lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #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 unsigned long mdt_num_threads;
 103
 104 /* ptlrpc request handler for MDT. All handlers are
 105  * grouped into several slices - struct mdt_opc_slice,
 106  * and stored in an array - mdt_handlers[].
 107  */
 108 struct mdt_handler {
 109         /* The name of this handler. */
 110         const char *mh_name;
 111         /* Fail id for this handler, checked at the beginning of this handler*/
 112         int         mh_fail_id;
 113         /* Operation code for this handler */
 114         __u32       mh_opc;
 115         /* flags are listed in enum mdt_handler_flags below. */
 116         __u32       mh_flags;
 117         /* The actual handler function to execute. */
 118         int (*mh_act)(struct mdt_thread_info *info);
 119         /* Request format for this request. */
 120         const struct req_format *mh_fmt;
 121 };
 122
 123 enum mdt_handler_flags {
 124         /*
 125          * struct mdt_body is passed in the incoming message, and object
 126          * identified by this fid exists on disk.
 127          *
 128          * "habeo corpus" == "I have a body"
 129          */
 130         HABEO_CORPUS = (1 << 0),
 131         /*
 132          * struct ldlm_request is passed in the incoming message.
 133          *
 134          * "habeo clavis" == "I have a key"
 135          */
 136         HABEO_CLAVIS = (1 << 1),
 137         /*
 138          * this request has fixed reply format, so that reply message can be
 139          * packed by generic code.
 140          *
 141          * "habeo refero" == "I have a reply"
 142          */
 143         HABEO_REFERO = (1 << 2),
 144         /*
 145          * this request will modify something, so check whether the filesystem
 146          * is readonly or not, then return -EROFS to client asap if necessary.
 147          *
 148          * "mutabor" == "I shall modify"
 149          */
 150         MUTABOR      = (1 << 3)
 151 };
 152
 153 struct mdt_opc_slice {
 154         __u32               mos_opc_start;
 155         int                 mos_opc_end;
 156         struct mdt_handler *mos_hs;
 157 };
 158
 159 static struct mdt_opc_slice mdt_regular_handlers[];
 160 static struct mdt_opc_slice mdt_readpage_handlers[];
 161 static struct mdt_opc_slice mdt_xmds_handlers[];
 162 static struct mdt_opc_slice mdt_seq_handlers[];
 163 static struct mdt_opc_slice mdt_fld_handlers[];
 164
 165 static struct mdt_device *mdt_dev(struct lu_device *d);
 166 static int mdt_regular_handle(struct ptlrpc_request *req);
 167 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 168 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 169                         struct getinfo_fid2path *fp);
 170
 171 static const struct lu_object_operations mdt_obj_ops;
 172
 173 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 174 {
 175         if (!rep)
 176                 return 0;
 177         return (rep->lock_policy_res1 & flag);
 178 }
 179
 180 void mdt_clear_disposition(struct mdt_thread_info *info,
 181                            struct ldlm_reply *rep, int flag)
 182 {
 183         if (info)
 184                 info->mti_opdata &= ~flag;
 185         if (rep)
 186                 rep->lock_policy_res1 &= ~flag;
 187 }
 188
 189 void mdt_set_disposition(struct mdt_thread_info *info,
 190                          struct ldlm_reply *rep, int flag)
 191 {
 192         if (info)
 193                 info->mti_opdata |= flag;
 194         if (rep)
 195                 rep->lock_policy_res1 |= flag;
 196 }
 197
 198 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 199 {
 200         lh->mlh_pdo_hash = 0;
 201         lh->mlh_reg_mode = lm;
 202         lh->mlh_type = MDT_REG_LOCK;
 203 }
 204
 205 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 206                        const char *name, int namelen)
 207 {
 208         lh->mlh_reg_mode = lm;
 209         lh->mlh_type = MDT_PDO_LOCK;
 210
 211         if (name != NULL) {
 212                 LASSERT(namelen > 0);
 213                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 214         } else {
 215                 LASSERT(namelen == 0);
 216                 lh->mlh_pdo_hash = 0ull;
 217         }
 218 }
 219
 220 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 221                               struct mdt_lock_handle *lh)
 222 {
 223         mdl_mode_t mode;
 224         ENTRY;
 225
 226         /*
 227          * Any dir access needs couple of locks:
 228          *
 229          * 1) on part of dir we gonna take lookup/modify;
 230          *
 231          * 2) on whole dir to protect it from concurrent splitting and/or to
 232          * flush client's cache for readdir().
 233          *
 234          * so, for a given mode and object this routine decides what lock mode
 235          * to use for lock #2:
 236          *
 237          * 1) if caller's gonna lookup in dir then we need to protect dir from
 238          * being splitted only - LCK_CR
 239          *
 240          * 2) if caller's gonna modify dir then we need to protect dir from
 241          * being splitted and to flush cache - LCK_CW
 242          *
 243          * 3) if caller's gonna modify dir and that dir seems ready for
 244          * splitting then we need to protect it from any type of access
 245          * (lookup/modify/split) - LCK_EX --bzzz
 246          */
 247
 248         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 249         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 250
 251         /*
 252          * Ask underlaying level its opinion about preferable PDO lock mode
 253          * having access type passed as regular lock mode:
 254          *
 255          * - MDL_MINMODE means that lower layer does not want to specify lock
 256          * mode;
 257          *
 258          * - MDL_NL means that no PDO lock should be taken. This is used in some
 259          * cases. Say, for non-splittable directories no need to use PDO locks
 260          * at all.
 261          */
 262         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 263                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 264
 265         if (mode != MDL_MINMODE) {
 266                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 267         } else {
 268                 /*
 269                  * Lower layer does not want to specify locking mode. We do it
 270                  * our selves. No special protection is needed, just flush
 271                  * client's cache on modification and allow concurrent
 272                  * mondification.
 273                  */
 274                 switch (lh->mlh_reg_mode) {
 275                 case LCK_EX:
 276                         lh->mlh_pdo_mode = LCK_EX;
 277                         break;
 278                 case LCK_PR:
 279                         lh->mlh_pdo_mode = LCK_CR;
 280                         break;
 281                 case LCK_PW:
 282                         lh->mlh_pdo_mode = LCK_CW;
 283                         break;
 284                 default:
 285                         CERROR("Not expected lock type (0x%x)\n",
 286                                (int)lh->mlh_reg_mode);
 287                         LBUG();
 288                 }
 289         }
 290
 291         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 292         EXIT;
 293 }
 294
 295 static int mdt_getstatus(struct mdt_thread_info *info)
 296 {
 297         struct mdt_device *mdt  = info->mti_mdt;
 298         struct md_device  *next = mdt->mdt_child;
 299         struct mdt_body   *repbody;
 300         int                rc;
 301
 302         ENTRY;
 303
 304         rc = mdt_check_ucred(info);
 305         if (rc)
 306                 RETURN(err_serious(rc));
 307
 308         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 309                 RETURN(err_serious(-ENOMEM));
 310
 311         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 312         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 313         if (rc != 0)
 314                 RETURN(rc);
 315
 316         repbody->valid |= OBD_MD_FLID;
 317
 318         if (mdt->mdt_opts.mo_mds_capa &&
 319             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 320                 struct mdt_object  *root;
 321                 struct lustre_capa *capa;
 322
 323                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 324                 if (IS_ERR(root))
 325                         RETURN(PTR_ERR(root));
 326
 327                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 328                 LASSERT(capa);
 329                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 330                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 331                                  0);
 332                 mdt_object_put(info->mti_env, root);
 333                 if (rc == 0)
 334                         repbody->valid |= OBD_MD_FLMDSCAPA;
 335         }
 336
 337         RETURN(rc);
 338 }
 339
 340 static int mdt_statfs(struct mdt_thread_info *info)
 341 {
 342         struct md_device      *next  = info->mti_mdt->mdt_child;
 343         struct ptlrpc_service *svc;
 344         struct obd_statfs     *osfs;
 345         int                    rc;
 346
 347         ENTRY;
 348
 349         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 350
 351         /* This will trigger a watchdog timeout */
 352         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 353                          (MDT_SERVICE_WATCHDOG_FACTOR *
 354                           at_get(&svc->srv_at_estimate)) + 1);
 355
 356         rc = mdt_check_ucred(info);
 357         if (rc)
 358                 RETURN(err_serious(rc));
 359
 360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 361                 rc = err_serious(-ENOMEM);
 362         } else {
 363                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 364                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 365                                               &info->mti_u.ksfs);
 366                 statfs_pack(osfs, &info->mti_u.ksfs);
 367         }
 368         RETURN(rc);
 369 }
 370
 371 /**
 372  * Pack SOM attributes into the reply.
 373  * Call under a DLM UPDATE lock.
 374  */
 375 static void mdt_pack_size2body(struct mdt_thread_info *info,
 376                                struct mdt_object *mo)
 377 {
 378         struct mdt_body *b;
 379         struct md_attr *ma = &info->mti_attr;
 380
 381         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 382         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 383
 384         /* Check if Size-on-MDS is supported, if this is a regular file,
 385          * if SOM is enabled on the object and if SOM cache exists and valid.
 386          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 387         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 388             !S_ISREG(ma->ma_attr.la_mode) ||
 389             !mdt_object_is_som_enabled(mo) ||
 390             !(ma->ma_valid & MA_SOM))
 391                 return;
 392
 393         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 394         b->size = ma->ma_som->msd_size;
 395         b->blocks = ma->ma_som->msd_blocks;
 396 }
 397
 398 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 399                         const struct lu_attr *attr, const struct lu_fid *fid)
 400 {
 401         /*XXX should pack the reply body according to lu_valid*/
 402         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 403                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 404                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 405                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 406
 407         if (!S_ISREG(attr->la_mode))
 408                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 409
 410         b->atime      = attr->la_atime;
 411         b->mtime      = attr->la_mtime;
 412         b->ctime      = attr->la_ctime;
 413         b->mode       = attr->la_mode;
 414         b->size       = attr->la_size;
 415         b->blocks     = attr->la_blocks;
 416         b->uid        = attr->la_uid;
 417         b->gid        = attr->la_gid;
 418         b->flags      = attr->la_flags;
 419         b->nlink      = attr->la_nlink;
 420         b->rdev       = attr->la_rdev;
 421
 422         if (fid) {
 423                 b->fid1 = *fid;
 424                 b->valid |= OBD_MD_FLID;
 425
 426                 /* FIXME: these should be fixed when new igif ready.*/
 427                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 428                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 429                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 430
 431                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 432                                 PFID(fid), b->nlink, b->mode, b->size);
 433         }
 434
 435         if (info)
 436                 mdt_body_reverse_idmap(info, b);
 437 }
 438
 439 static inline int mdt_body_has_lov(const struct lu_attr *la,
 440                                    const struct mdt_body *body)
 441 {
 442         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 443                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 444 }
 445
 446 static int mdt_getattr_internal(struct mdt_thread_info *info,
 447                                 struct mdt_object *o, int ma_need)
 448 {
 449         struct md_object        *next = mdt_object_child(o);
 450         const struct mdt_body   *reqbody = info->mti_body;
 451         struct ptlrpc_request   *req = mdt_info_req(info);
 452         struct md_attr          *ma = &info->mti_attr;
 453         struct lu_attr          *la = &ma->ma_attr;
 454         struct req_capsule      *pill = info->mti_pill;
 455         const struct lu_env     *env = info->mti_env;
 456         struct mdt_body         *repbody;
 457         struct lu_buf           *buffer = &info->mti_buf;
 458         int                     rc;
 459         ENTRY;
 460
 461         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 462                 RETURN(err_serious(-ENOMEM));
 463
 464         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 465
 466         ma->ma_valid = 0;
 467
 468         rc = mdt_object_exists(o);
 469         if (rc < 0) {
 470                 /* This object is located on remote node.*/
 471                 repbody->fid1 = *mdt_object_fid(o);
 472                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 473                 RETURN(0);
 474         }
 475
 476         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 477         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 478
 479         /* If it is dir object and client require MEA, then we got MEA */
 480         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 481             reqbody->valid & OBD_MD_MEA) {
 482                 /* Assumption: MDT_MD size is enough for lmv size. */
 483                 ma->ma_lmv = buffer->lb_buf;
 484                 ma->ma_lmv_size = buffer->lb_len;
 485                 ma->ma_need = MA_LMV | MA_INODE;
 486         } else {
 487                 ma->ma_lmm = buffer->lb_buf;
 488                 ma->ma_lmm_size = buffer->lb_len;
 489                 ma->ma_need = MA_LOV | MA_INODE;
 490         }
 491
 492         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 493             reqbody->valid & OBD_MD_FLDIREA  &&
 494             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 495                 /* get default stripe info for this dir. */
 496                 ma->ma_need |= MA_LOV_DEF;
 497         }
 498         ma->ma_need |= ma_need;
 499         if (ma->ma_need & MA_SOM)
 500                 ma->ma_som = &info->mti_u.som.data;
 501
 502         rc = mo_attr_get(env, next, ma);
 503         if (unlikely(rc)) {
 504                 CERROR("getattr error for "DFID": %d\n",
 505                         PFID(mdt_object_fid(o)), rc);
 506                 RETURN(rc);
 507         }
 508
 509         if (likely(ma->ma_valid & MA_INODE))
 510                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 511         else
 512                 RETURN(-EFAULT);
 513
 514         if (mdt_body_has_lov(la, reqbody)) {
 515                 if (ma->ma_valid & MA_LOV) {
 516                         LASSERT(ma->ma_lmm_size);
 517                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 518                         repbody->eadatasize = ma->ma_lmm_size;
 519                         if (S_ISDIR(la->la_mode))
 520                                 repbody->valid |= OBD_MD_FLDIREA;
 521                         else
 522                                 repbody->valid |= OBD_MD_FLEASIZE;
 523                 }
 524                 if (ma->ma_valid & MA_LMV) {
 525                         LASSERT(S_ISDIR(la->la_mode));
 526                         repbody->eadatasize = ma->ma_lmv_size;
 527                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 528                 }
 529                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 530                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 531                 }
 532         } else if (S_ISLNK(la->la_mode) &&
 533                    reqbody->valid & OBD_MD_LINKNAME) {
 534                 buffer->lb_buf = ma->ma_lmm;
 535                 buffer->lb_len = reqbody->eadatasize;
 536                 rc = mo_readlink(env, next, buffer);
 537                 if (unlikely(rc <= 0)) {
 538                         CERROR("readlink failed: %d\n", rc);
 539                         rc = -EFAULT;
 540                 } else {
 541                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 542                                  rc -= 2;
 543                         repbody->valid |= OBD_MD_LINKNAME;
 544                         repbody->eadatasize = rc;
 545                         /* NULL terminate */
 546                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 547                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 548                                (char*)ma->ma_lmm, rc);
 549                         rc = 0;
 550                 }
 551         }
 552
 553         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 554                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 555                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 556                 repbody->valid |= OBD_MD_FLMODEASIZE;
 557                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 558                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 559                        repbody->max_cookiesize);
 560         }
 561
 562         if (exp_connect_rmtclient(info->mti_exp) &&
 563             reqbody->valid & OBD_MD_FLRMTPERM) {
 564                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 565
 566                 /* mdt_getattr_lock only */
 567                 rc = mdt_pack_remote_perm(info, o, buf);
 568                 if (rc) {
 569                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 570                         repbody->aclsize = 0;
 571                         RETURN(rc);
 572                 } else {
 573                         repbody->valid |= OBD_MD_FLRMTPERM;
 574                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 575                 }
 576         }
 577 #ifdef CONFIG_FS_POSIX_ACL
 578         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 579                  (reqbody->valid & OBD_MD_FLACL)) {
 580                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 581                 buffer->lb_len = req_capsule_get_size(pill,
 582                                                       &RMF_ACL, RCL_SERVER);
 583                 if (buffer->lb_len > 0) {
 584                         rc = mo_xattr_get(env, next, buffer,
 585                                           XATTR_NAME_ACL_ACCESS);
 586                         if (rc < 0) {
 587                                 if (rc == -ENODATA) {
 588                                         repbody->aclsize = 0;
 589                                         repbody->valid |= OBD_MD_FLACL;
 590                                         rc = 0;
 591                                 } else if (rc == -EOPNOTSUPP) {
 592                                         rc = 0;
 593                                 } else {
 594                                         CERROR("got acl size: %d\n", rc);
 595                                 }
 596                         } else {
 597                                 repbody->aclsize = rc;
 598                                 repbody->valid |= OBD_MD_FLACL;
 599                                 rc = 0;
 600                         }
 601                 }
 602         }
 603 #endif
 604
 605         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 606             info->mti_mdt->mdt_opts.mo_mds_capa &&
 607             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 608                 struct lustre_capa *capa;
 609
 610                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 611                 LASSERT(capa);
 612                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 613                 rc = mo_capa_get(env, next, capa, 0);
 614                 if (rc)
 615                         RETURN(rc);
 616                 repbody->valid |= OBD_MD_FLMDSCAPA;
 617         }
 618         RETURN(rc);
 619 }
 620
 621 static int mdt_renew_capa(struct mdt_thread_info *info)
 622 {
 623         struct mdt_object  *obj = info->mti_object;
 624         struct mdt_body    *body;
 625         struct lustre_capa *capa, *c;
 626         int rc;
 627         ENTRY;
 628
 629         /* if object doesn't exist, or server has disabled capability,
 630          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 631          * flag not set.
 632          */
 633         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 634             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 635                 RETURN(0);
 636
 637         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 638         LASSERT(body != NULL);
 639
 640         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 641         LASSERT(c);
 642
 643         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 644         LASSERT(capa);
 645
 646         *capa = *c;
 647         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 648         if (rc == 0)
 649                 body->valid |= OBD_MD_FLOSSCAPA;
 650         RETURN(rc);
 651 }
 652
 653 static int mdt_getattr(struct mdt_thread_info *info)
 654 {
 655         struct mdt_object       *obj = info->mti_object;
 656         struct req_capsule      *pill = info->mti_pill;
 657         struct mdt_body         *reqbody;
 658         struct mdt_body         *repbody;
 659         mode_t                   mode;
 660         int                      md_size;
 661         int rc;
 662         ENTRY;
 663
 664         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 665         LASSERT(reqbody);
 666
 667         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 668                 rc = req_capsule_server_pack(pill);
 669                 if (unlikely(rc))
 670                         RETURN(err_serious(rc));
 671                 rc = mdt_renew_capa(info);
 672                 GOTO(out_shrink, rc);
 673         }
 674
 675         LASSERT(obj != NULL);
 676         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 677
 678         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 679         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 680             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 681                 md_size = reqbody->eadatasize;
 682         else
 683                 md_size = info->mti_mdt->mdt_max_mdsize;
 684
 685         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 686
 687         rc = req_capsule_server_pack(pill);
 688         if (unlikely(rc != 0))
 689                 RETURN(err_serious(rc));
 690
 691         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 692         LASSERT(repbody != NULL);
 693         repbody->eadatasize = 0;
 694         repbody->aclsize = 0;
 695
 696         if (reqbody->valid & OBD_MD_FLRMTPERM)
 697                 rc = mdt_init_ucred(info, reqbody);
 698         else
 699                 rc = mdt_check_ucred(info);
 700         if (unlikely(rc))
 701                 GOTO(out_shrink, rc);
 702
 703         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 704         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 705
 706         /*
 707          * Don't check capability at all, because rename might getattr for
 708          * remote obj, and at that time no capability is available.
 709          */
 710         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 711         rc = mdt_getattr_internal(info, obj, 0);
 712         if (reqbody->valid & OBD_MD_FLRMTPERM)
 713                 mdt_exit_ucred(info);
 714         EXIT;
 715 out_shrink:
 716         mdt_shrink_reply(info);
 717         return rc;
 718 }
 719
 720 static int mdt_is_subdir(struct mdt_thread_info *info)
 721 {
 722         struct mdt_object     *o = info->mti_object;
 723         struct req_capsule    *pill = info->mti_pill;
 724         const struct mdt_body *body = info->mti_body;
 725         struct mdt_body       *repbody;
 726         int                    rc;
 727         ENTRY;
 728
 729         LASSERT(o != NULL);
 730
 731         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 732
 733         /*
 734          * We save last checked parent fid to @repbody->fid1 for remote
 735          * directory case.
 736          */
 737         LASSERT(fid_is_sane(&body->fid2));
 738         LASSERT(mdt_object_exists(o) > 0);
 739         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 740                            &body->fid2, &repbody->fid1);
 741         if (rc == 0 || rc == -EREMOTE)
 742                 repbody->valid |= OBD_MD_FLID;
 743
 744         RETURN(rc);
 745 }
 746
 747 static int mdt_raw_lookup(struct mdt_thread_info *info,
 748                           struct mdt_object *parent,
 749                           const struct lu_name *lname,
 750                           struct ldlm_reply *ldlm_rep)
 751 {
 752         struct md_object *next = mdt_object_child(info->mti_object);
 753         const struct mdt_body *reqbody = info->mti_body;
 754         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 755         struct mdt_body *repbody;
 756         int rc;
 757         ENTRY;
 758
 759         if (reqbody->valid != OBD_MD_FLID)
 760                 RETURN(0);
 761
 762         LASSERT(!info->mti_cross_ref);
 763
 764         /* Only got the fid of this obj by name */
 765         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 766                         &info->mti_spec);
 767 #if 0
 768         /* XXX is raw_lookup possible as intent operation? */
 769         if (rc != 0) {
 770                 if (rc == -ENOENT)
 771                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 772                 RETURN(rc);
 773         } else
 774                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 775
 776         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 777 #endif
 778         if (rc == 0) {
 779                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 780                 repbody->fid1 = *child_fid;
 781                 repbody->valid = OBD_MD_FLID;
 782         }
 783         RETURN(1);
 784 }
 785
 786 /*
 787  * UPDATE lock should be taken against parent, and be release before exit;
 788  * child_bits lock should be taken against child, and be returned back:
 789  *            (1)normal request should release the child lock;
 790  *            (2)intent request will grant the lock to client.
 791  */
 792 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 793                                  struct mdt_lock_handle *lhc,
 794                                  __u64 child_bits,
 795                                  struct ldlm_reply *ldlm_rep)
 796 {
 797         struct ptlrpc_request  *req       = mdt_info_req(info);
 798         struct mdt_body        *reqbody   = NULL;
 799         struct mdt_object      *parent    = info->mti_object;
 800         struct mdt_object      *child;
 801         struct md_object       *next      = mdt_object_child(parent);
 802         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 803         struct lu_name         *lname     = NULL;
 804         const char             *name      = NULL;
 805         int                     namelen   = 0;
 806         struct mdt_lock_handle *lhp;
 807         struct ldlm_lock       *lock;
 808         struct ldlm_res_id     *res_id;
 809         int                     is_resent;
 810         int                     ma_need = 0;
 811         int                     rc;
 812
 813         ENTRY;
 814
 815         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 816         LASSERT(ergo(is_resent,
 817                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 818
 819         LASSERT(parent != NULL);
 820         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 821         if (name == NULL)
 822                 RETURN(err_serious(-EFAULT));
 823
 824         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 825                                        RCL_CLIENT) - 1;
 826         if (!info->mti_cross_ref) {
 827                 /*
 828                  * XXX: Check for "namelen == 0" is for getattr by fid
 829                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 830                  * that is the name must contain at least one character and
 831                  * the terminating '\0'
 832                  */
 833                 if (namelen == 0) {
 834                         reqbody = req_capsule_client_get(info->mti_pill,
 835                                                          &RMF_MDT_BODY);
 836                         LASSERT(fid_is_sane(&reqbody->fid2));
 837                         name = NULL;
 838
 839                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 840                                "ldlm_rep = %p\n",
 841                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 842                                ldlm_rep);
 843                 } else {
 844                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 845                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 846                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 847                                name, ldlm_rep);
 848                 }
 849         }
 850         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 851
 852         rc = mdt_object_exists(parent);
 853         if (unlikely(rc == 0)) {
 854                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 855                                 &parent->mot_obj.mo_lu,
 856                                 "Parent doesn't exist!\n");
 857                 RETURN(-ESTALE);
 858         } else if (!info->mti_cross_ref) {
 859                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 860                          PFID(mdt_object_fid(parent)));
 861         }
 862         if (lname) {
 863                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 864                 if (rc != 0) {
 865                         if (rc > 0)
 866                                 rc = 0;
 867                         RETURN(rc);
 868                 }
 869         }
 870
 871         if (info->mti_cross_ref) {
 872                 /* Only getattr on the child. Parent is on another node. */
 873                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 874                 child = parent;
 875                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 876                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 877
 878                 if (is_resent) {
 879                         /* Do not take lock for resent case. */
 880                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 881                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 882                                  lhc->mlh_reg_lh.cookie);
 883                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 884                                                 &lock->l_resource->lr_name));
 885                         LDLM_LOCK_PUT(lock);
 886                         rc = 0;
 887                 } else {
 888                         mdt_lock_handle_init(lhc);
 889                         mdt_lock_reg_init(lhc, LCK_PR);
 890
 891                         /*
 892                          * Object's name is on another MDS, no lookup lock is
 893                          * needed here but update is.
 894                          */
 895                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 896                         child_bits |= MDS_INODELOCK_UPDATE;
 897
 898                         rc = mdt_object_lock(info, child, lhc, child_bits,
 899                                              MDT_LOCAL_LOCK);
 900                 }
 901                 if (rc == 0) {
 902                         /* Finally, we can get attr for child. */
 903                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 904                                          BYPASS_CAPA);
 905                         rc = mdt_getattr_internal(info, child, 0);
 906                         if (unlikely(rc != 0))
 907                                 mdt_object_unlock(info, child, lhc, 1);
 908                 }
 909                 RETURN(rc);
 910         }
 911
 912         /* step 1: lock parent */
 913         lhp = &info->mti_lh[MDT_LH_PARENT];
 914         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 915         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 916                              MDT_LOCAL_LOCK);
 917
 918         if (unlikely(rc != 0))
 919                 RETURN(rc);
 920
 921         if (lname) {
 922                 /* step 2: lookup child's fid by name */
 923                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 924                                 &info->mti_spec);
 925
 926                 if (rc != 0) {
 927                         if (rc == -ENOENT)
 928                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 929                         GOTO(out_parent, rc);
 930                 } else
 931                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 932         } else {
 933                 *child_fid = reqbody->fid2;
 934                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 935         }
 936
 937         /*
 938          *step 3: find the child object by fid & lock it.
 939          *        regardless if it is local or remote.
 940          */
 941         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 942
 943         if (unlikely(IS_ERR(child)))
 944                 GOTO(out_parent, rc = PTR_ERR(child));
 945         if (is_resent) {
 946                 /* Do not take lock for resent case. */
 947                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 948                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 949                          lhc->mlh_reg_lh.cookie);
 950
 951                 res_id = &lock->l_resource->lr_name;
 952                 if (!fid_res_name_eq(mdt_object_fid(child),
 953                                     &lock->l_resource->lr_name)) {
 954                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 955                                                  &lock->l_resource->lr_name),
 956                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 957                                  (unsigned long)res_id->name[0],
 958                                  (unsigned long)res_id->name[1],
 959                                  (unsigned long)res_id->name[2],
 960                                  PFID(mdt_object_fid(parent)));
 961                           CWARN("Although resent, but still not get child lock"
 962                                 "parent:"DFID" child:"DFID"\n",
 963                                 PFID(mdt_object_fid(parent)),
 964                                 PFID(mdt_object_fid(child)));
 965                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 966                           LDLM_LOCK_PUT(lock);
 967                           GOTO(relock, 0);
 968                 }
 969                 LDLM_LOCK_PUT(lock);
 970                 rc = 0;
 971         } else {
 972                 struct md_attr *ma;
 973 relock:
 974                 ma = &info->mti_attr;
 975
 976                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
 977                 mdt_lock_handle_init(lhc);
 978                 mdt_lock_reg_init(lhc, LCK_PR);
 979
 980                 if (mdt_object_exists(child) == 0) {
 981                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 982                                         &child->mot_obj.mo_lu,
 983                                         "Object doesn't exist!\n");
 984                         GOTO(out_child, rc = -ESTALE);
 985                 }
 986
 987                 ma->ma_valid = 0;
 988                 ma->ma_need = MA_INODE;
 989                 rc = mo_attr_get(info->mti_env, next, ma);
 990                 if (unlikely(rc != 0))
 991                         GOTO(out_child, rc);
 992
 993                 /* If the file has not been changed for some time, we return
 994                  * not only a LOOKUP lock, but also an UPDATE lock and this
 995                  * might save us RPC on later STAT. For directories, it also
 996                  * let negative dentry starts working for this dir. */
 997                 if (ma->ma_valid & MA_INODE &&
 998                     ma->ma_attr.la_valid & LA_CTIME &&
 999                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1000                     ma->ma_attr.la_ctime < cfs_time_current_sec())
1001                         child_bits |= MDS_INODELOCK_UPDATE;
1002
1003                 rc = mdt_object_lock(info, child, lhc, child_bits,
1004                                      MDT_CROSS_LOCK);
1005
1006                 if (unlikely(rc != 0))
1007                         GOTO(out_child, rc);
1008         }
1009
1010         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1011         /* Get MA_SOM attributes if update lock is given. */
1012         if (lock &&
1013             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1014             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1015                 ma_need = MA_SOM;
1016
1017         /* finally, we can get attr for child. */
1018         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1019         rc = mdt_getattr_internal(info, child, ma_need);
1020         if (unlikely(rc != 0)) {
1021                 mdt_object_unlock(info, child, lhc, 1);
1022         } else if (lock) {
1023                 /* Debugging code. */
1024                 res_id = &lock->l_resource->lr_name;
1025                 LDLM_DEBUG(lock, "Returning lock to client");
1026                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1027                                          &lock->l_resource->lr_name),
1028                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1029                          (unsigned long)res_id->name[0],
1030                          (unsigned long)res_id->name[1],
1031                          (unsigned long)res_id->name[2],
1032                          PFID(mdt_object_fid(child)));
1033                 mdt_pack_size2body(info, child);
1034         }
1035         if (lock)
1036                 LDLM_LOCK_PUT(lock);
1037
1038         EXIT;
1039 out_child:
1040         mdt_object_put(info->mti_env, child);
1041 out_parent:
1042         mdt_object_unlock(info, parent, lhp, 1);
1043         return rc;
1044 }
1045
1046 /* normal handler: should release the child lock */
1047 static int mdt_getattr_name(struct mdt_thread_info *info)
1048 {
1049         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1050         struct mdt_body        *reqbody;
1051         struct mdt_body        *repbody;
1052         int rc;
1053         ENTRY;
1054
1055         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1056         LASSERT(reqbody != NULL);
1057         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1058         LASSERT(repbody != NULL);
1059
1060         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1061         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1062         repbody->eadatasize = 0;
1063         repbody->aclsize = 0;
1064
1065         rc = mdt_init_ucred(info, reqbody);
1066         if (unlikely(rc))
1067                 GOTO(out_shrink, rc);
1068
1069         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1070         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1071                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1072                 lhc->mlh_reg_lh.cookie = 0;
1073         }
1074         mdt_exit_ucred(info);
1075         EXIT;
1076 out_shrink:
1077         mdt_shrink_reply(info);
1078         return rc;
1079 }
1080
1081 static const struct lu_device_operations mdt_lu_ops;
1082
1083 static int lu_device_is_mdt(struct lu_device *d)
1084 {
1085         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1086 }
1087
1088 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1089                          void *karg, void *uarg);
1090
1091 static int mdt_set_info(struct mdt_thread_info *info)
1092 {
1093         struct ptlrpc_request *req = mdt_info_req(info);
1094         char *key;
1095         void *val;
1096         int keylen, vallen, rc = 0;
1097         ENTRY;
1098
1099         rc = req_capsule_server_pack(info->mti_pill);
1100         if (rc)
1101                 RETURN(rc);
1102
1103         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1104         if (key == NULL) {
1105                 DEBUG_REQ(D_HA, req, "no set_info key");
1106                 RETURN(-EFAULT);
1107         }
1108
1109         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1110                                       RCL_CLIENT);
1111
1112         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1113         if (val == NULL) {
1114                 DEBUG_REQ(D_HA, req, "no set_info val");
1115                 RETURN(-EFAULT);
1116         }
1117
1118         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1119                                       RCL_CLIENT);
1120
1121         /* Swab any part of val you need to here */
1122         if (KEY_IS(KEY_READ_ONLY)) {
1123                 req->rq_status = 0;
1124                 lustre_msg_set_status(req->rq_repmsg, 0);
1125
1126                 cfs_spin_lock(&req->rq_export->exp_lock);
1127                 if (*(__u32 *)val)
1128                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1129                 else
1130                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1131                 cfs_spin_unlock(&req->rq_export->exp_lock);
1132
1133         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1134                 struct changelog_setinfo *cs =
1135                         (struct changelog_setinfo *)val;
1136                 if (vallen != sizeof(*cs)) {
1137                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1138                         RETURN(-EINVAL);
1139                 }
1140                 if (ptlrpc_req_need_swab(req)) {
1141                         __swab64s(&cs->cs_recno);
1142                         __swab32s(&cs->cs_id);
1143                 }
1144
1145                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1146                                    vallen, val, NULL);
1147                 lustre_msg_set_status(req->rq_repmsg, rc);
1148
1149         } else {
1150                 RETURN(-EINVAL);
1151         }
1152         RETURN(0);
1153 }
1154
1155 static int mdt_connect(struct mdt_thread_info *info)
1156 {
1157         int rc;
1158         struct ptlrpc_request *req;
1159
1160         req = mdt_info_req(info);
1161         rc = target_handle_connect(req);
1162         if (rc == 0) {
1163                 LASSERT(req->rq_export != NULL);
1164                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1165                 rc = mdt_init_sec_level(info);
1166                 if (rc == 0)
1167                         rc = mdt_init_idmap(info);
1168                 if (rc != 0)
1169                         obd_disconnect(class_export_get(req->rq_export));
1170         } else {
1171                 rc = err_serious(rc);
1172         }
1173         return rc;
1174 }
1175
1176 static int mdt_disconnect(struct mdt_thread_info *info)
1177 {
1178         int rc;
1179         ENTRY;
1180
1181         rc = target_handle_disconnect(mdt_info_req(info));
1182         if (rc)
1183                 rc = err_serious(rc);
1184         RETURN(rc);
1185 }
1186
1187 static int mdt_sendpage(struct mdt_thread_info *info,
1188                         struct lu_rdpg *rdpg)
1189 {
1190         struct ptlrpc_request   *req = mdt_info_req(info);
1191         struct obd_export       *exp = req->rq_export;
1192         struct ptlrpc_bulk_desc *desc;
1193         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1194         int                      tmpcount;
1195         int                      tmpsize;
1196         int                      timeout;
1197         int                      i;
1198         int                      rc;
1199         ENTRY;
1200
1201         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1202                                     MDS_BULK_PORTAL);
1203         if (desc == NULL)
1204                 RETURN(-ENOMEM);
1205
1206         for (i = 0, tmpcount = rdpg->rp_count;
1207                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1208                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1209                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1210         }
1211
1212         LASSERT(desc->bd_nob == rdpg->rp_count);
1213         rc = sptlrpc_svc_wrap_bulk(req, desc);
1214         if (rc)
1215                 GOTO(free_desc, rc);
1216
1217         rc = ptlrpc_start_bulk_transfer(desc);
1218         if (rc)
1219                 GOTO(free_desc, rc);
1220
1221         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1222                 GOTO(abort_bulk, rc = 0);
1223
1224         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1225         if (timeout < 0)
1226                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1227                        req->rq_deadline, cfs_time_current_sec());
1228         *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1229                                     cfs_time_seconds(1), NULL, NULL);
1230         rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
1231                           exp->exp_failed || exp->exp_abort_active_req, lwi);
1232         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1233
1234         if (rc == 0) {
1235                 if (desc->bd_success &&
1236                     desc->bd_nob_transferred == rdpg->rp_count)
1237                         GOTO(free_desc, rc);
1238
1239                 rc = -ETIMEDOUT;
1240                 if (exp->exp_abort_active_req || exp->exp_failed)
1241                         GOTO(abort_bulk, rc);
1242         }
1243
1244         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1245                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1246                   desc->bd_nob_transferred, rdpg->rp_count,
1247                   exp->exp_client_uuid.uuid,
1248                   exp->exp_connection->c_remote_uuid.uuid);
1249
1250         class_fail_export(exp);
1251
1252         EXIT;
1253 abort_bulk:
1254         ptlrpc_abort_bulk(desc);
1255 free_desc:
1256         ptlrpc_free_bulk(desc);
1257         return rc;
1258 }
1259
1260 #ifdef HAVE_SPLIT_SUPPORT
1261 /*
1262  * Retrieve dir entry from the page and insert it to the slave object, actually,
1263  * this should be in osd layer, but since it will not in the final product, so
1264  * just do it here and do not define more moo api anymore for this.
1265  */
1266 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1267                               int size)
1268 {
1269         struct mdt_object *object = info->mti_object;
1270         struct lu_fid *lf = &info->mti_tmp_fid2;
1271         struct md_attr *ma = &info->mti_attr;
1272         struct lu_dirpage *dp;
1273         struct lu_dirent *ent;
1274         int rc = 0, offset = 0;
1275         ENTRY;
1276
1277         /* Make sure we have at least one entry. */
1278         if (size == 0)
1279                 RETURN(-EINVAL);
1280
1281         /*
1282          * Disable trans for this name insert, since it will include many trans
1283          * for this.
1284          */
1285         info->mti_no_need_trans = 1;
1286         /*
1287          * When write_dir_page, no need update parent's ctime,
1288          * and no permission check for name_insert.
1289          */
1290         ma->ma_attr.la_ctime = 0;
1291         ma->ma_attr.la_valid = LA_MODE;
1292         ma->ma_valid = MA_INODE;
1293
1294         cfs_kmap(page);
1295         dp = page_address(page);
1296         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1297
1298         for (ent = lu_dirent_start(dp); ent != NULL;
1299              ent = lu_dirent_next(ent)) {
1300                 struct lu_name *lname;
1301                 char *name;
1302
1303                 if (le16_to_cpu(ent->lde_namelen) == 0)
1304                         continue;
1305
1306                 fid_le_to_cpu(lf, &ent->lde_fid);
1307                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1308                         ma->ma_attr.la_mode = S_IFDIR;
1309                 else
1310                         ma->ma_attr.la_mode = 0;
1311                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1312                 if (name == NULL)
1313                         GOTO(out, rc = -ENOMEM);
1314
1315                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1316                 lname = mdt_name(info->mti_env, name,
1317                                  le16_to_cpu(ent->lde_namelen));
1318                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1319                 rc = mdo_name_insert(info->mti_env,
1320                                      md_object_next(&object->mot_obj),
1321                                      lname, lf, ma);
1322                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1323                 if (rc) {
1324                         CERROR("Can't insert %*.*s, rc %d\n",
1325                                le16_to_cpu(ent->lde_namelen),
1326                                le16_to_cpu(ent->lde_namelen),
1327                                ent->lde_name, rc);
1328                         GOTO(out, rc);
1329                 }
1330
1331                 offset += lu_dirent_size(ent);
1332                 if (offset >= size)
1333                         break;
1334         }
1335         EXIT;
1336 out:
1337         cfs_kunmap(page);
1338         return rc;
1339 }
1340
1341 static int mdt_bulk_timeout(void *data)
1342 {
1343         ENTRY;
1344
1345         CERROR("mdt bulk transfer timeout \n");
1346
1347         RETURN(1);
1348 }
1349
1350 static int mdt_writepage(struct mdt_thread_info *info)
1351 {
1352         struct ptlrpc_request   *req = mdt_info_req(info);
1353         struct mdt_body         *reqbody;
1354         struct l_wait_info      *lwi;
1355         struct ptlrpc_bulk_desc *desc;
1356         struct page             *page;
1357         int                rc;
1358         ENTRY;
1359
1360
1361         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1362         if (reqbody == NULL)
1363                 RETURN(err_serious(-EFAULT));
1364
1365         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1366         if (desc == NULL)
1367                 RETURN(err_serious(-ENOMEM));
1368
1369         /* allocate the page for the desc */
1370         page = cfs_alloc_page(CFS_ALLOC_STD);
1371         if (page == NULL)
1372                 GOTO(desc_cleanup, rc = -ENOMEM);
1373
1374         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1375                (int)reqbody->size, (int)reqbody->nlink);
1376
1377         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1378                               (int)reqbody->nlink);
1379
1380         rc = sptlrpc_svc_prep_bulk(req, desc);
1381         if (rc != 0)
1382                 GOTO(cleanup_page, rc);
1383         /*
1384          * Check if client was evicted while we were doing i/o before touching
1385          * network.
1386          */
1387         OBD_ALLOC_PTR(lwi);
1388         if (!lwi)
1389                 GOTO(cleanup_page, rc = -ENOMEM);
1390
1391         if (desc->bd_export->exp_failed)
1392                 rc = -ENOTCONN;
1393         else
1394                 rc = ptlrpc_start_bulk_transfer (desc);
1395         if (rc == 0) {
1396                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1397                                             mdt_bulk_timeout, desc);
1398                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1399                                   desc->bd_export->exp_failed, lwi);
1400                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1401                 if (rc == -ETIMEDOUT) {
1402                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1403                         ptlrpc_abort_bulk(desc);
1404                 } else if (desc->bd_export->exp_failed) {
1405                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1406                         rc = -ENOTCONN;
1407                         ptlrpc_abort_bulk(desc);
1408                 } else if (!desc->bd_success ||
1409                            desc->bd_nob_transferred != desc->bd_nob) {
1410                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1411                                   desc->bd_success ?
1412                                   "truncated" : "network error on",
1413                                   desc->bd_nob_transferred, desc->bd_nob);
1414                         /* XXX should this be a different errno? */
1415                         rc = -ETIMEDOUT;
1416                 }
1417         } else {
1418                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1419         }
1420         if (rc)
1421                 GOTO(cleanup_lwi, rc);
1422         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1423
1424 cleanup_lwi:
1425         OBD_FREE_PTR(lwi);
1426 cleanup_page:
1427         cfs_free_page(page);
1428 desc_cleanup:
1429         ptlrpc_free_bulk(desc);
1430         RETURN(rc);
1431 }
1432 #endif
1433
1434 static int mdt_readpage(struct mdt_thread_info *info)
1435 {
1436         struct mdt_object *object = info->mti_object;
1437         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1438         struct mdt_body   *reqbody;
1439         struct mdt_body   *repbody;
1440         int                rc;
1441         int                i;
1442         ENTRY;
1443
1444         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1445                 RETURN(err_serious(-ENOMEM));
1446
1447         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1448         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1449         if (reqbody == NULL || repbody == NULL)
1450                 RETURN(err_serious(-EFAULT));
1451
1452         /*
1453          * prepare @rdpg before calling lower layers and transfer itself. Here
1454          * reqbody->size contains offset of where to start to read and
1455          * reqbody->nlink contains number bytes to read.
1456          */
1457         rdpg->rp_hash = reqbody->size;
1458         if (rdpg->rp_hash != reqbody->size) {
1459                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1460                        rdpg->rp_hash, reqbody->size);
1461                 RETURN(-EFAULT);
1462         }
1463
1464         rdpg->rp_attrs = reqbody->mode;
1465         rdpg->rp_count  = reqbody->nlink;
1466         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1467         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1468         if (rdpg->rp_pages == NULL)
1469                 RETURN(-ENOMEM);
1470
1471         for (i = 0; i < rdpg->rp_npages; ++i) {
1472                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1473                 if (rdpg->rp_pages[i] == NULL)
1474                         GOTO(free_rdpg, rc = -ENOMEM);
1475         }
1476
1477         /* call lower layers to fill allocated pages with directory data */
1478         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1479         if (rc)
1480                 GOTO(free_rdpg, rc);
1481
1482         /* send pages to client */
1483         rc = mdt_sendpage(info, rdpg);
1484
1485         EXIT;
1486 free_rdpg:
1487
1488         for (i = 0; i < rdpg->rp_npages; i++)
1489                 if (rdpg->rp_pages[i] != NULL)
1490                         cfs_free_page(rdpg->rp_pages[i]);
1491         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1492
1493         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1494                 RETURN(0);
1495
1496         return rc;
1497 }
1498
1499 static int mdt_reint_internal(struct mdt_thread_info *info,
1500                               struct mdt_lock_handle *lhc,
1501                               __u32 op)
1502 {
1503         struct req_capsule      *pill = info->mti_pill;
1504         struct mdt_device       *mdt = info->mti_mdt;
1505         struct md_quota         *mq = md_quota(info->mti_env);
1506         struct mdt_body         *repbody;
1507         int                      rc = 0;
1508         ENTRY;
1509
1510         /* pack reply */
1511         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1512                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1513                                      mdt->mdt_max_mdsize);
1514         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1515                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1516                                      mdt->mdt_max_cookiesize);
1517
1518         rc = req_capsule_server_pack(pill);
1519         if (rc != 0) {
1520                 CERROR("Can't pack response, rc %d\n", rc);
1521                 RETURN(err_serious(rc));
1522         }
1523
1524         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1525                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1526                 LASSERT(repbody);
1527                 repbody->eadatasize = 0;
1528                 repbody->aclsize = 0;
1529         }
1530
1531         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1532                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1533
1534         rc = mdt_reint_unpack(info, op);
1535         if (rc != 0) {
1536                 CERROR("Can't unpack reint, rc %d\n", rc);
1537                 GOTO(out_shrink, rc = err_serious(rc));
1538         }
1539
1540         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1541
1542         /* for replay no cookkie / lmm need, because client have this already */
1543         if (info->mti_spec.no_create == 1)  {
1544                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1545                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1546
1547                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1548                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1549                                              0);
1550         }
1551
1552         rc = mdt_init_ucred_reint(info);
1553         if (rc)
1554                 GOTO(out_shrink, rc);
1555
1556         rc = mdt_fix_attr_ucred(info, op);
1557         if (rc != 0)
1558                 GOTO(out_ucred, rc = err_serious(rc));
1559
1560         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1561                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1562                 GOTO(out_ucred, rc);
1563         }
1564         mq->mq_exp = info->mti_exp;
1565         rc = mdt_reint_rec(info, lhc);
1566         EXIT;
1567 out_ucred:
1568         mdt_exit_ucred(info);
1569 out_shrink:
1570         mdt_shrink_reply(info);
1571         return rc;
1572 }
1573
1574 static long mdt_reint_opcode(struct mdt_thread_info *info,
1575                              const struct req_format **fmt)
1576 {
1577         struct mdt_rec_reint *rec;
1578         long opc;
1579
1580         opc = err_serious(-EFAULT);
1581         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1582         if (rec != NULL) {
1583                 opc = rec->rr_opcode;
1584                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1585                 if (opc < REINT_MAX && fmt[opc] != NULL)
1586                         req_capsule_extend(info->mti_pill, fmt[opc]);
1587                 else {
1588                         CERROR("Unsupported opc: %ld\n", opc);
1589                         opc = err_serious(opc);
1590                 }
1591         }
1592         return opc;
1593 }
1594
1595 static int mdt_reint(struct mdt_thread_info *info)
1596 {
1597         long opc;
1598         int  rc;
1599
1600         static const struct req_format *reint_fmts[REINT_MAX] = {
1601                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1602                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1603                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1604                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1605                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1606                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1607                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1608         };
1609
1610         ENTRY;
1611
1612         opc = mdt_reint_opcode(info, reint_fmts);
1613         if (opc >= 0) {
1614                 /*
1615                  * No lock possible here from client to pass it to reint code
1616                  * path.
1617                  */
1618                 rc = mdt_reint_internal(info, NULL, opc);
1619         } else {
1620                 rc = opc;
1621         }
1622
1623         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1624         RETURN(rc);
1625 }
1626
1627 /* this should sync the whole device */
1628 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1629 {
1630         struct dt_device *dt = mdt->mdt_bottom;
1631         int rc;
1632         ENTRY;
1633
1634         rc = dt->dd_ops->dt_sync(env, dt);
1635         RETURN(rc);
1636 }
1637
1638 /* this should sync this object */
1639 static int mdt_object_sync(struct mdt_thread_info *info)
1640 {
1641         struct md_object *next;
1642         int rc;
1643         ENTRY;
1644
1645         if (!mdt_object_exists(info->mti_object)) {
1646                 CWARN("Non existing object  "DFID"!\n",
1647                       PFID(mdt_object_fid(info->mti_object)));
1648                 RETURN(-ESTALE);
1649         }
1650         next = mdt_object_child(info->mti_object);
1651         rc = mo_object_sync(info->mti_env, next);
1652
1653         RETURN(rc);
1654 }
1655
1656 static int mdt_sync(struct mdt_thread_info *info)
1657 {
1658         struct req_capsule *pill = info->mti_pill;
1659         struct mdt_body *body;
1660         int rc;
1661         ENTRY;
1662
1663         /* The fid may be zero, so we req_capsule_set manually */
1664         req_capsule_set(pill, &RQF_MDS_SYNC);
1665
1666         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1667         if (body == NULL)
1668                 RETURN(err_serious(-EINVAL));
1669
1670         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1671                 RETURN(err_serious(-ENOMEM));
1672
1673         if (fid_seq(&body->fid1) == 0) {
1674                 /* sync the whole device */
1675                 rc = req_capsule_server_pack(pill);
1676                 if (rc == 0)
1677                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1678                 else
1679                         rc = err_serious(rc);
1680         } else {
1681                 /* sync an object */
1682                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1683                 if (rc == 0) {
1684                         rc = mdt_object_sync(info);
1685                         if (rc == 0) {
1686                                 struct md_object *next;
1687                                 const struct lu_fid *fid;
1688                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1689
1690                                 next = mdt_object_child(info->mti_object);
1691                                 info->mti_attr.ma_need = MA_INODE;
1692                                 info->mti_attr.ma_valid = 0;
1693                                 rc = mo_attr_get(info->mti_env, next,
1694                                                  &info->mti_attr);
1695                                 if (rc == 0) {
1696                                         body = req_capsule_server_get(pill,
1697                                                                 &RMF_MDT_BODY);
1698                                         fid = mdt_object_fid(info->mti_object);
1699                                         mdt_pack_attr2body(info, body, la, fid);
1700                                 }
1701                         }
1702                 } else
1703                         rc = err_serious(rc);
1704         }
1705         RETURN(rc);
1706 }
1707
1708 #ifdef HAVE_QUOTA_SUPPORT
1709 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1710 {
1711         struct obd_quotactl *oqctl;
1712         struct req_capsule *pill = info->mti_pill;
1713         struct obd_export *exp = info->mti_exp;
1714         struct md_quota *mq = md_quota(info->mti_env);
1715         struct md_device *next = info->mti_mdt->mdt_child;
1716         int rc;
1717         ENTRY;
1718
1719         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1720         if (oqctl == NULL)
1721                 RETURN(-EPROTO);
1722
1723         /* remote client has no permission for quotacheck */
1724         if (unlikely(exp_connect_rmtclient(exp)))
1725                 RETURN(-EPERM);
1726
1727         rc = req_capsule_server_pack(pill);
1728         if (rc)
1729                 RETURN(rc);
1730
1731         mq->mq_exp = exp;
1732         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1733                                                oqctl->qc_type);
1734         RETURN(rc);
1735 }
1736
1737 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1738 {
1739         struct obd_quotactl *oqctl, *repoqc;
1740         struct req_capsule *pill = info->mti_pill;
1741         struct obd_export *exp = info->mti_exp;
1742         struct md_quota *mq = md_quota(info->mti_env);
1743         struct md_device *next = info->mti_mdt->mdt_child;
1744         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1745         int id, rc;
1746         ENTRY;
1747
1748         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1749         if (oqctl == NULL)
1750                 RETURN(-EPROTO);
1751
1752         id = oqctl->qc_id;
1753         if (exp_connect_rmtclient(exp)) {
1754                 struct ptlrpc_request *req = mdt_info_req(info);
1755                 struct mdt_export_data *med = mdt_req2med(req);
1756                 struct lustre_idmap_table *idmap = med->med_idmap;
1757
1758                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1759                              oqctl->qc_cmd != Q_GETINFO))
1760                         RETURN(-EPERM);
1761
1762
1763                 if (oqctl->qc_type == USRQUOTA)
1764                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1765                                                      oqctl->qc_id);
1766                 else if (oqctl->qc_type == GRPQUOTA)
1767                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1768                                                      oqctl->qc_id);
1769                 else
1770                         RETURN(-EINVAL);
1771
1772                 if (id == CFS_IDMAP_NOTFOUND) {
1773                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1774                                oqctl->qc_id);
1775                         RETURN(-EACCES);
1776                 }
1777         }
1778
1779         rc = req_capsule_server_pack(pill);
1780         if (rc)
1781                 RETURN(rc);
1782
1783         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1784         LASSERT(repoqc != NULL);
1785
1786         mq->mq_exp = exp;
1787         switch (oqctl->qc_cmd) {
1788         case Q_QUOTAON:
1789                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1790                 break;
1791         case Q_QUOTAOFF:
1792                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1793                 break;
1794         case Q_SETINFO:
1795                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1796                                       &oqctl->qc_dqinfo);
1797                 break;
1798         case Q_GETINFO:
1799                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1800                                       &oqctl->qc_dqinfo);
1801                 break;
1802         case Q_SETQUOTA:
1803                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1804                                        &oqctl->qc_dqblk);
1805                 break;
1806         case Q_GETQUOTA:
1807                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1808                                        &oqctl->qc_dqblk);
1809                 break;
1810         case Q_GETOINFO:
1811                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1812                                        &oqctl->qc_dqinfo);
1813                 break;
1814         case Q_GETOQUOTA:
1815                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1816                                         &oqctl->qc_dqblk);
1817                 break;
1818         case LUSTRE_Q_INVALIDATE:
1819                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1820                 break;
1821         case LUSTRE_Q_FINVALIDATE:
1822                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1823                 break;
1824         default:
1825                 CERROR("unsupported mdt_quotactl command: %d\n",
1826                        oqctl->qc_cmd);
1827                 RETURN(-EFAULT);
1828         }
1829
1830         *repoqc = *oqctl;
1831         RETURN(rc);
1832 }
1833 #endif
1834
1835
1836 /*
1837  * OBD PING and other handlers.
1838  */
1839 static int mdt_obd_ping(struct mdt_thread_info *info)
1840 {
1841         int rc;
1842         ENTRY;
1843
1844         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1845
1846         rc = target_handle_ping(mdt_info_req(info));
1847         if (rc < 0)
1848                 rc = err_serious(rc);
1849         RETURN(rc);
1850 }
1851
1852 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1853 {
1854         return err_serious(-EOPNOTSUPP);
1855 }
1856
1857 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1858 {
1859         return err_serious(-EOPNOTSUPP);
1860 }
1861
1862
1863 /*
1864  * LLOG handlers.
1865  */
1866
1867 /** clone llog ctxt from child (mdd)
1868  * This allows remote llog (replicator) access.
1869  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1870  * context was originally set up, or we can handle them directly.
1871  * I choose the latter, but that means I need any llog
1872  * contexts set up by child to be accessable by the mdt.  So we clone the
1873  * context into our context list here.
1874  */
1875 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1876                                int idx)
1877 {
1878         struct md_device  *next = mdt->mdt_child;
1879         struct llog_ctxt *ctxt;
1880         int rc;
1881
1882         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1883                 return 0;
1884
1885         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1886         if (rc || ctxt == NULL) {
1887                 CERROR("Can't get mdd ctxt %d\n", rc);
1888                 return rc;
1889         }
1890
1891         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1892         if (rc)
1893                 CERROR("Can't set mdt ctxt %d\n", rc);
1894
1895         return rc;
1896 }
1897
1898 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1899                                  struct mdt_device *mdt, int idx)
1900 {
1901         struct llog_ctxt *ctxt;
1902
1903         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1904         if (ctxt == NULL)
1905                 return 0;
1906         /* Put once for the get we just did, and once for the clone */
1907         llog_ctxt_put(ctxt);
1908         llog_ctxt_put(ctxt);
1909         return 0;
1910 }
1911
1912 static int mdt_llog_create(struct mdt_thread_info *info)
1913 {
1914         int rc;
1915
1916         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1917         rc = llog_origin_handle_create(mdt_info_req(info));
1918         return (rc < 0 ? err_serious(rc) : rc);
1919 }
1920
1921 static int mdt_llog_destroy(struct mdt_thread_info *info)
1922 {
1923         int rc;
1924
1925         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1926         rc = llog_origin_handle_destroy(mdt_info_req(info));
1927         return (rc < 0 ? err_serious(rc) : rc);
1928 }
1929
1930 static int mdt_llog_read_header(struct mdt_thread_info *info)
1931 {
1932         int rc;
1933
1934         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1935         rc = llog_origin_handle_read_header(mdt_info_req(info));
1936         return (rc < 0 ? err_serious(rc) : rc);
1937 }
1938
1939 static int mdt_llog_next_block(struct mdt_thread_info *info)
1940 {
1941         int rc;
1942
1943         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1944         rc = llog_origin_handle_next_block(mdt_info_req(info));
1945         return (rc < 0 ? err_serious(rc) : rc);
1946 }
1947
1948 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1949 {
1950         int rc;
1951
1952         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1953         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1954         return (rc < 0 ? err_serious(rc) : rc);
1955 }
1956
1957
1958 /*
1959  * DLM handlers.
1960  */
1961 static struct ldlm_callback_suite cbs = {
1962         .lcs_completion = ldlm_server_completion_ast,
1963         .lcs_blocking   = ldlm_server_blocking_ast,
1964         .lcs_glimpse    = NULL
1965 };
1966
1967 static int mdt_enqueue(struct mdt_thread_info *info)
1968 {
1969         struct ptlrpc_request *req;
1970         __u64 req_bits;
1971         int rc;
1972
1973         /*
1974          * info->mti_dlm_req already contains swapped and (if necessary)
1975          * converted dlm request.
1976          */
1977         LASSERT(info->mti_dlm_req != NULL);
1978
1979         req = mdt_info_req(info);
1980
1981         /*
1982          * Lock without inodebits makes no sense and will oops later in
1983          * ldlm. Let's check it now to see if we have wrong lock from client or
1984          * bits get corrupted somewhere in mdt_intent_policy().
1985          */
1986         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1987         /* This is disabled because we need to support liblustre flock.
1988          * LASSERT(req_bits != 0);
1989          */
1990
1991         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1992                                   req, info->mti_dlm_req, &cbs);
1993         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1994         return rc ? err_serious(rc) : req->rq_status;
1995 }
1996
1997 static int mdt_convert(struct mdt_thread_info *info)
1998 {
1999         int rc;
2000         struct ptlrpc_request *req;
2001
2002         LASSERT(info->mti_dlm_req);
2003         req = mdt_info_req(info);
2004         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2005         return rc ? err_serious(rc) : req->rq_status;
2006 }
2007
2008 static int mdt_bl_callback(struct mdt_thread_info *info)
2009 {
2010         CERROR("bl callbacks should not happen on MDS\n");
2011         LBUG();
2012         return err_serious(-EOPNOTSUPP);
2013 }
2014
2015 static int mdt_cp_callback(struct mdt_thread_info *info)
2016 {
2017         CERROR("cp callbacks should not happen on MDS\n");
2018         LBUG();
2019         return err_serious(-EOPNOTSUPP);
2020 }
2021
2022 /*
2023  * sec context handlers
2024  */
2025 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2026 {
2027         int rc;
2028
2029         rc = mdt_handle_idmap(info);
2030
2031         if (unlikely(rc)) {
2032                 struct ptlrpc_request *req = mdt_info_req(info);
2033                 __u32                  opc;
2034
2035                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2036                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2037                         sptlrpc_svc_ctx_invalidate(req);
2038         }
2039
2040         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2041
2042         return rc;
2043 }
2044
2045 static struct mdt_object *mdt_obj(struct lu_object *o)
2046 {
2047         LASSERT(lu_device_is_mdt(o->lo_dev));
2048         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2049 }
2050
2051 struct mdt_object *mdt_object_find(const struct lu_env *env,
2052                                    struct mdt_device *d,
2053                                    const struct lu_fid *f)
2054 {
2055         struct lu_object *o;
2056         struct mdt_object *m;
2057         ENTRY;
2058
2059         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2060         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2061         if (unlikely(IS_ERR(o)))
2062                 m = (struct mdt_object *)o;
2063         else
2064                 m = mdt_obj(o);
2065         RETURN(m);
2066 }
2067
2068 /**
2069  * Asyncronous commit for mdt device.
2070  *
2071  * Pass asynchonous commit call down the MDS stack.
2072  *
2073  * \param env environment
2074  * \param mdt the mdt device
2075  */
2076 static void mdt_device_commit_async(const struct lu_env *env,
2077                                     struct mdt_device *mdt)
2078 {
2079         struct dt_device *dt = mdt->mdt_bottom;
2080         int rc;
2081
2082         rc = dt->dd_ops->dt_commit_async(env, dt);
2083         if (unlikely(rc != 0))
2084                 CWARN("async commit start failed with rc = %d", rc);
2085 }
2086
2087 /**
2088  * Mark the lock as "synchonous".
2089  *
2090  * Mark the lock to deffer transaction commit to the unlock time.
2091  *
2092  * \param lock the lock to mark as "synchonous"
2093  *
2094  * \see mdt_is_lock_sync
2095  * \see mdt_save_lock
2096  */
2097 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2098 {
2099         lock->l_ast_data = (void*)1;
2100 }
2101
2102 /**
2103  * Check whehter the lock "synchonous" or not.
2104  *
2105  * \param lock the lock to check
2106  * \retval 1 the lock is "synchonous"
2107  * \retval 0 the lock isn't "synchronous"
2108  *
2109  * \see mdt_set_lock_sync
2110  * \see mdt_save_lock
2111  */
2112 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2113 {
2114         return lock->l_ast_data != NULL;
2115 }
2116
2117 /**
2118  * Blocking AST for mdt locks.
2119  *
2120  * Starts transaction commit if in case of COS lock conflict or
2121  * deffers such a commit to the mdt_save_lock.
2122  *
2123  * \param lock the lock which blocks a request or cancelling lock
2124  * \param desc unused
2125  * \param data unused
2126  * \param flag indicates whether this cancelling or blocking callback
2127  * \retval 0
2128  * \see ldlm_blocking_ast_nocheck
2129  */
2130 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2131                      void *data, int flag)
2132 {
2133         struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2134         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2135         int rc;
2136         ENTRY;
2137
2138         if (flag == LDLM_CB_CANCELING)
2139                 RETURN(0);
2140         lock_res_and_lock(lock);
2141         if (lock->l_blocking_ast != mdt_blocking_ast) {
2142                 unlock_res_and_lock(lock);
2143                 RETURN(0);
2144         }
2145         if (mdt_cos_is_enabled(mdt) &&
2146             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2147             lock->l_blocking_lock != NULL &&
2148             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2149                 mdt_set_lock_sync(lock);
2150         }
2151         rc = ldlm_blocking_ast_nocheck(lock);
2152
2153         /* There is no lock conflict if l_blocking_lock == NULL,
2154          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2155          * when the last reference to a local lock was released */
2156         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2157                 struct lu_env env;
2158
2159                 rc = lu_env_init(&env, LCT_MD_THREAD);
2160                 if (unlikely(rc != 0))
2161                         CWARN("lu_env initialization failed with rc = %d,"
2162                               "cannot start asynchronous commit\n", rc);
2163                 else
2164                         mdt_device_commit_async(&env, mdt);
2165                 lu_env_fini(&env);
2166         }
2167         RETURN(rc);
2168 }
2169
2170 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2171                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2172 {
2173         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2174         ldlm_policy_data_t *policy = &info->mti_policy;
2175         struct ldlm_res_id *res_id = &info->mti_res_id;
2176         int rc;
2177         ENTRY;
2178
2179         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2180         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2181         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2182         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2183
2184         if (mdt_object_exists(o) < 0) {
2185                 if (locality == MDT_CROSS_LOCK) {
2186                         /* cross-ref object fix */
2187                         ibits &= ~MDS_INODELOCK_UPDATE;
2188                         ibits |= MDS_INODELOCK_LOOKUP;
2189                 } else {
2190                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2191                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2192                 }
2193                 /* No PDO lock on remote object */
2194                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2195         }
2196
2197         if (lh->mlh_type == MDT_PDO_LOCK) {
2198                 /* check for exists after object is locked */
2199                 if (mdt_object_exists(o) == 0) {
2200                         /* Non-existent object shouldn't have PDO lock */
2201                         RETURN(-ESTALE);
2202                 } else {
2203                         /* Non-dir object shouldn't have PDO lock */
2204                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2205                 }
2206         }
2207
2208         memset(policy, 0, sizeof(*policy));
2209         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2210
2211         /*
2212          * Take PDO lock on whole directory and build correct @res_id for lock
2213          * on part of directory.
2214          */
2215         if (lh->mlh_pdo_hash != 0) {
2216                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2217                 mdt_lock_pdo_mode(info, o, lh);
2218                 if (lh->mlh_pdo_mode != LCK_NL) {
2219                         /*
2220                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2221                          * is never going to be sent to client and we do not
2222                          * want it slowed down due to possible cancels.
2223                          */
2224                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2225                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2226                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2227                                           &info->mti_exp->exp_handle.h_cookie);
2228                         if (unlikely(rc))
2229                                 RETURN(rc);
2230                 }
2231
2232                 /*
2233                  * Finish res_id initializing by name hash marking part of
2234                  * directory which is taking modification.
2235                  */
2236                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2237         }
2238
2239         policy->l_inodebits.bits = ibits;
2240
2241         /*
2242          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2243          * going to be sent to client. If it is - mdt_intent_policy() path will
2244          * fix it up and turn FL_LOCAL flag off.
2245          */
2246         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2247                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2248                           &info->mti_exp->exp_handle.h_cookie);
2249         if (rc)
2250                 GOTO(out, rc);
2251
2252 out:
2253         if (rc)
2254                 mdt_object_unlock(info, o, lh, 1);
2255
2256
2257         RETURN(rc);
2258 }
2259
2260 /**
2261  * Save a lock within request object.
2262  *
2263  * Keep the lock referenced until whether client ACK or transaction
2264  * commit happens or release the lock immediately depending on input
2265  * parameters. If COS is ON, a write lock is converted to COS lock
2266  * before saving.
2267  *
2268  * \param info thead info object
2269  * \param h lock handle
2270  * \param mode lock mode
2271  * \param decref force immediate lock releasing
2272  */
2273 static
2274 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2275                    ldlm_mode_t mode, int decref)
2276 {
2277         ENTRY;
2278
2279         if (lustre_handle_is_used(h)) {
2280                 if (decref || !info->mti_has_trans ||
2281                     !(mode & (LCK_PW | LCK_EX))){
2282                         mdt_fid_unlock(h, mode);
2283                 } else {
2284                         struct mdt_device *mdt = info->mti_mdt;
2285                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2286                         struct ptlrpc_request *req = mdt_info_req(info);
2287                         int no_ack = 0;
2288
2289                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2290                                  h->cookie);
2291                         CDEBUG(D_HA, "request = %p reply state = %p"
2292                                " transno = "LPD64"\n",
2293                                req, req->rq_reply_state, req->rq_transno);
2294                         if (mdt_cos_is_enabled(mdt)) {
2295                                 no_ack = 1;
2296                                 ldlm_lock_downgrade(lock, LCK_COS);
2297                                 mode = LCK_COS;
2298                         }
2299                         ptlrpc_save_lock(req, h, mode, no_ack);
2300                         if (mdt_is_lock_sync(lock)) {
2301                                 CDEBUG(D_HA, "found sync-lock,"
2302                                        " async commit started\n");
2303                                 mdt_device_commit_async(info->mti_env,
2304                                                         mdt);
2305                         }
2306                         LDLM_LOCK_PUT(lock);
2307                 }
2308                 h->cookie = 0ull;
2309         }
2310
2311         EXIT;
2312 }
2313
2314 /**
2315  * Unlock mdt object.
2316  *
2317  * Immeditely release the regular lock and the PDO lock or save the
2318  * lock in reqeuest and keep them referenced until client ACK or
2319  * transaction commit.
2320  *
2321  * \param info thread info object
2322  * \param o mdt object
2323  * \param lh mdt lock handle referencing regular and PDO locks
2324  * \param decref force immediate lock releasing
2325  */
2326 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2327                        struct mdt_lock_handle *lh, int decref)
2328 {
2329         ENTRY;
2330
2331         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2332         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2333
2334         EXIT;
2335 }
2336
2337 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2338                                         const struct lu_fid *f,
2339                                         struct mdt_lock_handle *lh,
2340                                         __u64 ibits)
2341 {
2342         struct mdt_object *o;
2343
2344         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2345         if (!IS_ERR(o)) {
2346                 int rc;
2347
2348                 rc = mdt_object_lock(info, o, lh, ibits,
2349                                      MDT_LOCAL_LOCK);
2350                 if (rc != 0) {
2351                         mdt_object_put(info->mti_env, o);
2352                         o = ERR_PTR(rc);
2353                 }
2354         }
2355         return o;
2356 }
2357
2358 void mdt_object_unlock_put(struct mdt_thread_info * info,
2359                            struct mdt_object * o,
2360                            struct mdt_lock_handle *lh,
2361                            int decref)
2362 {
2363         mdt_object_unlock(info, o, lh, decref);
2364         mdt_object_put(info->mti_env, o);
2365 }
2366
2367 static struct mdt_handler *mdt_handler_find(__u32 opc,
2368                                             struct mdt_opc_slice *supported)
2369 {
2370         struct mdt_opc_slice *s;
2371         struct mdt_handler   *h;
2372
2373         h = NULL;
2374         for (s = supported; s->mos_hs != NULL; s++) {
2375                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2376                         h = s->mos_hs + (opc - s->mos_opc_start);
2377                         if (likely(h->mh_opc != 0))
2378                                 LASSERTF(h->mh_opc == opc,
2379                                          "opcode mismatch %d != %d\n",
2380                                          h->mh_opc, opc);
2381                         else
2382                                 h = NULL; /* unsupported opc */
2383                         break;
2384                 }
2385         }
2386         return h;
2387 }
2388
2389 static int mdt_lock_resname_compat(struct mdt_device *m,
2390                                    struct ldlm_request *req)
2391 {
2392         /* XXX something... later. */
2393         return 0;
2394 }
2395
2396 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2397 {
2398         /* XXX something... later. */
2399         return 0;
2400 }
2401
2402 /*
2403  * Generic code handling requests that have struct mdt_body passed in:
2404  *
2405  *  - extract mdt_body from request and save it in @info, if present;
2406  *
2407  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2408  *  @info;
2409  *
2410  *  - if HABEO_CORPUS flag is set for this request type check whether object
2411  *  actually exists on storage (lu_object_exists()).
2412  *
2413  */
2414 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2415 {
2416         const struct mdt_body    *body;
2417         struct mdt_object        *obj;
2418         const struct lu_env      *env;
2419         struct req_capsule       *pill;
2420         int                       rc;
2421         ENTRY;
2422
2423         env = info->mti_env;
2424         pill = info->mti_pill;
2425
2426         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2427         if (body == NULL)
2428                 RETURN(-EFAULT);
2429
2430         if (!(body->valid & OBD_MD_FLID))
2431                 RETURN(0);
2432
2433         if (!fid_is_sane(&body->fid1)) {
2434                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2435                 RETURN(-EINVAL);
2436         }
2437
2438         /*
2439          * Do not get size or any capa fields before we check that request
2440          * contains capa actually. There are some requests which do not, for
2441          * instance MDS_IS_SUBDIR.
2442          */
2443         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2444             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2445                 mdt_set_capainfo(info, 0, &body->fid1,
2446                                  req_capsule_client_get(pill, &RMF_CAPA1));
2447
2448         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2449         if (!IS_ERR(obj)) {
2450                 if ((flags & HABEO_CORPUS) &&
2451                     !mdt_object_exists(obj)) {
2452                         mdt_object_put(env, obj);
2453                         /* for capability renew ENOENT will be handled in
2454                          * mdt_renew_capa */
2455                         if (body->valid & OBD_MD_FLOSSCAPA)
2456                                 rc = 0;
2457                         else
2458                                 rc = -ENOENT;
2459                 } else {
2460                         info->mti_object = obj;
2461                         rc = 0;
2462                 }
2463         } else
2464                 rc = PTR_ERR(obj);
2465
2466         RETURN(rc);
2467 }
2468
2469 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2470 {
2471         struct req_capsule *pill = info->mti_pill;
2472         int rc;
2473         ENTRY;
2474
2475         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2476                 rc = mdt_body_unpack(info, flags);
2477         else
2478                 rc = 0;
2479
2480         if (rc == 0 && (flags & HABEO_REFERO)) {
2481                 struct mdt_device *mdt = info->mti_mdt;
2482
2483                 /* Pack reply. */
2484
2485                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2486                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2487                                              mdt->mdt_max_mdsize);
2488                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2489                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2490                                              mdt->mdt_max_cookiesize);
2491
2492                 rc = req_capsule_server_pack(pill);
2493         }
2494         RETURN(rc);
2495 }
2496
2497 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2498 {
2499         struct md_device *next = m->mdt_child;
2500
2501         return next->md_ops->mdo_init_capa_ctxt(env, next,
2502                                                 m->mdt_opts.mo_mds_capa,
2503                                                 m->mdt_capa_timeout,
2504                                                 m->mdt_capa_alg,
2505                                                 m->mdt_capa_keys);
2506 }
2507
2508 /*
2509  * Invoke handler for this request opc. Also do necessary preprocessing
2510  * (according to handler ->mh_flags), and post-processing (setting of
2511  * ->last_{xid,committed}).
2512  */
2513 static int mdt_req_handle(struct mdt_thread_info *info,
2514                           struct mdt_handler *h, struct ptlrpc_request *req)
2515 {
2516         int   rc, serious = 0;
2517         __u32 flags;
2518
2519         ENTRY;
2520
2521         LASSERT(h->mh_act != NULL);
2522         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2523         LASSERT(current->journal_info == NULL);
2524
2525         /*
2526          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2527          * to put same checks into handlers like mdt_close(), mdt_reint(),
2528          * etc., without talking to mdt authors first. Checking same thing
2529          * there again is useless and returning 0 error without packing reply
2530          * is buggy! Handlers either pack reply or return error.
2531          *
2532          * We return 0 here and do not send any reply in order to emulate
2533          * network failure. Do not send any reply in case any of NET related
2534          * fail_id has occured.
2535          */
2536         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2537                 RETURN(0);
2538
2539         rc = 0;
2540         flags = h->mh_flags;
2541         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2542
2543         if (h->mh_fmt != NULL) {
2544                 req_capsule_set(info->mti_pill, h->mh_fmt);
2545                 rc = mdt_unpack_req_pack_rep(info, flags);
2546         }
2547
2548         if (rc == 0 && flags & MUTABOR &&
2549             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2550                 /* should it be rq_status? */
2551                 rc = -EROFS;
2552
2553         if (rc == 0 && flags & HABEO_CLAVIS) {
2554                 struct ldlm_request *dlm_req;
2555
2556                 LASSERT(h->mh_fmt != NULL);
2557
2558                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2559                 if (dlm_req != NULL) {
2560                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2561                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2562                                                              dlm_req);
2563                         info->mti_dlm_req = dlm_req;
2564                 } else {
2565                         rc = -EFAULT;
2566                 }
2567         }
2568
2569         /* capability setting changed via /proc, needs reinitialize ctxt */
2570         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2571                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2572                 info->mti_mdt->mdt_capa_conf = 0;
2573         }
2574
2575         if (likely(rc == 0)) {
2576                 /*
2577                  * Process request, there can be two types of rc:
2578                  * 1) errors with msg unpack/pack, other failures outside the
2579                  * operation itself. This is counted as serious errors;
2580                  * 2) errors during fs operation, should be placed in rq_status
2581                  * only
2582                  */
2583                 rc = h->mh_act(info);
2584                 if (rc == 0 &&
2585                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2586                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2587                                   "pack reply and returned 0 error\n",
2588                                   h->mh_name);
2589                         LBUG();
2590                 }
2591                 serious = is_serious(rc);
2592                 rc = clear_serious(rc);
2593         } else
2594                 serious = 1;
2595
2596         req->rq_status = rc;
2597
2598         /*
2599          * ELDLM_* codes which > 0 should be in rq_status only as well as
2600          * all non-serious errors.
2601          */
2602         if (rc > 0 || !serious)
2603                 rc = 0;
2604
2605         LASSERT(current->journal_info == NULL);
2606
2607         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2608             info->mti_mdt->mdt_opts.mo_compat_resname) {
2609                 struct ldlm_reply *dlmrep;
2610
2611                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2612                 if (dlmrep != NULL)
2613                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2614         }
2615
2616         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2617         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2618                 target_committed_to_req(req);
2619
2620         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2621                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2622                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2623                 LBUG();
2624         }
2625
2626         target_send_reply(req, rc, info->mti_fail_id);
2627         RETURN(0);
2628 }
2629
2630 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2631 {
2632         lh->mlh_type = MDT_NUL_LOCK;
2633         lh->mlh_reg_lh.cookie = 0ull;
2634         lh->mlh_reg_mode = LCK_MINMODE;
2635         lh->mlh_pdo_lh.cookie = 0ull;
2636         lh->mlh_pdo_mode = LCK_MINMODE;
2637 }
2638
2639 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2640 {
2641         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2642         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2643 }
2644
2645 /*
2646  * Initialize fields of struct mdt_thread_info. Other fields are left in
2647  * uninitialized state, because it's too expensive to zero out whole
2648  * mdt_thread_info (> 1K) on each request arrival.
2649  */
2650 static void mdt_thread_info_init(struct ptlrpc_request *req,
2651                                  struct mdt_thread_info *info)
2652 {
2653         int i;
2654         struct md_capainfo *ci;
2655
2656         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2657         info->mti_pill = &req->rq_pill;
2658
2659         /* lock handle */
2660         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2661                 mdt_lock_handle_init(&info->mti_lh[i]);
2662
2663         /* mdt device: it can be NULL while CONNECT */
2664         if (req->rq_export) {
2665                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2666                 info->mti_exp = req->rq_export;
2667         } else
2668                 info->mti_mdt = NULL;
2669         info->mti_env = req->rq_svc_thread->t_env;
2670         ci = md_capainfo(info->mti_env);
2671         memset(ci, 0, sizeof *ci);
2672         if (req->rq_export) {
2673                 if (exp_connect_rmtclient(req->rq_export))
2674                         ci->mc_auth = LC_ID_CONVERT;
2675                 else if (req->rq_export->exp_connect_flags &
2676                          OBD_CONNECT_MDS_CAPA)
2677                         ci->mc_auth = LC_ID_PLAIN;
2678                 else
2679                         ci->mc_auth = LC_ID_NONE;
2680         }
2681
2682         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2683         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2684         info->mti_mos[0] = NULL;
2685         info->mti_mos[1] = NULL;
2686         info->mti_mos[2] = NULL;
2687         info->mti_mos[3] = NULL;
2688
2689         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2690         info->mti_body = NULL;
2691         info->mti_object = NULL;
2692         info->mti_dlm_req = NULL;
2693         info->mti_has_trans = 0;
2694         info->mti_no_need_trans = 0;
2695         info->mti_cross_ref = 0;
2696         info->mti_opdata = 0;
2697
2698         /* To not check for split by default. */
2699         info->mti_spec.sp_ck_split = 0;
2700         info->mti_spec.no_create = 0;
2701 }
2702
2703 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2704 {
2705         int i;
2706
2707         req_capsule_fini(info->mti_pill);
2708         if (info->mti_object != NULL) {
2709                 /*
2710                  * freeing an object may lead to OSD level transaction, do not
2711                  * let it mess with MDT. bz19385.
2712                  */
2713                 info->mti_no_need_trans = 1;
2714                 mdt_object_put(info->mti_env, info->mti_object);
2715                 info->mti_object = NULL;
2716         }
2717         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2718                 mdt_lock_handle_fini(&info->mti_lh[i]);
2719         info->mti_env = NULL;
2720 }
2721
2722 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2723                                        struct obd_device *obd, int *process)
2724 {
2725         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2726         case MDS_CONNECT: /* This will never get here, but for completeness. */
2727         case OST_CONNECT: /* This will never get here, but for completeness. */
2728         case MDS_DISCONNECT:
2729         case OST_DISCONNECT:
2730                *process = 1;
2731                RETURN(0);
2732
2733         case MDS_CLOSE:
2734         case MDS_DONE_WRITING:
2735         case MDS_SYNC: /* used in unmounting */
2736         case OBD_PING:
2737         case MDS_REINT:
2738         case SEQ_QUERY:
2739         case FLD_QUERY:
2740         case LDLM_ENQUEUE:
2741                 *process = target_queue_recovery_request(req, obd);
2742                 RETURN(0);
2743
2744         default:
2745                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2746                 *process = -EAGAIN;
2747                 RETURN(0);
2748         }
2749 }
2750
2751 /*
2752  * Handle recovery. Return:
2753  *        +1: continue request processing;
2754  *       -ve: abort immediately with the given error code;
2755  *         0: send reply with error code in req->rq_status;
2756  */
2757 static int mdt_recovery(struct mdt_thread_info *info)
2758 {
2759         struct ptlrpc_request *req = mdt_info_req(info);
2760         int recovering;
2761         struct obd_device *obd;
2762
2763         ENTRY;
2764
2765         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2766         case MDS_CONNECT:
2767         case SEC_CTX_INIT:
2768         case SEC_CTX_INIT_CONT:
2769         case SEC_CTX_FINI:
2770                 {
2771 #if 0
2772                         int rc;
2773
2774                         rc = mdt_handle_idmap(info);
2775                         if (rc)
2776                                 RETURN(rc);
2777                         else
2778 #endif
2779                                 RETURN(+1);
2780                 }
2781         }
2782
2783         if (unlikely(!class_connected_export(req->rq_export))) {
2784                 CERROR("operation %d on unconnected MDS from %s\n",
2785                        lustre_msg_get_opc(req->rq_reqmsg),
2786                        libcfs_id2str(req->rq_peer));
2787                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2788                  * mds_A will get -ENOTCONN(especially for ping req),
2789                  * which will cause that mds_A deactive timeout, then when
2790                  * mds_A cleanup, the cleanup process will be suspended since
2791                  * deactive timeout is not zero.
2792                  */
2793                 req->rq_status = -ENOTCONN;
2794                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2795                 RETURN(0);
2796         }
2797
2798         /* sanity check: if the xid matches, the request must be marked as a
2799          * resent or replayed */
2800         if (req_xid_is_last(req)) {
2801                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2802                       (MSG_RESENT | MSG_REPLAY))) {
2803                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2804                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2805                                   lustre_msg_get_flags(req->rq_reqmsg));
2806                         LBUG();
2807                         req->rq_status = -ENOTCONN;
2808                         RETURN(-ENOTCONN);
2809                 }
2810         }
2811
2812         /* else: note the opposite is not always true; a RESENT req after a
2813          * failover will usually not match the last_xid, since it was likely
2814          * never committed. A REPLAYed request will almost never match the
2815          * last xid, however it could for a committed, but still retained,
2816          * open. */
2817
2818         obd = req->rq_export->exp_obd;
2819
2820         /* Check for aborted recovery... */
2821         cfs_spin_lock_bh(&obd->obd_processing_task_lock);
2822         recovering = obd->obd_recovering;
2823         cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
2824         if (unlikely(recovering)) {
2825                 int rc;
2826                 int should_process;
2827                 DEBUG_REQ(D_INFO, req, "Got new replay");
2828                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2829                 if (rc != 0 || !should_process)
2830                         RETURN(rc);
2831                 else if (should_process < 0) {
2832                         req->rq_status = should_process;
2833                         rc = ptlrpc_error(req);
2834                         RETURN(rc);
2835                 }
2836         }
2837         RETURN(+1);
2838 }
2839
2840 static int mdt_msg_check_version(struct lustre_msg *msg)
2841 {
2842         int rc;
2843
2844         switch (lustre_msg_get_opc(msg)) {
2845         case MDS_CONNECT:
2846         case MDS_DISCONNECT:
2847         case MDS_SET_INFO:
2848         case OBD_PING:
2849         case SEC_CTX_INIT:
2850         case SEC_CTX_INIT_CONT:
2851         case SEC_CTX_FINI:
2852                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2853                 if (rc)
2854                         CERROR("bad opc %u version %08x, expecting %08x\n",
2855                                lustre_msg_get_opc(msg),
2856                                lustre_msg_get_version(msg),
2857                                LUSTRE_OBD_VERSION);
2858                 break;
2859         case MDS_GETSTATUS:
2860         case MDS_GETATTR:
2861         case MDS_GETATTR_NAME:
2862         case MDS_STATFS:
2863         case MDS_READPAGE:
2864         case MDS_WRITEPAGE:
2865         case MDS_IS_SUBDIR:
2866         case MDS_REINT:
2867         case MDS_CLOSE:
2868         case MDS_DONE_WRITING:
2869         case MDS_PIN:
2870         case MDS_SYNC:
2871         case MDS_GETXATTR:
2872         case MDS_SETXATTR:
2873         case MDS_GET_INFO:
2874         case MDS_QUOTACHECK:
2875         case MDS_QUOTACTL:
2876         case QUOTA_DQACQ:
2877         case QUOTA_DQREL:
2878         case SEQ_QUERY:
2879         case FLD_QUERY:
2880                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2881                 if (rc)
2882                         CERROR("bad opc %u version %08x, expecting %08x\n",
2883                                lustre_msg_get_opc(msg),
2884                                lustre_msg_get_version(msg),
2885                                LUSTRE_MDS_VERSION);
2886                 break;
2887         case LDLM_ENQUEUE:
2888         case LDLM_CONVERT:
2889         case LDLM_BL_CALLBACK:
2890         case LDLM_CP_CALLBACK:
2891                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2892                 if (rc)
2893                         CERROR("bad opc %u version %08x, expecting %08x\n",
2894                                lustre_msg_get_opc(msg),
2895                                lustre_msg_get_version(msg),
2896                                LUSTRE_DLM_VERSION);
2897                 break;
2898         case OBD_LOG_CANCEL:
2899         case LLOG_ORIGIN_HANDLE_CREATE:
2900         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2901         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2902         case LLOG_ORIGIN_HANDLE_CLOSE:
2903         case LLOG_ORIGIN_HANDLE_DESTROY:
2904         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2905         case LLOG_CATINFO:
2906                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2907                 if (rc)
2908                         CERROR("bad opc %u version %08x, expecting %08x\n",
2909                                lustre_msg_get_opc(msg),
2910                                lustre_msg_get_version(msg),
2911                                LUSTRE_LOG_VERSION);
2912                 break;
2913         default:
2914                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2915                 rc = -ENOTSUPP;
2916         }
2917         return rc;
2918 }
2919
2920 static int mdt_handle0(struct ptlrpc_request *req,
2921                        struct mdt_thread_info *info,
2922                        struct mdt_opc_slice *supported)
2923 {
2924         struct mdt_handler *h;
2925         struct lustre_msg  *msg;
2926         int                 rc;
2927
2928         ENTRY;
2929
2930         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2931                 RETURN(0);
2932
2933         LASSERT(current->journal_info == NULL);
2934
2935         msg = req->rq_reqmsg;
2936         rc = mdt_msg_check_version(msg);
2937         if (likely(rc == 0)) {
2938                 rc = mdt_recovery(info);
2939                 if (likely(rc == +1)) {
2940                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2941                                              supported);
2942                         if (likely(h != NULL)) {
2943                                 rc = mdt_req_handle(info, h, req);
2944                         } else {
2945                                 CERROR("The unsupported opc: 0x%x\n",
2946                                        lustre_msg_get_opc(msg) );
2947                                 req->rq_status = -ENOTSUPP;
2948                                 rc = ptlrpc_error(req);
2949                                 RETURN(rc);
2950                         }
2951                 }
2952         } else
2953                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2954         RETURN(rc);
2955 }
2956
2957 /*
2958  * MDT handler function called by ptlrpc service thread when request comes.
2959  *
2960  * XXX common "target" functionality should be factored into separate module
2961  * shared by mdt, ost and stand-alone services like fld.
2962  */
2963 static int mdt_handle_common(struct ptlrpc_request *req,
2964                              struct mdt_opc_slice *supported)
2965 {
2966         struct lu_env          *env;
2967         struct mdt_thread_info *info;
2968         int                     rc;
2969         ENTRY;
2970
2971         env = req->rq_svc_thread->t_env;
2972         LASSERT(env != NULL);
2973         LASSERT(env->le_ses != NULL);
2974         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2975         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2976         LASSERT(info != NULL);
2977
2978         mdt_thread_info_init(req, info);
2979
2980         rc = mdt_handle0(req, info, supported);
2981
2982         mdt_thread_info_fini(info);
2983         RETURN(rc);
2984 }
2985
2986 /*
2987  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2988  * as well.
2989  */
2990 int mdt_recovery_handle(struct ptlrpc_request *req)
2991 {
2992         int rc;
2993         ENTRY;
2994
2995         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2996         case FLD_QUERY:
2997                 rc = mdt_handle_common(req, mdt_fld_handlers);
2998                 break;
2999         case SEQ_QUERY:
3000                 rc = mdt_handle_common(req, mdt_seq_handlers);
3001                 break;
3002         default:
3003                 rc = mdt_handle_common(req, mdt_regular_handlers);
3004                 break;
3005         }
3006
3007         RETURN(rc);
3008 }
3009
3010 static int mdt_regular_handle(struct ptlrpc_request *req)
3011 {
3012         return mdt_handle_common(req, mdt_regular_handlers);
3013 }
3014
3015 static int mdt_readpage_handle(struct ptlrpc_request *req)
3016 {
3017         return mdt_handle_common(req, mdt_readpage_handlers);
3018 }
3019
3020 static int mdt_xmds_handle(struct ptlrpc_request *req)
3021 {
3022         return mdt_handle_common(req, mdt_xmds_handlers);
3023 }
3024
3025 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3026 {
3027         return mdt_handle_common(req, mdt_seq_handlers);
3028 }
3029
3030 static int mdt_mdss_handle(struct ptlrpc_request *req)
3031 {
3032         return mdt_handle_common(req, mdt_seq_handlers);
3033 }
3034
3035 static int mdt_dtss_handle(struct ptlrpc_request *req)
3036 {
3037         return mdt_handle_common(req, mdt_seq_handlers);
3038 }
3039
3040 static int mdt_fld_handle(struct ptlrpc_request *req)
3041 {
3042         return mdt_handle_common(req, mdt_fld_handlers);
3043 }
3044
3045 enum mdt_it_code {
3046         MDT_IT_OPEN,
3047         MDT_IT_OCREAT,
3048         MDT_IT_CREATE,
3049         MDT_IT_GETATTR,
3050         MDT_IT_READDIR,
3051         MDT_IT_LOOKUP,
3052         MDT_IT_UNLINK,
3053         MDT_IT_TRUNC,
3054         MDT_IT_GETXATTR,
3055         MDT_IT_NR
3056 };
3057
3058 static int mdt_intent_getattr(enum mdt_it_code opcode,
3059                               struct mdt_thread_info *info,
3060                               struct ldlm_lock **,
3061                               int);
3062 static int mdt_intent_reint(enum mdt_it_code opcode,
3063                             struct mdt_thread_info *info,
3064                             struct ldlm_lock **,
3065                             int);
3066
3067 static struct mdt_it_flavor {
3068         const struct req_format *it_fmt;
3069         __u32                    it_flags;
3070         int                    (*it_act)(enum mdt_it_code ,
3071                                          struct mdt_thread_info *,
3072                                          struct ldlm_lock **,
3073                                          int);
3074         long                     it_reint;
3075 } mdt_it_flavor[] = {
3076         [MDT_IT_OPEN]     = {
3077                 .it_fmt   = &RQF_LDLM_INTENT,
3078                 /*.it_flags = HABEO_REFERO,*/
3079                 .it_flags = 0,
3080                 .it_act   = mdt_intent_reint,
3081                 .it_reint = REINT_OPEN
3082         },
3083         [MDT_IT_OCREAT]   = {
3084                 .it_fmt   = &RQF_LDLM_INTENT,
3085                 .it_flags = MUTABOR,
3086                 .it_act   = mdt_intent_reint,
3087                 .it_reint = REINT_OPEN
3088         },
3089         [MDT_IT_CREATE]   = {
3090                 .it_fmt   = &RQF_LDLM_INTENT,
3091                 .it_flags = MUTABOR,
3092                 .it_act   = mdt_intent_reint,
3093                 .it_reint = REINT_CREATE
3094         },
3095         [MDT_IT_GETATTR]  = {
3096                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3097                 .it_flags = HABEO_REFERO,
3098                 .it_act   = mdt_intent_getattr
3099         },
3100         [MDT_IT_READDIR]  = {
3101                 .it_fmt   = NULL,
3102                 .it_flags = 0,
3103                 .it_act   = NULL
3104         },
3105         [MDT_IT_LOOKUP]   = {
3106                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3107                 .it_flags = HABEO_REFERO,
3108                 .it_act   = mdt_intent_getattr
3109         },
3110         [MDT_IT_UNLINK]   = {
3111                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3112                 .it_flags = MUTABOR,
3113                 .it_act   = NULL,
3114                 .it_reint = REINT_UNLINK
3115         },
3116         [MDT_IT_TRUNC]    = {
3117                 .it_fmt   = NULL,
3118                 .it_flags = MUTABOR,
3119                 .it_act   = NULL
3120         },
3121         [MDT_IT_GETXATTR] = {
3122                 .it_fmt   = NULL,
3123                 .it_flags = 0,
3124                 .it_act   = NULL
3125         }
3126 };
3127
3128 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3129                             struct ldlm_lock **lockp,
3130                             struct ldlm_lock *new_lock,
3131                             struct mdt_lock_handle *lh,
3132                             int flags)
3133 {
3134         struct ptlrpc_request  *req = mdt_info_req(info);
3135         struct ldlm_lock       *lock = *lockp;
3136
3137         /*
3138          * Get new lock only for cases when possible resent did not find any
3139          * lock.
3140          */
3141         if (new_lock == NULL)
3142                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3143
3144         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3145                 lh->mlh_reg_lh.cookie = 0;
3146                 RETURN(0);
3147         }
3148
3149         LASSERTF(new_lock != NULL,
3150                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3151
3152         /*
3153          * If we've already given this lock to a client once, then we should
3154          * have no readers or writers.  Otherwise, we should have one reader
3155          * _or_ writer ref (which will be zeroed below) before returning the
3156          * lock to a client.
3157          */
3158         if (new_lock->l_export == req->rq_export) {
3159                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3160         } else {
3161                 LASSERT(new_lock->l_export == NULL);
3162                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3163         }
3164
3165         *lockp = new_lock;
3166
3167         if (new_lock->l_export == req->rq_export) {
3168                 /*
3169                  * Already gave this to the client, which means that we
3170                  * reconstructed a reply.
3171                  */
3172                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3173                         MSG_RESENT);
3174                 lh->mlh_reg_lh.cookie = 0;
3175                 RETURN(ELDLM_LOCK_REPLACED);
3176         }
3177
3178         /*
3179          * Fixup the lock to be given to the client.
3180          */
3181         lock_res_and_lock(new_lock);
3182         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3183          * possible blocking AST. */
3184         while (new_lock->l_readers > 0) {
3185                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3186                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3187                 new_lock->l_readers--;
3188         }
3189         while (new_lock->l_writers > 0) {
3190                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3191                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3192                 new_lock->l_writers--;
3193         }
3194
3195         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3196         new_lock->l_blocking_ast = lock->l_blocking_ast;
3197         new_lock->l_completion_ast = lock->l_completion_ast;
3198         new_lock->l_remote_handle = lock->l_remote_handle;
3199         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3200
3201         unlock_res_and_lock(new_lock);
3202
3203         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3204                      &new_lock->l_remote_handle,
3205                      &new_lock->l_exp_hash);
3206
3207         LDLM_LOCK_RELEASE(new_lock);
3208         lh->mlh_reg_lh.cookie = 0;
3209
3210         RETURN(ELDLM_LOCK_REPLACED);
3211 }
3212
3213 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3214                                     struct ldlm_lock *new_lock,
3215                                     struct ldlm_lock **old_lock,
3216                                     struct mdt_lock_handle *lh)
3217 {
3218         struct ptlrpc_request  *req = mdt_info_req(info);
3219         struct obd_export      *exp = req->rq_export;
3220         struct lustre_handle    remote_hdl;
3221         struct ldlm_request    *dlmreq;
3222         struct ldlm_lock       *lock;
3223
3224         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3225                 return;
3226
3227         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3228         remote_hdl = dlmreq->lock_handle[0];
3229
3230         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3231         if (lock) {
3232                 if (lock != new_lock) {
3233                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3234                         lh->mlh_reg_mode = lock->l_granted_mode;
3235
3236                         LDLM_DEBUG(lock, "Restoring lock cookie");
3237                         DEBUG_REQ(D_DLMTRACE, req,
3238                                   "restoring lock cookie "LPX64,
3239                                   lh->mlh_reg_lh.cookie);
3240                         if (old_lock)
3241                                 *old_lock = LDLM_LOCK_GET(lock);
3242                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3243                         return;
3244                 }
3245
3246                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3247         }
3248
3249         /*
3250          * If the xid matches, then we know this is a resent request, and allow
3251          * it. (It's probably an OPEN, for which we don't send a lock.
3252          */
3253         if (req_xid_is_last(req))
3254                 return;
3255
3256         /*
3257          * This remote handle isn't enqueued, so we never received or processed
3258          * this request.  Clear MSG_RESENT, because it can be handled like any
3259          * normal request now.
3260          */
3261         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3262
3263         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3264                   remote_hdl.cookie);
3265 }
3266
3267 static int mdt_intent_getattr(enum mdt_it_code opcode,
3268                               struct mdt_thread_info *info,
3269                               struct ldlm_lock **lockp,
3270                               int flags)
3271 {
3272         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3273         struct ldlm_lock       *new_lock = NULL;
3274         __u64                   child_bits;
3275         struct ldlm_reply      *ldlm_rep;
3276         struct ptlrpc_request  *req;
3277         struct mdt_body        *reqbody;
3278         struct mdt_body        *repbody;
3279         int                     rc;
3280         ENTRY;
3281
3282         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3283         LASSERT(reqbody);
3284
3285         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3286         LASSERT(repbody);
3287
3288         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3289         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3290         repbody->eadatasize = 0;
3291         repbody->aclsize = 0;
3292
3293         switch (opcode) {
3294         case MDT_IT_LOOKUP:
3295                 child_bits = MDS_INODELOCK_LOOKUP;
3296                 break;
3297         case MDT_IT_GETATTR:
3298                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3299                 break;
3300         default:
3301                 CERROR("Unhandled till now");
3302                 GOTO(out_shrink, rc = -EINVAL);
3303         }
3304
3305         rc = mdt_init_ucred(info, reqbody);
3306         if (rc)
3307                 GOTO(out_shrink, rc);
3308
3309         req = info->mti_pill->rc_req;
3310         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3311         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3312
3313         /* Get lock from request for possible resent case. */
3314         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3315
3316         ldlm_rep->lock_policy_res2 =
3317                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3318
3319         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3320                 ldlm_rep->lock_policy_res2 = 0;
3321         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3322             ldlm_rep->lock_policy_res2) {
3323                 lhc->mlh_reg_lh.cookie = 0ull;
3324                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3325         }
3326
3327         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3328         EXIT;
3329 out_ucred:
3330         mdt_exit_ucred(info);
3331 out_shrink:
3332         mdt_shrink_reply(info);
3333         return rc;
3334 }
3335
3336 static int mdt_intent_reint(enum mdt_it_code opcode,
3337                             struct mdt_thread_info *info,
3338                             struct ldlm_lock **lockp,
3339                             int flags)
3340 {
3341         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3342         struct ldlm_reply      *rep = NULL;
3343         long                    opc;
3344         int                     rc;
3345
3346         static const struct req_format *intent_fmts[REINT_MAX] = {
3347                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3348                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3349         };
3350
3351         ENTRY;
3352
3353         opc = mdt_reint_opcode(info, intent_fmts);
3354         if (opc < 0)
3355                 RETURN(opc);
3356
3357         if (mdt_it_flavor[opcode].it_reint != opc) {
3358                 CERROR("Reint code %ld doesn't match intent: %d\n",
3359                        opc, opcode);
3360                 RETURN(err_serious(-EPROTO));
3361         }
3362
3363         /* Get lock from request for possible resent case. */
3364         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3365
3366         rc = mdt_reint_internal(info, lhc, opc);
3367
3368         /* Check whether the reply has been packed successfully. */
3369         if (mdt_info_req(info)->rq_repmsg != NULL)
3370                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3371         if (rep == NULL)
3372                 RETURN(err_serious(-EFAULT));
3373
3374         /* MDC expects this in any case */
3375         if (rc != 0)
3376                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3377
3378         /* Cross-ref case, the lock should be returned to the client */
3379         if (rc == -EREMOTE) {
3380                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3381                 rep->lock_policy_res2 = 0;
3382                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3383                 RETURN(rc);
3384         }
3385         rep->lock_policy_res2 = clear_serious(rc);
3386
3387         if (rc == -ENOTCONN || rc == -ENODEV ||
3388             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3389                 /*
3390                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3391                  * will be returned by rq_status, and client at ptlrpc layer
3392                  * will detect this, then disconnect, reconnect the import
3393                  * immediately, instead of impacting the following the rpc.
3394                  */
3395                 lhc->mlh_reg_lh.cookie = 0ull;
3396                 RETURN(rc);
3397         } else {
3398                 /*
3399                  * For other cases, the error will be returned by intent.
3400                  * and client will retrieve the result from intent.
3401                  */
3402                  /*
3403                   * FIXME: when open lock is finished, that should be
3404                   * checked here.
3405                   */
3406                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3407                         rep->lock_policy_res2 = 0;
3408                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3409                         RETURN(rc);
3410                 } else {
3411                         lhc->mlh_reg_lh.cookie = 0ull;
3412                         RETURN(ELDLM_LOCK_ABORTED);
3413                 }
3414         }
3415 }
3416
3417 static int mdt_intent_code(long itcode)
3418 {
3419         int rc;
3420
3421         switch(itcode) {
3422         case IT_OPEN:
3423                 rc = MDT_IT_OPEN;
3424                 break;
3425         case IT_OPEN|IT_CREAT:
3426                 rc = MDT_IT_OCREAT;
3427                 break;
3428         case IT_CREAT:
3429                 rc = MDT_IT_CREATE;
3430                 break;
3431         case IT_READDIR:
3432                 rc = MDT_IT_READDIR;
3433                 break;
3434         case IT_GETATTR:
3435                 rc = MDT_IT_GETATTR;
3436                 break;
3437         case IT_LOOKUP:
3438                 rc = MDT_IT_LOOKUP;
3439                 break;
3440         case IT_UNLINK:
3441                 rc = MDT_IT_UNLINK;
3442                 break;
3443         case IT_TRUNC:
3444                 rc = MDT_IT_TRUNC;
3445                 break;
3446         case IT_GETXATTR:
3447                 rc = MDT_IT_GETXATTR;
3448                 break;
3449         default:
3450                 CERROR("Unknown intent opcode: %ld\n", itcode);
3451                 rc = -EINVAL;
3452                 break;
3453         }
3454         return rc;
3455 }
3456
3457 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3458                           struct ldlm_lock **lockp, int flags)
3459 {
3460         struct req_capsule   *pill;
3461         struct mdt_it_flavor *flv;
3462         int opc;
3463         int rc;
3464         ENTRY;
3465
3466         opc = mdt_intent_code(itopc);
3467         if (opc < 0)
3468                 RETURN(-EINVAL);
3469
3470         pill = info->mti_pill;
3471         flv  = &mdt_it_flavor[opc];
3472
3473         if (flv->it_fmt != NULL)
3474                 req_capsule_extend(pill, flv->it_fmt);
3475
3476         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3477         if (rc == 0) {
3478                 struct ptlrpc_request *req = mdt_info_req(info);
3479                 if (flv->it_flags & MUTABOR &&
3480                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3481                         RETURN(-EROFS);
3482         }
3483         if (rc == 0 && flv->it_act != NULL) {
3484                 /* execute policy */
3485                 rc = flv->it_act(opc, info, lockp, flags);
3486         } else {
3487                 rc = -EOPNOTSUPP;
3488         }
3489         RETURN(rc);
3490 }
3491
3492 static int mdt_intent_policy(struct ldlm_namespace *ns,
3493                              struct ldlm_lock **lockp, void *req_cookie,
3494                              ldlm_mode_t mode, int flags, void *data)
3495 {
3496         struct mdt_thread_info *info;
3497         struct ptlrpc_request  *req  =  req_cookie;
3498         struct ldlm_intent     *it;
3499         struct req_capsule     *pill;
3500         int rc;
3501
3502         ENTRY;
3503
3504         LASSERT(req != NULL);
3505
3506         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3507                                   &mdt_thread_key);
3508         LASSERT(info != NULL);
3509         pill = info->mti_pill;
3510         LASSERT(pill->rc_req == req);
3511
3512         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3513                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3514                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3515                 if (it != NULL) {
3516                         const struct ldlm_request *dlmreq;
3517                         __u64 req_bits;
3518
3519                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3520                         if (rc == 0)
3521                                 rc = ELDLM_OK;
3522
3523                         /*
3524                          * Lock without inodebits makes no sense and will oops
3525                          * later in ldlm. Let's check it now to see if we have
3526                          * wrong lock from client or bits get corrupted
3527                          * somewhere in mdt_intent_opc().
3528                          */
3529                         dlmreq = info->mti_dlm_req;
3530                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3531                         LASSERT(req_bits != 0);
3532
3533                 } else
3534                         rc = err_serious(-EFAULT);
3535         } else {
3536                 /* No intent was provided */
3537                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3538                 rc = req_capsule_server_pack(pill);
3539                 if (rc)
3540                         rc = err_serious(rc);
3541         }
3542         RETURN(rc);
3543 }
3544
3545 /*
3546  * Seq wrappers
3547  */
3548 static void mdt_seq_adjust(const struct lu_env *env,
3549                           struct mdt_device *m, int lost)
3550 {
3551         struct md_site *ms = mdt_md_site(m);
3552         struct lu_seq_range out;
3553         ENTRY;
3554
3555         LASSERT(ms && ms->ms_server_seq);
3556         LASSERT(lost >= 0);
3557         /* get extra seq from seq_server, moving it's range up */
3558         while (lost-- > 0) {
3559                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3560         }
3561         EXIT;
3562 }
3563
3564 static int mdt_seq_fini(const struct lu_env *env,
3565                         struct mdt_device *m)
3566 {
3567         struct md_site *ms = mdt_md_site(m);
3568         ENTRY;
3569
3570         if (ms != NULL) {
3571                 if (ms->ms_server_seq) {
3572                         seq_server_fini(ms->ms_server_seq, env);
3573                         OBD_FREE_PTR(ms->ms_server_seq);
3574                         ms->ms_server_seq = NULL;
3575         }
3576
3577                 if (ms->ms_control_seq) {
3578                         seq_server_fini(ms->ms_control_seq, env);
3579                         OBD_FREE_PTR(ms->ms_control_seq);
3580                         ms->ms_control_seq = NULL;
3581         }
3582
3583                 if (ms->ms_client_seq) {
3584                         seq_client_fini(ms->ms_client_seq);
3585                         OBD_FREE_PTR(ms->ms_client_seq);
3586                         ms->ms_client_seq = NULL;
3587                 }
3588         }
3589
3590         RETURN(0);
3591 }
3592
3593 static int mdt_seq_init(const struct lu_env *env,
3594                         const char *uuid,
3595                         struct mdt_device *m)
3596 {
3597         struct md_site *ms;
3598         char *prefix;
3599         int rc;
3600         ENTRY;
3601
3602         ms = mdt_md_site(m);
3603
3604         /*
3605          * This is sequence-controller node. Init seq-controller server on local
3606          * MDT.
3607          */
3608         if (ms->ms_node_id == 0) {
3609                 LASSERT(ms->ms_control_seq == NULL);
3610
3611                 OBD_ALLOC_PTR(ms->ms_control_seq);
3612                 if (ms->ms_control_seq == NULL)
3613                         RETURN(-ENOMEM);
3614
3615                 rc = seq_server_init(ms->ms_control_seq,
3616                                      m->mdt_bottom, uuid,
3617                                      LUSTRE_SEQ_CONTROLLER,
3618                                      ms,
3619                                      env);
3620
3621                 if (rc)
3622                         GOTO(out_seq_fini, rc);
3623
3624                 OBD_ALLOC_PTR(ms->ms_client_seq);
3625                 if (ms->ms_client_seq == NULL)
3626                         GOTO(out_seq_fini, rc = -ENOMEM);
3627
3628                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3629                 if (prefix == NULL) {
3630                         OBD_FREE_PTR(ms->ms_client_seq);
3631                         GOTO(out_seq_fini, rc = -ENOMEM);
3632                 }
3633
3634                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3635                          uuid);
3636
3637                 /*
3638                  * Init seq-controller client after seq-controller server is
3639                  * ready. Pass ms->ms_control_seq to it for direct talking.
3640                  */
3641                 rc = seq_client_init(ms->ms_client_seq, NULL,
3642                                      LUSTRE_SEQ_METADATA, prefix,
3643                                      ms->ms_control_seq);
3644                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3645
3646                 if (rc)
3647                         GOTO(out_seq_fini, rc);
3648         }
3649
3650         /* Init seq-server on local MDT */
3651         LASSERT(ms->ms_server_seq == NULL);
3652
3653         OBD_ALLOC_PTR(ms->ms_server_seq);
3654         if (ms->ms_server_seq == NULL)
3655                 GOTO(out_seq_fini, rc = -ENOMEM);
3656
3657         rc = seq_server_init(ms->ms_server_seq,
3658                              m->mdt_bottom, uuid,
3659                              LUSTRE_SEQ_SERVER,
3660                              ms,
3661                              env);
3662         if (rc)
3663                 GOTO(out_seq_fini, rc = -ENOMEM);
3664
3665         /* Assign seq-controller client to local seq-server. */
3666         if (ms->ms_node_id == 0) {
3667                 LASSERT(ms->ms_client_seq != NULL);
3668
3669                 rc = seq_server_set_cli(ms->ms_server_seq,
3670                                         ms->ms_client_seq,
3671                                         env);
3672         }
3673
3674         EXIT;
3675 out_seq_fini:
3676         if (rc)
3677                 mdt_seq_fini(env, m);
3678
3679         return rc;
3680 }
3681 /*
3682  * Init client sequence manager which is used by local MDS to talk to sequence
3683  * controller on remote node.
3684  */
3685 static int mdt_seq_init_cli(const struct lu_env *env,
3686                             struct mdt_device *m,
3687                             struct lustre_cfg *cfg)
3688 {
3689         struct md_site    *ms = mdt_md_site(m);
3690         struct obd_device *mdc;
3691         struct obd_uuid   *uuidp, *mdcuuidp;
3692         char              *uuid_str, *mdc_uuid_str;
3693         int                rc;
3694         int                index;
3695         struct mdt_thread_info *info;
3696         char *p, *index_string = lustre_cfg_string(cfg, 2);
3697         ENTRY;
3698
3699         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3700         uuidp = &info->mti_u.uuid[0];
3701         mdcuuidp = &info->mti_u.uuid[1];
3702
3703         LASSERT(index_string);
3704
3705         index = simple_strtol(index_string, &p, 10);
3706         if (*p) {
3707                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3708                 RETURN(-EINVAL);
3709         }
3710
3711         /* check if this is adding the first MDC and controller is not yet
3712          * initialized. */
3713         if (index != 0 || ms->ms_client_seq)
3714                 RETURN(0);
3715
3716         uuid_str = lustre_cfg_string(cfg, 1);
3717         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3718         obd_str2uuid(uuidp, uuid_str);
3719         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3720
3721         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3722         if (!mdc) {
3723                 CERROR("can't find controller MDC by uuid %s\n",
3724                        uuid_str);
3725                 rc = -ENOENT;
3726         } else if (!mdc->obd_set_up) {
3727                 CERROR("target %s not set up\n", mdc->obd_name);
3728                 rc = -EINVAL;
3729         } else {
3730                 LASSERT(ms->ms_control_exp);
3731                 OBD_ALLOC_PTR(ms->ms_client_seq);
3732                 if (ms->ms_client_seq != NULL) {
3733                         char *prefix;
3734
3735                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3736                         if (!prefix)
3737                                 RETURN(-ENOMEM);
3738
3739                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3740                                  mdc->obd_name);
3741
3742                         rc = seq_client_init(ms->ms_client_seq,
3743                                              ms->ms_control_exp,
3744                                              LUSTRE_SEQ_METADATA,
3745                                              prefix, NULL);
3746                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3747                 } else
3748                         rc = -ENOMEM;
3749
3750                 if (rc)
3751                         RETURN(rc);
3752
3753                 LASSERT(ms->ms_server_seq != NULL);
3754                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3755                                         env);
3756         }
3757
3758         RETURN(rc);
3759 }
3760
3761 static void mdt_seq_fini_cli(struct mdt_device *m)
3762 {
3763         struct md_site *ms;
3764
3765         ENTRY;
3766
3767         ms = mdt_md_site(m);
3768
3769         if (ms != NULL) {
3770                 if (ms->ms_server_seq)
3771                         seq_server_set_cli(ms->ms_server_seq,
3772                                    NULL, NULL);
3773
3774                 if (ms->ms_control_exp) {
3775                         class_export_put(ms->ms_control_exp);
3776                         ms->ms_control_exp = NULL;
3777                 }
3778         }
3779         EXIT;
3780 }
3781
3782 /*
3783  * FLD wrappers
3784  */
3785 static int mdt_fld_fini(const struct lu_env *env,
3786                         struct mdt_device *m)
3787 {
3788         struct md_site *ms = mdt_md_site(m);
3789         ENTRY;
3790
3791         if (ms && ms->ms_server_fld) {
3792                 fld_server_fini(ms->ms_server_fld, env);
3793                 OBD_FREE_PTR(ms->ms_server_fld);
3794                 ms->ms_server_fld = NULL;
3795         }
3796
3797         RETURN(0);
3798 }
3799
3800 static int mdt_fld_init(const struct lu_env *env,
3801                         const char *uuid,
3802                         struct mdt_device *m)
3803 {
3804         struct md_site *ms;
3805         int rc;
3806         ENTRY;
3807
3808         ms = mdt_md_site(m);
3809
3810         OBD_ALLOC_PTR(ms->ms_server_fld);
3811         if (ms->ms_server_fld == NULL)
3812                 RETURN(rc = -ENOMEM);
3813
3814         rc = fld_server_init(ms->ms_server_fld,
3815                              m->mdt_bottom, uuid,
3816                              env, ms->ms_node_id);
3817         if (rc) {
3818                 OBD_FREE_PTR(ms->ms_server_fld);
3819                 ms->ms_server_fld = NULL;
3820                 RETURN(rc);
3821         }
3822
3823         RETURN(0);
3824 }
3825
3826 /* device init/fini methods */
3827 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3828 {
3829         ENTRY;
3830         if (m->mdt_regular_service != NULL) {
3831                 ptlrpc_unregister_service(m->mdt_regular_service);
3832                 m->mdt_regular_service = NULL;
3833         }
3834         if (m->mdt_readpage_service != NULL) {
3835                 ptlrpc_unregister_service(m->mdt_readpage_service);
3836                 m->mdt_readpage_service = NULL;
3837         }
3838         if (m->mdt_xmds_service != NULL) {
3839                 ptlrpc_unregister_service(m->mdt_xmds_service);
3840                 m->mdt_xmds_service = NULL;
3841         }
3842         if (m->mdt_setattr_service != NULL) {
3843                 ptlrpc_unregister_service(m->mdt_setattr_service);
3844                 m->mdt_setattr_service = NULL;
3845         }
3846         if (m->mdt_mdsc_service != NULL) {
3847                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3848                 m->mdt_mdsc_service = NULL;
3849         }
3850         if (m->mdt_mdss_service != NULL) {
3851                 ptlrpc_unregister_service(m->mdt_mdss_service);
3852                 m->mdt_mdss_service = NULL;
3853         }
3854         if (m->mdt_dtss_service != NULL) {
3855                 ptlrpc_unregister_service(m->mdt_dtss_service);
3856                 m->mdt_dtss_service = NULL;
3857         }
3858         if (m->mdt_fld_service != NULL) {
3859                 ptlrpc_unregister_service(m->mdt_fld_service);
3860                 m->mdt_fld_service = NULL;
3861         }
3862         EXIT;
3863 }
3864
3865 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3866 {
3867         int rc;
3868         static struct ptlrpc_service_conf conf;
3869         cfs_proc_dir_entry_t *procfs_entry;
3870         ENTRY;
3871
3872         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3873
3874         conf = (typeof(conf)) {
3875                 .psc_nbufs           = MDS_NBUFS,
3876                 .psc_bufsize         = MDS_BUFSIZE,
3877                 .psc_max_req_size    = MDS_MAXREQSIZE,
3878                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3879                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3880                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3881                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3882                 /*
3883                  * We'd like to have a mechanism to set this on a per-device
3884                  * basis, but alas...
3885                  */
3886                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3887                                           MDT_MAX_THREADS),
3888                 .psc_max_threads     = MDT_MAX_THREADS,
3889                 .psc_ctx_tags        = LCT_MD_THREAD
3890         };
3891
3892         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3893         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3894                            "mdt_ldlm_client", m->mdt_ldlm_client);
3895
3896         m->mdt_regular_service =
3897                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3898                                      procfs_entry, target_print_req,
3899                                      LUSTRE_MDT_NAME);
3900         if (m->mdt_regular_service == NULL)
3901                 RETURN(-ENOMEM);
3902
3903         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3904         if (rc)
3905                 GOTO(err_mdt_svc, rc);
3906
3907         /*
3908          * readpage service configuration. Parameters have to be adjusted,
3909          * ideally.
3910          */
3911         conf = (typeof(conf)) {
3912                 .psc_nbufs           = MDS_NBUFS,
3913                 .psc_bufsize         = MDS_BUFSIZE,
3914                 .psc_max_req_size    = MDS_MAXREQSIZE,
3915                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3916                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3917                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3918                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3919                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
3920                                           MDT_MAX_THREADS),
3921                 .psc_max_threads     = MDT_MAX_THREADS,
3922                 .psc_ctx_tags        = LCT_MD_THREAD
3923         };
3924         m->mdt_readpage_service =
3925                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3926                                      LUSTRE_MDT_NAME "_readpage",
3927                                      procfs_entry, target_print_req,"mdt_rdpg");
3928
3929         if (m->mdt_readpage_service == NULL) {
3930                 CERROR("failed to start readpage service\n");
3931                 GOTO(err_mdt_svc, rc = -ENOMEM);
3932         }
3933
3934         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3935
3936         /*
3937          * setattr service configuration.
3938          */
3939         conf = (typeof(conf)) {
3940                 .psc_nbufs           = MDS_NBUFS,
3941                 .psc_bufsize         = MDS_BUFSIZE,
3942                 .psc_max_req_size    = MDS_MAXREQSIZE,
3943                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3944                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3945                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3946                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3947                 .psc_min_threads   = min(max(mdt_num_threads, MDT_MIN_THREADS),
3948                                          MDT_MAX_THREADS),
3949                 .psc_max_threads     = MDT_MAX_THREADS,
3950                 .psc_ctx_tags        = LCT_MD_THREAD
3951         };
3952
3953         m->mdt_setattr_service =
3954                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3955                                      LUSTRE_MDT_NAME "_setattr",
3956                                      procfs_entry, target_print_req,"mdt_attr");
3957
3958         if (!m->mdt_setattr_service) {
3959                 CERROR("failed to start setattr service\n");
3960                 GOTO(err_mdt_svc, rc = -ENOMEM);
3961         }
3962
3963         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3964         if (rc)
3965                 GOTO(err_mdt_svc, rc);
3966
3967         /*
3968          * sequence controller service configuration
3969          */
3970         conf = (typeof(conf)) {
3971                 .psc_nbufs           = MDS_NBUFS,
3972                 .psc_bufsize         = MDS_BUFSIZE,
3973                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3974                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3975                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3976                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3977                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3978                 .psc_min_threads     = SEQ_NUM_THREADS,
3979                 .psc_max_threads     = SEQ_NUM_THREADS,
3980                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3981         };
3982
3983         m->mdt_mdsc_service =
3984                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3985                                      LUSTRE_MDT_NAME"_mdsc",
3986                                      procfs_entry, target_print_req,"mdt_mdsc");
3987         if (!m->mdt_mdsc_service) {
3988                 CERROR("failed to start seq controller service\n");
3989                 GOTO(err_mdt_svc, rc = -ENOMEM);
3990         }
3991
3992         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3993         if (rc)
3994                 GOTO(err_mdt_svc, rc);
3995
3996         /*
3997          * metadata sequence server service configuration
3998          */
3999         conf = (typeof(conf)) {
4000                 .psc_nbufs           = MDS_NBUFS,
4001                 .psc_bufsize         = MDS_BUFSIZE,
4002                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4003                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4004                 .psc_req_portal      = SEQ_METADATA_PORTAL,
4005                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4006                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4007                 .psc_min_threads     = SEQ_NUM_THREADS,
4008                 .psc_max_threads     = SEQ_NUM_THREADS,
4009                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4010         };
4011
4012         m->mdt_mdss_service =
4013                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4014                                      LUSTRE_MDT_NAME"_mdss",
4015                                      procfs_entry, target_print_req,"mdt_mdss");
4016         if (!m->mdt_mdss_service) {
4017                 CERROR("failed to start metadata seq server service\n");
4018                 GOTO(err_mdt_svc, rc = -ENOMEM);
4019         }
4020
4021         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4022         if (rc)
4023                 GOTO(err_mdt_svc, rc);
4024
4025
4026         /*
4027          * Data sequence server service configuration. We want to have really
4028          * cluster-wide sequences space. This is why we start only one sequence
4029          * controller which manages space.
4030          */
4031         conf = (typeof(conf)) {
4032                 .psc_nbufs           = MDS_NBUFS,
4033                 .psc_bufsize         = MDS_BUFSIZE,
4034                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4035                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4036                 .psc_req_portal      = SEQ_DATA_PORTAL,
4037                 .psc_rep_portal      = OSC_REPLY_PORTAL,
4038                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4039                 .psc_min_threads     = SEQ_NUM_THREADS,
4040                 .psc_max_threads     = SEQ_NUM_THREADS,
4041                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4042         };
4043
4044         m->mdt_dtss_service =
4045                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4046                                      LUSTRE_MDT_NAME"_dtss",
4047                                      procfs_entry, target_print_req,"mdt_dtss");
4048         if (!m->mdt_dtss_service) {
4049                 CERROR("failed to start data seq server service\n");
4050                 GOTO(err_mdt_svc, rc = -ENOMEM);
4051         }
4052
4053         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4054         if (rc)
4055                 GOTO(err_mdt_svc, rc);
4056
4057         /* FLD service start */
4058         conf = (typeof(conf)) {
4059                 .psc_nbufs           = MDS_NBUFS,
4060                 .psc_bufsize         = MDS_BUFSIZE,
4061                 .psc_max_req_size    = FLD_MAXREQSIZE,
4062                 .psc_max_reply_size  = FLD_MAXREPSIZE,
4063                 .psc_req_portal      = FLD_REQUEST_PORTAL,
4064                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4065                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4066                 .psc_min_threads     = FLD_NUM_THREADS,
4067                 .psc_max_threads     = FLD_NUM_THREADS,
4068                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
4069         };
4070
4071         m->mdt_fld_service =
4072                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4073                                      LUSTRE_MDT_NAME"_fld",
4074                                      procfs_entry, target_print_req, "mdt_fld");
4075         if (!m->mdt_fld_service) {
4076                 CERROR("failed to start fld service\n");
4077                 GOTO(err_mdt_svc, rc = -ENOMEM);
4078         }
4079
4080         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4081         if (rc)
4082                 GOTO(err_mdt_svc, rc);
4083
4084         /*
4085          * mds-mds service configuration. Separate portal is used to allow
4086          * mds-mds requests be not blocked during recovery.
4087          */
4088         conf = (typeof(conf)) {
4089                 .psc_nbufs           = MDS_NBUFS,
4090                 .psc_bufsize         = MDS_BUFSIZE,
4091                 .psc_max_req_size    = MDS_MAXREQSIZE,
4092                 .psc_max_reply_size  = MDS_MAXREPSIZE,
4093                 .psc_req_portal      = MDS_MDS_PORTAL,
4094                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4095                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4096                 .psc_min_threads    = min(max(mdt_num_threads, MDT_MIN_THREADS),
4097                                           MDT_MAX_THREADS),
4098                 .psc_max_threads     = MDT_MAX_THREADS,
4099                 .psc_ctx_tags        = LCT_MD_THREAD
4100         };
4101         m->mdt_xmds_service =
4102                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4103                                      LUSTRE_MDT_NAME "_mds",
4104                                      procfs_entry, target_print_req,"mdt_xmds");
4105
4106         if (m->mdt_xmds_service == NULL) {
4107                 CERROR("failed to start xmds service\n");
4108                 GOTO(err_mdt_svc, rc = -ENOMEM);
4109         }
4110
4111         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4112         if (rc)
4113                 GOTO(err_mdt_svc, rc);
4114
4115         EXIT;
4116 err_mdt_svc:
4117         if (rc)
4118                 mdt_stop_ptlrpc_service(m);
4119
4120         return rc;
4121 }
4122
4123 static void mdt_stack_fini(const struct lu_env *env,
4124                            struct mdt_device *m, struct lu_device *top)
4125 {
4126         struct obd_device       *obd = mdt2obd_dev(m);
4127         struct lustre_cfg_bufs  *bufs;
4128         struct lustre_cfg       *lcfg;
4129         struct mdt_thread_info  *info;
4130         char flags[3]="";
4131         ENTRY;
4132
4133         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4134         LASSERT(info != NULL);
4135
4136         bufs = &info->mti_u.bufs;
4137         /* process cleanup, pass mdt obd name to get obd umount flags */
4138         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4139         if (obd->obd_force)
4140                 strcat(flags, "F");
4141         if (obd->obd_fail)
4142                 strcat(flags, "A");
4143         lustre_cfg_bufs_set_string(bufs, 1, flags);
4144         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4145         if (!lcfg) {
4146                 CERROR("Cannot alloc lcfg!\n");
4147                 return;
4148         }
4149
4150         LASSERT(top);
4151         top->ld_ops->ldo_process_config(env, top, lcfg);
4152         lustre_cfg_free(lcfg);
4153
4154         lu_stack_fini(env, top);
4155         m->mdt_child = NULL;
4156         m->mdt_bottom = NULL;
4157 }
4158
4159 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4160                                          const char *typename,
4161                                          struct lu_device *child,
4162                                          struct lustre_cfg *cfg)
4163 {
4164         const char            *dev = lustre_cfg_string(cfg, 0);
4165         struct obd_type       *type;
4166         struct lu_device_type *ldt;
4167         struct lu_device      *d;
4168         int rc;
4169         ENTRY;
4170
4171         /* find the type */
4172         type = class_get_type(typename);
4173         if (!type) {
4174                 CERROR("Unknown type: '%s'\n", typename);
4175                 GOTO(out, rc = -ENODEV);
4176         }
4177
4178         rc = lu_env_refill((struct lu_env *)env);
4179         if (rc != 0) {
4180                 CERROR("Failure to refill session: '%d'\n", rc);
4181                 GOTO(out_type, rc);
4182         }
4183
4184         ldt = type->typ_lu;
4185         if (ldt == NULL) {
4186                 CERROR("type: '%s'\n", typename);
4187                 GOTO(out_type, rc = -EINVAL);
4188         }
4189
4190         ldt->ldt_obd_type = type;
4191         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4192         if (IS_ERR(d)) {
4193                 CERROR("Cannot allocate device: '%s'\n", typename);
4194                 GOTO(out_type, rc = -ENODEV);
4195         }
4196
4197         LASSERT(child->ld_site);
4198         d->ld_site = child->ld_site;
4199
4200         type->typ_refcnt++;
4201         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4202         if (rc) {
4203                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4204                 GOTO(out_alloc, rc);
4205         }
4206         lu_device_get(d);
4207         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4208
4209         RETURN(d);
4210
4211 out_alloc:
4212         ldt->ldt_ops->ldto_device_free(env, d);
4213         type->typ_refcnt--;
4214 out_type:
4215         class_put_type(type);
4216 out:
4217         return ERR_PTR(rc);
4218 }
4219
4220 static int mdt_stack_init(struct lu_env *env,
4221                           struct mdt_device *m,
4222                           struct lustre_cfg *cfg,
4223                           struct lustre_mount_info  *lmi)
4224 {
4225         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4226         struct lu_device  *tmp;
4227         struct md_device  *md;
4228         struct lu_device  *child_lu_dev;
4229         int rc;
4230         ENTRY;
4231
4232         /* init the stack */
4233         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4234         if (IS_ERR(tmp)) {
4235                 RETURN(PTR_ERR(tmp));
4236         }
4237         m->mdt_bottom = lu2dt_dev(tmp);
4238         d = tmp;
4239         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4240         if (IS_ERR(tmp)) {
4241                 GOTO(out, rc = PTR_ERR(tmp));
4242         }
4243         d = tmp;
4244         md = lu2md_dev(d);
4245
4246         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4247         if (IS_ERR(tmp)) {
4248                 GOTO(out, rc = PTR_ERR(tmp));
4249         }
4250         d = tmp;
4251         /*set mdd upcall device*/
4252         md_upcall_dev_set(md, lu2md_dev(d));
4253
4254         md = lu2md_dev(d);
4255         /*set cmm upcall device*/
4256         md_upcall_dev_set(md, &m->mdt_md_dev);
4257
4258         m->mdt_child = lu2md_dev(d);
4259
4260         /* process setup config */
4261         tmp = &m->mdt_md_dev.md_lu_dev;
4262         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4263         if (rc)
4264                 GOTO(out, rc);
4265
4266         /* initialize local objects */
4267         child_lu_dev = &m->mdt_child->md_lu_dev;
4268
4269         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4270                                                &m->mdt_md_dev.md_lu_dev,
4271                                                child_lu_dev);
4272 out:
4273         /* fini from last known good lu_device */
4274         if (rc)
4275                 mdt_stack_fini(env, m, d);
4276
4277         return rc;
4278 }
4279
4280 /**
4281  * setup CONFIG_ORIG context, used to access local config log.
4282  * this may need to be rewrite as part of llog rewrite for lu-api.
4283  */
4284 static int mdt_obd_llog_setup(struct obd_device *obd,
4285                               struct lustre_sb_info *lsi)
4286 {
4287         int     rc;
4288
4289         LASSERT(obd->obd_fsops == NULL);
4290
4291         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4292         if (IS_ERR(obd->obd_fsops))
4293                 return PTR_ERR(obd->obd_fsops);
4294
4295         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4296         if (rc) {
4297                 fsfilt_put_ops(obd->obd_fsops);
4298                 return rc;
4299         }
4300
4301         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4302         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4303         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4304         obd->obd_lvfs_ctxt.fs = get_ds();
4305
4306         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4307                         0, NULL, &llog_lvfs_ops);
4308         if (rc) {
4309                 CERROR("llog_setup() failed: %d\n", rc);
4310                 fsfilt_put_ops(obd->obd_fsops);
4311         }
4312
4313         return rc;
4314 }
4315
4316 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4317 {
4318         struct llog_ctxt *ctxt;
4319
4320         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4321         if (ctxt)
4322                 llog_cleanup(ctxt);
4323
4324         if (obd->obd_fsops) {
4325                 fsfilt_put_ops(obd->obd_fsops);
4326                 obd->obd_fsops = NULL;
4327         }
4328 }
4329
4330 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4331 {
4332         struct md_device  *next = m->mdt_child;
4333         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4334         struct lu_site    *ls   = d->ld_site;
4335         struct obd_device *obd = mdt2obd_dev(m);
4336         ENTRY;
4337
4338         target_recovery_fini(obd);
4339
4340         ping_evictor_stop();
4341
4342         mdt_stop_ptlrpc_service(m);
4343         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4344         mdt_obd_llog_cleanup(obd);
4345         obd_exports_barrier(obd);
4346         obd_zombie_barrier();
4347 #ifdef HAVE_QUOTA_SUPPORT
4348         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4349 #endif
4350         lut_fini(env, &m->mdt_lut);
4351         mdt_fs_cleanup(env, m);
4352         upcall_cache_cleanup(m->mdt_identity_cache);
4353         m->mdt_identity_cache = NULL;
4354
4355         if (m->mdt_namespace != NULL) {
4356                 ldlm_namespace_free(m->mdt_namespace, NULL,
4357                                     d->ld_obd->obd_force);
4358                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4359         }
4360
4361         cfs_free_nidlist(&m->mdt_nosquash_nids);
4362         if (m->mdt_nosquash_str) {
4363                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4364                 m->mdt_nosquash_str = NULL;
4365                 m->mdt_nosquash_strlen = 0;
4366         }
4367
4368         mdt_seq_fini(env, m);
4369         mdt_seq_fini_cli(m);
4370         mdt_fld_fini(env, m);
4371         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4372
4373         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4374         cfs_timer_disarm(&m->mdt_ck_timer);
4375         mdt_ck_thread_stop(m);
4376
4377         /*
4378          * Finish the stack
4379          */
4380         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4381
4382         lprocfs_free_per_client_stats(obd);
4383         lprocfs_free_obd_stats(obd);
4384         mdt_procfs_fini(m);
4385
4386         if (ls) {
4387                 struct md_site *mite;
4388
4389                 lu_site_fini(ls);
4390                 mite = lu_site2md(ls);
4391                 OBD_FREE_PTR(mite);
4392                 d->ld_site = NULL;
4393         }
4394         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4395
4396         EXIT;
4397 }
4398
4399 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4400 {
4401         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4402         struct sptlrpc_rule_set  tmp_rset;
4403         int                      rc;
4404
4405         sptlrpc_rule_set_init(&tmp_rset);
4406         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4407         if (rc) {
4408                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4409                        obd->obd_name, rc);
4410                 return rc;
4411         }
4412
4413         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4414
4415         cfs_write_lock(&m->mdt_sptlrpc_lock);
4416         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4417         m->mdt_sptlrpc_rset = tmp_rset;
4418         cfs_write_unlock(&m->mdt_sptlrpc_lock);
4419
4420         return 0;
4421 }
4422
4423 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4424 {
4425         char *p = options;
4426
4427         m->mdt_opts.mo_mds_capa = 1;
4428         m->mdt_opts.mo_oss_capa = 1;
4429 #ifdef CONFIG_FS_POSIX_ACL
4430         /* ACLs should be enabled by default (b=13829) */
4431         m->mdt_opts.mo_acl = 1;
4432         LCONSOLE_INFO("Enabling ACL\n");
4433 #else
4434         m->mdt_opts.mo_acl = 0;
4435         LCONSOLE_INFO("Disabling ACL\n");
4436 #endif
4437
4438         if (!options)
4439                 return;
4440
4441         while (*options) {
4442                 int len;
4443
4444                 while (*p && *p != ',')
4445                         p++;
4446
4447                 len = p - options;
4448                 if ((len == sizeof("user_xattr") - 1) &&
4449                     (memcmp(options, "user_xattr", len) == 0)) {
4450                         m->mdt_opts.mo_user_xattr = 1;
4451                         LCONSOLE_INFO("Enabling user_xattr\n");
4452                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4453                            (memcmp(options, "nouser_xattr", len) == 0)) {
4454                         m->mdt_opts.mo_user_xattr = 0;
4455                         LCONSOLE_INFO("Disabling user_xattr\n");
4456                 } else if ((len == sizeof("noacl") - 1) &&
4457                            (memcmp(options, "noacl", len) == 0)) {
4458                         m->mdt_opts.mo_acl = 0;
4459                         LCONSOLE_INFO("Disabling ACL\n");
4460                 }
4461
4462                 options = ++p;
4463         }
4464 }
4465
4466 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4467
4468 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4469                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4470 {
4471         struct mdt_thread_info    *info;
4472         struct obd_device         *obd;
4473         const char                *dev = lustre_cfg_string(cfg, 0);
4474         const char                *num = lustre_cfg_string(cfg, 2);
4475         struct lustre_mount_info  *lmi = NULL;
4476         struct lustre_sb_info     *lsi;
4477         struct lustre_disk_data   *ldd;
4478         struct lu_site            *s;
4479         struct md_site            *mite;
4480         const char                *identity_upcall = "NONE";
4481 #ifdef HAVE_QUOTA_SUPPORT
4482         struct md_device          *next;
4483 #endif
4484         int                        rc;
4485         int                        node_id;
4486         ENTRY;
4487
4488         md_device_init(&m->mdt_md_dev, ldt);
4489         /*
4490          * Environment (env) might be missing mdt_thread_key values at that
4491          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4492          * mode.
4493          *
4494          * Usually device allocation path doesn't use module key values, but
4495          * mdt has to do a lot of work here, so allocate key value.
4496          */
4497         rc = lu_env_refill((struct lu_env *)env);
4498         if (rc != 0)
4499                 RETURN(rc);
4500
4501         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4502         LASSERT(info != NULL);
4503
4504         obd = class_name2obd(dev);
4505         LASSERT(obd != NULL);
4506
4507         cfs_spin_lock_init(&m->mdt_transno_lock);
4508
4509         m->mdt_max_mdsize = MAX_MD_SIZE;
4510         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4511         m->mdt_som_conf = 0;
4512
4513         m->mdt_opts.mo_user_xattr = 0;
4514         m->mdt_opts.mo_acl = 0;
4515         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4516         lmi = server_get_mount_2(dev);
4517         if (lmi == NULL) {
4518                 CERROR("Cannot get mount info for %s!\n", dev);
4519                 RETURN(-EFAULT);
4520         } else {
4521                 lsi = s2lsi(lmi->lmi_sb);
4522                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4523                 /* CMD is supported only in IAM mode */
4524                 ldd = lsi->lsi_ldd;
4525                 LASSERT(num);
4526                 node_id = simple_strtol(num, NULL, 10);
4527                 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4528                         CERROR("CMD Operation not allowed in IOP mode\n");
4529                         GOTO(err_lmi, rc = -EINVAL);
4530                 }
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         cfs_spin_lock_init(&m->mdt_client_bitmap_lock);
4549
4550         OBD_ALLOC_PTR(mite);
4551         if (mite == NULL)
4552                 GOTO(err_lmi, rc = -ENOMEM);
4553
4554         s = &mite->ms_lu;
4555
4556         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4557         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4558         /* set this lu_device to obd, because error handling need it */
4559         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4560
4561         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4562         if (rc) {
4563                 CERROR("Can't init lu_site, rc %d\n", rc);
4564                 GOTO(err_free_site, rc);
4565         }
4566
4567         rc = mdt_procfs_init(m, dev);
4568         if (rc) {
4569                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4570                 GOTO(err_fini_proc, rc);
4571         }
4572
4573         /* set server index */
4574         lu_site2md(s)->ms_node_id = node_id;
4575
4576         /* failover is the default
4577          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4578          * assumed whose ms_node_id == 0 XXX
4579          * */
4580         obd->obd_replayable = 1;
4581         /* No connection accepted until configurations will finish */
4582         obd->obd_no_conn = 1;
4583
4584         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4585                 char *str = lustre_cfg_string(cfg, 4);
4586                 if (strchr(str, 'n')) {
4587                         CWARN("%s: recovery disabled\n", obd->obd_name);
4588                         obd->obd_replayable = 0;
4589                 }
4590         }
4591
4592         /* init the stack */
4593         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4594         if (rc) {
4595                 CERROR("Can't init device stack, rc %d\n", rc);
4596                 GOTO(err_fini_proc, rc);
4597         }
4598
4599         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4600         if (rc)
4601                 GOTO(err_fini_stack, rc);
4602
4603         rc = mdt_fld_init(env, obd->obd_name, m);
4604         if (rc)
4605                 GOTO(err_lut, rc);
4606
4607         rc = mdt_seq_init(env, obd->obd_name, m);
4608         if (rc)
4609                 GOTO(err_fini_fld, rc);
4610
4611         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4612                  LUSTRE_MDT_NAME"-%p", m);
4613         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4614                                               LDLM_NAMESPACE_SERVER,
4615                                               LDLM_NAMESPACE_GREEDY);
4616         if (m->mdt_namespace == NULL)
4617                 GOTO(err_fini_seq, rc = -ENOMEM);
4618
4619         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4620         /* set obd_namespace for compatibility with old code */
4621         obd->obd_namespace = m->mdt_namespace;
4622
4623         /* XXX: to support suppgid for ACL, we enable identity_upcall
4624          * by default, otherwise, maybe got unexpected -EACCESS. */
4625         if (m->mdt_opts.mo_acl)
4626                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4627
4628         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4629                                                   &mdt_identity_upcall_cache_ops);
4630         if (IS_ERR(m->mdt_identity_cache)) {
4631                 rc = PTR_ERR(m->mdt_identity_cache);
4632                 m->mdt_identity_cache = NULL;
4633                 GOTO(err_free_ns, rc);
4634         }
4635
4636         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4637
4638         rc = mdt_ck_thread_start(m);
4639         if (rc)
4640                 GOTO(err_free_ns, rc);
4641
4642         rc = mdt_fs_setup(env, m, obd, lsi);
4643         if (rc)
4644                 GOTO(err_capa, rc);
4645
4646         rc = mdt_obd_llog_setup(obd, lsi);
4647         if (rc)
4648                 GOTO(err_fs_cleanup, rc);
4649
4650         rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4651         if (rc)
4652                 GOTO(err_llog_cleanup, rc);
4653
4654         mdt_adapt_sptlrpc_conf(obd, 1);
4655
4656 #ifdef HAVE_QUOTA_SUPPORT
4657         next = m->mdt_child;
4658         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4659         if (rc)
4660                 GOTO(err_llog_cleanup, rc);
4661 #endif
4662
4663         server_put_mount_2(dev, lmi->lmi_mnt);
4664         lmi = NULL;
4665
4666         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4667
4668         rc = mdt_start_ptlrpc_service(m);
4669         if (rc)
4670                 GOTO(err_recovery, rc);
4671
4672         ping_evictor_start();
4673
4674         rc = lu_site_init_finish(s);
4675         if (rc)
4676                 GOTO(err_stop_service, rc);
4677
4678         if (obd->obd_recovering == 0)
4679                 mdt_postrecov(env, m);
4680
4681         mdt_init_capa_ctxt(env, m);
4682
4683         /* Reduce the initial timeout on an MDS because it doesn't need such
4684          * a long timeout as an OST does. Adaptive timeouts will adjust this
4685          * value appropriately. */
4686         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4687                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4688
4689         RETURN(0);
4690
4691 err_stop_service:
4692         ping_evictor_stop();
4693         mdt_stop_ptlrpc_service(m);
4694 err_recovery:
4695         target_recovery_fini(obd);
4696 #ifdef HAVE_QUOTA_SUPPORT
4697         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4698 #endif
4699 err_llog_cleanup:
4700         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4701         mdt_obd_llog_cleanup(obd);
4702 err_fs_cleanup:
4703         mdt_fs_cleanup(env, m);
4704 err_capa:
4705         cfs_timer_disarm(&m->mdt_ck_timer);
4706         mdt_ck_thread_stop(m);
4707 err_free_ns:
4708         upcall_cache_cleanup(m->mdt_identity_cache);
4709         m->mdt_identity_cache = NULL;
4710         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4711         obd->obd_namespace = m->mdt_namespace = NULL;
4712 err_fini_seq:
4713         mdt_seq_fini(env, m);
4714 err_fini_fld:
4715         mdt_fld_fini(env, m);
4716 err_lut:
4717         lut_fini(env, &m->mdt_lut);
4718 err_fini_stack:
4719         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4720 err_fini_proc:
4721         mdt_procfs_fini(m);
4722         lu_site_fini(s);
4723 err_free_site:
4724         OBD_FREE_PTR(mite);
4725 err_lmi:
4726         if (lmi)
4727                 server_put_mount_2(dev, lmi->lmi_mnt);
4728         return (rc);
4729 }
4730
4731 /* used by MGS to process specific configurations */
4732 static int mdt_process_config(const struct lu_env *env,
4733                               struct lu_device *d, struct lustre_cfg *cfg)
4734 {
4735         struct mdt_device *m = mdt_dev(d);
4736         struct md_device *md_next = m->mdt_child;
4737         struct lu_device *next = md2lu_dev(md_next);
4738         int rc = 0;
4739         ENTRY;
4740
4741         switch (cfg->lcfg_command) {
4742         case LCFG_PARAM: {
4743                 struct lprocfs_static_vars lvars;
4744                 struct obd_device *obd = d->ld_obd;
4745
4746                 /*
4747                  * For interoperability between 1.8 and 2.0,
4748                  * skip old "mdt.group_upcall" param.
4749                  */
4750                 {
4751                         char *param = lustre_cfg_string(cfg, 1);
4752                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4753                                 CWARN("For 1.8 interoperability, skip this"
4754                                        " mdt.group_upcall. It is obsolete\n");
4755                                 break;
4756                         }
4757                 }
4758
4759                 lprocfs_mdt_init_vars(&lvars);
4760                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4761                                               cfg, obd);
4762                 if (rc > 0 || rc == -ENOSYS)
4763                         /* we don't understand; pass it on */
4764                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4765                 break;
4766         }
4767         case LCFG_ADD_MDC:
4768                 /*
4769                  * Add mdc hook to get first MDT uuid and connect it to
4770                  * ls->controller to use for seq manager.
4771                  */
4772                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4773                 if (rc)
4774                         CERROR("Can't add mdc, rc %d\n", rc);
4775                 else
4776                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4777                 break;
4778         default:
4779                 /* others are passed further */
4780                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4781                 break;
4782         }
4783         RETURN(rc);
4784 }
4785
4786 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4787                                           const struct lu_object_header *hdr,
4788                                           struct lu_device *d)
4789 {
4790         struct mdt_object *mo;
4791
4792         ENTRY;
4793
4794         OBD_ALLOC_PTR(mo);
4795         if (mo != NULL) {
4796                 struct lu_object *o;
4797                 struct lu_object_header *h;
4798
4799                 o = &mo->mot_obj.mo_lu;
4800                 h = &mo->mot_header;
4801                 lu_object_header_init(h);
4802                 lu_object_init(o, h, d);
4803                 lu_object_add_top(h, o);
4804                 o->lo_ops = &mdt_obj_ops;
4805                 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4806                 RETURN(o);
4807         } else
4808                 RETURN(NULL);
4809 }
4810
4811 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4812                            const struct lu_object_conf *unused)
4813 {
4814         struct mdt_device *d = mdt_dev(o->lo_dev);
4815         struct lu_device  *under;
4816         struct lu_object  *below;
4817         int                rc = 0;
4818         ENTRY;
4819
4820         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4821                PFID(lu_object_fid(o)));
4822
4823         under = &d->mdt_child->md_lu_dev;
4824         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4825         if (below != NULL) {
4826                 lu_object_add(o, below);
4827         } else
4828                 rc = -ENOMEM;
4829
4830         RETURN(rc);
4831 }
4832
4833 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4834 {
4835         struct mdt_object *mo = mdt_obj(o);
4836         struct lu_object_header *h;
4837         ENTRY;
4838
4839         h = o->lo_header;
4840         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4841                PFID(lu_object_fid(o)));
4842
4843         lu_object_fini(o);
4844         lu_object_header_fini(h);
4845         OBD_FREE_PTR(mo);
4846         EXIT;
4847 }
4848
4849 static int mdt_object_print(const struct lu_env *env, void *cookie,
4850                             lu_printer_t p, const struct lu_object *o)
4851 {
4852         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4853         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
4854                     "flags=%llx, epochcount=%d, writecount=%d)",
4855                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4856                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4857 }
4858
4859 static const struct lu_device_operations mdt_lu_ops = {
4860         .ldo_object_alloc   = mdt_object_alloc,
4861         .ldo_process_config = mdt_process_config,
4862 };
4863
4864 static const struct lu_object_operations mdt_obj_ops = {
4865         .loo_object_init    = mdt_object_init,
4866         .loo_object_free    = mdt_object_free,
4867         .loo_object_print   = mdt_object_print
4868 };
4869
4870 static int mdt_obd_set_info_async(struct obd_export *exp,
4871                                   __u32 keylen, void *key,
4872                                   __u32 vallen, void *val,
4873                                   struct ptlrpc_request_set *set)
4874 {
4875         struct obd_device     *obd = exp->exp_obd;
4876         int                    rc;
4877         ENTRY;
4878
4879         LASSERT(obd);
4880
4881         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4882                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4883                 RETURN(rc);
4884         }
4885
4886         RETURN(0);
4887 }
4888
4889 /* mds_connect_internal */
4890 static int mdt_connect_internal(struct obd_export *exp,
4891                                 struct mdt_device *mdt,
4892                                 struct obd_connect_data *data)
4893 {
4894         if (data != NULL) {
4895                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4896                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4897
4898                 /* If no known bits (which should not happen, probably,
4899                    as everybody should support LOOKUP and UPDATE bits at least)
4900                    revert to compat mode with plain locks. */
4901                 if (!data->ocd_ibits_known &&
4902                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4903                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4904
4905                 if (!mdt->mdt_opts.mo_acl)
4906                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4907
4908                 if (!mdt->mdt_opts.mo_user_xattr)
4909                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4910
4911                 if (!mdt->mdt_som_conf)
4912                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4913
4914                 cfs_spin_lock(&exp->exp_lock);
4915                 exp->exp_connect_flags = data->ocd_connect_flags;
4916                 cfs_spin_unlock(&exp->exp_lock);
4917                 data->ocd_version = LUSTRE_VERSION_CODE;
4918                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4919         }
4920
4921 #if 0
4922         if (mdt->mdt_opts.mo_acl &&
4923             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4924                 CWARN("%s: MDS requires ACL support but client does not\n",
4925                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4926                 return -EBADE;
4927         }
4928 #endif
4929
4930         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4931                 CWARN("%s: MDS requires FID support, but client not\n",
4932                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4933                 return -EBADE;
4934         }
4935
4936         if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4937             !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4938                 CWARN("%s: MDS has SOM enabled, but client does not support "
4939                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4940                 return -EBADE;
4941         }
4942
4943         return 0;
4944 }
4945
4946 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4947                                      struct obd_export *exp,
4948                                      struct ptlrpc_request *req)
4949 {
4950         struct sptlrpc_flavor   flvr;
4951         int                     rc = 0;
4952
4953         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4954                 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4955                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4956                                              req->rq_sp_from,
4957                                              req->rq_peer.nid,
4958                                              &flvr);
4959                 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4960
4961                 cfs_spin_lock(&exp->exp_lock);
4962
4963                 exp->exp_sp_peer = req->rq_sp_from;
4964                 exp->exp_flvr = flvr;
4965
4966                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4967                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4968                         CERROR("unauthorized rpc flavor %x from %s, "
4969                                "expect %x\n", req->rq_flvr.sf_rpc,
4970                                libcfs_nid2str(req->rq_peer.nid),
4971                                exp->exp_flvr.sf_rpc);
4972                         rc = -EACCES;
4973                 }
4974
4975                 cfs_spin_unlock(&exp->exp_lock);
4976         } else {
4977                 if (exp->exp_sp_peer != req->rq_sp_from) {
4978                         CERROR("RPC source %s doesn't match %s\n",
4979                                sptlrpc_part2name(req->rq_sp_from),
4980                                sptlrpc_part2name(exp->exp_sp_peer));
4981                         rc = -EACCES;
4982                 } else {
4983                         rc = sptlrpc_target_export_check(exp, req);
4984                 }
4985         }
4986
4987         return rc;
4988 }
4989
4990 /* mds_connect copy */
4991 static int mdt_obd_connect(const struct lu_env *env,
4992                            struct obd_export **exp, struct obd_device *obd,
4993                            struct obd_uuid *cluuid,
4994                            struct obd_connect_data *data,
4995                            void *localdata)
4996 {
4997         struct mdt_thread_info *info;
4998         struct lsd_client_data *lcd;
4999         struct obd_export      *lexp;
5000         struct lustre_handle    conn = { 0 };
5001         struct mdt_device      *mdt;
5002         struct ptlrpc_request  *req;
5003         int                     rc;
5004         ENTRY;
5005
5006         LASSERT(env != NULL);
5007         if (!exp || !obd || !cluuid)
5008                 RETURN(-EINVAL);
5009
5010         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5011         req = info->mti_pill->rc_req;
5012         mdt = mdt_dev(obd->obd_lu_dev);
5013
5014         rc = class_connect(&conn, obd, cluuid);
5015         if (rc)
5016                 RETURN(rc);
5017
5018         lexp = class_conn2export(&conn);
5019         LASSERT(lexp != NULL);
5020
5021         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5022         if (rc)
5023                 GOTO(out, rc);
5024
5025         rc = mdt_connect_internal(lexp, mdt, data);
5026         if (rc == 0) {
5027                 OBD_ALLOC_PTR(lcd);
5028                 if (lcd != NULL) {
5029                         struct mdt_thread_info *mti;
5030                         mti = lu_context_key_get(&env->le_ctx,
5031                                                  &mdt_thread_key);
5032                         LASSERT(mti != NULL);
5033                         mti->mti_exp = lexp;
5034                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5035                         lexp->exp_mdt_data.med_lcd = lcd;
5036                         rc = mdt_client_new(env, mdt);
5037                         if (rc != 0) {
5038                                 OBD_FREE_PTR(lcd);
5039                                 lexp->exp_mdt_data.med_lcd = NULL;
5040                         } else {
5041                                 mdt_export_stats_init(obd, lexp, localdata);
5042                         }
5043                 } else
5044                         rc = -ENOMEM;
5045         }
5046
5047 out:
5048         if (rc != 0)
5049                 class_disconnect(lexp);
5050         else
5051                 *exp = lexp;
5052
5053         RETURN(rc);
5054 }
5055
5056 static int mdt_obd_reconnect(const struct lu_env *env,
5057                              struct obd_export *exp, struct obd_device *obd,
5058                              struct obd_uuid *cluuid,
5059                              struct obd_connect_data *data,
5060                              void *localdata)
5061 {
5062         struct mdt_thread_info *info;
5063         struct mdt_device      *mdt;
5064         struct ptlrpc_request  *req;
5065         int                     rc;
5066         ENTRY;
5067
5068         if (exp == NULL || obd == NULL || cluuid == NULL)
5069                 RETURN(-EINVAL);
5070
5071         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5072         req = info->mti_pill->rc_req;
5073         mdt = mdt_dev(obd->obd_lu_dev);
5074
5075         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5076         if (rc)
5077                 RETURN(rc);
5078
5079         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5080         if (rc == 0)
5081                 mdt_export_stats_init(obd, exp, localdata);
5082
5083         RETURN(rc);
5084 }
5085 static int mdt_export_cleanup(struct obd_export *exp)
5086 {
5087         struct mdt_export_data *med = &exp->exp_mdt_data;
5088         struct obd_device      *obd = exp->exp_obd;
5089         struct mdt_device      *mdt;
5090         struct mdt_thread_info *info;
5091         struct lu_env           env;
5092         CFS_LIST_HEAD(closing_list);
5093         struct mdt_file_data *mfd, *n;
5094         int rc = 0;
5095         ENTRY;
5096
5097         cfs_spin_lock(&med->med_open_lock);
5098         while (!cfs_list_empty(&med->med_open_head)) {
5099                 cfs_list_t *tmp = med->med_open_head.next;
5100                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5101
5102                 /* Remove mfd handle so it can't be found again.
5103                  * We are consuming the mfd_list reference here. */
5104                 class_handle_unhash(&mfd->mfd_handle);
5105                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5106         }
5107         cfs_spin_unlock(&med->med_open_lock);
5108         mdt = mdt_dev(obd->obd_lu_dev);
5109         LASSERT(mdt != NULL);
5110
5111         rc = lu_env_init(&env, LCT_MD_THREAD);
5112         if (rc)
5113                 RETURN(rc);
5114
5115         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5116         LASSERT(info != NULL);
5117         memset(info, 0, sizeof *info);
5118         info->mti_env = &env;
5119         info->mti_mdt = mdt;
5120         info->mti_exp = exp;
5121
5122         if (!cfs_list_empty(&closing_list)) {
5123                 struct md_attr *ma = &info->mti_attr;
5124                 int lmm_size;
5125                 int cookie_size;
5126
5127                 lmm_size = mdt->mdt_max_mdsize;
5128                 OBD_ALLOC(ma->ma_lmm, lmm_size);
5129                 if (ma->ma_lmm == NULL)
5130                         GOTO(out_lmm, rc = -ENOMEM);
5131
5132                 cookie_size = mdt->mdt_max_cookiesize;
5133                 OBD_ALLOC(ma->ma_cookie, cookie_size);
5134                 if (ma->ma_cookie == NULL)
5135                         GOTO(out_cookie, rc = -ENOMEM);
5136
5137                 /* Close any open files (which may also cause orphan unlinking). */
5138                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5139                         cfs_list_del_init(&mfd->mfd_list);
5140                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5141                         ma->ma_lmm_size = lmm_size;
5142                         ma->ma_cookie_size = cookie_size;
5143                         ma->ma_need = 0;
5144                         /* It is not for setattr, just tell MDD to send
5145                          * DESTROY RPC to OSS if needed */
5146                         ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5147                         ma->ma_valid = MA_FLAGS;
5148                         mdt_mfd_close(info, mfd);
5149                 }
5150                 OBD_FREE(ma->ma_cookie, cookie_size);
5151                 ma->ma_cookie = NULL;
5152 out_cookie:
5153                 OBD_FREE(ma->ma_lmm, lmm_size);
5154                 ma->ma_lmm = NULL;
5155         }
5156 out_lmm:
5157         info->mti_mdt = NULL;
5158         /* cleanup client slot early */
5159         mdt_client_del(&env, mdt);
5160         lu_env_fini(&env);
5161
5162         RETURN(rc);
5163 }
5164
5165 static int mdt_obd_disconnect(struct obd_export *exp)
5166 {
5167         int rc;
5168         ENTRY;
5169
5170         LASSERT(exp);
5171         class_export_get(exp);
5172
5173         rc = server_disconnect_export(exp);
5174         if (rc != 0)
5175                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5176
5177         rc = mdt_export_cleanup(exp);
5178         class_export_put(exp);
5179         RETURN(rc);
5180 }
5181
5182 /* FIXME: Can we avoid using these two interfaces? */
5183 static int mdt_init_export(struct obd_export *exp)
5184 {
5185         struct mdt_export_data *med = &exp->exp_mdt_data;
5186         int                     rc;
5187         ENTRY;
5188
5189         CFS_INIT_LIST_HEAD(&med->med_open_head);
5190         cfs_spin_lock_init(&med->med_open_lock);
5191         cfs_sema_init(&med->med_idmap_sem, 1);
5192         med->med_idmap = NULL;
5193         cfs_spin_lock(&exp->exp_lock);
5194         exp->exp_connecting = 1;
5195         cfs_spin_unlock(&exp->exp_lock);
5196         rc = ldlm_init_export(exp);
5197         if (rc)
5198                 CERROR("Error %d while initializing export\n", rc);
5199         RETURN(rc);
5200 }
5201
5202 static int mdt_destroy_export(struct obd_export *exp)
5203 {
5204         struct mdt_export_data *med;
5205         int rc = 0;
5206         ENTRY;
5207
5208         med = &exp->exp_mdt_data;
5209         if (exp_connect_rmtclient(exp))
5210                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5211
5212         target_destroy_export(exp);
5213         ldlm_destroy_export(exp);
5214
5215         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5216         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5217         if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5218                 RETURN(0);
5219
5220         RETURN(rc);
5221 }
5222
5223 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5224 {
5225         if (flag & CONFIG_LOG)
5226                 m->mdt_fl_cfglog = 1;
5227
5228         /* also notify active event */
5229         if (flag & CONFIG_SYNC)
5230                 m->mdt_fl_synced = 1;
5231
5232         if (m->mdt_fl_cfglog && m->mdt_fl_synced)
5233                 /* Open for clients */
5234                 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
5235 }
5236
5237 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5238                       enum md_upcall_event ev, void *data)
5239 {
5240         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5241         struct md_device  *next  = m->mdt_child;
5242         struct mdt_thread_info *mti;
5243         int rc = 0;
5244         ENTRY;
5245
5246         switch (ev) {
5247                 case MD_LOV_SYNC:
5248                         rc = next->md_ops->mdo_maxsize_get(env, next,
5249                                         &m->mdt_max_mdsize,
5250                                         &m->mdt_max_cookiesize);
5251                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5252                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5253                         mdt_allow_cli(m, CONFIG_SYNC);
5254                         if (data)
5255                                 (*(__u64 *)data) = m->mdt_mount_count;
5256                         break;
5257                 case MD_NO_TRANS:
5258                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5259                         mti->mti_no_need_trans = 1;
5260                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5261                         break;
5262                 case MD_LOV_CONFIG:
5263                         /* Check that MDT is not yet configured */
5264                         LASSERT(!m->mdt_fl_cfglog);
5265                         break;
5266 #ifdef HAVE_QUOTA_SUPPORT
5267                 case MD_LOV_QUOTA:
5268                         if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5269                             likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5270                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5271                         break;
5272 #endif
5273                 default:
5274                         CERROR("invalid event\n");
5275                         rc = -EINVAL;
5276                         break;
5277         }
5278         RETURN(rc);
5279 }
5280
5281 static int mdt_obd_notify(struct obd_device *host,
5282                           struct obd_device *watched,
5283                           enum obd_notify_event ev, void *data)
5284 {
5285         struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5286 #ifdef HAVE_QUOTA_SUPPORT
5287         struct md_device *next = mdt->mdt_child;
5288 #endif
5289         ENTRY;
5290
5291         switch (ev) {
5292         case OBD_NOTIFY_CONFIG:
5293                 mdt_allow_cli(mdt, (unsigned long)data);
5294
5295 #ifdef HAVE_QUOTA_SUPPORT
5296                /* quota_type has been processed, we can now handle
5297                 * incoming quota requests */
5298                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5299 #endif
5300                 break;
5301         default:
5302                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5303         }
5304         RETURN(0);
5305 }
5306
5307 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5308                             void *val, int vallen)
5309 {
5310         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5311         struct getinfo_fid2path *fpout, *fpin;
5312         int rc = 0;
5313
5314         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5315         fpout = val;
5316
5317         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5318                 lustre_swab_fid2path(fpin);
5319
5320         memcpy(fpout, fpin, sizeof(*fpin));
5321         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5322                 RETURN(-EINVAL);
5323
5324         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5325         RETURN(rc);
5326 }
5327
5328 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5329                         struct getinfo_fid2path *fp)
5330 {
5331         struct mdt_object *obj;
5332         int    rc;
5333         ENTRY;
5334
5335         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5336                PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5337
5338         if (!fid_is_sane(&fp->gf_fid))
5339                 RETURN(-EINVAL);
5340
5341         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5342         if (obj == NULL || IS_ERR(obj)) {
5343                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5344                        PTR_ERR(obj));
5345                 RETURN(-EINVAL);
5346         }
5347
5348         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5349         if (rc <= 0) {
5350                 if (rc == -1)
5351                         rc = -EREMOTE;
5352                 else
5353                         rc = -ENOENT;
5354                 mdt_object_put(env, obj);
5355                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5356                        PFID(&fp->gf_fid), rc);
5357                 RETURN(rc);
5358         }
5359
5360         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5361                      fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5362         mdt_object_put(env, obj);
5363
5364         RETURN(rc);
5365 }
5366
5367 static int mdt_get_info(struct mdt_thread_info *info)
5368 {
5369         struct ptlrpc_request *req = mdt_info_req(info);
5370         char *key;
5371         int keylen;
5372         __u32 *vallen;
5373         void *valout;
5374         int rc;
5375         ENTRY;
5376
5377         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5378         if (key == NULL) {
5379                 CDEBUG(D_IOCTL, "No GETINFO key");
5380                 RETURN(-EFAULT);
5381         }
5382         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5383                                       RCL_CLIENT);
5384
5385         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5386         if (vallen == NULL) {
5387                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5388                 RETURN(-EFAULT);
5389         }
5390
5391         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5392                              *vallen);
5393         rc = req_capsule_server_pack(info->mti_pill);
5394         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5395         if (valout == NULL) {
5396                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5397                 RETURN(-EFAULT);
5398         }
5399
5400         if (KEY_IS(KEY_FID2PATH))
5401                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5402         else
5403                 rc = -EINVAL;
5404
5405         lustre_msg_set_status(req->rq_repmsg, rc);
5406
5407         RETURN(rc);
5408 }
5409
5410 /* Pass the ioc down */
5411 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5412                          unsigned int cmd, int len, void *data)
5413 {
5414         struct lu_context ioctl_session;
5415         struct md_device *next = mdt->mdt_child;
5416         int rc;
5417         ENTRY;
5418
5419         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5420         if (rc)
5421                 RETURN(rc);
5422         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5423         lu_context_enter(&ioctl_session);
5424         env->le_ses = &ioctl_session;
5425
5426         LASSERT(next->md_ops->mdo_iocontrol);
5427         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5428
5429         lu_context_exit(&ioctl_session);
5430         lu_context_fini(&ioctl_session);
5431         RETURN(rc);
5432 }
5433
5434 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5435 {
5436         struct obd_ioctl_data *data = karg;
5437         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5438         __u64 version;
5439         struct mdt_object *obj;
5440         struct mdt_lock_handle  *lh;
5441         int rc;
5442         ENTRY;
5443         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5444         if (!fid_is_sane(fid))
5445                 RETURN(-EINVAL);
5446
5447         lh = &mti->mti_lh[MDT_LH_PARENT];
5448         mdt_lock_reg_init(lh, LCK_CR);
5449
5450         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5451         if (IS_ERR(obj))
5452                 RETURN(PTR_ERR(obj));
5453
5454         rc = mdt_object_exists(obj);
5455         if (rc < 0) {
5456                 rc = -EREMOTE;
5457                 /**
5458                  * before calling version get the correct MDS should be
5459                  * fid, this is error to find remote object here
5460                  */
5461                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5462         } else if (rc == 0) {
5463                 rc = -ENOENT;
5464                 CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
5465         } else {
5466                 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5467                 if (version < 0) {
5468                         rc = (int)version;
5469                 } else {
5470                         *(__u64 *)data->ioc_inlbuf2 = version;
5471                         rc = 0;
5472                 }
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         struct obd_device *obd = mdt2obd_dev(mdt);
5536 #ifdef HAVE_QUOTA_SUPPORT
5537         struct md_device *next = mdt->mdt_child;
5538 #endif
5539         int rc, lost;
5540         ENTRY;
5541         /* if some clients didn't participate in recovery then we can possibly
5542          * lost sequence. Now we should increase sequence for safe value */
5543         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5544         mdt_seq_adjust(env, mdt, lost);
5545
5546         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5547 #ifdef HAVE_QUOTA_SUPPORT
5548         if (likely(obd->obd_stopping == 0))
5549                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5550 #endif
5551         RETURN(rc);
5552 }
5553
5554 int mdt_obd_postrecov(struct obd_device *obd)
5555 {
5556         struct lu_env env;
5557         int rc;
5558
5559         rc = lu_env_init(&env, LCT_MD_THREAD);
5560         if (rc)
5561                 RETURN(rc);
5562         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5563         lu_env_fini(&env);
5564         return rc;
5565 }
5566
5567 /**
5568  * Send a copytool req to a client
5569  * Note this sends a request RPC from a server (MDT) to a client (MDC),
5570  * backwards of normal comms.
5571  */
5572 int mdt_hsm_copytool_send(struct obd_export *exp)
5573 {
5574         struct lnl_hdr *lh;
5575         struct hsm_action_list *hal;
5576         struct hsm_action_item *hai;
5577         int rc, len;
5578         ENTRY;
5579
5580         CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5581               libcfs_nid2str(exp->exp_connection->c_peer.nid));
5582
5583         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5584                 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5585         OBD_ALLOC(lh, len);
5586         if (lh == NULL)
5587                 RETURN(-ENOMEM);
5588
5589         lh->lnl_magic = LNL_MAGIC;
5590         lh->lnl_transport = LNL_TRANSPORT_HSM;
5591         lh->lnl_msgtype = HMT_ACTION_LIST;
5592         lh->lnl_msglen = len;
5593
5594         hal = (struct hsm_action_list *)(lh + 1);
5595         hal->hal_version = HAL_VERSION;
5596         hal->hal_archive_num = 1;
5597         obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5598                         MTI_NAME_MAXLEN);
5599
5600         /* mock up an action list */
5601         hal->hal_count = 2;
5602         hai = hai_zero(hal);
5603         hai->hai_action = HSMA_ARCHIVE;
5604         hai->hai_fid.f_oid = 0xA00A;
5605         hai->hai_len = sizeof(*hai);
5606         hai = hai_next(hai);
5607         hai->hai_action = HSMA_RESTORE;
5608         hai->hai_fid.f_oid = 0xB00B;
5609         hai->hai_len = sizeof(*hai);
5610
5611         /* Uses the ldlm reverse import; this rpc will be seen by
5612           the ldlm_callback_handler */
5613         rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO,
5614                                  sizeof(KEY_HSM_COPYTOOL_SEND),
5615                                  KEY_HSM_COPYTOOL_SEND,
5616                                  len, lh, NULL);
5617
5618         OBD_FREE(lh, len);
5619
5620         RETURN(rc);
5621 }
5622
5623 static struct obd_ops mdt_obd_device_ops = {
5624         .o_owner          = THIS_MODULE,
5625         .o_set_info_async = mdt_obd_set_info_async,
5626         .o_connect        = mdt_obd_connect,
5627         .o_reconnect      = mdt_obd_reconnect,
5628         .o_disconnect     = mdt_obd_disconnect,
5629         .o_init_export    = mdt_init_export,
5630         .o_destroy_export = mdt_destroy_export,
5631         .o_iocontrol      = mdt_iocontrol,
5632         .o_postrecov      = mdt_obd_postrecov,
5633         .o_notify         = mdt_obd_notify
5634 };
5635
5636 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5637                                          struct lu_device *d)
5638 {
5639         struct mdt_device *m = mdt_dev(d);
5640         ENTRY;
5641
5642         mdt_fini(env, m);
5643         RETURN(NULL);
5644 }
5645
5646 static struct lu_device *mdt_device_free(const struct lu_env *env,
5647                                          struct lu_device *d)
5648 {
5649         struct mdt_device *m = mdt_dev(d);
5650         ENTRY;
5651
5652         md_device_fini(&m->mdt_md_dev);
5653         OBD_FREE_PTR(m);
5654         RETURN(NULL);
5655 }
5656
5657 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5658                                           struct lu_device_type *t,
5659                                           struct lustre_cfg *cfg)
5660 {
5661         struct lu_device  *l;
5662         struct mdt_device *m;
5663
5664         OBD_ALLOC_PTR(m);
5665         if (m != NULL) {
5666                 int rc;
5667
5668                 l = &m->mdt_md_dev.md_lu_dev;
5669                 rc = mdt_init0(env, m, t, cfg);
5670                 if (rc != 0) {
5671                         mdt_device_free(env, l);
5672                         l = ERR_PTR(rc);
5673                         return l;
5674                 }
5675                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5676         } else
5677                 l = ERR_PTR(-ENOMEM);
5678         return l;
5679 }
5680
5681 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5682 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5683
5684 /* context key: mdt_thread_key */
5685 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5686
5687 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5688 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5689
5690 struct lu_context_key mdt_txn_key = {
5691         .lct_tags = LCT_TX_HANDLE,
5692         .lct_init = mdt_txn_key_init,
5693         .lct_fini = mdt_txn_key_fini
5694 };
5695
5696 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5697 {
5698         return md_ucred(info->mti_env);
5699 }
5700
5701 /**
5702  * Enable/disable COS (Commit On Sharing).
5703  *
5704  * Set/Clear the COS flag in mdt options.
5705  *
5706  * \param mdt mdt device
5707  * \param val 0 disables COS, other values enable COS
5708  */
5709 void mdt_enable_cos(struct mdt_device *mdt, int val)
5710 {
5711         struct lu_env env;
5712         int rc;
5713
5714         mdt->mdt_opts.mo_cos = !!val;
5715         rc = lu_env_init(&env, LCT_MD_THREAD);
5716         if (unlikely(rc != 0)) {
5717                 CWARN("lu_env initialization failed with rc = %d,"
5718                       "cannot sync\n", rc);
5719                 return;
5720         }
5721         mdt_device_sync(&env, mdt);
5722         lu_env_fini(&env);
5723 }
5724
5725 /**
5726  * Check COS (Commit On Sharing) status.
5727  *
5728  * Return COS flag status.
5729  *
5730  * \param mdt mdt device
5731  */
5732 int mdt_cos_is_enabled(struct mdt_device *mdt)
5733 {
5734         return mdt->mdt_opts.mo_cos != 0;
5735 }
5736
5737 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5738 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5739
5740 static struct lu_device_type_operations mdt_device_type_ops = {
5741         .ldto_init = mdt_type_init,
5742         .ldto_fini = mdt_type_fini,
5743
5744         .ldto_start = mdt_type_start,
5745         .ldto_stop  = mdt_type_stop,
5746
5747         .ldto_device_alloc = mdt_device_alloc,
5748         .ldto_device_free  = mdt_device_free,
5749         .ldto_device_fini  = mdt_device_fini
5750 };
5751
5752 static struct lu_device_type mdt_device_type = {
5753         .ldt_tags     = LU_DEVICE_MD,
5754         .ldt_name     = LUSTRE_MDT_NAME,
5755         .ldt_ops      = &mdt_device_type_ops,
5756         .ldt_ctx_tags = LCT_MD_THREAD
5757 };
5758
5759 static struct lu_local_obj_desc mdt_last_recv = {
5760         .llod_name      = LAST_RCVD,
5761         .llod_oid       = MDT_LAST_RECV_OID,
5762         .llod_is_index  = 0,
5763 };
5764
5765 static int __init mdt_mod_init(void)
5766 {
5767         struct lprocfs_static_vars lvars;
5768         int rc;
5769
5770         llo_local_obj_register(&mdt_last_recv);
5771
5772         mdt_num_threads = MDT_NUM_THREADS;
5773         lprocfs_mdt_init_vars(&lvars);
5774         rc = class_register_type(&mdt_obd_device_ops, NULL,
5775                                  lvars.module_vars, LUSTRE_MDT_NAME,
5776                                  &mdt_device_type);
5777
5778         return rc;
5779 }
5780
5781 static void __exit mdt_mod_exit(void)
5782 {
5783         llo_local_obj_unregister(&mdt_last_recv);
5784         class_unregister_type(LUSTRE_MDT_NAME);
5785 }
5786
5787
5788 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5789 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5790         .mh_name    = #opc,                                             \
5791         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5792         .mh_opc     = prefix ## _  ## opc,                              \
5793         .mh_flags   = flags,                                            \
5794         .mh_act     = fn,                                               \
5795         .mh_fmt     = fmt                                               \
5796 }
5797
5798 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5799         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5800
5801 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5802         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5803
5804 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5805         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5806 /*
5807  * Request with a format known in advance
5808  */
5809 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5810         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5811
5812 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5813         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5814
5815 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5816         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5817 /*
5818  * Request with a format we do not yet know
5819  */
5820 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5821         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5822
5823 static struct mdt_handler mdt_mds_ops[] = {
5824 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5825 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5826 DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
5827                                                              &RQF_OBD_SET_INFO),
5828 DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
5829 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5830 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5831 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5832 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5833 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5834 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5835 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5836 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5837 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5838 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5839 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5840 #ifdef HAVE_QUOTA_SUPPORT
5841 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5842 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5843 #endif
5844 };
5845
5846 #define DEF_OBD_HNDL(flags, name, fn)                   \
5847         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5848
5849
5850 static struct mdt_handler mdt_obd_ops[] = {
5851         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5852         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5853         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5854 };
5855
5856 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5857         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5858 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5859         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5860
5861 static struct mdt_handler mdt_dlm_ops[] = {
5862         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5863         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5864         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5865         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5866 };
5867
5868 #define DEF_LLOG_HNDL(flags, name, fn)                   \
5869         DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5870
5871 static struct mdt_handler mdt_llog_ops[] = {
5872         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
5873         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
5874         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5875         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
5876         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
5877         DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
5878         DEF_LLOG_HNDL(0, CATINFO,                   NULL),
5879         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
5880         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
5881 };
5882
5883 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5884         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5885
5886 static struct mdt_handler mdt_sec_ctx_ops[] = {
5887         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5888         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5889         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5890 };
5891
5892 static struct mdt_opc_slice mdt_regular_handlers[] = {
5893         {
5894                 .mos_opc_start = MDS_GETATTR,
5895                 .mos_opc_end   = MDS_LAST_OPC,
5896                 .mos_hs        = mdt_mds_ops
5897         },
5898         {
5899                 .mos_opc_start = OBD_PING,
5900                 .mos_opc_end   = OBD_LAST_OPC,
5901                 .mos_hs        = mdt_obd_ops
5902         },
5903         {
5904                 .mos_opc_start = LDLM_ENQUEUE,
5905                 .mos_opc_end   = LDLM_LAST_OPC,
5906                 .mos_hs        = mdt_dlm_ops
5907         },
5908         {
5909                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5910                 .mos_opc_end   = LLOG_LAST_OPC,
5911                 .mos_hs        = mdt_llog_ops
5912         },
5913         {
5914                 .mos_opc_start = SEC_CTX_INIT,
5915                 .mos_opc_end   = SEC_LAST_OPC,
5916                 .mos_hs        = mdt_sec_ctx_ops
5917         },
5918         {
5919                 .mos_hs        = NULL
5920         }
5921 };
5922
5923 static struct mdt_handler mdt_readpage_ops[] = {
5924         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5925         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5926 #ifdef HAVE_SPLIT_SUPPORT
5927         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5928 #endif
5929
5930         /*
5931          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5932          * detailed comments. --umka
5933          */
5934         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5935         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5936 };
5937
5938 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5939         {
5940                 .mos_opc_start = MDS_GETATTR,
5941                 .mos_opc_end   = MDS_LAST_OPC,
5942                 .mos_hs        = mdt_readpage_ops
5943         },
5944         {
5945                 .mos_hs        = NULL
5946         }
5947 };
5948
5949 static struct mdt_handler mdt_xmds_ops[] = {
5950         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5951         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5952         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5953         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5954 };
5955
5956 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5957         {
5958                 .mos_opc_start = MDS_GETATTR,
5959                 .mos_opc_end   = MDS_LAST_OPC,
5960                 .mos_hs        = mdt_xmds_ops
5961         },
5962         {
5963                 .mos_opc_start = OBD_PING,
5964                 .mos_opc_end   = OBD_LAST_OPC,
5965                 .mos_hs        = mdt_obd_ops
5966         },
5967         {
5968                 .mos_opc_start = SEC_CTX_INIT,
5969                 .mos_opc_end   = SEC_LAST_OPC,
5970                 .mos_hs        = mdt_sec_ctx_ops
5971         },
5972         {
5973                 .mos_hs        = NULL
5974         }
5975 };
5976
5977 static struct mdt_handler mdt_seq_ops[] = {
5978         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5979 };
5980
5981 static struct mdt_opc_slice mdt_seq_handlers[] = {
5982         {
5983                 .mos_opc_start = SEQ_QUERY,
5984                 .mos_opc_end   = SEQ_LAST_OPC,
5985                 .mos_hs        = mdt_seq_ops
5986         },
5987         {
5988                 .mos_hs        = NULL
5989         }
5990 };
5991
5992 static struct mdt_handler mdt_fld_ops[] = {
5993         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5994 };
5995
5996 static struct mdt_opc_slice mdt_fld_handlers[] = {
5997         {
5998                 .mos_opc_start = FLD_QUERY,
5999                 .mos_opc_end   = FLD_LAST_OPC,
6000                 .mos_hs        = mdt_fld_ops
6001         },
6002         {
6003                 .mos_hs        = NULL
6004         }
6005 };
6006
6007 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6008 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6009 MODULE_LICENSE("GPL");
6010
6011 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6012                 "number of mdt service threads to start");
6013
6014 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);