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 (c) 2007, 2010, Oracle and/or its affiliates. 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 #ifdef HAVE_QUOTA_SUPPORT
  71 # include <lustre_quota.h>
  72 #endif
  73 #include <lustre_acl.h>
  74 #include <lustre_param.h>
  75 #include <lustre_fsfilt.h>
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 ldlm_mode_t mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE] = LCK_MINMODE,
  90         [MDL_EX]      = LCK_EX,
  91         [MDL_PW]      = LCK_PW,
  92         [MDL_PR]      = LCK_PR,
  93         [MDL_CW]      = LCK_CW,
  94         [MDL_CR]      = LCK_CR,
  95         [MDL_NL]      = LCK_NL,
  96         [MDL_GROUP]   = LCK_GROUP
  97 };
  98
  99 /*
 100  * Initialized in mdt_mod_init().
 101  */
 102 static unsigned long mdt_num_threads;
 103 static unsigned long mdt_min_threads;
 104 static unsigned long mdt_max_threads;
 105
 106 /* ptlrpc request handler for MDT. All handlers are
 107  * grouped into several slices - struct mdt_opc_slice,
 108  * and stored in an array - mdt_handlers[].
 109  */
 110 struct mdt_handler {
 111         /* The name of this handler. */
 112         const char *mh_name;
 113         /* Fail id for this handler, checked at the beginning of this handler*/
 114         int         mh_fail_id;
 115         /* Operation code for this handler */
 116         __u32       mh_opc;
 117         /* flags are listed in enum mdt_handler_flags below. */
 118         __u32       mh_flags;
 119         /* The actual handler function to execute. */
 120         int (*mh_act)(struct mdt_thread_info *info);
 121         /* Request format for this request. */
 122         const struct req_format *mh_fmt;
 123 };
 124
 125 enum mdt_handler_flags {
 126         /*
 127          * struct mdt_body is passed in the incoming message, and object
 128          * identified by this fid exists on disk.
 129          *
 130          * "habeo corpus" == "I have a body"
 131          */
 132         HABEO_CORPUS = (1 << 0),
 133         /*
 134          * struct ldlm_request is passed in the incoming message.
 135          *
 136          * "habeo clavis" == "I have a key"
 137          */
 138         HABEO_CLAVIS = (1 << 1),
 139         /*
 140          * this request has fixed reply format, so that reply message can be
 141          * packed by generic code.
 142          *
 143          * "habeo refero" == "I have a reply"
 144          */
 145         HABEO_REFERO = (1 << 2),
 146         /*
 147          * this request will modify something, so check whether the filesystem
 148          * is readonly or not, then return -EROFS to client asap if necessary.
 149          *
 150          * "mutabor" == "I shall modify"
 151          */
 152         MUTABOR      = (1 << 3)
 153 };
 154
 155 struct mdt_opc_slice {
 156         __u32               mos_opc_start;
 157         int                 mos_opc_end;
 158         struct mdt_handler *mos_hs;
 159 };
 160
 161 static struct mdt_opc_slice mdt_regular_handlers[];
 162 static struct mdt_opc_slice mdt_readpage_handlers[];
 163 static struct mdt_opc_slice mdt_xmds_handlers[];
 164 static struct mdt_opc_slice mdt_seq_handlers[];
 165 static struct mdt_opc_slice mdt_fld_handlers[];
 166
 167 static struct mdt_device *mdt_dev(struct lu_device *d);
 168 static int mdt_regular_handle(struct ptlrpc_request *req);
 169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 171                         struct getinfo_fid2path *fp);
 172
 173 static const struct lu_object_operations mdt_obj_ops;
 174
 175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 176 {
 177         if (!rep)
 178                 return 0;
 179         return (rep->lock_policy_res1 & flag);
 180 }
 181
 182 void mdt_clear_disposition(struct mdt_thread_info *info,
 183                            struct ldlm_reply *rep, int flag)
 184 {
 185         if (info)
 186                 info->mti_opdata &= ~flag;
 187         if (rep)
 188                 rep->lock_policy_res1 &= ~flag;
 189 }
 190
 191 void mdt_set_disposition(struct mdt_thread_info *info,
 192                          struct ldlm_reply *rep, int flag)
 193 {
 194         if (info)
 195                 info->mti_opdata |= flag;
 196         if (rep)
 197                 rep->lock_policy_res1 |= flag;
 198 }
 199
 200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 201 {
 202         lh->mlh_pdo_hash = 0;
 203         lh->mlh_reg_mode = lm;
 204         lh->mlh_type = MDT_REG_LOCK;
 205 }
 206
 207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 208                        const char *name, int namelen)
 209 {
 210         lh->mlh_reg_mode = lm;
 211         lh->mlh_type = MDT_PDO_LOCK;
 212
 213         if (name != NULL) {
 214                 LASSERT(namelen > 0);
 215                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 216         } else {
 217                 LASSERT(namelen == 0);
 218                 lh->mlh_pdo_hash = 0ull;
 219         }
 220 }
 221
 222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 223                               struct mdt_lock_handle *lh)
 224 {
 225         mdl_mode_t mode;
 226         ENTRY;
 227
 228         /*
 229          * Any dir access needs couple of locks:
 230          *
 231          * 1) on part of dir we gonna take lookup/modify;
 232          *
 233          * 2) on whole dir to protect it from concurrent splitting and/or to
 234          * flush client's cache for readdir().
 235          *
 236          * so, for a given mode and object this routine decides what lock mode
 237          * to use for lock #2:
 238          *
 239          * 1) if caller's gonna lookup in dir then we need to protect dir from
 240          * being splitted only - LCK_CR
 241          *
 242          * 2) if caller's gonna modify dir then we need to protect dir from
 243          * being splitted and to flush cache - LCK_CW
 244          *
 245          * 3) if caller's gonna modify dir and that dir seems ready for
 246          * splitting then we need to protect it from any type of access
 247          * (lookup/modify/split) - LCK_EX --bzzz
 248          */
 249
 250         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 251         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 252
 253         /*
 254          * Ask underlaying level its opinion about preferable PDO lock mode
 255          * having access type passed as regular lock mode:
 256          *
 257          * - MDL_MINMODE means that lower layer does not want to specify lock
 258          * mode;
 259          *
 260          * - MDL_NL means that no PDO lock should be taken. This is used in some
 261          * cases. Say, for non-splittable directories no need to use PDO locks
 262          * at all.
 263          */
 264         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 265                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 266
 267         if (mode != MDL_MINMODE) {
 268                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 269         } else {
 270                 /*
 271                  * Lower layer does not want to specify locking mode. We do it
 272                  * our selves. No special protection is needed, just flush
 273                  * client's cache on modification and allow concurrent
 274                  * mondification.
 275                  */
 276                 switch (lh->mlh_reg_mode) {
 277                 case LCK_EX:
 278                         lh->mlh_pdo_mode = LCK_EX;
 279                         break;
 280                 case LCK_PR:
 281                         lh->mlh_pdo_mode = LCK_CR;
 282                         break;
 283                 case LCK_PW:
 284                         lh->mlh_pdo_mode = LCK_CW;
 285                         break;
 286                 default:
 287                         CERROR("Not expected lock type (0x%x)\n",
 288                                (int)lh->mlh_reg_mode);
 289                         LBUG();
 290                 }
 291         }
 292
 293         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 294         EXIT;
 295 }
 296
 297 static int mdt_getstatus(struct mdt_thread_info *info)
 298 {
 299         struct mdt_device *mdt  = info->mti_mdt;
 300         struct md_device  *next = mdt->mdt_child;
 301         struct mdt_body   *repbody;
 302         int                rc;
 303
 304         ENTRY;
 305
 306         rc = mdt_check_ucred(info);
 307         if (rc)
 308                 RETURN(err_serious(rc));
 309
 310         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 311                 RETURN(err_serious(-ENOMEM));
 312
 313         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 314         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 315         if (rc != 0)
 316                 RETURN(rc);
 317
 318         repbody->valid |= OBD_MD_FLID;
 319
 320         if (mdt->mdt_opts.mo_mds_capa &&
 321             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 322                 struct mdt_object  *root;
 323                 struct lustre_capa *capa;
 324
 325                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 326                 if (IS_ERR(root))
 327                         RETURN(PTR_ERR(root));
 328
 329                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 330                 LASSERT(capa);
 331                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 332                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 333                                  0);
 334                 mdt_object_put(info->mti_env, root);
 335                 if (rc == 0)
 336                         repbody->valid |= OBD_MD_FLMDSCAPA;
 337         }
 338
 339         RETURN(rc);
 340 }
 341
 342 static int mdt_statfs(struct mdt_thread_info *info)
 343 {
 344         struct md_device      *next  = info->mti_mdt->mdt_child;
 345         struct ptlrpc_service *svc;
 346         struct obd_statfs     *osfs;
 347         int                    rc;
 348
 349         ENTRY;
 350
 351         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 352
 353         /* This will trigger a watchdog timeout */
 354         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 355                          (MDT_SERVICE_WATCHDOG_FACTOR *
 356                           at_get(&svc->srv_at_estimate)) + 1);
 357
 358         rc = mdt_check_ucred(info);
 359         if (rc)
 360                 RETURN(err_serious(rc));
 361
 362         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 363                 rc = err_serious(-ENOMEM);
 364         } else {
 365                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 366                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 367                                               &info->mti_u.ksfs);
 368                 statfs_pack(osfs, &info->mti_u.ksfs);
 369         }
 370         RETURN(rc);
 371 }
 372
 373 /**
 374  * Pack SOM attributes into the reply.
 375  * Call under a DLM UPDATE lock.
 376  */
 377 static void mdt_pack_size2body(struct mdt_thread_info *info,
 378                                struct mdt_object *mo)
 379 {
 380         struct mdt_body *b;
 381         struct md_attr *ma = &info->mti_attr;
 382
 383         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 384         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 385
 386         /* Check if Size-on-MDS is supported, if this is a regular file,
 387          * if SOM is enabled on the object and if SOM cache exists and valid.
 388          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 389         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 390             !S_ISREG(ma->ma_attr.la_mode) ||
 391             !mdt_object_is_som_enabled(mo) ||
 392             !(ma->ma_valid & MA_SOM))
 393                 return;
 394
 395         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 396         b->size = ma->ma_som->msd_size;
 397         b->blocks = ma->ma_som->msd_blocks;
 398 }
 399
 400 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 401                         const struct lu_attr *attr, const struct lu_fid *fid)
 402 {
 403         struct md_attr *ma = &info->mti_attr;
 404
 405         LASSERT(ma->ma_valid & MA_INODE);
 406
 407         b->atime      = attr->la_atime;
 408         b->mtime      = attr->la_mtime;
 409         b->ctime      = attr->la_ctime;
 410         b->mode       = attr->la_mode;
 411         b->size       = attr->la_size;
 412         b->blocks     = attr->la_blocks;
 413         b->uid        = attr->la_uid;
 414         b->gid        = attr->la_gid;
 415         b->flags      = attr->la_flags;
 416         b->nlink      = attr->la_nlink;
 417         b->rdev       = attr->la_rdev;
 418
 419         /*XXX should pack the reply body according to lu_valid*/
 420         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 421                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 422                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 423                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 424
 425         if (!S_ISREG(attr->la_mode)) {
 426                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 427         } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
 428                 /* means no objects are allocated on osts. */
 429                 LASSERT(!(ma->ma_valid & MA_LOV));
 430                 /* just ignore blocks occupied by extend attributes on MDS */
 431                 b->blocks = 0;
 432                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 433                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 434         }
 435
 436         if (fid) {
 437                 b->fid1 = *fid;
 438                 b->valid |= OBD_MD_FLID;
 439
 440                 /* FIXME: these should be fixed when new igif ready.*/
 441                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 442                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 443                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 444
 445                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 446                                 PFID(fid), b->nlink, b->mode, b->size);
 447         }
 448
 449         if (info)
 450                 mdt_body_reverse_idmap(info, b);
 451
 452         if (b->valid & OBD_MD_FLSIZE)
 453                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 454                        PFID(fid), (unsigned long long)b->size);
 455 }
 456
 457 static inline int mdt_body_has_lov(const struct lu_attr *la,
 458                                    const struct mdt_body *body)
 459 {
 460         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 461                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 462 }
 463
 464 static int mdt_getattr_internal(struct mdt_thread_info *info,
 465                                 struct mdt_object *o, int ma_need)
 466 {
 467         struct md_object        *next = mdt_object_child(o);
 468         const struct mdt_body   *reqbody = info->mti_body;
 469         struct ptlrpc_request   *req = mdt_info_req(info);
 470         struct md_attr          *ma = &info->mti_attr;
 471         struct lu_attr          *la = &ma->ma_attr;
 472         struct req_capsule      *pill = info->mti_pill;
 473         const struct lu_env     *env = info->mti_env;
 474         struct mdt_body         *repbody;
 475         struct lu_buf           *buffer = &info->mti_buf;
 476         int                     rc;
 477         ENTRY;
 478
 479         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 480                 RETURN(err_serious(-ENOMEM));
 481
 482         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 483
 484         ma->ma_valid = 0;
 485
 486         rc = mdt_object_exists(o);
 487         if (rc < 0) {
 488                 /* This object is located on remote node.*/
 489                 repbody->fid1 = *mdt_object_fid(o);
 490                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 491                 RETURN(0);
 492         }
 493
 494         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 495         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 496
 497         /* If it is dir object and client require MEA, then we got MEA */
 498         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 499             reqbody->valid & OBD_MD_MEA) {
 500                 /* Assumption: MDT_MD size is enough for lmv size. */
 501                 ma->ma_lmv = buffer->lb_buf;
 502                 ma->ma_lmv_size = buffer->lb_len;
 503                 ma->ma_need = MA_LMV | MA_INODE;
 504         } else {
 505                 ma->ma_lmm = buffer->lb_buf;
 506                 ma->ma_lmm_size = buffer->lb_len;
 507                 ma->ma_need = MA_LOV | MA_INODE;
 508         }
 509
 510         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 511             reqbody->valid & OBD_MD_FLDIREA  &&
 512             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 513                 /* get default stripe info for this dir. */
 514                 ma->ma_need |= MA_LOV_DEF;
 515         }
 516         ma->ma_need |= ma_need;
 517         if (ma->ma_need & MA_SOM)
 518                 ma->ma_som = &info->mti_u.som.data;
 519
 520         rc = mo_attr_get(env, next, ma);
 521         if (unlikely(rc)) {
 522                 CERROR("getattr error for "DFID": %d\n",
 523                         PFID(mdt_object_fid(o)), rc);
 524                 RETURN(rc);
 525         }
 526
 527         if (likely(ma->ma_valid & MA_INODE))
 528                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 529         else
 530                 RETURN(-EFAULT);
 531
 532         if (mdt_body_has_lov(la, reqbody)) {
 533                 if (ma->ma_valid & MA_LOV) {
 534                         LASSERT(ma->ma_lmm_size);
 535                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 536                         repbody->eadatasize = ma->ma_lmm_size;
 537                         if (S_ISDIR(la->la_mode))
 538                                 repbody->valid |= OBD_MD_FLDIREA;
 539                         else
 540                                 repbody->valid |= OBD_MD_FLEASIZE;
 541                 }
 542                 if (ma->ma_valid & MA_LMV) {
 543                         LASSERT(S_ISDIR(la->la_mode));
 544                         repbody->eadatasize = ma->ma_lmv_size;
 545                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 546                 }
 547         } else if (S_ISLNK(la->la_mode) &&
 548                    reqbody->valid & OBD_MD_LINKNAME) {
 549                 buffer->lb_buf = ma->ma_lmm;
 550                 buffer->lb_len = reqbody->eadatasize;
 551                 rc = mo_readlink(env, next, buffer);
 552                 if (unlikely(rc <= 0)) {
 553                         CERROR("readlink failed: %d\n", rc);
 554                         rc = -EFAULT;
 555                 } else {
 556                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 557                                  rc -= 2;
 558                         repbody->valid |= OBD_MD_LINKNAME;
 559                         repbody->eadatasize = rc;
 560                         /* NULL terminate */
 561                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 562                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 563                                (char*)ma->ma_lmm, rc);
 564                         rc = 0;
 565                 }
 566         }
 567
 568         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 569                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 570                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 571                 repbody->valid |= OBD_MD_FLMODEASIZE;
 572                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 573                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 574                        repbody->max_cookiesize);
 575         }
 576
 577         if (exp_connect_rmtclient(info->mti_exp) &&
 578             reqbody->valid & OBD_MD_FLRMTPERM) {
 579                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 580
 581                 /* mdt_getattr_lock only */
 582                 rc = mdt_pack_remote_perm(info, o, buf);
 583                 if (rc) {
 584                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 585                         repbody->aclsize = 0;
 586                         RETURN(rc);
 587                 } else {
 588                         repbody->valid |= OBD_MD_FLRMTPERM;
 589                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 590                 }
 591         }
 592 #ifdef CONFIG_FS_POSIX_ACL
 593         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 594                  (reqbody->valid & OBD_MD_FLACL)) {
 595                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 596                 buffer->lb_len = req_capsule_get_size(pill,
 597                                                       &RMF_ACL, RCL_SERVER);
 598                 if (buffer->lb_len > 0) {
 599                         rc = mo_xattr_get(env, next, buffer,
 600                                           XATTR_NAME_ACL_ACCESS);
 601                         if (rc < 0) {
 602                                 if (rc == -ENODATA) {
 603                                         repbody->aclsize = 0;
 604                                         repbody->valid |= OBD_MD_FLACL;
 605                                         rc = 0;
 606                                 } else if (rc == -EOPNOTSUPP) {
 607                                         rc = 0;
 608                                 } else {
 609                                         CERROR("got acl size: %d\n", rc);
 610                                 }
 611                         } else {
 612                                 repbody->aclsize = rc;
 613                                 repbody->valid |= OBD_MD_FLACL;
 614                                 rc = 0;
 615                         }
 616                 }
 617         }
 618 #endif
 619
 620         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 621             info->mti_mdt->mdt_opts.mo_mds_capa &&
 622             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 623                 struct lustre_capa *capa;
 624
 625                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 626                 LASSERT(capa);
 627                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 628                 rc = mo_capa_get(env, next, capa, 0);
 629                 if (rc)
 630                         RETURN(rc);
 631                 repbody->valid |= OBD_MD_FLMDSCAPA;
 632         }
 633         RETURN(rc);
 634 }
 635
 636 static int mdt_renew_capa(struct mdt_thread_info *info)
 637 {
 638         struct mdt_object  *obj = info->mti_object;
 639         struct mdt_body    *body;
 640         struct lustre_capa *capa, *c;
 641         int rc;
 642         ENTRY;
 643
 644         /* if object doesn't exist, or server has disabled capability,
 645          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 646          * flag not set.
 647          */
 648         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 649             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 650                 RETURN(0);
 651
 652         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 653         LASSERT(body != NULL);
 654
 655         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 656         LASSERT(c);
 657
 658         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 659         LASSERT(capa);
 660
 661         *capa = *c;
 662         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 663         if (rc == 0)
 664                 body->valid |= OBD_MD_FLOSSCAPA;
 665         RETURN(rc);
 666 }
 667
 668 static int mdt_getattr(struct mdt_thread_info *info)
 669 {
 670         struct mdt_object       *obj = info->mti_object;
 671         struct req_capsule      *pill = info->mti_pill;
 672         struct mdt_body         *reqbody;
 673         struct mdt_body         *repbody;
 674         mode_t                   mode;
 675         int                      md_size;
 676         int rc;
 677         ENTRY;
 678
 679         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 680         LASSERT(reqbody);
 681
 682         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 683                 rc = req_capsule_server_pack(pill);
 684                 if (unlikely(rc))
 685                         RETURN(err_serious(rc));
 686                 rc = mdt_renew_capa(info);
 687                 GOTO(out_shrink, rc);
 688         }
 689
 690         LASSERT(obj != NULL);
 691         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 692
 693         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 694         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 695             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 696                 md_size = reqbody->eadatasize;
 697         else
 698                 md_size = info->mti_mdt->mdt_max_mdsize;
 699
 700         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 701
 702         rc = req_capsule_server_pack(pill);
 703         if (unlikely(rc != 0))
 704                 RETURN(err_serious(rc));
 705
 706         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 707         LASSERT(repbody != NULL);
 708         repbody->eadatasize = 0;
 709         repbody->aclsize = 0;
 710
 711         if (reqbody->valid & OBD_MD_FLRMTPERM)
 712                 rc = mdt_init_ucred(info, reqbody);
 713         else
 714                 rc = mdt_check_ucred(info);
 715         if (unlikely(rc))
 716                 GOTO(out_shrink, rc);
 717
 718         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 719         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 720
 721         /*
 722          * Don't check capability at all, because rename might getattr for
 723          * remote obj, and at that time no capability is available.
 724          */
 725         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 726         rc = mdt_getattr_internal(info, obj, 0);
 727         if (reqbody->valid & OBD_MD_FLRMTPERM)
 728                 mdt_exit_ucred(info);
 729         EXIT;
 730 out_shrink:
 731         mdt_shrink_reply(info);
 732         return rc;
 733 }
 734
 735 static int mdt_is_subdir(struct mdt_thread_info *info)
 736 {
 737         struct mdt_object     *o = info->mti_object;
 738         struct req_capsule    *pill = info->mti_pill;
 739         const struct mdt_body *body = info->mti_body;
 740         struct mdt_body       *repbody;
 741         int                    rc;
 742         ENTRY;
 743
 744         LASSERT(o != NULL);
 745
 746         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 747
 748         /*
 749          * We save last checked parent fid to @repbody->fid1 for remote
 750          * directory case.
 751          */
 752         LASSERT(fid_is_sane(&body->fid2));
 753         LASSERT(mdt_object_exists(o) > 0);
 754         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 755                            &body->fid2, &repbody->fid1);
 756         if (rc == 0 || rc == -EREMOTE)
 757                 repbody->valid |= OBD_MD_FLID;
 758
 759         RETURN(rc);
 760 }
 761
 762 static int mdt_raw_lookup(struct mdt_thread_info *info,
 763                           struct mdt_object *parent,
 764                           const struct lu_name *lname,
 765                           struct ldlm_reply *ldlm_rep)
 766 {
 767         struct md_object *next = mdt_object_child(info->mti_object);
 768         const struct mdt_body *reqbody = info->mti_body;
 769         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 770         struct mdt_body *repbody;
 771         int rc;
 772         ENTRY;
 773
 774         if (reqbody->valid != OBD_MD_FLID)
 775                 RETURN(0);
 776
 777         LASSERT(!info->mti_cross_ref);
 778
 779         /* Only got the fid of this obj by name */
 780         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 781                         &info->mti_spec);
 782 #if 0
 783         /* XXX is raw_lookup possible as intent operation? */
 784         if (rc != 0) {
 785                 if (rc == -ENOENT)
 786                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 787                 RETURN(rc);
 788         } else
 789                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 790
 791         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 792 #endif
 793         if (rc == 0) {
 794                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 795                 repbody->fid1 = *child_fid;
 796                 repbody->valid = OBD_MD_FLID;
 797         }
 798         RETURN(1);
 799 }
 800
 801 /*
 802  * UPDATE lock should be taken against parent, and be release before exit;
 803  * child_bits lock should be taken against child, and be returned back:
 804  *            (1)normal request should release the child lock;
 805  *            (2)intent request will grant the lock to client.
 806  */
 807 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 808                                  struct mdt_lock_handle *lhc,
 809                                  __u64 child_bits,
 810                                  struct ldlm_reply *ldlm_rep)
 811 {
 812         struct ptlrpc_request  *req       = mdt_info_req(info);
 813         struct mdt_body        *reqbody   = NULL;
 814         struct mdt_object      *parent    = info->mti_object;
 815         struct mdt_object      *child;
 816         struct md_object       *next      = mdt_object_child(parent);
 817         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 818         struct lu_name         *lname     = NULL;
 819         const char             *name      = NULL;
 820         int                     namelen   = 0;
 821         struct mdt_lock_handle *lhp;
 822         struct ldlm_lock       *lock;
 823         struct ldlm_res_id     *res_id;
 824         int                     is_resent;
 825         int                     ma_need = 0;
 826         int                     rc;
 827
 828         ENTRY;
 829
 830         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 831         LASSERT(ergo(is_resent,
 832                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 833
 834         LASSERT(parent != NULL);
 835         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 836         if (name == NULL)
 837                 RETURN(err_serious(-EFAULT));
 838
 839         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 840                                        RCL_CLIENT) - 1;
 841         if (!info->mti_cross_ref) {
 842                 /*
 843                  * XXX: Check for "namelen == 0" is for getattr by fid
 844                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 845                  * that is the name must contain at least one character and
 846                  * the terminating '\0'
 847                  */
 848                 if (namelen == 0) {
 849                         reqbody = req_capsule_client_get(info->mti_pill,
 850                                                          &RMF_MDT_BODY);
 851                         LASSERT(fid_is_sane(&reqbody->fid2));
 852                         name = NULL;
 853
 854                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 855                                "ldlm_rep = %p\n",
 856                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 857                                ldlm_rep);
 858                 } else {
 859                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 860                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 861                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 862                                name, ldlm_rep);
 863                 }
 864         }
 865         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 866
 867         rc = mdt_object_exists(parent);
 868         if (unlikely(rc == 0)) {
 869                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 870                                 &parent->mot_obj.mo_lu,
 871                                 "Parent doesn't exist!\n");
 872                 RETURN(-ESTALE);
 873         } else if (!info->mti_cross_ref) {
 874                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 875                          PFID(mdt_object_fid(parent)));
 876         }
 877         if (lname) {
 878                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 879                 if (rc != 0) {
 880                         if (rc > 0)
 881                                 rc = 0;
 882                         RETURN(rc);
 883                 }
 884         }
 885
 886         if (info->mti_cross_ref) {
 887                 /* Only getattr on the child. Parent is on another node. */
 888                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 889                 child = parent;
 890                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 891                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 892
 893                 if (is_resent) {
 894                         /* Do not take lock for resent case. */
 895                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 896                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 897                                  lhc->mlh_reg_lh.cookie);
 898                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 899                                                 &lock->l_resource->lr_name));
 900                         LDLM_LOCK_PUT(lock);
 901                         rc = 0;
 902                 } else {
 903                         mdt_lock_handle_init(lhc);
 904                         mdt_lock_reg_init(lhc, LCK_PR);
 905
 906                         /*
 907                          * Object's name is on another MDS, no lookup lock is
 908                          * needed here but update is.
 909                          */
 910                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 911                         child_bits |= MDS_INODELOCK_UPDATE;
 912
 913                         rc = mdt_object_lock(info, child, lhc, child_bits,
 914                                              MDT_LOCAL_LOCK);
 915                 }
 916                 if (rc == 0) {
 917                         /* Finally, we can get attr for child. */
 918                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 919                                          BYPASS_CAPA);
 920                         rc = mdt_getattr_internal(info, child, 0);
 921                         if (unlikely(rc != 0))
 922                                 mdt_object_unlock(info, child, lhc, 1);
 923                 }
 924                 RETURN(rc);
 925         }
 926
 927         /* step 1: lock parent */
 928         lhp = &info->mti_lh[MDT_LH_PARENT];
 929         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 930         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 931                              MDT_LOCAL_LOCK);
 932
 933         if (unlikely(rc != 0))
 934                 RETURN(rc);
 935
 936         if (lname) {
 937                 /* step 2: lookup child's fid by name */
 938                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 939                                 &info->mti_spec);
 940
 941                 if (rc != 0) {
 942                         if (rc == -ENOENT)
 943                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 944                         GOTO(out_parent, rc);
 945                 } else
 946                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 947         } else {
 948                 *child_fid = reqbody->fid2;
 949                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 950         }
 951
 952         /*
 953          *step 3: find the child object by fid & lock it.
 954          *        regardless if it is local or remote.
 955          */
 956         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 957
 958         if (unlikely(IS_ERR(child)))
 959                 GOTO(out_parent, rc = PTR_ERR(child));
 960         if (is_resent) {
 961                 /* Do not take lock for resent case. */
 962                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 963                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 964                          lhc->mlh_reg_lh.cookie);
 965
 966                 res_id = &lock->l_resource->lr_name;
 967                 if (!fid_res_name_eq(mdt_object_fid(child),
 968                                     &lock->l_resource->lr_name)) {
 969                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 970                                                  &lock->l_resource->lr_name),
 971                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 972                                  (unsigned long)res_id->name[0],
 973                                  (unsigned long)res_id->name[1],
 974                                  (unsigned long)res_id->name[2],
 975                                  PFID(mdt_object_fid(parent)));
 976                           CWARN("Although resent, but still not get child lock"
 977                                 "parent:"DFID" child:"DFID"\n",
 978                                 PFID(mdt_object_fid(parent)),
 979                                 PFID(mdt_object_fid(child)));
 980                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 981                           LDLM_LOCK_PUT(lock);
 982                           GOTO(relock, 0);
 983                 }
 984                 LDLM_LOCK_PUT(lock);
 985                 rc = 0;
 986         } else {
 987                 struct md_attr *ma;
 988 relock:
 989                 ma = &info->mti_attr;
 990
 991                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
 992                 mdt_lock_handle_init(lhc);
 993                 mdt_lock_reg_init(lhc, LCK_PR);
 994
 995                 if (mdt_object_exists(child) == 0) {
 996                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 997                                         &child->mot_obj.mo_lu,
 998                                         "Object doesn't exist!\n");
 999                         GOTO(out_child, rc = -ESTALE);
1000                 }
1001
1002                 ma->ma_valid = 0;
1003                 ma->ma_need = MA_INODE;
1004                 rc = mo_attr_get(info->mti_env, next, ma);
1005                 if (unlikely(rc != 0))
1006                         GOTO(out_child, rc);
1007
1008                 /* If the file has not been changed for some time, we return
1009                  * not only a LOOKUP lock, but also an UPDATE lock and this
1010                  * might save us RPC on later STAT. For directories, it also
1011                  * let negative dentry starts working for this dir. */
1012                 if (ma->ma_valid & MA_INODE &&
1013                     ma->ma_attr.la_valid & LA_CTIME &&
1014                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1015                     ma->ma_attr.la_ctime < cfs_time_current_sec())
1016                         child_bits |= MDS_INODELOCK_UPDATE;
1017
1018                 rc = mdt_object_lock(info, child, lhc, child_bits,
1019                                      MDT_CROSS_LOCK);
1020
1021                 if (unlikely(rc != 0))
1022                         GOTO(out_child, rc);
1023         }
1024
1025         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1026         /* Get MA_SOM attributes if update lock is given. */
1027         if (lock &&
1028             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1029             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1030                 ma_need = MA_SOM;
1031
1032         /* finally, we can get attr for child. */
1033         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1034         rc = mdt_getattr_internal(info, child, ma_need);
1035         if (unlikely(rc != 0)) {
1036                 mdt_object_unlock(info, child, lhc, 1);
1037         } else if (lock) {
1038                 /* Debugging code. */
1039                 res_id = &lock->l_resource->lr_name;
1040                 LDLM_DEBUG(lock, "Returning lock to client");
1041                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1042                                          &lock->l_resource->lr_name),
1043                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1044                          (unsigned long)res_id->name[0],
1045                          (unsigned long)res_id->name[1],
1046                          (unsigned long)res_id->name[2],
1047                          PFID(mdt_object_fid(child)));
1048                 mdt_pack_size2body(info, child);
1049         }
1050         if (lock)
1051                 LDLM_LOCK_PUT(lock);
1052
1053         EXIT;
1054 out_child:
1055         mdt_object_put(info->mti_env, child);
1056 out_parent:
1057         mdt_object_unlock(info, parent, lhp, 1);
1058         return rc;
1059 }
1060
1061 /* normal handler: should release the child lock */
1062 static int mdt_getattr_name(struct mdt_thread_info *info)
1063 {
1064         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1065         struct mdt_body        *reqbody;
1066         struct mdt_body        *repbody;
1067         int rc;
1068         ENTRY;
1069
1070         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1071         LASSERT(reqbody != NULL);
1072         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1073         LASSERT(repbody != NULL);
1074
1075         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1076         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1077         repbody->eadatasize = 0;
1078         repbody->aclsize = 0;
1079
1080         rc = mdt_init_ucred(info, reqbody);
1081         if (unlikely(rc))
1082                 GOTO(out_shrink, rc);
1083
1084         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1085         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1086                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1087                 lhc->mlh_reg_lh.cookie = 0;
1088         }
1089         mdt_exit_ucred(info);
1090         EXIT;
1091 out_shrink:
1092         mdt_shrink_reply(info);
1093         return rc;
1094 }
1095
1096 static const struct lu_device_operations mdt_lu_ops;
1097
1098 static int lu_device_is_mdt(struct lu_device *d)
1099 {
1100         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1101 }
1102
1103 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1104                          void *karg, void *uarg);
1105
1106 static int mdt_set_info(struct mdt_thread_info *info)
1107 {
1108         struct ptlrpc_request *req = mdt_info_req(info);
1109         char *key;
1110         void *val;
1111         int keylen, vallen, rc = 0;
1112         ENTRY;
1113
1114         rc = req_capsule_server_pack(info->mti_pill);
1115         if (rc)
1116                 RETURN(rc);
1117
1118         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1119         if (key == NULL) {
1120                 DEBUG_REQ(D_HA, req, "no set_info key");
1121                 RETURN(-EFAULT);
1122         }
1123
1124         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1125                                       RCL_CLIENT);
1126
1127         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1128         if (val == NULL) {
1129                 DEBUG_REQ(D_HA, req, "no set_info val");
1130                 RETURN(-EFAULT);
1131         }
1132
1133         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1134                                       RCL_CLIENT);
1135
1136         /* Swab any part of val you need to here */
1137         if (KEY_IS(KEY_READ_ONLY)) {
1138                 req->rq_status = 0;
1139                 lustre_msg_set_status(req->rq_repmsg, 0);
1140
1141                 cfs_spin_lock(&req->rq_export->exp_lock);
1142                 if (*(__u32 *)val)
1143                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1144                 else
1145                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1146                 cfs_spin_unlock(&req->rq_export->exp_lock);
1147
1148         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1149                 struct changelog_setinfo *cs =
1150                         (struct changelog_setinfo *)val;
1151                 if (vallen != sizeof(*cs)) {
1152                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1153                         RETURN(-EINVAL);
1154                 }
1155                 if (ptlrpc_req_need_swab(req)) {
1156                         __swab64s(&cs->cs_recno);
1157                         __swab32s(&cs->cs_id);
1158                 }
1159
1160                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1161                                    vallen, val, NULL);
1162                 lustre_msg_set_status(req->rq_repmsg, rc);
1163
1164         } else {
1165                 RETURN(-EINVAL);
1166         }
1167         RETURN(0);
1168 }
1169
1170 static int mdt_connect(struct mdt_thread_info *info)
1171 {
1172         int rc;
1173         struct ptlrpc_request *req;
1174
1175         req = mdt_info_req(info);
1176         rc = target_handle_connect(req);
1177         if (rc == 0) {
1178                 LASSERT(req->rq_export != NULL);
1179                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1180                 rc = mdt_init_sec_level(info);
1181                 if (rc == 0)
1182                         rc = mdt_init_idmap(info);
1183                 if (rc != 0)
1184                         obd_disconnect(class_export_get(req->rq_export));
1185         } else {
1186                 rc = err_serious(rc);
1187         }
1188         return rc;
1189 }
1190
1191 static int mdt_disconnect(struct mdt_thread_info *info)
1192 {
1193         int rc;
1194         ENTRY;
1195
1196         rc = target_handle_disconnect(mdt_info_req(info));
1197         if (rc)
1198                 rc = err_serious(rc);
1199         RETURN(rc);
1200 }
1201
1202 static int mdt_sendpage(struct mdt_thread_info *info,
1203                         struct lu_rdpg *rdpg)
1204 {
1205         struct ptlrpc_request   *req = mdt_info_req(info);
1206         struct obd_export       *exp = req->rq_export;
1207         struct ptlrpc_bulk_desc *desc;
1208         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1209         int                      tmpcount;
1210         int                      tmpsize;
1211         int                      timeout;
1212         int                      i;
1213         int                      rc;
1214         ENTRY;
1215
1216         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1217                                     MDS_BULK_PORTAL);
1218         if (desc == NULL)
1219                 RETURN(-ENOMEM);
1220
1221         for (i = 0, tmpcount = rdpg->rp_count;
1222                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1223                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1224                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1225         }
1226
1227         LASSERT(desc->bd_nob == rdpg->rp_count);
1228         rc = sptlrpc_svc_wrap_bulk(req, desc);
1229         if (rc)
1230                 GOTO(free_desc, rc);
1231
1232         rc = ptlrpc_start_bulk_transfer(desc);
1233         if (rc)
1234                 GOTO(free_desc, rc);
1235
1236         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1237                 GOTO(abort_bulk, rc = 0);
1238
1239         do {
1240                 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1241                 if (timeout < 0)
1242                         CERROR("Req deadline already passed %lu (now: %lu)\n",
1243                                req->rq_deadline, cfs_time_current_sec());
1244                 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1245                                             cfs_time_seconds(1), NULL, NULL);
1246                 rc = l_wait_event(desc->bd_waitq,
1247                                   !ptlrpc_server_bulk_active(desc) ||
1248                                   exp->exp_failed ||
1249                                   exp->exp_abort_active_req, lwi);
1250                 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1251         } while ((rc == -ETIMEDOUT) &&
1252                  (req->rq_deadline > cfs_time_current_sec()));
1253
1254         if (rc == 0) {
1255                 if (desc->bd_success &&
1256                     desc->bd_nob_transferred == rdpg->rp_count)
1257                         GOTO(free_desc, rc);
1258
1259                 rc = -ETIMEDOUT;
1260                 if (exp->exp_abort_active_req || exp->exp_failed)
1261                         GOTO(abort_bulk, rc);
1262         }
1263
1264         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1265                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1266                   desc->bd_nob_transferred, rdpg->rp_count,
1267                   exp->exp_client_uuid.uuid,
1268                   exp->exp_connection->c_remote_uuid.uuid);
1269
1270         class_fail_export(exp);
1271
1272         EXIT;
1273 abort_bulk:
1274         ptlrpc_abort_bulk(desc);
1275 free_desc:
1276         ptlrpc_free_bulk(desc);
1277         return rc;
1278 }
1279
1280 #ifdef HAVE_SPLIT_SUPPORT
1281 /*
1282  * Retrieve dir entry from the page and insert it to the slave object, actually,
1283  * this should be in osd layer, but since it will not in the final product, so
1284  * just do it here and do not define more moo api anymore for this.
1285  */
1286 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1287                               int size)
1288 {
1289         struct mdt_object *object = info->mti_object;
1290         struct lu_fid *lf = &info->mti_tmp_fid2;
1291         struct md_attr *ma = &info->mti_attr;
1292         struct lu_dirpage *dp;
1293         struct lu_dirent *ent;
1294         int rc = 0, offset = 0;
1295         ENTRY;
1296
1297         /* Make sure we have at least one entry. */
1298         if (size == 0)
1299                 RETURN(-EINVAL);
1300
1301         /*
1302          * Disable trans for this name insert, since it will include many trans
1303          * for this.
1304          */
1305         info->mti_no_need_trans = 1;
1306         /*
1307          * When write_dir_page, no need update parent's ctime,
1308          * and no permission check for name_insert.
1309          */
1310         ma->ma_attr.la_ctime = 0;
1311         ma->ma_attr.la_valid = LA_MODE;
1312         ma->ma_valid = MA_INODE;
1313
1314         cfs_kmap(page);
1315         dp = page_address(page);
1316         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1317
1318         for (ent = lu_dirent_start(dp); ent != NULL;
1319              ent = lu_dirent_next(ent)) {
1320                 struct lu_name *lname;
1321                 char *name;
1322
1323                 if (le16_to_cpu(ent->lde_namelen) == 0)
1324                         continue;
1325
1326                 fid_le_to_cpu(lf, &ent->lde_fid);
1327                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1328                         ma->ma_attr.la_mode = S_IFDIR;
1329                 else
1330                         ma->ma_attr.la_mode = 0;
1331                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1332                 if (name == NULL)
1333                         GOTO(out, rc = -ENOMEM);
1334
1335                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1336                 lname = mdt_name(info->mti_env, name,
1337                                  le16_to_cpu(ent->lde_namelen));
1338                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1339                 rc = mdo_name_insert(info->mti_env,
1340                                      md_object_next(&object->mot_obj),
1341                                      lname, lf, ma);
1342                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1343                 if (rc) {
1344                         CERROR("Can't insert %*.*s, rc %d\n",
1345                                le16_to_cpu(ent->lde_namelen),
1346                                le16_to_cpu(ent->lde_namelen),
1347                                ent->lde_name, rc);
1348                         GOTO(out, rc);
1349                 }
1350
1351                 offset += lu_dirent_size(ent);
1352                 if (offset >= size)
1353                         break;
1354         }
1355         EXIT;
1356 out:
1357         cfs_kunmap(page);
1358         return rc;
1359 }
1360
1361 static int mdt_bulk_timeout(void *data)
1362 {
1363         ENTRY;
1364
1365         CERROR("mdt bulk transfer timeout \n");
1366
1367         RETURN(1);
1368 }
1369
1370 static int mdt_writepage(struct mdt_thread_info *info)
1371 {
1372         struct ptlrpc_request   *req = mdt_info_req(info);
1373         struct mdt_body         *reqbody;
1374         struct l_wait_info      *lwi;
1375         struct ptlrpc_bulk_desc *desc;
1376         struct page             *page;
1377         int                rc;
1378         ENTRY;
1379
1380
1381         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1382         if (reqbody == NULL)
1383                 RETURN(err_serious(-EFAULT));
1384
1385         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1386         if (desc == NULL)
1387                 RETURN(err_serious(-ENOMEM));
1388
1389         /* allocate the page for the desc */
1390         page = cfs_alloc_page(CFS_ALLOC_STD);
1391         if (page == NULL)
1392                 GOTO(desc_cleanup, rc = -ENOMEM);
1393
1394         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1395                (int)reqbody->size, (int)reqbody->nlink);
1396
1397         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1398                               (int)reqbody->nlink);
1399
1400         rc = sptlrpc_svc_prep_bulk(req, desc);
1401         if (rc != 0)
1402                 GOTO(cleanup_page, rc);
1403         /*
1404          * Check if client was evicted while we were doing i/o before touching
1405          * network.
1406          */
1407         OBD_ALLOC_PTR(lwi);
1408         if (!lwi)
1409                 GOTO(cleanup_page, rc = -ENOMEM);
1410
1411         if (desc->bd_export->exp_failed)
1412                 rc = -ENOTCONN;
1413         else
1414                 rc = ptlrpc_start_bulk_transfer (desc);
1415         if (rc == 0) {
1416                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1417                                             mdt_bulk_timeout, desc);
1418                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1419                                   desc->bd_export->exp_failed, lwi);
1420                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1421                 if (rc == -ETIMEDOUT) {
1422                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1423                         ptlrpc_abort_bulk(desc);
1424                 } else if (desc->bd_export->exp_failed) {
1425                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1426                         rc = -ENOTCONN;
1427                         ptlrpc_abort_bulk(desc);
1428                 } else if (!desc->bd_success ||
1429                            desc->bd_nob_transferred != desc->bd_nob) {
1430                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1431                                   desc->bd_success ?
1432                                   "truncated" : "network error on",
1433                                   desc->bd_nob_transferred, desc->bd_nob);
1434                         /* XXX should this be a different errno? */
1435                         rc = -ETIMEDOUT;
1436                 }
1437         } else {
1438                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1439         }
1440         if (rc)
1441                 GOTO(cleanup_lwi, rc);
1442         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1443
1444 cleanup_lwi:
1445         OBD_FREE_PTR(lwi);
1446 cleanup_page:
1447         cfs_free_page(page);
1448 desc_cleanup:
1449         ptlrpc_free_bulk(desc);
1450         RETURN(rc);
1451 }
1452 #endif
1453
1454 static int mdt_readpage(struct mdt_thread_info *info)
1455 {
1456         struct mdt_object *object = info->mti_object;
1457         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1458         struct mdt_body   *reqbody;
1459         struct mdt_body   *repbody;
1460         int                rc;
1461         int                i;
1462         ENTRY;
1463
1464         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1465                 RETURN(err_serious(-ENOMEM));
1466
1467         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1468         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1469         if (reqbody == NULL || repbody == NULL)
1470                 RETURN(err_serious(-EFAULT));
1471
1472         /*
1473          * prepare @rdpg before calling lower layers and transfer itself. Here
1474          * reqbody->size contains offset of where to start to read and
1475          * reqbody->nlink contains number bytes to read.
1476          */
1477         rdpg->rp_hash = reqbody->size;
1478         if (rdpg->rp_hash != reqbody->size) {
1479                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1480                        rdpg->rp_hash, reqbody->size);
1481                 RETURN(-EFAULT);
1482         }
1483
1484         rdpg->rp_attrs = reqbody->mode;
1485         rdpg->rp_count  = reqbody->nlink;
1486         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1487         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1488         if (rdpg->rp_pages == NULL)
1489                 RETURN(-ENOMEM);
1490
1491         for (i = 0; i < rdpg->rp_npages; ++i) {
1492                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1493                 if (rdpg->rp_pages[i] == NULL)
1494                         GOTO(free_rdpg, rc = -ENOMEM);
1495         }
1496
1497         /* call lower layers to fill allocated pages with directory data */
1498         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1499         if (rc)
1500                 GOTO(free_rdpg, rc);
1501
1502         /* send pages to client */
1503         rc = mdt_sendpage(info, rdpg);
1504
1505         EXIT;
1506 free_rdpg:
1507
1508         for (i = 0; i < rdpg->rp_npages; i++)
1509                 if (rdpg->rp_pages[i] != NULL)
1510                         cfs_free_page(rdpg->rp_pages[i]);
1511         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1512
1513         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1514                 RETURN(0);
1515
1516         return rc;
1517 }
1518
1519 static int mdt_reint_internal(struct mdt_thread_info *info,
1520                               struct mdt_lock_handle *lhc,
1521                               __u32 op)
1522 {
1523         struct req_capsule      *pill = info->mti_pill;
1524         struct mdt_device       *mdt = info->mti_mdt;
1525         struct md_quota         *mq = md_quota(info->mti_env);
1526         struct mdt_body         *repbody;
1527         int                      rc = 0;
1528         ENTRY;
1529
1530         /* pack reply */
1531         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1532                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1533                                      mdt->mdt_max_mdsize);
1534         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1535                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1536                                      mdt->mdt_max_cookiesize);
1537
1538         rc = req_capsule_server_pack(pill);
1539         if (rc != 0) {
1540                 CERROR("Can't pack response, rc %d\n", rc);
1541                 RETURN(err_serious(rc));
1542         }
1543
1544         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1545                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1546                 LASSERT(repbody);
1547                 repbody->eadatasize = 0;
1548                 repbody->aclsize = 0;
1549         }
1550
1551         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1552                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1553
1554         rc = mdt_reint_unpack(info, op);
1555         if (rc != 0) {
1556                 CERROR("Can't unpack reint, rc %d\n", rc);
1557                 GOTO(out_shrink, rc = err_serious(rc));
1558         }
1559
1560         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1561
1562         /* for replay no cookkie / lmm need, because client have this already */
1563         if (info->mti_spec.no_create == 1)  {
1564                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1565                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1566
1567                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1568                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1569                                              0);
1570         }
1571
1572         rc = mdt_init_ucred_reint(info);
1573         if (rc)
1574                 GOTO(out_shrink, rc);
1575
1576         rc = mdt_fix_attr_ucred(info, op);
1577         if (rc != 0)
1578                 GOTO(out_ucred, rc = err_serious(rc));
1579
1580         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1581                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1582                 GOTO(out_ucred, rc);
1583         }
1584         mq->mq_exp = info->mti_exp;
1585         rc = mdt_reint_rec(info, lhc);
1586         EXIT;
1587 out_ucred:
1588         mdt_exit_ucred(info);
1589 out_shrink:
1590         mdt_shrink_reply(info);
1591         return rc;
1592 }
1593
1594 static long mdt_reint_opcode(struct mdt_thread_info *info,
1595                              const struct req_format **fmt)
1596 {
1597         struct mdt_rec_reint *rec;
1598         long opc;
1599
1600         opc = err_serious(-EFAULT);
1601         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1602         if (rec != NULL) {
1603                 opc = rec->rr_opcode;
1604                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1605                 if (opc < REINT_MAX && fmt[opc] != NULL)
1606                         req_capsule_extend(info->mti_pill, fmt[opc]);
1607                 else {
1608                         CERROR("Unsupported opc: %ld\n", opc);
1609                         opc = err_serious(opc);
1610                 }
1611         }
1612         return opc;
1613 }
1614
1615 static int mdt_reint(struct mdt_thread_info *info)
1616 {
1617         long opc;
1618         int  rc;
1619
1620         static const struct req_format *reint_fmts[REINT_MAX] = {
1621                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1622                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1623                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1624                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1625                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1626                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1627                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1628         };
1629
1630         ENTRY;
1631
1632         opc = mdt_reint_opcode(info, reint_fmts);
1633         if (opc >= 0) {
1634                 /*
1635                  * No lock possible here from client to pass it to reint code
1636                  * path.
1637                  */
1638                 rc = mdt_reint_internal(info, NULL, opc);
1639         } else {
1640                 rc = opc;
1641         }
1642
1643         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1644         RETURN(rc);
1645 }
1646
1647 /* this should sync the whole device */
1648 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1649 {
1650         struct dt_device *dt = mdt->mdt_bottom;
1651         int rc;
1652         ENTRY;
1653
1654         rc = dt->dd_ops->dt_sync(env, dt);
1655         RETURN(rc);
1656 }
1657
1658 /* this should sync this object */
1659 static int mdt_object_sync(struct mdt_thread_info *info)
1660 {
1661         struct md_object *next;
1662         int rc;
1663         ENTRY;
1664
1665         if (!mdt_object_exists(info->mti_object)) {
1666                 CWARN("Non existing object  "DFID"!\n",
1667                       PFID(mdt_object_fid(info->mti_object)));
1668                 RETURN(-ESTALE);
1669         }
1670         next = mdt_object_child(info->mti_object);
1671         rc = mo_object_sync(info->mti_env, next);
1672
1673         RETURN(rc);
1674 }
1675
1676 static int mdt_sync(struct mdt_thread_info *info)
1677 {
1678         struct req_capsule *pill = info->mti_pill;
1679         struct mdt_body *body;
1680         int rc;
1681         ENTRY;
1682
1683         /* The fid may be zero, so we req_capsule_set manually */
1684         req_capsule_set(pill, &RQF_MDS_SYNC);
1685
1686         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1687         if (body == NULL)
1688                 RETURN(err_serious(-EINVAL));
1689
1690         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1691                 RETURN(err_serious(-ENOMEM));
1692
1693         if (fid_seq(&body->fid1) == 0) {
1694                 /* sync the whole device */
1695                 rc = req_capsule_server_pack(pill);
1696                 if (rc == 0)
1697                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1698                 else
1699                         rc = err_serious(rc);
1700         } else {
1701                 /* sync an object */
1702                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1703                 if (rc == 0) {
1704                         rc = mdt_object_sync(info);
1705                         if (rc == 0) {
1706                                 struct md_object *next;
1707                                 const struct lu_fid *fid;
1708                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1709
1710                                 next = mdt_object_child(info->mti_object);
1711                                 info->mti_attr.ma_need = MA_INODE;
1712                                 info->mti_attr.ma_valid = 0;
1713                                 rc = mo_attr_get(info->mti_env, next,
1714                                                  &info->mti_attr);
1715                                 if (rc == 0) {
1716                                         body = req_capsule_server_get(pill,
1717                                                                 &RMF_MDT_BODY);
1718                                         fid = mdt_object_fid(info->mti_object);
1719                                         mdt_pack_attr2body(info, body, la, fid);
1720                                 }
1721                         }
1722                 } else
1723                         rc = err_serious(rc);
1724         }
1725         RETURN(rc);
1726 }
1727
1728 #ifdef HAVE_QUOTA_SUPPORT
1729 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1730 {
1731         struct obd_quotactl *oqctl;
1732         struct req_capsule *pill = info->mti_pill;
1733         struct obd_export *exp = info->mti_exp;
1734         struct md_quota *mq = md_quota(info->mti_env);
1735         struct md_device *next = info->mti_mdt->mdt_child;
1736         int rc;
1737         ENTRY;
1738
1739         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1740         if (oqctl == NULL)
1741                 RETURN(-EPROTO);
1742
1743         /* remote client has no permission for quotacheck */
1744         if (unlikely(exp_connect_rmtclient(exp)))
1745                 RETURN(-EPERM);
1746
1747         rc = req_capsule_server_pack(pill);
1748         if (rc)
1749                 RETURN(rc);
1750
1751         mq->mq_exp = exp;
1752         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1753                                                oqctl->qc_type);
1754         RETURN(rc);
1755 }
1756
1757 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1758 {
1759         struct obd_quotactl *oqctl, *repoqc;
1760         struct req_capsule *pill = info->mti_pill;
1761         struct obd_export *exp = info->mti_exp;
1762         struct md_quota *mq = md_quota(info->mti_env);
1763         struct md_device *next = info->mti_mdt->mdt_child;
1764         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1765         int id, rc;
1766         ENTRY;
1767
1768         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1769         if (oqctl == NULL)
1770                 RETURN(-EPROTO);
1771
1772         id = oqctl->qc_id;
1773         if (exp_connect_rmtclient(exp)) {
1774                 struct ptlrpc_request *req = mdt_info_req(info);
1775                 struct mdt_export_data *med = mdt_req2med(req);
1776                 struct lustre_idmap_table *idmap = med->med_idmap;
1777
1778                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1779                              oqctl->qc_cmd != Q_GETINFO))
1780                         RETURN(-EPERM);
1781
1782
1783                 if (oqctl->qc_type == USRQUOTA)
1784                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1785                                                      oqctl->qc_id);
1786                 else if (oqctl->qc_type == GRPQUOTA)
1787                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1788                                                      oqctl->qc_id);
1789                 else
1790                         RETURN(-EINVAL);
1791
1792                 if (id == CFS_IDMAP_NOTFOUND) {
1793                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1794                                oqctl->qc_id);
1795                         RETURN(-EACCES);
1796                 }
1797         }
1798
1799         rc = req_capsule_server_pack(pill);
1800         if (rc)
1801                 RETURN(rc);
1802
1803         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1804         LASSERT(repoqc != NULL);
1805
1806         mq->mq_exp = exp;
1807         switch (oqctl->qc_cmd) {
1808         case Q_QUOTAON:
1809                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1810                 break;
1811         case Q_QUOTAOFF:
1812                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1813                 break;
1814         case Q_SETINFO:
1815                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1816                                       &oqctl->qc_dqinfo);
1817                 break;
1818         case Q_GETINFO:
1819                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1820                                       &oqctl->qc_dqinfo);
1821                 break;
1822         case Q_SETQUOTA:
1823                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1824                                        &oqctl->qc_dqblk);
1825                 break;
1826         case Q_GETQUOTA:
1827                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1828                                        &oqctl->qc_dqblk);
1829                 break;
1830         case Q_GETOINFO:
1831                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1832                                        &oqctl->qc_dqinfo);
1833                 break;
1834         case Q_GETOQUOTA:
1835                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1836                                         &oqctl->qc_dqblk);
1837                 break;
1838         case LUSTRE_Q_INVALIDATE:
1839                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1840                 break;
1841         case LUSTRE_Q_FINVALIDATE:
1842                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1843                 break;
1844         default:
1845                 CERROR("unsupported mdt_quotactl command: %d\n",
1846                        oqctl->qc_cmd);
1847                 RETURN(-EFAULT);
1848         }
1849
1850         *repoqc = *oqctl;
1851         RETURN(rc);
1852 }
1853 #endif
1854
1855
1856 /*
1857  * OBD PING and other handlers.
1858  */
1859 static int mdt_obd_ping(struct mdt_thread_info *info)
1860 {
1861         int rc;
1862         ENTRY;
1863
1864         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1865
1866         rc = target_handle_ping(mdt_info_req(info));
1867         if (rc < 0)
1868                 rc = err_serious(rc);
1869         RETURN(rc);
1870 }
1871
1872 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1873 {
1874         return err_serious(-EOPNOTSUPP);
1875 }
1876
1877 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1878 {
1879         return err_serious(-EOPNOTSUPP);
1880 }
1881
1882
1883 /*
1884  * LLOG handlers.
1885  */
1886
1887 /** clone llog ctxt from child (mdd)
1888  * This allows remote llog (replicator) access.
1889  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1890  * context was originally set up, or we can handle them directly.
1891  * I choose the latter, but that means I need any llog
1892  * contexts set up by child to be accessable by the mdt.  So we clone the
1893  * context into our context list here.
1894  */
1895 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1896                                int idx)
1897 {
1898         struct md_device  *next = mdt->mdt_child;
1899         struct llog_ctxt *ctxt;
1900         int rc;
1901
1902         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1903                 return 0;
1904
1905         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1906         if (rc || ctxt == NULL) {
1907                 CERROR("Can't get mdd ctxt %d\n", rc);
1908                 return rc;
1909         }
1910
1911         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1912         if (rc)
1913                 CERROR("Can't set mdt ctxt %d\n", rc);
1914
1915         return rc;
1916 }
1917
1918 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1919                                  struct mdt_device *mdt, int idx)
1920 {
1921         struct llog_ctxt *ctxt;
1922
1923         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1924         if (ctxt == NULL)
1925                 return 0;
1926         /* Put once for the get we just did, and once for the clone */
1927         llog_ctxt_put(ctxt);
1928         llog_ctxt_put(ctxt);
1929         return 0;
1930 }
1931
1932 static int mdt_llog_create(struct mdt_thread_info *info)
1933 {
1934         int rc;
1935
1936         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1937         rc = llog_origin_handle_create(mdt_info_req(info));
1938         return (rc < 0 ? err_serious(rc) : rc);
1939 }
1940
1941 static int mdt_llog_destroy(struct mdt_thread_info *info)
1942 {
1943         int rc;
1944
1945         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1946         rc = llog_origin_handle_destroy(mdt_info_req(info));
1947         return (rc < 0 ? err_serious(rc) : rc);
1948 }
1949
1950 static int mdt_llog_read_header(struct mdt_thread_info *info)
1951 {
1952         int rc;
1953
1954         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1955         rc = llog_origin_handle_read_header(mdt_info_req(info));
1956         return (rc < 0 ? err_serious(rc) : rc);
1957 }
1958
1959 static int mdt_llog_next_block(struct mdt_thread_info *info)
1960 {
1961         int rc;
1962
1963         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1964         rc = llog_origin_handle_next_block(mdt_info_req(info));
1965         return (rc < 0 ? err_serious(rc) : rc);
1966 }
1967
1968 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1969 {
1970         int rc;
1971
1972         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1973         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1974         return (rc < 0 ? err_serious(rc) : rc);
1975 }
1976
1977
1978 /*
1979  * DLM handlers.
1980  */
1981 static struct ldlm_callback_suite cbs = {
1982         .lcs_completion = ldlm_server_completion_ast,
1983         .lcs_blocking   = ldlm_server_blocking_ast,
1984         .lcs_glimpse    = NULL
1985 };
1986
1987 static int mdt_enqueue(struct mdt_thread_info *info)
1988 {
1989         struct ptlrpc_request *req;
1990         int rc;
1991
1992         /*
1993          * info->mti_dlm_req already contains swapped and (if necessary)
1994          * converted dlm request.
1995          */
1996         LASSERT(info->mti_dlm_req != NULL);
1997
1998         req = mdt_info_req(info);
1999         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2000                                   req, info->mti_dlm_req, &cbs);
2001         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2002         return rc ? err_serious(rc) : req->rq_status;
2003 }
2004
2005 static int mdt_convert(struct mdt_thread_info *info)
2006 {
2007         int rc;
2008         struct ptlrpc_request *req;
2009
2010         LASSERT(info->mti_dlm_req);
2011         req = mdt_info_req(info);
2012         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2013         return rc ? err_serious(rc) : req->rq_status;
2014 }
2015
2016 static int mdt_bl_callback(struct mdt_thread_info *info)
2017 {
2018         CERROR("bl callbacks should not happen on MDS\n");
2019         LBUG();
2020         return err_serious(-EOPNOTSUPP);
2021 }
2022
2023 static int mdt_cp_callback(struct mdt_thread_info *info)
2024 {
2025         CERROR("cp callbacks should not happen on MDS\n");
2026         LBUG();
2027         return err_serious(-EOPNOTSUPP);
2028 }
2029
2030 /*
2031  * sec context handlers
2032  */
2033 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2034 {
2035         int rc;
2036
2037         rc = mdt_handle_idmap(info);
2038
2039         if (unlikely(rc)) {
2040                 struct ptlrpc_request *req = mdt_info_req(info);
2041                 __u32                  opc;
2042
2043                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2044                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2045                         sptlrpc_svc_ctx_invalidate(req);
2046         }
2047
2048         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2049
2050         return rc;
2051 }
2052
2053 static struct mdt_object *mdt_obj(struct lu_object *o)
2054 {
2055         LASSERT(lu_device_is_mdt(o->lo_dev));
2056         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2057 }
2058
2059 struct mdt_object *mdt_object_find(const struct lu_env *env,
2060                                    struct mdt_device *d,
2061                                    const struct lu_fid *f)
2062 {
2063         struct lu_object *o;
2064         struct mdt_object *m;
2065         ENTRY;
2066
2067         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2068         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2069         if (unlikely(IS_ERR(o)))
2070                 m = (struct mdt_object *)o;
2071         else
2072                 m = mdt_obj(o);
2073         RETURN(m);
2074 }
2075
2076 /**
2077  * Asyncronous commit for mdt device.
2078  *
2079  * Pass asynchonous commit call down the MDS stack.
2080  *
2081  * \param env environment
2082  * \param mdt the mdt device
2083  */
2084 static void mdt_device_commit_async(const struct lu_env *env,
2085                                     struct mdt_device *mdt)
2086 {
2087         struct dt_device *dt = mdt->mdt_bottom;
2088         int rc;
2089
2090         rc = dt->dd_ops->dt_commit_async(env, dt);
2091         if (unlikely(rc != 0))
2092                 CWARN("async commit start failed with rc = %d", rc);
2093 }
2094
2095 /**
2096  * Mark the lock as "synchonous".
2097  *
2098  * Mark the lock to deffer transaction commit to the unlock time.
2099  *
2100  * \param lock the lock to mark as "synchonous"
2101  *
2102  * \see mdt_is_lock_sync
2103  * \see mdt_save_lock
2104  */
2105 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2106 {
2107         lock->l_ast_data = (void*)1;
2108 }
2109
2110 /**
2111  * Check whehter the lock "synchonous" or not.
2112  *
2113  * \param lock the lock to check
2114  * \retval 1 the lock is "synchonous"
2115  * \retval 0 the lock isn't "synchronous"
2116  *
2117  * \see mdt_set_lock_sync
2118  * \see mdt_save_lock
2119  */
2120 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2121 {
2122         return lock->l_ast_data != NULL;
2123 }
2124
2125 /**
2126  * Blocking AST for mdt locks.
2127  *
2128  * Starts transaction commit if in case of COS lock conflict or
2129  * deffers such a commit to the mdt_save_lock.
2130  *
2131  * \param lock the lock which blocks a request or cancelling lock
2132  * \param desc unused
2133  * \param data unused
2134  * \param flag indicates whether this cancelling or blocking callback
2135  * \retval 0
2136  * \see ldlm_blocking_ast_nocheck
2137  */
2138 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2139                      void *data, int flag)
2140 {
2141         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2142         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2143         int rc;
2144         ENTRY;
2145
2146         if (flag == LDLM_CB_CANCELING)
2147                 RETURN(0);
2148         lock_res_and_lock(lock);
2149         if (lock->l_blocking_ast != mdt_blocking_ast) {
2150                 unlock_res_and_lock(lock);
2151                 RETURN(0);
2152         }
2153         if (mdt_cos_is_enabled(mdt) &&
2154             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2155             lock->l_blocking_lock != NULL &&
2156             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2157                 mdt_set_lock_sync(lock);
2158         }
2159         rc = ldlm_blocking_ast_nocheck(lock);
2160
2161         /* There is no lock conflict if l_blocking_lock == NULL,
2162          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2163          * when the last reference to a local lock was released */
2164         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2165                 struct lu_env env;
2166
2167                 rc = lu_env_init(&env, LCT_MD_THREAD);
2168                 if (unlikely(rc != 0))
2169                         CWARN("lu_env initialization failed with rc = %d,"
2170                               "cannot start asynchronous commit\n", rc);
2171                 else
2172                         mdt_device_commit_async(&env, mdt);
2173                 lu_env_fini(&env);
2174         }
2175         RETURN(rc);
2176 }
2177
2178 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2179                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2180 {
2181         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2182         ldlm_policy_data_t *policy = &info->mti_policy;
2183         struct ldlm_res_id *res_id = &info->mti_res_id;
2184         int rc;
2185         ENTRY;
2186
2187         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2188         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2189         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2190         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2191
2192         if (mdt_object_exists(o) < 0) {
2193                 if (locality == MDT_CROSS_LOCK) {
2194                         /* cross-ref object fix */
2195                         ibits &= ~MDS_INODELOCK_UPDATE;
2196                         ibits |= MDS_INODELOCK_LOOKUP;
2197                 } else {
2198                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2199                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2200                 }
2201                 /* No PDO lock on remote object */
2202                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2203         }
2204
2205         if (lh->mlh_type == MDT_PDO_LOCK) {
2206                 /* check for exists after object is locked */
2207                 if (mdt_object_exists(o) == 0) {
2208                         /* Non-existent object shouldn't have PDO lock */
2209                         RETURN(-ESTALE);
2210                 } else {
2211                         /* Non-dir object shouldn't have PDO lock */
2212                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2213                 }
2214         }
2215
2216         memset(policy, 0, sizeof(*policy));
2217         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2218
2219         /*
2220          * Take PDO lock on whole directory and build correct @res_id for lock
2221          * on part of directory.
2222          */
2223         if (lh->mlh_pdo_hash != 0) {
2224                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2225                 mdt_lock_pdo_mode(info, o, lh);
2226                 if (lh->mlh_pdo_mode != LCK_NL) {
2227                         /*
2228                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2229                          * is never going to be sent to client and we do not
2230                          * want it slowed down due to possible cancels.
2231                          */
2232                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2233                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2234                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2235                                           &info->mti_exp->exp_handle.h_cookie);
2236                         if (unlikely(rc))
2237                                 RETURN(rc);
2238                 }
2239
2240                 /*
2241                  * Finish res_id initializing by name hash marking part of
2242                  * directory which is taking modification.
2243                  */
2244                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2245         }
2246
2247         policy->l_inodebits.bits = ibits;
2248
2249         /*
2250          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2251          * going to be sent to client. If it is - mdt_intent_policy() path will
2252          * fix it up and turn FL_LOCAL flag off.
2253          */
2254         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2255                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2256                           &info->mti_exp->exp_handle.h_cookie);
2257         if (rc)
2258                 mdt_object_unlock(info, o, lh, 1);
2259         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2260                  lh->mlh_pdo_hash != 0 &&
2261                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2262                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2263         }
2264
2265         RETURN(rc);
2266 }
2267
2268 /**
2269  * Save a lock within request object.
2270  *
2271  * Keep the lock referenced until whether client ACK or transaction
2272  * commit happens or release the lock immediately depending on input
2273  * parameters. If COS is ON, a write lock is converted to COS lock
2274  * before saving.
2275  *
2276  * \param info thead info object
2277  * \param h lock handle
2278  * \param mode lock mode
2279  * \param decref force immediate lock releasing
2280  */
2281 static
2282 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2283                    ldlm_mode_t mode, int decref)
2284 {
2285         ENTRY;
2286
2287         if (lustre_handle_is_used(h)) {
2288                 if (decref || !info->mti_has_trans ||
2289                     !(mode & (LCK_PW | LCK_EX))){
2290                         mdt_fid_unlock(h, mode);
2291                 } else {
2292                         struct mdt_device *mdt = info->mti_mdt;
2293                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2294                         struct ptlrpc_request *req = mdt_info_req(info);
2295                         int no_ack = 0;
2296
2297                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2298                                  h->cookie);
2299                         CDEBUG(D_HA, "request = %p reply state = %p"
2300                                " transno = "LPD64"\n",
2301                                req, req->rq_reply_state, req->rq_transno);
2302                         if (mdt_cos_is_enabled(mdt)) {
2303                                 no_ack = 1;
2304                                 ldlm_lock_downgrade(lock, LCK_COS);
2305                                 mode = LCK_COS;
2306                         }
2307                         ptlrpc_save_lock(req, h, mode, no_ack);
2308                         if (mdt_is_lock_sync(lock)) {
2309                                 CDEBUG(D_HA, "found sync-lock,"
2310                                        " async commit started\n");
2311                                 mdt_device_commit_async(info->mti_env,
2312                                                         mdt);
2313                         }
2314                         LDLM_LOCK_PUT(lock);
2315                 }
2316                 h->cookie = 0ull;
2317         }
2318
2319         EXIT;
2320 }
2321
2322 /**
2323  * Unlock mdt object.
2324  *
2325  * Immeditely release the regular lock and the PDO lock or save the
2326  * lock in reqeuest and keep them referenced until client ACK or
2327  * transaction commit.
2328  *
2329  * \param info thread info object
2330  * \param o mdt object
2331  * \param lh mdt lock handle referencing regular and PDO locks
2332  * \param decref force immediate lock releasing
2333  */
2334 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2335                        struct mdt_lock_handle *lh, int decref)
2336 {
2337         ENTRY;
2338
2339         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2340         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2341
2342         EXIT;
2343 }
2344
2345 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2346                                         const struct lu_fid *f,
2347                                         struct mdt_lock_handle *lh,
2348                                         __u64 ibits)
2349 {
2350         struct mdt_object *o;
2351
2352         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2353         if (!IS_ERR(o)) {
2354                 int rc;
2355
2356                 rc = mdt_object_lock(info, o, lh, ibits,
2357                                      MDT_LOCAL_LOCK);
2358                 if (rc != 0) {
2359                         mdt_object_put(info->mti_env, o);
2360                         o = ERR_PTR(rc);
2361                 }
2362         }
2363         return o;
2364 }
2365
2366 void mdt_object_unlock_put(struct mdt_thread_info * info,
2367                            struct mdt_object * o,
2368                            struct mdt_lock_handle *lh,
2369                            int decref)
2370 {
2371         mdt_object_unlock(info, o, lh, decref);
2372         mdt_object_put(info->mti_env, o);
2373 }
2374
2375 static struct mdt_handler *mdt_handler_find(__u32 opc,
2376                                             struct mdt_opc_slice *supported)
2377 {
2378         struct mdt_opc_slice *s;
2379         struct mdt_handler   *h;
2380
2381         h = NULL;
2382         for (s = supported; s->mos_hs != NULL; s++) {
2383                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2384                         h = s->mos_hs + (opc - s->mos_opc_start);
2385                         if (likely(h->mh_opc != 0))
2386                                 LASSERTF(h->mh_opc == opc,
2387                                          "opcode mismatch %d != %d\n",
2388                                          h->mh_opc, opc);
2389                         else
2390                                 h = NULL; /* unsupported opc */
2391                         break;
2392                 }
2393         }
2394         return h;
2395 }
2396
2397 static int mdt_lock_resname_compat(struct mdt_device *m,
2398                                    struct ldlm_request *req)
2399 {
2400         /* XXX something... later. */
2401         return 0;
2402 }
2403
2404 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2405 {
2406         /* XXX something... later. */
2407         return 0;
2408 }
2409
2410 /*
2411  * Generic code handling requests that have struct mdt_body passed in:
2412  *
2413  *  - extract mdt_body from request and save it in @info, if present;
2414  *
2415  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2416  *  @info;
2417  *
2418  *  - if HABEO_CORPUS flag is set for this request type check whether object
2419  *  actually exists on storage (lu_object_exists()).
2420  *
2421  */
2422 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2423 {
2424         const struct mdt_body    *body;
2425         struct mdt_object        *obj;
2426         const struct lu_env      *env;
2427         struct req_capsule       *pill;
2428         int                       rc;
2429         ENTRY;
2430
2431         env = info->mti_env;
2432         pill = info->mti_pill;
2433
2434         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2435         if (body == NULL)
2436                 RETURN(-EFAULT);
2437
2438         if (!(body->valid & OBD_MD_FLID))
2439                 RETURN(0);
2440
2441         if (!fid_is_sane(&body->fid1)) {
2442                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2443                 RETURN(-EINVAL);
2444         }
2445
2446         /*
2447          * Do not get size or any capa fields before we check that request
2448          * contains capa actually. There are some requests which do not, for
2449          * instance MDS_IS_SUBDIR.
2450          */
2451         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2452             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2453                 mdt_set_capainfo(info, 0, &body->fid1,
2454                                  req_capsule_client_get(pill, &RMF_CAPA1));
2455
2456         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2457         if (!IS_ERR(obj)) {
2458                 if ((flags & HABEO_CORPUS) &&
2459                     !mdt_object_exists(obj)) {
2460                         mdt_object_put(env, obj);
2461                         /* for capability renew ENOENT will be handled in
2462                          * mdt_renew_capa */
2463                         if (body->valid & OBD_MD_FLOSSCAPA)
2464                                 rc = 0;
2465                         else
2466                                 rc = -ENOENT;
2467                 } else {
2468                         info->mti_object = obj;
2469                         rc = 0;
2470                 }
2471         } else
2472                 rc = PTR_ERR(obj);
2473
2474         RETURN(rc);
2475 }
2476
2477 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2478 {
2479         struct req_capsule *pill = info->mti_pill;
2480         int rc;
2481         ENTRY;
2482
2483         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2484                 rc = mdt_body_unpack(info, flags);
2485         else
2486                 rc = 0;
2487
2488         if (rc == 0 && (flags & HABEO_REFERO)) {
2489                 struct mdt_device *mdt = info->mti_mdt;
2490
2491                 /* Pack reply. */
2492
2493                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2494                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2495                                              mdt->mdt_max_mdsize);
2496                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2497                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2498                                              mdt->mdt_max_cookiesize);
2499
2500                 rc = req_capsule_server_pack(pill);
2501         }
2502         RETURN(rc);
2503 }
2504
2505 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2506 {
2507         struct md_device *next = m->mdt_child;
2508
2509         return next->md_ops->mdo_init_capa_ctxt(env, next,
2510                                                 m->mdt_opts.mo_mds_capa,
2511                                                 m->mdt_capa_timeout,
2512                                                 m->mdt_capa_alg,
2513                                                 m->mdt_capa_keys);
2514 }
2515
2516 /*
2517  * Invoke handler for this request opc. Also do necessary preprocessing
2518  * (according to handler ->mh_flags), and post-processing (setting of
2519  * ->last_{xid,committed}).
2520  */
2521 static int mdt_req_handle(struct mdt_thread_info *info,
2522                           struct mdt_handler *h, struct ptlrpc_request *req)
2523 {
2524         int   rc, serious = 0;
2525         __u32 flags;
2526
2527         ENTRY;
2528
2529         LASSERT(h->mh_act != NULL);
2530         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2531         LASSERT(current->journal_info == NULL);
2532
2533         /*
2534          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2535          * to put same checks into handlers like mdt_close(), mdt_reint(),
2536          * etc., without talking to mdt authors first. Checking same thing
2537          * there again is useless and returning 0 error without packing reply
2538          * is buggy! Handlers either pack reply or return error.
2539          *
2540          * We return 0 here and do not send any reply in order to emulate
2541          * network failure. Do not send any reply in case any of NET related
2542          * fail_id has occured.
2543          */
2544         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2545                 RETURN(0);
2546
2547         rc = 0;
2548         flags = h->mh_flags;
2549         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2550
2551         if (h->mh_fmt != NULL) {
2552                 req_capsule_set(info->mti_pill, h->mh_fmt);
2553                 rc = mdt_unpack_req_pack_rep(info, flags);
2554         }
2555
2556         if (rc == 0 && flags & MUTABOR &&
2557             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2558                 /* should it be rq_status? */
2559                 rc = -EROFS;
2560
2561         if (rc == 0 && flags & HABEO_CLAVIS) {
2562                 struct ldlm_request *dlm_req;
2563
2564                 LASSERT(h->mh_fmt != NULL);
2565
2566                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2567                 if (dlm_req != NULL) {
2568                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2569                                         LDLM_IBITS &&
2570                                      dlm_req->lock_desc.l_policy_data.\
2571                                         l_inodebits.bits == 0)) {
2572                                 /*
2573                                  * Lock without inodebits makes no sense and
2574                                  * will oops later in ldlm. If client miss to
2575                                  * set such bits, do not trigger ASSERTION.
2576                                  *
2577                                  * For liblustre flock case, it maybe zero.
2578                                  */
2579                                 rc = -EPROTO;
2580                         } else {
2581                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2582                                         rc = mdt_lock_resname_compat(
2583                                                                 info->mti_mdt,
2584                                                                 dlm_req);
2585                                 info->mti_dlm_req = dlm_req;
2586                         }
2587                 } else {
2588                         rc = -EFAULT;
2589                 }
2590         }
2591
2592         /* capability setting changed via /proc, needs reinitialize ctxt */
2593         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2594                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2595                 info->mti_mdt->mdt_capa_conf = 0;
2596         }
2597
2598         if (likely(rc == 0)) {
2599                 /*
2600                  * Process request, there can be two types of rc:
2601                  * 1) errors with msg unpack/pack, other failures outside the
2602                  * operation itself. This is counted as serious errors;
2603                  * 2) errors during fs operation, should be placed in rq_status
2604                  * only
2605                  */
2606                 rc = h->mh_act(info);
2607                 if (rc == 0 &&
2608                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2609                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2610                                   "pack reply and returned 0 error\n",
2611                                   h->mh_name);
2612                         LBUG();
2613                 }
2614                 serious = is_serious(rc);
2615                 rc = clear_serious(rc);
2616         } else
2617                 serious = 1;
2618
2619         req->rq_status = rc;
2620
2621         /*
2622          * ELDLM_* codes which > 0 should be in rq_status only as well as
2623          * all non-serious errors.
2624          */
2625         if (rc > 0 || !serious)
2626                 rc = 0;
2627
2628         LASSERT(current->journal_info == NULL);
2629
2630         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2631             info->mti_mdt->mdt_opts.mo_compat_resname) {
2632                 struct ldlm_reply *dlmrep;
2633
2634                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2635                 if (dlmrep != NULL)
2636                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2637         }
2638
2639         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2640         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2641                 target_committed_to_req(req);
2642
2643         if (unlikely(req_is_replay(req) &&
2644                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2645                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2646                 LBUG();
2647         }
2648
2649         target_send_reply(req, rc, info->mti_fail_id);
2650         RETURN(0);
2651 }
2652
2653 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2654 {
2655         lh->mlh_type = MDT_NUL_LOCK;
2656         lh->mlh_reg_lh.cookie = 0ull;
2657         lh->mlh_reg_mode = LCK_MINMODE;
2658         lh->mlh_pdo_lh.cookie = 0ull;
2659         lh->mlh_pdo_mode = LCK_MINMODE;
2660 }
2661
2662 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2663 {
2664         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2665         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2666 }
2667
2668 /*
2669  * Initialize fields of struct mdt_thread_info. Other fields are left in
2670  * uninitialized state, because it's too expensive to zero out whole
2671  * mdt_thread_info (> 1K) on each request arrival.
2672  */
2673 static void mdt_thread_info_init(struct ptlrpc_request *req,
2674                                  struct mdt_thread_info *info)
2675 {
2676         int i;
2677         struct md_capainfo *ci;
2678
2679         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2680         info->mti_pill = &req->rq_pill;
2681
2682         /* lock handle */
2683         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2684                 mdt_lock_handle_init(&info->mti_lh[i]);
2685
2686         /* mdt device: it can be NULL while CONNECT */
2687         if (req->rq_export) {
2688                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2689                 info->mti_exp = req->rq_export;
2690         } else
2691                 info->mti_mdt = NULL;
2692         info->mti_env = req->rq_svc_thread->t_env;
2693         ci = md_capainfo(info->mti_env);
2694         memset(ci, 0, sizeof *ci);
2695         if (req->rq_export) {
2696                 if (exp_connect_rmtclient(req->rq_export))
2697                         ci->mc_auth = LC_ID_CONVERT;
2698                 else if (req->rq_export->exp_connect_flags &
2699                          OBD_CONNECT_MDS_CAPA)
2700                         ci->mc_auth = LC_ID_PLAIN;
2701                 else
2702                         ci->mc_auth = LC_ID_NONE;
2703         }
2704
2705         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2706         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2707         info->mti_mos = NULL;
2708
2709         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2710         info->mti_body = NULL;
2711         info->mti_object = NULL;
2712         info->mti_dlm_req = NULL;
2713         info->mti_has_trans = 0;
2714         info->mti_no_need_trans = 0;
2715         info->mti_cross_ref = 0;
2716         info->mti_opdata = 0;
2717
2718         /* To not check for split by default. */
2719         info->mti_spec.sp_ck_split = 0;
2720         info->mti_spec.no_create = 0;
2721 }
2722
2723 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2724 {
2725         int i;
2726
2727         req_capsule_fini(info->mti_pill);
2728         if (info->mti_object != NULL) {
2729                 /*
2730                  * freeing an object may lead to OSD level transaction, do not
2731                  * let it mess with MDT. bz19385.
2732                  */
2733                 info->mti_no_need_trans = 1;
2734                 mdt_object_put(info->mti_env, info->mti_object);
2735                 info->mti_object = NULL;
2736         }
2737         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2738                 mdt_lock_handle_fini(&info->mti_lh[i]);
2739         info->mti_env = NULL;
2740 }
2741
2742 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2743                                        struct obd_device *obd, int *process)
2744 {
2745         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2746         case MDS_CONNECT: /* This will never get here, but for completeness. */
2747         case OST_CONNECT: /* This will never get here, but for completeness. */
2748         case MDS_DISCONNECT:
2749         case OST_DISCONNECT:
2750                *process = 1;
2751                RETURN(0);
2752
2753         case MDS_CLOSE:
2754         case MDS_DONE_WRITING:
2755         case MDS_SYNC: /* used in unmounting */
2756         case OBD_PING:
2757         case MDS_REINT:
2758         case SEQ_QUERY:
2759         case FLD_QUERY:
2760         case LDLM_ENQUEUE:
2761                 *process = target_queue_recovery_request(req, obd);
2762                 RETURN(0);
2763
2764         default:
2765                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2766                 *process = -EAGAIN;
2767                 RETURN(0);
2768         }
2769 }
2770
2771 /*
2772  * Handle recovery. Return:
2773  *        +1: continue request processing;
2774  *       -ve: abort immediately with the given error code;
2775  *         0: send reply with error code in req->rq_status;
2776  */
2777 static int mdt_recovery(struct mdt_thread_info *info)
2778 {
2779         struct ptlrpc_request *req = mdt_info_req(info);
2780         struct obd_device *obd;
2781
2782         ENTRY;
2783
2784         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2785         case MDS_CONNECT:
2786         case SEC_CTX_INIT:
2787         case SEC_CTX_INIT_CONT:
2788         case SEC_CTX_FINI:
2789                 {
2790 #if 0
2791                         int rc;
2792
2793                         rc = mdt_handle_idmap(info);
2794                         if (rc)
2795                                 RETURN(rc);
2796                         else
2797 #endif
2798                                 RETURN(+1);
2799                 }
2800         }
2801
2802         if (unlikely(!class_connected_export(req->rq_export))) {
2803                 CERROR("operation %d on unconnected MDS from %s\n",
2804                        lustre_msg_get_opc(req->rq_reqmsg),
2805                        libcfs_id2str(req->rq_peer));
2806                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2807                  * mds_A will get -ENOTCONN(especially for ping req),
2808                  * which will cause that mds_A deactive timeout, then when
2809                  * mds_A cleanup, the cleanup process will be suspended since
2810                  * deactive timeout is not zero.
2811                  */
2812                 req->rq_status = -ENOTCONN;
2813                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2814                 RETURN(0);
2815         }
2816
2817         /* sanity check: if the xid matches, the request must be marked as a
2818          * resent or replayed */
2819         if (req_xid_is_last(req)) {
2820                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2821                       (MSG_RESENT | MSG_REPLAY))) {
2822                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2823                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2824                                   lustre_msg_get_flags(req->rq_reqmsg));
2825                         LBUG();
2826                         req->rq_status = -ENOTCONN;
2827                         RETURN(-ENOTCONN);
2828                 }
2829         }
2830
2831         /* else: note the opposite is not always true; a RESENT req after a
2832          * failover will usually not match the last_xid, since it was likely
2833          * never committed. A REPLAYed request will almost never match the
2834          * last xid, however it could for a committed, but still retained,
2835          * open. */
2836
2837         obd = req->rq_export->exp_obd;
2838
2839         /* Check for aborted recovery... */
2840         if (unlikely(obd->obd_recovering)) {
2841                 int rc;
2842                 int should_process;
2843                 DEBUG_REQ(D_INFO, req, "Got new replay");
2844                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2845                 if (rc != 0 || !should_process)
2846                         RETURN(rc);
2847                 else if (should_process < 0) {
2848                         req->rq_status = should_process;
2849                         rc = ptlrpc_error(req);
2850                         RETURN(rc);
2851                 }
2852         }
2853         RETURN(+1);
2854 }
2855
2856 static int mdt_msg_check_version(struct lustre_msg *msg)
2857 {
2858         int rc;
2859
2860         switch (lustre_msg_get_opc(msg)) {
2861         case MDS_CONNECT:
2862         case MDS_DISCONNECT:
2863         case OBD_PING:
2864         case SEC_CTX_INIT:
2865         case SEC_CTX_INIT_CONT:
2866         case SEC_CTX_FINI:
2867                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2868                 if (rc)
2869                         CERROR("bad opc %u version %08x, expecting %08x\n",
2870                                lustre_msg_get_opc(msg),
2871                                lustre_msg_get_version(msg),
2872                                LUSTRE_OBD_VERSION);
2873                 break;
2874         case MDS_GETSTATUS:
2875         case MDS_GETATTR:
2876         case MDS_GETATTR_NAME:
2877         case MDS_STATFS:
2878         case MDS_READPAGE:
2879         case MDS_WRITEPAGE:
2880         case MDS_IS_SUBDIR:
2881         case MDS_REINT:
2882         case MDS_CLOSE:
2883         case MDS_DONE_WRITING:
2884         case MDS_PIN:
2885         case MDS_SYNC:
2886         case MDS_GETXATTR:
2887         case MDS_SETXATTR:
2888         case MDS_SET_INFO:
2889         case MDS_GET_INFO:
2890         case MDS_QUOTACHECK:
2891         case MDS_QUOTACTL:
2892         case QUOTA_DQACQ:
2893         case QUOTA_DQREL:
2894         case SEQ_QUERY:
2895         case FLD_QUERY:
2896                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2897                 if (rc)
2898                         CERROR("bad opc %u version %08x, expecting %08x\n",
2899                                lustre_msg_get_opc(msg),
2900                                lustre_msg_get_version(msg),
2901                                LUSTRE_MDS_VERSION);
2902                 break;
2903         case LDLM_ENQUEUE:
2904         case LDLM_CONVERT:
2905         case LDLM_BL_CALLBACK:
2906         case LDLM_CP_CALLBACK:
2907                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2908                 if (rc)
2909                         CERROR("bad opc %u version %08x, expecting %08x\n",
2910                                lustre_msg_get_opc(msg),
2911                                lustre_msg_get_version(msg),
2912                                LUSTRE_DLM_VERSION);
2913                 break;
2914         case OBD_LOG_CANCEL:
2915         case LLOG_ORIGIN_HANDLE_CREATE:
2916         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2917         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2918         case LLOG_ORIGIN_HANDLE_CLOSE:
2919         case LLOG_ORIGIN_HANDLE_DESTROY:
2920         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2921         case LLOG_CATINFO:
2922                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2923                 if (rc)
2924                         CERROR("bad opc %u version %08x, expecting %08x\n",
2925                                lustre_msg_get_opc(msg),
2926                                lustre_msg_get_version(msg),
2927                                LUSTRE_LOG_VERSION);
2928                 break;
2929         default:
2930                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2931                 rc = -ENOTSUPP;
2932         }
2933         return rc;
2934 }
2935
2936 static int mdt_handle0(struct ptlrpc_request *req,
2937                        struct mdt_thread_info *info,
2938                        struct mdt_opc_slice *supported)
2939 {
2940         struct mdt_handler *h;
2941         struct lustre_msg  *msg;
2942         int                 rc;
2943
2944         ENTRY;
2945
2946         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2947                 RETURN(0);
2948
2949         LASSERT(current->journal_info == NULL);
2950
2951         msg = req->rq_reqmsg;
2952         rc = mdt_msg_check_version(msg);
2953         if (likely(rc == 0)) {
2954                 rc = mdt_recovery(info);
2955                 if (likely(rc == +1)) {
2956                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2957                                              supported);
2958                         if (likely(h != NULL)) {
2959                                 rc = mdt_req_handle(info, h, req);
2960                         } else {
2961                                 CERROR("The unsupported opc: 0x%x\n",
2962                                        lustre_msg_get_opc(msg) );
2963                                 req->rq_status = -ENOTSUPP;
2964                                 rc = ptlrpc_error(req);
2965                                 RETURN(rc);
2966                         }
2967                 }
2968         } else
2969                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2970         RETURN(rc);
2971 }
2972
2973 /*
2974  * MDT handler function called by ptlrpc service thread when request comes.
2975  *
2976  * XXX common "target" functionality should be factored into separate module
2977  * shared by mdt, ost and stand-alone services like fld.
2978  */
2979 static int mdt_handle_common(struct ptlrpc_request *req,
2980                              struct mdt_opc_slice *supported)
2981 {
2982         struct lu_env          *env;
2983         struct mdt_thread_info *info;
2984         int                     rc;
2985         ENTRY;
2986
2987         env = req->rq_svc_thread->t_env;
2988         LASSERT(env != NULL);
2989         LASSERT(env->le_ses != NULL);
2990         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2991         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2992         LASSERT(info != NULL);
2993
2994         mdt_thread_info_init(req, info);
2995
2996         rc = mdt_handle0(req, info, supported);
2997
2998         mdt_thread_info_fini(info);
2999         RETURN(rc);
3000 }
3001
3002 /*
3003  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3004  * as well.
3005  */
3006 int mdt_recovery_handle(struct ptlrpc_request *req)
3007 {
3008         int rc;
3009         ENTRY;
3010
3011         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3012         case FLD_QUERY:
3013                 rc = mdt_handle_common(req, mdt_fld_handlers);
3014                 break;
3015         case SEQ_QUERY:
3016                 rc = mdt_handle_common(req, mdt_seq_handlers);
3017                 break;
3018         default:
3019                 rc = mdt_handle_common(req, mdt_regular_handlers);
3020                 break;
3021         }
3022
3023         RETURN(rc);
3024 }
3025
3026 static int mdt_regular_handle(struct ptlrpc_request *req)
3027 {
3028         return mdt_handle_common(req, mdt_regular_handlers);
3029 }
3030
3031 static int mdt_readpage_handle(struct ptlrpc_request *req)
3032 {
3033         return mdt_handle_common(req, mdt_readpage_handlers);
3034 }
3035
3036 static int mdt_xmds_handle(struct ptlrpc_request *req)
3037 {
3038         return mdt_handle_common(req, mdt_xmds_handlers);
3039 }
3040
3041 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3042 {
3043         return mdt_handle_common(req, mdt_seq_handlers);
3044 }
3045
3046 static int mdt_mdss_handle(struct ptlrpc_request *req)
3047 {
3048         return mdt_handle_common(req, mdt_seq_handlers);
3049 }
3050
3051 static int mdt_dtss_handle(struct ptlrpc_request *req)
3052 {
3053         return mdt_handle_common(req, mdt_seq_handlers);
3054 }
3055
3056 static int mdt_fld_handle(struct ptlrpc_request *req)
3057 {
3058         return mdt_handle_common(req, mdt_fld_handlers);
3059 }
3060
3061 enum mdt_it_code {
3062         MDT_IT_OPEN,
3063         MDT_IT_OCREAT,
3064         MDT_IT_CREATE,
3065         MDT_IT_GETATTR,
3066         MDT_IT_READDIR,
3067         MDT_IT_LOOKUP,
3068         MDT_IT_UNLINK,
3069         MDT_IT_TRUNC,
3070         MDT_IT_GETXATTR,
3071         MDT_IT_NR
3072 };
3073
3074 static int mdt_intent_getattr(enum mdt_it_code opcode,
3075                               struct mdt_thread_info *info,
3076                               struct ldlm_lock **,
3077                               int);
3078 static int mdt_intent_reint(enum mdt_it_code opcode,
3079                             struct mdt_thread_info *info,
3080                             struct ldlm_lock **,
3081                             int);
3082
3083 static struct mdt_it_flavor {
3084         const struct req_format *it_fmt;
3085         __u32                    it_flags;
3086         int                    (*it_act)(enum mdt_it_code ,
3087                                          struct mdt_thread_info *,
3088                                          struct ldlm_lock **,
3089                                          int);
3090         long                     it_reint;
3091 } mdt_it_flavor[] = {
3092         [MDT_IT_OPEN]     = {
3093                 .it_fmt   = &RQF_LDLM_INTENT,
3094                 /*.it_flags = HABEO_REFERO,*/
3095                 .it_flags = 0,
3096                 .it_act   = mdt_intent_reint,
3097                 .it_reint = REINT_OPEN
3098         },
3099         [MDT_IT_OCREAT]   = {
3100                 .it_fmt   = &RQF_LDLM_INTENT,
3101                 .it_flags = MUTABOR,
3102                 .it_act   = mdt_intent_reint,
3103                 .it_reint = REINT_OPEN
3104         },
3105         [MDT_IT_CREATE]   = {
3106                 .it_fmt   = &RQF_LDLM_INTENT,
3107                 .it_flags = MUTABOR,
3108                 .it_act   = mdt_intent_reint,
3109                 .it_reint = REINT_CREATE
3110         },
3111         [MDT_IT_GETATTR]  = {
3112                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3113                 .it_flags = HABEO_REFERO,
3114                 .it_act   = mdt_intent_getattr
3115         },
3116         [MDT_IT_READDIR]  = {
3117                 .it_fmt   = NULL,
3118                 .it_flags = 0,
3119                 .it_act   = NULL
3120         },
3121         [MDT_IT_LOOKUP]   = {
3122                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3123                 .it_flags = HABEO_REFERO,
3124                 .it_act   = mdt_intent_getattr
3125         },
3126         [MDT_IT_UNLINK]   = {
3127                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3128                 .it_flags = MUTABOR,
3129                 .it_act   = NULL,
3130                 .it_reint = REINT_UNLINK
3131         },
3132         [MDT_IT_TRUNC]    = {
3133                 .it_fmt   = NULL,
3134                 .it_flags = MUTABOR,
3135                 .it_act   = NULL
3136         },
3137         [MDT_IT_GETXATTR] = {
3138                 .it_fmt   = NULL,
3139                 .it_flags = 0,
3140                 .it_act   = NULL
3141         }
3142 };
3143
3144 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3145                             struct ldlm_lock **lockp,
3146                             struct ldlm_lock *new_lock,
3147                             struct mdt_lock_handle *lh,
3148                             int flags)
3149 {
3150         struct ptlrpc_request  *req = mdt_info_req(info);
3151         struct ldlm_lock       *lock = *lockp;
3152
3153         /*
3154          * Get new lock only for cases when possible resent did not find any
3155          * lock.
3156          */
3157         if (new_lock == NULL)
3158                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3159
3160         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3161                 lh->mlh_reg_lh.cookie = 0;
3162                 RETURN(0);
3163         }
3164
3165         LASSERTF(new_lock != NULL,
3166                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3167
3168         /*
3169          * If we've already given this lock to a client once, then we should
3170          * have no readers or writers.  Otherwise, we should have one reader
3171          * _or_ writer ref (which will be zeroed below) before returning the
3172          * lock to a client.
3173          */
3174         if (new_lock->l_export == req->rq_export) {
3175                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3176         } else {
3177                 LASSERT(new_lock->l_export == NULL);
3178                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3179         }
3180
3181         *lockp = new_lock;
3182
3183         if (new_lock->l_export == req->rq_export) {
3184                 /*
3185                  * Already gave this to the client, which means that we
3186                  * reconstructed a reply.
3187                  */
3188                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3189                         MSG_RESENT);
3190                 lh->mlh_reg_lh.cookie = 0;
3191                 RETURN(ELDLM_LOCK_REPLACED);
3192         }
3193
3194         /*
3195          * Fixup the lock to be given to the client.
3196          */
3197         lock_res_and_lock(new_lock);
3198         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3199          * possible blocking AST. */
3200         while (new_lock->l_readers > 0) {
3201                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3202                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3203                 new_lock->l_readers--;
3204         }
3205         while (new_lock->l_writers > 0) {
3206                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3207                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3208                 new_lock->l_writers--;
3209         }
3210
3211         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3212         new_lock->l_blocking_ast = lock->l_blocking_ast;
3213         new_lock->l_completion_ast = lock->l_completion_ast;
3214         new_lock->l_remote_handle = lock->l_remote_handle;
3215         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3216
3217         unlock_res_and_lock(new_lock);
3218
3219         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3220                      &new_lock->l_remote_handle,
3221                      &new_lock->l_exp_hash);
3222
3223         LDLM_LOCK_RELEASE(new_lock);
3224         lh->mlh_reg_lh.cookie = 0;
3225
3226         RETURN(ELDLM_LOCK_REPLACED);
3227 }
3228
3229 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3230                                     struct ldlm_lock *new_lock,
3231                                     struct ldlm_lock **old_lock,
3232                                     struct mdt_lock_handle *lh)
3233 {
3234         struct ptlrpc_request  *req = mdt_info_req(info);
3235         struct obd_export      *exp = req->rq_export;
3236         struct lustre_handle    remote_hdl;
3237         struct ldlm_request    *dlmreq;
3238         struct ldlm_lock       *lock;
3239
3240         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3241                 return;
3242
3243         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3244         remote_hdl = dlmreq->lock_handle[0];
3245
3246         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3247         if (lock) {
3248                 if (lock != new_lock) {
3249                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3250                         lh->mlh_reg_mode = lock->l_granted_mode;
3251
3252                         LDLM_DEBUG(lock, "Restoring lock cookie");
3253                         DEBUG_REQ(D_DLMTRACE, req,
3254                                   "restoring lock cookie "LPX64,
3255                                   lh->mlh_reg_lh.cookie);
3256                         if (old_lock)
3257                                 *old_lock = LDLM_LOCK_GET(lock);
3258                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3259                         return;
3260                 }
3261
3262                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3263         }
3264
3265         /*
3266          * If the xid matches, then we know this is a resent request, and allow
3267          * it. (It's probably an OPEN, for which we don't send a lock.
3268          */
3269         if (req_xid_is_last(req))
3270                 return;
3271
3272         /*
3273          * This remote handle isn't enqueued, so we never received or processed
3274          * this request.  Clear MSG_RESENT, because it can be handled like any
3275          * normal request now.
3276          */
3277         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3278
3279         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3280                   remote_hdl.cookie);
3281 }
3282
3283 static int mdt_intent_getattr(enum mdt_it_code opcode,
3284                               struct mdt_thread_info *info,
3285                               struct ldlm_lock **lockp,
3286                               int flags)
3287 {
3288         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3289         struct ldlm_lock       *new_lock = NULL;
3290         __u64                   child_bits;
3291         struct ldlm_reply      *ldlm_rep;
3292         struct ptlrpc_request  *req;
3293         struct mdt_body        *reqbody;
3294         struct mdt_body        *repbody;
3295         int                     rc;
3296         ENTRY;
3297
3298         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3299         LASSERT(reqbody);
3300
3301         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3302         LASSERT(repbody);
3303
3304         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3305         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3306         repbody->eadatasize = 0;
3307         repbody->aclsize = 0;
3308
3309         switch (opcode) {
3310         case MDT_IT_LOOKUP:
3311                 child_bits = MDS_INODELOCK_LOOKUP;
3312                 break;
3313         case MDT_IT_GETATTR:
3314                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3315                 break;
3316         default:
3317                 CERROR("Unhandled till now");
3318                 GOTO(out_shrink, rc = -EINVAL);
3319         }
3320
3321         rc = mdt_init_ucred(info, reqbody);
3322         if (rc)
3323                 GOTO(out_shrink, rc);
3324
3325         req = info->mti_pill->rc_req;
3326         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3327         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3328
3329         /* Get lock from request for possible resent case. */
3330         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3331
3332         ldlm_rep->lock_policy_res2 =
3333                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3334
3335         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3336                 ldlm_rep->lock_policy_res2 = 0;
3337         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3338             ldlm_rep->lock_policy_res2) {
3339                 lhc->mlh_reg_lh.cookie = 0ull;
3340                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3341         }
3342
3343         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3344         EXIT;
3345 out_ucred:
3346         mdt_exit_ucred(info);
3347 out_shrink:
3348         mdt_shrink_reply(info);
3349         return rc;
3350 }
3351
3352 static int mdt_intent_reint(enum mdt_it_code opcode,
3353                             struct mdt_thread_info *info,
3354                             struct ldlm_lock **lockp,
3355                             int flags)
3356 {
3357         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3358         struct ldlm_reply      *rep = NULL;
3359         long                    opc;
3360         int                     rc;
3361
3362         static const struct req_format *intent_fmts[REINT_MAX] = {
3363                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3364                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3365         };
3366
3367         ENTRY;
3368
3369         opc = mdt_reint_opcode(info, intent_fmts);
3370         if (opc < 0)
3371                 RETURN(opc);
3372
3373         if (mdt_it_flavor[opcode].it_reint != opc) {
3374                 CERROR("Reint code %ld doesn't match intent: %d\n",
3375                        opc, opcode);
3376                 RETURN(err_serious(-EPROTO));
3377         }
3378
3379         /* Get lock from request for possible resent case. */
3380         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3381
3382         rc = mdt_reint_internal(info, lhc, opc);
3383
3384         /* Check whether the reply has been packed successfully. */
3385         if (mdt_info_req(info)->rq_repmsg != NULL)
3386                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3387         if (rep == NULL)
3388                 RETURN(err_serious(-EFAULT));
3389
3390         /* MDC expects this in any case */
3391         if (rc != 0)
3392                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3393
3394         /* Cross-ref case, the lock should be returned to the client */
3395         if (rc == -EREMOTE) {
3396                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3397                 rep->lock_policy_res2 = 0;
3398                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3399                 RETURN(rc);
3400         }
3401         rep->lock_policy_res2 = clear_serious(rc);
3402
3403         if (rc == -ENOTCONN || rc == -ENODEV ||
3404             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3405                 /*
3406                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3407                  * will be returned by rq_status, and client at ptlrpc layer
3408                  * will detect this, then disconnect, reconnect the import
3409                  * immediately, instead of impacting the following the rpc.
3410                  */
3411                 lhc->mlh_reg_lh.cookie = 0ull;
3412                 RETURN(rc);
3413         } else {
3414                 /*
3415                  * For other cases, the error will be returned by intent.
3416                  * and client will retrieve the result from intent.
3417                  */
3418                  /*
3419                   * FIXME: when open lock is finished, that should be
3420                   * checked here.
3421                   */
3422                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3423                         LASSERTF(rc == 0, "Error occurred but lock handle "
3424                                  "is still in use\n");
3425                         rep->lock_policy_res2 = 0;
3426                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3427                         RETURN(rc);
3428                 } else {
3429                         lhc->mlh_reg_lh.cookie = 0ull;
3430                         RETURN(ELDLM_LOCK_ABORTED);
3431                 }
3432         }
3433 }
3434
3435 static int mdt_intent_code(long itcode)
3436 {
3437         int rc;
3438
3439         switch(itcode) {
3440         case IT_OPEN:
3441                 rc = MDT_IT_OPEN;
3442                 break;
3443         case IT_OPEN|IT_CREAT:
3444                 rc = MDT_IT_OCREAT;
3445                 break;
3446         case IT_CREAT:
3447                 rc = MDT_IT_CREATE;
3448                 break;
3449         case IT_READDIR:
3450                 rc = MDT_IT_READDIR;
3451                 break;
3452         case IT_GETATTR:
3453                 rc = MDT_IT_GETATTR;
3454                 break;
3455         case IT_LOOKUP:
3456                 rc = MDT_IT_LOOKUP;
3457                 break;
3458         case IT_UNLINK:
3459                 rc = MDT_IT_UNLINK;
3460                 break;
3461         case IT_TRUNC:
3462                 rc = MDT_IT_TRUNC;
3463                 break;
3464         case IT_GETXATTR:
3465                 rc = MDT_IT_GETXATTR;
3466                 break;
3467         default:
3468                 CERROR("Unknown intent opcode: %ld\n", itcode);
3469                 rc = -EINVAL;
3470                 break;
3471         }
3472         return rc;
3473 }
3474
3475 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3476                           struct ldlm_lock **lockp, int flags)
3477 {
3478         struct req_capsule   *pill;
3479         struct mdt_it_flavor *flv;
3480         int opc;
3481         int rc;
3482         ENTRY;
3483
3484         opc = mdt_intent_code(itopc);
3485         if (opc < 0)
3486                 RETURN(-EINVAL);
3487
3488         pill = info->mti_pill;
3489         flv  = &mdt_it_flavor[opc];
3490
3491         if (flv->it_fmt != NULL)
3492                 req_capsule_extend(pill, flv->it_fmt);
3493
3494         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3495         if (rc == 0) {
3496                 struct ptlrpc_request *req = mdt_info_req(info);
3497                 if (flv->it_flags & MUTABOR &&
3498                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3499                         RETURN(-EROFS);
3500         }
3501         if (rc == 0 && flv->it_act != NULL) {
3502                 /* execute policy */
3503                 rc = flv->it_act(opc, info, lockp, flags);
3504         } else {
3505                 rc = -EOPNOTSUPP;
3506         }
3507         RETURN(rc);
3508 }
3509
3510 static int mdt_intent_policy(struct ldlm_namespace *ns,
3511                              struct ldlm_lock **lockp, void *req_cookie,
3512                              ldlm_mode_t mode, int flags, void *data)
3513 {
3514         struct mdt_thread_info *info;
3515         struct ptlrpc_request  *req  =  req_cookie;
3516         struct ldlm_intent     *it;
3517         struct req_capsule     *pill;
3518         int rc;
3519
3520         ENTRY;
3521
3522         LASSERT(req != NULL);
3523
3524         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3525                                   &mdt_thread_key);
3526         LASSERT(info != NULL);
3527         pill = info->mti_pill;
3528         LASSERT(pill->rc_req == req);
3529
3530         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3531                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3532                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3533                 if (it != NULL) {
3534                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3535                         if (rc == 0)
3536                                 rc = ELDLM_OK;
3537
3538                         /* Lock without inodebits makes no sense and will oops
3539                          * later in ldlm. Let's check it now to see if we have
3540                          * ibits corrupted somewhere in mdt_intent_opc().
3541                          * The case for client miss to set ibits has been
3542                          * processed by others. */
3543                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3544                                         lr_type == LDLM_IBITS,
3545                                      info->mti_dlm_req->lock_desc.\
3546                                         l_policy_data.l_inodebits.bits != 0));
3547                 } else
3548                         rc = err_serious(-EFAULT);
3549         } else {
3550                 /* No intent was provided */
3551                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3552                 rc = req_capsule_server_pack(pill);
3553                 if (rc)
3554                         rc = err_serious(rc);
3555         }
3556         RETURN(rc);
3557 }
3558
3559 static int mdt_seq_fini(const struct lu_env *env,
3560                         struct mdt_device *m)
3561 {
3562         struct md_site *ms = mdt_md_site(m);
3563         ENTRY;
3564
3565         if (ms != NULL) {
3566                 if (ms->ms_server_seq) {
3567                         seq_server_fini(ms->ms_server_seq, env);
3568                         OBD_FREE_PTR(ms->ms_server_seq);
3569                         ms->ms_server_seq = NULL;
3570         }
3571
3572                 if (ms->ms_control_seq) {
3573                         seq_server_fini(ms->ms_control_seq, env);
3574                         OBD_FREE_PTR(ms->ms_control_seq);
3575                         ms->ms_control_seq = NULL;
3576         }
3577
3578                 if (ms->ms_client_seq) {
3579                         seq_client_fini(ms->ms_client_seq);
3580                         OBD_FREE_PTR(ms->ms_client_seq);
3581                         ms->ms_client_seq = NULL;
3582                 }
3583         }
3584
3585         RETURN(0);
3586 }
3587
3588 static int mdt_seq_init(const struct lu_env *env,
3589                         const char *uuid,
3590                         struct mdt_device *m)
3591 {
3592         struct md_site *ms;
3593         char *prefix;
3594         int rc;
3595         ENTRY;
3596
3597         ms = mdt_md_site(m);
3598
3599         /*
3600          * This is sequence-controller node. Init seq-controller server on local
3601          * MDT.
3602          */
3603         if (ms->ms_node_id == 0) {
3604                 LASSERT(ms->ms_control_seq == NULL);
3605
3606                 OBD_ALLOC_PTR(ms->ms_control_seq);
3607                 if (ms->ms_control_seq == NULL)
3608                         RETURN(-ENOMEM);
3609
3610                 rc = seq_server_init(ms->ms_control_seq,
3611                                      m->mdt_bottom, uuid,
3612                                      LUSTRE_SEQ_CONTROLLER,
3613                                      ms,
3614                                      env);
3615
3616                 if (rc)
3617                         GOTO(out_seq_fini, rc);
3618
3619                 OBD_ALLOC_PTR(ms->ms_client_seq);
3620                 if (ms->ms_client_seq == NULL)
3621                         GOTO(out_seq_fini, rc = -ENOMEM);
3622
3623                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3624                 if (prefix == NULL) {
3625                         OBD_FREE_PTR(ms->ms_client_seq);
3626                         GOTO(out_seq_fini, rc = -ENOMEM);
3627                 }
3628
3629                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3630                          uuid);
3631
3632                 /*
3633                  * Init seq-controller client after seq-controller server is
3634                  * ready. Pass ms->ms_control_seq to it for direct talking.
3635                  */
3636                 rc = seq_client_init(ms->ms_client_seq, NULL,
3637                                      LUSTRE_SEQ_METADATA, prefix,
3638                                      ms->ms_control_seq);
3639                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3640
3641                 if (rc)
3642                         GOTO(out_seq_fini, rc);
3643         }
3644
3645         /* Init seq-server on local MDT */
3646         LASSERT(ms->ms_server_seq == NULL);
3647
3648         OBD_ALLOC_PTR(ms->ms_server_seq);
3649         if (ms->ms_server_seq == NULL)
3650                 GOTO(out_seq_fini, rc = -ENOMEM);
3651
3652         rc = seq_server_init(ms->ms_server_seq,
3653                              m->mdt_bottom, uuid,
3654                              LUSTRE_SEQ_SERVER,
3655                              ms,
3656                              env);
3657         if (rc)
3658                 GOTO(out_seq_fini, rc = -ENOMEM);
3659
3660         /* Assign seq-controller client to local seq-server. */
3661         if (ms->ms_node_id == 0) {
3662                 LASSERT(ms->ms_client_seq != NULL);
3663
3664                 rc = seq_server_set_cli(ms->ms_server_seq,
3665                                         ms->ms_client_seq,
3666                                         env);
3667         }
3668
3669         EXIT;
3670 out_seq_fini:
3671         if (rc)
3672                 mdt_seq_fini(env, m);
3673
3674         return rc;
3675 }
3676 /*
3677  * Init client sequence manager which is used by local MDS to talk to sequence
3678  * controller on remote node.
3679  */
3680 static int mdt_seq_init_cli(const struct lu_env *env,
3681                             struct mdt_device *m,
3682                             struct lustre_cfg *cfg)
3683 {
3684         struct md_site    *ms = mdt_md_site(m);
3685         struct obd_device *mdc;
3686         struct obd_uuid   *uuidp, *mdcuuidp;
3687         char              *uuid_str, *mdc_uuid_str;
3688         int                rc;
3689         int                index;
3690         struct mdt_thread_info *info;
3691         char *p, *index_string = lustre_cfg_string(cfg, 2);
3692         ENTRY;
3693
3694         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3695         uuidp = &info->mti_u.uuid[0];
3696         mdcuuidp = &info->mti_u.uuid[1];
3697
3698         LASSERT(index_string);
3699
3700         index = simple_strtol(index_string, &p, 10);
3701         if (*p) {
3702                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3703                 RETURN(-EINVAL);
3704         }
3705
3706         /* check if this is adding the first MDC and controller is not yet
3707          * initialized. */
3708         if (index != 0 || ms->ms_client_seq)
3709                 RETURN(0);
3710
3711         uuid_str = lustre_cfg_string(cfg, 1);
3712         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3713         obd_str2uuid(uuidp, uuid_str);
3714         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3715
3716         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3717         if (!mdc) {
3718                 CERROR("can't find controller MDC by uuid %s\n",
3719                        uuid_str);
3720                 rc = -ENOENT;
3721         } else if (!mdc->obd_set_up) {
3722                 CERROR("target %s not set up\n", mdc->obd_name);
3723                 rc = -EINVAL;
3724         } else {
3725                 LASSERT(ms->ms_control_exp);
3726                 OBD_ALLOC_PTR(ms->ms_client_seq);
3727                 if (ms->ms_client_seq != NULL) {
3728                         char *prefix;
3729
3730                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3731                         if (!prefix)
3732                                 RETURN(-ENOMEM);
3733
3734                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3735                                  mdc->obd_name);
3736
3737                         rc = seq_client_init(ms->ms_client_seq,
3738                                              ms->ms_control_exp,
3739                                              LUSTRE_SEQ_METADATA,
3740                                              prefix, NULL);
3741                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3742                 } else
3743                         rc = -ENOMEM;
3744
3745                 if (rc)
3746                         RETURN(rc);
3747
3748                 LASSERT(ms->ms_server_seq != NULL);
3749                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3750                                         env);
3751         }
3752
3753         RETURN(rc);
3754 }
3755
3756 static void mdt_seq_fini_cli(struct mdt_device *m)
3757 {
3758         struct md_site *ms;
3759
3760         ENTRY;
3761
3762         ms = mdt_md_site(m);
3763
3764         if (ms != NULL) {
3765                 if (ms->ms_server_seq)
3766                         seq_server_set_cli(ms->ms_server_seq,
3767                                    NULL, NULL);
3768
3769                 if (ms->ms_control_exp) {
3770                         class_export_put(ms->ms_control_exp);
3771                         ms->ms_control_exp = NULL;
3772                 }
3773         }
3774         EXIT;
3775 }
3776
3777 /*
3778  * FLD wrappers
3779  */
3780 static int mdt_fld_fini(const struct lu_env *env,
3781                         struct mdt_device *m)
3782 {
3783         struct md_site *ms = mdt_md_site(m);
3784         ENTRY;
3785
3786         if (ms && ms->ms_server_fld) {
3787                 fld_server_fini(ms->ms_server_fld, env);
3788                 OBD_FREE_PTR(ms->ms_server_fld);
3789                 ms->ms_server_fld = NULL;
3790         }
3791
3792         RETURN(0);
3793 }
3794
3795 static int mdt_fld_init(const struct lu_env *env,
3796                         const char *uuid,
3797                         struct mdt_device *m)
3798 {
3799         struct md_site *ms;
3800         int rc;
3801         ENTRY;
3802
3803         ms = mdt_md_site(m);
3804
3805         OBD_ALLOC_PTR(ms->ms_server_fld);
3806         if (ms->ms_server_fld == NULL)
3807                 RETURN(rc = -ENOMEM);
3808
3809         rc = fld_server_init(ms->ms_server_fld,
3810                              m->mdt_bottom, uuid,
3811                              env, ms->ms_node_id);
3812         if (rc) {
3813                 OBD_FREE_PTR(ms->ms_server_fld);
3814                 ms->ms_server_fld = NULL;
3815                 RETURN(rc);
3816         }
3817
3818         RETURN(0);
3819 }
3820
3821 /* device init/fini methods */
3822 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3823 {
3824         ENTRY;
3825         if (m->mdt_regular_service != NULL) {
3826                 ptlrpc_unregister_service(m->mdt_regular_service);
3827                 m->mdt_regular_service = NULL;
3828         }
3829         if (m->mdt_readpage_service != NULL) {
3830                 ptlrpc_unregister_service(m->mdt_readpage_service);
3831                 m->mdt_readpage_service = NULL;
3832         }
3833         if (m->mdt_xmds_service != NULL) {
3834                 ptlrpc_unregister_service(m->mdt_xmds_service);
3835                 m->mdt_xmds_service = NULL;
3836         }
3837         if (m->mdt_setattr_service != NULL) {
3838                 ptlrpc_unregister_service(m->mdt_setattr_service);
3839                 m->mdt_setattr_service = NULL;
3840         }
3841         if (m->mdt_mdsc_service != NULL) {
3842                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3843                 m->mdt_mdsc_service = NULL;
3844         }
3845         if (m->mdt_mdss_service != NULL) {
3846                 ptlrpc_unregister_service(m->mdt_mdss_service);
3847                 m->mdt_mdss_service = NULL;
3848         }
3849         if (m->mdt_dtss_service != NULL) {
3850                 ptlrpc_unregister_service(m->mdt_dtss_service);
3851                 m->mdt_dtss_service = NULL;
3852         }
3853         if (m->mdt_fld_service != NULL) {
3854                 ptlrpc_unregister_service(m->mdt_fld_service);
3855                 m->mdt_fld_service = NULL;
3856         }
3857         EXIT;
3858 }
3859
3860 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3861 {
3862         int rc;
3863         static struct ptlrpc_service_conf conf;
3864         cfs_proc_dir_entry_t *procfs_entry;
3865         ENTRY;
3866
3867         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3868
3869         conf = (typeof(conf)) {
3870                 .psc_nbufs           = MDS_NBUFS,
3871                 .psc_bufsize         = MDS_BUFSIZE,
3872                 .psc_max_req_size    = MDS_MAXREQSIZE,
3873                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3874                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3875                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3876                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3877                 /*
3878                  * We'd like to have a mechanism to set this on a per-device
3879                  * basis, but alas...
3880                  */
3881                 .psc_min_threads     = mdt_min_threads,
3882                 .psc_max_threads     = mdt_max_threads,
3883                 .psc_ctx_tags        = LCT_MD_THREAD
3884         };
3885
3886         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3887         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3888                            "mdt_ldlm_client", m->mdt_ldlm_client);
3889
3890         m->mdt_regular_service =
3891                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3892                                      procfs_entry, target_print_req,
3893                                      LUSTRE_MDT_NAME);
3894         if (m->mdt_regular_service == NULL)
3895                 RETURN(-ENOMEM);
3896
3897         rc = ptlrpc_start_threads(m->mdt_regular_service);
3898         if (rc)
3899                 GOTO(err_mdt_svc, rc);
3900
3901         /*
3902          * readpage service configuration. Parameters have to be adjusted,
3903          * ideally.
3904          */
3905         conf = (typeof(conf)) {
3906                 .psc_nbufs           = MDS_NBUFS,
3907                 .psc_bufsize         = MDS_BUFSIZE,
3908                 .psc_max_req_size    = MDS_MAXREQSIZE,
3909                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3910                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3911                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3912                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3913                 .psc_min_threads     = mdt_min_threads,
3914                 .psc_max_threads     = mdt_max_threads,
3915                 .psc_ctx_tags        = LCT_MD_THREAD
3916         };
3917         m->mdt_readpage_service =
3918                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3919                                      LUSTRE_MDT_NAME "_readpage",
3920                                      procfs_entry, target_print_req,"mdt_rdpg");
3921
3922         if (m->mdt_readpage_service == NULL) {
3923                 CERROR("failed to start readpage service\n");
3924                 GOTO(err_mdt_svc, rc = -ENOMEM);
3925         }
3926
3927         rc = ptlrpc_start_threads(m->mdt_readpage_service);
3928
3929         /*
3930          * setattr service configuration.
3931          */
3932         conf = (typeof(conf)) {
3933                 .psc_nbufs           = MDS_NBUFS,
3934                 .psc_bufsize         = MDS_BUFSIZE,
3935                 .psc_max_req_size    = MDS_MAXREQSIZE,
3936                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3937                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3938                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3939                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3940                 .psc_min_threads     = mdt_min_threads,
3941                 .psc_max_threads     = mdt_max_threads,
3942                 .psc_ctx_tags        = LCT_MD_THREAD
3943         };
3944
3945         m->mdt_setattr_service =
3946                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3947                                      LUSTRE_MDT_NAME "_setattr",
3948                                      procfs_entry, target_print_req,"mdt_attr");
3949
3950         if (!m->mdt_setattr_service) {
3951                 CERROR("failed to start setattr service\n");
3952                 GOTO(err_mdt_svc, rc = -ENOMEM);
3953         }
3954
3955         rc = ptlrpc_start_threads(m->mdt_setattr_service);
3956         if (rc)
3957                 GOTO(err_mdt_svc, rc);
3958
3959         /*
3960          * sequence controller service configuration
3961          */
3962         conf = (typeof(conf)) {
3963                 .psc_nbufs           = MDS_NBUFS,
3964                 .psc_bufsize         = MDS_BUFSIZE,
3965                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3966                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3967                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3968                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3969                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3970                 .psc_min_threads     = mdt_min_threads,
3971                 .psc_max_threads     = mdt_max_threads,
3972                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3973         };
3974
3975         m->mdt_mdsc_service =
3976                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3977                                      LUSTRE_MDT_NAME"_mdsc",
3978                                      procfs_entry, target_print_req,"mdt_mdsc");
3979         if (!m->mdt_mdsc_service) {
3980                 CERROR("failed to start seq controller service\n");
3981                 GOTO(err_mdt_svc, rc = -ENOMEM);
3982         }
3983
3984         rc = ptlrpc_start_threads(m->mdt_mdsc_service);
3985         if (rc)
3986                 GOTO(err_mdt_svc, rc);
3987
3988         /*
3989          * metadata sequence server service configuration
3990          */
3991         conf = (typeof(conf)) {
3992                 .psc_nbufs           = MDS_NBUFS,
3993                 .psc_bufsize         = MDS_BUFSIZE,
3994                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3995                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3996                 .psc_req_portal      = SEQ_METADATA_PORTAL,
3997                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3998                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3999                 .psc_min_threads     = mdt_min_threads,
4000                 .psc_max_threads     = mdt_max_threads,
4001                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4002         };
4003
4004         m->mdt_mdss_service =
4005                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4006                                      LUSTRE_MDT_NAME"_mdss",
4007                                      procfs_entry, target_print_req,"mdt_mdss");
4008         if (!m->mdt_mdss_service) {
4009                 CERROR("failed to start metadata seq server service\n");
4010                 GOTO(err_mdt_svc, rc = -ENOMEM);
4011         }
4012
4013         rc = ptlrpc_start_threads(m->mdt_mdss_service);
4014         if (rc)
4015                 GOTO(err_mdt_svc, rc);
4016
4017
4018         /*
4019          * Data sequence server service configuration. We want to have really
4020          * cluster-wide sequences space. This is why we start only one sequence
4021          * controller which manages space.
4022          */
4023         conf = (typeof(conf)) {
4024                 .psc_nbufs           = MDS_NBUFS,
4025                 .psc_bufsize         = MDS_BUFSIZE,
4026                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4027                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4028                 .psc_req_portal      = SEQ_DATA_PORTAL,
4029                 .psc_rep_portal      = OSC_REPLY_PORTAL,
4030                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4031                 .psc_min_threads     = mdt_min_threads,
4032                 .psc_max_threads     = mdt_max_threads,
4033                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4034         };
4035
4036         m->mdt_dtss_service =
4037                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4038                                      LUSTRE_MDT_NAME"_dtss",
4039                                      procfs_entry, target_print_req,"mdt_dtss");
4040         if (!m->mdt_dtss_service) {
4041                 CERROR("failed to start data seq server service\n");
4042                 GOTO(err_mdt_svc, rc = -ENOMEM);
4043         }
4044
4045         rc = ptlrpc_start_threads(m->mdt_dtss_service);
4046         if (rc)
4047                 GOTO(err_mdt_svc, rc);
4048
4049         /* FLD service start */
4050         conf = (typeof(conf)) {
4051                 .psc_nbufs           = MDS_NBUFS,
4052                 .psc_bufsize         = MDS_BUFSIZE,
4053                 .psc_max_req_size    = FLD_MAXREQSIZE,
4054                 .psc_max_reply_size  = FLD_MAXREPSIZE,
4055                 .psc_req_portal      = FLD_REQUEST_PORTAL,
4056                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4057                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4058                 .psc_min_threads     = mdt_min_threads,
4059                 .psc_max_threads     = mdt_max_threads,
4060                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
4061         };
4062
4063         m->mdt_fld_service =
4064                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4065                                      LUSTRE_MDT_NAME"_fld",
4066                                      procfs_entry, target_print_req, "mdt_fld");
4067         if (!m->mdt_fld_service) {
4068                 CERROR("failed to start fld service\n");
4069                 GOTO(err_mdt_svc, rc = -ENOMEM);
4070         }
4071
4072         rc = ptlrpc_start_threads(m->mdt_fld_service);
4073         if (rc)
4074                 GOTO(err_mdt_svc, rc);
4075
4076         /*
4077          * mds-mds service configuration. Separate portal is used to allow
4078          * mds-mds requests be not blocked during recovery.
4079          */
4080         conf = (typeof(conf)) {
4081                 .psc_nbufs           = MDS_NBUFS,
4082                 .psc_bufsize         = MDS_BUFSIZE,
4083                 .psc_max_req_size    = MDS_MAXREQSIZE,
4084                 .psc_max_reply_size  = MDS_MAXREPSIZE,
4085                 .psc_req_portal      = MDS_MDS_PORTAL,
4086                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4087                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4088                 .psc_min_threads     = mdt_min_threads,
4089                 .psc_max_threads     = mdt_max_threads,
4090                 .psc_ctx_tags        = LCT_MD_THREAD
4091         };
4092         m->mdt_xmds_service =
4093                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4094                                      LUSTRE_MDT_NAME "_mds",
4095                                      procfs_entry, target_print_req,"mdt_xmds");
4096
4097         if (m->mdt_xmds_service == NULL) {
4098                 CERROR("failed to start xmds service\n");
4099                 GOTO(err_mdt_svc, rc = -ENOMEM);
4100         }
4101
4102         rc = ptlrpc_start_threads(m->mdt_xmds_service);
4103         if (rc)
4104                 GOTO(err_mdt_svc, rc);
4105
4106         EXIT;
4107 err_mdt_svc:
4108         if (rc)
4109                 mdt_stop_ptlrpc_service(m);
4110
4111         return rc;
4112 }
4113
4114 static void mdt_stack_fini(const struct lu_env *env,
4115                            struct mdt_device *m, struct lu_device *top)
4116 {
4117         struct obd_device       *obd = mdt2obd_dev(m);
4118         struct lustre_cfg_bufs  *bufs;
4119         struct lustre_cfg       *lcfg;
4120         struct mdt_thread_info  *info;
4121         char flags[3]="";
4122         ENTRY;
4123
4124         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4125         LASSERT(info != NULL);
4126
4127         bufs = &info->mti_u.bufs;
4128         /* process cleanup, pass mdt obd name to get obd umount flags */
4129         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4130         if (obd->obd_force)
4131                 strcat(flags, "F");
4132         if (obd->obd_fail)
4133                 strcat(flags, "A");
4134         lustre_cfg_bufs_set_string(bufs, 1, flags);
4135         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4136         if (!lcfg) {
4137                 CERROR("Cannot alloc lcfg!\n");
4138                 return;
4139         }
4140
4141         LASSERT(top);
4142         top->ld_ops->ldo_process_config(env, top, lcfg);
4143         lustre_cfg_free(lcfg);
4144
4145         lu_stack_fini(env, top);
4146         m->mdt_child = NULL;
4147         m->mdt_bottom = NULL;
4148 }
4149
4150 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4151                                          const char *typename,
4152                                          struct lu_device *child,
4153                                          struct lustre_cfg *cfg)
4154 {
4155         const char            *dev = lustre_cfg_string(cfg, 0);
4156         struct obd_type       *type;
4157         struct lu_device_type *ldt;
4158         struct lu_device      *d;
4159         int rc;
4160         ENTRY;
4161
4162         /* find the type */
4163         type = class_get_type(typename);
4164         if (!type) {
4165                 CERROR("Unknown type: '%s'\n", typename);
4166                 GOTO(out, rc = -ENODEV);
4167         }
4168
4169         rc = lu_env_refill((struct lu_env *)env);
4170         if (rc != 0) {
4171                 CERROR("Failure to refill session: '%d'\n", rc);
4172                 GOTO(out_type, rc);
4173         }
4174
4175         ldt = type->typ_lu;
4176         if (ldt == NULL) {
4177                 CERROR("type: '%s'\n", typename);
4178                 GOTO(out_type, rc = -EINVAL);
4179         }
4180
4181         ldt->ldt_obd_type = type;
4182         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4183         if (IS_ERR(d)) {
4184                 CERROR("Cannot allocate device: '%s'\n", typename);
4185                 GOTO(out_type, rc = -ENODEV);
4186         }
4187
4188         LASSERT(child->ld_site);
4189         d->ld_site = child->ld_site;
4190
4191         type->typ_refcnt++;
4192         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4193         if (rc) {
4194                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4195                 GOTO(out_alloc, rc);
4196         }
4197         lu_device_get(d);
4198         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4199
4200         RETURN(d);
4201
4202 out_alloc:
4203         ldt->ldt_ops->ldto_device_free(env, d);
4204         type->typ_refcnt--;
4205 out_type:
4206         class_put_type(type);
4207 out:
4208         return ERR_PTR(rc);
4209 }
4210
4211 static int mdt_stack_init(struct lu_env *env,
4212                           struct mdt_device *m,
4213                           struct lustre_cfg *cfg,
4214                           struct lustre_mount_info  *lmi)
4215 {
4216         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4217         struct lu_device  *tmp;
4218         struct md_device  *md;
4219         struct lu_device  *child_lu_dev;
4220         int rc;
4221         ENTRY;
4222
4223         /* init the stack */
4224         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4225         if (IS_ERR(tmp)) {
4226                 RETURN(PTR_ERR(tmp));
4227         }
4228         m->mdt_bottom = lu2dt_dev(tmp);
4229         d = tmp;
4230         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4231         if (IS_ERR(tmp)) {
4232                 GOTO(out, rc = PTR_ERR(tmp));
4233         }
4234         d = tmp;
4235         md = lu2md_dev(d);
4236
4237         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4238         if (IS_ERR(tmp)) {
4239                 GOTO(out, rc = PTR_ERR(tmp));
4240         }
4241         d = tmp;
4242         /*set mdd upcall device*/
4243         md_upcall_dev_set(md, lu2md_dev(d));
4244
4245         md = lu2md_dev(d);
4246         /*set cmm upcall device*/
4247         md_upcall_dev_set(md, &m->mdt_md_dev);
4248
4249         m->mdt_child = lu2md_dev(d);
4250
4251         /* process setup config */
4252         tmp = &m->mdt_md_dev.md_lu_dev;
4253         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4254         if (rc)
4255                 GOTO(out, rc);
4256
4257         /* initialize local objects */
4258         child_lu_dev = &m->mdt_child->md_lu_dev;
4259
4260         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4261                                                &m->mdt_md_dev.md_lu_dev,
4262                                                child_lu_dev);
4263 out:
4264         /* fini from last known good lu_device */
4265         if (rc)
4266                 mdt_stack_fini(env, m, d);
4267
4268         return rc;
4269 }
4270
4271 /**
4272  * setup CONFIG_ORIG context, used to access local config log.
4273  * this may need to be rewrite as part of llog rewrite for lu-api.
4274  */
4275 static int mdt_obd_llog_setup(struct obd_device *obd,
4276                               struct lustre_sb_info *lsi)
4277 {
4278         int     rc;
4279
4280         LASSERT(obd->obd_fsops == NULL);
4281
4282         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4283         if (IS_ERR(obd->obd_fsops))
4284                 return PTR_ERR(obd->obd_fsops);
4285
4286         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4287         if (rc) {
4288                 fsfilt_put_ops(obd->obd_fsops);
4289                 return rc;
4290         }
4291
4292         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4293         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4294         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4295         obd->obd_lvfs_ctxt.fs = get_ds();
4296
4297         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4298                         0, NULL, &llog_lvfs_ops);
4299         if (rc) {
4300                 CERROR("llog_setup() failed: %d\n", rc);
4301                 fsfilt_put_ops(obd->obd_fsops);
4302         }
4303
4304         return rc;
4305 }
4306
4307 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4308 {
4309         struct llog_ctxt *ctxt;
4310
4311         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4312         if (ctxt)
4313                 llog_cleanup(ctxt);
4314
4315         if (obd->obd_fsops) {
4316                 fsfilt_put_ops(obd->obd_fsops);
4317                 obd->obd_fsops = NULL;
4318         }
4319 }
4320
4321 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4322 {
4323         struct md_device  *next = m->mdt_child;
4324         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4325         struct lu_site    *ls   = d->ld_site;
4326         struct obd_device *obd = mdt2obd_dev(m);
4327         ENTRY;
4328
4329         target_recovery_fini(obd);
4330
4331         ping_evictor_stop();
4332
4333         mdt_stop_ptlrpc_service(m);
4334         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4335         mdt_obd_llog_cleanup(obd);
4336         obd_exports_barrier(obd);
4337         obd_zombie_barrier();
4338 #ifdef HAVE_QUOTA_SUPPORT
4339         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4340 #endif
4341         lut_fini(env, &m->mdt_lut);
4342         mdt_fs_cleanup(env, m);
4343         upcall_cache_cleanup(m->mdt_identity_cache);
4344         m->mdt_identity_cache = NULL;
4345
4346         if (m->mdt_namespace != NULL) {
4347                 ldlm_namespace_free(m->mdt_namespace, NULL,
4348                                     d->ld_obd->obd_force);
4349                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4350         }
4351
4352         cfs_free_nidlist(&m->mdt_nosquash_nids);
4353         if (m->mdt_nosquash_str) {
4354                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4355                 m->mdt_nosquash_str = NULL;
4356                 m->mdt_nosquash_strlen = 0;
4357         }
4358
4359         mdt_seq_fini(env, m);
4360         mdt_seq_fini_cli(m);
4361         mdt_fld_fini(env, m);
4362         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4363
4364         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4365         cfs_timer_disarm(&m->mdt_ck_timer);
4366         mdt_ck_thread_stop(m);
4367
4368         /*
4369          * Finish the stack
4370          */
4371         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4372
4373         lprocfs_free_per_client_stats(obd);
4374         lprocfs_free_obd_stats(obd);
4375         mdt_procfs_fini(m);
4376
4377         if (ls) {
4378                 struct md_site *mite;
4379
4380                 lu_site_fini(ls);
4381                 mite = lu_site2md(ls);
4382                 OBD_FREE_PTR(mite);
4383                 d->ld_site = NULL;
4384         }
4385         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4386
4387         EXIT;
4388 }
4389
4390 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4391 {
4392         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4393         struct sptlrpc_rule_set  tmp_rset;
4394         int                      rc;
4395
4396         sptlrpc_rule_set_init(&tmp_rset);
4397         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4398         if (rc) {
4399                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4400                        obd->obd_name, rc);
4401                 return rc;
4402         }
4403
4404         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4405
4406         cfs_write_lock(&m->mdt_sptlrpc_lock);
4407         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4408         m->mdt_sptlrpc_rset = tmp_rset;
4409         cfs_write_unlock(&m->mdt_sptlrpc_lock);
4410
4411         return 0;
4412 }
4413
4414 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4415 {
4416         char *p = options;
4417
4418         m->mdt_opts.mo_mds_capa = 1;
4419         m->mdt_opts.mo_oss_capa = 1;
4420 #ifdef CONFIG_FS_POSIX_ACL
4421         /* ACLs should be enabled by default (b=13829) */
4422         m->mdt_opts.mo_acl = 1;
4423         LCONSOLE_INFO("Enabling ACL\n");
4424 #else
4425         m->mdt_opts.mo_acl = 0;
4426         LCONSOLE_INFO("Disabling ACL\n");
4427 #endif
4428
4429         if (!options)
4430                 return;
4431
4432         while (*options) {
4433                 int len;
4434
4435                 while (*p && *p != ',')
4436                         p++;
4437
4438                 len = p - options;
4439                 if ((len == sizeof("user_xattr") - 1) &&
4440                     (memcmp(options, "user_xattr", len) == 0)) {
4441                         m->mdt_opts.mo_user_xattr = 1;
4442                         LCONSOLE_INFO("Enabling user_xattr\n");
4443                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4444                            (memcmp(options, "nouser_xattr", len) == 0)) {
4445                         m->mdt_opts.mo_user_xattr = 0;
4446                         LCONSOLE_INFO("Disabling user_xattr\n");
4447                 } else if ((len == sizeof("noacl") - 1) &&
4448                            (memcmp(options, "noacl", len) == 0)) {
4449                         m->mdt_opts.mo_acl = 0;
4450                         LCONSOLE_INFO("Disabling ACL\n");
4451                 }
4452
4453                 if (!*p)
4454                         break;
4455
4456                 options = ++p;
4457         }
4458 }
4459
4460 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4461
4462 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4463                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4464 {
4465         struct mdt_thread_info    *info;
4466         struct obd_device         *obd;
4467         const char                *dev = lustre_cfg_string(cfg, 0);
4468         const char                *num = lustre_cfg_string(cfg, 2);
4469         struct lustre_mount_info  *lmi = NULL;
4470         struct lustre_sb_info     *lsi;
4471         struct lustre_disk_data   *ldd;
4472         struct lu_site            *s;
4473         struct md_site            *mite;
4474         const char                *identity_upcall = "NONE";
4475 #ifdef HAVE_QUOTA_SUPPORT
4476         struct md_device          *next;
4477 #endif
4478         int                        rc;
4479         int                        node_id;
4480         ENTRY;
4481
4482         md_device_init(&m->mdt_md_dev, ldt);
4483         /*
4484          * Environment (env) might be missing mdt_thread_key values at that
4485          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4486          * mode.
4487          *
4488          * Usually device allocation path doesn't use module key values, but
4489          * mdt has to do a lot of work here, so allocate key value.
4490          */
4491         rc = lu_env_refill((struct lu_env *)env);
4492         if (rc != 0)
4493                 RETURN(rc);
4494
4495         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4496         LASSERT(info != NULL);
4497
4498         obd = class_name2obd(dev);
4499         LASSERT(obd != NULL);
4500
4501         m->mdt_max_mdsize = MAX_MD_SIZE;
4502         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4503         m->mdt_som_conf = 0;
4504
4505         m->mdt_opts.mo_user_xattr = 0;
4506         m->mdt_opts.mo_acl = 0;
4507         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4508         lmi = server_get_mount_2(dev);
4509         if (lmi == NULL) {
4510                 CERROR("Cannot get mount info for %s!\n", dev);
4511                 RETURN(-EFAULT);
4512         } else {
4513                 lsi = s2lsi(lmi->lmi_sb);
4514                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4515                 /* CMD is supported only in IAM mode */
4516                 ldd = lsi->lsi_ldd;
4517                 LASSERT(num);
4518                 node_id = simple_strtol(num, NULL, 10);
4519                 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4520                         CERROR("CMD Operation not allowed in IOP mode\n");
4521                         GOTO(err_lmi, rc = -EINVAL);
4522                 }
4523                 /* Read recovery timeouts */
4524                 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft)
4525                         obd->obd_recovery_timeout =
4526                                 lsi->lsi_lmd->lmd_recovery_time_soft;
4527
4528                 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard)
4529                         obd->obd_recovery_time_hard =
4530                                 lsi->lsi_lmd->lmd_recovery_time_hard;
4531         }
4532
4533         cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4534         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4535
4536         cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4537         m->mdt_opts.mo_compat_resname = 0;
4538         m->mdt_capa_timeout = CAPA_TIMEOUT;
4539         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4540         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4541         m->mdt_squash_uid = 0;
4542         m->mdt_squash_gid = 0;
4543         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4544         m->mdt_nosquash_str = NULL;
4545         m->mdt_nosquash_strlen = 0;
4546         cfs_init_rwsem(&m->mdt_squash_sem);
4547
4548         OBD_ALLOC_PTR(mite);
4549         if (mite == NULL)
4550                 GOTO(err_lmi, rc = -ENOMEM);
4551
4552         s = &mite->ms_lu;
4553
4554         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4555         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4556         /* set this lu_device to obd, because error handling need it */
4557         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4558
4559         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4560         if (rc) {
4561                 CERROR("Can't init lu_site, rc %d\n", rc);
4562                 GOTO(err_free_site, rc);
4563         }
4564
4565         rc = mdt_procfs_init(m, dev);
4566         if (rc) {
4567                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4568                 GOTO(err_fini_proc, rc);
4569         }
4570
4571         /* set server index */
4572         lu_site2md(s)->ms_node_id = node_id;
4573
4574         /* failover is the default
4575          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4576          * assumed whose ms_node_id == 0 XXX
4577          * */
4578         obd->obd_replayable = 1;
4579         /* No connection accepted until configurations will finish */
4580         obd->obd_no_conn = 1;
4581
4582         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4583                 char *str = lustre_cfg_string(cfg, 4);
4584                 if (strchr(str, 'n')) {
4585                         CWARN("%s: recovery disabled\n", obd->obd_name);
4586                         obd->obd_replayable = 0;
4587                 }
4588         }
4589
4590         /* init the stack */
4591         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4592         if (rc) {
4593                 CERROR("Can't init device stack, rc %d\n", rc);
4594                 GOTO(err_fini_proc, rc);
4595         }
4596
4597         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4598         if (rc)
4599                 GOTO(err_fini_stack, rc);
4600
4601         rc = mdt_fld_init(env, obd->obd_name, m);
4602         if (rc)
4603                 GOTO(err_lut, rc);
4604
4605         rc = mdt_seq_init(env, obd->obd_name, m);
4606         if (rc)
4607                 GOTO(err_fini_fld, rc);
4608
4609         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4610                  LUSTRE_MDT_NAME"-%p", m);
4611         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4612                                               LDLM_NAMESPACE_SERVER,
4613                                               LDLM_NAMESPACE_GREEDY,
4614                                               LDLM_NS_TYPE_MDT);
4615         if (m->mdt_namespace == NULL)
4616                 GOTO(err_fini_seq, rc = -ENOMEM);
4617
4618         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4619         /* set obd_namespace for compatibility with old code */
4620         obd->obd_namespace = m->mdt_namespace;
4621
4622         /* XXX: to support suppgid for ACL, we enable identity_upcall
4623          * by default, otherwise, maybe got unexpected -EACCESS. */
4624         if (m->mdt_opts.mo_acl)
4625                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4626
4627         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4628                                                   &mdt_identity_upcall_cache_ops);
4629         if (IS_ERR(m->mdt_identity_cache)) {
4630                 rc = PTR_ERR(m->mdt_identity_cache);
4631                 m->mdt_identity_cache = NULL;
4632                 GOTO(err_free_ns, rc);
4633         }
4634
4635         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4636
4637         rc = mdt_ck_thread_start(m);
4638         if (rc)
4639                 GOTO(err_free_ns, rc);
4640
4641         rc = mdt_fs_setup(env, m, obd, lsi);
4642         if (rc)
4643                 GOTO(err_capa, rc);
4644
4645         rc = mdt_obd_llog_setup(obd, lsi);
4646         if (rc)
4647                 GOTO(err_fs_cleanup, rc);
4648
4649         rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4650         if (rc)
4651                 GOTO(err_llog_cleanup, rc);
4652
4653         mdt_adapt_sptlrpc_conf(obd, 1);
4654
4655 #ifdef HAVE_QUOTA_SUPPORT
4656         next = m->mdt_child;
4657         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4658         if (rc)
4659                 GOTO(err_llog_cleanup, rc);
4660 #endif
4661
4662         server_put_mount_2(dev, lmi->lmi_mnt);
4663         lmi = NULL;
4664
4665         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4666
4667         rc = mdt_start_ptlrpc_service(m);
4668         if (rc)
4669                 GOTO(err_recovery, rc);
4670
4671         ping_evictor_start();
4672
4673         rc = lu_site_init_finish(s);
4674         if (rc)
4675                 GOTO(err_stop_service, rc);
4676
4677         if (obd->obd_recovering == 0)
4678                 mdt_postrecov(env, m);
4679
4680         mdt_init_capa_ctxt(env, m);
4681
4682         /* Reduce the initial timeout on an MDS because it doesn't need such
4683          * a long timeout as an OST does. Adaptive timeouts will adjust this
4684          * value appropriately. */
4685         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4686                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4687
4688         RETURN(0);
4689
4690 err_stop_service:
4691         ping_evictor_stop();
4692         mdt_stop_ptlrpc_service(m);
4693 err_recovery:
4694         target_recovery_fini(obd);
4695 #ifdef HAVE_QUOTA_SUPPORT
4696         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4697 #endif
4698 err_llog_cleanup:
4699         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4700         mdt_obd_llog_cleanup(obd);
4701 err_fs_cleanup:
4702         mdt_fs_cleanup(env, m);
4703 err_capa:
4704         cfs_timer_disarm(&m->mdt_ck_timer);
4705         mdt_ck_thread_stop(m);
4706 err_free_ns:
4707         upcall_cache_cleanup(m->mdt_identity_cache);
4708         m->mdt_identity_cache = NULL;
4709         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4710         obd->obd_namespace = m->mdt_namespace = NULL;
4711 err_fini_seq:
4712         mdt_seq_fini(env, m);
4713 err_fini_fld:
4714         mdt_fld_fini(env, m);
4715 err_lut:
4716         lut_fini(env, &m->mdt_lut);
4717 err_fini_stack:
4718         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4719 err_fini_proc:
4720         mdt_procfs_fini(m);
4721         lu_site_fini(s);
4722 err_free_site:
4723         OBD_FREE_PTR(mite);
4724 err_lmi:
4725         if (lmi)
4726                 server_put_mount_2(dev, lmi->lmi_mnt);
4727         return (rc);
4728 }
4729
4730 /* used by MGS to process specific configurations */
4731 static int mdt_process_config(const struct lu_env *env,
4732                               struct lu_device *d, struct lustre_cfg *cfg)
4733 {
4734         struct mdt_device *m = mdt_dev(d);
4735         struct md_device *md_next = m->mdt_child;
4736         struct lu_device *next = md2lu_dev(md_next);
4737         int rc = 0;
4738         ENTRY;
4739
4740         switch (cfg->lcfg_command) {
4741         case LCFG_PARAM: {
4742                 struct lprocfs_static_vars lvars;
4743                 struct obd_device *obd = d->ld_obd;
4744
4745                 /*
4746                  * For interoperability between 1.8 and 2.0,
4747                  * skip old "mdt.group_upcall" param.
4748                  */
4749                 {
4750                         char *param = lustre_cfg_string(cfg, 1);
4751                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4752                                 CWARN("For 1.8 interoperability, skip this"
4753                                        " mdt.group_upcall. It is obsolete\n");
4754                                 break;
4755                         }
4756                 }
4757
4758                 lprocfs_mdt_init_vars(&lvars);
4759                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4760                                               cfg, obd);
4761                 if (rc > 0 || rc == -ENOSYS)
4762                         /* we don't understand; pass it on */
4763                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4764                 break;
4765         }
4766         case LCFG_ADD_MDC:
4767                 /*
4768                  * Add mdc hook to get first MDT uuid and connect it to
4769                  * ls->controller to use for seq manager.
4770                  */
4771                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4772                 if (rc)
4773                         CERROR("Can't add mdc, rc %d\n", rc);
4774                 else
4775                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4776                 break;
4777         default:
4778                 /* others are passed further */
4779                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4780                 break;
4781         }
4782         RETURN(rc);
4783 }
4784
4785 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4786                                           const struct lu_object_header *hdr,
4787                                           struct lu_device *d)
4788 {
4789         struct mdt_object *mo;
4790
4791         ENTRY;
4792
4793         OBD_ALLOC_PTR(mo);
4794         if (mo != NULL) {
4795                 struct lu_object *o;
4796                 struct lu_object_header *h;
4797
4798                 o = &mo->mot_obj.mo_lu;
4799                 h = &mo->mot_header;
4800                 lu_object_header_init(h);
4801                 lu_object_init(o, h, d);
4802                 lu_object_add_top(h, o);
4803                 o->lo_ops = &mdt_obj_ops;
4804                 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4805                 RETURN(o);
4806         } else
4807                 RETURN(NULL);
4808 }
4809
4810 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4811                            const struct lu_object_conf *unused)
4812 {
4813         struct mdt_device *d = mdt_dev(o->lo_dev);
4814         struct lu_device  *under;
4815         struct lu_object  *below;
4816         int                rc = 0;
4817         ENTRY;
4818
4819         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4820                PFID(lu_object_fid(o)));
4821
4822         under = &d->mdt_child->md_lu_dev;
4823         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4824         if (below != NULL) {
4825                 lu_object_add(o, below);
4826         } else
4827                 rc = -ENOMEM;
4828
4829         RETURN(rc);
4830 }
4831
4832 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4833 {
4834         struct mdt_object *mo = mdt_obj(o);
4835         struct lu_object_header *h;
4836         ENTRY;
4837
4838         h = o->lo_header;
4839         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4840                PFID(lu_object_fid(o)));
4841
4842         lu_object_fini(o);
4843         lu_object_header_fini(h);
4844         OBD_FREE_PTR(mo);
4845         EXIT;
4846 }
4847
4848 static int mdt_object_print(const struct lu_env *env, void *cookie,
4849                             lu_printer_t p, const struct lu_object *o)
4850 {
4851         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4852         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4853                     "flags="LPX64", epochcount=%d, writecount=%d)",
4854                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4855                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4856 }
4857
4858 static const struct lu_device_operations mdt_lu_ops = {
4859         .ldo_object_alloc   = mdt_object_alloc,
4860         .ldo_process_config = mdt_process_config,
4861 };
4862
4863 static const struct lu_object_operations mdt_obj_ops = {
4864         .loo_object_init    = mdt_object_init,
4865         .loo_object_free    = mdt_object_free,
4866         .loo_object_print   = mdt_object_print
4867 };
4868
4869 static int mdt_obd_set_info_async(struct obd_export *exp,
4870                                   __u32 keylen, void *key,
4871                                   __u32 vallen, void *val,
4872                                   struct ptlrpc_request_set *set)
4873 {
4874         struct obd_device     *obd = exp->exp_obd;
4875         int                    rc;
4876         ENTRY;
4877
4878         LASSERT(obd);
4879
4880         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4881                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4882                 RETURN(rc);
4883         }
4884
4885         RETURN(0);
4886 }
4887
4888 /* mds_connect_internal */
4889 static int mdt_connect_internal(struct obd_export *exp,
4890                                 struct mdt_device *mdt,
4891                                 struct obd_connect_data *data)
4892 {
4893         if (data != NULL) {
4894                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4895                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4896
4897                 /* If no known bits (which should not happen, probably,
4898                    as everybody should support LOOKUP and UPDATE bits at least)
4899                    revert to compat mode with plain locks. */
4900                 if (!data->ocd_ibits_known &&
4901                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4902                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4903
4904                 if (!mdt->mdt_opts.mo_acl)
4905                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4906
4907                 if (!mdt->mdt_opts.mo_user_xattr)
4908                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4909
4910                 if (!mdt->mdt_som_conf)
4911                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4912
4913                 cfs_spin_lock(&exp->exp_lock);
4914                 exp->exp_connect_flags = data->ocd_connect_flags;
4915                 cfs_spin_unlock(&exp->exp_lock);
4916                 data->ocd_version = LUSTRE_VERSION_CODE;
4917                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4918         }
4919
4920 #if 0
4921         if (mdt->mdt_opts.mo_acl &&
4922             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4923                 CWARN("%s: MDS requires ACL support but client does not\n",
4924                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4925                 return -EBADE;
4926         }
4927 #endif
4928
4929         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4930                 CWARN("%s: MDS requires FID support, but client not\n",
4931                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4932                 return -EBADE;
4933         }
4934
4935         if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4936             !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4937                 CWARN("%s: MDS has SOM enabled, but client does not support "
4938                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4939                 return -EBADE;
4940         }
4941
4942         return 0;
4943 }
4944
4945 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4946                                      struct obd_export *exp,
4947                                      struct ptlrpc_request *req)
4948 {
4949         struct sptlrpc_flavor   flvr;
4950         int                     rc = 0;
4951
4952         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4953                 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4954                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4955                                              req->rq_sp_from,
4956                                              req->rq_peer.nid,
4957                                              &flvr);
4958                 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4959
4960                 cfs_spin_lock(&exp->exp_lock);
4961
4962                 exp->exp_sp_peer = req->rq_sp_from;
4963                 exp->exp_flvr = flvr;
4964
4965                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4966                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4967                         CERROR("unauthorized rpc flavor %x from %s, "
4968                                "expect %x\n", req->rq_flvr.sf_rpc,
4969                                libcfs_nid2str(req->rq_peer.nid),
4970                                exp->exp_flvr.sf_rpc);
4971                         rc = -EACCES;
4972                 }
4973
4974                 cfs_spin_unlock(&exp->exp_lock);
4975         } else {
4976                 if (exp->exp_sp_peer != req->rq_sp_from) {
4977                         CERROR("RPC source %s doesn't match %s\n",
4978                                sptlrpc_part2name(req->rq_sp_from),
4979                                sptlrpc_part2name(exp->exp_sp_peer));
4980                         rc = -EACCES;
4981                 } else {
4982                         rc = sptlrpc_target_export_check(exp, req);
4983                 }
4984         }
4985
4986         return rc;
4987 }
4988
4989 /* mds_connect copy */
4990 static int mdt_obd_connect(const struct lu_env *env,
4991                            struct obd_export **exp, struct obd_device *obd,
4992                            struct obd_uuid *cluuid,
4993                            struct obd_connect_data *data,
4994                            void *localdata)
4995 {
4996         struct mdt_thread_info *info;
4997         struct obd_export      *lexp;
4998         struct lustre_handle    conn = { 0 };
4999         struct mdt_device      *mdt;
5000         struct ptlrpc_request  *req;
5001         int                     rc;
5002         ENTRY;
5003
5004         LASSERT(env != NULL);
5005         if (!exp || !obd || !cluuid)
5006                 RETURN(-EINVAL);
5007
5008         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5009         req = info->mti_pill->rc_req;
5010         mdt = mdt_dev(obd->obd_lu_dev);
5011
5012         rc = class_connect(&conn, obd, cluuid);
5013         if (rc)
5014                 RETURN(rc);
5015
5016         lexp = class_conn2export(&conn);
5017         LASSERT(lexp != NULL);
5018
5019         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5020         if (rc)
5021                 GOTO(out, rc);
5022
5023         rc = mdt_connect_internal(lexp, mdt, data);
5024         if (rc == 0) {
5025                 struct mdt_thread_info *mti;
5026                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5027                 LASSERT(lcd);
5028                 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5029                 LASSERT(mti != NULL);
5030                 mti->mti_exp = lexp;
5031                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5032                 rc = mdt_client_new(env, mdt);
5033                 if (rc == 0)
5034                         mdt_export_stats_init(obd, lexp, 0, localdata);
5035         }
5036
5037 out:
5038         if (rc != 0) {
5039                 class_disconnect(lexp);
5040                 *exp = NULL;
5041         } else {
5042                 *exp = lexp;
5043         }
5044
5045         RETURN(rc);
5046 }
5047
5048 static int mdt_obd_reconnect(const struct lu_env *env,
5049                              struct obd_export *exp, struct obd_device *obd,
5050                              struct obd_uuid *cluuid,
5051                              struct obd_connect_data *data,
5052                              void *localdata)
5053 {
5054         struct mdt_thread_info *info;
5055         struct mdt_device      *mdt;
5056         struct ptlrpc_request  *req;
5057         int                     rc;
5058         ENTRY;
5059
5060         if (exp == NULL || obd == NULL || cluuid == NULL)
5061                 RETURN(-EINVAL);
5062
5063         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5064         req = info->mti_pill->rc_req;
5065         mdt = mdt_dev(obd->obd_lu_dev);
5066
5067         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5068         if (rc)
5069                 RETURN(rc);
5070
5071         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5072         if (rc == 0)
5073                 mdt_export_stats_init(obd, exp, 1, localdata);
5074
5075         RETURN(rc);
5076 }
5077 static int mdt_export_cleanup(struct obd_export *exp)
5078 {
5079         struct mdt_export_data *med = &exp->exp_mdt_data;
5080         struct obd_device      *obd = exp->exp_obd;
5081         struct mdt_device      *mdt;
5082         struct mdt_thread_info *info;
5083         struct lu_env           env;
5084         CFS_LIST_HEAD(closing_list);
5085         struct mdt_file_data *mfd, *n;
5086         int rc = 0;
5087         ENTRY;
5088
5089         cfs_spin_lock(&med->med_open_lock);
5090         while (!cfs_list_empty(&med->med_open_head)) {
5091                 cfs_list_t *tmp = med->med_open_head.next;
5092                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5093
5094                 /* Remove mfd handle so it can't be found again.
5095                  * We are consuming the mfd_list reference here. */
5096                 class_handle_unhash(&mfd->mfd_handle);
5097                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5098         }
5099         cfs_spin_unlock(&med->med_open_lock);
5100         mdt = mdt_dev(obd->obd_lu_dev);
5101         LASSERT(mdt != NULL);
5102
5103         rc = lu_env_init(&env, LCT_MD_THREAD);
5104         if (rc)
5105                 RETURN(rc);
5106
5107         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5108         LASSERT(info != NULL);
5109         memset(info, 0, sizeof *info);
5110         info->mti_env = &env;
5111         info->mti_mdt = mdt;
5112         info->mti_exp = exp;
5113
5114         if (!cfs_list_empty(&closing_list)) {
5115                 struct md_attr *ma = &info->mti_attr;
5116                 int lmm_size;
5117                 int cookie_size;
5118
5119                 lmm_size = mdt->mdt_max_mdsize;
5120                 OBD_ALLOC(ma->ma_lmm, lmm_size);
5121                 if (ma->ma_lmm == NULL)
5122                         GOTO(out_lmm, rc = -ENOMEM);
5123
5124                 cookie_size = mdt->mdt_max_cookiesize;
5125                 OBD_ALLOC(ma->ma_cookie, cookie_size);
5126                 if (ma->ma_cookie == NULL)
5127                         GOTO(out_cookie, rc = -ENOMEM);
5128
5129                 /* Close any open files (which may also cause orphan unlinking). */
5130                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5131                         cfs_list_del_init(&mfd->mfd_list);
5132                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5133                         ma->ma_lmm_size = lmm_size;
5134                         ma->ma_cookie_size = cookie_size;
5135                         ma->ma_need = 0;
5136                         /* It is not for setattr, just tell MDD to send
5137                          * DESTROY RPC to OSS if needed */
5138                         ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5139                         ma->ma_valid = MA_FLAGS;
5140                         mdt_mfd_close(info, mfd);
5141                 }
5142                 OBD_FREE(ma->ma_cookie, cookie_size);
5143                 ma->ma_cookie = NULL;
5144 out_cookie:
5145                 OBD_FREE(ma->ma_lmm, lmm_size);
5146                 ma->ma_lmm = NULL;
5147         }
5148 out_lmm:
5149         info->mti_mdt = NULL;
5150         /* cleanup client slot early */
5151         /* Do not erase record for recoverable client. */
5152         if (!obd->obd_fail || exp->exp_failed)
5153                 mdt_client_del(&env, mdt);
5154         lu_env_fini(&env);
5155
5156         RETURN(rc);
5157 }
5158
5159 static int mdt_obd_disconnect(struct obd_export *exp)
5160 {
5161         int rc;
5162         ENTRY;
5163
5164         LASSERT(exp);
5165         class_export_get(exp);
5166
5167         rc = server_disconnect_export(exp);
5168         if (rc != 0)
5169                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5170
5171         rc = mdt_export_cleanup(exp);
5172         class_export_put(exp);
5173         RETURN(rc);
5174 }
5175
5176 /* FIXME: Can we avoid using these two interfaces? */
5177 static int mdt_init_export(struct obd_export *exp)
5178 {
5179         struct mdt_export_data *med = &exp->exp_mdt_data;
5180         int                     rc;
5181         ENTRY;
5182
5183         CFS_INIT_LIST_HEAD(&med->med_open_head);
5184         cfs_spin_lock_init(&med->med_open_lock);
5185         cfs_sema_init(&med->med_idmap_sem, 1);
5186         med->med_idmap = NULL;
5187         cfs_spin_lock(&exp->exp_lock);
5188         exp->exp_connecting = 1;
5189         cfs_spin_unlock(&exp->exp_lock);
5190         rc = lut_client_alloc(exp);
5191         if (rc == 0)
5192                 rc = ldlm_init_export(exp);
5193
5194         if (rc)
5195                 CERROR("Error %d while initializing export\n", rc);
5196         RETURN(rc);
5197 }
5198
5199 static int mdt_destroy_export(struct obd_export *exp)
5200 {
5201         struct mdt_export_data *med;
5202         int rc = 0;
5203         ENTRY;
5204
5205         med = &exp->exp_mdt_data;
5206         if (exp_connect_rmtclient(exp))
5207                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5208
5209         target_destroy_export(exp);
5210         ldlm_destroy_export(exp);
5211         lut_client_free(exp);
5212
5213         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5214         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5215         if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5216                 RETURN(0);
5217
5218         RETURN(rc);
5219 }
5220
5221 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5222 {
5223         if (flag & CONFIG_LOG)
5224                 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5225
5226         /* also notify active event */
5227         if (flag & CONFIG_SYNC)
5228                 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5229
5230         if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5231             cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5232                 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5233
5234                 /* Open for clients */
5235                 if (obd->obd_no_conn) {
5236                         cfs_spin_lock(&obd->obd_dev_lock);
5237                         obd->obd_no_conn = 0;
5238                         cfs_spin_unlock(&obd->obd_dev_lock);
5239                 }
5240         }
5241 }
5242
5243 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5244                       enum md_upcall_event ev, void *data)
5245 {
5246         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5247         struct md_device  *next  = m->mdt_child;
5248         struct mdt_thread_info *mti;
5249         int rc = 0;
5250         ENTRY;
5251
5252         switch (ev) {
5253                 case MD_LOV_SYNC:
5254                         rc = next->md_ops->mdo_maxsize_get(env, next,
5255                                         &m->mdt_max_mdsize,
5256                                         &m->mdt_max_cookiesize);
5257                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5258                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5259                         mdt_allow_cli(m, CONFIG_SYNC);
5260                         if (data)
5261                                 (*(__u64 *)data) =
5262                                       m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5263                         break;
5264                 case MD_NO_TRANS:
5265                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5266                         mti->mti_no_need_trans = 1;
5267                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5268                         break;
5269                 case MD_LOV_CONFIG:
5270                         /* Check that MDT is not yet configured */
5271                         LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5272                         break;
5273 #ifdef HAVE_QUOTA_SUPPORT
5274                 case MD_LOV_QUOTA:
5275                         if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5276                             likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5277                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5278                         break;
5279 #endif
5280                 default:
5281                         CERROR("invalid event\n");
5282                         rc = -EINVAL;
5283                         break;
5284         }
5285         RETURN(rc);
5286 }
5287
5288 static int mdt_obd_notify(struct obd_device *host,
5289                           struct obd_device *watched,
5290                           enum obd_notify_event ev, void *data)
5291 {
5292         struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5293 #ifdef HAVE_QUOTA_SUPPORT
5294         struct md_device *next = mdt->mdt_child;
5295 #endif
5296         ENTRY;
5297
5298         switch (ev) {
5299         case OBD_NOTIFY_CONFIG:
5300                 mdt_allow_cli(mdt, (unsigned long)data);
5301
5302 #ifdef HAVE_QUOTA_SUPPORT
5303                /* quota_type has been processed, we can now handle
5304                 * incoming quota requests */
5305                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5306 #endif
5307                 break;
5308         default:
5309                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5310         }
5311         RETURN(0);
5312 }
5313
5314 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5315                             void *val, int vallen)
5316 {
5317         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5318         struct getinfo_fid2path *fpout, *fpin;
5319         int rc = 0;
5320
5321         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5322         fpout = val;
5323
5324         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5325                 lustre_swab_fid2path(fpin);
5326
5327         memcpy(fpout, fpin, sizeof(*fpin));
5328         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5329                 RETURN(-EINVAL);
5330
5331         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5332         RETURN(rc);
5333 }
5334
5335 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5336                         struct getinfo_fid2path *fp)
5337 {
5338         struct mdt_object *obj;
5339         int    rc;
5340         ENTRY;
5341
5342         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5343                PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5344
5345         if (!fid_is_sane(&fp->gf_fid))
5346                 RETURN(-EINVAL);
5347
5348         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5349         if (obj == NULL || IS_ERR(obj)) {
5350                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5351                        PTR_ERR(obj));
5352                 RETURN(-EINVAL);
5353         }
5354
5355         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5356         if (rc <= 0) {
5357                 if (rc == -1)
5358                         rc = -EREMOTE;
5359                 else
5360                         rc = -ENOENT;
5361                 mdt_object_put(env, obj);
5362                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5363                        PFID(&fp->gf_fid), rc);
5364                 RETURN(rc);
5365         }
5366
5367         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5368                      fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5369         mdt_object_put(env, obj);
5370
5371         RETURN(rc);
5372 }
5373
5374 static int mdt_get_info(struct mdt_thread_info *info)
5375 {
5376         struct ptlrpc_request *req = mdt_info_req(info);
5377         char *key;
5378         int keylen;
5379         __u32 *vallen;
5380         void *valout;
5381         int rc;
5382         ENTRY;
5383
5384         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5385         if (key == NULL) {
5386                 CDEBUG(D_IOCTL, "No GETINFO key");
5387                 RETURN(-EFAULT);
5388         }
5389         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5390                                       RCL_CLIENT);
5391
5392         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5393         if (vallen == NULL) {
5394                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5395                 RETURN(-EFAULT);
5396         }
5397
5398         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5399                              *vallen);
5400         rc = req_capsule_server_pack(info->mti_pill);
5401         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5402         if (valout == NULL) {
5403                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5404                 RETURN(-EFAULT);
5405         }
5406
5407         if (KEY_IS(KEY_FID2PATH))
5408                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5409         else
5410                 rc = -EINVAL;
5411
5412         lustre_msg_set_status(req->rq_repmsg, rc);
5413
5414         RETURN(rc);
5415 }
5416
5417 /* Pass the ioc down */
5418 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5419                          unsigned int cmd, int len, void *data)
5420 {
5421         struct lu_context ioctl_session;
5422         struct md_device *next = mdt->mdt_child;
5423         int rc;
5424         ENTRY;
5425
5426         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5427         if (rc)
5428                 RETURN(rc);
5429         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5430         lu_context_enter(&ioctl_session);
5431         env->le_ses = &ioctl_session;
5432
5433         LASSERT(next->md_ops->mdo_iocontrol);
5434         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5435
5436         lu_context_exit(&ioctl_session);
5437         lu_context_fini(&ioctl_session);
5438         RETURN(rc);
5439 }
5440
5441 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5442 {
5443         struct obd_ioctl_data *data = karg;
5444         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5445         __u64 version;
5446         struct mdt_object *obj;
5447         struct mdt_lock_handle  *lh;
5448         int rc;
5449         ENTRY;
5450         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5451         if (!fid_is_sane(fid))
5452                 RETURN(-EINVAL);
5453
5454         lh = &mti->mti_lh[MDT_LH_PARENT];
5455         mdt_lock_reg_init(lh, LCK_CR);
5456
5457         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5458         if (IS_ERR(obj))
5459                 RETURN(PTR_ERR(obj));
5460
5461         rc = mdt_object_exists(obj);
5462         if (rc < 0) {
5463                 rc = -EREMOTE;
5464                 /**
5465                  * before calling version get the correct MDS should be
5466                  * fid, this is error to find remote object here
5467                  */
5468                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5469         } else {
5470                 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5471                *(__u64 *)data->ioc_inlbuf2 = version;
5472                 rc = 0;
5473         }
5474         mdt_object_unlock_put(mti, obj, lh, 1);
5475         RETURN(rc);
5476 }
5477
5478 /* ioctls on obd dev */
5479 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5480                          void *karg, void *uarg)
5481 {
5482         struct lu_env      env;
5483         struct obd_device *obd = exp->exp_obd;
5484         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5485         struct dt_device  *dt = mdt->mdt_bottom;
5486         int rc;
5487
5488         ENTRY;
5489         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5490         rc = lu_env_init(&env, LCT_MD_THREAD);
5491         if (rc)
5492                 RETURN(rc);
5493
5494         switch (cmd) {
5495         case OBD_IOC_SYNC:
5496                 rc = mdt_device_sync(&env, mdt);
5497                 break;
5498         case OBD_IOC_SET_READONLY:
5499                 dt->dd_ops->dt_ro(&env, dt);
5500                 break;
5501         case OBD_IOC_ABORT_RECOVERY:
5502                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5503                 target_stop_recovery_thread(obd);
5504                 rc = 0;
5505                 break;
5506         case OBD_IOC_CHANGELOG_REG:
5507         case OBD_IOC_CHANGELOG_DEREG:
5508         case OBD_IOC_CHANGELOG_CLEAR:
5509                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5510                 break;
5511         case OBD_IOC_GET_OBJ_VERSION: {
5512                 struct mdt_thread_info *mti;
5513                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5514                 memset(mti, 0, sizeof *mti);
5515                 mti->mti_env = &env;
5516                 mti->mti_mdt = mdt;
5517                 mti->mti_exp = exp;
5518
5519                 rc = mdt_ioc_version_get(mti, karg);
5520                 break;
5521         }
5522         default:
5523                 CERROR("Not supported cmd = %d for device %s\n",
5524                        cmd, obd->obd_name);
5525                 rc = -EOPNOTSUPP;
5526         }
5527
5528         lu_env_fini(&env);
5529         RETURN(rc);
5530 }
5531
5532 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5533 {
5534         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5535 #ifdef HAVE_QUOTA_SUPPORT
5536         struct obd_device *obd = mdt2obd_dev(mdt);
5537         struct md_device *next = mdt->mdt_child;
5538 #endif
5539         int rc;
5540         ENTRY;
5541
5542         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5543 #ifdef HAVE_QUOTA_SUPPORT
5544         if (likely(obd->obd_stopping == 0))
5545                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5546 #endif
5547         RETURN(rc);
5548 }
5549
5550 int mdt_obd_postrecov(struct obd_device *obd)
5551 {
5552         struct lu_env env;
5553         int rc;
5554
5555         rc = lu_env_init(&env, LCT_MD_THREAD);
5556         if (rc)
5557                 RETURN(rc);
5558         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5559         lu_env_fini(&env);
5560         return rc;
5561 }
5562
5563 /**
5564  * Send a copytool req to a client
5565  * Note this sends a request RPC from a server (MDT) to a client (MDC),
5566  * backwards of normal comms.
5567  */
5568 int mdt_hsm_copytool_send(struct obd_export *exp)
5569 {
5570         struct kuc_hdr *lh;
5571         struct hsm_action_list *hal;
5572         struct hsm_action_item *hai;
5573         int rc, len;
5574         ENTRY;
5575
5576         CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5577               libcfs_nid2str(exp->exp_connection->c_peer.nid));
5578
5579         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5580                 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5581         OBD_ALLOC(lh, len);
5582         if (lh == NULL)
5583                 RETURN(-ENOMEM);
5584
5585         lh->kuc_magic = KUC_MAGIC;
5586         lh->kuc_transport = KUC_TRANSPORT_HSM;
5587         lh->kuc_msgtype = HMT_ACTION_LIST;
5588         lh->kuc_msglen = len;
5589
5590         hal = (struct hsm_action_list *)(lh + 1);
5591         hal->hal_version = HAL_VERSION;
5592         hal->hal_archive_num = 1;
5593         obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5594                         MTI_NAME_MAXLEN);
5595
5596         /* mock up an action list */
5597         hal->hal_count = 2;
5598         hai = hai_zero(hal);
5599         hai->hai_action = HSMA_ARCHIVE;
5600         hai->hai_fid.f_oid = 0xA00A;
5601         hai->hai_len = sizeof(*hai);
5602         hai = hai_next(hai);
5603         hai->hai_action = HSMA_RESTORE;
5604         hai->hai_fid.f_oid = 0xB00B;
5605         hai->hai_len = sizeof(*hai);
5606
5607         /* Uses the ldlm reverse import; this rpc will be seen by
5608           the ldlm_callback_handler */
5609         rc = do_set_info_async(exp->exp_imp_reverse,
5610                                LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5611                                sizeof(KEY_HSM_COPYTOOL_SEND),
5612                                KEY_HSM_COPYTOOL_SEND,
5613                                len, lh, NULL);
5614
5615         OBD_FREE(lh, len);
5616
5617         RETURN(rc);
5618 }
5619
5620 static struct obd_ops mdt_obd_device_ops = {
5621         .o_owner          = THIS_MODULE,
5622         .o_set_info_async = mdt_obd_set_info_async,
5623         .o_connect        = mdt_obd_connect,
5624         .o_reconnect      = mdt_obd_reconnect,
5625         .o_disconnect     = mdt_obd_disconnect,
5626         .o_init_export    = mdt_init_export,
5627         .o_destroy_export = mdt_destroy_export,
5628         .o_iocontrol      = mdt_iocontrol,
5629         .o_postrecov      = mdt_obd_postrecov,
5630         .o_notify         = mdt_obd_notify
5631 };
5632
5633 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5634                                          struct lu_device *d)
5635 {
5636         struct mdt_device *m = mdt_dev(d);
5637         ENTRY;
5638
5639         mdt_fini(env, m);
5640         RETURN(NULL);
5641 }
5642
5643 static struct lu_device *mdt_device_free(const struct lu_env *env,
5644                                          struct lu_device *d)
5645 {
5646         struct mdt_device *m = mdt_dev(d);
5647         ENTRY;
5648
5649         md_device_fini(&m->mdt_md_dev);
5650         OBD_FREE_PTR(m);
5651         RETURN(NULL);
5652 }
5653
5654 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5655                                           struct lu_device_type *t,
5656                                           struct lustre_cfg *cfg)
5657 {
5658         struct lu_device  *l;
5659         struct mdt_device *m;
5660
5661         OBD_ALLOC_PTR(m);
5662         if (m != NULL) {
5663                 int rc;
5664
5665                 l = &m->mdt_md_dev.md_lu_dev;
5666                 rc = mdt_init0(env, m, t, cfg);
5667                 if (rc != 0) {
5668                         mdt_device_free(env, l);
5669                         l = ERR_PTR(rc);
5670                         return l;
5671                 }
5672                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5673         } else
5674                 l = ERR_PTR(-ENOMEM);
5675         return l;
5676 }
5677
5678 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5679 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5680
5681 /* context key: mdt_thread_key */
5682 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5683
5684 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5685 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5686
5687 struct lu_context_key mdt_txn_key = {
5688         .lct_tags = LCT_TX_HANDLE,
5689         .lct_init = mdt_txn_key_init,
5690         .lct_fini = mdt_txn_key_fini
5691 };
5692
5693 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5694 {
5695         return md_ucred(info->mti_env);
5696 }
5697
5698 /**
5699  * Enable/disable COS (Commit On Sharing).
5700  *
5701  * Set/Clear the COS flag in mdt options.
5702  *
5703  * \param mdt mdt device
5704  * \param val 0 disables COS, other values enable COS
5705  */
5706 void mdt_enable_cos(struct mdt_device *mdt, int val)
5707 {
5708         struct lu_env env;
5709         int rc;
5710
5711         mdt->mdt_opts.mo_cos = !!val;
5712         rc = lu_env_init(&env, LCT_MD_THREAD);
5713         if (unlikely(rc != 0)) {
5714                 CWARN("lu_env initialization failed with rc = %d,"
5715                       "cannot sync\n", rc);
5716                 return;
5717         }
5718         mdt_device_sync(&env, mdt);
5719         lu_env_fini(&env);
5720 }
5721
5722 /**
5723  * Check COS (Commit On Sharing) status.
5724  *
5725  * Return COS flag status.
5726  *
5727  * \param mdt mdt device
5728  */
5729 int mdt_cos_is_enabled(struct mdt_device *mdt)
5730 {
5731         return mdt->mdt_opts.mo_cos != 0;
5732 }
5733
5734 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5735 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5736
5737 static struct lu_device_type_operations mdt_device_type_ops = {
5738         .ldto_init = mdt_type_init,
5739         .ldto_fini = mdt_type_fini,
5740
5741         .ldto_start = mdt_type_start,
5742         .ldto_stop  = mdt_type_stop,
5743
5744         .ldto_device_alloc = mdt_device_alloc,
5745         .ldto_device_free  = mdt_device_free,
5746         .ldto_device_fini  = mdt_device_fini
5747 };
5748
5749 static struct lu_device_type mdt_device_type = {
5750         .ldt_tags     = LU_DEVICE_MD,
5751         .ldt_name     = LUSTRE_MDT_NAME,
5752         .ldt_ops      = &mdt_device_type_ops,
5753         .ldt_ctx_tags = LCT_MD_THREAD
5754 };
5755
5756 static struct lu_local_obj_desc mdt_last_recv = {
5757         .llod_name      = LAST_RCVD,
5758         .llod_oid       = MDT_LAST_RECV_OID,
5759         .llod_is_index  = 0,
5760 };
5761
5762 static int __init mdt_mod_init(void)
5763 {
5764         struct lprocfs_static_vars lvars;
5765         int rc;
5766
5767         llo_local_obj_register(&mdt_last_recv);
5768
5769         if (mdt_num_threads > 0) {
5770                 if (mdt_num_threads > MDT_MAX_THREADS)
5771                         mdt_num_threads = MDT_MAX_THREADS;
5772                 if (mdt_num_threads < MDT_MIN_THREADS)
5773                         mdt_num_threads = MDT_MIN_THREADS;
5774                 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5775         } else {
5776                 mdt_max_threads = MDT_MAX_THREADS;
5777                 mdt_min_threads = MDT_MIN_THREADS;
5778                 if (mdt_min_threads < MDT_NUM_THREADS)
5779                         mdt_min_threads = MDT_NUM_THREADS;
5780         }
5781
5782         lprocfs_mdt_init_vars(&lvars);
5783         rc = class_register_type(&mdt_obd_device_ops, NULL,
5784                                  lvars.module_vars, LUSTRE_MDT_NAME,
5785                                  &mdt_device_type);
5786
5787         return rc;
5788 }
5789
5790 static void __exit mdt_mod_exit(void)
5791 {
5792         llo_local_obj_unregister(&mdt_last_recv);
5793         class_unregister_type(LUSTRE_MDT_NAME);
5794 }
5795
5796
5797 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5798 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5799         .mh_name    = #opc,                                             \
5800         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5801         .mh_opc     = prefix ## _  ## opc,                              \
5802         .mh_flags   = flags,                                            \
5803         .mh_act     = fn,                                               \
5804         .mh_fmt     = fmt                                               \
5805 }
5806
5807 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5808         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5809
5810 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5811         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5812
5813 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5814         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5815 /*
5816  * Request with a format known in advance
5817  */
5818 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5819         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5820
5821 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5822         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5823
5824 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5825         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5826 /*
5827  * Request with a format we do not yet know
5828  */
5829 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5830         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5831
5832 static struct mdt_handler mdt_mds_ops[] = {
5833 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5834 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5835 DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
5836                                                              &RQF_OBD_SET_INFO),
5837 DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
5838 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5839 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5840 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5841 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5842 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5843 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5844 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5845 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5846 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5847 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5848 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5849 #ifdef HAVE_QUOTA_SUPPORT
5850 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5851 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5852 #endif
5853 };
5854
5855 #define DEF_OBD_HNDL(flags, name, fn)                   \
5856         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5857
5858
5859 static struct mdt_handler mdt_obd_ops[] = {
5860         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5861         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5862         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5863 };
5864
5865 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5866         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5867 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5868         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5869
5870 static struct mdt_handler mdt_dlm_ops[] = {
5871         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5872         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5873         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5874         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5875 };
5876
5877 #define DEF_LLOG_HNDL(flags, name, fn)                   \
5878         DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5879
5880 static struct mdt_handler mdt_llog_ops[] = {
5881         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
5882         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
5883         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5884         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
5885         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
5886         DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
5887         DEF_LLOG_HNDL(0, CATINFO,                   NULL),
5888         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
5889         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
5890 };
5891
5892 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5893         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5894
5895 static struct mdt_handler mdt_sec_ctx_ops[] = {
5896         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5897         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5898         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5899 };
5900
5901 static struct mdt_opc_slice mdt_regular_handlers[] = {
5902         {
5903                 .mos_opc_start = MDS_GETATTR,
5904                 .mos_opc_end   = MDS_LAST_OPC,
5905                 .mos_hs        = mdt_mds_ops
5906         },
5907         {
5908                 .mos_opc_start = OBD_PING,
5909                 .mos_opc_end   = OBD_LAST_OPC,
5910                 .mos_hs        = mdt_obd_ops
5911         },
5912         {
5913                 .mos_opc_start = LDLM_ENQUEUE,
5914                 .mos_opc_end   = LDLM_LAST_OPC,
5915                 .mos_hs        = mdt_dlm_ops
5916         },
5917         {
5918                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5919                 .mos_opc_end   = LLOG_LAST_OPC,
5920                 .mos_hs        = mdt_llog_ops
5921         },
5922         {
5923                 .mos_opc_start = SEC_CTX_INIT,
5924                 .mos_opc_end   = SEC_LAST_OPC,
5925                 .mos_hs        = mdt_sec_ctx_ops
5926         },
5927         {
5928                 .mos_hs        = NULL
5929         }
5930 };
5931
5932 static struct mdt_handler mdt_readpage_ops[] = {
5933         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5934         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5935 #ifdef HAVE_SPLIT_SUPPORT
5936         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5937 #endif
5938
5939         /*
5940          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5941          * detailed comments. --umka
5942          */
5943         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5944         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5945 };
5946
5947 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5948         {
5949                 .mos_opc_start = MDS_GETATTR,
5950                 .mos_opc_end   = MDS_LAST_OPC,
5951                 .mos_hs        = mdt_readpage_ops
5952         },
5953         {
5954                 .mos_hs        = NULL
5955         }
5956 };
5957
5958 static struct mdt_handler mdt_xmds_ops[] = {
5959         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5960         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5961         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5962         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5963 };
5964
5965 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5966         {
5967                 .mos_opc_start = MDS_GETATTR,
5968                 .mos_opc_end   = MDS_LAST_OPC,
5969                 .mos_hs        = mdt_xmds_ops
5970         },
5971         {
5972                 .mos_opc_start = OBD_PING,
5973                 .mos_opc_end   = OBD_LAST_OPC,
5974                 .mos_hs        = mdt_obd_ops
5975         },
5976         {
5977                 .mos_opc_start = SEC_CTX_INIT,
5978                 .mos_opc_end   = SEC_LAST_OPC,
5979                 .mos_hs        = mdt_sec_ctx_ops
5980         },
5981         {
5982                 .mos_hs        = NULL
5983         }
5984 };
5985
5986 static struct mdt_handler mdt_seq_ops[] = {
5987         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5988 };
5989
5990 static struct mdt_opc_slice mdt_seq_handlers[] = {
5991         {
5992                 .mos_opc_start = SEQ_QUERY,
5993                 .mos_opc_end   = SEQ_LAST_OPC,
5994                 .mos_hs        = mdt_seq_ops
5995         },
5996         {
5997                 .mos_hs        = NULL
5998         }
5999 };
6000
6001 static struct mdt_handler mdt_fld_ops[] = {
6002         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6003 };
6004
6005 static struct mdt_opc_slice mdt_fld_handlers[] = {
6006         {
6007                 .mos_opc_start = FLD_QUERY,
6008                 .mos_opc_end   = FLD_LAST_OPC,
6009                 .mos_hs        = mdt_fld_ops
6010         },
6011         {
6012                 .mos_hs        = NULL
6013         }
6014 };
6015
6016 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6017 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6018 MODULE_LICENSE("GPL");
6019
6020 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6021                 "number of mdt service threads to start");
6022
6023 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);