lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #ifdef HAVE_QUOTA_SUPPORT
  71 # include <lustre_quota.h>
  72 #endif
  73 #include <lustre_acl.h>
  74 #include <lustre_param.h>
  75 #include <lustre_fsfilt.h>
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 ldlm_mode_t mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE] = LCK_MINMODE,
  90         [MDL_EX]      = LCK_EX,
  91         [MDL_PW]      = LCK_PW,
  92         [MDL_PR]      = LCK_PR,
  93         [MDL_CW]      = LCK_CW,
  94         [MDL_CR]      = LCK_CR,
  95         [MDL_NL]      = LCK_NL,
  96         [MDL_GROUP]   = LCK_GROUP
  97 };
  98
  99 /*
 100  * Initialized in mdt_mod_init().
 101  */
 102 static unsigned long mdt_num_threads;
 103 static unsigned long mdt_min_threads;
 104 static unsigned long mdt_max_threads;
 105
 106 /* ptlrpc request handler for MDT. All handlers are
 107  * grouped into several slices - struct mdt_opc_slice,
 108  * and stored in an array - mdt_handlers[].
 109  */
 110 struct mdt_handler {
 111         /* The name of this handler. */
 112         const char *mh_name;
 113         /* Fail id for this handler, checked at the beginning of this handler*/
 114         int         mh_fail_id;
 115         /* Operation code for this handler */
 116         __u32       mh_opc;
 117         /* flags are listed in enum mdt_handler_flags below. */
 118         __u32       mh_flags;
 119         /* The actual handler function to execute. */
 120         int (*mh_act)(struct mdt_thread_info *info);
 121         /* Request format for this request. */
 122         const struct req_format *mh_fmt;
 123 };
 124
 125 enum mdt_handler_flags {
 126         /*
 127          * struct mdt_body is passed in the incoming message, and object
 128          * identified by this fid exists on disk.
 129          *
 130          * "habeo corpus" == "I have a body"
 131          */
 132         HABEO_CORPUS = (1 << 0),
 133         /*
 134          * struct ldlm_request is passed in the incoming message.
 135          *
 136          * "habeo clavis" == "I have a key"
 137          */
 138         HABEO_CLAVIS = (1 << 1),
 139         /*
 140          * this request has fixed reply format, so that reply message can be
 141          * packed by generic code.
 142          *
 143          * "habeo refero" == "I have a reply"
 144          */
 145         HABEO_REFERO = (1 << 2),
 146         /*
 147          * this request will modify something, so check whether the filesystem
 148          * is readonly or not, then return -EROFS to client asap if necessary.
 149          *
 150          * "mutabor" == "I shall modify"
 151          */
 152         MUTABOR      = (1 << 3)
 153 };
 154
 155 struct mdt_opc_slice {
 156         __u32               mos_opc_start;
 157         int                 mos_opc_end;
 158         struct mdt_handler *mos_hs;
 159 };
 160
 161 static struct mdt_opc_slice mdt_regular_handlers[];
 162 static struct mdt_opc_slice mdt_readpage_handlers[];
 163 static struct mdt_opc_slice mdt_xmds_handlers[];
 164 static struct mdt_opc_slice mdt_seq_handlers[];
 165 static struct mdt_opc_slice mdt_fld_handlers[];
 166
 167 static struct mdt_device *mdt_dev(struct lu_device *d);
 168 static int mdt_regular_handle(struct ptlrpc_request *req);
 169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
 171                         struct getinfo_fid2path *fp);
 172
 173 static const struct lu_object_operations mdt_obj_ops;
 174
 175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 176 {
 177         if (!rep)
 178                 return 0;
 179         return (rep->lock_policy_res1 & flag);
 180 }
 181
 182 void mdt_clear_disposition(struct mdt_thread_info *info,
 183                            struct ldlm_reply *rep, int flag)
 184 {
 185         if (info)
 186                 info->mti_opdata &= ~flag;
 187         if (rep)
 188                 rep->lock_policy_res1 &= ~flag;
 189 }
 190
 191 void mdt_set_disposition(struct mdt_thread_info *info,
 192                          struct ldlm_reply *rep, int flag)
 193 {
 194         if (info)
 195                 info->mti_opdata |= flag;
 196         if (rep)
 197                 rep->lock_policy_res1 |= flag;
 198 }
 199
 200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 201 {
 202         lh->mlh_pdo_hash = 0;
 203         lh->mlh_reg_mode = lm;
 204         lh->mlh_type = MDT_REG_LOCK;
 205 }
 206
 207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 208                        const char *name, int namelen)
 209 {
 210         lh->mlh_reg_mode = lm;
 211         lh->mlh_type = MDT_PDO_LOCK;
 212
 213         if (name != NULL) {
 214                 LASSERT(namelen > 0);
 215                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 216         } else {
 217                 LASSERT(namelen == 0);
 218                 lh->mlh_pdo_hash = 0ull;
 219         }
 220 }
 221
 222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 223                               struct mdt_lock_handle *lh)
 224 {
 225         mdl_mode_t mode;
 226         ENTRY;
 227
 228         /*
 229          * Any dir access needs couple of locks:
 230          *
 231          * 1) on part of dir we gonna take lookup/modify;
 232          *
 233          * 2) on whole dir to protect it from concurrent splitting and/or to
 234          * flush client's cache for readdir().
 235          *
 236          * so, for a given mode and object this routine decides what lock mode
 237          * to use for lock #2:
 238          *
 239          * 1) if caller's gonna lookup in dir then we need to protect dir from
 240          * being splitted only - LCK_CR
 241          *
 242          * 2) if caller's gonna modify dir then we need to protect dir from
 243          * being splitted and to flush cache - LCK_CW
 244          *
 245          * 3) if caller's gonna modify dir and that dir seems ready for
 246          * splitting then we need to protect it from any type of access
 247          * (lookup/modify/split) - LCK_EX --bzzz
 248          */
 249
 250         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 251         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 252
 253         /*
 254          * Ask underlaying level its opinion about preferable PDO lock mode
 255          * having access type passed as regular lock mode:
 256          *
 257          * - MDL_MINMODE means that lower layer does not want to specify lock
 258          * mode;
 259          *
 260          * - MDL_NL means that no PDO lock should be taken. This is used in some
 261          * cases. Say, for non-splittable directories no need to use PDO locks
 262          * at all.
 263          */
 264         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 265                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 266
 267         if (mode != MDL_MINMODE) {
 268                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 269         } else {
 270                 /*
 271                  * Lower layer does not want to specify locking mode. We do it
 272                  * our selves. No special protection is needed, just flush
 273                  * client's cache on modification and allow concurrent
 274                  * mondification.
 275                  */
 276                 switch (lh->mlh_reg_mode) {
 277                 case LCK_EX:
 278                         lh->mlh_pdo_mode = LCK_EX;
 279                         break;
 280                 case LCK_PR:
 281                         lh->mlh_pdo_mode = LCK_CR;
 282                         break;
 283                 case LCK_PW:
 284                         lh->mlh_pdo_mode = LCK_CW;
 285                         break;
 286                 default:
 287                         CERROR("Not expected lock type (0x%x)\n",
 288                                (int)lh->mlh_reg_mode);
 289                         LBUG();
 290                 }
 291         }
 292
 293         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 294         EXIT;
 295 }
 296
 297 static int mdt_getstatus(struct mdt_thread_info *info)
 298 {
 299         struct mdt_device *mdt  = info->mti_mdt;
 300         struct md_device  *next = mdt->mdt_child;
 301         struct mdt_body   *repbody;
 302         int                rc;
 303
 304         ENTRY;
 305
 306         rc = mdt_check_ucred(info);
 307         if (rc)
 308                 RETURN(err_serious(rc));
 309
 310         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 311                 RETURN(err_serious(-ENOMEM));
 312
 313         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 314         rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
 315         if (rc != 0)
 316                 RETURN(rc);
 317
 318         repbody->valid |= OBD_MD_FLID;
 319
 320         if (mdt->mdt_opts.mo_mds_capa &&
 321             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 322                 struct mdt_object  *root;
 323                 struct lustre_capa *capa;
 324
 325                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 326                 if (IS_ERR(root))
 327                         RETURN(PTR_ERR(root));
 328
 329                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 330                 LASSERT(capa);
 331                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 332                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 333                                  0);
 334                 mdt_object_put(info->mti_env, root);
 335                 if (rc == 0)
 336                         repbody->valid |= OBD_MD_FLMDSCAPA;
 337         }
 338
 339         RETURN(rc);
 340 }
 341
 342 static int mdt_statfs(struct mdt_thread_info *info)
 343 {
 344         struct md_device      *next  = info->mti_mdt->mdt_child;
 345         struct ptlrpc_service *svc;
 346         struct obd_statfs     *osfs;
 347         int                    rc;
 348
 349         ENTRY;
 350
 351         svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
 352
 353         /* This will trigger a watchdog timeout */
 354         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 355                          (MDT_SERVICE_WATCHDOG_FACTOR *
 356                           at_get(&svc->srv_at_estimate)) + 1);
 357
 358         rc = mdt_check_ucred(info);
 359         if (rc)
 360                 RETURN(err_serious(rc));
 361
 362         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
 363                 rc = err_serious(-ENOMEM);
 364         } else {
 365                 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 366                 rc = next->md_ops->mdo_statfs(info->mti_env, next,
 367                                               &info->mti_u.ksfs);
 368                 statfs_pack(osfs, &info->mti_u.ksfs);
 369         }
 370         RETURN(rc);
 371 }
 372
 373 /**
 374  * Pack SOM attributes into the reply.
 375  * Call under a DLM UPDATE lock.
 376  */
 377 static void mdt_pack_size2body(struct mdt_thread_info *info,
 378                                struct mdt_object *mo)
 379 {
 380         struct mdt_body *b;
 381         struct md_attr *ma = &info->mti_attr;
 382
 383         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 384         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 385
 386         /* Check if Size-on-MDS is supported, if this is a regular file,
 387          * if SOM is enabled on the object and if SOM cache exists and valid.
 388          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 389         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 390             !S_ISREG(ma->ma_attr.la_mode) ||
 391             !mdt_object_is_som_enabled(mo) ||
 392             !(ma->ma_valid & MA_SOM))
 393                 return;
 394
 395         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 396         b->size = ma->ma_som->msd_size;
 397         b->blocks = ma->ma_som->msd_blocks;
 398 }
 399
 400 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 401                         const struct lu_attr *attr, const struct lu_fid *fid)
 402 {
 403         struct md_attr *ma = &info->mti_attr;
 404
 405         LASSERT(ma->ma_valid & MA_INODE);
 406
 407         b->atime      = attr->la_atime;
 408         b->mtime      = attr->la_mtime;
 409         b->ctime      = attr->la_ctime;
 410         b->mode       = attr->la_mode;
 411         b->size       = attr->la_size;
 412         b->blocks     = attr->la_blocks;
 413         b->uid        = attr->la_uid;
 414         b->gid        = attr->la_gid;
 415         b->flags      = attr->la_flags;
 416         b->nlink      = attr->la_nlink;
 417         b->rdev       = attr->la_rdev;
 418
 419         /*XXX should pack the reply body according to lu_valid*/
 420         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 421                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 422                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 423                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 424
 425         if (!S_ISREG(attr->la_mode)) {
 426                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 427         } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
 428                 /* means no objects are allocated on osts. */
 429                 LASSERT(!(ma->ma_valid & MA_LOV));
 430                 /* just ignore blocks occupied by extend attributes on MDS */
 431                 b->blocks = 0;
 432                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 433                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 434         }
 435
 436         if (fid) {
 437                 b->fid1 = *fid;
 438                 b->valid |= OBD_MD_FLID;
 439
 440                 /* FIXME: these should be fixed when new igif ready.*/
 441                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 442                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 443                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 444
 445                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 446                                 PFID(fid), b->nlink, b->mode, b->size);
 447         }
 448
 449         if (info)
 450                 mdt_body_reverse_idmap(info, b);
 451
 452         if (b->valid & OBD_MD_FLSIZE)
 453                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 454                        PFID(fid), (unsigned long long)b->size);
 455 }
 456
 457 static inline int mdt_body_has_lov(const struct lu_attr *la,
 458                                    const struct mdt_body *body)
 459 {
 460         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 461                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 462 }
 463
 464 static int mdt_getattr_internal(struct mdt_thread_info *info,
 465                                 struct mdt_object *o, int ma_need)
 466 {
 467         struct md_object        *next = mdt_object_child(o);
 468         const struct mdt_body   *reqbody = info->mti_body;
 469         struct ptlrpc_request   *req = mdt_info_req(info);
 470         struct md_attr          *ma = &info->mti_attr;
 471         struct lu_attr          *la = &ma->ma_attr;
 472         struct req_capsule      *pill = info->mti_pill;
 473         const struct lu_env     *env = info->mti_env;
 474         struct mdt_body         *repbody;
 475         struct lu_buf           *buffer = &info->mti_buf;
 476         int                     rc;
 477         ENTRY;
 478
 479         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 480                 RETURN(err_serious(-ENOMEM));
 481
 482         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 483
 484         ma->ma_valid = 0;
 485
 486         rc = mdt_object_exists(o);
 487         if (rc < 0) {
 488                 /* This object is located on remote node.*/
 489                 repbody->fid1 = *mdt_object_fid(o);
 490                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 491                 RETURN(0);
 492         }
 493
 494         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 495         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 496
 497         /* If it is dir object and client require MEA, then we got MEA */
 498         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 499             reqbody->valid & OBD_MD_MEA) {
 500                 /* Assumption: MDT_MD size is enough for lmv size. */
 501                 ma->ma_lmv = buffer->lb_buf;
 502                 ma->ma_lmv_size = buffer->lb_len;
 503                 ma->ma_need = MA_LMV | MA_INODE;
 504         } else {
 505                 ma->ma_lmm = buffer->lb_buf;
 506                 ma->ma_lmm_size = buffer->lb_len;
 507                 ma->ma_need = MA_LOV | MA_INODE;
 508         }
 509
 510         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 511             reqbody->valid & OBD_MD_FLDIREA  &&
 512             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 513                 /* get default stripe info for this dir. */
 514                 ma->ma_need |= MA_LOV_DEF;
 515         }
 516         ma->ma_need |= ma_need;
 517         if (ma->ma_need & MA_SOM)
 518                 ma->ma_som = &info->mti_u.som.data;
 519
 520         rc = mo_attr_get(env, next, ma);
 521         if (unlikely(rc)) {
 522                 CERROR("getattr error for "DFID": %d\n",
 523                         PFID(mdt_object_fid(o)), rc);
 524                 RETURN(rc);
 525         }
 526
 527         if (likely(ma->ma_valid & MA_INODE))
 528                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 529         else
 530                 RETURN(-EFAULT);
 531
 532         if (mdt_body_has_lov(la, reqbody)) {
 533                 if (ma->ma_valid & MA_LOV) {
 534                         LASSERT(ma->ma_lmm_size);
 535                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 536                         repbody->eadatasize = ma->ma_lmm_size;
 537                         if (S_ISDIR(la->la_mode))
 538                                 repbody->valid |= OBD_MD_FLDIREA;
 539                         else
 540                                 repbody->valid |= OBD_MD_FLEASIZE;
 541                 }
 542                 if (ma->ma_valid & MA_LMV) {
 543                         LASSERT(S_ISDIR(la->la_mode));
 544                         repbody->eadatasize = ma->ma_lmv_size;
 545                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 546                 }
 547         } else if (S_ISLNK(la->la_mode) &&
 548                    reqbody->valid & OBD_MD_LINKNAME) {
 549                 buffer->lb_buf = ma->ma_lmm;
 550                 buffer->lb_len = reqbody->eadatasize;
 551                 rc = mo_readlink(env, next, buffer);
 552                 if (unlikely(rc <= 0)) {
 553                         CERROR("readlink failed: %d\n", rc);
 554                         rc = -EFAULT;
 555                 } else {
 556                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 557                                  rc -= 2;
 558                         repbody->valid |= OBD_MD_LINKNAME;
 559                         repbody->eadatasize = rc;
 560                         /* NULL terminate */
 561                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 562                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 563                                (char*)ma->ma_lmm, rc);
 564                         rc = 0;
 565                 }
 566         }
 567
 568         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 569                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 570                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 571                 repbody->valid |= OBD_MD_FLMODEASIZE;
 572                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 573                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 574                        repbody->max_cookiesize);
 575         }
 576
 577         if (exp_connect_rmtclient(info->mti_exp) &&
 578             reqbody->valid & OBD_MD_FLRMTPERM) {
 579                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 580
 581                 /* mdt_getattr_lock only */
 582                 rc = mdt_pack_remote_perm(info, o, buf);
 583                 if (rc) {
 584                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 585                         repbody->aclsize = 0;
 586                         RETURN(rc);
 587                 } else {
 588                         repbody->valid |= OBD_MD_FLRMTPERM;
 589                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 590                 }
 591         }
 592 #ifdef CONFIG_FS_POSIX_ACL
 593         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 594                  (reqbody->valid & OBD_MD_FLACL)) {
 595                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 596                 buffer->lb_len = req_capsule_get_size(pill,
 597                                                       &RMF_ACL, RCL_SERVER);
 598                 if (buffer->lb_len > 0) {
 599                         rc = mo_xattr_get(env, next, buffer,
 600                                           XATTR_NAME_ACL_ACCESS);
 601                         if (rc < 0) {
 602                                 if (rc == -ENODATA) {
 603                                         repbody->aclsize = 0;
 604                                         repbody->valid |= OBD_MD_FLACL;
 605                                         rc = 0;
 606                                 } else if (rc == -EOPNOTSUPP) {
 607                                         rc = 0;
 608                                 } else {
 609                                         CERROR("got acl size: %d\n", rc);
 610                                 }
 611                         } else {
 612                                 repbody->aclsize = rc;
 613                                 repbody->valid |= OBD_MD_FLACL;
 614                                 rc = 0;
 615                         }
 616                 }
 617         }
 618 #endif
 619
 620         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 621             info->mti_mdt->mdt_opts.mo_mds_capa &&
 622             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 623                 struct lustre_capa *capa;
 624
 625                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 626                 LASSERT(capa);
 627                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 628                 rc = mo_capa_get(env, next, capa, 0);
 629                 if (rc)
 630                         RETURN(rc);
 631                 repbody->valid |= OBD_MD_FLMDSCAPA;
 632         }
 633         RETURN(rc);
 634 }
 635
 636 static int mdt_renew_capa(struct mdt_thread_info *info)
 637 {
 638         struct mdt_object  *obj = info->mti_object;
 639         struct mdt_body    *body;
 640         struct lustre_capa *capa, *c;
 641         int rc;
 642         ENTRY;
 643
 644         /* if object doesn't exist, or server has disabled capability,
 645          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 646          * flag not set.
 647          */
 648         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 649             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 650                 RETURN(0);
 651
 652         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 653         LASSERT(body != NULL);
 654
 655         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 656         LASSERT(c);
 657
 658         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 659         LASSERT(capa);
 660
 661         *capa = *c;
 662         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 663         if (rc == 0)
 664                 body->valid |= OBD_MD_FLOSSCAPA;
 665         RETURN(rc);
 666 }
 667
 668 static int mdt_getattr(struct mdt_thread_info *info)
 669 {
 670         struct mdt_object       *obj = info->mti_object;
 671         struct req_capsule      *pill = info->mti_pill;
 672         struct mdt_body         *reqbody;
 673         struct mdt_body         *repbody;
 674         mode_t                   mode;
 675         int                      md_size;
 676         int rc;
 677         ENTRY;
 678
 679         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 680         LASSERT(reqbody);
 681
 682         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 683                 rc = req_capsule_server_pack(pill);
 684                 if (unlikely(rc))
 685                         RETURN(err_serious(rc));
 686                 rc = mdt_renew_capa(info);
 687                 GOTO(out_shrink, rc);
 688         }
 689
 690         LASSERT(obj != NULL);
 691         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 692
 693         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 694         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 695             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 696                 md_size = reqbody->eadatasize;
 697         else
 698                 md_size = info->mti_mdt->mdt_max_mdsize;
 699
 700         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 701
 702         rc = req_capsule_server_pack(pill);
 703         if (unlikely(rc != 0))
 704                 RETURN(err_serious(rc));
 705
 706         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 707         LASSERT(repbody != NULL);
 708         repbody->eadatasize = 0;
 709         repbody->aclsize = 0;
 710
 711         if (reqbody->valid & OBD_MD_FLRMTPERM)
 712                 rc = mdt_init_ucred(info, reqbody);
 713         else
 714                 rc = mdt_check_ucred(info);
 715         if (unlikely(rc))
 716                 GOTO(out_shrink, rc);
 717
 718         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 719         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 720
 721         /*
 722          * Don't check capability at all, because rename might getattr for
 723          * remote obj, and at that time no capability is available.
 724          */
 725         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 726         rc = mdt_getattr_internal(info, obj, 0);
 727         if (reqbody->valid & OBD_MD_FLRMTPERM)
 728                 mdt_exit_ucred(info);
 729         EXIT;
 730 out_shrink:
 731         mdt_shrink_reply(info);
 732         return rc;
 733 }
 734
 735 static int mdt_is_subdir(struct mdt_thread_info *info)
 736 {
 737         struct mdt_object     *o = info->mti_object;
 738         struct req_capsule    *pill = info->mti_pill;
 739         const struct mdt_body *body = info->mti_body;
 740         struct mdt_body       *repbody;
 741         int                    rc;
 742         ENTRY;
 743
 744         LASSERT(o != NULL);
 745
 746         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 747
 748         /*
 749          * We save last checked parent fid to @repbody->fid1 for remote
 750          * directory case.
 751          */
 752         LASSERT(fid_is_sane(&body->fid2));
 753         LASSERT(mdt_object_exists(o) > 0);
 754         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 755                            &body->fid2, &repbody->fid1);
 756         if (rc == 0 || rc == -EREMOTE)
 757                 repbody->valid |= OBD_MD_FLID;
 758
 759         RETURN(rc);
 760 }
 761
 762 static int mdt_raw_lookup(struct mdt_thread_info *info,
 763                           struct mdt_object *parent,
 764                           const struct lu_name *lname,
 765                           struct ldlm_reply *ldlm_rep)
 766 {
 767         struct md_object *next = mdt_object_child(info->mti_object);
 768         const struct mdt_body *reqbody = info->mti_body;
 769         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 770         struct mdt_body *repbody;
 771         int rc;
 772         ENTRY;
 773
 774         if (reqbody->valid != OBD_MD_FLID)
 775                 RETURN(0);
 776
 777         LASSERT(!info->mti_cross_ref);
 778
 779         /* Only got the fid of this obj by name */
 780         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 781                         &info->mti_spec);
 782 #if 0
 783         /* XXX is raw_lookup possible as intent operation? */
 784         if (rc != 0) {
 785                 if (rc == -ENOENT)
 786                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 787                 RETURN(rc);
 788         } else
 789                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 790
 791         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 792 #endif
 793         if (rc == 0) {
 794                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 795                 repbody->fid1 = *child_fid;
 796                 repbody->valid = OBD_MD_FLID;
 797         }
 798         RETURN(1);
 799 }
 800
 801 /*
 802  * UPDATE lock should be taken against parent, and be release before exit;
 803  * child_bits lock should be taken against child, and be returned back:
 804  *            (1)normal request should release the child lock;
 805  *            (2)intent request will grant the lock to client.
 806  */
 807 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 808                                  struct mdt_lock_handle *lhc,
 809                                  __u64 child_bits,
 810                                  struct ldlm_reply *ldlm_rep)
 811 {
 812         struct ptlrpc_request  *req       = mdt_info_req(info);
 813         struct mdt_body        *reqbody   = NULL;
 814         struct mdt_object      *parent    = info->mti_object;
 815         struct mdt_object      *child;
 816         struct md_object       *next      = mdt_object_child(parent);
 817         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 818         struct lu_name         *lname     = NULL;
 819         const char             *name      = NULL;
 820         int                     namelen   = 0;
 821         struct mdt_lock_handle *lhp       = NULL;
 822         struct ldlm_lock       *lock;
 823         struct ldlm_res_id     *res_id;
 824         int                     is_resent;
 825         int                     ma_need = 0;
 826         int                     rc;
 827
 828         ENTRY;
 829
 830         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 831         LASSERT(ergo(is_resent,
 832                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 833
 834         LASSERT(parent != NULL);
 835         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 836         if (name == NULL)
 837                 RETURN(err_serious(-EFAULT));
 838
 839         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 840                                        RCL_CLIENT) - 1;
 841         if (!info->mti_cross_ref) {
 842                 /*
 843                  * XXX: Check for "namelen == 0" is for getattr by fid
 844                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 845                  * that is the name must contain at least one character and
 846                  * the terminating '\0'
 847                  */
 848                 if (namelen == 0) {
 849                         reqbody = req_capsule_client_get(info->mti_pill,
 850                                                          &RMF_MDT_BODY);
 851                         if (unlikely(reqbody == NULL))
 852                                 RETURN(err_serious(-EFAULT));
 853
 854                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
 855                                 RETURN(err_serious(-EINVAL));
 856
 857                         name = NULL;
 858                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 859                                "ldlm_rep = %p\n",
 860                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 861                                ldlm_rep);
 862                 } else {
 863                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 864                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 865                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 866                                name, ldlm_rep);
 867                 }
 868         }
 869         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 870
 871         rc = mdt_object_exists(parent);
 872         if (unlikely(rc == 0)) {
 873                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 874                                 &parent->mot_obj.mo_lu,
 875                                 "Parent doesn't exist!\n");
 876                 RETURN(-ESTALE);
 877         } else if (!info->mti_cross_ref) {
 878                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 879                          PFID(mdt_object_fid(parent)));
 880         }
 881         if (lname) {
 882                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 883                 if (rc != 0) {
 884                         if (rc > 0)
 885                                 rc = 0;
 886                         RETURN(rc);
 887                 }
 888         }
 889
 890         if (info->mti_cross_ref) {
 891                 /* Only getattr on the child. Parent is on another node. */
 892                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 893                 child = parent;
 894                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 895                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 896
 897                 if (is_resent) {
 898                         /* Do not take lock for resent case. */
 899                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 900                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 901                                  lhc->mlh_reg_lh.cookie);
 902                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 903                                                 &lock->l_resource->lr_name));
 904                         LDLM_LOCK_PUT(lock);
 905                         rc = 0;
 906                 } else {
 907                         mdt_lock_handle_init(lhc);
 908                         mdt_lock_reg_init(lhc, LCK_PR);
 909
 910                         /*
 911                          * Object's name is on another MDS, no lookup lock is
 912                          * needed here but update is.
 913                          */
 914                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 915                         child_bits |= MDS_INODELOCK_UPDATE;
 916
 917                         rc = mdt_object_lock(info, child, lhc, child_bits,
 918                                              MDT_LOCAL_LOCK);
 919                 }
 920                 if (rc == 0) {
 921                         /* Finally, we can get attr for child. */
 922                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 923                                          BYPASS_CAPA);
 924                         rc = mdt_getattr_internal(info, child, 0);
 925                         if (unlikely(rc != 0))
 926                                 mdt_object_unlock(info, child, lhc, 1);
 927                 }
 928                 RETURN(rc);
 929         }
 930
 931         if (lname) {
 932                 /* step 1: lock parent */
 933                 lhp = &info->mti_lh[MDT_LH_PARENT];
 934                 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 935                 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 936                                      MDT_LOCAL_LOCK);
 937                 if (unlikely(rc != 0))
 938                         RETURN(rc);
 939
 940                 /* step 2: lookup child's fid by name */
 941                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 942                                 &info->mti_spec);
 943
 944                 if (rc != 0) {
 945                         if (rc == -ENOENT)
 946                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 947                         GOTO(out_parent, rc);
 948                 } else
 949                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 950         } else {
 951                 *child_fid = reqbody->fid2;
 952                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 953         }
 954
 955         /*
 956          *step 3: find the child object by fid & lock it.
 957          *        regardless if it is local or remote.
 958          */
 959         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 960
 961         if (unlikely(IS_ERR(child)))
 962                 GOTO(out_parent, rc = PTR_ERR(child));
 963         if (is_resent) {
 964                 /* Do not take lock for resent case. */
 965                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 966                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 967                          lhc->mlh_reg_lh.cookie);
 968
 969                 res_id = &lock->l_resource->lr_name;
 970                 if (!fid_res_name_eq(mdt_object_fid(child),
 971                                     &lock->l_resource->lr_name)) {
 972                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 973                                                  &lock->l_resource->lr_name),
 974                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 975                                  (unsigned long)res_id->name[0],
 976                                  (unsigned long)res_id->name[1],
 977                                  (unsigned long)res_id->name[2],
 978                                  PFID(mdt_object_fid(parent)));
 979                           CWARN("Although resent, but still not get child lock"
 980                                 "parent:"DFID" child:"DFID"\n",
 981                                 PFID(mdt_object_fid(parent)),
 982                                 PFID(mdt_object_fid(child)));
 983                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 984                           LDLM_LOCK_PUT(lock);
 985                           GOTO(relock, 0);
 986                 }
 987                 LDLM_LOCK_PUT(lock);
 988                 rc = 0;
 989         } else {
 990 relock:
 991                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
 992                 mdt_lock_handle_init(lhc);
 993                 mdt_lock_reg_init(lhc, LCK_PR);
 994
 995                 if (mdt_object_exists(child) == 0) {
 996                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 997                                         &child->mot_obj.mo_lu,
 998                                         "Object doesn't exist!\n");
 999                         GOTO(out_child, rc = -ENOENT);
1000                 }
1001
1002                 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1003                         struct md_attr *ma = &info->mti_attr;
1004
1005                         ma->ma_valid = 0;
1006                         ma->ma_need = MA_INODE;
1007                         rc = mo_attr_get(info->mti_env,
1008                                          mdt_object_child(child), ma);
1009                         if (unlikely(rc != 0))
1010                                 GOTO(out_child, rc);
1011
1012                         /* If the file has not been changed for some time, we
1013                          * return not only a LOOKUP lock, but also an UPDATE
1014                          * lock and this might save us RPC on later STAT. For
1015                          * directories, it also let negative dentry starts
1016                          * working for this dir. */
1017                         if (ma->ma_valid & MA_INODE &&
1018                             ma->ma_attr.la_valid & LA_CTIME &&
1019                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1020                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1021                                 child_bits |= MDS_INODELOCK_UPDATE;
1022                 }
1023
1024                 rc = mdt_object_lock(info, child, lhc, child_bits,
1025                                      MDT_CROSS_LOCK);
1026
1027                 if (unlikely(rc != 0))
1028                         GOTO(out_child, rc);
1029         }
1030
1031         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1032         /* Get MA_SOM attributes if update lock is given. */
1033         if (lock &&
1034             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1035             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1036                 ma_need = MA_SOM;
1037
1038         /* finally, we can get attr for child. */
1039         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1040         rc = mdt_getattr_internal(info, child, ma_need);
1041         if (unlikely(rc != 0)) {
1042                 mdt_object_unlock(info, child, lhc, 1);
1043         } else if (lock) {
1044                 /* Debugging code. */
1045                 res_id = &lock->l_resource->lr_name;
1046                 LDLM_DEBUG(lock, "Returning lock to client");
1047                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1048                                          &lock->l_resource->lr_name),
1049                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1050                          (unsigned long)res_id->name[0],
1051                          (unsigned long)res_id->name[1],
1052                          (unsigned long)res_id->name[2],
1053                          PFID(mdt_object_fid(child)));
1054                 mdt_pack_size2body(info, child);
1055         }
1056         if (lock)
1057                 LDLM_LOCK_PUT(lock);
1058
1059         EXIT;
1060 out_child:
1061         mdt_object_put(info->mti_env, child);
1062 out_parent:
1063         if (lhp)
1064                 mdt_object_unlock(info, parent, lhp, 1);
1065         return rc;
1066 }
1067
1068 /* normal handler: should release the child lock */
1069 static int mdt_getattr_name(struct mdt_thread_info *info)
1070 {
1071         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1072         struct mdt_body        *reqbody;
1073         struct mdt_body        *repbody;
1074         int rc;
1075         ENTRY;
1076
1077         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1078         LASSERT(reqbody != NULL);
1079         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1080         LASSERT(repbody != NULL);
1081
1082         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1083         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1084         repbody->eadatasize = 0;
1085         repbody->aclsize = 0;
1086
1087         rc = mdt_init_ucred(info, reqbody);
1088         if (unlikely(rc))
1089                 GOTO(out_shrink, rc);
1090
1091         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1092         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1093                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1094                 lhc->mlh_reg_lh.cookie = 0;
1095         }
1096         mdt_exit_ucred(info);
1097         EXIT;
1098 out_shrink:
1099         mdt_shrink_reply(info);
1100         return rc;
1101 }
1102
1103 static const struct lu_device_operations mdt_lu_ops;
1104
1105 static int lu_device_is_mdt(struct lu_device *d)
1106 {
1107         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1108 }
1109
1110 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1111                          void *karg, void *uarg);
1112
1113 static int mdt_set_info(struct mdt_thread_info *info)
1114 {
1115         struct ptlrpc_request *req = mdt_info_req(info);
1116         char *key;
1117         void *val;
1118         int keylen, vallen, rc = 0;
1119         ENTRY;
1120
1121         rc = req_capsule_server_pack(info->mti_pill);
1122         if (rc)
1123                 RETURN(rc);
1124
1125         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1126         if (key == NULL) {
1127                 DEBUG_REQ(D_HA, req, "no set_info key");
1128                 RETURN(-EFAULT);
1129         }
1130
1131         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1132                                       RCL_CLIENT);
1133
1134         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1135         if (val == NULL) {
1136                 DEBUG_REQ(D_HA, req, "no set_info val");
1137                 RETURN(-EFAULT);
1138         }
1139
1140         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1141                                       RCL_CLIENT);
1142
1143         /* Swab any part of val you need to here */
1144         if (KEY_IS(KEY_READ_ONLY)) {
1145                 req->rq_status = 0;
1146                 lustre_msg_set_status(req->rq_repmsg, 0);
1147
1148                 cfs_spin_lock(&req->rq_export->exp_lock);
1149                 if (*(__u32 *)val)
1150                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1151                 else
1152                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1153                 cfs_spin_unlock(&req->rq_export->exp_lock);
1154
1155         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1156                 struct changelog_setinfo *cs =
1157                         (struct changelog_setinfo *)val;
1158                 if (vallen != sizeof(*cs)) {
1159                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1160                         RETURN(-EINVAL);
1161                 }
1162                 if (ptlrpc_req_need_swab(req)) {
1163                         __swab64s(&cs->cs_recno);
1164                         __swab32s(&cs->cs_id);
1165                 }
1166
1167                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1168                                    vallen, val, NULL);
1169                 lustre_msg_set_status(req->rq_repmsg, rc);
1170
1171         } else {
1172                 RETURN(-EINVAL);
1173         }
1174         RETURN(0);
1175 }
1176
1177 static int mdt_connect(struct mdt_thread_info *info)
1178 {
1179         int rc;
1180         struct ptlrpc_request *req;
1181
1182         req = mdt_info_req(info);
1183         rc = target_handle_connect(req);
1184         if (rc == 0) {
1185                 LASSERT(req->rq_export != NULL);
1186                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1187                 rc = mdt_init_sec_level(info);
1188                 if (rc == 0)
1189                         rc = mdt_init_idmap(info);
1190                 if (rc != 0)
1191                         obd_disconnect(class_export_get(req->rq_export));
1192         } else {
1193                 rc = err_serious(rc);
1194         }
1195         return rc;
1196 }
1197
1198 static int mdt_disconnect(struct mdt_thread_info *info)
1199 {
1200         int rc;
1201         ENTRY;
1202
1203         rc = target_handle_disconnect(mdt_info_req(info));
1204         if (rc)
1205                 rc = err_serious(rc);
1206         RETURN(rc);
1207 }
1208
1209 static int mdt_sendpage(struct mdt_thread_info *info,
1210                         struct lu_rdpg *rdpg)
1211 {
1212         struct ptlrpc_request   *req = mdt_info_req(info);
1213         struct obd_export       *exp = req->rq_export;
1214         struct ptlrpc_bulk_desc *desc;
1215         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1216         int                      tmpcount;
1217         int                      tmpsize;
1218         int                      timeout;
1219         int                      i;
1220         int                      rc;
1221         ENTRY;
1222
1223         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1224                                     MDS_BULK_PORTAL);
1225         if (desc == NULL)
1226                 RETURN(-ENOMEM);
1227
1228         for (i = 0, tmpcount = rdpg->rp_count;
1229                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1230                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1231                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1232         }
1233
1234         LASSERT(desc->bd_nob == rdpg->rp_count);
1235         rc = sptlrpc_svc_wrap_bulk(req, desc);
1236         if (rc)
1237                 GOTO(free_desc, rc);
1238
1239         rc = ptlrpc_start_bulk_transfer(desc);
1240         if (rc)
1241                 GOTO(free_desc, rc);
1242
1243         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1244                 GOTO(abort_bulk, rc = 0);
1245
1246         do {
1247                 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1248                 if (timeout < 0)
1249                         CERROR("Req deadline already passed %lu (now: %lu)\n",
1250                                req->rq_deadline, cfs_time_current_sec());
1251                 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1252                                             cfs_time_seconds(1), NULL, NULL);
1253                 rc = l_wait_event(desc->bd_waitq,
1254                                   !ptlrpc_server_bulk_active(desc) ||
1255                                   exp->exp_failed ||
1256                                   exp->exp_abort_active_req, lwi);
1257                 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1258         } while ((rc == -ETIMEDOUT) &&
1259                  (req->rq_deadline > cfs_time_current_sec()));
1260
1261         if (rc == 0) {
1262                 if (desc->bd_success &&
1263                     desc->bd_nob_transferred == rdpg->rp_count)
1264                         GOTO(free_desc, rc);
1265
1266                 rc = -ETIMEDOUT;
1267                 if (exp->exp_abort_active_req || exp->exp_failed)
1268                         GOTO(abort_bulk, rc);
1269         }
1270
1271         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1272                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1273                   desc->bd_nob_transferred, rdpg->rp_count,
1274                   exp->exp_client_uuid.uuid,
1275                   exp->exp_connection->c_remote_uuid.uuid);
1276
1277         class_fail_export(exp);
1278
1279         EXIT;
1280 abort_bulk:
1281         ptlrpc_abort_bulk(desc);
1282 free_desc:
1283         ptlrpc_free_bulk(desc);
1284         return rc;
1285 }
1286
1287 #ifdef HAVE_SPLIT_SUPPORT
1288 /*
1289  * Retrieve dir entry from the page and insert it to the slave object, actually,
1290  * this should be in osd layer, but since it will not in the final product, so
1291  * just do it here and do not define more moo api anymore for this.
1292  */
1293 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1294                               int size)
1295 {
1296         struct mdt_object *object = info->mti_object;
1297         struct lu_fid *lf = &info->mti_tmp_fid2;
1298         struct md_attr *ma = &info->mti_attr;
1299         struct lu_dirpage *dp;
1300         struct lu_dirent *ent;
1301         int rc = 0, offset = 0;
1302         ENTRY;
1303
1304         /* Make sure we have at least one entry. */
1305         if (size == 0)
1306                 RETURN(-EINVAL);
1307
1308         /*
1309          * Disable trans for this name insert, since it will include many trans
1310          * for this.
1311          */
1312         info->mti_no_need_trans = 1;
1313         /*
1314          * When write_dir_page, no need update parent's ctime,
1315          * and no permission check for name_insert.
1316          */
1317         ma->ma_attr.la_ctime = 0;
1318         ma->ma_attr.la_valid = LA_MODE;
1319         ma->ma_valid = MA_INODE;
1320
1321         cfs_kmap(page);
1322         dp = page_address(page);
1323         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1324
1325         for (ent = lu_dirent_start(dp); ent != NULL;
1326              ent = lu_dirent_next(ent)) {
1327                 struct lu_name *lname;
1328                 char *name;
1329
1330                 if (le16_to_cpu(ent->lde_namelen) == 0)
1331                         continue;
1332
1333                 fid_le_to_cpu(lf, &ent->lde_fid);
1334                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1335                         ma->ma_attr.la_mode = S_IFDIR;
1336                 else
1337                         ma->ma_attr.la_mode = 0;
1338                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1339                 if (name == NULL)
1340                         GOTO(out, rc = -ENOMEM);
1341
1342                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1343                 lname = mdt_name(info->mti_env, name,
1344                                  le16_to_cpu(ent->lde_namelen));
1345                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1346                 rc = mdo_name_insert(info->mti_env,
1347                                      md_object_next(&object->mot_obj),
1348                                      lname, lf, ma);
1349                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1350                 if (rc) {
1351                         CERROR("Can't insert %*.*s, rc %d\n",
1352                                le16_to_cpu(ent->lde_namelen),
1353                                le16_to_cpu(ent->lde_namelen),
1354                                ent->lde_name, rc);
1355                         GOTO(out, rc);
1356                 }
1357
1358                 offset += lu_dirent_size(ent);
1359                 if (offset >= size)
1360                         break;
1361         }
1362         EXIT;
1363 out:
1364         cfs_kunmap(page);
1365         return rc;
1366 }
1367
1368 static int mdt_bulk_timeout(void *data)
1369 {
1370         ENTRY;
1371
1372         CERROR("mdt bulk transfer timeout \n");
1373
1374         RETURN(1);
1375 }
1376
1377 static int mdt_writepage(struct mdt_thread_info *info)
1378 {
1379         struct ptlrpc_request   *req = mdt_info_req(info);
1380         struct mdt_body         *reqbody;
1381         struct l_wait_info      *lwi;
1382         struct ptlrpc_bulk_desc *desc;
1383         struct page             *page;
1384         int                rc;
1385         ENTRY;
1386
1387
1388         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1389         if (reqbody == NULL)
1390                 RETURN(err_serious(-EFAULT));
1391
1392         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1393         if (desc == NULL)
1394                 RETURN(err_serious(-ENOMEM));
1395
1396         /* allocate the page for the desc */
1397         page = cfs_alloc_page(CFS_ALLOC_STD);
1398         if (page == NULL)
1399                 GOTO(desc_cleanup, rc = -ENOMEM);
1400
1401         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1402                (int)reqbody->size, (int)reqbody->nlink);
1403
1404         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1405                               (int)reqbody->nlink);
1406
1407         rc = sptlrpc_svc_prep_bulk(req, desc);
1408         if (rc != 0)
1409                 GOTO(cleanup_page, rc);
1410         /*
1411          * Check if client was evicted while we were doing i/o before touching
1412          * network.
1413          */
1414         OBD_ALLOC_PTR(lwi);
1415         if (!lwi)
1416                 GOTO(cleanup_page, rc = -ENOMEM);
1417
1418         if (desc->bd_export->exp_failed)
1419                 rc = -ENOTCONN;
1420         else
1421                 rc = ptlrpc_start_bulk_transfer (desc);
1422         if (rc == 0) {
1423                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1424                                             mdt_bulk_timeout, desc);
1425                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1426                                   desc->bd_export->exp_failed, lwi);
1427                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1428                 if (rc == -ETIMEDOUT) {
1429                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1430                         ptlrpc_abort_bulk(desc);
1431                 } else if (desc->bd_export->exp_failed) {
1432                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1433                         rc = -ENOTCONN;
1434                         ptlrpc_abort_bulk(desc);
1435                 } else if (!desc->bd_success ||
1436                            desc->bd_nob_transferred != desc->bd_nob) {
1437                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1438                                   desc->bd_success ?
1439                                   "truncated" : "network error on",
1440                                   desc->bd_nob_transferred, desc->bd_nob);
1441                         /* XXX should this be a different errno? */
1442                         rc = -ETIMEDOUT;
1443                 }
1444         } else {
1445                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1446         }
1447         if (rc)
1448                 GOTO(cleanup_lwi, rc);
1449         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1450
1451 cleanup_lwi:
1452         OBD_FREE_PTR(lwi);
1453 cleanup_page:
1454         cfs_free_page(page);
1455 desc_cleanup:
1456         ptlrpc_free_bulk(desc);
1457         RETURN(rc);
1458 }
1459 #endif
1460
1461 static int mdt_readpage(struct mdt_thread_info *info)
1462 {
1463         struct mdt_object *object = info->mti_object;
1464         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1465         struct mdt_body   *reqbody;
1466         struct mdt_body   *repbody;
1467         int                rc;
1468         int                i;
1469         ENTRY;
1470
1471         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1472                 RETURN(err_serious(-ENOMEM));
1473
1474         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1475         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1476         if (reqbody == NULL || repbody == NULL)
1477                 RETURN(err_serious(-EFAULT));
1478
1479         /*
1480          * prepare @rdpg before calling lower layers and transfer itself. Here
1481          * reqbody->size contains offset of where to start to read and
1482          * reqbody->nlink contains number bytes to read.
1483          */
1484         rdpg->rp_hash = reqbody->size;
1485         if (rdpg->rp_hash != reqbody->size) {
1486                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1487                        rdpg->rp_hash, reqbody->size);
1488                 RETURN(-EFAULT);
1489         }
1490
1491         rdpg->rp_attrs = reqbody->mode;
1492         rdpg->rp_count  = reqbody->nlink;
1493         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1494         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1495         if (rdpg->rp_pages == NULL)
1496                 RETURN(-ENOMEM);
1497
1498         for (i = 0; i < rdpg->rp_npages; ++i) {
1499                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1500                 if (rdpg->rp_pages[i] == NULL)
1501                         GOTO(free_rdpg, rc = -ENOMEM);
1502         }
1503
1504         /* call lower layers to fill allocated pages with directory data */
1505         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1506         if (rc)
1507                 GOTO(free_rdpg, rc);
1508
1509         /* send pages to client */
1510         rc = mdt_sendpage(info, rdpg);
1511
1512         EXIT;
1513 free_rdpg:
1514
1515         for (i = 0; i < rdpg->rp_npages; i++)
1516                 if (rdpg->rp_pages[i] != NULL)
1517                         cfs_free_page(rdpg->rp_pages[i]);
1518         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1519
1520         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1521                 RETURN(0);
1522
1523         return rc;
1524 }
1525
1526 static int mdt_reint_internal(struct mdt_thread_info *info,
1527                               struct mdt_lock_handle *lhc,
1528                               __u32 op)
1529 {
1530         struct req_capsule      *pill = info->mti_pill;
1531         struct mdt_device       *mdt = info->mti_mdt;
1532         struct md_quota         *mq = md_quota(info->mti_env);
1533         struct mdt_body         *repbody;
1534         int                      rc = 0;
1535         ENTRY;
1536
1537         /* pack reply */
1538         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1539                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1540                                      mdt->mdt_max_mdsize);
1541         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1542                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1543                                      mdt->mdt_max_cookiesize);
1544
1545         rc = req_capsule_server_pack(pill);
1546         if (rc != 0) {
1547                 CERROR("Can't pack response, rc %d\n", rc);
1548                 RETURN(err_serious(rc));
1549         }
1550
1551         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1552                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1553                 LASSERT(repbody);
1554                 repbody->eadatasize = 0;
1555                 repbody->aclsize = 0;
1556         }
1557
1558         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1559                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1560
1561         rc = mdt_reint_unpack(info, op);
1562         if (rc != 0) {
1563                 CERROR("Can't unpack reint, rc %d\n", rc);
1564                 GOTO(out_shrink, rc = err_serious(rc));
1565         }
1566
1567         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1568
1569         /* for replay no cookkie / lmm need, because client have this already */
1570         if (info->mti_spec.no_create == 1)  {
1571                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1572                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1573
1574                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1575                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1576                                              0);
1577         }
1578
1579         rc = mdt_init_ucred_reint(info);
1580         if (rc)
1581                 GOTO(out_shrink, rc);
1582
1583         rc = mdt_fix_attr_ucred(info, op);
1584         if (rc != 0)
1585                 GOTO(out_ucred, rc = err_serious(rc));
1586
1587         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1588                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1589                 GOTO(out_ucred, rc);
1590         }
1591         mq->mq_exp = info->mti_exp;
1592         rc = mdt_reint_rec(info, lhc);
1593         EXIT;
1594 out_ucred:
1595         mdt_exit_ucred(info);
1596 out_shrink:
1597         mdt_shrink_reply(info);
1598         return rc;
1599 }
1600
1601 static long mdt_reint_opcode(struct mdt_thread_info *info,
1602                              const struct req_format **fmt)
1603 {
1604         struct mdt_rec_reint *rec;
1605         long opc;
1606
1607         opc = err_serious(-EFAULT);
1608         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1609         if (rec != NULL) {
1610                 opc = rec->rr_opcode;
1611                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1612                 if (opc < REINT_MAX && fmt[opc] != NULL)
1613                         req_capsule_extend(info->mti_pill, fmt[opc]);
1614                 else {
1615                         CERROR("Unsupported opc: %ld\n", opc);
1616                         opc = err_serious(opc);
1617                 }
1618         }
1619         return opc;
1620 }
1621
1622 static int mdt_reint(struct mdt_thread_info *info)
1623 {
1624         long opc;
1625         int  rc;
1626
1627         static const struct req_format *reint_fmts[REINT_MAX] = {
1628                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1629                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1630                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1631                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1632                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1633                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1634                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1635         };
1636
1637         ENTRY;
1638
1639         opc = mdt_reint_opcode(info, reint_fmts);
1640         if (opc >= 0) {
1641                 /*
1642                  * No lock possible here from client to pass it to reint code
1643                  * path.
1644                  */
1645                 rc = mdt_reint_internal(info, NULL, opc);
1646         } else {
1647                 rc = opc;
1648         }
1649
1650         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1651         RETURN(rc);
1652 }
1653
1654 /* this should sync the whole device */
1655 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1656 {
1657         struct dt_device *dt = mdt->mdt_bottom;
1658         int rc;
1659         ENTRY;
1660
1661         rc = dt->dd_ops->dt_sync(env, dt);
1662         RETURN(rc);
1663 }
1664
1665 /* this should sync this object */
1666 static int mdt_object_sync(struct mdt_thread_info *info)
1667 {
1668         struct md_object *next;
1669         int rc;
1670         ENTRY;
1671
1672         if (!mdt_object_exists(info->mti_object)) {
1673                 CWARN("Non existing object  "DFID"!\n",
1674                       PFID(mdt_object_fid(info->mti_object)));
1675                 RETURN(-ESTALE);
1676         }
1677         next = mdt_object_child(info->mti_object);
1678         rc = mo_object_sync(info->mti_env, next);
1679
1680         RETURN(rc);
1681 }
1682
1683 static int mdt_sync(struct mdt_thread_info *info)
1684 {
1685         struct req_capsule *pill = info->mti_pill;
1686         struct mdt_body *body;
1687         int rc;
1688         ENTRY;
1689
1690         /* The fid may be zero, so we req_capsule_set manually */
1691         req_capsule_set(pill, &RQF_MDS_SYNC);
1692
1693         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1694         if (body == NULL)
1695                 RETURN(err_serious(-EINVAL));
1696
1697         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1698                 RETURN(err_serious(-ENOMEM));
1699
1700         if (fid_seq(&body->fid1) == 0) {
1701                 /* sync the whole device */
1702                 rc = req_capsule_server_pack(pill);
1703                 if (rc == 0)
1704                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1705                 else
1706                         rc = err_serious(rc);
1707         } else {
1708                 /* sync an object */
1709                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1710                 if (rc == 0) {
1711                         rc = mdt_object_sync(info);
1712                         if (rc == 0) {
1713                                 struct md_object *next;
1714                                 const struct lu_fid *fid;
1715                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1716
1717                                 next = mdt_object_child(info->mti_object);
1718                                 info->mti_attr.ma_need = MA_INODE;
1719                                 info->mti_attr.ma_valid = 0;
1720                                 rc = mo_attr_get(info->mti_env, next,
1721                                                  &info->mti_attr);
1722                                 if (rc == 0) {
1723                                         body = req_capsule_server_get(pill,
1724                                                                 &RMF_MDT_BODY);
1725                                         fid = mdt_object_fid(info->mti_object);
1726                                         mdt_pack_attr2body(info, body, la, fid);
1727                                 }
1728                         }
1729                 } else
1730                         rc = err_serious(rc);
1731         }
1732         RETURN(rc);
1733 }
1734
1735 #ifdef HAVE_QUOTA_SUPPORT
1736 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1737 {
1738         struct obd_quotactl *oqctl;
1739         struct req_capsule *pill = info->mti_pill;
1740         struct obd_export *exp = info->mti_exp;
1741         struct md_quota *mq = md_quota(info->mti_env);
1742         struct md_device *next = info->mti_mdt->mdt_child;
1743         int rc;
1744         ENTRY;
1745
1746         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1747         if (oqctl == NULL)
1748                 RETURN(-EPROTO);
1749
1750         /* remote client has no permission for quotacheck */
1751         if (unlikely(exp_connect_rmtclient(exp)))
1752                 RETURN(-EPERM);
1753
1754         rc = req_capsule_server_pack(pill);
1755         if (rc)
1756                 RETURN(rc);
1757
1758         mq->mq_exp = exp;
1759         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1760                                                oqctl->qc_type);
1761         RETURN(rc);
1762 }
1763
1764 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1765 {
1766         struct obd_quotactl *oqctl, *repoqc;
1767         struct req_capsule *pill = info->mti_pill;
1768         struct obd_export *exp = info->mti_exp;
1769         struct md_quota *mq = md_quota(info->mti_env);
1770         struct md_device *next = info->mti_mdt->mdt_child;
1771         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1772         int id, rc;
1773         ENTRY;
1774
1775         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1776         if (oqctl == NULL)
1777                 RETURN(-EPROTO);
1778
1779         id = oqctl->qc_id;
1780         if (exp_connect_rmtclient(exp)) {
1781                 struct ptlrpc_request *req = mdt_info_req(info);
1782                 struct mdt_export_data *med = mdt_req2med(req);
1783                 struct lustre_idmap_table *idmap = med->med_idmap;
1784
1785                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1786                              oqctl->qc_cmd != Q_GETINFO))
1787                         RETURN(-EPERM);
1788
1789
1790                 if (oqctl->qc_type == USRQUOTA)
1791                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1792                                                      oqctl->qc_id);
1793                 else if (oqctl->qc_type == GRPQUOTA)
1794                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1795                                                      oqctl->qc_id);
1796                 else
1797                         RETURN(-EINVAL);
1798
1799                 if (id == CFS_IDMAP_NOTFOUND) {
1800                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1801                                oqctl->qc_id);
1802                         RETURN(-EACCES);
1803                 }
1804         }
1805
1806         rc = req_capsule_server_pack(pill);
1807         if (rc)
1808                 RETURN(rc);
1809
1810         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1811         LASSERT(repoqc != NULL);
1812
1813         mq->mq_exp = exp;
1814         switch (oqctl->qc_cmd) {
1815         case Q_QUOTAON:
1816                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1817                 break;
1818         case Q_QUOTAOFF:
1819                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1820                 break;
1821         case Q_SETINFO:
1822                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1823                                       &oqctl->qc_dqinfo);
1824                 break;
1825         case Q_GETINFO:
1826                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1827                                       &oqctl->qc_dqinfo);
1828                 break;
1829         case Q_SETQUOTA:
1830                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1831                                        &oqctl->qc_dqblk);
1832                 break;
1833         case Q_GETQUOTA:
1834                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1835                                        &oqctl->qc_dqblk);
1836                 break;
1837         case Q_GETOINFO:
1838                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1839                                        &oqctl->qc_dqinfo);
1840                 break;
1841         case Q_GETOQUOTA:
1842                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1843                                         &oqctl->qc_dqblk);
1844                 break;
1845         case LUSTRE_Q_INVALIDATE:
1846                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1847                 break;
1848         case LUSTRE_Q_FINVALIDATE:
1849                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1850                 break;
1851         default:
1852                 CERROR("unsupported mdt_quotactl command: %d\n",
1853                        oqctl->qc_cmd);
1854                 RETURN(-EFAULT);
1855         }
1856
1857         *repoqc = *oqctl;
1858         RETURN(rc);
1859 }
1860 #endif
1861
1862
1863 /*
1864  * OBD PING and other handlers.
1865  */
1866 static int mdt_obd_ping(struct mdt_thread_info *info)
1867 {
1868         int rc;
1869         ENTRY;
1870
1871         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1872
1873         rc = target_handle_ping(mdt_info_req(info));
1874         if (rc < 0)
1875                 rc = err_serious(rc);
1876         RETURN(rc);
1877 }
1878
1879 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1880 {
1881         return err_serious(-EOPNOTSUPP);
1882 }
1883
1884 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1885 {
1886         return err_serious(-EOPNOTSUPP);
1887 }
1888
1889
1890 /*
1891  * LLOG handlers.
1892  */
1893
1894 /** clone llog ctxt from child (mdd)
1895  * This allows remote llog (replicator) access.
1896  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1897  * context was originally set up, or we can handle them directly.
1898  * I choose the latter, but that means I need any llog
1899  * contexts set up by child to be accessable by the mdt.  So we clone the
1900  * context into our context list here.
1901  */
1902 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1903                                int idx)
1904 {
1905         struct md_device  *next = mdt->mdt_child;
1906         struct llog_ctxt *ctxt;
1907         int rc;
1908
1909         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1910                 return 0;
1911
1912         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1913         if (rc || ctxt == NULL) {
1914                 CERROR("Can't get mdd ctxt %d\n", rc);
1915                 return rc;
1916         }
1917
1918         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1919         if (rc)
1920                 CERROR("Can't set mdt ctxt %d\n", rc);
1921
1922         return rc;
1923 }
1924
1925 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1926                                  struct mdt_device *mdt, int idx)
1927 {
1928         struct llog_ctxt *ctxt;
1929
1930         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1931         if (ctxt == NULL)
1932                 return 0;
1933         /* Put once for the get we just did, and once for the clone */
1934         llog_ctxt_put(ctxt);
1935         llog_ctxt_put(ctxt);
1936         return 0;
1937 }
1938
1939 static int mdt_llog_create(struct mdt_thread_info *info)
1940 {
1941         int rc;
1942
1943         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1944         rc = llog_origin_handle_create(mdt_info_req(info));
1945         return (rc < 0 ? err_serious(rc) : rc);
1946 }
1947
1948 static int mdt_llog_destroy(struct mdt_thread_info *info)
1949 {
1950         int rc;
1951
1952         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1953         rc = llog_origin_handle_destroy(mdt_info_req(info));
1954         return (rc < 0 ? err_serious(rc) : rc);
1955 }
1956
1957 static int mdt_llog_read_header(struct mdt_thread_info *info)
1958 {
1959         int rc;
1960
1961         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1962         rc = llog_origin_handle_read_header(mdt_info_req(info));
1963         return (rc < 0 ? err_serious(rc) : rc);
1964 }
1965
1966 static int mdt_llog_next_block(struct mdt_thread_info *info)
1967 {
1968         int rc;
1969
1970         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1971         rc = llog_origin_handle_next_block(mdt_info_req(info));
1972         return (rc < 0 ? err_serious(rc) : rc);
1973 }
1974
1975 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1976 {
1977         int rc;
1978
1979         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1980         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1981         return (rc < 0 ? err_serious(rc) : rc);
1982 }
1983
1984
1985 /*
1986  * DLM handlers.
1987  */
1988 static struct ldlm_callback_suite cbs = {
1989         .lcs_completion = ldlm_server_completion_ast,
1990         .lcs_blocking   = ldlm_server_blocking_ast,
1991         .lcs_glimpse    = NULL
1992 };
1993
1994 static int mdt_enqueue(struct mdt_thread_info *info)
1995 {
1996         struct ptlrpc_request *req;
1997         int rc;
1998
1999         /*
2000          * info->mti_dlm_req already contains swapped and (if necessary)
2001          * converted dlm request.
2002          */
2003         LASSERT(info->mti_dlm_req != NULL);
2004
2005         req = mdt_info_req(info);
2006         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2007                                   req, info->mti_dlm_req, &cbs);
2008         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2009         return rc ? err_serious(rc) : req->rq_status;
2010 }
2011
2012 static int mdt_convert(struct mdt_thread_info *info)
2013 {
2014         int rc;
2015         struct ptlrpc_request *req;
2016
2017         LASSERT(info->mti_dlm_req);
2018         req = mdt_info_req(info);
2019         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2020         return rc ? err_serious(rc) : req->rq_status;
2021 }
2022
2023 static int mdt_bl_callback(struct mdt_thread_info *info)
2024 {
2025         CERROR("bl callbacks should not happen on MDS\n");
2026         LBUG();
2027         return err_serious(-EOPNOTSUPP);
2028 }
2029
2030 static int mdt_cp_callback(struct mdt_thread_info *info)
2031 {
2032         CERROR("cp callbacks should not happen on MDS\n");
2033         LBUG();
2034         return err_serious(-EOPNOTSUPP);
2035 }
2036
2037 /*
2038  * sec context handlers
2039  */
2040 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2041 {
2042         int rc;
2043
2044         rc = mdt_handle_idmap(info);
2045
2046         if (unlikely(rc)) {
2047                 struct ptlrpc_request *req = mdt_info_req(info);
2048                 __u32                  opc;
2049
2050                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2051                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2052                         sptlrpc_svc_ctx_invalidate(req);
2053         }
2054
2055         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2056
2057         return rc;
2058 }
2059
2060 static struct mdt_object *mdt_obj(struct lu_object *o)
2061 {
2062         LASSERT(lu_device_is_mdt(o->lo_dev));
2063         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2064 }
2065
2066 struct mdt_object *mdt_object_find(const struct lu_env *env,
2067                                    struct mdt_device *d,
2068                                    const struct lu_fid *f)
2069 {
2070         struct lu_object *o;
2071         struct mdt_object *m;
2072         ENTRY;
2073
2074         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2075         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2076         if (unlikely(IS_ERR(o)))
2077                 m = (struct mdt_object *)o;
2078         else
2079                 m = mdt_obj(o);
2080         RETURN(m);
2081 }
2082
2083 /**
2084  * Asyncronous commit for mdt device.
2085  *
2086  * Pass asynchonous commit call down the MDS stack.
2087  *
2088  * \param env environment
2089  * \param mdt the mdt device
2090  */
2091 static void mdt_device_commit_async(const struct lu_env *env,
2092                                     struct mdt_device *mdt)
2093 {
2094         struct dt_device *dt = mdt->mdt_bottom;
2095         int rc;
2096
2097         rc = dt->dd_ops->dt_commit_async(env, dt);
2098         if (unlikely(rc != 0))
2099                 CWARN("async commit start failed with rc = %d", rc);
2100 }
2101
2102 /**
2103  * Mark the lock as "synchonous".
2104  *
2105  * Mark the lock to deffer transaction commit to the unlock time.
2106  *
2107  * \param lock the lock to mark as "synchonous"
2108  *
2109  * \see mdt_is_lock_sync
2110  * \see mdt_save_lock
2111  */
2112 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2113 {
2114         lock->l_ast_data = (void*)1;
2115 }
2116
2117 /**
2118  * Check whehter the lock "synchonous" or not.
2119  *
2120  * \param lock the lock to check
2121  * \retval 1 the lock is "synchonous"
2122  * \retval 0 the lock isn't "synchronous"
2123  *
2124  * \see mdt_set_lock_sync
2125  * \see mdt_save_lock
2126  */
2127 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2128 {
2129         return lock->l_ast_data != NULL;
2130 }
2131
2132 /**
2133  * Blocking AST for mdt locks.
2134  *
2135  * Starts transaction commit if in case of COS lock conflict or
2136  * deffers such a commit to the mdt_save_lock.
2137  *
2138  * \param lock the lock which blocks a request or cancelling lock
2139  * \param desc unused
2140  * \param data unused
2141  * \param flag indicates whether this cancelling or blocking callback
2142  * \retval 0
2143  * \see ldlm_blocking_ast_nocheck
2144  */
2145 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2146                      void *data, int flag)
2147 {
2148         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2149         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2150         int rc;
2151         ENTRY;
2152
2153         if (flag == LDLM_CB_CANCELING)
2154                 RETURN(0);
2155         lock_res_and_lock(lock);
2156         if (lock->l_blocking_ast != mdt_blocking_ast) {
2157                 unlock_res_and_lock(lock);
2158                 RETURN(0);
2159         }
2160         if (mdt_cos_is_enabled(mdt) &&
2161             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2162             lock->l_blocking_lock != NULL &&
2163             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2164                 mdt_set_lock_sync(lock);
2165         }
2166         rc = ldlm_blocking_ast_nocheck(lock);
2167
2168         /* There is no lock conflict if l_blocking_lock == NULL,
2169          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2170          * when the last reference to a local lock was released */
2171         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2172                 struct lu_env env;
2173
2174                 rc = lu_env_init(&env, LCT_MD_THREAD);
2175                 if (unlikely(rc != 0))
2176                         CWARN("lu_env initialization failed with rc = %d,"
2177                               "cannot start asynchronous commit\n", rc);
2178                 else
2179                         mdt_device_commit_async(&env, mdt);
2180                 lu_env_fini(&env);
2181         }
2182         RETURN(rc);
2183 }
2184
2185 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2186                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2187 {
2188         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2189         ldlm_policy_data_t *policy = &info->mti_policy;
2190         struct ldlm_res_id *res_id = &info->mti_res_id;
2191         int rc;
2192         ENTRY;
2193
2194         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2195         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2196         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2197         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2198
2199         if (mdt_object_exists(o) < 0) {
2200                 if (locality == MDT_CROSS_LOCK) {
2201                         /* cross-ref object fix */
2202                         ibits &= ~MDS_INODELOCK_UPDATE;
2203                         ibits |= MDS_INODELOCK_LOOKUP;
2204                 } else {
2205                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2206                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2207                 }
2208                 /* No PDO lock on remote object */
2209                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2210         }
2211
2212         if (lh->mlh_type == MDT_PDO_LOCK) {
2213                 /* check for exists after object is locked */
2214                 if (mdt_object_exists(o) == 0) {
2215                         /* Non-existent object shouldn't have PDO lock */
2216                         RETURN(-ESTALE);
2217                 } else {
2218                         /* Non-dir object shouldn't have PDO lock */
2219                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2220                                 RETURN(-ENOTDIR);
2221                 }
2222         }
2223
2224         memset(policy, 0, sizeof(*policy));
2225         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2226
2227         /*
2228          * Take PDO lock on whole directory and build correct @res_id for lock
2229          * on part of directory.
2230          */
2231         if (lh->mlh_pdo_hash != 0) {
2232                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2233                 mdt_lock_pdo_mode(info, o, lh);
2234                 if (lh->mlh_pdo_mode != LCK_NL) {
2235                         /*
2236                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2237                          * is never going to be sent to client and we do not
2238                          * want it slowed down due to possible cancels.
2239                          */
2240                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2241                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2242                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2243                                           &info->mti_exp->exp_handle.h_cookie);
2244                         if (unlikely(rc))
2245                                 RETURN(rc);
2246                 }
2247
2248                 /*
2249                  * Finish res_id initializing by name hash marking part of
2250                  * directory which is taking modification.
2251                  */
2252                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2253         }
2254
2255         policy->l_inodebits.bits = ibits;
2256
2257         /*
2258          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2259          * going to be sent to client. If it is - mdt_intent_policy() path will
2260          * fix it up and turn FL_LOCAL flag off.
2261          */
2262         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2263                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2264                           &info->mti_exp->exp_handle.h_cookie);
2265         if (rc)
2266                 mdt_object_unlock(info, o, lh, 1);
2267         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2268                  lh->mlh_pdo_hash != 0 &&
2269                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2270                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2271         }
2272
2273         RETURN(rc);
2274 }
2275
2276 /**
2277  * Save a lock within request object.
2278  *
2279  * Keep the lock referenced until whether client ACK or transaction
2280  * commit happens or release the lock immediately depending on input
2281  * parameters. If COS is ON, a write lock is converted to COS lock
2282  * before saving.
2283  *
2284  * \param info thead info object
2285  * \param h lock handle
2286  * \param mode lock mode
2287  * \param decref force immediate lock releasing
2288  */
2289 static
2290 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2291                    ldlm_mode_t mode, int decref)
2292 {
2293         ENTRY;
2294
2295         if (lustre_handle_is_used(h)) {
2296                 if (decref || !info->mti_has_trans ||
2297                     !(mode & (LCK_PW | LCK_EX))){
2298                         mdt_fid_unlock(h, mode);
2299                 } else {
2300                         struct mdt_device *mdt = info->mti_mdt;
2301                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2302                         struct ptlrpc_request *req = mdt_info_req(info);
2303                         int no_ack = 0;
2304
2305                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2306                                  h->cookie);
2307                         CDEBUG(D_HA, "request = %p reply state = %p"
2308                                " transno = "LPD64"\n",
2309                                req, req->rq_reply_state, req->rq_transno);
2310                         if (mdt_cos_is_enabled(mdt)) {
2311                                 no_ack = 1;
2312                                 ldlm_lock_downgrade(lock, LCK_COS);
2313                                 mode = LCK_COS;
2314                         }
2315                         ptlrpc_save_lock(req, h, mode, no_ack);
2316                         if (mdt_is_lock_sync(lock)) {
2317                                 CDEBUG(D_HA, "found sync-lock,"
2318                                        " async commit started\n");
2319                                 mdt_device_commit_async(info->mti_env,
2320                                                         mdt);
2321                         }
2322                         LDLM_LOCK_PUT(lock);
2323                 }
2324                 h->cookie = 0ull;
2325         }
2326
2327         EXIT;
2328 }
2329
2330 /**
2331  * Unlock mdt object.
2332  *
2333  * Immeditely release the regular lock and the PDO lock or save the
2334  * lock in reqeuest and keep them referenced until client ACK or
2335  * transaction commit.
2336  *
2337  * \param info thread info object
2338  * \param o mdt object
2339  * \param lh mdt lock handle referencing regular and PDO locks
2340  * \param decref force immediate lock releasing
2341  */
2342 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2343                        struct mdt_lock_handle *lh, int decref)
2344 {
2345         ENTRY;
2346
2347         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2348         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2349
2350         EXIT;
2351 }
2352
2353 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2354                                         const struct lu_fid *f,
2355                                         struct mdt_lock_handle *lh,
2356                                         __u64 ibits)
2357 {
2358         struct mdt_object *o;
2359
2360         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2361         if (!IS_ERR(o)) {
2362                 int rc;
2363
2364                 rc = mdt_object_lock(info, o, lh, ibits,
2365                                      MDT_LOCAL_LOCK);
2366                 if (rc != 0) {
2367                         mdt_object_put(info->mti_env, o);
2368                         o = ERR_PTR(rc);
2369                 }
2370         }
2371         return o;
2372 }
2373
2374 void mdt_object_unlock_put(struct mdt_thread_info * info,
2375                            struct mdt_object * o,
2376                            struct mdt_lock_handle *lh,
2377                            int decref)
2378 {
2379         mdt_object_unlock(info, o, lh, decref);
2380         mdt_object_put(info->mti_env, o);
2381 }
2382
2383 static struct mdt_handler *mdt_handler_find(__u32 opc,
2384                                             struct mdt_opc_slice *supported)
2385 {
2386         struct mdt_opc_slice *s;
2387         struct mdt_handler   *h;
2388
2389         h = NULL;
2390         for (s = supported; s->mos_hs != NULL; s++) {
2391                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2392                         h = s->mos_hs + (opc - s->mos_opc_start);
2393                         if (likely(h->mh_opc != 0))
2394                                 LASSERTF(h->mh_opc == opc,
2395                                          "opcode mismatch %d != %d\n",
2396                                          h->mh_opc, opc);
2397                         else
2398                                 h = NULL; /* unsupported opc */
2399                         break;
2400                 }
2401         }
2402         return h;
2403 }
2404
2405 static int mdt_lock_resname_compat(struct mdt_device *m,
2406                                    struct ldlm_request *req)
2407 {
2408         /* XXX something... later. */
2409         return 0;
2410 }
2411
2412 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2413 {
2414         /* XXX something... later. */
2415         return 0;
2416 }
2417
2418 /*
2419  * Generic code handling requests that have struct mdt_body passed in:
2420  *
2421  *  - extract mdt_body from request and save it in @info, if present;
2422  *
2423  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2424  *  @info;
2425  *
2426  *  - if HABEO_CORPUS flag is set for this request type check whether object
2427  *  actually exists on storage (lu_object_exists()).
2428  *
2429  */
2430 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2431 {
2432         const struct mdt_body    *body;
2433         struct mdt_object        *obj;
2434         const struct lu_env      *env;
2435         struct req_capsule       *pill;
2436         int                       rc;
2437         ENTRY;
2438
2439         env = info->mti_env;
2440         pill = info->mti_pill;
2441
2442         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2443         if (body == NULL)
2444                 RETURN(-EFAULT);
2445
2446         if (!(body->valid & OBD_MD_FLID))
2447                 RETURN(0);
2448
2449         if (!fid_is_sane(&body->fid1)) {
2450                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2451                 RETURN(-EINVAL);
2452         }
2453
2454         /*
2455          * Do not get size or any capa fields before we check that request
2456          * contains capa actually. There are some requests which do not, for
2457          * instance MDS_IS_SUBDIR.
2458          */
2459         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2460             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2461                 mdt_set_capainfo(info, 0, &body->fid1,
2462                                  req_capsule_client_get(pill, &RMF_CAPA1));
2463
2464         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2465         if (!IS_ERR(obj)) {
2466                 if ((flags & HABEO_CORPUS) &&
2467                     !mdt_object_exists(obj)) {
2468                         mdt_object_put(env, obj);
2469                         /* for capability renew ENOENT will be handled in
2470                          * mdt_renew_capa */
2471                         if (body->valid & OBD_MD_FLOSSCAPA)
2472                                 rc = 0;
2473                         else
2474                                 rc = -ENOENT;
2475                 } else {
2476                         info->mti_object = obj;
2477                         rc = 0;
2478                 }
2479         } else
2480                 rc = PTR_ERR(obj);
2481
2482         RETURN(rc);
2483 }
2484
2485 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2486 {
2487         struct req_capsule *pill = info->mti_pill;
2488         int rc;
2489         ENTRY;
2490
2491         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2492                 rc = mdt_body_unpack(info, flags);
2493         else
2494                 rc = 0;
2495
2496         if (rc == 0 && (flags & HABEO_REFERO)) {
2497                 struct mdt_device *mdt = info->mti_mdt;
2498
2499                 /* Pack reply. */
2500
2501                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2502                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2503                                              mdt->mdt_max_mdsize);
2504                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2505                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2506                                              mdt->mdt_max_cookiesize);
2507
2508                 rc = req_capsule_server_pack(pill);
2509         }
2510         RETURN(rc);
2511 }
2512
2513 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2514 {
2515         struct md_device *next = m->mdt_child;
2516
2517         return next->md_ops->mdo_init_capa_ctxt(env, next,
2518                                                 m->mdt_opts.mo_mds_capa,
2519                                                 m->mdt_capa_timeout,
2520                                                 m->mdt_capa_alg,
2521                                                 m->mdt_capa_keys);
2522 }
2523
2524 /*
2525  * Invoke handler for this request opc. Also do necessary preprocessing
2526  * (according to handler ->mh_flags), and post-processing (setting of
2527  * ->last_{xid,committed}).
2528  */
2529 static int mdt_req_handle(struct mdt_thread_info *info,
2530                           struct mdt_handler *h, struct ptlrpc_request *req)
2531 {
2532         int   rc, serious = 0;
2533         __u32 flags;
2534
2535         ENTRY;
2536
2537         LASSERT(h->mh_act != NULL);
2538         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2539         LASSERT(current->journal_info == NULL);
2540
2541         /*
2542          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2543          * to put same checks into handlers like mdt_close(), mdt_reint(),
2544          * etc., without talking to mdt authors first. Checking same thing
2545          * there again is useless and returning 0 error without packing reply
2546          * is buggy! Handlers either pack reply or return error.
2547          *
2548          * We return 0 here and do not send any reply in order to emulate
2549          * network failure. Do not send any reply in case any of NET related
2550          * fail_id has occured.
2551          */
2552         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2553                 RETURN(0);
2554
2555         rc = 0;
2556         flags = h->mh_flags;
2557         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2558
2559         if (h->mh_fmt != NULL) {
2560                 req_capsule_set(info->mti_pill, h->mh_fmt);
2561                 rc = mdt_unpack_req_pack_rep(info, flags);
2562         }
2563
2564         if (rc == 0 && flags & MUTABOR &&
2565             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2566                 /* should it be rq_status? */
2567                 rc = -EROFS;
2568
2569         if (rc == 0 && flags & HABEO_CLAVIS) {
2570                 struct ldlm_request *dlm_req;
2571
2572                 LASSERT(h->mh_fmt != NULL);
2573
2574                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2575                 if (dlm_req != NULL) {
2576                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2577                                         LDLM_IBITS &&
2578                                      dlm_req->lock_desc.l_policy_data.\
2579                                         l_inodebits.bits == 0)) {
2580                                 /*
2581                                  * Lock without inodebits makes no sense and
2582                                  * will oops later in ldlm. If client miss to
2583                                  * set such bits, do not trigger ASSERTION.
2584                                  *
2585                                  * For liblustre flock case, it maybe zero.
2586                                  */
2587                                 rc = -EPROTO;
2588                         } else {
2589                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2590                                         rc = mdt_lock_resname_compat(
2591                                                                 info->mti_mdt,
2592                                                                 dlm_req);
2593                                 info->mti_dlm_req = dlm_req;
2594                         }
2595                 } else {
2596                         rc = -EFAULT;
2597                 }
2598         }
2599
2600         /* capability setting changed via /proc, needs reinitialize ctxt */
2601         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2602                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2603                 info->mti_mdt->mdt_capa_conf = 0;
2604         }
2605
2606         if (likely(rc == 0)) {
2607                 /*
2608                  * Process request, there can be two types of rc:
2609                  * 1) errors with msg unpack/pack, other failures outside the
2610                  * operation itself. This is counted as serious errors;
2611                  * 2) errors during fs operation, should be placed in rq_status
2612                  * only
2613                  */
2614                 rc = h->mh_act(info);
2615                 if (rc == 0 &&
2616                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2617                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2618                                   "pack reply and returned 0 error\n",
2619                                   h->mh_name);
2620                         LBUG();
2621                 }
2622                 serious = is_serious(rc);
2623                 rc = clear_serious(rc);
2624         } else
2625                 serious = 1;
2626
2627         req->rq_status = rc;
2628
2629         /*
2630          * ELDLM_* codes which > 0 should be in rq_status only as well as
2631          * all non-serious errors.
2632          */
2633         if (rc > 0 || !serious)
2634                 rc = 0;
2635
2636         LASSERT(current->journal_info == NULL);
2637
2638         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2639             info->mti_mdt->mdt_opts.mo_compat_resname) {
2640                 struct ldlm_reply *dlmrep;
2641
2642                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2643                 if (dlmrep != NULL)
2644                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2645         }
2646
2647         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2648         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2649                 target_committed_to_req(req);
2650
2651         if (unlikely(req_is_replay(req) &&
2652                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2653                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2654                 LBUG();
2655         }
2656
2657         target_send_reply(req, rc, info->mti_fail_id);
2658         RETURN(0);
2659 }
2660
2661 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2662 {
2663         lh->mlh_type = MDT_NUL_LOCK;
2664         lh->mlh_reg_lh.cookie = 0ull;
2665         lh->mlh_reg_mode = LCK_MINMODE;
2666         lh->mlh_pdo_lh.cookie = 0ull;
2667         lh->mlh_pdo_mode = LCK_MINMODE;
2668 }
2669
2670 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2671 {
2672         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2673         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2674 }
2675
2676 /*
2677  * Initialize fields of struct mdt_thread_info. Other fields are left in
2678  * uninitialized state, because it's too expensive to zero out whole
2679  * mdt_thread_info (> 1K) on each request arrival.
2680  */
2681 static void mdt_thread_info_init(struct ptlrpc_request *req,
2682                                  struct mdt_thread_info *info)
2683 {
2684         int i;
2685         struct md_capainfo *ci;
2686
2687         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2688         info->mti_pill = &req->rq_pill;
2689
2690         /* lock handle */
2691         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2692                 mdt_lock_handle_init(&info->mti_lh[i]);
2693
2694         /* mdt device: it can be NULL while CONNECT */
2695         if (req->rq_export) {
2696                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2697                 info->mti_exp = req->rq_export;
2698         } else
2699                 info->mti_mdt = NULL;
2700         info->mti_env = req->rq_svc_thread->t_env;
2701         ci = md_capainfo(info->mti_env);
2702         memset(ci, 0, sizeof *ci);
2703         if (req->rq_export) {
2704                 if (exp_connect_rmtclient(req->rq_export))
2705                         ci->mc_auth = LC_ID_CONVERT;
2706                 else if (req->rq_export->exp_connect_flags &
2707                          OBD_CONNECT_MDS_CAPA)
2708                         ci->mc_auth = LC_ID_PLAIN;
2709                 else
2710                         ci->mc_auth = LC_ID_NONE;
2711         }
2712
2713         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2714         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2715         info->mti_mos = NULL;
2716
2717         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2718         info->mti_body = NULL;
2719         info->mti_object = NULL;
2720         info->mti_dlm_req = NULL;
2721         info->mti_has_trans = 0;
2722         info->mti_no_need_trans = 0;
2723         info->mti_cross_ref = 0;
2724         info->mti_opdata = 0;
2725
2726         /* To not check for split by default. */
2727         info->mti_spec.sp_ck_split = 0;
2728         info->mti_spec.no_create = 0;
2729 }
2730
2731 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2732 {
2733         int i;
2734
2735         req_capsule_fini(info->mti_pill);
2736         if (info->mti_object != NULL) {
2737                 /*
2738                  * freeing an object may lead to OSD level transaction, do not
2739                  * let it mess with MDT. bz19385.
2740                  */
2741                 info->mti_no_need_trans = 1;
2742                 mdt_object_put(info->mti_env, info->mti_object);
2743                 info->mti_object = NULL;
2744         }
2745         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2746                 mdt_lock_handle_fini(&info->mti_lh[i]);
2747         info->mti_env = NULL;
2748 }
2749
2750 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2751                                        struct obd_device *obd, int *process)
2752 {
2753         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2754         case MDS_CONNECT: /* This will never get here, but for completeness. */
2755         case OST_CONNECT: /* This will never get here, but for completeness. */
2756         case MDS_DISCONNECT:
2757         case OST_DISCONNECT:
2758                *process = 1;
2759                RETURN(0);
2760
2761         case MDS_CLOSE:
2762         case MDS_DONE_WRITING:
2763         case MDS_SYNC: /* used in unmounting */
2764         case OBD_PING:
2765         case MDS_REINT:
2766         case SEQ_QUERY:
2767         case FLD_QUERY:
2768         case LDLM_ENQUEUE:
2769                 *process = target_queue_recovery_request(req, obd);
2770                 RETURN(0);
2771
2772         default:
2773                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2774                 *process = -EAGAIN;
2775                 RETURN(0);
2776         }
2777 }
2778
2779 /*
2780  * Handle recovery. Return:
2781  *        +1: continue request processing;
2782  *       -ve: abort immediately with the given error code;
2783  *         0: send reply with error code in req->rq_status;
2784  */
2785 static int mdt_recovery(struct mdt_thread_info *info)
2786 {
2787         struct ptlrpc_request *req = mdt_info_req(info);
2788         struct obd_device *obd;
2789
2790         ENTRY;
2791
2792         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2793         case MDS_CONNECT:
2794         case SEC_CTX_INIT:
2795         case SEC_CTX_INIT_CONT:
2796         case SEC_CTX_FINI:
2797                 {
2798 #if 0
2799                         int rc;
2800
2801                         rc = mdt_handle_idmap(info);
2802                         if (rc)
2803                                 RETURN(rc);
2804                         else
2805 #endif
2806                                 RETURN(+1);
2807                 }
2808         }
2809
2810         if (unlikely(!class_connected_export(req->rq_export))) {
2811                 CERROR("operation %d on unconnected MDS from %s\n",
2812                        lustre_msg_get_opc(req->rq_reqmsg),
2813                        libcfs_id2str(req->rq_peer));
2814                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2815                  * mds_A will get -ENOTCONN(especially for ping req),
2816                  * which will cause that mds_A deactive timeout, then when
2817                  * mds_A cleanup, the cleanup process will be suspended since
2818                  * deactive timeout is not zero.
2819                  */
2820                 req->rq_status = -ENOTCONN;
2821                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2822                 RETURN(0);
2823         }
2824
2825         /* sanity check: if the xid matches, the request must be marked as a
2826          * resent or replayed */
2827         if (req_xid_is_last(req)) {
2828                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2829                       (MSG_RESENT | MSG_REPLAY))) {
2830                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2831                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2832                                   lustre_msg_get_flags(req->rq_reqmsg));
2833                         LBUG();
2834                         req->rq_status = -ENOTCONN;
2835                         RETURN(-ENOTCONN);
2836                 }
2837         }
2838
2839         /* else: note the opposite is not always true; a RESENT req after a
2840          * failover will usually not match the last_xid, since it was likely
2841          * never committed. A REPLAYed request will almost never match the
2842          * last xid, however it could for a committed, but still retained,
2843          * open. */
2844
2845         obd = req->rq_export->exp_obd;
2846
2847         /* Check for aborted recovery... */
2848         if (unlikely(obd->obd_recovering)) {
2849                 int rc;
2850                 int should_process;
2851                 DEBUG_REQ(D_INFO, req, "Got new replay");
2852                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2853                 if (rc != 0 || !should_process)
2854                         RETURN(rc);
2855                 else if (should_process < 0) {
2856                         req->rq_status = should_process;
2857                         rc = ptlrpc_error(req);
2858                         RETURN(rc);
2859                 }
2860         }
2861         RETURN(+1);
2862 }
2863
2864 static int mdt_msg_check_version(struct lustre_msg *msg)
2865 {
2866         int rc;
2867
2868         switch (lustre_msg_get_opc(msg)) {
2869         case MDS_CONNECT:
2870         case MDS_DISCONNECT:
2871         case OBD_PING:
2872         case SEC_CTX_INIT:
2873         case SEC_CTX_INIT_CONT:
2874         case SEC_CTX_FINI:
2875                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2876                 if (rc)
2877                         CERROR("bad opc %u version %08x, expecting %08x\n",
2878                                lustre_msg_get_opc(msg),
2879                                lustre_msg_get_version(msg),
2880                                LUSTRE_OBD_VERSION);
2881                 break;
2882         case MDS_GETSTATUS:
2883         case MDS_GETATTR:
2884         case MDS_GETATTR_NAME:
2885         case MDS_STATFS:
2886         case MDS_READPAGE:
2887         case MDS_WRITEPAGE:
2888         case MDS_IS_SUBDIR:
2889         case MDS_REINT:
2890         case MDS_CLOSE:
2891         case MDS_DONE_WRITING:
2892         case MDS_PIN:
2893         case MDS_SYNC:
2894         case MDS_GETXATTR:
2895         case MDS_SETXATTR:
2896         case MDS_SET_INFO:
2897         case MDS_GET_INFO:
2898         case MDS_QUOTACHECK:
2899         case MDS_QUOTACTL:
2900         case QUOTA_DQACQ:
2901         case QUOTA_DQREL:
2902         case SEQ_QUERY:
2903         case FLD_QUERY:
2904                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2905                 if (rc)
2906                         CERROR("bad opc %u version %08x, expecting %08x\n",
2907                                lustre_msg_get_opc(msg),
2908                                lustre_msg_get_version(msg),
2909                                LUSTRE_MDS_VERSION);
2910                 break;
2911         case LDLM_ENQUEUE:
2912         case LDLM_CONVERT:
2913         case LDLM_BL_CALLBACK:
2914         case LDLM_CP_CALLBACK:
2915                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2916                 if (rc)
2917                         CERROR("bad opc %u version %08x, expecting %08x\n",
2918                                lustre_msg_get_opc(msg),
2919                                lustre_msg_get_version(msg),
2920                                LUSTRE_DLM_VERSION);
2921                 break;
2922         case OBD_LOG_CANCEL:
2923         case LLOG_ORIGIN_HANDLE_CREATE:
2924         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2925         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2926         case LLOG_ORIGIN_HANDLE_CLOSE:
2927         case LLOG_ORIGIN_HANDLE_DESTROY:
2928         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2929         case LLOG_CATINFO:
2930                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2931                 if (rc)
2932                         CERROR("bad opc %u version %08x, expecting %08x\n",
2933                                lustre_msg_get_opc(msg),
2934                                lustre_msg_get_version(msg),
2935                                LUSTRE_LOG_VERSION);
2936                 break;
2937         default:
2938                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2939                 rc = -ENOTSUPP;
2940         }
2941         return rc;
2942 }
2943
2944 static int mdt_handle0(struct ptlrpc_request *req,
2945                        struct mdt_thread_info *info,
2946                        struct mdt_opc_slice *supported)
2947 {
2948         struct mdt_handler *h;
2949         struct lustre_msg  *msg;
2950         int                 rc;
2951
2952         ENTRY;
2953
2954         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2955                 RETURN(0);
2956
2957         LASSERT(current->journal_info == NULL);
2958
2959         msg = req->rq_reqmsg;
2960         rc = mdt_msg_check_version(msg);
2961         if (likely(rc == 0)) {
2962                 rc = mdt_recovery(info);
2963                 if (likely(rc == +1)) {
2964                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2965                                              supported);
2966                         if (likely(h != NULL)) {
2967                                 rc = mdt_req_handle(info, h, req);
2968                         } else {
2969                                 CERROR("The unsupported opc: 0x%x\n",
2970                                        lustre_msg_get_opc(msg) );
2971                                 req->rq_status = -ENOTSUPP;
2972                                 rc = ptlrpc_error(req);
2973                                 RETURN(rc);
2974                         }
2975                 }
2976         } else
2977                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2978         RETURN(rc);
2979 }
2980
2981 /*
2982  * MDT handler function called by ptlrpc service thread when request comes.
2983  *
2984  * XXX common "target" functionality should be factored into separate module
2985  * shared by mdt, ost and stand-alone services like fld.
2986  */
2987 static int mdt_handle_common(struct ptlrpc_request *req,
2988                              struct mdt_opc_slice *supported)
2989 {
2990         struct lu_env          *env;
2991         struct mdt_thread_info *info;
2992         int                     rc;
2993         ENTRY;
2994
2995         env = req->rq_svc_thread->t_env;
2996         LASSERT(env != NULL);
2997         LASSERT(env->le_ses != NULL);
2998         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2999         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3000         LASSERT(info != NULL);
3001
3002         mdt_thread_info_init(req, info);
3003
3004         rc = mdt_handle0(req, info, supported);
3005
3006         mdt_thread_info_fini(info);
3007         RETURN(rc);
3008 }
3009
3010 /*
3011  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3012  * as well.
3013  */
3014 int mdt_recovery_handle(struct ptlrpc_request *req)
3015 {
3016         int rc;
3017         ENTRY;
3018
3019         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3020         case FLD_QUERY:
3021                 rc = mdt_handle_common(req, mdt_fld_handlers);
3022                 break;
3023         case SEQ_QUERY:
3024                 rc = mdt_handle_common(req, mdt_seq_handlers);
3025                 break;
3026         default:
3027                 rc = mdt_handle_common(req, mdt_regular_handlers);
3028                 break;
3029         }
3030
3031         RETURN(rc);
3032 }
3033
3034 static int mdt_regular_handle(struct ptlrpc_request *req)
3035 {
3036         return mdt_handle_common(req, mdt_regular_handlers);
3037 }
3038
3039 static int mdt_readpage_handle(struct ptlrpc_request *req)
3040 {
3041         return mdt_handle_common(req, mdt_readpage_handlers);
3042 }
3043
3044 static int mdt_xmds_handle(struct ptlrpc_request *req)
3045 {
3046         return mdt_handle_common(req, mdt_xmds_handlers);
3047 }
3048
3049 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3050 {
3051         return mdt_handle_common(req, mdt_seq_handlers);
3052 }
3053
3054 static int mdt_mdss_handle(struct ptlrpc_request *req)
3055 {
3056         return mdt_handle_common(req, mdt_seq_handlers);
3057 }
3058
3059 static int mdt_dtss_handle(struct ptlrpc_request *req)
3060 {
3061         return mdt_handle_common(req, mdt_seq_handlers);
3062 }
3063
3064 static int mdt_fld_handle(struct ptlrpc_request *req)
3065 {
3066         return mdt_handle_common(req, mdt_fld_handlers);
3067 }
3068
3069 enum mdt_it_code {
3070         MDT_IT_OPEN,
3071         MDT_IT_OCREAT,
3072         MDT_IT_CREATE,
3073         MDT_IT_GETATTR,
3074         MDT_IT_READDIR,
3075         MDT_IT_LOOKUP,
3076         MDT_IT_UNLINK,
3077         MDT_IT_TRUNC,
3078         MDT_IT_GETXATTR,
3079         MDT_IT_NR
3080 };
3081
3082 static int mdt_intent_getattr(enum mdt_it_code opcode,
3083                               struct mdt_thread_info *info,
3084                               struct ldlm_lock **,
3085                               int);
3086 static int mdt_intent_reint(enum mdt_it_code opcode,
3087                             struct mdt_thread_info *info,
3088                             struct ldlm_lock **,
3089                             int);
3090
3091 static struct mdt_it_flavor {
3092         const struct req_format *it_fmt;
3093         __u32                    it_flags;
3094         int                    (*it_act)(enum mdt_it_code ,
3095                                          struct mdt_thread_info *,
3096                                          struct ldlm_lock **,
3097                                          int);
3098         long                     it_reint;
3099 } mdt_it_flavor[] = {
3100         [MDT_IT_OPEN]     = {
3101                 .it_fmt   = &RQF_LDLM_INTENT,
3102                 /*.it_flags = HABEO_REFERO,*/
3103                 .it_flags = 0,
3104                 .it_act   = mdt_intent_reint,
3105                 .it_reint = REINT_OPEN
3106         },
3107         [MDT_IT_OCREAT]   = {
3108                 .it_fmt   = &RQF_LDLM_INTENT,
3109                 .it_flags = MUTABOR,
3110                 .it_act   = mdt_intent_reint,
3111                 .it_reint = REINT_OPEN
3112         },
3113         [MDT_IT_CREATE]   = {
3114                 .it_fmt   = &RQF_LDLM_INTENT,
3115                 .it_flags = MUTABOR,
3116                 .it_act   = mdt_intent_reint,
3117                 .it_reint = REINT_CREATE
3118         },
3119         [MDT_IT_GETATTR]  = {
3120                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3121                 .it_flags = HABEO_REFERO,
3122                 .it_act   = mdt_intent_getattr
3123         },
3124         [MDT_IT_READDIR]  = {
3125                 .it_fmt   = NULL,
3126                 .it_flags = 0,
3127                 .it_act   = NULL
3128         },
3129         [MDT_IT_LOOKUP]   = {
3130                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3131                 .it_flags = HABEO_REFERO,
3132                 .it_act   = mdt_intent_getattr
3133         },
3134         [MDT_IT_UNLINK]   = {
3135                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3136                 .it_flags = MUTABOR,
3137                 .it_act   = NULL,
3138                 .it_reint = REINT_UNLINK
3139         },
3140         [MDT_IT_TRUNC]    = {
3141                 .it_fmt   = NULL,
3142                 .it_flags = MUTABOR,
3143                 .it_act   = NULL
3144         },
3145         [MDT_IT_GETXATTR] = {
3146                 .it_fmt   = NULL,
3147                 .it_flags = 0,
3148                 .it_act   = NULL
3149         }
3150 };
3151
3152 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3153                             struct ldlm_lock **lockp,
3154                             struct ldlm_lock *new_lock,
3155                             struct mdt_lock_handle *lh,
3156                             int flags)
3157 {
3158         struct ptlrpc_request  *req = mdt_info_req(info);
3159         struct ldlm_lock       *lock = *lockp;
3160
3161         /*
3162          * Get new lock only for cases when possible resent did not find any
3163          * lock.
3164          */
3165         if (new_lock == NULL)
3166                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3167
3168         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3169                 lh->mlh_reg_lh.cookie = 0;
3170                 RETURN(0);
3171         }
3172
3173         LASSERTF(new_lock != NULL,
3174                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3175
3176         /*
3177          * If we've already given this lock to a client once, then we should
3178          * have no readers or writers.  Otherwise, we should have one reader
3179          * _or_ writer ref (which will be zeroed below) before returning the
3180          * lock to a client.
3181          */
3182         if (new_lock->l_export == req->rq_export) {
3183                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3184         } else {
3185                 LASSERT(new_lock->l_export == NULL);
3186                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3187         }
3188
3189         *lockp = new_lock;
3190
3191         if (new_lock->l_export == req->rq_export) {
3192                 /*
3193                  * Already gave this to the client, which means that we
3194                  * reconstructed a reply.
3195                  */
3196                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3197                         MSG_RESENT);
3198                 lh->mlh_reg_lh.cookie = 0;
3199                 RETURN(ELDLM_LOCK_REPLACED);
3200         }
3201
3202         /*
3203          * Fixup the lock to be given to the client.
3204          */
3205         lock_res_and_lock(new_lock);
3206         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3207          * possible blocking AST. */
3208         while (new_lock->l_readers > 0) {
3209                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3210                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3211                 new_lock->l_readers--;
3212         }
3213         while (new_lock->l_writers > 0) {
3214                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3215                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3216                 new_lock->l_writers--;
3217         }
3218
3219         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3220         new_lock->l_blocking_ast = lock->l_blocking_ast;
3221         new_lock->l_completion_ast = lock->l_completion_ast;
3222         new_lock->l_remote_handle = lock->l_remote_handle;
3223         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3224
3225         unlock_res_and_lock(new_lock);
3226
3227         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3228                      &new_lock->l_remote_handle,
3229                      &new_lock->l_exp_hash);
3230
3231         LDLM_LOCK_RELEASE(new_lock);
3232         lh->mlh_reg_lh.cookie = 0;
3233
3234         RETURN(ELDLM_LOCK_REPLACED);
3235 }
3236
3237 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3238                                     struct ldlm_lock *new_lock,
3239                                     struct ldlm_lock **old_lock,
3240                                     struct mdt_lock_handle *lh)
3241 {
3242         struct ptlrpc_request  *req = mdt_info_req(info);
3243         struct obd_export      *exp = req->rq_export;
3244         struct lustre_handle    remote_hdl;
3245         struct ldlm_request    *dlmreq;
3246         struct ldlm_lock       *lock;
3247
3248         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3249                 return;
3250
3251         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3252         remote_hdl = dlmreq->lock_handle[0];
3253
3254         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3255         if (lock) {
3256                 if (lock != new_lock) {
3257                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3258                         lh->mlh_reg_mode = lock->l_granted_mode;
3259
3260                         LDLM_DEBUG(lock, "Restoring lock cookie");
3261                         DEBUG_REQ(D_DLMTRACE, req,
3262                                   "restoring lock cookie "LPX64,
3263                                   lh->mlh_reg_lh.cookie);
3264                         if (old_lock)
3265                                 *old_lock = LDLM_LOCK_GET(lock);
3266                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3267                         return;
3268                 }
3269
3270                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3271         }
3272
3273         /*
3274          * If the xid matches, then we know this is a resent request, and allow
3275          * it. (It's probably an OPEN, for which we don't send a lock.
3276          */
3277         if (req_xid_is_last(req))
3278                 return;
3279
3280         /*
3281          * This remote handle isn't enqueued, so we never received or processed
3282          * this request.  Clear MSG_RESENT, because it can be handled like any
3283          * normal request now.
3284          */
3285         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3286
3287         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3288                   remote_hdl.cookie);
3289 }
3290
3291 static int mdt_intent_getattr(enum mdt_it_code opcode,
3292                               struct mdt_thread_info *info,
3293                               struct ldlm_lock **lockp,
3294                               int flags)
3295 {
3296         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3297         struct ldlm_lock       *new_lock = NULL;
3298         __u64                   child_bits;
3299         struct ldlm_reply      *ldlm_rep;
3300         struct ptlrpc_request  *req;
3301         struct mdt_body        *reqbody;
3302         struct mdt_body        *repbody;
3303         int                     rc;
3304         ENTRY;
3305
3306         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3307         LASSERT(reqbody);
3308
3309         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3310         LASSERT(repbody);
3311
3312         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3313         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3314         repbody->eadatasize = 0;
3315         repbody->aclsize = 0;
3316
3317         switch (opcode) {
3318         case MDT_IT_LOOKUP:
3319                 child_bits = MDS_INODELOCK_LOOKUP;
3320                 break;
3321         case MDT_IT_GETATTR:
3322                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3323                 break;
3324         default:
3325                 CERROR("Unhandled till now");
3326                 GOTO(out_shrink, rc = -EINVAL);
3327         }
3328
3329         rc = mdt_init_ucred(info, reqbody);
3330         if (rc)
3331                 GOTO(out_shrink, rc);
3332
3333         req = info->mti_pill->rc_req;
3334         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3335         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3336
3337         /* Get lock from request for possible resent case. */
3338         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3339
3340         ldlm_rep->lock_policy_res2 =
3341                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3342
3343         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3344                 ldlm_rep->lock_policy_res2 = 0;
3345         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3346             ldlm_rep->lock_policy_res2) {
3347                 lhc->mlh_reg_lh.cookie = 0ull;
3348                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3349         }
3350
3351         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3352         EXIT;
3353 out_ucred:
3354         mdt_exit_ucred(info);
3355 out_shrink:
3356         mdt_shrink_reply(info);
3357         return rc;
3358 }
3359
3360 static int mdt_intent_reint(enum mdt_it_code opcode,
3361                             struct mdt_thread_info *info,
3362                             struct ldlm_lock **lockp,
3363                             int flags)
3364 {
3365         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3366         struct ldlm_reply      *rep = NULL;
3367         long                    opc;
3368         int                     rc;
3369
3370         static const struct req_format *intent_fmts[REINT_MAX] = {
3371                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3372                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3373         };
3374
3375         ENTRY;
3376
3377         opc = mdt_reint_opcode(info, intent_fmts);
3378         if (opc < 0)
3379                 RETURN(opc);
3380
3381         if (mdt_it_flavor[opcode].it_reint != opc) {
3382                 CERROR("Reint code %ld doesn't match intent: %d\n",
3383                        opc, opcode);
3384                 RETURN(err_serious(-EPROTO));
3385         }
3386
3387         /* Get lock from request for possible resent case. */
3388         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3389
3390         rc = mdt_reint_internal(info, lhc, opc);
3391
3392         /* Check whether the reply has been packed successfully. */
3393         if (mdt_info_req(info)->rq_repmsg != NULL)
3394                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3395         if (rep == NULL)
3396                 RETURN(err_serious(-EFAULT));
3397
3398         /* MDC expects this in any case */
3399         if (rc != 0)
3400                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3401
3402         /* Cross-ref case, the lock should be returned to the client */
3403         if (rc == -EREMOTE) {
3404                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3405                 rep->lock_policy_res2 = 0;
3406                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3407                 RETURN(rc);
3408         }
3409         rep->lock_policy_res2 = clear_serious(rc);
3410
3411         if (rc == -ENOTCONN || rc == -ENODEV ||
3412             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3413                 /*
3414                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3415                  * will be returned by rq_status, and client at ptlrpc layer
3416                  * will detect this, then disconnect, reconnect the import
3417                  * immediately, instead of impacting the following the rpc.
3418                  */
3419                 lhc->mlh_reg_lh.cookie = 0ull;
3420                 RETURN(rc);
3421         } else {
3422                 /*
3423                  * For other cases, the error will be returned by intent.
3424                  * and client will retrieve the result from intent.
3425                  */
3426                  /*
3427                   * FIXME: when open lock is finished, that should be
3428                   * checked here.
3429                   */
3430                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3431                         LASSERTF(rc == 0, "Error occurred but lock handle "
3432                                  "is still in use\n");
3433                         rep->lock_policy_res2 = 0;
3434                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3435                         RETURN(rc);
3436                 } else {
3437                         lhc->mlh_reg_lh.cookie = 0ull;
3438                         RETURN(ELDLM_LOCK_ABORTED);
3439                 }
3440         }
3441 }
3442
3443 static int mdt_intent_code(long itcode)
3444 {
3445         int rc;
3446
3447         switch(itcode) {
3448         case IT_OPEN:
3449                 rc = MDT_IT_OPEN;
3450                 break;
3451         case IT_OPEN|IT_CREAT:
3452                 rc = MDT_IT_OCREAT;
3453                 break;
3454         case IT_CREAT:
3455                 rc = MDT_IT_CREATE;
3456                 break;
3457         case IT_READDIR:
3458                 rc = MDT_IT_READDIR;
3459                 break;
3460         case IT_GETATTR:
3461                 rc = MDT_IT_GETATTR;
3462                 break;
3463         case IT_LOOKUP:
3464                 rc = MDT_IT_LOOKUP;
3465                 break;
3466         case IT_UNLINK:
3467                 rc = MDT_IT_UNLINK;
3468                 break;
3469         case IT_TRUNC:
3470                 rc = MDT_IT_TRUNC;
3471                 break;
3472         case IT_GETXATTR:
3473                 rc = MDT_IT_GETXATTR;
3474                 break;
3475         default:
3476                 CERROR("Unknown intent opcode: %ld\n", itcode);
3477                 rc = -EINVAL;
3478                 break;
3479         }
3480         return rc;
3481 }
3482
3483 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3484                           struct ldlm_lock **lockp, int flags)
3485 {
3486         struct req_capsule   *pill;
3487         struct mdt_it_flavor *flv;
3488         int opc;
3489         int rc;
3490         ENTRY;
3491
3492         opc = mdt_intent_code(itopc);
3493         if (opc < 0)
3494                 RETURN(-EINVAL);
3495
3496         pill = info->mti_pill;
3497         flv  = &mdt_it_flavor[opc];
3498
3499         if (flv->it_fmt != NULL)
3500                 req_capsule_extend(pill, flv->it_fmt);
3501
3502         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3503         if (rc == 0) {
3504                 struct ptlrpc_request *req = mdt_info_req(info);
3505                 if (flv->it_flags & MUTABOR &&
3506                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3507                         RETURN(-EROFS);
3508         }
3509         if (rc == 0 && flv->it_act != NULL) {
3510                 /* execute policy */
3511                 rc = flv->it_act(opc, info, lockp, flags);
3512         } else {
3513                 rc = -EOPNOTSUPP;
3514         }
3515         RETURN(rc);
3516 }
3517
3518 static int mdt_intent_policy(struct ldlm_namespace *ns,
3519                              struct ldlm_lock **lockp, void *req_cookie,
3520                              ldlm_mode_t mode, int flags, void *data)
3521 {
3522         struct mdt_thread_info *info;
3523         struct ptlrpc_request  *req  =  req_cookie;
3524         struct ldlm_intent     *it;
3525         struct req_capsule     *pill;
3526         int rc;
3527
3528         ENTRY;
3529
3530         LASSERT(req != NULL);
3531
3532         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3533                                   &mdt_thread_key);
3534         LASSERT(info != NULL);
3535         pill = info->mti_pill;
3536         LASSERT(pill->rc_req == req);
3537
3538         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3539                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3540                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3541                 if (it != NULL) {
3542                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3543                         if (rc == 0)
3544                                 rc = ELDLM_OK;
3545
3546                         /* Lock without inodebits makes no sense and will oops
3547                          * later in ldlm. Let's check it now to see if we have
3548                          * ibits corrupted somewhere in mdt_intent_opc().
3549                          * The case for client miss to set ibits has been
3550                          * processed by others. */
3551                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3552                                         lr_type == LDLM_IBITS,
3553                                      info->mti_dlm_req->lock_desc.\
3554                                         l_policy_data.l_inodebits.bits != 0));
3555                 } else
3556                         rc = err_serious(-EFAULT);
3557         } else {
3558                 /* No intent was provided */
3559                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3560                 rc = req_capsule_server_pack(pill);
3561                 if (rc)
3562                         rc = err_serious(rc);
3563         }
3564         RETURN(rc);
3565 }
3566
3567 static int mdt_seq_fini(const struct lu_env *env,
3568                         struct mdt_device *m)
3569 {
3570         struct md_site *ms = mdt_md_site(m);
3571         ENTRY;
3572
3573         if (ms != NULL) {
3574                 if (ms->ms_server_seq) {
3575                         seq_server_fini(ms->ms_server_seq, env);
3576                         OBD_FREE_PTR(ms->ms_server_seq);
3577                         ms->ms_server_seq = NULL;
3578         }
3579
3580                 if (ms->ms_control_seq) {
3581                         seq_server_fini(ms->ms_control_seq, env);
3582                         OBD_FREE_PTR(ms->ms_control_seq);
3583                         ms->ms_control_seq = NULL;
3584         }
3585
3586                 if (ms->ms_client_seq) {
3587                         seq_client_fini(ms->ms_client_seq);
3588                         OBD_FREE_PTR(ms->ms_client_seq);
3589                         ms->ms_client_seq = NULL;
3590                 }
3591         }
3592
3593         RETURN(0);
3594 }
3595
3596 static int mdt_seq_init(const struct lu_env *env,
3597                         const char *uuid,
3598                         struct mdt_device *m)
3599 {
3600         struct md_site *ms;
3601         char *prefix;
3602         int rc;
3603         ENTRY;
3604
3605         ms = mdt_md_site(m);
3606
3607         /*
3608          * This is sequence-controller node. Init seq-controller server on local
3609          * MDT.
3610          */
3611         if (ms->ms_node_id == 0) {
3612                 LASSERT(ms->ms_control_seq == NULL);
3613
3614                 OBD_ALLOC_PTR(ms->ms_control_seq);
3615                 if (ms->ms_control_seq == NULL)
3616                         RETURN(-ENOMEM);
3617
3618                 rc = seq_server_init(ms->ms_control_seq,
3619                                      m->mdt_bottom, uuid,
3620                                      LUSTRE_SEQ_CONTROLLER,
3621                                      ms,
3622                                      env);
3623
3624                 if (rc)
3625                         GOTO(out_seq_fini, rc);
3626
3627                 OBD_ALLOC_PTR(ms->ms_client_seq);
3628                 if (ms->ms_client_seq == NULL)
3629                         GOTO(out_seq_fini, rc = -ENOMEM);
3630
3631                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3632                 if (prefix == NULL) {
3633                         OBD_FREE_PTR(ms->ms_client_seq);
3634                         GOTO(out_seq_fini, rc = -ENOMEM);
3635                 }
3636
3637                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3638                          uuid);
3639
3640                 /*
3641                  * Init seq-controller client after seq-controller server is
3642                  * ready. Pass ms->ms_control_seq to it for direct talking.
3643                  */
3644                 rc = seq_client_init(ms->ms_client_seq, NULL,
3645                                      LUSTRE_SEQ_METADATA, prefix,
3646                                      ms->ms_control_seq);
3647                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3648
3649                 if (rc)
3650                         GOTO(out_seq_fini, rc);
3651         }
3652
3653         /* Init seq-server on local MDT */
3654         LASSERT(ms->ms_server_seq == NULL);
3655
3656         OBD_ALLOC_PTR(ms->ms_server_seq);
3657         if (ms->ms_server_seq == NULL)
3658                 GOTO(out_seq_fini, rc = -ENOMEM);
3659
3660         rc = seq_server_init(ms->ms_server_seq,
3661                              m->mdt_bottom, uuid,
3662                              LUSTRE_SEQ_SERVER,
3663                              ms,
3664                              env);
3665         if (rc)
3666                 GOTO(out_seq_fini, rc = -ENOMEM);
3667
3668         /* Assign seq-controller client to local seq-server. */
3669         if (ms->ms_node_id == 0) {
3670                 LASSERT(ms->ms_client_seq != NULL);
3671
3672                 rc = seq_server_set_cli(ms->ms_server_seq,
3673                                         ms->ms_client_seq,
3674                                         env);
3675         }
3676
3677         EXIT;
3678 out_seq_fini:
3679         if (rc)
3680                 mdt_seq_fini(env, m);
3681
3682         return rc;
3683 }
3684 /*
3685  * Init client sequence manager which is used by local MDS to talk to sequence
3686  * controller on remote node.
3687  */
3688 static int mdt_seq_init_cli(const struct lu_env *env,
3689                             struct mdt_device *m,
3690                             struct lustre_cfg *cfg)
3691 {
3692         struct md_site    *ms = mdt_md_site(m);
3693         struct obd_device *mdc;
3694         struct obd_uuid   *uuidp, *mdcuuidp;
3695         char              *uuid_str, *mdc_uuid_str;
3696         int                rc;
3697         int                index;
3698         struct mdt_thread_info *info;
3699         char *p, *index_string = lustre_cfg_string(cfg, 2);
3700         ENTRY;
3701
3702         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3703         uuidp = &info->mti_u.uuid[0];
3704         mdcuuidp = &info->mti_u.uuid[1];
3705
3706         LASSERT(index_string);
3707
3708         index = simple_strtol(index_string, &p, 10);
3709         if (*p) {
3710                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3711                 RETURN(-EINVAL);
3712         }
3713
3714         /* check if this is adding the first MDC and controller is not yet
3715          * initialized. */
3716         if (index != 0 || ms->ms_client_seq)
3717                 RETURN(0);
3718
3719         uuid_str = lustre_cfg_string(cfg, 1);
3720         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3721         obd_str2uuid(uuidp, uuid_str);
3722         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3723
3724         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3725         if (!mdc) {
3726                 CERROR("can't find controller MDC by uuid %s\n",
3727                        uuid_str);
3728                 rc = -ENOENT;
3729         } else if (!mdc->obd_set_up) {
3730                 CERROR("target %s not set up\n", mdc->obd_name);
3731                 rc = -EINVAL;
3732         } else {
3733                 LASSERT(ms->ms_control_exp);
3734                 OBD_ALLOC_PTR(ms->ms_client_seq);
3735                 if (ms->ms_client_seq != NULL) {
3736                         char *prefix;
3737
3738                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3739                         if (!prefix)
3740                                 RETURN(-ENOMEM);
3741
3742                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3743                                  mdc->obd_name);
3744
3745                         rc = seq_client_init(ms->ms_client_seq,
3746                                              ms->ms_control_exp,
3747                                              LUSTRE_SEQ_METADATA,
3748                                              prefix, NULL);
3749                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3750                 } else
3751                         rc = -ENOMEM;
3752
3753                 if (rc)
3754                         RETURN(rc);
3755
3756                 LASSERT(ms->ms_server_seq != NULL);
3757                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3758                                         env);
3759         }
3760
3761         RETURN(rc);
3762 }
3763
3764 static void mdt_seq_fini_cli(struct mdt_device *m)
3765 {
3766         struct md_site *ms;
3767
3768         ENTRY;
3769
3770         ms = mdt_md_site(m);
3771
3772         if (ms != NULL) {
3773                 if (ms->ms_server_seq)
3774                         seq_server_set_cli(ms->ms_server_seq,
3775                                    NULL, NULL);
3776
3777                 if (ms->ms_control_exp) {
3778                         class_export_put(ms->ms_control_exp);
3779                         ms->ms_control_exp = NULL;
3780                 }
3781         }
3782         EXIT;
3783 }
3784
3785 /*
3786  * FLD wrappers
3787  */
3788 static int mdt_fld_fini(const struct lu_env *env,
3789                         struct mdt_device *m)
3790 {
3791         struct md_site *ms = mdt_md_site(m);
3792         ENTRY;
3793
3794         if (ms && ms->ms_server_fld) {
3795                 fld_server_fini(ms->ms_server_fld, env);
3796                 OBD_FREE_PTR(ms->ms_server_fld);
3797                 ms->ms_server_fld = NULL;
3798         }
3799
3800         RETURN(0);
3801 }
3802
3803 static int mdt_fld_init(const struct lu_env *env,
3804                         const char *uuid,
3805                         struct mdt_device *m)
3806 {
3807         struct md_site *ms;
3808         int rc;
3809         ENTRY;
3810
3811         ms = mdt_md_site(m);
3812
3813         OBD_ALLOC_PTR(ms->ms_server_fld);
3814         if (ms->ms_server_fld == NULL)
3815                 RETURN(rc = -ENOMEM);
3816
3817         rc = fld_server_init(ms->ms_server_fld,
3818                              m->mdt_bottom, uuid,
3819                              env, ms->ms_node_id);
3820         if (rc) {
3821                 OBD_FREE_PTR(ms->ms_server_fld);
3822                 ms->ms_server_fld = NULL;
3823                 RETURN(rc);
3824         }
3825
3826         RETURN(0);
3827 }
3828
3829 /* device init/fini methods */
3830 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3831 {
3832         ENTRY;
3833         if (m->mdt_regular_service != NULL) {
3834                 ptlrpc_unregister_service(m->mdt_regular_service);
3835                 m->mdt_regular_service = NULL;
3836         }
3837         if (m->mdt_readpage_service != NULL) {
3838                 ptlrpc_unregister_service(m->mdt_readpage_service);
3839                 m->mdt_readpage_service = NULL;
3840         }
3841         if (m->mdt_xmds_service != NULL) {
3842                 ptlrpc_unregister_service(m->mdt_xmds_service);
3843                 m->mdt_xmds_service = NULL;
3844         }
3845         if (m->mdt_setattr_service != NULL) {
3846                 ptlrpc_unregister_service(m->mdt_setattr_service);
3847                 m->mdt_setattr_service = NULL;
3848         }
3849         if (m->mdt_mdsc_service != NULL) {
3850                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3851                 m->mdt_mdsc_service = NULL;
3852         }
3853         if (m->mdt_mdss_service != NULL) {
3854                 ptlrpc_unregister_service(m->mdt_mdss_service);
3855                 m->mdt_mdss_service = NULL;
3856         }
3857         if (m->mdt_dtss_service != NULL) {
3858                 ptlrpc_unregister_service(m->mdt_dtss_service);
3859                 m->mdt_dtss_service = NULL;
3860         }
3861         if (m->mdt_fld_service != NULL) {
3862                 ptlrpc_unregister_service(m->mdt_fld_service);
3863                 m->mdt_fld_service = NULL;
3864         }
3865         EXIT;
3866 }
3867
3868 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3869 {
3870         int rc;
3871         static struct ptlrpc_service_conf conf;
3872         cfs_proc_dir_entry_t *procfs_entry;
3873         ENTRY;
3874
3875         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3876
3877         conf = (typeof(conf)) {
3878                 .psc_nbufs           = MDS_NBUFS,
3879                 .psc_bufsize         = MDS_BUFSIZE,
3880                 .psc_max_req_size    = MDS_MAXREQSIZE,
3881                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3882                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3883                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3884                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3885                 /*
3886                  * We'd like to have a mechanism to set this on a per-device
3887                  * basis, but alas...
3888                  */
3889                 .psc_min_threads     = mdt_min_threads,
3890                 .psc_max_threads     = mdt_max_threads,
3891                 .psc_ctx_tags        = LCT_MD_THREAD
3892         };
3893
3894         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3895         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3896                            "mdt_ldlm_client", m->mdt_ldlm_client);
3897
3898         m->mdt_regular_service =
3899                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3900                                      procfs_entry, target_print_req,
3901                                      LUSTRE_MDT_NAME);
3902         if (m->mdt_regular_service == NULL)
3903                 RETURN(-ENOMEM);
3904
3905         rc = ptlrpc_start_threads(m->mdt_regular_service);
3906         if (rc)
3907                 GOTO(err_mdt_svc, rc);
3908
3909         /*
3910          * readpage service configuration. Parameters have to be adjusted,
3911          * ideally.
3912          */
3913         conf = (typeof(conf)) {
3914                 .psc_nbufs           = MDS_NBUFS,
3915                 .psc_bufsize         = MDS_BUFSIZE,
3916                 .psc_max_req_size    = MDS_MAXREQSIZE,
3917                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3918                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3919                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3920                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3921                 .psc_min_threads     = mdt_min_threads,
3922                 .psc_max_threads     = mdt_max_threads,
3923                 .psc_ctx_tags        = LCT_MD_THREAD
3924         };
3925         m->mdt_readpage_service =
3926                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3927                                      LUSTRE_MDT_NAME "_readpage",
3928                                      procfs_entry, target_print_req,"mdt_rdpg");
3929
3930         if (m->mdt_readpage_service == NULL) {
3931                 CERROR("failed to start readpage service\n");
3932                 GOTO(err_mdt_svc, rc = -ENOMEM);
3933         }
3934
3935         rc = ptlrpc_start_threads(m->mdt_readpage_service);
3936
3937         /*
3938          * setattr service configuration.
3939          */
3940         conf = (typeof(conf)) {
3941                 .psc_nbufs           = MDS_NBUFS,
3942                 .psc_bufsize         = MDS_BUFSIZE,
3943                 .psc_max_req_size    = MDS_MAXREQSIZE,
3944                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3945                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3946                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3947                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3948                 .psc_min_threads     = mdt_min_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(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     = mdt_min_threads,
3979                 .psc_max_threads     = mdt_max_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(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     = mdt_min_threads,
4008                 .psc_max_threads     = mdt_max_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(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     = mdt_min_threads,
4040                 .psc_max_threads     = mdt_max_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(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     = mdt_min_threads,
4067                 .psc_max_threads     = mdt_max_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(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     = mdt_min_threads,
4097                 .psc_max_threads     = mdt_max_threads,
4098                 .psc_ctx_tags        = LCT_MD_THREAD
4099         };
4100         m->mdt_xmds_service =
4101                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4102                                      LUSTRE_MDT_NAME "_mds",
4103                                      procfs_entry, target_print_req,"mdt_xmds");
4104
4105         if (m->mdt_xmds_service == NULL) {
4106                 CERROR("failed to start xmds service\n");
4107                 GOTO(err_mdt_svc, rc = -ENOMEM);
4108         }
4109
4110         rc = ptlrpc_start_threads(m->mdt_xmds_service);
4111         if (rc)
4112                 GOTO(err_mdt_svc, rc);
4113
4114         EXIT;
4115 err_mdt_svc:
4116         if (rc)
4117                 mdt_stop_ptlrpc_service(m);
4118
4119         return rc;
4120 }
4121
4122 static void mdt_stack_fini(const struct lu_env *env,
4123                            struct mdt_device *m, struct lu_device *top)
4124 {
4125         struct obd_device       *obd = mdt2obd_dev(m);
4126         struct lustre_cfg_bufs  *bufs;
4127         struct lustre_cfg       *lcfg;
4128         struct mdt_thread_info  *info;
4129         char flags[3]="";
4130         ENTRY;
4131
4132         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4133         LASSERT(info != NULL);
4134
4135         bufs = &info->mti_u.bufs;
4136         /* process cleanup, pass mdt obd name to get obd umount flags */
4137         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4138         if (obd->obd_force)
4139                 strcat(flags, "F");
4140         if (obd->obd_fail)
4141                 strcat(flags, "A");
4142         lustre_cfg_bufs_set_string(bufs, 1, flags);
4143         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4144         if (!lcfg) {
4145                 CERROR("Cannot alloc lcfg!\n");
4146                 return;
4147         }
4148
4149         LASSERT(top);
4150         top->ld_ops->ldo_process_config(env, top, lcfg);
4151         lustre_cfg_free(lcfg);
4152
4153         lu_stack_fini(env, top);
4154         m->mdt_child = NULL;
4155         m->mdt_bottom = NULL;
4156 }
4157
4158 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4159                                          const char *typename,
4160                                          struct lu_device *child,
4161                                          struct lustre_cfg *cfg)
4162 {
4163         const char            *dev = lustre_cfg_string(cfg, 0);
4164         struct obd_type       *type;
4165         struct lu_device_type *ldt;
4166         struct lu_device      *d;
4167         int rc;
4168         ENTRY;
4169
4170         /* find the type */
4171         type = class_get_type(typename);
4172         if (!type) {
4173                 CERROR("Unknown type: '%s'\n", typename);
4174                 GOTO(out, rc = -ENODEV);
4175         }
4176
4177         rc = lu_env_refill((struct lu_env *)env);
4178         if (rc != 0) {
4179                 CERROR("Failure to refill session: '%d'\n", rc);
4180                 GOTO(out_type, rc);
4181         }
4182
4183         ldt = type->typ_lu;
4184         if (ldt == NULL) {
4185                 CERROR("type: '%s'\n", typename);
4186                 GOTO(out_type, rc = -EINVAL);
4187         }
4188
4189         ldt->ldt_obd_type = type;
4190         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4191         if (IS_ERR(d)) {
4192                 CERROR("Cannot allocate device: '%s'\n", typename);
4193                 GOTO(out_type, rc = -ENODEV);
4194         }
4195
4196         LASSERT(child->ld_site);
4197         d->ld_site = child->ld_site;
4198
4199         type->typ_refcnt++;
4200         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4201         if (rc) {
4202                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4203                 GOTO(out_alloc, rc);
4204         }
4205         lu_device_get(d);
4206         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4207
4208         RETURN(d);
4209
4210 out_alloc:
4211         ldt->ldt_ops->ldto_device_free(env, d);
4212         type->typ_refcnt--;
4213 out_type:
4214         class_put_type(type);
4215 out:
4216         return ERR_PTR(rc);
4217 }
4218
4219 static int mdt_stack_init(struct lu_env *env,
4220                           struct mdt_device *m,
4221                           struct lustre_cfg *cfg,
4222                           struct lustre_mount_info  *lmi)
4223 {
4224         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4225         struct lu_device  *tmp;
4226         struct md_device  *md;
4227         struct lu_device  *child_lu_dev;
4228         int rc;
4229         ENTRY;
4230
4231         /* init the stack */
4232         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4233         if (IS_ERR(tmp)) {
4234                 RETURN(PTR_ERR(tmp));
4235         }
4236         m->mdt_bottom = lu2dt_dev(tmp);
4237         d = tmp;
4238         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4239         if (IS_ERR(tmp)) {
4240                 GOTO(out, rc = PTR_ERR(tmp));
4241         }
4242         d = tmp;
4243         md = lu2md_dev(d);
4244
4245         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4246         if (IS_ERR(tmp)) {
4247                 GOTO(out, rc = PTR_ERR(tmp));
4248         }
4249         d = tmp;
4250         /*set mdd upcall device*/
4251         md_upcall_dev_set(md, lu2md_dev(d));
4252
4253         md = lu2md_dev(d);
4254         /*set cmm upcall device*/
4255         md_upcall_dev_set(md, &m->mdt_md_dev);
4256
4257         m->mdt_child = lu2md_dev(d);
4258
4259         /* process setup config */
4260         tmp = &m->mdt_md_dev.md_lu_dev;
4261         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4262         if (rc)
4263                 GOTO(out, rc);
4264
4265         /* initialize local objects */
4266         child_lu_dev = &m->mdt_child->md_lu_dev;
4267
4268         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4269                                                &m->mdt_md_dev.md_lu_dev,
4270                                                child_lu_dev);
4271 out:
4272         /* fini from last known good lu_device */
4273         if (rc)
4274                 mdt_stack_fini(env, m, d);
4275
4276         return rc;
4277 }
4278
4279 /**
4280  * setup CONFIG_ORIG context, used to access local config log.
4281  * this may need to be rewrite as part of llog rewrite for lu-api.
4282  */
4283 static int mdt_obd_llog_setup(struct obd_device *obd,
4284                               struct lustre_sb_info *lsi)
4285 {
4286         int     rc;
4287
4288         LASSERT(obd->obd_fsops == NULL);
4289
4290         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4291         if (IS_ERR(obd->obd_fsops))
4292                 return PTR_ERR(obd->obd_fsops);
4293
4294         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4295         if (rc) {
4296                 fsfilt_put_ops(obd->obd_fsops);
4297                 return rc;
4298         }
4299
4300         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4301         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4302         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4303         obd->obd_lvfs_ctxt.fs = get_ds();
4304
4305         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4306                         0, NULL, &llog_lvfs_ops);
4307         if (rc) {
4308                 CERROR("llog_setup() failed: %d\n", rc);
4309                 fsfilt_put_ops(obd->obd_fsops);
4310         }
4311
4312         return rc;
4313 }
4314
4315 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4316 {
4317         struct llog_ctxt *ctxt;
4318
4319         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4320         if (ctxt)
4321                 llog_cleanup(ctxt);
4322
4323         if (obd->obd_fsops) {
4324                 fsfilt_put_ops(obd->obd_fsops);
4325                 obd->obd_fsops = NULL;
4326         }
4327 }
4328
4329 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4330 {
4331         struct md_device  *next = m->mdt_child;
4332         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4333         struct lu_site    *ls   = d->ld_site;
4334         struct obd_device *obd = mdt2obd_dev(m);
4335         ENTRY;
4336
4337         target_recovery_fini(obd);
4338
4339         ping_evictor_stop();
4340
4341         mdt_stop_ptlrpc_service(m);
4342         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4343         mdt_obd_llog_cleanup(obd);
4344         obd_exports_barrier(obd);
4345         obd_zombie_barrier();
4346 #ifdef HAVE_QUOTA_SUPPORT
4347         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4348 #endif
4349         lut_fini(env, &m->mdt_lut);
4350         mdt_fs_cleanup(env, m);
4351         upcall_cache_cleanup(m->mdt_identity_cache);
4352         m->mdt_identity_cache = NULL;
4353
4354         if (m->mdt_namespace != NULL) {
4355                 ldlm_namespace_free(m->mdt_namespace, NULL,
4356                                     d->ld_obd->obd_force);
4357                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4358         }
4359
4360         cfs_free_nidlist(&m->mdt_nosquash_nids);
4361         if (m->mdt_nosquash_str) {
4362                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4363                 m->mdt_nosquash_str = NULL;
4364                 m->mdt_nosquash_strlen = 0;
4365         }
4366
4367         mdt_seq_fini(env, m);
4368         mdt_seq_fini_cli(m);
4369         mdt_fld_fini(env, m);
4370         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4371
4372         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4373         cfs_timer_disarm(&m->mdt_ck_timer);
4374         mdt_ck_thread_stop(m);
4375
4376         /*
4377          * Finish the stack
4378          */
4379         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4380
4381         lprocfs_free_per_client_stats(obd);
4382         lprocfs_free_obd_stats(obd);
4383         mdt_procfs_fini(m);
4384
4385         if (ls) {
4386                 struct md_site *mite;
4387
4388                 lu_site_fini(ls);
4389                 mite = lu_site2md(ls);
4390                 OBD_FREE_PTR(mite);
4391                 d->ld_site = NULL;
4392         }
4393         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4394
4395         EXIT;
4396 }
4397
4398 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4399 {
4400         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4401         struct sptlrpc_rule_set  tmp_rset;
4402         int                      rc;
4403
4404         sptlrpc_rule_set_init(&tmp_rset);
4405         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4406         if (rc) {
4407                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4408                        obd->obd_name, rc);
4409                 return rc;
4410         }
4411
4412         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4413
4414         cfs_write_lock(&m->mdt_sptlrpc_lock);
4415         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4416         m->mdt_sptlrpc_rset = tmp_rset;
4417         cfs_write_unlock(&m->mdt_sptlrpc_lock);
4418
4419         return 0;
4420 }
4421
4422 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4423 {
4424         char *p = options;
4425
4426         m->mdt_opts.mo_mds_capa = 1;
4427         m->mdt_opts.mo_oss_capa = 1;
4428 #ifdef CONFIG_FS_POSIX_ACL
4429         /* ACLs should be enabled by default (b=13829) */
4430         m->mdt_opts.mo_acl = 1;
4431         LCONSOLE_INFO("Enabling ACL\n");
4432 #else
4433         m->mdt_opts.mo_acl = 0;
4434         LCONSOLE_INFO("Disabling ACL\n");
4435 #endif
4436
4437         if (!options)
4438                 return;
4439
4440         while (*options) {
4441                 int len;
4442
4443                 while (*p && *p != ',')
4444                         p++;
4445
4446                 len = p - options;
4447                 if ((len == sizeof("user_xattr") - 1) &&
4448                     (memcmp(options, "user_xattr", len) == 0)) {
4449                         m->mdt_opts.mo_user_xattr = 1;
4450                         LCONSOLE_INFO("Enabling user_xattr\n");
4451                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4452                            (memcmp(options, "nouser_xattr", len) == 0)) {
4453                         m->mdt_opts.mo_user_xattr = 0;
4454                         LCONSOLE_INFO("Disabling user_xattr\n");
4455                 } else if ((len == sizeof("noacl") - 1) &&
4456                            (memcmp(options, "noacl", len) == 0)) {
4457                         m->mdt_opts.mo_acl = 0;
4458                         LCONSOLE_INFO("Disabling ACL\n");
4459                 }
4460
4461                 if (!*p)
4462                         break;
4463
4464                 options = ++p;
4465         }
4466 }
4467
4468 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4469
4470 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4471                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4472 {
4473         struct mdt_thread_info    *info;
4474         struct obd_device         *obd;
4475         const char                *dev = lustre_cfg_string(cfg, 0);
4476         const char                *num = lustre_cfg_string(cfg, 2);
4477         struct lustre_mount_info  *lmi = NULL;
4478         struct lustre_sb_info     *lsi;
4479         struct lustre_disk_data   *ldd;
4480         struct lu_site            *s;
4481         struct md_site            *mite;
4482         const char                *identity_upcall = "NONE";
4483 #ifdef HAVE_QUOTA_SUPPORT
4484         struct md_device          *next;
4485 #endif
4486         int                        rc;
4487         int                        node_id;
4488         ENTRY;
4489
4490         md_device_init(&m->mdt_md_dev, ldt);
4491         /*
4492          * Environment (env) might be missing mdt_thread_key values at that
4493          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4494          * mode.
4495          *
4496          * Usually device allocation path doesn't use module key values, but
4497          * mdt has to do a lot of work here, so allocate key value.
4498          */
4499         rc = lu_env_refill((struct lu_env *)env);
4500         if (rc != 0)
4501                 RETURN(rc);
4502
4503         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4504         LASSERT(info != NULL);
4505
4506         obd = class_name2obd(dev);
4507         LASSERT(obd != NULL);
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                 /* Read recovery timeouts */
4532                 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft)
4533                         obd->obd_recovery_timeout =
4534                                 lsi->lsi_lmd->lmd_recovery_time_soft;
4535
4536                 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard)
4537                         obd->obd_recovery_time_hard =
4538                                 lsi->lsi_lmd->lmd_recovery_time_hard;
4539         }
4540
4541         cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4542         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4543
4544         cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4545         m->mdt_opts.mo_compat_resname = 0;
4546         m->mdt_capa_timeout = CAPA_TIMEOUT;
4547         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4548         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4549         m->mdt_squash_uid = 0;
4550         m->mdt_squash_gid = 0;
4551         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4552         m->mdt_nosquash_str = NULL;
4553         m->mdt_nosquash_strlen = 0;
4554         cfs_init_rwsem(&m->mdt_squash_sem);
4555
4556         OBD_ALLOC_PTR(mite);
4557         if (mite == NULL)
4558                 GOTO(err_lmi, rc = -ENOMEM);
4559
4560         s = &mite->ms_lu;
4561
4562         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4563         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4564         /* set this lu_device to obd, because error handling need it */
4565         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4566
4567         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4568         if (rc) {
4569                 CERROR("Can't init lu_site, rc %d\n", rc);
4570                 GOTO(err_free_site, rc);
4571         }
4572
4573         rc = mdt_procfs_init(m, dev);
4574         if (rc) {
4575                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4576                 GOTO(err_fini_proc, rc);
4577         }
4578
4579         /* set server index */
4580         lu_site2md(s)->ms_node_id = node_id;
4581
4582         /* failover is the default
4583          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4584          * assumed whose ms_node_id == 0 XXX
4585          * */
4586         obd->obd_replayable = 1;
4587         /* No connection accepted until configurations will finish */
4588         obd->obd_no_conn = 1;
4589
4590         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4591                 char *str = lustre_cfg_string(cfg, 4);
4592                 if (strchr(str, 'n')) {
4593                         CWARN("%s: recovery disabled\n", obd->obd_name);
4594                         obd->obd_replayable = 0;
4595                 }
4596         }
4597
4598         /* init the stack */
4599         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4600         if (rc) {
4601                 CERROR("Can't init device stack, rc %d\n", rc);
4602                 GOTO(err_fini_proc, rc);
4603         }
4604
4605         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4606         if (rc)
4607                 GOTO(err_fini_stack, rc);
4608
4609         rc = mdt_fld_init(env, obd->obd_name, m);
4610         if (rc)
4611                 GOTO(err_lut, rc);
4612
4613         rc = mdt_seq_init(env, obd->obd_name, m);
4614         if (rc)
4615                 GOTO(err_fini_fld, rc);
4616
4617         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4618                  LUSTRE_MDT_NAME"-%p", m);
4619         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4620                                               LDLM_NAMESPACE_SERVER,
4621                                               LDLM_NAMESPACE_GREEDY,
4622                                               LDLM_NS_TYPE_MDT);
4623         if (m->mdt_namespace == NULL)
4624                 GOTO(err_fini_seq, rc = -ENOMEM);
4625
4626         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4627         /* set obd_namespace for compatibility with old code */
4628         obd->obd_namespace = m->mdt_namespace;
4629
4630         /* XXX: to support suppgid for ACL, we enable identity_upcall
4631          * by default, otherwise, maybe got unexpected -EACCESS. */
4632         if (m->mdt_opts.mo_acl)
4633                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4634
4635         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4636                                                   &mdt_identity_upcall_cache_ops);
4637         if (IS_ERR(m->mdt_identity_cache)) {
4638                 rc = PTR_ERR(m->mdt_identity_cache);
4639                 m->mdt_identity_cache = NULL;
4640                 GOTO(err_free_ns, rc);
4641         }
4642
4643         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4644
4645         rc = mdt_ck_thread_start(m);
4646         if (rc)
4647                 GOTO(err_free_ns, rc);
4648
4649         rc = mdt_fs_setup(env, m, obd, lsi);
4650         if (rc)
4651                 GOTO(err_capa, rc);
4652
4653         rc = mdt_obd_llog_setup(obd, lsi);
4654         if (rc)
4655                 GOTO(err_fs_cleanup, rc);
4656
4657         rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4658         if (rc)
4659                 GOTO(err_llog_cleanup, rc);
4660
4661         mdt_adapt_sptlrpc_conf(obd, 1);
4662
4663 #ifdef HAVE_QUOTA_SUPPORT
4664         next = m->mdt_child;
4665         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4666         if (rc)
4667                 GOTO(err_llog_cleanup, rc);
4668 #endif
4669
4670         server_put_mount_2(dev, lmi->lmi_mnt);
4671         lmi = NULL;
4672
4673         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4674
4675         rc = mdt_start_ptlrpc_service(m);
4676         if (rc)
4677                 GOTO(err_recovery, rc);
4678
4679         ping_evictor_start();
4680
4681         rc = lu_site_init_finish(s);
4682         if (rc)
4683                 GOTO(err_stop_service, rc);
4684
4685         if (obd->obd_recovering == 0)
4686                 mdt_postrecov(env, m);
4687
4688         mdt_init_capa_ctxt(env, m);
4689
4690         /* Reduce the initial timeout on an MDS because it doesn't need such
4691          * a long timeout as an OST does. Adaptive timeouts will adjust this
4692          * value appropriately. */
4693         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4694                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4695
4696         RETURN(0);
4697
4698 err_stop_service:
4699         ping_evictor_stop();
4700         mdt_stop_ptlrpc_service(m);
4701 err_recovery:
4702         target_recovery_fini(obd);
4703 #ifdef HAVE_QUOTA_SUPPORT
4704         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4705 #endif
4706 err_llog_cleanup:
4707         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4708         mdt_obd_llog_cleanup(obd);
4709 err_fs_cleanup:
4710         mdt_fs_cleanup(env, m);
4711 err_capa:
4712         cfs_timer_disarm(&m->mdt_ck_timer);
4713         mdt_ck_thread_stop(m);
4714 err_free_ns:
4715         upcall_cache_cleanup(m->mdt_identity_cache);
4716         m->mdt_identity_cache = NULL;
4717         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4718         obd->obd_namespace = m->mdt_namespace = NULL;
4719 err_fini_seq:
4720         mdt_seq_fini(env, m);
4721 err_fini_fld:
4722         mdt_fld_fini(env, m);
4723 err_lut:
4724         lut_fini(env, &m->mdt_lut);
4725 err_fini_stack:
4726         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4727 err_fini_proc:
4728         mdt_procfs_fini(m);
4729         lu_site_fini(s);
4730 err_free_site:
4731         OBD_FREE_PTR(mite);
4732 err_lmi:
4733         if (lmi)
4734                 server_put_mount_2(dev, lmi->lmi_mnt);
4735         return (rc);
4736 }
4737
4738 /* used by MGS to process specific configurations */
4739 static int mdt_process_config(const struct lu_env *env,
4740                               struct lu_device *d, struct lustre_cfg *cfg)
4741 {
4742         struct mdt_device *m = mdt_dev(d);
4743         struct md_device *md_next = m->mdt_child;
4744         struct lu_device *next = md2lu_dev(md_next);
4745         int rc = 0;
4746         ENTRY;
4747
4748         switch (cfg->lcfg_command) {
4749         case LCFG_PARAM: {
4750                 struct lprocfs_static_vars lvars;
4751                 struct obd_device *obd = d->ld_obd;
4752
4753                 /*
4754                  * For interoperability between 1.8 and 2.0,
4755                  * skip old "mdt.group_upcall" param.
4756                  */
4757                 {
4758                         char *param = lustre_cfg_string(cfg, 1);
4759                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4760                                 CWARN("For 1.8 interoperability, skip this"
4761                                        " mdt.group_upcall. It is obsolete\n");
4762                                 break;
4763                         }
4764                 }
4765
4766                 lprocfs_mdt_init_vars(&lvars);
4767                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4768                                               cfg, obd);
4769                 if (rc > 0 || rc == -ENOSYS)
4770                         /* we don't understand; pass it on */
4771                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4772                 break;
4773         }
4774         case LCFG_ADD_MDC:
4775                 /*
4776                  * Add mdc hook to get first MDT uuid and connect it to
4777                  * ls->controller to use for seq manager.
4778                  */
4779                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4780                 if (rc)
4781                         CERROR("Can't add mdc, rc %d\n", rc);
4782                 else
4783                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4784                 break;
4785         default:
4786                 /* others are passed further */
4787                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4788                 break;
4789         }
4790         RETURN(rc);
4791 }
4792
4793 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4794                                           const struct lu_object_header *hdr,
4795                                           struct lu_device *d)
4796 {
4797         struct mdt_object *mo;
4798
4799         ENTRY;
4800
4801         OBD_ALLOC_PTR(mo);
4802         if (mo != NULL) {
4803                 struct lu_object *o;
4804                 struct lu_object_header *h;
4805
4806                 o = &mo->mot_obj.mo_lu;
4807                 h = &mo->mot_header;
4808                 lu_object_header_init(h);
4809                 lu_object_init(o, h, d);
4810                 lu_object_add_top(h, o);
4811                 o->lo_ops = &mdt_obj_ops;
4812                 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4813                 RETURN(o);
4814         } else
4815                 RETURN(NULL);
4816 }
4817
4818 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4819                            const struct lu_object_conf *unused)
4820 {
4821         struct mdt_device *d = mdt_dev(o->lo_dev);
4822         struct lu_device  *under;
4823         struct lu_object  *below;
4824         int                rc = 0;
4825         ENTRY;
4826
4827         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4828                PFID(lu_object_fid(o)));
4829
4830         under = &d->mdt_child->md_lu_dev;
4831         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4832         if (below != NULL) {
4833                 lu_object_add(o, below);
4834         } else
4835                 rc = -ENOMEM;
4836
4837         RETURN(rc);
4838 }
4839
4840 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4841 {
4842         struct mdt_object *mo = mdt_obj(o);
4843         struct lu_object_header *h;
4844         ENTRY;
4845
4846         h = o->lo_header;
4847         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4848                PFID(lu_object_fid(o)));
4849
4850         lu_object_fini(o);
4851         lu_object_header_fini(h);
4852         OBD_FREE_PTR(mo);
4853         EXIT;
4854 }
4855
4856 static int mdt_object_print(const struct lu_env *env, void *cookie,
4857                             lu_printer_t p, const struct lu_object *o)
4858 {
4859         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4860         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4861                     "flags="LPX64", epochcount=%d, writecount=%d)",
4862                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4863                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4864 }
4865
4866 static const struct lu_device_operations mdt_lu_ops = {
4867         .ldo_object_alloc   = mdt_object_alloc,
4868         .ldo_process_config = mdt_process_config,
4869 };
4870
4871 static const struct lu_object_operations mdt_obj_ops = {
4872         .loo_object_init    = mdt_object_init,
4873         .loo_object_free    = mdt_object_free,
4874         .loo_object_print   = mdt_object_print
4875 };
4876
4877 static int mdt_obd_set_info_async(struct obd_export *exp,
4878                                   __u32 keylen, void *key,
4879                                   __u32 vallen, void *val,
4880                                   struct ptlrpc_request_set *set)
4881 {
4882         struct obd_device     *obd = exp->exp_obd;
4883         int                    rc;
4884         ENTRY;
4885
4886         LASSERT(obd);
4887
4888         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4889                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4890                 RETURN(rc);
4891         }
4892
4893         RETURN(0);
4894 }
4895
4896 /* mds_connect_internal */
4897 static int mdt_connect_internal(struct obd_export *exp,
4898                                 struct mdt_device *mdt,
4899                                 struct obd_connect_data *data)
4900 {
4901         if (data != NULL) {
4902                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4903                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4904
4905                 /* If no known bits (which should not happen, probably,
4906                    as everybody should support LOOKUP and UPDATE bits at least)
4907                    revert to compat mode with plain locks. */
4908                 if (!data->ocd_ibits_known &&
4909                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4910                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4911
4912                 if (!mdt->mdt_opts.mo_acl)
4913                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4914
4915                 if (!mdt->mdt_opts.mo_user_xattr)
4916                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4917
4918                 if (!mdt->mdt_som_conf)
4919                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4920
4921                 cfs_spin_lock(&exp->exp_lock);
4922                 exp->exp_connect_flags = data->ocd_connect_flags;
4923                 cfs_spin_unlock(&exp->exp_lock);
4924                 data->ocd_version = LUSTRE_VERSION_CODE;
4925                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4926         }
4927
4928 #if 0
4929         if (mdt->mdt_opts.mo_acl &&
4930             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4931                 CWARN("%s: MDS requires ACL support but client does not\n",
4932                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4933                 return -EBADE;
4934         }
4935 #endif
4936
4937         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4938                 CWARN("%s: MDS requires FID support, but client not\n",
4939                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4940                 return -EBADE;
4941         }
4942
4943         if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4944             !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4945                 CWARN("%s: MDS has SOM enabled, but client does not support "
4946                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4947                 return -EBADE;
4948         }
4949
4950         return 0;
4951 }
4952
4953 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4954                                      struct obd_export *exp,
4955                                      struct ptlrpc_request *req)
4956 {
4957         struct sptlrpc_flavor   flvr;
4958         int                     rc = 0;
4959
4960         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4961                 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4962                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4963                                              req->rq_sp_from,
4964                                              req->rq_peer.nid,
4965                                              &flvr);
4966                 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4967
4968                 cfs_spin_lock(&exp->exp_lock);
4969
4970                 exp->exp_sp_peer = req->rq_sp_from;
4971                 exp->exp_flvr = flvr;
4972
4973                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4974                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4975                         CERROR("unauthorized rpc flavor %x from %s, "
4976                                "expect %x\n", req->rq_flvr.sf_rpc,
4977                                libcfs_nid2str(req->rq_peer.nid),
4978                                exp->exp_flvr.sf_rpc);
4979                         rc = -EACCES;
4980                 }
4981
4982                 cfs_spin_unlock(&exp->exp_lock);
4983         } else {
4984                 if (exp->exp_sp_peer != req->rq_sp_from) {
4985                         CERROR("RPC source %s doesn't match %s\n",
4986                                sptlrpc_part2name(req->rq_sp_from),
4987                                sptlrpc_part2name(exp->exp_sp_peer));
4988                         rc = -EACCES;
4989                 } else {
4990                         rc = sptlrpc_target_export_check(exp, req);
4991                 }
4992         }
4993
4994         return rc;
4995 }
4996
4997 /* mds_connect copy */
4998 static int mdt_obd_connect(const struct lu_env *env,
4999                            struct obd_export **exp, struct obd_device *obd,
5000                            struct obd_uuid *cluuid,
5001                            struct obd_connect_data *data,
5002                            void *localdata)
5003 {
5004         struct mdt_thread_info *info;
5005         struct obd_export      *lexp;
5006         struct lustre_handle    conn = { 0 };
5007         struct mdt_device      *mdt;
5008         struct ptlrpc_request  *req;
5009         int                     rc;
5010         ENTRY;
5011
5012         LASSERT(env != NULL);
5013         if (!exp || !obd || !cluuid)
5014                 RETURN(-EINVAL);
5015
5016         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5017         req = info->mti_pill->rc_req;
5018         mdt = mdt_dev(obd->obd_lu_dev);
5019
5020         rc = class_connect(&conn, obd, cluuid);
5021         if (rc)
5022                 RETURN(rc);
5023
5024         lexp = class_conn2export(&conn);
5025         LASSERT(lexp != NULL);
5026
5027         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5028         if (rc)
5029                 GOTO(out, rc);
5030
5031         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5032                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5033
5034         rc = mdt_connect_internal(lexp, mdt, data);
5035         if (rc == 0) {
5036                 struct mdt_thread_info *mti;
5037                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5038                 LASSERT(lcd);
5039                 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5040                 LASSERT(mti != NULL);
5041                 mti->mti_exp = lexp;
5042                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5043                 rc = mdt_client_new(env, mdt);
5044                 if (rc == 0)
5045                         mdt_export_stats_init(obd, lexp, localdata);
5046         }
5047
5048 out:
5049         if (rc != 0) {
5050                 class_disconnect(lexp);
5051                 *exp = NULL;
5052         } else {
5053                 *exp = lexp;
5054         }
5055
5056         RETURN(rc);
5057 }
5058
5059 static int mdt_obd_reconnect(const struct lu_env *env,
5060                              struct obd_export *exp, struct obd_device *obd,
5061                              struct obd_uuid *cluuid,
5062                              struct obd_connect_data *data,
5063                              void *localdata)
5064 {
5065         struct mdt_thread_info *info;
5066         struct mdt_device      *mdt;
5067         struct ptlrpc_request  *req;
5068         int                     rc;
5069         ENTRY;
5070
5071         if (exp == NULL || obd == NULL || cluuid == NULL)
5072                 RETURN(-EINVAL);
5073
5074         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5075         req = info->mti_pill->rc_req;
5076         mdt = mdt_dev(obd->obd_lu_dev);
5077
5078         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5079         if (rc)
5080                 RETURN(rc);
5081
5082         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5083         if (rc == 0)
5084                 mdt_export_stats_init(obd, exp, localdata);
5085
5086         RETURN(rc);
5087 }
5088 static int mdt_export_cleanup(struct obd_export *exp)
5089 {
5090         struct mdt_export_data *med = &exp->exp_mdt_data;
5091         struct obd_device      *obd = exp->exp_obd;
5092         struct mdt_device      *mdt;
5093         struct mdt_thread_info *info;
5094         struct lu_env           env;
5095         CFS_LIST_HEAD(closing_list);
5096         struct mdt_file_data *mfd, *n;
5097         int rc = 0;
5098         ENTRY;
5099
5100         cfs_spin_lock(&med->med_open_lock);
5101         while (!cfs_list_empty(&med->med_open_head)) {
5102                 cfs_list_t *tmp = med->med_open_head.next;
5103                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5104
5105                 /* Remove mfd handle so it can't be found again.
5106                  * We are consuming the mfd_list reference here. */
5107                 class_handle_unhash(&mfd->mfd_handle);
5108                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5109         }
5110         cfs_spin_unlock(&med->med_open_lock);
5111         mdt = mdt_dev(obd->obd_lu_dev);
5112         LASSERT(mdt != NULL);
5113
5114         rc = lu_env_init(&env, LCT_MD_THREAD);
5115         if (rc)
5116                 RETURN(rc);
5117
5118         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5119         LASSERT(info != NULL);
5120         memset(info, 0, sizeof *info);
5121         info->mti_env = &env;
5122         info->mti_mdt = mdt;
5123         info->mti_exp = exp;
5124
5125         if (!cfs_list_empty(&closing_list)) {
5126                 struct md_attr *ma = &info->mti_attr;
5127                 int lmm_size;
5128                 int cookie_size;
5129
5130                 lmm_size = mdt->mdt_max_mdsize;
5131                 OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
5132                 if (ma->ma_lmm == NULL)
5133                         GOTO(out_lmm, rc = -ENOMEM);
5134
5135                 cookie_size = mdt->mdt_max_cookiesize;
5136                 OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
5137                 if (ma->ma_cookie == NULL)
5138                         GOTO(out_cookie, rc = -ENOMEM);
5139
5140                 /* Close any open files (which may also cause orphan unlinking). */
5141                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5142                         cfs_list_del_init(&mfd->mfd_list);
5143                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5144                         ma->ma_lmm_size = lmm_size;
5145                         ma->ma_cookie_size = cookie_size;
5146                         ma->ma_need = 0;
5147                         /* It is not for setattr, just tell MDD to send
5148                          * DESTROY RPC to OSS if needed */
5149                         ma->ma_valid = MA_FLAGS;
5150                         ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5151                         /* Don't unlink orphan on failover umount, LU-184 */
5152                         if (exp->exp_flags & OBD_OPT_FAILOVER)
5153                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5154                         mdt_mfd_close(info, mfd);
5155                 }
5156                 OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
5157                 ma->ma_cookie = NULL;
5158 out_cookie:
5159                 OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
5160                 ma->ma_lmm = NULL;
5161         }
5162 out_lmm:
5163         info->mti_mdt = NULL;
5164         /* cleanup client slot early */
5165         /* Do not erase record for recoverable client. */
5166         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5167                 mdt_client_del(&env, mdt);
5168         lu_env_fini(&env);
5169
5170         RETURN(rc);
5171 }
5172
5173 static int mdt_obd_disconnect(struct obd_export *exp)
5174 {
5175         int rc;
5176         ENTRY;
5177
5178         LASSERT(exp);
5179         class_export_get(exp);
5180
5181         rc = server_disconnect_export(exp);
5182         if (rc != 0)
5183                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5184
5185         rc = mdt_export_cleanup(exp);
5186         class_export_put(exp);
5187         RETURN(rc);
5188 }
5189
5190 /* FIXME: Can we avoid using these two interfaces? */
5191 static int mdt_init_export(struct obd_export *exp)
5192 {
5193         struct mdt_export_data *med = &exp->exp_mdt_data;
5194         int                     rc;
5195         ENTRY;
5196
5197         CFS_INIT_LIST_HEAD(&med->med_open_head);
5198         cfs_spin_lock_init(&med->med_open_lock);
5199         cfs_sema_init(&med->med_idmap_sem, 1);
5200         med->med_idmap = NULL;
5201         cfs_spin_lock(&exp->exp_lock);
5202         exp->exp_connecting = 1;
5203         cfs_spin_unlock(&exp->exp_lock);
5204         rc = lut_client_alloc(exp);
5205         if (rc == 0)
5206                 rc = ldlm_init_export(exp);
5207
5208         if (rc)
5209                 CERROR("Error %d while initializing export\n", rc);
5210         RETURN(rc);
5211 }
5212
5213 static int mdt_destroy_export(struct obd_export *exp)
5214 {
5215         struct mdt_export_data *med;
5216         int rc = 0;
5217         ENTRY;
5218
5219         med = &exp->exp_mdt_data;
5220         if (exp_connect_rmtclient(exp))
5221                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5222
5223         target_destroy_export(exp);
5224         ldlm_destroy_export(exp);
5225         lut_client_free(exp);
5226
5227         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5228         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5229         if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5230                 RETURN(0);
5231
5232         RETURN(rc);
5233 }
5234
5235 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5236 {
5237         if (flag & CONFIG_LOG)
5238                 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5239
5240         /* also notify active event */
5241         if (flag & CONFIG_SYNC)
5242                 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5243
5244         if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5245             cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5246                 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5247
5248                 /* Open for clients */
5249                 if (obd->obd_no_conn) {
5250                         cfs_spin_lock(&obd->obd_dev_lock);
5251                         obd->obd_no_conn = 0;
5252                         cfs_spin_unlock(&obd->obd_dev_lock);
5253                 }
5254         }
5255 }
5256
5257 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5258                       enum md_upcall_event ev, void *data)
5259 {
5260         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5261         struct md_device  *next  = m->mdt_child;
5262         struct mdt_thread_info *mti;
5263         int rc = 0;
5264         ENTRY;
5265
5266         switch (ev) {
5267                 case MD_LOV_SYNC:
5268                         rc = next->md_ops->mdo_maxsize_get(env, next,
5269                                         &m->mdt_max_mdsize,
5270                                         &m->mdt_max_cookiesize);
5271                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5272                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5273                         mdt_allow_cli(m, CONFIG_SYNC);
5274                         if (data)
5275                                 (*(__u64 *)data) =
5276                                       m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5277                         break;
5278                 case MD_NO_TRANS:
5279                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5280                         mti->mti_no_need_trans = 1;
5281                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5282                         break;
5283                 case MD_LOV_CONFIG:
5284                         /* Check that MDT is not yet configured */
5285                         LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5286                         break;
5287 #ifdef HAVE_QUOTA_SUPPORT
5288                 case MD_LOV_QUOTA:
5289                         if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5290                             likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5291                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5292                         break;
5293 #endif
5294                 default:
5295                         CERROR("invalid event\n");
5296                         rc = -EINVAL;
5297                         break;
5298         }
5299         RETURN(rc);
5300 }
5301
5302 static int mdt_obd_notify(struct obd_device *host,
5303                           struct obd_device *watched,
5304                           enum obd_notify_event ev, void *data)
5305 {
5306         struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5307 #ifdef HAVE_QUOTA_SUPPORT
5308         struct md_device *next = mdt->mdt_child;
5309 #endif
5310         ENTRY;
5311
5312         switch (ev) {
5313         case OBD_NOTIFY_CONFIG:
5314                 mdt_allow_cli(mdt, (unsigned long)data);
5315
5316 #ifdef HAVE_QUOTA_SUPPORT
5317                /* quota_type has been processed, we can now handle
5318                 * incoming quota requests */
5319                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5320 #endif
5321                 break;
5322         default:
5323                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5324         }
5325         RETURN(0);
5326 }
5327
5328 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5329                             void *val, int vallen)
5330 {
5331         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5332         struct getinfo_fid2path *fpout, *fpin;
5333         int rc = 0;
5334
5335         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5336         fpout = val;
5337
5338         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5339                 lustre_swab_fid2path(fpin);
5340
5341         memcpy(fpout, fpin, sizeof(*fpin));
5342         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5343                 RETURN(-EINVAL);
5344
5345         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5346         RETURN(rc);
5347 }
5348
5349 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5350                         struct getinfo_fid2path *fp)
5351 {
5352         struct mdt_object *obj;
5353         int    rc;
5354         ENTRY;
5355
5356         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5357                PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5358
5359         if (!fid_is_sane(&fp->gf_fid))
5360                 RETURN(-EINVAL);
5361
5362         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5363         if (obj == NULL || IS_ERR(obj)) {
5364                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5365                        PTR_ERR(obj));
5366                 RETURN(-EINVAL);
5367         }
5368
5369         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5370         if (rc <= 0) {
5371                 if (rc == -1)
5372                         rc = -EREMOTE;
5373                 else
5374                         rc = -ENOENT;
5375                 mdt_object_put(env, obj);
5376                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5377                        PFID(&fp->gf_fid), rc);
5378                 RETURN(rc);
5379         }
5380
5381         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5382                      fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5383         mdt_object_put(env, obj);
5384
5385         RETURN(rc);
5386 }
5387
5388 static int mdt_get_info(struct mdt_thread_info *info)
5389 {
5390         struct ptlrpc_request *req = mdt_info_req(info);
5391         char *key;
5392         int keylen;
5393         __u32 *vallen;
5394         void *valout;
5395         int rc;
5396         ENTRY;
5397
5398         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5399         if (key == NULL) {
5400                 CDEBUG(D_IOCTL, "No GETINFO key");
5401                 RETURN(-EFAULT);
5402         }
5403         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5404                                       RCL_CLIENT);
5405
5406         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5407         if (vallen == NULL) {
5408                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5409                 RETURN(-EFAULT);
5410         }
5411
5412         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5413                              *vallen);
5414         rc = req_capsule_server_pack(info->mti_pill);
5415         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5416         if (valout == NULL) {
5417                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5418                 RETURN(-EFAULT);
5419         }
5420
5421         if (KEY_IS(KEY_FID2PATH))
5422                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5423         else
5424                 rc = -EINVAL;
5425
5426         lustre_msg_set_status(req->rq_repmsg, rc);
5427
5428         RETURN(rc);
5429 }
5430
5431 /* Pass the ioc down */
5432 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5433                          unsigned int cmd, int len, void *data)
5434 {
5435         struct lu_context ioctl_session;
5436         struct md_device *next = mdt->mdt_child;
5437         int rc;
5438         ENTRY;
5439
5440         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5441         if (rc)
5442                 RETURN(rc);
5443         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5444         lu_context_enter(&ioctl_session);
5445         env->le_ses = &ioctl_session;
5446
5447         LASSERT(next->md_ops->mdo_iocontrol);
5448         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5449
5450         lu_context_exit(&ioctl_session);
5451         lu_context_fini(&ioctl_session);
5452         RETURN(rc);
5453 }
5454
5455 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5456 {
5457         struct obd_ioctl_data *data = karg;
5458         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5459         __u64 version;
5460         struct mdt_object *obj;
5461         struct mdt_lock_handle  *lh;
5462         int rc;
5463         ENTRY;
5464         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5465         if (!fid_is_sane(fid))
5466                 RETURN(-EINVAL);
5467
5468         lh = &mti->mti_lh[MDT_LH_PARENT];
5469         mdt_lock_reg_init(lh, LCK_CR);
5470
5471         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5472         if (IS_ERR(obj))
5473                 RETURN(PTR_ERR(obj));
5474
5475         rc = mdt_object_exists(obj);
5476         if (rc < 0) {
5477                 rc = -EREMOTE;
5478                 /**
5479                  * before calling version get the correct MDS should be
5480                  * fid, this is error to find remote object here
5481                  */
5482                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5483         } else {
5484                 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5485                *(__u64 *)data->ioc_inlbuf2 = version;
5486                 rc = 0;
5487         }
5488         mdt_object_unlock_put(mti, obj, lh, 1);
5489         RETURN(rc);
5490 }
5491
5492 /* ioctls on obd dev */
5493 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5494                          void *karg, void *uarg)
5495 {
5496         struct lu_env      env;
5497         struct obd_device *obd = exp->exp_obd;
5498         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5499         struct dt_device  *dt = mdt->mdt_bottom;
5500         int rc;
5501
5502         ENTRY;
5503         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5504         rc = lu_env_init(&env, LCT_MD_THREAD);
5505         if (rc)
5506                 RETURN(rc);
5507
5508         switch (cmd) {
5509         case OBD_IOC_SYNC:
5510                 rc = mdt_device_sync(&env, mdt);
5511                 break;
5512         case OBD_IOC_SET_READONLY:
5513                 dt->dd_ops->dt_ro(&env, dt);
5514                 break;
5515         case OBD_IOC_ABORT_RECOVERY:
5516                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5517                 target_stop_recovery_thread(obd);
5518                 rc = 0;
5519                 break;
5520         case OBD_IOC_CHANGELOG_REG:
5521         case OBD_IOC_CHANGELOG_DEREG:
5522         case OBD_IOC_CHANGELOG_CLEAR:
5523                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5524                 break;
5525         case OBD_IOC_GET_OBJ_VERSION: {
5526                 struct mdt_thread_info *mti;
5527                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5528                 memset(mti, 0, sizeof *mti);
5529                 mti->mti_env = &env;
5530                 mti->mti_mdt = mdt;
5531                 mti->mti_exp = exp;
5532
5533                 rc = mdt_ioc_version_get(mti, karg);
5534                 break;
5535         }
5536         default:
5537                 CERROR("Not supported cmd = %d for device %s\n",
5538                        cmd, obd->obd_name);
5539                 rc = -EOPNOTSUPP;
5540         }
5541
5542         lu_env_fini(&env);
5543         RETURN(rc);
5544 }
5545
5546 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5547 {
5548         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5549 #ifdef HAVE_QUOTA_SUPPORT
5550         struct obd_device *obd = mdt2obd_dev(mdt);
5551         struct md_device *next = mdt->mdt_child;
5552 #endif
5553         int rc;
5554         ENTRY;
5555
5556         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5557 #ifdef HAVE_QUOTA_SUPPORT
5558         if (likely(obd->obd_stopping == 0))
5559                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5560 #endif
5561         RETURN(rc);
5562 }
5563
5564 int mdt_obd_postrecov(struct obd_device *obd)
5565 {
5566         struct lu_env env;
5567         int rc;
5568
5569         rc = lu_env_init(&env, LCT_MD_THREAD);
5570         if (rc)
5571                 RETURN(rc);
5572         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5573         lu_env_fini(&env);
5574         return rc;
5575 }
5576
5577 /**
5578  * Send a copytool req to a client
5579  * Note this sends a request RPC from a server (MDT) to a client (MDC),
5580  * backwards of normal comms.
5581  */
5582 int mdt_hsm_copytool_send(struct obd_export *exp)
5583 {
5584         struct kuc_hdr *lh;
5585         struct hsm_action_list *hal;
5586         struct hsm_action_item *hai;
5587         int rc, len;
5588         ENTRY;
5589
5590         CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5591               libcfs_nid2str(exp->exp_connection->c_peer.nid));
5592
5593         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5594                 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5595         OBD_ALLOC(lh, len);
5596         if (lh == NULL)
5597                 RETURN(-ENOMEM);
5598
5599         lh->kuc_magic = KUC_MAGIC;
5600         lh->kuc_transport = KUC_TRANSPORT_HSM;
5601         lh->kuc_msgtype = HMT_ACTION_LIST;
5602         lh->kuc_msglen = len;
5603
5604         hal = (struct hsm_action_list *)(lh + 1);
5605         hal->hal_version = HAL_VERSION;
5606         hal->hal_archive_num = 1;
5607         obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5608                         MTI_NAME_MAXLEN);
5609
5610         /* mock up an action list */
5611         hal->hal_count = 2;
5612         hai = hai_zero(hal);
5613         hai->hai_action = HSMA_ARCHIVE;
5614         hai->hai_fid.f_oid = 0xA00A;
5615         hai->hai_len = sizeof(*hai);
5616         hai = hai_next(hai);
5617         hai->hai_action = HSMA_RESTORE;
5618         hai->hai_fid.f_oid = 0xB00B;
5619         hai->hai_len = sizeof(*hai);
5620
5621         /* Uses the ldlm reverse import; this rpc will be seen by
5622           the ldlm_callback_handler */
5623         rc = do_set_info_async(exp->exp_imp_reverse,
5624                                LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5625                                sizeof(KEY_HSM_COPYTOOL_SEND),
5626                                KEY_HSM_COPYTOOL_SEND,
5627                                len, lh, NULL);
5628
5629         OBD_FREE(lh, len);
5630
5631         RETURN(rc);
5632 }
5633
5634 static struct obd_ops mdt_obd_device_ops = {
5635         .o_owner          = THIS_MODULE,
5636         .o_set_info_async = mdt_obd_set_info_async,
5637         .o_connect        = mdt_obd_connect,
5638         .o_reconnect      = mdt_obd_reconnect,
5639         .o_disconnect     = mdt_obd_disconnect,
5640         .o_init_export    = mdt_init_export,
5641         .o_destroy_export = mdt_destroy_export,
5642         .o_iocontrol      = mdt_iocontrol,
5643         .o_postrecov      = mdt_obd_postrecov,
5644         .o_notify         = mdt_obd_notify
5645 };
5646
5647 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5648                                          struct lu_device *d)
5649 {
5650         struct mdt_device *m = mdt_dev(d);
5651         ENTRY;
5652
5653         mdt_fini(env, m);
5654         RETURN(NULL);
5655 }
5656
5657 static struct lu_device *mdt_device_free(const struct lu_env *env,
5658                                          struct lu_device *d)
5659 {
5660         struct mdt_device *m = mdt_dev(d);
5661         ENTRY;
5662
5663         md_device_fini(&m->mdt_md_dev);
5664         OBD_FREE_PTR(m);
5665         RETURN(NULL);
5666 }
5667
5668 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5669                                           struct lu_device_type *t,
5670                                           struct lustre_cfg *cfg)
5671 {
5672         struct lu_device  *l;
5673         struct mdt_device *m;
5674
5675         OBD_ALLOC_PTR(m);
5676         if (m != NULL) {
5677                 int rc;
5678
5679                 l = &m->mdt_md_dev.md_lu_dev;
5680                 rc = mdt_init0(env, m, t, cfg);
5681                 if (rc != 0) {
5682                         mdt_device_free(env, l);
5683                         l = ERR_PTR(rc);
5684                         return l;
5685                 }
5686                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5687         } else
5688                 l = ERR_PTR(-ENOMEM);
5689         return l;
5690 }
5691
5692 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5693 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5694
5695 /* context key: mdt_thread_key */
5696 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5697
5698 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5699 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5700
5701 struct lu_context_key mdt_txn_key = {
5702         .lct_tags = LCT_TX_HANDLE,
5703         .lct_init = mdt_txn_key_init,
5704         .lct_fini = mdt_txn_key_fini
5705 };
5706
5707 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5708 {
5709         return md_ucred(info->mti_env);
5710 }
5711
5712 /**
5713  * Enable/disable COS (Commit On Sharing).
5714  *
5715  * Set/Clear the COS flag in mdt options.
5716  *
5717  * \param mdt mdt device
5718  * \param val 0 disables COS, other values enable COS
5719  */
5720 void mdt_enable_cos(struct mdt_device *mdt, int val)
5721 {
5722         struct lu_env env;
5723         int rc;
5724
5725         mdt->mdt_opts.mo_cos = !!val;
5726         rc = lu_env_init(&env, LCT_MD_THREAD);
5727         if (unlikely(rc != 0)) {
5728                 CWARN("lu_env initialization failed with rc = %d,"
5729                       "cannot sync\n", rc);
5730                 return;
5731         }
5732         mdt_device_sync(&env, mdt);
5733         lu_env_fini(&env);
5734 }
5735
5736 /**
5737  * Check COS (Commit On Sharing) status.
5738  *
5739  * Return COS flag status.
5740  *
5741  * \param mdt mdt device
5742  */
5743 int mdt_cos_is_enabled(struct mdt_device *mdt)
5744 {
5745         return mdt->mdt_opts.mo_cos != 0;
5746 }
5747
5748 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5749 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5750
5751 static struct lu_device_type_operations mdt_device_type_ops = {
5752         .ldto_init = mdt_type_init,
5753         .ldto_fini = mdt_type_fini,
5754
5755         .ldto_start = mdt_type_start,
5756         .ldto_stop  = mdt_type_stop,
5757
5758         .ldto_device_alloc = mdt_device_alloc,
5759         .ldto_device_free  = mdt_device_free,
5760         .ldto_device_fini  = mdt_device_fini
5761 };
5762
5763 static struct lu_device_type mdt_device_type = {
5764         .ldt_tags     = LU_DEVICE_MD,
5765         .ldt_name     = LUSTRE_MDT_NAME,
5766         .ldt_ops      = &mdt_device_type_ops,
5767         .ldt_ctx_tags = LCT_MD_THREAD
5768 };
5769
5770 static struct lu_local_obj_desc mdt_last_recv = {
5771         .llod_name      = LAST_RCVD,
5772         .llod_oid       = MDT_LAST_RECV_OID,
5773         .llod_is_index  = 0,
5774 };
5775
5776 static int __init mdt_mod_init(void)
5777 {
5778         struct lprocfs_static_vars lvars;
5779         int rc;
5780
5781         llo_local_obj_register(&mdt_last_recv);
5782
5783         if (mdt_num_threads > 0) {
5784                 if (mdt_num_threads > MDT_MAX_THREADS)
5785                         mdt_num_threads = MDT_MAX_THREADS;
5786                 if (mdt_num_threads < MDT_MIN_THREADS)
5787                         mdt_num_threads = MDT_MIN_THREADS;
5788                 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5789         } else {
5790                 mdt_max_threads = MDT_MAX_THREADS;
5791                 mdt_min_threads = MDT_MIN_THREADS;
5792         }
5793
5794         lprocfs_mdt_init_vars(&lvars);
5795         rc = class_register_type(&mdt_obd_device_ops, NULL,
5796                                  lvars.module_vars, LUSTRE_MDT_NAME,
5797                                  &mdt_device_type);
5798
5799         return rc;
5800 }
5801
5802 static void __exit mdt_mod_exit(void)
5803 {
5804         llo_local_obj_unregister(&mdt_last_recv);
5805         class_unregister_type(LUSTRE_MDT_NAME);
5806 }
5807
5808
5809 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5810 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5811         .mh_name    = #opc,                                             \
5812         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5813         .mh_opc     = prefix ## _  ## opc,                              \
5814         .mh_flags   = flags,                                            \
5815         .mh_act     = fn,                                               \
5816         .mh_fmt     = fmt                                               \
5817 }
5818
5819 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5820         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5821
5822 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5823         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5824
5825 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5826         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5827 /*
5828  * Request with a format known in advance
5829  */
5830 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5831         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5832
5833 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5834         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5835
5836 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5837         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5838 /*
5839  * Request with a format we do not yet know
5840  */
5841 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5842         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5843
5844 static struct mdt_handler mdt_mds_ops[] = {
5845 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5846 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5847 DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
5848                                                              &RQF_OBD_SET_INFO),
5849 DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
5850 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5851 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5852 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5853 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5854 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5855 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5856 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5857 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5858 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5859 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5860 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5861 #ifdef HAVE_QUOTA_SUPPORT
5862 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5863 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5864 #endif
5865 };
5866
5867 #define DEF_OBD_HNDL(flags, name, fn)                   \
5868         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5869
5870
5871 static struct mdt_handler mdt_obd_ops[] = {
5872         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5873         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5874         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5875 };
5876
5877 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5878         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5879 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5880         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5881
5882 static struct mdt_handler mdt_dlm_ops[] = {
5883         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5884         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5885         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5886         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5887 };
5888
5889 #define DEF_LLOG_HNDL(flags, name, fn)                   \
5890         DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5891
5892 static struct mdt_handler mdt_llog_ops[] = {
5893         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
5894         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
5895         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5896         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
5897         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
5898         DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
5899         DEF_LLOG_HNDL(0, CATINFO,                   NULL),
5900         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
5901         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
5902 };
5903
5904 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5905         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5906
5907 static struct mdt_handler mdt_sec_ctx_ops[] = {
5908         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5909         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5910         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5911 };
5912
5913 static struct mdt_opc_slice mdt_regular_handlers[] = {
5914         {
5915                 .mos_opc_start = MDS_GETATTR,
5916                 .mos_opc_end   = MDS_LAST_OPC,
5917                 .mos_hs        = mdt_mds_ops
5918         },
5919         {
5920                 .mos_opc_start = OBD_PING,
5921                 .mos_opc_end   = OBD_LAST_OPC,
5922                 .mos_hs        = mdt_obd_ops
5923         },
5924         {
5925                 .mos_opc_start = LDLM_ENQUEUE,
5926                 .mos_opc_end   = LDLM_LAST_OPC,
5927                 .mos_hs        = mdt_dlm_ops
5928         },
5929         {
5930                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5931                 .mos_opc_end   = LLOG_LAST_OPC,
5932                 .mos_hs        = mdt_llog_ops
5933         },
5934         {
5935                 .mos_opc_start = SEC_CTX_INIT,
5936                 .mos_opc_end   = SEC_LAST_OPC,
5937                 .mos_hs        = mdt_sec_ctx_ops
5938         },
5939         {
5940                 .mos_hs        = NULL
5941         }
5942 };
5943
5944 static struct mdt_handler mdt_readpage_ops[] = {
5945         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5946         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5947 #ifdef HAVE_SPLIT_SUPPORT
5948         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5949 #endif
5950
5951         /*
5952          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5953          * detailed comments. --umka
5954          */
5955         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5956         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5957 };
5958
5959 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5960         {
5961                 .mos_opc_start = MDS_GETATTR,
5962                 .mos_opc_end   = MDS_LAST_OPC,
5963                 .mos_hs        = mdt_readpage_ops
5964         },
5965         {
5966                 .mos_hs        = NULL
5967         }
5968 };
5969
5970 static struct mdt_handler mdt_xmds_ops[] = {
5971         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5972         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5973         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5974         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5975 };
5976
5977 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5978         {
5979                 .mos_opc_start = MDS_GETATTR,
5980                 .mos_opc_end   = MDS_LAST_OPC,
5981                 .mos_hs        = mdt_xmds_ops
5982         },
5983         {
5984                 .mos_opc_start = OBD_PING,
5985                 .mos_opc_end   = OBD_LAST_OPC,
5986                 .mos_hs        = mdt_obd_ops
5987         },
5988         {
5989                 .mos_opc_start = SEC_CTX_INIT,
5990                 .mos_opc_end   = SEC_LAST_OPC,
5991                 .mos_hs        = mdt_sec_ctx_ops
5992         },
5993         {
5994                 .mos_hs        = NULL
5995         }
5996 };
5997
5998 static struct mdt_handler mdt_seq_ops[] = {
5999         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
6000 };
6001
6002 static struct mdt_opc_slice mdt_seq_handlers[] = {
6003         {
6004                 .mos_opc_start = SEQ_QUERY,
6005                 .mos_opc_end   = SEQ_LAST_OPC,
6006                 .mos_hs        = mdt_seq_ops
6007         },
6008         {
6009                 .mos_hs        = NULL
6010         }
6011 };
6012
6013 static struct mdt_handler mdt_fld_ops[] = {
6014         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6015 };
6016
6017 static struct mdt_opc_slice mdt_fld_handlers[] = {
6018         {
6019                 .mos_opc_start = FLD_QUERY,
6020                 .mos_opc_end   = FLD_LAST_OPC,
6021                 .mos_hs        = mdt_fld_ops
6022         },
6023         {
6024                 .mos_hs        = NULL
6025         }
6026 };
6027
6028 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6029 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6030 MODULE_LICENSE("GPL");
6031
6032 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6033                 "number of mdt service threads to start");
6034
6035 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);