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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LMV
  41 #ifdef __KERNEL__
  42 #include <linux/slab.h>
  43 #include <linux/module.h>
  44 #include <linux/init.h>
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <asm/div64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51 #else
  52 #include <liblustre.h>
  53 #endif
  54
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_log.h>
  57 #include <obd_support.h>
  58 #include <lustre_lib.h>
  59 #include <lustre_net.h>
  60 #include <obd_class.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_lite.h>
  63 #include <lustre_fid.h>
  64 #include "lmv_internal.h"
  65
  66 /* object cache. */
  67 cfs_mem_cache_t *lmv_object_cache;
  68 cfs_atomic_t lmv_object_count = CFS_ATOMIC_INIT(0);
  69
  70 static void lmv_activate_target(struct lmv_obd *lmv,
  71                                 struct lmv_tgt_desc *tgt,
  72                                 int activate)
  73 {
  74         if (tgt->ltd_active == activate)
  75                 return;
  76
  77         tgt->ltd_active = activate;
  78         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  79 }
  80
  81 /**
  82  * Error codes:
  83  *
  84  *  -EINVAL  : UUID can't be found in the LMV's target list
  85  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  86  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  87  */
  88 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  89                               int activate)
  90 {
  91         struct lmv_tgt_desc    *tgt;
  92         struct obd_device      *obd;
  93         int                     i;
  94         int                     rc = 0;
  95         ENTRY;
  96
  97         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  98                lmv, uuid->uuid, activate);
  99
 100         cfs_spin_lock(&lmv->lmv_lock);
 101         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 102                 if (tgt->ltd_exp == NULL)
 103                         continue;
 104
 105                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n",
 106                        i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 107
 108                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 109                         break;
 110         }
 111
 112         if (i == lmv->desc.ld_tgt_count)
 113                 GOTO(out_lmv_lock, rc = -EINVAL);
 114
 115         obd = class_exp2obd(tgt->ltd_exp);
 116         if (obd == NULL)
 117                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 118
 119         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 120                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 121                obd->obd_type->typ_name, i);
 122         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 123
 124         if (tgt->ltd_active == activate) {
 125                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 126                        activate ? "" : "in");
 127                 GOTO(out_lmv_lock, rc);
 128         }
 129
 130         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 131                activate ? "" : "in");
 132         lmv_activate_target(lmv, tgt, activate);
 133         EXIT;
 134
 135  out_lmv_lock:
 136         cfs_spin_unlock(&lmv->lmv_lock);
 137         return rc;
 138 }
 139
 140 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
 141                             struct obd_connect_data *data)
 142 {
 143         struct lmv_tgt_desc    *tgt;
 144         int                     i;
 145         ENTRY;
 146
 147         LASSERT(data != NULL);
 148
 149         cfs_spin_lock(&lmv->lmv_lock);
 150         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 151                 if (tgt->ltd_exp == NULL)
 152                         continue;
 153
 154                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 155                         lmv->datas[tgt->ltd_idx] = *data;
 156                         break;
 157                 }
 158         }
 159         cfs_spin_unlock(&lmv->lmv_lock);
 160         RETURN(0);
 161 }
 162
 163 struct obd_uuid *lmv_get_uuid(struct obd_export *exp) {
 164         struct obd_device *obd = exp->exp_obd;
 165         struct lmv_obd *lmv = &obd->u.lmv;
 166         return obd_get_uuid(lmv->tgts[0].ltd_exp);
 167 }
 168
 169 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 170                       enum obd_notify_event ev, void *data)
 171 {
 172         struct obd_connect_data *conn_data;
 173         struct lmv_obd          *lmv = &obd->u.lmv;
 174         struct obd_uuid         *uuid;
 175         int                      rc = 0;
 176         ENTRY;
 177
 178         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 179                 CERROR("unexpected notification of %s %s!\n",
 180                        watched->obd_type->typ_name,
 181                        watched->obd_name);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         uuid = &watched->u.cli.cl_target_uuid;
 186         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 187                 /*
 188                  * Set MDC as active before notifying the observer, so the
 189                  * observer can use the MDC normally.
 190                  */
 191                 rc = lmv_set_mdc_active(lmv, uuid,
 192                                         ev == OBD_NOTIFY_ACTIVE);
 193                 if (rc) {
 194                         CERROR("%sactivation of %s failed: %d\n",
 195                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 196                                uuid->uuid, rc);
 197                         RETURN(rc);
 198                 }
 199         } else if (ev == OBD_NOTIFY_OCD) {
 200                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 201
 202                 /*
 203                  * Set connect data to desired target, update exp_connect_flags.
 204                  */
 205                 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
 206                 if (rc) {
 207                         CERROR("can't set connect data to target %s, rc %d\n",
 208                                uuid->uuid, rc);
 209                         RETURN(rc);
 210                 }
 211
 212                 /*
 213                  * XXX: Make sure that ocd_connect_flags from all targets are
 214                  * the same. Otherwise one of MDTs runs wrong version or
 215                  * something like this.  --umka
 216                  */
 217                 obd->obd_self_export->exp_connect_flags =
 218                         conn_data->ocd_connect_flags;
 219         }
 220 #if 0
 221         else if (ev == OBD_NOTIFY_DISCON) {
 222                 /*
 223                  * For disconnect event, flush fld cache for failout MDS case.
 224                  */
 225                 fld_client_flush(&lmv->lmv_fld);
 226         }
 227 #endif
 228         /*
 229          * Pass the notification up the chain.
 230          */
 231         if (obd->obd_observer)
 232                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 233
 234         RETURN(rc);
 235 }
 236
 237 /**
 238  * This is fake connect function. Its purpose is to initialize lmv and say
 239  * caller that everything is okay. Real connection will be performed later.
 240  */
 241 static int lmv_connect(const struct lu_env *env,
 242                        struct obd_export **exp, struct obd_device *obd,
 243                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 244                        void *localdata)
 245 {
 246 #ifdef __KERNEL__
 247         struct proc_dir_entry *lmv_proc_dir;
 248 #endif
 249         struct lmv_obd        *lmv = &obd->u.lmv;
 250         struct lustre_handle  conn = { 0 };
 251         int                    rc = 0;
 252         ENTRY;
 253
 254         /*
 255          * We don't want to actually do the underlying connections more than
 256          * once, so keep track.
 257          */
 258         lmv->refcount++;
 259         if (lmv->refcount > 1) {
 260                 *exp = NULL;
 261                 RETURN(0);
 262         }
 263
 264         rc = class_connect(&conn, obd, cluuid);
 265         if (rc) {
 266                 CERROR("class_connection() returned %d\n", rc);
 267                 RETURN(rc);
 268         }
 269
 270         *exp = class_conn2export(&conn);
 271         class_export_get(*exp);
 272
 273         lmv->exp = *exp;
 274         lmv->connected = 0;
 275         lmv->cluuid = *cluuid;
 276
 277         if (data)
 278                 lmv->conn_data = *data;
 279
 280 #ifdef __KERNEL__
 281         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 282                                         NULL, NULL);
 283         if (IS_ERR(lmv_proc_dir)) {
 284                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 285                        obd->obd_type->typ_name, obd->obd_name);
 286                 lmv_proc_dir = NULL;
 287         }
 288 #endif
 289
 290         /*
 291          * All real clients should perform actual connection right away, because
 292          * it is possible, that LMV will not have opportunity to connect targets
 293          * and MDC stuff will be called directly, for instance while reading
 294          * ../mdc/../kbytesfree procfs file, etc.
 295          */
 296         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 297                 rc = lmv_check_connect(obd);
 298
 299 #ifdef __KERNEL__
 300         if (rc) {
 301                 if (lmv_proc_dir)
 302                         lprocfs_remove(&lmv_proc_dir);
 303         }
 304 #endif
 305
 306         RETURN(rc);
 307 }
 308
 309 static void lmv_set_timeouts(struct obd_device *obd)
 310 {
 311         struct lmv_tgt_desc   *tgts;
 312         struct lmv_obd        *lmv;
 313         int                    i;
 314
 315         lmv = &obd->u.lmv;
 316         if (lmv->server_timeout == 0)
 317                 return;
 318
 319         if (lmv->connected == 0)
 320                 return;
 321
 322         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 323                 if (tgts->ltd_exp == NULL)
 324                         continue;
 325
 326                 obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS),
 327                                    KEY_INTERMDS, 0, NULL, NULL);
 328         }
 329 }
 330
 331 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 332                             int def_easize, int cookiesize)
 333 {
 334         struct obd_device   *obd = exp->exp_obd;
 335         struct lmv_obd      *lmv = &obd->u.lmv;
 336         int                  i;
 337         int                  rc = 0;
 338         int                  change = 0;
 339         ENTRY;
 340
 341         if (lmv->max_easize < easize) {
 342                 lmv->max_easize = easize;
 343                 change = 1;
 344         }
 345         if (lmv->max_def_easize < def_easize) {
 346                 lmv->max_def_easize = def_easize;
 347                 change = 1;
 348         }
 349         if (lmv->max_cookiesize < cookiesize) {
 350                 lmv->max_cookiesize = cookiesize;
 351                 change = 1;
 352         }
 353         if (change == 0)
 354                 RETURN(0);
 355
 356         if (lmv->connected == 0)
 357                 RETURN(0);
 358
 359         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 360                 if (lmv->tgts[i].ltd_exp == NULL) {
 361                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 362                         continue;
 363                 }
 364
 365                 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
 366                                      cookiesize);
 367                 if (rc) {
 368                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 369                                "error %d.\n", i, rc);
 370                         break;
 371                 }
 372         }
 373         RETURN(rc);
 374 }
 375
 376 #define MAX_STRING_SIZE 128
 377
 378 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 379 {
 380 #ifdef __KERNEL__
 381         struct proc_dir_entry   *lmv_proc_dir;
 382 #endif
 383         struct lmv_obd          *lmv = &obd->u.lmv;
 384         struct obd_uuid         *cluuid = &lmv->cluuid;
 385         struct obd_connect_data *mdc_data = NULL;
 386         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 387         struct obd_device       *mdc_obd;
 388         struct obd_export       *mdc_exp;
 389         struct lu_fld_target     target;
 390         int                      rc;
 391         ENTRY;
 392
 393         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 394                                         &obd->obd_uuid);
 395         if (!mdc_obd) {
 396                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 397                 RETURN(-EINVAL);
 398         }
 399
 400         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 401                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 402                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 403                 cluuid->uuid);
 404
 405         if (!mdc_obd->obd_set_up) {
 406                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 407                 RETURN(-EINVAL);
 408         }
 409
 410         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 411                          &lmv->conn_data, NULL);
 412         if (rc) {
 413                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 414                 RETURN(rc);
 415         }
 416
 417         /*
 418          * Init fid sequence client for this mdc and add new fld target.
 419          */
 420         rc = obd_fid_init(mdc_exp);
 421         if (rc)
 422                 RETURN(rc);
 423
 424         target.ft_srv = NULL;
 425         target.ft_exp = mdc_exp;
 426         target.ft_idx = tgt->ltd_idx;
 427
 428         fld_client_add_target(&lmv->lmv_fld, &target);
 429
 430         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 431
 432         rc = obd_register_observer(mdc_obd, obd);
 433         if (rc) {
 434                 obd_disconnect(mdc_exp);
 435                 CERROR("target %s register_observer error %d\n",
 436                        tgt->ltd_uuid.uuid, rc);
 437                 RETURN(rc);
 438         }
 439
 440         if (obd->obd_observer) {
 441                 /*
 442                  * Tell the observer about the new target.
 443                  */
 444                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 445                                 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
 446                 if (rc) {
 447                         obd_disconnect(mdc_exp);
 448                         RETURN(rc);
 449                 }
 450         }
 451
 452         tgt->ltd_active = 1;
 453         tgt->ltd_exp = mdc_exp;
 454         lmv->desc.ld_active_tgt_count++;
 455
 456         /*
 457          * Copy connect data, it may be used later.
 458          */
 459         lmv->datas[tgt->ltd_idx] = *mdc_data;
 460
 461         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 462                         lmv->max_def_easize, lmv->max_cookiesize);
 463
 464         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 465                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 466                 cfs_atomic_read(&obd->obd_refcount));
 467
 468 #ifdef __KERNEL__
 469         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 470         if (lmv_proc_dir) {
 471                 struct proc_dir_entry *mdc_symlink;
 472
 473                 LASSERT(mdc_obd->obd_type != NULL);
 474                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 475                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 476                                                   lmv_proc_dir,
 477                                                   "../../../%s/%s",
 478                                                   mdc_obd->obd_type->typ_name,
 479                                                   mdc_obd->obd_name);
 480                 if (mdc_symlink == NULL) {
 481                         CERROR("Could not register LMV target "
 482                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 483                                obd->obd_type->typ_name, obd->obd_name,
 484                                mdc_obd->obd_name);
 485                         lprocfs_remove(&lmv_proc_dir);
 486                         lmv_proc_dir = NULL;
 487                 }
 488         }
 489 #endif
 490         RETURN(0);
 491 }
 492
 493 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 494 {
 495         struct lmv_obd      *lmv = &obd->u.lmv;
 496         struct lmv_tgt_desc *tgt;
 497         int                  rc = 0;
 498         ENTRY;
 499
 500         CDEBUG(D_CONFIG, "Target uuid: %s.\n", tgt_uuid->uuid);
 501
 502         lmv_init_lock(lmv);
 503
 504         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 505                 lmv_init_unlock(lmv);
 506                 CERROR("Can't add %s, LMV module compiled for %d MDCs. "
 507                        "That many MDCs already configured.\n",
 508                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 509                 RETURN(-EINVAL);
 510         }
 511         if (lmv->desc.ld_tgt_count == 0) {
 512                 struct obd_device *mdc_obd;
 513
 514                 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
 515                                                 &obd->obd_uuid);
 516                 if (!mdc_obd) {
 517                         lmv_init_unlock(lmv);
 518                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 519                         RETURN(-EINVAL);
 520                 }
 521
 522                 rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, NULL);
 523                 if (rc) {
 524                         lmv_init_unlock(lmv);
 525                         CERROR("lmv failed to setup llogging subsystems\n");
 526                 }
 527         }
 528         cfs_spin_lock(&lmv->lmv_lock);
 529         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 530         tgt->ltd_uuid = *tgt_uuid;
 531         cfs_spin_unlock(&lmv->lmv_lock);
 532
 533         if (lmv->connected) {
 534                 rc = lmv_connect_mdc(obd, tgt);
 535                 if (rc) {
 536                         cfs_spin_lock(&lmv->lmv_lock);
 537                         lmv->desc.ld_tgt_count--;
 538                         memset(tgt, 0, sizeof(*tgt));
 539                         cfs_spin_unlock(&lmv->lmv_lock);
 540                 } else {
 541                         int easize = sizeof(struct lmv_stripe_md) +
 542                                      lmv->desc.ld_tgt_count *
 543                                      sizeof(struct lu_fid);
 544                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 545                 }
 546         }
 547
 548         lmv_init_unlock(lmv);
 549         RETURN(rc);
 550 }
 551
 552 int lmv_check_connect(struct obd_device *obd)
 553 {
 554         struct lmv_obd       *lmv = &obd->u.lmv;
 555         struct lmv_tgt_desc  *tgt;
 556         int                   i;
 557         int                   rc;
 558         int                   easize;
 559         ENTRY;
 560
 561         if (lmv->connected)
 562                 RETURN(0);
 563
 564         lmv_init_lock(lmv);
 565         if (lmv->connected) {
 566                 lmv_init_unlock(lmv);
 567                 RETURN(0);
 568         }
 569
 570         if (lmv->desc.ld_tgt_count == 0) {
 571                 CERROR("%s: no targets configured.\n", obd->obd_name);
 572                 RETURN(-EINVAL);
 573         }
 574
 575         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 576                lmv->cluuid.uuid, obd->obd_name);
 577
 578         LASSERT(lmv->tgts != NULL);
 579
 580         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 581                 rc = lmv_connect_mdc(obd, tgt);
 582                 if (rc)
 583                         GOTO(out_disc, rc);
 584         }
 585
 586         lmv_set_timeouts(obd);
 587         class_export_put(lmv->exp);
 588         lmv->connected = 1;
 589         easize = lmv_get_easize(lmv);
 590         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 591         lmv_init_unlock(lmv);
 592         RETURN(0);
 593
 594  out_disc:
 595         while (i-- > 0) {
 596                 int rc2;
 597                 --tgt;
 598                 tgt->ltd_active = 0;
 599                 if (tgt->ltd_exp) {
 600                         --lmv->desc.ld_active_tgt_count;
 601                         rc2 = obd_disconnect(tgt->ltd_exp);
 602                         if (rc2) {
 603                                 CERROR("LMV target %s disconnect on "
 604                                        "MDC idx %d: error %d\n",
 605                                        tgt->ltd_uuid.uuid, i, rc2);
 606                         }
 607                 }
 608         }
 609         class_disconnect(lmv->exp);
 610         lmv_init_unlock(lmv);
 611         RETURN(rc);
 612 }
 613
 614 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 615 {
 616 #ifdef __KERNEL__
 617         struct proc_dir_entry  *lmv_proc_dir;
 618 #endif
 619         struct lmv_obd         *lmv = &obd->u.lmv;
 620         struct obd_device      *mdc_obd;
 621         int                     rc;
 622         ENTRY;
 623
 624         LASSERT(tgt != NULL);
 625         LASSERT(obd != NULL);
 626
 627         mdc_obd = class_exp2obd(tgt->ltd_exp);
 628
 629         if (mdc_obd)
 630                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 631
 632 #ifdef __KERNEL__
 633         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 634         if (lmv_proc_dir) {
 635                 struct proc_dir_entry *mdc_symlink;
 636
 637                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 638                 if (mdc_symlink) {
 639                         lprocfs_remove(&mdc_symlink);
 640                 } else {
 641                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 642                                obd->obd_type->typ_name, obd->obd_name,
 643                                mdc_obd->obd_name);
 644                 }
 645         }
 646 #endif
 647         rc = obd_fid_fini(tgt->ltd_exp);
 648         if (rc)
 649                 CERROR("Can't finanize fids factory\n");
 650
 651         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 652                tgt->ltd_exp->exp_obd->obd_name,
 653                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 654
 655         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 656         rc = obd_disconnect(tgt->ltd_exp);
 657         if (rc) {
 658                 if (tgt->ltd_active) {
 659                         CERROR("Target %s disconnect error %d\n",
 660                                tgt->ltd_uuid.uuid, rc);
 661                 }
 662         }
 663
 664         lmv_activate_target(lmv, tgt, 0);
 665         tgt->ltd_exp = NULL;
 666         RETURN(0);
 667 }
 668
 669 static int lmv_disconnect(struct obd_export *exp)
 670 {
 671         struct obd_device     *obd = class_exp2obd(exp);
 672 #ifdef __KERNEL__
 673         struct proc_dir_entry *lmv_proc_dir;
 674 #endif
 675         struct lmv_obd        *lmv = &obd->u.lmv;
 676         int                    rc;
 677         int                    i;
 678         ENTRY;
 679
 680         if (!lmv->tgts)
 681                 goto out_local;
 682
 683         /*
 684          * Only disconnect the underlying layers on the final disconnect.
 685          */
 686         lmv->refcount--;
 687         if (lmv->refcount != 0)
 688                 goto out_local;
 689
 690         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 691                 if (lmv->tgts[i].ltd_exp == NULL)
 692                         continue;
 693                 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
 694         }
 695
 696 #ifdef __KERNEL__
 697         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 698         if (lmv_proc_dir) {
 699                 lprocfs_remove(&lmv_proc_dir);
 700         } else {
 701                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 702                        obd->obd_type->typ_name, obd->obd_name);
 703         }
 704 #endif
 705
 706 out_local:
 707         /*
 708          * This is the case when no real connection is established by
 709          * lmv_check_connect().
 710          */
 711         if (!lmv->connected)
 712                 class_export_put(exp);
 713         rc = class_disconnect(exp);
 714         if (lmv->refcount == 0)
 715                 lmv->connected = 0;
 716         RETURN(rc);
 717 }
 718
 719 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 720                          int len, void *karg, void *uarg)
 721 {
 722         struct obd_device    *obddev = class_exp2obd(exp);
 723         struct lmv_obd       *lmv = &obddev->u.lmv;
 724         int                   i = 0;
 725         int                   rc = 0;
 726         int                   set = 0;
 727         int                   count = lmv->desc.ld_tgt_count;
 728         ENTRY;
 729
 730         if (count == 0)
 731                 RETURN(-ENOTTY);
 732
 733         switch (cmd) {
 734         case IOC_OBD_STATFS: {
 735                 struct obd_ioctl_data *data = karg;
 736                 struct obd_device *mdc_obd;
 737                 struct obd_statfs stat_buf = {0};
 738                 __u32 index;
 739
 740                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 741                 if ((index >= count))
 742                         RETURN(-ENODEV);
 743
 744                 if (!lmv->tgts[index].ltd_active)
 745                         RETURN(-ENODATA);
 746
 747                 mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp);
 748                 if (!mdc_obd)
 749                         RETURN(-EINVAL);
 750
 751                 rc = obd_statfs(mdc_obd, &stat_buf,
 752                                 cfs_time_current_64() - CFS_HZ, 0);
 753                 if (rc)
 754                         RETURN(rc);
 755                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 756                                      data->ioc_plen1))
 757                         RETURN(-EFAULT);
 758                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 759                                      data->ioc_plen2))
 760                         RETURN(-EFAULT);
 761                 break;
 762         }
 763         case OBD_IOC_QUOTACTL: {
 764                 struct if_quotactl *qctl = karg;
 765                 struct lmv_tgt_desc *tgt = NULL;
 766                 struct obd_quotactl *oqctl;
 767
 768                 if (qctl->qc_valid == QC_MDTIDX) {
 769                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 770                                 RETURN(-EINVAL);
 771
 772                         tgt = &lmv->tgts[qctl->qc_idx];
 773                         if (!tgt->ltd_exp)
 774                                 RETURN(-EINVAL);
 775                 } else if (qctl->qc_valid == QC_UUID) {
 776                         for (i = 0; i < count; i++) {
 777                                 tgt = &lmv->tgts[i];
 778                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 779                                                      &qctl->obd_uuid))
 780                                         continue;
 781
 782                                 if (tgt->ltd_exp == NULL)
 783                                         RETURN(-EINVAL);
 784
 785                                 break;
 786                         }
 787                 } else {
 788                         RETURN(-EINVAL);
 789                 }
 790
 791                 if (i >= count)
 792                         RETURN(-EAGAIN);
 793
 794                 LASSERT(tgt && tgt->ltd_exp);
 795                 OBD_ALLOC_PTR(oqctl);
 796                 if (!oqctl)
 797                         RETURN(-ENOMEM);
 798
 799                 QCTL_COPY(oqctl, qctl);
 800                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 801                 if (rc == 0) {
 802                         QCTL_COPY(qctl, oqctl);
 803                         qctl->qc_valid = QC_MDTIDX;
 804                         qctl->obd_uuid = tgt->ltd_uuid;
 805                 }
 806                 OBD_FREE_PTR(oqctl);
 807                 break;
 808         }
 809         case OBD_IOC_CHANGELOG_SEND:
 810         case OBD_IOC_CHANGELOG_CLEAR: {
 811                 struct ioc_changelog *icc = karg;
 812
 813                 if (icc->icc_mdtindex >= count)
 814                         RETURN(-ENODEV);
 815
 816                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex].ltd_exp,
 817                                    sizeof(*icc), icc, NULL);
 818                 break;
 819         }
 820
 821         default : {
 822                 for (i = 0; i < count; i++) {
 823                         int err;
 824
 825                         if (lmv->tgts[i].ltd_exp == NULL)
 826                                 continue;
 827
 828                         err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len,
 829                                             karg, uarg);
 830                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 831                                 RETURN(err);
 832                         } else if (err) {
 833                                 if (lmv->tgts[i].ltd_active) {
 834                                         CERROR("error: iocontrol MDC %s on MDT"
 835                                                "idx %d cmd %x: err = %d\n",
 836                                                 lmv->tgts[i].ltd_uuid.uuid,
 837                                                 i, cmd, err);
 838                                         if (!rc)
 839                                                 rc = err;
 840                                 }
 841                         } else
 842                                 set = 1;
 843                 }
 844                 if (!set && !rc)
 845                         rc = -EIO;
 846         }
 847         }
 848         RETURN(rc);
 849 }
 850
 851 static int lmv_all_chars_policy(int count, const char *name,
 852                                 int len)
 853 {
 854         unsigned int c = 0;
 855
 856         while (len > 0)
 857                 c += name[--len];
 858         c = c % count;
 859         return c;
 860 }
 861
 862 static int lmv_nid_policy(struct lmv_obd *lmv)
 863 {
 864         struct obd_import *imp;
 865         __u32              id;
 866
 867         /*
 868          * XXX: To get nid we assume that underlying obd device is mdc.
 869          */
 870         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 871         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 872         return id % lmv->desc.ld_tgt_count;
 873 }
 874
 875 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 876                           placement_policy_t placement)
 877 {
 878         switch (placement) {
 879         case PLACEMENT_CHAR_POLICY:
 880                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 881                                             op_data->op_name,
 882                                             op_data->op_namelen);
 883         case PLACEMENT_NID_POLICY:
 884                 return lmv_nid_policy(lmv);
 885
 886         default:
 887                 break;
 888         }
 889
 890         CERROR("Unsupported placement policy %x\n", placement);
 891         return -EINVAL;
 892 }
 893
 894 /**
 895  * This is _inode_ placement policy function (not name).
 896  */
 897 static int lmv_placement_policy(struct obd_device *obd,
 898                                 struct md_op_data *op_data,
 899                                 mdsno_t *mds)
 900 {
 901         struct lmv_obd          *lmv = &obd->u.lmv;
 902         struct lmv_object       *obj;
 903         int                      rc;
 904         ENTRY;
 905
 906         LASSERT(mds != NULL);
 907
 908         if (lmv->desc.ld_tgt_count == 1) {
 909                 *mds = 0;
 910                 RETURN(0);
 911         }
 912
 913         /*
 914          * Allocate new fid on target according to operation type and parent
 915          * home mds.
 916          */
 917         obj = lmv_object_find(obd, &op_data->op_fid1);
 918         if (obj != NULL || op_data->op_name == NULL ||
 919             op_data->op_opc != LUSTRE_OPC_MKDIR) {
 920                 /*
 921                  * Allocate fid for non-dir or for null name or for case parent
 922                  * dir is split.
 923                  */
 924                 if (obj) {
 925                         lmv_object_put(obj);
 926
 927                         /*
 928                          * If we have this flag turned on, and we see that
 929                          * parent dir is split, this means, that caller did not
 930                          * notice split yet. This is race and we would like to
 931                          * let caller know that.
 932                          */
 933                         if (op_data->op_bias & MDS_CHECK_SPLIT)
 934                                 RETURN(-ERESTART);
 935                 }
 936
 937                 /*
 938                  * Allocate new fid on same mds where parent fid is located and
 939                  * where operation will be sent. In case of split dir, ->op_fid1
 940                  * and ->op_mds here will contain fid and mds of slave directory
 941                  * object (assigned by caller).
 942                  */
 943                 *mds = op_data->op_mds;
 944                 rc = 0;
 945         } else {
 946                 /*
 947                  * Parent directory is not split and we want to create a
 948                  * directory in it. Let's calculate where to place it according
 949                  * to operation data @op_data.
 950                  */
 951                 *mds = lmv_choose_mds(lmv, op_data, lmv->lmv_placement);
 952                 rc = 0;
 953         }
 954
 955         if (rc) {
 956                 CERROR("Can't choose MDS, err = %d\n", rc);
 957         } else {
 958                 LASSERT(*mds < lmv->desc.ld_tgt_count);
 959         }
 960
 961         RETURN(rc);
 962 }
 963
 964 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
 965                     mdsno_t mds)
 966 {
 967         struct lmv_tgt_desc *tgt;
 968         int                  rc;
 969         ENTRY;
 970
 971         tgt = lmv_get_target(lmv, mds);
 972
 973         /*
 974          * New seq alloc and FLD setup should be atomic. Otherwise we may find
 975          * on server that seq in new allocated fid is not yet known.
 976          */
 977         cfs_down(&tgt->ltd_fid_sem);
 978
 979         if (!tgt->ltd_active)
 980                 GOTO(out, rc = -ENODEV);
 981
 982         /*
 983          * Asking underlaying tgt layer to allocate new fid.
 984          */
 985         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
 986         if (rc > 0) {
 987                 LASSERT(fid_is_sane(fid));
 988                 rc = 0;
 989         }
 990
 991         EXIT;
 992 out:
 993         cfs_up(&tgt->ltd_fid_sem);
 994         return rc;
 995 }
 996
 997 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
 998                   struct md_op_data *op_data)
 999 {
1000         struct obd_device     *obd = class_exp2obd(exp);
1001         struct lmv_obd        *lmv = &obd->u.lmv;
1002         mdsno_t                mds;
1003         int                    rc;
1004         ENTRY;
1005
1006         LASSERT(op_data != NULL);
1007         LASSERT(fid != NULL);
1008
1009         rc = lmv_placement_policy(obd, op_data, &mds);
1010         if (rc) {
1011                 CERROR("Can't get target for allocating fid, "
1012                        "rc %d\n", rc);
1013                 RETURN(rc);
1014         }
1015
1016         rc = __lmv_fid_alloc(lmv, fid, mds);
1017         if (rc) {
1018                 CERROR("Can't alloc new fid, rc %d\n", rc);
1019                 RETURN(rc);
1020         }
1021
1022         RETURN(rc);
1023 }
1024
1025 static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1026 {
1027         ENTRY;
1028         LASSERT(exp != NULL && fid != NULL);
1029         if (lmv_object_delete(exp, fid)) {
1030                 CDEBUG(D_INODE, "Object "DFID" is destroyed.\n",
1031                        PFID(fid));
1032         }
1033         RETURN(0);
1034 }
1035
1036 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1037 {
1038         struct lmv_obd             *lmv = &obd->u.lmv;
1039         struct lprocfs_static_vars  lvars;
1040         struct lmv_desc            *desc;
1041         int                         rc;
1042         int                         i = 0;
1043         ENTRY;
1044
1045         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1046                 CERROR("LMV setup requires a descriptor\n");
1047                 RETURN(-EINVAL);
1048         }
1049
1050         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1051         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1052                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1053                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1054                 RETURN(-EINVAL);
1055         }
1056
1057         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
1058
1059         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
1060         if (lmv->tgts == NULL)
1061                 RETURN(-ENOMEM);
1062
1063         for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
1064                 cfs_sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
1065                 lmv->tgts[i].ltd_idx = i;
1066         }
1067
1068         lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
1069
1070         OBD_ALLOC(lmv->datas, lmv->datas_size);
1071         if (lmv->datas == NULL)
1072                 GOTO(out_free_tgts, rc = -ENOMEM);
1073
1074         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1075         lmv->desc.ld_tgt_count = 0;
1076         lmv->desc.ld_active_tgt_count = 0;
1077         lmv->max_cookiesize = 0;
1078         lmv->max_def_easize = 0;
1079         lmv->max_easize = 0;
1080         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1081
1082         cfs_spin_lock_init(&lmv->lmv_lock);
1083         cfs_sema_init(&lmv->init_sem, 1);
1084
1085         rc = lmv_object_setup(obd);
1086         if (rc) {
1087                 CERROR("Can't setup LMV object manager, error %d.\n", rc);
1088                 GOTO(out_free_datas, rc);
1089         }
1090
1091         lprocfs_lmv_init_vars(&lvars);
1092         lprocfs_obd_setup(obd, lvars.obd_vars);
1093 #ifdef LPROCFS
1094         {
1095                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status",
1096                                         0444, &lmv_proc_target_fops, obd);
1097                 if (rc)
1098                         CWARN("Error adding target_obd_stats file (%d)\n", rc);
1099        }
1100 #endif
1101         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1102                              LUSTRE_CLI_FLD_HASH_DHT);
1103         if (rc) {
1104                 CERROR("Can't init FLD, err %d\n", rc);
1105                 GOTO(out_free_datas, rc);
1106         }
1107
1108         RETURN(0);
1109
1110 out_free_datas:
1111         OBD_FREE(lmv->datas, lmv->datas_size);
1112         lmv->datas = NULL;
1113 out_free_tgts:
1114         OBD_FREE(lmv->tgts, lmv->tgts_size);
1115         lmv->tgts = NULL;
1116         return rc;
1117 }
1118
1119 static int lmv_cleanup(struct obd_device *obd)
1120 {
1121         struct lmv_obd   *lmv = &obd->u.lmv;
1122         ENTRY;
1123
1124         fld_client_fini(&lmv->lmv_fld);
1125         lprocfs_obd_cleanup(obd);
1126         lmv_object_cleanup(obd);
1127         OBD_FREE(lmv->datas, lmv->datas_size);
1128         OBD_FREE(lmv->tgts, lmv->tgts_size);
1129
1130         RETURN(0);
1131 }
1132
1133 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1134 {
1135         struct lustre_cfg     *lcfg = buf;
1136         struct obd_uuid        tgt_uuid;
1137         int                    rc;
1138         ENTRY;
1139
1140         switch(lcfg->lcfg_command) {
1141         case LCFG_ADD_MDC:
1142                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1143                         GOTO(out, rc = -EINVAL);
1144
1145                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1146                 rc = lmv_add_target(obd, &tgt_uuid);
1147                 GOTO(out, rc);
1148         default: {
1149                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1150                 GOTO(out, rc = -EINVAL);
1151         }
1152         }
1153 out:
1154         RETURN(rc);
1155 }
1156
1157 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1158                       __u64 max_age, __u32 flags)
1159 {
1160         struct lmv_obd        *lmv = &obd->u.lmv;
1161         struct obd_statfs     *temp;
1162         int                    rc = 0;
1163         int                    i;
1164         ENTRY;
1165
1166         rc = lmv_check_connect(obd);
1167         if (rc)
1168                 RETURN(rc);
1169
1170         OBD_ALLOC(temp, sizeof(*temp));
1171         if (temp == NULL)
1172                 RETURN(-ENOMEM);
1173
1174         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1175                 if (lmv->tgts[i].ltd_exp == NULL)
1176                         continue;
1177
1178                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp,
1179                                 max_age, flags);
1180                 if (rc) {
1181                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1182                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1183                                rc);
1184                         GOTO(out_free_temp, rc);
1185                 }
1186                 if (i == 0) {
1187                         *osfs = *temp;
1188                 } else {
1189                         osfs->os_bavail += temp->os_bavail;
1190                         osfs->os_blocks += temp->os_blocks;
1191                         osfs->os_ffree += temp->os_ffree;
1192                         osfs->os_files += temp->os_files;
1193                 }
1194         }
1195
1196         EXIT;
1197 out_free_temp:
1198         OBD_FREE(temp, sizeof(*temp));
1199         return rc;
1200 }
1201
1202 static int lmv_getstatus(struct obd_export *exp,
1203                          struct lu_fid *fid,
1204                          struct obd_capa **pc)
1205 {
1206         struct obd_device    *obd = exp->exp_obd;
1207         struct lmv_obd       *lmv = &obd->u.lmv;
1208         int                   rc;
1209         ENTRY;
1210
1211         rc = lmv_check_connect(obd);
1212         if (rc)
1213                 RETURN(rc);
1214
1215         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1216         RETURN(rc);
1217 }
1218
1219 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1220                         struct obd_capa *oc, obd_valid valid, const char *name,
1221                         const char *input, int input_size, int output_size,
1222                         int flags, struct ptlrpc_request **request)
1223 {
1224         struct obd_device      *obd = exp->exp_obd;
1225         struct lmv_obd         *lmv = &obd->u.lmv;
1226         struct lmv_tgt_desc    *tgt;
1227         int                     rc;
1228         ENTRY;
1229
1230         rc = lmv_check_connect(obd);
1231         if (rc)
1232                 RETURN(rc);
1233
1234         tgt = lmv_find_target(lmv, fid);
1235         if (IS_ERR(tgt))
1236                 RETURN(PTR_ERR(tgt));
1237
1238         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1239                          input_size, output_size, flags, request);
1240
1241         RETURN(rc);
1242 }
1243
1244 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1245                         struct obd_capa *oc, obd_valid valid, const char *name,
1246                         const char *input, int input_size, int output_size,
1247                         int flags, __u32 suppgid,
1248                         struct ptlrpc_request **request)
1249 {
1250         struct obd_device      *obd = exp->exp_obd;
1251         struct lmv_obd         *lmv = &obd->u.lmv;
1252         struct lmv_tgt_desc    *tgt;
1253         int                     rc;
1254         ENTRY;
1255
1256         rc = lmv_check_connect(obd);
1257         if (rc)
1258                 RETURN(rc);
1259
1260         tgt = lmv_find_target(lmv, fid);
1261         if (IS_ERR(tgt))
1262                 RETURN(PTR_ERR(tgt));
1263
1264         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1265                          input_size, output_size, flags, suppgid,
1266                          request);
1267
1268         RETURN(rc);
1269 }
1270
1271 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1272                        struct ptlrpc_request **request)
1273 {
1274         struct obd_device       *obd = exp->exp_obd;
1275         struct lmv_obd          *lmv = &obd->u.lmv;
1276         struct lmv_tgt_desc     *tgt;
1277         struct lmv_object       *obj;
1278         int                      rc;
1279         int                      i;
1280         ENTRY;
1281
1282         rc = lmv_check_connect(obd);
1283         if (rc)
1284                 RETURN(rc);
1285
1286         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1287         if (IS_ERR(tgt))
1288                 RETURN(PTR_ERR(tgt));
1289
1290         if (op_data->op_valid & OBD_MD_MDTIDX) {
1291                 op_data->op_mds = tgt->ltd_idx;
1292                 RETURN(0);
1293         }
1294
1295         rc = md_getattr(tgt->ltd_exp, op_data, request);
1296         if (rc)
1297                 RETURN(rc);
1298
1299         obj = lmv_object_find_lock(obd, &op_data->op_fid1);
1300
1301         CDEBUG(D_INODE, "GETATTR for "DFID" %s\n", PFID(&op_data->op_fid1),
1302                obj ? "(split)" : "");
1303
1304         /*
1305          * If object is split, then we loop over all the slaves and gather size
1306          * attribute. In ideal world we would have to gather also mds field from
1307          * all slaves, as object is spread over the cluster and this is
1308          * definitely interesting information and it is not good to loss it,
1309          * but...
1310          */
1311         if (obj) {
1312                 struct mdt_body *body;
1313
1314                 if (*request == NULL) {
1315                         lmv_object_put(obj);
1316                         RETURN(rc);
1317                 }
1318
1319                 body = req_capsule_server_get(&(*request)->rq_pill,
1320                                               &RMF_MDT_BODY);
1321                 LASSERT(body != NULL);
1322
1323                 for (i = 0; i < obj->lo_objcount; i++) {
1324                         if (lmv->tgts[i].ltd_exp == NULL) {
1325                                 CWARN("%s: NULL export for %d\n",
1326                                       obd->obd_name, i);
1327                                 continue;
1328                         }
1329
1330                         /*
1331                          * Skip master object.
1332                          */
1333                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid))
1334                                 continue;
1335
1336                         body->size += obj->lo_stripes[i].ls_size;
1337                 }
1338
1339                 lmv_object_put_unlock(obj);
1340         }
1341
1342         RETURN(rc);
1343 }
1344
1345 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1346                              ldlm_iterator_t it, void *data)
1347 {
1348         struct obd_device   *obd = exp->exp_obd;
1349         struct lmv_obd      *lmv = &obd->u.lmv;
1350         int                  i;
1351         int                  rc;
1352         ENTRY;
1353
1354         rc = lmv_check_connect(obd);
1355         if (rc)
1356                 RETURN(rc);
1357
1358         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1359
1360         /*
1361          * With CMD every object can have two locks in different namespaces:
1362          * lookup lock in space of mds storing direntry and update/open lock in
1363          * space of mds storing inode.
1364          */
1365         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1366                 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1367
1368         RETURN(0);
1369 }
1370
1371 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1372                      struct md_open_data *mod, struct ptlrpc_request **request)
1373 {
1374         struct obd_device     *obd = exp->exp_obd;
1375         struct lmv_obd        *lmv = &obd->u.lmv;
1376         struct lmv_tgt_desc   *tgt;
1377         int                    rc;
1378         ENTRY;
1379
1380         rc = lmv_check_connect(obd);
1381         if (rc)
1382                 RETURN(rc);
1383
1384         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1385         if (IS_ERR(tgt))
1386                 RETURN(PTR_ERR(tgt));
1387
1388         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1389         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1390         RETURN(rc);
1391 }
1392
1393 /**
1394  * Called in the case MDS returns -ERESTART on create on open, what means that
1395  * directory is split and its LMV presentation object has to be updated.
1396  */
1397 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1398 {
1399         struct obd_device       *obd = exp->exp_obd;
1400         struct lmv_obd          *lmv = &obd->u.lmv;
1401         struct ptlrpc_request   *req = NULL;
1402         struct lmv_tgt_desc     *tgt;
1403         struct lmv_object       *obj;
1404         struct lustre_md         md;
1405         struct md_op_data       *op_data;
1406         int                      mealen;
1407         int                      rc;
1408         __u64                    valid;
1409         ENTRY;
1410
1411         md.mea = NULL;
1412         mealen = lmv_get_easize(lmv);
1413
1414         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1415
1416         tgt = lmv_find_target(lmv, fid);
1417         if (IS_ERR(tgt))
1418                 RETURN(PTR_ERR(tgt));
1419
1420         /*
1421          * Time to update mea of parent fid.
1422          */
1423
1424         OBD_ALLOC_PTR(op_data);
1425         if (op_data == NULL)
1426                 RETURN(-ENOMEM);
1427
1428         op_data->op_fid1 = *fid;
1429         op_data->op_mode = mealen;
1430         op_data->op_valid = valid;
1431
1432         rc = md_getattr(tgt->ltd_exp, op_data, &req);
1433         OBD_FREE_PTR(op_data);
1434         if (rc) {
1435                 CERROR("md_getattr() failed, error %d\n", rc);
1436                 GOTO(cleanup, rc);
1437         }
1438
1439         rc = md_get_lustre_md(tgt->ltd_exp, req, NULL, exp, &md);
1440         if (rc) {
1441                 CERROR("md_get_lustre_md() failed, error %d\n", rc);
1442                 GOTO(cleanup, rc);
1443         }
1444
1445         if (md.mea == NULL)
1446                 GOTO(cleanup, rc = -ENODATA);
1447
1448         obj = lmv_object_create(exp, fid, md.mea);
1449         if (IS_ERR(obj))
1450                 rc = PTR_ERR(obj);
1451         else
1452                 lmv_object_put(obj);
1453
1454         obd_free_memmd(exp, (void *)&md.mea);
1455         EXIT;
1456 cleanup:
1457         if (req)
1458                 ptlrpc_req_finished(req);
1459         return rc;
1460 }
1461
1462 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1463                const void *data, int datalen, int mode, __u32 uid,
1464                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1465                struct ptlrpc_request **request)
1466 {
1467         struct obd_device       *obd = exp->exp_obd;
1468         struct lmv_obd          *lmv = &obd->u.lmv;
1469         struct lmv_tgt_desc     *tgt;
1470         struct lmv_object       *obj;
1471         int                      rc;
1472         int                      loop = 0;
1473         int                      sidx;
1474         ENTRY;
1475
1476         rc = lmv_check_connect(obd);
1477         if (rc)
1478                 RETURN(rc);
1479
1480         if (!lmv->desc.ld_active_tgt_count)
1481                 RETURN(-EIO);
1482 repeat:
1483         ++loop;
1484         LASSERT(loop <= 2);
1485
1486         obj = lmv_object_find(obd, &op_data->op_fid1);
1487         if (obj) {
1488                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1489                                        op_data->op_name, op_data->op_namelen);
1490                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1491                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1492                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1493                 tgt = lmv_get_target(lmv, op_data->op_mds);
1494                 lmv_object_put(obj);
1495         } else {
1496                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1497                 op_data->op_bias |= MDS_CHECK_SPLIT;
1498                 op_data->op_mds = tgt->ltd_idx;
1499         }
1500
1501         if (IS_ERR(tgt))
1502                 RETURN(PTR_ERR(tgt));
1503
1504         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1505         if (rc == -ERESTART)
1506                 goto repeat;
1507         else if (rc)
1508                 RETURN(rc);
1509
1510         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1511                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1512                op_data->op_mds);
1513
1514         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1515         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1516                        cap_effective, rdev, request);
1517         if (rc == 0) {
1518                 if (*request == NULL)
1519                         RETURN(rc);
1520                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1521         } else if (rc == -ERESTART) {
1522                 LASSERT(*request != NULL);
1523                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1524                           "Got -ERESTART during create!\n");
1525                 ptlrpc_req_finished(*request);
1526                 *request = NULL;
1527
1528                 /*
1529                  * Directory got split. Time to update local object and repeat
1530                  * the request with proper MDS.
1531                  */
1532                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1533                 if (rc == 0) {
1534                         rc = lmv_allocate_slaves(obd, &op_data->op_fid1,
1535                                                  op_data, &op_data->op_fid2);
1536                         if (rc)
1537                                 RETURN(rc);
1538                         goto repeat;
1539                 }
1540         }
1541         RETURN(rc);
1542 }
1543
1544 static int lmv_done_writing(struct obd_export *exp,
1545                             struct md_op_data *op_data,
1546                             struct md_open_data *mod)
1547 {
1548         struct obd_device     *obd = exp->exp_obd;
1549         struct lmv_obd        *lmv = &obd->u.lmv;
1550         struct lmv_tgt_desc   *tgt;
1551         int                    rc;
1552         ENTRY;
1553
1554         rc = lmv_check_connect(obd);
1555         if (rc)
1556                 RETURN(rc);
1557
1558         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1559         if (IS_ERR(tgt))
1560                 RETURN(PTR_ERR(tgt));
1561
1562         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1563         RETURN(rc);
1564 }
1565
1566 static int
1567 lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1568                    struct lookup_intent *it, struct md_op_data *op_data,
1569                    struct lustre_handle *lockh, void *lmm, int lmmsize)
1570 {
1571         struct obd_device     *obd = exp->exp_obd;
1572         struct lmv_obd        *lmv = &obd->u.lmv;
1573         struct lmv_stripe_md  *mea = op_data->op_mea1;
1574         struct md_op_data     *op_data2;
1575         struct lmv_tgt_desc   *tgt;
1576         int                    i;
1577         int                    rc = 0;
1578         ENTRY;
1579
1580         OBD_ALLOC_PTR(op_data2);
1581         if (op_data2 == NULL)
1582                 RETURN(-ENOMEM);
1583
1584         LASSERT(mea != NULL);
1585         for (i = 0; i < mea->mea_count; i++) {
1586                 memset(op_data2, 0, sizeof(*op_data2));
1587                 op_data2->op_fid1 = mea->mea_ids[i];
1588                 op_data2->op_bias = 0;
1589
1590                 tgt = lmv_find_target(lmv, &op_data2->op_fid1);
1591                 if (IS_ERR(tgt))
1592                         GOTO(cleanup, rc = PTR_ERR(tgt));
1593
1594                 if (tgt->ltd_exp == NULL)
1595                         continue;
1596
1597                 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data2,
1598                                 lockh + i, lmm, lmmsize, NULL, 0);
1599
1600                 CDEBUG(D_INODE, "Take lock on slave "DFID" -> %d/%d\n",
1601                        PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1602
1603                 if (rc)
1604                         GOTO(cleanup, rc);
1605
1606                 if (it->d.lustre.it_data) {
1607                         struct ptlrpc_request *req;
1608                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1609                         ptlrpc_req_finished(req);
1610                 }
1611
1612                 if (it->d.lustre.it_status)
1613                         GOTO(cleanup, rc = it->d.lustre.it_status);
1614         }
1615
1616         EXIT;
1617 cleanup:
1618         OBD_FREE_PTR(op_data2);
1619
1620         if (rc != 0) {
1621                 /*
1622                  * Drop all taken locks.
1623                  */
1624                 while (--i >= 0) {
1625                         if (lockh[i].cookie)
1626                                 ldlm_lock_decref(lockh + i, einfo->ei_mode);
1627                         lockh[i].cookie = 0;
1628                 }
1629         }
1630         return rc;
1631 }
1632
1633 static int
1634 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1635                    struct lookup_intent *it, struct md_op_data *op_data,
1636                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1637                    int extra_lock_flags)
1638 {
1639         struct ptlrpc_request      *req = it->d.lustre.it_data;
1640         struct obd_device          *obd = exp->exp_obd;
1641         struct lmv_obd             *lmv = &obd->u.lmv;
1642         struct lustre_handle        plock;
1643         struct lmv_tgt_desc        *tgt;
1644         struct md_op_data          *rdata;
1645         struct lu_fid               fid1;
1646         struct mdt_body            *body;
1647         int                         rc = 0;
1648         int                         pmode;
1649         ENTRY;
1650
1651         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1652         LASSERT(body != NULL);
1653
1654         if (!(body->valid & OBD_MD_MDS))
1655                 RETURN(0);
1656
1657         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1658                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1659
1660         /*
1661          * We got LOOKUP lock, but we really need attrs.
1662          */
1663         pmode = it->d.lustre.it_lock_mode;
1664         LASSERT(pmode != 0);
1665         memcpy(&plock, lockh, sizeof(plock));
1666         it->d.lustre.it_lock_mode = 0;
1667         it->d.lustre.it_data = NULL;
1668         fid1 = body->fid1;
1669
1670         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1671         ptlrpc_req_finished(req);
1672
1673         tgt = lmv_find_target(lmv, &fid1);
1674         if (IS_ERR(tgt))
1675                 GOTO(out, rc = PTR_ERR(tgt));
1676
1677         OBD_ALLOC_PTR(rdata);
1678         if (rdata == NULL)
1679                 GOTO(out, rc = -ENOMEM);
1680
1681         rdata->op_fid1 = fid1;
1682         rdata->op_bias = MDS_CROSS_REF;
1683
1684         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1685                         lmm, lmmsize, NULL, extra_lock_flags);
1686         OBD_FREE_PTR(rdata);
1687         EXIT;
1688 out:
1689         ldlm_lock_decref(&plock, pmode);
1690         return rc;
1691 }
1692
1693 static int
1694 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1695             struct lookup_intent *it, struct md_op_data *op_data,
1696             struct lustre_handle *lockh, void *lmm, int lmmsize,
1697             struct ptlrpc_request **req, int extra_lock_flags)
1698 {
1699         struct obd_device        *obd = exp->exp_obd;
1700         struct lmv_obd           *lmv = &obd->u.lmv;
1701         struct lmv_tgt_desc      *tgt;
1702         struct lmv_object        *obj;
1703         int                       sidx;
1704         int                       rc;
1705         ENTRY;
1706
1707         rc = lmv_check_connect(obd);
1708         if (rc)
1709                 RETURN(rc);
1710
1711         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1712                LL_IT2STR(it), PFID(&op_data->op_fid1));
1713
1714         if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) {
1715                 rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
1716                                         lockh, lmm, lmmsize);
1717                 RETURN(rc);
1718         }
1719
1720         obj = lmv_object_find(obd, &op_data->op_fid1);
1721         if (obj && op_data->op_namelen) {
1722                 sidx = raw_name2idx(obj->lo_hashtype,
1723                                        obj->lo_objcount,
1724                                        (char *)op_data->op_name,
1725                                        op_data->op_namelen);
1726                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1727                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1728         } else {
1729                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1730         }
1731         if (obj)
1732                 lmv_object_put(obj);
1733
1734         if (IS_ERR(tgt))
1735                 RETURN(PTR_ERR(tgt));
1736
1737         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1738                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1739
1740         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1741                         lmm, lmmsize, req, extra_lock_flags);
1742
1743         if (rc == 0 && it && it->it_op == IT_OPEN) {
1744                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1745                                         lmm, lmmsize, extra_lock_flags);
1746         }
1747         RETURN(rc);
1748 }
1749
1750 static int
1751 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1752                  struct ptlrpc_request **request)
1753 {
1754         struct ptlrpc_request   *req = NULL;
1755         struct obd_device       *obd = exp->exp_obd;
1756         struct lmv_obd          *lmv = &obd->u.lmv;
1757         struct lu_fid            rid = op_data->op_fid1;
1758         struct lmv_tgt_desc     *tgt;
1759         struct mdt_body         *body;
1760         struct lmv_object       *obj;
1761         obd_valid                valid = op_data->op_valid;
1762         int                      rc;
1763         int                      loop = 0;
1764         int                      sidx;
1765         ENTRY;
1766
1767         rc = lmv_check_connect(obd);
1768         if (rc)
1769                 RETURN(rc);
1770
1771 repeat:
1772         ++loop;
1773         LASSERT(loop <= 2);
1774         obj = lmv_object_find(obd, &rid);
1775         if (obj) {
1776                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1777                                     op_data->op_name, op_data->op_namelen);
1778                 rid = obj->lo_stripes[sidx].ls_fid;
1779                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1780                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1781                 valid &= ~OBD_MD_FLCKSPLIT;
1782                 lmv_object_put(obj);
1783         } else {
1784                 tgt = lmv_find_target(lmv, &rid);
1785                 valid |= OBD_MD_FLCKSPLIT;
1786                 op_data->op_mds = tgt->ltd_idx;
1787         }
1788         if (IS_ERR(tgt))
1789                 RETURN(PTR_ERR(tgt));
1790
1791         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" - "DFID" -> mds #%d\n",
1792                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1793                PFID(&rid), tgt->ltd_idx);
1794
1795         op_data->op_valid = valid;
1796         op_data->op_fid1 = rid;
1797         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1798         if (rc == 0) {
1799                 body = req_capsule_server_get(&(*request)->rq_pill,
1800                                               &RMF_MDT_BODY);
1801                 LASSERT(body != NULL);
1802
1803                 if (body->valid & OBD_MD_MDS) {
1804                         rid = body->fid1;
1805                         CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1806                                PFID(&rid));
1807
1808                         tgt = lmv_find_target(lmv, &rid);
1809                         if (IS_ERR(tgt)) {
1810                                 ptlrpc_req_finished(*request);
1811                                 RETURN(PTR_ERR(tgt));
1812                         }
1813
1814                         op_data->op_fid1 = rid;
1815                         op_data->op_valid |= OBD_MD_FLCROSSREF;
1816                         op_data->op_namelen = 0;
1817                         op_data->op_name = NULL;
1818                         rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1819                         ptlrpc_req_finished(*request);
1820                         *request = req;
1821                 }
1822         } else if (rc == -ERESTART) {
1823                 LASSERT(*request != NULL);
1824                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1825                           "Got -ERESTART during getattr!\n");
1826                 ptlrpc_req_finished(*request);
1827                 *request = NULL;
1828
1829                 /*
1830                  * Directory got split. Time to update local object and repeat
1831                  * the request with proper MDS.
1832                  */
1833                 rc = lmv_handle_split(exp, &rid);
1834                 if (rc == 0)
1835                         goto repeat;
1836         }
1837         RETURN(rc);
1838 }
1839
1840 #define md_op_data_fid(op_data, fl)                     \
1841         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1842          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1843          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1844          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1845          NULL)
1846
1847 static int lmv_early_cancel_slaves(struct obd_export *exp,
1848                                    struct md_op_data *op_data, int op_tgt,
1849                                    ldlm_mode_t mode, int bits, int flag)
1850 {
1851         struct obd_device      *obd = exp->exp_obd;
1852         struct lmv_obd         *lmv = &obd->u.lmv;
1853         ldlm_policy_data_t      policy = {{0}};
1854         struct lu_fid          *op_fid;
1855         struct lu_fid          *st_fid;
1856         struct lmv_tgt_desc    *tgt;
1857         struct lmv_object      *obj;
1858         int                     rc = 0;
1859         int                     i;
1860         ENTRY;
1861
1862         op_fid = md_op_data_fid(op_data, flag);
1863         if (!fid_is_sane(op_fid))
1864                 RETURN(0);
1865
1866         obj = lmv_object_find(obd, op_fid);
1867         if (obj == NULL)
1868                 RETURN(-EALREADY);
1869
1870         policy.l_inodebits.bits = bits;
1871         for (i = 0; i < obj->lo_objcount; i++) {
1872                 tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
1873                 st_fid = &obj->lo_stripes[i].ls_fid;
1874                 if (op_tgt != tgt->ltd_idx) {
1875                         CDEBUG(D_INODE, "EARLY_CANCEL slave "DFID" -> mds #%d\n",
1876                                PFID(st_fid), tgt->ltd_idx);
1877                         rc = md_cancel_unused(tgt->ltd_exp, st_fid, &policy,
1878                                               mode, LDLM_FL_ASYNC, NULL);
1879                         if (rc)
1880                                 GOTO(out_put_obj, rc);
1881                 } else {
1882                         CDEBUG(D_INODE,
1883                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1884                                op_tgt, PFID(st_fid));
1885                         /*
1886                          * Do not cancel locks for operation target, they will
1887                          * be handled later in underlaying layer when calling
1888                          * function we run on behalf of.
1889                          */
1890                         *op_fid = *st_fid;
1891                         op_data->op_flags |= flag;
1892                 }
1893         }
1894         EXIT;
1895 out_put_obj:
1896         lmv_object_put(obj);
1897         return rc;
1898 }
1899
1900 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1901                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1902 {
1903         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1904         struct obd_device      *obd = exp->exp_obd;
1905         struct lmv_obd         *lmv = &obd->u.lmv;
1906         struct lmv_tgt_desc    *tgt;
1907         ldlm_policy_data_t      policy = {{0}};
1908         struct lmv_object      *obj;
1909         int                     rc = 0;
1910         ENTRY;
1911
1912         if (!fid_is_sane(fid))
1913                 RETURN(0);
1914
1915         obj = lmv_object_find(obd, fid);
1916         if (obj) {
1917                 rc = lmv_early_cancel_slaves(exp, op_data, op_tgt, mode,
1918                                              bits, flag);
1919                 lmv_object_put(obj);
1920         } else {
1921                 tgt = lmv_find_target(lmv, fid);
1922                 if (IS_ERR(tgt))
1923                         RETURN(PTR_ERR(tgt));
1924
1925                 if (tgt->ltd_idx != op_tgt) {
1926                         CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1927                         policy.l_inodebits.bits = bits;
1928                         rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1929                                               mode, LDLM_FL_ASYNC, NULL);
1930                 } else {
1931                         CDEBUG(D_INODE,
1932                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1933                                op_tgt, PFID(fid));
1934                         op_data->op_flags |= flag;
1935                         rc = 0;
1936                 }
1937
1938         }
1939         RETURN(rc);
1940 }
1941
1942 /*
1943  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1944  * op_data->op_fid2
1945  */
1946 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1947                     struct ptlrpc_request **request)
1948 {
1949         struct obd_device       *obd = exp->exp_obd;
1950         struct lmv_obd          *lmv = &obd->u.lmv;
1951         struct lmv_tgt_desc     *tgt;
1952         struct lmv_object       *obj;
1953         int                      rc;
1954         int                      loop = 0;
1955         mdsno_t                  mds;
1956         int                      sidx;
1957         ENTRY;
1958
1959         rc = lmv_check_connect(obd);
1960         if (rc)
1961                 RETURN(rc);
1962
1963 repeat:
1964         ++loop;
1965         LASSERT(loop <= 2);
1966         LASSERT(op_data->op_namelen != 0);
1967
1968         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1969                PFID(&op_data->op_fid2), op_data->op_namelen,
1970                op_data->op_name, PFID(&op_data->op_fid1));
1971
1972         obj = lmv_object_find(obd, &op_data->op_fid2);
1973         if (obj) {
1974                 sidx = raw_name2idx(obj->lo_hashtype,
1975                                     obj->lo_objcount,
1976                                     op_data->op_name,
1977                                     op_data->op_namelen);
1978                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
1979                 mds = obj->lo_stripes[sidx].ls_mds;
1980                 lmv_object_put(obj);
1981         } else {
1982                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
1983                 if (rc)
1984                         RETURN(rc);
1985         }
1986
1987         CDEBUG(D_INODE, "Forward to mds #%x ("DFID")\n",
1988                mds, PFID(&op_data->op_fid1));
1989
1990         op_data->op_fsuid = cfs_curproc_fsuid();
1991         op_data->op_fsgid = cfs_curproc_fsgid();
1992         op_data->op_cap = cfs_curproc_cap_pack();
1993         tgt = lmv_get_target(lmv, mds);
1994
1995         /*
1996          * Cancel UPDATE lock on child (fid1).
1997          */
1998         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1999         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2000                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2001         if (rc == 0)
2002                 rc = md_link(tgt->ltd_exp, op_data, request);
2003         if (rc == -ERESTART) {
2004                 LASSERT(*request != NULL);
2005                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2006                           "Got -ERESTART during link!\n");
2007                 ptlrpc_req_finished(*request);
2008                 *request = NULL;
2009
2010                 /*
2011                  * Directory got split. Time to update local object and repeat
2012                  * the request with proper MDS.
2013                  */
2014                 rc = lmv_handle_split(exp, &op_data->op_fid2);
2015                 if (rc == 0)
2016                         goto repeat;
2017         }
2018
2019         RETURN(rc);
2020 }
2021
2022 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2023                       const char *old, int oldlen, const char *new, int newlen,
2024                       struct ptlrpc_request **request)
2025 {
2026         struct obd_device       *obd = exp->exp_obd;
2027         struct lmv_obd          *lmv = &obd->u.lmv;
2028         struct lmv_tgt_desc     *src_tgt;
2029         struct lmv_tgt_desc     *tgt_tgt;
2030         int                      rc;
2031         int                      sidx;
2032         int                      loop = 0;
2033         struct lmv_object       *obj;
2034         mdsno_t                  mds1;
2035         mdsno_t                  mds2;
2036         ENTRY;
2037
2038         LASSERT(oldlen != 0);
2039
2040         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2041                oldlen, old, PFID(&op_data->op_fid1),
2042                newlen, new, PFID(&op_data->op_fid2));
2043
2044         rc = lmv_check_connect(obd);
2045         if (rc)
2046                 RETURN(rc);
2047
2048 repeat:
2049         ++loop;
2050         LASSERT(loop <= 2);
2051         obj = lmv_object_find(obd, &op_data->op_fid1);
2052         if (obj) {
2053                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2054                                     (char *)old, oldlen);
2055                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2056                 mds1 = obj->lo_stripes[sidx].ls_mds;
2057                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
2058                 lmv_object_put(obj);
2059         } else {
2060                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
2061                 if (rc)
2062                         RETURN(rc);
2063         }
2064
2065         obj = lmv_object_find(obd, &op_data->op_fid2);
2066         if (obj) {
2067                 /*
2068                  * Directory is already split, so we have to forward request to
2069                  * the right MDS.
2070                  */
2071                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2072                                        (char *)new, newlen);
2073
2074                 mds2 = obj->lo_stripes[sidx].ls_mds;
2075                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2076                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
2077                 lmv_object_put(obj);
2078         } else {
2079                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
2080                 if (rc)
2081                         RETURN(rc);
2082         }
2083
2084         op_data->op_fsuid = cfs_curproc_fsuid();
2085         op_data->op_fsgid = cfs_curproc_fsgid();
2086         op_data->op_cap = cfs_curproc_cap_pack();
2087
2088         src_tgt = lmv_get_target(lmv, mds1);
2089         tgt_tgt = lmv_get_target(lmv, mds2);
2090
2091         /*
2092          * LOOKUP lock on src child (fid3) should also be cancelled for
2093          * src_tgt in mdc_rename.
2094          */
2095         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2096
2097         /*
2098          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2099          * own target.
2100          */
2101         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2102                               LCK_EX, MDS_INODELOCK_UPDATE,
2103                               MF_MDC_CANCEL_FID2);
2104
2105         /*
2106          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2107          */
2108         if (rc == 0) {
2109                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2110                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2111                                       MF_MDC_CANCEL_FID4);
2112         }
2113
2114         /*
2115          * Cancel all the locks on tgt child (fid4).
2116          */
2117         if (rc == 0)
2118                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2119                                       LCK_EX, MDS_INODELOCK_FULL,
2120                                       MF_MDC_CANCEL_FID4);
2121
2122         if (rc == 0)
2123                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2124                                new, newlen, request);
2125
2126         if (rc == -ERESTART) {
2127                 LASSERT(*request != NULL);
2128                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2129                           "Got -ERESTART during rename!\n");
2130                 ptlrpc_req_finished(*request);
2131                 *request = NULL;
2132
2133                 /*
2134                  * Directory got split. Time to update local object and repeat
2135                  * the request with proper MDS.
2136                  */
2137                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2138                 if (rc == 0)
2139                         goto repeat;
2140         }
2141         RETURN(rc);
2142 }
2143
2144 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2145                        void *ea, int ealen, void *ea2, int ea2len,
2146                        struct ptlrpc_request **request,
2147                        struct md_open_data **mod)
2148 {
2149         struct obd_device       *obd = exp->exp_obd;
2150         struct lmv_obd          *lmv = &obd->u.lmv;
2151         struct ptlrpc_request   *req;
2152         struct lmv_tgt_desc     *tgt;
2153         struct lmv_object       *obj;
2154         int                      rc = 0;
2155         int                      i;
2156         ENTRY;
2157
2158         rc = lmv_check_connect(obd);
2159         if (rc)
2160                 RETURN(rc);
2161
2162         obj = lmv_object_find(obd, &op_data->op_fid1);
2163
2164         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x%s\n",
2165                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2166                obj ? ", split" : "");
2167
2168         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2169         if (obj) {
2170                 for (i = 0; i < obj->lo_objcount; i++) {
2171                         op_data->op_fid1 = obj->lo_stripes[i].ls_fid;
2172
2173                         tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
2174                         if (IS_ERR(tgt)) {
2175                                 rc = PTR_ERR(tgt);
2176                                 break;
2177                         }
2178
2179                         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen,
2180                                         ea2, ea2len, &req, mod);
2181
2182                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) {
2183                                 /*
2184                                  * This is master object and this request should
2185                                  * be returned back to llite.
2186                                  */
2187                                 *request = req;
2188                         } else {
2189                                 ptlrpc_req_finished(req);
2190                         }
2191
2192                         if (rc)
2193                                 break;
2194                 }
2195                 lmv_object_put(obj);
2196         } else {
2197                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2198                 if (IS_ERR(tgt))
2199                         RETURN(PTR_ERR(tgt));
2200
2201                 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2202                                 ea2len, request, mod);
2203         }
2204         RETURN(rc);
2205 }
2206
2207 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2208                     struct obd_capa *oc, struct ptlrpc_request **request)
2209 {
2210         struct obd_device         *obd = exp->exp_obd;
2211         struct lmv_obd            *lmv = &obd->u.lmv;
2212         struct lmv_tgt_desc       *tgt;
2213         int                        rc;
2214         ENTRY;
2215
2216         rc = lmv_check_connect(obd);
2217         if (rc)
2218                 RETURN(rc);
2219
2220         tgt = lmv_find_target(lmv, fid);
2221         if (IS_ERR(tgt))
2222                 RETURN(PTR_ERR(tgt));
2223
2224         rc = md_sync(tgt->ltd_exp, fid, oc, request);
2225         RETURN(rc);
2226 }
2227
2228 /**
2229  * Main purpose of LMV blocking ast is to remove split directory LMV
2230  * presentation object (struct lmv_object) attached to the lock being revoked.
2231  */
2232 int lmv_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2233                      void *data, int flag)
2234 {
2235         struct lustre_handle    lockh;
2236         struct lmv_object      *obj;
2237         int                     rc;
2238         ENTRY;
2239
2240         switch (flag) {
2241         case LDLM_CB_BLOCKING:
2242                 ldlm_lock2handle(lock, &lockh);
2243                 rc = ldlm_cli_cancel(&lockh);
2244                 if (rc < 0) {
2245                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2246                         RETURN(rc);
2247                 }
2248                 break;
2249         case LDLM_CB_CANCELING:
2250                 /*
2251                  * Time to drop cached attrs for split directory object
2252                  */
2253                 obj = lock->l_ast_data;
2254                 if (obj) {
2255                         CDEBUG(D_INODE, "Cancel %s on "LPU64"/"LPU64
2256                                ", master "DFID"\n",
2257                                lock->l_resource->lr_name.name[3] == 1 ?
2258                                "LOOKUP" : "UPDATE",
2259                                lock->l_resource->lr_name.name[0],
2260                                lock->l_resource->lr_name.name[1],
2261                                PFID(&obj->lo_fid));
2262                         lmv_object_put(obj);
2263                 }
2264                 break;
2265         default:
2266                 LBUG();
2267         }
2268         RETURN(0);
2269 }
2270
2271 static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj)
2272 {
2273         __u64         val;
2274
2275         val = le64_to_cpu(*hash);
2276         if (val < hash_adj)
2277                 val += MAX_HASH_SIZE;
2278         if (val != DIR_END_OFF)
2279                 *hash = cpu_to_le64(val - hash_adj);
2280 }
2281
2282 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2283 {
2284         __u64              id;
2285         struct obd_import *imp;
2286
2287         /*
2288          * XXX: to get nid we assume that underlying obd device is mdc.
2289          */
2290         imp  = class_exp2cliimp(exp);
2291         id   = imp->imp_connection->c_self + fid_flatten(fid);
2292
2293         CDEBUG(D_INODE, "Readpage node rank: "LPX64" "DFID" "LPX64" "LPX64"\n",
2294                imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2295
2296         return id ^ (id >> 32);
2297 }
2298
2299 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2300                         struct obd_capa *oc, __u64 offset64, struct page *page,
2301                         struct ptlrpc_request **request)
2302 {
2303         struct obd_device       *obd = exp->exp_obd;
2304         struct lmv_obd          *lmv = &obd->u.lmv;
2305         struct lu_fid            rid = *fid;
2306         struct lmv_object       *obj;
2307         __u64                    offset;
2308         __u64                    hash_adj = 0;
2309         __u32                    rank = 0;
2310         __u64                    seg_size = 0;
2311         __u64                    tgt_tmp = 0;
2312         int                      tgt_idx = 0;
2313         int                      tgt0_idx = 0;
2314         int                      rc;
2315         int                      nr = 0;
2316         struct lmv_stripe       *los;
2317         struct lmv_tgt_desc     *tgt;
2318         struct lu_dirpage       *dp;
2319         struct lu_dirent        *ent;
2320         ENTRY;
2321
2322         offset = offset64;
2323
2324         rc = lmv_check_connect(obd);
2325         if (rc)
2326                 RETURN(rc);
2327
2328         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid));
2329
2330         /*
2331          * This case handle directory lookup in clustered metadata case (i.e.
2332          * split directory is located on multiple md servers.)
2333          * each server keeps directory entries for certain range of hashes.
2334          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
2335          * first server will keep records with hashes [ 0 ... MAX_HASH / N  - 1],
2336          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
2337          * so on....
2338          *      readdir can simply start reading entries from 0 - N server in
2339          * order but that will not scale well as all client will request dir in
2340          * to server in same order.
2341          * Following algorithm does optimization:
2342          * Instead of doing readdir in 1, 2, ...., N order, client with a
2343          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
2344          * (every client has rank R)
2345          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
2346          * since client ask dir from MDS{R} client has pages with offsets
2347          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
2348          * on hash  values that we get.
2349          */
2350         obj = lmv_object_find_lock(obd, fid);
2351         if (obj) {
2352                 nr       = obj->lo_objcount;
2353                 LASSERT(nr > 0);
2354                 seg_size = MAX_HASH_SIZE;
2355                 do_div(seg_size, nr);
2356                 los      = obj->lo_stripes;
2357                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
2358                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
2359                 tgt_tmp  = offset;
2360                 do_div(tgt_tmp, seg_size);
2361                 tgt0_idx = do_div(tgt_tmp,  nr);
2362                 tgt_idx  = (tgt0_idx + rank) % nr;
2363
2364                 if (tgt_idx < tgt0_idx)
2365                         /*
2366                          * Wrap around.
2367                          *
2368                          * Last segment has unusual length due to division
2369                          * rounding.
2370                          */
2371                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
2372                 else
2373                         hash_adj = 0;
2374
2375                 hash_adj += rank * seg_size;
2376
2377                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
2378                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
2379                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
2380
2381                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
2382                 rid = obj->lo_stripes[tgt_idx].ls_fid;
2383                 tgt = lmv_get_target(lmv, los[tgt_idx].ls_mds);
2384
2385                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
2386                        PFID(&rid), (unsigned long)offset, tgt_idx);
2387         } else
2388                 tgt = lmv_find_target(lmv, &rid);
2389
2390         if (IS_ERR(tgt))
2391                 GOTO(cleanup, rc = PTR_ERR(tgt));
2392
2393         rc = md_readpage(tgt->ltd_exp, &rid, oc, offset, page, request);
2394         if (rc)
2395                 GOTO(cleanup, rc);
2396         if (obj) {
2397                 dp = cfs_kmap(page);
2398
2399                 lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2400                 lmv_hash_adjust(&dp->ldp_hash_end,   hash_adj);
2401                 LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64);
2402
2403                 for (ent = lu_dirent_start(dp); ent != NULL;
2404                      ent = lu_dirent_next(ent))
2405                         lmv_hash_adjust(&ent->lde_hash, hash_adj);
2406
2407                 if (tgt0_idx != nr - 1) {
2408                         __u64 end;
2409
2410                         end = le64_to_cpu(dp->ldp_hash_end);
2411                         if (end == DIR_END_OFF) {
2412                                 dp->ldp_hash_end = cpu_to_le32(seg_size *
2413                                                                (tgt0_idx + 1));
2414                                 CDEBUG(D_INODE,
2415                                        ""DFID" reset end "LPX64" tgt %d\n",
2416                                        PFID(&rid),
2417                                        (__u64)le64_to_cpu(dp->ldp_hash_end), tgt_idx);
2418                         }
2419                 }
2420                 cfs_kunmap(page);
2421         }
2422         EXIT;
2423 cleanup:
2424         if (obj)
2425                 lmv_object_put_unlock(obj);
2426         return rc;
2427 }
2428
2429 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2430                       struct ptlrpc_request **request)
2431 {
2432         struct obd_device       *obd = exp->exp_obd;
2433         struct lmv_obd          *lmv = &obd->u.lmv;
2434         struct lmv_tgt_desc     *tgt = NULL;
2435         struct lmv_object       *obj;
2436         int                      rc;
2437         int                      sidx;
2438         int                      loop = 0;
2439         ENTRY;
2440
2441         rc = lmv_check_connect(obd);
2442         if (rc)
2443                 RETURN(rc);
2444
2445 repeat:
2446         ++loop;
2447         LASSERT(loop <= 2);
2448         LASSERT(op_data->op_namelen != 0);
2449
2450         obj = lmv_object_find(obd, &op_data->op_fid1);
2451         if (obj) {
2452                 sidx = raw_name2idx(obj->lo_hashtype,
2453                                     obj->lo_objcount,
2454                                     op_data->op_name,
2455                                     op_data->op_namelen);
2456                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
2457                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2458                 tgt = lmv_get_target(lmv,
2459                                      obj->lo_stripes[sidx].ls_mds);
2460                 lmv_object_put(obj);
2461                 CDEBUG(D_INODE, "UNLINK '%*s' in "DFID" -> %u\n",
2462                        op_data->op_namelen, op_data->op_name,
2463                        PFID(&op_data->op_fid1), sidx);
2464         }
2465
2466         if (tgt == NULL) {
2467                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2468                 if (IS_ERR(tgt))
2469                         RETURN(PTR_ERR(tgt));
2470                 op_data->op_bias |= MDS_CHECK_SPLIT;
2471         }
2472
2473         op_data->op_fsuid = cfs_curproc_fsuid();
2474         op_data->op_fsgid = cfs_curproc_fsgid();
2475         op_data->op_cap = cfs_curproc_cap_pack();
2476
2477         /*
2478          * If child's fid is given, cancel unused locks for it if it is from
2479          * another export than parent.
2480          *
2481          * LOOKUP lock for child (fid3) should also be cancelled on parent
2482          * tgt_tgt in mdc_unlink().
2483          */
2484         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2485
2486         /*
2487          * Cancel FULL locks on child (fid3).
2488          */
2489         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2490                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2491
2492         if (rc == 0)
2493                 rc = md_unlink(tgt->ltd_exp, op_data, request);
2494
2495         if (rc == -ERESTART) {
2496                 LASSERT(*request != NULL);
2497                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2498                           "Got -ERESTART during unlink!\n");
2499                 ptlrpc_req_finished(*request);
2500                 *request = NULL;
2501
2502                 /*
2503                  * Directory got split. Time to update local object and repeat
2504                  * the request with proper MDS.
2505                  */
2506                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2507                 if (rc == 0)
2508                         goto repeat;
2509         }
2510         RETURN(rc);
2511 }
2512
2513 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2514 {
2515         int        rc = 0;
2516
2517         switch (stage) {
2518         case OBD_CLEANUP_EARLY:
2519                 /* XXX: here should be calling obd_precleanup() down to
2520                  * stack. */
2521                 break;
2522         case OBD_CLEANUP_EXPORTS:
2523                 rc = obd_llog_finish(obd, 0);
2524                 if (rc != 0)
2525                         CERROR("failed to cleanup llogging subsystems\n");
2526                 break;
2527         default:
2528                 break;
2529         }
2530         RETURN(rc);
2531 }
2532
2533 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2534                         void *key, __u32 *vallen, void *val,
2535                         struct lov_stripe_md *lsm)
2536 {
2537         struct obd_device       *obd;
2538         struct lmv_obd          *lmv;
2539         int                      rc = 0;
2540         ENTRY;
2541
2542         obd = class_exp2obd(exp);
2543         if (obd == NULL) {
2544                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2545                        exp->exp_handle.h_cookie);
2546                 RETURN(-EINVAL);
2547         }
2548
2549         lmv = &obd->u.lmv;
2550         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2551                 struct lmv_tgt_desc *tgts;
2552                 int i;
2553
2554                 rc = lmv_check_connect(obd);
2555                 if (rc)
2556                         RETURN(rc);
2557
2558                 LASSERT(*vallen == sizeof(__u32));
2559                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2560                      i++, tgts++) {
2561
2562                         /*
2563                          * All tgts should be connected when this gets called.
2564                          */
2565                         if (!tgts || !tgts->ltd_exp) {
2566                                 CERROR("target not setup?\n");
2567                                 continue;
2568                         }
2569
2570                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2571                                           vallen, val, NULL))
2572                                 RETURN(0);
2573                 }
2574                 RETURN(-EINVAL);
2575         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2576                 rc = lmv_check_connect(obd);
2577                 if (rc)
2578                         RETURN(rc);
2579
2580                 /*
2581                  * Forwarding this request to first MDS, it should know LOV
2582                  * desc.
2583                  */
2584                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2585                                   vallen, val, NULL);
2586                 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2587                         exp->exp_connect_flags =
2588                         ((struct obd_connect_data *)val)->ocd_connect_flags;
2589                 }
2590                 RETURN(rc);
2591         }
2592
2593         CDEBUG(D_IOCTL, "Invalid key\n");
2594         RETURN(-EINVAL);
2595 }
2596
2597 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2598                        void *key, obd_count vallen, void *val,
2599                        struct ptlrpc_request_set *set)
2600 {
2601         struct lmv_tgt_desc    *tgt;
2602         struct obd_device      *obd;
2603         struct lmv_obd         *lmv;
2604         int rc = 0;
2605         ENTRY;
2606
2607         obd = class_exp2obd(exp);
2608         if (obd == NULL) {
2609                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2610                        exp->exp_handle.h_cookie);
2611                 RETURN(-EINVAL);
2612         }
2613         lmv = &obd->u.lmv;
2614
2615         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2616             KEY_IS(KEY_INIT_RECOV_BACKUP)) {
2617                 int i, err = 0;
2618
2619                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2620                         tgt = &lmv->tgts[i];
2621
2622                         if (!tgt->ltd_exp)
2623                                 continue;
2624
2625                         err = obd_set_info_async(tgt->ltd_exp,
2626                                                  keylen, key, vallen, val, set);
2627                         if (err && rc == 0)
2628                                 rc = err;
2629                 }
2630
2631                 RETURN(rc);
2632         }
2633
2634         RETURN(-EINVAL);
2635 }
2636
2637 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2638                struct lov_stripe_md *lsm)
2639 {
2640         struct obd_device         *obd = class_exp2obd(exp);
2641         struct lmv_obd            *lmv = &obd->u.lmv;
2642         struct lmv_stripe_md      *meap;
2643         struct lmv_stripe_md      *lsmp;
2644         int                        mea_size;
2645         int                        i;
2646         ENTRY;
2647
2648         mea_size = lmv_get_easize(lmv);
2649         if (!lmmp)
2650                 RETURN(mea_size);
2651
2652         if (*lmmp && !lsm) {
2653                 OBD_FREE(*lmmp, mea_size);
2654                 *lmmp = NULL;
2655                 RETURN(0);
2656         }
2657
2658         if (*lmmp == NULL) {
2659                 OBD_ALLOC(*lmmp, mea_size);
2660                 if (*lmmp == NULL)
2661                         RETURN(-ENOMEM);
2662         }
2663
2664         if (!lsm)
2665                 RETURN(mea_size);
2666
2667         lsmp = (struct lmv_stripe_md *)lsm;
2668         meap = (struct lmv_stripe_md *)*lmmp;
2669
2670         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2671             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2672                 RETURN(-EINVAL);
2673
2674         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2675         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2676         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2677
2678         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2679                 meap->mea_ids[i] = meap->mea_ids[i];
2680                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2681         }
2682
2683         RETURN(mea_size);
2684 }
2685
2686 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2687                  struct lov_mds_md *lmm, int lmm_size)
2688 {
2689         struct obd_device          *obd = class_exp2obd(exp);
2690         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2691         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2692         struct lmv_obd             *lmv = &obd->u.lmv;
2693         int                         mea_size;
2694         int                         i;
2695         __u32                       magic;
2696         ENTRY;
2697
2698         mea_size = lmv_get_easize(lmv);
2699         if (lsmp == NULL)
2700                 return mea_size;
2701
2702         if (*lsmp != NULL && lmm == NULL) {
2703                 OBD_FREE(*tmea, mea_size);
2704                 *lsmp = NULL;
2705                 RETURN(0);
2706         }
2707
2708         LASSERT(mea_size == lmm_size);
2709
2710         OBD_ALLOC(*tmea, mea_size);
2711         if (*tmea == NULL)
2712                 RETURN(-ENOMEM);
2713
2714         if (!lmm)
2715                 RETURN(mea_size);
2716
2717         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2718             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2719             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2720         {
2721                 magic = le32_to_cpu(mea->mea_magic);
2722         } else {
2723                 /*
2724                  * Old mea is not handled here.
2725                  */
2726                 CERROR("Old not supportable EA is found\n");
2727                 LBUG();
2728         }
2729
2730         (*tmea)->mea_magic = magic;
2731         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2732         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2733
2734         for (i = 0; i < (*tmea)->mea_count; i++) {
2735                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2736                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2737         }
2738         RETURN(mea_size);
2739 }
2740
2741 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2742                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2743                              int flags, void *opaque)
2744 {
2745         struct obd_device       *obd = exp->exp_obd;
2746         struct lmv_obd          *lmv = &obd->u.lmv;
2747         int                      rc = 0;
2748         int                      err;
2749         int                      i;
2750         ENTRY;
2751
2752         LASSERT(fid != NULL);
2753
2754         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2755                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2756                         continue;
2757
2758                 err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
2759                                        policy, mode, flags, opaque);
2760                 if (!rc)
2761                         rc = err;
2762         }
2763         RETURN(rc);
2764 }
2765
2766 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2767                       __u32 *bits)
2768 {
2769         struct obd_device       *obd = exp->exp_obd;
2770         struct lmv_obd          *lmv = &obd->u.lmv;
2771         int                      rc;
2772         ENTRY;
2773
2774         rc =  md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data, bits);
2775         RETURN(rc);
2776 }
2777
2778 ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
2779                            const struct lu_fid *fid, ldlm_type_t type,
2780                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2781                            struct lustre_handle *lockh)
2782 {
2783         struct obd_device       *obd = exp->exp_obd;
2784         struct lmv_obd          *lmv = &obd->u.lmv;
2785         ldlm_mode_t              rc;
2786         int                      i;
2787         ENTRY;
2788
2789         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2790
2791         /*
2792          * With CMD every object can have two locks in different namespaces:
2793          * lookup lock in space of mds storing direntry and update/open lock in
2794          * space of mds storing inode. Thus we check all targets, not only that
2795          * one fid was created in.
2796          */
2797         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2798                 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2799                                    type, policy, mode, lockh);
2800                 if (rc)
2801                         RETURN(rc);
2802         }
2803
2804         RETURN(0);
2805 }
2806
2807 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2808                       struct obd_export *dt_exp, struct obd_export *md_exp,
2809                       struct lustre_md *md)
2810 {
2811         struct obd_device       *obd = exp->exp_obd;
2812         struct lmv_obd          *lmv = &obd->u.lmv;
2813         int                      rc;
2814         ENTRY;
2815         rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
2816         RETURN(rc);
2817 }
2818
2819 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2820 {
2821         struct obd_device       *obd = exp->exp_obd;
2822         struct lmv_obd          *lmv = &obd->u.lmv;
2823         ENTRY;
2824
2825         if (md->mea)
2826                 obd_free_memmd(exp, (void *)&md->mea);
2827         RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2828 }
2829
2830 int lmv_set_open_replay_data(struct obd_export *exp,
2831                              struct obd_client_handle *och,
2832                              struct ptlrpc_request *open_req)
2833 {
2834         struct obd_device       *obd = exp->exp_obd;
2835         struct lmv_obd          *lmv = &obd->u.lmv;
2836         struct lmv_tgt_desc     *tgt;
2837         ENTRY;
2838
2839         tgt = lmv_find_target(lmv, &och->och_fid);
2840         if (IS_ERR(tgt))
2841                 RETURN(PTR_ERR(tgt));
2842
2843         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2844 }
2845
2846 int lmv_clear_open_replay_data(struct obd_export *exp,
2847                                struct obd_client_handle *och)
2848 {
2849         struct obd_device       *obd = exp->exp_obd;
2850         struct lmv_obd          *lmv = &obd->u.lmv;
2851         struct lmv_tgt_desc     *tgt;
2852         ENTRY;
2853
2854         tgt = lmv_find_target(lmv, &och->och_fid);
2855         if (IS_ERR(tgt))
2856                 RETURN(PTR_ERR(tgt));
2857
2858         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2859 }
2860
2861 static int lmv_get_remote_perm(struct obd_export *exp,
2862                                const struct lu_fid *fid,
2863                                struct obd_capa *oc, __u32 suppgid,
2864                                struct ptlrpc_request **request)
2865 {
2866         struct obd_device       *obd = exp->exp_obd;
2867         struct lmv_obd          *lmv = &obd->u.lmv;
2868         struct lmv_tgt_desc     *tgt;
2869         int                      rc;
2870         ENTRY;
2871
2872         rc = lmv_check_connect(obd);
2873         if (rc)
2874                 RETURN(rc);
2875
2876         tgt = lmv_find_target(lmv, fid);
2877         if (IS_ERR(tgt))
2878                 RETURN(PTR_ERR(tgt));
2879
2880         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2881         RETURN(rc);
2882 }
2883
2884 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2885                           renew_capa_cb_t cb)
2886 {
2887         struct obd_device       *obd = exp->exp_obd;
2888         struct lmv_obd          *lmv = &obd->u.lmv;
2889         struct lmv_tgt_desc     *tgt;
2890         int                      rc;
2891         ENTRY;
2892
2893         rc = lmv_check_connect(obd);
2894         if (rc)
2895                 RETURN(rc);
2896
2897         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2898         if (IS_ERR(tgt))
2899                 RETURN(PTR_ERR(tgt));
2900
2901         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2902         RETURN(rc);
2903 }
2904
2905 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2906                     const struct req_msg_field *field, struct obd_capa **oc)
2907 {
2908         struct obd_device *obd = exp->exp_obd;
2909         struct lmv_obd *lmv = &obd->u.lmv;
2910         int rc;
2911
2912         ENTRY;
2913         rc = md_unpack_capa(lmv->tgts[0].ltd_exp, req, field, oc);
2914         RETURN(rc);
2915 }
2916
2917 int lmv_intent_getattr_async(struct obd_export *exp,
2918                              struct md_enqueue_info *minfo,
2919                              struct ldlm_enqueue_info *einfo)
2920 {
2921         struct md_op_data       *op_data = &minfo->mi_data;
2922         struct obd_device       *obd = exp->exp_obd;
2923         struct lmv_obd          *lmv = &obd->u.lmv;
2924         struct lmv_object       *obj;
2925         struct lmv_tgt_desc     *tgt;
2926         int                      rc;
2927         int                      sidx;
2928         ENTRY;
2929
2930         rc = lmv_check_connect(obd);
2931         if (rc)
2932                 RETURN(rc);
2933
2934         if (!fid_is_sane(&op_data->op_fid2)) {
2935                 obj = lmv_object_find(obd, &op_data->op_fid1);
2936                 if (obj && op_data->op_namelen) {
2937                         sidx = raw_name2idx(obj->lo_hashtype,
2938                                             obj->lo_objcount,
2939                                             (char *)op_data->op_name,
2940                                             op_data->op_namelen);
2941                         op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2942                         tgt = lmv_get_target(lmv,
2943                                              obj->lo_stripes[sidx].ls_mds);
2944                         CDEBUG(D_INODE,
2945                                "Choose slave dir ("DFID") -> mds #%d\n",
2946                                PFID(&op_data->op_fid1), tgt->ltd_idx);
2947                 } else {
2948                         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2949                 }
2950                 if (obj)
2951                         lmv_object_put(obj);
2952         } else {
2953                 op_data->op_fid1 = op_data->op_fid2;
2954                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2955                 op_data->op_bias = MDS_CROSS_REF;
2956                 /*
2957                  * Unfortunately, we have to lie to MDC/MDS to retrieve
2958                  * attributes llite needs.
2959                 */
2960                 if (minfo->mi_it.it_op & IT_LOOKUP)
2961                         minfo->mi_it.it_op = IT_GETATTR;
2962         }
2963
2964         if (IS_ERR(tgt))
2965                 RETURN(PTR_ERR(tgt));
2966
2967         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2968         RETURN(rc);
2969 }
2970
2971 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2972                         struct lu_fid *fid, __u32 *bits)
2973 {
2974         struct obd_device       *obd = exp->exp_obd;
2975         struct lmv_obd          *lmv = &obd->u.lmv;
2976         struct lmv_tgt_desc     *tgt;
2977         int                      rc;
2978         ENTRY;
2979
2980         rc = lmv_check_connect(obd);
2981         if (rc)
2982                 RETURN(rc);
2983
2984         tgt = lmv_find_target(lmv, fid);
2985         if (IS_ERR(tgt))
2986                 RETURN(PTR_ERR(tgt));
2987
2988         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2989         RETURN(rc);
2990 }
2991
2992
2993 struct obd_ops lmv_obd_ops = {
2994         .o_owner                = THIS_MODULE,
2995         .o_setup                = lmv_setup,
2996         .o_cleanup              = lmv_cleanup,
2997         .o_precleanup           = lmv_precleanup,
2998         .o_process_config       = lmv_process_config,
2999         .o_connect              = lmv_connect,
3000         .o_disconnect           = lmv_disconnect,
3001         .o_statfs               = lmv_statfs,
3002         .o_get_info             = lmv_get_info,
3003         .o_set_info_async       = lmv_set_info_async,
3004         .o_packmd               = lmv_packmd,
3005         .o_unpackmd             = lmv_unpackmd,
3006         .o_notify               = lmv_notify,
3007         .o_get_uuid             = lmv_get_uuid,
3008         .o_iocontrol            = lmv_iocontrol,
3009         .o_fid_delete           = lmv_fid_delete
3010 };
3011
3012 struct md_ops lmv_md_ops = {
3013         .m_getstatus            = lmv_getstatus,
3014         .m_change_cbdata        = lmv_change_cbdata,
3015         .m_close                = lmv_close,
3016         .m_create               = lmv_create,
3017         .m_done_writing         = lmv_done_writing,
3018         .m_enqueue              = lmv_enqueue,
3019         .m_getattr              = lmv_getattr,
3020         .m_getxattr             = lmv_getxattr,
3021         .m_getattr_name         = lmv_getattr_name,
3022         .m_intent_lock          = lmv_intent_lock,
3023         .m_link                 = lmv_link,
3024         .m_rename               = lmv_rename,
3025         .m_setattr              = lmv_setattr,
3026         .m_setxattr             = lmv_setxattr,
3027         .m_sync                 = lmv_sync,
3028         .m_readpage             = lmv_readpage,
3029         .m_unlink               = lmv_unlink,
3030         .m_init_ea_size         = lmv_init_ea_size,
3031         .m_cancel_unused        = lmv_cancel_unused,
3032         .m_set_lock_data        = lmv_set_lock_data,
3033         .m_lock_match           = lmv_lock_match,
3034         .m_get_lustre_md        = lmv_get_lustre_md,
3035         .m_free_lustre_md       = lmv_free_lustre_md,
3036         .m_set_open_replay_data = lmv_set_open_replay_data,
3037         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3038         .m_renew_capa           = lmv_renew_capa,
3039         .m_unpack_capa          = lmv_unpack_capa,
3040         .m_get_remote_perm      = lmv_get_remote_perm,
3041         .m_intent_getattr_async = lmv_intent_getattr_async,
3042         .m_revalidate_lock      = lmv_revalidate_lock
3043 };
3044
3045 static quota_interface_t *quota_interface;
3046 extern quota_interface_t lmv_quota_interface;
3047
3048 int __init lmv_init(void)
3049 {
3050         struct lprocfs_static_vars lvars;
3051         int                        rc;
3052
3053         lmv_object_cache = cfs_mem_cache_create("lmv_objects",
3054                                                 sizeof(struct lmv_object),
3055                                                 0, 0);
3056         if (!lmv_object_cache) {
3057                 CERROR("Error allocating lmv objects cache\n");
3058                 return -ENOMEM;
3059         }
3060
3061         lprocfs_lmv_init_vars(&lvars);
3062
3063         cfs_request_module("lquota");
3064         quota_interface = PORTAL_SYMBOL_GET(lmv_quota_interface);
3065         init_obd_quota_ops(quota_interface, &lmv_obd_ops);
3066
3067         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
3068                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
3069         if (rc) {
3070                 if (quota_interface)
3071                         PORTAL_SYMBOL_PUT(lmv_quota_interface);
3072                 cfs_mem_cache_destroy(lmv_object_cache);
3073         }
3074
3075         return rc;
3076 }
3077
3078 #ifdef __KERNEL__
3079 static void lmv_exit(void)
3080 {
3081         if (quota_interface)
3082                 PORTAL_SYMBOL_PUT(lmv_quota_interface);
3083
3084         class_unregister_type(LUSTRE_LMV_NAME);
3085
3086         LASSERTF(cfs_atomic_read(&lmv_object_count) == 0,
3087                  "Can't free lmv objects cache, %d object(s) busy\n",
3088                  cfs_atomic_read(&lmv_object_count));
3089         cfs_mem_cache_destroy(lmv_object_cache);
3090 }
3091
3092 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3093 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3094 MODULE_LICENSE("GPL");
3095
3096 module_init(lmv_init);
3097 module_exit(lmv_exit);
3098 #endif