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 #include <linux/seq_file.h>
  34 #else
  35 #include <liblustre.h>
  36 #endif
  37 #include <linux/ext2_fs.h>
  38
  39 #include <linux/obd_support.h>
  40 #include <linux/lustre_lib.h>
  41 #include <linux/lustre_net.h>
  42 #include <linux/lustre_idl.h>
  43 #include <linux/lustre_dlm.h>
  44 #include <linux/lustre_mds.h>
  45 #include <linux/obd_class.h>
  46 #include <linux/obd_ost.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 /* object cache. */
  53 kmem_cache_t *obj_cache;
  54 atomic_t obj_cache_count = ATOMIC_INIT(0);
  55
  56 static void lmv_activate_target(struct lmv_obd *lmv,
  57                                 struct lmv_tgt_desc *tgt,
  58                                 int activate)
  59 {
  60         if (tgt->active == activate)
  61                 return;
  62
  63         tgt->active = activate;
  64         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  65 }
  66
  67 /* Error codes:
  68  *
  69  *  -EINVAL  : UUID can't be found in the LMV's target list
  70  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  71  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  72  */
  73 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  74                               int activate)
  75 {
  76         struct lmv_tgt_desc *tgt;
  77         struct obd_device *obd;
  78         int i, rc = 0;
  79         ENTRY;
  80
  81         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  82                lmv, uuid->uuid, activate);
  83
  84         spin_lock(&lmv->lmv_lock);
  85         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  86                 if (tgt->ltd_exp == NULL)
  87                         continue;
  88
  89                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  90                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  91
  92                 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
  93                         break;
  94         }
  95
  96         if (i == lmv->desc.ld_tgt_count)
  97                 GOTO(out_lmv_lock, rc = -EINVAL);
  98
  99         obd = class_exp2obd(tgt->ltd_exp);
 100         if (obd == NULL)
 101                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 102
 103         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 104                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 105                obd->obd_type->typ_name, i);
 106         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 107
 108         if (tgt->active == activate) {
 109                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 110                        activate ? "" : "in");
 111                 GOTO(out_lmv_lock, rc);
 112         }
 113
 114         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 115                obd, activate ? "" : "in");
 116
 117         lmv_activate_target(lmv, tgt, activate);
 118
 119         EXIT;
 120
 121  out_lmv_lock:
 122         spin_unlock(&lmv->lmv_lock);
 123         return rc;
 124 }
 125
 126 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 127                       int active, void *data)
 128 {
 129         struct obd_uuid *uuid;
 130         int rc;
 131         ENTRY;
 132
 133         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 134                 CERROR("unexpected notification of %s %s!\n",
 135                        watched->obd_type->typ_name,
 136                        watched->obd_name);
 137                 RETURN(-EINVAL);
 138         }
 139         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 140
 141         /* Set MDC as active before notifying the observer, so the observer can
 142          * use the MDC normally.
 143          */
 144         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
 145         if (rc) {
 146                 CERROR("%sactivation of %s failed: %d\n",
 147                        active ? "" : "de", uuid->uuid, rc);
 148                 RETURN(rc);
 149         }
 150
 151         if (obd->obd_observer)
 152                 /* Pass the notification up the chain. */
 153                 rc = obd_notify(obd->obd_observer, watched, active, data);
 154
 155         RETURN(rc);
 156 }
 157
 158 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
 159 {
 160         struct lprocfs_static_vars lvars;
 161         int rc;
 162         ENTRY;
 163
 164         lprocfs_init_vars(lmv, &lvars);
 165         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 166 #ifdef __KERNEL__
 167         if (rc == 0) {
 168                 struct proc_dir_entry *entry;
 169
 170                 entry = create_proc_entry("target_obd_status", 0444,
 171                                            dev->obd_proc_entry);
 172                 if (entry == NULL)
 173                         RETURN(-ENOMEM);
 174                 entry->proc_fops = &lmv_proc_target_fops;
 175                 entry->data = dev;
 176        }
 177 #endif
 178         RETURN (rc);
 179 }
 180
 181 int lmv_detach(struct obd_device *dev)
 182 {
 183         return lprocfs_obd_detach(dev);
 184 }
 185
 186 /* this is fake connect function. Its purpose is to initialize lmv and say
 187  * caller that everything is okay. Real connection will be performed later. */
 188 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
 189                        struct obd_uuid *cluuid, unsigned long flags)
 190 {
 191 #ifdef __KERNEL__
 192         struct proc_dir_entry *lmv_proc_dir;
 193 #endif
 194         struct lmv_obd *lmv = &obd->u.lmv;
 195         struct obd_export *exp;
 196         int rc = 0;
 197         ENTRY;
 198
 199         rc = class_connect(conn, obd, cluuid);
 200         if (rc) {
 201                 CERROR("class_connection() returned %d\n", rc);
 202                 RETURN(rc);
 203         }
 204
 205         exp = class_conn2export(conn);
 206
 207         /* we don't want to actually do the underlying connections more than
 208          * once, so keep track. */
 209         lmv->refcount++;
 210         if (lmv->refcount > 1) {
 211                 class_export_put(exp);
 212                 RETURN(0);
 213         }
 214
 215         lmv->exp = exp;
 216         lmv->connected = 0;
 217         lmv->cluuid = *cluuid;
 218         lmv->connect_flags = flags;
 219         sema_init(&lmv->init_sem, 1);
 220
 221 #ifdef __KERNEL__
 222         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 223                                         NULL, NULL);
 224         if (IS_ERR(lmv_proc_dir)) {
 225                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 226                        obd->obd_type->typ_name, obd->obd_name);
 227                 lmv_proc_dir = NULL;
 228         }
 229 #endif
 230
 231         /*
 232          * all real clients shouls perform actual connection rightaway, because
 233          * it is possible, that LMV will not have opportunity to connect
 234          * targets, as MDC stuff will bit called directly, for instance while
 235          * reading ../mdc/../kbytesfree procfs file, etc.
 236          */
 237         if (flags & OBD_OPT_REAL_CLIENT)
 238                 rc = lmv_check_connect(obd);
 239
 240 #ifdef __KERNEL__
 241         if (rc) {
 242                 if (lmv_proc_dir)
 243                         lprocfs_remove(lmv_proc_dir);
 244         }
 245 #endif
 246
 247         RETURN(rc);
 248 }
 249
 250 void lmv_set_timeouts(struct obd_device *obd)
 251 {
 252         struct lmv_tgt_desc *tgts;
 253         struct lmv_obd *lmv;
 254         int i;
 255
 256         lmv = &obd->u.lmv;
 257         if (lmv->server_timeout == 0)
 258                 return;
 259
 260         if (lmv->connected == 0)
 261                 return;
 262
 263         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 264                 if (tgts->ltd_exp == NULL)
 265                         continue;
 266
 267                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 268                              "inter_mds", 0, NULL);
 269         }
 270 }
 271
 272 #define MAX_STRING_SIZE 128
 273
 274 /* performs a check if passed obd is connected. If no - connect it. */
 275 int lmv_check_connect(struct obd_device *obd)
 276 {
 277 #ifdef __KERNEL__
 278         struct proc_dir_entry *lmv_proc_dir;
 279 #endif
 280         struct lmv_obd *lmv = &obd->u.lmv;
 281         struct lmv_tgt_desc *tgts;
 282         struct obd_uuid *cluuid;
 283         struct obd_export *exp;
 284         int rc, rc2, i;
 285         ENTRY;
 286
 287         if (lmv->connected)
 288                 RETURN(0);
 289
 290         down(&lmv->init_sem);
 291         if (lmv->connected) {
 292                 up(&lmv->init_sem);
 293                 RETURN(0);
 294         }
 295
 296         cluuid = &lmv->cluuid;
 297         exp = lmv->exp;
 298
 299         CDEBUG(D_OTHER, "time to connect %s to %s\n",
 300                cluuid->uuid, obd->obd_name);
 301
 302         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 303                 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 304                 struct lustre_handle conn = {0, };
 305                 struct obd_device *tgt_obd;
 306
 307                 LASSERT(tgts != NULL);
 308
 309                 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
 310                                                 &obd->obd_uuid);
 311                 if (!tgt_obd) {
 312                         CERROR("target %s not attached\n", tgts->uuid.uuid);
 313                         GOTO(out_disc, rc = -EINVAL);
 314                 }
 315
 316                 /* for MDS: don't connect to yourself */
 317                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
 318                         CDEBUG(D_OTHER, "don't connect back to %s\n",
 319                                cluuid->uuid);
 320                         tgts->ltd_exp = NULL;
 321                         continue;
 322                 }
 323
 324                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
 325                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 326                         tgts->uuid.uuid, obd->obd_uuid.uuid,
 327                         cluuid->uuid);
 328
 329                 if (!tgt_obd->obd_set_up) {
 330                         CERROR("target %s not set up\n", tgts->uuid.uuid);
 331                         GOTO(out_disc, rc = -EINVAL);
 332                 }
 333
 334                 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
 335                                  lmv->connect_flags);
 336                 if (rc) {
 337                         CERROR("target %s connect error %d\n",
 338                                 tgts->uuid.uuid, rc);
 339                         GOTO(out_disc, rc);
 340                 }
 341                 tgts->ltd_exp = class_conn2export(&conn);
 342
 343                 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
 344                                  lmv->max_cookiesize);
 345
 346                 rc = obd_register_observer(tgt_obd, obd);
 347                 if (rc) {
 348                         CERROR("target %s register_observer error %d\n",
 349                                tgts->uuid.uuid, rc);
 350                         obd_disconnect(tgts->ltd_exp, 0);
 351                         GOTO(out_disc, rc);
 352                 }
 353
 354                 lmv->desc.ld_active_tgt_count++;
 355                 tgts->active = 1;
 356
 357                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
 358                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 359                         atomic_read(&obd->obd_refcount));
 360
 361 #ifdef __KERNEL__
 362                 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 363                 if (lmv_proc_dir) {
 364                         struct obd_device *mdc_obd = class_conn2obd(&conn);
 365                         struct proc_dir_entry *mdc_symlink;
 366                         char name[MAX_STRING_SIZE + 1];
 367
 368                         LASSERT(mdc_obd != NULL);
 369                         LASSERT(mdc_obd->obd_type != NULL);
 370                         LASSERT(mdc_obd->obd_type->typ_name != NULL);
 371                         name[MAX_STRING_SIZE] = '\0';
 372                         snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 373                                  mdc_obd->obd_type->typ_name,
 374                                  mdc_obd->obd_name);
 375                         mdc_symlink = proc_symlink(mdc_obd->obd_name,
 376                                                    lmv_proc_dir, name);
 377                         if (mdc_symlink == NULL) {
 378                                 CERROR("could not register LMV target "
 379                                        "/proc/fs/lustre/%s/%s/target_obds/%s.",
 380                                        obd->obd_type->typ_name, obd->obd_name,
 381                                        mdc_obd->obd_name);
 382                                 lprocfs_remove(lmv_proc_dir);
 383                                 lmv_proc_dir = NULL;
 384                         }
 385                 }
 386 #endif
 387         }
 388
 389         lmv_set_timeouts(obd);
 390         class_export_put(exp);
 391         lmv->connected = 1;
 392         up(&lmv->init_sem);
 393         RETURN(0);
 394
 395  out_disc:
 396         while (i-- > 0) {
 397                 struct obd_uuid uuid;
 398                 --tgts;
 399                 --lmv->desc.ld_active_tgt_count;
 400                 tgts->active = 0;
 401                 /* save for CERROR below; (we know it's terminated) */
 402                 uuid = tgts->uuid;
 403                 rc2 = obd_disconnect(tgts->ltd_exp, 0);
 404                 if (rc2)
 405                         CERROR("error: LMV target %s disconnect on MDC idx %d: "
 406                                "error %d\n", uuid.uuid, i, rc2);
 407         }
 408         class_disconnect(exp, 0);
 409         up(&lmv->init_sem);
 410         return rc;
 411 }
 412
 413 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 414 {
 415         struct obd_device *obd = class_exp2obd(exp);
 416         struct lmv_obd *lmv = &obd->u.lmv;
 417
 418 #ifdef __KERNEL__
 419         struct proc_dir_entry *lmv_proc_dir;
 420 #endif
 421         int rc, i;
 422         ENTRY;
 423
 424         if (!lmv->tgts)
 425                 goto out_local;
 426
 427         /* Only disconnect the underlying layers on the final disconnect. */
 428         lmv->refcount--;
 429         if (lmv->refcount != 0)
 430                 goto out_local;
 431
 432 #ifdef __KERNEL__
 433         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 434 #endif
 435
 436         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 437                 struct obd_device *mdc_obd;
 438
 439                 if (lmv->tgts[i].ltd_exp == NULL)
 440                         continue;
 441
 442                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 443
 444                 if (mdc_obd)
 445                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 446
 447 #ifdef __KERNEL__
 448                 if (lmv_proc_dir) {
 449                         struct proc_dir_entry *mdc_symlink;
 450
 451                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 452                         if (mdc_symlink) {
 453                                 lprocfs_remove(mdc_symlink);
 454                         } else {
 455                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 456                                        obd->obd_type->typ_name, obd->obd_name,
 457                                        mdc_obd->obd_name);
 458                         }
 459                 }
 460 #endif
 461                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 462                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 463                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 464
 465                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 466                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 467                 if (rc) {
 468                         if (lmv->tgts[i].active) {
 469                                 CERROR("Target %s disconnect error %d\n",
 470                                        lmv->tgts[i].uuid.uuid, rc);
 471                         }
 472                         rc = 0;
 473                 }
 474
 475                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 476                 lmv->tgts[i].ltd_exp = NULL;
 477         }
 478
 479 #ifdef __KERNEL__
 480         if (lmv_proc_dir) {
 481                 lprocfs_remove(lmv_proc_dir);
 482         } else {
 483                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 484                        obd->obd_type->typ_name, obd->obd_name);
 485         }
 486 #endif
 487
 488 out_local:
 489         /* this is the case when no real connection is established by
 490          * lmv_check_connect(). */
 491         if (!lmv->connected)
 492                 class_export_put(exp);
 493         rc = class_disconnect(exp, 0);
 494         if (lmv->refcount == 0)
 495                 lmv->connected = 0;
 496         RETURN(rc);
 497 }
 498
 499 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 500                          int len, void *karg, void *uarg)
 501 {
 502         struct obd_device *obddev = class_exp2obd(exp);
 503         struct lmv_obd *lmv = &obddev->u.lmv;
 504         int i, rc = 0, set = 0;
 505         ENTRY;
 506
 507         if (lmv->desc.ld_tgt_count == 0)
 508                 RETURN(-ENOTTY);
 509
 510         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 511                 int err;
 512
 513                 if (lmv->tgts[i].ltd_exp == NULL)
 514                         continue;
 515
 516                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 517                 if (err) {
 518                         if (lmv->tgts[i].active) {
 519                                 CERROR("error: iocontrol MDC %s on MDT"
 520                                        "idx %d: err = %d\n",
 521                                        lmv->tgts[i].uuid.uuid, i, err);
 522                                 if (!rc)
 523                                         rc = err;
 524                         }
 525                 } else
 526                         set = 1;
 527         }
 528         if (!set && !rc)
 529                 rc = -EIO;
 530
 531         RETURN(rc);
 532 }
 533
 534 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 535 {
 536         int i, rc = 0;
 537         struct lmv_desc *desc;
 538         struct obd_uuid *uuids;
 539         struct lmv_tgt_desc *tgts;
 540         struct obd_device *tgt_obd;
 541         struct lustre_cfg *lcfg = buf;
 542         struct lmv_obd *lmv = &obd->u.lmv;
 543         ENTRY;
 544
 545         if (lcfg->lcfg_inllen1 < 1) {
 546                 CERROR("LMV setup requires a descriptor\n");
 547                 RETURN(-EINVAL);
 548         }
 549
 550         if (lcfg->lcfg_inllen2 < 1) {
 551                 CERROR("LMV setup requires an MDT UUID list\n");
 552                 RETURN(-EINVAL);
 553         }
 554
 555         desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
 556         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
 557                 CERROR("descriptor size wrong: %d > %d\n",
 558                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
 559                 RETURN(-EINVAL);
 560         }
 561
 562         uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
 563         if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
 564                 CERROR("UUID array size wrong: %u * %u != %u\n",
 565                        sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
 566                 RETURN(-EINVAL);
 567         }
 568
 569         lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
 570         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 571         if (lmv->tgts == NULL) {
 572                 CERROR("Out of memory\n");
 573                 RETURN(-ENOMEM);
 574         }
 575
 576         lmv->desc = *desc;
 577         spin_lock_init(&lmv->lmv_lock);
 578
 579         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
 580                 tgts->uuid = uuids[i];
 581
 582         lmv->max_cookiesize = 0;
 583
 584         lmv->max_easize = sizeof(struct lustre_id) *
 585                 desc->ld_tgt_count + sizeof(struct mea);
 586
 587         rc = lmv_setup_mgr(obd);
 588         if (rc) {
 589                 CERROR("Can't setup LMV object manager, "
 590                        "error %d.\n", rc);
 591                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 592         }
 593
 594         tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
 595                                         &obd->obd_uuid);
 596         if (!tgt_obd) {
 597                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
 598                 RETURN(-EINVAL);
 599         }
 600
 601         RETURN(rc);
 602 }
 603
 604 static int lmv_cleanup(struct obd_device *obd, int flags)
 605 {
 606         struct lmv_obd *lmv = &obd->u.lmv;
 607         ENTRY;
 608
 609         lmv_cleanup_mgr(obd);
 610         OBD_FREE(lmv->tgts, lmv->tgts_size);
 611
 612         RETURN(0);
 613 }
 614
 615 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 616                       unsigned long max_age)
 617 {
 618         struct lmv_obd *lmv = &obd->u.lmv;
 619         struct obd_statfs *temp;
 620         int rc = 0, i;
 621         ENTRY;
 622
 623         rc = lmv_check_connect(obd);
 624         if (rc)
 625                 RETURN(rc);
 626
 627         OBD_ALLOC(temp, sizeof(*temp));
 628         if (temp == NULL)
 629                 RETURN(-ENOMEM);
 630
 631         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 632                 if (lmv->tgts[i].ltd_exp == NULL) {
 633                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 634                         continue;
 635                 }
 636
 637                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 638                 if (rc) {
 639                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 640                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 641                                rc);
 642                         GOTO(out_free_temp, rc);
 643                 }
 644                 if (i == 0) {
 645                         memcpy(osfs, temp, sizeof(*temp));
 646                 } else {
 647                         osfs->os_bavail += temp->os_bavail;
 648                         osfs->os_blocks += temp->os_blocks;
 649                         osfs->os_ffree += temp->os_ffree;
 650                         osfs->os_files += temp->os_files;
 651                 }
 652         }
 653
 654         EXIT;
 655 out_free_temp:
 656         OBD_FREE(temp, sizeof(*temp));
 657         return rc;
 658 }
 659
 660 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 661 {
 662         struct obd_device *obd = exp->exp_obd;
 663         struct lmv_obd *lmv = &obd->u.lmv;
 664         int rc;
 665         ENTRY;
 666
 667         rc = lmv_check_connect(obd);
 668         if (rc)
 669                 RETURN(rc);
 670
 671         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 672         id_group(id) = 0;
 673
 674         RETURN(rc);
 675 }
 676
 677 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 678                        __u64 valid, unsigned int ea_size,
 679                        struct ptlrpc_request **request)
 680 {
 681         struct obd_device *obd = exp->exp_obd;
 682         struct lmv_obd *lmv = &obd->u.lmv;
 683         int rc, i = id_group(id);
 684         struct lmv_obj *obj;
 685         ENTRY;
 686
 687         rc = lmv_check_connect(obd);
 688         if (rc)
 689                 RETURN(rc);
 690
 691         LASSERT(i < lmv->desc.ld_tgt_count);
 692
 693         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 694                         ea_size, request);
 695         if (rc)
 696                 RETURN(rc);
 697
 698         obj = lmv_grab_obj(obd, id);
 699
 700         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 701                OLID4(id), obj ? "(splitted)" : "");
 702
 703         /*
 704          * if object is splitted, then we loop over all the slaves and gather
 705          * size attribute. In ideal world we would have to gather also mds field
 706          * from all slaves, as object is spread over the cluster and this is
 707          * definitely interesting information and it is not good to loss it,
 708          * but...
 709          */
 710         if (obj) {
 711                 struct mds_body *body;
 712
 713                 if (*request == NULL) {
 714                         lmv_put_obj(obj);
 715                         RETURN(rc);
 716                 }
 717
 718                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 719                                       sizeof(*body));
 720                 LASSERT(body != NULL);
 721
 722                 lmv_lock_obj(obj);
 723
 724                 for (i = 0; i < obj->objcount; i++) {
 725
 726                         if (lmv->tgts[i].ltd_exp == NULL) {
 727                                 CWARN("%s: NULL export for %d\n",
 728                                       obd->obd_name, i);
 729                                 continue;
 730                         }
 731
 732                         /* skip master obj. */
 733                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 734                                 continue;
 735
 736                         body->size += obj->objs[i].size;
 737                 }
 738
 739                 lmv_unlock_obj(obj);
 740                 lmv_put_obj(obj);
 741         }
 742
 743         RETURN(rc);
 744 }
 745
 746 static int lmv_change_cbdata(struct obd_export *exp,
 747                              struct lustre_id *id,
 748                              ldlm_iterator_t it,
 749                              void *data)
 750 {
 751         struct obd_device *obd = exp->exp_obd;
 752         struct lmv_obd *lmv = &obd->u.lmv;
 753         int rc = 0;
 754         ENTRY;
 755
 756         rc = lmv_check_connect(obd);
 757         if (rc)
 758                 RETURN(rc);
 759
 760         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 761         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 762
 763         rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
 764                               id, it, data);
 765
 766         RETURN(rc);
 767 }
 768
 769 static int lmv_change_cbdata_name(struct obd_export *exp,
 770                                   struct lustre_id *pid,
 771                                   char *name, int len,
 772                                   struct lustre_id *cid,
 773                                   ldlm_iterator_t it,
 774                                   void *data)
 775 {
 776         struct obd_device *obd = exp->exp_obd;
 777         struct lmv_obd *lmv = &obd->u.lmv;
 778         struct lustre_id rcid = *cid;
 779         struct lmv_obj *obj;
 780         int rc = 0, mds;
 781         ENTRY;
 782
 783         rc = lmv_check_connect(obd);
 784         if (rc)
 785                 RETURN(rc);
 786
 787         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 788         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 789
 790         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 791                OLID4(pid), len, name, OLID4(cid));
 792
 793         /* this is default mds for directory name belongs to. */
 794         mds = id_group(pid);
 795         obj = lmv_grab_obj(obd, pid);
 796         if (obj) {
 797                 /* directory is splitted. look for right mds for this name. */
 798                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 799                 rcid = obj->objs[mds].id;
 800                 mds = id_group(&rcid);
 801                 lmv_put_obj(obj);
 802         }
 803         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 804         RETURN(rc);
 805 }
 806
 807 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 808 {
 809         struct obd_device *obd = exp->exp_obd;
 810         struct lmv_obd *lmv = &obd->u.lmv;
 811         int rc = 0;
 812         ENTRY;
 813
 814         rc = lmv_check_connect(obd);
 815         if (rc)
 816                 RETURN(rc);
 817
 818         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 819         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 820         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 821         RETURN(rc);
 822 }
 823
 824 int lmv_close(struct obd_export *exp, struct obdo *obdo,
 825               struct obd_client_handle *och,
 826               struct ptlrpc_request **request)
 827 {
 828         struct obd_device *obd = exp->exp_obd;
 829         struct lmv_obd *lmv = &obd->u.lmv;
 830         int rc, i = obdo->o_mds;
 831         ENTRY;
 832
 833         rc = lmv_check_connect(obd);
 834         if (rc)
 835                 RETURN(rc);
 836
 837         LASSERT(i < lmv->desc.ld_tgt_count);
 838         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 839                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 840         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 841         RETURN(rc);
 842 }
 843
 844 int lmv_get_mea_and_update_object(struct obd_export *exp,
 845                                   struct lustre_id *id)
 846 {
 847         struct obd_device *obd = exp->exp_obd;
 848         struct lmv_obd *lmv = &obd->u.lmv;
 849         struct ptlrpc_request *req = NULL;
 850         struct lmv_obj *obj;
 851         struct lustre_md md;
 852         int mealen, rc;
 853         __u64 valid;
 854         ENTRY;
 855
 856         md.mea = NULL;
 857         mealen = MEA_SIZE_LMV(lmv);
 858
 859         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 860
 861         /* time to update mea of parent id */
 862         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 863                         id, valid, mealen, &req);
 864         if (rc) {
 865                 CERROR("md_getattr() failed, error %d\n", rc);
 866                 GOTO(cleanup, rc);
 867         }
 868
 869         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 870         if (rc) {
 871                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 872                 GOTO(cleanup, rc);
 873         }
 874
 875         if (md.mea == NULL)
 876                 GOTO(cleanup, rc = -ENODATA);
 877
 878         obj = lmv_create_obj(exp, id, md.mea);
 879         if (IS_ERR(obj))
 880                 rc = PTR_ERR(obj);
 881
 882         lmv_put_obj(obj);
 883         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 884
 885         EXIT;
 886 cleanup:
 887         if (req)
 888                 ptlrpc_req_finished(req);
 889         return rc;
 890 }
 891
 892 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 893                const void *data, int datalen, int mode, __u32 uid,
 894                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 895 {
 896         struct obd_device *obd = exp->exp_obd;
 897         struct lmv_obd *lmv = &obd->u.lmv;
 898         struct mds_body *body;
 899         struct lmv_obj *obj;
 900         int rc, mds, loop = 0;
 901         ENTRY;
 902
 903         rc = lmv_check_connect(obd);
 904         if (rc)
 905                 RETURN(rc);
 906
 907         if (!lmv->desc.ld_active_tgt_count)
 908                 RETURN(-EIO);
 909 repeat:
 910         LASSERT(++loop <= 2);
 911         obj = lmv_grab_obj(obd, &op_data->id1);
 912         if (obj) {
 913                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 914                                    op_data->name, op_data->namelen);
 915                 op_data->id1 = obj->objs[mds].id;
 916                 lmv_put_obj(obj);
 917         }
 918
 919         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 920                op_data->name, OLID4(&op_data->id1));
 921
 922         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 923                        op_data, data, datalen, mode, uid, gid, rdev,
 924                        request);
 925         if (rc == 0) {
 926                 if (*request == NULL)
 927                         RETURN(rc);
 928
 929                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 930                                       sizeof(*body));
 931                 if (body == NULL)
 932                         RETURN(-ENOMEM);
 933
 934                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 935         } else if (rc == -ERESTART) {
 936                 /*
 937                  * directory got splitted. time to update local object and
 938                  * repeat the request with proper MDS.
 939                  */
 940                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 941                 if (rc == 0) {
 942                         ptlrpc_req_finished(*request);
 943                         goto repeat;
 944                 }
 945         }
 946         RETURN(rc);
 947 }
 948
 949 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 950 {
 951         struct obd_device *obd = exp->exp_obd;
 952         struct lmv_obd *lmv = &obd->u.lmv;
 953         int rc;
 954         ENTRY;
 955
 956         rc = lmv_check_connect(obd);
 957         if (rc)
 958                 RETURN(rc);
 959
 960         /* FIXME: choose right MDC here */
 961         CWARN("this method isn't implemented yet\n");
 962         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 963         RETURN(rc);
 964 }
 965
 966 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
 967                        struct lookup_intent *it, int lockmode,
 968                        struct mdc_op_data *data, struct lustre_handle *lockh,
 969                        void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
 970                        ldlm_blocking_callback cb_blocking, void *cb_data)
 971 {
 972         struct obd_device *obd = exp->exp_obd;
 973         struct lmv_obd *lmv = &obd->u.lmv;
 974         struct mea *mea = data->mea1;
 975         struct mdc_op_data *data2;
 976         int i, rc, mds;
 977         ENTRY;
 978
 979         OBD_ALLOC(data2, sizeof(*data2));
 980         if (data2 == NULL)
 981                 RETURN(-ENOMEM);
 982
 983         LASSERT(mea != NULL);
 984         for (i = 0; i < mea->mea_count; i++) {
 985                 memset(data2, 0, sizeof(*data2));
 986                 data2->id1 = mea->mea_ids[i];
 987                 mds = id_group(&data2->id1);
 988
 989                 if (lmv->tgts[mds].ltd_exp == NULL)
 990                         continue;
 991
 992                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
 993                                 lockmode, data2, lockh + i, lmm, lmmsize,
 994                                 cb_compl, cb_blocking, cb_data);
 995
 996                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
 997                        OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
 998                 if (rc)
 999                         GOTO(cleanup, rc);
1000                 if (it->d.lustre.it_data) {
1001                         struct ptlrpc_request *req;
1002                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1003                         ptlrpc_req_finished(req);
1004                 }
1005
1006                 if (it->d.lustre.it_status)
1007                         GOTO(cleanup, rc = it->d.lustre.it_status);
1008         }
1009
1010         OBD_FREE(data2, sizeof(*data2));
1011         RETURN(0);
1012 cleanup:
1013         OBD_FREE(data2, sizeof(*data2));
1014
1015         /* drop all taken locks */
1016         while (--i >= 0) {
1017                 if (lockh[i].cookie)
1018                         ldlm_lock_decref(lockh + i, lockmode);
1019                 lockh[i].cookie = 0;
1020         }
1021         return rc;
1022 }
1023
1024 int lmv_enqueue(struct obd_export *exp, int lock_type,
1025                 struct lookup_intent *it, int lock_mode,
1026                 struct mdc_op_data *data, struct lustre_handle *lockh,
1027                 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1028                 ldlm_blocking_callback cb_blocking, void *cb_data)
1029 {
1030         struct obd_device *obd = exp->exp_obd;
1031         struct lmv_obd *lmv = &obd->u.lmv;
1032         struct lmv_obj *obj;
1033         int rc, mds;
1034         ENTRY;
1035
1036         rc = lmv_check_connect(obd);
1037         if (rc)
1038                 RETURN(rc);
1039
1040         if (data->mea1 && it->it_op == IT_UNLINK) {
1041                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1042                                         data, lockh, lmm, lmmsize,
1043                                         cb_compl, cb_blocking, cb_data);
1044                 RETURN(rc);
1045         }
1046
1047         if (data->namelen) {
1048                 obj = lmv_grab_obj(obd, &data->id1);
1049                 if (obj) {
1050                         /* directory is splitted. look for right mds for this
1051                          * name */
1052                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1053                                            (char *)data->name, data->namelen);
1054                         data->id1 = obj->objs[mds].id;
1055                         lmv_put_obj(obj);
1056                 }
1057         }
1058         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1059                OLID4(&data->id1));
1060
1061         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1062                         lock_type, it, lock_mode, data, lockh, lmm,
1063                         lmmsize, cb_compl, cb_blocking, cb_data);
1064         RETURN(rc);
1065 }
1066
1067 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1068                      char *filename, int namelen, __u64 valid,
1069                      unsigned int ea_size, struct ptlrpc_request **request)
1070 {
1071         int rc, mds = id_group(id), loop = 0;
1072         struct obd_device *obd = exp->exp_obd;
1073         struct lmv_obd *lmv = &obd->u.lmv;
1074         struct lustre_id rid = *id;
1075         struct mds_body *body;
1076         struct lmv_obj *obj;
1077         __u64 old_valid;
1078         ENTRY;
1079
1080         rc = lmv_check_connect(obd);
1081         if (rc)
1082                 RETURN(rc);
1083 repeat:
1084         LASSERT(++loop <= 2);
1085         obj = lmv_grab_obj(obd, id);
1086         if (obj) {
1087                 /* directory is splitted. look for right mds for this name */
1088                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1089                                    filename, namelen - 1);
1090                 rid = obj->objs[mds].id;
1091                 lmv_put_obj(obj);
1092         }
1093
1094         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1095                namelen, filename, OLID4(id), OLID4(&rid));
1096
1097         old_valid = valid;
1098
1099         /*
1100          * here should be applied OBD_MD_FID to ->valid, because otherwise,
1101          * mds_getattr_lock() will not fetch fid component of lustre_id and
1102          * thus, next call to md_getattr_lock() will be performed to wrong mds.
1103          */
1104         if (!(old_valid & OBD_MD_FID))
1105                 valid |= OBD_MD_FID;
1106
1107         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1108                              &rid, filename, namelen, valid,
1109                              ea_size, request);
1110         if (rc == 0) {
1111                 /*
1112                  * this could be cross-node reference. in this case all we have
1113                  * right now is lustre_id triple. we'd like to find other
1114                  * attributes.
1115                  */
1116                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1117                 LASSERT(body != NULL);
1118                 if (body->valid & OBD_MD_MDS) {
1119                         struct ptlrpc_request *req = NULL;
1120
1121                         rid = body->id1;
1122                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1123
1124                         /*
1125                          * turning OBD_MD_FID fetching off, as we already have
1126                          * full lustre_id and do need to fetch fid component
1127                          * again. This will help to make thing slightly faster.
1128                          */
1129                         if (!(old_valid & OBD_MD_FID))
1130                                 valid &= ~OBD_MD_FID;
1131
1132                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1133                                              &rid, NULL, 1, valid, ea_size, &req);
1134                         ptlrpc_req_finished(*request);
1135                         *request = req;
1136                 }
1137         } else if (rc == -ERESTART) {
1138                 /* directory got splitted. time to update local object and
1139                  * repeat the request with proper MDS */
1140                 rc = lmv_get_mea_and_update_object(exp, &rid);
1141                 if (rc == 0) {
1142                         ptlrpc_req_finished(*request);
1143                         goto repeat;
1144                 }
1145         }
1146         RETURN(rc);
1147 }
1148
1149 /*
1150  * llite passes id of an target inode in data->id1 and id of directory in
1151  * data->id2
1152  */
1153 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1154              struct ptlrpc_request **request)
1155 {
1156         struct obd_device *obd = exp->exp_obd;
1157         struct lmv_obd *lmv = &obd->u.lmv;
1158         struct lmv_obj *obj;
1159         int rc;
1160         ENTRY;
1161
1162         rc = lmv_check_connect(obd);
1163         if (rc)
1164                 RETURN(rc);
1165
1166         if (data->namelen != 0) {
1167                 /* usual link request */
1168                 obj = lmv_grab_obj(obd, &data->id1);
1169                 if (obj) {
1170                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1171                                           data->name, data->namelen);
1172                         data->id1 = obj->objs[rc].id;
1173                         lmv_put_obj(obj);
1174                 }
1175
1176                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1177                        OLID4(&data->id2), data->namelen, data->name,
1178                        OLID4(&data->id1));
1179         } else {
1180                 /* request from MDS to acquire i_links for inode by id1 */
1181                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1182                        OLID4(&data->id1));
1183         }
1184
1185         rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1186                      data, request);
1187         RETURN(rc);
1188 }
1189
1190 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1191                const char *old, int oldlen, const char *new, int newlen,
1192                struct ptlrpc_request **request)
1193 {
1194         struct obd_device *obd = exp->exp_obd;
1195         struct lmv_obd *lmv = &obd->u.lmv;
1196         struct lmv_obj *obj;
1197         int rc, mds;
1198         ENTRY;
1199
1200         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1201                oldlen, old, OLID4(&data->id1), newlen, new,
1202                OLID4(&data->id2));
1203
1204         rc = lmv_check_connect(obd);
1205         if (rc)
1206                 RETURN(rc);
1207
1208         if (oldlen == 0) {
1209                 /*
1210                  * MDS with old dir entry is asking another MDS to create name
1211                  * there.
1212                  */
1213                 CDEBUG(D_OTHER,
1214                        "create %*s(%d/%d) in "DLID4" pointing "
1215                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1216                        OLID4(&data->id2), OLID4(&data->id1));
1217
1218                 mds = id_group(&data->id2);
1219
1220                 /*
1221                  * target directory can be splitted, sowe should forward request
1222                  * to the right MDS.
1223                  */
1224                 obj = lmv_grab_obj(obd, &data->id2);
1225                 if (obj) {
1226                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1227                                            (char *)new, newlen);
1228                         data->id2 = obj->objs[mds].id;
1229                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1230                                OLID4(&data->id2));
1231                         lmv_put_obj(obj);
1232                 }
1233                 goto request;
1234         }
1235
1236         obj = lmv_grab_obj(obd, &data->id1);
1237         if (obj) {
1238                 /*
1239                  * directory is already splitted, so we have to forward request
1240                  * to the right MDS.
1241                  */
1242                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1243                                    (char *)old, oldlen);
1244                 data->id1 = obj->objs[mds].id;
1245                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1246                        OLID4(&data->id1));
1247                 lmv_put_obj(obj);
1248         }
1249
1250         obj = lmv_grab_obj(obd, &data->id2);
1251         if (obj) {
1252                 /*
1253                  * directory is already splitted, so we have to forward request
1254                  * to the right MDS.
1255                  */
1256                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1257                                    (char *)new, newlen);
1258
1259                 data->id2 = obj->objs[mds].id;
1260                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1261                        OLID4(&data->id2));
1262                 lmv_put_obj(obj);
1263         }
1264
1265         mds = id_group(&data->id1);
1266
1267 request:
1268         if (id_group(&data->id1) != id_group(&data->id2)) {
1269                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1270                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1271                        newlen, new);
1272         }
1273
1274         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1275                        new, newlen, request);
1276         RETURN(rc);
1277 }
1278
1279 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1280                 struct iattr *iattr, void *ea, int ealen, void *ea2,
1281                 int ea2len, struct ptlrpc_request **request)
1282 {
1283         struct obd_device *obd = exp->exp_obd;
1284         struct lmv_obd *lmv = &obd->u.lmv;
1285         struct ptlrpc_request *req;
1286         struct mds_body *body;
1287         struct lmv_obj *obj;
1288         int rc = 0, i;
1289         ENTRY;
1290
1291         rc = lmv_check_connect(obd);
1292         if (rc)
1293                 RETURN(rc);
1294
1295         obj = lmv_grab_obj(obd, &data->id1);
1296
1297         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1298                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1299
1300         if (obj) {
1301                 for (i = 0; i < obj->objcount; i++) {
1302                         data->id1 = obj->objs[i].id;
1303
1304                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1305                                         data, iattr, ea, ealen, ea2, ea2len, &req);
1306
1307                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1308                                 /*
1309                                  * this is master object and this request should
1310                                  * be returned back to llite.
1311                                  */
1312                                 *request = req;
1313                         } else {
1314                                 ptlrpc_req_finished(req);
1315                         }
1316
1317                         if (rc)
1318                                 break;
1319                 }
1320                 lmv_put_obj(obj);
1321         } else {
1322                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1323                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1324                                 data, iattr, ea, ealen, ea2, ea2len, request);
1325                 if (rc == 0) {
1326                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1327                                               sizeof(*body));
1328                         LASSERT(body != NULL);
1329                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1330                 }
1331         }
1332         RETURN(rc);
1333 }
1334
1335 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1336              struct ptlrpc_request **request)
1337 {
1338         struct obd_device *obd = exp->exp_obd;
1339         struct lmv_obd *lmv = &obd->u.lmv;
1340         int rc;
1341         ENTRY;
1342
1343         rc = lmv_check_connect(obd);
1344         if (rc)
1345                 RETURN(rc);
1346
1347         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1348                      id, request);
1349         RETURN(rc);
1350 }
1351
1352 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1353                             struct ldlm_lock_desc *desc,
1354                             void *data, int flag)
1355 {
1356         struct lustre_handle lockh;
1357         struct lmv_obj *obj;
1358         int rc;
1359         ENTRY;
1360
1361         switch (flag) {
1362         case LDLM_CB_BLOCKING:
1363                 ldlm_lock2handle(lock, &lockh);
1364                 rc = ldlm_cli_cancel(&lockh);
1365                 if (rc < 0) {
1366                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1367                         RETURN(rc);
1368                 }
1369                 break;
1370         case LDLM_CB_CANCELING:
1371                 /* time to drop cached attrs for dirobj */
1372                 obj = lock->l_ast_data;
1373                 if (obj) {
1374                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1375                                ", master "DLID4"\n",
1376                                lock->l_resource->lr_name.name[3] == 1 ?
1377                                "LOOKUP" : "UPDATE",
1378                                lock->l_resource->lr_name.name[0],
1379                                lock->l_resource->lr_name.name[1],
1380                                OLID4(&obj->id));
1381                         lmv_put_obj(obj);
1382                 }
1383                 break;
1384         default:
1385                 LBUG();
1386         }
1387         RETURN(0);
1388 }
1389
1390 void lmv_remove_dots(struct page *page)
1391 {
1392         unsigned limit = PAGE_CACHE_SIZE;
1393         char *kaddr = page_address(page);
1394         struct ext2_dir_entry_2 *p;
1395         unsigned offs, rec_len;
1396
1397         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1398                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1399                 rec_len = le16_to_cpu(p->rec_len);
1400
1401                 if ((p->name_len == 1 && p->name[0] == '.') ||
1402                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1403                         p->inode = 0;
1404         }
1405 }
1406
1407 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1408                  __u64 offset, struct page *page,
1409                  struct ptlrpc_request **request)
1410 {
1411         struct obd_device *obd = exp->exp_obd;
1412         struct lmv_obd *lmv = &obd->u.lmv;
1413         struct lustre_id rid = *id;
1414         struct lmv_obj *obj;
1415         int rc, i;
1416         ENTRY;
1417
1418 #warning "we need well-desgined readdir() implementation"
1419         rc = lmv_check_connect(obd);
1420         if (rc)
1421                 RETURN(rc);
1422
1423         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1424         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1425                offset, OLID4(&rid));
1426
1427         obj = lmv_grab_obj(obd, id);
1428         if (obj) {
1429                 lmv_lock_obj(obj);
1430
1431                 /* find dirobj containing page with requested offset. */
1432                 for (i = 0; i < obj->objcount; i++) {
1433                         if (offset < obj->objs[i].size)
1434                                 break;
1435                         offset -= obj->objs[i].size;
1436                 }
1437                 rid = obj->objs[i].id;
1438
1439                 lmv_unlock_obj(obj);
1440                 lmv_put_obj(obj);
1441
1442                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1443                        OLID4(&rid), (unsigned long)offset);
1444         }
1445         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1446                          offset, page, request);
1447
1448         if (rc == 0 && !id_equal_fid(&rid, id))
1449                 /* this page isn't from master object. To avoid "." and ".."
1450                  * duplication in directory, we have to remove them from all
1451                  * slave objects */
1452                 lmv_remove_dots(page);
1453
1454         RETURN(rc);
1455 }
1456
1457 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1458                       struct ptlrpc_request **req)
1459 {
1460         struct obd_device *obd = exp->exp_obd;
1461         struct lmv_obd *lmv = &obd->u.lmv;
1462         struct mea *mea = data->mea1;
1463         struct mdc_op_data *data2;
1464         int i, rc = 0;
1465         ENTRY;
1466
1467         OBD_ALLOC(data2, sizeof(*data2));
1468         if (data2 == NULL)
1469                 RETURN(-ENOMEM);
1470
1471         LASSERT(mea != NULL);
1472         for (i = 0; i < mea->mea_count; i++) {
1473                 memset(data2, 0, sizeof(*data2));
1474                 data2->id1 = mea->mea_ids[i];
1475                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1476
1477                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1478                         continue;
1479
1480                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1481                                data2, req);
1482
1483                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1484                        OLID4(&mea->mea_ids[i]), rc);
1485
1486                 if (*req) {
1487                         ptlrpc_req_finished(*req);
1488                         *req = NULL;
1489                 }
1490                 if (rc)
1491                         RETURN(rc);
1492         }
1493         OBD_FREE(data2, sizeof(*data2));
1494         RETURN(rc);
1495 }
1496
1497 int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1498 {
1499         ENTRY;
1500
1501         LASSERT(exp && id);
1502         if (lmv_delete_obj(exp, id)) {
1503                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1504                        OLID4(id));
1505         }
1506         RETURN(0);
1507 }
1508
1509 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1510                struct ptlrpc_request **request)
1511 {
1512         struct obd_device *obd = exp->exp_obd;
1513         struct lmv_obd *lmv = &obd->u.lmv;
1514         int rc, i = 0;
1515         ENTRY;
1516
1517         rc = lmv_check_connect(obd);
1518         if (rc)
1519                 RETURN(rc);
1520
1521         if (data->namelen == 0 && data->mea1 != NULL) {
1522                 /* mds asks to remove slave objects */
1523                 rc = lmv_unlink_slaves(exp, data, request);
1524                 RETURN(rc);
1525         }
1526
1527         if (data->namelen != 0) {
1528                 struct lmv_obj *obj;
1529
1530                 obj = lmv_grab_obj(obd, &data->id1);
1531                 if (obj) {
1532                         i = raw_name2idx(obj->hashtype, obj->objcount,
1533                                          data->name, data->namelen);
1534                         data->id1 = obj->objs[i].id;
1535                         lmv_put_obj(obj);
1536                 }
1537                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1538                        data->namelen, data->name, OLID4(&data->id1),
1539                        i);
1540         } else {
1541                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1542                        OLID4(&data->id1));
1543         }
1544         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1545                        data, request);
1546         RETURN(rc);
1547 }
1548
1549 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1550                                     char *name, int len)
1551 {
1552         struct obd_device *obd = exp->exp_obd;
1553         struct lmv_obd *lmv = &obd->u.lmv;
1554         int rc;
1555         ENTRY;
1556
1557         rc = lmv_check_connect(obd);
1558         if (rc)
1559                 RETURN(ERR_PTR(rc));
1560         obd = lmv->tgts[0].ltd_exp->exp_obd;
1561         EXIT;
1562
1563         return obd;
1564 }
1565
1566 int lmv_init_ea_size(struct obd_export *exp, int easize,
1567                      int cookiesize)
1568 {
1569         struct obd_device *obd = exp->exp_obd;
1570         struct lmv_obd *lmv = &obd->u.lmv;
1571         int i, rc = 0, change = 0;
1572         ENTRY;
1573
1574         if (lmv->max_easize < easize) {
1575                 lmv->max_easize = easize;
1576                 change = 1;
1577         }
1578         if (lmv->max_cookiesize < cookiesize) {
1579                 lmv->max_cookiesize = cookiesize;
1580                 change = 1;
1581         }
1582         if (change == 0)
1583                 RETURN(0);
1584
1585         if (lmv->connected == 0)
1586                 RETURN(0);
1587
1588         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1589                 if (lmv->tgts[i].ltd_exp == NULL) {
1590                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1591                         continue;
1592                 }
1593
1594                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1595                 if (rc) {
1596                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1597                                "error %d.\n", i, rc);
1598                         break;
1599                 }
1600         }
1601         RETURN(rc);
1602 }
1603
1604 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1605                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
1606 {
1607         struct obd_device *obd = exp->exp_obd;
1608         struct lmv_obd *lmv = &obd->u.lmv;
1609         struct lov_stripe_md obj_md;
1610         struct lov_stripe_md *obj_mdp = &obj_md;
1611         int rc = 0;
1612         ENTRY;
1613
1614         LASSERT(ea == NULL);
1615         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1616
1617         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
1618                         oa, &obj_mdp, oti);
1619
1620         RETURN(rc);
1621 }
1622
1623 int lmv_getready(struct obd_export *exp)
1624 {
1625         struct obd_device *obd = exp->exp_obd;
1626         int rc = 0;
1627
1628         ENTRY;
1629         rc = lmv_check_connect(obd);
1630         RETURN(rc);
1631 }
1632
1633 /*
1634  * to be called from MDS only. @oa should have correct store cookie and o_fid
1635  * values for "master" object, as it will be used.
1636  */
1637 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1638                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1639 {
1640         struct obd_device *obd = exp->exp_obd;
1641         struct lmv_obd *lmv = &obd->u.lmv;
1642         struct lustre_id mid;
1643         int i, c, rc = 0;
1644         struct mea *mea;
1645         ENTRY;
1646
1647         rc = lmv_check_connect(obd);
1648         if (rc)
1649                 RETURN(rc);
1650
1651         LASSERT(oa != NULL);
1652
1653         if (ea == NULL) {
1654                 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1655                 if (rc)
1656                         CERROR("Can't create object, rc = %d\n", rc);
1657                 RETURN(rc);
1658         }
1659
1660         if (*ea == NULL) {
1661                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1662                 if (rc < 0) {
1663                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1664                                rc);
1665                         RETURN(rc);
1666                 } else
1667                         rc = 0;
1668
1669                 if (*ea == NULL)
1670                         RETURN(-ENOMEM);
1671         }
1672
1673         /*
1674          * here we should take care about splitted dir, so store cookie and fid
1675          * for "master" object should already be allocated and passed in @oa.
1676          */
1677         LASSERT(oa->o_id != 0);
1678         LASSERT(oa->o_fid != 0);
1679
1680         /* save "master" object id */
1681         obdo2id(&mid, oa);
1682
1683         mea = (struct mea *)*ea;
1684         mea->mea_master = -1;
1685         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1686
1687         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1688                 mea->mea_count = lmv->desc.ld_tgt_count;
1689
1690         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1691                 struct lov_stripe_md obj_md;
1692                 struct lov_stripe_md *obj_mdp = &obj_md;
1693
1694                 if (lmv->tgts[i].ltd_exp == NULL) {
1695                         /* this is "master" MDS */
1696                         mea->mea_master = i;
1697                         mea->mea_ids[c] = mid;
1698                         c++;
1699                         continue;
1700                 }
1701
1702                 /*
1703                  * "master" MDS should always be part of stripped dir,
1704                  * so scan for it.
1705                  */
1706                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1707                         continue;
1708
1709                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1710                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1711
1712                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1713                 if (rc) {
1714                         CERROR("obd_create() failed on MDT target %d, "
1715                                "error %d\n", c, rc);
1716                         RETURN(rc);
1717                 }
1718
1719                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1720                        i, oa->o_id, oa->o_generation);
1721
1722
1723                 /*
1724                  * here, when object is created (or it is master and was passed
1725                  * from caller) on desired MDS we save its fid to local mea_ids.
1726                  */
1727                 LASSERT(oa->o_fid);
1728
1729                 /*
1730                  * store cookie should be defined here for both cases (master
1731                  * object and not master), because master is already created.
1732                  */
1733                 LASSERT(oa->o_id);
1734
1735                 /* fill mea by store cookie and fid */
1736                 obdo2id(&mea->mea_ids[c], oa);
1737                 c++;
1738         }
1739         LASSERT(c == mea->mea_count);
1740
1741         CDEBUG(D_OTHER, "%d dirobjects created\n",
1742                (int)mea->mea_count);
1743
1744         RETURN(rc);
1745 }
1746
1747 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1748                          struct obd_device *tgt, int count,
1749                          struct llog_catid *logid)
1750 {
1751         struct llog_ctxt *ctxt;
1752         int rc;
1753         ENTRY;
1754
1755         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1756                             &llog_client_ops);
1757         if (rc == 0) {
1758                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1759                 ctxt->loc_imp = tgt->u.cli.cl_import;
1760         }
1761
1762         RETURN(rc);
1763 }
1764
1765 static int lmv_llog_finish(struct obd_device *obd,
1766                            struct obd_llogs *llogs, int count)
1767 {
1768         int rc;
1769         ENTRY;
1770
1771         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1772         RETURN(rc);
1773 }
1774
1775 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1776                         void *key, __u32 *vallen, void *val)
1777 {
1778         struct obd_device *obd;
1779         struct lmv_obd *lmv;
1780         int rc = 0;
1781         ENTRY;
1782
1783         obd = class_exp2obd(exp);
1784         if (obd == NULL) {
1785                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1786                        exp->exp_handle.h_cookie);
1787                 RETURN(-EINVAL);
1788         }
1789
1790         lmv = &obd->u.lmv;
1791         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1792                 __u32 *mdsize = val;
1793                 *vallen = sizeof(__u32);
1794                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1795                         + sizeof(struct mea);
1796                 RETURN(0);
1797         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1798                 struct obd_uuid *cluuid = &lmv->cluuid;
1799                 struct lmv_tgt_desc *tgts;
1800                 __u32 *mdsnum = val;
1801                 int i;
1802
1803                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1804                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1805                                 *vallen = sizeof(__u32);
1806                                 *mdsnum = i;
1807                                 RETURN(0);
1808                         }
1809                 }
1810                 LASSERT(0);
1811         } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1812                 /* getting rootid from first MDS. */
1813                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1814                                   vallen, val);
1815                 RETURN(rc);
1816         } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1817                 struct lmv_desc *desc_ret = val;
1818                 *desc_ret = lmv->desc;
1819                 RETURN(0);
1820         }
1821
1822         CDEBUG(D_IOCTL, "invalid key\n");
1823         RETURN(-EINVAL);
1824 }
1825
1826 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1827                  void *key, obd_count vallen, void *val)
1828 {
1829         struct obd_device *obd;
1830         struct lmv_obd *lmv;
1831         ENTRY;
1832
1833         obd = class_exp2obd(exp);
1834         if (obd == NULL) {
1835                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1836                        exp->exp_handle.h_cookie);
1837                 RETURN(-EINVAL);
1838         }
1839         lmv = &obd->u.lmv;
1840
1841         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1842                 lmv->server_timeout = 1;
1843                 lmv_set_timeouts(obd);
1844                 RETURN(0);
1845         }
1846
1847         RETURN(-EINVAL);
1848 }
1849
1850 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1851                struct lov_stripe_md *lsm)
1852 {
1853         struct obd_device *obd = class_exp2obd(exp);
1854         struct lmv_obd *lmv = &obd->u.lmv;
1855         struct mea *meap, *lsmp;
1856         int mea_size, i;
1857         ENTRY;
1858
1859         mea_size = (sizeof(struct lustre_id) *
1860                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
1861         if (!lmmp)
1862                 RETURN(mea_size);
1863
1864         if (*lmmp && !lsm) {
1865                 OBD_FREE(*lmmp, mea_size);
1866                 *lmmp = NULL;
1867                 RETURN(0);
1868         }
1869
1870         if (*lmmp == NULL) {
1871                 OBD_ALLOC(*lmmp, mea_size);
1872                 if (*lmmp == NULL)
1873                         RETURN(-ENOMEM);
1874         }
1875
1876         if (!lsm)
1877                 RETURN(mea_size);
1878
1879         lsmp = (struct mea *)lsm;
1880         meap = (struct mea *)*lmmp;
1881
1882         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
1883         meap->mea_count = cpu_to_le32(lsmp->mea_count);
1884         meap->mea_master = cpu_to_le32(lsmp->mea_master);
1885
1886         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1887                 meap->mea_ids[i] = meap->mea_ids[i];
1888                 id_cpu_to_le(&meap->mea_ids[i]);
1889         }
1890
1891         RETURN(mea_size);
1892 }
1893
1894 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1895                  struct lov_mds_md *disk_src, int mdsize)
1896 {
1897         struct obd_device *obd = class_exp2obd(exp);
1898         struct mea **tmea = (struct mea **)mem_tgt;
1899         struct mea *mea = (struct mea *)disk_src;
1900         struct lmv_obd *lmv = &obd->u.lmv;
1901         int mea_size, i;
1902         ENTRY;
1903
1904         mea_size = sizeof(struct lustre_id) *
1905                 lmv->desc.ld_tgt_count + sizeof(struct mea);
1906         if (mem_tgt == NULL)
1907                 return mea_size;
1908
1909         if (*mem_tgt != NULL && disk_src == NULL) {
1910                 OBD_FREE(*tmea, mea_size);
1911                 RETURN(0);
1912         }
1913
1914         LASSERT(mea_size == mdsize);
1915
1916         OBD_ALLOC(*tmea, mea_size);
1917         if (*tmea == NULL)
1918                 RETURN(-ENOMEM);
1919
1920         if (!disk_src)
1921                 RETURN(mea_size);
1922
1923         (*tmea)->mea_magic = le32_to_cpu(mea->mea_magic);
1924         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
1925         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
1926
1927         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1928                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
1929                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
1930         }
1931
1932         RETURN(mea_size);
1933 }
1934
1935 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1936             struct lov_stripe_md *ea, obd_count oa_bufs,
1937             struct brw_page *pgarr, struct obd_trans_info *oti)
1938 {
1939         struct obd_device *obd = exp->exp_obd;
1940         struct lmv_obd *lmv = &obd->u.lmv;
1941         struct mea *mea = (struct mea *) ea;
1942         int err;
1943
1944         LASSERT(oa != NULL);
1945         LASSERT(ea != NULL);
1946         LASSERT(pgarr != NULL);
1947         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1948
1949         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1950         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1951         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1952
1953         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1954                       oa, NULL, oa_bufs, pgarr, oti);
1955         RETURN(err);
1956 }
1957
1958 struct obd_ops lmv_obd_ops = {
1959         .o_owner                = THIS_MODULE,
1960         .o_attach               = lmv_attach,
1961         .o_detach               = lmv_detach,
1962         .o_setup                = lmv_setup,
1963         .o_cleanup              = lmv_cleanup,
1964         .o_connect              = lmv_connect,
1965         .o_disconnect           = lmv_disconnect,
1966         .o_statfs               = lmv_statfs,
1967         .o_llog_init            = lmv_llog_init,
1968         .o_llog_finish          = lmv_llog_finish,
1969         .o_get_info             = lmv_get_info,
1970         .o_set_info             = lmv_set_info,
1971         .o_create               = lmv_obd_create,
1972         .o_packmd               = lmv_packmd,
1973         .o_unpackmd             = lmv_unpackmd,
1974         .o_brw                  = lmv_brw,
1975         .o_init_ea_size         = lmv_init_ea_size,
1976         .o_notify               = lmv_notify,
1977         .o_iocontrol            = lmv_iocontrol,
1978         .o_getready             = lmv_getready,
1979 };
1980
1981 struct md_ops lmv_md_ops = {
1982         .m_getstatus           = lmv_getstatus,
1983         .m_getattr             = lmv_getattr,
1984         .m_change_cbdata       = lmv_change_cbdata,
1985         .m_change_cbdata_name  = lmv_change_cbdata_name,
1986         .m_close               = lmv_close,
1987         .m_create              = lmv_create,
1988         .m_done_writing        = lmv_done_writing,
1989         .m_enqueue             = lmv_enqueue,
1990         .m_getattr_lock        = lmv_getattr_lock,
1991         .m_intent_lock         = lmv_intent_lock,
1992         .m_link                = lmv_link,
1993         .m_rename              = lmv_rename,
1994         .m_setattr             = lmv_setattr,
1995         .m_sync                = lmv_sync,
1996         .m_readpage            = lmv_readpage,
1997         .m_unlink              = lmv_unlink,
1998         .m_get_real_obd        = lmv_get_real_obd,
1999         .m_valid_attrs         = lmv_valid_attrs,
2000         .m_delete_inode        = lmv_delete_inode,
2001 };
2002
2003 int __init lmv_init(void)
2004 {
2005         struct lprocfs_static_vars lvars;
2006         int rc;
2007
2008         obj_cache = kmem_cache_create("lmv_objects",
2009                                       sizeof(struct lmv_obj),
2010                                       0, 0, NULL, NULL);
2011         if (!obj_cache) {
2012                 CERROR("error allocating lmv objects cache\n");
2013                 return -ENOMEM;
2014         }
2015
2016         lprocfs_init_vars(lmv, &lvars);
2017         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2018                                  lvars.module_vars,
2019                                  OBD_LMV_DEVICENAME);
2020         if (rc)
2021                 kmem_cache_destroy(obj_cache);
2022
2023         return rc;
2024 }
2025
2026 #ifdef __KERNEL__
2027 static void lmv_exit(void)
2028 {
2029         class_unregister_type(OBD_LMV_DEVICENAME);
2030
2031         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2032                  "can't free lmv objects cache, %d object(s)"
2033                  "still in use\n", atomic_read(&obj_cache_count));
2034 }
2035
2036 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2037 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2038 MODULE_LICENSE("GPL");
2039
2040 module_init(lmv_init);
2041 module_exit(lmv_exit);
2042 #endif