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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #include <lustre_acl.h>
  71 #include <lustre_param.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 /*
  96  * Initialized in mdt_mod_init().
  97  */
  98 unsigned long mdt_num_threads;
  99
 100 /* ptlrpc request handler for MDT. All handlers are
 101  * grouped into several slices - struct mdt_opc_slice,
 102  * and stored in an array - mdt_handlers[].
 103  */
 104 struct mdt_handler {
 105         /* The name of this handler. */
 106         const char *mh_name;
 107         /* Fail id for this handler, checked at the beginning of this handler*/
 108         int         mh_fail_id;
 109         /* Operation code for this handler */
 110         __u32       mh_opc;
 111         /* flags are listed in enum mdt_handler_flags below. */
 112         __u32       mh_flags;
 113         /* The actual handler function to execute. */
 114         int (*mh_act)(struct mdt_thread_info *info);
 115         /* Request format for this request. */
 116         const struct req_format *mh_fmt;
 117 };
 118
 119 enum mdt_handler_flags {
 120         /*
 121          * struct mdt_body is passed in the incoming message, and object
 122          * identified by this fid exists on disk.
 123          *
 124          * "habeo corpus" == "I have a body"
 125          */
 126         HABEO_CORPUS = (1 << 0),
 127         /*
 128          * struct ldlm_request is passed in the incoming message.
 129          *
 130          * "habeo clavis" == "I have a key"
 131          */
 132         HABEO_CLAVIS = (1 << 1),
 133         /*
 134          * this request has fixed reply format, so that reply message can be
 135          * packed by generic code.
 136          *
 137          * "habeo refero" == "I have a reply"
 138          */
 139         HABEO_REFERO = (1 << 2),
 140         /*
 141          * this request will modify something, so check whether the filesystem
 142          * is readonly or not, then return -EROFS to client asap if necessary.
 143          *
 144          * "mutabor" == "I shall modify"
 145          */
 146         MUTABOR      = (1 << 3)
 147 };
 148
 149 struct mdt_opc_slice {
 150         __u32               mos_opc_start;
 151         int                 mos_opc_end;
 152         struct mdt_handler *mos_hs;
 153 };
 154
 155 static struct mdt_opc_slice mdt_regular_handlers[];
 156 static struct mdt_opc_slice mdt_readpage_handlers[];
 157 static struct mdt_opc_slice mdt_xmds_handlers[];
 158 static struct mdt_opc_slice mdt_seq_handlers[];
 159 static struct mdt_opc_slice mdt_fld_handlers[];
 160
 161 static struct mdt_device *mdt_dev(struct lu_device *d);
 162 static int mdt_regular_handle(struct ptlrpc_request *req);
 163 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 164
 165 static struct lu_object_operations mdt_obj_ops;
 166
 167 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 168 {
 169         if (!rep)
 170                 return 0;
 171         return (rep->lock_policy_res1 & flag);
 172 }
 173
 174 void mdt_clear_disposition(struct mdt_thread_info *info,
 175                            struct ldlm_reply *rep, int flag)
 176 {
 177         if (info)
 178                 info->mti_opdata &= ~flag;
 179         if (rep)
 180                 rep->lock_policy_res1 &= ~flag;
 181 }
 182
 183 void mdt_set_disposition(struct mdt_thread_info *info,
 184                          struct ldlm_reply *rep, int flag)
 185 {
 186         if (info)
 187                 info->mti_opdata |= flag;
 188         if (rep)
 189                 rep->lock_policy_res1 |= flag;
 190 }
 191
 192 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 193 {
 194         lh->mlh_pdo_hash = 0;
 195         lh->mlh_reg_mode = lm;
 196         lh->mlh_type = MDT_REG_LOCK;
 197 }
 198
 199 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 200                        const char *name, int namelen)
 201 {
 202         lh->mlh_reg_mode = lm;
 203         lh->mlh_type = MDT_PDO_LOCK;
 204
 205         if (name != NULL) {
 206                 LASSERT(namelen > 0);
 207                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 208         } else {
 209                 LASSERT(namelen == 0);
 210                 lh->mlh_pdo_hash = 0ull;
 211         }
 212 }
 213
 214 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 215                               struct mdt_lock_handle *lh)
 216 {
 217         mdl_mode_t mode;
 218         ENTRY;
 219
 220         /*
 221          * Any dir access needs couple of locks:
 222          *
 223          * 1) on part of dir we gonna take lookup/modify;
 224          *
 225          * 2) on whole dir to protect it from concurrent splitting and/or to
 226          * flush client's cache for readdir().
 227          *
 228          * so, for a given mode and object this routine decides what lock mode
 229          * to use for lock #2:
 230          *
 231          * 1) if caller's gonna lookup in dir then we need to protect dir from
 232          * being splitted only - LCK_CR
 233          *
 234          * 2) if caller's gonna modify dir then we need to protect dir from
 235          * being splitted and to flush cache - LCK_CW
 236          *
 237          * 3) if caller's gonna modify dir and that dir seems ready for
 238          * splitting then we need to protect it from any type of access
 239          * (lookup/modify/split) - LCK_EX --bzzz
 240          */
 241
 242         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 243         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 244
 245         /*
 246          * Ask underlaying level its opinion about preferable PDO lock mode
 247          * having access type passed as regular lock mode:
 248          *
 249          * - MDL_MINMODE means that lower layer does not want to specify lock
 250          * mode;
 251          *
 252          * - MDL_NL means that no PDO lock should be taken. This is used in some
 253          * cases. Say, for non-splittable directories no need to use PDO locks
 254          * at all.
 255          */
 256         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 257                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 258
 259         if (mode != MDL_MINMODE) {
 260                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 261         } else {
 262                 /*
 263                  * Lower layer does not want to specify locking mode. We do it
 264                  * our selves. No special protection is needed, just flush
 265                  * client's cache on modification and allow concurrent
 266                  * mondification.
 267                  */
 268                 switch (lh->mlh_reg_mode) {
 269                 case LCK_EX:
 270                         lh->mlh_pdo_mode = LCK_EX;
 271                         break;
 272                 case LCK_PR:
 273                         lh->mlh_pdo_mode = LCK_CR;
 274                         break;
 275                 case LCK_PW:
 276                         lh->mlh_pdo_mode = LCK_CW;
 277                         break;
 278                 default:
 279                         CERROR("Not expected lock type (0x%x)\n",
 280                                (int)lh->mlh_reg_mode);
 281                         LBUG();
 282                 }
 283         }
 284
 285         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 286         EXIT;
 287 }
 288
 289 static int mdt_getstatus(struct mdt_thread_info *info)
 290 {
 291         struct mdt_device *mdt  = info->mti_mdt;
 292         struct md_device  *next = mdt->mdt_child;
 293         struct mdt_body   *repbody;
 294         int                rc;
 295
 296         ENTRY;
 297
 298         rc = mdt_check_ucred(info);
 299         if (rc)
 300                 RETURN(err_serious(rc));
 301
 302         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 303                 RETURN(err_serious(-ENOMEM));
 304
 305         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 306         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 307         if (rc != 0)
 308                 RETURN(rc);
 309
 310         repbody->valid |= OBD_MD_FLID;
 311
 312         if (mdt->mdt_opts.mo_mds_capa) {
 313                 struct mdt_object  *root;
 314                 struct lustre_capa *capa;
 315
 316                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 317                 if (IS_ERR(root))
 318                         RETURN(PTR_ERR(root));
 319
 320                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 321                 LASSERT(capa);
 322                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 323
 324                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 325                                  0);
 326                 mdt_object_put(info->mti_env, root);
 327                 if (rc == 0)
 328                         repbody->valid |= OBD_MD_FLMDSCAPA;
 329         }
 330
 331         RETURN(rc);
 332 }
 333
 334 static int mdt_statfs(struct mdt_thread_info *info)
 335 {
 336         struct md_device      *next  = info->mti_mdt->mdt_child;
 337         struct ptlrpc_service *svc;
 338         struct obd_statfs     *osfs;
 339         int                    rc;
 340
 341         ENTRY;
 342
 343         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 344
 345         /* This will trigger a watchdog timeout */
 346         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 347                          (MDT_SERVICE_WATCHDOG_FACTOR *
 348                           at_get(&svc->srv_at_estimate) / 1000) + 1);
 349
 350         rc = mdt_check_ucred(info);
 351         if (rc)
 352                 RETURN(err_serious(rc));
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 355                 rc = err_serious(-ENOMEM);
 356         } else {
 357                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 358                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 359                                               &info->mti_u.ksfs);
 360                 statfs_pack(osfs, &info->mti_u.ksfs);
 361         }
 362         RETURN(rc);
 363 }
 364
 365 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
 366 {
 367         struct mdt_body *b;
 368         struct lu_attr *attr = &info->mti_attr.ma_attr;
 369
 370         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371
 372         /* Check if Size-on-MDS is enabled. */
 373         if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
 374             S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
 375                 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
 376                 b->size = attr->la_size;
 377                 b->blocks = attr->la_blocks;
 378         }
 379 }
 380
 381 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 382                         const struct lu_attr *attr, const struct lu_fid *fid)
 383 {
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode))
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392
 393         b->atime      = attr->la_atime;
 394         b->mtime      = attr->la_mtime;
 395         b->ctime      = attr->la_ctime;
 396         b->mode       = attr->la_mode;
 397         b->size       = attr->la_size;
 398         b->blocks     = attr->la_blocks;
 399         b->uid        = attr->la_uid;
 400         b->gid        = attr->la_gid;
 401         b->flags      = attr->la_flags;
 402         b->nlink      = attr->la_nlink;
 403         b->rdev       = attr->la_rdev;
 404
 405         if (fid) {
 406                 b->fid1 = *fid;
 407                 b->valid |= OBD_MD_FLID;
 408
 409                 /* FIXME: these should be fixed when new igif ready.*/
 410                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 411                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 412                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 413
 414                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 415                                 PFID(fid), b->nlink, b->mode, b->size);
 416         }
 417
 418         if (info)
 419                 mdt_body_reverse_idmap(info, b);
 420 }
 421
 422 static inline int mdt_body_has_lov(const struct lu_attr *la,
 423                                    const struct mdt_body *body)
 424 {
 425         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 426                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 427 }
 428
 429 static int mdt_getattr_internal(struct mdt_thread_info *info,
 430                                 struct mdt_object *o)
 431 {
 432         struct md_object        *next = mdt_object_child(o);
 433         const struct mdt_body   *reqbody = info->mti_body;
 434         struct ptlrpc_request   *req = mdt_info_req(info);
 435         struct mdt_export_data  *med = &req->rq_export->exp_mdt_data;
 436         struct md_attr          *ma = &info->mti_attr;
 437         struct lu_attr          *la = &ma->ma_attr;
 438         struct req_capsule      *pill = info->mti_pill;
 439         const struct lu_env     *env = info->mti_env;
 440         struct mdt_body         *repbody;
 441         struct lu_buf           *buffer = &info->mti_buf;
 442         int                     rc;
 443         ENTRY;
 444
 445         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 446                 RETURN(err_serious(-ENOMEM));
 447
 448         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 449
 450         ma->ma_valid = 0;
 451
 452         rc = mdt_object_exists(o);
 453         if (rc < 0) {
 454                 /* This object is located on remote node.*/
 455                 repbody->fid1 = *mdt_object_fid(o);
 456                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 457                 RETURN(0);
 458         }
 459
 460         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 461         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 462
 463         /* If it is dir object and client require MEA, then we got MEA */
 464         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 465             reqbody->valid & OBD_MD_MEA) {
 466                 /* Assumption: MDT_MD size is enough for lmv size. */
 467                 ma->ma_lmv = buffer->lb_buf;
 468                 ma->ma_lmv_size = buffer->lb_len;
 469                 ma->ma_need = MA_LMV | MA_INODE;
 470         } else {
 471                 ma->ma_lmm = buffer->lb_buf;
 472                 ma->ma_lmm_size = buffer->lb_len;
 473                 ma->ma_need = MA_LOV | MA_INODE;
 474         }
 475
 476         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 477             reqbody->valid & OBD_MD_FLDIREA  &&
 478             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 479                 /* get default stripe info for this dir. */
 480                 ma->ma_need |= MA_LOV_DEF;
 481         }
 482         rc = mo_attr_get(env, next, ma);
 483         if (unlikely(rc)) {
 484                 CERROR("getattr error for "DFID": %d\n",
 485                         PFID(mdt_object_fid(o)), rc);
 486                 RETURN(rc);
 487         }
 488
 489         if (likely(ma->ma_valid & MA_INODE))
 490                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 491         else
 492                 RETURN(-EFAULT);
 493
 494         if (mdt_body_has_lov(la, reqbody)) {
 495                 if (ma->ma_valid & MA_LOV) {
 496                         LASSERT(ma->ma_lmm_size);
 497                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 498                         repbody->eadatasize = ma->ma_lmm_size;
 499                         if (S_ISDIR(la->la_mode))
 500                                 repbody->valid |= OBD_MD_FLDIREA;
 501                         else
 502                                 repbody->valid |= OBD_MD_FLEASIZE;
 503                 }
 504                 if (ma->ma_valid & MA_LMV) {
 505                         LASSERT(S_ISDIR(la->la_mode));
 506                         repbody->eadatasize = ma->ma_lmv_size;
 507                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 508                 }
 509                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 510                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 511                 }
 512         } else if (S_ISLNK(la->la_mode) &&
 513                    reqbody->valid & OBD_MD_LINKNAME) {
 514                 buffer->lb_buf = ma->ma_lmm;
 515                 buffer->lb_len = reqbody->eadatasize;
 516                 rc = mo_readlink(env, next, buffer);
 517                 if (unlikely(rc <= 0)) {
 518                         CERROR("readlink failed: %d\n", rc);
 519                         rc = -EFAULT;
 520                 } else {
 521                         repbody->valid |= OBD_MD_LINKNAME;
 522                         repbody->eadatasize = rc;
 523                         /* NULL terminate */
 524                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 525                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 526                                (char*)ma->ma_lmm, rc);
 527                         rc = 0;
 528                 }
 529         }
 530
 531         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 532                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 533                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 534                 repbody->valid |= OBD_MD_FLMODEASIZE;
 535                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 536                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 537                        repbody->max_cookiesize);
 538         }
 539
 540         if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
 541                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 542
 543                 /* mdt_getattr_lock only */
 544                 rc = mdt_pack_remote_perm(info, o, buf);
 545                 if (rc) {
 546                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 547                         repbody->aclsize = 0;
 548                         RETURN(rc);
 549                 } else {
 550                         repbody->valid |= OBD_MD_FLRMTPERM;
 551                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 552                 }
 553         }
 554 #ifdef CONFIG_FS_POSIX_ACL
 555         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 556                  (reqbody->valid & OBD_MD_FLACL)) {
 557                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 558                 buffer->lb_len = req_capsule_get_size(pill,
 559                                                       &RMF_ACL, RCL_SERVER);
 560                 if (buffer->lb_len > 0) {
 561                         rc = mo_xattr_get(env, next, buffer,
 562                                           XATTR_NAME_ACL_ACCESS);
 563                         if (rc < 0) {
 564                                 if (rc == -ENODATA) {
 565                                         repbody->aclsize = 0;
 566                                         repbody->valid |= OBD_MD_FLACL;
 567                                         rc = 0;
 568                                 } else if (rc == -EOPNOTSUPP) {
 569                                         rc = 0;
 570                                 } else {
 571                                         CERROR("got acl size: %d\n", rc);
 572                                 }
 573                         } else {
 574                                 repbody->aclsize = rc;
 575                                 repbody->valid |= OBD_MD_FLACL;
 576                                 rc = 0;
 577                         }
 578                 }
 579         }
 580 #endif
 581
 582         if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
 583             info->mti_mdt->mdt_opts.mo_mds_capa) {
 584                 struct lustre_capa *capa;
 585
 586                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 587                 LASSERT(capa);
 588                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 589                 rc = mo_capa_get(env, next, capa, 0);
 590                 if (rc)
 591                         RETURN(rc);
 592                 repbody->valid |= OBD_MD_FLMDSCAPA;
 593         }
 594         RETURN(rc);
 595 }
 596
 597 static int mdt_renew_capa(struct mdt_thread_info *info)
 598 {
 599         struct mdt_device  *mdt = info->mti_mdt;
 600         struct mdt_object  *obj = info->mti_object;
 601         struct mdt_body    *body;
 602         struct lustre_capa *capa, *c;
 603         int rc;
 604         ENTRY;
 605
 606         /* if object doesn't exist, or server has disabled capability,
 607          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 608          * flag not set.
 609          */
 610         if (!obj || !mdt->mdt_opts.mo_mds_capa)
 611                 RETURN(0);
 612
 613         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 614         LASSERT(body != NULL);
 615
 616         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 617         LASSERT(c);
 618
 619         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 620         LASSERT(capa);
 621
 622         *capa = *c;
 623         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 624         if (rc == 0)
 625                 body->valid |= OBD_MD_FLOSSCAPA;
 626         RETURN(rc);
 627 }
 628
 629 static int mdt_getattr(struct mdt_thread_info *info)
 630 {
 631         struct mdt_object       *obj = info->mti_object;
 632         struct req_capsule      *pill = info->mti_pill;
 633         struct mdt_body         *reqbody;
 634         struct mdt_body         *repbody;
 635         mode_t                   mode;
 636         int                      md_size;
 637         int rc;
 638         ENTRY;
 639
 640         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 641         LASSERT(reqbody);
 642
 643         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 644                 rc = req_capsule_server_pack(pill);
 645                 if (unlikely(rc))
 646                         RETURN(err_serious(rc));
 647                 rc = mdt_renew_capa(info);
 648                 GOTO(out_shrink, rc);
 649         }
 650
 651         LASSERT(obj != NULL);
 652         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 653
 654         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 655         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 656             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 657                 md_size = reqbody->eadatasize;
 658         else
 659                 md_size = info->mti_mdt->mdt_max_mdsize;
 660
 661         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 662
 663         rc = req_capsule_server_pack(pill);
 664         if (unlikely(rc != 0))
 665                 RETURN(err_serious(rc));
 666
 667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 668         LASSERT(repbody != NULL);
 669         repbody->eadatasize = 0;
 670         repbody->aclsize = 0;
 671
 672         if (reqbody->valid & OBD_MD_FLRMTPERM)
 673                 rc = mdt_init_ucred(info, reqbody);
 674         else
 675                 rc = mdt_check_ucred(info);
 676         if (unlikely(rc))
 677                 GOTO(out_shrink, rc);
 678
 679         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 680         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 681
 682         /*
 683          * Don't check capability at all, because rename might getattr for
 684          * remote obj, and at that time no capability is available.
 685          */
 686         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 687         rc = mdt_getattr_internal(info, obj);
 688         if (reqbody->valid & OBD_MD_FLRMTPERM)
 689                 mdt_exit_ucred(info);
 690         EXIT;
 691 out_shrink:
 692         mdt_shrink_reply(info);
 693         return rc;
 694 }
 695
 696 static int mdt_is_subdir(struct mdt_thread_info *info)
 697 {
 698         struct mdt_object     *o = info->mti_object;
 699         struct req_capsule    *pill = info->mti_pill;
 700         const struct mdt_body *body = info->mti_body;
 701         struct mdt_body       *repbody;
 702         int                    rc;
 703         ENTRY;
 704
 705         LASSERT(o != NULL);
 706
 707         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 708
 709         /*
 710          * We save last checked parent fid to @repbody->fid1 for remote
 711          * directory case.
 712          */
 713         LASSERT(fid_is_sane(&body->fid2));
 714         LASSERT(mdt_object_exists(o) > 0);
 715         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 716                            &body->fid2, &repbody->fid1);
 717         if (rc == 0 || rc == -EREMOTE)
 718                 repbody->valid |= OBD_MD_FLID;
 719
 720         RETURN(rc);
 721 }
 722
 723 static int mdt_raw_lookup(struct mdt_thread_info *info,
 724                           struct mdt_object *parent,
 725                           const struct lu_name *lname,
 726                           struct ldlm_reply *ldlm_rep)
 727 {
 728         struct md_object *next = mdt_object_child(info->mti_object);
 729         const struct mdt_body *reqbody = info->mti_body;
 730         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 731         struct mdt_body *repbody;
 732         int rc;
 733         ENTRY;
 734
 735         if (reqbody->valid != OBD_MD_FLID)
 736                 RETURN(0);
 737
 738         LASSERT(!info->mti_cross_ref);
 739
 740         /* Only got the fid of this obj by name */
 741         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 742                         &info->mti_spec);
 743 #if 0
 744         /* XXX is raw_lookup possible as intent operation? */
 745         if (rc != 0) {
 746                 if (rc == -ENOENT)
 747                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 748                 RETURN(rc);
 749         } else
 750                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 751
 752         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 753 #endif
 754         if (rc == 0) {
 755                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 756                 repbody->fid1 = *child_fid;
 757                 repbody->valid = OBD_MD_FLID;
 758         }
 759         RETURN(1);
 760 }
 761
 762 /*
 763  * UPDATE lock should be taken against parent, and be release before exit;
 764  * child_bits lock should be taken against child, and be returned back:
 765  *            (1)normal request should release the child lock;
 766  *            (2)intent request will grant the lock to client.
 767  */
 768 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 769                                  struct mdt_lock_handle *lhc,
 770                                  __u64 child_bits,
 771                                  struct ldlm_reply *ldlm_rep)
 772 {
 773         struct ptlrpc_request  *req       = mdt_info_req(info);
 774         struct mdt_body        *reqbody   = NULL;
 775         struct mdt_object      *parent    = info->mti_object;
 776         struct mdt_object      *child;
 777         struct md_object       *next      = mdt_object_child(parent);
 778         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 779         struct lu_name         *lname     = NULL;
 780         const char             *name;
 781         int                     namelen   = 0;
 782         struct mdt_lock_handle *lhp;
 783         struct ldlm_lock       *lock;
 784         struct ldlm_res_id     *res_id;
 785         int                     is_resent;
 786         int                     rc;
 787
 788         ENTRY;
 789
 790         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 791         LASSERT(ergo(is_resent,
 792                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 793
 794         LASSERT(parent != NULL);
 795         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 796         if (name == NULL)
 797                 RETURN(err_serious(-EFAULT));
 798
 799         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 800                                        RCL_CLIENT) - 1;
 801         LASSERT(namelen >= 0);
 802
 803         /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
 804          * otherwise do not allow empty name, that is the name must contain
 805          * at least one character and the terminating '\0'*/
 806         if (namelen == 0) {
 807                 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 808                 LASSERT(fid_is_sane(&reqbody->fid2));
 809                 name = NULL;
 810
 811                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 812                        "ldlm_rep = %p\n",
 813                        PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 814                        ldlm_rep);
 815         } else {
 816                 lname = mdt_name(info->mti_env, (char *)name, namelen);
 817                 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 818                        "ldlm_rep = %p\n",
 819                        PFID(mdt_object_fid(parent)), name, ldlm_rep);
 820         }
 821
 822         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 823
 824         rc = mdt_object_exists(parent);
 825         if (unlikely(rc == 0)) {
 826                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 827                                 &parent->mot_obj.mo_lu,
 828                                 "Parent doesn't exist!\n");
 829                 RETURN(-ESTALE);
 830         } else
 831                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 832                          PFID(mdt_object_fid(parent)));
 833
 834         if (lname) {
 835                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 836                 if (rc != 0) {
 837                         if (rc > 0)
 838                                 rc = 0;
 839                         RETURN(rc);
 840                 }
 841         }
 842
 843         if (info->mti_cross_ref) {
 844                 /* Only getattr on the child. Parent is on another node. */
 845                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 846                 child = parent;
 847                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 848                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 849
 850                 if (is_resent) {
 851                         /* Do not take lock for resent case. */
 852                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 853                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 854                                  lhc->mlh_reg_lh.cookie);
 855                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 856                                                 &lock->l_resource->lr_name));
 857                         LDLM_LOCK_PUT(lock);
 858                         rc = 0;
 859                 } else {
 860                         mdt_lock_handle_init(lhc);
 861                         mdt_lock_reg_init(lhc, LCK_PR);
 862
 863                         /*
 864                          * Object's name is on another MDS, no lookup lock is
 865                          * needed here but update is.
 866                          */
 867                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 868                         child_bits |= MDS_INODELOCK_UPDATE;
 869
 870                         rc = mdt_object_lock(info, child, lhc, child_bits,
 871                                              MDT_LOCAL_LOCK);
 872                 }
 873                 if (rc == 0) {
 874                         /* Finally, we can get attr for child. */
 875                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 876                                          BYPASS_CAPA);
 877                         rc = mdt_getattr_internal(info, child);
 878                         if (unlikely(rc != 0))
 879                                 mdt_object_unlock(info, child, lhc, 1);
 880                 }
 881                 RETURN(rc);
 882         }
 883
 884         /* step 1: lock parent */
 885         lhp = &info->mti_lh[MDT_LH_PARENT];
 886         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 887         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 888                              MDT_LOCAL_LOCK);
 889
 890         if (unlikely(rc != 0))
 891                 RETURN(rc);
 892
 893         if (lname) {
 894                 /* step 2: lookup child's fid by name */
 895                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 896                                 &info->mti_spec);
 897
 898                 if (rc != 0) {
 899                         if (rc == -ENOENT)
 900                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 901                         GOTO(out_parent, rc);
 902                 } else
 903                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 904         } else {
 905                 *child_fid = reqbody->fid2;
 906                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 907         }
 908
 909         /*
 910          *step 3: find the child object by fid & lock it.
 911          *        regardless if it is local or remote.
 912          */
 913         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 914
 915         if (unlikely(IS_ERR(child)))
 916                 GOTO(out_parent, rc = PTR_ERR(child));
 917         if (is_resent) {
 918                 /* Do not take lock for resent case. */
 919                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 920                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 921                          lhc->mlh_reg_lh.cookie);
 922
 923                 res_id = &lock->l_resource->lr_name;
 924                 if (!fid_res_name_eq(mdt_object_fid(child),
 925                                     &lock->l_resource->lr_name)) {
 926                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 927                                                  &lock->l_resource->lr_name),
 928                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 929                                  (unsigned long)res_id->name[0],
 930                                  (unsigned long)res_id->name[1],
 931                                  (unsigned long)res_id->name[2],
 932                                  PFID(mdt_object_fid(parent)));
 933                           CWARN("Although resent, but still not get child lock"
 934                                 "parent:"DFID" child:"DFID"\n",
 935                                 PFID(mdt_object_fid(parent)),
 936                                 PFID(mdt_object_fid(child)));
 937                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 938                           LDLM_LOCK_PUT(lock);
 939                           GOTO(relock, 0);
 940                 }
 941                 LDLM_LOCK_PUT(lock);
 942                 rc = 0;
 943         } else {
 944                 struct md_attr *ma;
 945 relock:
 946                 ma = &info->mti_attr;
 947
 948                 mdt_lock_handle_init(lhc);
 949                 mdt_lock_reg_init(lhc, LCK_PR);
 950
 951                 if (mdt_object_exists(child) == 0) {
 952                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 953                                         &child->mot_obj.mo_lu,
 954                                         "Object doesn't exist!\n");
 955                         GOTO(out_child, rc = -ESTALE);
 956                 }
 957
 958                 ma->ma_valid = 0;
 959                 ma->ma_need = MA_INODE;
 960                 rc = mo_attr_get(info->mti_env, next, ma);
 961                 if (unlikely(rc != 0))
 962                         GOTO(out_child, rc);
 963
 964                 /* If the file has not been changed for some time, we return
 965                  * not only a LOOKUP lock, but also an UPDATE lock and this
 966                  * might save us RPC on later STAT. For directories, it also
 967                  * let negative dentry starts working for this dir. */
 968                 if (ma->ma_valid & MA_INODE &&
 969                     ma->ma_attr.la_valid & LA_CTIME &&
 970                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
 971                     ma->ma_attr.la_ctime < cfs_time_current_sec())
 972                         child_bits |= MDS_INODELOCK_UPDATE;
 973
 974                 rc = mdt_object_lock(info, child, lhc, child_bits,
 975                                      MDT_CROSS_LOCK);
 976
 977                 if (unlikely(rc != 0))
 978                         GOTO(out_child, rc);
 979         }
 980
 981         /* finally, we can get attr for child. */
 982         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
 983         rc = mdt_getattr_internal(info, child);
 984         if (unlikely(rc != 0)) {
 985                 mdt_object_unlock(info, child, lhc, 1);
 986         } else {
 987                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 988                 if (lock) {
 989                         struct mdt_body *repbody;
 990
 991                         /* Debugging code. */
 992                         res_id = &lock->l_resource->lr_name;
 993                         LDLM_DEBUG(lock, "Returning lock to client\n");
 994                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
 995                                                  &lock->l_resource->lr_name),
 996                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 997                                  (unsigned long)res_id->name[0],
 998                                  (unsigned long)res_id->name[1],
 999                                  (unsigned long)res_id->name[2],
1000                                  PFID(mdt_object_fid(child)));
1001                         /*
1002                          * Pack Size-on-MDS inode attributes to the body if
1003                          * update lock is given.
1004                          */
1005                         repbody = req_capsule_server_get(info->mti_pill,
1006                                                          &RMF_MDT_BODY);
1007                         if (lock->l_policy_data.l_inodebits.bits &
1008                             MDS_INODELOCK_UPDATE)
1009                                 mdt_pack_size2body(info, child);
1010                         LDLM_LOCK_PUT(lock);
1011                 }
1012         }
1013         EXIT;
1014 out_child:
1015         mdt_object_put(info->mti_env, child);
1016 out_parent:
1017         mdt_object_unlock(info, parent, lhp, 1);
1018         return rc;
1019 }
1020
1021 /* normal handler: should release the child lock */
1022 static int mdt_getattr_name(struct mdt_thread_info *info)
1023 {
1024         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1025         struct mdt_body        *reqbody;
1026         struct mdt_body        *repbody;
1027         int rc;
1028         ENTRY;
1029
1030         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1031         LASSERT(reqbody != NULL);
1032         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1033         LASSERT(repbody != NULL);
1034
1035         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1036         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1037         repbody->eadatasize = 0;
1038         repbody->aclsize = 0;
1039
1040         rc = mdt_init_ucred(info, reqbody);
1041         if (unlikely(rc))
1042                 GOTO(out_shrink, rc);
1043
1044         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1045         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1046                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1047                 lhc->mlh_reg_lh.cookie = 0;
1048         }
1049         mdt_exit_ucred(info);
1050         EXIT;
1051 out_shrink:
1052         mdt_shrink_reply(info);
1053         return rc;
1054 }
1055
1056 static struct lu_device_operations mdt_lu_ops;
1057
1058 static int lu_device_is_mdt(struct lu_device *d)
1059 {
1060         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1061 }
1062
1063 static int mdt_set_info(struct mdt_thread_info *info)
1064 {
1065         struct ptlrpc_request *req = mdt_info_req(info);
1066         char *key;
1067         __u32 *val;
1068         int keylen, rc = 0;
1069         ENTRY;
1070
1071         rc = req_capsule_server_pack(info->mti_pill);
1072         if (rc)
1073                 RETURN(rc);
1074
1075         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1076         if (key == NULL) {
1077                 DEBUG_REQ(D_HA, req, "no set_info key");
1078                 RETURN(-EFAULT);
1079         }
1080
1081         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1082                                       RCL_CLIENT);
1083
1084         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1085         if (val == NULL) {
1086                 DEBUG_REQ(D_HA, req, "no set_info val");
1087                 RETURN(-EFAULT);
1088         }
1089
1090         if (!KEY_IS(KEY_READ_ONLY))
1091                 RETURN(-EINVAL);
1092
1093         req->rq_status = 0;
1094         lustre_msg_set_status(req->rq_repmsg, 0);
1095
1096         spin_lock(&req->rq_export->exp_lock);
1097         if (*val)
1098                 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1099         else
1100                 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1101         spin_unlock(&req->rq_export->exp_lock);
1102
1103         RETURN(0);
1104 }
1105
1106 static int mdt_connect(struct mdt_thread_info *info)
1107 {
1108         int rc;
1109         struct ptlrpc_request *req;
1110
1111         req = mdt_info_req(info);
1112         rc = target_handle_connect(req);
1113         if (rc == 0) {
1114                 LASSERT(req->rq_export != NULL);
1115                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1116                 rc = mdt_init_idmap(info);
1117                 if (rc != 0)
1118                         /* if mdt_init_idmap failed, revocation for connect */
1119                         obd_disconnect(class_export_get(req->rq_export));
1120         } else
1121                 rc = err_serious(rc);
1122         return rc;
1123 }
1124
1125 static int mdt_disconnect(struct mdt_thread_info *info)
1126 {
1127         int rc;
1128         ENTRY;
1129
1130         rc = target_handle_disconnect(mdt_info_req(info));
1131         if (rc)
1132                 rc = err_serious(rc);
1133         RETURN(rc);
1134 }
1135
1136 static int mdt_sendpage(struct mdt_thread_info *info,
1137                         struct lu_rdpg *rdpg)
1138 {
1139         struct ptlrpc_request   *req = mdt_info_req(info);
1140         struct ptlrpc_bulk_desc *desc;
1141         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1142         int                      tmpcount;
1143         int                      tmpsize;
1144         int                      timeout;
1145         int                      i;
1146         int                      rc;
1147         ENTRY;
1148
1149         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1150                                     MDS_BULK_PORTAL);
1151         if (desc == NULL)
1152                 RETURN(-ENOMEM);
1153
1154         for (i = 0, tmpcount = rdpg->rp_count;
1155                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1156                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1157                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1158         }
1159
1160         LASSERT(desc->bd_nob == rdpg->rp_count);
1161         rc = ptlrpc_start_bulk_transfer(desc);
1162         if (rc)
1163                 GOTO(free_desc, rc);
1164
1165         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1166                 GOTO(abort_bulk, rc = 0);
1167
1168         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1169         if (timeout < 0)
1170                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1171                        req->rq_deadline, cfs_time_current_sec());
1172         *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL);
1173         rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1174         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1175
1176         if (rc == 0) {
1177                 if (desc->bd_success &&
1178                     desc->bd_nob_transferred == rdpg->rp_count)
1179                         GOTO(free_desc, rc);
1180
1181                 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1182         }
1183
1184         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1185                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1186                   desc->bd_nob_transferred, rdpg->rp_count,
1187                   req->rq_export->exp_client_uuid.uuid,
1188                   req->rq_export->exp_connection->c_remote_uuid.uuid);
1189
1190         class_fail_export(req->rq_export);
1191
1192         EXIT;
1193 abort_bulk:
1194         ptlrpc_abort_bulk(desc);
1195 free_desc:
1196         ptlrpc_free_bulk(desc);
1197         return rc;
1198 }
1199
1200 #ifdef HAVE_SPLIT_SUPPORT
1201 /*
1202  * Retrieve dir entry from the page and insert it to the slave object, actually,
1203  * this should be in osd layer, but since it will not in the final product, so
1204  * just do it here and do not define more moo api anymore for this.
1205  */
1206 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1207                               int size)
1208 {
1209         struct mdt_object *object = info->mti_object;
1210         struct lu_fid *lf = &info->mti_tmp_fid2;
1211         struct md_attr *ma = &info->mti_attr;
1212         struct lu_dirpage *dp;
1213         struct lu_dirent *ent;
1214         int rc = 0, offset = 0;
1215         ENTRY;
1216
1217         /* Make sure we have at least one entry. */
1218         if (size == 0)
1219                 RETURN(-EINVAL);
1220
1221         /*
1222          * Disable trans for this name insert, since it will include many trans
1223          * for this.
1224          */
1225         info->mti_no_need_trans = 1;
1226         /*
1227          * When write_dir_page, no need update parent's ctime,
1228          * and no permission check for name_insert.
1229          */
1230         ma->ma_attr.la_ctime = 0;
1231         ma->ma_attr.la_valid = LA_MODE;
1232         ma->ma_valid = MA_INODE;
1233
1234         cfs_kmap(page);
1235         dp = page_address(page);
1236         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1237
1238         for (ent = lu_dirent_start(dp); ent != NULL;
1239              ent = lu_dirent_next(ent)) {
1240                 struct lu_name *lname;
1241                 char *name;
1242
1243                 if (le16_to_cpu(ent->lde_namelen) == 0)
1244                         continue;
1245
1246                 fid_le_to_cpu(lf, &ent->lde_fid);
1247                 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1248                         ma->ma_attr.la_mode = S_IFDIR;
1249                 else
1250                         ma->ma_attr.la_mode = 0;
1251                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1252                 if (name == NULL)
1253                         GOTO(out, rc = -ENOMEM);
1254
1255                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1256                 lname = mdt_name(info->mti_env, name,
1257                                  le16_to_cpu(ent->lde_namelen) + 1);
1258                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1259                 rc = mdo_name_insert(info->mti_env,
1260                                      md_object_next(&object->mot_obj),
1261                                      lname, lf, ma);
1262                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1263                 if (rc) {
1264                         CERROR("Can't insert %*.*s, rc %d\n",
1265                                le16_to_cpu(ent->lde_namelen),
1266                                le16_to_cpu(ent->lde_namelen),
1267                                ent->lde_name, rc);
1268                         GOTO(out, rc);
1269                 }
1270
1271                 offset += lu_dirent_size(ent);
1272                 if (offset >= size)
1273                         break;
1274         }
1275         EXIT;
1276 out:
1277         cfs_kunmap(page);
1278         return rc;
1279 }
1280
1281 static int mdt_bulk_timeout(void *data)
1282 {
1283         ENTRY;
1284
1285         CERROR("mdt bulk transfer timeout \n");
1286
1287         RETURN(1);
1288 }
1289
1290 static int mdt_writepage(struct mdt_thread_info *info)
1291 {
1292         struct ptlrpc_request   *req = mdt_info_req(info);
1293         struct mdt_body         *reqbody;
1294         struct l_wait_info      *lwi;
1295         struct ptlrpc_bulk_desc *desc;
1296         struct page             *page;
1297         int                rc;
1298         ENTRY;
1299
1300
1301         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1302         if (reqbody == NULL)
1303                 RETURN(err_serious(-EFAULT));
1304
1305         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1306         if (desc == NULL)
1307                 RETURN(err_serious(-ENOMEM));
1308
1309         /* allocate the page for the desc */
1310         page = cfs_alloc_page(CFS_ALLOC_STD);
1311         if (page == NULL)
1312                 GOTO(desc_cleanup, rc = -ENOMEM);
1313
1314         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1315                (int)reqbody->size, (int)reqbody->nlink);
1316
1317         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1318                               (int)reqbody->nlink);
1319
1320         /*
1321          * Check if client was evicted while we were doing i/o before touching
1322          * network.
1323          */
1324         OBD_ALLOC_PTR(lwi);
1325         if (!lwi)
1326                 GOTO(cleanup_page, rc = -ENOMEM);
1327
1328         if (desc->bd_export->exp_failed)
1329                 rc = -ENOTCONN;
1330         else
1331                 rc = ptlrpc_start_bulk_transfer (desc);
1332         if (rc == 0) {
1333                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1334                                             mdt_bulk_timeout, desc);
1335                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1336                                   desc->bd_export->exp_failed, lwi);
1337                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1338                 if (rc == -ETIMEDOUT) {
1339                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1340                         ptlrpc_abort_bulk(desc);
1341                 } else if (desc->bd_export->exp_failed) {
1342                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1343                         rc = -ENOTCONN;
1344                         ptlrpc_abort_bulk(desc);
1345                 } else if (!desc->bd_success ||
1346                            desc->bd_nob_transferred != desc->bd_nob) {
1347                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1348                                   desc->bd_success ?
1349                                   "truncated" : "network error on",
1350                                   desc->bd_nob_transferred, desc->bd_nob);
1351                         /* XXX should this be a different errno? */
1352                         rc = -ETIMEDOUT;
1353                 }
1354         } else {
1355                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1356         }
1357         if (rc)
1358                 GOTO(cleanup_lwi, rc);
1359         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1360
1361 cleanup_lwi:
1362         OBD_FREE_PTR(lwi);
1363 cleanup_page:
1364         cfs_free_page(page);
1365 desc_cleanup:
1366         ptlrpc_free_bulk(desc);
1367         RETURN(rc);
1368 }
1369 #endif
1370
1371 static int mdt_readpage(struct mdt_thread_info *info)
1372 {
1373         struct mdt_object *object = info->mti_object;
1374         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1375         struct mdt_body   *reqbody;
1376         struct mdt_body   *repbody;
1377         int                rc;
1378         int                i;
1379         ENTRY;
1380
1381         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1382                 RETURN(err_serious(-ENOMEM));
1383
1384         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1385         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1386         if (reqbody == NULL || repbody == NULL)
1387                 RETURN(err_serious(-EFAULT));
1388
1389         /*
1390          * prepare @rdpg before calling lower layers and transfer itself. Here
1391          * reqbody->size contains offset of where to start to read and
1392          * reqbody->nlink contains number bytes to read.
1393          */
1394         rdpg->rp_hash = reqbody->size;
1395         if ((__u64)rdpg->rp_hash != reqbody->size) {
1396                 CERROR("Invalid hash: %#llx != %#llx\n",
1397                        (__u64)rdpg->rp_hash, reqbody->size);
1398                 RETURN(-EFAULT);
1399         }
1400         rdpg->rp_count  = reqbody->nlink;
1401         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1402         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1403         if (rdpg->rp_pages == NULL)
1404                 RETURN(-ENOMEM);
1405
1406         for (i = 0; i < rdpg->rp_npages; ++i) {
1407                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1408                 if (rdpg->rp_pages[i] == NULL)
1409                         GOTO(free_rdpg, rc = -ENOMEM);
1410         }
1411
1412         /* call lower layers to fill allocated pages with directory data */
1413         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1414         if (rc)
1415                 GOTO(free_rdpg, rc);
1416
1417         /* send pages to client */
1418         rc = mdt_sendpage(info, rdpg);
1419
1420         EXIT;
1421 free_rdpg:
1422
1423         for (i = 0; i < rdpg->rp_npages; i++)
1424                 if (rdpg->rp_pages[i] != NULL)
1425                         cfs_free_page(rdpg->rp_pages[i]);
1426         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1427
1428         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1429                 RETURN(0);
1430
1431         return rc;
1432 }
1433
1434 static int mdt_reint_internal(struct mdt_thread_info *info,
1435                               struct mdt_lock_handle *lhc,
1436                               __u32 op)
1437 {
1438         struct req_capsule      *pill = info->mti_pill;
1439         struct mdt_device       *mdt = info->mti_mdt;
1440         struct mdt_body         *repbody;
1441         int                      rc = 0;
1442         ENTRY;
1443
1444         /* pack reply */
1445         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1446                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1447                                      mdt->mdt_max_mdsize);
1448         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1449                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1450                                      mdt->mdt_max_cookiesize);
1451
1452         rc = req_capsule_server_pack(pill);
1453         if (rc != 0) {
1454                 CERROR("Can't pack response, rc %d\n", rc);
1455                 RETURN(err_serious(rc));
1456         }
1457
1458         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1459                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1460                 LASSERT(repbody);
1461                 repbody->eadatasize = 0;
1462                 repbody->aclsize = 0;
1463         }
1464
1465         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1466                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1467
1468         rc = mdt_reint_unpack(info, op);
1469         if (rc != 0) {
1470                 CERROR("Can't unpack reint, rc %d\n", rc);
1471                 GOTO(out_shrink, rc = err_serious(rc));
1472         }
1473
1474         rc = mdt_init_ucred_reint(info);
1475         if (rc)
1476                 GOTO(out_shrink, rc);
1477
1478         rc = mdt_fix_attr_ucred(info, op);
1479         if (rc != 0)
1480                 GOTO(out_ucred, rc = err_serious(rc));
1481
1482         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1483                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1484                 GOTO(out_ucred, rc);
1485         }
1486         rc = mdt_reint_rec(info, lhc);
1487         EXIT;
1488 out_ucred:
1489         mdt_exit_ucred(info);
1490 out_shrink:
1491         mdt_shrink_reply(info);
1492         return rc;
1493 }
1494
1495 static long mdt_reint_opcode(struct mdt_thread_info *info,
1496                              const struct req_format **fmt)
1497 {
1498         struct mdt_rec_reint *rec;
1499         long opc;
1500
1501         opc = err_serious(-EFAULT);
1502         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1503         if (rec != NULL) {
1504                 opc = rec->rr_opcode;
1505                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1506                 if (opc < REINT_MAX && fmt[opc] != NULL)
1507                         req_capsule_extend(info->mti_pill, fmt[opc]);
1508                 else {
1509                         CERROR("Unsupported opc: %ld\n", opc);
1510                         opc = err_serious(opc);
1511                 }
1512         }
1513         return opc;
1514 }
1515
1516 static int mdt_reint(struct mdt_thread_info *info)
1517 {
1518         long opc;
1519         int  rc;
1520
1521         static const struct req_format *reint_fmts[REINT_MAX] = {
1522                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1523                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1524                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1525                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1526                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1527                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1528                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1529         };
1530
1531         ENTRY;
1532
1533         opc = mdt_reint_opcode(info, reint_fmts);
1534         if (opc >= 0) {
1535                 /*
1536                  * No lock possible here from client to pass it to reint code
1537                  * path.
1538                  */
1539                 rc = mdt_reint_internal(info, NULL, opc);
1540         } else {
1541                 rc = opc;
1542         }
1543
1544         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1545         RETURN(rc);
1546 }
1547
1548 /* TODO these two methods not available now. */
1549
1550 /* this should sync the whole device */
1551 static int mdt_device_sync(struct mdt_thread_info *info)
1552 {
1553         return 0;
1554 }
1555
1556 /* this should sync this object */
1557 static int mdt_object_sync(struct mdt_thread_info *info)
1558 {
1559         return 0;
1560 }
1561
1562 static int mdt_sync(struct mdt_thread_info *info)
1563 {
1564         struct req_capsule *pill = info->mti_pill;
1565         struct mdt_body *body;
1566         int rc;
1567         ENTRY;
1568
1569         /* The fid may be zero, so we req_capsule_set manually */
1570         req_capsule_set(pill, &RQF_MDS_SYNC);
1571
1572         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1573         if (body == NULL)
1574                 RETURN(err_serious(-EINVAL));
1575
1576         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1577                 RETURN(err_serious(-ENOMEM));
1578
1579         if (fid_seq(&body->fid1) == 0) {
1580                 /* sync the whole device */
1581                 rc = req_capsule_server_pack(pill);
1582                 if (rc == 0)
1583                         rc = mdt_device_sync(info);
1584                 else
1585                         rc = err_serious(rc);
1586         } else {
1587                 /* sync an object */
1588                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1589                 if (rc == 0) {
1590                         rc = mdt_object_sync(info);
1591                         if (rc == 0) {
1592                                 struct md_object *next;
1593                                 const struct lu_fid *fid;
1594                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1595
1596                                 next = mdt_object_child(info->mti_object);
1597                                 info->mti_attr.ma_need = MA_INODE;
1598                                 info->mti_attr.ma_valid = 0;
1599                                 rc = mo_attr_get(info->mti_env, next,
1600                                                  &info->mti_attr);
1601                                 if (rc == 0) {
1602                                         body = req_capsule_server_get(pill,
1603                                                                 &RMF_MDT_BODY);
1604                                         fid = mdt_object_fid(info->mti_object);
1605                                         mdt_pack_attr2body(info, body, la, fid);
1606                                 }
1607                         }
1608                 } else
1609                         rc = err_serious(rc);
1610         }
1611         RETURN(rc);
1612 }
1613
1614 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1615 {
1616         return err_serious(-EOPNOTSUPP);
1617 }
1618
1619 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1620 {
1621         return err_serious(-EOPNOTSUPP);
1622 }
1623
1624 /*
1625  * OBD PING and other handlers.
1626  */
1627 static int mdt_obd_ping(struct mdt_thread_info *info)
1628 {
1629         int rc;
1630         ENTRY;
1631
1632         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1633
1634         rc = target_handle_ping(mdt_info_req(info));
1635         if (rc < 0)
1636                 rc = err_serious(rc);
1637         RETURN(rc);
1638 }
1639
1640 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1641 {
1642         return err_serious(-EOPNOTSUPP);
1643 }
1644
1645 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1646 {
1647         return err_serious(-EOPNOTSUPP);
1648 }
1649
1650
1651 /*
1652  * DLM handlers.
1653  */
1654 static struct ldlm_callback_suite cbs = {
1655         .lcs_completion = ldlm_server_completion_ast,
1656         .lcs_blocking   = ldlm_server_blocking_ast,
1657         .lcs_glimpse    = NULL
1658 };
1659
1660 static int mdt_enqueue(struct mdt_thread_info *info)
1661 {
1662         struct ptlrpc_request *req;
1663         __u64 req_bits;
1664         int rc;
1665
1666         /*
1667          * info->mti_dlm_req already contains swapped and (if necessary)
1668          * converted dlm request.
1669          */
1670         LASSERT(info->mti_dlm_req != NULL);
1671
1672         req = mdt_info_req(info);
1673
1674         /*
1675          * Lock without inodebits makes no sense and will oops later in
1676          * ldlm. Let's check it now to see if we have wrong lock from client or
1677          * bits get corrupted somewhere in mdt_intent_policy().
1678          */
1679         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1680         /* This is disabled because we need to support liblustre flock.
1681          * LASSERT(req_bits != 0);
1682          */
1683
1684         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1685                                   req, info->mti_dlm_req, &cbs);
1686         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1687         return rc ? err_serious(rc) : req->rq_status;
1688 }
1689
1690 static int mdt_convert(struct mdt_thread_info *info)
1691 {
1692         int rc;
1693         struct ptlrpc_request *req;
1694
1695         LASSERT(info->mti_dlm_req);
1696         req = mdt_info_req(info);
1697         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1698         return rc ? err_serious(rc) : req->rq_status;
1699 }
1700
1701 static int mdt_bl_callback(struct mdt_thread_info *info)
1702 {
1703         CERROR("bl callbacks should not happen on MDS\n");
1704         LBUG();
1705         return err_serious(-EOPNOTSUPP);
1706 }
1707
1708 static int mdt_cp_callback(struct mdt_thread_info *info)
1709 {
1710         CERROR("cp callbacks should not happen on MDS\n");
1711         LBUG();
1712         return err_serious(-EOPNOTSUPP);
1713 }
1714
1715 /*
1716  * sec context handlers
1717  */
1718 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1719 {
1720         int rc;
1721
1722         rc = mdt_handle_idmap(info);
1723
1724         if (unlikely(rc)) {
1725                 struct ptlrpc_request *req = mdt_info_req(info);
1726                 __u32                  opc;
1727
1728                 opc = lustre_msg_get_opc(req->rq_reqmsg);
1729                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1730                         sptlrpc_svc_ctx_invalidate(req);
1731         }
1732
1733         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
1734
1735         return rc;
1736 }
1737
1738 static struct mdt_object *mdt_obj(struct lu_object *o)
1739 {
1740         LASSERT(lu_device_is_mdt(o->lo_dev));
1741         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1742 }
1743
1744 struct mdt_object *mdt_object_find(const struct lu_env *env,
1745                                    struct mdt_device *d,
1746                                    const struct lu_fid *f)
1747 {
1748         struct lu_object *o;
1749         struct mdt_object *m;
1750         ENTRY;
1751
1752         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1753         o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1754         if (unlikely(IS_ERR(o)))
1755                 m = (struct mdt_object *)o;
1756         else
1757                 m = mdt_obj(o);
1758         RETURN(m);
1759 }
1760
1761 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1762                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
1763 {
1764         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1765         ldlm_policy_data_t *policy = &info->mti_policy;
1766         struct ldlm_res_id *res_id = &info->mti_res_id;
1767         int rc;
1768         ENTRY;
1769
1770         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1771         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1772         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1773         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1774
1775         if (mdt_object_exists(o) < 0) {
1776                 if (locality == MDT_CROSS_LOCK) {
1777                         /* cross-ref object fix */
1778                         ibits &= ~MDS_INODELOCK_UPDATE;
1779                         ibits |= MDS_INODELOCK_LOOKUP;
1780                 } else {
1781                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1782                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1783                 }
1784                 /* No PDO lock on remote object */
1785                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1786         }
1787
1788         memset(policy, 0, sizeof(*policy));
1789         fid_build_reg_res_name(mdt_object_fid(o), res_id);
1790
1791         /*
1792          * Take PDO lock on whole directory and build correct @res_id for lock
1793          * on part of directory.
1794          */
1795         if (lh->mlh_pdo_hash != 0) {
1796                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1797                 mdt_lock_pdo_mode(info, o, lh);
1798                 if (lh->mlh_pdo_mode != LCK_NL) {
1799                         /*
1800                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1801                          * is never going to be sent to client and we do not
1802                          * want it slowed down due to possible cancels.
1803                          */
1804                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1805                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1806                                           policy, res_id, LDLM_FL_ATOMIC_CB);
1807                         if (unlikely(rc))
1808                                 RETURN(rc);
1809                 }
1810
1811                 /*
1812                  * Finish res_id initializing by name hash marking patr of
1813                  * directory which is taking modification.
1814                  */
1815                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1816         }
1817
1818         policy->l_inodebits.bits = ibits;
1819
1820         /*
1821          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1822          * going to be sent to client. If it is - mdt_intent_policy() path will
1823          * fix it up and turns FL_LOCAL flag off.
1824          */
1825         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1826                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1827
1828         if (rc)
1829                 GOTO(out, rc);
1830
1831         if (lh->mlh_type == MDT_PDO_LOCK) {
1832                 /* check for exists after object is locked */
1833                 if (mdt_object_exists(o) == 0) {
1834                         /* Non-existent object shouldn't have PDO lock */
1835                         rc = -ESTALE;
1836                 } else {
1837                         /* Non-dir object shouldn't have PDO lock */
1838                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1839                 }
1840         }
1841 out:
1842         if (rc)
1843                 mdt_object_unlock(info, o, lh, 1);
1844
1845
1846         RETURN(rc);
1847 }
1848
1849 static inline
1850 void mdt_save_lock(struct ptlrpc_request *req, struct lustre_handle *h,
1851                    ldlm_mode_t mode, int decref)
1852 {
1853         ENTRY;
1854
1855         if (lustre_handle_is_used(h)) {
1856                 if (decref)
1857                         mdt_fid_unlock(h, mode);
1858                 else
1859                         ptlrpc_save_lock(req, h, mode);
1860                 h->cookie = 0ull;
1861         }
1862
1863         EXIT;
1864 }
1865
1866 /*
1867  * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1868  * to save this lock in req.  when transaction committed, req will be released,
1869  * and lock will, too.
1870  */
1871 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1872                        struct mdt_lock_handle *lh, int decref)
1873 {
1874         struct ptlrpc_request *req = mdt_info_req(info);
1875         ENTRY;
1876
1877         mdt_save_lock(req, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
1878         mdt_save_lock(req, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
1879
1880         EXIT;
1881 }
1882
1883 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1884                                         const struct lu_fid *f,
1885                                         struct mdt_lock_handle *lh,
1886                                         __u64 ibits)
1887 {
1888         struct mdt_object *o;
1889
1890         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1891         if (!IS_ERR(o)) {
1892                 int rc;
1893
1894                 rc = mdt_object_lock(info, o, lh, ibits,
1895                                      MDT_LOCAL_LOCK);
1896                 if (rc != 0) {
1897                         mdt_object_put(info->mti_env, o);
1898                         o = ERR_PTR(rc);
1899                 }
1900         }
1901         return o;
1902 }
1903
1904 void mdt_object_unlock_put(struct mdt_thread_info * info,
1905                            struct mdt_object * o,
1906                            struct mdt_lock_handle *lh,
1907                            int decref)
1908 {
1909         mdt_object_unlock(info, o, lh, decref);
1910         mdt_object_put(info->mti_env, o);
1911 }
1912
1913 static struct mdt_handler *mdt_handler_find(__u32 opc,
1914                                             struct mdt_opc_slice *supported)
1915 {
1916         struct mdt_opc_slice *s;
1917         struct mdt_handler   *h;
1918
1919         h = NULL;
1920         for (s = supported; s->mos_hs != NULL; s++) {
1921                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1922                         h = s->mos_hs + (opc - s->mos_opc_start);
1923                         if (likely(h->mh_opc != 0))
1924                                 LASSERT(h->mh_opc == opc);
1925                         else
1926                                 h = NULL; /* unsupported opc */
1927                         break;
1928                 }
1929         }
1930         return h;
1931 }
1932
1933 static int mdt_lock_resname_compat(struct mdt_device *m,
1934                                    struct ldlm_request *req)
1935 {
1936         /* XXX something... later. */
1937         return 0;
1938 }
1939
1940 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1941 {
1942         /* XXX something... later. */
1943         return 0;
1944 }
1945
1946 /*
1947  * Generic code handling requests that have struct mdt_body passed in:
1948  *
1949  *  - extract mdt_body from request and save it in @info, if present;
1950  *
1951  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
1952  *  @info;
1953  *
1954  *  - if HABEO_CORPUS flag is set for this request type check whether object
1955  *  actually exists on storage (lu_object_exists()).
1956  *
1957  */
1958 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1959 {
1960         const struct mdt_body    *body;
1961         struct mdt_object        *obj;
1962         const struct lu_env      *env;
1963         struct req_capsule       *pill;
1964         int                       rc;
1965         ENTRY;
1966
1967         env = info->mti_env;
1968         pill = info->mti_pill;
1969
1970         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1971         if (body == NULL)
1972                 RETURN(-EFAULT);
1973
1974         if (!(body->valid & OBD_MD_FLID))
1975                 RETURN(0);
1976
1977         if (!fid_is_sane(&body->fid1)) {
1978                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1979                 RETURN(-EINVAL);
1980         }
1981
1982         /*
1983          * Do not get size or any capa fields before we check that request
1984          * contains capa actually. There are some requests which do not, for
1985          * instance MDS_IS_SUBDIR.
1986          */
1987         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1988             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1989                 mdt_set_capainfo(info, 0, &body->fid1,
1990                                  req_capsule_client_get(pill, &RMF_CAPA1));
1991
1992         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1993         if (!IS_ERR(obj)) {
1994                 if ((flags & HABEO_CORPUS) &&
1995                     !mdt_object_exists(obj)) {
1996                         mdt_object_put(env, obj);
1997                         /* for capability renew ENOENT will be handled in
1998                          * mdt_renew_capa */
1999                         if (body->valid & OBD_MD_FLOSSCAPA)
2000                                 rc = 0;
2001                         else
2002                                 rc = -ENOENT;
2003                 } else {
2004                         info->mti_object = obj;
2005                         rc = 0;
2006                 }
2007         } else
2008                 rc = PTR_ERR(obj);
2009
2010         RETURN(rc);
2011 }
2012
2013 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2014 {
2015         struct req_capsule *pill = info->mti_pill;
2016         int rc;
2017         ENTRY;
2018
2019         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2020                 rc = mdt_body_unpack(info, flags);
2021         else
2022                 rc = 0;
2023
2024         if (rc == 0 && (flags & HABEO_REFERO)) {
2025                 struct mdt_device *mdt = info->mti_mdt;
2026
2027                 /* Pack reply. */
2028                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2029                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2030                                              mdt->mdt_max_mdsize);
2031                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2032                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2033                                              mdt->mdt_max_cookiesize);
2034
2035                 rc = req_capsule_server_pack(pill);
2036         }
2037         RETURN(rc);
2038 }
2039
2040 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2041 {
2042         struct md_device *next = m->mdt_child;
2043
2044         return next->md_ops->mdo_init_capa_ctxt(env, next,
2045                                                 m->mdt_opts.mo_mds_capa,
2046                                                 m->mdt_capa_timeout,
2047                                                 m->mdt_capa_alg,
2048                                                 m->mdt_capa_keys);
2049 }
2050
2051 /*
2052  * Invoke handler for this request opc. Also do necessary preprocessing
2053  * (according to handler ->mh_flags), and post-processing (setting of
2054  * ->last_{xid,committed}).
2055  */
2056 static int mdt_req_handle(struct mdt_thread_info *info,
2057                           struct mdt_handler *h, struct ptlrpc_request *req)
2058 {
2059         int   rc, serious = 0;
2060         __u32 flags;
2061
2062         ENTRY;
2063
2064         LASSERT(h->mh_act != NULL);
2065         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2066         LASSERT(current->journal_info == NULL);
2067
2068         /*
2069          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2070          * to put same checks into handlers like mdt_close(), mdt_reint(),
2071          * etc., without talking to mdt authors first. Checking same thing
2072          * there again is useless and returning 0 error wihtout packing reply
2073          * is buggy! Handlers either pack reply or return error.
2074          *
2075          * We return 0 here and do not send any reply in order to emulate
2076          * network failure. Do not send any reply in case any of NET related
2077          * fail_id has occured.
2078          */
2079         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2080                 RETURN(0);
2081
2082         rc = 0;
2083         flags = h->mh_flags;
2084         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2085
2086         if (h->mh_fmt != NULL) {
2087                 req_capsule_set(info->mti_pill, h->mh_fmt);
2088                 rc = mdt_unpack_req_pack_rep(info, flags);
2089         }
2090
2091         if (rc == 0 && flags & MUTABOR &&
2092             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2093                 /* should it be rq_status? */
2094                 rc = -EROFS;
2095
2096         if (rc == 0 && flags & HABEO_CLAVIS) {
2097                 struct ldlm_request *dlm_req;
2098
2099                 LASSERT(h->mh_fmt != NULL);
2100
2101                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2102                 if (dlm_req != NULL) {
2103                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2104                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2105                                                              dlm_req);
2106                         info->mti_dlm_req = dlm_req;
2107                 } else {
2108                         rc = -EFAULT;
2109                 }
2110         }
2111
2112         /* capability setting changed via /proc, needs reinitialize ctxt */
2113         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2114                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2115                 info->mti_mdt->mdt_capa_conf = 0;
2116         }
2117
2118         if (likely(rc == 0)) {
2119                 /*
2120                  * Process request, there can be two types of rc:
2121                  * 1) errors with msg unpack/pack, other failures outside the
2122                  * operation itself. This is counted as serious errors;
2123                  * 2) errors during fs operation, should be placed in rq_status
2124                  * only
2125                  */
2126                 rc = h->mh_act(info);
2127                 if (rc == 0 &&
2128                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2129                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2130                                   "pack reply and returned 0 error\n",
2131                                   h->mh_name);
2132                         LBUG();
2133                 }
2134                 serious = is_serious(rc);
2135                 rc = clear_serious(rc);
2136         } else
2137                 serious = 1;
2138
2139         req->rq_status = rc;
2140
2141         /*
2142          * ELDLM_* codes which > 0 should be in rq_status only as well as
2143          * all non-serious errors.
2144          */
2145         if (rc > 0 || !serious)
2146                 rc = 0;
2147
2148         LASSERT(current->journal_info == NULL);
2149
2150         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2151             info->mti_mdt->mdt_opts.mo_compat_resname) {
2152                 struct ldlm_reply *dlmrep;
2153
2154                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2155                 if (dlmrep != NULL)
2156                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2157         }
2158
2159         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2160         if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2161                 target_committed_to_req(req);
2162
2163         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2164                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2165                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2166                 LBUG();
2167         }
2168
2169         target_send_reply(req, rc, info->mti_fail_id);
2170         RETURN(0);
2171 }
2172
2173 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2174 {
2175         lh->mlh_type = MDT_NUL_LOCK;
2176         lh->mlh_reg_lh.cookie = 0ull;
2177         lh->mlh_reg_mode = LCK_MINMODE;
2178         lh->mlh_pdo_lh.cookie = 0ull;
2179         lh->mlh_pdo_mode = LCK_MINMODE;
2180 }
2181
2182 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2183 {
2184         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2185         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2186 }
2187
2188 /*
2189  * Initialize fields of struct mdt_thread_info. Other fields are left in
2190  * uninitialized state, because it's too expensive to zero out whole
2191  * mdt_thread_info (> 1K) on each request arrival.
2192  */
2193 static void mdt_thread_info_init(struct ptlrpc_request *req,
2194                                  struct mdt_thread_info *info)
2195 {
2196         int i;
2197         struct md_capainfo *ci;
2198
2199         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2200         info->mti_pill = &req->rq_pill;
2201
2202         /* lock handle */
2203         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2204                 mdt_lock_handle_init(&info->mti_lh[i]);
2205
2206         /* mdt device: it can be NULL while CONNECT */
2207         if (req->rq_export) {
2208                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2209                 info->mti_exp = req->rq_export;
2210         } else
2211                 info->mti_mdt = NULL;
2212         info->mti_env = req->rq_svc_thread->t_env;
2213         ci = md_capainfo(info->mti_env);
2214         memset(ci, 0, sizeof *ci);
2215
2216         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2217         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2218
2219         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2220         info->mti_body = NULL;
2221         info->mti_object = NULL;
2222         info->mti_dlm_req = NULL;
2223         info->mti_has_trans = 0;
2224         info->mti_no_need_trans = 0;
2225         info->mti_cross_ref = 0;
2226         info->mti_opdata = 0;
2227
2228         /* To not check for split by default. */
2229         info->mti_spec.sp_ck_split = 0;
2230 }
2231
2232 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2233 {
2234         int i;
2235
2236         req_capsule_fini(info->mti_pill);
2237         if (info->mti_object != NULL) {
2238                 mdt_object_put(info->mti_env, info->mti_object);
2239                 info->mti_object = NULL;
2240         }
2241         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2242                 mdt_lock_handle_fini(&info->mti_lh[i]);
2243         info->mti_env = NULL;
2244 }
2245
2246 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2247                                        struct obd_device *obd, int *process)
2248 {
2249         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2250         case MDS_CONNECT: /* This will never get here, but for completeness. */
2251         case OST_CONNECT: /* This will never get here, but for completeness. */
2252         case MDS_DISCONNECT:
2253         case OST_DISCONNECT:
2254                *process = 1;
2255                RETURN(0);
2256
2257         case MDS_CLOSE:
2258         case MDS_DONE_WRITING:
2259         case MDS_SYNC: /* used in unmounting */
2260         case OBD_PING:
2261         case MDS_REINT:
2262         case SEQ_QUERY:
2263         case FLD_QUERY:
2264         case LDLM_ENQUEUE:
2265                 *process = target_queue_recovery_request(req, obd);
2266                 RETURN(0);
2267
2268         default:
2269                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2270                 *process = -EAGAIN;
2271                 RETURN(0);
2272         }
2273 }
2274
2275 /*
2276  * Handle recovery. Return:
2277  *        +1: continue request processing;
2278  *       -ve: abort immediately with the given error code;
2279  *         0: send reply with error code in req->rq_status;
2280  */
2281 static int mdt_recovery(struct mdt_thread_info *info)
2282 {
2283         struct ptlrpc_request *req = mdt_info_req(info);
2284         int recovering;
2285         struct obd_device *obd;
2286
2287         ENTRY;
2288
2289         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2290         case MDS_CONNECT:
2291         case SEC_CTX_INIT:
2292         case SEC_CTX_INIT_CONT:
2293         case SEC_CTX_FINI:
2294                 {
2295 #if 0
2296                         int rc;
2297
2298                         rc = mdt_handle_idmap(info);
2299                         if (rc)
2300                                 RETURN(rc);
2301                         else
2302 #endif
2303                                 RETURN(+1);
2304                 }
2305         }
2306
2307         if (unlikely(req->rq_export == NULL)) {
2308                 CERROR("operation %d on unconnected MDS from %s\n",
2309                        lustre_msg_get_opc(req->rq_reqmsg),
2310                        libcfs_id2str(req->rq_peer));
2311                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2312                  * mds_A will get -ENOTCONN(especially for ping req),
2313                  * which will cause that mds_A deactive timeout, then when
2314                  * mds_A cleanup, the cleanup process will be suspended since
2315                  * deactive timeout is not zero.
2316                  */
2317                 req->rq_status = -ENOTCONN;
2318                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2319                 RETURN(0);
2320         }
2321
2322         /* sanity check: if the xid matches, the request must be marked as a
2323          * resent or replayed */
2324         if (req_xid_is_last(req)) {
2325                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2326                       (MSG_RESENT | MSG_REPLAY))) {
2327                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2328                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2329                                   lustre_msg_get_flags(req->rq_reqmsg));
2330                         LBUG();
2331                         req->rq_status = -ENOTCONN;
2332                         RETURN(-ENOTCONN);
2333                 }
2334         }
2335
2336         /* else: note the opposite is not always true; a RESENT req after a
2337          * failover will usually not match the last_xid, since it was likely
2338          * never committed. A REPLAYed request will almost never match the
2339          * last xid, however it could for a committed, but still retained,
2340          * open. */
2341
2342         obd = req->rq_export->exp_obd;
2343
2344         /* Check for aborted recovery... */
2345         spin_lock_bh(&obd->obd_processing_task_lock);
2346         recovering = obd->obd_recovering;
2347         spin_unlock_bh(&obd->obd_processing_task_lock);
2348         if (unlikely(recovering)) {
2349                 int rc;
2350                 int should_process;
2351                 DEBUG_REQ(D_INFO, req, "Got new replay");
2352                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2353                 if (rc != 0 || !should_process)
2354                         RETURN(rc);
2355                 else if (should_process < 0) {
2356                         req->rq_status = should_process;
2357                         rc = ptlrpc_error(req);
2358                         RETURN(rc);
2359                 }
2360         }
2361         RETURN(+1);
2362 }
2363
2364 static int mdt_msg_check_version(struct lustre_msg *msg)
2365 {
2366         int rc;
2367
2368         switch (lustre_msg_get_opc(msg)) {
2369         case MDS_CONNECT:
2370         case MDS_DISCONNECT:
2371         case OBD_PING:
2372         case SEC_CTX_INIT:
2373         case SEC_CTX_INIT_CONT:
2374         case SEC_CTX_FINI:
2375                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2376                 if (rc)
2377                         CERROR("bad opc %u version %08x, expecting %08x\n",
2378                                lustre_msg_get_opc(msg),
2379                                lustre_msg_get_version(msg),
2380                                LUSTRE_OBD_VERSION);
2381                 break;
2382         case MDS_GETSTATUS:
2383         case MDS_GETATTR:
2384         case MDS_GETATTR_NAME:
2385         case MDS_STATFS:
2386         case MDS_READPAGE:
2387         case MDS_WRITEPAGE:
2388         case MDS_IS_SUBDIR:
2389         case MDS_REINT:
2390         case MDS_CLOSE:
2391         case MDS_DONE_WRITING:
2392         case MDS_PIN:
2393         case MDS_SYNC:
2394         case MDS_GETXATTR:
2395         case MDS_SETXATTR:
2396         case MDS_SET_INFO:
2397         case MDS_QUOTACHECK:
2398         case MDS_QUOTACTL:
2399         case QUOTA_DQACQ:
2400         case QUOTA_DQREL:
2401         case SEQ_QUERY:
2402         case FLD_QUERY:
2403                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2404                 if (rc)
2405                         CERROR("bad opc %u version %08x, expecting %08x\n",
2406                                lustre_msg_get_opc(msg),
2407                                lustre_msg_get_version(msg),
2408                                LUSTRE_MDS_VERSION);
2409                 break;
2410         case LDLM_ENQUEUE:
2411         case LDLM_CONVERT:
2412         case LDLM_BL_CALLBACK:
2413         case LDLM_CP_CALLBACK:
2414                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2415                 if (rc)
2416                         CERROR("bad opc %u version %08x, expecting %08x\n",
2417                                lustre_msg_get_opc(msg),
2418                                lustre_msg_get_version(msg),
2419                                LUSTRE_DLM_VERSION);
2420                 break;
2421         case OBD_LOG_CANCEL:
2422         case LLOG_ORIGIN_HANDLE_CREATE:
2423         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2424         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2425         case LLOG_ORIGIN_HANDLE_CLOSE:
2426         case LLOG_ORIGIN_HANDLE_DESTROY:
2427         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2428         case LLOG_CATINFO:
2429                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2430                 if (rc)
2431                         CERROR("bad opc %u version %08x, expecting %08x\n",
2432                                lustre_msg_get_opc(msg),
2433                                lustre_msg_get_version(msg),
2434                                LUSTRE_LOG_VERSION);
2435                 break;
2436         default:
2437                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2438                 rc = -ENOTSUPP;
2439         }
2440         return rc;
2441 }
2442
2443 static int mdt_handle0(struct ptlrpc_request *req,
2444                        struct mdt_thread_info *info,
2445                        struct mdt_opc_slice *supported)
2446 {
2447         struct mdt_handler *h;
2448         struct lustre_msg  *msg;
2449         int                 rc;
2450
2451         ENTRY;
2452
2453         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2454                 RETURN(0);
2455
2456         LASSERT(current->journal_info == NULL);
2457
2458         msg = req->rq_reqmsg;
2459         rc = mdt_msg_check_version(msg);
2460         if (likely(rc == 0)) {
2461                 rc = mdt_recovery(info);
2462                 if (likely(rc == +1)) {
2463                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2464                                              supported);
2465                         if (likely(h != NULL)) {
2466                                 rc = mdt_req_handle(info, h, req);
2467                         } else {
2468                                 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2469                                 req->rq_status = -ENOTSUPP;
2470                                 rc = ptlrpc_error(req);
2471                                 RETURN(rc);
2472                         }
2473                 }
2474         } else
2475                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2476         RETURN(rc);
2477 }
2478
2479 /*
2480  * MDT handler function called by ptlrpc service thread when request comes.
2481  *
2482  * XXX common "target" functionality should be factored into separate module
2483  * shared by mdt, ost and stand-alone services like fld.
2484  */
2485 static int mdt_handle_common(struct ptlrpc_request *req,
2486                              struct mdt_opc_slice *supported)
2487 {
2488         struct lu_env          *env;
2489         struct mdt_thread_info *info;
2490         int                     rc;
2491         ENTRY;
2492
2493         env = req->rq_svc_thread->t_env;
2494         LASSERT(env != NULL);
2495         LASSERT(env->le_ses != NULL);
2496         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2497         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2498         LASSERT(info != NULL);
2499
2500         mdt_thread_info_init(req, info);
2501
2502         rc = mdt_handle0(req, info, supported);
2503
2504         mdt_thread_info_fini(info);
2505         RETURN(rc);
2506 }
2507
2508 /*
2509  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2510  * as well.
2511  */
2512 int mdt_recovery_handle(struct ptlrpc_request *req)
2513 {
2514         int rc;
2515         ENTRY;
2516
2517         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2518         case FLD_QUERY:
2519                 rc = mdt_handle_common(req, mdt_fld_handlers);
2520                 break;
2521         case SEQ_QUERY:
2522                 rc = mdt_handle_common(req, mdt_seq_handlers);
2523                 break;
2524         default:
2525                 rc = mdt_handle_common(req, mdt_regular_handlers);
2526                 break;
2527         }
2528
2529         RETURN(rc);
2530 }
2531
2532 static int mdt_regular_handle(struct ptlrpc_request *req)
2533 {
2534         return mdt_handle_common(req, mdt_regular_handlers);
2535 }
2536
2537 static int mdt_readpage_handle(struct ptlrpc_request *req)
2538 {
2539         return mdt_handle_common(req, mdt_readpage_handlers);
2540 }
2541
2542 static int mdt_xmds_handle(struct ptlrpc_request *req)
2543 {
2544         return mdt_handle_common(req, mdt_xmds_handlers);
2545 }
2546
2547 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2548 {
2549         return mdt_handle_common(req, mdt_seq_handlers);
2550 }
2551
2552 static int mdt_mdss_handle(struct ptlrpc_request *req)
2553 {
2554         return mdt_handle_common(req, mdt_seq_handlers);
2555 }
2556
2557 static int mdt_dtss_handle(struct ptlrpc_request *req)
2558 {
2559         return mdt_handle_common(req, mdt_seq_handlers);
2560 }
2561
2562 static int mdt_fld_handle(struct ptlrpc_request *req)
2563 {
2564         return mdt_handle_common(req, mdt_fld_handlers);
2565 }
2566
2567 enum mdt_it_code {
2568         MDT_IT_OPEN,
2569         MDT_IT_OCREAT,
2570         MDT_IT_CREATE,
2571         MDT_IT_GETATTR,
2572         MDT_IT_READDIR,
2573         MDT_IT_LOOKUP,
2574         MDT_IT_UNLINK,
2575         MDT_IT_TRUNC,
2576         MDT_IT_GETXATTR,
2577         MDT_IT_NR
2578 };
2579
2580 static int mdt_intent_getattr(enum mdt_it_code opcode,
2581                               struct mdt_thread_info *info,
2582                               struct ldlm_lock **,
2583                               int);
2584 static int mdt_intent_reint(enum mdt_it_code opcode,
2585                             struct mdt_thread_info *info,
2586                             struct ldlm_lock **,
2587                             int);
2588
2589 static struct mdt_it_flavor {
2590         const struct req_format *it_fmt;
2591         __u32                    it_flags;
2592         int                    (*it_act)(enum mdt_it_code ,
2593                                          struct mdt_thread_info *,
2594                                          struct ldlm_lock **,
2595                                          int);
2596         long                     it_reint;
2597 } mdt_it_flavor[] = {
2598         [MDT_IT_OPEN]     = {
2599                 .it_fmt   = &RQF_LDLM_INTENT,
2600                 /*.it_flags = HABEO_REFERO,*/
2601                 .it_flags = 0,
2602                 .it_act   = mdt_intent_reint,
2603                 .it_reint = REINT_OPEN
2604         },
2605         [MDT_IT_OCREAT]   = {
2606                 .it_fmt   = &RQF_LDLM_INTENT,
2607                 .it_flags = MUTABOR,
2608                 .it_act   = mdt_intent_reint,
2609                 .it_reint = REINT_OPEN
2610         },
2611         [MDT_IT_CREATE]   = {
2612                 .it_fmt   = &RQF_LDLM_INTENT,
2613                 .it_flags = MUTABOR,
2614                 .it_act   = mdt_intent_reint,
2615                 .it_reint = REINT_CREATE
2616         },
2617         [MDT_IT_GETATTR]  = {
2618                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2619                 .it_flags = HABEO_REFERO,
2620                 .it_act   = mdt_intent_getattr
2621         },
2622         [MDT_IT_READDIR]  = {
2623                 .it_fmt   = NULL,
2624                 .it_flags = 0,
2625                 .it_act   = NULL
2626         },
2627         [MDT_IT_LOOKUP]   = {
2628                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2629                 .it_flags = HABEO_REFERO,
2630                 .it_act   = mdt_intent_getattr
2631         },
2632         [MDT_IT_UNLINK]   = {
2633                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2634                 .it_flags = MUTABOR,
2635                 .it_act   = NULL,
2636                 .it_reint = REINT_UNLINK
2637         },
2638         [MDT_IT_TRUNC]    = {
2639                 .it_fmt   = NULL,
2640                 .it_flags = MUTABOR,
2641                 .it_act   = NULL
2642         },
2643         [MDT_IT_GETXATTR] = {
2644                 .it_fmt   = NULL,
2645                 .it_flags = 0,
2646                 .it_act   = NULL
2647         }
2648 };
2649
2650 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2651                             struct ldlm_lock **lockp,
2652                             struct ldlm_lock *new_lock,
2653                             struct mdt_lock_handle *lh,
2654                             int flags)
2655 {
2656         struct ptlrpc_request  *req = mdt_info_req(info);
2657         struct ldlm_lock       *lock = *lockp;
2658
2659         /*
2660          * Get new lock only for cases when possible resent did not find any
2661          * lock.
2662          */
2663         if (new_lock == NULL)
2664                 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2665
2666         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2667                 lh->mlh_reg_lh.cookie = 0;
2668                 RETURN(0);
2669         }
2670
2671         LASSERTF(new_lock != NULL,
2672                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2673
2674         /*
2675          * If we've already given this lock to a client once, then we should
2676          * have no readers or writers.  Otherwise, we should have one reader
2677          * _or_ writer ref (which will be zeroed below) before returning the
2678          * lock to a client.
2679          */
2680         if (new_lock->l_export == req->rq_export) {
2681                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2682         } else {
2683                 LASSERT(new_lock->l_export == NULL);
2684                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2685         }
2686
2687         *lockp = new_lock;
2688
2689         if (new_lock->l_export == req->rq_export) {
2690                 /*
2691                  * Already gave this to the client, which means that we
2692                  * reconstructed a reply.
2693                  */
2694                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2695                         MSG_RESENT);
2696                 lh->mlh_reg_lh.cookie = 0;
2697                 RETURN(ELDLM_LOCK_REPLACED);
2698         }
2699
2700         /* This lock might already be given to the client by an resent req,
2701          * in this case we should return ELDLM_LOCK_ABORTED,
2702          * so we should check led_held_locks here, but it will affect
2703          * performance, FIXME
2704          */
2705         /* Fixup the lock to be given to the client */
2706         lock_res_and_lock(new_lock);
2707         new_lock->l_readers = 0;
2708         new_lock->l_writers = 0;
2709
2710         new_lock->l_export = class_export_get(req->rq_export);
2711         spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2712         list_add(&new_lock->l_export_chain,
2713                  &new_lock->l_export->exp_ldlm_data.led_held_locks);
2714         spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2715
2716         new_lock->l_blocking_ast = lock->l_blocking_ast;
2717         new_lock->l_completion_ast = lock->l_completion_ast;
2718         new_lock->l_remote_handle = lock->l_remote_handle;
2719         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2720
2721         unlock_res_and_lock(new_lock);
2722         LDLM_LOCK_PUT(new_lock);
2723         lh->mlh_reg_lh.cookie = 0;
2724
2725         RETURN(ELDLM_LOCK_REPLACED);
2726 }
2727
2728 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2729                                     struct ldlm_lock *new_lock,
2730                                     struct ldlm_lock **old_lock,
2731                                     struct mdt_lock_handle *lh)
2732 {
2733         struct ptlrpc_request  *req = mdt_info_req(info);
2734         struct obd_export      *exp = req->rq_export;
2735         struct lustre_handle    remote_hdl;
2736         struct ldlm_request    *dlmreq;
2737         struct list_head       *iter;
2738
2739         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2740                 return;
2741
2742         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2743         remote_hdl = dlmreq->lock_handle[0];
2744
2745         spin_lock(&exp->exp_ldlm_data.led_lock);
2746         list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2747                 struct ldlm_lock *lock;
2748                 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2749                 if (lock == new_lock)
2750                         continue;
2751                 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2752                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2753                         lh->mlh_reg_mode = lock->l_granted_mode;
2754
2755                         LDLM_DEBUG(lock, "restoring lock cookie");
2756                         DEBUG_REQ(D_DLMTRACE, req,
2757                                   "restoring lock cookie "LPX64,
2758                                   lh->mlh_reg_lh.cookie);
2759                         if (old_lock)
2760                                 *old_lock = LDLM_LOCK_GET(lock);
2761                         spin_unlock(&exp->exp_ldlm_data.led_lock);
2762                         return;
2763                 }
2764         }
2765         spin_unlock(&exp->exp_ldlm_data.led_lock);
2766
2767         /*
2768          * If the xid matches, then we know this is a resent request, and allow
2769          * it. (It's probably an OPEN, for which we don't send a lock.
2770          */
2771         if (req_xid_is_last(req))
2772                 return;
2773
2774         /*
2775          * This remote handle isn't enqueued, so we never received or processed
2776          * this request.  Clear MSG_RESENT, because it can be handled like any
2777          * normal request now.
2778          */
2779         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2780
2781         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2782                   remote_hdl.cookie);
2783 }
2784
2785 static int mdt_intent_getattr(enum mdt_it_code opcode,
2786                               struct mdt_thread_info *info,
2787                               struct ldlm_lock **lockp,
2788                               int flags)
2789 {
2790         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2791         struct ldlm_lock       *new_lock = NULL;
2792         __u64                   child_bits;
2793         struct ldlm_reply      *ldlm_rep;
2794         struct ptlrpc_request  *req;
2795         struct mdt_body        *reqbody;
2796         struct mdt_body        *repbody;
2797         int                     rc;
2798         ENTRY;
2799
2800         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2801         LASSERT(reqbody);
2802
2803         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2804         LASSERT(repbody);
2805
2806         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2807         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2808         repbody->eadatasize = 0;
2809         repbody->aclsize = 0;
2810
2811         switch (opcode) {
2812         case MDT_IT_LOOKUP:
2813                 child_bits = MDS_INODELOCK_LOOKUP;
2814                 break;
2815         case MDT_IT_GETATTR:
2816                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2817                 break;
2818         default:
2819                 CERROR("Unhandled till now");
2820                 GOTO(out_shrink, rc = -EINVAL);
2821         }
2822
2823         rc = mdt_init_ucred(info, reqbody);
2824         if (rc)
2825                 GOTO(out_shrink, rc);
2826
2827         req = info->mti_pill->rc_req;
2828         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2829         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2830
2831         /* Get lock from request for possible resent case. */
2832         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2833
2834         ldlm_rep->lock_policy_res2 =
2835                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2836
2837         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2838                 ldlm_rep->lock_policy_res2 = 0;
2839         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2840             ldlm_rep->lock_policy_res2) {
2841                 lhc->mlh_reg_lh.cookie = 0ull;
2842                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2843         }
2844
2845         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2846         EXIT;
2847 out_ucred:
2848         mdt_exit_ucred(info);
2849 out_shrink:
2850         mdt_shrink_reply(info);
2851         return rc;
2852 }
2853
2854 static int mdt_intent_reint(enum mdt_it_code opcode,
2855                             struct mdt_thread_info *info,
2856                             struct ldlm_lock **lockp,
2857                             int flags)
2858 {
2859         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2860         struct ldlm_reply      *rep = NULL;
2861         long                    opc;
2862         int                     rc;
2863
2864         static const struct req_format *intent_fmts[REINT_MAX] = {
2865                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
2866                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
2867         };
2868
2869         ENTRY;
2870
2871         opc = mdt_reint_opcode(info, intent_fmts);
2872         if (opc < 0)
2873                 RETURN(opc);
2874
2875         if (mdt_it_flavor[opcode].it_reint != opc) {
2876                 CERROR("Reint code %ld doesn't match intent: %d\n",
2877                        opc, opcode);
2878                 RETURN(err_serious(-EPROTO));
2879         }
2880
2881         /* Get lock from request for possible resent case. */
2882         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2883
2884         rc = mdt_reint_internal(info, lhc, opc);
2885
2886         /* Check whether the reply has been packed successfully. */
2887         if (mdt_info_req(info)->rq_repmsg != NULL)
2888                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2889         if (rep == NULL)
2890                 RETURN(err_serious(-EFAULT));
2891
2892         /* MDC expects this in any case */
2893         if (rc != 0)
2894                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2895
2896         /* Cross-ref case, the lock should be returned to the client */
2897         if (rc == -EREMOTE) {
2898                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2899                 rep->lock_policy_res2 = 0;
2900                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2901                 RETURN(rc);
2902         }
2903         rep->lock_policy_res2 = clear_serious(rc);
2904
2905         lhc->mlh_reg_lh.cookie = 0ull;
2906         if (rc == -ENOTCONN || rc == -ENODEV) {
2907                 /*
2908                  * If it is the disconnect error (ENODEV & ENOCONN), the error
2909                  * will be returned by rq_status, and client at ptlrpc layer
2910                  * will detect this, then disconnect, reconnect the import
2911                  * immediately, instead of impacting the following the rpc.
2912                  */
2913                 RETURN(rc);
2914         } else {
2915                 /*
2916                  * For other cases, the error will be returned by intent.
2917                  * and client will retrieve the result from intent.
2918                  */
2919                  /*
2920                   * FIXME: when open lock is finished, that should be
2921                   * checked here.
2922                   */
2923                 RETURN(ELDLM_LOCK_ABORTED);
2924         }
2925 }
2926
2927 static int mdt_intent_code(long itcode)
2928 {
2929         int rc;
2930
2931         switch(itcode) {
2932         case IT_OPEN:
2933                 rc = MDT_IT_OPEN;
2934                 break;
2935         case IT_OPEN|IT_CREAT:
2936                 rc = MDT_IT_OCREAT;
2937                 break;
2938         case IT_CREAT:
2939                 rc = MDT_IT_CREATE;
2940                 break;
2941         case IT_READDIR:
2942                 rc = MDT_IT_READDIR;
2943                 break;
2944         case IT_GETATTR:
2945                 rc = MDT_IT_GETATTR;
2946                 break;
2947         case IT_LOOKUP:
2948                 rc = MDT_IT_LOOKUP;
2949                 break;
2950         case IT_UNLINK:
2951                 rc = MDT_IT_UNLINK;
2952                 break;
2953         case IT_TRUNC:
2954                 rc = MDT_IT_TRUNC;
2955                 break;
2956         case IT_GETXATTR:
2957                 rc = MDT_IT_GETXATTR;
2958                 break;
2959         default:
2960                 CERROR("Unknown intent opcode: %ld\n", itcode);
2961                 rc = -EINVAL;
2962                 break;
2963         }
2964         return rc;
2965 }
2966
2967 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2968                           struct ldlm_lock **lockp, int flags)
2969 {
2970         struct req_capsule   *pill;
2971         struct mdt_it_flavor *flv;
2972         int opc;
2973         int rc;
2974         ENTRY;
2975
2976         opc = mdt_intent_code(itopc);
2977         if (opc < 0)
2978                 RETURN(-EINVAL);
2979
2980         pill = info->mti_pill;
2981         flv  = &mdt_it_flavor[opc];
2982
2983         if (flv->it_fmt != NULL)
2984                 req_capsule_extend(pill, flv->it_fmt);
2985
2986         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2987         if (rc == 0) {
2988                 struct ptlrpc_request *req = mdt_info_req(info);
2989                 if (flv->it_flags & MUTABOR &&
2990                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2991                         RETURN(-EROFS);
2992         }
2993         if (rc == 0 && flv->it_act != NULL) {
2994                 /* execute policy */
2995                 rc = flv->it_act(opc, info, lockp, flags);
2996         } else {
2997                 rc = -EOPNOTSUPP;
2998         }
2999         RETURN(rc);
3000 }
3001
3002 static int mdt_intent_policy(struct ldlm_namespace *ns,
3003                              struct ldlm_lock **lockp, void *req_cookie,
3004                              ldlm_mode_t mode, int flags, void *data)
3005 {
3006         struct mdt_thread_info *info;
3007         struct ptlrpc_request  *req  =  req_cookie;
3008         struct ldlm_intent     *it;
3009         struct req_capsule     *pill;
3010         int rc;
3011
3012         ENTRY;
3013
3014         LASSERT(req != NULL);
3015
3016         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3017                                   &mdt_thread_key);
3018         LASSERT(info != NULL);
3019         pill = info->mti_pill;
3020         LASSERT(pill->rc_req == req);
3021
3022         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3023                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3024                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3025                 if (it != NULL) {
3026                         const struct ldlm_request *dlmreq;
3027                         __u64 req_bits;
3028 #if 0
3029                         struct ldlm_lock       *lock = *lockp;
3030
3031                         LDLM_DEBUG(lock, "intent policy opc: %s\n",
3032                                    ldlm_it2str(it->opc));
3033 #endif
3034
3035                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3036                         if (rc == 0)
3037                                 rc = ELDLM_OK;
3038
3039                         /*
3040                          * Lock without inodebits makes no sense and will oops
3041                          * later in ldlm. Let's check it now to see if we have
3042                          * wrong lock from client or bits get corrupted
3043                          * somewhere in mdt_intent_opc().
3044                          */
3045                         dlmreq = info->mti_dlm_req;
3046                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3047                         LASSERT(req_bits != 0);
3048
3049                 } else
3050                         rc = err_serious(-EFAULT);
3051         } else {
3052                 /* No intent was provided */
3053                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3054                 rc = req_capsule_server_pack(pill);
3055                 if (rc)
3056                         rc = err_serious(rc);
3057         }
3058         RETURN(rc);
3059 }
3060
3061 /*
3062  * Seq wrappers
3063  */
3064 static void mdt_seq_adjust(const struct lu_env *env,
3065                           struct mdt_device *m, int lost)
3066 {
3067         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3068         struct lu_range out;
3069         ENTRY;
3070
3071         LASSERT(ls && ls->ls_server_seq);
3072         LASSERT(lost >= 0);
3073         /* get extra seq from seq_server, moving it's range up */
3074         while (lost-- > 0) {
3075                 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3076         }
3077         EXIT;
3078 }
3079
3080 static int mdt_seq_fini(const struct lu_env *env,
3081                         struct mdt_device *m)
3082 {
3083         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3084         ENTRY;
3085
3086         if (ls && ls->ls_server_seq) {
3087                 seq_server_fini(ls->ls_server_seq, env);
3088                 OBD_FREE_PTR(ls->ls_server_seq);
3089                 ls->ls_server_seq = NULL;
3090         }
3091
3092         if (ls && ls->ls_control_seq) {
3093                 seq_server_fini(ls->ls_control_seq, env);
3094                 OBD_FREE_PTR(ls->ls_control_seq);
3095                 ls->ls_control_seq = NULL;
3096         }
3097
3098         if (ls && ls->ls_client_seq) {
3099                 seq_client_fini(ls->ls_client_seq);
3100                 OBD_FREE_PTR(ls->ls_client_seq);
3101                 ls->ls_client_seq = NULL;
3102         }
3103
3104         RETURN(0);
3105 }
3106
3107 static int mdt_seq_init(const struct lu_env *env,
3108                         const char *uuid,
3109                         struct mdt_device *m)
3110 {
3111         struct lu_site *ls;
3112         char *prefix;
3113         int rc;
3114         ENTRY;
3115
3116         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3117
3118         /*
3119          * This is sequence-controller node. Init seq-controller server on local
3120          * MDT.
3121          */
3122         if (ls->ls_node_id == 0) {
3123                 LASSERT(ls->ls_control_seq == NULL);
3124
3125                 OBD_ALLOC_PTR(ls->ls_control_seq);
3126                 if (ls->ls_control_seq == NULL)
3127                         RETURN(-ENOMEM);
3128
3129                 rc = seq_server_init(ls->ls_control_seq,
3130                                      m->mdt_bottom, uuid,
3131                                      LUSTRE_SEQ_CONTROLLER,
3132                                      env);
3133
3134                 if (rc)
3135                         GOTO(out_seq_fini, rc);
3136
3137                 OBD_ALLOC_PTR(ls->ls_client_seq);
3138                 if (ls->ls_client_seq == NULL)
3139                         GOTO(out_seq_fini, rc = -ENOMEM);
3140
3141                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3142                 if (prefix == NULL) {
3143                         OBD_FREE_PTR(ls->ls_client_seq);
3144                         GOTO(out_seq_fini, rc = -ENOMEM);
3145                 }
3146
3147                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3148                          uuid);
3149
3150                 /*
3151                  * Init seq-controller client after seq-controller server is
3152                  * ready. Pass ls->ls_control_seq to it for direct talking.
3153                  */
3154                 rc = seq_client_init(ls->ls_client_seq, NULL,
3155                                      LUSTRE_SEQ_METADATA, prefix,
3156                                      ls->ls_control_seq);
3157                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3158
3159                 if (rc)
3160                         GOTO(out_seq_fini, rc);
3161         }
3162
3163         /* Init seq-server on local MDT */
3164         LASSERT(ls->ls_server_seq == NULL);
3165
3166         OBD_ALLOC_PTR(ls->ls_server_seq);
3167         if (ls->ls_server_seq == NULL)
3168                 GOTO(out_seq_fini, rc = -ENOMEM);
3169
3170         rc = seq_server_init(ls->ls_server_seq,
3171                              m->mdt_bottom, uuid,
3172                              LUSTRE_SEQ_SERVER,
3173                              env);
3174         if (rc)
3175                 GOTO(out_seq_fini, rc = -ENOMEM);
3176
3177         /* Assign seq-controller client to local seq-server. */
3178         if (ls->ls_node_id == 0) {
3179                 LASSERT(ls->ls_client_seq != NULL);
3180
3181                 rc = seq_server_set_cli(ls->ls_server_seq,
3182                                         ls->ls_client_seq,
3183                                         env);
3184         }
3185
3186         EXIT;
3187 out_seq_fini:
3188         if (rc)
3189                 mdt_seq_fini(env, m);
3190
3191         return rc;
3192 }
3193 /*
3194  * Init client sequence manager which is used by local MDS to talk to sequence
3195  * controller on remote node.
3196  */
3197 static int mdt_seq_init_cli(const struct lu_env *env,
3198                             struct mdt_device *m,
3199                             struct lustre_cfg *cfg)
3200 {
3201         struct lu_site    *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3202         struct obd_device *mdc;
3203         struct obd_uuid   *uuidp, *mdcuuidp;
3204         char              *uuid_str, *mdc_uuid_str;
3205         int                rc;
3206         int                index;
3207         struct mdt_thread_info *info;
3208         char *p, *index_string = lustre_cfg_string(cfg, 2);
3209         ENTRY;
3210
3211         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3212         uuidp = &info->mti_u.uuid[0];
3213         mdcuuidp = &info->mti_u.uuid[1];
3214
3215         LASSERT(index_string);
3216
3217         index = simple_strtol(index_string, &p, 10);
3218         if (*p) {
3219                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3220                 RETURN(-EINVAL);
3221         }
3222
3223         /* check if this is adding the first MDC and controller is not yet
3224          * initialized. */
3225         if (index != 0 || ls->ls_client_seq)
3226                 RETURN(0);
3227
3228         uuid_str = lustre_cfg_string(cfg, 1);
3229         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3230         obd_str2uuid(uuidp, uuid_str);
3231         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3232
3233         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3234         if (!mdc) {
3235                 CERROR("can't find controller MDC by uuid %s\n",
3236                        uuid_str);
3237                 rc = -ENOENT;
3238         } else if (!mdc->obd_set_up) {
3239                 CERROR("target %s not set up\n", mdc->obd_name);
3240                 rc = -EINVAL;
3241         } else {
3242                 LASSERT(ls->ls_control_exp);
3243                 OBD_ALLOC_PTR(ls->ls_client_seq);
3244                 if (ls->ls_client_seq != NULL) {
3245                         char *prefix;
3246
3247                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3248                         if (!prefix)
3249                                 RETURN(-ENOMEM);
3250
3251                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3252                                  mdc->obd_name);
3253
3254                         rc = seq_client_init(ls->ls_client_seq,
3255                                              ls->ls_control_exp,
3256                                              LUSTRE_SEQ_METADATA,
3257                                              prefix, NULL);
3258                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3259                 } else
3260                         rc = -ENOMEM;
3261
3262                 if (rc)
3263                         RETURN(rc);
3264
3265                 LASSERT(ls->ls_server_seq != NULL);
3266                 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3267                                         env);
3268         }
3269
3270         RETURN(rc);
3271 }
3272
3273 static void mdt_seq_fini_cli(struct mdt_device *m)
3274 {
3275         struct lu_site *ls;
3276
3277         ENTRY;
3278
3279         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3280
3281         if (ls && ls->ls_server_seq)
3282                 seq_server_set_cli(ls->ls_server_seq,
3283                                    NULL, NULL);
3284
3285         if (ls && ls->ls_control_exp) {
3286                 class_export_put(ls->ls_control_exp);
3287                 ls->ls_control_exp = NULL;
3288         }
3289         EXIT;
3290 }
3291
3292 /*
3293  * FLD wrappers
3294  */
3295 static int mdt_fld_fini(const struct lu_env *env,
3296                         struct mdt_device *m)
3297 {
3298         struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3299         ENTRY;
3300
3301         if (ls && ls->ls_server_fld) {
3302                 fld_server_fini(ls->ls_server_fld, env);
3303                 OBD_FREE_PTR(ls->ls_server_fld);
3304                 ls->ls_server_fld = NULL;
3305         }
3306
3307         RETURN(0);
3308 }
3309
3310 static int mdt_fld_init(const struct lu_env *env,
3311                         const char *uuid,
3312                         struct mdt_device *m)
3313 {
3314         struct lu_site *ls;
3315         int rc;
3316         ENTRY;
3317
3318         ls = m->mdt_md_dev.md_lu_dev.ld_site;
3319
3320         OBD_ALLOC_PTR(ls->ls_server_fld);
3321         if (ls->ls_server_fld == NULL)
3322                 RETURN(rc = -ENOMEM);
3323
3324         rc = fld_server_init(ls->ls_server_fld,
3325                              m->mdt_bottom, uuid, env);
3326         if (rc) {
3327                 OBD_FREE_PTR(ls->ls_server_fld);
3328                 ls->ls_server_fld = NULL;
3329                 RETURN(rc);
3330         }
3331
3332         RETURN(0);
3333 }
3334
3335 /* device init/fini methods */
3336 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3337 {
3338         ENTRY;
3339         if (m->mdt_regular_service != NULL) {
3340                 ptlrpc_unregister_service(m->mdt_regular_service);
3341                 m->mdt_regular_service = NULL;
3342         }
3343         if (m->mdt_readpage_service != NULL) {
3344                 ptlrpc_unregister_service(m->mdt_readpage_service);
3345                 m->mdt_readpage_service = NULL;
3346         }
3347         if (m->mdt_xmds_service != NULL) {
3348                 ptlrpc_unregister_service(m->mdt_xmds_service);
3349                 m->mdt_xmds_service = NULL;
3350         }
3351         if (m->mdt_setattr_service != NULL) {
3352                 ptlrpc_unregister_service(m->mdt_setattr_service);
3353                 m->mdt_setattr_service = NULL;
3354         }
3355         if (m->mdt_mdsc_service != NULL) {
3356                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3357                 m->mdt_mdsc_service = NULL;
3358         }
3359         if (m->mdt_mdss_service != NULL) {
3360                 ptlrpc_unregister_service(m->mdt_mdss_service);
3361                 m->mdt_mdss_service = NULL;
3362         }
3363         if (m->mdt_dtss_service != NULL) {
3364                 ptlrpc_unregister_service(m->mdt_dtss_service);
3365                 m->mdt_dtss_service = NULL;
3366         }
3367         if (m->mdt_fld_service != NULL) {
3368                 ptlrpc_unregister_service(m->mdt_fld_service);
3369                 m->mdt_fld_service = NULL;
3370         }
3371         ENTRY;
3372 }
3373
3374 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3375 {
3376         int rc;
3377         static struct ptlrpc_service_conf conf;
3378         cfs_proc_dir_entry_t *procfs_entry;
3379         ENTRY;
3380
3381         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3382
3383         conf = (typeof(conf)) {
3384                 .psc_nbufs           = MDS_NBUFS,
3385                 .psc_bufsize         = MDS_BUFSIZE,
3386                 .psc_max_req_size    = MDS_MAXREQSIZE,
3387                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3388                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3389                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3390                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3391                 /*
3392                  * We'd like to have a mechanism to set this on a per-device
3393                  * basis, but alas...
3394                  */
3395                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3396                                           MDT_MAX_THREADS),
3397                 .psc_max_threads     = MDT_MAX_THREADS,
3398                 .psc_ctx_tags        = LCT_MD_THREAD
3399         };
3400
3401         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3402         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3403                            "mdt_ldlm_client", m->mdt_ldlm_client);
3404
3405         m->mdt_regular_service =
3406                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3407                                      procfs_entry, target_print_req,
3408                                      LUSTRE_MDT_NAME);
3409         if (m->mdt_regular_service == NULL)
3410                 RETURN(-ENOMEM);
3411
3412         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3413         if (rc)
3414                 GOTO(err_mdt_svc, rc);
3415
3416         /*
3417          * readpage service configuration. Parameters have to be adjusted,
3418          * ideally.
3419          */
3420         conf = (typeof(conf)) {
3421                 .psc_nbufs           = MDS_NBUFS,
3422                 .psc_bufsize         = MDS_BUFSIZE,
3423                 .psc_max_req_size    = MDS_MAXREQSIZE,
3424                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3425                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3426                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3427                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3428                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3429                                           MDT_MAX_THREADS),
3430                 .psc_max_threads     = MDT_MAX_THREADS,
3431                 .psc_ctx_tags        = LCT_MD_THREAD
3432         };
3433         m->mdt_readpage_service =
3434                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3435                                      LUSTRE_MDT_NAME "_readpage",
3436                                      procfs_entry, target_print_req,"mdt_rdpg");
3437
3438         if (m->mdt_readpage_service == NULL) {
3439                 CERROR("failed to start readpage service\n");
3440                 GOTO(err_mdt_svc, rc = -ENOMEM);
3441         }
3442
3443         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3444
3445         /*
3446          * setattr service configuration.
3447          */
3448         conf = (typeof(conf)) {
3449                 .psc_nbufs           = MDS_NBUFS,
3450                 .psc_bufsize         = MDS_BUFSIZE,
3451                 .psc_max_req_size    = MDS_MAXREQSIZE,
3452                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3453                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3454                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3455                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3456                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3457                                          MDT_MAX_THREADS),
3458                 .psc_max_threads     = MDT_MAX_THREADS,
3459                 .psc_ctx_tags        = LCT_MD_THREAD
3460         };
3461
3462         m->mdt_setattr_service =
3463                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3464                                      LUSTRE_MDT_NAME "_setattr",
3465                                      procfs_entry, target_print_req,"mdt_attr");
3466
3467         if (!m->mdt_setattr_service) {
3468                 CERROR("failed to start setattr service\n");
3469                 GOTO(err_mdt_svc, rc = -ENOMEM);
3470         }
3471
3472         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3473         if (rc)
3474                 GOTO(err_mdt_svc, rc);
3475
3476         /*
3477          * sequence controller service configuration
3478          */
3479         conf = (typeof(conf)) {
3480                 .psc_nbufs           = MDS_NBUFS,
3481                 .psc_bufsize         = MDS_BUFSIZE,
3482                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3483                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3484                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3485                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3486                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3487                 .psc_min_threads     = SEQ_NUM_THREADS,
3488                 .psc_max_threads     = SEQ_NUM_THREADS,
3489                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3490         };
3491
3492         m->mdt_mdsc_service =
3493                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3494                                      LUSTRE_MDT_NAME"_mdsc",
3495                                      procfs_entry, target_print_req,"mdt_mdsc");
3496         if (!m->mdt_mdsc_service) {
3497                 CERROR("failed to start seq controller service\n");
3498                 GOTO(err_mdt_svc, rc = -ENOMEM);
3499         }
3500
3501         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3502         if (rc)
3503                 GOTO(err_mdt_svc, rc);
3504
3505         /*
3506          * metadata sequence server service configuration
3507          */
3508         conf = (typeof(conf)) {
3509                 .psc_nbufs           = MDS_NBUFS,
3510                 .psc_bufsize         = MDS_BUFSIZE,
3511                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3512                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3513                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3514                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3515                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3516                 .psc_min_threads     = SEQ_NUM_THREADS,
3517                 .psc_max_threads     = SEQ_NUM_THREADS,
3518                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3519         };
3520
3521         m->mdt_mdss_service =
3522                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3523                                      LUSTRE_MDT_NAME"_mdss",
3524                                      procfs_entry, target_print_req,"mdt_mdss");
3525         if (!m->mdt_mdss_service) {
3526                 CERROR("failed to start metadata seq server service\n");
3527                 GOTO(err_mdt_svc, rc = -ENOMEM);
3528         }
3529
3530         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3531         if (rc)
3532                 GOTO(err_mdt_svc, rc);
3533
3534
3535         /*
3536          * Data sequence server service configuration. We want to have really
3537          * cluster-wide sequences space. This is why we start only one sequence
3538          * controller which manages space.
3539          */
3540         conf = (typeof(conf)) {
3541                 .psc_nbufs           = MDS_NBUFS,
3542                 .psc_bufsize         = MDS_BUFSIZE,
3543                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3544                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3545                 .psc_req_portal      = SEQ_DATA_PORTAL,
3546                 .psc_rep_portal      = OSC_REPLY_PORTAL,
3547                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3548                 .psc_min_threads     = SEQ_NUM_THREADS,
3549                 .psc_max_threads     = SEQ_NUM_THREADS,
3550                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3551         };
3552
3553         m->mdt_dtss_service =
3554                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3555                                      LUSTRE_MDT_NAME"_dtss",
3556                                      procfs_entry, target_print_req,"mdt_dtss");
3557         if (!m->mdt_dtss_service) {
3558                 CERROR("failed to start data seq server service\n");
3559                 GOTO(err_mdt_svc, rc = -ENOMEM);
3560         }
3561
3562         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3563         if (rc)
3564                 GOTO(err_mdt_svc, rc);
3565
3566         /* FLD service start */
3567         conf = (typeof(conf)) {
3568                 .psc_nbufs           = MDS_NBUFS,
3569                 .psc_bufsize         = MDS_BUFSIZE,
3570                 .psc_max_req_size    = FLD_MAXREQSIZE,
3571                 .psc_max_reply_size  = FLD_MAXREPSIZE,
3572                 .psc_req_portal      = FLD_REQUEST_PORTAL,
3573                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3574                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3575                 .psc_min_threads     = FLD_NUM_THREADS,
3576                 .psc_max_threads     = FLD_NUM_THREADS,
3577                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
3578         };
3579
3580         m->mdt_fld_service =
3581                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3582                                      LUSTRE_MDT_NAME"_fld",
3583                                      procfs_entry, target_print_req, "mdt_fld");
3584         if (!m->mdt_fld_service) {
3585                 CERROR("failed to start fld service\n");
3586                 GOTO(err_mdt_svc, rc = -ENOMEM);
3587         }
3588
3589         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3590         if (rc)
3591                 GOTO(err_mdt_svc, rc);
3592
3593         /*
3594          * mds-mds service configuration. Separate portal is used to allow
3595          * mds-mds requests be not blocked during recovery.
3596          */
3597         conf = (typeof(conf)) {
3598                 .psc_nbufs           = MDS_NBUFS,
3599                 .psc_bufsize         = MDS_BUFSIZE,
3600                 .psc_max_req_size    = MDS_MAXREQSIZE,
3601                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3602                 .psc_req_portal      = MDS_MDS_PORTAL,
3603                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3604                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3605                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3606                                           MDT_MAX_THREADS),
3607                 .psc_max_threads     = MDT_MAX_THREADS,
3608                 .psc_ctx_tags        = LCT_MD_THREAD
3609         };
3610         m->mdt_xmds_service =
3611                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3612                                      LUSTRE_MDT_NAME "_mds",
3613                                      procfs_entry, target_print_req,"mdt_xmds");
3614
3615         if (m->mdt_xmds_service == NULL) {
3616                 CERROR("failed to start readpage service\n");
3617                 GOTO(err_mdt_svc, rc = -ENOMEM);
3618         }
3619
3620         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3621         if (rc)
3622                 GOTO(err_mdt_svc, rc);
3623
3624         EXIT;
3625 err_mdt_svc:
3626         if (rc)
3627                 mdt_stop_ptlrpc_service(m);
3628
3629         return rc;
3630 }
3631
3632 static void mdt_stack_fini(const struct lu_env *env,
3633                            struct mdt_device *m, struct lu_device *top)
3634 {
3635         struct obd_device       *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3636         struct lustre_cfg_bufs  *bufs;
3637         struct lustre_cfg       *lcfg;
3638         struct mdt_thread_info  *info;
3639         char flags[3]="";
3640         ENTRY;
3641
3642         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3643         LASSERT(info != NULL);
3644
3645         bufs = &info->mti_u.bufs;
3646         /* process cleanup, pass mdt obd name to get obd umount flags */
3647         lustre_cfg_bufs_reset(bufs, obd->obd_name);
3648         if (obd->obd_force)
3649                 strcat(flags, "F");
3650         if (obd->obd_fail)
3651                 strcat(flags, "A");
3652         lustre_cfg_bufs_set_string(bufs, 1, flags);
3653         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3654         if (!lcfg) {
3655                 CERROR("Cannot alloc lcfg!\n");
3656                 return;
3657         }
3658
3659         LASSERT(top);
3660         top->ld_ops->ldo_process_config(env, top, lcfg);
3661         lustre_cfg_free(lcfg);
3662
3663         lu_stack_fini(env, top);
3664         m->mdt_child = NULL;
3665         m->mdt_bottom = NULL;
3666 }
3667
3668 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3669                                          const char *typename,
3670                                          struct lu_device *child,
3671                                          struct lustre_cfg *cfg)
3672 {
3673         const char            *dev = lustre_cfg_string(cfg, 0);
3674         struct obd_type       *type;
3675         struct lu_device_type *ldt;
3676         struct lu_device      *d;
3677         int rc;
3678         ENTRY;
3679
3680         /* find the type */
3681         type = class_get_type(typename);
3682         if (!type) {
3683                 CERROR("Unknown type: '%s'\n", typename);
3684                 GOTO(out, rc = -ENODEV);
3685         }
3686
3687         rc = lu_context_refill(&env->le_ctx);
3688         if (rc != 0) {
3689                 CERROR("Failure to refill context: '%d'\n", rc);
3690                 GOTO(out_type, rc);
3691         }
3692
3693         if (env->le_ses != NULL) {
3694                 rc = lu_context_refill(env->le_ses);
3695                 if (rc != 0) {
3696                         CERROR("Failure to refill session: '%d'\n", rc);
3697                         GOTO(out_type, rc);
3698                 }
3699         }
3700
3701         ldt = type->typ_lu;
3702         if (ldt == NULL) {
3703                 CERROR("type: '%s'\n", typename);
3704                 GOTO(out_type, rc = -EINVAL);
3705         }
3706
3707         ldt->ldt_obd_type = type;
3708         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3709         if (IS_ERR(d)) {
3710                 CERROR("Cannot allocate device: '%s'\n", typename);
3711                 GOTO(out_type, rc = -ENODEV);
3712         }
3713
3714         LASSERT(child->ld_site);
3715         d->ld_site = child->ld_site;
3716
3717         type->typ_refcnt++;
3718         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3719         if (rc) {
3720                 CERROR("can't init device '%s', rc %d\n", typename, rc);
3721                 GOTO(out_alloc, rc);
3722         }
3723         lu_device_get(d);
3724
3725         RETURN(d);
3726
3727 out_alloc:
3728         ldt->ldt_ops->ldto_device_free(env, d);
3729         type->typ_refcnt--;
3730 out_type:
3731         class_put_type(type);
3732 out:
3733         return ERR_PTR(rc);
3734 }
3735
3736 static int mdt_stack_init(const struct lu_env *env,
3737                           struct mdt_device *m, struct lustre_cfg *cfg)
3738 {
3739         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
3740         struct lu_device  *tmp;
3741         struct md_device  *md;
3742         int rc;
3743         ENTRY;
3744
3745         /* init the stack */
3746         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3747         if (IS_ERR(tmp)) {
3748                 RETURN(PTR_ERR(tmp));
3749         }
3750         m->mdt_bottom = lu2dt_dev(tmp);
3751         d = tmp;
3752         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3753         if (IS_ERR(tmp)) {
3754                 GOTO(out, rc = PTR_ERR(tmp));
3755         }
3756         d = tmp;
3757         md = lu2md_dev(d);
3758
3759         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3760         if (IS_ERR(tmp)) {
3761                 GOTO(out, rc = PTR_ERR(tmp));
3762         }
3763         d = tmp;
3764         /*set mdd upcall device*/
3765         md_upcall_dev_set(md, lu2md_dev(d));
3766
3767         md = lu2md_dev(d);
3768         /*set cmm upcall device*/
3769         md_upcall_dev_set(md, &m->mdt_md_dev);
3770
3771         m->mdt_child = lu2md_dev(d);
3772
3773         /* process setup config */
3774         tmp = &m->mdt_md_dev.md_lu_dev;
3775         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3776         GOTO(out, rc);
3777 out:
3778         /* fini from last known good lu_device */
3779         if (rc)
3780                 mdt_stack_fini(env, m, d);
3781
3782         return rc;
3783 }
3784
3785 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3786 {
3787         struct md_device *next = m->mdt_child;
3788         struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
3789         struct lu_site   *ls   = d->ld_site;
3790         struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3791         ENTRY;
3792
3793         ping_evictor_stop();
3794
3795         target_recovery_fini(obd);
3796         mdt_stop_ptlrpc_service(m);
3797
3798         mdt_fs_cleanup(env, m);
3799
3800         upcall_cache_cleanup(m->mdt_identity_cache);
3801         m->mdt_identity_cache = NULL;
3802
3803         if (m->mdt_namespace != NULL) {
3804                 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3805                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3806         }
3807
3808         mdt_seq_fini(env, m);
3809         mdt_seq_fini_cli(m);
3810         mdt_fld_fini(env, m);
3811         mdt_procfs_fini(m);
3812         ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3813         lprocfs_obd_cleanup(d->ld_obd);
3814
3815         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3816
3817         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3818         cfs_timer_disarm(&m->mdt_ck_timer);
3819         mdt_ck_thread_stop(m);
3820
3821         /* finish the stack */
3822         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3823
3824         if (ls) {
3825                 lu_site_fini(ls);
3826                 OBD_FREE_PTR(ls);
3827                 d->ld_site = NULL;
3828         }
3829         LASSERT(atomic_read(&d->ld_ref) == 0);
3830         md_device_fini(&m->mdt_md_dev);
3831
3832         EXIT;
3833 }
3834
3835 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3836 {
3837         char *p = options;
3838
3839 #ifdef CONFIG_FS_POSIX_ACL
3840         /* ACLs should be enabled by default (b=13829) */
3841         m->mdt_opts.mo_acl = 1;
3842         LCONSOLE_INFO("Enabling ACL\n");
3843 #else
3844         m->mdt_opts.mo_acl = 0;
3845         LCONSOLE_INFO("Disabling ACL\n");
3846 #endif
3847
3848         if (!options)
3849                 return;
3850
3851         while (*options) {
3852                 int len;
3853
3854                 while (*p && *p != ',')
3855                         p++;
3856
3857                 len = p - options;
3858                 if ((len == sizeof("user_xattr") - 1) &&
3859                     (memcmp(options, "user_xattr", len) == 0)) {
3860                         m->mdt_opts.mo_user_xattr = 1;
3861                         LCONSOLE_INFO("Enabling user_xattr\n");
3862                 } else if ((len == sizeof("nouser_xattr") - 1) &&
3863                            (memcmp(options, "nouser_xattr", len) == 0)) {
3864                         m->mdt_opts.mo_user_xattr = 0;
3865                         LCONSOLE_INFO("Disabling user_xattr\n");
3866                 } else if ((len == sizeof("noacl") - 1) &&
3867                            (memcmp(options, "noacl", len) == 0)) {
3868                         m->mdt_opts.mo_acl = 0;
3869                         LCONSOLE_INFO("Disabling ACL\n");
3870                 }
3871
3872                 options = ++p;
3873         }
3874 }
3875
3876 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3877
3878 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3879                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
3880 {
3881         struct lprocfs_static_vars lvars;
3882         struct mdt_thread_info    *info;
3883         struct obd_device         *obd;
3884         const char                *dev = lustre_cfg_string(cfg, 0);
3885         const char                *num = lustre_cfg_string(cfg, 2);
3886         struct lustre_mount_info  *lmi;
3887         struct lustre_sb_info     *lsi;
3888         struct lu_site            *s;
3889         const char                *identity_upcall = "NONE";
3890         int                        rc;
3891         ENTRY;
3892
3893         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3894         LASSERT(info != NULL);
3895
3896         obd = class_name2obd(dev);
3897         LASSERT(obd != NULL);
3898
3899         spin_lock_init(&m->mdt_transno_lock);
3900
3901         m->mdt_max_mdsize = MAX_MD_SIZE;
3902         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3903
3904         m->mdt_opts.mo_user_xattr = 0;
3905         m->mdt_opts.mo_acl = 0;
3906         lmi = server_get_mount_2(dev);
3907         if (lmi == NULL) {
3908                 CERROR("Cannot get mount info for %s!\n", dev);
3909                 RETURN(-EFAULT);
3910         } else {
3911                 lsi = s2lsi(lmi->lmi_sb);
3912                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3913                 server_put_mount_2(dev, lmi->lmi_mnt);
3914         }
3915
3916         m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3917         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3918
3919         spin_lock_init(&m->mdt_ioepoch_lock);
3920         m->mdt_opts.mo_compat_resname = 0;
3921         m->mdt_capa_timeout = CAPA_TIMEOUT;
3922         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3923         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3924
3925         spin_lock_init(&m->mdt_client_bitmap_lock);
3926
3927         OBD_ALLOC_PTR(s);
3928         if (s == NULL)
3929                 RETURN(-ENOMEM);
3930
3931         md_device_init(&m->mdt_md_dev, ldt);
3932         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3933         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3934         /* set this lu_device to obd, because error handling need it */
3935         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3936
3937         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3938         if (rc) {
3939                 CERROR("Can't init lu_site, rc %d\n", rc);
3940                 GOTO(err_free_site, rc);
3941         }
3942
3943         lprocfs_mdt_init_vars(&lvars);
3944         rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3945         if (rc) {
3946                 CERROR("Can't init lprocfs, rc %d\n", rc);
3947                 GOTO(err_fini_site, rc);
3948         }
3949         ptlrpc_lprocfs_register_obd(obd);
3950
3951         rc = mdt_procfs_init(m, dev);
3952         if (rc) {
3953                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3954                 GOTO(err_fini_proc, rc);
3955         }
3956
3957         /* set server index */
3958         LASSERT(num);
3959         s->ls_node_id = simple_strtol(num, NULL, 10);
3960
3961         /* failover is the default
3962          * FIXME: we do not failout mds0/mgs, which may cause some problems.
3963          * assumed whose ls_node_id == 0 XXX
3964          * */
3965         obd->obd_replayable = 1;
3966         /* No connection accepted until configurations will finish */
3967         obd->obd_no_conn = 1;
3968
3969         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3970                 char *str = lustre_cfg_string(cfg, 4);
3971                 if (strchr(str, 'n')) {
3972                         CWARN("%s: recovery disabled\n", obd->obd_name);
3973                         obd->obd_replayable = 0;
3974                 }
3975         }
3976
3977         /* init the stack */
3978         rc = mdt_stack_init(env, m, cfg);
3979         if (rc) {
3980                 CERROR("Can't init device stack, rc %d\n", rc);
3981                 GOTO(err_fini_proc, rc);
3982         }
3983
3984         rc = mdt_fld_init(env, obd->obd_name, m);
3985         if (rc)
3986                 GOTO(err_fini_stack, rc);
3987
3988         rc = mdt_seq_init(env, obd->obd_name, m);
3989         if (rc)
3990                 GOTO(err_fini_fld, rc);
3991
3992         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3993                  LUSTRE_MDT_NAME"-%p", m);
3994         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
3995                                               LDLM_NAMESPACE_SERVER,
3996                                               LDLM_NAMESPACE_GREEDY);
3997         if (m->mdt_namespace == NULL)
3998                 GOTO(err_fini_seq, rc = -ENOMEM);
3999
4000         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4001         /* set obd_namespace for compatibility with old code */
4002         obd->obd_namespace = m->mdt_namespace;
4003
4004         /* XXX: to support suppgid for ACL, we enable identity_upcall
4005          * by default, otherwise, maybe got unexpected -EACCESS. */
4006         if (m->mdt_opts.mo_acl)
4007                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4008
4009         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4010                                                   &mdt_identity_upcall_cache_ops);
4011         if (IS_ERR(m->mdt_identity_cache)) {
4012                 rc = PTR_ERR(m->mdt_identity_cache);
4013                 m->mdt_identity_cache = NULL;
4014                 GOTO(err_free_ns, rc);
4015         }
4016
4017         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4018
4019         rc = mdt_ck_thread_start(m);
4020         if (rc)
4021                 GOTO(err_free_ns, rc);
4022
4023         rc = mdt_fs_setup(env, m, obd);
4024         if (rc)
4025                 GOTO(err_capa, rc);
4026
4027         target_recovery_init(obd, mdt_recovery_handle);
4028
4029         rc = mdt_start_ptlrpc_service(m);
4030         if (rc)
4031                 GOTO(err_fs_cleanup, rc);
4032
4033         ping_evictor_start();
4034
4035         rc = lu_site_init_finish(s);
4036         if (rc)
4037                 GOTO(err_stop_service, rc);
4038
4039         if (obd->obd_recovering == 0)
4040                 mdt_postrecov(env, m);
4041
4042         mdt_init_capa_ctxt(env, m);
4043
4044         /* Reduce the initial timeout on an MDS because it doesn't need such
4045          * a long timeout as an OST does. Adaptive timeouts will adjust this
4046          * value appropriately. */
4047         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4048                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4049
4050         RETURN(0);
4051
4052 err_stop_service:
4053         ping_evictor_stop();
4054         mdt_stop_ptlrpc_service(m);
4055 err_fs_cleanup:
4056         target_recovery_fini(obd);
4057         mdt_fs_cleanup(env, m);
4058 err_capa:
4059         cfs_timer_disarm(&m->mdt_ck_timer);
4060         mdt_ck_thread_stop(m);
4061 err_free_ns:
4062         upcall_cache_cleanup(m->mdt_identity_cache);
4063         m->mdt_identity_cache = NULL;
4064         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4065         obd->obd_namespace = m->mdt_namespace = NULL;
4066 err_fini_seq:
4067         mdt_seq_fini(env, m);
4068 err_fini_fld:
4069         mdt_fld_fini(env, m);
4070 err_fini_stack:
4071         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4072 err_fini_proc:
4073         mdt_procfs_fini(m);
4074         ptlrpc_lprocfs_unregister_obd(obd);
4075         lprocfs_obd_cleanup(obd);
4076 err_fini_site:
4077         lu_site_fini(s);
4078 err_free_site:
4079         OBD_FREE_PTR(s);
4080
4081         md_device_fini(&m->mdt_md_dev);
4082         return (rc);
4083 }
4084
4085 /* used by MGS to process specific configurations */
4086 static int mdt_process_config(const struct lu_env *env,
4087                               struct lu_device *d, struct lustre_cfg *cfg)
4088 {
4089         struct mdt_device *m = mdt_dev(d);
4090         struct md_device *md_next = m->mdt_child;
4091         struct lu_device *next = md2lu_dev(md_next);
4092         int rc = 0;
4093         ENTRY;
4094
4095         switch (cfg->lcfg_command) {
4096         case LCFG_SPTLRPC_CONF: {
4097                 struct sptlrpc_conf_log *log;
4098                 struct sptlrpc_rule_set  tmp_rset;
4099
4100                 log = sptlrpc_conf_log_extract(cfg);
4101                 if (IS_ERR(log)) {
4102                         rc = PTR_ERR(log);
4103                         break;
4104                 }
4105
4106                 sptlrpc_rule_set_init(&tmp_rset);
4107
4108                 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4109                 if (rc) {
4110                         CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4111                         break;
4112                 }
4113
4114                 write_lock(&m->mdt_sptlrpc_lock);
4115                 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4116                 m->mdt_sptlrpc_rset = tmp_rset;
4117                 write_unlock(&m->mdt_sptlrpc_lock);
4118
4119                 sptlrpc_target_update_exp_flavor(
4120                                 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4121
4122                 break;
4123         }
4124         case LCFG_PARAM: {
4125                 struct lprocfs_static_vars lvars;
4126                 struct obd_device *obd = d->ld_obd;
4127
4128                 lprocfs_mdt_init_vars(&lvars);
4129                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4130                 if (rc)
4131                         /* others are passed further */
4132                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4133                 break;
4134         }
4135         case LCFG_ADD_MDC:
4136                 /*
4137                  * Add mdc hook to get first MDT uuid and connect it to
4138                  * ls->controller to use for seq manager.
4139                  */
4140                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4141                 if (rc)
4142                         CERROR("Can't add mdc, rc %d\n", rc);
4143                 else
4144                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4145                 break;
4146         default:
4147                 /* others are passed further */
4148                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4149                 break;
4150         }
4151         RETURN(rc);
4152 }
4153
4154 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4155                                           const struct lu_object_header *hdr,
4156                                           struct lu_device *d)
4157 {
4158         struct mdt_object *mo;
4159
4160         ENTRY;
4161
4162         OBD_ALLOC_PTR(mo);
4163         if (mo != NULL) {
4164                 struct lu_object *o;
4165                 struct lu_object_header *h;
4166
4167                 o = &mo->mot_obj.mo_lu;
4168                 h = &mo->mot_header;
4169                 lu_object_header_init(h);
4170                 lu_object_init(o, h, d);
4171                 lu_object_add_top(h, o);
4172                 o->lo_ops = &mdt_obj_ops;
4173                 RETURN(o);
4174         } else
4175                 RETURN(NULL);
4176 }
4177
4178 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4179 {
4180         struct mdt_device *d = mdt_dev(o->lo_dev);
4181         struct lu_device  *under;
4182         struct lu_object  *below;
4183         int                rc = 0;
4184         ENTRY;
4185
4186         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4187                PFID(lu_object_fid(o)));
4188
4189         under = &d->mdt_child->md_lu_dev;
4190         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4191         if (below != NULL) {
4192                 lu_object_add(o, below);
4193         } else
4194                 rc = -ENOMEM;
4195
4196         RETURN(rc);
4197 }
4198
4199 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4200 {
4201         struct mdt_object *mo = mdt_obj(o);
4202         struct lu_object_header *h;
4203         ENTRY;
4204
4205         h = o->lo_header;
4206         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4207                PFID(lu_object_fid(o)));
4208
4209         lu_object_fini(o);
4210         lu_object_header_fini(h);
4211         OBD_FREE_PTR(mo);
4212         EXIT;
4213 }
4214
4215 static int mdt_object_print(const struct lu_env *env, void *cookie,
4216                             lu_printer_t p, const struct lu_object *o)
4217 {
4218         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4219 }
4220
4221 static struct lu_device_operations mdt_lu_ops = {
4222         .ldo_object_alloc   = mdt_object_alloc,
4223         .ldo_process_config = mdt_process_config
4224 };
4225
4226 static struct lu_object_operations mdt_obj_ops = {
4227         .loo_object_init    = mdt_object_init,
4228         .loo_object_free    = mdt_object_free,
4229         .loo_object_print   = mdt_object_print
4230 };
4231
4232 /* mds_connect_internal */
4233 static int mdt_connect_internal(struct obd_export *exp,
4234                                 struct mdt_device *mdt,
4235                                 struct obd_connect_data *data)
4236 {
4237         __u64 flags;
4238
4239         if (data != NULL) {
4240                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4241                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4242
4243                 /* If no known bits (which should not happen, probably,
4244                    as everybody should support LOOKUP and UPDATE bits at least)
4245                    revert to compat mode with plain locks. */
4246                 if (!data->ocd_ibits_known &&
4247                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4248                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4249
4250                 if (!mdt->mdt_opts.mo_acl)
4251                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4252
4253                 if (!mdt->mdt_opts.mo_user_xattr)
4254                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4255
4256                 if (!mdt->mdt_opts.mo_mds_capa)
4257                         data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4258
4259                 if (!mdt->mdt_opts.mo_oss_capa)
4260                         data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4261
4262                 spin_lock(&exp->exp_lock);
4263                 exp->exp_connect_flags = data->ocd_connect_flags;
4264                 spin_unlock(&exp->exp_lock);
4265                 data->ocd_version = LUSTRE_VERSION_CODE;
4266                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4267         }
4268
4269 #if 0
4270         if (mdt->mdt_opts.mo_acl &&
4271             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4272                 CWARN("%s: MDS requires ACL support but client does not\n",
4273                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4274                 return -EBADE;
4275         }
4276 #endif
4277
4278         flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4279         if ((exp->exp_connect_flags & flags) == flags) {
4280                 CWARN("%s: both local and remote client flags are set\n",
4281                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4282                 return -EBADE;
4283         }
4284
4285         if (mdt->mdt_opts.mo_mds_capa &&
4286             ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4287                 CWARN("%s: MDS requires capability support, but client not\n",
4288                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4289                 return -EBADE;
4290         }
4291
4292         if (mdt->mdt_opts.mo_oss_capa &&
4293             ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4294                 CWARN("%s: MDS requires OSS capability support, "
4295                       "but client not\n",
4296                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4297                 return -EBADE;
4298         }
4299
4300         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4301                 CWARN("%s: MDS requires FID support, but client not\n",
4302                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4303                 return -EBADE;
4304         }
4305
4306         return 0;
4307 }
4308
4309 /* mds_connect copy */
4310 static int mdt_obd_connect(const struct lu_env *env,
4311                            struct lustre_handle *conn, struct obd_device *obd,
4312                            struct obd_uuid *cluuid,
4313                            struct obd_connect_data *data,
4314                            void *localdata)
4315 {
4316         struct mdt_thread_info *info;
4317         struct lsd_client_data *lcd;
4318         struct obd_export      *exp;
4319         struct mdt_device      *mdt;
4320         struct ptlrpc_request  *req;
4321         int                     rc;
4322         ENTRY;
4323
4324         LASSERT(env != NULL);
4325         if (!conn || !obd || !cluuid)
4326                 RETURN(-EINVAL);
4327
4328         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4329         req = info->mti_pill->rc_req;
4330         mdt = mdt_dev(obd->obd_lu_dev);
4331
4332         rc = class_connect(conn, obd, cluuid);
4333         if (rc)
4334                 RETURN(rc);
4335
4336         exp = class_conn2export(conn);
4337         LASSERT(exp != NULL);
4338
4339         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4340
4341         spin_lock(&exp->exp_lock);
4342         exp->exp_sp_peer = req->rq_sp_from;
4343
4344         read_lock(&mdt->mdt_sptlrpc_lock);
4345         sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4346                                 req->rq_peer.nid, &exp->exp_flvr);
4347         read_unlock(&mdt->mdt_sptlrpc_lock);
4348
4349         if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4350                 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4351                        req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4352                        libcfs_nid2str(req->rq_peer.nid));
4353                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4354                 spin_unlock(&exp->exp_lock);
4355                 RETURN(-EACCES);
4356         }
4357         spin_unlock(&exp->exp_lock);
4358
4359         rc = mdt_connect_internal(exp, mdt, data);
4360         if (rc == 0) {
4361                 OBD_ALLOC_PTR(lcd);
4362                 if (lcd != NULL) {
4363                         struct mdt_thread_info *mti;
4364                         mti = lu_context_key_get(&env->le_ctx,
4365                                                  &mdt_thread_key);
4366                         LASSERT(mti != NULL);
4367                         mti->mti_exp = exp;
4368                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4369                         exp->exp_mdt_data.med_lcd = lcd;
4370                         rc = mdt_client_new(env, mdt);
4371                         if (rc != 0) {
4372                                 OBD_FREE_PTR(lcd);
4373                                 exp->exp_mdt_data.med_lcd = NULL;
4374                         }
4375                 } else
4376                         rc = -ENOMEM;
4377         }
4378
4379         if (rc != 0)
4380                 class_disconnect(exp);
4381         else
4382                 class_export_put(exp);
4383
4384         RETURN(rc);
4385 }
4386
4387 static int mdt_obd_reconnect(const struct lu_env *env,
4388                              struct obd_export *exp, struct obd_device *obd,
4389                              struct obd_uuid *cluuid,
4390                              struct obd_connect_data *data)
4391 {
4392         struct mdt_thread_info *info;
4393         struct mdt_device      *mdt;
4394         struct ptlrpc_request  *req;
4395         int                     rc;
4396         ENTRY;
4397
4398         if (exp == NULL || obd == NULL || cluuid == NULL)
4399                 RETURN(-EINVAL);
4400
4401         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4402         req = info->mti_pill->rc_req;
4403         mdt = mdt_dev(obd->obd_lu_dev);
4404
4405         CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4406
4407         spin_lock(&exp->exp_lock);
4408         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4409                 exp->exp_sp_peer = req->rq_sp_from;
4410
4411                 read_lock(&mdt->mdt_sptlrpc_lock);
4412                 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4413                                         exp->exp_sp_peer,
4414                                         req->rq_peer.nid, &exp->exp_flvr);
4415                 read_unlock(&mdt->mdt_sptlrpc_lock);
4416
4417                 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4418                         CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4419                                req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4420                                libcfs_nid2str(req->rq_peer.nid));
4421                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4422                         spin_unlock(&exp->exp_lock);
4423                         RETURN(-EACCES);
4424                 }
4425         }
4426         spin_unlock(&exp->exp_lock);
4427
4428         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4429
4430         RETURN(rc);
4431 }
4432
4433 static int mdt_obd_disconnect(struct obd_export *exp)
4434 {
4435         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4436         int rc;
4437         ENTRY;
4438
4439         LASSERT(exp);
4440         class_export_get(exp);
4441
4442         /* Disconnect early so that clients can't keep using export */
4443         rc = class_disconnect(exp);
4444         if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4445                 ldlm_cancel_locks_for_export(exp);
4446
4447         /* complete all outstanding replies */
4448         spin_lock(&exp->exp_lock);
4449         while (!list_empty(&exp->exp_outstanding_replies)) {
4450                 struct ptlrpc_reply_state *rs =
4451                         list_entry(exp->exp_outstanding_replies.next,
4452                                    struct ptlrpc_reply_state, rs_exp_list);
4453                 struct ptlrpc_service *svc = rs->rs_service;
4454
4455                 spin_lock(&svc->srv_lock);
4456                 list_del_init(&rs->rs_exp_list);
4457                 ptlrpc_schedule_difficult_reply(rs);
4458                 spin_unlock(&svc->srv_lock);
4459         }
4460         spin_unlock(&exp->exp_lock);
4461
4462         class_export_put(exp);
4463         RETURN(rc);
4464 }
4465
4466 /* FIXME: Can we avoid using these two interfaces? */
4467 static int mdt_init_export(struct obd_export *exp)
4468 {
4469         struct mdt_export_data *med = &exp->exp_mdt_data;
4470         ENTRY;
4471
4472         CFS_INIT_LIST_HEAD(&med->med_open_head);
4473         spin_lock_init(&med->med_open_lock);
4474         sema_init(&med->med_idmap_sem, 1);
4475         med->med_idmap = NULL;
4476         spin_lock(&exp->exp_lock);
4477         exp->exp_connecting = 1;
4478         spin_unlock(&exp->exp_lock);
4479         RETURN(0);
4480 }
4481
4482 static int mdt_destroy_export(struct obd_export *export)
4483 {
4484         struct mdt_export_data *med;
4485         struct obd_device      *obd = export->exp_obd;
4486         struct mdt_device      *mdt;
4487         struct mdt_thread_info *info;
4488         struct lu_env           env;
4489         struct md_attr         *ma;
4490         int lmm_size;
4491         int cookie_size;
4492         int rc = 0;
4493         ENTRY;
4494
4495         med = &export->exp_mdt_data;
4496         if (med->med_rmtclient)
4497                 mdt_cleanup_idmap(med);
4498
4499         target_destroy_export(export);
4500
4501         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4502                 RETURN(0);
4503
4504         mdt = mdt_dev(obd->obd_lu_dev);
4505         LASSERT(mdt != NULL);
4506
4507         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4508         if (rc)
4509                 RETURN(rc);
4510
4511         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4512         LASSERT(info != NULL);
4513         memset(info, 0, sizeof *info);
4514         info->mti_env = &env;
4515         info->mti_mdt = mdt;
4516         info->mti_exp = export;
4517
4518         ma = &info->mti_attr;
4519         lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4520         cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4521         OBD_ALLOC(ma->ma_lmm, lmm_size);
4522         OBD_ALLOC(ma->ma_cookie, cookie_size);
4523
4524         if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4525                 GOTO(out, rc = -ENOMEM);
4526         ma->ma_need = MA_LOV | MA_COOKIE;
4527         ma->ma_valid = 0;
4528         /* Close any open files (which may also cause orphan unlinking). */
4529         spin_lock(&med->med_open_lock);
4530         while (!list_empty(&med->med_open_head)) {
4531                 struct list_head *tmp = med->med_open_head.next;
4532                 struct mdt_file_data *mfd =
4533                         list_entry(tmp, struct mdt_file_data, mfd_list);
4534
4535                 /* Remove mfd handle so it can't be found again.
4536                  * We are consuming the mfd_list reference here. */
4537                 class_handle_unhash(&mfd->mfd_handle);
4538                 list_del_init(&mfd->mfd_list);
4539                 spin_unlock(&med->med_open_lock);
4540                 mdt_mfd_close(info, mfd);
4541                 /* TODO: if we close the unlinked file,
4542                  * we need to remove it's objects from OST */
4543                 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4544                 spin_lock(&med->med_open_lock);
4545                 ma->ma_lmm_size = lmm_size;
4546                 ma->ma_cookie_size = cookie_size;
4547                 ma->ma_need = MA_LOV | MA_COOKIE;
4548                 ma->ma_valid = 0;
4549         }
4550         spin_unlock(&med->med_open_lock);
4551         info->mti_mdt = NULL;
4552         mdt_client_del(&env, mdt);
4553
4554         EXIT;
4555 out:
4556         if (lmm_size) {
4557                 OBD_FREE(ma->ma_lmm, lmm_size);
4558                 ma->ma_lmm = NULL;
4559         }
4560         if (cookie_size) {
4561                 OBD_FREE(ma->ma_cookie, cookie_size);
4562                 ma->ma_cookie = NULL;
4563         }
4564         lu_env_fini(&env);
4565
4566         return rc;
4567 }
4568
4569 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4570 {
4571         if (flag & CONFIG_LOG)
4572                 m->mdt_fl_cfglog = 1;
4573         if (flag & CONFIG_SYNC)
4574                 m->mdt_fl_synced = 1;
4575
4576         if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4577                 /* Open for clients */
4578                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4579 }
4580
4581 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4582                       enum md_upcall_event ev)
4583 {
4584         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4585         struct md_device  *next  = m->mdt_child;
4586         struct mdt_thread_info *mti;
4587         int rc = 0;
4588         ENTRY;
4589
4590         switch (ev) {
4591                 case MD_LOV_SYNC:
4592                         rc = next->md_ops->mdo_maxsize_get(env, next,
4593                                         &m->mdt_max_mdsize,
4594                                         &m->mdt_max_cookiesize);
4595                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4596                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
4597                         mdt_allow_cli(m, CONFIG_SYNC);
4598                         break;
4599                 case MD_NO_TRANS:
4600                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4601                         mti->mti_no_need_trans = 1;
4602                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4603                         break;
4604                 case MD_LOV_CONFIG:
4605                         /* Check that MDT is not yet configured */
4606                         LASSERT(!m->mdt_fl_cfglog);
4607                         break;
4608                 default:
4609                         CERROR("invalid event\n");
4610                         rc = -EINVAL;
4611                         break;
4612         }
4613         RETURN(rc);
4614 }
4615
4616 static int mdt_obd_notify(struct obd_device *host,
4617                           struct obd_device *watched,
4618                           enum obd_notify_event ev, void *data)
4619 {
4620         ENTRY;
4621
4622         switch (ev) {
4623         case OBD_NOTIFY_CONFIG:
4624                 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data);
4625                 break;
4626         default:
4627                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4628         }
4629         RETURN(0);
4630 }
4631
4632 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4633                          void *karg, void *uarg)
4634 {
4635         struct lu_env      env;
4636         struct obd_device *obd= exp->exp_obd;
4637         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4638         struct dt_device  *dt = mdt->mdt_bottom;
4639         int rc;
4640
4641         ENTRY;
4642         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4643         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4644         if (rc)
4645                 RETURN(rc);
4646
4647         switch (cmd) {
4648         case OBD_IOC_SYNC:
4649                 rc = dt->dd_ops->dt_sync(&env, dt);
4650                 break;
4651         case OBD_IOC_SET_READONLY:
4652                 rc = dt->dd_ops->dt_sync(&env, dt);
4653                 dt->dd_ops->dt_ro(&env, dt);
4654                 break;
4655         case OBD_IOC_ABORT_RECOVERY:
4656                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4657                 target_stop_recovery_thread(obd);
4658                 rc = 0;
4659                 break;
4660         default:
4661                 CERROR("Not supported cmd = %d for device %s\n",
4662                        cmd, obd->obd_name);
4663                 rc = -EOPNOTSUPP;
4664         }
4665
4666         lu_env_fini(&env);
4667         RETURN(rc);
4668 }
4669
4670 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4671 {
4672         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4673         struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4674         int rc, lost;
4675         ENTRY;
4676         /* if some clients didn't participate in recovery then we can possibly
4677          * lost sequence. Now we should increase sequence for safe value */
4678         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4679         mdt_seq_adjust(env, mdt, lost);
4680
4681         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4682         RETURN(rc);
4683 }
4684
4685 int mdt_obd_postrecov(struct obd_device *obd)
4686 {
4687         struct lu_env env;
4688         int rc;
4689
4690         rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4691         if (rc)
4692                 RETURN(rc);
4693         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4694         lu_env_fini(&env);
4695         return rc;
4696 }
4697
4698 static struct obd_ops mdt_obd_device_ops = {
4699         .o_owner          = THIS_MODULE,
4700         .o_connect        = mdt_obd_connect,
4701         .o_reconnect      = mdt_obd_reconnect,
4702         .o_disconnect     = mdt_obd_disconnect,
4703         .o_init_export    = mdt_init_export,
4704         .o_destroy_export = mdt_destroy_export,
4705         .o_iocontrol      = mdt_iocontrol,
4706         .o_postrecov      = mdt_obd_postrecov,
4707         .o_notify         = mdt_obd_notify
4708 };
4709
4710 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4711                                          struct lu_device *d)
4712 {
4713         struct mdt_device *m = mdt_dev(d);
4714         ENTRY;
4715
4716         mdt_fini(env, m);
4717         RETURN(NULL);
4718 }
4719
4720 static struct lu_device *mdt_device_free(const struct lu_env *env,
4721                                          struct lu_device *d)
4722 {
4723         struct mdt_device *m = mdt_dev(d);
4724         ENTRY;
4725
4726         OBD_FREE_PTR(m);
4727         RETURN(NULL);
4728 }
4729
4730 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4731                                           struct lu_device_type *t,
4732                                           struct lustre_cfg *cfg)
4733 {
4734         struct lu_device  *l;
4735         struct mdt_device *m;
4736
4737         OBD_ALLOC_PTR(m);
4738         if (m != NULL) {
4739                 int rc;
4740
4741                 l = &m->mdt_md_dev.md_lu_dev;
4742                 rc = mdt_init0(env, m, t, cfg);
4743                 if (rc != 0) {
4744                         OBD_FREE_PTR(m);
4745                         l = ERR_PTR(rc);
4746                         return l;
4747                 }
4748                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4749         } else
4750                 l = ERR_PTR(-ENOMEM);
4751         return l;
4752 }
4753
4754 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4755 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4756
4757 /* context key: mdt_thread_key */
4758 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4759
4760 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4761 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4762
4763 struct lu_context_key mdt_txn_key = {
4764         .lct_tags = LCT_TX_HANDLE,
4765         .lct_init = mdt_txn_key_init,
4766         .lct_fini = mdt_txn_key_fini
4767 };
4768
4769 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4770 {
4771         return md_ucred(info->mti_env);
4772 }
4773
4774 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4775 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4776
4777 static struct lu_device_type_operations mdt_device_type_ops = {
4778         .ldto_init = mdt_type_init,
4779         .ldto_fini = mdt_type_fini,
4780
4781         .ldto_device_alloc = mdt_device_alloc,
4782         .ldto_device_free  = mdt_device_free,
4783         .ldto_device_fini  = mdt_device_fini
4784 };
4785
4786 static struct lu_device_type mdt_device_type = {
4787         .ldt_tags     = LU_DEVICE_MD,
4788         .ldt_name     = LUSTRE_MDT_NAME,
4789         .ldt_ops      = &mdt_device_type_ops,
4790         .ldt_ctx_tags = LCT_MD_THREAD
4791 };
4792
4793 static int __init mdt_mod_init(void)
4794 {
4795         struct lprocfs_static_vars lvars;
4796         int rc;
4797
4798         mdt_num_threads = MDT_NUM_THREADS;
4799         lprocfs_mdt_init_vars(&lvars);
4800         rc = class_register_type(&mdt_obd_device_ops, NULL,
4801                                  lvars.module_vars, LUSTRE_MDT_NAME,
4802                                  &mdt_device_type);
4803
4804         return rc;
4805 }
4806
4807 static void __exit mdt_mod_exit(void)
4808 {
4809         class_unregister_type(LUSTRE_MDT_NAME);
4810 }
4811
4812
4813 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
4814 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
4815         .mh_name    = #opc,                                             \
4816         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
4817         .mh_opc     = prefix ## _  ## opc,                              \
4818         .mh_flags   = flags,                                            \
4819         .mh_act     = fn,                                               \
4820         .mh_fmt     = fmt                                               \
4821 }
4822
4823 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
4824         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4825
4826 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
4827         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4828
4829 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
4830         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4831 /*
4832  * Request with a format known in advance
4833  */
4834 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
4835         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4836
4837 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
4838         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4839
4840 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
4841         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4842 /*
4843  * Request with a format we do not yet know
4844  */
4845 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
4846         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4847
4848 static struct mdt_handler mdt_mds_ops[] = {
4849 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4850 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
4851 DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
4852 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
4853 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
4854 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4855 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
4856 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
4857 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
4858 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
4859 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
4860 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
4861 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
4862 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4863 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
4864 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
4865 };
4866
4867 #define DEF_OBD_HNDL(flags, name, fn)                   \
4868         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4869
4870
4871 static struct mdt_handler mdt_obd_ops[] = {
4872         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
4873         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
4874         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
4875 };
4876
4877 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
4878         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4879 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
4880         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4881
4882 static struct mdt_handler mdt_dlm_ops[] = {
4883         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
4884         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
4885         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
4886         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
4887 };
4888
4889 static struct mdt_handler mdt_llog_ops[] = {
4890 };
4891
4892 #define DEF_SEC_CTX_HNDL(name, fn)                      \
4893         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4894
4895 static struct mdt_handler mdt_sec_ctx_ops[] = {
4896         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
4897         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
4898         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
4899 };
4900
4901 static struct mdt_opc_slice mdt_regular_handlers[] = {
4902         {
4903                 .mos_opc_start = MDS_GETATTR,
4904                 .mos_opc_end   = MDS_LAST_OPC,
4905                 .mos_hs        = mdt_mds_ops
4906         },
4907         {
4908                 .mos_opc_start = OBD_PING,
4909                 .mos_opc_end   = OBD_LAST_OPC,
4910                 .mos_hs        = mdt_obd_ops
4911         },
4912         {
4913                 .mos_opc_start = LDLM_ENQUEUE,
4914                 .mos_opc_end   = LDLM_LAST_OPC,
4915                 .mos_hs        = mdt_dlm_ops
4916         },
4917         {
4918                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4919                 .mos_opc_end   = LLOG_LAST_OPC,
4920                 .mos_hs        = mdt_llog_ops
4921         },
4922         {
4923                 .mos_opc_start = SEC_CTX_INIT,
4924                 .mos_opc_end   = SEC_LAST_OPC,
4925                 .mos_hs        = mdt_sec_ctx_ops
4926         },
4927         {
4928                 .mos_hs        = NULL
4929         }
4930 };
4931
4932 static struct mdt_handler mdt_readpage_ops[] = {
4933         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
4934         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4935 #ifdef HAVE_SPLIT_SUPPORT
4936         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4937 #endif
4938
4939         /*
4940          * XXX: this is ugly and should be fixed one day, see mdc_close() for
4941          * detailed comments. --umka
4942          */
4943         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
4944         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
4945 };
4946
4947 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4948         {
4949                 .mos_opc_start = MDS_GETATTR,
4950                 .mos_opc_end   = MDS_LAST_OPC,
4951                 .mos_hs        = mdt_readpage_ops
4952         },
4953         {
4954                 .mos_hs        = NULL
4955         }
4956 };
4957
4958 static struct mdt_handler mdt_xmds_ops[] = {
4959         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
4960         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
4961         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
4962         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
4963 };
4964
4965 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4966         {
4967                 .mos_opc_start = MDS_GETATTR,
4968                 .mos_opc_end   = MDS_LAST_OPC,
4969                 .mos_hs        = mdt_xmds_ops
4970         },
4971         {
4972                 .mos_opc_start = OBD_PING,
4973                 .mos_opc_end   = OBD_LAST_OPC,
4974                 .mos_hs        = mdt_obd_ops
4975         },
4976         {
4977                 .mos_opc_start = SEC_CTX_INIT,
4978                 .mos_opc_end   = SEC_LAST_OPC,
4979                 .mos_hs        = mdt_sec_ctx_ops
4980         },
4981         {
4982                 .mos_hs        = NULL
4983         }
4984 };
4985
4986 static struct mdt_handler mdt_seq_ops[] = {
4987         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4988 };
4989
4990 static struct mdt_opc_slice mdt_seq_handlers[] = {
4991         {
4992                 .mos_opc_start = SEQ_QUERY,
4993                 .mos_opc_end   = SEQ_LAST_OPC,
4994                 .mos_hs        = mdt_seq_ops
4995         },
4996         {
4997                 .mos_hs        = NULL
4998         }
4999 };
5000
5001 static struct mdt_handler mdt_fld_ops[] = {
5002         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5003 };
5004
5005 static struct mdt_opc_slice mdt_fld_handlers[] = {
5006         {
5007                 .mos_opc_start = FLD_QUERY,
5008                 .mos_opc_end   = FLD_LAST_OPC,
5009                 .mos_hs        = mdt_fld_ops
5010         },
5011         {
5012                 .mos_hs        = NULL
5013         }
5014 };
5015
5016 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5017 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
5018 MODULE_LICENSE("GPL");
5019
5020 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
5021                 "number of mdt service threads to start");
5022
5023 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);