lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/mdt/mdt_handler.c
   5  *  Lustre Metadata Target (mdt) request handler
   6  *
   7  *  Copyright (c) 2006 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *   Author: Mike Shaver <shaver@clusterfs.com>
  12  *   Author: Nikita Danilov <nikita@clusterfs.com>
  13  *   Author: Huang Hua <huanghua@clusterfs.com>
  14  *   Author: Yury Umanets <umka@clusterfs.com>
  15  *
  16  *   This file is part of the Lustre file system, http://www.lustre.org
  17  *   Lustre is a trademark of Cluster File Systems, Inc.
  18  *
  19  *   You may have signed or agreed to another license before downloading
  20  *   this software.  If so, you are bound by the terms and conditions
  21  *   of that agreement, and the following does not apply to you.  See the
  22  *   LICENSE file included with this distribution for more information.
  23  *
  24  *   If you did not agree to a different license, then this copy of Lustre
  25  *   is open source software; you can redistribute it and/or modify it
  26  *   under the terms of version 2 of the GNU General Public License as
  27  *   published by the Free Software Foundation.
  28  *
  29  *   In either case, Lustre is distributed in the hope that it will be
  30  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  31  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  32  *   license text for more details.
  33  */
  34
  35 #ifndef EXPORT_SYMTAB
  36 # define EXPORT_SYMTAB
  37 #endif
  38 #define DEBUG_SUBSYSTEM S_MDS
  39
  40 #include <linux/module.h>
  41 /*
  42  * struct OBD_{ALLOC,FREE}*()
  43  */
  44 #include <obd_support.h>
  45 /* struct ptlrpc_request */
  46 #include <lustre_net.h>
  47 /* struct obd_export */
  48 #include <lustre_export.h>
  49 /* struct obd_device */
  50 #include <obd.h>
  51 /* lu2dt_dev() */
  52 #include <dt_object.h>
  53 #include <lustre_mds.h>
  54 #include <lustre_mdt.h>
  55 #include "mdt_internal.h"
  56 #include <linux/lustre_acl.h>
  57 #include <lustre_param.h>
  58
  59 mdl_mode_t mdt_mdl_lock_modes[] = {
  60         [LCK_MINMODE] = MDL_MINMODE,
  61         [LCK_EX]      = MDL_EX,
  62         [LCK_PW]      = MDL_PW,
  63         [LCK_PR]      = MDL_PR,
  64         [LCK_CW]      = MDL_CW,
  65         [LCK_CR]      = MDL_CR,
  66         [LCK_NL]      = MDL_NL,
  67         [LCK_GROUP]   = MDL_GROUP
  68 };
  69
  70 ldlm_mode_t mdt_dlm_lock_modes[] = {
  71         [MDL_MINMODE] = LCK_MINMODE,
  72         [MDL_EX]      = LCK_EX,
  73         [MDL_PW]      = LCK_PW,
  74         [MDL_PR]      = LCK_PR,
  75         [MDL_CW]      = LCK_CW,
  76         [MDL_CR]      = LCK_CR,
  77         [MDL_NL]      = LCK_NL,
  78         [MDL_GROUP]   = LCK_GROUP
  79 };
  80
  81 /*
  82  * Initialized in mdt_mod_init().
  83  */
  84 unsigned long mdt_num_threads;
  85
  86 /* ptlrpc request handler for MDT. All handlers are
  87  * grouped into several slices - struct mdt_opc_slice,
  88  * and stored in an array - mdt_handlers[].
  89  */
  90 struct mdt_handler {
  91         /* The name of this handler. */
  92         const char *mh_name;
  93         /* Fail id for this handler, checked at the beginning of this handler*/
  94         int         mh_fail_id;
  95         /* Operation code for this handler */
  96         __u32       mh_opc;
  97         /* flags are listed in enum mdt_handler_flags below. */
  98         __u32       mh_flags;
  99         /* The actual handler function to execute. */
 100         int (*mh_act)(struct mdt_thread_info *info);
 101         /* Request format for this request. */
 102         const struct req_format *mh_fmt;
 103 };
 104
 105 enum mdt_handler_flags {
 106         /*
 107          * struct mdt_body is passed in the incoming message, and object
 108          * identified by this fid exists on disk.
 109          *
 110          * "habeo corpus" == "I have a body"
 111          */
 112         HABEO_CORPUS = (1 << 0),
 113         /*
 114          * struct ldlm_request is passed in the incoming message.
 115          *
 116          * "habeo clavis" == "I have a key"
 117          */
 118         HABEO_CLAVIS = (1 << 1),
 119         /*
 120          * this request has fixed reply format, so that reply message can be
 121          * packed by generic code.
 122          *
 123          * "habeo refero" == "I have a reply"
 124          */
 125         HABEO_REFERO = (1 << 2),
 126         /*
 127          * this request will modify something, so check whether the filesystem
 128          * is readonly or not, then return -EROFS to client asap if necessary.
 129          *
 130          * "mutabor" == "I shall modify"
 131          */
 132         MUTABOR      = (1 << 3)
 133 };
 134
 135 struct mdt_opc_slice {
 136         __u32               mos_opc_start;
 137         int                 mos_opc_end;
 138         struct mdt_handler *mos_hs;
 139 };
 140
 141 static struct mdt_opc_slice mdt_regular_handlers[];
 142 static struct mdt_opc_slice mdt_readpage_handlers[];
 143 static struct mdt_opc_slice mdt_xmds_handlers[];
 144 static struct mdt_opc_slice mdt_seq_handlers[];
 145 static struct mdt_opc_slice mdt_fld_handlers[];
 146
 147 static struct mdt_device *mdt_dev(struct lu_device *d);
 148 static int mdt_regular_handle(struct ptlrpc_request *req);
 149 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 150
 151 static struct lu_object_operations mdt_obj_ops;
 152
 153 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 154 {
 155         if (!rep)
 156                 return 0;
 157         return (rep->lock_policy_res1 & flag);
 158 }
 159
 160 void mdt_clear_disposition(struct mdt_thread_info *info,
 161                            struct ldlm_reply *rep, int flag)
 162 {
 163         if (info)
 164                 info->mti_opdata &= ~flag;
 165         if (rep)
 166                 rep->lock_policy_res1 &= ~flag;
 167 }
 168
 169 void mdt_set_disposition(struct mdt_thread_info *info,
 170                          struct ldlm_reply *rep, int flag)
 171 {
 172         if (info)
 173                 info->mti_opdata |= flag;
 174         if (rep)
 175                 rep->lock_policy_res1 |= flag;
 176 }
 177
 178 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 179 {
 180         lh->mlh_pdo_hash = 0;
 181         lh->mlh_reg_mode = lm;
 182         lh->mlh_type = MDT_REG_LOCK;
 183 }
 184
 185 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 186                        const char *name, int namelen)
 187 {
 188         lh->mlh_reg_mode = lm;
 189         lh->mlh_type = MDT_PDO_LOCK;
 190
 191         if (name != NULL) {
 192                 LASSERT(namelen > 0);
 193                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 194         } else {
 195                 LASSERT(namelen == 0);
 196                 lh->mlh_pdo_hash = 0ull;
 197         }
 198 }
 199
 200 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 201                               struct mdt_lock_handle *lh)
 202 {
 203         mdl_mode_t mode;
 204         ENTRY;
 205
 206         /*
 207          * Any dir access needs couple of locks:
 208          *
 209          * 1) on part of dir we gonna take lookup/modify;
 210          *
 211          * 2) on whole dir to protect it from concurrent splitting and/or to
 212          * flush client's cache for readdir().
 213          *
 214          * so, for a given mode and object this routine decides what lock mode
 215          * to use for lock #2:
 216          *
 217          * 1) if caller's gonna lookup in dir then we need to protect dir from
 218          * being splitted only - LCK_CR
 219          *
 220          * 2) if caller's gonna modify dir then we need to protect dir from
 221          * being splitted and to flush cache - LCK_CW
 222          *
 223          * 3) if caller's gonna modify dir and that dir seems ready for
 224          * splitting then we need to protect it from any type of access
 225          * (lookup/modify/split) - LCK_EX --bzzz
 226          */
 227
 228         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 229         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 230
 231         /*
 232          * Ask underlaying level its opinion about preferable PDO lock mode
 233          * having access type passed as regular lock mode:
 234          *
 235          * - MDL_MINMODE means that lower layer does not want to specify lock
 236          * mode;
 237          *
 238          * - MDL_NL means that no PDO lock should be taken. This is used in some
 239          * cases. Say, for non-splittable directories no need to use PDO locks
 240          * at all.
 241          */
 242         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 243                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 244
 245         if (mode != MDL_MINMODE) {
 246                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 247         } else {
 248                 /*
 249                  * Lower layer does not want to specify locking mode. We do it
 250                  * our selves. No special protection is needed, just flush
 251                  * client's cache on modification and allow concurrent
 252                  * mondification.
 253                  */
 254                 switch (lh->mlh_reg_mode) {
 255                 case LCK_EX:
 256                         lh->mlh_pdo_mode = LCK_EX;
 257                         break;
 258                 case LCK_PR:
 259                         lh->mlh_pdo_mode = LCK_CR;
 260                         break;
 261                 case LCK_PW:
 262                         lh->mlh_pdo_mode = LCK_CW;
 263                         break;
 264                 default:
 265                         CERROR("Not expected lock type (0x%x)\n",
 266                                (int)lh->mlh_reg_mode);
 267                         LBUG();
 268                 }
 269         }
 270
 271         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 272         EXIT;
 273 }
 274
 275 static int mdt_getstatus(struct mdt_thread_info *info)
 276 {
 277         struct mdt_device *mdt  = info->mti_mdt;
 278         struct md_device  *next = mdt->mdt_child;
 279         struct mdt_body   *repbody;
 280         int                rc;
 281
 282         ENTRY;
 283
 284         rc = mdt_check_ucred(info);
 285         if (rc)
 286                 RETURN(err_serious(rc));
 287
 288         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 289                 RETURN(err_serious(-ENOMEM));
 290
 291         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 292         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 293         if (rc != 0)
 294                 RETURN(rc);
 295
 296         repbody->valid |= OBD_MD_FLID;
 297
 298         if (mdt->mdt_opts.mo_mds_capa) {
 299                 struct mdt_object  *root;
 300                 struct lustre_capa *capa;
 301
 302                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 303                 if (IS_ERR(root))
 304                         RETURN(PTR_ERR(root));
 305
 306                 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
 307                 LASSERT(capa);
 308                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 309
 310                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 311                                  0);
 312                 mdt_object_put(info->mti_env, root);
 313                 if (rc == 0)
 314                         repbody->valid |= OBD_MD_FLMDSCAPA;
 315         }
 316
 317         RETURN(rc);
 318 }
 319
 320 static int mdt_statfs(struct mdt_thread_info *info)
 321 {
 322         struct md_device  *next  = info->mti_mdt->mdt_child;
 323         struct obd_statfs *osfs;
 324         int                rc;
 325
 326         ENTRY;
 327
 328         /* This will trigger a watchdog timeout */
 329         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 330                          (MDT_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
 331
 332         rc = mdt_check_ucred(info);
 333         if (rc)
 334                 RETURN(err_serious(rc));
 335
 336         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 337                 rc = err_serious(-ENOMEM);
 338         } else {
 339                 osfs = req_capsule_server_get(&info->mti_pill,&RMF_OBD_STATFS);
 340                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 341                                               &info->mti_u.ksfs);
 342                 statfs_pack(osfs, &info->mti_u.ksfs);
 343         }
 344         RETURN(rc);
 345 }
 346
 347 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 348 {
 349         struct mdt_body *b;
 350         struct lu_attr *attr = &info->mti_attr.ma_attr;
 351
 352         b = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 353
 354         /* Check if Size-on-MDS is enabled. */
 355         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 356             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 357                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 358                 b->size = attr->la_size;
 359                 b->blocks = attr->la_blocks;
 360         }
 361 }
 362
 363 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 364                         const struct lu_attr *attr, const struct lu_fid *fid)
 365 {
 366         /*XXX should pack the reply body according to lu_valid*/
 367         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 368                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 369                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 370                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 371
 372         if (!S_ISREG(attr->la_mode))
 373                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 374
 375         b->atime      = attr->la_atime;
 376         b->mtime      = attr->la_mtime;
 377         b->ctime      = attr->la_ctime;
 378         b->mode       = attr->la_mode;
 379         b->size       = attr->la_size;
 380         b->blocks     = attr->la_blocks;
 381         b->uid        = attr->la_uid;
 382         b->gid        = attr->la_gid;
 383         b->flags      = attr->la_flags;
 384         b->nlink      = attr->la_nlink;
 385         b->rdev       = attr->la_rdev;
 386
 387         if (fid) {
 388                 b->fid1 = *fid;
 389                 b->valid |= OBD_MD_FLID;
 390                 CDEBUG(D_INODE, ""DFID": nlink=%d, mode=%o, size="LPU64"\n",
 391                                 PFID(fid), b->nlink, b->mode, b->size);
 392         }
 393
 394         if (info)
 395                 mdt_body_reverse_idmap(info, b);
 396 }
 397
 398 static inline int mdt_body_has_lov(const struct lu_attr *la,
 399                                    const struct mdt_body *body)
 400 {
 401         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 402                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 403 }
 404
 405 static int mdt_getattr_internal(struct mdt_thread_info *info,
 406                                 struct mdt_object *o)
 407 {
 408         struct md_object        *next = mdt_object_child(o);
 409         const struct mdt_body   *reqbody = info->mti_body;
 410         struct ptlrpc_request   *req = mdt_info_req(info);
 411         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 412         struct md_attr          *ma = &info->mti_attr;
 413         struct lu_attr          *la = &ma->ma_attr;
 414         struct req_capsule      *pill = &info->mti_pill;
 415         const struct lu_env     *env = info->mti_env;
 416         struct mdt_body         *repbody;
 417         struct lu_buf           *buffer = &info->mti_buf;
 418         int                     rc;
 419         ENTRY;
 420
 421         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 422                 RETURN(err_serious(-ENOMEM));
 423
 424         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 425
 426         ma->ma_valid = 0;
 427
 428         rc = mdt_object_exists(o);
 429         if (rc < 0) {
 430                 /* This object is located on remote node.*/
 431                 repbody->fid1 = *mdt_object_fid(o);
 432                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 433                 RETURN(0);
 434         }
 435
 436         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 437         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 438
 439         /* If it is dir object and client require MEA, then we got MEA */
 440         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 441             reqbody->valid & OBD_MD_MEA) {
 442                 /* Assumption: MDT_MD size is enough for lmv size. */
 443                 ma->ma_lmv = buffer->lb_buf;
 444                 ma->ma_lmv_size = buffer->lb_len;
 445                 ma->ma_need = MA_LMV | MA_INODE;
 446         } else {
 447                 ma->ma_lmm = buffer->lb_buf;
 448                 ma->ma_lmm_size = buffer->lb_len;
 449                 ma->ma_need = MA_LOV | MA_INODE;
 450         }
 451
 452         rc = mo_attr_get(env, next, ma);
 453         if (unlikely(rc)) {
 454                 CERROR("getattr error for "DFID": %d\n",
 455                         PFID(mdt_object_fid(o)), rc);
 456                 RETURN(rc);
 457         }
 458
 459         if (likely(ma->ma_valid & MA_INODE))
 460                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 461         else
 462                 RETURN(-EFAULT);
 463
 464         if (mdt_body_has_lov(la, reqbody)) {
 465                 if (ma->ma_valid & MA_LOV) {
 466                         LASSERT(ma->ma_lmm_size);
 467                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 468                         repbody->eadatasize = ma->ma_lmm_size;
 469                         if (S_ISDIR(la->la_mode))
 470                                 repbody->valid |= OBD_MD_FLDIREA;
 471                         else
 472                                 repbody->valid |= OBD_MD_FLEASIZE;
 473                 }
 474                 if (ma->ma_valid & MA_LMV) {
 475                         LASSERT(S_ISDIR(la->la_mode));
 476                         repbody->eadatasize = ma->ma_lmv_size;
 477                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 478                 }
 479                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 480                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 481                 }
 482         } else if (S_ISLNK(la->la_mode) &&
 483                    reqbody->valid & OBD_MD_LINKNAME) {
 484                 buffer->lb_buf = ma->ma_lmm;
 485                 buffer->lb_len = reqbody->eadatasize;
 486                 rc = mo_readlink(env, next, buffer);
 487                 if (unlikely(rc <= 0)) {
 488                         CERROR("readlink failed: %d\n", rc);
 489                         rc = -EFAULT;
 490                 } else {
 491                         repbody->valid |= OBD_MD_LINKNAME;
 492                         repbody->eadatasize = rc;
 493                         /* NULL terminate */
 494                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 495                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 496                                (char*)ma->ma_lmm, rc);
 497                         rc = 0;
 498                 }
 499         }
 500
 501         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 502                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 503                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 504                 repbody->valid |= OBD_MD_FLMODEASIZE;
 505                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 506                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 507                        repbody->max_cookiesize);
 508         }
 509
 510         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 511                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 512
 513                 /* mdt_getattr_lock only */
 514                 rc = mdt_pack_remote_perm(info, o, buf);
 515                 if (rc) {
 516                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 517                         repbody->aclsize = 0;
 518                         RETURN(rc);
 519                 } else {
 520                         repbody->valid |= OBD_MD_FLRMTPERM;
 521                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 522                 }
 523         }
 524 #ifdef CONFIG_FS_POSIX_ACL
 525         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 526                  (reqbody->valid & OBD_MD_FLACL)) {
 527                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 528                 buffer->lb_len = req_capsule_get_size(pill,
 529                                                       &RMF_ACL, RCL_SERVER);
 530                 if (buffer->lb_len > 0) {
 531                         rc = mo_xattr_get(env, next, buffer,
 532                                           XATTR_NAME_ACL_ACCESS);
 533                         if (rc < 0) {
 534                                 if (rc == -ENODATA) {
 535                                         repbody->aclsize = 0;
 536                                         repbody->valid |= OBD_MD_FLACL;
 537                                         rc = 0;
 538                                 } else if (rc == -EOPNOTSUPP) {
 539                                         rc = 0;
 540                                 } else {
 541                                         CERROR("got acl size: %d\n", rc);
 542                                 }
 543                         } else {
 544                                 repbody->aclsize = rc;
 545                                 repbody->valid |= OBD_MD_FLACL;
 546                                 rc = 0;
 547                         }
 548                 }
 549         }
 550 #endif
 551
 552         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 553             info->mti_mdt->mdt_opts.mo_mds_capa) {
 554                 struct lustre_capa *capa;
 555
 556                 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
 557                 LASSERT(capa);
 558                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 559                 rc = mo_capa_get(env, next, capa, 0);
 560                 if (rc)
 561                         RETURN(rc);
 562                 repbody->valid |= OBD_MD_FLMDSCAPA;
 563         }
 564         RETURN(rc);
 565 }
 566
 567 static int mdt_renew_capa(struct mdt_thread_info *info)
 568 {
 569         struct mdt_device  *mdt = info->mti_mdt;
 570         struct mdt_object  *obj = info->mti_object;
 571         struct mdt_body    *body;
 572         struct lustre_capa *capa, *c;
 573         int rc;
 574         ENTRY;
 575
 576         /* if object doesn't exist, or server has disabled capability,
 577          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 578          * flag not set.
 579          */
 580         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 581                 RETURN(0);
 582
 583         body = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 584         LASSERT(body != NULL);
 585
 586         c = req_capsule_client_get(&info->mti_pill, &RMF_CAPA1);
 587         LASSERT(c);
 588
 589         capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
 590         LASSERT(capa);
 591
 592         *capa = *c;
 593         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 594         if (rc == 0)
 595                 body->valid |= OBD_MD_FLOSSCAPA;
 596
 597         RETURN(rc);
 598 }
 599
 600 static int mdt_getattr(struct mdt_thread_info *info)
 601 {
 602         struct mdt_object       *obj = info->mti_object;
 603         struct req_capsule      *pill = &info->mti_pill;
 604         struct mdt_body         *reqbody;
 605         struct mdt_body         *repbody;
 606         mode_t                   mode;
 607         int                      md_size;
 608         int rc;
 609         ENTRY;
 610
 611         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 612         LASSERT(reqbody);
 613
 614         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 615                 rc = req_capsule_pack(pill);
 616                 if (unlikely(rc))
 617                         rc = err_serious(rc);
 618                 else {
 619                         rc = mdt_renew_capa(info);
 620                         mdt_shrink_reply(info);
 621                 }
 622                 GOTO(out, rc);
 623         }
 624
 625         LASSERT(obj != NULL);
 626         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 627
 628         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 629         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 630             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 631                 md_size = reqbody->eadatasize;
 632         else
 633                 md_size = info->mti_mdt->mdt_max_mdsize;
 634
 635         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 636
 637         rc = req_capsule_pack(pill);
 638         if (unlikely(rc != 0))
 639                 GOTO(out, rc = err_serious(rc));
 640
 641         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 642         LASSERT(repbody != NULL);
 643         repbody->eadatasize = 0;
 644         repbody->aclsize = 0;
 645
 646         if (reqbody->valid & OBD_MD_FLRMTPERM)
 647                 rc = mdt_init_ucred(info, reqbody);
 648         else
 649                 rc = mdt_check_ucred(info);
 650         if (unlikely(rc))
 651                 GOTO(out_shrink, rc);
 652
 653         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 654         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 655
 656         /*
 657          * Don't check capability at all, because rename might getattr for
 658          * remote obj, and at that time no capability is available.
 659          */
 660         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 661         rc = mdt_getattr_internal(info, obj);
 662         if (reqbody->valid & OBD_MD_FLRMTPERM)
 663                 mdt_exit_ucred(info);
 664         EXIT;
 665 out_shrink:
 666         mdt_shrink_reply(info);
 667 out:
 668         return rc;
 669 }
 670
 671 static int mdt_is_subdir(struct mdt_thread_info *info)
 672 {
 673         struct mdt_object     *o = info->mti_object;
 674         struct req_capsule    *pill = &info->mti_pill;
 675         const struct mdt_body *body = info->mti_body;
 676         struct mdt_body       *repbody;
 677         int                    rc;
 678         ENTRY;
 679
 680         LASSERT(o != NULL);
 681
 682         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 683
 684         /*
 685          * We save last checked parent fid to @repbody->fid1 for remote
 686          * directory case.
 687          */
 688         LASSERT(fid_is_sane(&body->fid2));
 689         LASSERT(mdt_object_exists(o) > 0);
 690         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 691                            &body->fid2, &repbody->fid1);
 692         if (rc == 0 || rc == -EREMOTE)
 693                 repbody->valid |= OBD_MD_FLID;
 694
 695         RETURN(rc);
 696 }
 697
 698 static int mdt_raw_lookup(struct mdt_thread_info *info,
 699                           struct mdt_object *parent,
 700                           const struct lu_name *lname,
 701                           struct ldlm_reply *ldlm_rep)
 702 {
 703         struct md_object *next = mdt_object_child(info->mti_object);
 704         const struct mdt_body *reqbody = info->mti_body;
 705         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 706         struct mdt_body *repbody;
 707         int rc;
 708         ENTRY;
 709
 710         if (reqbody->valid != OBD_MD_FLID)
 711                 RETURN(0);
 712
 713         LASSERT(!info->mti_cross_ref);
 714
 715         /* Only got the fid of this obj by name */
 716         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 717                         &info->mti_spec);
 718 #if 0
 719         /* XXX is raw_lookup possible as intent operation? */
 720         if (rc != 0) {
 721                 if (rc == -ENOENT)
 722                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 723                 RETURN(rc);
 724         } else
 725                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 726
 727         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 728 #endif
 729         if (rc == 0) {
 730                 repbody = req_capsule_server_get(&info->mti_pill,
 731                                                  &RMF_MDT_BODY);
 732                 repbody->fid1 = *child_fid;
 733                 repbody->valid = OBD_MD_FLID;
 734         }
 735         RETURN(1);
 736 }
 737
 738 /*
 739  * UPDATE lock should be taken against parent, and be release before exit;
 740  * child_bits lock should be taken against child, and be returned back:
 741  *            (1)normal request should release the child lock;
 742  *            (2)intent request will grant the lock to client.
 743  */
 744 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 745                                  struct mdt_lock_handle *lhc,
 746                                  __u64 child_bits,
 747                                  struct ldlm_reply *ldlm_rep)
 748 {
 749         struct ptlrpc_request  *req       = mdt_info_req(info);
 750         struct mdt_body        *reqbody   = NULL;
 751         struct mdt_object      *parent    = info->mti_object;
 752         struct mdt_object      *child;
 753         struct md_object       *next      = mdt_object_child(parent);
 754         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 755         struct lu_name         *lname     = NULL;
 756         const char             *name;
 757         int                     namelen   = 0;
 758         struct mdt_lock_handle *lhp;
 759         struct ldlm_lock       *lock;
 760         struct ldlm_res_id     *res_id;
 761         int                     is_resent;
 762         int                     rc;
 763
 764         ENTRY;
 765
 766         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 767         LASSERT(ergo(is_resent,
 768                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 769
 770         LASSERT(parent != NULL);
 771         name = req_capsule_client_get(&info->mti_pill, &RMF_NAME);
 772         if (name == NULL)
 773                 RETURN(err_serious(-EFAULT));
 774
 775         namelen = req_capsule_get_size(&info->mti_pill, &RMF_NAME,
 776                                        RCL_CLIENT) - 1;
 777         LASSERT(namelen >= 0);
 778
 779         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 780          * otherwise do not allow empty name, that is the name must contain
 781          * at least one character and the terminating '\0'*/
 782         if (namelen == 0) {
 783                 reqbody =req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
 784                 LASSERT(fid_is_sane(&reqbody->fid2));
 785                 name = NULL;
 786
 787                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 788                        "ldlm_rep = %p\n",
 789                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 790                        ldlm_rep);
 791         } else {
 792                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 793                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 794                        "ldlm_rep = %p\n",
 795                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 796         }
 797
 798         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 799
 800         rc = mdt_object_exists(parent);
 801         if (unlikely(rc == 0)) {
 802                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 803                                 &parent->mot_obj.mo_lu,
 804                                 "Parent doesn't exist!\n");
 805                 RETURN(-ESTALE);
 806         } else
 807                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 808                          PFID(mdt_object_fid(parent)));
 809
 810         if (lname) {
 811                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 812                 if (rc != 0) {
 813                         if (rc > 0)
 814                                 rc = 0;
 815                         RETURN(rc);
 816                 }
 817         }
 818
 819         if (info->mti_cross_ref) {
 820                 /* Only getattr on the child. Parent is on another node. */
 821                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 822                 child = parent;
 823                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 824                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 825
 826                 if (is_resent) {
 827                         /* Do not take lock for resent case. */
 828                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 829                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 830                                  lhc->mlh_reg_lh.cookie);
 831                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 832                                                 &lock->l_resource->lr_name));
 833                         LDLM_LOCK_PUT(lock);
 834                         rc = 0;
 835                 } else {
 836                         mdt_lock_handle_init(lhc);
 837                         mdt_lock_reg_init(lhc, LCK_PR);
 838
 839                         /*
 840                          * Object's name is on another MDS, no lookup lock is
 841                          * needed here but update is.
 842                          */
 843                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 844                         child_bits |= MDS_INODELOCK_UPDATE;
 845
 846                         rc = mdt_object_lock(info, child, lhc, child_bits,
 847                                              MDT_LOCAL_LOCK);
 848                 }
 849                 if (rc == 0) {
 850                         /* Finally, we can get attr for child. */
 851                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 852                                          BYPASS_CAPA);
 853                         rc = mdt_getattr_internal(info, child);
 854                         if (unlikely(rc != 0))
 855                                 mdt_object_unlock(info, child, lhc, 1);
 856                 }
 857                 GOTO(out, rc);
 858         }
 859
 860         /* step 1: lock parent */
 861         lhp = &info->mti_lh[MDT_LH_PARENT];
 862         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 863         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 864                              MDT_LOCAL_LOCK);
 865
 866         if (unlikely(rc != 0))
 867                 RETURN(rc);
 868
 869         if (lname) {
 870                 /* step 2: lookup child's fid by name */
 871                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 872                                 &info->mti_spec);
 873
 874                 if (rc != 0) {
 875                         if (rc == -ENOENT)
 876                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 877                         GOTO(out_parent, rc);
 878                 } else
 879                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 880         } else {
 881                 *child_fid = reqbody->fid2;
 882                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 883         }
 884
 885         /*
 886          *step 3: find the child object by fid & lock it.
 887          *        regardless if it is local or remote.
 888          */
 889         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 890
 891         if (unlikely(IS_ERR(child)))
 892                 GOTO(out_parent, rc = PTR_ERR(child));
 893         if (is_resent) {
 894                 /* Do not take lock for resent case. */
 895                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 896                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 897                          lhc->mlh_reg_lh.cookie);
 898
 899                 res_id = &lock->l_resource->lr_name;
 900                 if (!fid_res_name_eq(mdt_object_fid(child),
 901                                     &lock->l_resource->lr_name)) {
 902                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 903                                                  &lock->l_resource->lr_name),
 904                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 905                                  (unsigned long)res_id->name[0],
 906                                  (unsigned long)res_id->name[1],
 907                                  (unsigned long)res_id->name[2],
 908                                  PFID(mdt_object_fid(parent)));
 909                           CWARN("Although resent, but still not get child lock"
 910                                 "parent:"DFID" child:"DFID"\n",
 911                                 PFID(mdt_object_fid(parent)),
 912                                 PFID(mdt_object_fid(child)));
 913                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 914                           LDLM_LOCK_PUT(lock);
 915                           GOTO(relock, 0);
 916                 }
 917                 LDLM_LOCK_PUT(lock);
 918                 rc = 0;
 919         } else {
 920 relock:
 921                 mdt_lock_handle_init(lhc);
 922                 mdt_lock_reg_init(lhc, LCK_PR);
 923
 924                 if (mdt_object_exists(child) == 0) {
 925                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 926                                         &child->mot_obj.mo_lu,
 927                                         "Object doesn't exist!\n");
 928                 }
 929                 rc = mdt_object_lock(info, child, lhc, child_bits,
 930                                      MDT_CROSS_LOCK);
 931
 932                 if (unlikely(rc != 0))
 933                         GOTO(out_child, rc);
 934         }
 935
 936         /* finally, we can get attr for child. */
 937         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 938         rc = mdt_getattr_internal(info, child);
 939         if (unlikely(rc != 0)) {
 940                 mdt_object_unlock(info, child, lhc, 1);
 941         } else {
 942                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 943                 if (lock) {
 944                         struct mdt_body *repbody;
 945                         struct lu_attr *ma;
 946
 947                         /* Debugging code. */
 948                         res_id = &lock->l_resource->lr_name;
 949                         LDLM_DEBUG(lock, "Returning lock to client\n");
 950                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 951                                                  &lock->l_resource->lr_name),
 952                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 953                                  (unsigned long)res_id->name[0],
 954                                  (unsigned long)res_id->name[1],
 955                                  (unsigned long)res_id->name[2],
 956                                  PFID(mdt_object_fid(child)));
 957                         /*
 958                          * Pack Size-on-MDS inode attributes to the body if
 959                          * update lock is given.
 960                          */
 961                         repbody = req_capsule_server_get(&info->mti_pill,
 962                                                          &RMF_MDT_BODY);
 963                         ma = &info->mti_attr.ma_attr;
 964                         if (lock->l_policy_data.l_inodebits.bits &
 965                             MDS_INODELOCK_UPDATE)
 966                                 mdt_pack_size2body(info, child);
 967                         LDLM_LOCK_PUT(lock);
 968                 }
 969         }
 970         EXIT;
 971 out_child:
 972         mdt_object_put(info->mti_env, child);
 973 out_parent:
 974         mdt_object_unlock(info, parent, lhp, 1);
 975 out:
 976         return rc;
 977 }
 978
 979 /* normal handler: should release the child lock */
 980 static int mdt_getattr_name(struct mdt_thread_info *info)
 981 {
 982         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
 983         struct mdt_body        *reqbody;
 984         struct mdt_body        *repbody;
 985         int rc;
 986         ENTRY;
 987
 988         reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
 989         LASSERT(reqbody != NULL);
 990         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
 991         LASSERT(repbody != NULL);
 992
 993         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 994         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 995         repbody->eadatasize = 0;
 996         repbody->aclsize = 0;
 997
 998         rc = mdt_init_ucred(info, reqbody);
 999         if (unlikely(rc))
1000                 GOTO(out, rc);
1001
1002         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1003         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1004                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1005                 lhc->mlh_reg_lh.cookie = 0;
1006         }
1007         mdt_exit_ucred(info);
1008         EXIT;
1009 out:
1010         mdt_shrink_reply(info);
1011         return rc;
1012 }
1013
1014 static struct lu_device_operations mdt_lu_ops;
1015
1016 static int lu_device_is_mdt(struct lu_device *d)
1017 {
1018         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1019 }
1020
1021 static int mdt_set_info(struct mdt_thread_info *info)
1022 {
1023         struct ptlrpc_request *req = mdt_info_req(info);
1024         char *key;
1025         __u32 *val;
1026         int keylen, rc = 0;
1027         ENTRY;
1028
1029         rc = lustre_pack_reply(req, 1, NULL, NULL);
1030         if (rc)
1031                 RETURN(rc);
1032
1033         key = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_KEY);
1034         if (key == NULL) {
1035                 DEBUG_REQ(D_HA, req, "no set_info key");
1036                 RETURN(-EFAULT);
1037         }
1038
1039         keylen = req_capsule_get_size(&info->mti_pill, &RMF_SETINFO_KEY,
1040                                       RCL_CLIENT);
1041
1042         val = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_VAL);
1043         if (val == NULL) {
1044                 DEBUG_REQ(D_HA, req, "no set_info val");
1045                 RETURN(-EFAULT);
1046         }
1047
1048         if (keylen != (sizeof(KEY_READ_ONLY) - 1) ||
1049             memcmp(key, KEY_READ_ONLY, keylen) != 0)
1050                 RETURN(-EINVAL);
1051
1052         req->rq_status = 0;
1053         lustre_msg_set_status(req->rq_repmsg, 0);
1054
1055         spin_lock(&req->rq_export->exp_lock);
1056         if (*val)
1057                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1058         else
1059                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1060         spin_unlock(&req->rq_export->exp_lock);
1061
1062         RETURN(0);
1063 }
1064
1065 static int mdt_connect(struct mdt_thread_info *info)
1066 {
1067         int rc;
1068         struct ptlrpc_request *req;
1069
1070         req = mdt_info_req(info);
1071         rc = target_handle_connect(req);
1072         if (rc == 0) {
1073                 LASSERT(req->rq_export != NULL);
1074                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1075                 rc = mdt_init_idmap(info);
1076                 if (rc != 0)
1077                         /* if mdt_init_idmap failed, revocation for connect */
1078                         obd_disconnect(class_export_get(req->rq_export));
1079         } else
1080                 rc = err_serious(rc);
1081         return rc;
1082 }
1083
1084 static int mdt_disconnect(struct mdt_thread_info *info)
1085 {
1086         int rc;
1087         ENTRY;
1088
1089         rc = target_handle_disconnect(mdt_info_req(info));
1090         if (rc)
1091                 rc = err_serious(rc);
1092         RETURN(rc);
1093 }
1094
1095 static int mdt_sendpage(struct mdt_thread_info *info,
1096                         struct lu_rdpg *rdpg)
1097 {
1098         struct ptlrpc_request   *req = mdt_info_req(info);
1099         struct ptlrpc_bulk_desc *desc;
1100         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1101         int                      tmpcount;
1102         int                      tmpsize;
1103         int                      i;
1104         int                      rc;
1105         ENTRY;
1106
1107         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1108                                     MDS_BULK_PORTAL);
1109         if (desc == NULL)
1110                 GOTO(out, rc = -ENOMEM);
1111
1112         for (i = 0, tmpcount = rdpg->rp_count;
1113                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1114                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1115                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1116         }
1117
1118         LASSERT(desc->bd_nob == rdpg->rp_count);
1119         rc = ptlrpc_start_bulk_transfer(desc);
1120         if (rc)
1121                 GOTO(free_desc, rc);
1122
1123         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1124                 GOTO(abort_bulk, rc);
1125
1126         *lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
1127         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1128         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1129
1130         if (rc == 0) {
1131                 if (desc->bd_success &&
1132                     desc->bd_nob_transferred == rdpg->rp_count)
1133                         GOTO(free_desc, rc);
1134
1135                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1136         }
1137
1138         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1139                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1140                   desc->bd_nob_transferred, rdpg->rp_count,
1141                   req->rq_export->exp_client_uuid.uuid,
1142                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1143
1144         class_fail_export(req->rq_export);
1145
1146         EXIT;
1147 abort_bulk:
1148         ptlrpc_abort_bulk(desc);
1149 free_desc:
1150         ptlrpc_free_bulk(desc);
1151 out:
1152         return rc;
1153 }
1154
1155 #ifdef HAVE_SPLIT_SUPPORT
1156 /*
1157  * Retrieve dir entry from the page and insert it to the slave object, actually,
1158  * this should be in osd layer, but since it will not in the final product, so
1159  * just do it here and do not define more moo api anymore for this.
1160  */
1161 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1162                               int size)
1163 {
1164         struct mdt_object *object = info->mti_object;
1165         struct lu_fid *lf = &info->mti_tmp_fid2;
1166         struct md_attr *ma = &info->mti_attr;
1167         struct lu_dirpage *dp;
1168         struct lu_dirent *ent;
1169         int rc = 0, offset = 0;
1170         ENTRY;
1171
1172         /* Make sure we have at least one entry. */
1173         if (size == 0)
1174                 RETURN(-EINVAL);
1175
1176         /*
1177          * Disable trans for this name insert, since it will include many trans
1178          * for this.
1179          */
1180         info->mti_no_need_trans = 1;
1181         /*
1182          * When write_dir_page, no need update parent's ctime,
1183          * and no permission check for name_insert.
1184          */
1185         ma->ma_attr.la_ctime = 0;
1186         ma->ma_attr.la_valid = LA_MODE;
1187         ma->ma_valid = MA_INODE;
1188
1189         kmap(page);
1190         dp = page_address(page);
1191         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1192
1193         for (ent = lu_dirent_start(dp); ent != NULL;
1194              ent = lu_dirent_next(ent)) {
1195                 struct lu_name *lname;
1196                 char *name;
1197
1198                 if (le16_to_cpu(ent->lde_namelen) == 0)
1199                         continue;
1200
1201                 fid_le_to_cpu(lf, &ent->lde_fid);
1202                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1203                         ma->ma_attr.la_mode = S_IFDIR;
1204                 else
1205                         ma->ma_attr.la_mode = 0;
1206                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1207                 if (name == NULL)
1208                         GOTO(out, rc = -ENOMEM);
1209
1210                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1211                 lname = mdt_name(info->mti_env, name,
1212                                  le16_to_cpu(ent->lde_namelen) + 1);
1213                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1214                 rc = mdo_name_insert(info->mti_env,
1215                                      md_object_next(&object->mot_obj),
1216                                      lname, lf, ma);
1217                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1218                 if (rc) {
1219                         CERROR("Can't insert %*.*s, rc %d\n",
1220                                le16_to_cpu(ent->lde_namelen),
1221                                le16_to_cpu(ent->lde_namelen),
1222                                ent->lde_name, rc);
1223                         GOTO(out, rc);
1224                 }
1225
1226                 offset += lu_dirent_size(ent);
1227                 if (offset >= size)
1228                         break;
1229         }
1230         EXIT;
1231 out:
1232         kunmap(page);
1233         return rc;
1234 }
1235
1236 static int mdt_bulk_timeout(void *data)
1237 {
1238         ENTRY;
1239
1240         CERROR("mdt bulk transfer timeout \n");
1241
1242         RETURN(1);
1243 }
1244
1245 static int mdt_writepage(struct mdt_thread_info *info)
1246 {
1247         struct ptlrpc_request   *req = mdt_info_req(info);
1248         struct mdt_body         *reqbody;
1249         struct l_wait_info      *lwi;
1250         struct ptlrpc_bulk_desc *desc;
1251         struct page             *page;
1252         int                rc;
1253         ENTRY;
1254
1255
1256         reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
1257         if (reqbody == NULL)
1258                 RETURN(err_serious(-EFAULT));
1259
1260         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1261         if (desc == NULL)
1262                 RETURN(err_serious(-ENOMEM));
1263
1264         /* allocate the page for the desc */
1265         page = cfs_alloc_page(CFS_ALLOC_STD);
1266         if (page == NULL)
1267                 GOTO(desc_cleanup, rc = -ENOMEM);
1268
1269         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1270                (int)reqbody->size, (int)reqbody->nlink);
1271
1272         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1273                               (int)reqbody->nlink);
1274
1275         /*
1276          * Check if client was evicted while we were doing i/o before touching
1277          * network.
1278          */
1279         OBD_ALLOC_PTR(lwi);
1280         if (!lwi)
1281                 GOTO(cleanup_page, rc = -ENOMEM);
1282
1283         if (desc->bd_export->exp_failed)
1284                 rc = -ENOTCONN;
1285         else
1286                 rc = ptlrpc_start_bulk_transfer (desc);
1287         if (rc == 0) {
1288                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1289                                             mdt_bulk_timeout, desc);
1290                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1291                                   desc->bd_export->exp_failed, lwi);
1292                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1293                 if (rc == -ETIMEDOUT) {
1294                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1295                         ptlrpc_abort_bulk(desc);
1296                 } else if (desc->bd_export->exp_failed) {
1297                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1298                         rc = -ENOTCONN;
1299                         ptlrpc_abort_bulk(desc);
1300                 } else if (!desc->bd_success ||
1301                            desc->bd_nob_transferred != desc->bd_nob) {
1302                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1303                                   desc->bd_success ?
1304                                   "truncated" : "network error on",
1305                                   desc->bd_nob_transferred, desc->bd_nob);
1306                         /* XXX should this be a different errno? */
1307                         rc = -ETIMEDOUT;
1308                 }
1309         } else {
1310                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1311         }
1312         if (rc)
1313                 GOTO(cleanup_lwi, rc);
1314         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1315
1316 cleanup_lwi:
1317         OBD_FREE_PTR(lwi);
1318 cleanup_page:
1319         __cfs_free_page(page);
1320 desc_cleanup:
1321         ptlrpc_free_bulk(desc);
1322         RETURN(rc);
1323 }
1324 #endif
1325
1326 static int mdt_readpage(struct mdt_thread_info *info)
1327 {
1328         struct mdt_object *object = info->mti_object;
1329         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1330         struct mdt_body   *reqbody;
1331         struct mdt_body   *repbody;
1332         int                rc;
1333         int                i;
1334         ENTRY;
1335
1336         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1337                 RETURN(err_serious(-ENOMEM));
1338
1339         reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
1340         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
1341         if (reqbody == NULL || repbody == NULL)
1342                 RETURN(err_serious(-EFAULT));
1343
1344         rc = mdt_check_ucred(info);
1345         if (rc)
1346                 RETURN(err_serious(rc));
1347
1348         /*
1349          * prepare @rdpg before calling lower layers and transfer itself. Here
1350          * reqbody->size contains offset of where to start to read and
1351          * reqbody->nlink contains number bytes to read.
1352          */
1353         rdpg->rp_hash = reqbody->size;
1354         if ((__u64)rdpg->rp_hash != reqbody->size) {
1355                 CERROR("Invalid hash: %#llx != %#llx\n",
1356                        (__u64)rdpg->rp_hash, reqbody->size);
1357                 RETURN(-EFAULT);
1358         }
1359         rdpg->rp_count  = reqbody->nlink;
1360         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1361         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1362         if (rdpg->rp_pages == NULL)
1363                 RETURN(-ENOMEM);
1364
1365         for (i = 0; i < rdpg->rp_npages; ++i) {
1366                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1367                 if (rdpg->rp_pages[i] == NULL)
1368                         GOTO(free_rdpg, rc = -ENOMEM);
1369         }
1370
1371         /* call lower layers to fill allocated pages with directory data */
1372         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1373         if (rc)
1374                 GOTO(free_rdpg, rc);
1375
1376         /* send pages to client */
1377         rc = mdt_sendpage(info, rdpg);
1378
1379         EXIT;
1380 free_rdpg:
1381
1382         for (i = 0; i < rdpg->rp_npages; i++)
1383                 if (rdpg->rp_pages[i] != NULL)
1384                         __cfs_free_page(rdpg->rp_pages[i]);
1385         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1386
1387         OBD_FAIL_RETURN(OBD_FAIL_MDS_SENDPAGE, 0);
1388
1389         return rc;
1390 }
1391
1392 static int mdt_reint_internal(struct mdt_thread_info *info,
1393                               struct mdt_lock_handle *lhc,
1394                               __u32 op)
1395 {
1396         struct req_capsule      *pill = &info->mti_pill;
1397         struct mdt_device       *mdt = info->mti_mdt;
1398         struct mdt_body         *repbody;
1399         int                      need_shrink = 0;
1400         int                      rc;
1401         ENTRY;
1402
1403         /* pack reply */
1404         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
1405                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1406                                      mdt->mdt_max_mdsize);
1407                 need_shrink = 1;
1408         }
1409         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
1410                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1411                                      mdt->mdt_max_cookiesize);
1412                 need_shrink = 1;
1413         }
1414         rc = req_capsule_pack(pill);
1415         if (rc != 0) {
1416                 CERROR("Can't pack response, rc %d\n", rc);
1417                 RETURN(err_serious(rc));
1418         }
1419
1420         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1421                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1422                 LASSERT(repbody);
1423                 repbody->eadatasize = 0;
1424                 repbody->aclsize = 0;
1425         }
1426
1427         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1428                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1429
1430         rc = mdt_reint_unpack(info, op);
1431         if (rc != 0) {
1432                 CERROR("Can't unpack reint, rc %d\n", rc);
1433                 GOTO(out_shrink, rc = err_serious(rc));
1434         }
1435
1436         rc = mdt_init_ucred_reint(info);
1437         if (rc)
1438                 GOTO(out_shrink, rc);
1439
1440         rc = mdt_fix_attr_ucred(info, op);
1441         if (rc != 0)
1442                 GOTO(out_ucred, rc = err_serious(rc));
1443
1444         need_shrink = 0;
1445         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1446                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1447                 GOTO(out_ucred, rc);
1448         }
1449
1450         rc = mdt_reint_rec(info, lhc);
1451         EXIT;
1452 out_ucred:
1453         mdt_exit_ucred(info);
1454 out_shrink:
1455         if (need_shrink)
1456                 mdt_shrink_reply(info);
1457         return rc;
1458 }
1459
1460 static long mdt_reint_opcode(struct mdt_thread_info *info,
1461                              const struct req_format **fmt)
1462 {
1463         __u32 *ptr;
1464         long opc;
1465
1466         opc = err_serious(-EFAULT);
1467         ptr = req_capsule_client_get(&info->mti_pill, &RMF_REINT_OPC);
1468         if (ptr != NULL) {
1469                 opc = *ptr;
1470                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1471                 if (opc < REINT_MAX && fmt[opc] != NULL)
1472                         req_capsule_extend(&info->mti_pill, fmt[opc]);
1473                 else {
1474                         CERROR("Unsupported opc: %ld\n", opc);
1475                         opc = err_serious(opc);
1476                 }
1477         }
1478         return opc;
1479 }
1480
1481 static int mdt_reint(struct mdt_thread_info *info)
1482 {
1483         long opc;
1484         int  rc;
1485
1486         static const struct req_format *reint_fmts[REINT_MAX] = {
1487                 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1488                 [REINT_CREATE]  = &RQF_MDS_REINT_CREATE,
1489                 [REINT_LINK]    = &RQF_MDS_REINT_LINK,
1490                 [REINT_UNLINK]  = &RQF_MDS_REINT_UNLINK,
1491                 [REINT_RENAME]  = &RQF_MDS_REINT_RENAME,
1492                 [REINT_OPEN]    = &RQF_MDS_REINT_OPEN
1493         };
1494
1495         ENTRY;
1496
1497         opc = mdt_reint_opcode(info, reint_fmts);
1498         if (opc >= 0) {
1499                 /*
1500                  * No lock possible here from client to pass it to reint code
1501                  * path.
1502                  */
1503                 rc = mdt_reint_internal(info, NULL, opc);
1504         } else {
1505                 rc = opc;
1506         }
1507
1508         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1509         RETURN(rc);
1510 }
1511
1512 /* TODO these two methods not available now. */
1513
1514 /* this should sync the whole device */
1515 static int mdt_device_sync(struct mdt_thread_info *info)
1516 {
1517         return 0;
1518 }
1519
1520 /* this should sync this object */
1521 static int mdt_object_sync(struct mdt_thread_info *info)
1522 {
1523         return 0;
1524 }
1525
1526 static int mdt_sync(struct mdt_thread_info *info)
1527 {
1528         struct req_capsule *pill = &info->mti_pill;
1529         struct mdt_body *body;
1530         int rc;
1531         ENTRY;
1532
1533         /* The fid may be zero, so we req_capsule_set manually */
1534         req_capsule_set(pill, &RQF_MDS_SYNC);
1535
1536         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1537         if (body == NULL)
1538                 RETURN(err_serious(-EINVAL));
1539
1540         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1541                 RETURN(err_serious(-ENOMEM));
1542
1543         if (fid_seq(&body->fid1) == 0) {
1544                 /* sync the whole device */
1545                 rc = req_capsule_pack(pill);
1546                 if (rc == 0)
1547                         rc = mdt_device_sync(info);
1548                 else
1549                         rc = err_serious(rc);
1550         } else {
1551                 /* sync an object */
1552                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1553                 if (rc == 0) {
1554                         rc = mdt_object_sync(info);
1555                         if (rc == 0) {
1556                                 struct md_object *next;
1557                                 const struct lu_fid *fid;
1558                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1559
1560                                 next = mdt_object_child(info->mti_object);
1561                                 info->mti_attr.ma_need = MA_INODE;
1562                                 info->mti_attr.ma_valid = 0;
1563                                 rc = mo_attr_get(info->mti_env, next,
1564                                                  &info->mti_attr);
1565                                 if (rc == 0) {
1566                                         body = req_capsule_server_get(pill,
1567                                                                 &RMF_MDT_BODY);
1568                                         fid = mdt_object_fid(info->mti_object);
1569                                         mdt_pack_attr2body(info, body, la, fid);
1570                                 }
1571                         }
1572                 } else
1573                         rc = err_serious(rc);
1574         }
1575         RETURN(rc);
1576 }
1577
1578 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1579 {
1580         return err_serious(-EOPNOTSUPP);
1581 }
1582
1583 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1584 {
1585         return err_serious(-EOPNOTSUPP);
1586 }
1587
1588 /*
1589  * OBD PING and other handlers.
1590  */
1591 static int mdt_obd_ping(struct mdt_thread_info *info)
1592 {
1593         int rc;
1594         ENTRY;
1595         rc = target_handle_ping(mdt_info_req(info));
1596         if (rc < 0)
1597                 rc = err_serious(rc);
1598         RETURN(rc);
1599 }
1600
1601 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1602 {
1603         return err_serious(-EOPNOTSUPP);
1604 }
1605
1606 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1607 {
1608         return err_serious(-EOPNOTSUPP);
1609 }
1610
1611
1612 /*
1613  * DLM handlers.
1614  */
1615 static struct ldlm_callback_suite cbs = {
1616         .lcs_completion = ldlm_server_completion_ast,
1617         .lcs_blocking   = ldlm_server_blocking_ast,
1618         .lcs_glimpse    = NULL
1619 };
1620
1621 static int mdt_enqueue(struct mdt_thread_info *info)
1622 {
1623         struct ptlrpc_request *req;
1624         __u64 req_bits;
1625         int rc;
1626
1627         /*
1628          * info->mti_dlm_req already contains swapped and (if necessary)
1629          * converted dlm request.
1630          */
1631         LASSERT(info->mti_dlm_req != NULL);
1632
1633         if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) {
1634                 info->mti_fail_id = OBD_FAIL_LDLM_ENQUEUE;
1635                 return 0;
1636         }
1637
1638         req = mdt_info_req(info);
1639
1640         /*
1641          * Lock without inodebits makes no sense and will oops later in
1642          * ldlm. Let's check it now to see if we have wrong lock from client or
1643          * bits get corrupted somewhere in mdt_intent_policy().
1644          */
1645         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1646         LASSERT(req_bits != 0);
1647
1648         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1649                                   req, info->mti_dlm_req, &cbs);
1650         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1651         return rc ? err_serious(rc) : req->rq_status;
1652 }
1653
1654 static int mdt_convert(struct mdt_thread_info *info)
1655 {
1656         int rc;
1657         struct ptlrpc_request *req;
1658
1659         LASSERT(info->mti_dlm_req);
1660         req = mdt_info_req(info);
1661         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1662         return rc ? err_serious(rc) : req->rq_status;
1663 }
1664
1665 static int mdt_bl_callback(struct mdt_thread_info *info)
1666 {
1667         CERROR("bl callbacks should not happen on MDS\n");
1668         LBUG();
1669         return err_serious(-EOPNOTSUPP);
1670 }
1671
1672 static int mdt_cp_callback(struct mdt_thread_info *info)
1673 {
1674         CERROR("cp callbacks should not happen on MDS\n");
1675         LBUG();
1676         return err_serious(-EOPNOTSUPP);
1677 }
1678
1679 /*
1680  * sec context handlers
1681  */
1682 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1683 {
1684         int rc;
1685
1686         rc = mdt_handle_idmap(info);
1687
1688         if (unlikely(rc)) {
1689                 struct ptlrpc_request *req = mdt_info_req(info);
1690                 __u32                  opc;
1691
1692                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1693                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1694                         sptlrpc_svc_ctx_invalidate(req);
1695         }
1696
1697         return rc;
1698 }
1699
1700 static struct mdt_object *mdt_obj(struct lu_object *o)
1701 {
1702         LASSERT(lu_device_is_mdt(o->lo_dev));
1703         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1704 }
1705
1706 struct mdt_object *mdt_object_find(const struct lu_env *env,
1707                                    struct mdt_device *d,
1708                                    const struct lu_fid *f)
1709 {
1710         struct lu_object *o;
1711         struct mdt_object *m;
1712         ENTRY;
1713
1714         o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1715         if (unlikely(IS_ERR(o)))
1716                 m = (struct mdt_object *)o;
1717         else
1718                 m = mdt_obj(o);
1719         RETURN(m);
1720 }
1721
1722 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1723                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1724 {
1725         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1726         ldlm_policy_data_t *policy = &info->mti_policy;
1727         struct ldlm_res_id *res_id = &info->mti_res_id;
1728         int rc;
1729         ENTRY;
1730
1731         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1732         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1733         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1734         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1735
1736         if (mdt_object_exists(o) < 0) {
1737                 if (locality == MDT_CROSS_LOCK) {
1738                         /* cross-ref object fix */
1739                         ibits &= ~MDS_INODELOCK_UPDATE;
1740                         ibits |= MDS_INODELOCK_LOOKUP;
1741                 } else {
1742                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1743                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1744                 }
1745                 /* No PDO lock on remote object */
1746                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1747         }
1748
1749         memset(policy, 0, sizeof(*policy));
1750         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1751
1752         /*
1753          * Take PDO lock on whole directory and build correct @res_id for lock
1754          * on part of directory.
1755          */
1756         if (lh->mlh_pdo_hash != 0) {
1757                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1758                 mdt_lock_pdo_mode(info, o, lh);
1759                 if (lh->mlh_pdo_mode != LCK_NL) {
1760                         /*
1761                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1762                          * is never going to be sent to client and we do not
1763                          * want it slowed down due to possible cancels.
1764                          */
1765                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1766                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1767                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1768                         if (unlikely(rc))
1769                                 RETURN(rc);
1770                 }
1771
1772                 /*
1773                  * Finish res_id initializing by name hash marking patr of
1774                  * directory which is taking modification.
1775                  */
1776                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1777         }
1778
1779         policy->l_inodebits.bits = ibits;
1780
1781         /*
1782          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1783          * going to be sent to client. If it is - mdt_intent_policy() path will
1784          * fix it up and turns FL_LOCAL flag off.
1785          */
1786         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1787                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1788
1789         if (rc)
1790                 GOTO(out, rc);
1791
1792         if (lh->mlh_type == MDT_PDO_LOCK) {
1793                 /* check for exists after object is locked */
1794                 if (mdt_object_exists(o) == 0) {
1795                         /* Non-existent object shouldn't have PDO lock */
1796                         rc = -ESTALE;
1797                 } else {
1798                         /* Non-dir object shouldn't have PDO lock */
1799                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1800                 }
1801         }
1802 out:
1803         if (rc)
1804                 mdt_object_unlock(info, o, lh, 1);
1805
1806
1807         RETURN(rc);
1808 }
1809
1810 /*
1811  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1812  * to save this lock in req.  when transaction committed, req will be released,
1813  * and lock will, too.
1814  */
1815 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1816                        struct mdt_lock_handle *lh, int decref)
1817 {
1818         struct ptlrpc_request *req = mdt_info_req(info);
1819         ENTRY;
1820
1821         if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
1822                 /* Do not save PDO locks to request, just decref. */
1823                 mdt_fid_unlock(&lh->mlh_pdo_lh,
1824                                lh->mlh_pdo_mode);
1825                 lh->mlh_pdo_lh.cookie = 0ull;
1826         }
1827
1828         if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
1829                 if (decref) {
1830                         mdt_fid_unlock(&lh->mlh_reg_lh,
1831                                        lh->mlh_reg_mode);
1832                 } else {
1833                         ptlrpc_save_lock(req, &lh->mlh_reg_lh,
1834                                          lh->mlh_reg_mode);
1835                 }
1836                 lh->mlh_reg_lh.cookie = 0ull;
1837         }
1838
1839         EXIT;
1840 }
1841
1842 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1843                                         const struct lu_fid *f,
1844                                         struct mdt_lock_handle *lh,
1845                                         __u64 ibits)
1846 {
1847         struct mdt_object *o;
1848
1849         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1850         if (!IS_ERR(o)) {
1851                 int rc;
1852
1853                 rc = mdt_object_lock(info, o, lh, ibits,
1854                                      MDT_LOCAL_LOCK);
1855                 if (rc != 0) {
1856                         mdt_object_put(info->mti_env, o);
1857                         o = ERR_PTR(rc);
1858                 }
1859         }
1860         return o;
1861 }
1862
1863 void mdt_object_unlock_put(struct mdt_thread_info * info,
1864                            struct mdt_object * o,
1865                            struct mdt_lock_handle *lh,
1866                            int decref)
1867 {
1868         mdt_object_unlock(info, o, lh, decref);
1869         mdt_object_put(info->mti_env, o);
1870 }
1871
1872 static struct mdt_handler *mdt_handler_find(__u32 opc,
1873                                             struct mdt_opc_slice *supported)
1874 {
1875         struct mdt_opc_slice *s;
1876         struct mdt_handler   *h;
1877
1878         h = NULL;
1879         for (s = supported; s->mos_hs != NULL; s++) {
1880                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1881                         h = s->mos_hs + (opc - s->mos_opc_start);
1882                         if (likely(h->mh_opc != 0))
1883                                 LASSERT(h->mh_opc == opc);
1884                         else
1885                                 h = NULL; /* unsupported opc */
1886                         break;
1887                 }
1888         }
1889         return h;
1890 }
1891
1892 static int mdt_lock_resname_compat(struct mdt_device *m,
1893                                    struct ldlm_request *req)
1894 {
1895         /* XXX something... later. */
1896         return 0;
1897 }
1898
1899 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1900 {
1901         /* XXX something... later. */
1902         return 0;
1903 }
1904
1905 /*
1906  * Generic code handling requests that have struct mdt_body passed in:
1907  *
1908  *  - extract mdt_body from request and save it in @info, if present;
1909  *
1910  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1911  *  @info;
1912  *
1913  *  - if HABEO_CORPUS flag is set for this request type check whether object
1914  *  actually exists on storage (lu_object_exists()).
1915  *
1916  */
1917 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1918 {
1919         const struct mdt_body    *body;
1920         struct mdt_object        *obj;
1921         const struct lu_env      *env;
1922         struct req_capsule       *pill;
1923         int                       rc;
1924
1925         env = info->mti_env;
1926         pill = &info->mti_pill;
1927
1928         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1929         if (body == NULL)
1930                 return -EFAULT;
1931
1932         if (!fid_is_sane(&body->fid1)) {
1933                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1934                 return -EINVAL;
1935         }
1936
1937         /*
1938          * Do not get size or any capa fields before we check that request
1939          * contains capa actually. There are some requests which do not, for
1940          * instance MDS_IS_SUBDIR.
1941          */
1942         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1943             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1944                 mdt_set_capainfo(info, 0, &body->fid1,
1945                                  req_capsule_client_get(pill, &RMF_CAPA1));
1946
1947         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1948         if (!IS_ERR(obj)) {
1949                 if ((flags & HABEO_CORPUS) &&
1950                     !mdt_object_exists(obj)) {
1951                         mdt_object_put(env, obj);
1952                         /* for capability renew ENOENT will be handled in
1953                          * mdt_renew_capa */
1954                         if (body->valid & OBD_MD_FLOSSCAPA)
1955                                 rc = 0;
1956                         else
1957                                 rc = -ENOENT;
1958                 } else {
1959                         info->mti_object = obj;
1960                         rc = 0;
1961                 }
1962         } else
1963                 rc = PTR_ERR(obj);
1964
1965         return rc;
1966 }
1967
1968 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
1969 {
1970         struct req_capsule *pill;
1971         int rc;
1972
1973         ENTRY;
1974         pill = &info->mti_pill;
1975
1976         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
1977                 rc = mdt_body_unpack(info, flags);
1978         else
1979                 rc = 0;
1980
1981         if (rc == 0 && (flags & HABEO_REFERO)) {
1982                 struct mdt_device *mdt = info->mti_mdt;
1983
1984                 /* Pack reply. */
1985                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1986                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1987                                              mdt->mdt_max_mdsize);
1988                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1989                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1990                                              mdt->mdt_max_cookiesize);
1991
1992                 rc = req_capsule_pack(pill);
1993         }
1994         RETURN(rc);
1995 }
1996
1997 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
1998 {
1999         struct md_device *next = m->mdt_child;
2000
2001         return next->md_ops->mdo_init_capa_ctxt(env, next,
2002                                                 m->mdt_opts.mo_mds_capa,
2003                                                 m->mdt_capa_timeout,
2004                                                 m->mdt_capa_alg,
2005                                                 m->mdt_capa_keys);
2006 }
2007
2008 /*
2009  * Invoke handler for this request opc. Also do necessary preprocessing
2010  * (according to handler ->mh_flags), and post-processing (setting of
2011  * ->last_{xid,committed}).
2012  */
2013 static int mdt_req_handle(struct mdt_thread_info *info,
2014                           struct mdt_handler *h, struct ptlrpc_request *req)
2015 {
2016         int   rc, serious = 0;
2017         __u32 flags;
2018
2019         ENTRY;
2020
2021         LASSERT(h->mh_act != NULL);
2022         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2023         LASSERT(current->journal_info == NULL);
2024
2025         /*
2026          * Mask out OBD_FAIL_ONCE, because that will stop
2027          * correct handling of failed req later in ldlm due to doing
2028          * obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED without actually
2029          * correct actions like it is done in target_send_reply_msg().
2030          */
2031         if (h->mh_fail_id != 0) {
2032                 /*
2033                  * Set to info->mti_fail_id to handler fail_id, it will be used
2034                  * later, and better than use default fail_id.
2035                  */
2036                 if (OBD_FAIL_CHECK(h->mh_fail_id && OBD_FAIL_MASK_LOC)) {
2037                         info->mti_fail_id = h->mh_fail_id;
2038                         RETURN(0);
2039                 }
2040         }
2041
2042         rc = 0;
2043         flags = h->mh_flags;
2044         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2045
2046         if (h->mh_fmt != NULL) {
2047                 req_capsule_set(&info->mti_pill, h->mh_fmt);
2048                 rc = mdt_unpack_req_pack_rep(info, flags);
2049         }
2050
2051         if (rc == 0 && flags & MUTABOR &&
2052             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2053                 /* should it be rq_status? */
2054                 rc = -EROFS;
2055
2056         if (rc == 0 && flags & HABEO_CLAVIS) {
2057                 struct ldlm_request *dlm_req;
2058
2059                 LASSERT(h->mh_fmt != NULL);
2060
2061                 dlm_req = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ);
2062                 if (dlm_req != NULL) {
2063                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2064                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2065                                                              dlm_req);
2066                         info->mti_dlm_req = dlm_req;
2067                 } else {
2068                         CERROR("Can't unpack dlm request\n");
2069                         rc = -EFAULT;
2070                 }
2071         }
2072
2073         /* capability setting changed via /proc, needs reinitialize ctxt */
2074         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2075                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2076                 info->mti_mdt->mdt_capa_conf = 0;
2077         }
2078
2079         if (likely(rc == 0)) {
2080                 /*
2081                  * Process request, there can be two types of rc:
2082                  * 1) errors with msg unpack/pack, other failures outside the
2083                  * operation itself. This is counted as serious errors;
2084                  * 2) errors during fs operation, should be placed in rq_status
2085                  * only
2086                  */
2087                 rc = h->mh_act(info);
2088                 serious = is_serious(rc);
2089                 rc = clear_serious(rc);
2090         } else
2091                 serious = 1;
2092
2093         req->rq_status = rc;
2094
2095         /*
2096          * ELDLM_* codes which > 0 should be in rq_status only as well as
2097          * all non-serious errors.
2098          */
2099         if (rc > 0 || !serious)
2100                 rc = 0;
2101
2102         LASSERT(current->journal_info == NULL);
2103
2104         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2105             info->mti_mdt->mdt_opts.mo_compat_resname) {
2106                 struct ldlm_reply *dlmrep;
2107
2108                 dlmrep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2109                 if (dlmrep != NULL)
2110                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2111         }
2112
2113         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2114         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2115                 target_committed_to_req(req);
2116
2117         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2118                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2119                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2120                 LBUG();
2121         }
2122
2123         RETURN(rc);
2124 }
2125
2126 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2127 {
2128         lh->mlh_type = MDT_NUL_LOCK;
2129         lh->mlh_reg_lh.cookie = 0ull;
2130         lh->mlh_reg_mode = LCK_MINMODE;
2131         lh->mlh_pdo_lh.cookie = 0ull;
2132         lh->mlh_pdo_mode = LCK_MINMODE;
2133 }
2134
2135 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2136 {
2137         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2138         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2139 }
2140
2141 /*
2142  * Initialize fields of struct mdt_thread_info. Other fields are left in
2143  * uninitialized state, because it's too expensive to zero out whole
2144  * mdt_thread_info (> 1K) on each request arrival.
2145  */
2146 static void mdt_thread_info_init(struct ptlrpc_request *req,
2147                                  struct mdt_thread_info *info)
2148 {
2149         int i;
2150         struct md_capainfo *ci;
2151
2152         info->mti_rep_buf_nr = ARRAY_SIZE(info->mti_rep_buf_size);
2153         for (i = 0; i < ARRAY_SIZE(info->mti_rep_buf_size); i++)
2154                 info->mti_rep_buf_size[i] = -1;
2155         req_capsule_init(&info->mti_pill, req, RCL_SERVER,
2156                          info->mti_rep_buf_size);
2157
2158         /* lock handle */
2159         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2160                 mdt_lock_handle_init(&info->mti_lh[i]);
2161
2162         /* mdt device: it can be NULL while CONNECT */
2163         if (req->rq_export) {
2164                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2165                 info->mti_exp = req->rq_export;
2166         } else
2167                 info->mti_mdt = NULL;
2168         info->mti_env = req->rq_svc_thread->t_env;
2169         ci = md_capainfo(info->mti_env);
2170         memset(ci, 0, sizeof *ci);
2171
2172         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2173         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2174
2175         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2176         info->mti_body = NULL;
2177         info->mti_object = NULL;
2178         info->mti_dlm_req = NULL;
2179         info->mti_has_trans = 0;
2180         info->mti_no_need_trans = 0;
2181         info->mti_cross_ref = 0;
2182         info->mti_opdata = 0;
2183
2184         /* To not check for split by default. */
2185         info->mti_spec.sp_ck_split = 0;
2186 }
2187
2188 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2189 {
2190         int i;
2191
2192         req_capsule_fini(&info->mti_pill);
2193         if (info->mti_object != NULL) {
2194                 mdt_object_put(info->mti_env, info->mti_object);
2195                 info->mti_object = NULL;
2196         }
2197         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2198                 mdt_lock_handle_fini(&info->mti_lh[i]);
2199         info->mti_env = NULL;
2200 }
2201
2202 /* mds/handler.c */
2203 extern int mds_filter_recovery_request(struct ptlrpc_request *req,
2204                                        struct obd_device *obd, int *process);
2205 /*
2206  * Handle recovery. Return:
2207  *        +1: continue request processing;
2208  *       -ve: abort immediately with the given error code;
2209  *         0: send reply with error code in req->rq_status;
2210  */
2211 static int mdt_recovery(struct mdt_thread_info *info)
2212 {
2213         struct ptlrpc_request *req = mdt_info_req(info);
2214         int recovering;
2215         struct obd_device *obd;
2216
2217         ENTRY;
2218
2219         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2220         case MDS_CONNECT:
2221         case SEC_CTX_INIT:
2222         case SEC_CTX_INIT_CONT:
2223         case SEC_CTX_FINI:
2224                 {
2225 #if 0
2226                         int rc;
2227
2228                         rc = mdt_handle_idmap(info);
2229                         if (rc)
2230                                 RETURN(rc);
2231                         else
2232 #endif
2233                                 RETURN(+1);
2234                 }
2235         }
2236
2237         if (unlikely(req->rq_export == NULL)) {
2238                 CERROR("operation %d on unconnected MDS from %s\n",
2239                        lustre_msg_get_opc(req->rq_reqmsg),
2240                        libcfs_id2str(req->rq_peer));
2241                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2242                  * mds_A will get -ENOTCONN(especially for ping req),
2243                  * which will cause that mds_A deactive timeout, then when
2244                  * mds_A cleanup, the cleanup process will be suspended since
2245                  * deactive timeout is not zero.
2246                  */
2247                 req->rq_status = -ENOTCONN;
2248                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2249                 RETURN(0);
2250         }
2251
2252         /* sanity check: if the xid matches, the request must be marked as a
2253          * resent or replayed */
2254         if (req_xid_is_last(req)) {
2255                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2256                       (MSG_RESENT | MSG_REPLAY))) {
2257                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2258                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2259                                   lustre_msg_get_flags(req->rq_reqmsg));
2260                         LBUG();
2261                         req->rq_status = -ENOTCONN;
2262                         RETURN(-ENOTCONN);
2263                 }
2264         }
2265
2266         /* else: note the opposite is not always true; a RESENT req after a
2267          * failover will usually not match the last_xid, since it was likely
2268          * never committed. A REPLAYed request will almost never match the
2269          * last xid, however it could for a committed, but still retained,
2270          * open. */
2271
2272         obd = req->rq_export->exp_obd;
2273
2274         /* Check for aborted recovery... */
2275         spin_lock_bh(&obd->obd_processing_task_lock);
2276         recovering = obd->obd_recovering;
2277         spin_unlock_bh(&obd->obd_processing_task_lock);
2278         if (unlikely(recovering)) {
2279                 int rc;
2280                 int should_process;
2281                 DEBUG_REQ(D_INFO, req, "Got new replay");
2282                 rc = mds_filter_recovery_request(req, obd, &should_process);
2283                 if (rc != 0 || !should_process)
2284                         RETURN(rc);
2285                 else if (should_process < 0) {
2286                         req->rq_status = should_process;
2287                         rc = ptlrpc_error(req);
2288                         RETURN(rc);
2289                 }
2290         }
2291         RETURN(+1);
2292 }
2293
2294 static int mdt_reply(struct ptlrpc_request *req, int rc,
2295                      struct mdt_thread_info *info)
2296 {
2297         ENTRY;
2298
2299 #if 0
2300         if (req->rq_reply_state == NULL && rc == 0) {
2301                 req->rq_status = rc;
2302                 lustre_pack_reply(req, 1, NULL, NULL);
2303         }
2304 #endif
2305         target_send_reply(req, rc, info->mti_fail_id);
2306         RETURN(0);
2307 }
2308
2309 /* mds/handler.c */
2310 extern int mds_msg_check_version(struct lustre_msg *msg);
2311
2312 static int mdt_handle0(struct ptlrpc_request *req,
2313                        struct mdt_thread_info *info,
2314                        struct mdt_opc_slice *supported)
2315 {
2316         struct mdt_handler *h;
2317         struct lustre_msg  *msg;
2318         int                 rc;
2319
2320         ENTRY;
2321
2322         OBD_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0);
2323
2324         LASSERT(current->journal_info == NULL);
2325
2326         msg = req->rq_reqmsg;
2327         rc = mds_msg_check_version(msg);
2328         if (likely(rc == 0)) {
2329                 rc = mdt_recovery(info);
2330                 if (likely(rc == +1)) {
2331                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2332                                              supported);
2333                         if (likely(h != NULL)) {
2334                                 rc = mdt_req_handle(info, h, req);
2335                                 rc = mdt_reply(req, rc, info);
2336                         } else {
2337                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2338                                 req->rq_status = -ENOTSUPP;
2339                                 rc = ptlrpc_error(req);
2340                                 RETURN(rc);
2341                         }
2342                 }
2343         } else
2344                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2345         RETURN(rc);
2346 }
2347
2348 /*
2349  * MDT handler function called by ptlrpc service thread when request comes.
2350  *
2351  * XXX common "target" functionality should be factored into separate module
2352  * shared by mdt, ost and stand-alone services like fld.
2353  */
2354 static int mdt_handle_common(struct ptlrpc_request *req,
2355                              struct mdt_opc_slice *supported)
2356 {
2357         struct lu_env          *env;
2358         struct mdt_thread_info *info;
2359         int                     rc;
2360         ENTRY;
2361
2362         env = req->rq_svc_thread->t_env;
2363         LASSERT(env != NULL);
2364         LASSERT(env->le_ses != NULL);
2365         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2366         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2367         LASSERT(info != NULL);
2368
2369         mdt_thread_info_init(req, info);
2370
2371         rc = mdt_handle0(req, info, supported);
2372
2373         mdt_thread_info_fini(info);
2374         RETURN(rc);
2375 }
2376
2377 /*
2378  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2379  * as well.
2380  */
2381 int mdt_recovery_handle(struct ptlrpc_request *req)
2382 {
2383         int rc;
2384         ENTRY;
2385
2386         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2387         case FLD_QUERY:
2388                 rc = mdt_handle_common(req, mdt_fld_handlers);
2389                 break;
2390         case SEQ_QUERY:
2391                 rc = mdt_handle_common(req, mdt_seq_handlers);
2392                 break;
2393         default:
2394                 rc = mdt_handle_common(req, mdt_regular_handlers);
2395                 break;
2396         }
2397
2398         RETURN(rc);
2399 }
2400
2401 static int mdt_regular_handle(struct ptlrpc_request *req)
2402 {
2403         return mdt_handle_common(req, mdt_regular_handlers);
2404 }
2405
2406 static int mdt_readpage_handle(struct ptlrpc_request *req)
2407 {
2408         return mdt_handle_common(req, mdt_readpage_handlers);
2409 }
2410
2411 static int mdt_xmds_handle(struct ptlrpc_request *req)
2412 {
2413         return mdt_handle_common(req, mdt_xmds_handlers);
2414 }
2415
2416 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2417 {
2418         return mdt_handle_common(req, mdt_seq_handlers);
2419 }
2420
2421 static int mdt_mdss_handle(struct ptlrpc_request *req)
2422 {
2423         return mdt_handle_common(req, mdt_seq_handlers);
2424 }
2425
2426 static int mdt_dtss_handle(struct ptlrpc_request *req)
2427 {
2428         return mdt_handle_common(req, mdt_seq_handlers);
2429 }
2430
2431 static int mdt_fld_handle(struct ptlrpc_request *req)
2432 {
2433         return mdt_handle_common(req, mdt_fld_handlers);
2434 }
2435
2436 enum mdt_it_code {
2437         MDT_IT_OPEN,
2438         MDT_IT_OCREAT,
2439         MDT_IT_CREATE,
2440         MDT_IT_GETATTR,
2441         MDT_IT_READDIR,
2442         MDT_IT_LOOKUP,
2443         MDT_IT_UNLINK,
2444         MDT_IT_TRUNC,
2445         MDT_IT_GETXATTR,
2446         MDT_IT_NR
2447 };
2448
2449 static int mdt_intent_getattr(enum mdt_it_code opcode,
2450                               struct mdt_thread_info *info,
2451                               struct ldlm_lock **,
2452                               int);
2453 static int mdt_intent_reint(enum mdt_it_code opcode,
2454                             struct mdt_thread_info *info,
2455                             struct ldlm_lock **,
2456                             int);
2457
2458 static struct mdt_it_flavor {
2459         const struct req_format *it_fmt;
2460         __u32                    it_flags;
2461         int                    (*it_act)(enum mdt_it_code ,
2462                                          struct mdt_thread_info *,
2463                                          struct ldlm_lock **,
2464                                          int);
2465         long                     it_reint;
2466 } mdt_it_flavor[] = {
2467         [MDT_IT_OPEN]     = {
2468                 .it_fmt   = &RQF_LDLM_INTENT,
2469                 /*.it_flags = HABEO_REFERO,*/
2470                 .it_flags = 0,
2471                 .it_act   = mdt_intent_reint,
2472                 .it_reint = REINT_OPEN
2473         },
2474         [MDT_IT_OCREAT]   = {
2475                 .it_fmt   = &RQF_LDLM_INTENT,
2476                 .it_flags = MUTABOR,
2477                 .it_act   = mdt_intent_reint,
2478                 .it_reint = REINT_OPEN
2479         },
2480         [MDT_IT_CREATE]   = {
2481                 .it_fmt   = &RQF_LDLM_INTENT,
2482                 .it_flags = MUTABOR,
2483                 .it_act   = mdt_intent_reint,
2484                 .it_reint = REINT_CREATE
2485         },
2486         [MDT_IT_GETATTR]  = {
2487                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2488                 .it_flags = HABEO_REFERO,
2489                 .it_act   = mdt_intent_getattr
2490         },
2491         [MDT_IT_READDIR]  = {
2492                 .it_fmt   = NULL,
2493                 .it_flags = 0,
2494                 .it_act   = NULL
2495         },
2496         [MDT_IT_LOOKUP]   = {
2497                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2498                 .it_flags = HABEO_REFERO,
2499                 .it_act   = mdt_intent_getattr
2500         },
2501         [MDT_IT_UNLINK]   = {
2502                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2503                 .it_flags = MUTABOR,
2504                 .it_act   = NULL,
2505                 .it_reint = REINT_UNLINK
2506         },
2507         [MDT_IT_TRUNC]    = {
2508                 .it_fmt   = NULL,
2509                 .it_flags = MUTABOR,
2510                 .it_act   = NULL
2511         },
2512         [MDT_IT_GETXATTR] = {
2513                 .it_fmt   = NULL,
2514                 .it_flags = 0,
2515                 .it_act   = NULL
2516         }
2517 };
2518
2519 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2520                             struct ldlm_lock **lockp,
2521                             struct ldlm_lock *new_lock,
2522                             struct mdt_lock_handle *lh,
2523                             int flags)
2524 {
2525         struct ptlrpc_request  *req = mdt_info_req(info);
2526         struct ldlm_lock       *lock = *lockp;
2527
2528         /*
2529          * Get new lock only for cases when possible resent did not find any
2530          * lock.
2531          */
2532         if (new_lock == NULL)
2533                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2534
2535         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2536                 lh->mlh_reg_lh.cookie = 0;
2537                 RETURN(0);
2538         }
2539
2540         LASSERTF(new_lock != NULL,
2541                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2542
2543         /*
2544          * If we've already given this lock to a client once, then we should
2545          * have no readers or writers.  Otherwise, we should have one reader
2546          * _or_ writer ref (which will be zeroed below) before returning the
2547          * lock to a client.
2548          */
2549         if (new_lock->l_export == req->rq_export) {
2550                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2551         } else {
2552                 LASSERT(new_lock->l_export == NULL);
2553                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2554         }
2555
2556         *lockp = new_lock;
2557
2558         if (new_lock->l_export == req->rq_export) {
2559                 /*
2560                  * Already gave this to the client, which means that we
2561                  * reconstructed a reply.
2562                  */
2563                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2564                         MSG_RESENT);
2565                 lh->mlh_reg_lh.cookie = 0;
2566                 RETURN(ELDLM_LOCK_REPLACED);
2567         }
2568
2569         /* This lock might already be given to the client by an resent req,
2570          * in this case we should return ELDLM_LOCK_ABORTED,
2571          * so we should check led_held_locks here, but it will affect
2572          * performance, FIXME
2573          */
2574         /* Fixup the lock to be given to the client */
2575         lock_res_and_lock(new_lock);
2576         new_lock->l_readers = 0;
2577         new_lock->l_writers = 0;
2578
2579         new_lock->l_export = class_export_get(req->rq_export);
2580         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2581         list_add(&new_lock->l_export_chain,
2582                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2583         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2584
2585         new_lock->l_blocking_ast = lock->l_blocking_ast;
2586         new_lock->l_completion_ast = lock->l_completion_ast;
2587         new_lock->l_remote_handle = lock->l_remote_handle;
2588         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2589
2590         unlock_res_and_lock(new_lock);
2591         LDLM_LOCK_PUT(new_lock);
2592         lh->mlh_reg_lh.cookie = 0;
2593
2594         RETURN(ELDLM_LOCK_REPLACED);
2595 }
2596
2597 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2598                                     struct ldlm_lock *new_lock,
2599                                     struct ldlm_lock **old_lock,
2600                                     struct mdt_lock_handle *lh)
2601 {
2602         struct ptlrpc_request  *req = mdt_info_req(info);
2603         struct obd_export      *exp = req->rq_export;
2604         struct lustre_handle    remote_hdl;
2605         struct ldlm_request    *dlmreq;
2606         struct list_head       *iter;
2607
2608         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2609                 return;
2610
2611         dlmreq = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ);
2612         remote_hdl = dlmreq->lock_handle[0];
2613
2614         spin_lock(&exp->exp_ldlm_data.led_lock);
2615         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2616                 struct ldlm_lock *lock;
2617                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2618                 if (lock == new_lock)
2619                         continue;
2620                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2621                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2622                         lh->mlh_reg_mode = lock->l_granted_mode;
2623
2624                         LDLM_DEBUG(lock, "restoring lock cookie");
2625                         DEBUG_REQ(D_DLMTRACE, req,
2626                                   "restoring lock cookie "LPX64,
2627                                   lh->mlh_reg_lh.cookie);
2628                         if (old_lock)
2629                                 *old_lock = LDLM_LOCK_GET(lock);
2630                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2631                         return;
2632                 }
2633         }
2634         spin_unlock(&exp->exp_ldlm_data.led_lock);
2635
2636         /*
2637          * If the xid matches, then we know this is a resent request, and allow
2638          * it. (It's probably an OPEN, for which we don't send a lock.
2639          */
2640         if (req_xid_is_last(req))
2641                 return;
2642
2643         /*
2644          * This remote handle isn't enqueued, so we never received or processed
2645          * this request.  Clear MSG_RESENT, because it can be handled like any
2646          * normal request now.
2647          */
2648         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2649
2650         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2651                   remote_hdl.cookie);
2652 }
2653
2654 static int mdt_intent_getattr(enum mdt_it_code opcode,
2655                               struct mdt_thread_info *info,
2656                               struct ldlm_lock **lockp,
2657                               int flags)
2658 {
2659         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2660         struct ldlm_lock       *new_lock = NULL;
2661         __u64                   child_bits;
2662         struct ldlm_reply      *ldlm_rep;
2663         struct ptlrpc_request  *req;
2664         struct mdt_body        *reqbody;
2665         struct mdt_body        *repbody;
2666         int                     rc;
2667         ENTRY;
2668
2669         reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
2670         LASSERT(reqbody);
2671
2672         repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
2673         LASSERT(repbody);
2674
2675         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2676         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2677         repbody->eadatasize = 0;
2678         repbody->aclsize = 0;
2679
2680         switch (opcode) {
2681         case MDT_IT_LOOKUP:
2682                 child_bits = MDS_INODELOCK_LOOKUP;
2683                 break;
2684         case MDT_IT_GETATTR:
2685                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2686                 break;
2687         default:
2688                 CERROR("Unhandled till now");
2689                 GOTO(out, rc = -EINVAL);
2690         }
2691
2692         rc = mdt_init_ucred(info, reqbody);
2693         if (rc)
2694                 GOTO(out, rc);
2695
2696         req = info->mti_pill.rc_req;
2697         ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2698         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2699
2700         /* Get lock from request for possible resent case. */
2701         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2702
2703         ldlm_rep->lock_policy_res2 =
2704                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2705
2706         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2707                 ldlm_rep->lock_policy_res2 = 0;
2708         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2709             ldlm_rep->lock_policy_res2) {
2710                 lhc->mlh_reg_lh.cookie = 0ull;
2711                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2712         }
2713
2714         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2715         EXIT;
2716 out_ucred:
2717         mdt_exit_ucred(info);
2718 out:
2719         mdt_shrink_reply(info);
2720         return rc;
2721 }
2722
2723 static int mdt_intent_reint(enum mdt_it_code opcode,
2724                             struct mdt_thread_info *info,
2725                             struct ldlm_lock **lockp,
2726                             int flags)
2727 {
2728         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2729         struct ldlm_reply      *rep = NULL;
2730         long                    opc;
2731         int                     rc;
2732
2733         static const struct req_format *intent_fmts[REINT_MAX] = {
2734                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2735                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2736         };
2737
2738         ENTRY;
2739
2740         opc = mdt_reint_opcode(info, intent_fmts);
2741         if (opc < 0)
2742                 GOTO(out, rc = opc);
2743
2744         if (mdt_it_flavor[opcode].it_reint != opc) {
2745                 CERROR("Reint code %ld doesn't match intent: %d\n",
2746                        opc, opcode);
2747                 GOTO(out, rc = err_serious(-EPROTO));
2748         }
2749
2750         /* Get lock from request for possible resent case. */
2751         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2752
2753         rc = mdt_reint_internal(info, lhc, opc);
2754
2755         /* Check whether the reply has been packed successfully. */
2756         if (mdt_info_req(info)->rq_repmsg != NULL)
2757                 rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2758         if (rep == NULL)
2759                 GOTO(out, rc = err_serious(-EFAULT));
2760
2761         /* MDC expects this in any case */
2762         if (rc != 0)
2763                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2764
2765         /* Cross-ref case, the lock should be returned to the client */
2766         if (rc == -EREMOTE) {
2767                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2768                 rep->lock_policy_res2 = 0;
2769                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2770                 GOTO(out, rc);
2771         }
2772         rep->lock_policy_res2 = clear_serious(rc);
2773
2774         lhc->mlh_reg_lh.cookie = 0ull;
2775         rc = ELDLM_LOCK_ABORTED;
2776         EXIT;
2777 out:
2778         return rc;
2779 }
2780
2781 static int mdt_intent_code(long itcode)
2782 {
2783         int rc;
2784
2785         switch(itcode) {
2786         case IT_OPEN:
2787                 rc = MDT_IT_OPEN;
2788                 break;
2789         case IT_OPEN|IT_CREAT:
2790                 rc = MDT_IT_OCREAT;
2791                 break;
2792         case IT_CREAT:
2793                 rc = MDT_IT_CREATE;
2794                 break;
2795         case IT_READDIR:
2796                 rc = MDT_IT_READDIR;
2797                 break;
2798         case IT_GETATTR:
2799                 rc = MDT_IT_GETATTR;
2800                 break;
2801         case IT_LOOKUP:
2802                 rc = MDT_IT_LOOKUP;
2803                 break;
2804         case IT_UNLINK:
2805                 rc = MDT_IT_UNLINK;
2806                 break;
2807         case IT_TRUNC:
2808                 rc = MDT_IT_TRUNC;
2809                 break;
2810         case IT_GETXATTR:
2811                 rc = MDT_IT_GETXATTR;
2812                 break;
2813         default:
2814                 CERROR("Unknown intent opcode: %ld\n", itcode);
2815                 rc = -EINVAL;
2816                 break;
2817         }
2818         return rc;
2819 }
2820
2821 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2822                           struct ldlm_lock **lockp, int flags)
2823 {
2824         struct req_capsule   *pill;
2825         struct mdt_it_flavor *flv;
2826         int opc;
2827         int rc;
2828         ENTRY;
2829
2830         opc = mdt_intent_code(itopc);
2831         if (opc < 0)
2832                 RETURN(-EINVAL);
2833
2834         pill = &info->mti_pill;
2835         flv  = &mdt_it_flavor[opc];
2836
2837         if (flv->it_fmt != NULL)
2838                 req_capsule_extend(pill, flv->it_fmt);
2839
2840         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2841         if (rc == 0) {
2842                 struct ptlrpc_request *req = mdt_info_req(info);
2843                 if (flv->it_flags & MUTABOR &&
2844                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2845                         rc = -EROFS;
2846         }
2847         if (rc == 0 && flv->it_act != NULL) {
2848                 /* execute policy */
2849                 rc = flv->it_act(opc, info, lockp, flags);
2850         } else
2851                 rc = -EOPNOTSUPP;
2852         RETURN(rc);
2853 }
2854
2855 static int mdt_intent_policy(struct ldlm_namespace *ns,
2856                              struct ldlm_lock **lockp, void *req_cookie,
2857                              ldlm_mode_t mode, int flags, void *data)
2858 {
2859         struct mdt_thread_info *info;
2860         struct ptlrpc_request  *req  =  req_cookie;
2861         struct ldlm_intent     *it;
2862         struct req_capsule     *pill;
2863         int rc;
2864
2865         ENTRY;
2866
2867         LASSERT(req != NULL);
2868
2869         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2870                                   &mdt_thread_key);
2871         LASSERT(info != NULL);
2872         pill = &info->mti_pill;
2873         LASSERT(pill->rc_req == req);
2874
2875         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
2876                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
2877                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
2878                 if (it != NULL) {
2879                         const struct ldlm_request *dlmreq;
2880                         __u64 req_bits;
2881 #if 0
2882                         struct ldlm_lock       *lock = *lockp;
2883
2884                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
2885                                    ldlm_it2str(it->opc));
2886 #endif
2887
2888                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
2889                         if (rc == 0)
2890                                 rc = ELDLM_OK;
2891
2892                         /*
2893                          * Lock without inodebits makes no sense and will oops
2894                          * later in ldlm. Let's check it now to see if we have
2895                          * wrong lock from client or bits get corrupted
2896                          * somewhere in mdt_intent_opc().
2897                          */
2898                         dlmreq = info->mti_dlm_req;
2899                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
2900                         LASSERT(req_bits != 0);
2901
2902                 } else
2903                         rc = err_serious(-EFAULT);
2904         } else {
2905                 /* No intent was provided */
2906                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
2907                 rc = req_capsule_pack(pill);
2908                 if (rc)
2909                         rc = err_serious(rc);
2910         }
2911         RETURN(rc);
2912 }
2913
2914 /*
2915  * Seq wrappers
2916  */
2917 static void mdt_seq_adjust(const struct lu_env *env,
2918                           struct mdt_device *m, int lost)
2919 {
2920         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
2921         struct lu_range out;
2922         ENTRY;
2923
2924         LASSERT(ls && ls->ls_server_seq);
2925         LASSERT(lost >= 0);
2926         /* get extra seq from seq_server, moving it's range up */
2927         while (lost-- > 0) {
2928                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
2929         }
2930         EXIT;
2931 }
2932
2933 static int mdt_seq_fini(const struct lu_env *env,
2934                         struct mdt_device *m)
2935 {
2936         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
2937         ENTRY;
2938
2939         if (ls && ls->ls_server_seq) {
2940                 seq_server_fini(ls->ls_server_seq, env);
2941                 OBD_FREE_PTR(ls->ls_server_seq);
2942                 ls->ls_server_seq = NULL;
2943         }
2944
2945         if (ls && ls->ls_control_seq) {
2946                 seq_server_fini(ls->ls_control_seq, env);
2947                 OBD_FREE_PTR(ls->ls_control_seq);
2948                 ls->ls_control_seq = NULL;
2949         }
2950
2951         if (ls && ls->ls_client_seq) {
2952                 seq_client_fini(ls->ls_client_seq);
2953                 OBD_FREE_PTR(ls->ls_client_seq);
2954                 ls->ls_client_seq = NULL;
2955         }
2956
2957         RETURN(0);
2958 }
2959
2960 static int mdt_seq_init(const struct lu_env *env,
2961                         const char *uuid,
2962                         struct mdt_device *m)
2963 {
2964         struct lu_site *ls;
2965         char *prefix;
2966         int rc;
2967         ENTRY;
2968
2969         ls = m->mdt_md_dev.md_lu_dev.ld_site;
2970
2971         /*
2972          * This is sequence-controller node. Init seq-controller server on local
2973          * MDT.
2974          */
2975         if (ls->ls_node_id == 0) {
2976                 LASSERT(ls->ls_control_seq == NULL);
2977
2978                 OBD_ALLOC_PTR(ls->ls_control_seq);
2979                 if (ls->ls_control_seq == NULL)
2980                         RETURN(-ENOMEM);
2981
2982                 rc = seq_server_init(ls->ls_control_seq,
2983                                      m->mdt_bottom, uuid,
2984                                      LUSTRE_SEQ_CONTROLLER,
2985                                      env);
2986
2987                 if (rc)
2988                         GOTO(out_seq_fini, rc);
2989
2990                 OBD_ALLOC_PTR(ls->ls_client_seq);
2991                 if (ls->ls_client_seq == NULL)
2992                         GOTO(out_seq_fini, rc = -ENOMEM);
2993
2994                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
2995                 if (prefix == NULL) {
2996                         OBD_FREE_PTR(ls->ls_client_seq);
2997                         GOTO(out_seq_fini, rc = -ENOMEM);
2998                 }
2999
3000                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3001                          uuid);
3002
3003                 /*
3004                  * Init seq-controller client after seq-controller server is
3005                  * ready. Pass ls->ls_control_seq to it for direct talking.
3006                  */
3007                 rc = seq_client_init(ls->ls_client_seq, NULL,
3008                                      LUSTRE_SEQ_METADATA, prefix,
3009                                      ls->ls_control_seq);
3010                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3011
3012                 if (rc)
3013                         GOTO(out_seq_fini, rc);
3014         }
3015
3016         /* Init seq-server on local MDT */
3017         LASSERT(ls->ls_server_seq == NULL);
3018
3019         OBD_ALLOC_PTR(ls->ls_server_seq);
3020         if (ls->ls_server_seq == NULL)
3021                 GOTO(out_seq_fini, rc = -ENOMEM);
3022
3023         rc = seq_server_init(ls->ls_server_seq,
3024                              m->mdt_bottom, uuid,
3025                              LUSTRE_SEQ_SERVER,
3026                              env);
3027         if (rc)
3028                 GOTO(out_seq_fini, rc = -ENOMEM);
3029
3030         /* Assign seq-controller client to local seq-server. */
3031         if (ls->ls_node_id == 0) {
3032                 LASSERT(ls->ls_client_seq != NULL);
3033
3034                 rc = seq_server_set_cli(ls->ls_server_seq,
3035                                         ls->ls_client_seq,
3036                                         env);
3037         }
3038
3039         EXIT;
3040 out_seq_fini:
3041         if (rc)
3042                 mdt_seq_fini(env, m);
3043
3044         return rc;
3045 }
3046 /*
3047  * Init client sequence manager which is used by local MDS to talk to sequence
3048  * controller on remote node.
3049  */
3050 static int mdt_seq_init_cli(const struct lu_env *env,
3051                             struct mdt_device *m,
3052                             struct lustre_cfg *cfg)
3053 {
3054         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3055         struct obd_device *mdc;
3056         struct obd_uuid   *uuidp, *mdcuuidp;
3057         char              *uuid_str, *mdc_uuid_str;
3058         int                rc;
3059         int                index;
3060         struct mdt_thread_info *info;
3061         char *p, *index_string = lustre_cfg_string(cfg, 2);
3062         ENTRY;
3063
3064         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3065         uuidp = &info->mti_u.uuid[0];
3066         mdcuuidp = &info->mti_u.uuid[1];
3067
3068         LASSERT(index_string);
3069
3070         index = simple_strtol(index_string, &p, 10);
3071         if (*p) {
3072                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3073                 RETURN(-EINVAL);
3074         }
3075
3076         /* check if this is adding the first MDC and controller is not yet
3077          * initialized. */
3078         if (index != 0 || ls->ls_client_seq)
3079                 RETURN(0);
3080
3081         uuid_str = lustre_cfg_string(cfg, 1);
3082         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3083         obd_str2uuid(uuidp, uuid_str);
3084         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3085
3086         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3087         if (!mdc) {
3088                 CERROR("can't find controller MDC by uuid %s\n",
3089                        uuid_str);
3090                 rc = -ENOENT;
3091         } else if (!mdc->obd_set_up) {
3092                 CERROR("target %s not set up\n", mdc->obd_name);
3093                 rc = -EINVAL;
3094         } else {
3095                 LASSERT(ls->ls_control_exp);
3096                 OBD_ALLOC_PTR(ls->ls_client_seq);
3097                 if (ls->ls_client_seq != NULL) {
3098                         char *prefix;
3099
3100                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3101                         if (!prefix)
3102                                 RETURN(-ENOMEM);
3103
3104                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3105                                  mdc->obd_name);
3106
3107                         rc = seq_client_init(ls->ls_client_seq,
3108                                              ls->ls_control_exp,
3109                                              LUSTRE_SEQ_METADATA,
3110                                              prefix, NULL);
3111                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3112                 } else
3113                         rc = -ENOMEM;
3114
3115                 if (rc)
3116                         RETURN(rc);
3117
3118                 LASSERT(ls->ls_server_seq != NULL);
3119                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3120                                         env);
3121         }
3122
3123         RETURN(rc);
3124 }
3125
3126 static void mdt_seq_fini_cli(struct mdt_device *m)
3127 {
3128         struct lu_site *ls;
3129
3130         ENTRY;
3131
3132         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3133
3134         if (ls && ls->ls_server_seq)
3135                 seq_server_set_cli(ls->ls_server_seq,
3136                                    NULL, NULL);
3137
3138         if (ls && ls->ls_control_exp) {
3139                 class_export_put(ls->ls_control_exp);
3140                 ls->ls_control_exp = NULL;
3141         }
3142         EXIT;
3143 }
3144
3145 /*
3146  * FLD wrappers
3147  */
3148 static int mdt_fld_fini(const struct lu_env *env,
3149                         struct mdt_device *m)
3150 {
3151         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3152         ENTRY;
3153
3154         if (ls && ls->ls_server_fld) {
3155                 fld_server_fini(ls->ls_server_fld, env);
3156                 OBD_FREE_PTR(ls->ls_server_fld);
3157                 ls->ls_server_fld = NULL;
3158         }
3159
3160         RETURN(0);
3161 }
3162
3163 static int mdt_fld_init(const struct lu_env *env,
3164                         const char *uuid,
3165                         struct mdt_device *m)
3166 {
3167         struct lu_site *ls;
3168         int rc;
3169         ENTRY;
3170
3171         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3172
3173         OBD_ALLOC_PTR(ls->ls_server_fld);
3174         if (ls->ls_server_fld == NULL)
3175                 RETURN(rc = -ENOMEM);
3176
3177         rc = fld_server_init(ls->ls_server_fld,
3178                              m->mdt_bottom, uuid, env);
3179         if (rc) {
3180                 OBD_FREE_PTR(ls->ls_server_fld);
3181                 ls->ls_server_fld = NULL;
3182                 RETURN(rc);
3183         }
3184
3185         RETURN(0);
3186 }
3187
3188 /* device init/fini methods */
3189 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3190 {
3191         ENTRY;
3192         if (m->mdt_regular_service != NULL) {
3193                 ptlrpc_unregister_service(m->mdt_regular_service);
3194                 m->mdt_regular_service = NULL;
3195         }
3196         if (m->mdt_readpage_service != NULL) {
3197                 ptlrpc_unregister_service(m->mdt_readpage_service);
3198                 m->mdt_readpage_service = NULL;
3199         }
3200         if (m->mdt_xmds_service != NULL) {
3201                 ptlrpc_unregister_service(m->mdt_xmds_service);
3202                 m->mdt_xmds_service = NULL;
3203         }
3204         if (m->mdt_setattr_service != NULL) {
3205                 ptlrpc_unregister_service(m->mdt_setattr_service);
3206                 m->mdt_setattr_service = NULL;
3207         }
3208         if (m->mdt_mdsc_service != NULL) {
3209                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3210                 m->mdt_mdsc_service = NULL;
3211         }
3212         if (m->mdt_mdss_service != NULL) {
3213                 ptlrpc_unregister_service(m->mdt_mdss_service);
3214                 m->mdt_mdss_service = NULL;
3215         }
3216         if (m->mdt_dtss_service != NULL) {
3217                 ptlrpc_unregister_service(m->mdt_dtss_service);
3218                 m->mdt_dtss_service = NULL;
3219         }
3220         if (m->mdt_fld_service != NULL) {
3221                 ptlrpc_unregister_service(m->mdt_fld_service);
3222                 m->mdt_fld_service = NULL;
3223         }
3224         ENTRY;
3225 }
3226
3227 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3228 {
3229         int rc;
3230         static struct ptlrpc_service_conf conf;
3231         cfs_proc_dir_entry_t *procfs_entry;
3232         ENTRY;
3233
3234         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3235
3236         conf = (typeof(conf)) {
3237                 .psc_nbufs            = MDS_NBUFS,
3238                 .psc_bufsize          = MDS_BUFSIZE,
3239                 .psc_max_req_size     = MDS_MAXREQSIZE,
3240                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3241                 .psc_req_portal       = MDS_REQUEST_PORTAL,
3242                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3243                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3244                 /*
3245                  * We'd like to have a mechanism to set this on a per-device
3246                  * basis, but alas...
3247                  */
3248                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3249                                        MDT_MAX_THREADS),
3250                 .psc_max_threads   = MDT_MAX_THREADS,
3251                 .psc_ctx_tags      = LCT_MD_THREAD
3252         };
3253
3254         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3255         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3256                            "mdt_ldlm_client", m->mdt_ldlm_client);
3257
3258         m->mdt_regular_service =
3259                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3260                                      procfs_entry, NULL, LUSTRE_MDT_NAME);
3261         if (m->mdt_regular_service == NULL)
3262                 RETURN(-ENOMEM);
3263
3264         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3265         if (rc)
3266                 GOTO(err_mdt_svc, rc);
3267
3268         /*
3269          * readpage service configuration. Parameters have to be adjusted,
3270          * ideally.
3271          */
3272         conf = (typeof(conf)) {
3273                 .psc_nbufs            = MDS_NBUFS,
3274                 .psc_bufsize          = MDS_BUFSIZE,
3275                 .psc_max_req_size     = MDS_MAXREQSIZE,
3276                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3277                 .psc_req_portal       = MDS_READPAGE_PORTAL,
3278                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3279                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3280                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3281                                        MDT_MAX_THREADS),
3282                 .psc_max_threads   = MDT_MAX_THREADS,
3283                 .psc_ctx_tags      = LCT_MD_THREAD
3284         };
3285         m->mdt_readpage_service =
3286                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3287                                      LUSTRE_MDT_NAME "_readpage",
3288                                      procfs_entry, NULL, "mdt_rdpg");
3289
3290         if (m->mdt_readpage_service == NULL) {
3291                 CERROR("failed to start readpage service\n");
3292                 GOTO(err_mdt_svc, rc = -ENOMEM);
3293         }
3294
3295         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3296
3297         /*
3298          * setattr service configuration.
3299          */
3300         conf = (typeof(conf)) {
3301                 .psc_nbufs            = MDS_NBUFS,
3302                 .psc_bufsize          = MDS_BUFSIZE,
3303                 .psc_max_req_size     = MDS_MAXREQSIZE,
3304                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3305                 .psc_req_portal       = MDS_SETATTR_PORTAL,
3306                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3307                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3308                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3309                                        MDT_MAX_THREADS),
3310                 .psc_max_threads   = MDT_MAX_THREADS,
3311                 .psc_ctx_tags      = LCT_MD_THREAD
3312         };
3313
3314         m->mdt_setattr_service =
3315                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3316                                      LUSTRE_MDT_NAME "_setattr",
3317                                      procfs_entry, NULL, "mdt_attr");
3318
3319         if (!m->mdt_setattr_service) {
3320                 CERROR("failed to start setattr service\n");
3321                 GOTO(err_mdt_svc, rc = -ENOMEM);
3322         }
3323
3324         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3325         if (rc)
3326                 GOTO(err_mdt_svc, rc);
3327
3328         /*
3329          * sequence controller service configuration
3330          */
3331         conf = (typeof(conf)) {
3332                 .psc_nbufs = MDS_NBUFS,
3333                 .psc_bufsize = MDS_BUFSIZE,
3334                 .psc_max_req_size = SEQ_MAXREQSIZE,
3335                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3336                 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3337                 .psc_rep_portal = MDC_REPLY_PORTAL,
3338                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3339                 .psc_min_threads = SEQ_NUM_THREADS,
3340                 .psc_max_threads = SEQ_NUM_THREADS,
3341                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3342         };
3343
3344         m->mdt_mdsc_service =
3345                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3346                                      LUSTRE_MDT_NAME"_mdsc",
3347                                      procfs_entry, NULL, "mdt_mdsc");
3348         if (!m->mdt_mdsc_service) {
3349                 CERROR("failed to start seq controller service\n");
3350                 GOTO(err_mdt_svc, rc = -ENOMEM);
3351         }
3352
3353         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3354         if (rc)
3355                 GOTO(err_mdt_svc, rc);
3356
3357         /*
3358          * metadata sequence server service configuration
3359          */
3360         conf = (typeof(conf)) {
3361                 .psc_nbufs = MDS_NBUFS,
3362                 .psc_bufsize = MDS_BUFSIZE,
3363                 .psc_max_req_size = SEQ_MAXREQSIZE,
3364                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3365                 .psc_req_portal = SEQ_METADATA_PORTAL,
3366                 .psc_rep_portal = MDC_REPLY_PORTAL,
3367                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3368                 .psc_min_threads = SEQ_NUM_THREADS,
3369                 .psc_max_threads = SEQ_NUM_THREADS,
3370                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3371         };
3372
3373         m->mdt_mdss_service =
3374                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3375                                      LUSTRE_MDT_NAME"_mdss",
3376                                      procfs_entry, NULL, "mdt_mdss");
3377         if (!m->mdt_mdss_service) {
3378                 CERROR("failed to start metadata seq server service\n");
3379                 GOTO(err_mdt_svc, rc = -ENOMEM);
3380         }
3381
3382         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3383         if (rc)
3384                 GOTO(err_mdt_svc, rc);
3385
3386
3387         /*
3388          * Data sequence server service configuration. We want to have really
3389          * cluster-wide sequences space. This is why we start only one sequence
3390          * controller which manages space.
3391          */
3392         conf = (typeof(conf)) {
3393                 .psc_nbufs = MDS_NBUFS,
3394                 .psc_bufsize = MDS_BUFSIZE,
3395                 .psc_max_req_size = SEQ_MAXREQSIZE,
3396                 .psc_max_reply_size = SEQ_MAXREPSIZE,
3397                 .psc_req_portal = SEQ_DATA_PORTAL,
3398                 .psc_rep_portal = OSC_REPLY_PORTAL,
3399                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3400                 .psc_min_threads = SEQ_NUM_THREADS,
3401                 .psc_max_threads = SEQ_NUM_THREADS,
3402                 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3403         };
3404
3405         m->mdt_dtss_service =
3406                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3407                                      LUSTRE_MDT_NAME"_dtss",
3408                                      procfs_entry, NULL, "mdt_dtss");
3409         if (!m->mdt_dtss_service) {
3410                 CERROR("failed to start data seq server service\n");
3411                 GOTO(err_mdt_svc, rc = -ENOMEM);
3412         }
3413
3414         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3415         if (rc)
3416                 GOTO(err_mdt_svc, rc);
3417
3418         /* FLD service start */
3419         conf = (typeof(conf)) {
3420                 .psc_nbufs            = MDS_NBUFS,
3421                 .psc_bufsize          = MDS_BUFSIZE,
3422                 .psc_max_req_size     = FLD_MAXREQSIZE,
3423                 .psc_max_reply_size   = FLD_MAXREPSIZE,
3424                 .psc_req_portal       = FLD_REQUEST_PORTAL,
3425                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3426                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3427                 .psc_min_threads      = FLD_NUM_THREADS,
3428                 .psc_max_threads      = FLD_NUM_THREADS,
3429                 .psc_ctx_tags         = LCT_DT_THREAD|LCT_MD_THREAD
3430         };
3431
3432         m->mdt_fld_service =
3433                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3434                                      LUSTRE_MDT_NAME"_fld",
3435                                      procfs_entry, NULL, "mdt_fld");
3436         if (!m->mdt_fld_service) {
3437                 CERROR("failed to start fld service\n");
3438                 GOTO(err_mdt_svc, rc = -ENOMEM);
3439         }
3440
3441         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3442         if (rc)
3443                 GOTO(err_mdt_svc, rc);
3444
3445         /*
3446          * mds-mds service configuration. Separate portal is used to allow
3447          * mds-mds requests be not blocked during recovery.
3448          */
3449         conf = (typeof(conf)) {
3450                 .psc_nbufs            = MDS_NBUFS,
3451                 .psc_bufsize          = MDS_BUFSIZE,
3452                 .psc_max_req_size     = MDS_MAXREQSIZE,
3453                 .psc_max_reply_size   = MDS_MAXREPSIZE,
3454                 .psc_req_portal       = MDS_MDS_PORTAL,
3455                 .psc_rep_portal       = MDC_REPLY_PORTAL,
3456                 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3457                 .psc_min_threads      = min(max(mdt_num_threads, MDT_MIN_THREADS),
3458                                             MDT_MAX_THREADS),
3459                 .psc_max_threads      = MDT_MAX_THREADS,
3460                 .psc_ctx_tags         = LCT_MD_THREAD
3461         };
3462         m->mdt_xmds_service = ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3463                                                   LUSTRE_MDT_NAME "_mds",
3464                                                   procfs_entry, NULL, "mdt_xmds");
3465
3466         if (m->mdt_xmds_service == NULL) {
3467                 CERROR("failed to start readpage service\n");
3468                 GOTO(err_mdt_svc, rc = -ENOMEM);
3469         }
3470
3471         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3472         if (rc)
3473                 GOTO(err_mdt_svc, rc);
3474
3475         EXIT;
3476 err_mdt_svc:
3477         if (rc)
3478                 mdt_stop_ptlrpc_service(m);
3479
3480         return rc;
3481 }
3482
3483 static void mdt_stack_fini(const struct lu_env *env,
3484                            struct mdt_device *m, struct lu_device *top)
3485 {
3486         struct lu_device        *d = top, *n;
3487         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3488         struct lustre_cfg_bufs  *bufs;
3489         struct lustre_cfg       *lcfg;
3490         struct mdt_thread_info  *info;
3491         char flags[3]="";
3492         ENTRY;
3493
3494         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3495         LASSERT(info != NULL);
3496
3497         bufs = &info->mti_u.bufs;
3498         /* process cleanup, pass mdt obd name to get obd umount flags */
3499         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3500         if (obd->obd_force)
3501                 strcat(flags, "F");
3502         if (obd->obd_fail)
3503                 strcat(flags, "A");
3504         lustre_cfg_bufs_set_string(bufs, 1, flags);
3505         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3506         if (!lcfg) {
3507                 CERROR("Cannot alloc lcfg!\n");
3508                 return;
3509         }
3510
3511         LASSERT(top);
3512         top->ld_ops->ldo_process_config(env, top, lcfg);
3513         lustre_cfg_free(lcfg);
3514
3515         lu_site_purge(env, top->ld_site, ~0);
3516         while (d != NULL) {
3517                 struct obd_type *type;
3518                 struct lu_device_type *ldt = d->ld_type;
3519
3520                 /* each fini() returns next device in stack of layers
3521                  * * so we can avoid the recursion */
3522                 n = ldt->ldt_ops->ldto_device_fini(env, d);
3523                 lu_device_put(d);
3524                 ldt->ldt_ops->ldto_device_free(env, d);
3525                 type = ldt->ldt_obd_type;
3526                 type->typ_refcnt--;
3527                 class_put_type(type);
3528
3529                 /* switch to the next device in the layer */
3530                 d = n;
3531         }
3532         m->mdt_child = NULL;
3533         m->mdt_bottom = NULL;
3534 }
3535
3536 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3537                                          const char *typename,
3538                                          struct lu_device *child,
3539                                          struct lustre_cfg *cfg)
3540 {
3541         const char            *dev = lustre_cfg_string(cfg, 0);
3542         struct obd_type       *type;
3543         struct lu_device_type *ldt;
3544         struct lu_device      *d;
3545         int rc;
3546         ENTRY;
3547
3548         /* find the type */
3549         type = class_get_type(typename);
3550         if (!type) {
3551                 CERROR("Unknown type: '%s'\n", typename);
3552                 GOTO(out, rc = -ENODEV);
3553         }
3554
3555         rc = lu_context_refill(&env->le_ctx);
3556         if (rc != 0) {
3557                 CERROR("Failure to refill context: '%d'\n", rc);
3558                 GOTO(out_type, rc);
3559         }
3560
3561         if (env->le_ses != NULL) {
3562                 rc = lu_context_refill(env->le_ses);
3563                 if (rc != 0) {
3564                         CERROR("Failure to refill session: '%d'\n", rc);
3565                         GOTO(out_type, rc);
3566                 }
3567         }
3568
3569         ldt = type->typ_lu;
3570         if (ldt == NULL) {
3571                 CERROR("type: '%s'\n", typename);
3572                 GOTO(out_type, rc = -EINVAL);
3573         }
3574
3575         ldt->ldt_obd_type = type;
3576         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3577         if (IS_ERR(d)) {
3578                 CERROR("Cannot allocate device: '%s'\n", typename);
3579                 GOTO(out_type, rc = -ENODEV);
3580         }
3581
3582         LASSERT(child->ld_site);
3583         d->ld_site = child->ld_site;
3584
3585         type->typ_refcnt++;
3586         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3587         if (rc) {
3588                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3589                 GOTO(out_alloc, rc);
3590         }
3591         lu_device_get(d);
3592
3593         RETURN(d);
3594
3595 out_alloc:
3596         ldt->ldt_ops->ldto_device_free(env, d);
3597         type->typ_refcnt--;
3598 out_type:
3599         class_put_type(type);
3600 out:
3601         return ERR_PTR(rc);
3602 }
3603
3604 static int mdt_stack_init(const struct lu_env *env,
3605                           struct mdt_device *m, struct lustre_cfg *cfg)
3606 {
3607         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3608         struct lu_device  *tmp;
3609         struct md_device  *md;
3610         int rc;
3611         ENTRY;
3612
3613         /* init the stack */
3614         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3615         if (IS_ERR(tmp)) {
3616                 RETURN(PTR_ERR(tmp));
3617         }
3618         m->mdt_bottom = lu2dt_dev(tmp);
3619         d = tmp;
3620         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3621         if (IS_ERR(tmp)) {
3622                 GOTO(out, rc = PTR_ERR(tmp));
3623         }
3624         d = tmp;
3625         md = lu2md_dev(d);
3626
3627         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3628         if (IS_ERR(tmp)) {
3629                 GOTO(out, rc = PTR_ERR(tmp));
3630         }
3631         d = tmp;
3632         /*set mdd upcall device*/
3633         md_upcall_dev_set(md, lu2md_dev(d));
3634
3635         md = lu2md_dev(d);
3636         /*set cmm upcall device*/
3637         md_upcall_dev_set(md, &m->mdt_md_dev);
3638
3639         m->mdt_child = lu2md_dev(d);
3640
3641         /* process setup config */
3642         tmp = &m->mdt_md_dev.md_lu_dev;
3643         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3644         GOTO(out, rc);
3645 out:
3646         /* fini from last known good lu_device */
3647         if (rc)
3648                 mdt_stack_fini(env, m, d);
3649
3650         return rc;
3651 }
3652
3653 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3654 {
3655         struct md_device *next = m->mdt_child;
3656         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3657         struct lu_site   *ls   = d->ld_site;
3658         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3659         ENTRY;
3660
3661         ping_evictor_stop();
3662
3663         target_recovery_fini(obd);
3664         mdt_stop_ptlrpc_service(m);
3665
3666         mdt_fs_cleanup(env, m);
3667
3668         upcall_cache_cleanup(m->mdt_rmtacl_cache);
3669         m->mdt_rmtacl_cache = NULL;
3670
3671         upcall_cache_cleanup(m->mdt_identity_cache);
3672         m->mdt_identity_cache = NULL;
3673
3674         if (m->mdt_namespace != NULL) {
3675                 ldlm_namespace_free(m->mdt_namespace, d->ld_obd->obd_force);
3676                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3677         }
3678
3679         mdt_seq_fini(env, m);
3680         mdt_seq_fini_cli(m);
3681         mdt_fld_fini(env, m);
3682         mdt_procfs_fini(m);
3683         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3684         lprocfs_obd_cleanup(d->ld_obd);
3685
3686         if (m->mdt_rootsquash_info) {
3687                 OBD_FREE_PTR(m->mdt_rootsquash_info);
3688                 m->mdt_rootsquash_info = NULL;
3689         }
3690
3691         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3692         del_timer(&m->mdt_ck_timer);
3693         mdt_ck_thread_stop(m);
3694
3695         /* finish the stack */
3696         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3697
3698         if (ls) {
3699                 if (!list_empty(&ls->ls_lru) || ls->ls_total != 0) {
3700                         /*
3701                          * Uh-oh, objects still exist.
3702                          */
3703                         static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR);
3704
3705                         lu_site_print(env, ls, &cookie, lu_cdebug_printer);
3706                 }
3707
3708                 lu_site_fini(ls);
3709                 OBD_FREE_PTR(ls);
3710                 d->ld_site = NULL;
3711         }
3712         LASSERT(atomic_read(&d->ld_ref) == 0);
3713         md_device_fini(&m->mdt_md_dev);
3714
3715         EXIT;
3716 }
3717
3718 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3719 {
3720         char *p = options;
3721
3722         if (!options)
3723                 return;
3724
3725         while (*options) {
3726                 int len;
3727
3728                 while (*p && *p != ',')
3729                         p++;
3730
3731                 len = p - options;
3732                 if ((len == sizeof("user_xattr") - 1) &&
3733                     (memcmp(options, "user_xattr", len) == 0)) {
3734                         m->mdt_opts.mo_user_xattr = 1;
3735                         LCONSOLE_INFO("Enabling user_xattr\n");
3736                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3737                            (memcmp(options, "nouser_xattr", len) == 0)) {
3738                         m->mdt_opts.mo_user_xattr = 0;
3739                         LCONSOLE_INFO("Disabling user_xattr\n");
3740                 } else if ((len == sizeof("acl") - 1) &&
3741                            (memcmp(options, "acl", len) == 0)) {
3742 #ifdef CONFIG_FS_POSIX_ACL
3743                         m->mdt_opts.mo_acl = 1;
3744                         LCONSOLE_INFO("Enabling ACL\n");
3745 #else
3746                         m->mdt_opts.mo_acl = 0;
3747                         CWARN("ignoring unsupported acl mount option\n");
3748                         LCONSOLE_INFO("Disabling ACL\n");
3749 #endif
3750                 } else if ((len == sizeof("noacl") - 1) &&
3751                            (memcmp(options, "noacl", len) == 0)) {
3752                         m->mdt_opts.mo_acl = 0;
3753                         LCONSOLE_INFO("Disabling ACL\n");
3754                 }
3755
3756                 options = ++p;
3757         }
3758 }
3759
3760 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3761
3762 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3763                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3764 {
3765         struct lprocfs_static_vars lvars;
3766         struct mdt_thread_info    *info;
3767         struct obd_device         *obd;
3768         const char                *dev = lustre_cfg_string(cfg, 0);
3769         const char                *num = lustre_cfg_string(cfg, 2);
3770         struct lustre_mount_info  *lmi;
3771         struct lustre_sb_info     *lsi;
3772         struct lu_site            *s;
3773         int                        rc;
3774         ENTRY;
3775
3776         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3777         LASSERT(info != NULL);
3778
3779         obd = class_name2obd(dev);
3780         LASSERT(obd != NULL);
3781
3782         spin_lock_init(&m->mdt_transno_lock);
3783
3784         m->mdt_max_mdsize = MAX_MD_SIZE;
3785         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3786
3787         m->mdt_opts.mo_user_xattr = 0;
3788         m->mdt_opts.mo_acl = 0;
3789         lmi = server_get_mount_2(dev);
3790         if (lmi == NULL) {
3791                 CERROR("Cannot get mount info for %s!\n", dev);
3792                 RETURN(-EFAULT);
3793         } else {
3794                 lsi = s2lsi(lmi->lmi_sb);
3795                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3796                 server_put_mount_2(dev, lmi->lmi_mnt);
3797         }
3798
3799         spin_lock_init(&m->mdt_ioepoch_lock);
3800         m->mdt_opts.mo_compat_resname = 0;
3801         m->mdt_capa_timeout = CAPA_TIMEOUT;
3802         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3803         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3804
3805         spin_lock_init(&m->mdt_client_bitmap_lock);
3806
3807         OBD_ALLOC_PTR(s);
3808         if (s == NULL)
3809                 RETURN(-ENOMEM);
3810
3811         md_device_init(&m->mdt_md_dev, ldt);
3812         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3813         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3814         /* set this lu_device to obd, because error handling need it */
3815         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3816
3817         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3818         if (rc) {
3819                 CERROR("Can't init lu_site, rc %d\n", rc);
3820                 GOTO(err_free_site, rc);
3821         }
3822
3823         lprocfs_init_vars(mdt, &lvars);
3824         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3825         if (rc) {
3826                 CERROR("Can't init lprocfs, rc %d\n", rc);
3827                 GOTO(err_fini_site, rc);
3828         }
3829         ptlrpc_lprocfs_register_obd(obd);
3830
3831         rc = mdt_procfs_init(m, dev);
3832         if (rc) {
3833                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3834                 GOTO(err_fini_proc, rc);
3835         }
3836
3837         /* set server index */
3838         LASSERT(num);
3839         s->ls_node_id = simple_strtol(num, NULL, 10);
3840
3841         /* failover is the default
3842          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3843          * assumed whose ls_node_id == 0 XXX
3844          * */
3845         obd->obd_replayable = 1;
3846         /* No connection accepted until configurations will finish */
3847         obd->obd_no_conn = 1;
3848
3849         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3850                 char *str = lustre_cfg_string(cfg, 4);
3851                 if (strchr(str, 'n')) {
3852                         CWARN("%s: recovery disabled\n", obd->obd_name);
3853                         obd->obd_replayable = 0;
3854                 }
3855         }
3856
3857         /* init the stack */
3858         rc = mdt_stack_init(env, m, cfg);
3859         if (rc) {
3860                 CERROR("Can't init device stack, rc %d\n", rc);
3861                 GOTO(err_fini_proc, rc);
3862         }
3863
3864         rc = mdt_fld_init(env, obd->obd_name, m);
3865         if (rc)
3866                 GOTO(err_fini_stack, rc);
3867
3868         rc = mdt_seq_init(env, obd->obd_name, m);
3869         if (rc)
3870                 GOTO(err_fini_fld, rc);
3871
3872         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3873                  LUSTRE_MDT_NAME"-%p", m);
3874         m->mdt_namespace = ldlm_namespace_new(info->mti_u.ns_name,
3875                                               LDLM_NAMESPACE_SERVER,
3876                                               LDLM_NAMESPACE_GREEDY);
3877         if (m->mdt_namespace == NULL)
3878                 GOTO(err_fini_seq, rc = -ENOMEM);
3879
3880         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3881         /* set obd_namespace for compatibility with old code */
3882         obd->obd_namespace = m->mdt_namespace;
3883
3884         m->mdt_identity_cache = upcall_cache_init(obd->obd_name,
3885                                                   "NONE",
3886                                                   &mdt_identity_upcall_cache_ops);
3887         if (IS_ERR(m->mdt_identity_cache)) {
3888                 rc = PTR_ERR(m->mdt_identity_cache);
3889                 m->mdt_identity_cache = NULL;
3890                 GOTO(err_free_ns, rc);
3891         }
3892
3893         m->mdt_rmtacl_cache = upcall_cache_init(obd->obd_name,
3894                                                 MDT_RMTACL_UPCALL_PATH,
3895                                                 &mdt_rmtacl_upcall_cache_ops);
3896         if (IS_ERR(m->mdt_rmtacl_cache)) {
3897                 rc = PTR_ERR(m->mdt_rmtacl_cache);
3898                 m->mdt_rmtacl_cache = NULL;
3899                 GOTO(err_free_ns, rc);
3900         }
3901
3902         m->mdt_ck_timer.function = mdt_ck_timer_callback;
3903         m->mdt_ck_timer.data = (unsigned long)m;
3904         init_timer(&m->mdt_ck_timer);
3905         rc = mdt_ck_thread_start(m);
3906         if (rc)
3907                 GOTO(err_free_ns, rc);
3908
3909         rc = mdt_fs_setup(env, m, obd);
3910         if (rc)
3911                 GOTO(err_capa, rc);
3912
3913         target_recovery_init(obd, mdt_recovery_handle);
3914
3915         rc = mdt_start_ptlrpc_service(m);
3916         if (rc)
3917                 GOTO(err_fs_cleanup, rc);
3918
3919         ping_evictor_start();
3920
3921         rc = lu_site_init_finish(s);
3922         if (rc)
3923                 GOTO(err_stop_service, rc);
3924
3925         if (obd->obd_recovering == 0)
3926                 mdt_postrecov(env, m);
3927
3928         mdt_init_capa_ctxt(env, m);
3929
3930         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
3931                 ldlm_timeout = 6;
3932
3933         RETURN(0);
3934
3935 err_stop_service:
3936         ping_evictor_stop();
3937         mdt_stop_ptlrpc_service(m);
3938 err_fs_cleanup:
3939         target_recovery_fini(obd);
3940         mdt_fs_cleanup(env, m);
3941 err_capa:
3942         del_timer(&m->mdt_ck_timer);
3943         mdt_ck_thread_stop(m);
3944 err_free_ns:
3945         upcall_cache_cleanup(m->mdt_rmtacl_cache);
3946         m->mdt_rmtacl_cache = NULL;
3947         upcall_cache_cleanup(m->mdt_identity_cache);
3948         m->mdt_identity_cache = NULL;
3949         ldlm_namespace_free(m->mdt_namespace, 0);
3950         obd->obd_namespace = m->mdt_namespace = NULL;
3951 err_fini_seq:
3952         mdt_seq_fini(env, m);
3953 err_fini_fld:
3954         mdt_fld_fini(env, m);
3955 err_fini_stack:
3956         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3957 err_fini_proc:
3958         mdt_procfs_fini(m);
3959         lprocfs_obd_cleanup(obd);
3960 err_fini_site:
3961         lu_site_fini(s);
3962 err_free_site:
3963         OBD_FREE_PTR(s);
3964
3965         md_device_fini(&m->mdt_md_dev);
3966         return (rc);
3967 }
3968
3969 /* used by MGS to process specific configurations */
3970 static int mdt_process_config(const struct lu_env *env,
3971                               struct lu_device *d, struct lustre_cfg *cfg)
3972 {
3973         struct mdt_device *m = mdt_dev(d);
3974         struct md_device *md_next = m->mdt_child;
3975         struct lu_device *next = md2lu_dev(md_next);
3976         int rc = 0;
3977         ENTRY;
3978
3979         switch (cfg->lcfg_command) {
3980         case LCFG_PARAM: {
3981                 struct lprocfs_static_vars lvars;
3982                 struct obd_device *obd = d->ld_obd;
3983
3984                 lprocfs_init_vars(mdt, &lvars);
3985                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
3986                 if (rc)
3987                         /* others are passed further */
3988                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
3989                 break;
3990         }
3991         case LCFG_ADD_MDC:
3992                 /*
3993                  * Add mdc hook to get first MDT uuid and connect it to
3994                  * ls->controller to use for seq manager.
3995                  */
3996                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
3997                 if (rc)
3998                         CERROR("Can't add mdc, rc %d\n", rc);
3999                 else
4000                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4001                 break;
4002         default:
4003                 /* others are passed further */
4004                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4005                 break;
4006         }
4007         RETURN(rc);
4008 }
4009
4010 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4011                                           const struct lu_object_header *hdr,
4012                                           struct lu_device *d)
4013 {
4014         struct mdt_object *mo;
4015
4016         ENTRY;
4017
4018         OBD_ALLOC_PTR(mo);
4019         if (mo != NULL) {
4020                 struct lu_object *o;
4021                 struct lu_object_header *h;
4022
4023                 o = &mo->mot_obj.mo_lu;
4024                 h = &mo->mot_header;
4025                 lu_object_header_init(h);
4026                 lu_object_init(o, h, d);
4027                 lu_object_add_top(h, o);
4028                 o->lo_ops = &mdt_obj_ops;
4029                 RETURN(o);
4030         } else
4031                 RETURN(NULL);
4032 }
4033
4034 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4035 {
4036         struct mdt_device *d = mdt_dev(o->lo_dev);
4037         struct lu_device  *under;
4038         struct lu_object  *below;
4039         int                rc = 0;
4040         ENTRY;
4041
4042         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4043                PFID(lu_object_fid(o)));
4044
4045         under = &d->mdt_child->md_lu_dev;
4046         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4047         if (below != NULL) {
4048                 lu_object_add(o, below);
4049         } else
4050                 rc = -ENOMEM;
4051
4052         RETURN(rc);
4053 }
4054
4055 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4056 {
4057         struct mdt_object *mo = mdt_obj(o);
4058         struct lu_object_header *h;
4059         ENTRY;
4060
4061         h = o->lo_header;
4062         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4063                PFID(lu_object_fid(o)));
4064
4065         lu_object_fini(o);
4066         lu_object_header_fini(h);
4067         OBD_FREE_PTR(mo);
4068         EXIT;
4069 }
4070
4071 static int mdt_object_print(const struct lu_env *env, void *cookie,
4072                             lu_printer_t p, const struct lu_object *o)
4073 {
4074         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4075 }
4076
4077 static struct lu_device_operations mdt_lu_ops = {
4078         .ldo_object_alloc   = mdt_object_alloc,
4079         .ldo_process_config = mdt_process_config
4080 };
4081
4082 static struct lu_object_operations mdt_obj_ops = {
4083         .loo_object_init    = mdt_object_init,
4084         .loo_object_free    = mdt_object_free,
4085         .loo_object_print   = mdt_object_print
4086 };
4087
4088 /* mds_connect_internal */
4089 static int mdt_connect_internal(struct obd_export *exp,
4090                                 struct mdt_device *mdt,
4091                                 struct obd_connect_data *data)
4092 {
4093         __u64 flags;
4094
4095         if (data != NULL) {
4096                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4097                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4098
4099                 /* If no known bits (which should not happen, probably,
4100                    as everybody should support LOOKUP and UPDATE bits at least)
4101                    revert to compat mode with plain locks. */
4102                 if (!data->ocd_ibits_known &&
4103                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4104                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4105
4106                 if (!mdt->mdt_opts.mo_acl)
4107                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4108
4109                 if (!mdt->mdt_opts.mo_user_xattr)
4110                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4111
4112                 if (!mdt->mdt_opts.mo_mds_capa)
4113                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4114
4115                 if (!mdt->mdt_opts.mo_oss_capa)
4116                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4117
4118                 spin_lock(&exp->exp_lock);
4119                 exp->exp_connect_flags = data->ocd_connect_flags;
4120                 spin_unlock(&exp->exp_lock);
4121                 data->ocd_version = LUSTRE_VERSION_CODE;
4122                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4123         }
4124
4125 #if 0
4126         if (mdt->mdt_opts.mo_acl &&
4127             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4128                 CWARN("%s: MDS requires ACL support but client does not\n",
4129                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4130                 return -EBADE;
4131         }
4132 #endif
4133
4134         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4135         if ((exp->exp_connect_flags & flags) == flags) {
4136                 CWARN("%s: both local and remote client flags are set\n",
4137                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4138                 return -EBADE;
4139         }
4140
4141         if (mdt->mdt_opts.mo_mds_capa &&
4142             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4143                 CWARN("%s: MDS requires capability support, but client not\n",
4144                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4145                 return -EBADE;
4146         }
4147
4148         if (mdt->mdt_opts.mo_oss_capa &&
4149             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4150                 CWARN("%s: MDS requires OSS capability support, "
4151                       "but client not\n",
4152                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4153                 return -EBADE;
4154         }
4155
4156         return 0;
4157 }
4158
4159 /* mds_connect copy */
4160 static int mdt_obd_connect(const struct lu_env *env,
4161                            struct lustre_handle *conn, struct obd_device *obd,
4162                            struct obd_uuid *cluuid,
4163                            struct obd_connect_data *data)
4164 {
4165         struct mdt_client_data *mcd;
4166         struct obd_export      *exp;
4167         struct mdt_device      *mdt;
4168         int                     rc;
4169         ENTRY;
4170
4171         LASSERT(env != NULL);
4172         if (!conn || !obd || !cluuid)
4173                 RETURN(-EINVAL);
4174
4175         mdt = mdt_dev(obd->obd_lu_dev);
4176
4177         rc = class_connect(conn, obd, cluuid);
4178         if (rc)
4179                 RETURN(rc);
4180
4181         exp = class_conn2export(conn);
4182         LASSERT(exp != NULL);
4183
4184         rc = mdt_connect_internal(exp, mdt, data);
4185         if (rc == 0) {
4186                 OBD_ALLOC_PTR(mcd);
4187                 if (mcd != NULL) {
4188                         struct mdt_thread_info *mti;
4189                         mti = lu_context_key_get(&env->le_ctx,
4190                                                  &mdt_thread_key);
4191                         LASSERT(mti != NULL);
4192                         mti->mti_exp = exp;
4193                         memcpy(mcd->mcd_uuid, cluuid, sizeof mcd->mcd_uuid);
4194                         exp->exp_mdt_data.med_mcd = mcd;
4195                         rc = mdt_client_new(env, mdt);
4196                         if (rc != 0) {
4197                                 OBD_FREE_PTR(mcd);
4198                                 exp->exp_mdt_data.med_mcd = NULL;
4199                         }
4200                 } else
4201                         rc = -ENOMEM;
4202         }
4203
4204         if (rc != 0)
4205                 class_disconnect(exp);
4206         else
4207                 class_export_put(exp);
4208
4209         RETURN(rc);
4210 }
4211
4212 static int mdt_obd_reconnect(struct obd_export *exp, struct obd_device *obd,
4213                              struct obd_uuid *cluuid,
4214                              struct obd_connect_data *data)
4215 {
4216         int rc;
4217         ENTRY;
4218
4219         if (exp == NULL || obd == NULL || cluuid == NULL)
4220                 RETURN(-EINVAL);
4221
4222         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4223
4224         RETURN(rc);
4225 }
4226
4227 static int mdt_obd_disconnect(struct obd_export *exp)
4228 {
4229         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4230         int rc;
4231         ENTRY;
4232
4233         LASSERT(exp);
4234         class_export_get(exp);
4235
4236         /* Disconnect early so that clients can't keep using export */
4237         rc = class_disconnect(exp);
4238         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4239                 ldlm_cancel_locks_for_export(exp);
4240
4241         /* complete all outstanding replies */
4242         spin_lock(&exp->exp_lock);
4243         while (!list_empty(&exp->exp_outstanding_replies)) {
4244                 struct ptlrpc_reply_state *rs =
4245                         list_entry(exp->exp_outstanding_replies.next,
4246                                    struct ptlrpc_reply_state, rs_exp_list);
4247                 struct ptlrpc_service *svc = rs->rs_service;
4248
4249                 spin_lock(&svc->srv_lock);
4250                 list_del_init(&rs->rs_exp_list);
4251                 ptlrpc_schedule_difficult_reply(rs);
4252                 spin_unlock(&svc->srv_lock);
4253         }
4254         spin_unlock(&exp->exp_lock);
4255
4256         class_export_put(exp);
4257         RETURN(rc);
4258 }
4259
4260 /* FIXME: Can we avoid using these two interfaces? */
4261 static int mdt_init_export(struct obd_export *exp)
4262 {
4263         struct mdt_export_data *med = &exp->exp_mdt_data;
4264         ENTRY;
4265
4266         INIT_LIST_HEAD(&med->med_open_head);
4267         spin_lock_init(&med->med_open_lock);
4268         spin_lock(&exp->exp_lock);
4269         exp->exp_connecting = 1;
4270         spin_unlock(&exp->exp_lock);
4271         RETURN(0);
4272 }
4273
4274 static int mdt_destroy_export(struct obd_export *export)
4275 {
4276         struct mdt_export_data *med;
4277         struct obd_device      *obd = export->exp_obd;
4278         struct mdt_device      *mdt;
4279         struct mdt_thread_info *info;
4280         struct lu_env           env;
4281         struct md_attr         *ma;
4282         int lmm_size;
4283         int cookie_size;
4284         int rc = 0;
4285         ENTRY;
4286
4287         med = &export->exp_mdt_data;
4288         if (med->med_rmtclient)
4289                 mdt_cleanup_idmap(med);
4290
4291         target_destroy_export(export);
4292
4293         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4294                 RETURN(0);
4295
4296         mdt = mdt_dev(obd->obd_lu_dev);
4297         LASSERT(mdt != NULL);
4298
4299         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4300         if (rc)
4301                 RETURN(rc);
4302
4303         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4304         LASSERT(info != NULL);
4305         memset(info, 0, sizeof *info);
4306         info->mti_env = &env;
4307         info->mti_mdt = mdt;
4308         info->mti_exp = export;
4309
4310         ma = &info->mti_attr;
4311         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4312         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4313         OBD_ALLOC(ma->ma_lmm, lmm_size);
4314         OBD_ALLOC(ma->ma_cookie, cookie_size);
4315
4316         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4317                 GOTO(out, rc = -ENOMEM);
4318         ma->ma_need = MA_LOV | MA_COOKIE;
4319         ma->ma_valid = 0;
4320         /* Close any open files (which may also cause orphan unlinking). */
4321         spin_lock(&med->med_open_lock);
4322         while (!list_empty(&med->med_open_head)) {
4323                 struct list_head *tmp = med->med_open_head.next;
4324                 struct mdt_file_data *mfd =
4325                         list_entry(tmp, struct mdt_file_data, mfd_list);
4326
4327                 /* Remove mfd handle so it can't be found again.
4328                  * We are consuming the mfd_list reference here. */
4329                 class_handle_unhash(&mfd->mfd_handle);
4330                 list_del_init(&mfd->mfd_list);
4331                 spin_unlock(&med->med_open_lock);
4332                 mdt_mfd_close(info, mfd);
4333                 /* TODO: if we close the unlinked file,
4334                  * we need to remove it's objects from OST */
4335                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4336                 spin_lock(&med->med_open_lock);
4337                 ma->ma_lmm_size = lmm_size;
4338                 ma->ma_cookie_size = cookie_size;
4339                 ma->ma_need = MA_LOV | MA_COOKIE;
4340                 ma->ma_valid = 0;
4341         }
4342         spin_unlock(&med->med_open_lock);
4343         info->mti_mdt = NULL;
4344         mdt_client_del(&env, mdt);
4345
4346         EXIT;
4347 out:
4348         if (lmm_size) {
4349                 OBD_FREE(ma->ma_lmm, lmm_size);
4350                 ma->ma_lmm = NULL;
4351         }
4352         if (cookie_size) {
4353                 OBD_FREE(ma->ma_cookie, cookie_size);
4354                 ma->ma_cookie = NULL;
4355         }
4356         lu_env_fini(&env);
4357
4358         return rc;
4359 }
4360
4361 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4362 {
4363         if (flag & CONFIG_LOG)
4364                 m->mdt_fl_cfglog = 1;
4365         if (flag & CONFIG_SYNC)
4366                 m->mdt_fl_synced = 1;
4367
4368         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4369                 /* Open for clients */
4370                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4371 }
4372
4373 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4374                       enum md_upcall_event ev)
4375 {
4376         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4377         struct md_device  *next  = m->mdt_child;
4378         struct mdt_thread_info *mti;
4379         int rc = 0;
4380         ENTRY;
4381
4382         switch (ev) {
4383                 case MD_LOV_SYNC:
4384                         rc = next->md_ops->mdo_maxsize_get(env, next,
4385                                         &m->mdt_max_mdsize,
4386                                         &m->mdt_max_cookiesize);
4387                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4388                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4389                         mdt_allow_cli(m, CONFIG_SYNC);
4390                         break;
4391                 case MD_NO_TRANS:
4392                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4393                         mti->mti_no_need_trans = 1;
4394                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4395                         break;
4396                 case MD_LOV_CONFIG:
4397                         /* Check that MDT is not yet configured */
4398                         LASSERT(!m->mdt_fl_cfglog);
4399                         break;
4400                 default:
4401                         CERROR("invalid event\n");
4402                         rc = -EINVAL;
4403                         break;
4404         }
4405         RETURN(rc);
4406 }
4407
4408 static int mdt_obd_notify(struct obd_device *host,
4409                           struct obd_device *watched,
4410                           enum obd_notify_event ev, void *data)
4411 {
4412         ENTRY;
4413
4414         switch (ev) {
4415         case OBD_NOTIFY_CONFIG:
4416                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned int)data);
4417                 break;
4418         default:
4419                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4420         }
4421         RETURN(0);
4422 }
4423
4424 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4425                          void *karg, void *uarg)
4426 {
4427         struct lu_env      env;
4428         struct obd_device *obd= exp->exp_obd;
4429         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4430         struct dt_device  *dt = mdt->mdt_bottom;
4431         int rc;
4432
4433         ENTRY;
4434         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4435         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4436         if (rc)
4437                 RETURN(rc);
4438
4439         switch (cmd) {
4440         case OBD_IOC_SYNC:
4441                 rc = dt->dd_ops->dt_sync(&env, dt);
4442                 break;
4443         case OBD_IOC_SET_READONLY:
4444                 rc = dt->dd_ops->dt_sync(&env, dt);
4445                 dt->dd_ops->dt_ro(&env, dt);
4446                 break;
4447         case OBD_IOC_ABORT_RECOVERY:
4448                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4449                 target_stop_recovery_thread(obd);
4450                 rc = 0;
4451                 break;
4452         default:
4453                 CERROR("Not supported cmd = %d for device %s\n",
4454                        cmd, obd->obd_name);
4455                 rc = -EOPNOTSUPP;
4456         }
4457
4458         lu_env_fini(&env);
4459         RETURN(rc);
4460 }
4461
4462 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4463 {
4464         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4465         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4466         int rc, lost;
4467         ENTRY;
4468         /* if some clients didn't participate in recovery then we can possibly
4469          * lost sequence. Now we should increase sequence for safe value */
4470         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4471         mdt_seq_adjust(env, mdt, lost);
4472
4473         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4474         RETURN(rc);
4475 }
4476
4477 int mdt_obd_postrecov(struct obd_device *obd)
4478 {
4479         struct lu_env env;
4480         int rc;
4481
4482         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4483         if (rc)
4484                 RETURN(rc);
4485         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4486         lu_env_fini(&env);
4487         return rc;
4488 }
4489
4490 static struct obd_ops mdt_obd_device_ops = {
4491         .o_owner          = THIS_MODULE,
4492         .o_connect        = mdt_obd_connect,
4493         .o_reconnect      = mdt_obd_reconnect,
4494         .o_disconnect     = mdt_obd_disconnect,
4495         .o_init_export    = mdt_init_export,
4496         .o_destroy_export = mdt_destroy_export,
4497         .o_iocontrol      = mdt_iocontrol,
4498         .o_postrecov      = mdt_obd_postrecov,
4499         .o_notify         = mdt_obd_notify
4500 };
4501
4502 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4503                                          struct lu_device *d)
4504 {
4505         struct mdt_device *m = mdt_dev(d);
4506         ENTRY;
4507
4508         mdt_fini(env, m);
4509         RETURN(NULL);
4510 }
4511
4512 static void mdt_device_free(const struct lu_env *env, struct lu_device *d)
4513 {
4514         struct mdt_device *m = mdt_dev(d);
4515
4516         OBD_FREE_PTR(m);
4517 }
4518
4519 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4520                                           struct lu_device_type *t,
4521                                           struct lustre_cfg *cfg)
4522 {
4523         struct lu_device  *l;
4524         struct mdt_device *m;
4525
4526         OBD_ALLOC_PTR(m);
4527         if (m != NULL) {
4528                 int rc;
4529
4530                 l = &m->mdt_md_dev.md_lu_dev;
4531                 rc = mdt_init0(env, m, t, cfg);
4532                 if (rc != 0) {
4533                         OBD_FREE_PTR(m);
4534                         l = ERR_PTR(rc);
4535                         return l;
4536                 }
4537                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4538         } else
4539                 l = ERR_PTR(-ENOMEM);
4540         return l;
4541 }
4542
4543 /*
4544  * context key constructor/destructor
4545  */
4546 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4547
4548 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4549
4550 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4551
4552 struct lu_context_key mdt_txn_key = {
4553         .lct_tags = LCT_TX_HANDLE,
4554         .lct_init = mdt_txn_key_init,
4555         .lct_fini = mdt_txn_key_fini
4556 };
4557
4558 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4559 {
4560         return md_ucred(info->mti_env);
4561 }
4562
4563 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4564
4565 static struct lu_device_type_operations mdt_device_type_ops = {
4566         .ldto_init = mdt_type_init,
4567         .ldto_fini = mdt_type_fini,
4568
4569         .ldto_device_alloc = mdt_device_alloc,
4570         .ldto_device_free  = mdt_device_free,
4571         .ldto_device_fini  = mdt_device_fini
4572 };
4573
4574 static struct lu_device_type mdt_device_type = {
4575         .ldt_tags     = LU_DEVICE_MD,
4576         .ldt_name     = LUSTRE_MDT_NAME,
4577         .ldt_ops      = &mdt_device_type_ops,
4578         .ldt_ctx_tags = LCT_MD_THREAD
4579 };
4580
4581 static int __init mdt_mod_init(void)
4582 {
4583         struct lprocfs_static_vars lvars;
4584         int rc;
4585
4586         mdt_num_threads = MDT_NUM_THREADS;
4587         lprocfs_init_vars(mdt, &lvars);
4588         rc = class_register_type(&mdt_obd_device_ops, NULL,
4589                                  lvars.module_vars, LUSTRE_MDT_NAME,
4590                                  &mdt_device_type);
4591
4592         return rc;
4593 }
4594
4595 static void __exit mdt_mod_exit(void)
4596 {
4597         class_unregister_type(LUSTRE_MDT_NAME);
4598 }
4599
4600
4601 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4602 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4603         .mh_name    = #opc,                                             \
4604         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4605         .mh_opc     = prefix ## _  ## opc,                              \
4606         .mh_flags   = flags,                                            \
4607         .mh_act     = fn,                                               \
4608         .mh_fmt     = fmt                                               \
4609 }
4610
4611 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4612         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4613
4614 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4615         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4616
4617 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4618         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4619 /*
4620  * Request with a format known in advance
4621  */
4622 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4623         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4624
4625 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4626         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4627
4628 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4629         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4630 /*
4631  * Request with a format we do not yet know
4632  */
4633 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4634         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4635
4636 static struct mdt_handler mdt_mds_ops[] = {
4637 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4638 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4639 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4640 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4641 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4642 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4643 DEF_MDT_HNDL_F(HABEO_CORPUS|MUTABOR,      SETXATTR,     mdt_setxattr),
4644 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4645 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4646 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4647 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4648 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4649 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4650 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4651 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4652 DEF_MDT_HNDL_0(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4653 DEF_MDT_HNDL_0(0,                         QUOTACTL,     mdt_quotactl_handle)
4654 };
4655
4656 #define DEF_OBD_HNDL(flags, name, fn)                   \
4657         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4658
4659
4660 static struct mdt_handler mdt_obd_ops[] = {
4661         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4662         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4663         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4664 };
4665
4666 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4667         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4668 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4669         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4670
4671 static struct mdt_handler mdt_dlm_ops[] = {
4672         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4673         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4674         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4675         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4676 };
4677
4678 static struct mdt_handler mdt_llog_ops[] = {
4679 };
4680
4681 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4682         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4683
4684 static struct mdt_handler mdt_sec_ctx_ops[] = {
4685         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4686         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4687         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4688 };
4689
4690 static struct mdt_opc_slice mdt_regular_handlers[] = {
4691         {
4692                 .mos_opc_start = MDS_GETATTR,
4693                 .mos_opc_end   = MDS_LAST_OPC,
4694                 .mos_hs        = mdt_mds_ops
4695         },
4696         {
4697                 .mos_opc_start = OBD_PING,
4698                 .mos_opc_end   = OBD_LAST_OPC,
4699                 .mos_hs        = mdt_obd_ops
4700         },
4701         {
4702                 .mos_opc_start = LDLM_ENQUEUE,
4703                 .mos_opc_end   = LDLM_LAST_OPC,
4704                 .mos_hs        = mdt_dlm_ops
4705         },
4706         {
4707                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4708                 .mos_opc_end   = LLOG_LAST_OPC,
4709                 .mos_hs        = mdt_llog_ops
4710         },
4711         {
4712                 .mos_opc_start = SEC_CTX_INIT,
4713                 .mos_opc_end   = SEC_LAST_OPC,
4714                 .mos_hs        = mdt_sec_ctx_ops
4715         },
4716         {
4717                 .mos_hs        = NULL
4718         }
4719 };
4720
4721 static struct mdt_handler mdt_readpage_ops[] = {
4722         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4723         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4724 #ifdef HAVE_SPLIT_SUPPORT
4725         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4726 #endif
4727
4728         /*
4729          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4730          * detailed comments. --umka
4731          */
4732         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4733         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4734 };
4735
4736 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4737         {
4738                 .mos_opc_start = MDS_GETATTR,
4739                 .mos_opc_end   = MDS_LAST_OPC,
4740                 .mos_hs        = mdt_readpage_ops
4741         },
4742         {
4743                 .mos_hs        = NULL
4744         }
4745 };
4746
4747 static struct mdt_handler mdt_xmds_ops[] = {
4748         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4749         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4750         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4751         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4752 };
4753
4754 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4755         {
4756                 .mos_opc_start = MDS_GETATTR,
4757                 .mos_opc_end   = MDS_LAST_OPC,
4758                 .mos_hs        = mdt_xmds_ops
4759         },
4760         {
4761                 .mos_opc_start = OBD_PING,
4762                 .mos_opc_end   = OBD_LAST_OPC,
4763                 .mos_hs        = mdt_obd_ops
4764         },
4765         {
4766                 .mos_hs        = NULL
4767         }
4768 };
4769
4770 static struct mdt_handler mdt_seq_ops[] = {
4771         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4772 };
4773
4774 static struct mdt_opc_slice mdt_seq_handlers[] = {
4775         {
4776                 .mos_opc_start = SEQ_QUERY,
4777                 .mos_opc_end   = SEQ_LAST_OPC,
4778                 .mos_hs        = mdt_seq_ops
4779         },
4780         {
4781                 .mos_hs        = NULL
4782         }
4783 };
4784
4785 static struct mdt_handler mdt_fld_ops[] = {
4786         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4787 };
4788
4789 static struct mdt_opc_slice mdt_fld_handlers[] = {
4790         {
4791                 .mos_opc_start = FLD_QUERY,
4792                 .mos_opc_end   = FLD_LAST_OPC,
4793                 .mos_hs        = mdt_fld_ops
4794         },
4795         {
4796                 .mos_hs        = NULL
4797         }
4798 };
4799
4800 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4801 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4802 MODULE_LICENSE("GPL");
4803
4804 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
4805                 "number of mdt service threads to start");
4806
4807 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);