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_mdt.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_linkea.h>
  72 #include <lustre_lfsck.h>
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 ldlm_mode_t mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE] = LCK_MINMODE,
  87         [MDL_EX]      = LCK_EX,
  88         [MDL_PW]      = LCK_PW,
  89         [MDL_PR]      = LCK_PR,
  90         [MDL_CW]      = LCK_CW,
  91         [MDL_CR]      = LCK_CR,
  92         [MDL_NL]      = LCK_NL,
  93         [MDL_GROUP]   = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 static struct lu_kmem_descr mdt_caches[] = {
 105         {
 106                 .ckd_cache = &mdt_object_kmem,
 107                 .ckd_name  = "mdt_obj",
 108                 .ckd_size  = sizeof(struct mdt_object)
 109         },
 110         {
 111                 .ckd_cache = NULL
 112         }
 113 };
 114
 115 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 116 {
 117         if (!rep)
 118                 return 0;
 119         return (rep->lock_policy_res1 & flag);
 120 }
 121
 122 void mdt_clear_disposition(struct mdt_thread_info *info,
 123                            struct ldlm_reply *rep, int flag)
 124 {
 125         if (info)
 126                 info->mti_opdata &= ~flag;
 127         if (rep)
 128                 rep->lock_policy_res1 &= ~flag;
 129 }
 130
 131 void mdt_set_disposition(struct mdt_thread_info *info,
 132                          struct ldlm_reply *rep, int flag)
 133 {
 134         if (info)
 135                 info->mti_opdata |= flag;
 136         if (rep)
 137                 rep->lock_policy_res1 |= flag;
 138 }
 139
 140 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 141 {
 142         lh->mlh_pdo_hash = 0;
 143         lh->mlh_reg_mode = lm;
 144         lh->mlh_rreg_mode = lm;
 145         lh->mlh_type = MDT_REG_LOCK;
 146 }
 147
 148 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 149                        const char *name, int namelen)
 150 {
 151         lh->mlh_reg_mode = lm;
 152         lh->mlh_rreg_mode = lm;
 153         lh->mlh_type = MDT_PDO_LOCK;
 154
 155         if (name != NULL && (name[0] != '\0')) {
 156                 LASSERT(namelen > 0);
 157                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 158                 /* XXX Workaround for LU-2856
 159                  * Zero is a valid return value of full_name_hash, but several
 160                  * users of mlh_pdo_hash assume a non-zero hash value. We
 161                  * therefore map zero onto an arbitrary, but consistent
 162                  * value (1) to avoid problems further down the road. */
 163                 if (unlikely(!lh->mlh_pdo_hash))
 164                         lh->mlh_pdo_hash = 1;
 165         } else {
 166                 LASSERT(namelen == 0);
 167                 lh->mlh_pdo_hash = 0ull;
 168         }
 169 }
 170
 171 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 172                               struct mdt_lock_handle *lh)
 173 {
 174         mdl_mode_t mode;
 175         ENTRY;
 176
 177         /*
 178          * Any dir access needs couple of locks:
 179          *
 180          * 1) on part of dir we gonna take lookup/modify;
 181          *
 182          * 2) on whole dir to protect it from concurrent splitting and/or to
 183          * flush client's cache for readdir().
 184          *
 185          * so, for a given mode and object this routine decides what lock mode
 186          * to use for lock #2:
 187          *
 188          * 1) if caller's gonna lookup in dir then we need to protect dir from
 189          * being splitted only - LCK_CR
 190          *
 191          * 2) if caller's gonna modify dir then we need to protect dir from
 192          * being splitted and to flush cache - LCK_CW
 193          *
 194          * 3) if caller's gonna modify dir and that dir seems ready for
 195          * splitting then we need to protect it from any type of access
 196          * (lookup/modify/split) - LCK_EX --bzzz
 197          */
 198
 199         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 200         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 201
 202         /*
 203          * Ask underlaying level its opinion about preferable PDO lock mode
 204          * having access type passed as regular lock mode:
 205          *
 206          * - MDL_MINMODE means that lower layer does not want to specify lock
 207          * mode;
 208          *
 209          * - MDL_NL means that no PDO lock should be taken. This is used in some
 210          * cases. Say, for non-splittable directories no need to use PDO locks
 211          * at all.
 212          */
 213         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 214                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 215
 216         if (mode != MDL_MINMODE) {
 217                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 218         } else {
 219                 /*
 220                  * Lower layer does not want to specify locking mode. We do it
 221                  * our selves. No special protection is needed, just flush
 222                  * client's cache on modification and allow concurrent
 223                  * mondification.
 224                  */
 225                 switch (lh->mlh_reg_mode) {
 226                 case LCK_EX:
 227                         lh->mlh_pdo_mode = LCK_EX;
 228                         break;
 229                 case LCK_PR:
 230                         lh->mlh_pdo_mode = LCK_CR;
 231                         break;
 232                 case LCK_PW:
 233                         lh->mlh_pdo_mode = LCK_CW;
 234                         break;
 235                 default:
 236                         CERROR("Not expected lock type (0x%x)\n",
 237                                (int)lh->mlh_reg_mode);
 238                         LBUG();
 239                 }
 240         }
 241
 242         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 243         EXIT;
 244 }
 245
 246 int mdt_getstatus(struct mdt_thread_info *info)
 247 {
 248         struct mdt_device       *mdt  = info->mti_mdt;
 249         struct mdt_body         *repbody;
 250         int                     rc;
 251         ENTRY;
 252
 253         rc = mdt_check_ucred(info);
 254         if (rc)
 255                 RETURN(err_serious(rc));
 256
 257         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 258                 RETURN(err_serious(-ENOMEM));
 259
 260         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 261         repbody->fid1 = mdt->mdt_md_root_fid;
 262         repbody->valid |= OBD_MD_FLID;
 263
 264         if (mdt->mdt_opts.mo_mds_capa &&
 265             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 266                 struct mdt_object  *root;
 267                 struct lustre_capa *capa;
 268
 269                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 270                 if (IS_ERR(root))
 271                         RETURN(PTR_ERR(root));
 272
 273                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 274                 LASSERT(capa);
 275                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 276                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 277                                  0);
 278                 mdt_object_put(info->mti_env, root);
 279                 if (rc == 0)
 280                         repbody->valid |= OBD_MD_FLMDSCAPA;
 281         }
 282
 283         RETURN(rc);
 284 }
 285
 286 int mdt_statfs(struct mdt_thread_info *info)
 287 {
 288         struct ptlrpc_request           *req = mdt_info_req(info);
 289         struct md_device                *next = info->mti_mdt->mdt_child;
 290         struct ptlrpc_service_part      *svcpt;
 291         struct obd_statfs               *osfs;
 292         int                             rc;
 293
 294         ENTRY;
 295
 296         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
 297
 298         /* This will trigger a watchdog timeout */
 299         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 300                          (MDT_SERVICE_WATCHDOG_FACTOR *
 301                           at_get(&svcpt->scp_at_estimate)) + 1);
 302
 303         rc = mdt_check_ucred(info);
 304         if (rc)
 305                 RETURN(err_serious(rc));
 306
 307         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 308                 RETURN(err_serious(-ENOMEM));
 309
 310         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 311         if (!osfs)
 312                 RETURN(-EPROTO);
 313
 314         /** statfs information are cached in the mdt_device */
 315         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 316                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 317                 /** statfs data is too old, get up-to-date one */
 318                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 319                 if (rc)
 320                         RETURN(rc);
 321                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 322                 info->mti_mdt->mdt_osfs = *osfs;
 323                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 324                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 325         } else {
 326                 /** use cached statfs data */
 327                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 328                 *osfs = info->mti_mdt->mdt_osfs;
 329                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 330         }
 331
 332         if (rc == 0)
 333                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 334
 335         RETURN(rc);
 336 }
 337
 338 /**
 339  * Pack SOM attributes into the reply.
 340  * Call under a DLM UPDATE lock.
 341  */
 342 static void mdt_pack_size2body(struct mdt_thread_info *info,
 343                                struct mdt_object *mo)
 344 {
 345         struct mdt_body *b;
 346         struct md_attr *ma = &info->mti_attr;
 347
 348         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 349         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 350
 351         /* Check if Size-on-MDS is supported, if this is a regular file,
 352          * if SOM is enabled on the object and if SOM cache exists and valid.
 353          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 354         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 355             !S_ISREG(ma->ma_attr.la_mode) ||
 356             !mdt_object_is_som_enabled(mo) ||
 357             !(ma->ma_valid & MA_SOM))
 358                 return;
 359
 360         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 361         b->size = ma->ma_som->msd_size;
 362         b->blocks = ma->ma_som->msd_blocks;
 363 }
 364
 365 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 366                         const struct lu_attr *attr, const struct lu_fid *fid)
 367 {
 368         struct md_attr *ma = &info->mti_attr;
 369
 370         LASSERT(ma->ma_valid & MA_INODE);
 371
 372         b->atime      = attr->la_atime;
 373         b->mtime      = attr->la_mtime;
 374         b->ctime      = attr->la_ctime;
 375         b->mode       = attr->la_mode;
 376         b->size       = attr->la_size;
 377         b->blocks     = attr->la_blocks;
 378         b->uid        = attr->la_uid;
 379         b->gid        = attr->la_gid;
 380         b->flags      = attr->la_flags;
 381         b->nlink      = attr->la_nlink;
 382         b->rdev       = attr->la_rdev;
 383
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode)) {
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 393                 /* means no objects are allocated on osts. */
 394                 LASSERT(!(ma->ma_valid & MA_LOV));
 395                 /* just ignore blocks occupied by extend attributes on MDS */
 396                 b->blocks = 0;
 397                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 398                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 399         }
 400
 401         if (fid) {
 402                 b->fid1 = *fid;
 403                 b->valid |= OBD_MD_FLID;
 404                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 405                                 PFID(fid), b->nlink, b->mode, b->size);
 406         }
 407
 408         if (info)
 409                 mdt_body_reverse_idmap(info, b);
 410
 411         if (b->valid & OBD_MD_FLSIZE)
 412                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 413                        PFID(fid), (unsigned long long)b->size);
 414 }
 415
 416 static inline int mdt_body_has_lov(const struct lu_attr *la,
 417                                    const struct mdt_body *body)
 418 {
 419         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 420                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 421 }
 422
 423 void mdt_client_compatibility(struct mdt_thread_info *info)
 424 {
 425         struct mdt_body       *body;
 426         struct ptlrpc_request *req = mdt_info_req(info);
 427         struct obd_export     *exp = req->rq_export;
 428         struct md_attr        *ma = &info->mti_attr;
 429         struct lu_attr        *la = &ma->ma_attr;
 430         ENTRY;
 431
 432         if (exp_connect_layout(exp))
 433                 /* the client can deal with 16-bit lmm_stripe_count */
 434                 RETURN_EXIT;
 435
 436         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 437
 438         if (!mdt_body_has_lov(la, body))
 439                 RETURN_EXIT;
 440
 441         /* now we have a reply with a lov for a client not compatible with the
 442          * layout lock so we have to clean the layout generation number */
 443         if (S_ISREG(la->la_mode))
 444                 ma->ma_lmm->lmm_layout_gen = 0;
 445         EXIT;
 446 }
 447
 448 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 449                              char *name)
 450 {
 451         const struct lu_env *env = info->mti_env;
 452         int rc;
 453         ENTRY;
 454
 455         LASSERT(info->mti_big_lmm_used == 0);
 456         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 457         if (rc < 0)
 458                 RETURN(rc);
 459
 460         /* big_lmm may need to be grown */
 461         if (info->mti_big_lmmsize < rc) {
 462                 int size = size_roundup_power2(rc);
 463
 464                 if (info->mti_big_lmmsize > 0) {
 465                         /* free old buffer */
 466                         LASSERT(info->mti_big_lmm);
 467                         OBD_FREE_LARGE(info->mti_big_lmm,
 468                                        info->mti_big_lmmsize);
 469                         info->mti_big_lmm = NULL;
 470                         info->mti_big_lmmsize = 0;
 471                 }
 472
 473                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 474                 if (info->mti_big_lmm == NULL)
 475                         RETURN(-ENOMEM);
 476                 info->mti_big_lmmsize = size;
 477         }
 478         LASSERT(info->mti_big_lmmsize >= rc);
 479
 480         info->mti_buf.lb_buf = info->mti_big_lmm;
 481         info->mti_buf.lb_len = info->mti_big_lmmsize;
 482         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 483
 484         RETURN(rc);
 485 }
 486
 487 int mdt_attr_get_lov(struct mdt_thread_info *info,
 488                      struct mdt_object *o, struct md_attr *ma)
 489 {
 490         struct md_object *next = mdt_object_child(o);
 491         struct lu_buf    *buf = &info->mti_buf;
 492         int rc;
 493
 494         buf->lb_buf = ma->ma_lmm;
 495         buf->lb_len = ma->ma_lmm_size;
 496         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 497         if (rc > 0) {
 498                 ma->ma_lmm_size = rc;
 499                 ma->ma_valid |= MA_LOV;
 500                 rc = 0;
 501         } else if (rc == -ENODATA) {
 502                 /* no LOV EA */
 503                 rc = 0;
 504         } else if (rc == -ERANGE) {
 505                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 506                 if (rc > 0) {
 507                         info->mti_big_lmm_used = 1;
 508                         ma->ma_valid |= MA_LOV;
 509                         ma->ma_lmm = info->mti_big_lmm;
 510                         ma->ma_lmm_size = rc;
 511                         /* update mdt_max_mdsize so all clients
 512                          * will be aware about that */
 513                         if (info->mti_mdt->mdt_max_mdsize < rc)
 514                                 info->mti_mdt->mdt_max_mdsize = rc;
 515                         rc = 0;
 516                 }
 517         }
 518
 519         return rc;
 520 }
 521
 522 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 523                       struct mdt_object *o, struct lu_fid *pfid)
 524 {
 525         struct lu_buf           *buf = &info->mti_buf;
 526         struct link_ea_header   *leh;
 527         struct link_ea_entry    *lee;
 528         int                      rc;
 529         ENTRY;
 530
 531         buf->lb_buf = info->mti_big_lmm;
 532         buf->lb_len = info->mti_big_lmmsize;
 533         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 534                           buf, XATTR_NAME_LINK);
 535         /* ignore errors, MA_PFID won't be set and it is
 536          * up to the caller to treat this as an error */
 537         if (rc == -ERANGE || buf->lb_len == 0) {
 538                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 539                 buf->lb_buf = info->mti_big_lmm;
 540                 buf->lb_len = info->mti_big_lmmsize;
 541         }
 542
 543         if (rc < 0)
 544                 RETURN(rc);
 545         if (rc < sizeof(*leh)) {
 546                 CERROR("short LinkEA on "DFID": rc = %d\n",
 547                        PFID(mdt_object_fid(o)), rc);
 548                 RETURN(-ENODATA);
 549         }
 550
 551         leh = (struct link_ea_header *) buf->lb_buf;
 552         lee = (struct link_ea_entry *)(leh + 1);
 553         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 554                 leh->leh_magic = LINK_EA_MAGIC;
 555                 leh->leh_reccount = __swab32(leh->leh_reccount);
 556                 leh->leh_len = __swab64(leh->leh_len);
 557         }
 558         if (leh->leh_magic != LINK_EA_MAGIC)
 559                 RETURN(-EINVAL);
 560         if (leh->leh_reccount == 0)
 561                 RETURN(-ENODATA);
 562
 563         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 564         fid_be_to_cpu(pfid, pfid);
 565
 566         RETURN(0);
 567 }
 568
 569 int mdt_attr_get_complex(struct mdt_thread_info *info,
 570                          struct mdt_object *o, struct md_attr *ma)
 571 {
 572         const struct lu_env *env = info->mti_env;
 573         struct md_object    *next = mdt_object_child(o);
 574         struct lu_buf       *buf = &info->mti_buf;
 575         u32                  mode = lu_object_attr(&next->mo_lu);
 576         int                  need = ma->ma_need;
 577         int                  rc = 0, rc2;
 578         ENTRY;
 579
 580         ma->ma_valid = 0;
 581
 582         if (need & MA_INODE) {
 583                 ma->ma_need = MA_INODE;
 584                 rc = mo_attr_get(env, next, ma);
 585                 if (rc)
 586                         GOTO(out, rc);
 587                 ma->ma_valid |= MA_INODE;
 588         }
 589
 590         if (need & MA_PFID) {
 591                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 592                 if (rc == 0)
 593                         ma->ma_valid |= MA_PFID;
 594                 /* ignore this error, parent fid is not mandatory */
 595                 rc = 0;
 596         }
 597
 598         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 599                 rc = mdt_attr_get_lov(info, o, ma);
 600                 if (rc)
 601                         GOTO(out, rc);
 602         }
 603
 604         if (need & MA_LMV && S_ISDIR(mode)) {
 605                 buf->lb_buf = ma->ma_lmv;
 606                 buf->lb_len = ma->ma_lmv_size;
 607                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 608                 if (rc2 > 0) {
 609                         ma->ma_lmv_size = rc2;
 610                         ma->ma_valid |= MA_LMV;
 611                 } else if (rc2 == -ENODATA) {
 612                         /* no LMV EA */
 613                         ma->ma_lmv_size = 0;
 614                 } else
 615                         GOTO(out, rc = rc2);
 616         }
 617
 618         if (need & MA_SOM && S_ISREG(mode)) {
 619                 buf->lb_buf = info->mti_xattr_buf;
 620                 buf->lb_len = sizeof(info->mti_xattr_buf);
 621                 CLASSERT(sizeof(struct som_attrs) <=
 622                          sizeof(info->mti_xattr_buf));
 623                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 624                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 625                 if (rc2 == 0)
 626                         ma->ma_valid |= MA_SOM;
 627                 else if (rc2 < 0 && rc2 != -ENODATA)
 628                         GOTO(out, rc = rc2);
 629         }
 630
 631         if (need & MA_HSM && S_ISREG(mode)) {
 632                 buf->lb_buf = info->mti_xattr_buf;
 633                 buf->lb_len = sizeof(info->mti_xattr_buf);
 634                 CLASSERT(sizeof(struct hsm_attrs) <=
 635                          sizeof(info->mti_xattr_buf));
 636                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 637                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 638                 if (rc2 == 0)
 639                         ma->ma_valid |= MA_HSM;
 640                 else if (rc2 < 0 && rc2 != -ENODATA)
 641                         GOTO(out, rc = rc2);
 642         }
 643
 644 #ifdef CONFIG_FS_POSIX_ACL
 645         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 646                 buf->lb_buf = ma->ma_acl;
 647                 buf->lb_len = ma->ma_acl_size;
 648                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 649                 if (rc2 > 0) {
 650                         ma->ma_acl_size = rc2;
 651                         ma->ma_valid |= MA_ACL_DEF;
 652                 } else if (rc2 == -ENODATA) {
 653                         /* no ACLs */
 654                         ma->ma_acl_size = 0;
 655                 } else
 656                         GOTO(out, rc = rc2);
 657         }
 658 #endif
 659 out:
 660         ma->ma_need = need;
 661         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 662                rc, ma->ma_valid, ma->ma_lmm);
 663         RETURN(rc);
 664 }
 665
 666 static int mdt_getattr_internal(struct mdt_thread_info *info,
 667                                 struct mdt_object *o, int ma_need)
 668 {
 669         struct md_object        *next = mdt_object_child(o);
 670         const struct mdt_body   *reqbody = info->mti_body;
 671         struct ptlrpc_request   *req = mdt_info_req(info);
 672         struct md_attr          *ma = &info->mti_attr;
 673         struct lu_attr          *la = &ma->ma_attr;
 674         struct req_capsule      *pill = info->mti_pill;
 675         const struct lu_env     *env = info->mti_env;
 676         struct mdt_body         *repbody;
 677         struct lu_buf           *buffer = &info->mti_buf;
 678         int                     rc;
 679         int                     is_root;
 680         ENTRY;
 681
 682         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 683                 RETURN(err_serious(-ENOMEM));
 684
 685         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 686
 687         ma->ma_valid = 0;
 688
 689         if (mdt_object_remote(o)) {
 690                 /* This object is located on remote node.*/
 691                 /* Return -EIO for old client */
 692                 if (!mdt_is_dne_client(req->rq_export))
 693                         GOTO(out, rc = -EIO);
 694
 695                 repbody->fid1 = *mdt_object_fid(o);
 696                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 697                 GOTO(out, rc = 0);
 698         }
 699
 700         buffer->lb_len = reqbody->eadatasize;
 701         if (buffer->lb_len > 0)
 702                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 703         else
 704                 buffer->lb_buf = NULL;
 705
 706         /* If it is dir object and client require MEA, then we got MEA */
 707         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 708             reqbody->valid & OBD_MD_MEA) {
 709                 /* Assumption: MDT_MD size is enough for lmv size. */
 710                 ma->ma_lmv = buffer->lb_buf;
 711                 ma->ma_lmv_size = buffer->lb_len;
 712                 ma->ma_need = MA_LMV | MA_INODE;
 713         } else {
 714                 ma->ma_lmm = buffer->lb_buf;
 715                 ma->ma_lmm_size = buffer->lb_len;
 716                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 717         }
 718
 719         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 720             reqbody->valid & OBD_MD_FLDIREA  &&
 721             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 722                 /* get default stripe info for this dir. */
 723                 ma->ma_need |= MA_LOV_DEF;
 724         }
 725         ma->ma_need |= ma_need;
 726         if (ma->ma_need & MA_SOM)
 727                 ma->ma_som = &info->mti_u.som.data;
 728
 729         rc = mdt_attr_get_complex(info, o, ma);
 730         if (unlikely(rc)) {
 731                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 732                        mdt_obd_name(info->mti_mdt),
 733                        PFID(mdt_object_fid(o)), rc);
 734                 RETURN(rc);
 735         }
 736
 737         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 738
 739         /* the Lustre protocol supposes to return default striping
 740          * on the user-visible root if explicitly requested */
 741         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 742             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
 743                 struct lu_fid      rootfid;
 744                 struct mdt_object *root;
 745                 struct mdt_device *mdt = info->mti_mdt;
 746
 747                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 748                 if (rc)
 749                         RETURN(rc);
 750                 root = mdt_object_find(env, mdt, &rootfid);
 751                 if (IS_ERR(root))
 752                         RETURN(PTR_ERR(root));
 753                 rc = mdt_attr_get_lov(info, root, ma);
 754                 mdt_object_put(info->mti_env, root);
 755                 if (unlikely(rc)) {
 756                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 757                                mdt_obd_name(info->mti_mdt),
 758                                PFID(mdt_object_fid(o)), rc);
 759                         RETURN(rc);
 760                 }
 761         }
 762
 763         if (likely(ma->ma_valid & MA_INODE))
 764                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 765         else
 766                 RETURN(-EFAULT);
 767
 768         if (mdt_body_has_lov(la, reqbody)) {
 769                 if (ma->ma_valid & MA_LOV) {
 770                         LASSERT(ma->ma_lmm_size);
 771                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 772                         repbody->eadatasize = ma->ma_lmm_size;
 773                         if (S_ISDIR(la->la_mode))
 774                                 repbody->valid |= OBD_MD_FLDIREA;
 775                         else
 776                                 repbody->valid |= OBD_MD_FLEASIZE;
 777                 }
 778                 if (ma->ma_valid & MA_LMV) {
 779                         LASSERT(S_ISDIR(la->la_mode));
 780                         repbody->eadatasize = ma->ma_lmv_size;
 781                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 782                 }
 783         } else if (S_ISLNK(la->la_mode) &&
 784                    reqbody->valid & OBD_MD_LINKNAME) {
 785                 buffer->lb_buf = ma->ma_lmm;
 786                 /* eadatasize from client includes NULL-terminator, so
 787                  * there is no need to read it */
 788                 buffer->lb_len = reqbody->eadatasize - 1;
 789                 rc = mo_readlink(env, next, buffer);
 790                 if (unlikely(rc <= 0)) {
 791                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 792                                mdt_obd_name(info->mti_mdt),
 793                                PFID(mdt_object_fid(o)), rc);
 794                         rc = -EFAULT;
 795                 } else {
 796                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 797
 798                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 799                                 rc -= 2;
 800                         repbody->valid |= OBD_MD_LINKNAME;
 801                         /* we need to report back size with NULL-terminator
 802                          * because client expects that */
 803                         repbody->eadatasize = rc + 1;
 804                         if (repbody->eadatasize != reqbody->eadatasize)
 805                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 806                                        "on "DFID ", expected %d\n",
 807                                        mdt_obd_name(info->mti_mdt),
 808                                        rc, PFID(mdt_object_fid(o)),
 809                                        reqbody->eadatasize - 1);
 810                         /* NULL terminate */
 811                         ((char *)ma->ma_lmm)[rc] = 0;
 812
 813                         /* If the total CDEBUG() size is larger than a page, it
 814                          * will print a warning to the console, avoid this by
 815                          * printing just the last part of the symlink. */
 816                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 817                                print_limit < rc ? "..." : "", print_limit,
 818                                (char *)ma->ma_lmm + rc - print_limit, rc);
 819                         rc = 0;
 820                 }
 821         }
 822
 823         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 824                 repbody->max_cookiesize = 0;
 825                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 826                 repbody->valid |= OBD_MD_FLMODEASIZE;
 827                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 828                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 829                        repbody->max_cookiesize);
 830         }
 831
 832         if (exp_connect_rmtclient(info->mti_exp) &&
 833             reqbody->valid & OBD_MD_FLRMTPERM) {
 834                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 835
 836                 /* mdt_getattr_lock only */
 837                 rc = mdt_pack_remote_perm(info, o, buf);
 838                 if (rc) {
 839                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 840                         repbody->aclsize = 0;
 841                         RETURN(rc);
 842                 } else {
 843                         repbody->valid |= OBD_MD_FLRMTPERM;
 844                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 845                 }
 846         }
 847 #ifdef CONFIG_FS_POSIX_ACL
 848         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 849                  (reqbody->valid & OBD_MD_FLACL)) {
 850                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 851                 buffer->lb_len = req_capsule_get_size(pill,
 852                                                       &RMF_ACL, RCL_SERVER);
 853                 if (buffer->lb_len > 0) {
 854                         rc = mo_xattr_get(env, next, buffer,
 855                                           XATTR_NAME_ACL_ACCESS);
 856                         if (rc < 0) {
 857                                 if (rc == -ENODATA) {
 858                                         repbody->aclsize = 0;
 859                                         repbody->valid |= OBD_MD_FLACL;
 860                                         rc = 0;
 861                                 } else if (rc == -EOPNOTSUPP) {
 862                                         rc = 0;
 863                                 } else {
 864                                         CERROR("%s: unable to read "DFID
 865                                                " ACL: rc = %d\n",
 866                                                mdt_obd_name(info->mti_mdt),
 867                                                PFID(mdt_object_fid(o)), rc);
 868                                 }
 869                         } else {
 870                                 repbody->aclsize = rc;
 871                                 repbody->valid |= OBD_MD_FLACL;
 872                                 rc = 0;
 873                         }
 874                 }
 875         }
 876 #endif
 877
 878         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 879             info->mti_mdt->mdt_opts.mo_mds_capa &&
 880             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 881                 struct lustre_capa *capa;
 882
 883                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 884                 LASSERT(capa);
 885                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 886                 rc = mo_capa_get(env, next, capa, 0);
 887                 if (rc)
 888                         RETURN(rc);
 889                 repbody->valid |= OBD_MD_FLMDSCAPA;
 890         }
 891
 892 out:
 893         if (rc == 0)
 894                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 895
 896         RETURN(rc);
 897 }
 898
 899 static int mdt_renew_capa(struct mdt_thread_info *info)
 900 {
 901         struct mdt_object  *obj = info->mti_object;
 902         struct mdt_body    *body;
 903         struct lustre_capa *capa, *c;
 904         int rc;
 905         ENTRY;
 906
 907         /* if object doesn't exist, or server has disabled capability,
 908          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 909          * flag not set.
 910          */
 911         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 912             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 913                 RETURN(0);
 914
 915         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 916         LASSERT(body != NULL);
 917
 918         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 919         LASSERT(c);
 920
 921         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 922         LASSERT(capa);
 923
 924         *capa = *c;
 925         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 926         if (rc == 0)
 927                 body->valid |= OBD_MD_FLOSSCAPA;
 928         RETURN(rc);
 929 }
 930
 931 int mdt_getattr(struct mdt_thread_info *info)
 932 {
 933         struct mdt_object       *obj = info->mti_object;
 934         struct req_capsule      *pill = info->mti_pill;
 935         struct mdt_body         *reqbody;
 936         struct mdt_body         *repbody;
 937         mode_t                   mode;
 938         int rc, rc2;
 939         ENTRY;
 940
 941         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 942         LASSERT(reqbody);
 943
 944         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 945                 rc = req_capsule_server_pack(pill);
 946                 if (unlikely(rc))
 947                         RETURN(err_serious(rc));
 948                 rc = mdt_renew_capa(info);
 949                 GOTO(out_shrink, rc);
 950         }
 951
 952         LASSERT(obj != NULL);
 953         LASSERT(lu_object_assert_exists(&obj->mot_obj));
 954
 955         mode = lu_object_attr(&obj->mot_obj);
 956
 957         /* old clients may not report needed easize, use max value then */
 958         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 959                              reqbody->eadatasize == 0 ?
 960                              info->mti_mdt->mdt_max_mdsize :
 961                              reqbody->eadatasize);
 962
 963         rc = req_capsule_server_pack(pill);
 964         if (unlikely(rc != 0))
 965                 RETURN(err_serious(rc));
 966
 967         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 968         LASSERT(repbody != NULL);
 969         repbody->eadatasize = 0;
 970         repbody->aclsize = 0;
 971
 972         if (reqbody->valid & OBD_MD_FLRMTPERM)
 973                 rc = mdt_init_ucred(info, reqbody);
 974         else
 975                 rc = mdt_check_ucred(info);
 976         if (unlikely(rc))
 977                 GOTO(out_shrink, rc);
 978
 979         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 980
 981         /*
 982          * Don't check capability at all, because rename might getattr for
 983          * remote obj, and at that time no capability is available.
 984          */
 985         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 986         rc = mdt_getattr_internal(info, obj, 0);
 987         if (reqbody->valid & OBD_MD_FLRMTPERM)
 988                 mdt_exit_ucred(info);
 989         EXIT;
 990 out_shrink:
 991         mdt_client_compatibility(info);
 992         rc2 = mdt_fix_reply(info);
 993         if (rc == 0)
 994                 rc = rc2;
 995         return rc;
 996 }
 997
 998 int mdt_is_subdir(struct mdt_thread_info *info)
 999 {
1000         struct mdt_object     *o = info->mti_object;
1001         struct req_capsule    *pill = info->mti_pill;
1002         const struct mdt_body *body = info->mti_body;
1003         struct mdt_body       *repbody;
1004         int                    rc;
1005         ENTRY;
1006
1007         LASSERT(o != NULL);
1008
1009         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1010
1011         /*
1012          * We save last checked parent fid to @repbody->fid1 for remote
1013          * directory case.
1014          */
1015         LASSERT(fid_is_sane(&body->fid2));
1016         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1017         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1018                            &body->fid2, &repbody->fid1);
1019         if (rc == 0 || rc == -EREMOTE)
1020                 repbody->valid |= OBD_MD_FLID;
1021
1022         RETURN(rc);
1023 }
1024
1025 int mdt_swap_layouts(struct mdt_thread_info *info)
1026 {
1027         struct ptlrpc_request   *req = mdt_info_req(info);
1028         struct obd_export       *exp = req->rq_export;
1029         struct mdt_object       *o1, *o2, *o;
1030         struct mdt_lock_handle  *lh1, *lh2;
1031         struct mdc_swap_layouts *msl;
1032         int                      rc;
1033         ENTRY;
1034
1035         /* client does not support layout lock, so layout swaping
1036          * is disabled.
1037          * FIXME: there is a problem for old clients which don't support
1038          * layout lock yet. If those clients have already opened the file
1039          * they won't be notified at all so that old layout may still be
1040          * used to do IO. This can be fixed after file release is landed by
1041          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1042         if (!exp_connect_layout(exp))
1043                 RETURN(-EOPNOTSUPP);
1044
1045         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1046                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1047                                  req_capsule_client_get(info->mti_pill,
1048                                                         &RMF_CAPA1));
1049
1050         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1051                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1052                                  req_capsule_client_get(info->mti_pill,
1053                                                         &RMF_CAPA2));
1054
1055         o1 = info->mti_object;
1056         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1057                                 &info->mti_body->fid2);
1058         if (IS_ERR(o))
1059                 GOTO(out, rc = PTR_ERR(o));
1060
1061         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1062                 GOTO(put, rc = -ENOENT);
1063
1064         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1065         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1066                 GOTO(put, rc);
1067
1068         if (rc < 0)
1069                 swap(o1, o2);
1070
1071         /* permission check. Make sure the calling process having permission
1072          * to write both files. */
1073         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1074                                 MAY_WRITE);
1075         if (rc < 0)
1076                 GOTO(put, rc);
1077
1078         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1079                                 MAY_WRITE);
1080         if (rc < 0)
1081                 GOTO(put, rc);
1082
1083         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1084         if (msl == NULL)
1085                 GOTO(put, rc = -EPROTO);
1086
1087         lh1 = &info->mti_lh[MDT_LH_NEW];
1088         mdt_lock_reg_init(lh1, LCK_EX);
1089         lh2 = &info->mti_lh[MDT_LH_OLD];
1090         mdt_lock_reg_init(lh2, LCK_EX);
1091
1092         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT,
1093                              MDT_LOCAL_LOCK);
1094         if (rc < 0)
1095                 GOTO(put, rc);
1096
1097         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT,
1098                              MDT_LOCAL_LOCK);
1099         if (rc < 0)
1100                 GOTO(unlock1, rc);
1101
1102         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1103                              mdt_object_child(o2), msl->msl_flags);
1104         GOTO(unlock2, rc);
1105 unlock2:
1106         mdt_object_unlock(info, o2, lh2, rc);
1107 unlock1:
1108         mdt_object_unlock(info, o1, lh1, rc);
1109 put:
1110         mdt_object_put(info->mti_env, o);
1111 out:
1112         RETURN(rc);
1113 }
1114
1115 static int mdt_raw_lookup(struct mdt_thread_info *info,
1116                           struct mdt_object *parent,
1117                           const struct lu_name *lname,
1118                           struct ldlm_reply *ldlm_rep)
1119 {
1120         struct md_object *next = mdt_object_child(info->mti_object);
1121         const struct mdt_body *reqbody = info->mti_body;
1122         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1123         struct mdt_body *repbody;
1124         int rc;
1125         ENTRY;
1126
1127         if (reqbody->valid != OBD_MD_FLID)
1128                 RETURN(0);
1129
1130         LASSERT(!info->mti_cross_ref);
1131
1132         /* Only got the fid of this obj by name */
1133         fid_zero(child_fid);
1134         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1135                         &info->mti_spec);
1136 #if 0
1137         /* XXX is raw_lookup possible as intent operation? */
1138         if (rc != 0) {
1139                 if (rc == -ENOENT)
1140                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1141                 RETURN(rc);
1142         } else
1143                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1144
1145         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1146 #endif
1147         if (rc == 0) {
1148                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1149                 repbody->fid1 = *child_fid;
1150                 repbody->valid = OBD_MD_FLID;
1151         }
1152         RETURN(1);
1153 }
1154
1155 /*
1156  * UPDATE lock should be taken against parent, and be release before exit;
1157  * child_bits lock should be taken against child, and be returned back:
1158  *            (1)normal request should release the child lock;
1159  *            (2)intent request will grant the lock to client.
1160  */
1161 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1162                                  struct mdt_lock_handle *lhc,
1163                                  __u64 child_bits,
1164                                  struct ldlm_reply *ldlm_rep)
1165 {
1166         struct ptlrpc_request  *req       = mdt_info_req(info);
1167         struct mdt_body        *reqbody   = NULL;
1168         struct mdt_object      *parent    = info->mti_object;
1169         struct mdt_object      *child;
1170         struct md_object       *next      = mdt_object_child(parent);
1171         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1172         struct lu_name         *lname     = NULL;
1173         const char             *name      = NULL;
1174         int                     namelen   = 0;
1175         struct mdt_lock_handle *lhp       = NULL;
1176         struct ldlm_lock       *lock;
1177         struct ldlm_res_id     *res_id;
1178         int                     is_resent;
1179         int                     ma_need = 0;
1180         int                     rc;
1181
1182         ENTRY;
1183
1184         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1185         LASSERT(ergo(is_resent,
1186                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1187
1188         LASSERT(parent != NULL);
1189         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1190         if (name == NULL)
1191                 RETURN(err_serious(-EFAULT));
1192
1193         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1194                                        RCL_CLIENT) - 1;
1195         if (!info->mti_cross_ref) {
1196                 /*
1197                  * XXX: Check for "namelen == 0" is for getattr by fid
1198                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1199                  * that is the name must contain at least one character and
1200                  * the terminating '\0'
1201                  */
1202                 if (namelen == 0) {
1203                         reqbody = req_capsule_client_get(info->mti_pill,
1204                                                          &RMF_MDT_BODY);
1205                         if (unlikely(reqbody == NULL))
1206                                 RETURN(err_serious(-EFAULT));
1207
1208                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1209                                 RETURN(err_serious(-EINVAL));
1210
1211                         name = NULL;
1212                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1213                                "ldlm_rep = %p\n",
1214                                PFID(mdt_object_fid(parent)),
1215                                PFID(&reqbody->fid2), ldlm_rep);
1216                 } else {
1217                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1218                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1219                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1220                                name, ldlm_rep);
1221                 }
1222         }
1223         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1224
1225         if (unlikely(!mdt_object_exists(parent))) {
1226                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1227                                 &parent->mot_obj,
1228                                 "Parent doesn't exist!\n");
1229                 RETURN(-ESTALE);
1230         } else if (!info->mti_cross_ref) {
1231                 LASSERTF(!mdt_object_remote(parent),
1232                          "Parent "DFID" is on remote server\n",
1233                          PFID(mdt_object_fid(parent)));
1234         }
1235         if (lname) {
1236                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1237                 if (rc != 0) {
1238                         if (rc > 0)
1239                                 rc = 0;
1240                         RETURN(rc);
1241                 }
1242         }
1243
1244         if (info->mti_cross_ref) {
1245                 /* Only getattr on the child. Parent is on another node. */
1246                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1247                 child = parent;
1248                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1249                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1250
1251                 if (is_resent) {
1252                         /* Do not take lock for resent case. */
1253                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1254                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1255                                  lhc->mlh_reg_lh.cookie);
1256                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1257                                                 &lock->l_resource->lr_name));
1258                         LDLM_LOCK_PUT(lock);
1259                         rc = 0;
1260                 } else {
1261                         mdt_lock_handle_init(lhc);
1262                         mdt_lock_reg_init(lhc, LCK_PR);
1263
1264                         /*
1265                          * Object's name is on another MDS, no lookup lock is
1266                          * needed here but update is.
1267                          */
1268                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1269                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1270
1271                         rc = mdt_object_lock(info, child, lhc, child_bits,
1272                                              MDT_LOCAL_LOCK);
1273                 }
1274                 if (rc == 0) {
1275                         /* Finally, we can get attr for child. */
1276                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1277                                          BYPASS_CAPA);
1278                         rc = mdt_getattr_internal(info, child, 0);
1279                         if (unlikely(rc != 0))
1280                                 mdt_object_unlock(info, child, lhc, 1);
1281                 }
1282                 RETURN(rc);
1283         }
1284
1285         if (lname) {
1286                 /* step 1: lock parent only if parent is a directory */
1287                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1288                         lhp = &info->mti_lh[MDT_LH_PARENT];
1289                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1290                         rc = mdt_object_lock(info, parent, lhp,
1291                                              MDS_INODELOCK_UPDATE,
1292                                              MDT_LOCAL_LOCK);
1293                         if (unlikely(rc != 0))
1294                                 RETURN(rc);
1295                 }
1296
1297                 /* step 2: lookup child's fid by name */
1298                 fid_zero(child_fid);
1299                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1300                                 &info->mti_spec);
1301
1302                 if (rc != 0) {
1303                         if (rc == -ENOENT)
1304                                 mdt_set_disposition(info, ldlm_rep,
1305                                                     DISP_LOOKUP_NEG);
1306                         GOTO(out_parent, rc);
1307                 } else
1308                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1309         } else {
1310                 *child_fid = reqbody->fid2;
1311                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1312         }
1313
1314         /*
1315          *step 3: find the child object by fid & lock it.
1316          *        regardless if it is local or remote.
1317          */
1318         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1319
1320         if (unlikely(IS_ERR(child)))
1321                 GOTO(out_parent, rc = PTR_ERR(child));
1322         if (is_resent) {
1323                 /* Do not take lock for resent case. */
1324                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1325                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1326                          lhc->mlh_reg_lh.cookie);
1327
1328                 res_id = &lock->l_resource->lr_name;
1329                 if (!fid_res_name_eq(mdt_object_fid(child),
1330                                     &lock->l_resource->lr_name)) {
1331                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1332                                                  &lock->l_resource->lr_name),
1333                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1334                                  (unsigned long)res_id->name[0],
1335                                  (unsigned long)res_id->name[1],
1336                                  (unsigned long)res_id->name[2],
1337                                  PFID(mdt_object_fid(parent)));
1338                           CWARN("Although resent, but still not get child lock"
1339                                 "parent:"DFID" child:"DFID"\n",
1340                                 PFID(mdt_object_fid(parent)),
1341                                 PFID(mdt_object_fid(child)));
1342                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1343                           LDLM_LOCK_PUT(lock);
1344                           GOTO(relock, 0);
1345                 }
1346                 LDLM_LOCK_PUT(lock);
1347                 rc = 0;
1348         } else {
1349                 bool try_layout = false;
1350
1351 relock:
1352                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1353                 mdt_lock_handle_init(lhc);
1354                 mdt_lock_reg_init(lhc, LCK_PR);
1355
1356                 if (!mdt_object_exists(child)) {
1357                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1358                                         &child->mot_obj,
1359                                         "Object doesn't exist!\n");
1360                         GOTO(out_child, rc = -ENOENT);
1361                 }
1362
1363                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1364                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1365                         struct md_attr *ma = &info->mti_attr;
1366
1367                         ma->ma_valid = 0;
1368                         ma->ma_need = MA_INODE;
1369                         rc = mdt_attr_get_complex(info, child, ma);
1370                         if (unlikely(rc != 0))
1371                                 GOTO(out_child, rc);
1372
1373                         /* If the file has not been changed for some time, we
1374                          * return not only a LOOKUP lock, but also an UPDATE
1375                          * lock and this might save us RPC on later STAT. For
1376                          * directories, it also let negative dentry starts
1377                          * working for this dir. */
1378                         if (ma->ma_valid & MA_INODE &&
1379                             ma->ma_attr.la_valid & LA_CTIME &&
1380                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1381                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1382                                 child_bits |= MDS_INODELOCK_UPDATE;
1383                 }
1384
1385                 /* layout lock must be granted in a best-effort way
1386                  * for IT operations */
1387                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1388                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1389                     exp_connect_layout(info->mti_exp) &&
1390                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1391                     ldlm_rep != NULL) {
1392                         /* try to grant layout lock for regular file. */
1393                         try_layout = true;
1394                 }
1395
1396                 rc = 0;
1397                 if (try_layout) {
1398                         child_bits |= MDS_INODELOCK_LAYOUT;
1399                         /* try layout lock, it may fail to be granted due to
1400                          * contention at LOOKUP or UPDATE */
1401                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1402                                                  MDT_CROSS_LOCK)) {
1403                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1404                                 LASSERT(child_bits != 0);
1405                                 rc = mdt_object_lock(info, child, lhc,
1406                                                 child_bits, MDT_CROSS_LOCK);
1407                         } else {
1408                                 ma_need |= MA_LOV;
1409                         }
1410                 } else {
1411                         rc = mdt_object_lock(info, child, lhc, child_bits,
1412                                                 MDT_CROSS_LOCK);
1413                 }
1414                 if (unlikely(rc != 0))
1415                         GOTO(out_child, rc);
1416         }
1417
1418         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1419         /* Get MA_SOM attributes if update lock is given. */
1420         if (lock &&
1421             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1422             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1423                 ma_need |= MA_SOM;
1424
1425         /* finally, we can get attr for child. */
1426         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1427         rc = mdt_getattr_internal(info, child, ma_need);
1428         if (unlikely(rc != 0)) {
1429                 mdt_object_unlock(info, child, lhc, 1);
1430         } else if (lock) {
1431                 /* Debugging code. */
1432                 res_id = &lock->l_resource->lr_name;
1433                 LDLM_DEBUG(lock, "Returning lock to client");
1434                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1435                                          &lock->l_resource->lr_name),
1436                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1437                          (unsigned long)res_id->name[0],
1438                          (unsigned long)res_id->name[1],
1439                          (unsigned long)res_id->name[2],
1440                          PFID(mdt_object_fid(child)));
1441                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1442                         mdt_pack_size2body(info, child);
1443         }
1444         if (lock)
1445                 LDLM_LOCK_PUT(lock);
1446
1447         EXIT;
1448 out_child:
1449         mdt_object_put(info->mti_env, child);
1450 out_parent:
1451         if (lhp)
1452                 mdt_object_unlock(info, parent, lhp, 1);
1453         return rc;
1454 }
1455
1456 /* normal handler: should release the child lock */
1457 int mdt_getattr_name(struct mdt_thread_info *info)
1458 {
1459         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1460         struct mdt_body        *reqbody;
1461         struct mdt_body        *repbody;
1462         int rc, rc2;
1463         ENTRY;
1464
1465         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1466         LASSERT(reqbody != NULL);
1467         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1468         LASSERT(repbody != NULL);
1469
1470         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1471         repbody->eadatasize = 0;
1472         repbody->aclsize = 0;
1473
1474         rc = mdt_init_ucred(info, reqbody);
1475         if (unlikely(rc))
1476                 GOTO(out_shrink, rc);
1477
1478         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1479         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1480                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1481                 lhc->mlh_reg_lh.cookie = 0;
1482         }
1483         mdt_exit_ucred(info);
1484         EXIT;
1485 out_shrink:
1486         mdt_client_compatibility(info);
1487         rc2 = mdt_fix_reply(info);
1488         if (rc == 0)
1489                 rc = rc2;
1490         return rc;
1491 }
1492
1493 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1494                          void *karg, void *uarg);
1495
1496 int mdt_set_info(struct mdt_thread_info *info)
1497 {
1498         struct ptlrpc_request *req = mdt_info_req(info);
1499         char *key;
1500         void *val;
1501         int keylen, vallen, rc = 0;
1502         ENTRY;
1503
1504         rc = req_capsule_server_pack(info->mti_pill);
1505         if (rc)
1506                 RETURN(rc);
1507
1508         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1509         if (key == NULL) {
1510                 DEBUG_REQ(D_HA, req, "no set_info key");
1511                 RETURN(-EFAULT);
1512         }
1513
1514         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1515                                       RCL_CLIENT);
1516
1517         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1518         if (val == NULL) {
1519                 DEBUG_REQ(D_HA, req, "no set_info val");
1520                 RETURN(-EFAULT);
1521         }
1522
1523         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1524                                       RCL_CLIENT);
1525
1526         /* Swab any part of val you need to here */
1527         if (KEY_IS(KEY_READ_ONLY)) {
1528                 req->rq_status = 0;
1529                 lustre_msg_set_status(req->rq_repmsg, 0);
1530
1531                 spin_lock(&req->rq_export->exp_lock);
1532                 if (*(__u32 *)val)
1533                         *exp_connect_flags_ptr(req->rq_export) |=
1534                                 OBD_CONNECT_RDONLY;
1535                 else
1536                         *exp_connect_flags_ptr(req->rq_export) &=
1537                                 ~OBD_CONNECT_RDONLY;
1538                 spin_unlock(&req->rq_export->exp_lock);
1539
1540         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1541                 struct changelog_setinfo *cs =
1542                         (struct changelog_setinfo *)val;
1543                 if (vallen != sizeof(*cs)) {
1544                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1545                         RETURN(-EINVAL);
1546                 }
1547                 if (ptlrpc_req_need_swab(req)) {
1548                         __swab64s(&cs->cs_recno);
1549                         __swab32s(&cs->cs_id);
1550                 }
1551
1552                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1553                                    vallen, val, NULL);
1554                 lustre_msg_set_status(req->rq_repmsg, rc);
1555
1556         } else {
1557                 RETURN(-EINVAL);
1558         }
1559         RETURN(0);
1560 }
1561
1562 /**
1563  * Top-level handler for MDT connection requests.
1564  */
1565 int mdt_connect(struct mdt_thread_info *info)
1566 {
1567         int rc;
1568         struct obd_connect_data *reply;
1569         struct obd_export *exp;
1570         struct ptlrpc_request *req = mdt_info_req(info);
1571
1572         rc = target_handle_connect(req);
1573         if (rc != 0)
1574                 return err_serious(rc);
1575
1576         LASSERT(req->rq_export != NULL);
1577         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1578         rc = mdt_init_sec_level(info);
1579         if (rc != 0) {
1580                 obd_disconnect(class_export_get(req->rq_export));
1581                 return rc;
1582         }
1583
1584         /* To avoid exposing partially initialized connection flags, changes up
1585          * to this point have been staged in reply->ocd_connect_flags. Now that
1586          * connection handling has completed successfully, atomically update
1587          * the connect flags in the shared export data structure. LU-1623 */
1588         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1589         exp = req->rq_export;
1590         spin_lock(&exp->exp_lock);
1591         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1592         spin_unlock(&exp->exp_lock);
1593
1594         rc = mdt_init_idmap(info);
1595         if (rc != 0)
1596                 obd_disconnect(class_export_get(req->rq_export));
1597
1598         return rc;
1599 }
1600
1601 int mdt_disconnect(struct mdt_thread_info *info)
1602 {
1603         int rc;
1604         ENTRY;
1605
1606         rc = target_handle_disconnect(mdt_info_req(info));
1607         if (rc)
1608                 rc = err_serious(rc);
1609         RETURN(rc);
1610 }
1611
1612 static int mdt_sendpage(struct mdt_thread_info *info,
1613                         struct lu_rdpg *rdpg, int nob)
1614 {
1615         struct ptlrpc_request   *req = mdt_info_req(info);
1616         struct obd_export       *exp = req->rq_export;
1617         struct ptlrpc_bulk_desc *desc;
1618         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1619         int                      tmpcount;
1620         int                      tmpsize;
1621         int                      i;
1622         int                      rc;
1623         ENTRY;
1624
1625         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1626                                     MDS_BULK_PORTAL);
1627         if (desc == NULL)
1628                 RETURN(-ENOMEM);
1629
1630         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1631                 /* old client requires reply size in it's PAGE_SIZE,
1632                  * which is rdpg->rp_count */
1633                 nob = rdpg->rp_count;
1634
1635         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1636              i++, tmpcount -= tmpsize) {
1637                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1638                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1639         }
1640
1641         LASSERT(desc->bd_nob == nob);
1642         rc = target_bulk_io(exp, desc, lwi);
1643         ptlrpc_free_bulk_pin(desc);
1644         RETURN(rc);
1645 }
1646
1647 int mdt_readpage(struct mdt_thread_info *info)
1648 {
1649         struct mdt_object *object = info->mti_object;
1650         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1651         struct mdt_body   *reqbody;
1652         struct mdt_body   *repbody;
1653         int                rc;
1654         int                i;
1655         ENTRY;
1656
1657         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1658                 RETURN(err_serious(-ENOMEM));
1659
1660         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1661         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1662         if (reqbody == NULL || repbody == NULL)
1663                 RETURN(err_serious(-EFAULT));
1664
1665         /*
1666          * prepare @rdpg before calling lower layers and transfer itself. Here
1667          * reqbody->size contains offset of where to start to read and
1668          * reqbody->nlink contains number bytes to read.
1669          */
1670         rdpg->rp_hash = reqbody->size;
1671         if (rdpg->rp_hash != reqbody->size) {
1672                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1673                        rdpg->rp_hash, reqbody->size);
1674                 RETURN(-EFAULT);
1675         }
1676
1677         rdpg->rp_attrs = reqbody->mode;
1678         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1679                 rdpg->rp_attrs |= LUDA_64BITHASH;
1680         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1681                                 exp_max_brw_size(info->mti_exp));
1682         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1683                           PAGE_CACHE_SHIFT;
1684         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1685         if (rdpg->rp_pages == NULL)
1686                 RETURN(-ENOMEM);
1687
1688         for (i = 0; i < rdpg->rp_npages; ++i) {
1689                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1690                 if (rdpg->rp_pages[i] == NULL)
1691                         GOTO(free_rdpg, rc = -ENOMEM);
1692         }
1693
1694         /* call lower layers to fill allocated pages with directory data */
1695         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1696         if (rc < 0)
1697                 GOTO(free_rdpg, rc);
1698
1699         /* send pages to client */
1700         rc = mdt_sendpage(info, rdpg, rc);
1701
1702         EXIT;
1703 free_rdpg:
1704
1705         for (i = 0; i < rdpg->rp_npages; i++)
1706                 if (rdpg->rp_pages[i] != NULL)
1707                         __free_page(rdpg->rp_pages[i]);
1708         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1709
1710         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1711                 RETURN(0);
1712
1713         return rc;
1714 }
1715
1716 static int mdt_reint_internal(struct mdt_thread_info *info,
1717                               struct mdt_lock_handle *lhc,
1718                               __u32 op)
1719 {
1720         struct req_capsule      *pill = info->mti_pill;
1721         struct mdt_body         *repbody;
1722         int                      rc = 0, rc2;
1723         ENTRY;
1724
1725
1726         rc = mdt_reint_unpack(info, op);
1727         if (rc != 0) {
1728                 CERROR("Can't unpack reint, rc %d\n", rc);
1729                 RETURN(err_serious(rc));
1730         }
1731
1732         /* for replay (no_create) lmm is not needed, client has it already */
1733         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1734                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1735                                      info->mti_rr.rr_eadatalen);
1736
1737         /* llog cookies are always 0, the field is kept for compatibility */
1738         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1739                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1740
1741         rc = req_capsule_server_pack(pill);
1742         if (rc != 0) {
1743                 CERROR("Can't pack response, rc %d\n", rc);
1744                 RETURN(err_serious(rc));
1745         }
1746
1747         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1748                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1749                 LASSERT(repbody);
1750                 repbody->eadatasize = 0;
1751                 repbody->aclsize = 0;
1752         }
1753
1754         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1755
1756         /* for replay no cookkie / lmm need, because client have this already */
1757         if (info->mti_spec.no_create)
1758                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1759                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1760
1761         rc = mdt_init_ucred_reint(info);
1762         if (rc)
1763                 GOTO(out_shrink, rc);
1764
1765         rc = mdt_fix_attr_ucred(info, op);
1766         if (rc != 0)
1767                 GOTO(out_ucred, rc = err_serious(rc));
1768
1769         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1770                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1771                 GOTO(out_ucred, rc);
1772         }
1773         rc = mdt_reint_rec(info, lhc);
1774         EXIT;
1775 out_ucred:
1776         mdt_exit_ucred(info);
1777 out_shrink:
1778         mdt_client_compatibility(info);
1779         rc2 = mdt_fix_reply(info);
1780         if (rc == 0)
1781                 rc = rc2;
1782         return rc;
1783 }
1784
1785 static long mdt_reint_opcode(struct mdt_thread_info *info,
1786                              const struct req_format **fmt)
1787 {
1788         struct mdt_rec_reint *rec;
1789         long opc;
1790
1791         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1792         if (rec != NULL) {
1793                 opc = rec->rr_opcode;
1794                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1795                 if (opc < REINT_MAX && fmt[opc] != NULL)
1796                         req_capsule_extend(info->mti_pill, fmt[opc]);
1797                 else {
1798                         CERROR("%s: Unsupported opcode '%ld' from client '%s': "
1799                                "rc = %d\n", mdt_obd_name(info->mti_mdt), opc,
1800                                info->mti_mdt->mdt_ldlm_client->cli_name,
1801                                -EFAULT);
1802                         opc = err_serious(-EFAULT);
1803                 }
1804         } else {
1805                 opc = err_serious(-EFAULT);
1806         }
1807         return opc;
1808 }
1809
1810 int mdt_reint(struct mdt_thread_info *info)
1811 {
1812         long opc;
1813         int  rc;
1814
1815         static const struct req_format *reint_fmts[REINT_MAX] = {
1816                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1817                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1818                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1819                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1820                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1821                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1822                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1823                 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1824         };
1825
1826         ENTRY;
1827
1828         opc = mdt_reint_opcode(info, reint_fmts);
1829         if (opc >= 0) {
1830                 /*
1831                  * No lock possible here from client to pass it to reint code
1832                  * path.
1833                  */
1834                 rc = mdt_reint_internal(info, NULL, opc);
1835         } else {
1836                 rc = opc;
1837         }
1838
1839         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1840         RETURN(rc);
1841 }
1842
1843 /* this should sync the whole device */
1844 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1845 {
1846         struct dt_device *dt = mdt->mdt_bottom;
1847         int rc;
1848         ENTRY;
1849
1850         rc = dt->dd_ops->dt_sync(env, dt);
1851         RETURN(rc);
1852 }
1853
1854 /* this should sync this object */
1855 static int mdt_object_sync(struct mdt_thread_info *info)
1856 {
1857         struct md_object *next;
1858         int rc;
1859         ENTRY;
1860
1861         if (!mdt_object_exists(info->mti_object)) {
1862                 CWARN("Non existing object  "DFID"!\n",
1863                       PFID(mdt_object_fid(info->mti_object)));
1864                 RETURN(-ESTALE);
1865         }
1866         next = mdt_object_child(info->mti_object);
1867         rc = mo_object_sync(info->mti_env, next);
1868
1869         RETURN(rc);
1870 }
1871
1872 int mdt_sync(struct mdt_thread_info *info)
1873 {
1874         struct ptlrpc_request *req = mdt_info_req(info);
1875         struct req_capsule *pill = info->mti_pill;
1876         struct mdt_body *body;
1877         int rc;
1878         ENTRY;
1879
1880         /* The fid may be zero, so we req_capsule_set manually */
1881         req_capsule_set(pill, &RQF_MDS_SYNC);
1882
1883         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1884         if (body == NULL)
1885                 RETURN(err_serious(-EINVAL));
1886
1887         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1888                 RETURN(err_serious(-ENOMEM));
1889
1890         if (fid_seq(&body->fid1) == 0) {
1891                 /* sync the whole device */
1892                 rc = req_capsule_server_pack(pill);
1893                 if (rc == 0)
1894                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1895                 else
1896                         rc = err_serious(rc);
1897         } else {
1898                 /* sync an object */
1899                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1900                 if (rc == 0) {
1901                         rc = mdt_object_sync(info);
1902                         if (rc == 0) {
1903                                 const struct lu_fid *fid;
1904                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1905
1906                                 info->mti_attr.ma_need = MA_INODE;
1907                                 info->mti_attr.ma_valid = 0;
1908                                 rc = mdt_attr_get_complex(info, info->mti_object,
1909                                                           &info->mti_attr);
1910                                 if (rc == 0) {
1911                                         body = req_capsule_server_get(pill,
1912                                                                 &RMF_MDT_BODY);
1913                                         fid = mdt_object_fid(info->mti_object);
1914                                         mdt_pack_attr2body(info, body, la, fid);
1915                                 }
1916                         }
1917                 } else
1918                         rc = err_serious(rc);
1919         }
1920         if (rc == 0)
1921                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1922
1923         RETURN(rc);
1924 }
1925
1926 /*
1927  * Quotacheck handler.
1928  * in-kernel quotacheck isn't supported any more.
1929  */
1930 int mdt_quotacheck(struct mdt_thread_info *info)
1931 {
1932         struct obd_quotactl     *oqctl;
1933         int                      rc;
1934         ENTRY;
1935
1936         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1937         if (oqctl == NULL)
1938                 RETURN(err_serious(-EPROTO));
1939
1940         rc = req_capsule_server_pack(info->mti_pill);
1941         if (rc)
1942                 RETURN(err_serious(rc));
1943
1944         /* deprecated, not used any more */
1945         RETURN(-EOPNOTSUPP);
1946 }
1947
1948 /*
1949  * Handle quota control requests to consult current usage/limit, but also
1950  * to configure quota enforcement
1951  */
1952 int mdt_quotactl(struct mdt_thread_info *info)
1953 {
1954         struct obd_export       *exp  = info->mti_exp;
1955         struct req_capsule      *pill = info->mti_pill;
1956         struct obd_quotactl     *oqctl, *repoqc;
1957         int                      id, rc;
1958         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1959         ENTRY;
1960
1961         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1962         if (oqctl == NULL)
1963                 RETURN(err_serious(-EPROTO));
1964
1965         rc = req_capsule_server_pack(pill);
1966         if (rc)
1967                 RETURN(err_serious(rc));
1968
1969         switch (oqctl->qc_cmd) {
1970         case Q_QUOTACHECK:
1971         case LUSTRE_Q_INVALIDATE:
1972         case LUSTRE_Q_FINVALIDATE:
1973         case Q_QUOTAON:
1974         case Q_QUOTAOFF:
1975         case Q_INITQUOTA:
1976                 /* deprecated, not used any more */
1977                 RETURN(-EOPNOTSUPP);
1978                 /* master quotactl */
1979         case Q_GETINFO:
1980         case Q_SETINFO:
1981         case Q_SETQUOTA:
1982         case Q_GETQUOTA:
1983                 if (qmt == NULL)
1984                         RETURN(-EOPNOTSUPP);
1985                 /* slave quotactl */
1986         case Q_GETOINFO:
1987         case Q_GETOQUOTA:
1988                 break;
1989         default:
1990                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1991                 RETURN(-EFAULT);
1992         }
1993
1994         /* map uid/gid for remote client */
1995         id = oqctl->qc_id;
1996         if (exp_connect_rmtclient(exp)) {
1997                 struct lustre_idmap_table *idmap;
1998
1999                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
2000
2001                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2002                              oqctl->qc_cmd != Q_GETINFO))
2003                         RETURN(-EPERM);
2004
2005                 if (oqctl->qc_type == USRQUOTA)
2006                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2007                                                      oqctl->qc_id);
2008                 else if (oqctl->qc_type == GRPQUOTA)
2009                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2010                                                      oqctl->qc_id);
2011                 else
2012                         RETURN(-EINVAL);
2013
2014                 if (id == CFS_IDMAP_NOTFOUND) {
2015                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2016                         RETURN(-EACCES);
2017                 }
2018         }
2019
2020         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2021         if (repoqc == NULL)
2022                 RETURN(err_serious(-EFAULT));
2023
2024         if (oqctl->qc_id != id)
2025                 swap(oqctl->qc_id, id);
2026
2027         switch (oqctl->qc_cmd) {
2028
2029         case Q_GETINFO:
2030         case Q_SETINFO:
2031         case Q_SETQUOTA:
2032         case Q_GETQUOTA:
2033                 /* forward quotactl request to QMT */
2034                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
2035                 break;
2036
2037         case Q_GETOINFO:
2038         case Q_GETOQUOTA:
2039                 /* slave quotactl */
2040                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
2041                                    oqctl);
2042                 break;
2043
2044         default:
2045                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2046                 RETURN(-EFAULT);
2047         }
2048
2049         if (oqctl->qc_id != id)
2050                 swap(oqctl->qc_id, id);
2051
2052         *repoqc = *oqctl;
2053         RETURN(rc);
2054 }
2055
2056 /*
2057  * OBD PING and other handlers.
2058  */
2059 int mdt_obd_ping(struct mdt_thread_info *info)
2060 {
2061         int rc;
2062         ENTRY;
2063
2064         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
2065
2066         rc = target_handle_ping(mdt_info_req(info));
2067         if (rc < 0)
2068                 rc = err_serious(rc);
2069         RETURN(rc);
2070 }
2071
2072 /*
2073  * OBD_IDX_READ handler
2074  */
2075 int mdt_obd_idx_read(struct mdt_thread_info *info)
2076 {
2077         struct mdt_device       *mdt = info->mti_mdt;
2078         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2079         struct idx_info         *req_ii, *rep_ii;
2080         int                      rc, i;
2081         ENTRY;
2082
2083         memset(rdpg, 0, sizeof(*rdpg));
2084         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
2085
2086         /* extract idx_info buffer from request & reply */
2087         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
2088         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
2089                 RETURN(err_serious(-EPROTO));
2090
2091         rc = req_capsule_server_pack(info->mti_pill);
2092         if (rc)
2093                 RETURN(err_serious(rc));
2094
2095         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
2096         if (rep_ii == NULL)
2097                 RETURN(err_serious(-EFAULT));
2098         rep_ii->ii_magic = IDX_INFO_MAGIC;
2099
2100         /* extract hash to start with */
2101         rdpg->rp_hash = req_ii->ii_hash_start;
2102
2103         /* extract requested attributes */
2104         rdpg->rp_attrs = req_ii->ii_attrs;
2105
2106         /* check that fid packed in request is valid and supported */
2107         if (!fid_is_sane(&req_ii->ii_fid))
2108                 RETURN(-EINVAL);
2109         rep_ii->ii_fid = req_ii->ii_fid;
2110
2111         /* copy flags */
2112         rep_ii->ii_flags = req_ii->ii_flags;
2113
2114         /* compute number of pages to allocate, ii_count is the number of 4KB
2115          * containers */
2116         if (req_ii->ii_count <= 0)
2117                 GOTO(out, rc = -EFAULT);
2118         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2119                                exp_max_brw_size(info->mti_exp));
2120         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
2121                                 PAGE_CACHE_SHIFT;
2122
2123         /* allocate pages to store the containers */
2124         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2125         if (rdpg->rp_pages == NULL)
2126                 GOTO(out, rc = -ENOMEM);
2127         for (i = 0; i < rdpg->rp_npages; i++) {
2128                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
2129                 if (rdpg->rp_pages[i] == NULL)
2130                         GOTO(out, rc = -ENOMEM);
2131         }
2132
2133         /* populate pages with key/record pairs */
2134         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2135         if (rc < 0)
2136                 GOTO(out, rc);
2137
2138         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2139                  "asked %d > %d\n", rc, rdpg->rp_count);
2140
2141         /* send pages to client */
2142         rc = mdt_sendpage(info, rdpg, rc);
2143
2144         GOTO(out, rc);
2145 out:
2146         if (rdpg->rp_pages) {
2147                 for (i = 0; i < rdpg->rp_npages; i++)
2148                         if (rdpg->rp_pages[i])
2149                                 __free_page(rdpg->rp_pages[i]);
2150                 OBD_FREE(rdpg->rp_pages,
2151                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2152         }
2153         return rc;
2154 }
2155
2156 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2157 {
2158         return err_serious(-EOPNOTSUPP);
2159 }
2160
2161 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2162 {
2163         return err_serious(-EOPNOTSUPP);
2164 }
2165
2166 /*
2167  * LLOG handlers.
2168  */
2169
2170 /** clone llog ctxt from child (mdd)
2171  * This allows remote llog (replicator) access.
2172  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2173  * context was originally set up, or we can handle them directly.
2174  * I choose the latter, but that means I need any llog
2175  * contexts set up by child to be accessable by the mdt.  So we clone the
2176  * context into our context list here.
2177  */
2178 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2179                                int idx)
2180 {
2181         struct md_device  *next = mdt->mdt_child;
2182         struct llog_ctxt *ctxt;
2183         int rc;
2184
2185         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2186                 return 0;
2187
2188         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2189         if (rc || ctxt == NULL) {
2190                 return 0;
2191         }
2192
2193         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2194         if (rc)
2195                 CERROR("Can't set mdt ctxt %d\n", rc);
2196
2197         return rc;
2198 }
2199
2200 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2201                                  struct mdt_device *mdt, int idx)
2202 {
2203         struct llog_ctxt *ctxt;
2204
2205         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2206         if (ctxt == NULL)
2207                 return 0;
2208         /* Put once for the get we just did, and once for the clone */
2209         llog_ctxt_put(ctxt);
2210         llog_ctxt_put(ctxt);
2211         return 0;
2212 }
2213
2214 int mdt_llog_create(struct mdt_thread_info *info)
2215 {
2216         int rc;
2217
2218         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2219         rc = llog_origin_handle_open(mdt_info_req(info));
2220         return (rc < 0 ? err_serious(rc) : rc);
2221 }
2222
2223 int mdt_llog_destroy(struct mdt_thread_info *info)
2224 {
2225         int rc;
2226
2227         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2228         rc = llog_origin_handle_destroy(mdt_info_req(info));
2229         return (rc < 0 ? err_serious(rc) : rc);
2230 }
2231
2232 int mdt_llog_read_header(struct mdt_thread_info *info)
2233 {
2234         int rc;
2235
2236         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2237         rc = llog_origin_handle_read_header(mdt_info_req(info));
2238         return (rc < 0 ? err_serious(rc) : rc);
2239 }
2240
2241 int mdt_llog_next_block(struct mdt_thread_info *info)
2242 {
2243         int rc;
2244
2245         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2246         rc = llog_origin_handle_next_block(mdt_info_req(info));
2247         return (rc < 0 ? err_serious(rc) : rc);
2248 }
2249
2250 int mdt_llog_prev_block(struct mdt_thread_info *info)
2251 {
2252         int rc;
2253
2254         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2255         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2256         return (rc < 0 ? err_serious(rc) : rc);
2257 }
2258
2259
2260 /*
2261  * DLM handlers.
2262  */
2263
2264 static struct ldlm_callback_suite cbs = {
2265         .lcs_completion = ldlm_server_completion_ast,
2266         .lcs_blocking   = ldlm_server_blocking_ast,
2267         .lcs_glimpse    = ldlm_server_glimpse_ast
2268 };
2269
2270 int mdt_enqueue(struct mdt_thread_info *info)
2271 {
2272         struct ptlrpc_request *req;
2273         int rc;
2274
2275         /*
2276          * info->mti_dlm_req already contains swapped and (if necessary)
2277          * converted dlm request.
2278          */
2279         LASSERT(info->mti_dlm_req != NULL);
2280
2281         req = mdt_info_req(info);
2282         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2283                                   req, info->mti_dlm_req, &cbs);
2284         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2285         return rc ? err_serious(rc) : req->rq_status;
2286 }
2287
2288 int mdt_convert(struct mdt_thread_info *info)
2289 {
2290         int rc;
2291         struct ptlrpc_request *req;
2292
2293         LASSERT(info->mti_dlm_req);
2294         req = mdt_info_req(info);
2295         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2296         return rc ? err_serious(rc) : req->rq_status;
2297 }
2298
2299 int mdt_bl_callback(struct mdt_thread_info *info)
2300 {
2301         CERROR("bl callbacks should not happen on MDS\n");
2302         LBUG();
2303         return err_serious(-EOPNOTSUPP);
2304 }
2305
2306 int mdt_cp_callback(struct mdt_thread_info *info)
2307 {
2308         CERROR("cp callbacks should not happen on MDS\n");
2309         LBUG();
2310         return err_serious(-EOPNOTSUPP);
2311 }
2312
2313 /*
2314  * sec context handlers
2315  */
2316 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2317 {
2318         int rc;
2319
2320         rc = mdt_handle_idmap(info);
2321
2322         if (unlikely(rc)) {
2323                 struct ptlrpc_request *req = mdt_info_req(info);
2324                 __u32                  opc;
2325
2326                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2327                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2328                         sptlrpc_svc_ctx_invalidate(req);
2329         }
2330
2331         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2332
2333         return rc;
2334 }
2335
2336 /*
2337  * quota request handlers
2338  */
2339 int mdt_quota_dqacq(struct mdt_thread_info *info)
2340 {
2341         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2342         int                      rc;
2343         ENTRY;
2344
2345         if (qmt == NULL)
2346                 RETURN(err_serious(-EOPNOTSUPP));
2347
2348         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2349         RETURN(rc);
2350 }
2351
2352 static struct mdt_object *mdt_obj(struct lu_object *o)
2353 {
2354         LASSERT(lu_device_is_mdt(o->lo_dev));
2355         return container_of0(o, struct mdt_object, mot_obj);
2356 }
2357
2358 struct mdt_object *mdt_object_new(const struct lu_env *env,
2359                                   struct mdt_device *d,
2360                                   const struct lu_fid *f)
2361 {
2362         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2363         struct lu_object *o;
2364         struct mdt_object *m;
2365         ENTRY;
2366
2367         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2368         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2369         if (unlikely(IS_ERR(o)))
2370                 m = (struct mdt_object *)o;
2371         else
2372                 m = mdt_obj(o);
2373         RETURN(m);
2374 }
2375
2376 struct mdt_object *mdt_object_find(const struct lu_env *env,
2377                                    struct mdt_device *d,
2378                                    const struct lu_fid *f)
2379 {
2380         struct lu_object *o;
2381         struct mdt_object *m;
2382         ENTRY;
2383
2384         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2385         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2386         if (unlikely(IS_ERR(o)))
2387                 m = (struct mdt_object *)o;
2388         else
2389                 m = mdt_obj(o);
2390
2391         RETURN(m);
2392 }
2393
2394 /**
2395  * Asyncronous commit for mdt device.
2396  *
2397  * Pass asynchonous commit call down the MDS stack.
2398  *
2399  * \param env environment
2400  * \param mdt the mdt device
2401  */
2402 static void mdt_device_commit_async(const struct lu_env *env,
2403                                     struct mdt_device *mdt)
2404 {
2405         struct dt_device *dt = mdt->mdt_bottom;
2406         int rc;
2407
2408         rc = dt->dd_ops->dt_commit_async(env, dt);
2409         if (unlikely(rc != 0))
2410                 CWARN("async commit start failed with rc = %d", rc);
2411 }
2412
2413 /**
2414  * Mark the lock as "synchonous".
2415  *
2416  * Mark the lock to deffer transaction commit to the unlock time.
2417  *
2418  * \param lock the lock to mark as "synchonous"
2419  *
2420  * \see mdt_is_lock_sync
2421  * \see mdt_save_lock
2422  */
2423 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2424 {
2425         lock->l_ast_data = (void*)1;
2426 }
2427
2428 /**
2429  * Check whehter the lock "synchonous" or not.
2430  *
2431  * \param lock the lock to check
2432  * \retval 1 the lock is "synchonous"
2433  * \retval 0 the lock isn't "synchronous"
2434  *
2435  * \see mdt_set_lock_sync
2436  * \see mdt_save_lock
2437  */
2438 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2439 {
2440         return lock->l_ast_data != NULL;
2441 }
2442
2443 /**
2444  * Blocking AST for mdt locks.
2445  *
2446  * Starts transaction commit if in case of COS lock conflict or
2447  * deffers such a commit to the mdt_save_lock.
2448  *
2449  * \param lock the lock which blocks a request or cancelling lock
2450  * \param desc unused
2451  * \param data unused
2452  * \param flag indicates whether this cancelling or blocking callback
2453  * \retval 0
2454  * \see ldlm_blocking_ast_nocheck
2455  */
2456 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2457                      void *data, int flag)
2458 {
2459         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2460         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2461         int rc;
2462         ENTRY;
2463
2464         if (flag == LDLM_CB_CANCELING)
2465                 RETURN(0);
2466         lock_res_and_lock(lock);
2467         if (lock->l_blocking_ast != mdt_blocking_ast) {
2468                 unlock_res_and_lock(lock);
2469                 RETURN(0);
2470         }
2471         if (mdt_cos_is_enabled(mdt) &&
2472             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2473             lock->l_blocking_lock != NULL &&
2474             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2475                 mdt_set_lock_sync(lock);
2476         }
2477         rc = ldlm_blocking_ast_nocheck(lock);
2478
2479         /* There is no lock conflict if l_blocking_lock == NULL,
2480          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2481          * when the last reference to a local lock was released */
2482         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2483                 struct lu_env env;
2484
2485                 rc = lu_env_init(&env, LCT_LOCAL);
2486                 if (unlikely(rc != 0))
2487                         CWARN("lu_env initialization failed with rc = %d,"
2488                               "cannot start asynchronous commit\n", rc);
2489                 else
2490                         mdt_device_commit_async(&env, mdt);
2491                 lu_env_fini(&env);
2492         }
2493         RETURN(rc);
2494 }
2495
2496 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2497                         void *data, int flag)
2498 {
2499         struct lustre_handle lockh;
2500         int               rc;
2501
2502         switch (flag) {
2503         case LDLM_CB_BLOCKING:
2504                 ldlm_lock2handle(lock, &lockh);
2505                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2506                 if (rc < 0) {
2507                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2508                         RETURN(rc);
2509                 }
2510                 break;
2511         case LDLM_CB_CANCELING:
2512                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2513                 break;
2514         default:
2515                 LBUG();
2516         }
2517         RETURN(0);
2518 }
2519
2520 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2521                            struct mdt_object *o, struct lustre_handle *lh,
2522                            ldlm_mode_t mode, __u64 ibits)
2523 {
2524         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2525         ldlm_policy_data_t *policy = &mti->mti_policy;
2526         int rc = 0;
2527         ENTRY;
2528
2529         LASSERT(mdt_object_remote(o));
2530
2531         LASSERT((ibits & MDS_INODELOCK_UPDATE));
2532
2533         memset(einfo, 0, sizeof(*einfo));
2534         einfo->ei_type = LDLM_IBITS;
2535         einfo->ei_mode = mode;
2536         einfo->ei_cb_bl = mdt_md_blocking_ast;
2537         einfo->ei_cb_cp = ldlm_completion_ast;
2538
2539         memset(policy, 0, sizeof(*policy));
2540         policy->l_inodebits.bits = ibits;
2541
2542         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2543                             policy);
2544         RETURN(rc);
2545 }
2546
2547 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2548                             struct mdt_lock_handle *lh, __u64 ibits,
2549                             bool nonblock, int locality)
2550 {
2551         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2552         ldlm_policy_data_t *policy = &info->mti_policy;
2553         struct ldlm_res_id *res_id = &info->mti_res_id;
2554         __u64 dlmflags;
2555         int rc;
2556         ENTRY;
2557
2558         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2559         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2560         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2561         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2562
2563         if (mdt_object_remote(o)) {
2564                 if (locality == MDT_CROSS_LOCK) {
2565                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM);
2566                         ibits |= MDS_INODELOCK_LOOKUP;
2567                 } else {
2568                         LASSERTF(!(ibits &
2569                                   (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)),
2570                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2571                                 mdt_obd_name(info->mti_mdt), ibits,
2572                                 PFID(mdt_object_fid(o)));
2573                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2574                 }
2575                 /* No PDO lock on remote object */
2576                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2577         }
2578
2579         if (lh->mlh_type == MDT_PDO_LOCK) {
2580                 /* check for exists after object is locked */
2581                 if (mdt_object_exists(o) == 0) {
2582                         /* Non-existent object shouldn't have PDO lock */
2583                         RETURN(-ESTALE);
2584                 } else {
2585                         /* Non-dir object shouldn't have PDO lock */
2586                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2587                                 RETURN(-ENOTDIR);
2588                 }
2589         }
2590
2591         memset(policy, 0, sizeof(*policy));
2592         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2593
2594         dlmflags = LDLM_FL_ATOMIC_CB;
2595         if (nonblock)
2596                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2597
2598         /*
2599          * Take PDO lock on whole directory and build correct @res_id for lock
2600          * on part of directory.
2601          */
2602         if (lh->mlh_pdo_hash != 0) {
2603                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2604                 mdt_lock_pdo_mode(info, o, lh);
2605                 if (lh->mlh_pdo_mode != LCK_NL) {
2606                         /*
2607                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2608                          * is never going to be sent to client and we do not
2609                          * want it slowed down due to possible cancels.
2610                          */
2611                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2612                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2613                                           policy, res_id, dlmflags,
2614                                           &info->mti_exp->exp_handle.h_cookie);
2615                         if (unlikely(rc))
2616                                 RETURN(rc);
2617                 }
2618
2619                 /*
2620                  * Finish res_id initializing by name hash marking part of
2621                  * directory which is taking modification.
2622                  */
2623                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2624         }
2625
2626         policy->l_inodebits.bits = ibits;
2627
2628         /*
2629          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2630          * going to be sent to client. If it is - mdt_intent_policy() path will
2631          * fix it up and turn FL_LOCAL flag off.
2632          */
2633         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2634                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2635                           &info->mti_exp->exp_handle.h_cookie);
2636         if (rc)
2637                 mdt_object_unlock(info, o, lh, 1);
2638         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2639                  lh->mlh_pdo_hash != 0 &&
2640                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2641                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2642         }
2643
2644         RETURN(rc);
2645 }
2646
2647 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2648                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2649 {
2650         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2651 }
2652
2653 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2654                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2655 {
2656         struct mdt_lock_handle tmp = *lh;
2657         int rc;
2658
2659         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2660         if (rc == 0)
2661                 *lh = tmp;
2662
2663         return rc == 0;
2664 }
2665
2666 /**
2667  * Save a lock within request object.
2668  *
2669  * Keep the lock referenced until whether client ACK or transaction
2670  * commit happens or release the lock immediately depending on input
2671  * parameters. If COS is ON, a write lock is converted to COS lock
2672  * before saving.
2673  *
2674  * \param info thead info object
2675  * \param h lock handle
2676  * \param mode lock mode
2677  * \param decref force immediate lock releasing
2678  */
2679 static
2680 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2681                    ldlm_mode_t mode, int decref)
2682 {
2683         ENTRY;
2684
2685         if (lustre_handle_is_used(h)) {
2686                 if (decref || !info->mti_has_trans ||
2687                     !(mode & (LCK_PW | LCK_EX))){
2688                         mdt_fid_unlock(h, mode);
2689                 } else {
2690                         struct mdt_device *mdt = info->mti_mdt;
2691                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2692                         struct ptlrpc_request *req = mdt_info_req(info);
2693                         int no_ack = 0;
2694
2695                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2696                                  h->cookie);
2697                         CDEBUG(D_HA, "request = %p reply state = %p"
2698                                " transno = "LPD64"\n",
2699                                req, req->rq_reply_state, req->rq_transno);
2700                         if (mdt_cos_is_enabled(mdt)) {
2701                                 no_ack = 1;
2702                                 ldlm_lock_downgrade(lock, LCK_COS);
2703                                 mode = LCK_COS;
2704                         }
2705                         ptlrpc_save_lock(req, h, mode, no_ack);
2706                         if (mdt_is_lock_sync(lock)) {
2707                                 CDEBUG(D_HA, "found sync-lock,"
2708                                        " async commit started\n");
2709                                 mdt_device_commit_async(info->mti_env,
2710                                                         mdt);
2711                         }
2712                         LDLM_LOCK_PUT(lock);
2713                 }
2714                 h->cookie = 0ull;
2715         }
2716
2717         EXIT;
2718 }
2719
2720 /**
2721  * Unlock mdt object.
2722  *
2723  * Immeditely release the regular lock and the PDO lock or save the
2724  * lock in reqeuest and keep them referenced until client ACK or
2725  * transaction commit.
2726  *
2727  * \param info thread info object
2728  * \param o mdt object
2729  * \param lh mdt lock handle referencing regular and PDO locks
2730  * \param decref force immediate lock releasing
2731  */
2732 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2733                        struct mdt_lock_handle *lh, int decref)
2734 {
2735         ENTRY;
2736
2737         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2738         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2739
2740         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2741                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2742
2743         EXIT;
2744 }
2745
2746 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2747                                         const struct lu_fid *f,
2748                                         struct mdt_lock_handle *lh,
2749                                         __u64 ibits)
2750 {
2751         struct mdt_object *o;
2752
2753         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2754         if (!IS_ERR(o)) {
2755                 int rc;
2756
2757                 rc = mdt_object_lock(info, o, lh, ibits,
2758                                      MDT_LOCAL_LOCK);
2759                 if (rc != 0) {
2760                         mdt_object_put(info->mti_env, o);
2761                         o = ERR_PTR(rc);
2762                 }
2763         }
2764         return o;
2765 }
2766
2767 void mdt_object_unlock_put(struct mdt_thread_info * info,
2768                            struct mdt_object * o,
2769                            struct mdt_lock_handle *lh,
2770                            int decref)
2771 {
2772         mdt_object_unlock(info, o, lh, decref);
2773         mdt_object_put(info->mti_env, o);
2774 }
2775
2776 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2777 {
2778         struct mdt_opc_slice *s;
2779         struct mdt_handler   *h;
2780
2781         h = NULL;
2782         for (s = supported; s->mos_hs != NULL; s++) {
2783                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2784                         h = s->mos_hs + (opc - s->mos_opc_start);
2785                         if (likely(h->mh_opc != 0))
2786                                 LASSERTF(h->mh_opc == opc,
2787                                          "opcode mismatch %d != %d\n",
2788                                          h->mh_opc, opc);
2789                         else
2790                                 h = NULL; /* unsupported opc */
2791                         break;
2792                 }
2793         }
2794         return h;
2795 }
2796
2797 static int mdt_lock_resname_compat(struct mdt_device *m,
2798                                    struct ldlm_request *req)
2799 {
2800         /* XXX something... later. */
2801         return 0;
2802 }
2803
2804 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2805 {
2806         /* XXX something... later. */
2807         return 0;
2808 }
2809
2810 /*
2811  * Generic code handling requests that have struct mdt_body passed in:
2812  *
2813  *  - extract mdt_body from request and save it in @info, if present;
2814  *
2815  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2816  *  @info;
2817  *
2818  *  - if HABEO_CORPUS flag is set for this request type check whether object
2819  *  actually exists on storage (lu_object_exists()).
2820  *
2821  */
2822 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2823 {
2824         const struct mdt_body    *body;
2825         struct mdt_object        *obj;
2826         const struct lu_env      *env;
2827         struct req_capsule       *pill;
2828         int                       rc;
2829         ENTRY;
2830
2831         env = info->mti_env;
2832         pill = info->mti_pill;
2833
2834         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2835         if (body == NULL)
2836                 RETURN(-EFAULT);
2837
2838         if (!(body->valid & OBD_MD_FLID))
2839                 RETURN(0);
2840
2841         if (!fid_is_sane(&body->fid1)) {
2842                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2843                 RETURN(-EINVAL);
2844         }
2845
2846         /*
2847          * Do not get size or any capa fields before we check that request
2848          * contains capa actually. There are some requests which do not, for
2849          * instance MDS_IS_SUBDIR.
2850          */
2851         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2852             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2853                 mdt_set_capainfo(info, 0, &body->fid1,
2854                                  req_capsule_client_get(pill, &RMF_CAPA1));
2855
2856         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2857         if (!IS_ERR(obj)) {
2858                 if ((flags & HABEO_CORPUS) &&
2859                     !mdt_object_exists(obj)) {
2860                         mdt_object_put(env, obj);
2861                         /* for capability renew ENOENT will be handled in
2862                          * mdt_renew_capa */
2863                         if (body->valid & OBD_MD_FLOSSCAPA)
2864                                 rc = 0;
2865                         else
2866                                 rc = -ENOENT;
2867                 } else {
2868                         info->mti_object = obj;
2869                         rc = 0;
2870                 }
2871         } else
2872                 rc = PTR_ERR(obj);
2873
2874         RETURN(rc);
2875 }
2876
2877 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2878 {
2879         struct req_capsule *pill = info->mti_pill;
2880         int rc;
2881         ENTRY;
2882
2883         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2884                 rc = mdt_body_unpack(info, flags);
2885         else
2886                 rc = 0;
2887
2888         if (rc == 0 && (flags & HABEO_REFERO)) {
2889                 /* Pack reply. */
2890                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2891                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2892                                              info->mti_body->eadatasize);
2893                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2894                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2895                                              RCL_SERVER, 0);
2896
2897                 rc = req_capsule_server_pack(pill);
2898         }
2899         RETURN(rc);
2900 }
2901
2902 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2903 {
2904         struct md_device *next = m->mdt_child;
2905
2906         return next->md_ops->mdo_init_capa_ctxt(env, next,
2907                                                 m->mdt_opts.mo_mds_capa,
2908                                                 m->mdt_capa_timeout,
2909                                                 m->mdt_capa_alg,
2910                                                 m->mdt_capa_keys);
2911 }
2912
2913 /*
2914  * Invoke handler for this request opc. Also do necessary preprocessing
2915  * (according to handler ->mh_flags), and post-processing (setting of
2916  * ->last_{xid,committed}).
2917  */
2918 static int mdt_req_handle(struct mdt_thread_info *info,
2919                           struct mdt_handler *h, struct ptlrpc_request *req)
2920 {
2921         int   rc, serious = 0;
2922         __u32 flags;
2923
2924         ENTRY;
2925
2926         LASSERT(h->mh_act != NULL);
2927         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2928         LASSERT(current->journal_info == NULL);
2929
2930         /*
2931          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2932          * to put same checks into handlers like mdt_close(), mdt_reint(),
2933          * etc., without talking to mdt authors first. Checking same thing
2934          * there again is useless and returning 0 error without packing reply
2935          * is buggy! Handlers either pack reply or return error.
2936          *
2937          * We return 0 here and do not send any reply in order to emulate
2938          * network failure. Do not send any reply in case any of NET related
2939          * fail_id has occured.
2940          */
2941         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2942                 RETURN(0);
2943
2944         rc = 0;
2945         flags = h->mh_flags;
2946         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2947
2948         if (h->mh_fmt != NULL) {
2949                 req_capsule_set(info->mti_pill, h->mh_fmt);
2950                 rc = mdt_unpack_req_pack_rep(info, flags);
2951         }
2952
2953         if (rc == 0 && flags & MUTABOR &&
2954             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2955                 /* should it be rq_status? */
2956                 rc = -EROFS;
2957
2958         if (rc == 0 && flags & HABEO_CLAVIS) {
2959                 struct ldlm_request *dlm_req;
2960
2961                 LASSERT(h->mh_fmt != NULL);
2962
2963                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2964                 if (dlm_req != NULL) {
2965                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2966                                         LDLM_IBITS &&
2967                                      dlm_req->lock_desc.l_policy_data.\
2968                                         l_inodebits.bits == 0)) {
2969                                 /*
2970                                  * Lock without inodebits makes no sense and
2971                                  * will oops later in ldlm. If client miss to
2972                                  * set such bits, do not trigger ASSERTION.
2973                                  *
2974                                  * For liblustre flock case, it maybe zero.
2975                                  */
2976                                 rc = -EPROTO;
2977                         } else {
2978                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2979                                         rc = mdt_lock_resname_compat(
2980                                                                 info->mti_mdt,
2981                                                                 dlm_req);
2982                                 info->mti_dlm_req = dlm_req;
2983                         }
2984                 } else {
2985                         rc = -EFAULT;
2986                 }
2987         }
2988
2989         /* capability setting changed via /proc, needs reinitialize ctxt */
2990         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2991                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2992                 info->mti_mdt->mdt_capa_conf = 0;
2993         }
2994
2995         if (likely(rc == 0)) {
2996                 /*
2997                  * Process request, there can be two types of rc:
2998                  * 1) errors with msg unpack/pack, other failures outside the
2999                  * operation itself. This is counted as serious errors;
3000                  * 2) errors during fs operation, should be placed in rq_status
3001                  * only
3002                  */
3003                 rc = h->mh_act(info);
3004                 if (rc == 0 &&
3005                     !req->rq_no_reply && req->rq_reply_state == NULL) {
3006                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
3007                                   "pack reply and returned 0 error\n",
3008                                   h->mh_name);
3009                         LBUG();
3010                 }
3011                 serious = is_serious(rc);
3012                 rc = clear_serious(rc);
3013         } else
3014                 serious = 1;
3015
3016         req->rq_status = rc;
3017
3018         /*
3019          * ELDLM_* codes which > 0 should be in rq_status only as well as
3020          * all non-serious errors.
3021          */
3022         if (rc > 0 || !serious)
3023                 rc = 0;
3024
3025         LASSERT(current->journal_info == NULL);
3026
3027         if (rc == 0 && (flags & HABEO_CLAVIS) &&
3028             info->mti_mdt->mdt_opts.mo_compat_resname) {
3029                 struct ldlm_reply *dlmrep;
3030
3031                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3032                 if (dlmrep != NULL)
3033                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
3034         }
3035
3036         /* If we're DISCONNECTing, the mdt_export_data is already freed */
3037         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
3038                 target_committed_to_req(req);
3039
3040         if (unlikely(req_is_replay(req) &&
3041                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
3042                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
3043                 LBUG();
3044         }
3045
3046         target_send_reply(req, rc, info->mti_fail_id);
3047         RETURN(0);
3048 }
3049
3050 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3051 {
3052         lh->mlh_type = MDT_NUL_LOCK;
3053         lh->mlh_reg_lh.cookie = 0ull;
3054         lh->mlh_reg_mode = LCK_MINMODE;
3055         lh->mlh_pdo_lh.cookie = 0ull;
3056         lh->mlh_pdo_mode = LCK_MINMODE;
3057         lh->mlh_rreg_lh.cookie = 0ull;
3058         lh->mlh_rreg_mode = LCK_MINMODE;
3059 }
3060
3061 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3062 {
3063         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3064         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3065 }
3066
3067 /*
3068  * Initialize fields of struct mdt_thread_info. Other fields are left in
3069  * uninitialized state, because it's too expensive to zero out whole
3070  * mdt_thread_info (> 1K) on each request arrival.
3071  */
3072 static void mdt_thread_info_init(struct ptlrpc_request *req,
3073                                  struct mdt_thread_info *info)
3074 {
3075         int i;
3076
3077         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
3078         info->mti_pill = &req->rq_pill;
3079
3080         /* lock handle */
3081         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3082                 mdt_lock_handle_init(&info->mti_lh[i]);
3083
3084         /* mdt device: it can be NULL while CONNECT */
3085         if (req->rq_export) {
3086                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3087                 info->mti_exp = req->rq_export;
3088         } else
3089                 info->mti_mdt = NULL;
3090         info->mti_env = req->rq_svc_thread->t_env;
3091         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
3092         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3093         info->mti_mos = NULL;
3094
3095         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3096         info->mti_big_buf = LU_BUF_NULL;
3097         info->mti_body = NULL;
3098         info->mti_object = NULL;
3099         info->mti_dlm_req = NULL;
3100         info->mti_has_trans = 0;
3101         info->mti_cross_ref = 0;
3102         info->mti_opdata = 0;
3103         info->mti_big_lmm_used = 0;
3104
3105         /* To not check for split by default. */
3106         info->mti_spec.no_create = 0;
3107         info->mti_spec.sp_rm_entry = 0;
3108 }
3109
3110 static void mdt_thread_info_fini(struct mdt_thread_info *info)
3111 {
3112         int i;
3113
3114         req_capsule_fini(info->mti_pill);
3115         if (info->mti_object != NULL) {
3116                 mdt_object_put(info->mti_env, info->mti_object);
3117                 info->mti_object = NULL;
3118         }
3119
3120         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3121                 mdt_lock_handle_fini(&info->mti_lh[i]);
3122         info->mti_env = NULL;
3123
3124         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3125                 lu_buf_free(&info->mti_big_buf);
3126 }
3127
3128 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3129                                        struct obd_device *obd, int *process)
3130 {
3131         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3132         case MDS_CONNECT: /* This will never get here, but for completeness. */
3133         case OST_CONNECT: /* This will never get here, but for completeness. */
3134         case MDS_DISCONNECT:
3135         case OST_DISCONNECT:
3136         case OBD_IDX_READ:
3137                *process = 1;
3138                RETURN(0);
3139
3140         case MDS_CLOSE:
3141         case MDS_DONE_WRITING:
3142         case MDS_SYNC: /* used in unmounting */
3143         case OBD_PING:
3144         case MDS_REINT:
3145         case UPDATE_OBJ:
3146         case SEQ_QUERY:
3147         case FLD_QUERY:
3148         case LDLM_ENQUEUE:
3149                 *process = target_queue_recovery_request(req, obd);
3150                 RETURN(0);
3151
3152         default:
3153                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3154                 *process = -EAGAIN;
3155                 RETURN(0);
3156         }
3157 }
3158
3159 /*
3160  * Handle recovery. Return:
3161  *        +1: continue request processing;
3162  *       -ve: abort immediately with the given error code;
3163  *         0: send reply with error code in req->rq_status;
3164  */
3165 static int mdt_recovery(struct mdt_thread_info *info)
3166 {
3167         struct ptlrpc_request *req = mdt_info_req(info);
3168         struct obd_device *obd;
3169
3170         ENTRY;
3171
3172         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3173         case MDS_CONNECT:
3174         case SEC_CTX_INIT:
3175         case SEC_CTX_INIT_CONT:
3176         case SEC_CTX_FINI:
3177                 {
3178 #if 0
3179                         int rc;
3180
3181                         rc = mdt_handle_idmap(info);
3182                         if (rc)
3183                                 RETURN(rc);
3184                         else
3185 #endif
3186                                 RETURN(+1);
3187                 }
3188         }
3189
3190         if (unlikely(!class_connected_export(req->rq_export))) {
3191                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3192                        lustre_msg_get_opc(req->rq_reqmsg),
3193                        libcfs_id2str(req->rq_peer));
3194                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3195                  * mds_A will get -ENOTCONN(especially for ping req),
3196                  * which will cause that mds_A deactive timeout, then when
3197                  * mds_A cleanup, the cleanup process will be suspended since
3198                  * deactive timeout is not zero.
3199                  */
3200                 req->rq_status = -ENOTCONN;
3201                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3202                 RETURN(0);
3203         }
3204
3205         /* sanity check: if the xid matches, the request must be marked as a
3206          * resent or replayed */
3207         if (req_xid_is_last(req)) {
3208                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3209                       (MSG_RESENT | MSG_REPLAY))) {
3210                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3211                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3212                                   lustre_msg_get_flags(req->rq_reqmsg));
3213                         LBUG();
3214                         req->rq_status = -ENOTCONN;
3215                         RETURN(-ENOTCONN);
3216                 }
3217         }
3218
3219         /* else: note the opposite is not always true; a RESENT req after a
3220          * failover will usually not match the last_xid, since it was likely
3221          * never committed. A REPLAYed request will almost never match the
3222          * last xid, however it could for a committed, but still retained,
3223          * open. */
3224
3225         obd = req->rq_export->exp_obd;
3226
3227         /* Check for aborted recovery... */
3228         if (unlikely(obd->obd_recovering)) {
3229                 int rc;
3230                 int should_process;
3231                 DEBUG_REQ(D_INFO, req, "Got new replay");
3232                 rc = mdt_filter_recovery_request(req, obd, &should_process);
3233                 if (rc != 0 || !should_process)
3234                         RETURN(rc);
3235                 else if (should_process < 0) {
3236                         req->rq_status = should_process;
3237                         rc = ptlrpc_error(req);
3238                         RETURN(rc);
3239                 }
3240         }
3241         RETURN(+1);
3242 }
3243
3244 static int mdt_msg_check_version(struct lustre_msg *msg)
3245 {
3246         int rc;
3247
3248         switch (lustre_msg_get_opc(msg)) {
3249         case MDS_CONNECT:
3250         case MDS_DISCONNECT:
3251         case OBD_PING:
3252         case SEC_CTX_INIT:
3253         case SEC_CTX_INIT_CONT:
3254         case SEC_CTX_FINI:
3255         case OBD_IDX_READ:
3256                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3257                 if (rc)
3258                         CERROR("bad opc %u version %08x, expecting %08x\n",
3259                                lustre_msg_get_opc(msg),
3260                                lustre_msg_get_version(msg),
3261                                LUSTRE_OBD_VERSION);
3262                 break;
3263         case MDS_GETSTATUS:
3264         case MDS_GETATTR:
3265         case MDS_GETATTR_NAME:
3266         case MDS_STATFS:
3267         case MDS_READPAGE:
3268         case MDS_WRITEPAGE:
3269         case MDS_IS_SUBDIR:
3270         case MDS_REINT:
3271         case MDS_CLOSE:
3272         case MDS_DONE_WRITING:
3273         case MDS_PIN:
3274         case MDS_SYNC:
3275         case MDS_GETXATTR:
3276         case MDS_SETXATTR:
3277         case MDS_SET_INFO:
3278         case MDS_GET_INFO:
3279         case MDS_HSM_PROGRESS:
3280         case MDS_HSM_REQUEST:
3281         case MDS_HSM_CT_REGISTER:
3282         case MDS_HSM_CT_UNREGISTER:
3283         case MDS_HSM_STATE_GET:
3284         case MDS_HSM_STATE_SET:
3285         case MDS_HSM_ACTION:
3286         case MDS_QUOTACHECK:
3287         case MDS_QUOTACTL:
3288         case UPDATE_OBJ:
3289         case MDS_SWAP_LAYOUTS:
3290         case QUOTA_DQACQ:
3291         case QUOTA_DQREL:
3292         case SEQ_QUERY:
3293         case FLD_QUERY:
3294                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3295                 if (rc)
3296                         CERROR("bad opc %u version %08x, expecting %08x\n",
3297                                lustre_msg_get_opc(msg),
3298                                lustre_msg_get_version(msg),
3299                                LUSTRE_MDS_VERSION);
3300                 break;
3301         case LDLM_ENQUEUE:
3302         case LDLM_CONVERT:
3303         case LDLM_BL_CALLBACK:
3304         case LDLM_CP_CALLBACK:
3305                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3306                 if (rc)
3307                         CERROR("bad opc %u version %08x, expecting %08x\n",
3308                                lustre_msg_get_opc(msg),
3309                                lustre_msg_get_version(msg),
3310                                LUSTRE_DLM_VERSION);
3311                 break;
3312         case OBD_LOG_CANCEL:
3313         case LLOG_ORIGIN_HANDLE_CREATE:
3314         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3315         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3316         case LLOG_ORIGIN_HANDLE_CLOSE:
3317         case LLOG_ORIGIN_HANDLE_DESTROY:
3318         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3319         case LLOG_CATINFO:
3320                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3321                 if (rc)
3322                         CERROR("bad opc %u version %08x, expecting %08x\n",
3323                                lustre_msg_get_opc(msg),
3324                                lustre_msg_get_version(msg),
3325                                LUSTRE_LOG_VERSION);
3326                 break;
3327         default:
3328                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3329                 rc = -ENOTSUPP;
3330         }
3331         return rc;
3332 }
3333
3334 static int mdt_handle0(struct ptlrpc_request *req,
3335                        struct mdt_thread_info *info,
3336                        struct mdt_opc_slice *supported)
3337 {
3338         struct mdt_handler *h;
3339         struct lustre_msg  *msg;
3340         int                 rc;
3341
3342         ENTRY;
3343
3344         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3345                 RETURN(0);
3346
3347         LASSERT(current->journal_info == NULL);
3348
3349         msg = req->rq_reqmsg;
3350         rc = mdt_msg_check_version(msg);
3351         if (likely(rc == 0)) {
3352                 rc = mdt_recovery(info);
3353                 if (likely(rc == +1)) {
3354                         h = mdt_handler_find(lustre_msg_get_opc(msg),
3355                                              supported);
3356                         if (likely(h != NULL)) {
3357                                 rc = mdt_req_handle(info, h, req);
3358                         } else {
3359                                 CERROR("%s: opc unsupported: 0x%x\n",
3360                                         mdt_obd_name(info->mti_mdt),
3361                                         lustre_msg_get_opc(msg));
3362                                 req->rq_status = -ENOTSUPP;
3363                                 rc = ptlrpc_error(req);
3364                                 RETURN(rc);
3365                         }
3366                 }
3367         } else {
3368                 CDEBUG(D_INFO, "%s: drops mal-formed request: rc = %d\n",
3369                         mdt_obd_name(info->mti_mdt), rc);
3370                 req->rq_status = rc;
3371                 rc = ptlrpc_error(req);
3372         }
3373         RETURN(rc);
3374 }
3375
3376 /*
3377  * MDT handler function called by ptlrpc service thread when request comes.
3378  *
3379  * XXX common "target" functionality should be factored into separate module
3380  * shared by mdt, ost and stand-alone services like fld.
3381  */
3382 int mdt_handle_common(struct ptlrpc_request *req,
3383                       struct mdt_opc_slice *supported)
3384 {
3385         struct lu_env          *env;
3386         struct mdt_thread_info *info;
3387         int                     rc;
3388         ENTRY;
3389
3390         env = req->rq_svc_thread->t_env;
3391         /* Refill(initilize) the context(mdt_thread_info), in case it is
3392          * not initialized yet. Usually it happens during start up, after
3393          * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3394          * before MDT_thread_info is initialized */
3395         lu_env_refill(env);
3396         LASSERT(env != NULL);
3397         LASSERT(env->le_ses != NULL);
3398         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3399         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3400         LASSERT(info != NULL);
3401
3402         mdt_thread_info_init(req, info);
3403
3404         rc = mdt_handle0(req, info, supported);
3405
3406         mdt_thread_info_fini(info);
3407         RETURN(rc);
3408 }
3409
3410 /*
3411  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3412  * as well.
3413  */
3414 int mdt_recovery_handle(struct ptlrpc_request *req)
3415 {
3416         int rc;
3417         ENTRY;
3418
3419         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3420         case FLD_QUERY:
3421                 rc = mdt_handle_common(req, mdt_fld_handlers);
3422                 break;
3423         case SEQ_QUERY:
3424                 rc = mdt_handle_common(req, mdt_seq_handlers);
3425                 break;
3426         default:
3427                 rc = mdt_handle_common(req, mdt_regular_handlers);
3428                 break;
3429         }
3430
3431         RETURN(rc);
3432 }
3433
3434 enum mdt_it_code {
3435         MDT_IT_OPEN,
3436         MDT_IT_OCREAT,
3437         MDT_IT_CREATE,
3438         MDT_IT_GETATTR,
3439         MDT_IT_READDIR,
3440         MDT_IT_LOOKUP,
3441         MDT_IT_UNLINK,
3442         MDT_IT_TRUNC,
3443         MDT_IT_GETXATTR,
3444         MDT_IT_LAYOUT,
3445         MDT_IT_QUOTA,
3446         MDT_IT_NR
3447 };
3448
3449 static int mdt_intent_getattr(enum mdt_it_code opcode,
3450                               struct mdt_thread_info *info,
3451                               struct ldlm_lock **,
3452                               __u64);
3453 static int mdt_intent_layout(enum mdt_it_code opcode,
3454                              struct mdt_thread_info *info,
3455                              struct ldlm_lock **,
3456                              __u64);
3457 static int mdt_intent_reint(enum mdt_it_code opcode,
3458                             struct mdt_thread_info *info,
3459                             struct ldlm_lock **,
3460                             __u64);
3461
3462 static struct mdt_it_flavor {
3463         const struct req_format *it_fmt;
3464         __u32                    it_flags;
3465         int                    (*it_act)(enum mdt_it_code ,
3466                                          struct mdt_thread_info *,
3467                                          struct ldlm_lock **,
3468                                          __u64);
3469         long                     it_reint;
3470 } mdt_it_flavor[] = {
3471         [MDT_IT_OPEN]     = {
3472                 .it_fmt   = &RQF_LDLM_INTENT,
3473                 /*.it_flags = HABEO_REFERO,*/
3474                 .it_flags = 0,
3475                 .it_act   = mdt_intent_reint,
3476                 .it_reint = REINT_OPEN
3477         },
3478         [MDT_IT_OCREAT]   = {
3479                 .it_fmt   = &RQF_LDLM_INTENT,
3480                 .it_flags = MUTABOR,
3481                 .it_act   = mdt_intent_reint,
3482                 .it_reint = REINT_OPEN
3483         },
3484         [MDT_IT_CREATE]   = {
3485                 .it_fmt   = &RQF_LDLM_INTENT,
3486                 .it_flags = MUTABOR,
3487                 .it_act   = mdt_intent_reint,
3488                 .it_reint = REINT_CREATE
3489         },
3490         [MDT_IT_GETATTR]  = {
3491                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3492                 .it_flags = HABEO_REFERO,
3493                 .it_act   = mdt_intent_getattr
3494         },
3495         [MDT_IT_READDIR]  = {
3496                 .it_fmt   = NULL,
3497                 .it_flags = 0,
3498                 .it_act   = NULL
3499         },
3500         [MDT_IT_LOOKUP]   = {
3501                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3502                 .it_flags = HABEO_REFERO,
3503                 .it_act   = mdt_intent_getattr
3504         },
3505         [MDT_IT_UNLINK]   = {
3506                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3507                 .it_flags = MUTABOR,
3508                 .it_act   = NULL,
3509                 .it_reint = REINT_UNLINK
3510         },
3511         [MDT_IT_TRUNC]    = {
3512                 .it_fmt   = NULL,
3513                 .it_flags = MUTABOR,
3514                 .it_act   = NULL
3515         },
3516         [MDT_IT_GETXATTR] = {
3517                 .it_fmt   = NULL,
3518                 .it_flags = 0,
3519                 .it_act   = NULL
3520         },
3521         [MDT_IT_LAYOUT] = {
3522                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3523                 .it_flags = 0,
3524                 .it_act   = mdt_intent_layout
3525         }
3526 };
3527
3528 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3529                             struct ldlm_lock **lockp,
3530                             struct ldlm_lock *new_lock,
3531                             struct mdt_lock_handle *lh,
3532                             __u64 flags)
3533 {
3534         struct ptlrpc_request  *req = mdt_info_req(info);
3535         struct ldlm_lock       *lock = *lockp;
3536
3537         /*
3538          * Get new lock only for cases when possible resent did not find any
3539          * lock.
3540          */
3541         if (new_lock == NULL)
3542                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3543
3544         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3545                 lh->mlh_reg_lh.cookie = 0;
3546                 RETURN(0);
3547         }
3548
3549         LASSERTF(new_lock != NULL,
3550                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3551
3552         /*
3553          * If we've already given this lock to a client once, then we should
3554          * have no readers or writers.  Otherwise, we should have one reader
3555          * _or_ writer ref (which will be zeroed below) before returning the
3556          * lock to a client.
3557          */
3558         if (new_lock->l_export == req->rq_export) {
3559                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3560         } else {
3561                 LASSERT(new_lock->l_export == NULL);
3562                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3563         }
3564
3565         *lockp = new_lock;
3566
3567         if (new_lock->l_export == req->rq_export) {
3568                 /*
3569                  * Already gave this to the client, which means that we
3570                  * reconstructed a reply.
3571                  */
3572                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3573                         MSG_RESENT);
3574                 lh->mlh_reg_lh.cookie = 0;
3575                 RETURN(ELDLM_LOCK_REPLACED);
3576         }
3577
3578         /*
3579          * Fixup the lock to be given to the client.
3580          */
3581         lock_res_and_lock(new_lock);
3582         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3583          * possible blocking AST. */
3584         while (new_lock->l_readers > 0) {
3585                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3586                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3587                 new_lock->l_readers--;
3588         }
3589         while (new_lock->l_writers > 0) {
3590                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3591                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3592                 new_lock->l_writers--;
3593         }
3594
3595         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3596         new_lock->l_blocking_ast = lock->l_blocking_ast;
3597         new_lock->l_completion_ast = lock->l_completion_ast;
3598         new_lock->l_remote_handle = lock->l_remote_handle;
3599         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3600
3601         unlock_res_and_lock(new_lock);
3602
3603         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3604                      &new_lock->l_remote_handle,
3605                      &new_lock->l_exp_hash);
3606
3607         LDLM_LOCK_RELEASE(new_lock);
3608         lh->mlh_reg_lh.cookie = 0;
3609
3610         RETURN(ELDLM_LOCK_REPLACED);
3611 }
3612
3613 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3614                                     struct ldlm_lock *new_lock,
3615                                     struct ldlm_lock **old_lock,
3616                                     struct mdt_lock_handle *lh)
3617 {
3618         struct ptlrpc_request  *req = mdt_info_req(info);
3619         struct obd_export      *exp = req->rq_export;
3620         struct lustre_handle    remote_hdl;
3621         struct ldlm_request    *dlmreq;
3622         struct ldlm_lock       *lock;
3623
3624         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3625                 return;
3626
3627         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3628         remote_hdl = dlmreq->lock_handle[0];
3629
3630         /* In the function below, .hs_keycmp resolves to
3631          * ldlm_export_lock_keycmp() */
3632         /* coverity[overrun-buffer-val] */
3633         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3634         if (lock) {
3635                 if (lock != new_lock) {
3636                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3637                         lh->mlh_reg_mode = lock->l_granted_mode;
3638
3639                         LDLM_DEBUG(lock, "Restoring lock cookie");
3640                         DEBUG_REQ(D_DLMTRACE, req,
3641                                   "restoring lock cookie "LPX64,
3642                                   lh->mlh_reg_lh.cookie);
3643                         if (old_lock)
3644                                 *old_lock = LDLM_LOCK_GET(lock);
3645                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3646                         return;
3647                 }
3648
3649                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3650         }
3651
3652         /*
3653          * If the xid matches, then we know this is a resent request, and allow
3654          * it. (It's probably an OPEN, for which we don't send a lock.
3655          */
3656         if (req_xid_is_last(req))
3657                 return;
3658
3659         /*
3660          * This remote handle isn't enqueued, so we never received or processed
3661          * this request.  Clear MSG_RESENT, because it can be handled like any
3662          * normal request now.
3663          */
3664         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3665
3666         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3667                   remote_hdl.cookie);
3668 }
3669
3670 static int mdt_intent_getattr(enum mdt_it_code opcode,
3671                               struct mdt_thread_info *info,
3672                               struct ldlm_lock **lockp,
3673                               __u64 flags)
3674 {
3675         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3676         struct ldlm_lock       *new_lock = NULL;
3677         __u64                   child_bits;
3678         struct ldlm_reply      *ldlm_rep;
3679         struct ptlrpc_request  *req;
3680         struct mdt_body        *reqbody;
3681         struct mdt_body        *repbody;
3682         int                     rc, rc2;
3683         ENTRY;
3684
3685         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3686         LASSERT(reqbody);
3687
3688         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3689         LASSERT(repbody);
3690
3691         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3692         repbody->eadatasize = 0;
3693         repbody->aclsize = 0;
3694
3695         switch (opcode) {
3696         case MDT_IT_LOOKUP:
3697                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3698                 break;
3699         case MDT_IT_GETATTR:
3700                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3701                              MDS_INODELOCK_PERM;
3702                 break;
3703         default:
3704                 CERROR("Unsupported intent (%d)\n", opcode);
3705                 GOTO(out_shrink, rc = -EINVAL);
3706         }
3707
3708         rc = mdt_init_ucred(info, reqbody);
3709         if (rc)
3710                 GOTO(out_shrink, rc);
3711
3712         req = info->mti_pill->rc_req;
3713         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3714         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3715
3716         /* Get lock from request for possible resent case. */
3717         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3718
3719         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3720         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3721
3722         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3723                 ldlm_rep->lock_policy_res2 = 0;
3724         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3725             ldlm_rep->lock_policy_res2) {
3726                 lhc->mlh_reg_lh.cookie = 0ull;
3727                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3728         }
3729
3730         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3731         EXIT;
3732 out_ucred:
3733         mdt_exit_ucred(info);
3734 out_shrink:
3735         mdt_client_compatibility(info);
3736         rc2 = mdt_fix_reply(info);
3737         if (rc == 0)
3738                 rc = rc2;
3739         return rc;
3740 }
3741
3742 static int mdt_intent_layout(enum mdt_it_code opcode,
3743                              struct mdt_thread_info *info,
3744                              struct ldlm_lock **lockp,
3745                              __u64 flags)
3746 {
3747         struct layout_intent *layout;
3748         struct lu_fid *fid;
3749         struct mdt_object *obj = NULL;
3750         struct md_object *child = NULL;
3751         int rc;
3752         ENTRY;
3753
3754         if (opcode != MDT_IT_LAYOUT) {
3755                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3756                         opcode);
3757                 RETURN(-EINVAL);
3758         }
3759
3760         fid = &info->mti_tmp_fid2;
3761         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3762
3763         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3764         if (IS_ERR(obj))
3765                 RETURN(PTR_ERR(obj));
3766
3767         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3768                 child = mdt_object_child(obj);
3769
3770                 /* get the length of lsm */
3771                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3772                                   XATTR_NAME_LOV);
3773
3774                 if (rc > info->mti_mdt->mdt_max_mdsize)
3775                         info->mti_mdt->mdt_max_mdsize = rc;
3776         }
3777
3778         mdt_object_put(info->mti_env, obj);
3779
3780         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3781         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3782                         ldlm_lvbo_size(*lockp));
3783         rc = req_capsule_server_pack(info->mti_pill);
3784         if (rc != 0)
3785                 RETURN(-EINVAL);
3786
3787         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3788         LASSERT(layout != NULL);
3789         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3790                 /* return to normal ldlm handling */
3791                 RETURN(0);
3792
3793         CERROR("%s: Unsupported layout intent (%d)\n",
3794                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3795         RETURN(-EINVAL);
3796 }
3797
3798 static int mdt_intent_reint(enum mdt_it_code opcode,
3799                             struct mdt_thread_info *info,
3800                             struct ldlm_lock **lockp,
3801                             __u64 flags)
3802 {
3803         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3804         struct ldlm_reply      *rep = NULL;
3805         long                    opc;
3806         int                     rc;
3807
3808         static const struct req_format *intent_fmts[REINT_MAX] = {
3809                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3810                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3811         };
3812
3813         ENTRY;
3814
3815         opc = mdt_reint_opcode(info, intent_fmts);
3816         if (opc < 0)
3817                 RETURN(opc);
3818
3819         if (mdt_it_flavor[opcode].it_reint != opc) {
3820                 CERROR("Reint code %ld doesn't match intent: %d\n",
3821                        opc, opcode);
3822                 RETURN(err_serious(-EPROTO));
3823         }
3824
3825         /* Get lock from request for possible resent case. */
3826         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3827
3828         rc = mdt_reint_internal(info, lhc, opc);
3829
3830         /* Check whether the reply has been packed successfully. */
3831         if (mdt_info_req(info)->rq_repmsg != NULL)
3832                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3833         if (rep == NULL)
3834                 RETURN(err_serious(-EFAULT));
3835
3836         /* MDC expects this in any case */
3837         if (rc != 0)
3838                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3839
3840         /* the open lock or the lock for cross-ref object should be
3841          * returned to the client */
3842         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3843                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3844                 rep->lock_policy_res2 = 0;
3845                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3846                 RETURN(rc);
3847         }
3848
3849         rep->lock_policy_res2 = clear_serious(rc);
3850
3851         if (rep->lock_policy_res2 == -ENOENT &&
3852             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3853                 rep->lock_policy_res2 = 0;
3854
3855         if (rc == -ENOTCONN || rc == -ENODEV ||
3856             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3857                 /*
3858                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3859                  * will be returned by rq_status, and client at ptlrpc layer
3860                  * will detect this, then disconnect, reconnect the import
3861                  * immediately, instead of impacting the following the rpc.
3862                  */
3863                 lhc->mlh_reg_lh.cookie = 0ull;
3864                 RETURN(rc);
3865         } else {
3866                 /*
3867                  * For other cases, the error will be returned by intent.
3868                  * and client will retrieve the result from intent.
3869                  */
3870                  /*
3871                   * FIXME: when open lock is finished, that should be
3872                   * checked here.
3873                   */
3874                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3875                         LASSERTF(rc == 0, "Error occurred but lock handle "
3876                                  "is still in use, rc = %d\n", rc);
3877                         rep->lock_policy_res2 = 0;
3878                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3879                         RETURN(rc);
3880                 } else {
3881                         lhc->mlh_reg_lh.cookie = 0ull;
3882                         RETURN(ELDLM_LOCK_ABORTED);
3883                 }
3884         }
3885 }
3886
3887 static int mdt_intent_code(long itcode)
3888 {
3889         int rc;
3890
3891         switch(itcode) {
3892         case IT_OPEN:
3893                 rc = MDT_IT_OPEN;
3894                 break;
3895         case IT_OPEN|IT_CREAT:
3896                 rc = MDT_IT_OCREAT;
3897                 break;
3898         case IT_CREAT:
3899                 rc = MDT_IT_CREATE;
3900                 break;
3901         case IT_READDIR:
3902                 rc = MDT_IT_READDIR;
3903                 break;
3904         case IT_GETATTR:
3905                 rc = MDT_IT_GETATTR;
3906                 break;
3907         case IT_LOOKUP:
3908                 rc = MDT_IT_LOOKUP;
3909                 break;
3910         case IT_UNLINK:
3911                 rc = MDT_IT_UNLINK;
3912                 break;
3913         case IT_TRUNC:
3914                 rc = MDT_IT_TRUNC;
3915                 break;
3916         case IT_GETXATTR:
3917                 rc = MDT_IT_GETXATTR;
3918                 break;
3919         case IT_LAYOUT:
3920                 rc = MDT_IT_LAYOUT;
3921                 break;
3922         case IT_QUOTA_DQACQ:
3923         case IT_QUOTA_CONN:
3924                 rc = MDT_IT_QUOTA;
3925                 break;
3926         default:
3927                 CERROR("Unknown intent opcode: %ld\n", itcode);
3928                 rc = -EINVAL;
3929                 break;
3930         }
3931         return rc;
3932 }
3933
3934 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3935                           struct ldlm_lock **lockp, __u64 flags)
3936 {
3937         struct req_capsule   *pill;
3938         struct mdt_it_flavor *flv;
3939         int opc;
3940         int rc;
3941         ENTRY;
3942
3943         opc = mdt_intent_code(itopc);
3944         if (opc < 0)
3945                 RETURN(-EINVAL);
3946
3947         pill = info->mti_pill;
3948
3949         if (opc == MDT_IT_QUOTA) {
3950                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3951
3952                 if (qmt == NULL)
3953                         RETURN(-EOPNOTSUPP);
3954
3955                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3956                 /* pass the request to quota master */
3957                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3958                                                  mdt_info_req(info), lockp,
3959                                                  flags);
3960                 RETURN(rc);
3961         }
3962
3963         flv  = &mdt_it_flavor[opc];
3964         if (flv->it_fmt != NULL)
3965                 req_capsule_extend(pill, flv->it_fmt);
3966
3967         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3968         if (rc == 0) {
3969                 struct ptlrpc_request *req = mdt_info_req(info);
3970                 if (flv->it_flags & MUTABOR &&
3971                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3972                         RETURN(-EROFS);
3973         }
3974         if (rc == 0 && flv->it_act != NULL) {
3975                 struct ldlm_reply *rep;
3976
3977                 /* execute policy */
3978                 rc = flv->it_act(opc, info, lockp, flags);
3979
3980                 /* Check whether the reply has been packed successfully. */
3981                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3982                         rep = req_capsule_server_get(info->mti_pill,
3983                                                      &RMF_DLM_REP);
3984                         rep->lock_policy_res2 =
3985                                 ptlrpc_status_hton(rep->lock_policy_res2);
3986                 }
3987         } else {
3988                 rc = -EOPNOTSUPP;
3989         }
3990         RETURN(rc);
3991 }
3992
3993 static int mdt_intent_policy(struct ldlm_namespace *ns,
3994                              struct ldlm_lock **lockp, void *req_cookie,
3995                              ldlm_mode_t mode, __u64 flags, void *data)
3996 {
3997         struct mdt_thread_info *info;
3998         struct ptlrpc_request  *req  =  req_cookie;
3999         struct ldlm_intent     *it;
4000         struct req_capsule     *pill;
4001         int rc;
4002
4003         ENTRY;
4004
4005         LASSERT(req != NULL);
4006
4007         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
4008                                   &mdt_thread_key);
4009         LASSERT(info != NULL);
4010         pill = info->mti_pill;
4011         LASSERT(pill->rc_req == req);
4012
4013         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4014                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4015                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4016                 if (it != NULL) {
4017                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4018                         if (rc == 0)
4019                                 rc = ELDLM_OK;
4020
4021                         /* Lock without inodebits makes no sense and will oops
4022                          * later in ldlm. Let's check it now to see if we have
4023                          * ibits corrupted somewhere in mdt_intent_opc().
4024                          * The case for client miss to set ibits has been
4025                          * processed by others. */
4026                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
4027                                         lr_type == LDLM_IBITS,
4028                                      info->mti_dlm_req->lock_desc.\
4029                                         l_policy_data.l_inodebits.bits != 0));
4030                 } else
4031                         rc = err_serious(-EFAULT);
4032         } else {
4033                 /* No intent was provided */
4034                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
4035                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4036                 rc = req_capsule_server_pack(pill);
4037                 if (rc)
4038                         rc = err_serious(rc);
4039         }
4040         RETURN(rc);
4041 }
4042
4043 static int mdt_seq_fini(const struct lu_env *env,
4044                         struct mdt_device *m)
4045 {
4046         return seq_site_fini(env, mdt_seq_site(m));
4047 }
4048
4049 static int mdt_seq_init(const struct lu_env *env,
4050                         const char *uuid,
4051                         struct mdt_device *m)
4052 {
4053         struct seq_server_site *ss;
4054         char *prefix;
4055         int rc;
4056         ENTRY;
4057
4058         ss = mdt_seq_site(m);
4059
4060         /*
4061          * This is sequence-controller node. Init seq-controller server on local
4062          * MDT.
4063          */
4064         if (ss->ss_node_id == 0) {
4065                 LASSERT(ss->ss_control_seq == NULL);
4066
4067                 OBD_ALLOC_PTR(ss->ss_control_seq);
4068                 if (ss->ss_control_seq == NULL)
4069                         RETURN(-ENOMEM);
4070
4071                 rc = seq_server_init(ss->ss_control_seq,
4072                                      m->mdt_bottom, uuid,
4073                                      LUSTRE_SEQ_CONTROLLER,
4074                                      ss,
4075                                      env);
4076
4077                 if (rc)
4078                         GOTO(out_seq_fini, rc);
4079
4080                 OBD_ALLOC_PTR(ss->ss_client_seq);
4081                 if (ss->ss_client_seq == NULL)
4082                         GOTO(out_seq_fini, rc = -ENOMEM);
4083
4084                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4085                 if (prefix == NULL) {
4086                         OBD_FREE_PTR(ss->ss_client_seq);
4087                         GOTO(out_seq_fini, rc = -ENOMEM);
4088                 }
4089
4090                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
4091                          uuid);
4092
4093                 /*
4094                  * Init seq-controller client after seq-controller server is
4095                  * ready. Pass ss->ss_control_seq to it for direct talking.
4096                  */
4097                 rc = seq_client_init(ss->ss_client_seq, NULL,
4098                                      LUSTRE_SEQ_METADATA, prefix,
4099                                      ss->ss_control_seq);
4100                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
4101
4102                 if (rc)
4103                         GOTO(out_seq_fini, rc);
4104         }
4105
4106         /* Init seq-server on local MDT */
4107         LASSERT(ss->ss_server_seq == NULL);
4108
4109         OBD_ALLOC_PTR(ss->ss_server_seq);
4110         if (ss->ss_server_seq == NULL)
4111                 GOTO(out_seq_fini, rc = -ENOMEM);
4112
4113         rc = seq_server_init(ss->ss_server_seq,
4114                              m->mdt_bottom, uuid,
4115                              LUSTRE_SEQ_SERVER,
4116                              ss,
4117                              env);
4118         if (rc)
4119                 GOTO(out_seq_fini, rc = -ENOMEM);
4120
4121         /* Assign seq-controller client to local seq-server. */
4122         if (ss->ss_node_id == 0) {
4123                 LASSERT(ss->ss_client_seq != NULL);
4124
4125                 rc = seq_server_set_cli(ss->ss_server_seq,
4126                                         ss->ss_client_seq,
4127                                         env);
4128         }
4129
4130         EXIT;
4131 out_seq_fini:
4132         if (rc)
4133                 mdt_seq_fini(env, m);
4134
4135         return rc;
4136 }
4137
4138 /*
4139  * FLD wrappers
4140  */
4141 static int mdt_fld_fini(const struct lu_env *env,
4142                         struct mdt_device *m)
4143 {
4144         struct seq_server_site *ss = mdt_seq_site(m);
4145         ENTRY;
4146
4147         if (ss && ss->ss_server_fld) {
4148                 fld_server_fini(env, ss->ss_server_fld);
4149                 OBD_FREE_PTR(ss->ss_server_fld);
4150                 ss->ss_server_fld = NULL;
4151         }
4152
4153         RETURN(0);
4154 }
4155
4156 static int mdt_fld_init(const struct lu_env *env,
4157                         const char *uuid,
4158                         struct mdt_device *m)
4159 {
4160         struct seq_server_site *ss;
4161         int rc;
4162         ENTRY;
4163
4164         ss = mdt_seq_site(m);
4165
4166         OBD_ALLOC_PTR(ss->ss_server_fld);
4167         if (ss->ss_server_fld == NULL)
4168                 RETURN(rc = -ENOMEM);
4169
4170         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4171                              ss->ss_node_id, LU_SEQ_RANGE_MDT);
4172         if (rc) {
4173                 OBD_FREE_PTR(ss->ss_server_fld);
4174                 ss->ss_server_fld = NULL;
4175                 RETURN(rc);
4176         }
4177
4178         RETURN(0);
4179 }
4180
4181 static void mdt_stack_pre_fini(const struct lu_env *env,
4182                            struct mdt_device *m, struct lu_device *top)
4183 {
4184         struct obd_device       *obd;
4185         struct lustre_cfg_bufs  *bufs;
4186         struct lustre_cfg       *lcfg;
4187         struct mdt_thread_info  *info;
4188         ENTRY;
4189
4190         LASSERT(top);
4191
4192         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4193         LASSERT(info != NULL);
4194
4195         bufs = &info->mti_u.bufs;
4196
4197         LASSERT(m->mdt_child_exp);
4198         LASSERT(m->mdt_child_exp->exp_obd);
4199         obd = m->mdt_child_exp->exp_obd;
4200
4201         /* process cleanup, pass mdt obd name to get obd umount flags */
4202         /* XXX: this is needed because all layers are referenced by
4203          * objects (some of them are pinned by osd, for example *
4204          * the proper solution should be a model where object used
4205          * by osd only doesn't have mdt/mdd slices -bzzz */
4206         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4207         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4208         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4209         if (!lcfg) {
4210                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
4211                 return;
4212         }
4213         top->ld_ops->ldo_process_config(env, top, lcfg);
4214         lustre_cfg_free(lcfg);
4215         EXIT;
4216 }
4217
4218 static void mdt_stack_fini(const struct lu_env *env,
4219                            struct mdt_device *m, struct lu_device *top)
4220 {
4221         struct obd_device       *obd = mdt2obd_dev(m);
4222         struct lustre_cfg_bufs  *bufs;
4223         struct lustre_cfg       *lcfg;
4224         struct mdt_thread_info  *info;
4225         char                     flags[3] = "";
4226         ENTRY;
4227
4228         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4229         LASSERT(info != NULL);
4230
4231         lu_dev_del_linkage(top->ld_site, top);
4232
4233         lu_site_purge(env, top->ld_site, -1);
4234
4235         bufs = &info->mti_u.bufs;
4236         /* process cleanup, pass mdt obd name to get obd umount flags */
4237         /* another purpose is to let all layers to release their objects */
4238         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4239         if (obd->obd_force)
4240                 strcat(flags, "F");
4241         if (obd->obd_fail)
4242                 strcat(flags, "A");
4243         lustre_cfg_bufs_set_string(bufs, 1, flags);
4244         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4245         if (!lcfg) {
4246                 CERROR("Cannot alloc lcfg!\n");
4247                 return;
4248         }
4249         LASSERT(top);
4250         top->ld_ops->ldo_process_config(env, top, lcfg);
4251         lustre_cfg_free(lcfg);
4252
4253         lu_site_purge(env, top->ld_site, -1);
4254
4255         m->mdt_child = NULL;
4256         m->mdt_bottom = NULL;
4257
4258         obd_disconnect(m->mdt_child_exp);
4259         m->mdt_child_exp = NULL;
4260
4261         obd_disconnect(m->mdt_bottom_exp);
4262         m->mdt_child_exp = NULL;
4263 }
4264
4265 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4266                                const char *next, struct obd_export **exp)
4267 {
4268         struct obd_connect_data *data = NULL;
4269         struct obd_device       *obd;
4270         int                      rc;
4271         ENTRY;
4272
4273         OBD_ALLOC_PTR(data);
4274         if (data == NULL)
4275                 GOTO(out, rc = -ENOMEM);
4276
4277         obd = class_name2obd(next);
4278         if (obd == NULL) {
4279                 CERROR("%s: can't locate next device: %s\n",
4280                        mdt_obd_name(m), next);
4281                 GOTO(out, rc = -ENOTCONN);
4282         }
4283
4284         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4285         data->ocd_version = LUSTRE_VERSION_CODE;
4286
4287         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4288         if (rc) {
4289                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4290                        mdt_obd_name(m), next, rc);
4291                 GOTO(out, rc);
4292         }
4293
4294 out:
4295         if (data)
4296                 OBD_FREE_PTR(data);
4297         RETURN(rc);
4298 }
4299
4300 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4301                           struct lustre_cfg *cfg)
4302 {
4303         char                   *dev = lustre_cfg_string(cfg, 0);
4304         int                     rc, name_size, uuid_size;
4305         char                   *name, *uuid, *p;
4306         struct lustre_cfg_bufs *bufs;
4307         struct lustre_cfg      *lcfg;
4308         struct obd_device      *obd;
4309         struct lustre_profile  *lprof;
4310         struct lu_site         *site;
4311         ENTRY;
4312
4313         /* in 1.8 we had the only device in the stack - MDS.
4314          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4315          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4316          * to generate names and setup MDT, MDD. MDT will be using
4317          * generated name to connect to MDD. for MDD the next device
4318          * will be LOD with name taken from so called "profile" which
4319          * is generated by mount_option line
4320          *
4321          * 1.8 MGS generates config. commands like this:
4322          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4323          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4324          * 2.0 MGS generates config. commands like this:
4325          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4326          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4327          *                    3:lustre-MDT0000-mdtlov  4:f
4328          *
4329          * we generate MDD name from MDT one, just replacing T with D
4330          *
4331          * after all the preparations, the logical equivalent will be
4332          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4333          *                    3:lustre-MDT0000-mdtlov  4:f
4334          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4335          *                    3:lustre-MDD0000  4:f
4336          *
4337          *  notice we build the stack from down to top: MDD first, then MDT */
4338
4339         name_size = MAX_OBD_NAME;
4340         uuid_size = MAX_OBD_NAME;
4341
4342         OBD_ALLOC(name, name_size);
4343         OBD_ALLOC(uuid, uuid_size);
4344         if (name == NULL || uuid == NULL)
4345                 GOTO(cleanup_mem, rc = -ENOMEM);
4346
4347         OBD_ALLOC_PTR(bufs);
4348         if (!bufs)
4349                 GOTO(cleanup_mem, rc = -ENOMEM);
4350
4351         strcpy(name, dev);
4352         p = strstr(name, "-MDT");
4353         if (p == NULL)
4354                 GOTO(cleanup_mem, rc = -ENOMEM);
4355         p[3] = 'D';
4356
4357         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4358
4359         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4360         if (lprof == NULL || lprof->lp_dt == NULL) {
4361                 CERROR("can't find the profile: %s\n",
4362                        lustre_cfg_string(cfg, 0));
4363                 GOTO(cleanup_mem, rc = -EINVAL);
4364         }
4365
4366         lustre_cfg_bufs_reset(bufs, name);
4367         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4368         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4369         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4370
4371         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4372         if (!lcfg)
4373                 GOTO(free_bufs, rc = -ENOMEM);
4374
4375         rc = class_attach(lcfg);
4376         if (rc)
4377                 GOTO(lcfg_cleanup, rc);
4378
4379         obd = class_name2obd(name);
4380         if (!obd) {
4381                 CERROR("Can not find obd %s (%s in config)\n",
4382                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4383                 GOTO(class_detach, rc = -EINVAL);
4384         }
4385
4386         lustre_cfg_free(lcfg);
4387
4388         lustre_cfg_bufs_reset(bufs, name);
4389         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4390         lustre_cfg_bufs_set_string(bufs, 2, dev);
4391         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4392
4393         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4394
4395         rc = class_setup(obd, lcfg);
4396         if (rc)
4397                 GOTO(class_detach, rc);
4398
4399         /* connect to MDD we just setup */
4400         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4401         if (rc)
4402                 RETURN(rc);
4403
4404         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4405         LASSERT(site);
4406         LASSERT(mdt_lu_site(mdt) == NULL);
4407         mdt->mdt_lu_dev.ld_site = site;
4408         site->ls_top_dev = &mdt->mdt_lu_dev;
4409         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4410
4411
4412         /* now connect to bottom OSD */
4413         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4414         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4415         if (rc)
4416                 RETURN(rc);
4417         mdt->mdt_bottom =
4418                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4419
4420
4421         rc = lu_env_refill((struct lu_env *)env);
4422         if (rc != 0)
4423                 CERROR("Failure to refill session: '%d'\n", rc);
4424
4425         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4426
4427         EXIT;
4428 class_detach:
4429         if (rc)
4430                 class_detach(obd, lcfg);
4431 lcfg_cleanup:
4432         lustre_cfg_free(lcfg);
4433 free_bufs:
4434         OBD_FREE_PTR(bufs);
4435 cleanup_mem:
4436         if (name)
4437                 OBD_FREE(name, name_size);
4438         if (uuid)
4439                 OBD_FREE(uuid, uuid_size);
4440         RETURN(rc);
4441 }
4442
4443 /* setup quota master target on MDT0 */
4444 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4445                           struct lustre_cfg *cfg)
4446 {
4447         struct obd_device       *obd;
4448         char                    *dev = lustre_cfg_string(cfg, 0);
4449         char                    *qmtname, *uuid, *p;
4450         struct lustre_cfg_bufs  *bufs;
4451         struct lustre_cfg       *lcfg;
4452         struct lustre_profile   *lprof;
4453         struct obd_connect_data *data;
4454         int                      rc;
4455         ENTRY;
4456
4457         LASSERT(mdt->mdt_qmt_exp == NULL);
4458         LASSERT(mdt->mdt_qmt_dev == NULL);
4459
4460         /* quota master is on MDT0 only for now */
4461         if (mdt->mdt_seq_site.ss_node_id != 0)
4462                 RETURN(0);
4463
4464         /* MGS generates config commands which look as follows:
4465          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4466          *                    3:lustre-MDT0000-mdtlov  4:f
4467          *
4468          * We generate the QMT name from the MDT one, just replacing MD with QM
4469          * after all the preparations, the logical equivalent will be:
4470          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4471          *                    3:lustre-MDT0000-osd  4:f */
4472         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4473         OBD_ALLOC(uuid, UUID_MAX);
4474         OBD_ALLOC_PTR(bufs);
4475         OBD_ALLOC_PTR(data);
4476         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4477                 GOTO(cleanup_mem, rc = -ENOMEM);
4478
4479         strcpy(qmtname, dev);
4480         p = strstr(qmtname, "-MDT");
4481         if (p == NULL)
4482                 GOTO(cleanup_mem, rc = -ENOMEM);
4483         /* replace MD with QM */
4484         p[1] = 'Q';
4485         p[2] = 'M';
4486
4487         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4488
4489         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4490         if (lprof == NULL || lprof->lp_dt == NULL) {
4491                 CERROR("can't find profile for %s\n",
4492                        lustre_cfg_string(cfg, 0));
4493                 GOTO(cleanup_mem, rc = -EINVAL);
4494         }
4495
4496         lustre_cfg_bufs_reset(bufs, qmtname);
4497         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4498         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4499         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4500
4501         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4502         if (!lcfg)
4503                 GOTO(cleanup_mem, rc = -ENOMEM);
4504
4505         rc = class_attach(lcfg);
4506         if (rc)
4507                 GOTO(lcfg_cleanup, rc);
4508
4509         obd = class_name2obd(qmtname);
4510         if (!obd) {
4511                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4512                        lustre_cfg_string(cfg, 0));
4513                 GOTO(class_detach, rc = -EINVAL);
4514         }
4515
4516         lustre_cfg_free(lcfg);
4517
4518         lustre_cfg_bufs_reset(bufs, qmtname);
4519         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4520         lustre_cfg_bufs_set_string(bufs, 2, dev);
4521
4522         /* for quota, the next device should be the OSD device */
4523         lustre_cfg_bufs_set_string(bufs, 3,
4524                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4525
4526         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4527
4528         rc = class_setup(obd, lcfg);
4529         if (rc)
4530                 GOTO(class_detach, rc);
4531
4532         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4533
4534         /* configure local quota objects */
4535         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4536                                                    &mdt->mdt_lu_dev,
4537                                                    mdt->mdt_qmt_dev);
4538         if (rc)
4539                 GOTO(class_cleanup, rc);
4540
4541         /* connect to quota master target */
4542         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4543         data->ocd_version = LUSTRE_VERSION_CODE;
4544         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4545                          data, NULL);
4546         if (rc) {
4547                 CERROR("cannot connect to quota master device %s (%d)\n",
4548                        qmtname, rc);
4549                 GOTO(class_cleanup, rc);
4550         }
4551
4552         EXIT;
4553 class_cleanup:
4554         if (rc) {
4555                 class_manual_cleanup(obd);
4556                 mdt->mdt_qmt_dev = NULL;
4557         }
4558 class_detach:
4559         if (rc)
4560                 class_detach(obd, lcfg);
4561 lcfg_cleanup:
4562         lustre_cfg_free(lcfg);
4563 cleanup_mem:
4564         if (bufs)
4565                 OBD_FREE_PTR(bufs);
4566         if (qmtname)
4567                 OBD_FREE(qmtname, MAX_OBD_NAME);
4568         if (uuid)
4569                 OBD_FREE(uuid, UUID_MAX);
4570         if (data)
4571                 OBD_FREE_PTR(data);
4572         return rc;
4573 }
4574
4575 /* Shutdown quota master target associated with mdt */
4576 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4577 {
4578         ENTRY;
4579
4580         if (mdt->mdt_qmt_exp == NULL)
4581                 RETURN_EXIT;
4582         LASSERT(mdt->mdt_qmt_dev != NULL);
4583
4584         /* the qmt automatically shuts down when the mdt disconnects */
4585         obd_disconnect(mdt->mdt_qmt_exp);
4586         mdt->mdt_qmt_exp = NULL;
4587         mdt->mdt_qmt_dev = NULL;
4588         EXIT;
4589 }
4590
4591 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4592 {
4593         struct md_device  *next = m->mdt_child;
4594         struct lu_device  *d    = &m->mdt_lu_dev;
4595         struct obd_device *obd = mdt2obd_dev(m);
4596         ENTRY;
4597
4598         target_recovery_fini(obd);
4599
4600         ping_evictor_stop();
4601
4602
4603         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4604         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4605         obd_exports_barrier(obd);
4606         obd_zombie_barrier();
4607
4608         mdt_procfs_fini(m);
4609
4610         tgt_fini(env, &m->mdt_lut);
4611         mdt_fs_cleanup(env, m);
4612         upcall_cache_cleanup(m->mdt_identity_cache);
4613         m->mdt_identity_cache = NULL;
4614
4615         if (m->mdt_namespace != NULL) {
4616                 ldlm_namespace_free(m->mdt_namespace, NULL,
4617                                     d->ld_obd->obd_force);
4618                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4619         }
4620
4621         mdt_quota_fini(env, m);
4622
4623         cfs_free_nidlist(&m->mdt_nosquash_nids);
4624         if (m->mdt_nosquash_str) {
4625                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4626                 m->mdt_nosquash_str = NULL;
4627                 m->mdt_nosquash_strlen = 0;
4628         }
4629
4630         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
4631
4632         mdt_seq_fini(env, m);
4633         mdt_fld_fini(env, m);
4634         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4635
4636         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4637         cfs_timer_disarm(&m->mdt_ck_timer);
4638         mdt_ck_thread_stop(m);
4639
4640         /*
4641          * Finish the stack
4642          */
4643         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4644
4645         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4646
4647         server_put_mount(mdt_obd_name(m), NULL);
4648
4649         EXIT;
4650 }
4651
4652 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4653 {
4654         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4655         struct sptlrpc_rule_set  tmp_rset;
4656         int                      rc;
4657
4658         sptlrpc_rule_set_init(&tmp_rset);
4659         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4660         if (rc) {
4661                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4662                        mdt_obd_name(m), rc);
4663                 return rc;
4664         }
4665
4666         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4667
4668         write_lock(&m->mdt_sptlrpc_lock);
4669         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4670         m->mdt_sptlrpc_rset = tmp_rset;
4671         write_unlock(&m->mdt_sptlrpc_lock);
4672
4673         return 0;
4674 }
4675
4676 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4677
4678 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4679                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4680 {
4681         struct mdt_thread_info    *info;
4682         struct obd_device         *obd;
4683         const char                *dev = lustre_cfg_string(cfg, 0);
4684         const char                *num = lustre_cfg_string(cfg, 2);
4685         struct lustre_mount_info  *lmi = NULL;
4686         struct lustre_sb_info     *lsi;
4687         struct lu_site            *s;
4688         struct seq_server_site    *ss_site;
4689         const char                *identity_upcall = "NONE";
4690         struct md_device          *next;
4691         int                        rc;
4692         int                        node_id;
4693         mntopt_t                   mntopts;
4694         ENTRY;
4695
4696         lu_device_init(&m->mdt_lu_dev, ldt);
4697         /*
4698          * Environment (env) might be missing mdt_thread_key values at that
4699          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4700          * mode.
4701          *
4702          * Usually device allocation path doesn't use module key values, but
4703          * mdt has to do a lot of work here, so allocate key value.
4704          */
4705         rc = lu_env_refill((struct lu_env *)env);
4706         if (rc != 0)
4707                 RETURN(rc);
4708
4709         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4710         LASSERT(info != NULL);
4711
4712         obd = class_name2obd(dev);
4713         LASSERT(obd != NULL);
4714
4715         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4716
4717         m->mdt_som_conf = 0;
4718
4719         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4720         lmi = server_get_mount(dev);
4721         if (lmi == NULL) {
4722                 CERROR("Cannot get mount info for %s!\n", dev);
4723                 RETURN(-EFAULT);
4724         } else {
4725                 lsi = s2lsi(lmi->lmi_sb);
4726                 /* CMD is supported only in IAM mode */
4727                 LASSERT(num);
4728                 node_id = simple_strtol(num, NULL, 10);
4729                 obd->u.obt.obt_magic = OBT_MAGIC;
4730         }
4731
4732         rwlock_init(&m->mdt_sptlrpc_lock);
4733         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4734
4735         spin_lock_init(&m->mdt_ioepoch_lock);
4736         m->mdt_opts.mo_compat_resname = 0;
4737         m->mdt_opts.mo_mds_capa = 1;
4738         m->mdt_opts.mo_oss_capa = 1;
4739         m->mdt_capa_timeout = CAPA_TIMEOUT;
4740         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4741         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4742         m->mdt_squash_uid = 0;
4743         m->mdt_squash_gid = 0;
4744         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4745         m->mdt_nosquash_str = NULL;
4746         m->mdt_nosquash_strlen = 0;
4747         init_rwsem(&m->mdt_squash_sem);
4748         spin_lock_init(&m->mdt_osfs_lock);
4749         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4750         m->mdt_enable_remote_dir = 0;
4751         m->mdt_enable_remote_dir_gid = 0;
4752
4753         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4754         m->mdt_lu_dev.ld_obd = obd;
4755         /* Set this lu_device to obd for error handling purposes. */
4756         obd->obd_lu_dev = &m->mdt_lu_dev;
4757
4758         /* init the stack */
4759         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4760         if (rc) {
4761                 CERROR("%s: Can't init device stack, rc %d\n",
4762                        mdt_obd_name(m), rc);
4763                 GOTO(err_lmi, rc);
4764         }
4765
4766         s = mdt_lu_site(m);
4767         ss_site = mdt_seq_site(m);
4768         s->ld_seq_site = ss_site;
4769         ss_site->ss_lu = s;
4770
4771         /* set server index */
4772         ss_site->ss_node_id = node_id;
4773
4774         /* failover is the default
4775          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4776          * assumed whose ss_node_id == 0 XXX
4777          * */
4778         obd->obd_replayable = 1;
4779         /* No connection accepted until configurations will finish */
4780         obd->obd_no_conn = 1;
4781
4782         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4783                 char *str = lustre_cfg_string(cfg, 4);
4784                 if (strchr(str, 'n')) {
4785                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4786                         obd->obd_replayable = 0;
4787                 }
4788         }
4789
4790         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4791         if (rc)
4792                 GOTO(err_fini_stack, rc);
4793
4794         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4795         if (rc)
4796                 GOTO(err_lut, rc);
4797
4798         rc = mdt_seq_init(env, mdt_obd_name(m), m);
4799         if (rc)
4800                 GOTO(err_fini_fld, rc);
4801
4802         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4803                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4804         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4805                                               LDLM_NAMESPACE_SERVER,
4806                                               LDLM_NAMESPACE_GREEDY,
4807                                               LDLM_NS_TYPE_MDT);
4808         if (m->mdt_namespace == NULL)
4809                 GOTO(err_fini_seq, rc = -ENOMEM);
4810
4811         m->mdt_namespace->ns_lvbp = m;
4812         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4813
4814         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4815         /* set obd_namespace for compatibility with old code */
4816         obd->obd_namespace = m->mdt_namespace;
4817
4818         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4819
4820         rc = mdt_ck_thread_start(m);
4821         if (rc)
4822                 GOTO(err_free_ns, rc);
4823
4824         rc = mdt_fs_setup(env, m, obd, lsi);
4825         if (rc)
4826                 GOTO(err_capa, rc);
4827
4828         mdt_adapt_sptlrpc_conf(obd, 1);
4829
4830         next = m->mdt_child;
4831         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4832                                          &mntopts);
4833         if (rc)
4834                 GOTO(err_llog_cleanup, rc);
4835
4836         if (mntopts & MNTOPT_USERXATTR)
4837                 m->mdt_opts.mo_user_xattr = 1;
4838         else
4839                 m->mdt_opts.mo_user_xattr = 0;
4840
4841         if (mntopts & MNTOPT_ACL)
4842                 m->mdt_opts.mo_acl = 1;
4843         else
4844                 m->mdt_opts.mo_acl = 0;
4845
4846         /* XXX: to support suppgid for ACL, we enable identity_upcall
4847          * by default, otherwise, maybe got unexpected -EACCESS. */
4848         if (m->mdt_opts.mo_acl)
4849                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4850
4851         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4852                                                 identity_upcall,
4853                                                 &mdt_identity_upcall_cache_ops);
4854         if (IS_ERR(m->mdt_identity_cache)) {
4855                 rc = PTR_ERR(m->mdt_identity_cache);
4856                 m->mdt_identity_cache = NULL;
4857                 GOTO(err_llog_cleanup, rc);
4858         }
4859
4860         rc = mdt_procfs_init(m, dev);
4861         if (rc) {
4862                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4863                 GOTO(err_recovery, rc);
4864         }
4865
4866         rc = mdt_quota_init(env, m, cfg);
4867         if (rc)
4868                 GOTO(err_procfs, rc);
4869
4870         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4871         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4872                            "mdt_ldlm_client", m->mdt_ldlm_client);
4873
4874         ping_evictor_start();
4875
4876         /* recovery will be started upon mdt_prepare()
4877          * when the whole stack is complete and ready
4878          * to serve the requests */
4879
4880         mdt_init_capa_ctxt(env, m);
4881
4882         /* Reduce the initial timeout on an MDS because it doesn't need such
4883          * a long timeout as an OST does. Adaptive timeouts will adjust this
4884          * value appropriately. */
4885         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4886                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4887
4888         RETURN(0);
4889
4890 err_procfs:
4891         mdt_procfs_fini(m);
4892 err_recovery:
4893         target_recovery_fini(obd);
4894         upcall_cache_cleanup(m->mdt_identity_cache);
4895         m->mdt_identity_cache = NULL;
4896 err_llog_cleanup:
4897         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4898         mdt_fs_cleanup(env, m);
4899 err_capa:
4900         cfs_timer_disarm(&m->mdt_ck_timer);
4901         mdt_ck_thread_stop(m);
4902 err_free_ns:
4903         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4904         obd->obd_namespace = m->mdt_namespace = NULL;
4905 err_fini_seq:
4906         mdt_seq_fini(env, m);
4907 err_fini_fld:
4908         mdt_fld_fini(env, m);
4909 err_lut:
4910         tgt_fini(env, &m->mdt_lut);
4911 err_fini_stack:
4912         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4913 err_lmi:
4914         if (lmi)
4915                 server_put_mount(dev, lmi->lmi_mnt);
4916         return (rc);
4917 }
4918
4919 /* For interoperability, the left element is old parameter, the right one
4920  * is the new version of the parameter, if some parameter is deprecated,
4921  * the new version should be set as NULL. */
4922 static struct cfg_interop_param mdt_interop_param[] = {
4923         { "mdt.group_upcall",   NULL },
4924         { "mdt.quota_type",     NULL },
4925         { "mdd.quota_type",     NULL },
4926         { "mdt.rootsquash",     "mdt.root_squash" },
4927         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4928         { NULL }
4929 };
4930
4931 /* used by MGS to process specific configurations */
4932 static int mdt_process_config(const struct lu_env *env,
4933                               struct lu_device *d, struct lustre_cfg *cfg)
4934 {
4935         struct mdt_device *m = mdt_dev(d);
4936         struct md_device *md_next = m->mdt_child;
4937         struct lu_device *next = md2lu_dev(md_next);
4938         int rc;
4939         ENTRY;
4940
4941         switch (cfg->lcfg_command) {
4942         case LCFG_PARAM: {
4943                 struct lprocfs_static_vars  lvars;
4944                 struct obd_device          *obd = d->ld_obd;
4945
4946                 /* For interoperability */
4947                 struct cfg_interop_param   *ptr = NULL;
4948                 struct lustre_cfg          *old_cfg = NULL;
4949                 char                       *param = NULL;
4950
4951                 param = lustre_cfg_string(cfg, 1);
4952                 if (param == NULL) {
4953                         CERROR("param is empty\n");
4954                         rc = -EINVAL;
4955                         break;
4956                 }
4957
4958                 ptr = class_find_old_param(param, mdt_interop_param);
4959                 if (ptr != NULL) {
4960                         if (ptr->new_param == NULL) {
4961                                 rc = 0;
4962                                 CWARN("For interoperability, skip this %s."
4963                                       " It is obsolete.\n", ptr->old_param);
4964                                 break;
4965                         }
4966
4967                         CWARN("Found old param %s, changed it to %s.\n",
4968                               ptr->old_param, ptr->new_param);
4969
4970                         old_cfg = cfg;
4971                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4972                         if (IS_ERR(cfg)) {
4973                                 rc = PTR_ERR(cfg);
4974                                 break;
4975                         }
4976                 }
4977
4978                 lprocfs_mdt_init_vars(&lvars);
4979                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4980                                               cfg, obd);
4981                 if (rc > 0 || rc == -ENOSYS)
4982                         /* we don't understand; pass it on */
4983                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4984
4985                 if (old_cfg != NULL)
4986                         lustre_cfg_free(cfg);
4987
4988                 break;
4989         }
4990         default:
4991                 /* others are passed further */
4992                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4993                 break;
4994         }
4995         RETURN(rc);
4996 }
4997
4998 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4999                                           const struct lu_object_header *hdr,
5000                                           struct lu_device *d)
5001 {
5002         struct mdt_object *mo;
5003
5004         ENTRY;
5005
5006         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
5007         if (mo != NULL) {
5008                 struct lu_object *o;
5009                 struct lu_object_header *h;
5010
5011                 o = &mo->mot_obj;
5012                 h = &mo->mot_header;
5013                 lu_object_header_init(h);
5014                 lu_object_init(o, h, d);
5015                 lu_object_add_top(h, o);
5016                 o->lo_ops = &mdt_obj_ops;
5017                 mutex_init(&mo->mot_ioepoch_mutex);
5018                 mutex_init(&mo->mot_lov_mutex);
5019                 RETURN(o);
5020         } else
5021                 RETURN(NULL);
5022 }
5023
5024 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5025                            const struct lu_object_conf *unused)
5026 {
5027         struct mdt_device *d = mdt_dev(o->lo_dev);
5028         struct lu_device  *under;
5029         struct lu_object  *below;
5030         int                rc = 0;
5031         ENTRY;
5032
5033         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5034                PFID(lu_object_fid(o)));
5035
5036         under = &d->mdt_child->md_lu_dev;
5037         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5038         if (below != NULL) {
5039                 lu_object_add(o, below);
5040         } else
5041                 rc = -ENOMEM;
5042
5043         RETURN(rc);
5044 }
5045
5046 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5047 {
5048         struct mdt_object *mo = mdt_obj(o);
5049         struct lu_object_header *h;
5050         ENTRY;
5051
5052         h = o->lo_header;
5053         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5054                PFID(lu_object_fid(o)));
5055
5056         lu_object_fini(o);
5057         lu_object_header_fini(h);
5058         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5059
5060         EXIT;
5061 }
5062
5063 static int mdt_object_print(const struct lu_env *env, void *cookie,
5064                             lu_printer_t p, const struct lu_object *o)
5065 {
5066         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5067         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
5068                     "flags="LPX64", epochcount=%d, writecount=%d)",
5069                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
5070                     mdto->mot_ioepoch_count, mdto->mot_writecount);
5071 }
5072
5073 static int mdt_prepare(const struct lu_env *env,
5074                 struct lu_device *pdev,
5075                 struct lu_device *cdev)
5076 {
5077         struct mdt_device *mdt = mdt_dev(cdev);
5078         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5079         struct obd_device *obd = cdev->ld_obd;
5080         struct lfsck_start_param lsp;
5081         int rc;
5082
5083         ENTRY;
5084
5085         LASSERT(obd);
5086
5087         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5088         if (rc)
5089                 RETURN(rc);
5090
5091         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5092         if (rc)
5093                 RETURN(rc);
5094
5095         lsp.lsp_start = NULL;
5096         lsp.lsp_namespace = mdt->mdt_namespace;
5097         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5098                                                    OBD_IOC_START_LFSCK,
5099                                                    0, &lsp);
5100         if (rc != 0) {
5101                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5102                       mdt_obd_name(mdt), rc);
5103                 rc = 0;
5104         }
5105
5106         if (mdt->mdt_seq_site.ss_node_id == 0) {
5107                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5108                                                          &mdt->mdt_md_root_fid);
5109                 if (rc)
5110                         RETURN(rc);
5111         }
5112         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5113         target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
5114         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5115         LASSERT(obd->obd_no_conn);
5116         spin_lock(&obd->obd_dev_lock);
5117         obd->obd_no_conn = 0;
5118         spin_unlock(&obd->obd_dev_lock);
5119
5120         if (obd->obd_recovering == 0)
5121                 mdt_postrecov(env, mdt);
5122
5123         RETURN(rc);
5124 }
5125
5126 const struct lu_device_operations mdt_lu_ops = {
5127         .ldo_object_alloc   = mdt_object_alloc,
5128         .ldo_process_config = mdt_process_config,
5129         .ldo_prepare        = mdt_prepare,
5130 };
5131
5132 static const struct lu_object_operations mdt_obj_ops = {
5133         .loo_object_init    = mdt_object_init,
5134         .loo_object_free    = mdt_object_free,
5135         .loo_object_print   = mdt_object_print
5136 };
5137
5138 static int mdt_obd_set_info_async(const struct lu_env *env,
5139                                   struct obd_export *exp,
5140                                   __u32 keylen, void *key,
5141                                   __u32 vallen, void *val,
5142                                   struct ptlrpc_request_set *set)
5143 {
5144         struct obd_device     *obd = exp->exp_obd;
5145         int                    rc;
5146         ENTRY;
5147
5148         LASSERT(obd);
5149
5150         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5151                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
5152                 RETURN(rc);
5153         }
5154
5155         RETURN(0);
5156 }
5157
5158 /**
5159  * Match client and server connection feature flags.
5160  *
5161  * Compute the compatibility flags for a connection request based on
5162  * features mutually supported by client and server.
5163  *
5164  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5165  * must not be updated here, otherwise a partially initialized value may
5166  * be exposed. After the connection request is successfully processed,
5167  * the top-level MDT connect request handler atomically updates the export
5168  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5169  * reply. \see mdt_connect().
5170  *
5171  * \param exp   the obd_export associated with this client/target pair
5172  * \param mdt   the target device for the connection
5173  * \param data  stores data for this connect request
5174  *
5175  * \retval 0       success
5176  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5177  * \retval -EBADE  client and server feature requirements are incompatible
5178  */
5179 static int mdt_connect_internal(struct obd_export *exp,
5180                                 struct mdt_device *mdt,
5181                                 struct obd_connect_data *data)
5182 {
5183         LASSERT(data != NULL);
5184
5185         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5186         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5187
5188         /* If no known bits (which should not happen, probably,
5189            as everybody should support LOOKUP and UPDATE bits at least)
5190            revert to compat mode with plain locks. */
5191         if (!data->ocd_ibits_known &&
5192             data->ocd_connect_flags & OBD_CONNECT_IBITS)
5193                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
5194
5195         if (!mdt->mdt_opts.mo_acl)
5196                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5197
5198         if (!mdt->mdt_opts.mo_user_xattr)
5199                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5200
5201         if (!mdt->mdt_som_conf)
5202                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
5203
5204         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5205                 data->ocd_brw_size = min(data->ocd_brw_size,
5206                                          (__u32)MD_MAX_BRW_SIZE);
5207                 if (data->ocd_brw_size == 0) {
5208                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5209                                " ocd_version: %x ocd_grant: %d "
5210                                "ocd_index: %u ocd_brw_size is "
5211                                "unexpectedly zero, network data "
5212                                "corruption? Refusing connection of this"
5213                                " client\n",
5214                                mdt_obd_name(mdt),
5215                                exp->exp_client_uuid.uuid,
5216                                exp, data->ocd_connect_flags, data->ocd_version,
5217                                data->ocd_grant, data->ocd_index);
5218                         return -EPROTO;
5219                 }
5220         }
5221
5222         /* NB: Disregard the rule against updating
5223          * exp_connect_data.ocd_connect_flags in this case, since
5224          * tgt_client_new() needs to know if this is a lightweight
5225          * connection, and it is safe to expose this flag before
5226          * connection processing completes. */
5227         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5228                 spin_lock(&exp->exp_lock);
5229                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5230                 spin_unlock(&exp->exp_lock);
5231         }
5232
5233         data->ocd_version = LUSTRE_VERSION_CODE;
5234         exp->exp_connect_data = *data;
5235         exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
5236
5237         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5238                 CWARN("%s: MDS requires FID support, but client not\n",
5239                       mdt_obd_name(mdt));
5240                 return -EBADE;
5241         }
5242
5243         if (mdt->mdt_som_conf &&
5244             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5245                                          OBD_CONNECT_MDS_MDS |
5246                                          OBD_CONNECT_SOM))) {
5247                 CWARN("%s: MDS has SOM enabled, but client does not support "
5248                       "it\n", mdt_obd_name(mdt));
5249                 return -EBADE;
5250         }
5251
5252         if (OCD_HAS_FLAG(data, PINGLESS)) {
5253                 if (ptlrpc_pinger_suppress_pings()) {
5254                         spin_lock(&exp->exp_obd->obd_dev_lock);
5255                         list_del_init(&exp->exp_obd_chain_timed);
5256                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5257                 } else {
5258                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5259                 }
5260         }
5261
5262         return 0;
5263 }
5264
5265 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
5266                                      struct obd_export *exp,
5267                                      struct ptlrpc_request *req)
5268 {
5269         struct sptlrpc_flavor   flvr;
5270         int                     rc = 0;
5271
5272         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
5273                 read_lock(&mdt->mdt_sptlrpc_lock);
5274                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
5275                                              req->rq_sp_from,
5276                                              req->rq_peer.nid,
5277                                              &flvr);
5278                 read_unlock(&mdt->mdt_sptlrpc_lock);
5279
5280                 spin_lock(&exp->exp_lock);
5281
5282                 exp->exp_sp_peer = req->rq_sp_from;
5283                 exp->exp_flvr = flvr;
5284
5285                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
5286                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
5287                         CERROR("unauthorized rpc flavor %x from %s, "
5288                                "expect %x\n", req->rq_flvr.sf_rpc,
5289                                libcfs_nid2str(req->rq_peer.nid),
5290                                exp->exp_flvr.sf_rpc);
5291                         rc = -EACCES;
5292                 }
5293
5294                 spin_unlock(&exp->exp_lock);
5295         } else {
5296                 if (exp->exp_sp_peer != req->rq_sp_from) {
5297                         CERROR("RPC source %s doesn't match %s\n",
5298                                sptlrpc_part2name(req->rq_sp_from),
5299                                sptlrpc_part2name(exp->exp_sp_peer));
5300                         rc = -EACCES;
5301                 } else {
5302                         rc = sptlrpc_target_export_check(exp, req);
5303                 }
5304         }
5305
5306         return rc;
5307 }
5308
5309 /* mds_connect copy */
5310 static int mdt_obd_connect(const struct lu_env *env,
5311                            struct obd_export **exp, struct obd_device *obd,
5312                            struct obd_uuid *cluuid,
5313                            struct obd_connect_data *data,
5314                            void *localdata)
5315 {
5316         struct mdt_thread_info *info;
5317         struct obd_export      *lexp;
5318         struct lustre_handle    conn = { 0 };
5319         struct mdt_device      *mdt;
5320         struct ptlrpc_request  *req;
5321         int                     rc;
5322         ENTRY;
5323
5324         LASSERT(env != NULL);
5325         if (!exp || !obd || !cluuid)
5326                 RETURN(-EINVAL);
5327
5328         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5329         req = info->mti_pill->rc_req;
5330         mdt = mdt_dev(obd->obd_lu_dev);
5331
5332         /*
5333          * first, check whether the stack is ready to handle requests
5334          * XXX: probably not very appropriate method is used now
5335          *      at some point we should find a better one
5336          */
5337         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5338             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5339                 rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd);
5340                 if (rc)
5341                         RETURN(-EAGAIN);
5342                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5343         }
5344
5345         rc = class_connect(&conn, obd, cluuid);
5346         if (rc)
5347                 RETURN(rc);
5348
5349         lexp = class_conn2export(&conn);
5350         LASSERT(lexp != NULL);
5351
5352         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5353         if (rc)
5354                 GOTO(out, rc);
5355
5356         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5357                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5358
5359         rc = mdt_connect_internal(lexp, mdt, data);
5360         if (rc == 0) {
5361                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5362
5363                 LASSERT(lcd);
5364                 info->mti_exp = lexp;
5365                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5366                 rc = tgt_client_new(env, lexp);
5367                 if (rc == 0)
5368                         mdt_export_stats_init(obd, lexp, localdata);
5369         }
5370
5371 out:
5372         if (rc != 0) {
5373                 class_disconnect(lexp);
5374                 *exp = NULL;
5375         } else {
5376                 *exp = lexp;
5377         }
5378
5379         RETURN(rc);
5380 }
5381
5382 static int mdt_obd_reconnect(const struct lu_env *env,
5383                              struct obd_export *exp, struct obd_device *obd,
5384                              struct obd_uuid *cluuid,
5385                              struct obd_connect_data *data,
5386                              void *localdata)
5387 {
5388         struct mdt_thread_info *info;
5389         struct mdt_device      *mdt;
5390         struct ptlrpc_request  *req;
5391         int                     rc;
5392         ENTRY;
5393
5394         if (exp == NULL || obd == NULL || cluuid == NULL)
5395                 RETURN(-EINVAL);
5396
5397         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5398         req = info->mti_pill->rc_req;
5399         mdt = mdt_dev(obd->obd_lu_dev);
5400
5401         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5402         if (rc)
5403                 RETURN(rc);
5404
5405         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5406         if (rc == 0)
5407                 mdt_export_stats_init(obd, exp, localdata);
5408
5409         RETURN(rc);
5410 }
5411
5412 static int mdt_export_cleanup(struct obd_export *exp)
5413 {
5414         struct mdt_export_data *med = &exp->exp_mdt_data;
5415         struct obd_device      *obd = exp->exp_obd;
5416         struct mdt_device      *mdt;
5417         struct mdt_thread_info *info;
5418         struct lu_env           env;
5419         CFS_LIST_HEAD(closing_list);
5420         struct mdt_file_data *mfd, *n;
5421         int rc = 0;
5422         ENTRY;
5423
5424         spin_lock(&med->med_open_lock);
5425         while (!cfs_list_empty(&med->med_open_head)) {
5426                 cfs_list_t *tmp = med->med_open_head.next;
5427                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5428
5429                 /* Remove mfd handle so it can't be found again.
5430                  * We are consuming the mfd_list reference here. */
5431                 class_handle_unhash(&mfd->mfd_handle);
5432                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5433         }
5434         spin_unlock(&med->med_open_lock);
5435         mdt = mdt_dev(obd->obd_lu_dev);
5436         LASSERT(mdt != NULL);
5437
5438         rc = lu_env_init(&env, LCT_MD_THREAD);
5439         if (rc)
5440                 RETURN(rc);
5441
5442         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5443         LASSERT(info != NULL);
5444         memset(info, 0, sizeof *info);
5445         info->mti_env = &env;
5446         info->mti_mdt = mdt;
5447         info->mti_exp = exp;
5448
5449         if (!cfs_list_empty(&closing_list)) {
5450                 struct md_attr *ma = &info->mti_attr;
5451
5452                 /* Close any open files (which may also cause orphan unlinking). */
5453                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5454                         cfs_list_del_init(&mfd->mfd_list);
5455                         ma->ma_need = ma->ma_valid = 0;
5456                         /* Don't unlink orphan on failover umount, LU-184 */
5457                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5458                                 ma->ma_valid = MA_FLAGS;
5459                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5460                         }
5461                         mdt_mfd_close(info, mfd);
5462                 }
5463         }
5464         info->mti_mdt = NULL;
5465         /* cleanup client slot early */
5466         /* Do not erase record for recoverable client. */
5467         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5468                 tgt_client_del(&env, exp);
5469         lu_env_fini(&env);
5470
5471         RETURN(rc);
5472 }
5473
5474 static int mdt_obd_disconnect(struct obd_export *exp)
5475 {
5476         int rc;
5477         ENTRY;
5478
5479         LASSERT(exp);
5480         class_export_get(exp);
5481
5482         rc = server_disconnect_export(exp);
5483         if (rc != 0)
5484                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5485
5486         rc = mdt_export_cleanup(exp);
5487         class_export_put(exp);
5488         RETURN(rc);
5489 }
5490
5491 /* FIXME: Can we avoid using these two interfaces? */
5492 static int mdt_init_export(struct obd_export *exp)
5493 {
5494         struct mdt_export_data *med = &exp->exp_mdt_data;
5495         int                     rc;
5496         ENTRY;
5497
5498         CFS_INIT_LIST_HEAD(&med->med_open_head);
5499         spin_lock_init(&med->med_open_lock);
5500         mutex_init(&med->med_idmap_mutex);
5501         med->med_idmap = NULL;
5502         spin_lock(&exp->exp_lock);
5503         exp->exp_connecting = 1;
5504         spin_unlock(&exp->exp_lock);
5505
5506         /* self-export doesn't need client data and ldlm initialization */
5507         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5508                                      &exp->exp_client_uuid)))
5509                 RETURN(0);
5510
5511         rc = tgt_client_alloc(exp);
5512         if (rc)
5513                 GOTO(err, rc);
5514
5515         rc = ldlm_init_export(exp);
5516         if (rc)
5517                 GOTO(err_free, rc);
5518
5519         RETURN(rc);
5520
5521 err_free:
5522         tgt_client_free(exp);
5523 err:
5524         CERROR("%s: Failed to initialize export: rc = %d\n",
5525                exp->exp_obd->obd_name, rc);
5526         return rc;
5527 }
5528
5529 static int mdt_destroy_export(struct obd_export *exp)
5530 {
5531         ENTRY;
5532
5533         if (exp_connect_rmtclient(exp))
5534                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5535
5536         target_destroy_export(exp);
5537         /* destroy can be called from failed obd_setup, so
5538          * checking uuid is safer than obd_self_export */
5539         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5540                                      &exp->exp_client_uuid)))
5541                 RETURN(0);
5542
5543         ldlm_destroy_export(exp);
5544         tgt_client_free(exp);
5545
5546         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5547         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5548
5549         RETURN(0);
5550 }
5551
5552 /** The maximum depth that fid2path() will search.
5553  * This is limited only because we want to store the fids for
5554  * historical path lookup purposes.
5555  */
5556 #define MAX_PATH_DEPTH 100
5557
5558 /** mdt_path() lookup structure. */
5559 struct path_lookup_info {
5560         __u64                   pli_recno;      /**< history point */
5561         __u64                   pli_currec;     /**< current record */
5562         struct lu_fid           pli_fid;
5563         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5564         struct mdt_object       *pli_mdt_obj;
5565         char                    *pli_path;      /**< full path */
5566         int                     pli_pathlen;
5567         int                     pli_linkno;     /**< which hardlink to follow */
5568         int                     pli_fidcount;   /**< number of \a pli_fids */
5569 };
5570
5571 static int mdt_links_read(struct mdt_thread_info *info,
5572                           struct mdt_object *mdt_obj, struct linkea_data *ldata)
5573 {
5574         int rc;
5575
5576         LASSERT(ldata->ld_buf->lb_buf != NULL);
5577
5578         if (!mdt_object_exists(mdt_obj))
5579                 return -ENODATA;
5580
5581         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5582                           ldata->ld_buf, XATTR_NAME_LINK);
5583         if (rc == -ERANGE) {
5584                 /* Buf was too small, figure out what we need. */
5585                 lu_buf_free(ldata->ld_buf);
5586                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5587                                   ldata->ld_buf, XATTR_NAME_LINK);
5588                 if (rc < 0)
5589                         return rc;
5590                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5591                 if (ldata->ld_buf->lb_buf == NULL)
5592                         return -ENOMEM;
5593                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5594                                   ldata->ld_buf, XATTR_NAME_LINK);
5595         }
5596         if (rc < 0)
5597                 return rc;
5598
5599         return linkea_init(ldata);
5600 }
5601
5602 static int mdt_path_current(struct mdt_thread_info *info,
5603                             struct path_lookup_info *pli)
5604 {
5605         struct mdt_device       *mdt = info->mti_mdt;
5606         struct mdt_object       *mdt_obj;
5607         struct link_ea_header   *leh;
5608         struct link_ea_entry    *lee;
5609         struct lu_name          *tmpname = &info->mti_name;
5610         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5611         struct lu_buf           *buf = &info->mti_big_buf;
5612         char                    *ptr;
5613         int                     reclen;
5614         struct linkea_data      ldata = { 0 };
5615         int                     rc = 0;
5616         ENTRY;
5617
5618         /* temp buffer for path element, the buffer will be finally freed
5619          * in mdt_thread_info_fini */
5620         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5621         if (buf->lb_buf == NULL)
5622                 RETURN(-ENOMEM);
5623
5624         ldata.ld_buf = buf;
5625         ptr = pli->pli_path + pli->pli_pathlen - 1;
5626         *ptr = 0;
5627         --ptr;
5628         pli->pli_fidcount = 0;
5629         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5630
5631         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5632          * so checking root_fid can only happen on MDT0. */
5633         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5634                           &pli->pli_fids[pli->pli_fidcount])) {
5635                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5636                                           &pli->pli_fids[pli->pli_fidcount]);
5637                 if (IS_ERR(mdt_obj))
5638                         GOTO(out, rc = PTR_ERR(mdt_obj));
5639                 if (mdt_object_remote(mdt_obj)) {
5640                         mdt_object_put(info->mti_env, mdt_obj);
5641                         GOTO(remote_out, rc = -EREMOTE);
5642                 }
5643                 if (!mdt_object_exists(mdt_obj)) {
5644                         mdt_object_put(info->mti_env, mdt_obj);
5645                         GOTO(out, rc = -ENOENT);
5646                 }
5647
5648                 rc = mdt_links_read(info, mdt_obj, &ldata);
5649                 mdt_object_put(info->mti_env, mdt_obj);
5650                 if (rc != 0)
5651                         GOTO(out, rc);
5652
5653                 leh = buf->lb_buf;
5654                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5655                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5656                 /* If set, use link #linkno for path lookup, otherwise use
5657                    link #0.  Only do this for the final path element. */
5658                 if ((pli->pli_fidcount == 0) &&
5659                     (pli->pli_linkno < leh->leh_reccount)) {
5660                         int count;
5661                         for (count = 0; count < pli->pli_linkno; count++) {
5662                                 lee = (struct link_ea_entry *)
5663                                      ((char *)lee + reclen);
5664                                 linkea_entry_unpack(lee, &reclen, tmpname,
5665                                                     tmpfid);
5666                         }
5667                         if (pli->pli_linkno < leh->leh_reccount - 1)
5668                                 /* indicate to user there are more links */
5669                                 pli->pli_linkno++;
5670                 }
5671
5672                 /* Pack the name in the end of the buffer */
5673                 ptr -= tmpname->ln_namelen;
5674                 if (ptr - 1 <= pli->pli_path)
5675                         GOTO(out, rc = -EOVERFLOW);
5676                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5677                 *(--ptr) = '/';
5678
5679                 /* Store the parent fid for historic lookup */
5680                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5681                         GOTO(out, rc = -EOVERFLOW);
5682                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5683         }
5684
5685 remote_out:
5686         ptr++; /* skip leading / */
5687         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5688
5689         EXIT;
5690 out:
5691         return rc;
5692 }
5693
5694 /* Returns the full path to this fid, as of changelog record recno. */
5695 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5696                     char *path, int pathlen, __u64 *recno, int *linkno,
5697                     struct lu_fid *fid)
5698 {
5699         struct mdt_device       *mdt = info->mti_mdt;
5700         struct path_lookup_info *pli;
5701         int                     tries = 3;
5702         int                     rc = -EAGAIN;
5703         ENTRY;
5704
5705         if (pathlen < 3)
5706                 RETURN(-EOVERFLOW);
5707
5708         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5709                 path[0] = '\0';
5710                 RETURN(0);
5711         }
5712
5713         OBD_ALLOC_PTR(pli);
5714         if (pli == NULL)
5715                 RETURN(-ENOMEM);
5716
5717         pli->pli_mdt_obj = obj;
5718         pli->pli_recno = *recno;
5719         pli->pli_path = path;
5720         pli->pli_pathlen = pathlen;
5721         pli->pli_linkno = *linkno;
5722
5723         /* Retry multiple times in case file is being moved */
5724         while (tries-- && rc == -EAGAIN)
5725                 rc = mdt_path_current(info, pli);
5726
5727         /* return the last resolved fids to the client, so the client will
5728          * build the left path on another MDT for remote object */
5729         *fid = pli->pli_fids[pli->pli_fidcount];
5730
5731         *recno = pli->pli_currec;
5732         /* Return next link index to caller */
5733         *linkno = pli->pli_linkno;
5734
5735         OBD_FREE_PTR(pli);
5736
5737         RETURN(rc);
5738 }
5739
5740 static int mdt_fid2path(struct mdt_thread_info *info,
5741                         struct getinfo_fid2path *fp)
5742 {
5743         struct mdt_device *mdt = info->mti_mdt;
5744         struct mdt_object *obj;
5745         int    rc;
5746         ENTRY;
5747
5748         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5749                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5750
5751         if (!fid_is_sane(&fp->gf_fid))
5752                 RETURN(-EINVAL);
5753
5754         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5755                 CWARN("%s: "DFID" is invalid, sequence should be "
5756                       ">= "LPX64"\n", mdt_obd_name(mdt),
5757                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5758                 RETURN(-EINVAL);
5759         }
5760
5761         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5762         if (obj == NULL || IS_ERR(obj)) {
5763                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5764                        PTR_ERR(obj));
5765                 RETURN(-EINVAL);
5766         }
5767
5768         if (mdt_object_remote(obj))
5769                 rc = -EREMOTE;
5770         else if (!mdt_object_exists(obj))
5771                 rc = -ENOENT;
5772         else
5773                 rc = 0;
5774
5775         if (rc < 0) {
5776                 mdt_object_put(info->mti_env, obj);
5777                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5778                        PFID(&fp->gf_fid), rc);
5779                 RETURN(rc);
5780         }
5781
5782         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5783                       &fp->gf_linkno, &fp->gf_fid);
5784
5785         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5786                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5787
5788         mdt_object_put(info->mti_env, obj);
5789
5790         RETURN(rc);
5791 }
5792
5793 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5794                             void *val, int vallen)
5795 {
5796         struct getinfo_fid2path *fpout, *fpin;
5797         int rc = 0;
5798
5799         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5800         fpout = val;
5801
5802         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5803                 lustre_swab_fid2path(fpin);
5804
5805         memcpy(fpout, fpin, sizeof(*fpin));
5806         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5807                 RETURN(-EINVAL);
5808
5809         rc = mdt_fid2path(info, fpout);
5810         RETURN(rc);
5811 }
5812
5813 int mdt_get_info(struct mdt_thread_info *info)
5814 {
5815         struct ptlrpc_request *req = mdt_info_req(info);
5816         char *key;
5817         int keylen;
5818         __u32 *vallen;
5819         void *valout;
5820         int rc;
5821         ENTRY;
5822
5823         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5824         if (key == NULL) {
5825                 CDEBUG(D_IOCTL, "No GETINFO key");
5826                 RETURN(-EFAULT);
5827         }
5828         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5829                                       RCL_CLIENT);
5830
5831         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5832         if (vallen == NULL) {
5833                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5834                 RETURN(-EFAULT);
5835         }
5836
5837         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5838                              *vallen);
5839         rc = req_capsule_server_pack(info->mti_pill);
5840         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5841         if (valout == NULL) {
5842                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5843                 RETURN(-EFAULT);
5844         }
5845
5846         if (KEY_IS(KEY_FID2PATH))
5847                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5848         else
5849                 rc = -EINVAL;
5850
5851         lustre_msg_set_status(req->rq_repmsg, rc);
5852
5853         RETURN(rc);
5854 }
5855
5856 /* Pass the ioc down */
5857 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5858                          unsigned int cmd, int len, void *data)
5859 {
5860         struct lu_context ioctl_session;
5861         struct md_device *next = mdt->mdt_child;
5862         int rc;
5863         ENTRY;
5864
5865         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5866         if (rc)
5867                 RETURN(rc);
5868         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5869         lu_context_enter(&ioctl_session);
5870         env->le_ses = &ioctl_session;
5871
5872         LASSERT(next->md_ops->mdo_iocontrol);
5873         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5874
5875         lu_context_exit(&ioctl_session);
5876         lu_context_fini(&ioctl_session);
5877         RETURN(rc);
5878 }
5879
5880 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5881 {
5882         struct obd_ioctl_data *data = karg;
5883         struct lu_fid *fid;
5884         __u64 version;
5885         struct mdt_object *obj;
5886         struct mdt_lock_handle  *lh;
5887         int rc;
5888         ENTRY;
5889
5890         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5891             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5892                 RETURN(-EINVAL);
5893
5894         fid = (struct lu_fid *)data->ioc_inlbuf1;
5895
5896         if (!fid_is_sane(fid))
5897                 RETURN(-EINVAL);
5898
5899         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5900
5901         lh = &mti->mti_lh[MDT_LH_PARENT];
5902         mdt_lock_reg_init(lh, LCK_CR);
5903
5904         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5905         if (IS_ERR(obj))
5906                 RETURN(PTR_ERR(obj));
5907
5908         if (mdt_object_remote(obj)) {
5909                 rc = -EREMOTE;
5910                 /**
5911                  * before calling version get the correct MDS should be
5912                  * fid, this is error to find remote object here
5913                  */
5914                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5915         } else if (!mdt_object_exists(obj)) {
5916                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5917                 rc = -ENOENT;
5918         } else {
5919                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5920                *(__u64 *)data->ioc_inlbuf2 = version;
5921                 rc = 0;
5922         }
5923         mdt_object_unlock_put(mti, obj, lh, 1);
5924         RETURN(rc);
5925 }
5926
5927 /* ioctls on obd dev */
5928 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5929                          void *karg, void *uarg)
5930 {
5931         struct lu_env      env;
5932         struct obd_device *obd = exp->exp_obd;
5933         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5934         struct dt_device  *dt = mdt->mdt_bottom;
5935         int rc;
5936
5937         ENTRY;
5938         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5939         rc = lu_env_init(&env, LCT_MD_THREAD);
5940         if (rc)
5941                 RETURN(rc);
5942
5943         switch (cmd) {
5944         case OBD_IOC_SYNC:
5945                 rc = mdt_device_sync(&env, mdt);
5946                 break;
5947         case OBD_IOC_SET_READONLY:
5948                 rc = dt->dd_ops->dt_ro(&env, dt);
5949                 break;
5950         case OBD_IOC_ABORT_RECOVERY:
5951                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5952                 target_stop_recovery_thread(obd);
5953                 rc = 0;
5954                 break;
5955         case OBD_IOC_CHANGELOG_REG:
5956         case OBD_IOC_CHANGELOG_DEREG:
5957         case OBD_IOC_CHANGELOG_CLEAR:
5958                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5959                 break;
5960         case OBD_IOC_START_LFSCK: {
5961                 struct md_device *next = mdt->mdt_child;
5962                 struct obd_ioctl_data *data = karg;
5963                 struct lfsck_start_param lsp;
5964
5965                 if (unlikely(data == NULL)) {
5966                         rc = -EINVAL;
5967                         break;
5968                 }
5969
5970                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5971                 lsp.lsp_namespace = mdt->mdt_namespace;
5972                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5973                 break;
5974         }
5975         case OBD_IOC_STOP_LFSCK: {
5976                 struct md_device *next = mdt->mdt_child;
5977
5978                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
5979                 break;
5980         }
5981         case OBD_IOC_GET_OBJ_VERSION: {
5982                 struct mdt_thread_info *mti;
5983                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5984                 memset(mti, 0, sizeof *mti);
5985                 mti->mti_env = &env;
5986                 mti->mti_mdt = mdt;
5987                 mti->mti_exp = exp;
5988
5989                 rc = mdt_ioc_version_get(mti, karg);
5990                 break;
5991         }
5992         default:
5993                 rc = -EOPNOTSUPP;
5994                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5995                         mdt_obd_name(mdt), cmd, rc);
5996         }
5997
5998         lu_env_fini(&env);
5999         RETURN(rc);
6000 }
6001
6002 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6003 {
6004         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6005         int rc;
6006         ENTRY;
6007
6008         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6009         RETURN(rc);
6010 }
6011
6012 int mdt_obd_postrecov(struct obd_device *obd)
6013 {
6014         struct lu_env env;
6015         int rc;
6016
6017         rc = lu_env_init(&env, LCT_MD_THREAD);
6018         if (rc)
6019                 RETURN(rc);
6020         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6021         lu_env_fini(&env);
6022         return rc;
6023 }
6024
6025 static struct obd_ops mdt_obd_device_ops = {
6026         .o_owner          = THIS_MODULE,
6027         .o_set_info_async = mdt_obd_set_info_async,
6028         .o_connect        = mdt_obd_connect,
6029         .o_reconnect      = mdt_obd_reconnect,
6030         .o_disconnect     = mdt_obd_disconnect,
6031         .o_init_export    = mdt_init_export,
6032         .o_destroy_export = mdt_destroy_export,
6033         .o_iocontrol      = mdt_iocontrol,
6034         .o_postrecov      = mdt_obd_postrecov,
6035 };
6036
6037 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6038                                          struct lu_device *d)
6039 {
6040         struct mdt_device *m = mdt_dev(d);
6041         ENTRY;
6042
6043         mdt_fini(env, m);
6044         RETURN(NULL);
6045 }
6046
6047 static struct lu_device *mdt_device_free(const struct lu_env *env,
6048                                          struct lu_device *d)
6049 {
6050         struct mdt_device *m = mdt_dev(d);
6051         ENTRY;
6052
6053         lu_device_fini(&m->mdt_lu_dev);
6054         OBD_FREE_PTR(m);
6055
6056         RETURN(NULL);
6057 }
6058
6059 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6060                                           struct lu_device_type *t,
6061                                           struct lustre_cfg *cfg)
6062 {
6063         struct lu_device  *l;
6064         struct mdt_device *m;
6065
6066         OBD_ALLOC_PTR(m);
6067         if (m != NULL) {
6068                 int rc;
6069
6070                 l = &m->mdt_lu_dev;
6071                 rc = mdt_init0(env, m, t, cfg);
6072                 if (rc != 0) {
6073                         mdt_device_free(env, l);
6074                         l = ERR_PTR(rc);
6075                         return l;
6076                 }
6077         } else
6078                 l = ERR_PTR(-ENOMEM);
6079         return l;
6080 }
6081
6082 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6083 LU_KEY_INIT(mdt, struct mdt_thread_info);
6084
6085 static void mdt_key_fini(const struct lu_context *ctx,
6086                          struct lu_context_key *key, void* data)
6087 {
6088         struct mdt_thread_info *info = data;
6089
6090         if (info->mti_big_lmm) {
6091                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6092                 info->mti_big_lmm = NULL;
6093                 info->mti_big_lmmsize = 0;
6094         }
6095         OBD_FREE_PTR(info);
6096 }
6097
6098 /* context key: mdt_thread_key */
6099 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6100
6101 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6102 {
6103         return lu_ucred(info->mti_env);
6104 }
6105
6106 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6107 {
6108         return lu_ucred_check(info->mti_env);
6109 }
6110
6111 /**
6112  * Enable/disable COS (Commit On Sharing).
6113  *
6114  * Set/Clear the COS flag in mdt options.
6115  *
6116  * \param mdt mdt device
6117  * \param val 0 disables COS, other values enable COS
6118  */
6119 void mdt_enable_cos(struct mdt_device *mdt, int val)
6120 {
6121         struct lu_env env;
6122         int rc;
6123
6124         mdt->mdt_opts.mo_cos = !!val;
6125         rc = lu_env_init(&env, LCT_LOCAL);
6126         if (unlikely(rc != 0)) {
6127                 CWARN("lu_env initialization failed with rc = %d,"
6128                       "cannot sync\n", rc);
6129                 return;
6130         }
6131         mdt_device_sync(&env, mdt);
6132         lu_env_fini(&env);
6133 }
6134
6135 /**
6136  * Check COS (Commit On Sharing) status.
6137  *
6138  * Return COS flag status.
6139  *
6140  * \param mdt mdt device
6141  */
6142 int mdt_cos_is_enabled(struct mdt_device *mdt)
6143 {
6144         return mdt->mdt_opts.mo_cos != 0;
6145 }
6146
6147 static struct lu_device_type_operations mdt_device_type_ops = {
6148         .ldto_device_alloc = mdt_device_alloc,
6149         .ldto_device_free  = mdt_device_free,
6150         .ldto_device_fini  = mdt_device_fini
6151 };
6152
6153 static struct lu_device_type mdt_device_type = {
6154         .ldt_tags     = LU_DEVICE_MD,
6155         .ldt_name     = LUSTRE_MDT_NAME,
6156         .ldt_ops      = &mdt_device_type_ops,
6157         .ldt_ctx_tags = LCT_MD_THREAD
6158 };
6159
6160 static int __init mdt_mod_init(void)
6161 {
6162         struct lprocfs_static_vars lvars;
6163         int rc;
6164
6165         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6166                  FID_NOBRACE_LEN + 1);
6167         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6168                  FID_LEN + 1);
6169         rc = lu_kmem_init(mdt_caches);
6170         if (rc)
6171                 return rc;
6172
6173         rc = mds_mod_init();
6174         if (rc)
6175                 GOTO(lu_fini, rc);
6176
6177         lprocfs_mdt_init_vars(&lvars);
6178         rc = class_register_type(&mdt_obd_device_ops, NULL,
6179                                  lvars.module_vars, LUSTRE_MDT_NAME,
6180                                  &mdt_device_type);
6181         if (rc)
6182                 GOTO(mds_fini, rc);
6183 lu_fini:
6184         if (rc)
6185                 lu_kmem_fini(mdt_caches);
6186 mds_fini:
6187         if (rc)
6188                 mds_mod_exit();
6189         return rc;
6190 }
6191
6192 static void __exit mdt_mod_exit(void)
6193 {
6194         class_unregister_type(LUSTRE_MDT_NAME);
6195         mds_mod_exit();
6196         lu_kmem_fini(mdt_caches);
6197 }
6198
6199 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6200 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6201 MODULE_LICENSE("GPL");
6202
6203 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);