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 #ifdef __KERNEL__
 445                 if (lmv_proc_dir) {
 446                         struct proc_dir_entry *mdc_symlink;
 447
 448                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 449                         if (mdc_symlink) {
 450                                 lprocfs_remove(mdc_symlink);
 451                         } else {
 452                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 453                                        obd->obd_type->typ_name, obd->obd_name,
 454                                        mdc_obd->obd_name);
 455                         }
 456                 }
 457 #endif
 458                 if (obd->obd_no_recov) {
 459                         if (mdc_obd)
 460                                 mdc_obd->obd_no_recov = 1;
 461                 }
 462                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 463                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 464                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 465
 466                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 467                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 468                 if (rc) {
 469                         if (lmv->tgts[i].active) {
 470                                 CERROR("Target %s disconnect error %d\n",
 471                                        lmv->tgts[i].uuid.uuid, rc);
 472                         }
 473                         rc = 0;
 474                 }
 475
 476                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 477                 lmv->tgts[i].ltd_exp = NULL;
 478         }
 479
 480 #ifdef __KERNEL__
 481         if (lmv_proc_dir) {
 482                 lprocfs_remove(lmv_proc_dir);
 483         } else {
 484                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 485                        obd->obd_type->typ_name, obd->obd_name);
 486         }
 487 #endif
 488
 489 out_local:
 490         /* this is the case when no real connection is established by
 491          * lmv_check_connect(). */
 492         if (!lmv->connected)
 493                 class_export_put(exp);
 494         rc = class_disconnect(exp, 0);
 495         if (lmv->refcount == 0)
 496                 lmv->connected = 0;
 497         RETURN(rc);
 498 }
 499
 500 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 501                          int len, void *karg, void *uarg)
 502 {
 503         struct obd_device *obddev = class_exp2obd(exp);
 504         struct lmv_obd *lmv = &obddev->u.lmv;
 505         int i, rc = 0, set = 0;
 506         ENTRY;
 507
 508         if (lmv->desc.ld_tgt_count == 0)
 509                 RETURN(-ENOTTY);
 510
 511         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 512                 int err;
 513
 514                 if (lmv->tgts[i].ltd_exp == NULL)
 515                         continue;
 516
 517                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 518                 if (err) {
 519                         if (lmv->tgts[i].active) {
 520                                 CERROR("error: iocontrol MDC %s on MDT"
 521                                        "idx %d: err = %d\n",
 522                                        lmv->tgts[i].uuid.uuid, i, err);
 523                                 if (!rc)
 524                                         rc = err;
 525                         }
 526                 } else
 527                         set = 1;
 528         }
 529         if (!set && !rc)
 530                 rc = -EIO;
 531
 532         RETURN(rc);
 533 }
 534
 535 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 536 {
 537         int i, rc = 0;
 538         struct lmv_desc *desc;
 539         struct obd_uuid *uuids;
 540         struct lmv_tgt_desc *tgts;
 541         struct obd_device *tgt_obd;
 542         struct lustre_cfg *lcfg = buf;
 543         struct lmv_obd *lmv = &obd->u.lmv;
 544         ENTRY;
 545
 546         if (lcfg->lcfg_inllen1 < 1) {
 547                 CERROR("LMV setup requires a descriptor\n");
 548                 RETURN(-EINVAL);
 549         }
 550
 551         if (lcfg->lcfg_inllen2 < 1) {
 552                 CERROR("LMV setup requires an MDT UUID list\n");
 553                 RETURN(-EINVAL);
 554         }
 555
 556         desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
 557         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
 558                 CERROR("descriptor size wrong: %d > %d\n",
 559                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
 560                 RETURN(-EINVAL);
 561         }
 562
 563         uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
 564         if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
 565                 CERROR("UUID array size wrong: %u * %u != %u\n",
 566                        sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
 567                 RETURN(-EINVAL);
 568         }
 569
 570         lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
 571         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 572         if (lmv->tgts == NULL) {
 573                 CERROR("Out of memory\n");
 574                 RETURN(-ENOMEM);
 575         }
 576
 577         lmv->desc = *desc;
 578         spin_lock_init(&lmv->lmv_lock);
 579
 580         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
 581                 tgts->uuid = uuids[i];
 582
 583         lmv->max_cookiesize = 0;
 584
 585         lmv->max_easize = sizeof(struct lustre_id) *
 586                 desc->ld_tgt_count + sizeof(struct mea);
 587
 588         rc = lmv_setup_mgr(obd);
 589         if (rc) {
 590                 CERROR("Can't setup LMV object manager, "
 591                        "error %d.\n", rc);
 592                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 593         }
 594
 595         tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
 596                                         &obd->obd_uuid);
 597         if (!tgt_obd) {
 598                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
 599                 RETURN(-EINVAL);
 600         }
 601
 602         RETURN(rc);
 603 }
 604
 605 static int lmv_cleanup(struct obd_device *obd, int flags)
 606 {
 607         struct lmv_obd *lmv = &obd->u.lmv;
 608         ENTRY;
 609
 610         lmv_cleanup_mgr(obd);
 611         OBD_FREE(lmv->tgts, lmv->tgts_size);
 612
 613         RETURN(0);
 614 }
 615
 616 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 617                       unsigned long max_age)
 618 {
 619         struct lmv_obd *lmv = &obd->u.lmv;
 620         struct obd_statfs *temp;
 621         int rc = 0, i;
 622         ENTRY;
 623
 624         rc = lmv_check_connect(obd);
 625         if (rc)
 626                 RETURN(rc);
 627
 628         OBD_ALLOC(temp, sizeof(*temp));
 629         if (temp == NULL)
 630                 RETURN(-ENOMEM);
 631
 632         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 633                 if (lmv->tgts[i].ltd_exp == NULL) {
 634                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 635                         continue;
 636                 }
 637
 638                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 639                 if (rc) {
 640                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 641                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 642                                rc);
 643                         GOTO(out_free_temp, rc);
 644                 }
 645                 if (i == 0) {
 646                         memcpy(osfs, temp, sizeof(*temp));
 647                 } else {
 648                         osfs->os_bavail += temp->os_bavail;
 649                         osfs->os_blocks += temp->os_blocks;
 650                         osfs->os_ffree += temp->os_ffree;
 651                         osfs->os_files += temp->os_files;
 652                 }
 653         }
 654
 655         EXIT;
 656 out_free_temp:
 657         OBD_FREE(temp, sizeof(*temp));
 658         return rc;
 659 }
 660
 661 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 662 {
 663         struct obd_device *obd = exp->exp_obd;
 664         struct lmv_obd *lmv = &obd->u.lmv;
 665         int rc;
 666         ENTRY;
 667
 668         rc = lmv_check_connect(obd);
 669         if (rc)
 670                 RETURN(rc);
 671
 672         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 673         id_group(id) = 0;
 674
 675         RETURN(rc);
 676 }
 677
 678 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 679                        __u64 valid, unsigned int ea_size,
 680                        struct ptlrpc_request **request)
 681 {
 682         struct obd_device *obd = exp->exp_obd;
 683         struct lmv_obd *lmv = &obd->u.lmv;
 684         int rc, i = id_group(id);
 685         struct lmv_obj *obj;
 686         ENTRY;
 687
 688         rc = lmv_check_connect(obd);
 689         if (rc)
 690                 RETURN(rc);
 691
 692         LASSERT(i < lmv->desc.ld_tgt_count);
 693
 694         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 695                         ea_size, request);
 696         if (rc)
 697                 RETURN(rc);
 698
 699         obj = lmv_grab_obj(obd, id);
 700
 701         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 702                OLID4(id), obj ? "(splitted)" : "");
 703
 704         /*
 705          * if object is splitted, then we loop over all the slaves and gather
 706          * size attribute. In ideal world we would have to gather also mds field
 707          * from all slaves, as object is spread over the cluster and this is
 708          * definitely interesting information and it is not good to loss it,
 709          * but...
 710          */
 711         if (obj) {
 712                 struct mds_body *body;
 713
 714                 if (*request == NULL) {
 715                         lmv_put_obj(obj);
 716                         RETURN(rc);
 717                 }
 718
 719                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 720                                       sizeof(*body));
 721                 LASSERT(body != NULL);
 722
 723                 lmv_lock_obj(obj);
 724
 725                 for (i = 0; i < obj->objcount; i++) {
 726
 727                         if (lmv->tgts[i].ltd_exp == NULL) {
 728                                 CWARN("%s: NULL export for %d\n",
 729                                       obd->obd_name, i);
 730                                 continue;
 731                         }
 732
 733                         /* skip master obj. */
 734                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 735                                 continue;
 736
 737                         body->size += obj->objs[i].size;
 738                 }
 739
 740                 lmv_unlock_obj(obj);
 741                 lmv_put_obj(obj);
 742         }
 743
 744         RETURN(rc);
 745 }
 746
 747 static int lmv_change_cbdata(struct obd_export *exp,
 748                              struct lustre_id *id,
 749                              ldlm_iterator_t it,
 750                              void *data)
 751 {
 752         struct obd_device *obd = exp->exp_obd;
 753         struct lmv_obd *lmv = &obd->u.lmv;
 754         int rc = 0;
 755         ENTRY;
 756
 757         rc = lmv_check_connect(obd);
 758         if (rc)
 759                 RETURN(rc);
 760
 761         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 762         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 763
 764         rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
 765                               id, it, data);
 766
 767         RETURN(rc);
 768 }
 769
 770 static int lmv_change_cbdata_name(struct obd_export *exp,
 771                                   struct lustre_id *pid,
 772                                   char *name, int len,
 773                                   struct lustre_id *cid,
 774                                   ldlm_iterator_t it,
 775                                   void *data)
 776 {
 777         struct obd_device *obd = exp->exp_obd;
 778         struct lmv_obd *lmv = &obd->u.lmv;
 779         struct lustre_id rcid = *cid;
 780         struct lmv_obj *obj;
 781         int rc = 0, mds;
 782         ENTRY;
 783
 784         rc = lmv_check_connect(obd);
 785         if (rc)
 786                 RETURN(rc);
 787
 788         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 789         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 790
 791         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 792                OLID4(pid), len, name, OLID4(cid));
 793
 794         /* this is default mds for directory name belongs to. */
 795         mds = id_group(pid);
 796         obj = lmv_grab_obj(obd, pid);
 797         if (obj) {
 798                 /* directory is splitted. look for right mds for this name. */
 799                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 800                 rcid = obj->objs[mds].id;
 801                 mds = id_group(&rcid);
 802                 lmv_put_obj(obj);
 803         }
 804         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 805         RETURN(rc);
 806 }
 807
 808 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 809 {
 810         struct obd_device *obd = exp->exp_obd;
 811         struct lmv_obd *lmv = &obd->u.lmv;
 812         int rc = 0;
 813         ENTRY;
 814
 815         rc = lmv_check_connect(obd);
 816         if (rc)
 817                 RETURN(rc);
 818
 819         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 820         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 821         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 822         RETURN(rc);
 823 }
 824
 825 int lmv_close(struct obd_export *exp, struct obdo *obdo,
 826               struct obd_client_handle *och,
 827               struct ptlrpc_request **request)
 828 {
 829         struct obd_device *obd = exp->exp_obd;
 830         struct lmv_obd *lmv = &obd->u.lmv;
 831         int rc, i = obdo->o_mds;
 832         ENTRY;
 833
 834         rc = lmv_check_connect(obd);
 835         if (rc)
 836                 RETURN(rc);
 837
 838         LASSERT(i < lmv->desc.ld_tgt_count);
 839         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 840                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 841         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 842         RETURN(rc);
 843 }
 844
 845 int lmv_get_mea_and_update_object(struct obd_export *exp,
 846                                   struct lustre_id *id)
 847 {
 848         struct obd_device *obd = exp->exp_obd;
 849         struct lmv_obd *lmv = &obd->u.lmv;
 850         struct ptlrpc_request *req = NULL;
 851         struct lmv_obj *obj;
 852         struct lustre_md md;
 853         int mealen, rc;
 854         __u64 valid;
 855         ENTRY;
 856
 857         md.mea = NULL;
 858         mealen = MEA_SIZE_LMV(lmv);
 859
 860         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 861
 862         /* time to update mea of parent id */
 863         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 864                         id, valid, mealen, &req);
 865         if (rc) {
 866                 CERROR("md_getattr() failed, error %d\n", rc);
 867                 GOTO(cleanup, rc);
 868         }
 869
 870         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 871         if (rc) {
 872                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 873                 GOTO(cleanup, rc);
 874         }
 875
 876         if (md.mea == NULL)
 877                 GOTO(cleanup, rc = -ENODATA);
 878
 879         obj = lmv_create_obj(exp, id, md.mea);
 880         if (IS_ERR(obj))
 881                 rc = PTR_ERR(obj);
 882
 883         lmv_put_obj(obj);
 884         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 885
 886         EXIT;
 887 cleanup:
 888         if (req)
 889                 ptlrpc_req_finished(req);
 890         return rc;
 891 }
 892
 893 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 894                const void *data, int datalen, int mode, __u32 uid,
 895                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 896 {
 897         struct obd_device *obd = exp->exp_obd;
 898         struct lmv_obd *lmv = &obd->u.lmv;
 899         struct mds_body *body;
 900         struct lmv_obj *obj;
 901         int rc, mds, loop = 0;
 902         ENTRY;
 903
 904         rc = lmv_check_connect(obd);
 905         if (rc)
 906                 RETURN(rc);
 907
 908         if (!lmv->desc.ld_active_tgt_count)
 909                 RETURN(-EIO);
 910 repeat:
 911         LASSERT(++loop <= 2);
 912         obj = lmv_grab_obj(obd, &op_data->id1);
 913         if (obj) {
 914                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 915                                    op_data->name, op_data->namelen);
 916                 op_data->id1 = obj->objs[mds].id;
 917                 lmv_put_obj(obj);
 918         }
 919
 920         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 921                op_data->name, OLID4(&op_data->id1));
 922
 923         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 924                        op_data, data, datalen, mode, uid, gid, rdev,
 925                        request);
 926         if (rc == 0) {
 927                 if (*request == NULL)
 928                         RETURN(rc);
 929
 930                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 931                                       sizeof(*body));
 932                 if (body == NULL)
 933                         RETURN(-ENOMEM);
 934
 935                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 936         } else if (rc == -ERESTART) {
 937                 /*
 938                  * directory got splitted. time to update local object and
 939                  * repeat the request with proper MDS.
 940                  */
 941                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 942                 if (rc == 0) {
 943                         ptlrpc_req_finished(*request);
 944                         goto repeat;
 945                 }
 946         }
 947         RETURN(rc);
 948 }
 949
 950 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 951 {
 952         struct obd_device *obd = exp->exp_obd;
 953         struct lmv_obd *lmv = &obd->u.lmv;
 954         int rc;
 955         ENTRY;
 956
 957         rc = lmv_check_connect(obd);
 958         if (rc)
 959                 RETURN(rc);
 960
 961         /* FIXME: choose right MDC here */
 962         CWARN("this method isn't implemented yet\n");
 963         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 964         RETURN(rc);
 965 }
 966
 967 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
 968                        struct lookup_intent *it, int lockmode,
 969                        struct mdc_op_data *data, struct lustre_handle *lockh,
 970                        void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
 971                        ldlm_blocking_callback cb_blocking, void *cb_data)
 972 {
 973         struct obd_device *obd = exp->exp_obd;
 974         struct lmv_obd *lmv = &obd->u.lmv;
 975         struct mea *mea = data->mea1;
 976         struct mdc_op_data *data2;
 977         int i, rc, mds;
 978         ENTRY;
 979
 980         OBD_ALLOC(data2, sizeof(*data2));
 981         if (data2 == NULL)
 982                 RETURN(-ENOMEM);
 983
 984         LASSERT(mea != NULL);
 985         for (i = 0; i < mea->mea_count; i++) {
 986                 memset(data2, 0, sizeof(*data2));
 987                 data2->id1 = mea->mea_ids[i];
 988                 mds = id_group(&data2->id1);
 989
 990                 if (lmv->tgts[mds].ltd_exp == NULL)
 991                         continue;
 992
 993                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
 994                                 lockmode, data2, lockh + i, lmm, lmmsize,
 995                                 cb_compl, cb_blocking, cb_data);
 996
 997                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
 998                        OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
 999                 if (rc)
1000                         GOTO(cleanup, rc);
1001                 if (it->d.lustre.it_data) {
1002                         struct ptlrpc_request *req;
1003                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1004                         ptlrpc_req_finished(req);
1005                 }
1006
1007                 if (it->d.lustre.it_status)
1008                         GOTO(cleanup, rc = it->d.lustre.it_status);
1009         }
1010
1011         OBD_FREE(data2, sizeof(*data2));
1012         RETURN(0);
1013 cleanup:
1014         OBD_FREE(data2, sizeof(*data2));
1015
1016         /* drop all taken locks */
1017         while (--i >= 0) {
1018                 if (lockh[i].cookie)
1019                         ldlm_lock_decref(lockh + i, lockmode);
1020                 lockh[i].cookie = 0;
1021         }
1022         return rc;
1023 }
1024
1025 int lmv_enqueue(struct obd_export *exp, int lock_type,
1026                 struct lookup_intent *it, int lock_mode,
1027                 struct mdc_op_data *data, struct lustre_handle *lockh,
1028                 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1029                 ldlm_blocking_callback cb_blocking, void *cb_data)
1030 {
1031         struct obd_device *obd = exp->exp_obd;
1032         struct lmv_obd *lmv = &obd->u.lmv;
1033         struct lmv_obj *obj;
1034         int rc, mds;
1035         ENTRY;
1036
1037         rc = lmv_check_connect(obd);
1038         if (rc)
1039                 RETURN(rc);
1040
1041         if (data->mea1 && it->it_op == IT_UNLINK) {
1042                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1043                                         data, lockh, lmm, lmmsize,
1044                                         cb_compl, cb_blocking, cb_data);
1045                 RETURN(rc);
1046         }
1047
1048         if (data->namelen) {
1049                 obj = lmv_grab_obj(obd, &data->id1);
1050                 if (obj) {
1051                         /* directory is splitted. look for right mds for this
1052                          * name */
1053                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1054                                            (char *)data->name, data->namelen);
1055                         data->id1 = obj->objs[mds].id;
1056                         lmv_put_obj(obj);
1057                 }
1058         }
1059         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1060                OLID4(&data->id1));
1061
1062         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1063                         lock_type, it, lock_mode, data, lockh, lmm,
1064                         lmmsize, cb_compl, cb_blocking, cb_data);
1065         RETURN(rc);
1066 }
1067
1068 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1069                      char *filename, int namelen, __u64 valid,
1070                      unsigned int ea_size, struct ptlrpc_request **request)
1071 {
1072         int rc, mds = id_group(id), loop = 0;
1073         struct obd_device *obd = exp->exp_obd;
1074         struct lmv_obd *lmv = &obd->u.lmv;
1075         struct lustre_id rid = *id;
1076         struct mds_body *body;
1077         struct lmv_obj *obj;
1078         __u64 old_valid;
1079         ENTRY;
1080
1081         rc = lmv_check_connect(obd);
1082         if (rc)
1083                 RETURN(rc);
1084 repeat:
1085         LASSERT(++loop <= 2);
1086         obj = lmv_grab_obj(obd, id);
1087         if (obj) {
1088                 /* directory is splitted. look for right mds for this name */
1089                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1090                                    filename, namelen - 1);
1091                 rid = obj->objs[mds].id;
1092                 lmv_put_obj(obj);
1093         }
1094
1095         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1096                namelen, filename, OLID4(id), OLID4(&rid));
1097
1098         old_valid = valid;
1099
1100         /*
1101          * here should be applied OBD_MD_FID to ->valid, because otherwise,
1102          * mds_getattr_lock() will not fetch fid component of lustre_id and
1103          * thus, next call to md_getattr_lock() will be performed to wrong mds.
1104          */
1105         if (!(old_valid & OBD_MD_FID))
1106                 valid |= OBD_MD_FID;
1107
1108         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1109                              &rid, filename, namelen, valid,
1110                              ea_size, request);
1111         if (rc == 0) {
1112                 /*
1113                  * this could be cross-node reference. in this case all we have
1114                  * right now is lustre_id triple. we'd like to find other
1115                  * attributes.
1116                  */
1117                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1118                 LASSERT(body != NULL);
1119                 if (body->valid & OBD_MD_MDS) {
1120                         struct ptlrpc_request *req = NULL;
1121
1122                         rid = body->id1;
1123                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1124
1125                         /*
1126                          * turning OBD_MD_FID fetching off, as we already have
1127                          * full lustre_id and do need to fetch fid component
1128                          * again. This will help to make thing slightly faster.
1129                          */
1130                         if (!(old_valid & OBD_MD_FID))
1131                                 valid &= ~OBD_MD_FID;
1132
1133                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1134                                              &rid, NULL, 1, valid, ea_size, &req);
1135                         ptlrpc_req_finished(*request);
1136                         *request = req;
1137                 }
1138         } else if (rc == -ERESTART) {
1139                 /* directory got splitted. time to update local object and
1140                  * repeat the request with proper MDS */
1141                 rc = lmv_get_mea_and_update_object(exp, &rid);
1142                 if (rc == 0) {
1143                         ptlrpc_req_finished(*request);
1144                         goto repeat;
1145                 }
1146         }
1147         RETURN(rc);
1148 }
1149
1150 /*
1151  * llite passes id of an target inode in data->id1 and id of directory in
1152  * data->id2
1153  */
1154 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1155              struct ptlrpc_request **request)
1156 {
1157         struct obd_device *obd = exp->exp_obd;
1158         struct lmv_obd *lmv = &obd->u.lmv;
1159         struct lmv_obj *obj;
1160         int rc;
1161         ENTRY;
1162
1163         rc = lmv_check_connect(obd);
1164         if (rc)
1165                 RETURN(rc);
1166
1167         if (data->namelen != 0) {
1168                 /* usual link request */
1169                 obj = lmv_grab_obj(obd, &data->id1);
1170                 if (obj) {
1171                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1172                                           data->name, data->namelen);
1173                         data->id1 = obj->objs[rc].id;
1174                         lmv_put_obj(obj);
1175                 }
1176
1177                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1178                        OLID4(&data->id2), data->namelen, data->name,
1179                        OLID4(&data->id1));
1180         } else {
1181                 /* request from MDS to acquire i_links for inode by id1 */
1182                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1183                        OLID4(&data->id1));
1184         }
1185
1186         rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1187                      data, request);
1188         RETURN(rc);
1189 }
1190
1191 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1192                const char *old, int oldlen, const char *new, int newlen,
1193                struct ptlrpc_request **request)
1194 {
1195         struct obd_device *obd = exp->exp_obd;
1196         struct lmv_obd *lmv = &obd->u.lmv;
1197         struct lmv_obj *obj;
1198         int rc, mds;
1199         ENTRY;
1200
1201         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1202                oldlen, old, OLID4(&data->id1), newlen, new,
1203                OLID4(&data->id2));
1204
1205         rc = lmv_check_connect(obd);
1206         if (rc)
1207                 RETURN(rc);
1208
1209         if (oldlen == 0) {
1210                 /*
1211                  * MDS with old dir entry is asking another MDS to create name
1212                  * there.
1213                  */
1214                 CDEBUG(D_OTHER,
1215                        "create %*s(%d/%d) in "DLID4" pointing "
1216                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1217                        OLID4(&data->id2), OLID4(&data->id1));
1218
1219                 mds = id_group(&data->id2);
1220
1221                 /*
1222                  * target directory can be splitted, sowe should forward request
1223                  * to the right MDS.
1224                  */
1225                 obj = lmv_grab_obj(obd, &data->id2);
1226                 if (obj) {
1227                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1228                                            (char *)new, newlen);
1229                         data->id2 = obj->objs[mds].id;
1230                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1231                                OLID4(&data->id2));
1232                         lmv_put_obj(obj);
1233                 }
1234                 goto request;
1235         }
1236
1237         obj = lmv_grab_obj(obd, &data->id1);
1238         if (obj) {
1239                 /*
1240                  * directory is already splitted, so we have to forward request
1241                  * to the right MDS.
1242                  */
1243                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1244                                    (char *)old, oldlen);
1245                 data->id1 = obj->objs[mds].id;
1246                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1247                        OLID4(&data->id1));
1248                 lmv_put_obj(obj);
1249         }
1250
1251         obj = lmv_grab_obj(obd, &data->id2);
1252         if (obj) {
1253                 /*
1254                  * directory is already splitted, so we have to forward request
1255                  * to the right MDS.
1256                  */
1257                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1258                                    (char *)new, newlen);
1259
1260                 data->id2 = obj->objs[mds].id;
1261                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1262                        OLID4(&data->id2));
1263                 lmv_put_obj(obj);
1264         }
1265
1266         mds = id_group(&data->id1);
1267
1268 request:
1269         if (id_group(&data->id1) != id_group(&data->id2)) {
1270                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1271                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1272                        newlen, new);
1273         }
1274
1275         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1276                        new, newlen, request);
1277         RETURN(rc);
1278 }
1279
1280 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1281                 struct iattr *iattr, void *ea, int ealen, void *ea2,
1282                 int ea2len, struct ptlrpc_request **request)
1283 {
1284         struct obd_device *obd = exp->exp_obd;
1285         struct lmv_obd *lmv = &obd->u.lmv;
1286         struct ptlrpc_request *req;
1287         struct mds_body *body;
1288         struct lmv_obj *obj;
1289         int rc = 0, i;
1290         ENTRY;
1291
1292         rc = lmv_check_connect(obd);
1293         if (rc)
1294                 RETURN(rc);
1295
1296         obj = lmv_grab_obj(obd, &data->id1);
1297
1298         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1299                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1300
1301         if (obj) {
1302                 for (i = 0; i < obj->objcount; i++) {
1303                         data->id1 = obj->objs[i].id;
1304
1305                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1306                                         data, iattr, ea, ealen, ea2, ea2len, &req);
1307
1308                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1309                                 /*
1310                                  * this is master object and this request should
1311                                  * be returned back to llite.
1312                                  */
1313                                 *request = req;
1314                         } else {
1315                                 ptlrpc_req_finished(req);
1316                         }
1317
1318                         if (rc)
1319                                 break;
1320                 }
1321                 lmv_put_obj(obj);
1322         } else {
1323                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1324                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1325                                 data, iattr, ea, ealen, ea2, ea2len, request);
1326                 if (rc == 0) {
1327                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1328                                               sizeof(*body));
1329                         LASSERT(body != NULL);
1330                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1331                 }
1332         }
1333         RETURN(rc);
1334 }
1335
1336 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1337              struct ptlrpc_request **request)
1338 {
1339         struct obd_device *obd = exp->exp_obd;
1340         struct lmv_obd *lmv = &obd->u.lmv;
1341         int rc;
1342         ENTRY;
1343
1344         rc = lmv_check_connect(obd);
1345         if (rc)
1346                 RETURN(rc);
1347
1348         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1349                      id, request);
1350         RETURN(rc);
1351 }
1352
1353 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1354                             struct ldlm_lock_desc *desc,
1355                             void *data, int flag)
1356 {
1357         struct lustre_handle lockh;
1358         struct lmv_obj *obj;
1359         int rc;
1360         ENTRY;
1361
1362         switch (flag) {
1363         case LDLM_CB_BLOCKING:
1364                 ldlm_lock2handle(lock, &lockh);
1365                 rc = ldlm_cli_cancel(&lockh);
1366                 if (rc < 0) {
1367                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1368                         RETURN(rc);
1369                 }
1370                 break;
1371         case LDLM_CB_CANCELING:
1372                 /* time to drop cached attrs for dirobj */
1373                 obj = lock->l_ast_data;
1374                 if (obj) {
1375                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1376                                ", master "DLID4"\n",
1377                                lock->l_resource->lr_name.name[3] == 1 ?
1378                                "LOOKUP" : "UPDATE",
1379                                lock->l_resource->lr_name.name[0],
1380                                lock->l_resource->lr_name.name[1],
1381                                OLID4(&obj->id));
1382                         lmv_put_obj(obj);
1383                 }
1384                 break;
1385         default:
1386                 LBUG();
1387         }
1388         RETURN(0);
1389 }
1390
1391 void lmv_remove_dots(struct page *page)
1392 {
1393         unsigned limit = PAGE_CACHE_SIZE;
1394         char *kaddr = page_address(page);
1395         struct ext2_dir_entry_2 *p;
1396         unsigned offs, rec_len;
1397
1398         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1399                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1400                 rec_len = le16_to_cpu(p->rec_len);
1401
1402                 if ((p->name_len == 1 && p->name[0] == '.') ||
1403                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1404                         p->inode = 0;
1405         }
1406 }
1407
1408 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1409                  __u64 offset, struct page *page,
1410                  struct ptlrpc_request **request)
1411 {
1412         struct obd_device *obd = exp->exp_obd;
1413         struct lmv_obd *lmv = &obd->u.lmv;
1414         struct lustre_id rid = *id;
1415         struct lmv_obj *obj;
1416         int rc, i;
1417         ENTRY;
1418
1419 #warning "we need well-desgined readdir() implementation"
1420         rc = lmv_check_connect(obd);
1421         if (rc)
1422                 RETURN(rc);
1423
1424         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1425         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1426                offset, OLID4(&rid));
1427
1428         obj = lmv_grab_obj(obd, id);
1429         if (obj) {
1430                 lmv_lock_obj(obj);
1431
1432                 /* find dirobj containing page with requested offset. */
1433                 for (i = 0; i < obj->objcount; i++) {
1434                         if (offset < obj->objs[i].size)
1435                                 break;
1436                         offset -= obj->objs[i].size;
1437                 }
1438                 rid = obj->objs[i].id;
1439
1440                 lmv_unlock_obj(obj);
1441                 lmv_put_obj(obj);
1442
1443                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1444                        OLID4(&rid), (unsigned long)offset);
1445         }
1446         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1447                          offset, page, request);
1448
1449         if (rc == 0 && !id_equal_fid(&rid, id))
1450                 /* this page isn't from master object. To avoid "." and ".."
1451                  * duplication in directory, we have to remove them from all
1452                  * slave objects */
1453                 lmv_remove_dots(page);
1454
1455         RETURN(rc);
1456 }
1457
1458 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1459                       struct ptlrpc_request **req)
1460 {
1461         struct obd_device *obd = exp->exp_obd;
1462         struct lmv_obd *lmv = &obd->u.lmv;
1463         struct mea *mea = data->mea1;
1464         struct mdc_op_data *data2;
1465         int i, rc = 0;
1466         ENTRY;
1467
1468         OBD_ALLOC(data2, sizeof(*data2));
1469         if (data2 == NULL)
1470                 RETURN(-ENOMEM);
1471
1472         LASSERT(mea != NULL);
1473         for (i = 0; i < mea->mea_count; i++) {
1474                 memset(data2, 0, sizeof(*data2));
1475                 data2->id1 = mea->mea_ids[i];
1476                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1477
1478                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1479                         continue;
1480
1481                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1482                                data2, req);
1483
1484                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1485                        OLID4(&mea->mea_ids[i]), rc);
1486
1487                 if (*req) {
1488                         ptlrpc_req_finished(*req);
1489                         *req = NULL;
1490                 }
1491                 if (rc)
1492                         RETURN(rc);
1493         }
1494         OBD_FREE(data2, sizeof(*data2));
1495         RETURN(rc);
1496 }
1497
1498 int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1499 {
1500         ENTRY;
1501
1502         LASSERT(exp && id);
1503         if (lmv_delete_obj(exp, id)) {
1504                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1505                        OLID4(id));
1506         }
1507         RETURN(0);
1508 }
1509
1510 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1511                struct ptlrpc_request **request)
1512 {
1513         struct obd_device *obd = exp->exp_obd;
1514         struct lmv_obd *lmv = &obd->u.lmv;
1515         int rc, i = 0;
1516         ENTRY;
1517
1518         rc = lmv_check_connect(obd);
1519         if (rc)
1520                 RETURN(rc);
1521
1522         if (data->namelen == 0 && data->mea1 != NULL) {
1523                 /* mds asks to remove slave objects */
1524                 rc = lmv_unlink_slaves(exp, data, request);
1525                 RETURN(rc);
1526         }
1527
1528         if (data->namelen != 0) {
1529                 struct lmv_obj *obj;
1530
1531                 obj = lmv_grab_obj(obd, &data->id1);
1532                 if (obj) {
1533                         i = raw_name2idx(obj->hashtype, obj->objcount,
1534                                          data->name, data->namelen);
1535                         data->id1 = obj->objs[i].id;
1536                         lmv_put_obj(obj);
1537                 }
1538                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1539                        data->namelen, data->name, OLID4(&data->id1),
1540                        i);
1541         } else {
1542                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1543                        OLID4(&data->id1));
1544         }
1545         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1546                        data, request);
1547         RETURN(rc);
1548 }
1549
1550 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1551                                     char *name, int len)
1552 {
1553         struct obd_device *obd = exp->exp_obd;
1554         struct lmv_obd *lmv = &obd->u.lmv;
1555         int rc;
1556         ENTRY;
1557
1558         rc = lmv_check_connect(obd);
1559         if (rc)
1560                 RETURN(ERR_PTR(rc));
1561         obd = lmv->tgts[0].ltd_exp->exp_obd;
1562         EXIT;
1563
1564         return obd;
1565 }
1566
1567 int lmv_init_ea_size(struct obd_export *exp, int easize,
1568                      int cookiesize)
1569 {
1570         struct obd_device *obd = exp->exp_obd;
1571         struct lmv_obd *lmv = &obd->u.lmv;
1572         int i, rc = 0, change = 0;
1573         ENTRY;
1574
1575         if (lmv->max_easize < easize) {
1576                 lmv->max_easize = easize;
1577                 change = 1;
1578         }
1579         if (lmv->max_cookiesize < cookiesize) {
1580                 lmv->max_cookiesize = cookiesize;
1581                 change = 1;
1582         }
1583         if (change == 0)
1584                 RETURN(0);
1585
1586         if (lmv->connected == 0)
1587                 RETURN(0);
1588
1589         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1590                 if (lmv->tgts[i].ltd_exp == NULL) {
1591                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1592                         continue;
1593                 }
1594
1595                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1596                 if (rc) {
1597                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1598                                "error %d.\n", i, rc);
1599                         break;
1600                 }
1601         }
1602         RETURN(rc);
1603 }
1604
1605 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1606                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
1607 {
1608         struct obd_device *obd = exp->exp_obd;
1609         struct lmv_obd *lmv = &obd->u.lmv;
1610         struct lov_stripe_md obj_md;
1611         struct lov_stripe_md *obj_mdp = &obj_md;
1612         int rc = 0;
1613         ENTRY;
1614
1615         LASSERT(ea == NULL);
1616         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1617
1618         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
1619                         oa, &obj_mdp, oti);
1620
1621         RETURN(rc);
1622 }
1623
1624 int lmv_getready(struct obd_export *exp)
1625 {
1626         struct obd_device *obd = exp->exp_obd;
1627         int rc = 0;
1628
1629         ENTRY;
1630         rc = lmv_check_connect(obd);
1631         RETURN(rc);
1632 }
1633
1634 /*
1635  * to be called from MDS only. @oa should have correct store cookie and o_fid
1636  * values for "master" object, as it will be used.
1637  */
1638 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1639                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1640 {
1641         struct obd_device *obd = exp->exp_obd;
1642         struct lmv_obd *lmv = &obd->u.lmv;
1643         struct lustre_id mid;
1644         int i, c, rc = 0;
1645         struct mea *mea;
1646         ENTRY;
1647
1648         rc = lmv_check_connect(obd);
1649         if (rc)
1650                 RETURN(rc);
1651
1652         LASSERT(oa != NULL);
1653
1654         if (ea == NULL) {
1655                 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1656                 if (rc)
1657                         CERROR("Can't create object, rc = %d\n", rc);
1658                 RETURN(rc);
1659         }
1660
1661         if (*ea == NULL) {
1662                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1663                 if (rc < 0) {
1664                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1665                                rc);
1666                         RETURN(rc);
1667                 } else
1668                         rc = 0;
1669
1670                 if (*ea == NULL)
1671                         RETURN(-ENOMEM);
1672         }
1673
1674         /*
1675          * here we should take care about splitted dir, so store cookie and fid
1676          * for "master" object should already be allocated and passed in @oa.
1677          */
1678         LASSERT(oa->o_id != 0);
1679         LASSERT(oa->o_fid != 0);
1680
1681         /* save "master" object id */
1682         obdo2id(&mid, oa);
1683
1684         mea = (struct mea *)*ea;
1685         mea->mea_master = -1;
1686         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1687
1688         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1689                 mea->mea_count = lmv->desc.ld_tgt_count;
1690
1691         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1692                 struct lov_stripe_md obj_md;
1693                 struct lov_stripe_md *obj_mdp = &obj_md;
1694
1695                 if (lmv->tgts[i].ltd_exp == NULL) {
1696                         /* this is "master" MDS */
1697                         mea->mea_master = i;
1698                         mea->mea_ids[c] = mid;
1699                         c++;
1700                         continue;
1701                 }
1702
1703                 /*
1704                  * "master" MDS should always be part of stripped dir,
1705                  * so scan for it.
1706                  */
1707                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1708                         continue;
1709
1710                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1711                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1712
1713                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1714                 if (rc) {
1715                         CERROR("obd_create() failed on MDT target %d, "
1716                                "error %d\n", c, rc);
1717                         RETURN(rc);
1718                 }
1719
1720                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1721                        i, oa->o_id, oa->o_generation);
1722
1723
1724                 /*
1725                  * here, when object is created (or it is master and was passed
1726                  * from caller) on desired MDS we save its fid to local mea_ids.
1727                  */
1728                 LASSERT(oa->o_fid);
1729
1730                 /*
1731                  * store cookie should be defined here for both cases (master
1732                  * object and not master), because master is already created.
1733                  */
1734                 LASSERT(oa->o_id);
1735
1736                 /* fill mea by store cookie and fid */
1737                 obdo2id(&mea->mea_ids[c], oa);
1738                 c++;
1739         }
1740         LASSERT(c == mea->mea_count);
1741
1742         CDEBUG(D_OTHER, "%d dirobjects created\n",
1743                (int)mea->mea_count);
1744
1745         RETURN(rc);
1746 }
1747
1748 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1749                          struct obd_device *tgt, int count,
1750                          struct llog_catid *logid)
1751 {
1752         struct llog_ctxt *ctxt;
1753         int rc;
1754         ENTRY;
1755
1756         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1757                             &llog_client_ops);
1758         if (rc == 0) {
1759                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1760                 ctxt->loc_imp = tgt->u.cli.cl_import;
1761         }
1762
1763         RETURN(rc);
1764 }
1765
1766 static int lmv_llog_finish(struct obd_device *obd,
1767                            struct obd_llogs *llogs, int count)
1768 {
1769         int rc;
1770         ENTRY;
1771
1772         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1773         RETURN(rc);
1774 }
1775
1776 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1777                         void *key, __u32 *vallen, void *val)
1778 {
1779         struct obd_device *obd;
1780         struct lmv_obd *lmv;
1781         int rc = 0;
1782         ENTRY;
1783
1784         obd = class_exp2obd(exp);
1785         if (obd == NULL) {
1786                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1787                        exp->exp_handle.h_cookie);
1788                 RETURN(-EINVAL);
1789         }
1790
1791         lmv = &obd->u.lmv;
1792         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1793                 __u32 *mdsize = val;
1794                 *vallen = sizeof(__u32);
1795                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1796                         + sizeof(struct mea);
1797                 RETURN(0);
1798         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1799                 struct obd_uuid *cluuid = &lmv->cluuid;
1800                 struct lmv_tgt_desc *tgts;
1801                 __u32 *mdsnum = val;
1802                 int i;
1803
1804                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1805                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1806                                 *vallen = sizeof(__u32);
1807                                 *mdsnum = i;
1808                                 RETURN(0);
1809                         }
1810                 }
1811                 LASSERT(0);
1812         } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1813                 /* getting rootid from first MDS. */
1814                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1815                                   vallen, val);
1816                 RETURN(rc);
1817         } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1818                 struct lmv_desc *desc_ret = val;
1819                 *desc_ret = lmv->desc;
1820                 RETURN(0);
1821         }
1822
1823         CDEBUG(D_IOCTL, "invalid key\n");
1824         RETURN(-EINVAL);
1825 }
1826
1827 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1828                  void *key, obd_count vallen, void *val)
1829 {
1830         struct obd_device *obd;
1831         struct lmv_obd *lmv;
1832         ENTRY;
1833
1834         obd = class_exp2obd(exp);
1835         if (obd == NULL) {
1836                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1837                        exp->exp_handle.h_cookie);
1838                 RETURN(-EINVAL);
1839         }
1840         lmv = &obd->u.lmv;
1841
1842         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1843                 lmv->server_timeout = 1;
1844                 lmv_set_timeouts(obd);
1845                 RETURN(0);
1846         }
1847
1848         RETURN(-EINVAL);
1849 }
1850
1851 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1852                struct lov_stripe_md *lsm)
1853 {
1854         struct obd_device *obd = class_exp2obd(exp);
1855         struct lmv_obd *lmv = &obd->u.lmv;
1856         struct mea *meap, *lsmp;
1857         int mea_size, i;
1858         ENTRY;
1859
1860         mea_size = (sizeof(struct lustre_id) *
1861                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
1862         if (!lmmp)
1863                 RETURN(mea_size);
1864
1865         if (*lmmp && !lsm) {
1866                 OBD_FREE(*lmmp, mea_size);
1867                 *lmmp = NULL;
1868                 RETURN(0);
1869         }
1870
1871         if (*lmmp == NULL) {
1872                 OBD_ALLOC(*lmmp, mea_size);
1873                 if (*lmmp == NULL)
1874                         RETURN(-ENOMEM);
1875         }
1876
1877         if (!lsm)
1878                 RETURN(mea_size);
1879
1880         lsmp = (struct mea *)lsm;
1881         meap = (struct mea *)*lmmp;
1882
1883         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
1884         meap->mea_count = cpu_to_le32(lsmp->mea_count);
1885         meap->mea_master = cpu_to_le32(lsmp->mea_master);
1886
1887         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1888                 meap->mea_ids[i] = meap->mea_ids[i];
1889                 id_cpu_to_le(&meap->mea_ids[i]);
1890         }
1891
1892         RETURN(mea_size);
1893 }
1894
1895 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1896                  struct lov_mds_md *disk_src, int mdsize)
1897 {
1898         struct obd_device *obd = class_exp2obd(exp);
1899         struct mea **tmea = (struct mea **)mem_tgt;
1900         struct mea *mea = (struct mea *)disk_src;
1901         struct lmv_obd *lmv = &obd->u.lmv;
1902         int mea_size, i;
1903         ENTRY;
1904
1905         mea_size = sizeof(struct lustre_id) *
1906                 lmv->desc.ld_tgt_count + sizeof(struct mea);
1907         if (mem_tgt == NULL)
1908                 return mea_size;
1909
1910         if (*mem_tgt != NULL && disk_src == NULL) {
1911                 OBD_FREE(*tmea, mea_size);
1912                 RETURN(0);
1913         }
1914
1915         LASSERT(mea_size == mdsize);
1916
1917         OBD_ALLOC(*tmea, mea_size);
1918         if (*tmea == NULL)
1919                 RETURN(-ENOMEM);
1920
1921         if (!disk_src)
1922                 RETURN(mea_size);
1923
1924         (*tmea)->mea_magic = le32_to_cpu(mea->mea_magic);
1925         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
1926         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
1927
1928         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1929                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
1930                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
1931         }
1932
1933         RETURN(mea_size);
1934 }
1935
1936 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1937             struct lov_stripe_md *ea, obd_count oa_bufs,
1938             struct brw_page *pgarr, struct obd_trans_info *oti)
1939 {
1940         struct obd_device *obd = exp->exp_obd;
1941         struct lmv_obd *lmv = &obd->u.lmv;
1942         struct mea *mea = (struct mea *) ea;
1943         int err;
1944
1945         LASSERT(oa != NULL);
1946         LASSERT(ea != NULL);
1947         LASSERT(pgarr != NULL);
1948         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1949
1950         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
1951         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
1952         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1953
1954         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
1955                       oa, NULL, oa_bufs, pgarr, oti);
1956         RETURN(err);
1957 }
1958
1959 struct obd_ops lmv_obd_ops = {
1960         .o_owner                = THIS_MODULE,
1961         .o_attach               = lmv_attach,
1962         .o_detach               = lmv_detach,
1963         .o_setup                = lmv_setup,
1964         .o_cleanup              = lmv_cleanup,
1965         .o_connect              = lmv_connect,
1966         .o_disconnect           = lmv_disconnect,
1967         .o_statfs               = lmv_statfs,
1968         .o_llog_init            = lmv_llog_init,
1969         .o_llog_finish          = lmv_llog_finish,
1970         .o_get_info             = lmv_get_info,
1971         .o_set_info             = lmv_set_info,
1972         .o_create               = lmv_obd_create,
1973         .o_packmd               = lmv_packmd,
1974         .o_unpackmd             = lmv_unpackmd,
1975         .o_brw                  = lmv_brw,
1976         .o_init_ea_size         = lmv_init_ea_size,
1977         .o_notify               = lmv_notify,
1978         .o_iocontrol            = lmv_iocontrol,
1979         .o_getready             = lmv_getready,
1980 };
1981
1982 struct md_ops lmv_md_ops = {
1983         .m_getstatus           = lmv_getstatus,
1984         .m_getattr             = lmv_getattr,
1985         .m_change_cbdata       = lmv_change_cbdata,
1986         .m_change_cbdata_name  = lmv_change_cbdata_name,
1987         .m_close               = lmv_close,
1988         .m_create              = lmv_create,
1989         .m_done_writing        = lmv_done_writing,
1990         .m_enqueue             = lmv_enqueue,
1991         .m_getattr_lock        = lmv_getattr_lock,
1992         .m_intent_lock         = lmv_intent_lock,
1993         .m_link                = lmv_link,
1994         .m_rename              = lmv_rename,
1995         .m_setattr             = lmv_setattr,
1996         .m_sync                = lmv_sync,
1997         .m_readpage            = lmv_readpage,
1998         .m_unlink              = lmv_unlink,
1999         .m_get_real_obd        = lmv_get_real_obd,
2000         .m_valid_attrs         = lmv_valid_attrs,
2001         .m_delete_inode        = lmv_delete_inode,
2002 };
2003
2004 int __init lmv_init(void)
2005 {
2006         struct lprocfs_static_vars lvars;
2007         int rc;
2008
2009         obj_cache = kmem_cache_create("lmv_objects",
2010                                       sizeof(struct lmv_obj),
2011                                       0, 0, NULL, NULL);
2012         if (!obj_cache) {
2013                 CERROR("error allocating lmv objects cache\n");
2014                 return -ENOMEM;
2015         }
2016
2017         lprocfs_init_vars(lmv, &lvars);
2018         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2019                                  lvars.module_vars,
2020                                  OBD_LMV_DEVICENAME);
2021         if (rc)
2022                 kmem_cache_destroy(obj_cache);
2023
2024         return rc;
2025 }
2026
2027 #ifdef __KERNEL__
2028 static void lmv_exit(void)
2029 {
2030         class_unregister_type(OBD_LMV_DEVICENAME);
2031
2032         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2033                  "can't free lmv objects cache, %d object(s)"
2034                  "still in use\n", atomic_read(&obj_cache_count));
2035 }
2036
2037 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2038 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2039 MODULE_LICENSE("GPL");
2040
2041 module_init(lmv_init);
2042 module_exit(lmv_exit);
2043 #endif