lustre/mdt/mdt_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #ifndef EXPORT_SYMTAB
  50 # define EXPORT_SYMTAB
  51 #endif
  52 #define DEBUG_SUBSYSTEM S_MDS
  53
  54 #include <linux/module.h>
  55 /*
  56  * struct OBD_{ALLOC,FREE}*()
  57  */
  58 #include <obd_support.h>
  59 /* struct ptlrpc_request */
  60 #include <lustre_net.h>
  61 /* struct obd_export */
  62 #include <lustre_export.h>
  63 /* struct obd_device */
  64 #include <obd.h>
  65 /* lu2dt_dev() */
  66 #include <dt_object.h>
  67 #include <lustre_mds.h>
  68 #include <lustre_mdt.h>
  69 #include "mdt_internal.h"
  70 #ifdef HAVE_QUOTA_SUPPORT
  71 # include <lustre_quota.h>
  72 #endif
  73 #include <lustre_acl.h>
  74 #include <lustre_param.h>
  75 #include <lustre_fsfilt.h>
  76
  77 mdl_mode_t mdt_mdl_lock_modes[] = {
  78         [LCK_MINMODE] = MDL_MINMODE,
  79         [LCK_EX]      = MDL_EX,
  80         [LCK_PW]      = MDL_PW,
  81         [LCK_PR]      = MDL_PR,
  82         [LCK_CW]      = MDL_CW,
  83         [LCK_CR]      = MDL_CR,
  84         [LCK_NL]      = MDL_NL,
  85         [LCK_GROUP]   = MDL_GROUP
  86 };
  87
  88 ldlm_mode_t mdt_dlm_lock_modes[] = {
  89         [MDL_MINMODE] = LCK_MINMODE,
  90         [MDL_EX]      = LCK_EX,
  91         [MDL_PW]      = LCK_PW,
  92         [MDL_PR]      = LCK_PR,
  93         [MDL_CW]      = LCK_CW,
  94         [MDL_CR]      = LCK_CR,
  95         [MDL_NL]      = LCK_NL,
  96         [MDL_GROUP]   = LCK_GROUP
  97 };
  98
  99 /*
 100  * Initialized in mdt_mod_init().
 101  */
 102 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         /*XXX should pack the reply body according to lu_valid*/
 406         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 407                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 408                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 409                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 410
 411         if (!S_ISREG(attr->la_mode))
 412                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 413
 414         /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 415         if (ma->ma_lmm_size == 0)
 416                 b->valid |= OBD_MD_FLSIZE;
 417
 418         b->atime      = attr->la_atime;
 419         b->mtime      = attr->la_mtime;
 420         b->ctime      = attr->la_ctime;
 421         b->mode       = attr->la_mode;
 422         b->size       = attr->la_size;
 423         b->blocks     = attr->la_blocks;
 424         b->uid        = attr->la_uid;
 425         b->gid        = attr->la_gid;
 426         b->flags      = attr->la_flags;
 427         b->nlink      = attr->la_nlink;
 428         b->rdev       = attr->la_rdev;
 429
 430         if (fid) {
 431                 b->fid1 = *fid;
 432                 b->valid |= OBD_MD_FLID;
 433
 434                 /* FIXME: these should be fixed when new igif ready.*/
 435                 b->ino  =  fid_oid(fid);       /* 1.6 compatibility */
 436                 b->generation = fid_ver(fid);  /* 1.6 compatibility */
 437                 b->valid |= OBD_MD_FLGENER;    /* 1.6 compatibility */
 438
 439                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 440                                 PFID(fid), b->nlink, b->mode, b->size);
 441         }
 442
 443         if (info)
 444                 mdt_body_reverse_idmap(info, b);
 445 }
 446
 447 static inline int mdt_body_has_lov(const struct lu_attr *la,
 448                                    const struct mdt_body *body)
 449 {
 450         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 451                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 452 }
 453
 454 static int mdt_getattr_internal(struct mdt_thread_info *info,
 455                                 struct mdt_object *o, int ma_need)
 456 {
 457         struct md_object        *next = mdt_object_child(o);
 458         const struct mdt_body   *reqbody = info->mti_body;
 459         struct ptlrpc_request   *req = mdt_info_req(info);
 460         struct md_attr          *ma = &info->mti_attr;
 461         struct lu_attr          *la = &ma->ma_attr;
 462         struct req_capsule      *pill = info->mti_pill;
 463         const struct lu_env     *env = info->mti_env;
 464         struct mdt_body         *repbody;
 465         struct lu_buf           *buffer = &info->mti_buf;
 466         int                     rc;
 467         ENTRY;
 468
 469         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 470                 RETURN(err_serious(-ENOMEM));
 471
 472         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 473
 474         ma->ma_valid = 0;
 475
 476         rc = mdt_object_exists(o);
 477         if (rc < 0) {
 478                 /* This object is located on remote node.*/
 479                 repbody->fid1 = *mdt_object_fid(o);
 480                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 481                 RETURN(0);
 482         }
 483
 484         buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 485         buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
 486
 487         /* If it is dir object and client require MEA, then we got MEA */
 488         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 489             reqbody->valid & OBD_MD_MEA) {
 490                 /* Assumption: MDT_MD size is enough for lmv size. */
 491                 ma->ma_lmv = buffer->lb_buf;
 492                 ma->ma_lmv_size = buffer->lb_len;
 493                 ma->ma_need = MA_LMV | MA_INODE;
 494         } else {
 495                 ma->ma_lmm = buffer->lb_buf;
 496                 ma->ma_lmm_size = buffer->lb_len;
 497                 ma->ma_need = MA_LOV | MA_INODE;
 498         }
 499
 500         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 501             reqbody->valid & OBD_MD_FLDIREA  &&
 502             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 503                 /* get default stripe info for this dir. */
 504                 ma->ma_need |= MA_LOV_DEF;
 505         }
 506         ma->ma_need |= ma_need;
 507         if (ma->ma_need & MA_SOM)
 508                 ma->ma_som = &info->mti_u.som.data;
 509
 510         rc = mo_attr_get(env, next, ma);
 511         if (unlikely(rc)) {
 512                 CERROR("getattr error for "DFID": %d\n",
 513                         PFID(mdt_object_fid(o)), rc);
 514                 RETURN(rc);
 515         }
 516
 517         if (likely(ma->ma_valid & MA_INODE))
 518                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 519         else
 520                 RETURN(-EFAULT);
 521
 522         if (mdt_body_has_lov(la, reqbody)) {
 523                 if (ma->ma_valid & MA_LOV) {
 524                         LASSERT(ma->ma_lmm_size);
 525                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 526                         repbody->eadatasize = ma->ma_lmm_size;
 527                         if (S_ISDIR(la->la_mode))
 528                                 repbody->valid |= OBD_MD_FLDIREA;
 529                         else
 530                                 repbody->valid |= OBD_MD_FLEASIZE;
 531                 }
 532                 if (ma->ma_valid & MA_LMV) {
 533                         LASSERT(S_ISDIR(la->la_mode));
 534                         repbody->eadatasize = ma->ma_lmv_size;
 535                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 536                 }
 537                 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
 538                         repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 539                 }
 540         } else if (S_ISLNK(la->la_mode) &&
 541                    reqbody->valid & OBD_MD_LINKNAME) {
 542                 buffer->lb_buf = ma->ma_lmm;
 543                 buffer->lb_len = reqbody->eadatasize;
 544                 rc = mo_readlink(env, next, buffer);
 545                 if (unlikely(rc <= 0)) {
 546                         CERROR("readlink failed: %d\n", rc);
 547                         rc = -EFAULT;
 548                 } else {
 549                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 550                                  rc -= 2;
 551                         repbody->valid |= OBD_MD_LINKNAME;
 552                         repbody->eadatasize = rc;
 553                         /* NULL terminate */
 554                         ((char*)ma->ma_lmm)[rc - 1] = 0;
 555                         CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
 556                                (char*)ma->ma_lmm, rc);
 557                         rc = 0;
 558                 }
 559         }
 560
 561         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 562                 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
 563                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 564                 repbody->valid |= OBD_MD_FLMODEASIZE;
 565                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 566                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 567                        repbody->max_cookiesize);
 568         }
 569
 570         if (exp_connect_rmtclient(info->mti_exp) &&
 571             reqbody->valid & OBD_MD_FLRMTPERM) {
 572                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 573
 574                 /* mdt_getattr_lock only */
 575                 rc = mdt_pack_remote_perm(info, o, buf);
 576                 if (rc) {
 577                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 578                         repbody->aclsize = 0;
 579                         RETURN(rc);
 580                 } else {
 581                         repbody->valid |= OBD_MD_FLRMTPERM;
 582                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 583                 }
 584         }
 585 #ifdef CONFIG_FS_POSIX_ACL
 586         else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
 587                  (reqbody->valid & OBD_MD_FLACL)) {
 588                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 589                 buffer->lb_len = req_capsule_get_size(pill,
 590                                                       &RMF_ACL, RCL_SERVER);
 591                 if (buffer->lb_len > 0) {
 592                         rc = mo_xattr_get(env, next, buffer,
 593                                           XATTR_NAME_ACL_ACCESS);
 594                         if (rc < 0) {
 595                                 if (rc == -ENODATA) {
 596                                         repbody->aclsize = 0;
 597                                         repbody->valid |= OBD_MD_FLACL;
 598                                         rc = 0;
 599                                 } else if (rc == -EOPNOTSUPP) {
 600                                         rc = 0;
 601                                 } else {
 602                                         CERROR("got acl size: %d\n", rc);
 603                                 }
 604                         } else {
 605                                 repbody->aclsize = rc;
 606                                 repbody->valid |= OBD_MD_FLACL;
 607                                 rc = 0;
 608                         }
 609                 }
 610         }
 611 #endif
 612
 613         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 614             info->mti_mdt->mdt_opts.mo_mds_capa &&
 615             info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
 616                 struct lustre_capa *capa;
 617
 618                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 619                 LASSERT(capa);
 620                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 621                 rc = mo_capa_get(env, next, capa, 0);
 622                 if (rc)
 623                         RETURN(rc);
 624                 repbody->valid |= OBD_MD_FLMDSCAPA;
 625         }
 626         RETURN(rc);
 627 }
 628
 629 static int mdt_renew_capa(struct mdt_thread_info *info)
 630 {
 631         struct mdt_object  *obj = info->mti_object;
 632         struct mdt_body    *body;
 633         struct lustre_capa *capa, *c;
 634         int rc;
 635         ENTRY;
 636
 637         /* if object doesn't exist, or server has disabled capability,
 638          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 639          * flag not set.
 640          */
 641         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 642             !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
 643                 RETURN(0);
 644
 645         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 646         LASSERT(body != NULL);
 647
 648         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 649         LASSERT(c);
 650
 651         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 652         LASSERT(capa);
 653
 654         *capa = *c;
 655         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 656         if (rc == 0)
 657                 body->valid |= OBD_MD_FLOSSCAPA;
 658         RETURN(rc);
 659 }
 660
 661 static int mdt_getattr(struct mdt_thread_info *info)
 662 {
 663         struct mdt_object       *obj = info->mti_object;
 664         struct req_capsule      *pill = info->mti_pill;
 665         struct mdt_body         *reqbody;
 666         struct mdt_body         *repbody;
 667         mode_t                   mode;
 668         int                      md_size;
 669         int rc;
 670         ENTRY;
 671
 672         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 673         LASSERT(reqbody);
 674
 675         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 676                 rc = req_capsule_server_pack(pill);
 677                 if (unlikely(rc))
 678                         RETURN(err_serious(rc));
 679                 rc = mdt_renew_capa(info);
 680                 GOTO(out_shrink, rc);
 681         }
 682
 683         LASSERT(obj != NULL);
 684         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 685
 686         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 687         if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
 688             (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
 689                 md_size = reqbody->eadatasize;
 690         else
 691                 md_size = info->mti_mdt->mdt_max_mdsize;
 692
 693         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
 694
 695         rc = req_capsule_server_pack(pill);
 696         if (unlikely(rc != 0))
 697                 RETURN(err_serious(rc));
 698
 699         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 700         LASSERT(repbody != NULL);
 701         repbody->eadatasize = 0;
 702         repbody->aclsize = 0;
 703
 704         if (reqbody->valid & OBD_MD_FLRMTPERM)
 705                 rc = mdt_init_ucred(info, reqbody);
 706         else
 707                 rc = mdt_check_ucred(info);
 708         if (unlikely(rc))
 709                 GOTO(out_shrink, rc);
 710
 711         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
 712         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 713
 714         /*
 715          * Don't check capability at all, because rename might getattr for
 716          * remote obj, and at that time no capability is available.
 717          */
 718         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 719         rc = mdt_getattr_internal(info, obj, 0);
 720         if (reqbody->valid & OBD_MD_FLRMTPERM)
 721                 mdt_exit_ucred(info);
 722         EXIT;
 723 out_shrink:
 724         mdt_shrink_reply(info);
 725         return rc;
 726 }
 727
 728 static int mdt_is_subdir(struct mdt_thread_info *info)
 729 {
 730         struct mdt_object     *o = info->mti_object;
 731         struct req_capsule    *pill = info->mti_pill;
 732         const struct mdt_body *body = info->mti_body;
 733         struct mdt_body       *repbody;
 734         int                    rc;
 735         ENTRY;
 736
 737         LASSERT(o != NULL);
 738
 739         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 740
 741         /*
 742          * We save last checked parent fid to @repbody->fid1 for remote
 743          * directory case.
 744          */
 745         LASSERT(fid_is_sane(&body->fid2));
 746         LASSERT(mdt_object_exists(o) > 0);
 747         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
 748                            &body->fid2, &repbody->fid1);
 749         if (rc == 0 || rc == -EREMOTE)
 750                 repbody->valid |= OBD_MD_FLID;
 751
 752         RETURN(rc);
 753 }
 754
 755 static int mdt_raw_lookup(struct mdt_thread_info *info,
 756                           struct mdt_object *parent,
 757                           const struct lu_name *lname,
 758                           struct ldlm_reply *ldlm_rep)
 759 {
 760         struct md_object *next = mdt_object_child(info->mti_object);
 761         const struct mdt_body *reqbody = info->mti_body;
 762         struct lu_fid *child_fid = &info->mti_tmp_fid1;
 763         struct mdt_body *repbody;
 764         int rc;
 765         ENTRY;
 766
 767         if (reqbody->valid != OBD_MD_FLID)
 768                 RETURN(0);
 769
 770         LASSERT(!info->mti_cross_ref);
 771
 772         /* Only got the fid of this obj by name */
 773         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 774                         &info->mti_spec);
 775 #if 0
 776         /* XXX is raw_lookup possible as intent operation? */
 777         if (rc != 0) {
 778                 if (rc == -ENOENT)
 779                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 780                 RETURN(rc);
 781         } else
 782                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 783
 784         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 785 #endif
 786         if (rc == 0) {
 787                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 788                 repbody->fid1 = *child_fid;
 789                 repbody->valid = OBD_MD_FLID;
 790         }
 791         RETURN(1);
 792 }
 793
 794 /*
 795  * UPDATE lock should be taken against parent, and be release before exit;
 796  * child_bits lock should be taken against child, and be returned back:
 797  *            (1)normal request should release the child lock;
 798  *            (2)intent request will grant the lock to client.
 799  */
 800 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 801                                  struct mdt_lock_handle *lhc,
 802                                  __u64 child_bits,
 803                                  struct ldlm_reply *ldlm_rep)
 804 {
 805         struct ptlrpc_request  *req       = mdt_info_req(info);
 806         struct mdt_body        *reqbody   = NULL;
 807         struct mdt_object      *parent    = info->mti_object;
 808         struct mdt_object      *child;
 809         struct md_object       *next      = mdt_object_child(parent);
 810         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
 811         struct lu_name         *lname     = NULL;
 812         const char             *name      = NULL;
 813         int                     namelen   = 0;
 814         struct mdt_lock_handle *lhp;
 815         struct ldlm_lock       *lock;
 816         struct ldlm_res_id     *res_id;
 817         int                     is_resent;
 818         int                     ma_need = 0;
 819         int                     rc;
 820
 821         ENTRY;
 822
 823         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
 824         LASSERT(ergo(is_resent,
 825                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 826
 827         LASSERT(parent != NULL);
 828         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 829         if (name == NULL)
 830                 RETURN(err_serious(-EFAULT));
 831
 832         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
 833                                        RCL_CLIENT) - 1;
 834         if (!info->mti_cross_ref) {
 835                 /*
 836                  * XXX: Check for "namelen == 0" is for getattr by fid
 837                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
 838                  * that is the name must contain at least one character and
 839                  * the terminating '\0'
 840                  */
 841                 if (namelen == 0) {
 842                         reqbody = req_capsule_client_get(info->mti_pill,
 843                                                          &RMF_MDT_BODY);
 844                         LASSERT(fid_is_sane(&reqbody->fid2));
 845                         name = NULL;
 846
 847                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
 848                                "ldlm_rep = %p\n",
 849                                PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
 850                                ldlm_rep);
 851                 } else {
 852                         lname = mdt_name(info->mti_env, (char *)name, namelen);
 853                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
 854                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
 855                                name, ldlm_rep);
 856                 }
 857         }
 858         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
 859
 860         rc = mdt_object_exists(parent);
 861         if (unlikely(rc == 0)) {
 862                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 863                                 &parent->mot_obj.mo_lu,
 864                                 "Parent doesn't exist!\n");
 865                 RETURN(-ESTALE);
 866         } else if (!info->mti_cross_ref) {
 867                 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
 868                          PFID(mdt_object_fid(parent)));
 869         }
 870         if (lname) {
 871                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
 872                 if (rc != 0) {
 873                         if (rc > 0)
 874                                 rc = 0;
 875                         RETURN(rc);
 876                 }
 877         }
 878
 879         if (info->mti_cross_ref) {
 880                 /* Only getattr on the child. Parent is on another node. */
 881                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 882                 child = parent;
 883                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
 884                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
 885
 886                 if (is_resent) {
 887                         /* Do not take lock for resent case. */
 888                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 889                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 890                                  lhc->mlh_reg_lh.cookie);
 891                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
 892                                                 &lock->l_resource->lr_name));
 893                         LDLM_LOCK_PUT(lock);
 894                         rc = 0;
 895                 } else {
 896                         mdt_lock_handle_init(lhc);
 897                         mdt_lock_reg_init(lhc, LCK_PR);
 898
 899                         /*
 900                          * Object's name is on another MDS, no lookup lock is
 901                          * needed here but update is.
 902                          */
 903                         child_bits &= ~MDS_INODELOCK_LOOKUP;
 904                         child_bits |= MDS_INODELOCK_UPDATE;
 905
 906                         rc = mdt_object_lock(info, child, lhc, child_bits,
 907                                              MDT_LOCAL_LOCK);
 908                 }
 909                 if (rc == 0) {
 910                         /* Finally, we can get attr for child. */
 911                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
 912                                          BYPASS_CAPA);
 913                         rc = mdt_getattr_internal(info, child, 0);
 914                         if (unlikely(rc != 0))
 915                                 mdt_object_unlock(info, child, lhc, 1);
 916                 }
 917                 RETURN(rc);
 918         }
 919
 920         /* step 1: lock parent */
 921         lhp = &info->mti_lh[MDT_LH_PARENT];
 922         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
 923         rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 924                              MDT_LOCAL_LOCK);
 925
 926         if (unlikely(rc != 0))
 927                 RETURN(rc);
 928
 929         if (lname) {
 930                 /* step 2: lookup child's fid by name */
 931                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
 932                                 &info->mti_spec);
 933
 934                 if (rc != 0) {
 935                         if (rc == -ENOENT)
 936                                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 937                         GOTO(out_parent, rc);
 938                 } else
 939                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 940         } else {
 941                 *child_fid = reqbody->fid2;
 942                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
 943         }
 944
 945         /*
 946          *step 3: find the child object by fid & lock it.
 947          *        regardless if it is local or remote.
 948          */
 949         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
 950
 951         if (unlikely(IS_ERR(child)))
 952                 GOTO(out_parent, rc = PTR_ERR(child));
 953         if (is_resent) {
 954                 /* Do not take lock for resent case. */
 955                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
 956                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
 957                          lhc->mlh_reg_lh.cookie);
 958
 959                 res_id = &lock->l_resource->lr_name;
 960                 if (!fid_res_name_eq(mdt_object_fid(child),
 961                                     &lock->l_resource->lr_name)) {
 962                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
 963                                                  &lock->l_resource->lr_name),
 964                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
 965                                  (unsigned long)res_id->name[0],
 966                                  (unsigned long)res_id->name[1],
 967                                  (unsigned long)res_id->name[2],
 968                                  PFID(mdt_object_fid(parent)));
 969                           CWARN("Although resent, but still not get child lock"
 970                                 "parent:"DFID" child:"DFID"\n",
 971                                 PFID(mdt_object_fid(parent)),
 972                                 PFID(mdt_object_fid(child)));
 973                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
 974                           LDLM_LOCK_PUT(lock);
 975                           GOTO(relock, 0);
 976                 }
 977                 LDLM_LOCK_PUT(lock);
 978                 rc = 0;
 979         } else {
 980                 struct md_attr *ma;
 981 relock:
 982                 ma = &info->mti_attr;
 983
 984                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
 985                 mdt_lock_handle_init(lhc);
 986                 mdt_lock_reg_init(lhc, LCK_PR);
 987
 988                 if (mdt_object_exists(child) == 0) {
 989                         LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
 990                                         &child->mot_obj.mo_lu,
 991                                         "Object doesn't exist!\n");
 992                         GOTO(out_child, rc = -ESTALE);
 993                 }
 994
 995                 ma->ma_valid = 0;
 996                 ma->ma_need = MA_INODE;
 997                 rc = mo_attr_get(info->mti_env, next, ma);
 998                 if (unlikely(rc != 0))
 999                         GOTO(out_child, rc);
1000
1001                 /* If the file has not been changed for some time, we return
1002                  * not only a LOOKUP lock, but also an UPDATE lock and this
1003                  * might save us RPC on later STAT. For directories, it also
1004                  * let negative dentry starts working for this dir. */
1005                 if (ma->ma_valid & MA_INODE &&
1006                     ma->ma_attr.la_valid & LA_CTIME &&
1007                     info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1008                     ma->ma_attr.la_ctime < cfs_time_current_sec())
1009                         child_bits |= MDS_INODELOCK_UPDATE;
1010
1011                 rc = mdt_object_lock(info, child, lhc, child_bits,
1012                                      MDT_CROSS_LOCK);
1013
1014                 if (unlikely(rc != 0))
1015                         GOTO(out_child, rc);
1016         }
1017
1018         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1019         /* Get MA_SOM attributes if update lock is given. */
1020         if (lock &&
1021             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1022             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1023                 ma_need = MA_SOM;
1024
1025         /* finally, we can get attr for child. */
1026         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1027         rc = mdt_getattr_internal(info, child, ma_need);
1028         if (unlikely(rc != 0)) {
1029                 mdt_object_unlock(info, child, lhc, 1);
1030         } else if (lock) {
1031                 /* Debugging code. */
1032                 res_id = &lock->l_resource->lr_name;
1033                 LDLM_DEBUG(lock, "Returning lock to client");
1034                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1035                                          &lock->l_resource->lr_name),
1036                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1037                          (unsigned long)res_id->name[0],
1038                          (unsigned long)res_id->name[1],
1039                          (unsigned long)res_id->name[2],
1040                          PFID(mdt_object_fid(child)));
1041                 mdt_pack_size2body(info, child);
1042         }
1043         if (lock)
1044                 LDLM_LOCK_PUT(lock);
1045
1046         EXIT;
1047 out_child:
1048         mdt_object_put(info->mti_env, child);
1049 out_parent:
1050         mdt_object_unlock(info, parent, lhp, 1);
1051         return rc;
1052 }
1053
1054 /* normal handler: should release the child lock */
1055 static int mdt_getattr_name(struct mdt_thread_info *info)
1056 {
1057         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1058         struct mdt_body        *reqbody;
1059         struct mdt_body        *repbody;
1060         int rc;
1061         ENTRY;
1062
1063         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1064         LASSERT(reqbody != NULL);
1065         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1066         LASSERT(repbody != NULL);
1067
1068         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1069         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1070         repbody->eadatasize = 0;
1071         repbody->aclsize = 0;
1072
1073         rc = mdt_init_ucred(info, reqbody);
1074         if (unlikely(rc))
1075                 GOTO(out_shrink, rc);
1076
1077         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1078         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1079                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1080                 lhc->mlh_reg_lh.cookie = 0;
1081         }
1082         mdt_exit_ucred(info);
1083         EXIT;
1084 out_shrink:
1085         mdt_shrink_reply(info);
1086         return rc;
1087 }
1088
1089 static const struct lu_device_operations mdt_lu_ops;
1090
1091 static int lu_device_is_mdt(struct lu_device *d)
1092 {
1093         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1094 }
1095
1096 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1097                          void *karg, void *uarg);
1098
1099 static int mdt_set_info(struct mdt_thread_info *info)
1100 {
1101         struct ptlrpc_request *req = mdt_info_req(info);
1102         char *key;
1103         void *val;
1104         int keylen, vallen, rc = 0;
1105         ENTRY;
1106
1107         rc = req_capsule_server_pack(info->mti_pill);
1108         if (rc)
1109                 RETURN(rc);
1110
1111         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1112         if (key == NULL) {
1113                 DEBUG_REQ(D_HA, req, "no set_info key");
1114                 RETURN(-EFAULT);
1115         }
1116
1117         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1118                                       RCL_CLIENT);
1119
1120         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1121         if (val == NULL) {
1122                 DEBUG_REQ(D_HA, req, "no set_info val");
1123                 RETURN(-EFAULT);
1124         }
1125
1126         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1127                                       RCL_CLIENT);
1128
1129         /* Swab any part of val you need to here */
1130         if (KEY_IS(KEY_READ_ONLY)) {
1131                 req->rq_status = 0;
1132                 lustre_msg_set_status(req->rq_repmsg, 0);
1133
1134                 cfs_spin_lock(&req->rq_export->exp_lock);
1135                 if (*(__u32 *)val)
1136                         req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1137                 else
1138                         req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1139                 cfs_spin_unlock(&req->rq_export->exp_lock);
1140
1141         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1142                 struct changelog_setinfo *cs =
1143                         (struct changelog_setinfo *)val;
1144                 if (vallen != sizeof(*cs)) {
1145                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1146                         RETURN(-EINVAL);
1147                 }
1148                 if (ptlrpc_req_need_swab(req)) {
1149                         __swab64s(&cs->cs_recno);
1150                         __swab32s(&cs->cs_id);
1151                 }
1152
1153                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1154                                    vallen, val, NULL);
1155                 lustre_msg_set_status(req->rq_repmsg, rc);
1156
1157         } else {
1158                 RETURN(-EINVAL);
1159         }
1160         RETURN(0);
1161 }
1162
1163 static int mdt_connect(struct mdt_thread_info *info)
1164 {
1165         int rc;
1166         struct ptlrpc_request *req;
1167
1168         req = mdt_info_req(info);
1169         rc = target_handle_connect(req);
1170         if (rc == 0) {
1171                 LASSERT(req->rq_export != NULL);
1172                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1173                 rc = mdt_init_sec_level(info);
1174                 if (rc == 0)
1175                         rc = mdt_init_idmap(info);
1176                 if (rc != 0)
1177                         obd_disconnect(class_export_get(req->rq_export));
1178         } else {
1179                 rc = err_serious(rc);
1180         }
1181         return rc;
1182 }
1183
1184 static int mdt_disconnect(struct mdt_thread_info *info)
1185 {
1186         int rc;
1187         ENTRY;
1188
1189         rc = target_handle_disconnect(mdt_info_req(info));
1190         if (rc)
1191                 rc = err_serious(rc);
1192         RETURN(rc);
1193 }
1194
1195 static int mdt_sendpage(struct mdt_thread_info *info,
1196                         struct lu_rdpg *rdpg)
1197 {
1198         struct ptlrpc_request   *req = mdt_info_req(info);
1199         struct obd_export       *exp = req->rq_export;
1200         struct ptlrpc_bulk_desc *desc;
1201         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1202         int                      tmpcount;
1203         int                      tmpsize;
1204         int                      timeout;
1205         int                      i;
1206         int                      rc;
1207         ENTRY;
1208
1209         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1210                                     MDS_BULK_PORTAL);
1211         if (desc == NULL)
1212                 RETURN(-ENOMEM);
1213
1214         for (i = 0, tmpcount = rdpg->rp_count;
1215                 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1216                 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1217                 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1218         }
1219
1220         LASSERT(desc->bd_nob == rdpg->rp_count);
1221         rc = sptlrpc_svc_wrap_bulk(req, desc);
1222         if (rc)
1223                 GOTO(free_desc, rc);
1224
1225         rc = ptlrpc_start_bulk_transfer(desc);
1226         if (rc)
1227                 GOTO(free_desc, rc);
1228
1229         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1230                 GOTO(abort_bulk, rc = 0);
1231
1232         timeout = (int) req->rq_deadline - cfs_time_current_sec();
1233         if (timeout < 0)
1234                 CERROR("Req deadline already passed %lu (now: %lu)\n",
1235                        req->rq_deadline, cfs_time_current_sec());
1236         *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1237                                     cfs_time_seconds(1), NULL, NULL);
1238         rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
1239                           exp->exp_failed || exp->exp_abort_active_req, lwi);
1240         LASSERT (rc == 0 || rc == -ETIMEDOUT);
1241
1242         if (rc == 0) {
1243                 if (desc->bd_success &&
1244                     desc->bd_nob_transferred == rdpg->rp_count)
1245                         GOTO(free_desc, rc);
1246
1247                 rc = -ETIMEDOUT;
1248                 if (exp->exp_abort_active_req || exp->exp_failed)
1249                         GOTO(abort_bulk, rc);
1250         }
1251
1252         DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1253                   (rc == -ETIMEDOUT) ? "timeout" : "network error",
1254                   desc->bd_nob_transferred, rdpg->rp_count,
1255                   exp->exp_client_uuid.uuid,
1256                   exp->exp_connection->c_remote_uuid.uuid);
1257
1258         class_fail_export(exp);
1259
1260         EXIT;
1261 abort_bulk:
1262         ptlrpc_abort_bulk(desc);
1263 free_desc:
1264         ptlrpc_free_bulk(desc);
1265         return rc;
1266 }
1267
1268 #ifdef HAVE_SPLIT_SUPPORT
1269 /*
1270  * Retrieve dir entry from the page and insert it to the slave object, actually,
1271  * this should be in osd layer, but since it will not in the final product, so
1272  * just do it here and do not define more moo api anymore for this.
1273  */
1274 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1275                               int size)
1276 {
1277         struct mdt_object *object = info->mti_object;
1278         struct lu_fid *lf = &info->mti_tmp_fid2;
1279         struct md_attr *ma = &info->mti_attr;
1280         struct lu_dirpage *dp;
1281         struct lu_dirent *ent;
1282         int rc = 0, offset = 0;
1283         ENTRY;
1284
1285         /* Make sure we have at least one entry. */
1286         if (size == 0)
1287                 RETURN(-EINVAL);
1288
1289         /*
1290          * Disable trans for this name insert, since it will include many trans
1291          * for this.
1292          */
1293         info->mti_no_need_trans = 1;
1294         /*
1295          * When write_dir_page, no need update parent's ctime,
1296          * and no permission check for name_insert.
1297          */
1298         ma->ma_attr.la_ctime = 0;
1299         ma->ma_attr.la_valid = LA_MODE;
1300         ma->ma_valid = MA_INODE;
1301
1302         cfs_kmap(page);
1303         dp = page_address(page);
1304         offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1305
1306         for (ent = lu_dirent_start(dp); ent != NULL;
1307              ent = lu_dirent_next(ent)) {
1308                 struct lu_name *lname;
1309                 char *name;
1310
1311                 if (le16_to_cpu(ent->lde_namelen) == 0)
1312                         continue;
1313
1314                 fid_le_to_cpu(lf, &ent->lde_fid);
1315                 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1316                         ma->ma_attr.la_mode = S_IFDIR;
1317                 else
1318                         ma->ma_attr.la_mode = 0;
1319                 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1320                 if (name == NULL)
1321                         GOTO(out, rc = -ENOMEM);
1322
1323                 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1324                 lname = mdt_name(info->mti_env, name,
1325                                  le16_to_cpu(ent->lde_namelen));
1326                 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1327                 rc = mdo_name_insert(info->mti_env,
1328                                      md_object_next(&object->mot_obj),
1329                                      lname, lf, ma);
1330                 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1331                 if (rc) {
1332                         CERROR("Can't insert %*.*s, rc %d\n",
1333                                le16_to_cpu(ent->lde_namelen),
1334                                le16_to_cpu(ent->lde_namelen),
1335                                ent->lde_name, rc);
1336                         GOTO(out, rc);
1337                 }
1338
1339                 offset += lu_dirent_size(ent);
1340                 if (offset >= size)
1341                         break;
1342         }
1343         EXIT;
1344 out:
1345         cfs_kunmap(page);
1346         return rc;
1347 }
1348
1349 static int mdt_bulk_timeout(void *data)
1350 {
1351         ENTRY;
1352
1353         CERROR("mdt bulk transfer timeout \n");
1354
1355         RETURN(1);
1356 }
1357
1358 static int mdt_writepage(struct mdt_thread_info *info)
1359 {
1360         struct ptlrpc_request   *req = mdt_info_req(info);
1361         struct mdt_body         *reqbody;
1362         struct l_wait_info      *lwi;
1363         struct ptlrpc_bulk_desc *desc;
1364         struct page             *page;
1365         int                rc;
1366         ENTRY;
1367
1368
1369         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1370         if (reqbody == NULL)
1371                 RETURN(err_serious(-EFAULT));
1372
1373         desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1374         if (desc == NULL)
1375                 RETURN(err_serious(-ENOMEM));
1376
1377         /* allocate the page for the desc */
1378         page = cfs_alloc_page(CFS_ALLOC_STD);
1379         if (page == NULL)
1380                 GOTO(desc_cleanup, rc = -ENOMEM);
1381
1382         CDEBUG(D_INFO, "Received page offset %d size %d \n",
1383                (int)reqbody->size, (int)reqbody->nlink);
1384
1385         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1386                               (int)reqbody->nlink);
1387
1388         rc = sptlrpc_svc_prep_bulk(req, desc);
1389         if (rc != 0)
1390                 GOTO(cleanup_page, rc);
1391         /*
1392          * Check if client was evicted while we were doing i/o before touching
1393          * network.
1394          */
1395         OBD_ALLOC_PTR(lwi);
1396         if (!lwi)
1397                 GOTO(cleanup_page, rc = -ENOMEM);
1398
1399         if (desc->bd_export->exp_failed)
1400                 rc = -ENOTCONN;
1401         else
1402                 rc = ptlrpc_start_bulk_transfer (desc);
1403         if (rc == 0) {
1404                 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1405                                             mdt_bulk_timeout, desc);
1406                 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1407                                   desc->bd_export->exp_failed, lwi);
1408                 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1409                 if (rc == -ETIMEDOUT) {
1410                         DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1411                         ptlrpc_abort_bulk(desc);
1412                 } else if (desc->bd_export->exp_failed) {
1413                         DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1414                         rc = -ENOTCONN;
1415                         ptlrpc_abort_bulk(desc);
1416                 } else if (!desc->bd_success ||
1417                            desc->bd_nob_transferred != desc->bd_nob) {
1418                         DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1419                                   desc->bd_success ?
1420                                   "truncated" : "network error on",
1421                                   desc->bd_nob_transferred, desc->bd_nob);
1422                         /* XXX should this be a different errno? */
1423                         rc = -ETIMEDOUT;
1424                 }
1425         } else {
1426                 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1427         }
1428         if (rc)
1429                 GOTO(cleanup_lwi, rc);
1430         rc = mdt_write_dir_page(info, page, reqbody->nlink);
1431
1432 cleanup_lwi:
1433         OBD_FREE_PTR(lwi);
1434 cleanup_page:
1435         cfs_free_page(page);
1436 desc_cleanup:
1437         ptlrpc_free_bulk(desc);
1438         RETURN(rc);
1439 }
1440 #endif
1441
1442 static int mdt_readpage(struct mdt_thread_info *info)
1443 {
1444         struct mdt_object *object = info->mti_object;
1445         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1446         struct mdt_body   *reqbody;
1447         struct mdt_body   *repbody;
1448         int                rc;
1449         int                i;
1450         ENTRY;
1451
1452         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1453                 RETURN(err_serious(-ENOMEM));
1454
1455         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1456         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1457         if (reqbody == NULL || repbody == NULL)
1458                 RETURN(err_serious(-EFAULT));
1459
1460         /*
1461          * prepare @rdpg before calling lower layers and transfer itself. Here
1462          * reqbody->size contains offset of where to start to read and
1463          * reqbody->nlink contains number bytes to read.
1464          */
1465         rdpg->rp_hash = reqbody->size;
1466         if (rdpg->rp_hash != reqbody->size) {
1467                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1468                        rdpg->rp_hash, reqbody->size);
1469                 RETURN(-EFAULT);
1470         }
1471
1472         rdpg->rp_attrs = reqbody->mode;
1473         rdpg->rp_count  = reqbody->nlink;
1474         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1475         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1476         if (rdpg->rp_pages == NULL)
1477                 RETURN(-ENOMEM);
1478
1479         for (i = 0; i < rdpg->rp_npages; ++i) {
1480                 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1481                 if (rdpg->rp_pages[i] == NULL)
1482                         GOTO(free_rdpg, rc = -ENOMEM);
1483         }
1484
1485         /* call lower layers to fill allocated pages with directory data */
1486         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1487         if (rc)
1488                 GOTO(free_rdpg, rc);
1489
1490         /* send pages to client */
1491         rc = mdt_sendpage(info, rdpg);
1492
1493         EXIT;
1494 free_rdpg:
1495
1496         for (i = 0; i < rdpg->rp_npages; i++)
1497                 if (rdpg->rp_pages[i] != NULL)
1498                         cfs_free_page(rdpg->rp_pages[i]);
1499         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1500
1501         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1502                 RETURN(0);
1503
1504         return rc;
1505 }
1506
1507 static int mdt_reint_internal(struct mdt_thread_info *info,
1508                               struct mdt_lock_handle *lhc,
1509                               __u32 op)
1510 {
1511         struct req_capsule      *pill = info->mti_pill;
1512         struct mdt_device       *mdt = info->mti_mdt;
1513         struct md_quota         *mq = md_quota(info->mti_env);
1514         struct mdt_body         *repbody;
1515         int                      rc = 0;
1516         ENTRY;
1517
1518         /* pack reply */
1519         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1520                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1521                                      mdt->mdt_max_mdsize);
1522         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1523                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1524                                      mdt->mdt_max_cookiesize);
1525
1526         rc = req_capsule_server_pack(pill);
1527         if (rc != 0) {
1528                 CERROR("Can't pack response, rc %d\n", rc);
1529                 RETURN(err_serious(rc));
1530         }
1531
1532         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1533                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1534                 LASSERT(repbody);
1535                 repbody->eadatasize = 0;
1536                 repbody->aclsize = 0;
1537         }
1538
1539         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1540                 GOTO(out_shrink, rc = err_serious(-EFAULT));
1541
1542         rc = mdt_reint_unpack(info, op);
1543         if (rc != 0) {
1544                 CERROR("Can't unpack reint, rc %d\n", rc);
1545                 GOTO(out_shrink, rc = err_serious(rc));
1546         }
1547
1548         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1549
1550         /* for replay no cookkie / lmm need, because client have this already */
1551         if (info->mti_spec.no_create == 1)  {
1552                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1553                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1554
1555                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1556                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1557                                              0);
1558         }
1559
1560         rc = mdt_init_ucred_reint(info);
1561         if (rc)
1562                 GOTO(out_shrink, rc);
1563
1564         rc = mdt_fix_attr_ucred(info, op);
1565         if (rc != 0)
1566                 GOTO(out_ucred, rc = err_serious(rc));
1567
1568         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1569                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1570                 GOTO(out_ucred, rc);
1571         }
1572         mq->mq_exp = info->mti_exp;
1573         rc = mdt_reint_rec(info, lhc);
1574         EXIT;
1575 out_ucred:
1576         mdt_exit_ucred(info);
1577 out_shrink:
1578         mdt_shrink_reply(info);
1579         return rc;
1580 }
1581
1582 static long mdt_reint_opcode(struct mdt_thread_info *info,
1583                              const struct req_format **fmt)
1584 {
1585         struct mdt_rec_reint *rec;
1586         long opc;
1587
1588         opc = err_serious(-EFAULT);
1589         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1590         if (rec != NULL) {
1591                 opc = rec->rr_opcode;
1592                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1593                 if (opc < REINT_MAX && fmt[opc] != NULL)
1594                         req_capsule_extend(info->mti_pill, fmt[opc]);
1595                 else {
1596                         CERROR("Unsupported opc: %ld\n", opc);
1597                         opc = err_serious(opc);
1598                 }
1599         }
1600         return opc;
1601 }
1602
1603 static int mdt_reint(struct mdt_thread_info *info)
1604 {
1605         long opc;
1606         int  rc;
1607
1608         static const struct req_format *reint_fmts[REINT_MAX] = {
1609                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1610                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1611                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1612                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1613                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1614                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1615                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1616         };
1617
1618         ENTRY;
1619
1620         opc = mdt_reint_opcode(info, reint_fmts);
1621         if (opc >= 0) {
1622                 /*
1623                  * No lock possible here from client to pass it to reint code
1624                  * path.
1625                  */
1626                 rc = mdt_reint_internal(info, NULL, opc);
1627         } else {
1628                 rc = opc;
1629         }
1630
1631         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1632         RETURN(rc);
1633 }
1634
1635 /* this should sync the whole device */
1636 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1637 {
1638         struct dt_device *dt = mdt->mdt_bottom;
1639         int rc;
1640         ENTRY;
1641
1642         rc = dt->dd_ops->dt_sync(env, dt);
1643         RETURN(rc);
1644 }
1645
1646 /* this should sync this object */
1647 static int mdt_object_sync(struct mdt_thread_info *info)
1648 {
1649         struct md_object *next;
1650         int rc;
1651         ENTRY;
1652
1653         if (!mdt_object_exists(info->mti_object)) {
1654                 CWARN("Non existing object  "DFID"!\n",
1655                       PFID(mdt_object_fid(info->mti_object)));
1656                 RETURN(-ESTALE);
1657         }
1658         next = mdt_object_child(info->mti_object);
1659         rc = mo_object_sync(info->mti_env, next);
1660
1661         RETURN(rc);
1662 }
1663
1664 static int mdt_sync(struct mdt_thread_info *info)
1665 {
1666         struct req_capsule *pill = info->mti_pill;
1667         struct mdt_body *body;
1668         int rc;
1669         ENTRY;
1670
1671         /* The fid may be zero, so we req_capsule_set manually */
1672         req_capsule_set(pill, &RQF_MDS_SYNC);
1673
1674         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1675         if (body == NULL)
1676                 RETURN(err_serious(-EINVAL));
1677
1678         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1679                 RETURN(err_serious(-ENOMEM));
1680
1681         if (fid_seq(&body->fid1) == 0) {
1682                 /* sync the whole device */
1683                 rc = req_capsule_server_pack(pill);
1684                 if (rc == 0)
1685                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1686                 else
1687                         rc = err_serious(rc);
1688         } else {
1689                 /* sync an object */
1690                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1691                 if (rc == 0) {
1692                         rc = mdt_object_sync(info);
1693                         if (rc == 0) {
1694                                 struct md_object *next;
1695                                 const struct lu_fid *fid;
1696                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1697
1698                                 next = mdt_object_child(info->mti_object);
1699                                 info->mti_attr.ma_need = MA_INODE;
1700                                 info->mti_attr.ma_valid = 0;
1701                                 rc = mo_attr_get(info->mti_env, next,
1702                                                  &info->mti_attr);
1703                                 if (rc == 0) {
1704                                         body = req_capsule_server_get(pill,
1705                                                                 &RMF_MDT_BODY);
1706                                         fid = mdt_object_fid(info->mti_object);
1707                                         mdt_pack_attr2body(info, body, la, fid);
1708                                 }
1709                         }
1710                 } else
1711                         rc = err_serious(rc);
1712         }
1713         RETURN(rc);
1714 }
1715
1716 #ifdef HAVE_QUOTA_SUPPORT
1717 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1718 {
1719         struct obd_quotactl *oqctl;
1720         struct req_capsule *pill = info->mti_pill;
1721         struct obd_export *exp = info->mti_exp;
1722         struct md_quota *mq = md_quota(info->mti_env);
1723         struct md_device *next = info->mti_mdt->mdt_child;
1724         int rc;
1725         ENTRY;
1726
1727         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1728         if (oqctl == NULL)
1729                 RETURN(-EPROTO);
1730
1731         /* remote client has no permission for quotacheck */
1732         if (unlikely(exp_connect_rmtclient(exp)))
1733                 RETURN(-EPERM);
1734
1735         rc = req_capsule_server_pack(pill);
1736         if (rc)
1737                 RETURN(rc);
1738
1739         mq->mq_exp = exp;
1740         rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1741                                                oqctl->qc_type);
1742         RETURN(rc);
1743 }
1744
1745 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1746 {
1747         struct obd_quotactl *oqctl, *repoqc;
1748         struct req_capsule *pill = info->mti_pill;
1749         struct obd_export *exp = info->mti_exp;
1750         struct md_quota *mq = md_quota(info->mti_env);
1751         struct md_device *next = info->mti_mdt->mdt_child;
1752         const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1753         int id, rc;
1754         ENTRY;
1755
1756         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1757         if (oqctl == NULL)
1758                 RETURN(-EPROTO);
1759
1760         id = oqctl->qc_id;
1761         if (exp_connect_rmtclient(exp)) {
1762                 struct ptlrpc_request *req = mdt_info_req(info);
1763                 struct mdt_export_data *med = mdt_req2med(req);
1764                 struct lustre_idmap_table *idmap = med->med_idmap;
1765
1766                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1767                              oqctl->qc_cmd != Q_GETINFO))
1768                         RETURN(-EPERM);
1769
1770
1771                 if (oqctl->qc_type == USRQUOTA)
1772                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1773                                                      oqctl->qc_id);
1774                 else if (oqctl->qc_type == GRPQUOTA)
1775                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1776                                                      oqctl->qc_id);
1777                 else
1778                         RETURN(-EINVAL);
1779
1780                 if (id == CFS_IDMAP_NOTFOUND) {
1781                         CDEBUG(D_QUOTA, "no mapping for id %u\n",
1782                                oqctl->qc_id);
1783                         RETURN(-EACCES);
1784                 }
1785         }
1786
1787         rc = req_capsule_server_pack(pill);
1788         if (rc)
1789                 RETURN(rc);
1790
1791         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1792         LASSERT(repoqc != NULL);
1793
1794         mq->mq_exp = exp;
1795         switch (oqctl->qc_cmd) {
1796         case Q_QUOTAON:
1797                 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1798                 break;
1799         case Q_QUOTAOFF:
1800                 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1801                 break;
1802         case Q_SETINFO:
1803                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1804                                       &oqctl->qc_dqinfo);
1805                 break;
1806         case Q_GETINFO:
1807                 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1808                                       &oqctl->qc_dqinfo);
1809                 break;
1810         case Q_SETQUOTA:
1811                 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1812                                        &oqctl->qc_dqblk);
1813                 break;
1814         case Q_GETQUOTA:
1815                 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1816                                        &oqctl->qc_dqblk);
1817                 break;
1818         case Q_GETOINFO:
1819                 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1820                                        &oqctl->qc_dqinfo);
1821                 break;
1822         case Q_GETOQUOTA:
1823                 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1824                                         &oqctl->qc_dqblk);
1825                 break;
1826         case LUSTRE_Q_INVALIDATE:
1827                 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1828                 break;
1829         case LUSTRE_Q_FINVALIDATE:
1830                 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1831                 break;
1832         default:
1833                 CERROR("unsupported mdt_quotactl command: %d\n",
1834                        oqctl->qc_cmd);
1835                 RETURN(-EFAULT);
1836         }
1837
1838         *repoqc = *oqctl;
1839         RETURN(rc);
1840 }
1841 #endif
1842
1843
1844 /*
1845  * OBD PING and other handlers.
1846  */
1847 static int mdt_obd_ping(struct mdt_thread_info *info)
1848 {
1849         int rc;
1850         ENTRY;
1851
1852         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1853
1854         rc = target_handle_ping(mdt_info_req(info));
1855         if (rc < 0)
1856                 rc = err_serious(rc);
1857         RETURN(rc);
1858 }
1859
1860 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1861 {
1862         return err_serious(-EOPNOTSUPP);
1863 }
1864
1865 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1866 {
1867         return err_serious(-EOPNOTSUPP);
1868 }
1869
1870
1871 /*
1872  * LLOG handlers.
1873  */
1874
1875 /** clone llog ctxt from child (mdd)
1876  * This allows remote llog (replicator) access.
1877  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1878  * context was originally set up, or we can handle them directly.
1879  * I choose the latter, but that means I need any llog
1880  * contexts set up by child to be accessable by the mdt.  So we clone the
1881  * context into our context list here.
1882  */
1883 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1884                                int idx)
1885 {
1886         struct md_device  *next = mdt->mdt_child;
1887         struct llog_ctxt *ctxt;
1888         int rc;
1889
1890         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1891                 return 0;
1892
1893         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1894         if (rc || ctxt == NULL) {
1895                 CERROR("Can't get mdd ctxt %d\n", rc);
1896                 return rc;
1897         }
1898
1899         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1900         if (rc)
1901                 CERROR("Can't set mdt ctxt %d\n", rc);
1902
1903         return rc;
1904 }
1905
1906 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1907                                  struct mdt_device *mdt, int idx)
1908 {
1909         struct llog_ctxt *ctxt;
1910
1911         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1912         if (ctxt == NULL)
1913                 return 0;
1914         /* Put once for the get we just did, and once for the clone */
1915         llog_ctxt_put(ctxt);
1916         llog_ctxt_put(ctxt);
1917         return 0;
1918 }
1919
1920 static int mdt_llog_create(struct mdt_thread_info *info)
1921 {
1922         int rc;
1923
1924         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1925         rc = llog_origin_handle_create(mdt_info_req(info));
1926         return (rc < 0 ? err_serious(rc) : rc);
1927 }
1928
1929 static int mdt_llog_destroy(struct mdt_thread_info *info)
1930 {
1931         int rc;
1932
1933         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1934         rc = llog_origin_handle_destroy(mdt_info_req(info));
1935         return (rc < 0 ? err_serious(rc) : rc);
1936 }
1937
1938 static int mdt_llog_read_header(struct mdt_thread_info *info)
1939 {
1940         int rc;
1941
1942         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1943         rc = llog_origin_handle_read_header(mdt_info_req(info));
1944         return (rc < 0 ? err_serious(rc) : rc);
1945 }
1946
1947 static int mdt_llog_next_block(struct mdt_thread_info *info)
1948 {
1949         int rc;
1950
1951         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1952         rc = llog_origin_handle_next_block(mdt_info_req(info));
1953         return (rc < 0 ? err_serious(rc) : rc);
1954 }
1955
1956 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1957 {
1958         int rc;
1959
1960         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1961         rc = llog_origin_handle_prev_block(mdt_info_req(info));
1962         return (rc < 0 ? err_serious(rc) : rc);
1963 }
1964
1965
1966 /*
1967  * DLM handlers.
1968  */
1969 static struct ldlm_callback_suite cbs = {
1970         .lcs_completion = ldlm_server_completion_ast,
1971         .lcs_blocking   = ldlm_server_blocking_ast,
1972         .lcs_glimpse    = NULL
1973 };
1974
1975 static int mdt_enqueue(struct mdt_thread_info *info)
1976 {
1977         struct ptlrpc_request *req;
1978         __u64 req_bits;
1979         int rc;
1980
1981         /*
1982          * info->mti_dlm_req already contains swapped and (if necessary)
1983          * converted dlm request.
1984          */
1985         LASSERT(info->mti_dlm_req != NULL);
1986
1987         req = mdt_info_req(info);
1988
1989         /*
1990          * Lock without inodebits makes no sense and will oops later in
1991          * ldlm. Let's check it now to see if we have wrong lock from client or
1992          * bits get corrupted somewhere in mdt_intent_policy().
1993          */
1994         req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1995         /* This is disabled because we need to support liblustre flock.
1996          * LASSERT(req_bits != 0);
1997          */
1998
1999         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2000                                   req, info->mti_dlm_req, &cbs);
2001         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2002         return rc ? err_serious(rc) : req->rq_status;
2003 }
2004
2005 static int mdt_convert(struct mdt_thread_info *info)
2006 {
2007         int rc;
2008         struct ptlrpc_request *req;
2009
2010         LASSERT(info->mti_dlm_req);
2011         req = mdt_info_req(info);
2012         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2013         return rc ? err_serious(rc) : req->rq_status;
2014 }
2015
2016 static int mdt_bl_callback(struct mdt_thread_info *info)
2017 {
2018         CERROR("bl callbacks should not happen on MDS\n");
2019         LBUG();
2020         return err_serious(-EOPNOTSUPP);
2021 }
2022
2023 static int mdt_cp_callback(struct mdt_thread_info *info)
2024 {
2025         CERROR("cp callbacks should not happen on MDS\n");
2026         LBUG();
2027         return err_serious(-EOPNOTSUPP);
2028 }
2029
2030 /*
2031  * sec context handlers
2032  */
2033 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2034 {
2035         int rc;
2036
2037         rc = mdt_handle_idmap(info);
2038
2039         if (unlikely(rc)) {
2040                 struct ptlrpc_request *req = mdt_info_req(info);
2041                 __u32                  opc;
2042
2043                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2044                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2045                         sptlrpc_svc_ctx_invalidate(req);
2046         }
2047
2048         OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2049
2050         return rc;
2051 }
2052
2053 static struct mdt_object *mdt_obj(struct lu_object *o)
2054 {
2055         LASSERT(lu_device_is_mdt(o->lo_dev));
2056         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2057 }
2058
2059 struct mdt_object *mdt_object_find(const struct lu_env *env,
2060                                    struct mdt_device *d,
2061                                    const struct lu_fid *f)
2062 {
2063         struct lu_object *o;
2064         struct mdt_object *m;
2065         ENTRY;
2066
2067         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2068         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2069         if (unlikely(IS_ERR(o)))
2070                 m = (struct mdt_object *)o;
2071         else
2072                 m = mdt_obj(o);
2073         RETURN(m);
2074 }
2075
2076 /**
2077  * Asyncronous commit for mdt device.
2078  *
2079  * Pass asynchonous commit call down the MDS stack.
2080  *
2081  * \param env environment
2082  * \param mdt the mdt device
2083  */
2084 static void mdt_device_commit_async(const struct lu_env *env,
2085                                     struct mdt_device *mdt)
2086 {
2087         struct dt_device *dt = mdt->mdt_bottom;
2088         int rc;
2089
2090         rc = dt->dd_ops->dt_commit_async(env, dt);
2091         if (unlikely(rc != 0))
2092                 CWARN("async commit start failed with rc = %d", rc);
2093 }
2094
2095 /**
2096  * Mark the lock as "synchonous".
2097  *
2098  * Mark the lock to deffer transaction commit to the unlock time.
2099  *
2100  * \param lock the lock to mark as "synchonous"
2101  *
2102  * \see mdt_is_lock_sync
2103  * \see mdt_save_lock
2104  */
2105 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2106 {
2107         lock->l_ast_data = (void*)1;
2108 }
2109
2110 /**
2111  * Check whehter the lock "synchonous" or not.
2112  *
2113  * \param lock the lock to check
2114  * \retval 1 the lock is "synchonous"
2115  * \retval 0 the lock isn't "synchronous"
2116  *
2117  * \see mdt_set_lock_sync
2118  * \see mdt_save_lock
2119  */
2120 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2121 {
2122         return lock->l_ast_data != NULL;
2123 }
2124
2125 /**
2126  * Blocking AST for mdt locks.
2127  *
2128  * Starts transaction commit if in case of COS lock conflict or
2129  * deffers such a commit to the mdt_save_lock.
2130  *
2131  * \param lock the lock which blocks a request or cancelling lock
2132  * \param desc unused
2133  * \param data unused
2134  * \param flag indicates whether this cancelling or blocking callback
2135  * \retval 0
2136  * \see ldlm_blocking_ast_nocheck
2137  */
2138 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2139                      void *data, int flag)
2140 {
2141         struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2142         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2143         int rc;
2144         ENTRY;
2145
2146         if (flag == LDLM_CB_CANCELING)
2147                 RETURN(0);
2148         lock_res_and_lock(lock);
2149         if (lock->l_blocking_ast != mdt_blocking_ast) {
2150                 unlock_res_and_lock(lock);
2151                 RETURN(0);
2152         }
2153         if (mdt_cos_is_enabled(mdt) &&
2154             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2155             lock->l_blocking_lock != NULL &&
2156             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2157                 mdt_set_lock_sync(lock);
2158         }
2159         rc = ldlm_blocking_ast_nocheck(lock);
2160
2161         /* There is no lock conflict if l_blocking_lock == NULL,
2162          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2163          * when the last reference to a local lock was released */
2164         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2165                 struct lu_env env;
2166
2167                 rc = lu_env_init(&env, LCT_MD_THREAD);
2168                 if (unlikely(rc != 0))
2169                         CWARN("lu_env initialization failed with rc = %d,"
2170                               "cannot start asynchronous commit\n", rc);
2171                 else
2172                         mdt_device_commit_async(&env, mdt);
2173                 lu_env_fini(&env);
2174         }
2175         RETURN(rc);
2176 }
2177
2178 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2179                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2180 {
2181         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2182         ldlm_policy_data_t *policy = &info->mti_policy;
2183         struct ldlm_res_id *res_id = &info->mti_res_id;
2184         int rc;
2185         ENTRY;
2186
2187         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2188         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2189         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2190         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2191
2192         if (mdt_object_exists(o) < 0) {
2193                 if (locality == MDT_CROSS_LOCK) {
2194                         /* cross-ref object fix */
2195                         ibits &= ~MDS_INODELOCK_UPDATE;
2196                         ibits |= MDS_INODELOCK_LOOKUP;
2197                 } else {
2198                         LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2199                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2200                 }
2201                 /* No PDO lock on remote object */
2202                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2203         }
2204
2205         if (lh->mlh_type == MDT_PDO_LOCK) {
2206                 /* check for exists after object is locked */
2207                 if (mdt_object_exists(o) == 0) {
2208                         /* Non-existent object shouldn't have PDO lock */
2209                         RETURN(-ESTALE);
2210                 } else {
2211                         /* Non-dir object shouldn't have PDO lock */
2212                         LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2213                 }
2214         }
2215
2216         memset(policy, 0, sizeof(*policy));
2217         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2218
2219         /*
2220          * Take PDO lock on whole directory and build correct @res_id for lock
2221          * on part of directory.
2222          */
2223         if (lh->mlh_pdo_hash != 0) {
2224                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2225                 mdt_lock_pdo_mode(info, o, lh);
2226                 if (lh->mlh_pdo_mode != LCK_NL) {
2227                         /*
2228                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2229                          * is never going to be sent to client and we do not
2230                          * want it slowed down due to possible cancels.
2231                          */
2232                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2233                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2234                                           policy, res_id, LDLM_FL_ATOMIC_CB,
2235                                           &info->mti_exp->exp_handle.h_cookie);
2236                         if (unlikely(rc))
2237                                 RETURN(rc);
2238                 }
2239
2240                 /*
2241                  * Finish res_id initializing by name hash marking part of
2242                  * directory which is taking modification.
2243                  */
2244                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2245         }
2246
2247         policy->l_inodebits.bits = ibits;
2248
2249         /*
2250          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2251          * going to be sent to client. If it is - mdt_intent_policy() path will
2252          * fix it up and turn FL_LOCAL flag off.
2253          */
2254         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2255                           res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2256                           &info->mti_exp->exp_handle.h_cookie);
2257         if (rc)
2258                 GOTO(out, rc);
2259
2260 out:
2261         if (rc)
2262                 mdt_object_unlock(info, o, lh, 1);
2263
2264
2265         RETURN(rc);
2266 }
2267
2268 /**
2269  * Save a lock within request object.
2270  *
2271  * Keep the lock referenced until whether client ACK or transaction
2272  * commit happens or release the lock immediately depending on input
2273  * parameters. If COS is ON, a write lock is converted to COS lock
2274  * before saving.
2275  *
2276  * \param info thead info object
2277  * \param h lock handle
2278  * \param mode lock mode
2279  * \param decref force immediate lock releasing
2280  */
2281 static
2282 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2283                    ldlm_mode_t mode, int decref)
2284 {
2285         ENTRY;
2286
2287         if (lustre_handle_is_used(h)) {
2288                 if (decref || !info->mti_has_trans ||
2289                     !(mode & (LCK_PW | LCK_EX))){
2290                         mdt_fid_unlock(h, mode);
2291                 } else {
2292                         struct mdt_device *mdt = info->mti_mdt;
2293                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2294                         struct ptlrpc_request *req = mdt_info_req(info);
2295                         int no_ack = 0;
2296
2297                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2298                                  h->cookie);
2299                         CDEBUG(D_HA, "request = %p reply state = %p"
2300                                " transno = "LPD64"\n",
2301                                req, req->rq_reply_state, req->rq_transno);
2302                         if (mdt_cos_is_enabled(mdt)) {
2303                                 no_ack = 1;
2304                                 ldlm_lock_downgrade(lock, LCK_COS);
2305                                 mode = LCK_COS;
2306                         }
2307                         ptlrpc_save_lock(req, h, mode, no_ack);
2308                         if (mdt_is_lock_sync(lock)) {
2309                                 CDEBUG(D_HA, "found sync-lock,"
2310                                        " async commit started\n");
2311                                 mdt_device_commit_async(info->mti_env,
2312                                                         mdt);
2313                         }
2314                         LDLM_LOCK_PUT(lock);
2315                 }
2316                 h->cookie = 0ull;
2317         }
2318
2319         EXIT;
2320 }
2321
2322 /**
2323  * Unlock mdt object.
2324  *
2325  * Immeditely release the regular lock and the PDO lock or save the
2326  * lock in reqeuest and keep them referenced until client ACK or
2327  * transaction commit.
2328  *
2329  * \param info thread info object
2330  * \param o mdt object
2331  * \param lh mdt lock handle referencing regular and PDO locks
2332  * \param decref force immediate lock releasing
2333  */
2334 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2335                        struct mdt_lock_handle *lh, int decref)
2336 {
2337         ENTRY;
2338
2339         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2340         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2341
2342         EXIT;
2343 }
2344
2345 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2346                                         const struct lu_fid *f,
2347                                         struct mdt_lock_handle *lh,
2348                                         __u64 ibits)
2349 {
2350         struct mdt_object *o;
2351
2352         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2353         if (!IS_ERR(o)) {
2354                 int rc;
2355
2356                 rc = mdt_object_lock(info, o, lh, ibits,
2357                                      MDT_LOCAL_LOCK);
2358                 if (rc != 0) {
2359                         mdt_object_put(info->mti_env, o);
2360                         o = ERR_PTR(rc);
2361                 }
2362         }
2363         return o;
2364 }
2365
2366 void mdt_object_unlock_put(struct mdt_thread_info * info,
2367                            struct mdt_object * o,
2368                            struct mdt_lock_handle *lh,
2369                            int decref)
2370 {
2371         mdt_object_unlock(info, o, lh, decref);
2372         mdt_object_put(info->mti_env, o);
2373 }
2374
2375 static struct mdt_handler *mdt_handler_find(__u32 opc,
2376                                             struct mdt_opc_slice *supported)
2377 {
2378         struct mdt_opc_slice *s;
2379         struct mdt_handler   *h;
2380
2381         h = NULL;
2382         for (s = supported; s->mos_hs != NULL; s++) {
2383                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2384                         h = s->mos_hs + (opc - s->mos_opc_start);
2385                         if (likely(h->mh_opc != 0))
2386                                 LASSERTF(h->mh_opc == opc,
2387                                          "opcode mismatch %d != %d\n",
2388                                          h->mh_opc, opc);
2389                         else
2390                                 h = NULL; /* unsupported opc */
2391                         break;
2392                 }
2393         }
2394         return h;
2395 }
2396
2397 static int mdt_lock_resname_compat(struct mdt_device *m,
2398                                    struct ldlm_request *req)
2399 {
2400         /* XXX something... later. */
2401         return 0;
2402 }
2403
2404 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2405 {
2406         /* XXX something... later. */
2407         return 0;
2408 }
2409
2410 /*
2411  * Generic code handling requests that have struct mdt_body passed in:
2412  *
2413  *  - extract mdt_body from request and save it in @info, if present;
2414  *
2415  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2416  *  @info;
2417  *
2418  *  - if HABEO_CORPUS flag is set for this request type check whether object
2419  *  actually exists on storage (lu_object_exists()).
2420  *
2421  */
2422 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2423 {
2424         const struct mdt_body    *body;
2425         struct mdt_object        *obj;
2426         const struct lu_env      *env;
2427         struct req_capsule       *pill;
2428         int                       rc;
2429         ENTRY;
2430
2431         env = info->mti_env;
2432         pill = info->mti_pill;
2433
2434         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2435         if (body == NULL)
2436                 RETURN(-EFAULT);
2437
2438         if (!(body->valid & OBD_MD_FLID))
2439                 RETURN(0);
2440
2441         if (!fid_is_sane(&body->fid1)) {
2442                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2443                 RETURN(-EINVAL);
2444         }
2445
2446         /*
2447          * Do not get size or any capa fields before we check that request
2448          * contains capa actually. There are some requests which do not, for
2449          * instance MDS_IS_SUBDIR.
2450          */
2451         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2452             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2453                 mdt_set_capainfo(info, 0, &body->fid1,
2454                                  req_capsule_client_get(pill, &RMF_CAPA1));
2455
2456         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2457         if (!IS_ERR(obj)) {
2458                 if ((flags & HABEO_CORPUS) &&
2459                     !mdt_object_exists(obj)) {
2460                         mdt_object_put(env, obj);
2461                         /* for capability renew ENOENT will be handled in
2462                          * mdt_renew_capa */
2463                         if (body->valid & OBD_MD_FLOSSCAPA)
2464                                 rc = 0;
2465                         else
2466                                 rc = -ENOENT;
2467                 } else {
2468                         info->mti_object = obj;
2469                         rc = 0;
2470                 }
2471         } else
2472                 rc = PTR_ERR(obj);
2473
2474         RETURN(rc);
2475 }
2476
2477 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2478 {
2479         struct req_capsule *pill = info->mti_pill;
2480         int rc;
2481         ENTRY;
2482
2483         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2484                 rc = mdt_body_unpack(info, flags);
2485         else
2486                 rc = 0;
2487
2488         if (rc == 0 && (flags & HABEO_REFERO)) {
2489                 struct mdt_device *mdt = info->mti_mdt;
2490
2491                 /* Pack reply. */
2492
2493                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2494                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2495                                              mdt->mdt_max_mdsize);
2496                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2497                         req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2498                                              mdt->mdt_max_cookiesize);
2499
2500                 rc = req_capsule_server_pack(pill);
2501         }
2502         RETURN(rc);
2503 }
2504
2505 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2506 {
2507         struct md_device *next = m->mdt_child;
2508
2509         return next->md_ops->mdo_init_capa_ctxt(env, next,
2510                                                 m->mdt_opts.mo_mds_capa,
2511                                                 m->mdt_capa_timeout,
2512                                                 m->mdt_capa_alg,
2513                                                 m->mdt_capa_keys);
2514 }
2515
2516 /*
2517  * Invoke handler for this request opc. Also do necessary preprocessing
2518  * (according to handler ->mh_flags), and post-processing (setting of
2519  * ->last_{xid,committed}).
2520  */
2521 static int mdt_req_handle(struct mdt_thread_info *info,
2522                           struct mdt_handler *h, struct ptlrpc_request *req)
2523 {
2524         int   rc, serious = 0;
2525         __u32 flags;
2526
2527         ENTRY;
2528
2529         LASSERT(h->mh_act != NULL);
2530         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2531         LASSERT(current->journal_info == NULL);
2532
2533         /*
2534          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2535          * to put same checks into handlers like mdt_close(), mdt_reint(),
2536          * etc., without talking to mdt authors first. Checking same thing
2537          * there again is useless and returning 0 error without packing reply
2538          * is buggy! Handlers either pack reply or return error.
2539          *
2540          * We return 0 here and do not send any reply in order to emulate
2541          * network failure. Do not send any reply in case any of NET related
2542          * fail_id has occured.
2543          */
2544         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2545                 RETURN(0);
2546
2547         rc = 0;
2548         flags = h->mh_flags;
2549         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2550
2551         if (h->mh_fmt != NULL) {
2552                 req_capsule_set(info->mti_pill, h->mh_fmt);
2553                 rc = mdt_unpack_req_pack_rep(info, flags);
2554         }
2555
2556         if (rc == 0 && flags & MUTABOR &&
2557             req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2558                 /* should it be rq_status? */
2559                 rc = -EROFS;
2560
2561         if (rc == 0 && flags & HABEO_CLAVIS) {
2562                 struct ldlm_request *dlm_req;
2563
2564                 LASSERT(h->mh_fmt != NULL);
2565
2566                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2567                 if (dlm_req != NULL) {
2568                         if (info->mti_mdt->mdt_opts.mo_compat_resname)
2569                                 rc = mdt_lock_resname_compat(info->mti_mdt,
2570                                                              dlm_req);
2571                         info->mti_dlm_req = dlm_req;
2572                 } else {
2573                         rc = -EFAULT;
2574                 }
2575         }
2576
2577         /* capability setting changed via /proc, needs reinitialize ctxt */
2578         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2579                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2580                 info->mti_mdt->mdt_capa_conf = 0;
2581         }
2582
2583         if (likely(rc == 0)) {
2584                 /*
2585                  * Process request, there can be two types of rc:
2586                  * 1) errors with msg unpack/pack, other failures outside the
2587                  * operation itself. This is counted as serious errors;
2588                  * 2) errors during fs operation, should be placed in rq_status
2589                  * only
2590                  */
2591                 rc = h->mh_act(info);
2592                 if (rc == 0 &&
2593                     !req->rq_no_reply && req->rq_reply_state == NULL) {
2594                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2595                                   "pack reply and returned 0 error\n",
2596                                   h->mh_name);
2597                         LBUG();
2598                 }
2599                 serious = is_serious(rc);
2600                 rc = clear_serious(rc);
2601         } else
2602                 serious = 1;
2603
2604         req->rq_status = rc;
2605
2606         /*
2607          * ELDLM_* codes which > 0 should be in rq_status only as well as
2608          * all non-serious errors.
2609          */
2610         if (rc > 0 || !serious)
2611                 rc = 0;
2612
2613         LASSERT(current->journal_info == NULL);
2614
2615         if (rc == 0 && (flags & HABEO_CLAVIS) &&
2616             info->mti_mdt->mdt_opts.mo_compat_resname) {
2617                 struct ldlm_reply *dlmrep;
2618
2619                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2620                 if (dlmrep != NULL)
2621                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2622         }
2623
2624         /* If we're DISCONNECTing, the mdt_export_data is already freed */
2625         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2626                 target_committed_to_req(req);
2627
2628         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2629                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2630                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2631                 LBUG();
2632         }
2633
2634         target_send_reply(req, rc, info->mti_fail_id);
2635         RETURN(0);
2636 }
2637
2638 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2639 {
2640         lh->mlh_type = MDT_NUL_LOCK;
2641         lh->mlh_reg_lh.cookie = 0ull;
2642         lh->mlh_reg_mode = LCK_MINMODE;
2643         lh->mlh_pdo_lh.cookie = 0ull;
2644         lh->mlh_pdo_mode = LCK_MINMODE;
2645 }
2646
2647 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2648 {
2649         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2650         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2651 }
2652
2653 /*
2654  * Initialize fields of struct mdt_thread_info. Other fields are left in
2655  * uninitialized state, because it's too expensive to zero out whole
2656  * mdt_thread_info (> 1K) on each request arrival.
2657  */
2658 static void mdt_thread_info_init(struct ptlrpc_request *req,
2659                                  struct mdt_thread_info *info)
2660 {
2661         int i;
2662         struct md_capainfo *ci;
2663
2664         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2665         info->mti_pill = &req->rq_pill;
2666
2667         /* lock handle */
2668         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2669                 mdt_lock_handle_init(&info->mti_lh[i]);
2670
2671         /* mdt device: it can be NULL while CONNECT */
2672         if (req->rq_export) {
2673                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2674                 info->mti_exp = req->rq_export;
2675         } else
2676                 info->mti_mdt = NULL;
2677         info->mti_env = req->rq_svc_thread->t_env;
2678         ci = md_capainfo(info->mti_env);
2679         memset(ci, 0, sizeof *ci);
2680         if (req->rq_export) {
2681                 if (exp_connect_rmtclient(req->rq_export))
2682                         ci->mc_auth = LC_ID_CONVERT;
2683                 else if (req->rq_export->exp_connect_flags &
2684                          OBD_CONNECT_MDS_CAPA)
2685                         ci->mc_auth = LC_ID_PLAIN;
2686                 else
2687                         ci->mc_auth = LC_ID_NONE;
2688         }
2689
2690         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2691         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2692         info->mti_mos[0] = NULL;
2693         info->mti_mos[1] = NULL;
2694         info->mti_mos[2] = NULL;
2695         info->mti_mos[3] = NULL;
2696
2697         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2698         info->mti_body = NULL;
2699         info->mti_object = NULL;
2700         info->mti_dlm_req = NULL;
2701         info->mti_has_trans = 0;
2702         info->mti_no_need_trans = 0;
2703         info->mti_cross_ref = 0;
2704         info->mti_opdata = 0;
2705
2706         /* To not check for split by default. */
2707         info->mti_spec.sp_ck_split = 0;
2708         info->mti_spec.no_create = 0;
2709 }
2710
2711 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2712 {
2713         int i;
2714
2715         req_capsule_fini(info->mti_pill);
2716         if (info->mti_object != NULL) {
2717                 /*
2718                  * freeing an object may lead to OSD level transaction, do not
2719                  * let it mess with MDT. bz19385.
2720                  */
2721                 info->mti_no_need_trans = 1;
2722                 mdt_object_put(info->mti_env, info->mti_object);
2723                 info->mti_object = NULL;
2724         }
2725         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2726                 mdt_lock_handle_fini(&info->mti_lh[i]);
2727         info->mti_env = NULL;
2728 }
2729
2730 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2731                                        struct obd_device *obd, int *process)
2732 {
2733         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2734         case MDS_CONNECT: /* This will never get here, but for completeness. */
2735         case OST_CONNECT: /* This will never get here, but for completeness. */
2736         case MDS_DISCONNECT:
2737         case OST_DISCONNECT:
2738                *process = 1;
2739                RETURN(0);
2740
2741         case MDS_CLOSE:
2742         case MDS_DONE_WRITING:
2743         case MDS_SYNC: /* used in unmounting */
2744         case OBD_PING:
2745         case MDS_REINT:
2746         case SEQ_QUERY:
2747         case FLD_QUERY:
2748         case LDLM_ENQUEUE:
2749                 *process = target_queue_recovery_request(req, obd);
2750                 RETURN(0);
2751
2752         default:
2753                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2754                 *process = -EAGAIN;
2755                 RETURN(0);
2756         }
2757 }
2758
2759 /*
2760  * Handle recovery. Return:
2761  *        +1: continue request processing;
2762  *       -ve: abort immediately with the given error code;
2763  *         0: send reply with error code in req->rq_status;
2764  */
2765 static int mdt_recovery(struct mdt_thread_info *info)
2766 {
2767         struct ptlrpc_request *req = mdt_info_req(info);
2768         struct obd_device *obd;
2769
2770         ENTRY;
2771
2772         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2773         case MDS_CONNECT:
2774         case SEC_CTX_INIT:
2775         case SEC_CTX_INIT_CONT:
2776         case SEC_CTX_FINI:
2777                 {
2778 #if 0
2779                         int rc;
2780
2781                         rc = mdt_handle_idmap(info);
2782                         if (rc)
2783                                 RETURN(rc);
2784                         else
2785 #endif
2786                                 RETURN(+1);
2787                 }
2788         }
2789
2790         if (unlikely(!class_connected_export(req->rq_export))) {
2791                 CERROR("operation %d on unconnected MDS from %s\n",
2792                        lustre_msg_get_opc(req->rq_reqmsg),
2793                        libcfs_id2str(req->rq_peer));
2794                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2795                  * mds_A will get -ENOTCONN(especially for ping req),
2796                  * which will cause that mds_A deactive timeout, then when
2797                  * mds_A cleanup, the cleanup process will be suspended since
2798                  * deactive timeout is not zero.
2799                  */
2800                 req->rq_status = -ENOTCONN;
2801                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2802                 RETURN(0);
2803         }
2804
2805         /* sanity check: if the xid matches, the request must be marked as a
2806          * resent or replayed */
2807         if (req_xid_is_last(req)) {
2808                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2809                       (MSG_RESENT | MSG_REPLAY))) {
2810                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2811                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2812                                   lustre_msg_get_flags(req->rq_reqmsg));
2813                         LBUG();
2814                         req->rq_status = -ENOTCONN;
2815                         RETURN(-ENOTCONN);
2816                 }
2817         }
2818
2819         /* else: note the opposite is not always true; a RESENT req after a
2820          * failover will usually not match the last_xid, since it was likely
2821          * never committed. A REPLAYed request will almost never match the
2822          * last xid, however it could for a committed, but still retained,
2823          * open. */
2824
2825         obd = req->rq_export->exp_obd;
2826
2827         /* Check for aborted recovery... */
2828         if (unlikely(obd->obd_recovering)) {
2829                 int rc;
2830                 int should_process;
2831                 DEBUG_REQ(D_INFO, req, "Got new replay");
2832                 rc = mdt_filter_recovery_request(req, obd, &should_process);
2833                 if (rc != 0 || !should_process)
2834                         RETURN(rc);
2835                 else if (should_process < 0) {
2836                         req->rq_status = should_process;
2837                         rc = ptlrpc_error(req);
2838                         RETURN(rc);
2839                 }
2840         }
2841         RETURN(+1);
2842 }
2843
2844 static int mdt_msg_check_version(struct lustre_msg *msg)
2845 {
2846         int rc;
2847
2848         switch (lustre_msg_get_opc(msg)) {
2849         case MDS_CONNECT:
2850         case MDS_DISCONNECT:
2851         case OBD_PING:
2852         case SEC_CTX_INIT:
2853         case SEC_CTX_INIT_CONT:
2854         case SEC_CTX_FINI:
2855                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2856                 if (rc)
2857                         CERROR("bad opc %u version %08x, expecting %08x\n",
2858                                lustre_msg_get_opc(msg),
2859                                lustre_msg_get_version(msg),
2860                                LUSTRE_OBD_VERSION);
2861                 break;
2862         case MDS_GETSTATUS:
2863         case MDS_GETATTR:
2864         case MDS_GETATTR_NAME:
2865         case MDS_STATFS:
2866         case MDS_READPAGE:
2867         case MDS_WRITEPAGE:
2868         case MDS_IS_SUBDIR:
2869         case MDS_REINT:
2870         case MDS_CLOSE:
2871         case MDS_DONE_WRITING:
2872         case MDS_PIN:
2873         case MDS_SYNC:
2874         case MDS_GETXATTR:
2875         case MDS_SETXATTR:
2876         case MDS_SET_INFO:
2877         case MDS_GET_INFO:
2878         case MDS_QUOTACHECK:
2879         case MDS_QUOTACTL:
2880         case QUOTA_DQACQ:
2881         case QUOTA_DQREL:
2882         case SEQ_QUERY:
2883         case FLD_QUERY:
2884                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2885                 if (rc)
2886                         CERROR("bad opc %u version %08x, expecting %08x\n",
2887                                lustre_msg_get_opc(msg),
2888                                lustre_msg_get_version(msg),
2889                                LUSTRE_MDS_VERSION);
2890                 break;
2891         case LDLM_ENQUEUE:
2892         case LDLM_CONVERT:
2893         case LDLM_BL_CALLBACK:
2894         case LDLM_CP_CALLBACK:
2895                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2896                 if (rc)
2897                         CERROR("bad opc %u version %08x, expecting %08x\n",
2898                                lustre_msg_get_opc(msg),
2899                                lustre_msg_get_version(msg),
2900                                LUSTRE_DLM_VERSION);
2901                 break;
2902         case OBD_LOG_CANCEL:
2903         case LLOG_ORIGIN_HANDLE_CREATE:
2904         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2905         case LLOG_ORIGIN_HANDLE_READ_HEADER:
2906         case LLOG_ORIGIN_HANDLE_CLOSE:
2907         case LLOG_ORIGIN_HANDLE_DESTROY:
2908         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2909         case LLOG_CATINFO:
2910                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2911                 if (rc)
2912                         CERROR("bad opc %u version %08x, expecting %08x\n",
2913                                lustre_msg_get_opc(msg),
2914                                lustre_msg_get_version(msg),
2915                                LUSTRE_LOG_VERSION);
2916                 break;
2917         default:
2918                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2919                 rc = -ENOTSUPP;
2920         }
2921         return rc;
2922 }
2923
2924 static int mdt_handle0(struct ptlrpc_request *req,
2925                        struct mdt_thread_info *info,
2926                        struct mdt_opc_slice *supported)
2927 {
2928         struct mdt_handler *h;
2929         struct lustre_msg  *msg;
2930         int                 rc;
2931
2932         ENTRY;
2933
2934         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2935                 RETURN(0);
2936
2937         LASSERT(current->journal_info == NULL);
2938
2939         msg = req->rq_reqmsg;
2940         rc = mdt_msg_check_version(msg);
2941         if (likely(rc == 0)) {
2942                 rc = mdt_recovery(info);
2943                 if (likely(rc == +1)) {
2944                         h = mdt_handler_find(lustre_msg_get_opc(msg),
2945                                              supported);
2946                         if (likely(h != NULL)) {
2947                                 rc = mdt_req_handle(info, h, req);
2948                         } else {
2949                                 CERROR("The unsupported opc: 0x%x\n",
2950                                        lustre_msg_get_opc(msg) );
2951                                 req->rq_status = -ENOTSUPP;
2952                                 rc = ptlrpc_error(req);
2953                                 RETURN(rc);
2954                         }
2955                 }
2956         } else
2957                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2958         RETURN(rc);
2959 }
2960
2961 /*
2962  * MDT handler function called by ptlrpc service thread when request comes.
2963  *
2964  * XXX common "target" functionality should be factored into separate module
2965  * shared by mdt, ost and stand-alone services like fld.
2966  */
2967 static int mdt_handle_common(struct ptlrpc_request *req,
2968                              struct mdt_opc_slice *supported)
2969 {
2970         struct lu_env          *env;
2971         struct mdt_thread_info *info;
2972         int                     rc;
2973         ENTRY;
2974
2975         env = req->rq_svc_thread->t_env;
2976         LASSERT(env != NULL);
2977         LASSERT(env->le_ses != NULL);
2978         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2979         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2980         LASSERT(info != NULL);
2981
2982         mdt_thread_info_init(req, info);
2983
2984         rc = mdt_handle0(req, info, supported);
2985
2986         mdt_thread_info_fini(info);
2987         RETURN(rc);
2988 }
2989
2990 /*
2991  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2992  * as well.
2993  */
2994 int mdt_recovery_handle(struct ptlrpc_request *req)
2995 {
2996         int rc;
2997         ENTRY;
2998
2999         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3000         case FLD_QUERY:
3001                 rc = mdt_handle_common(req, mdt_fld_handlers);
3002                 break;
3003         case SEQ_QUERY:
3004                 rc = mdt_handle_common(req, mdt_seq_handlers);
3005                 break;
3006         default:
3007                 rc = mdt_handle_common(req, mdt_regular_handlers);
3008                 break;
3009         }
3010
3011         RETURN(rc);
3012 }
3013
3014 static int mdt_regular_handle(struct ptlrpc_request *req)
3015 {
3016         return mdt_handle_common(req, mdt_regular_handlers);
3017 }
3018
3019 static int mdt_readpage_handle(struct ptlrpc_request *req)
3020 {
3021         return mdt_handle_common(req, mdt_readpage_handlers);
3022 }
3023
3024 static int mdt_xmds_handle(struct ptlrpc_request *req)
3025 {
3026         return mdt_handle_common(req, mdt_xmds_handlers);
3027 }
3028
3029 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3030 {
3031         return mdt_handle_common(req, mdt_seq_handlers);
3032 }
3033
3034 static int mdt_mdss_handle(struct ptlrpc_request *req)
3035 {
3036         return mdt_handle_common(req, mdt_seq_handlers);
3037 }
3038
3039 static int mdt_dtss_handle(struct ptlrpc_request *req)
3040 {
3041         return mdt_handle_common(req, mdt_seq_handlers);
3042 }
3043
3044 static int mdt_fld_handle(struct ptlrpc_request *req)
3045 {
3046         return mdt_handle_common(req, mdt_fld_handlers);
3047 }
3048
3049 enum mdt_it_code {
3050         MDT_IT_OPEN,
3051         MDT_IT_OCREAT,
3052         MDT_IT_CREATE,
3053         MDT_IT_GETATTR,
3054         MDT_IT_READDIR,
3055         MDT_IT_LOOKUP,
3056         MDT_IT_UNLINK,
3057         MDT_IT_TRUNC,
3058         MDT_IT_GETXATTR,
3059         MDT_IT_NR
3060 };
3061
3062 static int mdt_intent_getattr(enum mdt_it_code opcode,
3063                               struct mdt_thread_info *info,
3064                               struct ldlm_lock **,
3065                               int);
3066 static int mdt_intent_reint(enum mdt_it_code opcode,
3067                             struct mdt_thread_info *info,
3068                             struct ldlm_lock **,
3069                             int);
3070
3071 static struct mdt_it_flavor {
3072         const struct req_format *it_fmt;
3073         __u32                    it_flags;
3074         int                    (*it_act)(enum mdt_it_code ,
3075                                          struct mdt_thread_info *,
3076                                          struct ldlm_lock **,
3077                                          int);
3078         long                     it_reint;
3079 } mdt_it_flavor[] = {
3080         [MDT_IT_OPEN]     = {
3081                 .it_fmt   = &RQF_LDLM_INTENT,
3082                 /*.it_flags = HABEO_REFERO,*/
3083                 .it_flags = 0,
3084                 .it_act   = mdt_intent_reint,
3085                 .it_reint = REINT_OPEN
3086         },
3087         [MDT_IT_OCREAT]   = {
3088                 .it_fmt   = &RQF_LDLM_INTENT,
3089                 .it_flags = MUTABOR,
3090                 .it_act   = mdt_intent_reint,
3091                 .it_reint = REINT_OPEN
3092         },
3093         [MDT_IT_CREATE]   = {
3094                 .it_fmt   = &RQF_LDLM_INTENT,
3095                 .it_flags = MUTABOR,
3096                 .it_act   = mdt_intent_reint,
3097                 .it_reint = REINT_CREATE
3098         },
3099         [MDT_IT_GETATTR]  = {
3100                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3101                 .it_flags = HABEO_REFERO,
3102                 .it_act   = mdt_intent_getattr
3103         },
3104         [MDT_IT_READDIR]  = {
3105                 .it_fmt   = NULL,
3106                 .it_flags = 0,
3107                 .it_act   = NULL
3108         },
3109         [MDT_IT_LOOKUP]   = {
3110                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3111                 .it_flags = HABEO_REFERO,
3112                 .it_act   = mdt_intent_getattr
3113         },
3114         [MDT_IT_UNLINK]   = {
3115                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3116                 .it_flags = MUTABOR,
3117                 .it_act   = NULL,
3118                 .it_reint = REINT_UNLINK
3119         },
3120         [MDT_IT_TRUNC]    = {
3121                 .it_fmt   = NULL,
3122                 .it_flags = MUTABOR,
3123                 .it_act   = NULL
3124         },
3125         [MDT_IT_GETXATTR] = {
3126                 .it_fmt   = NULL,
3127                 .it_flags = 0,
3128                 .it_act   = NULL
3129         }
3130 };
3131
3132 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3133                             struct ldlm_lock **lockp,
3134                             struct ldlm_lock *new_lock,
3135                             struct mdt_lock_handle *lh,
3136                             int flags)
3137 {
3138         struct ptlrpc_request  *req = mdt_info_req(info);
3139         struct ldlm_lock       *lock = *lockp;
3140
3141         /*
3142          * Get new lock only for cases when possible resent did not find any
3143          * lock.
3144          */
3145         if (new_lock == NULL)
3146                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3147
3148         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3149                 lh->mlh_reg_lh.cookie = 0;
3150                 RETURN(0);
3151         }
3152
3153         LASSERTF(new_lock != NULL,
3154                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3155
3156         /*
3157          * If we've already given this lock to a client once, then we should
3158          * have no readers or writers.  Otherwise, we should have one reader
3159          * _or_ writer ref (which will be zeroed below) before returning the
3160          * lock to a client.
3161          */
3162         if (new_lock->l_export == req->rq_export) {
3163                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3164         } else {
3165                 LASSERT(new_lock->l_export == NULL);
3166                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3167         }
3168
3169         *lockp = new_lock;
3170
3171         if (new_lock->l_export == req->rq_export) {
3172                 /*
3173                  * Already gave this to the client, which means that we
3174                  * reconstructed a reply.
3175                  */
3176                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3177                         MSG_RESENT);
3178                 lh->mlh_reg_lh.cookie = 0;
3179                 RETURN(ELDLM_LOCK_REPLACED);
3180         }
3181
3182         /*
3183          * Fixup the lock to be given to the client.
3184          */
3185         lock_res_and_lock(new_lock);
3186         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3187          * possible blocking AST. */
3188         while (new_lock->l_readers > 0) {
3189                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3190                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3191                 new_lock->l_readers--;
3192         }
3193         while (new_lock->l_writers > 0) {
3194                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3195                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3196                 new_lock->l_writers--;
3197         }
3198
3199         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3200         new_lock->l_blocking_ast = lock->l_blocking_ast;
3201         new_lock->l_completion_ast = lock->l_completion_ast;
3202         new_lock->l_remote_handle = lock->l_remote_handle;
3203         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3204
3205         unlock_res_and_lock(new_lock);
3206
3207         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3208                      &new_lock->l_remote_handle,
3209                      &new_lock->l_exp_hash);
3210
3211         LDLM_LOCK_RELEASE(new_lock);
3212         lh->mlh_reg_lh.cookie = 0;
3213
3214         RETURN(ELDLM_LOCK_REPLACED);
3215 }
3216
3217 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3218                                     struct ldlm_lock *new_lock,
3219                                     struct ldlm_lock **old_lock,
3220                                     struct mdt_lock_handle *lh)
3221 {
3222         struct ptlrpc_request  *req = mdt_info_req(info);
3223         struct obd_export      *exp = req->rq_export;
3224         struct lustre_handle    remote_hdl;
3225         struct ldlm_request    *dlmreq;
3226         struct ldlm_lock       *lock;
3227
3228         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3229                 return;
3230
3231         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3232         remote_hdl = dlmreq->lock_handle[0];
3233
3234         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3235         if (lock) {
3236                 if (lock != new_lock) {
3237                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3238                         lh->mlh_reg_mode = lock->l_granted_mode;
3239
3240                         LDLM_DEBUG(lock, "Restoring lock cookie");
3241                         DEBUG_REQ(D_DLMTRACE, req,
3242                                   "restoring lock cookie "LPX64,
3243                                   lh->mlh_reg_lh.cookie);
3244                         if (old_lock)
3245                                 *old_lock = LDLM_LOCK_GET(lock);
3246                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3247                         return;
3248                 }
3249
3250                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3251         }
3252
3253         /*
3254          * If the xid matches, then we know this is a resent request, and allow
3255          * it. (It's probably an OPEN, for which we don't send a lock.
3256          */
3257         if (req_xid_is_last(req))
3258                 return;
3259
3260         /*
3261          * This remote handle isn't enqueued, so we never received or processed
3262          * this request.  Clear MSG_RESENT, because it can be handled like any
3263          * normal request now.
3264          */
3265         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3266
3267         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3268                   remote_hdl.cookie);
3269 }
3270
3271 static int mdt_intent_getattr(enum mdt_it_code opcode,
3272                               struct mdt_thread_info *info,
3273                               struct ldlm_lock **lockp,
3274                               int flags)
3275 {
3276         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3277         struct ldlm_lock       *new_lock = NULL;
3278         __u64                   child_bits;
3279         struct ldlm_reply      *ldlm_rep;
3280         struct ptlrpc_request  *req;
3281         struct mdt_body        *reqbody;
3282         struct mdt_body        *repbody;
3283         int                     rc;
3284         ENTRY;
3285
3286         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3287         LASSERT(reqbody);
3288
3289         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3290         LASSERT(repbody);
3291
3292         info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3293         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3294         repbody->eadatasize = 0;
3295         repbody->aclsize = 0;
3296
3297         switch (opcode) {
3298         case MDT_IT_LOOKUP:
3299                 child_bits = MDS_INODELOCK_LOOKUP;
3300                 break;
3301         case MDT_IT_GETATTR:
3302                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3303                 break;
3304         default:
3305                 CERROR("Unhandled till now");
3306                 GOTO(out_shrink, rc = -EINVAL);
3307         }
3308
3309         rc = mdt_init_ucred(info, reqbody);
3310         if (rc)
3311                 GOTO(out_shrink, rc);
3312
3313         req = info->mti_pill->rc_req;
3314         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3315         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3316
3317         /* Get lock from request for possible resent case. */
3318         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3319
3320         ldlm_rep->lock_policy_res2 =
3321                 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3322
3323         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3324                 ldlm_rep->lock_policy_res2 = 0;
3325         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3326             ldlm_rep->lock_policy_res2) {
3327                 lhc->mlh_reg_lh.cookie = 0ull;
3328                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3329         }
3330
3331         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3332         EXIT;
3333 out_ucred:
3334         mdt_exit_ucred(info);
3335 out_shrink:
3336         mdt_shrink_reply(info);
3337         return rc;
3338 }
3339
3340 static int mdt_intent_reint(enum mdt_it_code opcode,
3341                             struct mdt_thread_info *info,
3342                             struct ldlm_lock **lockp,
3343                             int flags)
3344 {
3345         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3346         struct ldlm_reply      *rep = NULL;
3347         long                    opc;
3348         int                     rc;
3349
3350         static const struct req_format *intent_fmts[REINT_MAX] = {
3351                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3352                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3353         };
3354
3355         ENTRY;
3356
3357         opc = mdt_reint_opcode(info, intent_fmts);
3358         if (opc < 0)
3359                 RETURN(opc);
3360
3361         if (mdt_it_flavor[opcode].it_reint != opc) {
3362                 CERROR("Reint code %ld doesn't match intent: %d\n",
3363                        opc, opcode);
3364                 RETURN(err_serious(-EPROTO));
3365         }
3366
3367         /* Get lock from request for possible resent case. */
3368         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3369
3370         rc = mdt_reint_internal(info, lhc, opc);
3371
3372         /* Check whether the reply has been packed successfully. */
3373         if (mdt_info_req(info)->rq_repmsg != NULL)
3374                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3375         if (rep == NULL)
3376                 RETURN(err_serious(-EFAULT));
3377
3378         /* MDC expects this in any case */
3379         if (rc != 0)
3380                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3381
3382         /* Cross-ref case, the lock should be returned to the client */
3383         if (rc == -EREMOTE) {
3384                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3385                 rep->lock_policy_res2 = 0;
3386                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3387                 RETURN(rc);
3388         }
3389         rep->lock_policy_res2 = clear_serious(rc);
3390
3391         if (rc == -ENOTCONN || rc == -ENODEV ||
3392             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3393                 /*
3394                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3395                  * will be returned by rq_status, and client at ptlrpc layer
3396                  * will detect this, then disconnect, reconnect the import
3397                  * immediately, instead of impacting the following the rpc.
3398                  */
3399                 lhc->mlh_reg_lh.cookie = 0ull;
3400                 RETURN(rc);
3401         } else {
3402                 /*
3403                  * For other cases, the error will be returned by intent.
3404                  * and client will retrieve the result from intent.
3405                  */
3406                  /*
3407                   * FIXME: when open lock is finished, that should be
3408                   * checked here.
3409                   */
3410                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3411                         rep->lock_policy_res2 = 0;
3412                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3413                         RETURN(rc);
3414                 } else {
3415                         lhc->mlh_reg_lh.cookie = 0ull;
3416                         RETURN(ELDLM_LOCK_ABORTED);
3417                 }
3418         }
3419 }
3420
3421 static int mdt_intent_code(long itcode)
3422 {
3423         int rc;
3424
3425         switch(itcode) {
3426         case IT_OPEN:
3427                 rc = MDT_IT_OPEN;
3428                 break;
3429         case IT_OPEN|IT_CREAT:
3430                 rc = MDT_IT_OCREAT;
3431                 break;
3432         case IT_CREAT:
3433                 rc = MDT_IT_CREATE;
3434                 break;
3435         case IT_READDIR:
3436                 rc = MDT_IT_READDIR;
3437                 break;
3438         case IT_GETATTR:
3439                 rc = MDT_IT_GETATTR;
3440                 break;
3441         case IT_LOOKUP:
3442                 rc = MDT_IT_LOOKUP;
3443                 break;
3444         case IT_UNLINK:
3445                 rc = MDT_IT_UNLINK;
3446                 break;
3447         case IT_TRUNC:
3448                 rc = MDT_IT_TRUNC;
3449                 break;
3450         case IT_GETXATTR:
3451                 rc = MDT_IT_GETXATTR;
3452                 break;
3453         default:
3454                 CERROR("Unknown intent opcode: %ld\n", itcode);
3455                 rc = -EINVAL;
3456                 break;
3457         }
3458         return rc;
3459 }
3460
3461 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3462                           struct ldlm_lock **lockp, int flags)
3463 {
3464         struct req_capsule   *pill;
3465         struct mdt_it_flavor *flv;
3466         int opc;
3467         int rc;
3468         ENTRY;
3469
3470         opc = mdt_intent_code(itopc);
3471         if (opc < 0)
3472                 RETURN(-EINVAL);
3473
3474         pill = info->mti_pill;
3475         flv  = &mdt_it_flavor[opc];
3476
3477         if (flv->it_fmt != NULL)
3478                 req_capsule_extend(pill, flv->it_fmt);
3479
3480         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3481         if (rc == 0) {
3482                 struct ptlrpc_request *req = mdt_info_req(info);
3483                 if (flv->it_flags & MUTABOR &&
3484                     req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3485                         RETURN(-EROFS);
3486         }
3487         if (rc == 0 && flv->it_act != NULL) {
3488                 /* execute policy */
3489                 rc = flv->it_act(opc, info, lockp, flags);
3490         } else {
3491                 rc = -EOPNOTSUPP;
3492         }
3493         RETURN(rc);
3494 }
3495
3496 static int mdt_intent_policy(struct ldlm_namespace *ns,
3497                              struct ldlm_lock **lockp, void *req_cookie,
3498                              ldlm_mode_t mode, int flags, void *data)
3499 {
3500         struct mdt_thread_info *info;
3501         struct ptlrpc_request  *req  =  req_cookie;
3502         struct ldlm_intent     *it;
3503         struct req_capsule     *pill;
3504         int rc;
3505
3506         ENTRY;
3507
3508         LASSERT(req != NULL);
3509
3510         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3511                                   &mdt_thread_key);
3512         LASSERT(info != NULL);
3513         pill = info->mti_pill;
3514         LASSERT(pill->rc_req == req);
3515
3516         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3517                 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3518                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3519                 if (it != NULL) {
3520                         const struct ldlm_request *dlmreq;
3521                         __u64 req_bits;
3522
3523                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3524                         if (rc == 0)
3525                                 rc = ELDLM_OK;
3526
3527                         /*
3528                          * Lock without inodebits makes no sense and will oops
3529                          * later in ldlm. Let's check it now to see if we have
3530                          * wrong lock from client or bits get corrupted
3531                          * somewhere in mdt_intent_opc().
3532                          */
3533                         dlmreq = info->mti_dlm_req;
3534                         req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3535                         LASSERT(req_bits != 0);
3536
3537                 } else
3538                         rc = err_serious(-EFAULT);
3539         } else {
3540                 /* No intent was provided */
3541                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3542                 rc = req_capsule_server_pack(pill);
3543                 if (rc)
3544                         rc = err_serious(rc);
3545         }
3546         RETURN(rc);
3547 }
3548
3549 /*
3550  * Seq wrappers
3551  */
3552 static void mdt_seq_adjust(const struct lu_env *env,
3553                           struct mdt_device *m, int lost)
3554 {
3555         struct md_site *ms = mdt_md_site(m);
3556         struct lu_seq_range out;
3557         ENTRY;
3558
3559         LASSERT(ms && ms->ms_server_seq);
3560         LASSERT(lost >= 0);
3561         /* get extra seq from seq_server, moving it's range up */
3562         while (lost-- > 0) {
3563                 seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env);
3564         }
3565         EXIT;
3566 }
3567
3568 static int mdt_seq_fini(const struct lu_env *env,
3569                         struct mdt_device *m)
3570 {
3571         struct md_site *ms = mdt_md_site(m);
3572         ENTRY;
3573
3574         if (ms != NULL) {
3575                 if (ms->ms_server_seq) {
3576                         seq_server_fini(ms->ms_server_seq, env);
3577                         OBD_FREE_PTR(ms->ms_server_seq);
3578                         ms->ms_server_seq = NULL;
3579         }
3580
3581                 if (ms->ms_control_seq) {
3582                         seq_server_fini(ms->ms_control_seq, env);
3583                         OBD_FREE_PTR(ms->ms_control_seq);
3584                         ms->ms_control_seq = NULL;
3585         }
3586
3587                 if (ms->ms_client_seq) {
3588                         seq_client_fini(ms->ms_client_seq);
3589                         OBD_FREE_PTR(ms->ms_client_seq);
3590                         ms->ms_client_seq = NULL;
3591                 }
3592         }
3593
3594         RETURN(0);
3595 }
3596
3597 static int mdt_seq_init(const struct lu_env *env,
3598                         const char *uuid,
3599                         struct mdt_device *m)
3600 {
3601         struct md_site *ms;
3602         char *prefix;
3603         int rc;
3604         ENTRY;
3605
3606         ms = mdt_md_site(m);
3607
3608         /*
3609          * This is sequence-controller node. Init seq-controller server on local
3610          * MDT.
3611          */
3612         if (ms->ms_node_id == 0) {
3613                 LASSERT(ms->ms_control_seq == NULL);
3614
3615                 OBD_ALLOC_PTR(ms->ms_control_seq);
3616                 if (ms->ms_control_seq == NULL)
3617                         RETURN(-ENOMEM);
3618
3619                 rc = seq_server_init(ms->ms_control_seq,
3620                                      m->mdt_bottom, uuid,
3621                                      LUSTRE_SEQ_CONTROLLER,
3622                                      ms,
3623                                      env);
3624
3625                 if (rc)
3626                         GOTO(out_seq_fini, rc);
3627
3628                 OBD_ALLOC_PTR(ms->ms_client_seq);
3629                 if (ms->ms_client_seq == NULL)
3630                         GOTO(out_seq_fini, rc = -ENOMEM);
3631
3632                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3633                 if (prefix == NULL) {
3634                         OBD_FREE_PTR(ms->ms_client_seq);
3635                         GOTO(out_seq_fini, rc = -ENOMEM);
3636                 }
3637
3638                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3639                          uuid);
3640
3641                 /*
3642                  * Init seq-controller client after seq-controller server is
3643                  * ready. Pass ms->ms_control_seq to it for direct talking.
3644                  */
3645                 rc = seq_client_init(ms->ms_client_seq, NULL,
3646                                      LUSTRE_SEQ_METADATA, prefix,
3647                                      ms->ms_control_seq);
3648                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3649
3650                 if (rc)
3651                         GOTO(out_seq_fini, rc);
3652         }
3653
3654         /* Init seq-server on local MDT */
3655         LASSERT(ms->ms_server_seq == NULL);
3656
3657         OBD_ALLOC_PTR(ms->ms_server_seq);
3658         if (ms->ms_server_seq == NULL)
3659                 GOTO(out_seq_fini, rc = -ENOMEM);
3660
3661         rc = seq_server_init(ms->ms_server_seq,
3662                              m->mdt_bottom, uuid,
3663                              LUSTRE_SEQ_SERVER,
3664                              ms,
3665                              env);
3666         if (rc)
3667                 GOTO(out_seq_fini, rc = -ENOMEM);
3668
3669         /* Assign seq-controller client to local seq-server. */
3670         if (ms->ms_node_id == 0) {
3671                 LASSERT(ms->ms_client_seq != NULL);
3672
3673                 rc = seq_server_set_cli(ms->ms_server_seq,
3674                                         ms->ms_client_seq,
3675                                         env);
3676         }
3677
3678         EXIT;
3679 out_seq_fini:
3680         if (rc)
3681                 mdt_seq_fini(env, m);
3682
3683         return rc;
3684 }
3685 /*
3686  * Init client sequence manager which is used by local MDS to talk to sequence
3687  * controller on remote node.
3688  */
3689 static int mdt_seq_init_cli(const struct lu_env *env,
3690                             struct mdt_device *m,
3691                             struct lustre_cfg *cfg)
3692 {
3693         struct md_site    *ms = mdt_md_site(m);
3694         struct obd_device *mdc;
3695         struct obd_uuid   *uuidp, *mdcuuidp;
3696         char              *uuid_str, *mdc_uuid_str;
3697         int                rc;
3698         int                index;
3699         struct mdt_thread_info *info;
3700         char *p, *index_string = lustre_cfg_string(cfg, 2);
3701         ENTRY;
3702
3703         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3704         uuidp = &info->mti_u.uuid[0];
3705         mdcuuidp = &info->mti_u.uuid[1];
3706
3707         LASSERT(index_string);
3708
3709         index = simple_strtol(index_string, &p, 10);
3710         if (*p) {
3711                 CERROR("Invalid index in lustre_cgf, offset 2\n");
3712                 RETURN(-EINVAL);
3713         }
3714
3715         /* check if this is adding the first MDC and controller is not yet
3716          * initialized. */
3717         if (index != 0 || ms->ms_client_seq)
3718                 RETURN(0);
3719
3720         uuid_str = lustre_cfg_string(cfg, 1);
3721         mdc_uuid_str = lustre_cfg_string(cfg, 4);
3722         obd_str2uuid(uuidp, uuid_str);
3723         obd_str2uuid(mdcuuidp, mdc_uuid_str);
3724
3725         mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3726         if (!mdc) {
3727                 CERROR("can't find controller MDC by uuid %s\n",
3728                        uuid_str);
3729                 rc = -ENOENT;
3730         } else if (!mdc->obd_set_up) {
3731                 CERROR("target %s not set up\n", mdc->obd_name);
3732                 rc = -EINVAL;
3733         } else {
3734                 LASSERT(ms->ms_control_exp);
3735                 OBD_ALLOC_PTR(ms->ms_client_seq);
3736                 if (ms->ms_client_seq != NULL) {
3737                         char *prefix;
3738
3739                         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3740                         if (!prefix)
3741                                 RETURN(-ENOMEM);
3742
3743                         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3744                                  mdc->obd_name);
3745
3746                         rc = seq_client_init(ms->ms_client_seq,
3747                                              ms->ms_control_exp,
3748                                              LUSTRE_SEQ_METADATA,
3749                                              prefix, NULL);
3750                         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3751                 } else
3752                         rc = -ENOMEM;
3753
3754                 if (rc)
3755                         RETURN(rc);
3756
3757                 LASSERT(ms->ms_server_seq != NULL);
3758                 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3759                                         env);
3760         }
3761
3762         RETURN(rc);
3763 }
3764
3765 static void mdt_seq_fini_cli(struct mdt_device *m)
3766 {
3767         struct md_site *ms;
3768
3769         ENTRY;
3770
3771         ms = mdt_md_site(m);
3772
3773         if (ms != NULL) {
3774                 if (ms->ms_server_seq)
3775                         seq_server_set_cli(ms->ms_server_seq,
3776                                    NULL, NULL);
3777
3778                 if (ms->ms_control_exp) {
3779                         class_export_put(ms->ms_control_exp);
3780                         ms->ms_control_exp = NULL;
3781                 }
3782         }
3783         EXIT;
3784 }
3785
3786 /*
3787  * FLD wrappers
3788  */
3789 static int mdt_fld_fini(const struct lu_env *env,
3790                         struct mdt_device *m)
3791 {
3792         struct md_site *ms = mdt_md_site(m);
3793         ENTRY;
3794
3795         if (ms && ms->ms_server_fld) {
3796                 fld_server_fini(ms->ms_server_fld, env);
3797                 OBD_FREE_PTR(ms->ms_server_fld);
3798                 ms->ms_server_fld = NULL;
3799         }
3800
3801         RETURN(0);
3802 }
3803
3804 static int mdt_fld_init(const struct lu_env *env,
3805                         const char *uuid,
3806                         struct mdt_device *m)
3807 {
3808         struct md_site *ms;
3809         int rc;
3810         ENTRY;
3811
3812         ms = mdt_md_site(m);
3813
3814         OBD_ALLOC_PTR(ms->ms_server_fld);
3815         if (ms->ms_server_fld == NULL)
3816                 RETURN(rc = -ENOMEM);
3817
3818         rc = fld_server_init(ms->ms_server_fld,
3819                              m->mdt_bottom, uuid,
3820                              env, ms->ms_node_id);
3821         if (rc) {
3822                 OBD_FREE_PTR(ms->ms_server_fld);
3823                 ms->ms_server_fld = NULL;
3824                 RETURN(rc);
3825         }
3826
3827         RETURN(0);
3828 }
3829
3830 /* device init/fini methods */
3831 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3832 {
3833         ENTRY;
3834         if (m->mdt_regular_service != NULL) {
3835                 ptlrpc_unregister_service(m->mdt_regular_service);
3836                 m->mdt_regular_service = NULL;
3837         }
3838         if (m->mdt_readpage_service != NULL) {
3839                 ptlrpc_unregister_service(m->mdt_readpage_service);
3840                 m->mdt_readpage_service = NULL;
3841         }
3842         if (m->mdt_xmds_service != NULL) {
3843                 ptlrpc_unregister_service(m->mdt_xmds_service);
3844                 m->mdt_xmds_service = NULL;
3845         }
3846         if (m->mdt_setattr_service != NULL) {
3847                 ptlrpc_unregister_service(m->mdt_setattr_service);
3848                 m->mdt_setattr_service = NULL;
3849         }
3850         if (m->mdt_mdsc_service != NULL) {
3851                 ptlrpc_unregister_service(m->mdt_mdsc_service);
3852                 m->mdt_mdsc_service = NULL;
3853         }
3854         if (m->mdt_mdss_service != NULL) {
3855                 ptlrpc_unregister_service(m->mdt_mdss_service);
3856                 m->mdt_mdss_service = NULL;
3857         }
3858         if (m->mdt_dtss_service != NULL) {
3859                 ptlrpc_unregister_service(m->mdt_dtss_service);
3860                 m->mdt_dtss_service = NULL;
3861         }
3862         if (m->mdt_fld_service != NULL) {
3863                 ptlrpc_unregister_service(m->mdt_fld_service);
3864                 m->mdt_fld_service = NULL;
3865         }
3866         EXIT;
3867 }
3868
3869 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3870 {
3871         int rc;
3872         static struct ptlrpc_service_conf conf;
3873         cfs_proc_dir_entry_t *procfs_entry;
3874         ENTRY;
3875
3876         procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3877
3878         conf = (typeof(conf)) {
3879                 .psc_nbufs           = MDS_NBUFS,
3880                 .psc_bufsize         = MDS_BUFSIZE,
3881                 .psc_max_req_size    = MDS_MAXREQSIZE,
3882                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3883                 .psc_req_portal      = MDS_REQUEST_PORTAL,
3884                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3885                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3886                 /*
3887                  * We'd like to have a mechanism to set this on a per-device
3888                  * basis, but alas...
3889                  */
3890                 .psc_min_threads     = mdt_min_threads,
3891                 .psc_max_threads     = mdt_max_threads,
3892                 .psc_ctx_tags        = LCT_MD_THREAD
3893         };
3894
3895         m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3896         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3897                            "mdt_ldlm_client", m->mdt_ldlm_client);
3898
3899         m->mdt_regular_service =
3900                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3901                                      procfs_entry, target_print_req,
3902                                      LUSTRE_MDT_NAME);
3903         if (m->mdt_regular_service == NULL)
3904                 RETURN(-ENOMEM);
3905
3906         rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3907         if (rc)
3908                 GOTO(err_mdt_svc, rc);
3909
3910         /*
3911          * readpage service configuration. Parameters have to be adjusted,
3912          * ideally.
3913          */
3914         conf = (typeof(conf)) {
3915                 .psc_nbufs           = MDS_NBUFS,
3916                 .psc_bufsize         = MDS_BUFSIZE,
3917                 .psc_max_req_size    = MDS_MAXREQSIZE,
3918                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3919                 .psc_req_portal      = MDS_READPAGE_PORTAL,
3920                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3921                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3922                 .psc_min_threads     = mdt_min_threads,
3923                 .psc_max_threads     = mdt_max_threads,
3924                 .psc_ctx_tags        = LCT_MD_THREAD
3925         };
3926         m->mdt_readpage_service =
3927                 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3928                                      LUSTRE_MDT_NAME "_readpage",
3929                                      procfs_entry, target_print_req,"mdt_rdpg");
3930
3931         if (m->mdt_readpage_service == NULL) {
3932                 CERROR("failed to start readpage service\n");
3933                 GOTO(err_mdt_svc, rc = -ENOMEM);
3934         }
3935
3936         rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3937
3938         /*
3939          * setattr service configuration.
3940          */
3941         conf = (typeof(conf)) {
3942                 .psc_nbufs           = MDS_NBUFS,
3943                 .psc_bufsize         = MDS_BUFSIZE,
3944                 .psc_max_req_size    = MDS_MAXREQSIZE,
3945                 .psc_max_reply_size  = MDS_MAXREPSIZE,
3946                 .psc_req_portal      = MDS_SETATTR_PORTAL,
3947                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3948                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3949                 .psc_min_threads     = mdt_min_threads,
3950                 .psc_max_threads     = mdt_max_threads,
3951                 .psc_ctx_tags        = LCT_MD_THREAD
3952         };
3953
3954         m->mdt_setattr_service =
3955                 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3956                                      LUSTRE_MDT_NAME "_setattr",
3957                                      procfs_entry, target_print_req,"mdt_attr");
3958
3959         if (!m->mdt_setattr_service) {
3960                 CERROR("failed to start setattr service\n");
3961                 GOTO(err_mdt_svc, rc = -ENOMEM);
3962         }
3963
3964         rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3965         if (rc)
3966                 GOTO(err_mdt_svc, rc);
3967
3968         /*
3969          * sequence controller service configuration
3970          */
3971         conf = (typeof(conf)) {
3972                 .psc_nbufs           = MDS_NBUFS,
3973                 .psc_bufsize         = MDS_BUFSIZE,
3974                 .psc_max_req_size    = SEQ_MAXREQSIZE,
3975                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
3976                 .psc_req_portal      = SEQ_CONTROLLER_PORTAL,
3977                 .psc_rep_portal      = MDC_REPLY_PORTAL,
3978                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3979                 .psc_min_threads     = mdt_min_threads,
3980                 .psc_max_threads     = mdt_max_threads,
3981                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
3982         };
3983
3984         m->mdt_mdsc_service =
3985                 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3986                                      LUSTRE_MDT_NAME"_mdsc",
3987                                      procfs_entry, target_print_req,"mdt_mdsc");
3988         if (!m->mdt_mdsc_service) {
3989                 CERROR("failed to start seq controller service\n");
3990                 GOTO(err_mdt_svc, rc = -ENOMEM);
3991         }
3992
3993         rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3994         if (rc)
3995                 GOTO(err_mdt_svc, rc);
3996
3997         /*
3998          * metadata sequence server service configuration
3999          */
4000         conf = (typeof(conf)) {
4001                 .psc_nbufs           = MDS_NBUFS,
4002                 .psc_bufsize         = MDS_BUFSIZE,
4003                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4004                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4005                 .psc_req_portal      = SEQ_METADATA_PORTAL,
4006                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4007                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4008                 .psc_min_threads     = mdt_min_threads,
4009                 .psc_max_threads     = mdt_max_threads,
4010                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4011         };
4012
4013         m->mdt_mdss_service =
4014                 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
4015                                      LUSTRE_MDT_NAME"_mdss",
4016                                      procfs_entry, target_print_req,"mdt_mdss");
4017         if (!m->mdt_mdss_service) {
4018                 CERROR("failed to start metadata seq server service\n");
4019                 GOTO(err_mdt_svc, rc = -ENOMEM);
4020         }
4021
4022         rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4023         if (rc)
4024                 GOTO(err_mdt_svc, rc);
4025
4026
4027         /*
4028          * Data sequence server service configuration. We want to have really
4029          * cluster-wide sequences space. This is why we start only one sequence
4030          * controller which manages space.
4031          */
4032         conf = (typeof(conf)) {
4033                 .psc_nbufs           = MDS_NBUFS,
4034                 .psc_bufsize         = MDS_BUFSIZE,
4035                 .psc_max_req_size    = SEQ_MAXREQSIZE,
4036                 .psc_max_reply_size  = SEQ_MAXREPSIZE,
4037                 .psc_req_portal      = SEQ_DATA_PORTAL,
4038                 .psc_rep_portal      = OSC_REPLY_PORTAL,
4039                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4040                 .psc_min_threads     = mdt_min_threads,
4041                 .psc_max_threads     = mdt_max_threads,
4042                 .psc_ctx_tags        = LCT_MD_THREAD|LCT_DT_THREAD
4043         };
4044
4045         m->mdt_dtss_service =
4046                 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4047                                      LUSTRE_MDT_NAME"_dtss",
4048                                      procfs_entry, target_print_req,"mdt_dtss");
4049         if (!m->mdt_dtss_service) {
4050                 CERROR("failed to start data seq server service\n");
4051                 GOTO(err_mdt_svc, rc = -ENOMEM);
4052         }
4053
4054         rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4055         if (rc)
4056                 GOTO(err_mdt_svc, rc);
4057
4058         /* FLD service start */
4059         conf = (typeof(conf)) {
4060                 .psc_nbufs           = MDS_NBUFS,
4061                 .psc_bufsize         = MDS_BUFSIZE,
4062                 .psc_max_req_size    = FLD_MAXREQSIZE,
4063                 .psc_max_reply_size  = FLD_MAXREPSIZE,
4064                 .psc_req_portal      = FLD_REQUEST_PORTAL,
4065                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4066                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4067                 .psc_min_threads     = mdt_min_threads,
4068                 .psc_max_threads     = mdt_max_threads,
4069                 .psc_ctx_tags        = LCT_DT_THREAD|LCT_MD_THREAD
4070         };
4071
4072         m->mdt_fld_service =
4073                 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4074                                      LUSTRE_MDT_NAME"_fld",
4075                                      procfs_entry, target_print_req, "mdt_fld");
4076         if (!m->mdt_fld_service) {
4077                 CERROR("failed to start fld service\n");
4078                 GOTO(err_mdt_svc, rc = -ENOMEM);
4079         }
4080
4081         rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4082         if (rc)
4083                 GOTO(err_mdt_svc, rc);
4084
4085         /*
4086          * mds-mds service configuration. Separate portal is used to allow
4087          * mds-mds requests be not blocked during recovery.
4088          */
4089         conf = (typeof(conf)) {
4090                 .psc_nbufs           = MDS_NBUFS,
4091                 .psc_bufsize         = MDS_BUFSIZE,
4092                 .psc_max_req_size    = MDS_MAXREQSIZE,
4093                 .psc_max_reply_size  = MDS_MAXREPSIZE,
4094                 .psc_req_portal      = MDS_MDS_PORTAL,
4095                 .psc_rep_portal      = MDC_REPLY_PORTAL,
4096                 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4097                 .psc_min_threads     = mdt_min_threads,
4098                 .psc_max_threads     = mdt_max_threads,
4099                 .psc_ctx_tags        = LCT_MD_THREAD
4100         };
4101         m->mdt_xmds_service =
4102                 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4103                                      LUSTRE_MDT_NAME "_mds",
4104                                      procfs_entry, target_print_req,"mdt_xmds");
4105
4106         if (m->mdt_xmds_service == NULL) {
4107                 CERROR("failed to start xmds service\n");
4108                 GOTO(err_mdt_svc, rc = -ENOMEM);
4109         }
4110
4111         rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4112         if (rc)
4113                 GOTO(err_mdt_svc, rc);
4114
4115         EXIT;
4116 err_mdt_svc:
4117         if (rc)
4118                 mdt_stop_ptlrpc_service(m);
4119
4120         return rc;
4121 }
4122
4123 static void mdt_stack_fini(const struct lu_env *env,
4124                            struct mdt_device *m, struct lu_device *top)
4125 {
4126         struct obd_device       *obd = mdt2obd_dev(m);
4127         struct lustre_cfg_bufs  *bufs;
4128         struct lustre_cfg       *lcfg;
4129         struct mdt_thread_info  *info;
4130         char flags[3]="";
4131         ENTRY;
4132
4133         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4134         LASSERT(info != NULL);
4135
4136         bufs = &info->mti_u.bufs;
4137         /* process cleanup, pass mdt obd name to get obd umount flags */
4138         lustre_cfg_bufs_reset(bufs, obd->obd_name);
4139         if (obd->obd_force)
4140                 strcat(flags, "F");
4141         if (obd->obd_fail)
4142                 strcat(flags, "A");
4143         lustre_cfg_bufs_set_string(bufs, 1, flags);
4144         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4145         if (!lcfg) {
4146                 CERROR("Cannot alloc lcfg!\n");
4147                 return;
4148         }
4149
4150         LASSERT(top);
4151         top->ld_ops->ldo_process_config(env, top, lcfg);
4152         lustre_cfg_free(lcfg);
4153
4154         lu_stack_fini(env, top);
4155         m->mdt_child = NULL;
4156         m->mdt_bottom = NULL;
4157 }
4158
4159 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4160                                          const char *typename,
4161                                          struct lu_device *child,
4162                                          struct lustre_cfg *cfg)
4163 {
4164         const char            *dev = lustre_cfg_string(cfg, 0);
4165         struct obd_type       *type;
4166         struct lu_device_type *ldt;
4167         struct lu_device      *d;
4168         int rc;
4169         ENTRY;
4170
4171         /* find the type */
4172         type = class_get_type(typename);
4173         if (!type) {
4174                 CERROR("Unknown type: '%s'\n", typename);
4175                 GOTO(out, rc = -ENODEV);
4176         }
4177
4178         rc = lu_env_refill((struct lu_env *)env);
4179         if (rc != 0) {
4180                 CERROR("Failure to refill session: '%d'\n", rc);
4181                 GOTO(out_type, rc);
4182         }
4183
4184         ldt = type->typ_lu;
4185         if (ldt == NULL) {
4186                 CERROR("type: '%s'\n", typename);
4187                 GOTO(out_type, rc = -EINVAL);
4188         }
4189
4190         ldt->ldt_obd_type = type;
4191         d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4192         if (IS_ERR(d)) {
4193                 CERROR("Cannot allocate device: '%s'\n", typename);
4194                 GOTO(out_type, rc = -ENODEV);
4195         }
4196
4197         LASSERT(child->ld_site);
4198         d->ld_site = child->ld_site;
4199
4200         type->typ_refcnt++;
4201         rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4202         if (rc) {
4203                 CERROR("can't init device '%s', rc %d\n", typename, rc);
4204                 GOTO(out_alloc, rc);
4205         }
4206         lu_device_get(d);
4207         lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4208
4209         RETURN(d);
4210
4211 out_alloc:
4212         ldt->ldt_ops->ldto_device_free(env, d);
4213         type->typ_refcnt--;
4214 out_type:
4215         class_put_type(type);
4216 out:
4217         return ERR_PTR(rc);
4218 }
4219
4220 static int mdt_stack_init(struct lu_env *env,
4221                           struct mdt_device *m,
4222                           struct lustre_cfg *cfg,
4223                           struct lustre_mount_info  *lmi)
4224 {
4225         struct lu_device  *d = &m->mdt_md_dev.md_lu_dev;
4226         struct lu_device  *tmp;
4227         struct md_device  *md;
4228         struct lu_device  *child_lu_dev;
4229         int rc;
4230         ENTRY;
4231
4232         /* init the stack */
4233         tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4234         if (IS_ERR(tmp)) {
4235                 RETURN(PTR_ERR(tmp));
4236         }
4237         m->mdt_bottom = lu2dt_dev(tmp);
4238         d = tmp;
4239         tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4240         if (IS_ERR(tmp)) {
4241                 GOTO(out, rc = PTR_ERR(tmp));
4242         }
4243         d = tmp;
4244         md = lu2md_dev(d);
4245
4246         tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4247         if (IS_ERR(tmp)) {
4248                 GOTO(out, rc = PTR_ERR(tmp));
4249         }
4250         d = tmp;
4251         /*set mdd upcall device*/
4252         md_upcall_dev_set(md, lu2md_dev(d));
4253
4254         md = lu2md_dev(d);
4255         /*set cmm upcall device*/
4256         md_upcall_dev_set(md, &m->mdt_md_dev);
4257
4258         m->mdt_child = lu2md_dev(d);
4259
4260         /* process setup config */
4261         tmp = &m->mdt_md_dev.md_lu_dev;
4262         rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4263         if (rc)
4264                 GOTO(out, rc);
4265
4266         /* initialize local objects */
4267         child_lu_dev = &m->mdt_child->md_lu_dev;
4268
4269         rc = child_lu_dev->ld_ops->ldo_prepare(env,
4270                                                &m->mdt_md_dev.md_lu_dev,
4271                                                child_lu_dev);
4272 out:
4273         /* fini from last known good lu_device */
4274         if (rc)
4275                 mdt_stack_fini(env, m, d);
4276
4277         return rc;
4278 }
4279
4280 /**
4281  * setup CONFIG_ORIG context, used to access local config log.
4282  * this may need to be rewrite as part of llog rewrite for lu-api.
4283  */
4284 static int mdt_obd_llog_setup(struct obd_device *obd,
4285                               struct lustre_sb_info *lsi)
4286 {
4287         int     rc;
4288
4289         LASSERT(obd->obd_fsops == NULL);
4290
4291         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4292         if (IS_ERR(obd->obd_fsops))
4293                 return PTR_ERR(obd->obd_fsops);
4294
4295         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4296         if (rc) {
4297                 fsfilt_put_ops(obd->obd_fsops);
4298                 return rc;
4299         }
4300
4301         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4302         obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4303         obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4304         obd->obd_lvfs_ctxt.fs = get_ds();
4305
4306         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4307                         0, NULL, &llog_lvfs_ops);
4308         if (rc) {
4309                 CERROR("llog_setup() failed: %d\n", rc);
4310                 fsfilt_put_ops(obd->obd_fsops);
4311         }
4312
4313         return rc;
4314 }
4315
4316 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4317 {
4318         struct llog_ctxt *ctxt;
4319
4320         ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4321         if (ctxt)
4322                 llog_cleanup(ctxt);
4323
4324         if (obd->obd_fsops) {
4325                 fsfilt_put_ops(obd->obd_fsops);
4326                 obd->obd_fsops = NULL;
4327         }
4328 }
4329
4330 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4331 {
4332         struct md_device  *next = m->mdt_child;
4333         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4334         struct lu_site    *ls   = d->ld_site;
4335         struct obd_device *obd = mdt2obd_dev(m);
4336         ENTRY;
4337
4338         target_recovery_fini(obd);
4339
4340         ping_evictor_stop();
4341
4342         mdt_stop_ptlrpc_service(m);
4343         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4344         mdt_obd_llog_cleanup(obd);
4345         obd_exports_barrier(obd);
4346         obd_zombie_barrier();
4347 #ifdef HAVE_QUOTA_SUPPORT
4348         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4349 #endif
4350         lut_fini(env, &m->mdt_lut);
4351         mdt_fs_cleanup(env, m);
4352         upcall_cache_cleanup(m->mdt_identity_cache);
4353         m->mdt_identity_cache = NULL;
4354
4355         if (m->mdt_namespace != NULL) {
4356                 ldlm_namespace_free(m->mdt_namespace, NULL,
4357                                     d->ld_obd->obd_force);
4358                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4359         }
4360
4361         cfs_free_nidlist(&m->mdt_nosquash_nids);
4362         if (m->mdt_nosquash_str) {
4363                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4364                 m->mdt_nosquash_str = NULL;
4365                 m->mdt_nosquash_strlen = 0;
4366         }
4367
4368         mdt_seq_fini(env, m);
4369         mdt_seq_fini_cli(m);
4370         mdt_fld_fini(env, m);
4371         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4372
4373         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4374         cfs_timer_disarm(&m->mdt_ck_timer);
4375         mdt_ck_thread_stop(m);
4376
4377         /*
4378          * Finish the stack
4379          */
4380         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4381
4382         lprocfs_free_per_client_stats(obd);
4383         lprocfs_free_obd_stats(obd);
4384         mdt_procfs_fini(m);
4385
4386         if (ls) {
4387                 struct md_site *mite;
4388
4389                 lu_site_fini(ls);
4390                 mite = lu_site2md(ls);
4391                 OBD_FREE_PTR(mite);
4392                 d->ld_site = NULL;
4393         }
4394         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4395
4396         EXIT;
4397 }
4398
4399 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4400 {
4401         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4402         struct sptlrpc_rule_set  tmp_rset;
4403         int                      rc;
4404
4405         sptlrpc_rule_set_init(&tmp_rset);
4406         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4407         if (rc) {
4408                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4409                        obd->obd_name, rc);
4410                 return rc;
4411         }
4412
4413         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4414
4415         cfs_write_lock(&m->mdt_sptlrpc_lock);
4416         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4417         m->mdt_sptlrpc_rset = tmp_rset;
4418         cfs_write_unlock(&m->mdt_sptlrpc_lock);
4419
4420         return 0;
4421 }
4422
4423 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4424 {
4425         char *p = options;
4426
4427         m->mdt_opts.mo_mds_capa = 1;
4428         m->mdt_opts.mo_oss_capa = 1;
4429 #ifdef CONFIG_FS_POSIX_ACL
4430         /* ACLs should be enabled by default (b=13829) */
4431         m->mdt_opts.mo_acl = 1;
4432         LCONSOLE_INFO("Enabling ACL\n");
4433 #else
4434         m->mdt_opts.mo_acl = 0;
4435         LCONSOLE_INFO("Disabling ACL\n");
4436 #endif
4437
4438         if (!options)
4439                 return;
4440
4441         while (*options) {
4442                 int len;
4443
4444                 while (*p && *p != ',')
4445                         p++;
4446
4447                 len = p - options;
4448                 if ((len == sizeof("user_xattr") - 1) &&
4449                     (memcmp(options, "user_xattr", len) == 0)) {
4450                         m->mdt_opts.mo_user_xattr = 1;
4451                         LCONSOLE_INFO("Enabling user_xattr\n");
4452                 } else if ((len == sizeof("nouser_xattr") - 1) &&
4453                            (memcmp(options, "nouser_xattr", len) == 0)) {
4454                         m->mdt_opts.mo_user_xattr = 0;
4455                         LCONSOLE_INFO("Disabling user_xattr\n");
4456                 } else if ((len == sizeof("noacl") - 1) &&
4457                            (memcmp(options, "noacl", len) == 0)) {
4458                         m->mdt_opts.mo_acl = 0;
4459                         LCONSOLE_INFO("Disabling ACL\n");
4460                 }
4461
4462                 options = ++p;
4463         }
4464 }
4465
4466 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4467
4468 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4469                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4470 {
4471         struct mdt_thread_info    *info;
4472         struct obd_device         *obd;
4473         const char                *dev = lustre_cfg_string(cfg, 0);
4474         const char                *num = lustre_cfg_string(cfg, 2);
4475         struct lustre_mount_info  *lmi = NULL;
4476         struct lustre_sb_info     *lsi;
4477         struct lustre_disk_data   *ldd;
4478         struct lu_site            *s;
4479         struct md_site            *mite;
4480         const char                *identity_upcall = "NONE";
4481 #ifdef HAVE_QUOTA_SUPPORT
4482         struct md_device          *next;
4483 #endif
4484         int                        rc;
4485         int                        node_id;
4486         ENTRY;
4487
4488         md_device_init(&m->mdt_md_dev, ldt);
4489         /*
4490          * Environment (env) might be missing mdt_thread_key values at that
4491          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4492          * mode.
4493          *
4494          * Usually device allocation path doesn't use module key values, but
4495          * mdt has to do a lot of work here, so allocate key value.
4496          */
4497         rc = lu_env_refill((struct lu_env *)env);
4498         if (rc != 0)
4499                 RETURN(rc);
4500
4501         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4502         LASSERT(info != NULL);
4503
4504         obd = class_name2obd(dev);
4505         LASSERT(obd != NULL);
4506
4507         m->mdt_max_mdsize = MAX_MD_SIZE;
4508         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4509         m->mdt_som_conf = 0;
4510
4511         m->mdt_opts.mo_user_xattr = 0;
4512         m->mdt_opts.mo_acl = 0;
4513         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4514         lmi = server_get_mount_2(dev);
4515         if (lmi == NULL) {
4516                 CERROR("Cannot get mount info for %s!\n", dev);
4517                 RETURN(-EFAULT);
4518         } else {
4519                 lsi = s2lsi(lmi->lmi_sb);
4520                 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4521                 /* CMD is supported only in IAM mode */
4522                 ldd = lsi->lsi_ldd;
4523                 LASSERT(num);
4524                 node_id = simple_strtol(num, NULL, 10);
4525                 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4526                         CERROR("CMD Operation not allowed in IOP mode\n");
4527                         GOTO(err_lmi, rc = -EINVAL);
4528                 }
4529         }
4530
4531         cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4532         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4533
4534         cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4535         m->mdt_opts.mo_compat_resname = 0;
4536         m->mdt_capa_timeout = CAPA_TIMEOUT;
4537         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4538         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4539         m->mdt_squash_uid = 0;
4540         m->mdt_squash_gid = 0;
4541         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4542         m->mdt_nosquash_str = NULL;
4543         m->mdt_nosquash_strlen = 0;
4544         cfs_init_rwsem(&m->mdt_squash_sem);
4545
4546         OBD_ALLOC_PTR(mite);
4547         if (mite == NULL)
4548                 GOTO(err_lmi, rc = -ENOMEM);
4549
4550         s = &mite->ms_lu;
4551
4552         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4553         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4554         /* set this lu_device to obd, because error handling need it */
4555         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4556
4557         rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4558         if (rc) {
4559                 CERROR("Can't init lu_site, rc %d\n", rc);
4560                 GOTO(err_free_site, rc);
4561         }
4562
4563         rc = mdt_procfs_init(m, dev);
4564         if (rc) {
4565                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4566                 GOTO(err_fini_proc, rc);
4567         }
4568
4569         /* set server index */
4570         lu_site2md(s)->ms_node_id = node_id;
4571
4572         /* failover is the default
4573          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4574          * assumed whose ms_node_id == 0 XXX
4575          * */
4576         obd->obd_replayable = 1;
4577         /* No connection accepted until configurations will finish */
4578         obd->obd_no_conn = 1;
4579
4580         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4581                 char *str = lustre_cfg_string(cfg, 4);
4582                 if (strchr(str, 'n')) {
4583                         CWARN("%s: recovery disabled\n", obd->obd_name);
4584                         obd->obd_replayable = 0;
4585                 }
4586         }
4587
4588         /* init the stack */
4589         rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4590         if (rc) {
4591                 CERROR("Can't init device stack, rc %d\n", rc);
4592                 GOTO(err_fini_proc, rc);
4593         }
4594
4595         rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4596         if (rc)
4597                 GOTO(err_fini_stack, rc);
4598
4599         rc = mdt_fld_init(env, obd->obd_name, m);
4600         if (rc)
4601                 GOTO(err_lut, rc);
4602
4603         rc = mdt_seq_init(env, obd->obd_name, m);
4604         if (rc)
4605                 GOTO(err_fini_fld, rc);
4606
4607         snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4608                  LUSTRE_MDT_NAME"-%p", m);
4609         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4610                                               LDLM_NAMESPACE_SERVER,
4611                                               LDLM_NAMESPACE_GREEDY);
4612         if (m->mdt_namespace == NULL)
4613                 GOTO(err_fini_seq, rc = -ENOMEM);
4614
4615         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4616         /* set obd_namespace for compatibility with old code */
4617         obd->obd_namespace = m->mdt_namespace;
4618
4619         /* XXX: to support suppgid for ACL, we enable identity_upcall
4620          * by default, otherwise, maybe got unexpected -EACCESS. */
4621         if (m->mdt_opts.mo_acl)
4622                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4623
4624         m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4625                                                   &mdt_identity_upcall_cache_ops);
4626         if (IS_ERR(m->mdt_identity_cache)) {
4627                 rc = PTR_ERR(m->mdt_identity_cache);
4628                 m->mdt_identity_cache = NULL;
4629                 GOTO(err_free_ns, rc);
4630         }
4631
4632         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4633
4634         rc = mdt_ck_thread_start(m);
4635         if (rc)
4636                 GOTO(err_free_ns, rc);
4637
4638         rc = mdt_fs_setup(env, m, obd, lsi);
4639         if (rc)
4640                 GOTO(err_capa, rc);
4641
4642         rc = mdt_obd_llog_setup(obd, lsi);
4643         if (rc)
4644                 GOTO(err_fs_cleanup, rc);
4645
4646         rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4647         if (rc)
4648                 GOTO(err_llog_cleanup, rc);
4649
4650         mdt_adapt_sptlrpc_conf(obd, 1);
4651
4652 #ifdef HAVE_QUOTA_SUPPORT
4653         next = m->mdt_child;
4654         rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4655         if (rc)
4656                 GOTO(err_llog_cleanup, rc);
4657 #endif
4658
4659         server_put_mount_2(dev, lmi->lmi_mnt);
4660         lmi = NULL;
4661
4662         target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4663
4664         rc = mdt_start_ptlrpc_service(m);
4665         if (rc)
4666                 GOTO(err_recovery, rc);
4667
4668         ping_evictor_start();
4669
4670         rc = lu_site_init_finish(s);
4671         if (rc)
4672                 GOTO(err_stop_service, rc);
4673
4674         if (obd->obd_recovering == 0)
4675                 mdt_postrecov(env, m);
4676
4677         mdt_init_capa_ctxt(env, m);
4678
4679         /* Reduce the initial timeout on an MDS because it doesn't need such
4680          * a long timeout as an OST does. Adaptive timeouts will adjust this
4681          * value appropriately. */
4682         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4683                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4684
4685         RETURN(0);
4686
4687 err_stop_service:
4688         ping_evictor_stop();
4689         mdt_stop_ptlrpc_service(m);
4690 err_recovery:
4691         target_recovery_fini(obd);
4692 #ifdef HAVE_QUOTA_SUPPORT
4693         next->md_ops->mdo_quota.mqo_cleanup(env, next);
4694 #endif
4695 err_llog_cleanup:
4696         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4697         mdt_obd_llog_cleanup(obd);
4698 err_fs_cleanup:
4699         mdt_fs_cleanup(env, m);
4700 err_capa:
4701         cfs_timer_disarm(&m->mdt_ck_timer);
4702         mdt_ck_thread_stop(m);
4703 err_free_ns:
4704         upcall_cache_cleanup(m->mdt_identity_cache);
4705         m->mdt_identity_cache = NULL;
4706         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4707         obd->obd_namespace = m->mdt_namespace = NULL;
4708 err_fini_seq:
4709         mdt_seq_fini(env, m);
4710 err_fini_fld:
4711         mdt_fld_fini(env, m);
4712 err_lut:
4713         lut_fini(env, &m->mdt_lut);
4714 err_fini_stack:
4715         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4716 err_fini_proc:
4717         mdt_procfs_fini(m);
4718         lu_site_fini(s);
4719 err_free_site:
4720         OBD_FREE_PTR(mite);
4721 err_lmi:
4722         if (lmi)
4723                 server_put_mount_2(dev, lmi->lmi_mnt);
4724         return (rc);
4725 }
4726
4727 /* used by MGS to process specific configurations */
4728 static int mdt_process_config(const struct lu_env *env,
4729                               struct lu_device *d, struct lustre_cfg *cfg)
4730 {
4731         struct mdt_device *m = mdt_dev(d);
4732         struct md_device *md_next = m->mdt_child;
4733         struct lu_device *next = md2lu_dev(md_next);
4734         int rc = 0;
4735         ENTRY;
4736
4737         switch (cfg->lcfg_command) {
4738         case LCFG_PARAM: {
4739                 struct lprocfs_static_vars lvars;
4740                 struct obd_device *obd = d->ld_obd;
4741
4742                 /*
4743                  * For interoperability between 1.8 and 2.0,
4744                  * skip old "mdt.group_upcall" param.
4745                  */
4746                 {
4747                         char *param = lustre_cfg_string(cfg, 1);
4748                         if (param && !strncmp("mdt.group_upcall", param, 16)) {
4749                                 CWARN("For 1.8 interoperability, skip this"
4750                                        " mdt.group_upcall. It is obsolete\n");
4751                                 break;
4752                         }
4753                 }
4754
4755                 lprocfs_mdt_init_vars(&lvars);
4756                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4757                                               cfg, obd);
4758                 if (rc > 0 || rc == -ENOSYS)
4759                         /* we don't understand; pass it on */
4760                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4761                 break;
4762         }
4763         case LCFG_ADD_MDC:
4764                 /*
4765                  * Add mdc hook to get first MDT uuid and connect it to
4766                  * ls->controller to use for seq manager.
4767                  */
4768                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4769                 if (rc)
4770                         CERROR("Can't add mdc, rc %d\n", rc);
4771                 else
4772                         rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4773                 break;
4774         default:
4775                 /* others are passed further */
4776                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4777                 break;
4778         }
4779         RETURN(rc);
4780 }
4781
4782 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4783                                           const struct lu_object_header *hdr,
4784                                           struct lu_device *d)
4785 {
4786         struct mdt_object *mo;
4787
4788         ENTRY;
4789
4790         OBD_ALLOC_PTR(mo);
4791         if (mo != NULL) {
4792                 struct lu_object *o;
4793                 struct lu_object_header *h;
4794
4795                 o = &mo->mot_obj.mo_lu;
4796                 h = &mo->mot_header;
4797                 lu_object_header_init(h);
4798                 lu_object_init(o, h, d);
4799                 lu_object_add_top(h, o);
4800                 o->lo_ops = &mdt_obj_ops;
4801                 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4802                 RETURN(o);
4803         } else
4804                 RETURN(NULL);
4805 }
4806
4807 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4808                            const struct lu_object_conf *unused)
4809 {
4810         struct mdt_device *d = mdt_dev(o->lo_dev);
4811         struct lu_device  *under;
4812         struct lu_object  *below;
4813         int                rc = 0;
4814         ENTRY;
4815
4816         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4817                PFID(lu_object_fid(o)));
4818
4819         under = &d->mdt_child->md_lu_dev;
4820         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4821         if (below != NULL) {
4822                 lu_object_add(o, below);
4823         } else
4824                 rc = -ENOMEM;
4825
4826         RETURN(rc);
4827 }
4828
4829 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4830 {
4831         struct mdt_object *mo = mdt_obj(o);
4832         struct lu_object_header *h;
4833         ENTRY;
4834
4835         h = o->lo_header;
4836         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4837                PFID(lu_object_fid(o)));
4838
4839         lu_object_fini(o);
4840         lu_object_header_fini(h);
4841         OBD_FREE_PTR(mo);
4842         EXIT;
4843 }
4844
4845 static int mdt_object_print(const struct lu_env *env, void *cookie,
4846                             lu_printer_t p, const struct lu_object *o)
4847 {
4848         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4849         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4850                     "flags="LPX64", epochcount=%d, writecount=%d)",
4851                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4852                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4853 }
4854
4855 static const struct lu_device_operations mdt_lu_ops = {
4856         .ldo_object_alloc   = mdt_object_alloc,
4857         .ldo_process_config = mdt_process_config,
4858 };
4859
4860 static const struct lu_object_operations mdt_obj_ops = {
4861         .loo_object_init    = mdt_object_init,
4862         .loo_object_free    = mdt_object_free,
4863         .loo_object_print   = mdt_object_print
4864 };
4865
4866 static int mdt_obd_set_info_async(struct obd_export *exp,
4867                                   __u32 keylen, void *key,
4868                                   __u32 vallen, void *val,
4869                                   struct ptlrpc_request_set *set)
4870 {
4871         struct obd_device     *obd = exp->exp_obd;
4872         int                    rc;
4873         ENTRY;
4874
4875         LASSERT(obd);
4876
4877         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4878                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4879                 RETURN(rc);
4880         }
4881
4882         RETURN(0);
4883 }
4884
4885 /* mds_connect_internal */
4886 static int mdt_connect_internal(struct obd_export *exp,
4887                                 struct mdt_device *mdt,
4888                                 struct obd_connect_data *data)
4889 {
4890         if (data != NULL) {
4891                 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4892                 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4893
4894                 /* If no known bits (which should not happen, probably,
4895                    as everybody should support LOOKUP and UPDATE bits at least)
4896                    revert to compat mode with plain locks. */
4897                 if (!data->ocd_ibits_known &&
4898                     data->ocd_connect_flags & OBD_CONNECT_IBITS)
4899                         data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4900
4901                 if (!mdt->mdt_opts.mo_acl)
4902                         data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4903
4904                 if (!mdt->mdt_opts.mo_user_xattr)
4905                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4906
4907                 if (!mdt->mdt_som_conf)
4908                         data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4909
4910                 cfs_spin_lock(&exp->exp_lock);
4911                 exp->exp_connect_flags = data->ocd_connect_flags;
4912                 cfs_spin_unlock(&exp->exp_lock);
4913                 data->ocd_version = LUSTRE_VERSION_CODE;
4914                 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4915         }
4916
4917 #if 0
4918         if (mdt->mdt_opts.mo_acl &&
4919             ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4920                 CWARN("%s: MDS requires ACL support but client does not\n",
4921                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4922                 return -EBADE;
4923         }
4924 #endif
4925
4926         if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4927                 CWARN("%s: MDS requires FID support, but client not\n",
4928                       mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4929                 return -EBADE;
4930         }
4931
4932         if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4933             !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4934                 CWARN("%s: MDS has SOM enabled, but client does not support "
4935                       "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4936                 return -EBADE;
4937         }
4938
4939         return 0;
4940 }
4941
4942 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4943                                      struct obd_export *exp,
4944                                      struct ptlrpc_request *req)
4945 {
4946         struct sptlrpc_flavor   flvr;
4947         int                     rc = 0;
4948
4949         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4950                 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4951                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4952                                              req->rq_sp_from,
4953                                              req->rq_peer.nid,
4954                                              &flvr);
4955                 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4956
4957                 cfs_spin_lock(&exp->exp_lock);
4958
4959                 exp->exp_sp_peer = req->rq_sp_from;
4960                 exp->exp_flvr = flvr;
4961
4962                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4963                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4964                         CERROR("unauthorized rpc flavor %x from %s, "
4965                                "expect %x\n", req->rq_flvr.sf_rpc,
4966                                libcfs_nid2str(req->rq_peer.nid),
4967                                exp->exp_flvr.sf_rpc);
4968                         rc = -EACCES;
4969                 }
4970
4971                 cfs_spin_unlock(&exp->exp_lock);
4972         } else {
4973                 if (exp->exp_sp_peer != req->rq_sp_from) {
4974                         CERROR("RPC source %s doesn't match %s\n",
4975                                sptlrpc_part2name(req->rq_sp_from),
4976                                sptlrpc_part2name(exp->exp_sp_peer));
4977                         rc = -EACCES;
4978                 } else {
4979                         rc = sptlrpc_target_export_check(exp, req);
4980                 }
4981         }
4982
4983         return rc;
4984 }
4985
4986 /* mds_connect copy */
4987 static int mdt_obd_connect(const struct lu_env *env,
4988                            struct obd_export **exp, struct obd_device *obd,
4989                            struct obd_uuid *cluuid,
4990                            struct obd_connect_data *data,
4991                            void *localdata)
4992 {
4993         struct mdt_thread_info *info;
4994         struct lsd_client_data *lcd;
4995         struct obd_export      *lexp;
4996         struct lustre_handle    conn = { 0 };
4997         struct mdt_device      *mdt;
4998         struct ptlrpc_request  *req;
4999         int                     rc;
5000         ENTRY;
5001
5002         LASSERT(env != NULL);
5003         if (!exp || !obd || !cluuid)
5004                 RETURN(-EINVAL);
5005
5006         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5007         req = info->mti_pill->rc_req;
5008         mdt = mdt_dev(obd->obd_lu_dev);
5009
5010         rc = class_connect(&conn, obd, cluuid);
5011         if (rc)
5012                 RETURN(rc);
5013
5014         lexp = class_conn2export(&conn);
5015         LASSERT(lexp != NULL);
5016
5017         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5018         if (rc)
5019                 GOTO(out, rc);
5020
5021         rc = mdt_connect_internal(lexp, mdt, data);
5022         if (rc == 0) {
5023                 OBD_ALLOC_PTR(lcd);
5024                 if (lcd != NULL) {
5025                         struct mdt_thread_info *mti;
5026                         mti = lu_context_key_get(&env->le_ctx,
5027                                                  &mdt_thread_key);
5028                         LASSERT(mti != NULL);
5029                         mti->mti_exp = lexp;
5030                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5031                         lexp->exp_mdt_data.med_lcd = lcd;
5032                         rc = mdt_client_new(env, mdt);
5033                         if (rc == 0)
5034                                 mdt_export_stats_init(obd, lexp, localdata);
5035                 } else {
5036                         rc = -ENOMEM;
5037                 }
5038         }
5039
5040 out:
5041         if (rc != 0)
5042                 class_disconnect(lexp);
5043         else
5044                 *exp = lexp;
5045
5046         RETURN(rc);
5047 }
5048
5049 static int mdt_obd_reconnect(const struct lu_env *env,
5050                              struct obd_export *exp, struct obd_device *obd,
5051                              struct obd_uuid *cluuid,
5052                              struct obd_connect_data *data,
5053                              void *localdata)
5054 {
5055         struct mdt_thread_info *info;
5056         struct mdt_device      *mdt;
5057         struct ptlrpc_request  *req;
5058         int                     rc;
5059         ENTRY;
5060
5061         if (exp == NULL || obd == NULL || cluuid == NULL)
5062                 RETURN(-EINVAL);
5063
5064         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5065         req = info->mti_pill->rc_req;
5066         mdt = mdt_dev(obd->obd_lu_dev);
5067
5068         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5069         if (rc)
5070                 RETURN(rc);
5071
5072         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5073         if (rc == 0)
5074                 mdt_export_stats_init(obd, exp, localdata);
5075
5076         RETURN(rc);
5077 }
5078 static int mdt_export_cleanup(struct obd_export *exp)
5079 {
5080         struct mdt_export_data *med = &exp->exp_mdt_data;
5081         struct obd_device      *obd = exp->exp_obd;
5082         struct mdt_device      *mdt;
5083         struct mdt_thread_info *info;
5084         struct lu_env           env;
5085         CFS_LIST_HEAD(closing_list);
5086         struct mdt_file_data *mfd, *n;
5087         int rc = 0;
5088         ENTRY;
5089
5090         cfs_spin_lock(&med->med_open_lock);
5091         while (!cfs_list_empty(&med->med_open_head)) {
5092                 cfs_list_t *tmp = med->med_open_head.next;
5093                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5094
5095                 /* Remove mfd handle so it can't be found again.
5096                  * We are consuming the mfd_list reference here. */
5097                 class_handle_unhash(&mfd->mfd_handle);
5098                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5099         }
5100         cfs_spin_unlock(&med->med_open_lock);
5101         mdt = mdt_dev(obd->obd_lu_dev);
5102         LASSERT(mdt != NULL);
5103
5104         rc = lu_env_init(&env, LCT_MD_THREAD);
5105         if (rc)
5106                 RETURN(rc);
5107
5108         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5109         LASSERT(info != NULL);
5110         memset(info, 0, sizeof *info);
5111         info->mti_env = &env;
5112         info->mti_mdt = mdt;
5113         info->mti_exp = exp;
5114
5115         if (!cfs_list_empty(&closing_list)) {
5116                 struct md_attr *ma = &info->mti_attr;
5117                 int lmm_size;
5118                 int cookie_size;
5119
5120                 lmm_size = mdt->mdt_max_mdsize;
5121                 OBD_ALLOC(ma->ma_lmm, lmm_size);
5122                 if (ma->ma_lmm == NULL)
5123                         GOTO(out_lmm, rc = -ENOMEM);
5124
5125                 cookie_size = mdt->mdt_max_cookiesize;
5126                 OBD_ALLOC(ma->ma_cookie, cookie_size);
5127                 if (ma->ma_cookie == NULL)
5128                         GOTO(out_cookie, rc = -ENOMEM);
5129
5130                 /* Close any open files (which may also cause orphan unlinking). */
5131                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5132                         cfs_list_del_init(&mfd->mfd_list);
5133                         memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5134                         ma->ma_lmm_size = lmm_size;
5135                         ma->ma_cookie_size = cookie_size;
5136                         ma->ma_need = 0;
5137                         /* It is not for setattr, just tell MDD to send
5138                          * DESTROY RPC to OSS if needed */
5139                         ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5140                         ma->ma_valid = MA_FLAGS;
5141                         mdt_mfd_close(info, mfd);
5142                 }
5143                 OBD_FREE(ma->ma_cookie, cookie_size);
5144                 ma->ma_cookie = NULL;
5145 out_cookie:
5146                 OBD_FREE(ma->ma_lmm, lmm_size);
5147                 ma->ma_lmm = NULL;
5148         }
5149 out_lmm:
5150         info->mti_mdt = NULL;
5151         /* cleanup client slot early */
5152         mdt_client_del(&env, mdt);
5153         lu_env_fini(&env);
5154
5155         RETURN(rc);
5156 }
5157
5158 static int mdt_obd_disconnect(struct obd_export *exp)
5159 {
5160         int rc;
5161         ENTRY;
5162
5163         LASSERT(exp);
5164         class_export_get(exp);
5165
5166         rc = server_disconnect_export(exp);
5167         if (rc != 0)
5168                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5169
5170         rc = mdt_export_cleanup(exp);
5171         class_export_put(exp);
5172         RETURN(rc);
5173 }
5174
5175 /* FIXME: Can we avoid using these two interfaces? */
5176 static int mdt_init_export(struct obd_export *exp)
5177 {
5178         struct mdt_export_data *med = &exp->exp_mdt_data;
5179         int                     rc;
5180         ENTRY;
5181
5182         CFS_INIT_LIST_HEAD(&med->med_open_head);
5183         cfs_spin_lock_init(&med->med_open_lock);
5184         cfs_sema_init(&med->med_idmap_sem, 1);
5185         med->med_idmap = NULL;
5186         cfs_spin_lock(&exp->exp_lock);
5187         exp->exp_connecting = 1;
5188         cfs_spin_unlock(&exp->exp_lock);
5189         rc = ldlm_init_export(exp);
5190         if (rc)
5191                 CERROR("Error %d while initializing export\n", rc);
5192         RETURN(rc);
5193 }
5194
5195 static int mdt_destroy_export(struct obd_export *exp)
5196 {
5197         struct mdt_export_data *med;
5198         int rc = 0;
5199         ENTRY;
5200
5201         med = &exp->exp_mdt_data;
5202         if (exp_connect_rmtclient(exp))
5203                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5204
5205         target_destroy_export(exp);
5206         ldlm_destroy_export(exp);
5207
5208         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5209         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5210         if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5211                 RETURN(0);
5212
5213         lut_client_free(exp);
5214         RETURN(rc);
5215 }
5216
5217 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5218 {
5219         if (flag & CONFIG_LOG)
5220                 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5221
5222         /* also notify active event */
5223         if (flag & CONFIG_SYNC)
5224                 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5225
5226         if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5227             cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5228                 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5229
5230                 /* Open for clients */
5231                 if (obd->obd_no_conn) {
5232                         cfs_spin_lock_bh(&obd->obd_processing_task_lock);
5233                         obd->obd_no_conn = 0;
5234                         cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
5235                 }
5236         }
5237 }
5238
5239 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5240                       enum md_upcall_event ev, void *data)
5241 {
5242         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5243         struct md_device  *next  = m->mdt_child;
5244         struct mdt_thread_info *mti;
5245         int rc = 0;
5246         ENTRY;
5247
5248         switch (ev) {
5249                 case MD_LOV_SYNC:
5250                         rc = next->md_ops->mdo_maxsize_get(env, next,
5251                                         &m->mdt_max_mdsize,
5252                                         &m->mdt_max_cookiesize);
5253                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5254                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
5255                         mdt_allow_cli(m, CONFIG_SYNC);
5256                         if (data)
5257                                 (*(__u64 *)data) = m->mdt_mount_count;
5258                         break;
5259                 case MD_NO_TRANS:
5260                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5261                         mti->mti_no_need_trans = 1;
5262                         CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5263                         break;
5264                 case MD_LOV_CONFIG:
5265                         /* Check that MDT is not yet configured */
5266                         LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5267                         break;
5268 #ifdef HAVE_QUOTA_SUPPORT
5269                 case MD_LOV_QUOTA:
5270                         if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5271                             likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5272                                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5273                         break;
5274 #endif
5275                 default:
5276                         CERROR("invalid event\n");
5277                         rc = -EINVAL;
5278                         break;
5279         }
5280         RETURN(rc);
5281 }
5282
5283 static int mdt_obd_notify(struct obd_device *host,
5284                           struct obd_device *watched,
5285                           enum obd_notify_event ev, void *data)
5286 {
5287         struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5288 #ifdef HAVE_QUOTA_SUPPORT
5289         struct md_device *next = mdt->mdt_child;
5290 #endif
5291         ENTRY;
5292
5293         switch (ev) {
5294         case OBD_NOTIFY_CONFIG:
5295                 mdt_allow_cli(mdt, (unsigned long)data);
5296
5297 #ifdef HAVE_QUOTA_SUPPORT
5298                /* quota_type has been processed, we can now handle
5299                 * incoming quota requests */
5300                 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5301 #endif
5302                 break;
5303         default:
5304                 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5305         }
5306         RETURN(0);
5307 }
5308
5309 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5310                             void *val, int vallen)
5311 {
5312         struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5313         struct getinfo_fid2path *fpout, *fpin;
5314         int rc = 0;
5315
5316         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5317         fpout = val;
5318
5319         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5320                 lustre_swab_fid2path(fpin);
5321
5322         memcpy(fpout, fpin, sizeof(*fpin));
5323         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5324                 RETURN(-EINVAL);
5325
5326         rc = mdt_fid2path(info->mti_env, mdt, fpout);
5327         RETURN(rc);
5328 }
5329
5330 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5331                         struct getinfo_fid2path *fp)
5332 {
5333         struct mdt_object *obj;
5334         int    rc;
5335         ENTRY;
5336
5337         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5338                PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5339
5340         if (!fid_is_sane(&fp->gf_fid))
5341                 RETURN(-EINVAL);
5342
5343         obj = mdt_object_find(env, mdt, &fp->gf_fid);
5344         if (obj == NULL || IS_ERR(obj)) {
5345                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5346                        PTR_ERR(obj));
5347                 RETURN(-EINVAL);
5348         }
5349
5350         rc = lu_object_exists(&obj->mot_obj.mo_lu);
5351         if (rc <= 0) {
5352                 if (rc == -1)
5353                         rc = -EREMOTE;
5354                 else
5355                         rc = -ENOENT;
5356                 mdt_object_put(env, obj);
5357                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5358                        PFID(&fp->gf_fid), rc);
5359                 RETURN(rc);
5360         }
5361
5362         rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5363                      fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5364         mdt_object_put(env, obj);
5365
5366         RETURN(rc);
5367 }
5368
5369 static int mdt_get_info(struct mdt_thread_info *info)
5370 {
5371         struct ptlrpc_request *req = mdt_info_req(info);
5372         char *key;
5373         int keylen;
5374         __u32 *vallen;
5375         void *valout;
5376         int rc;
5377         ENTRY;
5378
5379         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5380         if (key == NULL) {
5381                 CDEBUG(D_IOCTL, "No GETINFO key");
5382                 RETURN(-EFAULT);
5383         }
5384         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5385                                       RCL_CLIENT);
5386
5387         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5388         if (vallen == NULL) {
5389                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5390                 RETURN(-EFAULT);
5391         }
5392
5393         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5394                              *vallen);
5395         rc = req_capsule_server_pack(info->mti_pill);
5396         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5397         if (valout == NULL) {
5398                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5399                 RETURN(-EFAULT);
5400         }
5401
5402         if (KEY_IS(KEY_FID2PATH))
5403                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5404         else
5405                 rc = -EINVAL;
5406
5407         lustre_msg_set_status(req->rq_repmsg, rc);
5408
5409         RETURN(rc);
5410 }
5411
5412 /* Pass the ioc down */
5413 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5414                          unsigned int cmd, int len, void *data)
5415 {
5416         struct lu_context ioctl_session;
5417         struct md_device *next = mdt->mdt_child;
5418         int rc;
5419         ENTRY;
5420
5421         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5422         if (rc)
5423                 RETURN(rc);
5424         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5425         lu_context_enter(&ioctl_session);
5426         env->le_ses = &ioctl_session;
5427
5428         LASSERT(next->md_ops->mdo_iocontrol);
5429         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5430
5431         lu_context_exit(&ioctl_session);
5432         lu_context_fini(&ioctl_session);
5433         RETURN(rc);
5434 }
5435
5436 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5437 {
5438         struct obd_ioctl_data *data = karg;
5439         struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5440         __u64 version;
5441         struct mdt_object *obj;
5442         struct mdt_lock_handle  *lh;
5443         int rc;
5444         ENTRY;
5445         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5446         if (!fid_is_sane(fid))
5447                 RETURN(-EINVAL);
5448
5449         lh = &mti->mti_lh[MDT_LH_PARENT];
5450         mdt_lock_reg_init(lh, LCK_CR);
5451
5452         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5453         if (IS_ERR(obj))
5454                 RETURN(PTR_ERR(obj));
5455
5456         rc = mdt_object_exists(obj);
5457         if (rc < 0) {
5458                 rc = -EREMOTE;
5459                 /**
5460                  * before calling version get the correct MDS should be
5461                  * fid, this is error to find remote object here
5462                  */
5463                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5464         } else if (rc == 0) {
5465                 rc = -ENOENT;
5466                 CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
5467         } else {
5468                 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5469                 if (version < 0) {
5470                         rc = (int)version;
5471                 } else {
5472                         *(__u64 *)data->ioc_inlbuf2 = version;
5473                         rc = 0;
5474                 }
5475         }
5476         mdt_object_unlock_put(mti, obj, lh, 1);
5477         RETURN(rc);
5478 }
5479
5480 /* ioctls on obd dev */
5481 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5482                          void *karg, void *uarg)
5483 {
5484         struct lu_env      env;
5485         struct obd_device *obd = exp->exp_obd;
5486         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5487         struct dt_device  *dt = mdt->mdt_bottom;
5488         int rc;
5489
5490         ENTRY;
5491         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5492         rc = lu_env_init(&env, LCT_MD_THREAD);
5493         if (rc)
5494                 RETURN(rc);
5495
5496         switch (cmd) {
5497         case OBD_IOC_SYNC:
5498                 rc = mdt_device_sync(&env, mdt);
5499                 break;
5500         case OBD_IOC_SET_READONLY:
5501                 dt->dd_ops->dt_ro(&env, dt);
5502                 break;
5503         case OBD_IOC_ABORT_RECOVERY:
5504                 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5505                 target_stop_recovery_thread(obd);
5506                 rc = 0;
5507                 break;
5508         case OBD_IOC_CHANGELOG_REG:
5509         case OBD_IOC_CHANGELOG_DEREG:
5510         case OBD_IOC_CHANGELOG_CLEAR:
5511                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5512                 break;
5513         case OBD_IOC_GET_OBJ_VERSION: {
5514                 struct mdt_thread_info *mti;
5515                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5516                 memset(mti, 0, sizeof *mti);
5517                 mti->mti_env = &env;
5518                 mti->mti_mdt = mdt;
5519                 mti->mti_exp = exp;
5520
5521                 rc = mdt_ioc_version_get(mti, karg);
5522                 break;
5523         }
5524         default:
5525                 CERROR("Not supported cmd = %d for device %s\n",
5526                        cmd, obd->obd_name);
5527                 rc = -EOPNOTSUPP;
5528         }
5529
5530         lu_env_fini(&env);
5531         RETURN(rc);
5532 }
5533
5534 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5535 {
5536         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5537         struct obd_device *obd = mdt2obd_dev(mdt);
5538 #ifdef HAVE_QUOTA_SUPPORT
5539         struct md_device *next = mdt->mdt_child;
5540 #endif
5541         int rc, lost;
5542         ENTRY;
5543         /* if some clients didn't participate in recovery then we can possibly
5544          * lost sequence. Now we should increase sequence for safe value */
5545         lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
5546         mdt_seq_adjust(env, mdt, lost);
5547
5548         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5549 #ifdef HAVE_QUOTA_SUPPORT
5550         if (likely(obd->obd_stopping == 0))
5551                 next->md_ops->mdo_quota.mqo_recovery(env, next);
5552 #endif
5553         RETURN(rc);
5554 }
5555
5556 int mdt_obd_postrecov(struct obd_device *obd)
5557 {
5558         struct lu_env env;
5559         int rc;
5560
5561         rc = lu_env_init(&env, LCT_MD_THREAD);
5562         if (rc)
5563                 RETURN(rc);
5564         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5565         lu_env_fini(&env);
5566         return rc;
5567 }
5568
5569 /**
5570  * Send a copytool req to a client
5571  * Note this sends a request RPC from a server (MDT) to a client (MDC),
5572  * backwards of normal comms.
5573  */
5574 int mdt_hsm_copytool_send(struct obd_export *exp)
5575 {
5576         struct kuc_hdr *lh;
5577         struct hsm_action_list *hal;
5578         struct hsm_action_item *hai;
5579         int rc, len;
5580         ENTRY;
5581
5582         CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5583               libcfs_nid2str(exp->exp_connection->c_peer.nid));
5584
5585         len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5586                 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5587         OBD_ALLOC(lh, len);
5588         if (lh == NULL)
5589                 RETURN(-ENOMEM);
5590
5591         lh->kuc_magic = KUC_MAGIC;
5592         lh->kuc_transport = KUC_TRANSPORT_HSM;
5593         lh->kuc_msgtype = HMT_ACTION_LIST;
5594         lh->kuc_msglen = len;
5595
5596         hal = (struct hsm_action_list *)(lh + 1);
5597         hal->hal_version = HAL_VERSION;
5598         hal->hal_archive_num = 1;
5599         obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5600                         MTI_NAME_MAXLEN);
5601
5602         /* mock up an action list */
5603         hal->hal_count = 2;
5604         hai = hai_zero(hal);
5605         hai->hai_action = HSMA_ARCHIVE;
5606         hai->hai_fid.f_oid = 0xA00A;
5607         hai->hai_len = sizeof(*hai);
5608         hai = hai_next(hai);
5609         hai->hai_action = HSMA_RESTORE;
5610         hai->hai_fid.f_oid = 0xB00B;
5611         hai->hai_len = sizeof(*hai);
5612
5613         /* Uses the ldlm reverse import; this rpc will be seen by
5614           the ldlm_callback_handler */
5615         rc = do_set_info_async(exp->exp_imp_reverse,
5616                                LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5617                                sizeof(KEY_HSM_COPYTOOL_SEND),
5618                                KEY_HSM_COPYTOOL_SEND,
5619                                len, lh, NULL);
5620
5621         OBD_FREE(lh, len);
5622
5623         RETURN(rc);
5624 }
5625
5626 static struct obd_ops mdt_obd_device_ops = {
5627         .o_owner          = THIS_MODULE,
5628         .o_set_info_async = mdt_obd_set_info_async,
5629         .o_connect        = mdt_obd_connect,
5630         .o_reconnect      = mdt_obd_reconnect,
5631         .o_disconnect     = mdt_obd_disconnect,
5632         .o_init_export    = mdt_init_export,
5633         .o_destroy_export = mdt_destroy_export,
5634         .o_iocontrol      = mdt_iocontrol,
5635         .o_postrecov      = mdt_obd_postrecov,
5636         .o_notify         = mdt_obd_notify
5637 };
5638
5639 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5640                                          struct lu_device *d)
5641 {
5642         struct mdt_device *m = mdt_dev(d);
5643         ENTRY;
5644
5645         mdt_fini(env, m);
5646         RETURN(NULL);
5647 }
5648
5649 static struct lu_device *mdt_device_free(const struct lu_env *env,
5650                                          struct lu_device *d)
5651 {
5652         struct mdt_device *m = mdt_dev(d);
5653         ENTRY;
5654
5655         md_device_fini(&m->mdt_md_dev);
5656         OBD_FREE_PTR(m);
5657         RETURN(NULL);
5658 }
5659
5660 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5661                                           struct lu_device_type *t,
5662                                           struct lustre_cfg *cfg)
5663 {
5664         struct lu_device  *l;
5665         struct mdt_device *m;
5666
5667         OBD_ALLOC_PTR(m);
5668         if (m != NULL) {
5669                 int rc;
5670
5671                 l = &m->mdt_md_dev.md_lu_dev;
5672                 rc = mdt_init0(env, m, t, cfg);
5673                 if (rc != 0) {
5674                         mdt_device_free(env, l);
5675                         l = ERR_PTR(rc);
5676                         return l;
5677                 }
5678                 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5679         } else
5680                 l = ERR_PTR(-ENOMEM);
5681         return l;
5682 }
5683
5684 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5685 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5686
5687 /* context key: mdt_thread_key */
5688 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5689
5690 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5691 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5692
5693 struct lu_context_key mdt_txn_key = {
5694         .lct_tags = LCT_TX_HANDLE,
5695         .lct_init = mdt_txn_key_init,
5696         .lct_fini = mdt_txn_key_fini
5697 };
5698
5699 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5700 {
5701         return md_ucred(info->mti_env);
5702 }
5703
5704 /**
5705  * Enable/disable COS (Commit On Sharing).
5706  *
5707  * Set/Clear the COS flag in mdt options.
5708  *
5709  * \param mdt mdt device
5710  * \param val 0 disables COS, other values enable COS
5711  */
5712 void mdt_enable_cos(struct mdt_device *mdt, int val)
5713 {
5714         struct lu_env env;
5715         int rc;
5716
5717         mdt->mdt_opts.mo_cos = !!val;
5718         rc = lu_env_init(&env, LCT_MD_THREAD);
5719         if (unlikely(rc != 0)) {
5720                 CWARN("lu_env initialization failed with rc = %d,"
5721                       "cannot sync\n", rc);
5722                 return;
5723         }
5724         mdt_device_sync(&env, mdt);
5725         lu_env_fini(&env);
5726 }
5727
5728 /**
5729  * Check COS (Commit On Sharing) status.
5730  *
5731  * Return COS flag status.
5732  *
5733  * \param mdt mdt device
5734  */
5735 int mdt_cos_is_enabled(struct mdt_device *mdt)
5736 {
5737         return mdt->mdt_opts.mo_cos != 0;
5738 }
5739
5740 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5741 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5742
5743 static struct lu_device_type_operations mdt_device_type_ops = {
5744         .ldto_init = mdt_type_init,
5745         .ldto_fini = mdt_type_fini,
5746
5747         .ldto_start = mdt_type_start,
5748         .ldto_stop  = mdt_type_stop,
5749
5750         .ldto_device_alloc = mdt_device_alloc,
5751         .ldto_device_free  = mdt_device_free,
5752         .ldto_device_fini  = mdt_device_fini
5753 };
5754
5755 static struct lu_device_type mdt_device_type = {
5756         .ldt_tags     = LU_DEVICE_MD,
5757         .ldt_name     = LUSTRE_MDT_NAME,
5758         .ldt_ops      = &mdt_device_type_ops,
5759         .ldt_ctx_tags = LCT_MD_THREAD
5760 };
5761
5762 static struct lu_local_obj_desc mdt_last_recv = {
5763         .llod_name      = LAST_RCVD,
5764         .llod_oid       = MDT_LAST_RECV_OID,
5765         .llod_is_index  = 0,
5766 };
5767
5768 static int __init mdt_mod_init(void)
5769 {
5770         struct lprocfs_static_vars lvars;
5771         int rc;
5772
5773         llo_local_obj_register(&mdt_last_recv);
5774
5775         if (mdt_num_threads > 0) {
5776                 if (mdt_num_threads > MDT_MAX_THREADS)
5777                         mdt_num_threads = MDT_MAX_THREADS;
5778                 if (mdt_num_threads < MDT_MIN_THREADS)
5779                         mdt_num_threads = MDT_MIN_THREADS;
5780                 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5781         } else {
5782                 mdt_max_threads = MDT_MAX_THREADS;
5783                 mdt_min_threads = MDT_MIN_THREADS;
5784                 if (mdt_min_threads < MDT_NUM_THREADS)
5785                         mdt_min_threads = MDT_NUM_THREADS;
5786         }
5787
5788         lprocfs_mdt_init_vars(&lvars);
5789         rc = class_register_type(&mdt_obd_device_ops, NULL,
5790                                  lvars.module_vars, LUSTRE_MDT_NAME,
5791                                  &mdt_device_type);
5792
5793         return rc;
5794 }
5795
5796 static void __exit mdt_mod_exit(void)
5797 {
5798         llo_local_obj_unregister(&mdt_last_recv);
5799         class_unregister_type(LUSTRE_MDT_NAME);
5800 }
5801
5802
5803 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt)             \
5804 [prefix ## _ ## opc - prefix ## _ ## base] = {                          \
5805         .mh_name    = #opc,                                             \
5806         .mh_fail_id = OBD_FAIL_ ## prefix ## _  ## opc ## suffix,       \
5807         .mh_opc     = prefix ## _  ## opc,                              \
5808         .mh_flags   = flags,                                            \
5809         .mh_act     = fn,                                               \
5810         .mh_fmt     = fmt                                               \
5811 }
5812
5813 #define DEF_MDT_HNDL(flags, name, fn, fmt)                                  \
5814         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5815
5816 #define DEF_SEQ_HNDL(flags, name, fn, fmt)                      \
5817         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5818
5819 #define DEF_FLD_HNDL(flags, name, fn, fmt)                      \
5820         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5821 /*
5822  * Request with a format known in advance
5823  */
5824 #define DEF_MDT_HNDL_F(flags, name, fn)                                 \
5825         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5826
5827 #define DEF_SEQ_HNDL_F(flags, name, fn)                                 \
5828         DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5829
5830 #define DEF_FLD_HNDL_F(flags, name, fn)                                 \
5831         DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5832 /*
5833  * Request with a format we do not yet know
5834  */
5835 #define DEF_MDT_HNDL_0(flags, name, fn)                                 \
5836         DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5837
5838 static struct mdt_handler mdt_mds_ops[] = {
5839 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5840 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
5841 DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
5842                                                              &RQF_OBD_SET_INFO),
5843 DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
5844 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
5845 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
5846 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5847 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETXATTR,     mdt_getxattr),
5848 DEF_MDT_HNDL_F(0           |HABEO_REFERO, STATFS,       mdt_statfs),
5849 DEF_MDT_HNDL_F(0           |MUTABOR,      REINT,        mdt_reint),
5850 DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,        mdt_close),
5851 DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING, mdt_done_writing),
5852 DEF_MDT_HNDL_F(0           |HABEO_REFERO, PIN,          mdt_pin),
5853 DEF_MDT_HNDL_0(0,                         SYNC,         mdt_sync),
5854 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5855 #ifdef HAVE_QUOTA_SUPPORT
5856 DEF_MDT_HNDL_F(0,                         QUOTACHECK,   mdt_quotacheck_handle),
5857 DEF_MDT_HNDL_F(0,                         QUOTACTL,     mdt_quotactl_handle)
5858 #endif
5859 };
5860
5861 #define DEF_OBD_HNDL(flags, name, fn)                   \
5862         DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5863
5864
5865 static struct mdt_handler mdt_obd_ops[] = {
5866         DEF_OBD_HNDL(0, PING,           mdt_obd_ping),
5867         DEF_OBD_HNDL(0, LOG_CANCEL,     mdt_obd_log_cancel),
5868         DEF_OBD_HNDL(0, QC_CALLBACK,    mdt_obd_qc_callback)
5869 };
5870
5871 #define DEF_DLM_HNDL_0(flags, name, fn)                   \
5872         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5873 #define DEF_DLM_HNDL_F(flags, name, fn)                   \
5874         DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5875
5876 static struct mdt_handler mdt_dlm_ops[] = {
5877         DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE,        mdt_enqueue),
5878         DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT,        mdt_convert),
5879         DEF_DLM_HNDL_0(0,            BL_CALLBACK,    mdt_bl_callback),
5880         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
5881 };
5882
5883 #define DEF_LLOG_HNDL(flags, name, fn)                   \
5884         DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5885
5886 static struct mdt_handler mdt_llog_ops[] = {
5887         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
5888         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
5889         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5890         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
5891         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
5892         DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
5893         DEF_LLOG_HNDL(0, CATINFO,                   NULL),
5894         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
5895         DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
5896 };
5897
5898 #define DEF_SEC_CTX_HNDL(name, fn)                      \
5899         DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5900
5901 static struct mdt_handler mdt_sec_ctx_ops[] = {
5902         DEF_SEC_CTX_HNDL(INIT,          mdt_sec_ctx_handle),
5903         DEF_SEC_CTX_HNDL(INIT_CONT,     mdt_sec_ctx_handle),
5904         DEF_SEC_CTX_HNDL(FINI,          mdt_sec_ctx_handle)
5905 };
5906
5907 static struct mdt_opc_slice mdt_regular_handlers[] = {
5908         {
5909                 .mos_opc_start = MDS_GETATTR,
5910                 .mos_opc_end   = MDS_LAST_OPC,
5911                 .mos_hs        = mdt_mds_ops
5912         },
5913         {
5914                 .mos_opc_start = OBD_PING,
5915                 .mos_opc_end   = OBD_LAST_OPC,
5916                 .mos_hs        = mdt_obd_ops
5917         },
5918         {
5919                 .mos_opc_start = LDLM_ENQUEUE,
5920                 .mos_opc_end   = LDLM_LAST_OPC,
5921                 .mos_hs        = mdt_dlm_ops
5922         },
5923         {
5924                 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5925                 .mos_opc_end   = LLOG_LAST_OPC,
5926                 .mos_hs        = mdt_llog_ops
5927         },
5928         {
5929                 .mos_opc_start = SEC_CTX_INIT,
5930                 .mos_opc_end   = SEC_LAST_OPC,
5931                 .mos_hs        = mdt_sec_ctx_ops
5932         },
5933         {
5934                 .mos_hs        = NULL
5935         }
5936 };
5937
5938 static struct mdt_handler mdt_readpage_ops[] = {
5939         DEF_MDT_HNDL_F(0,                         CONNECT,  mdt_connect),
5940         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5941 #ifdef HAVE_SPLIT_SUPPORT
5942         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5943 #endif
5944
5945         /*
5946          * XXX: this is ugly and should be fixed one day, see mdc_close() for
5947          * detailed comments. --umka
5948          */
5949         DEF_MDT_HNDL_F(HABEO_CORPUS,              CLOSE,    mdt_close),
5950         DEF_MDT_HNDL_F(HABEO_CORPUS,              DONE_WRITING,    mdt_done_writing),
5951 };
5952
5953 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5954         {
5955                 .mos_opc_start = MDS_GETATTR,
5956                 .mos_opc_end   = MDS_LAST_OPC,
5957                 .mos_hs        = mdt_readpage_ops
5958         },
5959         {
5960                 .mos_hs        = NULL
5961         }
5962 };
5963
5964 static struct mdt_handler mdt_xmds_ops[] = {
5965         DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
5966         DEF_MDT_HNDL_F(HABEO_CORPUS             , GETATTR,      mdt_getattr),
5967         DEF_MDT_HNDL_F(0 | MUTABOR              , REINT,        mdt_reint),
5968         DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR,    mdt_is_subdir),
5969 };
5970
5971 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5972         {
5973                 .mos_opc_start = MDS_GETATTR,
5974                 .mos_opc_end   = MDS_LAST_OPC,
5975                 .mos_hs        = mdt_xmds_ops
5976         },
5977         {
5978                 .mos_opc_start = OBD_PING,
5979                 .mos_opc_end   = OBD_LAST_OPC,
5980                 .mos_hs        = mdt_obd_ops
5981         },
5982         {
5983                 .mos_opc_start = SEC_CTX_INIT,
5984                 .mos_opc_end   = SEC_LAST_OPC,
5985                 .mos_hs        = mdt_sec_ctx_ops
5986         },
5987         {
5988                 .mos_hs        = NULL
5989         }
5990 };
5991
5992 static struct mdt_handler mdt_seq_ops[] = {
5993         DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5994 };
5995
5996 static struct mdt_opc_slice mdt_seq_handlers[] = {
5997         {
5998                 .mos_opc_start = SEQ_QUERY,
5999                 .mos_opc_end   = SEQ_LAST_OPC,
6000                 .mos_hs        = mdt_seq_ops
6001         },
6002         {
6003                 .mos_hs        = NULL
6004         }
6005 };
6006
6007 static struct mdt_handler mdt_fld_ops[] = {
6008         DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
6009 };
6010
6011 static struct mdt_opc_slice mdt_fld_handlers[] = {
6012         {
6013                 .mos_opc_start = FLD_QUERY,
6014                 .mos_opc_end   = FLD_LAST_OPC,
6015                 .mos_hs        = mdt_fld_ops
6016         },
6017         {
6018                 .mos_hs        = NULL
6019         }
6020 };
6021
6022 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6023 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6024 MODULE_LICENSE("GPL");
6025
6026 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6027                 "number of mdt service threads to start");
6028
6029 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);