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