obt.obt_magic = OBT_MAGIC;
4322         }
4323
4324         spin_lock_init(&m->mdt_ioepoch_lock);
4325         m->mdt_capa_timeout = CAPA_TIMEOUT;
4326         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4327         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4328         m->mdt_squash_uid = 0;
4329         m->mdt_squash_gid = 0;
4330         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4331         m->mdt_nosquash_str = NULL;
4332         m->mdt_nosquash_strlen = 0;
4333         init_rwsem(&m->mdt_squash_sem);
4334         spin_lock_init(&m->mdt_osfs_lock);
4335         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4336         m->mdt_enable_remote_dir = 0;
4337         m->mdt_enable_remote_dir_gid = 0;
4338
4339         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4340         m->mdt_lu_dev.ld_obd = obd;
4341         /* Set this lu_device to obd for error handling purposes. */
4342         obd->obd_lu_dev = &m->mdt_lu_dev;
4343
4344         /* init the stack */
4345         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4346         if (rc) {
4347                 CERROR("%s: Can't init device stack, rc %d\n",
4348                        mdt_obd_name(m), rc);
4349                 GOTO(err_lmi, rc);
4350         }
4351
4352         s = mdt_lu_site(m);
4353         ss_site = mdt_seq_site(m);
4354         s->ld_seq_site = ss_site;
4355         ss_site->ss_lu = s;
4356
4357         /* set server index */
4358         ss_site->ss_node_id = node_id;
4359
4360         /* failover is the default
4361          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4362          * assumed whose ss_node_id == 0 XXX
4363          * */
4364         obd->obd_replayable = 1;
4365         /* No connection accepted until configurations will finish */
4366         obd->obd_no_conn = 1;
4367
4368         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4369                 char *str = lustre_cfg_string(cfg, 4);
4370                 if (strchr(str, 'n')) {
4371                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4372                         obd->obd_replayable = 0;
4373                 }
4374         }
4375
4376         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4377         if (rc)
4378                 GOTO(err_fini_stack, rc);
4379
4380         rc = mdt_seq_init(env, m);
4381         if (rc)
4382                 GOTO(err_fini_fld, rc);
4383
4384         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4385                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4386         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4387                                               LDLM_NAMESPACE_SERVER,
4388                                               LDLM_NAMESPACE_GREEDY,
4389                                               LDLM_NS_TYPE_MDT);
4390         if (m->mdt_namespace == NULL)
4391                 GOTO(err_fini_seq, rc = -ENOMEM);
4392
4393         m->mdt_namespace->ns_lvbp = m;
4394         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4395
4396         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4397         /* set obd_namespace for compatibility with old code */
4398         obd->obd_namespace = m->mdt_namespace;
4399
4400         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4401
4402         rc = mdt_hsm_cdt_init(m);
4403         if (rc != 0) {
4404                 CERROR("%s: error initializing coordinator, rc %d\n",
4405                        mdt_obd_name(m), rc);
4406                 GOTO(err_free_ns, rc);
4407         }
4408
4409         rc = mdt_ck_thread_start(m);
4410         if (rc)
4411                 GOTO(err_free_hsm, rc);
4412
4413         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4414                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4415                       OBD_FAIL_MDS_ALL_REPLY_NET);
4416         if (rc)
4417                 GOTO(err_capa, rc);
4418
4419         rc = mdt_fs_setup(env, m, obd, lsi);
4420         if (rc)
4421                 GOTO(err_tgt, rc);
4422
4423         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4424
4425         next = m->mdt_child;
4426         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4427                                          &mntopts);
4428         if (rc)
4429                 GOTO(err_fs_cleanup, rc);
4430
4431         if (mntopts & MNTOPT_USERXATTR)
4432                 m->mdt_opts.mo_user_xattr = 1;
4433         else
4434                 m->mdt_opts.mo_user_xattr = 0;
4435
4436         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4437         if (rc)
4438                 GOTO(err_fs_cleanup, rc);
4439
4440         if (mntopts & MNTOPT_ACL)
4441                 m->mdt_opts.mo_acl = 1;
4442         else
4443                 m->mdt_opts.mo_acl = 0;
4444
4445         /* XXX: to support suppgid for ACL, we enable identity_upcall
4446          * by default, otherwise, maybe got unexpected -EACCESS. */
4447         if (m->mdt_opts.mo_acl)
4448                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4449
4450         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4451                                                 identity_upcall,
4452                                                 &mdt_identity_upcall_cache_ops);
4453         if (IS_ERR(m->mdt_identity_cache)) {
4454                 rc = PTR_ERR(m->mdt_identity_cache);
4455                 m->mdt_identity_cache = NULL;
4456                 GOTO(err_fs_cleanup, rc);
4457         }
4458
4459         rc = mdt_procfs_init(m, dev);
4460         if (rc) {
4461                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4462                 GOTO(err_recovery, rc);
4463         }
4464
4465         rc = mdt_quota_init(env, m, cfg);
4466         if (rc)
4467                 GOTO(err_procfs, rc);
4468
4469         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4470         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4471                            "mdt_ldlm_client", m->mdt_ldlm_client);
4472
4473         ping_evictor_start();
4474
4475         /* recovery will be started upon mdt_prepare()
4476          * when the whole stack is complete and ready
4477          * to serve the requests */
4478
4479         mdt_init_capa_ctxt(env, m);
4480
4481         /* Reduce the initial timeout on an MDS because it doesn't need such
4482          * a long timeout as an OST does. Adaptive timeouts will adjust this
4483          * value appropriately. */
4484         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4485                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4486
4487         RETURN(0);
4488 err_procfs:
4489         mdt_procfs_fini(m);
4490 err_recovery:
4491         target_recovery_fini(obd);
4492         upcall_cache_cleanup(m->mdt_identity_cache);
4493         m->mdt_identity_cache = NULL;
4494 err_fs_cleanup:
4495         mdt_fs_cleanup(env, m);
4496 err_tgt:
4497         tgt_fini(env, &m->mdt_lut);
4498 err_capa:
4499         cfs_timer_disarm(&m->mdt_ck_timer);
4500         mdt_ck_thread_stop(m);
4501 err_free_hsm:
4502         mdt_hsm_cdt_fini(m);
4503 err_free_ns:
4504         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4505         obd->obd_namespace = m->mdt_namespace = NULL;
4506 err_fini_seq:
4507         mdt_seq_fini(env, m);
4508 err_fini_fld:
4509         mdt_fld_fini(env, m);
4510 err_fini_stack:
4511         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4512 err_lmi:
4513         if (lmi)
4514                 server_put_mount(dev);
4515         return(rc);
4516 }
4517
4518 /* For interoperability, the left element is old parameter, the right one
4519  * is the new version of the parameter, if some parameter is deprecated,
4520  * the new version should be set as NULL. */
4521 static struct cfg_interop_param mdt_interop_param[] = {
4522         { "mdt.group_upcall",   NULL },
4523         { "mdt.quota_type",     NULL },
4524         { "mdd.quota_type",     NULL },
4525         { "mdt.rootsquash",     "mdt.root_squash" },
4526         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4527         { NULL }
4528 };
4529
4530 /* used by MGS to process specific configurations */
4531 static int mdt_process_config(const struct lu_env *env,
4532                               struct lu_device *d, struct lustre_cfg *cfg)
4533 {
4534         struct mdt_device *m = mdt_dev(d);
4535         struct md_device *md_next = m->mdt_child;
4536         struct lu_device *next = md2lu_dev(md_next);
4537         int rc;
4538         ENTRY;
4539
4540         switch (cfg->lcfg_command) {
4541         case LCFG_PARAM: {
4542                 struct lprocfs_static_vars  lvars;
4543                 struct obd_device          *obd = d->ld_obd;
4544
4545                 /* For interoperability */
4546                 struct cfg_interop_param   *ptr = NULL;
4547                 struct lustre_cfg          *old_cfg = NULL;
4548                 char                       *param = NULL;
4549
4550                 param = lustre_cfg_string(cfg, 1);
4551                 if (param == NULL) {
4552                         CERROR("param is empty\n");
4553                         rc = -EINVAL;
4554                         break;
4555                 }
4556
4557                 ptr = class_find_old_param(param, mdt_interop_param);
4558                 if (ptr != NULL) {
4559                         if (ptr->new_param == NULL) {
4560                                 rc = 0;
4561                                 CWARN("For interoperability, skip this %s."
4562                                       " It is obsolete.\n", ptr->old_param);
4563                                 break;
4564                         }
4565
4566                         CWARN("Found old param %s, changed it to %s.\n",
4567                               ptr->old_param, ptr->new_param);
4568
4569                         old_cfg = cfg;
4570                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4571                         if (IS_ERR(cfg)) {
4572                                 rc = PTR_ERR(cfg);
4573                                 break;
4574                         }
4575                 }
4576
4577                 lprocfs_mdt_init_vars(&lvars);
4578                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4579                                               cfg, obd);
4580                 if (rc > 0 || rc == -ENOSYS) {
4581                         /* is it an HSM var ? */
4582                         rc = class_process_proc_param(PARAM_HSM,
4583                                                       hsm_cdt_get_proc_vars(),
4584                                                       cfg, obd);
4585                         if (rc > 0 || rc == -ENOSYS)
4586                                 /* we don't understand; pass it on */
4587                                 rc = next->ld_ops->ldo_process_config(env, next,
4588                                                                       cfg);
4589                 }
4590
4591                 if (old_cfg != NULL)
4592                         lustre_cfg_free(cfg);
4593
4594                 break;
4595         }
4596         default:
4597                 /* others are passed further */
4598                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4599                 break;
4600         }
4601         RETURN(rc);
4602 }
4603
4604 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4605                                           const struct lu_object_header *hdr,
4606                                           struct lu_device *d)
4607 {
4608         struct mdt_object *mo;
4609
4610         ENTRY;
4611
4612         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
4613         if (mo != NULL) {
4614                 struct lu_object *o;
4615                 struct lu_object_header *h;
4616
4617                 o = &mo->mot_obj;
4618                 h = &mo->mot_header;
4619                 lu_object_header_init(h);
4620                 lu_object_init(o, h, d);
4621                 lu_object_add_top(h, o);
4622                 o->lo_ops = &mdt_obj_ops;
4623                 mutex_init(&mo->mot_ioepoch_mutex);
4624                 mutex_init(&mo->mot_lov_mutex);
4625                 init_rwsem(&mo->mot_open_sem);
4626                 RETURN(o);
4627         }
4628         RETURN(NULL);
4629 }
4630
4631 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4632                            const struct lu_object_conf *unused)
4633 {
4634         struct mdt_device *d = mdt_dev(o->lo_dev);
4635         struct lu_device  *under;
4636         struct lu_object  *below;
4637         int                rc = 0;
4638         ENTRY;
4639
4640         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4641                PFID(lu_object_fid(o)));
4642
4643         under = &d->mdt_child->md_lu_dev;
4644         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4645         if (below != NULL) {
4646                 lu_object_add(o, below);
4647         } else
4648                 rc = -ENOMEM;
4649
4650         RETURN(rc);
4651 }
4652
4653 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4654 {
4655         struct mdt_object *mo = mdt_obj(o);
4656         struct lu_object_header *h;
4657         ENTRY;
4658
4659         h = o->lo_header;
4660         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4661                PFID(lu_object_fid(o)));
4662
4663         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4664         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4665
4666         lu_object_fini(o);
4667         lu_object_header_fini(h);
4668         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4669
4670         EXIT;
4671 }
4672
4673 static int mdt_object_print(const struct lu_env *env, void *cookie,
4674                             lu_printer_t p, const struct lu_object *o)
4675 {
4676         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4677         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4678                     "flags="LPX64", epochcount=%d, writecount=%d)",
4679                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4680                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4681 }
4682
4683 static int mdt_prepare(const struct lu_env *env,
4684                 struct lu_device *pdev,
4685                 struct lu_device *cdev)
4686 {
4687         struct mdt_device *mdt = mdt_dev(cdev);
4688         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4689         struct obd_device *obd = cdev->ld_obd;
4690         struct lfsck_start_param lsp;
4691         int rc;
4692
4693         ENTRY;
4694
4695         LASSERT(obd);
4696
4697         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4698         if (rc)
4699                 RETURN(rc);
4700
4701         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4702         if (rc)
4703                 RETURN(rc);
4704
4705         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4706         if (rc)
4707                 RETURN(rc);
4708
4709         lsp.lsp_namespace = mdt->mdt_namespace;
4710         lsp.lsp_start = NULL;
4711         lsp.lsp_index_valid = 0;
4712         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
4713                                                    OBD_IOC_START_LFSCK,
4714                                                    0, &lsp);
4715         if (rc != 0) {
4716                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
4717                       mdt_obd_name(mdt), rc);
4718                 rc = 0;
4719         }
4720
4721         if (mdt->mdt_seq_site.ss_node_id == 0) {
4722                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4723                                                          &mdt->mdt_md_root_fid);
4724                 if (rc)
4725                         RETURN(rc);
4726         }
4727
4728         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4729         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4730         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4731         LASSERT(obd->obd_no_conn);
4732         spin_lock(&obd->obd_dev_lock);
4733         obd->obd_no_conn = 0;
4734         spin_unlock(&obd->obd_dev_lock);
4735
4736         if (obd->obd_recovering == 0)
4737                 mdt_postrecov(env, mdt);
4738
4739         RETURN(rc);
4740 }
4741
4742 const struct lu_device_operations mdt_lu_ops = {
4743         .ldo_object_alloc   = mdt_object_alloc,
4744         .ldo_process_config = mdt_process_config,
4745         .ldo_prepare        = mdt_prepare,
4746 };
4747
4748 static const struct lu_object_operations mdt_obj_ops = {
4749         .loo_object_init    = mdt_object_init,
4750         .loo_object_free    = mdt_object_free,
4751         .loo_object_print   = mdt_object_print
4752 };
4753
4754 static int mdt_obd_set_info_async(const struct lu_env *env,
4755                                   struct obd_export *exp,
4756                                   __u32 keylen, void *key,
4757                                   __u32 vallen, void *val,
4758                                   struct ptlrpc_request_set *set)
4759 {
4760         int rc;
4761
4762         ENTRY;
4763
4764         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4765                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4766                 RETURN(rc);
4767         }
4768
4769         RETURN(0);
4770 }
4771
4772 /**
4773  * Match client and server connection feature flags.
4774  *
4775  * Compute the compatibility flags for a connection request based on
4776  * features mutually supported by client and server.
4777  *
4778  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4779  * must not be updated here, otherwise a partially initialized value may
4780  * be exposed. After the connection request is successfully processed,
4781  * the top-level MDT connect request handler atomically updates the export
4782  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4783  * reply. \see mdt_connect().
4784  *
4785  * \param exp   the obd_export associated with this client/target pair
4786  * \param mdt   the target device for the connection
4787  * \param data  stores data for this connect request
4788  *
4789  * \retval 0       success
4790  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4791  * \retval -EBADE  client and server feature requirements are incompatible
4792  */
4793 static int mdt_connect_internal(struct obd_export *exp,
4794                                 struct mdt_device *mdt,
4795                                 struct obd_connect_data *data)
4796 {
4797         LASSERT(data != NULL);
4798
4799         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4800         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4801
4802         /* If no known bits (which should not happen, probably,
4803            as everybody should support LOOKUP and UPDATE bits at least)
4804            revert to compat mode with plain locks. */
4805         if (!data->ocd_ibits_known &&
4806             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4807                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4808
4809         if (!mdt->mdt_opts.mo_acl)
4810                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4811
4812         if (!mdt->mdt_opts.mo_user_xattr)
4813                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4814
4815         if (!mdt->mdt_som_conf)
4816                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4817
4818         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4819                 data->ocd_brw_size = min(data->ocd_brw_size,
4820                                          (__u32)MD_MAX_BRW_SIZE);
4821                 if (data->ocd_brw_size == 0) {
4822                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4823                                " ocd_version: %x ocd_grant: %d "
4824                                "ocd_index: %u ocd_brw_size is "
4825                                "unexpectedly zero, network data "
4826                                "corruption? Refusing connection of this"
4827                                " client\n",
4828                                mdt_obd_name(mdt),
4829                                exp->exp_client_uuid.uuid,
4830                                exp, data->ocd_connect_flags, data->ocd_version,
4831                                data->ocd_grant, data->ocd_index);
4832                         return -EPROTO;
4833                 }
4834         }
4835
4836         /* NB: Disregard the rule against updating
4837          * exp_connect_data.ocd_connect_flags in this case, since
4838          * tgt_client_new() needs to know if this is a lightweight
4839          * connection, and it is safe to expose this flag before
4840          * connection processing completes. */
4841         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4842                 spin_lock(&exp->exp_lock);
4843                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4844                 spin_unlock(&exp->exp_lock);
4845         }
4846
4847         data->ocd_version = LUSTRE_VERSION_CODE;
4848
4849         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4850                 CWARN("%s: MDS requires FID support, but client not\n",
4851                       mdt_obd_name(mdt));
4852                 return -EBADE;
4853         }
4854
4855         if (mdt->mdt_som_conf &&
4856             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4857                                          OBD_CONNECT_MDS_MDS |
4858                                          OBD_CONNECT_SOM))) {
4859                 CWARN("%s: MDS has SOM enabled, but client does not support "
4860                       "it\n", mdt_obd_name(mdt));
4861                 return -EBADE;
4862         }
4863
4864         if (OCD_HAS_FLAG(data, PINGLESS)) {
4865                 if (ptlrpc_pinger_suppress_pings()) {
4866                         spin_lock(&exp->exp_obd->obd_dev_lock);
4867                         list_del_init(&exp->exp_obd_chain_timed);
4868                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4869                 } else {
4870                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4871                 }
4872         }
4873
4874         data->ocd_max_easize = mdt->mdt_max_ea_size;
4875
4876         return 0;
4877 }
4878
4879 /* mds_connect copy */
4880 static int mdt_obd_connect(const struct lu_env *env,
4881                            struct obd_export **exp, struct obd_device *obd,
4882                            struct obd_uuid *cluuid,
4883                            struct obd_connect_data *data,
4884                            void *localdata)
4885 {
4886         struct obd_export      *lexp;
4887         struct lustre_handle    conn = { 0 };
4888         struct mdt_device      *mdt;
4889         int                     rc;
4890         ENTRY;
4891
4892         LASSERT(env != NULL);
4893         if (!exp || !obd || !cluuid)
4894                 RETURN(-EINVAL);
4895
4896         mdt = mdt_dev(obd->obd_lu_dev);
4897
4898         /*
4899          * first, check whether the stack is ready to handle requests
4900          * XXX: probably not very appropriate method is used now
4901          *      at some point we should find a better one
4902          */
4903         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
4904             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
4905                 rc = obd_get_info(env, mdt->mdt_child_exp,
4906                                   sizeof(KEY_OSP_CONNECTED),
4907                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
4908                 if (rc)
4909                         RETURN(-EAGAIN);
4910                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
4911         }
4912
4913         rc = class_connect(&conn, obd, cluuid);
4914         if (rc)
4915                 RETURN(rc);
4916
4917         lexp = class_conn2export(&conn);
4918         LASSERT(lexp != NULL);
4919
4920         rc = mdt_connect_internal(lexp, mdt, data);
4921         if (rc == 0) {
4922                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
4923
4924                 LASSERT(lcd);
4925                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4926                 rc = tgt_client_new(env, lexp);
4927                 if (rc == 0)
4928                         mdt_export_stats_init(obd, lexp, localdata);
4929
4930                 /* For phase I, sync for cross-ref operation. */
4931                 lexp->exp_keep_sync = 1;
4932         }
4933
4934         if (rc != 0) {
4935                 class_disconnect(lexp);
4936                 *exp = NULL;
4937         } else {
4938                 *exp = lexp;
4939         }
4940
4941         RETURN(rc);
4942 }
4943
4944 static int mdt_obd_reconnect(const struct lu_env *env,
4945                              struct obd_export *exp, struct obd_device *obd,
4946                              struct obd_uuid *cluuid,
4947                              struct obd_connect_data *data,
4948                              void *localdata)
4949 {
4950         int                     rc;
4951         ENTRY;
4952
4953         if (exp == NULL || obd == NULL || cluuid == NULL)
4954                 RETURN(-EINVAL);
4955
4956         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4957         if (rc == 0)
4958                 mdt_export_stats_init(obd, exp, localdata);
4959
4960         RETURN(rc);
4961 }
4962
4963 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4964                                    struct mdt_thread_info *info,
4965                                    struct mdt_file_data *mfd)
4966 {
4967         struct lu_context ses;
4968         int rc;
4969         ENTRY;
4970
4971         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4972         if (rc)
4973                 RETURN(rc);
4974
4975         env->le_ses = &ses;
4976         lu_context_enter(&ses);
4977
4978         mdt_ucred(info)->uc_valid = UCRED_OLD;
4979         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4980
4981         lu_context_exit(&ses);
4982         lu_context_fini(&ses);
4983         env->le_ses = NULL;
4984
4985         RETURN(rc);
4986 }
4987
4988 static int mdt_export_cleanup(struct obd_export *exp)
4989 {
4990         struct mdt_export_data *med = &exp->exp_mdt_data;
4991         struct obd_device      *obd = exp->exp_obd;
4992         struct mdt_device      *mdt;
4993         struct mdt_thread_info *info;
4994         struct lu_env           env;
4995         CFS_LIST_HEAD(closing_list);
4996         struct mdt_file_data *mfd, *n;
4997         int rc = 0;
4998         ENTRY;
4999
5000         spin_lock(&med->med_open_lock);
5001         while (!cfs_list_empty(&med->med_open_head)) {
5002                 cfs_list_t *tmp = med->med_open_head.next;
5003                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5004
5005                 /* Remove mfd handle so it can't be found again.
5006                  * We are consuming the mfd_list reference here. */
5007                 class_handle_unhash(&mfd->mfd_handle);
5008                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5009         }
5010         spin_unlock(&med->med_open_lock);
5011         mdt = mdt_dev(obd->obd_lu_dev);
5012         LASSERT(mdt != NULL);
5013
5014         rc = lu_env_init(&env, LCT_MD_THREAD);
5015         if (rc)
5016                 RETURN(rc);
5017
5018         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5019         LASSERT(info != NULL);
5020         memset(info, 0, sizeof *info);
5021         info->mti_env = &env;
5022         info->mti_mdt = mdt;
5023         info->mti_exp = exp;
5024
5025         if (!cfs_list_empty(&closing_list)) {
5026                 struct md_attr *ma = &info->mti_attr;
5027
5028                 /* Close any open files (which may also cause orphan unlinking). */
5029                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5030                         cfs_list_del_init(&mfd->mfd_list);
5031                         ma->ma_need = ma->ma_valid = 0;
5032
5033                         /* This file is being closed due to an eviction, it
5034                          * could have been modified and now dirty regarding to
5035                          * HSM archive, check this!
5036                          * The logic here is to mark a file dirty if there's a
5037                          * chance it was dirtied before the client was evicted,
5038                          * so that we don't have to wait for a release attempt
5039                          * before finding out the file was actually dirty and
5040                          * fail the release. Aggressively marking it dirty here
5041                          * will cause the policy engine to attempt to
5042                          * re-archive it; when rearchiving, we can compare the
5043                          * current version to the HSM data_version and make the
5044                          * archive request into a noop if it's not actually
5045                          * dirty.
5046                          */
5047                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5048                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5049
5050                         /* Don't unlink orphan on failover umount, LU-184 */
5051                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5052                                 ma->ma_valid = MA_FLAGS;
5053                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5054                         }
5055                         mdt_mfd_close(info, mfd);
5056                 }
5057         }
5058         info->mti_mdt = NULL;
5059         /* cleanup client slot early */
5060         /* Do not erase record for recoverable client. */
5061         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5062                 tgt_client_del(&env, exp);
5063         lu_env_fini(&env);
5064
5065         RETURN(rc);
5066 }
5067
5068 static int mdt_obd_disconnect(struct obd_export *exp)
5069 {
5070         int rc;
5071         ENTRY;
5072
5073         LASSERT(exp);
5074         class_export_get(exp);
5075
5076         rc = server_disconnect_export(exp);
5077         if (rc != 0)
5078                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5079
5080         rc = mdt_export_cleanup(exp);
5081         class_export_put(exp);
5082         RETURN(rc);
5083 }
5084
5085 /* FIXME: Can we avoid using these two interfaces? */
5086 static int mdt_init_export(struct obd_export *exp)
5087 {
5088         struct mdt_export_data *med = &exp->exp_mdt_data;
5089         int                     rc;
5090         ENTRY;
5091
5092         CFS_INIT_LIST_HEAD(&med->med_open_head);
5093         spin_lock_init(&med->med_open_lock);
5094         mutex_init(&med->med_idmap_mutex);
5095         med->med_idmap = NULL;
5096         spin_lock(&exp->exp_lock);
5097         exp->exp_connecting = 1;
5098         spin_unlock(&exp->exp_lock);
5099
5100         /* self-export doesn't need client data and ldlm initialization */
5101         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5102                                      &exp->exp_client_uuid)))
5103                 RETURN(0);
5104
5105         rc = tgt_client_alloc(exp);
5106         if (rc)
5107                 GOTO(err, rc);
5108
5109         rc = ldlm_init_export(exp);
5110         if (rc)
5111                 GOTO(err_free, rc);
5112
5113         RETURN(rc);
5114
5115 err_free:
5116         tgt_client_free(exp);
5117 err:
5118         CERROR("%s: Failed to initialize export: rc = %d\n",
5119                exp->exp_obd->obd_name, rc);
5120         return rc;
5121 }
5122
5123 static int mdt_destroy_export(struct obd_export *exp)
5124 {
5125         ENTRY;
5126
5127         if (exp_connect_rmtclient(exp))
5128                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5129
5130         target_destroy_export(exp);
5131         /* destroy can be called from failed obd_setup, so
5132          * checking uuid is safer than obd_self_export */
5133         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5134                                      &exp->exp_client_uuid)))
5135                 RETURN(0);
5136
5137         ldlm_destroy_export(exp);
5138         tgt_client_free(exp);
5139
5140         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5141         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5142
5143         RETURN(0);
5144 }
5145
5146 /** The maximum depth that fid2path() will search.
5147  * This is limited only because we want to store the fids for
5148  * historical path lookup purposes.
5149  */
5150 #define MAX_PATH_DEPTH 100
5151
5152 /** mdt_path() lookup structure. */
5153 struct path_lookup_info {
5154         __u64                   pli_recno;      /**< history point */
5155         __u64                   pli_currec;     /**< current record */
5156         struct lu_fid           pli_fid;
5157         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5158         struct mdt_object       *pli_mdt_obj;
5159         char                    *pli_path;      /**< full path */
5160         int                     pli_pathlen;
5161         int                     pli_linkno;     /**< which hardlink to follow */
5162         int                     pli_fidcount;   /**< number of \a pli_fids */
5163 };
5164
5165 static int mdt_links_read(struct mdt_thread_info *info,
5166                           struct mdt_object *mdt_obj, struct linkea_data *ldata)
5167 {
5168         int rc;
5169
5170         LASSERT(ldata->ld_buf->lb_buf != NULL);
5171
5172         if (!mdt_object_exists(mdt_obj))
5173                 return -ENODATA;
5174
5175         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5176                           ldata->ld_buf, XATTR_NAME_LINK);
5177         if (rc == -ERANGE) {
5178                 /* Buf was too small, figure out what we need. */
5179                 lu_buf_free(ldata->ld_buf);
5180                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5181                                   ldata->ld_buf, XATTR_NAME_LINK);
5182                 if (rc < 0)
5183                         return rc;
5184                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5185                 if (ldata->ld_buf->lb_buf == NULL)
5186                         return -ENOMEM;
5187                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5188                                   ldata->ld_buf, XATTR_NAME_LINK);
5189         }
5190         if (rc < 0)
5191                 return rc;
5192
5193         return linkea_init(ldata);
5194 }
5195
5196 static int mdt_path_current(struct mdt_thread_info *info,
5197                             struct path_lookup_info *pli)
5198 {
5199         struct mdt_device       *mdt = info->mti_mdt;
5200         struct mdt_object       *mdt_obj;
5201         struct link_ea_header   *leh;
5202         struct link_ea_entry    *lee;
5203         struct lu_name          *tmpname = &info->mti_name;
5204         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5205         struct lu_buf           *buf = &info->mti_big_buf;
5206         char                    *ptr;
5207         int                     reclen;
5208         struct linkea_data      ldata = { 0 };
5209         int                     rc = 0;
5210         ENTRY;
5211
5212         /* temp buffer for path element, the buffer will be finally freed
5213          * in mdt_thread_info_fini */
5214         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5215         if (buf->lb_buf == NULL)
5216                 RETURN(-ENOMEM);
5217
5218         ldata.ld_buf = buf;
5219         ptr = pli->pli_path + pli->pli_pathlen - 1;
5220         *ptr = 0;
5221         --ptr;
5222         pli->pli_fidcount = 0;
5223         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5224
5225         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5226          * so checking root_fid can only happen on MDT0. */
5227         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5228                           &pli->pli_fids[pli->pli_fidcount])) {
5229                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5230                                           &pli->pli_fids[pli->pli_fidcount]);
5231                 if (IS_ERR(mdt_obj))
5232                         GOTO(out, rc = PTR_ERR(mdt_obj));
5233                 if (mdt_object_remote(mdt_obj)) {
5234                         mdt_object_put(info->mti_env, mdt_obj);
5235                         GOTO(remote_out, rc = -EREMOTE);
5236                 }
5237                 if (!mdt_object_exists(mdt_obj)) {
5238                         mdt_object_put(info->mti_env, mdt_obj);
5239                         GOTO(out, rc = -ENOENT);
5240                 }
5241
5242                 rc = mdt_links_read(info, mdt_obj, &ldata);
5243                 mdt_object_put(info->mti_env, mdt_obj);
5244                 if (rc != 0)
5245                         GOTO(out, rc);
5246
5247                 leh = buf->lb_buf;
5248                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5249                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5250                 /* If set, use link #linkno for path lookup, otherwise use
5251                    link #0.  Only do this for the final path element. */
5252                 if (pli->pli_fidcount == 0 &&
5253                     pli->pli_linkno < leh->leh_reccount) {
5254                         int count;
5255                         for (count = 0; count < pli->pli_linkno; count++) {
5256                                 lee = (struct link_ea_entry *)
5257                                      ((char *)lee + reclen);
5258                                 linkea_entry_unpack(lee, &reclen, tmpname,
5259                                                     tmpfid);
5260                         }
5261                         if (pli->pli_linkno < leh->leh_reccount - 1)
5262                                 /* indicate to user there are more links */
5263                                 pli->pli_linkno++;
5264                 }
5265
5266                 /* Pack the name in the end of the buffer */
5267                 ptr -= tmpname->ln_namelen;
5268                 if (ptr - 1 <= pli->pli_path)
5269                         GOTO(out, rc = -EOVERFLOW);
5270                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5271                 *(--ptr) = '/';
5272
5273                 /* Store the parent fid for historic lookup */
5274                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5275                         GOTO(out, rc = -EOVERFLOW);
5276                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5277         }
5278
5279 remote_out:
5280         ptr++; /* skip leading / */
5281         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5282
5283         EXIT;
5284 out:
5285         return rc;
5286 }
5287
5288 /* Returns the full path to this fid, as of changelog record recno. */
5289 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5290                     char *path, int pathlen, __u64 *recno, int *linkno,
5291                     struct lu_fid *fid)
5292 {
5293         struct mdt_device       *mdt = info->mti_mdt;
5294         struct path_lookup_info *pli;
5295         int                     tries = 3;
5296         int                     rc = -EAGAIN;
5297         ENTRY;
5298
5299         if (pathlen < 3)
5300                 RETURN(-EOVERFLOW);
5301
5302         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5303                 path[0] = '\0';
5304                 RETURN(0);
5305         }
5306
5307         OBD_ALLOC_PTR(pli);
5308         if (pli == NULL)
5309                 RETURN(-ENOMEM);
5310
5311         pli->pli_mdt_obj = obj;
5312         pli->pli_recno = *recno;
5313         pli->pli_path = path;
5314         pli->pli_pathlen = pathlen;
5315         pli->pli_linkno = *linkno;
5316
5317         /* Retry multiple times in case file is being moved */
5318         while (tries-- && rc == -EAGAIN)
5319                 rc = mdt_path_current(info, pli);
5320
5321         /* return the last resolved fids to the client, so the client will
5322          * build the left path on another MDT for remote object */
5323         *fid = pli->pli_fids[pli->pli_fidcount];
5324
5325         *recno = pli->pli_currec;
5326         /* Return next link index to caller */
5327         *linkno = pli->pli_linkno;
5328
5329         OBD_FREE_PTR(pli);
5330
5331         RETURN(rc);
5332 }
5333
5334 static int mdt_fid2path(struct mdt_thread_info *info,
5335                         struct getinfo_fid2path *fp)
5336 {
5337         struct mdt_device *mdt = info->mti_mdt;
5338         struct mdt_object *obj;
5339         int    rc;
5340         ENTRY;
5341
5342         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5343                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5344
5345         if (!fid_is_sane(&fp->gf_fid))
5346                 RETURN(-EINVAL);
5347
5348         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5349                 CWARN("%s: "DFID" is invalid, sequence should be "
5350                       ">= "LPX64"\n", mdt_obd_name(mdt),
5351                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5352                 RETURN(-EINVAL);
5353         }
5354
5355         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5356         if (obj == NULL || IS_ERR(obj)) {
5357                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5358                        PTR_ERR(obj));
5359                 RETURN(-EINVAL);
5360         }
5361
5362         if (mdt_object_remote(obj))
5363                 rc = -EREMOTE;
5364         else if (!mdt_object_exists(obj))
5365                 rc = -ENOENT;
5366         else
5367                 rc = 0;
5368
5369         if (rc < 0) {
5370                 mdt_object_put(info->mti_env, obj);
5371                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5372                        PFID(&fp->gf_fid), rc);
5373                 RETURN(rc);
5374         }
5375
5376         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5377                       &fp->gf_linkno, &fp->gf_fid);
5378
5379         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5380                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5381
5382         mdt_object_put(info->mti_env, obj);
5383
5384         RETURN(rc);
5385 }
5386
5387 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5388                             void *val, int vallen)
5389 {
5390         struct getinfo_fid2path *fpout, *fpin;
5391         int rc = 0;
5392
5393         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5394         fpout = val;
5395
5396         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5397                 lustre_swab_fid2path(fpin);
5398
5399         memcpy(fpout, fpin, sizeof(*fpin));
5400         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5401                 RETURN(-EINVAL);
5402
5403         rc = mdt_fid2path(info, fpout);
5404         RETURN(rc);
5405 }
5406
5407 int mdt_get_info(struct tgt_session_info *tsi)
5408 {
5409         char    *key;
5410         int      keylen;
5411         __u32   *vallen;
5412         void    *valout;
5413         int      rc;
5414
5415         ENTRY;
5416
5417         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5418         if (key == NULL) {
5419                 CDEBUG(D_IOCTL, "No GETINFO key");
5420                 RETURN(err_serious(-EFAULT));
5421         }
5422         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5423                                       RCL_CLIENT);
5424
5425         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5426         if (vallen == NULL) {
5427                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5428                 RETURN(err_serious(-EFAULT));
5429         }
5430
5431         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5432                              *vallen);
5433         rc = req_capsule_server_pack(tsi->tsi_pill);
5434         if (rc)
5435                 RETURN(err_serious(rc));
5436
5437         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5438         if (valout == NULL) {
5439                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5440                 RETURN(err_serious(-EFAULT));
5441         }
5442
5443         if (KEY_IS(KEY_FID2PATH)) {
5444                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5445
5446                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5447                 mdt_thread_info_fini(info);
5448         } else {
5449                 rc = -EINVAL;
5450         }
5451         RETURN(rc);
5452 }
5453
5454 /* Pass the ioc down */
5455 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5456                          unsigned int cmd, int len, void *data)
5457 {
5458         struct lu_context ioctl_session;
5459         struct md_device *next = mdt->mdt_child;
5460         int rc;
5461         ENTRY;
5462
5463         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5464         if (rc)
5465                 RETURN(rc);
5466         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5467         lu_context_enter(&ioctl_session);
5468         env->le_ses = &ioctl_session;
5469
5470         LASSERT(next->md_ops->mdo_iocontrol);
5471         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5472
5473         lu_context_exit(&ioctl_session);
5474         lu_context_fini(&ioctl_session);
5475         RETURN(rc);
5476 }
5477
5478 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5479 {
5480         struct obd_ioctl_data *data = karg;
5481         struct lu_fid *fid;
5482         __u64 version;
5483         struct mdt_object *obj;
5484         struct mdt_lock_handle  *lh;
5485         int rc;
5486         ENTRY;
5487
5488         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5489             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5490                 RETURN(-EINVAL);
5491
5492         fid = (struct lu_fid *)data->ioc_inlbuf1;
5493
5494         if (!fid_is_sane(fid))
5495                 RETURN(-EINVAL);
5496
5497         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5498
5499         lh = &mti->mti_lh[MDT_LH_PARENT];
5500         mdt_lock_reg_init(lh, LCK_CR);
5501
5502         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5503         if (IS_ERR(obj))
5504                 RETURN(PTR_ERR(obj));
5505
5506         if (mdt_object_remote(obj)) {
5507                 rc = -EREMOTE;
5508                 /**
5509                  * before calling version get the correct MDS should be
5510                  * fid, this is error to find remote object here
5511                  */
5512                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5513         } else if (!mdt_object_exists(obj)) {
5514                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5515                 rc = -ENOENT;
5516         } else {
5517                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5518                *(__u64 *)data->ioc_inlbuf2 = version;
5519                 rc = 0;
5520         }
5521         mdt_object_unlock_put(mti, obj, lh, 1);
5522         RETURN(rc);
5523 }
5524
5525 /* ioctls on obd dev */
5526 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5527                          void *karg, void *uarg)
5528 {
5529         struct lu_env      env;
5530         struct obd_device *obd = exp->exp_obd;
5531         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5532         struct dt_device  *dt = mdt->mdt_bottom;
5533         int rc;
5534
5535         ENTRY;
5536         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5537         rc = lu_env_init(&env, LCT_MD_THREAD);
5538         if (rc)
5539                 RETURN(rc);
5540
5541         switch (cmd) {
5542         case OBD_IOC_SYNC:
5543                 rc = mdt_device_sync(&env, mdt);
5544                 break;
5545         case OBD_IOC_SET_READONLY:
5546                 rc = dt->dd_ops->dt_ro(&env, dt);
5547                 break;
5548         case OBD_IOC_ABORT_RECOVERY:
5549                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5550                 target_stop_recovery_thread(obd);
5551                 rc = 0;
5552                 break;
5553         case OBD_IOC_CHANGELOG_REG:
5554         case OBD_IOC_CHANGELOG_DEREG:
5555         case OBD_IOC_CHANGELOG_CLEAR:
5556                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5557                 break;
5558         case OBD_IOC_START_LFSCK: {
5559                 struct md_device *next = mdt->mdt_child;
5560                 struct obd_ioctl_data *data = karg;
5561                 struct lfsck_start_param lsp;
5562
5563                 if (unlikely(data == NULL)) {
5564                         rc = -EINVAL;
5565                         break;
5566                 }
5567
5568                 lsp.lsp_namespace = mdt->mdt_namespace;
5569                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5570                 lsp.lsp_index_valid = 0;
5571                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5572                 break;
5573         }
5574         case OBD_IOC_STOP_LFSCK: {
5575                 struct md_device        *next = mdt->mdt_child;
5576                 struct obd_ioctl_data   *data = karg;
5577                 struct lfsck_stop       *stop =
5578                                 (struct lfsck_stop *)(data->ioc_inlbuf1);
5579
5580                 stop->ls_status = LS_STOPPED;
5581                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, stop);
5582                 break;
5583         }
5584         case OBD_IOC_GET_OBJ_VERSION: {
5585                 struct mdt_thread_info *mti;
5586                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5587                 memset(mti, 0, sizeof *mti);
5588                 mti->mti_env = &env;
5589                 mti->mti_mdt = mdt;
5590                 mti->mti_exp = exp;
5591
5592                 rc = mdt_ioc_version_get(mti, karg);
5593                 break;
5594         }
5595         case OBD_IOC_CATLOGLIST:
5596                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg);
5597                 break;
5598         default:
5599                 rc = -EOPNOTSUPP;
5600                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5601                         mdt_obd_name(mdt), cmd, rc);
5602         }
5603
5604         lu_env_fini(&env);
5605         RETURN(rc);
5606 }
5607
5608 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5609 {
5610         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5611         int rc;
5612         ENTRY;
5613
5614         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5615         RETURN(rc);
5616 }
5617
5618 int mdt_obd_postrecov(struct obd_device *obd)
5619 {
5620         struct lu_env env;
5621         int rc;
5622
5623         rc = lu_env_init(&env, LCT_MD_THREAD);
5624         if (rc)
5625                 RETURN(rc);
5626         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5627         lu_env_fini(&env);
5628         return rc;
5629 }
5630
5631 static struct obd_ops mdt_obd_device_ops = {
5632         .o_owner          = THIS_MODULE,
5633         .o_set_info_async = mdt_obd_set_info_async,
5634         .o_connect        = mdt_obd_connect,
5635         .o_reconnect      = mdt_obd_reconnect,
5636         .o_disconnect     = mdt_obd_disconnect,
5637         .o_init_export    = mdt_init_export,
5638         .o_destroy_export = mdt_destroy_export,
5639         .o_iocontrol      = mdt_iocontrol,
5640         .o_postrecov      = mdt_obd_postrecov,
5641 };
5642
5643 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5644                                          struct lu_device *d)
5645 {
5646         struct mdt_device *m = mdt_dev(d);
5647         ENTRY;
5648
5649         mdt_fini(env, m);
5650         RETURN(NULL);
5651 }
5652
5653 static struct lu_device *mdt_device_free(const struct lu_env *env,
5654                                          struct lu_device *d)
5655 {
5656         struct mdt_device *m = mdt_dev(d);
5657         ENTRY;
5658
5659         lu_device_fini(&m->mdt_lu_dev);
5660         OBD_FREE_PTR(m);
5661
5662         RETURN(NULL);
5663 }
5664
5665 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5666                                           struct lu_device_type *t,
5667                                           struct lustre_cfg *cfg)
5668 {
5669         struct lu_device  *l;
5670         struct mdt_device *m;
5671
5672         OBD_ALLOC_PTR(m);
5673         if (m != NULL) {
5674                 int rc;
5675
5676                 l = &m->mdt_lu_dev;
5677                 rc = mdt_init0(env, m, t, cfg);
5678                 if (rc != 0) {
5679                         mdt_device_free(env, l);
5680                         l = ERR_PTR(rc);
5681                         return l;
5682                 }
5683         } else
5684                 l = ERR_PTR(-ENOMEM);
5685         return l;
5686 }
5687
5688 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5689 LU_KEY_INIT(mdt, struct mdt_thread_info);
5690
5691 static void mdt_key_fini(const struct lu_context *ctx,
5692                          struct lu_context_key *key, void* data)
5693 {
5694         struct mdt_thread_info *info = data;
5695
5696         if (info->mti_big_lmm) {
5697                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5698                 info->mti_big_lmm = NULL;
5699                 info->mti_big_lmmsize = 0;
5700         }
5701         OBD_FREE_PTR(info);
5702 }
5703
5704 /* context key: mdt_thread_key */
5705 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5706
5707 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5708 {
5709         return lu_ucred(info->mti_env);
5710 }
5711
5712 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5713 {
5714         return lu_ucred_check(info->mti_env);
5715 }
5716
5717 /**
5718  * Enable/disable COS (Commit On Sharing).
5719  *
5720  * Set/Clear the COS flag in mdt options.
5721  *
5722  * \param mdt mdt device
5723  * \param val 0 disables COS, other values enable COS
5724  */
5725 void mdt_enable_cos(struct mdt_device *mdt, int val)
5726 {
5727         struct lu_env env;
5728         int rc;
5729
5730         mdt->mdt_opts.mo_cos = !!val;
5731         rc = lu_env_init(&env, LCT_LOCAL);
5732         if (unlikely(rc != 0)) {
5733                 CWARN("lu_env initialization failed with rc = %d,"
5734                       "cannot sync\n", rc);
5735                 return;
5736         }
5737         mdt_device_sync(&env, mdt);
5738         lu_env_fini(&env);
5739 }
5740
5741 /**
5742  * Check COS (Commit On Sharing) status.
5743  *
5744  * Return COS flag status.
5745  *
5746  * \param mdt mdt device
5747  */
5748 int mdt_cos_is_enabled(struct mdt_device *mdt)
5749 {
5750         return mdt->mdt_opts.mo_cos != 0;
5751 }
5752
5753 static struct lu_device_type_operations mdt_device_type_ops = {
5754         .ldto_device_alloc = mdt_device_alloc,
5755         .ldto_device_free  = mdt_device_free,
5756         .ldto_device_fini  = mdt_device_fini
5757 };
5758
5759 static struct lu_device_type mdt_device_type = {
5760         .ldt_tags     = LU_DEVICE_MD,
5761         .ldt_name     = LUSTRE_MDT_NAME,
5762         .ldt_ops      = &mdt_device_type_ops,
5763         .ldt_ctx_tags = LCT_MD_THREAD
5764 };
5765
5766 static int __init mdt_mod_init(void)
5767 {
5768         struct lprocfs_static_vars lvars;
5769         int rc;
5770
5771         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5772                  FID_NOBRACE_LEN + 1);
5773         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5774                  FID_LEN + 1);
5775         rc = lu_kmem_init(mdt_caches);
5776         if (rc)
5777                 return rc;
5778
5779         rc = mds_mod_init();
5780         if (rc)
5781                 GOTO(lu_fini, rc);
5782
5783         lprocfs_mdt_init_vars(&lvars);
5784         rc = class_register_type(&mdt_obd_device_ops, NULL, NULL,
5785 #ifndef HAVE_ONLY_PROCFS_SEQ
5786                                 lvars.module_vars,
5787 #endif
5788                                 LUSTRE_MDT_NAME, &mdt_device_type);
5789         if (rc)
5790                 GOTO(mds_fini, rc);
5791 lu_fini:
5792         if (rc)
5793                 lu_kmem_fini(mdt_caches);
5794 mds_fini:
5795         if (rc)
5796                 mds_mod_exit();
5797         return rc;
5798 }
5799
5800 static void __exit mdt_mod_exit(void)
5801 {
5802         class_unregister_type(LUSTRE_MDT_NAME);
5803         mds_mod_exit();
5804         lu_kmem_fini(mdt_caches);
5805 }
5806
5807 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5808 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5809 MODULE_LICENSE("GPL");
5810
5811 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);