lustre/lmv/lmv_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/div64.h>
  33 #else
  34 #include <liblustre.h>
  35 #endif
  36 #include <linux/ext2_fs.h>
  37
  38 #include <linux/obd_support.h>
  39 #include <linux/lustre_lib.h>
  40 #include <linux/lustre_net.h>
  41 #include <linux/lustre_idl.h>
  42 #include <linux/lustre_dlm.h>
  43 #include <linux/lustre_mds.h>
  44 #include <linux/obd_class.h>
  45 #include <linux/obd_ost.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/lprocfs_status.h>
  48 #include <linux/lustre_fsfilt.h>
  49 #include <linux/obd_lmv.h>
  50 #include "lmv_internal.h"
  51
  52 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
  53 {
  54         struct lprocfs_static_vars lvars;
  55         struct proc_dir_entry *entry;
  56         int rc;
  57         ENTRY;
  58
  59         lprocfs_init_vars(lmv, &lvars);
  60         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
  61         if (rc)
  62                 RETURN (rc);
  63
  64         entry = create_proc_entry("target_obd", 0444, dev->obd_proc_entry);
  65         if (entry == NULL)
  66                 RETURN(-ENOMEM);
  67         /* entry->proc_fops = &lmv_proc_target_fops; */
  68         entry->data = dev;
  69
  70         RETURN (rc);
  71 }
  72
  73 int lmv_detach(struct obd_device *dev)
  74 {
  75         return lprocfs_obd_detach(dev);
  76 }
  77
  78 static int lmv_connect_fake(struct lustre_handle *conn,
  79                             struct obd_device *obd,
  80                             struct obd_uuid *cluuid)
  81 {
  82         struct lmv_obd *lmv = &obd->u.lmv;
  83         int rc;
  84         ENTRY;
  85
  86         rc = class_connect(conn, obd, cluuid);
  87         if (rc) {
  88                 CERROR("class_connection() returned %d\n", rc);
  89                 RETURN(rc);
  90         }
  91
  92         lmv->exp = class_conn2export(conn);
  93         LASSERT(lmv->exp != NULL);
  94
  95         lmv->cluuid = *cluuid;
  96         lmv->connected = 0;
  97
  98         RETURN(0);
  99 }
 100
 101 int lmv_connect(struct obd_device *obd)
 102 {
 103         struct lmv_obd *lmv = &obd->u.lmv;
 104         struct obd_uuid *cluuid;
 105         struct lmv_tgt_desc *tgts;
 106         struct obd_export *exp;
 107         int rc, i;
 108         ENTRY;
 109
 110         if (lmv->connected)
 111                 RETURN(0);
 112
 113         lmv->connected = 1;
 114         cluuid = &lmv->cluuid;
 115         exp = lmv->exp;
 116         CDEBUG(D_OTHER, "time to connect %s to %s\n",
 117                         cluuid->uuid, obd->obd_name);
 118
 119         /* We don't want to actually do the underlying connections more than
 120          * once, so keep track. */
 121         lmv->refcount++;
 122         if (lmv->refcount > 1) {
 123                 class_export_put(exp);
 124                 RETURN(0);
 125         }
 126
 127         for (i = 0, tgts = lmv->tgts; i < lmv->count; i++, tgts++) {
 128                 struct obd_device *tgt_obd;
 129                 struct obd_uuid lmv_osc_uuid = { "LMV_OSC_UUID" };
 130                 struct lustre_handle conn = {0, };
 131
 132                 LASSERT(tgts != NULL);
 133
 134                 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
 135                                                 &obd->obd_uuid);
 136                 if (!tgt_obd) {
 137                         CERROR("Target %s not attached\n", tgts->uuid.uuid);
 138                         GOTO(out_disc, rc = -EINVAL);
 139                 }
 140
 141                 /* for MDS: don't connect to yourself */
 142                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
 143                         CDEBUG(D_OTHER, "don't connect back to %s\n",
 144                                cluuid->uuid);
 145                         tgts->exp = NULL;
 146                         continue;
 147                 }
 148
 149                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
 150                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 151                         tgts->uuid.uuid, obd->obd_uuid.uuid,
 152                         cluuid->uuid);
 153
 154                 if (!tgt_obd->obd_set_up) {
 155                         CERROR("Target %s not set up\n", tgts->uuid.uuid);
 156                         GOTO(out_disc, rc = -EINVAL);
 157                 }
 158
 159                 rc = obd_connect(&conn, tgt_obd, &lmv_osc_uuid);
 160                 if (rc) {
 161                         CERROR("Target %s connect error %d\n",
 162                                 tgts->uuid.uuid, rc);
 163                         GOTO(out_disc, rc);
 164                 }
 165                 tgts->exp = class_conn2export(&conn);
 166
 167                 obd_init_ea_size(tgts->exp, lmv->max_easize,
 168                                         lmv->max_cookiesize);
 169
 170                 rc = obd_register_observer(tgt_obd, obd);
 171                 if (rc) {
 172                         CERROR("Target %s register_observer error %d\n",
 173                                tgts->uuid.uuid, rc);
 174                         obd_disconnect(tgts->exp, 0);
 175                         GOTO(out_disc, rc);
 176                 }
 177
 178                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
 179                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 180                         atomic_read(&obd->obd_refcount));
 181         }
 182
 183         class_export_put(exp);
 184         RETURN (0);
 185
 186  out_disc:
 187         /* FIXME: cleanup here */
 188         class_disconnect(exp, 0);
 189         RETURN (rc);
 190 }
 191
 192 static int lmv_disconnect(struct obd_export *exp, int flags)
 193 {
 194         struct obd_device *obd = class_exp2obd(exp);
 195         struct lmv_obd *lmv = &obd->u.lmv;
 196         int rc, i;
 197         ENTRY;
 198
 199         if (!lmv->tgts)
 200                 goto out_local;
 201
 202         /* Only disconnect the underlying layers on the final disconnect. */
 203         lmv->refcount--;
 204         if (lmv->refcount != 0)
 205                 goto out_local;
 206
 207         for (i = 0; i < lmv->count; i++) {
 208                 if (lmv->tgts[i].exp == NULL)
 209                         continue;
 210
 211                 if (obd->obd_no_recov) {
 212                         /* Pass it on to our clients.
 213                          * XXX This should be an argument to disconnect,
 214                          * XXX not a back-door flag on the OBD.  Ah well.
 215                          */
 216                         struct obd_device *mdc_obd;
 217                         mdc_obd = class_exp2obd(lmv->tgts[i].exp);
 218                         if (mdc_obd)
 219                                 mdc_obd->obd_no_recov = 1;
 220                 }
 221
 222                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 223                         lmv->tgts[i].exp->exp_obd->obd_name,
 224                         lmv->tgts[i].exp->exp_obd->obd_uuid.uuid);
 225
 226                 obd_register_observer(lmv->tgts[i].exp->exp_obd, NULL);
 227
 228                 rc = obd_disconnect(lmv->tgts[i].exp, flags);
 229                 lmv->tgts[i].exp = NULL;
 230         }
 231
 232  out_local:
 233         /* FIXME: cleanup here */
 234         if (!lmv->connected)
 235                 class_export_put(exp);
 236         rc = class_disconnect(exp, 0);
 237         RETURN(rc);
 238 }
 239
 240 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 241 {
 242         struct lustre_cfg *lcfg = buf;
 243         struct lmv_desc *desc;
 244         struct lmv_obd *lmv = &obd->u.lmv;
 245         struct obd_uuid *uuids;
 246         struct lmv_tgt_desc *tgts;
 247         int i;
 248         int count;
 249         int rc = 0;
 250         ENTRY;
 251
 252         if (lcfg->lcfg_inllen1 < 1) {
 253                 CERROR("LMV setup requires a descriptor\n");
 254                 RETURN(-EINVAL);
 255         }
 256
 257         if (lcfg->lcfg_inllen2 < 1) {
 258                 CERROR("LMV setup requires an OST UUID list\n");
 259                 RETURN(-EINVAL);
 260         }
 261
 262         desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
 263         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
 264                 CERROR("descriptor size wrong: %d > %d\n",
 265                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
 266                 RETURN(-EINVAL);
 267         }
 268
 269         count = desc->ld_count;
 270         uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
 271         if (sizeof(*uuids) * count != lcfg->lcfg_inllen2) {
 272                 CERROR("UUID array size wrong: %u * %u != %u\n",
 273                        sizeof(*uuids), count, lcfg->lcfg_inllen2);
 274                 RETURN(-EINVAL);
 275         }
 276
 277         lmv->bufsize = sizeof(struct lmv_tgt_desc) * count;
 278         OBD_ALLOC(lmv->tgts, lmv->bufsize);
 279         if (lmv->tgts == NULL) {
 280                 CERROR("Out of memory\n");
 281                 RETURN(-EINVAL);
 282         }
 283
 284         for (i = 0, tgts = lmv->tgts; i < count; i++, tgts++) {
 285                 tgts->uuid = uuids[i];
 286                 lmv->count++;
 287         }
 288
 289         lmv->max_easize = sizeof(struct ll_fid) * lmv->count
 290                                         + sizeof(struct mea);
 291         lmv->max_cookiesize = 0;
 292
 293         RETURN(rc);
 294 }
 295
 296 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 297                       unsigned long max_age)
 298 {
 299         struct lmv_obd *lmv = &obd->u.lmv;
 300         struct obd_statfs temp;
 301         int rc = 0, i;
 302         ENTRY;
 303         lmv_connect(obd);
 304         for (i = 0; i < lmv->count; i++) {
 305                 rc = obd_statfs(lmv->tgts[i].exp->exp_obd, &temp, max_age);
 306                 if (rc) {
 307                         CERROR("can't stat MDS #%d (%s)\n", i,
 308                                lmv->tgts[i].exp->exp_obd->obd_name);
 309                         RETURN(rc);
 310                 }
 311                 if (i == 0) {
 312                         memcpy(osfs, &temp, sizeof(temp));
 313                 } else {
 314                         osfs->os_bavail += temp.os_bavail;
 315                         osfs->os_blocks += temp.os_blocks;
 316                         osfs->os_ffree += temp.os_ffree;
 317                         osfs->os_files += temp.os_files;
 318                 }
 319         }
 320         RETURN(rc);
 321 }
 322
 323 static int lmv_cleanup(struct obd_device *obd, int flags)
 324 {
 325         struct lmv_obd *lmv = &obd->u.lmv;
 326         ENTRY;
 327         lmv_cleanup_objs(obd);
 328         OBD_FREE(lmv->tgts, lmv->bufsize);
 329         RETURN(0);
 330 }
 331
 332 static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid)
 333 {
 334         struct obd_device *obd = exp->exp_obd;
 335         struct lmv_obd *lmv = &obd->u.lmv;
 336         int rc;
 337         ENTRY;
 338         lmv_connect(obd);
 339         rc = md_getstatus(lmv->tgts[0].exp, fid);
 340         fid->mds = 0;
 341         RETURN(rc);
 342 }
 343
 344 static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid,
 345                 unsigned long valid, unsigned int ea_size,
 346                 struct ptlrpc_request **request)
 347 {
 348         struct obd_device *obd = exp->exp_obd;
 349         struct lmv_obd *lmv = &obd->u.lmv;
 350         int rc, i = fid->mds;
 351         struct lmv_obj *obj;
 352         ENTRY;
 353         lmv_connect(obd);
 354         obj = lmv_grab_obj(obd, fid, 0);
 355         CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n",
 356                (unsigned long) fid->mds,
 357                (unsigned long) fid->id,
 358                (unsigned long) fid->generation,
 359                obj ? "(splitted)" : "");
 360
 361         LASSERT(fid->mds < lmv->count);
 362         rc = md_getattr(lmv->tgts[i].exp, fid,
 363                              valid, ea_size, request);
 364         if (rc == 0 && obj) {
 365                 /* we have to loop over dirobjs here and gather attrs
 366                  * for all the slaves */
 367 #warning "attrs gathering here"
 368         }
 369         lmv_put_obj(obj);
 370         RETURN(rc);
 371 }
 372
 373 static int lmv_change_cbdata(struct obd_export *exp,
 374                                  struct ll_fid *fid,
 375                                  ldlm_iterator_t it, void *data)
 376 {
 377         struct obd_device *obd = exp->exp_obd;
 378         struct lmv_obd *lmv = &obd->u.lmv;
 379         int rc = 0;
 380         ENTRY;
 381         lmv_connect(obd);
 382         CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n",
 383                (unsigned long) fid->mds,
 384                (unsigned long) fid->id,
 385                (unsigned long) fid->generation);
 386         LASSERT(fid->mds < lmv->count);
 387         rc = md_change_cbdata(lmv->tgts[fid->mds].exp, fid, it, data);
 388         RETURN(rc);
 389 }
 390
 391 static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
 392                                   char *name, int len, struct ll_fid *cfid,
 393                                   ldlm_iterator_t it, void *data)
 394 {
 395         struct obd_device *obd = exp->exp_obd;
 396         struct lmv_obd *lmv = &obd->u.lmv;
 397         struct lmv_obj *obj;
 398         int rc = 0, mds;
 399         ENTRY;
 400         lmv_connect(obd);
 401         LASSERT(pfid->mds < lmv->count);
 402         LASSERT(cfid->mds < lmv->count);
 403         CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n",
 404                (unsigned long) pfid->mds, (unsigned long) pfid->id,
 405                (unsigned long) pfid->generation, len, name,
 406                (unsigned long) cfid->mds, (unsigned long) cfid->id,
 407                (unsigned long) cfid->generation);
 408
 409         /* this is default mds for directory name belongs to */
 410         mds = pfid->mds;
 411         obj = lmv_grab_obj(obd, pfid, 0);
 412         if (obj) {
 413                 /* directory is splitted. look for right mds for this name */
 414                 mds = raw_name2idx(obj->objcount, name, len);
 415                 lmv_put_obj(obj);
 416         }
 417         rc = md_change_cbdata(lmv->tgts[mds].exp, cfid, it, data);
 418         RETURN(rc);
 419 }
 420
 421 static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid)
 422 {
 423         struct obd_device *obd = exp->exp_obd;
 424         struct lmv_obd *lmv = &obd->u.lmv;
 425         int rc = 0;
 426         ENTRY;
 427         lmv_connect(obd);
 428         CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n",
 429                (unsigned long) fid->mds,
 430                (unsigned long) fid->id,
 431                (unsigned long) fid->generation);
 432         LASSERT(fid->mds < lmv->count);
 433         rc = md_valid_attrs(lmv->tgts[fid->mds].exp, fid);
 434         RETURN(rc);
 435 }
 436
 437 int lmv_close(struct obd_export *exp, struct obdo *obdo,
 438                   struct obd_client_handle *och,
 439                   struct ptlrpc_request **request)
 440 {
 441         struct obd_device *obd = exp->exp_obd;
 442         struct lmv_obd *lmv = &obd->u.lmv;
 443         int rc, i = obdo->o_mds;
 444         ENTRY;
 445         lmv_connect(obd);
 446         LASSERT(i < lmv->count);
 447         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds,
 448                (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation);
 449         rc = md_close(lmv->tgts[i].exp, obdo, och, request);
 450         RETURN(rc);
 451 }
 452
 453 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 454                    const void *data, int datalen, int mode, __u32 uid,
 455                    __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 456 {
 457         struct obd_device *obd = exp->exp_obd;
 458         struct lmv_obd *lmv = &obd->u.lmv;
 459         struct mea *mea = op_data->mea1;
 460         struct mds_body *mds_body;
 461         int rc, i, mds, free_mea = 0;
 462         struct lmv_obj *obj;
 463         ENTRY;
 464         lmv_connect(obd);
 465         /* TODO: where to create new directories?
 466          * current design don't support directory on a slave MDS,
 467          * but we lookup by name may forward any request in slave
 468          */
 469 repeat:
 470         obj = lmv_grab_obj(obd, &op_data->fid1, 0);
 471         if (obj) {
 472                 mds = raw_name2idx(obj->objcount, op_data->name,
 473                                         op_data->namelen - 1);
 474                 op_data->fid1 = obj->objs[mds].fid;
 475                 lmv_put_obj(obj);
 476         }
 477
 478         CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu (mea 0x%p)\n",
 479                         op_data->namelen, op_data->name,
 480                         (unsigned long) op_data->fid1.mds,
 481                         (unsigned long) op_data->fid1.id,
 482                         (unsigned long) op_data->fid1.generation, mea);
 483         rc = md_create(lmv->tgts[op_data->fid1.mds].exp, op_data, data,
 484                        datalen, mode, uid, gid, rdev, request);
 485         if (rc == 0) {
 486                 if (*request == NULL)
 487                      RETURN(rc);
 488                 mds_body = lustre_msg_buf((*request)->rq_repmsg, 0,
 489                                           sizeof(*mds_body));
 490                 LASSERT(mds_body != NULL);
 491                 CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, mds = %d\n",
 492                        (unsigned long) mds_body->fid1.id,
 493                        (unsigned long) mds_body->fid1.generation,
 494                        op_data->fid1.mds);
 495                 LASSERT(mds_body->valid & OBD_MD_MDS ||
 496                                 mds_body->mds == op_data->fid1.mds);
 497         } else if (rc == -ESTALE) {
 498                 struct ptlrpc_request *req = NULL;
 499                 struct lustre_md md;
 500                 int mealen;
 501
 502                 LBUG(); /* FIXME ASAP */
 503                 CDEBUG(D_OTHER, "it seems MDS splitted dir\n");
 504                 LASSERT(mea == NULL);
 505
 506                 mealen = sizeof(struct ll_fid)*lmv->count + sizeof(struct mea);
 507                 /* time to update mea of parent fid */
 508                 i = op_data->fid1.mds;
 509                 rc = md_getattr(lmv->tgts[i].exp, &op_data->fid1,
 510                                         OBD_MD_FLEASIZE, mealen, &req);
 511                 LASSERT(rc == 0);
 512                 md.mea = NULL;
 513                 rc = mdc_req2lustre_md(req, 0, NULL, exp, &md);
 514                 LASSERT(rc == 0);
 515                 LASSERT(md.mea != NULL);
 516                 mea = md.mea;
 517                 ptlrpc_req_finished(req);
 518                 free_mea = 1;
 519
 520                 goto repeat;
 521         }
 522         if (free_mea)
 523                 obd_free_memmd(exp, (struct lov_stripe_md**) &mea);
 524         RETURN(rc);
 525 }
 526
 527 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 528 {
 529         struct obd_device *obd = exp->exp_obd;
 530         struct lmv_obd *lmv = &obd->u.lmv;
 531         int rc;
 532         ENTRY;
 533         lmv_connect(obd);
 534         /* FIXME: choose right MDC here */
 535         rc = md_done_writing(lmv->tgts[0].exp, obdo);
 536         RETURN(rc);
 537 }
 538
 539 int lmv_enqueue(struct obd_export *exp, int lock_type,
 540                     struct lookup_intent *it, int lock_mode,
 541                     struct mdc_op_data *data, struct lustre_handle *lockh,
 542                     void *lmm, int lmmsize,
 543                     ldlm_completion_callback cb_completion,
 544                     ldlm_blocking_callback cb_blocking, void *cb_data)
 545 {
 546         struct obd_device *obd = exp->exp_obd;
 547         struct lmv_obd *lmv = &obd->u.lmv;
 548         struct lmv_obj *obj;
 549         int rc, mds;
 550         ENTRY;
 551         lmv_connect(obd);
 552         if (data->namelen) {
 553                 obj = lmv_grab_obj(obd, &data->fid1, 0);
 554                 if (obj) {
 555                         /* directory is splitted. look for
 556                          * right mds for this name */
 557                         mds = raw_name2idx(obj->objcount, data->name,
 558                                                 data->namelen);
 559                         data->fid1 = obj->objs[mds].fid;
 560                         lmv_put_obj(obj);
 561                 }
 562         }
 563         CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n",
 564                LL_IT2STR(it), (unsigned long) data->fid1.id,
 565                (unsigned long) data->fid1.generation);
 566         rc = md_enqueue(lmv->tgts[data->fid1.mds].exp, lock_type, it,
 567                         lock_mode, data, lockh, lmm, lmmsize, cb_completion,
 568                         cb_blocking, cb_data);
 569
 570         RETURN(rc);
 571 }
 572
 573 int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid,
 574                          char *filename, int namelen, unsigned long valid,
 575                          unsigned int ea_size, struct ptlrpc_request **request)
 576 {
 577         struct obd_device *obd = exp->exp_obd;
 578         struct lmv_obd *lmv = &obd->u.lmv;
 579         struct ll_fid rfid = *fid;
 580         int rc, mds = fid->mds;
 581         struct lmv_obj *obj;
 582         ENTRY;
 583         lmv_connect(obd);
 584         CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu\n",
 585                namelen - 1, filename, (unsigned long) fid->mds,
 586                (unsigned long) fid->id, (unsigned long) fid->generation);
 587         obj = lmv_grab_obj(obd, fid, 0);
 588         if (obj) {
 589                 /* directory is splitted. look for right mds for this name */
 590                 mds = raw_name2idx(obj->objcount, filename, namelen - 1);
 591                 rfid = obj->objs[mds].fid;
 592                 lmv_put_obj(obj);
 593         }
 594         rc = md_getattr_name(lmv->tgts[mds].exp, &rfid, filename, namelen,
 595                                   valid, ea_size, request);
 596         RETURN(rc);
 597 }
 598
 599
 600 /*
 601  * llite passes fid of an target inode in data->fid1 and
 602  * fid of directory in data->fid2
 603  */
 604 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
 605              struct ptlrpc_request **request)
 606 {
 607         struct obd_device *obd = exp->exp_obd;
 608         struct lmv_obd *lmv = &obd->u.lmv;
 609         struct lmv_obj *obj;
 610         int rc;
 611         ENTRY;
 612         lmv_connect(obd);
 613         if (data->namelen != 0) {
 614                 /* usual link request */
 615                 obj = lmv_grab_obj(obd, &data->fid1, 0);
 616                 if (obj) {
 617                         rc = raw_name2idx(obj->objcount, data->name,
 618                                          data->namelen);
 619                         data->fid1 = obj->objs[rc].fid;
 620                         lmv_put_obj(obj);
 621                 }
 622                 CDEBUG(D_OTHER,"link %u/%u/%u:%*s to %u/%u/%u mds %d\n",
 623                        (unsigned) data->fid2.mds, (unsigned) data->fid2.id,
 624                        (unsigned) data->fid2.generation, data->namelen,
 625                        data->name, (unsigned) data->fid1.mds,
 626                        (unsigned) data->fid1.id,
 627                        (unsigned) data->fid1.generation, data->fid1.mds);
 628         } else {
 629                 /* request from MDS to acquire i_links for inode by fid1 */
 630                 CDEBUG(D_OTHER, "inc i_nlinks for %u/%u/%u\n",
 631                        (unsigned) data->fid1.mds, (unsigned) data->fid1.id,
 632                        (unsigned) data->fid1.generation);
 633         }
 634
 635         rc = md_link(lmv->tgts[data->fid1.mds].exp, data, request);
 636         RETURN(rc);
 637 }
 638
 639 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
 640                const char *old, int oldlen, const char *new, int newlen,
 641                struct ptlrpc_request **request)
 642 {
 643         struct obd_device *obd = exp->exp_obd;
 644         struct lmv_obd *lmv = &obd->u.lmv;
 645         struct lmv_obj *obj;
 646         int rc, mds;
 647         ENTRY;
 648
 649         CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n",
 650                oldlen, old, (unsigned long) data->fid1.mds,
 651                (unsigned long) data->fid1.id,
 652                (unsigned long) data->fid1.generation,
 653                newlen, new, (unsigned long) data->fid2.mds,
 654                (unsigned long) data->fid2.id,
 655                (unsigned long) data->fid2.generation);
 656
 657         lmv_connect(obd);
 658
 659         if (oldlen == 0) {
 660                 /* MDS with old dir entry is asking another MDS
 661                  * to create name there */
 662                 CDEBUG(D_OTHER,
 663                        "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n",
 664                        newlen, new, oldlen, newlen,
 665                        (unsigned long) data->fid2.mds,
 666                        (unsigned long) data->fid2.id,
 667                        (unsigned long) data->fid2.generation,
 668                        (unsigned long) data->fid1.mds,
 669                        (unsigned long) data->fid1.id,
 670                        (unsigned long) data->fid1.generation);
 671                 mds = data->fid2.mds;
 672                 goto request;
 673         }
 674
 675         obj = lmv_grab_obj(obd, &data->fid1, 0);
 676         if (obj) {
 677                 /* directory is already splitted, so we have to forward
 678                  * request to the right MDS */
 679                 mds = raw_name2idx(obj->objcount, old, oldlen);
 680                 data->fid1 = obj->objs[mds].fid;
 681                 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
 682                        (unsigned long) obj->objs[mds].fid.mds,
 683                        (unsigned long) obj->objs[mds].fid.id,
 684                        (unsigned long) obj->objs[mds].fid.generation);
 685         }
 686         lmv_put_obj(obj);
 687
 688         obj = lmv_grab_obj(obd, &data->fid2, 0);
 689         if (obj) {
 690                 /* directory is already splitted, so we have to forward
 691                  * request to the right MDS */
 692                 mds = raw_name2idx(obj->objcount, new, newlen);
 693                 data->fid2 = obj->objs[mds].fid;
 694                 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
 695                        (unsigned long) obj->objs[mds].fid.mds,
 696                        (unsigned long) obj->objs[mds].fid.id,
 697                        (unsigned long) obj->objs[mds].fid.generation);
 698         }
 699         lmv_put_obj(obj);
 700
 701         mds = data->fid1.mds;
 702
 703 request:
 704         rc = md_rename(lmv->tgts[mds].exp, data, old, oldlen,
 705                             new, newlen, request);
 706         RETURN(rc);
 707 }
 708
 709 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
 710                 struct iattr *iattr, void *ea, int ealen, void *ea2, int ea2len,
 711                 struct ptlrpc_request **request)
 712 {
 713         struct obd_device *obd = exp->exp_obd;
 714         struct lmv_obd *lmv = &obd->u.lmv;
 715         int rc = 0, i = data->fid1.mds;
 716         struct ptlrpc_request *req;
 717         struct mds_body *mds_body;
 718         struct lmv_obj *obj;
 719         ENTRY;
 720         lmv_connect(obd);
 721         obj = lmv_grab_obj(obd, &data->fid1, 0);
 722         CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n",
 723                (unsigned long) data->fid1.mds,
 724                (unsigned long) data->fid1.id,
 725                (unsigned long) data->fid1.generation, iattr->ia_valid,
 726                obj ? ", splitted" : "");
 727         if (obj) {
 728                 for (i = 0; i < obj->objcount; i++) {
 729                         data->fid1 = obj->objs[i].fid;
 730                         rc = md_setattr(lmv->tgts[i].exp, data, iattr, ea,
 731                                         ealen, ea2, ea2len, &req);
 732                         LASSERT(rc == 0);
 733                         if (fid_equal(&obj->fid, &obj->objs[i].fid)) {
 734                                 /* this is master object and this request
 735                                  * should be returned back to llite */
 736                                 *request = req;
 737                         } else {
 738                                 ptlrpc_req_finished(req);
 739                         }
 740                 }
 741                 lmv_put_obj(obj);
 742         } else {
 743                 LASSERT(data->fid1.mds < lmv->count);
 744                 rc = md_setattr(lmv->tgts[i].exp, data, iattr, ea, ealen,
 745                                 ea2, ea2len, request);
 746                 if (rc == 0) {
 747                         mds_body = lustre_msg_buf((*request)->rq_repmsg, 0,
 748                                         sizeof(*mds_body));
 749                         LASSERT(mds_body != NULL);
 750                         LASSERT(mds_body->mds == i);
 751                 }
 752         }
 753         RETURN(rc);
 754 }
 755
 756 int lmv_sync(struct obd_export *exp, struct ll_fid *fid,
 757              struct ptlrpc_request **request)
 758 {
 759         struct obd_device *obd = exp->exp_obd;
 760         struct lmv_obd *lmv = &obd->u.lmv;
 761         int rc;
 762         ENTRY;
 763         lmv_connect(obd);
 764         rc = md_sync(lmv->tgts[0].exp, fid, request);
 765         RETURN(rc);
 766 }
 767
 768 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
 769                             struct ldlm_lock_desc *desc, void *data, int flag)
 770 {
 771         struct lustre_handle lockh;
 772         struct lmv_obj *obj;
 773         int rc;
 774         ENTRY;
 775
 776         switch (flag) {
 777         case LDLM_CB_BLOCKING:
 778                 ldlm_lock2handle(lock, &lockh);
 779                 rc = ldlm_cli_cancel(&lockh);
 780                 if (rc < 0) {
 781                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
 782                         RETURN(rc);
 783                 }
 784                 break;
 785         case LDLM_CB_CANCELING:
 786                 /* time to drop cached attrs for dirobj */
 787                 obj = lock->l_ast_data;
 788                 if (!obj)
 789                         break;
 790
 791                 CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n",
 792                        lock->l_resource->lr_name.name[3] == 1 ?
 793                                 "LOOKUP" : "UPDATE",
 794                        (unsigned long) lock->l_resource->lr_name.name[0],
 795                        (unsigned long) lock->l_resource->lr_name.name[1],
 796                        (unsigned long) obj->fid.mds,
 797                        (unsigned long) obj->fid.id,
 798                        (unsigned long) obj->fid.generation);
 799                 break;
 800         default:
 801                 LBUG();
 802         }
 803         RETURN(0);
 804 }
 805
 806 void lmv_remove_dots(struct page *page)
 807 {
 808         char *kaddr = page_address(page);
 809         unsigned limit = PAGE_CACHE_SIZE;
 810         unsigned offs, rec_len;
 811         struct ext2_dir_entry_2 *p;
 812
 813         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
 814                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
 815                 rec_len = le16_to_cpu(p->rec_len);
 816
 817                 if ((p->name_len == 1 && p->name[0] == '.') ||
 818                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
 819                         p->inode = 0;
 820         }
 821 }
 822
 823 int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
 824                  __u64 offset, struct page *page,
 825                  struct ptlrpc_request **request)
 826 {
 827         struct obd_device *obd = exp->exp_obd;
 828         struct lmv_obd *lmv = &obd->u.lmv;
 829         struct ll_fid rfid = *mdc_fid;
 830         struct lmv_obj *obj;
 831         int rc, i;
 832         ENTRY;
 833         lmv_connect(obd);
 834
 835         LASSERT(mdc_fid->mds < lmv->count);
 836         CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n",
 837                offset, (unsigned long) rfid.mds,
 838                (unsigned long) rfid.id,
 839                (unsigned long) rfid.generation);
 840
 841         obj = lmv_grab_obj(obd, mdc_fid, 0);
 842         if (obj) {
 843                 /* find dirobj containing page with requested offset */
 844                 /* FIXME: what about protecting cached attrs here? */
 845                 for (i = 0; i < obj->objcount; i++) {
 846                         if (offset < obj->objs[i].size)
 847                                 break;
 848                         offset -= obj->objs[i].size;
 849                 }
 850                 rfid = obj->objs[i].fid;
 851                 CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n",
 852                        (unsigned long) rfid.mds,
 853                        (unsigned long) rfid.id,
 854                        (unsigned long) rfid.generation,
 855                        (unsigned long) offset);
 856         }
 857         rc = md_readpage(lmv->tgts[rfid.mds].exp, &rfid, offset, page, request);
 858         if (rc == 0 && !fid_equal(&rfid, mdc_fid)) {
 859                 /* this page isn't from master object. to avoid
 860                  * ./.. duplication in directory, we have to remove them
 861                  * from all slave objects */
 862                 lmv_remove_dots(page);
 863         }
 864
 865         lmv_put_obj(obj);
 866
 867         RETURN(rc);
 868 }
 869
 870 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
 871                struct ptlrpc_request **request)
 872 {
 873         struct obd_device *obd = exp->exp_obd;
 874         struct lmv_obd *lmv = &obd->u.lmv;
 875         int rc, i = 0;
 876         ENTRY;
 877         lmv_connect(obd);
 878         if (data->namelen != 0) {
 879                 struct lmv_obj *obj;
 880                 obj = lmv_grab_obj(obd, &data->fid1, 0);
 881                 if (obj) {
 882                         i = raw_name2idx(obj->objcount, data->name,
 883                                          data->namelen);
 884                         data->fid1 = obj->objs[i].fid;
 885                         lmv_put_obj(obj);
 886                 }
 887                 CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n",
 888                        data->namelen, data->name,
 889                        (unsigned long) data->fid1.mds,
 890                        (unsigned long) data->fid1.id,
 891                        (unsigned long) data->fid1.generation, i);
 892         } else {
 893                 CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n",
 894                        (unsigned long) data->fid1.mds,
 895                        (unsigned long) data->fid1.id,
 896                        (unsigned long) data->fid1.generation);
 897         }
 898         rc = md_unlink(lmv->tgts[data->fid1.mds].exp, data, request);
 899         RETURN(rc);
 900 }
 901
 902 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
 903                                         char *name, int len)
 904 {
 905         struct obd_device *obd = exp->exp_obd;
 906         struct lmv_obd *lmv = &obd->u.lmv;
 907         ENTRY;
 908         lmv_connect(obd);
 909         obd = lmv->tgts[0].exp->exp_obd;
 910         EXIT;
 911         return obd;
 912 }
 913
 914 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
 915 {
 916         struct obd_device *obd = exp->exp_obd;
 917         struct lmv_obd *lmv = &obd->u.lmv;
 918         int i, rc = 0, change = 0;
 919         ENTRY;
 920
 921         if (lmv->max_easize < easize) {
 922                 lmv->max_easize = easize;
 923                 change = 1;
 924         }
 925         if (lmv->max_cookiesize < cookiesize) {
 926                 lmv->max_cookiesize = cookiesize;
 927                 change = 1;
 928         }
 929         if (change == 0)
 930                 RETURN(0);
 931
 932         if (lmv->connected == 0)
 933                 RETURN(0);
 934
 935         /* FIXME: error handling? */
 936         for (i = 0; i < lmv->count; i++)
 937                 rc = obd_init_ea_size(lmv->tgts[i].exp, easize, cookiesize);
 938         RETURN(rc);
 939 }
 940
 941 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
 942                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
 943 {
 944         struct obd_device *obd = exp->exp_obd;
 945         struct lmv_obd *lmv = &obd->u.lmv;
 946         struct lov_stripe_md obj_md;
 947         struct lov_stripe_md *obj_mdp = &obj_md;
 948         int rc = 0;
 949         ENTRY;
 950         lmv_connect(obd);
 951
 952         LASSERT(ea == NULL);
 953         LASSERT(oa->o_mds < lmv->count);
 954
 955         rc = obd_create(lmv->tgts[oa->o_mds].exp, oa, &obj_mdp, oti);
 956         LASSERT(rc == 0);
 957
 958         RETURN(rc);
 959 }
 960
 961 /*
 962  * to be called from MDS only
 963  */
 964 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
 965                struct lov_stripe_md **ea, struct obd_trans_info *oti)
 966 {
 967         struct obd_device *obd = exp->exp_obd;
 968         struct lmv_obd *lmv = &obd->u.lmv;
 969         struct mea *mea;
 970         int i, c, rc = 0;
 971         struct ll_fid mfid;
 972         ENTRY;
 973         lmv_connect(obd);
 974
 975         LASSERT(oa != NULL);
 976
 977         if (ea == NULL) {
 978                 rc = lmv_obd_create_single(exp, oa, NULL, oti);
 979                 RETURN(rc);
 980         }
 981
 982         if (*ea == NULL) {
 983                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **) ea);
 984                 LASSERT(*ea != NULL);
 985         }
 986
 987         mea = (struct mea *) *ea;
 988         mfid.id = oa->o_id;
 989         mfid.generation = oa->o_generation;
 990         rc = 0;
 991         if (!mea->mea_count || mea->mea_count > lmv->count)
 992                 mea->mea_count = lmv->count;
 993
 994         mea->mea_master = -1;
 995
 996         /* FIXME: error handling? */
 997         for (i = 0, c = 0; c < mea->mea_count && i < lmv->count; i++) {
 998                 struct lov_stripe_md obj_md;
 999                 struct lov_stripe_md *obj_mdp = &obj_md;
1000
1001                 if (lmv->tgts[i].exp == NULL) {
1002                         /* this is master MDS */
1003                         mea->mea_fids[c].id = mfid.id;
1004                         mea->mea_fids[c].generation = mfid.generation;
1005                         mea->mea_fids[c].mds = i;
1006                         mea->mea_master = i;
1007                         c++;
1008                         continue;
1009                 }
1010
1011                 /* "Master" MDS should always be part of stripped dir, so
1012                    scan for it */
1013                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1014                         continue;
1015
1016                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE
1017                                 | OBD_MD_FLUID | OBD_MD_FLGID;
1018
1019                 rc = obd_create(lmv->tgts[c].exp, oa, &obj_mdp, oti);
1020                 /* FIXME: error handling here */
1021                 LASSERT(rc == 0);
1022
1023                 mea->mea_fids[c].id = oa->o_id;
1024                 mea->mea_fids[c].generation = oa->o_generation;
1025                 mea->mea_fids[c].mds = i;
1026                 c++;
1027                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1028                        i, oa->o_id, oa->o_generation);
1029         }
1030         LASSERT(c == mea->mea_count);
1031         CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1032
1033         RETURN(rc);
1034 }
1035
1036 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1037                            void *key, __u32 *vallen, void *val)
1038 {
1039         struct obd_device *obd;
1040         struct lmv_obd *lmv;
1041         ENTRY;
1042
1043         obd = class_exp2obd(exp);
1044         if (obd == NULL) {
1045                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1046                        exp->exp_handle.h_cookie);
1047                 RETURN(-EINVAL);
1048         }
1049
1050         lmv = &obd->u.lmv;
1051         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1052                 __u32 *mdsize = val;
1053                 *vallen = sizeof(__u32);
1054                 *mdsize = sizeof(struct ll_fid) * lmv->count
1055                                 + sizeof(struct mea);
1056                 RETURN(0);
1057         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1058                 struct obd_uuid *cluuid = &lmv->cluuid;
1059                 struct lmv_tgt_desc *tgts;
1060                 __u32 *mdsnum = val;
1061                 int i;
1062
1063                 for (i = 0, tgts = lmv->tgts; i < lmv->count; i++, tgts++) {
1064                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1065                                 *vallen = sizeof(__u32);
1066                                 *mdsnum = i;
1067                                 RETURN(0);
1068                         }
1069                 }
1070                 LASSERT(0);
1071         }
1072
1073         CDEBUG(D_IOCTL, "invalid key\n");
1074         RETURN(-EINVAL);
1075 }
1076
1077 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1078                  void *key, obd_count vallen, void *val)
1079 {
1080         struct obd_device *obd;
1081         struct lmv_obd *lmv;
1082         ENTRY;
1083
1084         obd = class_exp2obd(exp);
1085         if (obd == NULL) {
1086                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1087                        exp->exp_handle.h_cookie);
1088                 RETURN(-EINVAL);
1089         }
1090         lmv = &obd->u.lmv;
1091         lmv_connect(obd);
1092
1093         if (keylen >= strlen("client") && strcmp(key, "client") == 0) {
1094                 struct lmv_tgt_desc *tgts;
1095                 int i, rc;
1096
1097                 for (i = 0, tgts = lmv->tgts; i < lmv->count; i++, tgts++) {
1098                         rc = obd_set_info(tgts->exp, keylen, key, vallen, val);
1099                         if (rc)
1100                                 RETURN(rc);
1101                 }
1102                 RETURN(0);
1103         }
1104
1105         RETURN(-EINVAL);
1106 }
1107
1108 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1109                struct lov_stripe_md *lsm)
1110 {
1111         struct obd_device *obd = class_exp2obd(exp);
1112         struct lmv_obd *lmv = &obd->u.lmv;
1113         int mea_size;
1114         ENTRY;
1115
1116         mea_size = sizeof(struct ll_fid) * lmv->count + sizeof(struct mea);
1117         if (!lmmp)
1118                 RETURN(mea_size);
1119
1120         if (*lmmp && !lsm) {
1121                 OBD_FREE(*lmmp, mea_size);
1122                 *lmmp = NULL;
1123                 RETURN(0);
1124         }
1125
1126         if (!*lmmp) {
1127                 OBD_ALLOC(*lmmp, mea_size);
1128                 if (!*lmmp)
1129                         RETURN(-ENOMEM);
1130         }
1131
1132         if (!lsm)
1133                 RETURN(mea_size);
1134
1135 #warning "MEA packing/convertation must be here! -bzzz"
1136         memcpy(*lmmp, lsm, mea_size);
1137         RETURN(mea_size);
1138 }
1139
1140 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1141                         struct lov_mds_md *disk_src, int mdsize)
1142 {
1143         struct obd_device *obd = class_exp2obd(exp);
1144         struct lmv_obd *lmv = &obd->u.lmv;
1145         struct mea **tmea = (struct mea **) mem_tgt;
1146         struct mea *mea = (void *) disk_src;
1147         int mea_size;
1148         ENTRY;
1149
1150         mea_size = sizeof(struct ll_fid) * lmv->count + sizeof(struct mea);
1151         if (mem_tgt == NULL)
1152                 return mea_size;
1153
1154         if (*mem_tgt != NULL && disk_src == NULL) {
1155                 OBD_FREE(*tmea, mea_size);
1156                 RETURN(0);
1157         }
1158
1159         LASSERT(mea_size == mdsize);
1160
1161         OBD_ALLOC(*tmea, mea_size);
1162         /* FIXME: error handling here */
1163         LASSERT(*tmea != NULL);
1164
1165         if (!disk_src)
1166                 RETURN(mea_size);
1167
1168 #warning "MEA unpacking/convertation must be here! -bzzz"
1169         memcpy(*tmea, mea, mdsize);
1170         RETURN(mea_size);
1171 }
1172
1173 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1174                 struct lov_stripe_md *ea, obd_count oa_bufs,
1175                 struct brw_page *pgarr, struct obd_trans_info *oti)
1176 {
1177         struct obd_device *obd = exp->exp_obd;
1178         struct lmv_obd *lmv = &obd->u.lmv;
1179         struct mea *mea = (struct mea *) ea;
1180         int err;
1181
1182         LASSERT(oa != NULL);
1183         LASSERT(ea != NULL);
1184         LASSERT(pgarr != NULL);
1185         LASSERT(oa->o_mds < lmv->count);
1186
1187         oa->o_gr = mea->mea_fids[oa->o_mds].generation;
1188         oa->o_id = mea->mea_fids[oa->o_mds].id;
1189         oa->o_valid =  OBD_MD_FLID | OBD_MD_FLGROUP;
1190         err = obd_brw(rw, lmv->tgts[oa->o_mds].exp, oa,
1191                         NULL, oa_bufs, pgarr, oti);
1192         RETURN(err);
1193 }
1194
1195 struct obd_ops lmv_obd_ops = {
1196         o_owner:                THIS_MODULE,
1197         o_attach:               lmv_attach,
1198         o_detach:               lmv_detach,
1199         o_setup:                lmv_setup,
1200         o_cleanup:              lmv_cleanup,
1201         o_connect:              lmv_connect_fake,
1202         o_disconnect:           lmv_disconnect,
1203         o_statfs:               lmv_statfs,
1204         o_get_info:             lmv_get_info,
1205         o_set_info:             lmv_set_info,
1206         o_create:               lmv_obd_create,
1207         o_packmd:               lmv_packmd,
1208         o_unpackmd:             lmv_unpackmd,
1209         o_brw:                  lmv_brw,
1210         o_init_ea_size:         lmv_init_ea_size,
1211 };
1212
1213 struct md_ops lmv_md_ops = {
1214         m_getstatus:            lmv_getstatus,
1215         m_getattr:              lmv_getattr,
1216         m_change_cbdata:        lmv_change_cbdata,
1217         m_change_cbdata_name:   lmv_change_cbdata_name,
1218         m_close:                lmv_close,
1219         m_create:               lmv_create,
1220         m_done_writing:         lmv_done_writing,
1221         m_enqueue:              lmv_enqueue,
1222         m_getattr_name:         lmv_getattr_name,
1223         m_intent_lock:          lmv_intent_lock,
1224         m_link:                 lmv_link,
1225         m_rename:               lmv_rename,
1226         m_setattr:              lmv_setattr,
1227         m_sync:                 lmv_sync,
1228         m_readpage:             lmv_readpage,
1229         m_unlink:               lmv_unlink,
1230         m_get_real_obd:         lmv_get_real_obd,
1231         m_valid_attrs:          lmv_valid_attrs,
1232 };
1233
1234 //#ifndef LPROCFS
1235 static struct lprocfs_vars lprocfs_module_vars[] = { {0} };
1236 static struct lprocfs_vars lprocfs_obd_vars[] = { {0} };
1237 //#else
1238 LPROCFS_INIT_VARS(lmv, lprocfs_module_vars, lprocfs_obd_vars)
1239
1240 int __init lmv_init(void)
1241 {
1242         struct lprocfs_static_vars lvars;
1243         int rc;
1244
1245         lprocfs_init_vars(lmv, &lvars);
1246         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1247                                  lvars.module_vars, OBD_LMV_DEVICENAME);
1248         RETURN(rc);
1249 }
1250
1251 static void lmv_exit(void)
1252 {
1253         class_unregister_type(OBD_LMV_DEVICENAME);
1254 }
1255
1256 #ifdef __KERNEL__
1257 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1258 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1259 MODULE_LICENSE("GPL");
1260
1261 module_init(lmv_init);
1262 module_exit(lmv_exit);
1263 #endif
1264