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 (c) 2004, 2010, Oracle and/or its affiliates. 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_log.h>
  56 #include <obd_support.h>
  57 #include <lustre_lib.h>
  58 #include <lustre_net.h>
  59 #include <obd_class.h>
  60 #include <lprocfs_status.h>
  61 #include <lustre_lite.h>
  62 #include <lustre_fid.h>
  63 #include "lmv_internal.h"
  64
  65 /* object cache. */
  66 cfs_mem_cache_t *lmv_object_cache;
  67 cfs_atomic_t lmv_object_count = CFS_ATOMIC_INIT(0);
  68
  69 static void lmv_activate_target(struct lmv_obd *lmv,
  70                                 struct lmv_tgt_desc *tgt,
  71                                 int activate)
  72 {
  73         if (tgt->ltd_active == activate)
  74                 return;
  75
  76         tgt->ltd_active = activate;
  77         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  78 }
  79
  80 /**
  81  * Error codes:
  82  *
  83  *  -EINVAL  : UUID can't be found in the LMV's target list
  84  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  85  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  86  */
  87 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  88                               int activate)
  89 {
  90         struct lmv_tgt_desc    *tgt;
  91         struct obd_device      *obd;
  92         int                     i;
  93         int                     rc = 0;
  94         ENTRY;
  95
  96         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  97                lmv, uuid->uuid, activate);
  98
  99         cfs_spin_lock(&lmv->lmv_lock);
 100         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 101                 if (tgt->ltd_exp == NULL)
 102                         continue;
 103
 104                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n",
 105                        i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 106
 107                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 108                         break;
 109         }
 110
 111         if (i == lmv->desc.ld_tgt_count)
 112                 GOTO(out_lmv_lock, rc = -EINVAL);
 113
 114         obd = class_exp2obd(tgt->ltd_exp);
 115         if (obd == NULL)
 116                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 117
 118         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 119                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 120                obd->obd_type->typ_name, i);
 121         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 122
 123         if (tgt->ltd_active == activate) {
 124                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 125                        activate ? "" : "in");
 126                 GOTO(out_lmv_lock, rc);
 127         }
 128
 129         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 130                activate ? "" : "in");
 131         lmv_activate_target(lmv, tgt, activate);
 132         EXIT;
 133
 134  out_lmv_lock:
 135         cfs_spin_unlock(&lmv->lmv_lock);
 136         return rc;
 137 }
 138
 139 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
 140                             struct obd_connect_data *data)
 141 {
 142         struct lmv_tgt_desc    *tgt;
 143         int                     i;
 144         ENTRY;
 145
 146         LASSERT(data != NULL);
 147
 148         cfs_spin_lock(&lmv->lmv_lock);
 149         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 150                 if (tgt->ltd_exp == NULL)
 151                         continue;
 152
 153                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 154                         lmv->datas[tgt->ltd_idx] = *data;
 155                         break;
 156                 }
 157         }
 158         cfs_spin_unlock(&lmv->lmv_lock);
 159         RETURN(0);
 160 }
 161
 162 struct obd_uuid *lmv_get_uuid(struct obd_export *exp) {
 163         struct obd_device *obd = exp->exp_obd;
 164         struct lmv_obd *lmv = &obd->u.lmv;
 165         return obd_get_uuid(lmv->tgts[0].ltd_exp);
 166 }
 167
 168 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 169                       enum obd_notify_event ev, void *data)
 170 {
 171         struct obd_connect_data *conn_data;
 172         struct lmv_obd          *lmv = &obd->u.lmv;
 173         struct obd_uuid         *uuid;
 174         int                      rc = 0;
 175         ENTRY;
 176
 177         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 178                 CERROR("unexpected notification of %s %s!\n",
 179                        watched->obd_type->typ_name,
 180                        watched->obd_name);
 181                 RETURN(-EINVAL);
 182         }
 183
 184         uuid = &watched->u.cli.cl_target_uuid;
 185         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 186                 /*
 187                  * Set MDC as active before notifying the observer, so the
 188                  * observer can use the MDC normally.
 189                  */
 190                 rc = lmv_set_mdc_active(lmv, uuid,
 191                                         ev == OBD_NOTIFY_ACTIVE);
 192                 if (rc) {
 193                         CERROR("%sactivation of %s failed: %d\n",
 194                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 195                                uuid->uuid, rc);
 196                         RETURN(rc);
 197                 }
 198         } else if (ev == OBD_NOTIFY_OCD) {
 199                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 200
 201                 /*
 202                  * Set connect data to desired target, update exp_connect_flags.
 203                  */
 204                 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
 205                 if (rc) {
 206                         CERROR("can't set connect data to target %s, rc %d\n",
 207                                uuid->uuid, rc);
 208                         RETURN(rc);
 209                 }
 210
 211                 /*
 212                  * XXX: Make sure that ocd_connect_flags from all targets are
 213                  * the same. Otherwise one of MDTs runs wrong version or
 214                  * something like this.  --umka
 215                  */
 216                 obd->obd_self_export->exp_connect_flags =
 217                         conn_data->ocd_connect_flags;
 218         }
 219 #if 0
 220         else if (ev == OBD_NOTIFY_DISCON) {
 221                 /*
 222                  * For disconnect event, flush fld cache for failout MDS case.
 223                  */
 224                 fld_client_flush(&lmv->lmv_fld);
 225         }
 226 #endif
 227         /*
 228          * Pass the notification up the chain.
 229          */
 230         if (obd->obd_observer)
 231                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 232
 233         RETURN(rc);
 234 }
 235
 236 /**
 237  * This is fake connect function. Its purpose is to initialize lmv and say
 238  * caller that everything is okay. Real connection will be performed later.
 239  */
 240 static int lmv_connect(const struct lu_env *env,
 241                        struct obd_export **exp, struct obd_device *obd,
 242                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 243                        void *localdata)
 244 {
 245 #ifdef __KERNEL__
 246         struct proc_dir_entry *lmv_proc_dir;
 247 #endif
 248         struct lmv_obd        *lmv = &obd->u.lmv;
 249         struct lustre_handle  conn = { 0 };
 250         int                    rc = 0;
 251         ENTRY;
 252
 253         /*
 254          * We don't want to actually do the underlying connections more than
 255          * once, so keep track.
 256          */
 257         lmv->refcount++;
 258         if (lmv->refcount > 1) {
 259                 *exp = NULL;
 260                 RETURN(0);
 261         }
 262
 263         rc = class_connect(&conn, obd, cluuid);
 264         if (rc) {
 265                 CERROR("class_connection() returned %d\n", rc);
 266                 RETURN(rc);
 267         }
 268
 269         *exp = class_conn2export(&conn);
 270         class_export_get(*exp);
 271
 272         lmv->exp = *exp;
 273         lmv->connected = 0;
 274         lmv->cluuid = *cluuid;
 275
 276         if (data)
 277                 lmv->conn_data = *data;
 278
 279 #ifdef __KERNEL__
 280         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 281                                         NULL, NULL);
 282         if (IS_ERR(lmv_proc_dir)) {
 283                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 284                        obd->obd_type->typ_name, obd->obd_name);
 285                 lmv_proc_dir = NULL;
 286         }
 287 #endif
 288
 289         /*
 290          * All real clients should perform actual connection right away, because
 291          * it is possible, that LMV will not have opportunity to connect targets
 292          * and MDC stuff will be called directly, for instance while reading
 293          * ../mdc/../kbytesfree procfs file, etc.
 294          */
 295         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 296                 rc = lmv_check_connect(obd);
 297
 298 #ifdef __KERNEL__
 299         if (rc) {
 300                 if (lmv_proc_dir)
 301                         lprocfs_remove(&lmv_proc_dir);
 302         }
 303 #endif
 304
 305         RETURN(rc);
 306 }
 307
 308 static void lmv_set_timeouts(struct obd_device *obd)
 309 {
 310         struct lmv_tgt_desc   *tgts;
 311         struct lmv_obd        *lmv;
 312         int                    i;
 313
 314         lmv = &obd->u.lmv;
 315         if (lmv->server_timeout == 0)
 316                 return;
 317
 318         if (lmv->connected == 0)
 319                 return;
 320
 321         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 322                 if (tgts->ltd_exp == NULL)
 323                         continue;
 324
 325                 obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS),
 326                                    KEY_INTERMDS, 0, NULL, NULL);
 327         }
 328 }
 329
 330 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 331                             int def_easize, int cookiesize)
 332 {
 333         struct obd_device   *obd = exp->exp_obd;
 334         struct lmv_obd      *lmv = &obd->u.lmv;
 335         int                  i;
 336         int                  rc = 0;
 337         int                  change = 0;
 338         ENTRY;
 339
 340         if (lmv->max_easize < easize) {
 341                 lmv->max_easize = easize;
 342                 change = 1;
 343         }
 344         if (lmv->max_def_easize < def_easize) {
 345                 lmv->max_def_easize = def_easize;
 346                 change = 1;
 347         }
 348         if (lmv->max_cookiesize < cookiesize) {
 349                 lmv->max_cookiesize = cookiesize;
 350                 change = 1;
 351         }
 352         if (change == 0)
 353                 RETURN(0);
 354
 355         if (lmv->connected == 0)
 356                 RETURN(0);
 357
 358         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 359                 if (lmv->tgts[i].ltd_exp == NULL) {
 360                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 361                         continue;
 362                 }
 363
 364                 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
 365                                      cookiesize);
 366                 if (rc) {
 367                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 368                                "error %d.\n", i, rc);
 369                         break;
 370                 }
 371         }
 372         RETURN(rc);
 373 }
 374
 375 #define MAX_STRING_SIZE 128
 376
 377 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 378 {
 379 #ifdef __KERNEL__
 380         struct proc_dir_entry   *lmv_proc_dir;
 381 #endif
 382         struct lmv_obd          *lmv = &obd->u.lmv;
 383         struct obd_uuid         *cluuid = &lmv->cluuid;
 384         struct obd_connect_data *mdc_data = NULL;
 385         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 386         struct obd_device       *mdc_obd;
 387         struct obd_export       *mdc_exp;
 388         struct lu_fld_target     target;
 389         int                      rc;
 390         ENTRY;
 391
 392         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 393                                         &obd->obd_uuid);
 394         if (!mdc_obd) {
 395                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 396                 RETURN(-EINVAL);
 397         }
 398
 399         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 400                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 401                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 402                 cluuid->uuid);
 403
 404         if (!mdc_obd->obd_set_up) {
 405                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 406                 RETURN(-EINVAL);
 407         }
 408
 409         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 410                          &lmv->conn_data, NULL);
 411         if (rc) {
 412                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 413                 RETURN(rc);
 414         }
 415
 416         /*
 417          * Init fid sequence client for this mdc and add new fld target.
 418          */
 419         rc = obd_fid_init(mdc_exp);
 420         if (rc)
 421                 RETURN(rc);
 422
 423         target.ft_srv = NULL;
 424         target.ft_exp = mdc_exp;
 425         target.ft_idx = tgt->ltd_idx;
 426
 427         fld_client_add_target(&lmv->lmv_fld, &target);
 428
 429         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 430
 431         rc = obd_register_observer(mdc_obd, obd);
 432         if (rc) {
 433                 obd_disconnect(mdc_exp);
 434                 CERROR("target %s register_observer error %d\n",
 435                        tgt->ltd_uuid.uuid, rc);
 436                 RETURN(rc);
 437         }
 438
 439         if (obd->obd_observer) {
 440                 /*
 441                  * Tell the observer about the new target.
 442                  */
 443                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 444                                 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
 445                 if (rc) {
 446                         obd_disconnect(mdc_exp);
 447                         RETURN(rc);
 448                 }
 449         }
 450
 451         tgt->ltd_active = 1;
 452         tgt->ltd_exp = mdc_exp;
 453         lmv->desc.ld_active_tgt_count++;
 454
 455         /*
 456          * Copy connect data, it may be used later.
 457          */
 458         lmv->datas[tgt->ltd_idx] = *mdc_data;
 459
 460         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 461                         lmv->max_def_easize, lmv->max_cookiesize);
 462
 463         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 464                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 465                 cfs_atomic_read(&obd->obd_refcount));
 466
 467 #ifdef __KERNEL__
 468         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 469         if (lmv_proc_dir) {
 470                 struct proc_dir_entry *mdc_symlink;
 471
 472                 LASSERT(mdc_obd->obd_type != NULL);
 473                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 474                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 475                                                   lmv_proc_dir,
 476                                                   "../../../%s/%s",
 477                                                   mdc_obd->obd_type->typ_name,
 478                                                   mdc_obd->obd_name);
 479                 if (mdc_symlink == NULL) {
 480                         CERROR("Could not register LMV target "
 481                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 482                                obd->obd_type->typ_name, obd->obd_name,
 483                                mdc_obd->obd_name);
 484                         lprocfs_remove(&lmv_proc_dir);
 485                         lmv_proc_dir = NULL;
 486                 }
 487         }
 488 #endif
 489         RETURN(0);
 490 }
 491
 492 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 493 {
 494         struct lmv_obd      *lmv = &obd->u.lmv;
 495         struct lmv_tgt_desc *tgt;
 496         int                  rc = 0;
 497         ENTRY;
 498
 499         CDEBUG(D_CONFIG, "Target uuid: %s.\n", tgt_uuid->uuid);
 500
 501         lmv_init_lock(lmv);
 502
 503         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 504                 lmv_init_unlock(lmv);
 505                 CERROR("Can't add %s, LMV module compiled for %d MDCs. "
 506                        "That many MDCs already configured.\n",
 507                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 508                 RETURN(-EINVAL);
 509         }
 510         if (lmv->desc.ld_tgt_count == 0) {
 511                 struct obd_device *mdc_obd;
 512
 513                 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
 514                                                 &obd->obd_uuid);
 515                 if (!mdc_obd) {
 516                         lmv_init_unlock(lmv);
 517                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 518                         RETURN(-EINVAL);
 519                 }
 520
 521                 rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, NULL);
 522                 if (rc) {
 523                         lmv_init_unlock(lmv);
 524                         CERROR("lmv failed to setup llogging subsystems\n");
 525                 }
 526         }
 527         cfs_spin_lock(&lmv->lmv_lock);
 528         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 529         tgt->ltd_uuid = *tgt_uuid;
 530         cfs_spin_unlock(&lmv->lmv_lock);
 531
 532         if (lmv->connected) {
 533                 rc = lmv_connect_mdc(obd, tgt);
 534                 if (rc) {
 535                         cfs_spin_lock(&lmv->lmv_lock);
 536                         lmv->desc.ld_tgt_count--;
 537                         memset(tgt, 0, sizeof(*tgt));
 538                         cfs_spin_unlock(&lmv->lmv_lock);
 539                 } else {
 540                         int easize = sizeof(struct lmv_stripe_md) +
 541                                      lmv->desc.ld_tgt_count *
 542                                      sizeof(struct lu_fid);
 543                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 544                 }
 545         }
 546
 547         lmv_init_unlock(lmv);
 548         RETURN(rc);
 549 }
 550
 551 int lmv_check_connect(struct obd_device *obd)
 552 {
 553         struct lmv_obd       *lmv = &obd->u.lmv;
 554         struct lmv_tgt_desc  *tgt;
 555         int                   i;
 556         int                   rc;
 557         int                   easize;
 558         ENTRY;
 559
 560         if (lmv->connected)
 561                 RETURN(0);
 562
 563         lmv_init_lock(lmv);
 564         if (lmv->connected) {
 565                 lmv_init_unlock(lmv);
 566                 RETURN(0);
 567         }
 568
 569         if (lmv->desc.ld_tgt_count == 0) {
 570                 CERROR("%s: no targets configured.\n", obd->obd_name);
 571                 RETURN(-EINVAL);
 572         }
 573
 574         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 575                lmv->cluuid.uuid, obd->obd_name);
 576
 577         LASSERT(lmv->tgts != NULL);
 578
 579         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 580                 rc = lmv_connect_mdc(obd, tgt);
 581                 if (rc)
 582                         GOTO(out_disc, rc);
 583         }
 584
 585         lmv_set_timeouts(obd);
 586         class_export_put(lmv->exp);
 587         lmv->connected = 1;
 588         easize = lmv_get_easize(lmv);
 589         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 590         lmv_init_unlock(lmv);
 591         RETURN(0);
 592
 593  out_disc:
 594         while (i-- > 0) {
 595                 int rc2;
 596                 --tgt;
 597                 tgt->ltd_active = 0;
 598                 if (tgt->ltd_exp) {
 599                         --lmv->desc.ld_active_tgt_count;
 600                         rc2 = obd_disconnect(tgt->ltd_exp);
 601                         if (rc2) {
 602                                 CERROR("LMV target %s disconnect on "
 603                                        "MDC idx %d: error %d\n",
 604                                        tgt->ltd_uuid.uuid, i, rc2);
 605                         }
 606                 }
 607         }
 608         class_disconnect(lmv->exp);
 609         lmv_init_unlock(lmv);
 610         RETURN(rc);
 611 }
 612
 613 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 614 {
 615 #ifdef __KERNEL__
 616         struct proc_dir_entry  *lmv_proc_dir;
 617 #endif
 618         struct lmv_obd         *lmv = &obd->u.lmv;
 619         struct obd_device      *mdc_obd;
 620         int                     rc;
 621         ENTRY;
 622
 623         LASSERT(tgt != NULL);
 624         LASSERT(obd != NULL);
 625
 626         mdc_obd = class_exp2obd(tgt->ltd_exp);
 627
 628         if (mdc_obd) {
 629                 mdc_obd->obd_force = obd->obd_force;
 630                 mdc_obd->obd_fail = obd->obd_fail;
 631                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 632         }
 633
 634 #ifdef __KERNEL__
 635         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 636         if (lmv_proc_dir) {
 637                 struct proc_dir_entry *mdc_symlink;
 638
 639                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 640                 if (mdc_symlink) {
 641                         lprocfs_remove(&mdc_symlink);
 642                 } else {
 643                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 644                                obd->obd_type->typ_name, obd->obd_name,
 645                                mdc_obd->obd_name);
 646                 }
 647         }
 648 #endif
 649         rc = obd_fid_fini(tgt->ltd_exp);
 650         if (rc)
 651                 CERROR("Can't finanize fids factory\n");
 652
 653         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 654                tgt->ltd_exp->exp_obd->obd_name,
 655                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 656
 657         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 658         rc = obd_disconnect(tgt->ltd_exp);
 659         if (rc) {
 660                 if (tgt->ltd_active) {
 661                         CERROR("Target %s disconnect error %d\n",
 662                                tgt->ltd_uuid.uuid, rc);
 663                 }
 664         }
 665
 666         lmv_activate_target(lmv, tgt, 0);
 667         tgt->ltd_exp = NULL;
 668         RETURN(0);
 669 }
 670
 671 static int lmv_disconnect(struct obd_export *exp)
 672 {
 673         struct obd_device     *obd = class_exp2obd(exp);
 674 #ifdef __KERNEL__
 675         struct proc_dir_entry *lmv_proc_dir;
 676 #endif
 677         struct lmv_obd        *lmv = &obd->u.lmv;
 678         int                    rc;
 679         int                    i;
 680         ENTRY;
 681
 682         if (!lmv->tgts)
 683                 goto out_local;
 684
 685         /*
 686          * Only disconnect the underlying layers on the final disconnect.
 687          */
 688         lmv->refcount--;
 689         if (lmv->refcount != 0)
 690                 goto out_local;
 691
 692         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 693                 if (lmv->tgts[i].ltd_exp == NULL)
 694                         continue;
 695                 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
 696         }
 697
 698 #ifdef __KERNEL__
 699         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 700         if (lmv_proc_dir) {
 701                 lprocfs_remove(&lmv_proc_dir);
 702         } else {
 703                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 704                        obd->obd_type->typ_name, obd->obd_name);
 705         }
 706 #endif
 707
 708 out_local:
 709         /*
 710          * This is the case when no real connection is established by
 711          * lmv_check_connect().
 712          */
 713         if (!lmv->connected)
 714                 class_export_put(exp);
 715         rc = class_disconnect(exp);
 716         if (lmv->refcount == 0)
 717                 lmv->connected = 0;
 718         RETURN(rc);
 719 }
 720
 721 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 722                          int len, void *karg, void *uarg)
 723 {
 724         struct obd_device    *obddev = class_exp2obd(exp);
 725         struct lmv_obd       *lmv = &obddev->u.lmv;
 726         int                   i = 0;
 727         int                   rc = 0;
 728         int                   set = 0;
 729         int                   count = lmv->desc.ld_tgt_count;
 730         ENTRY;
 731
 732         if (count == 0)
 733                 RETURN(-ENOTTY);
 734
 735         switch (cmd) {
 736         case IOC_OBD_STATFS: {
 737                 struct obd_ioctl_data *data = karg;
 738                 struct obd_device *mdc_obd;
 739                 struct obd_statfs stat_buf = {0};
 740                 __u32 index;
 741
 742                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 743                 if ((index >= count))
 744                         RETURN(-ENODEV);
 745
 746                 if (!lmv->tgts[index].ltd_active)
 747                         RETURN(-ENODATA);
 748
 749                 mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp);
 750                 if (!mdc_obd)
 751                         RETURN(-EINVAL);
 752
 753                 /* copy UUID */
 754                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 755                                      min((int) data->ioc_plen2,
 756                                          (int) sizeof(struct obd_uuid))))
 757                         RETURN(-EFAULT);
 758
 759                 rc = obd_statfs(mdc_obd, &stat_buf,
 760                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 761                                 0);
 762                 if (rc)
 763                         RETURN(rc);
 764                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 765                                      min((int) data->ioc_plen1,
 766                                          (int) sizeof(stat_buf))))
 767                         RETURN(-EFAULT);
 768                 break;
 769         }
 770         case OBD_IOC_QUOTACTL: {
 771                 struct if_quotactl *qctl = karg;
 772                 struct lmv_tgt_desc *tgt = NULL;
 773                 struct obd_quotactl *oqctl;
 774
 775                 if (qctl->qc_valid == QC_MDTIDX) {
 776                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 777                                 RETURN(-EINVAL);
 778
 779                         tgt = &lmv->tgts[qctl->qc_idx];
 780                         if (!tgt->ltd_exp)
 781                                 RETURN(-EINVAL);
 782                 } else if (qctl->qc_valid == QC_UUID) {
 783                         for (i = 0; i < count; i++) {
 784                                 tgt = &lmv->tgts[i];
 785                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 786                                                      &qctl->obd_uuid))
 787                                         continue;
 788
 789                                 if (tgt->ltd_exp == NULL)
 790                                         RETURN(-EINVAL);
 791
 792                                 break;
 793                         }
 794                 } else {
 795                         RETURN(-EINVAL);
 796                 }
 797
 798                 if (i >= count)
 799                         RETURN(-EAGAIN);
 800
 801                 LASSERT(tgt && tgt->ltd_exp);
 802                 OBD_ALLOC_PTR(oqctl);
 803                 if (!oqctl)
 804                         RETURN(-ENOMEM);
 805
 806                 QCTL_COPY(oqctl, qctl);
 807                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 808                 if (rc == 0) {
 809                         QCTL_COPY(qctl, oqctl);
 810                         qctl->qc_valid = QC_MDTIDX;
 811                         qctl->obd_uuid = tgt->ltd_uuid;
 812                 }
 813                 OBD_FREE_PTR(oqctl);
 814                 break;
 815         }
 816         case OBD_IOC_CHANGELOG_SEND:
 817         case OBD_IOC_CHANGELOG_CLEAR: {
 818                 struct ioc_changelog *icc = karg;
 819
 820                 if (icc->icc_mdtindex >= count)
 821                         RETURN(-ENODEV);
 822
 823                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex].ltd_exp,
 824                                    sizeof(*icc), icc, NULL);
 825                 break;
 826         }
 827         case LL_IOC_GET_CONNECT_FLAGS: {
 828                 rc = obd_iocontrol(cmd, lmv->tgts[0].ltd_exp, len, karg, uarg);
 829                 break;
 830         }
 831
 832         default : {
 833                 for (i = 0; i < count; i++) {
 834                         int err;
 835                         struct obd_device *mdc_obd;
 836
 837                         if (lmv->tgts[i].ltd_exp == NULL)
 838                                 continue;
 839                         /* ll_umount_begin() sets force flag but for lmv, not
 840                          * mdc. Let's pass it through */
 841                         mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 842                         mdc_obd->obd_force = obddev->obd_force;
 843                         err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len,
 844                                             karg, uarg);
 845                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 846                                 RETURN(err);
 847                         } else if (err) {
 848                                 if (lmv->tgts[i].ltd_active) {
 849                                         CERROR("error: iocontrol MDC %s on MDT"
 850                                                "idx %d cmd %x: err = %d\n",
 851                                                 lmv->tgts[i].ltd_uuid.uuid,
 852                                                 i, cmd, err);
 853                                         if (!rc)
 854                                                 rc = err;
 855                                 }
 856                         } else
 857                                 set = 1;
 858                 }
 859                 if (!set && !rc)
 860                         rc = -EIO;
 861         }
 862         }
 863         RETURN(rc);
 864 }
 865
 866 static int lmv_all_chars_policy(int count, const char *name,
 867                                 int len)
 868 {
 869         unsigned int c = 0;
 870
 871         while (len > 0)
 872                 c += name[--len];
 873         c = c % count;
 874         return c;
 875 }
 876
 877 static int lmv_nid_policy(struct lmv_obd *lmv)
 878 {
 879         struct obd_import *imp;
 880         __u32              id;
 881
 882         /*
 883          * XXX: To get nid we assume that underlying obd device is mdc.
 884          */
 885         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 886         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 887         return id % lmv->desc.ld_tgt_count;
 888 }
 889
 890 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 891                           placement_policy_t placement)
 892 {
 893         switch (placement) {
 894         case PLACEMENT_CHAR_POLICY:
 895                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 896                                             op_data->op_name,
 897                                             op_data->op_namelen);
 898         case PLACEMENT_NID_POLICY:
 899                 return lmv_nid_policy(lmv);
 900
 901         default:
 902                 break;
 903         }
 904
 905         CERROR("Unsupported placement policy %x\n", placement);
 906         return -EINVAL;
 907 }
 908
 909 /**
 910  * This is _inode_ placement policy function (not name).
 911  */
 912 static int lmv_placement_policy(struct obd_device *obd,
 913                                 struct md_op_data *op_data,
 914                                 mdsno_t *mds)
 915 {
 916         struct lmv_obd          *lmv = &obd->u.lmv;
 917         struct lmv_object       *obj;
 918         int                      rc;
 919         ENTRY;
 920
 921         LASSERT(mds != NULL);
 922
 923         if (lmv->desc.ld_tgt_count == 1) {
 924                 *mds = 0;
 925                 RETURN(0);
 926         }
 927
 928         /*
 929          * Allocate new fid on target according to operation type and parent
 930          * home mds.
 931          */
 932         obj = lmv_object_find(obd, &op_data->op_fid1);
 933         if (obj != NULL || op_data->op_name == NULL ||
 934             op_data->op_opc != LUSTRE_OPC_MKDIR) {
 935                 /*
 936                  * Allocate fid for non-dir or for null name or for case parent
 937                  * dir is split.
 938                  */
 939                 if (obj) {
 940                         lmv_object_put(obj);
 941
 942                         /*
 943                          * If we have this flag turned on, and we see that
 944                          * parent dir is split, this means, that caller did not
 945                          * notice split yet. This is race and we would like to
 946                          * let caller know that.
 947                          */
 948                         if (op_data->op_bias & MDS_CHECK_SPLIT)
 949                                 RETURN(-ERESTART);
 950                 }
 951
 952                 /*
 953                  * Allocate new fid on same mds where parent fid is located and
 954                  * where operation will be sent. In case of split dir, ->op_fid1
 955                  * and ->op_mds here will contain fid and mds of slave directory
 956                  * object (assigned by caller).
 957                  */
 958                 *mds = op_data->op_mds;
 959                 rc = 0;
 960         } else {
 961                 /*
 962                  * Parent directory is not split and we want to create a
 963                  * directory in it. Let's calculate where to place it according
 964                  * to operation data @op_data.
 965                  */
 966                 *mds = lmv_choose_mds(lmv, op_data, lmv->lmv_placement);
 967                 rc = 0;
 968         }
 969
 970         if (rc) {
 971                 CERROR("Can't choose MDS, err = %d\n", rc);
 972         } else {
 973                 LASSERT(*mds < lmv->desc.ld_tgt_count);
 974         }
 975
 976         RETURN(rc);
 977 }
 978
 979 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
 980                     mdsno_t mds)
 981 {
 982         struct lmv_tgt_desc *tgt;
 983         int                  rc;
 984         ENTRY;
 985
 986         tgt = lmv_get_target(lmv, mds);
 987
 988         /*
 989          * New seq alloc and FLD setup should be atomic. Otherwise we may find
 990          * on server that seq in new allocated fid is not yet known.
 991          */
 992         cfs_down(&tgt->ltd_fid_sem);
 993
 994         if (!tgt->ltd_active)
 995                 GOTO(out, rc = -ENODEV);
 996
 997         /*
 998          * Asking underlaying tgt layer to allocate new fid.
 999          */
1000         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1001         if (rc > 0) {
1002                 LASSERT(fid_is_sane(fid));
1003                 rc = 0;
1004         }
1005
1006         EXIT;
1007 out:
1008         cfs_up(&tgt->ltd_fid_sem);
1009         return rc;
1010 }
1011
1012 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1013                   struct md_op_data *op_data)
1014 {
1015         struct obd_device     *obd = class_exp2obd(exp);
1016         struct lmv_obd        *lmv = &obd->u.lmv;
1017         mdsno_t                mds;
1018         int                    rc;
1019         ENTRY;
1020
1021         LASSERT(op_data != NULL);
1022         LASSERT(fid != NULL);
1023
1024         rc = lmv_placement_policy(obd, op_data, &mds);
1025         if (rc) {
1026                 CERROR("Can't get target for allocating fid, "
1027                        "rc %d\n", rc);
1028                 RETURN(rc);
1029         }
1030
1031         rc = __lmv_fid_alloc(lmv, fid, mds);
1032         if (rc) {
1033                 CERROR("Can't alloc new fid, rc %d\n", rc);
1034                 RETURN(rc);
1035         }
1036
1037         RETURN(rc);
1038 }
1039
1040 static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1041 {
1042         ENTRY;
1043         LASSERT(exp != NULL && fid != NULL);
1044         if (lmv_object_delete(exp, fid)) {
1045                 CDEBUG(D_INODE, "Object "DFID" is destroyed.\n",
1046                        PFID(fid));
1047         }
1048         RETURN(0);
1049 }
1050
1051 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1052 {
1053         struct lmv_obd             *lmv = &obd->u.lmv;
1054         struct lprocfs_static_vars  lvars;
1055         struct lmv_desc            *desc;
1056         int                         rc;
1057         int                         i = 0;
1058         ENTRY;
1059
1060         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1061                 CERROR("LMV setup requires a descriptor\n");
1062                 RETURN(-EINVAL);
1063         }
1064
1065         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1066         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1067                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1068                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1069                 RETURN(-EINVAL);
1070         }
1071
1072         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
1073
1074         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
1075         if (lmv->tgts == NULL)
1076                 RETURN(-ENOMEM);
1077
1078         for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
1079                 cfs_sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
1080                 lmv->tgts[i].ltd_idx = i;
1081         }
1082
1083         lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
1084
1085         OBD_ALLOC(lmv->datas, lmv->datas_size);
1086         if (lmv->datas == NULL)
1087                 GOTO(out_free_tgts, rc = -ENOMEM);
1088
1089         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1090         lmv->desc.ld_tgt_count = 0;
1091         lmv->desc.ld_active_tgt_count = 0;
1092         lmv->max_cookiesize = 0;
1093         lmv->max_def_easize = 0;
1094         lmv->max_easize = 0;
1095         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1096
1097         cfs_spin_lock_init(&lmv->lmv_lock);
1098         cfs_sema_init(&lmv->init_sem, 1);
1099
1100         rc = lmv_object_setup(obd);
1101         if (rc) {
1102                 CERROR("Can't setup LMV object manager, error %d.\n", rc);
1103                 GOTO(out_free_datas, rc);
1104         }
1105
1106         lprocfs_lmv_init_vars(&lvars);
1107         lprocfs_obd_setup(obd, lvars.obd_vars);
1108 #ifdef LPROCFS
1109         {
1110                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status",
1111                                         0444, &lmv_proc_target_fops, obd);
1112                 if (rc)
1113                         CWARN("Error adding target_obd_stats file (%d)\n", rc);
1114        }
1115 #endif
1116         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1117                              LUSTRE_CLI_FLD_HASH_DHT);
1118         if (rc) {
1119                 CERROR("Can't init FLD, err %d\n", rc);
1120                 GOTO(out_free_datas, rc);
1121         }
1122
1123         RETURN(0);
1124
1125 out_free_datas:
1126         OBD_FREE(lmv->datas, lmv->datas_size);
1127         lmv->datas = NULL;
1128 out_free_tgts:
1129         OBD_FREE(lmv->tgts, lmv->tgts_size);
1130         lmv->tgts = NULL;
1131         return rc;
1132 }
1133
1134 static int lmv_cleanup(struct obd_device *obd)
1135 {
1136         struct lmv_obd   *lmv = &obd->u.lmv;
1137         ENTRY;
1138
1139         fld_client_fini(&lmv->lmv_fld);
1140         lprocfs_obd_cleanup(obd);
1141         lmv_object_cleanup(obd);
1142         OBD_FREE(lmv->datas, lmv->datas_size);
1143         OBD_FREE(lmv->tgts, lmv->tgts_size);
1144
1145         RETURN(0);
1146 }
1147
1148 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1149 {
1150         struct lustre_cfg     *lcfg = buf;
1151         struct obd_uuid        tgt_uuid;
1152         int                    rc;
1153         ENTRY;
1154
1155         switch(lcfg->lcfg_command) {
1156         case LCFG_ADD_MDC:
1157                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1158                         GOTO(out, rc = -EINVAL);
1159
1160                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1161                 rc = lmv_add_target(obd, &tgt_uuid);
1162                 GOTO(out, rc);
1163         default: {
1164                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1165                 GOTO(out, rc = -EINVAL);
1166         }
1167         }
1168 out:
1169         RETURN(rc);
1170 }
1171
1172 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1173                       __u64 max_age, __u32 flags)
1174 {
1175         struct lmv_obd        *lmv = &obd->u.lmv;
1176         struct obd_statfs     *temp;
1177         int                    rc = 0;
1178         int                    i;
1179         ENTRY;
1180
1181         rc = lmv_check_connect(obd);
1182         if (rc)
1183                 RETURN(rc);
1184
1185         OBD_ALLOC(temp, sizeof(*temp));
1186         if (temp == NULL)
1187                 RETURN(-ENOMEM);
1188
1189         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1190                 if (lmv->tgts[i].ltd_exp == NULL)
1191                         continue;
1192
1193                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp,
1194                                 max_age, flags);
1195                 if (rc) {
1196                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1197                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1198                                rc);
1199                         GOTO(out_free_temp, rc);
1200                 }
1201                 if (i == 0) {
1202                         *osfs = *temp;
1203                 } else {
1204                         osfs->os_bavail += temp->os_bavail;
1205                         osfs->os_blocks += temp->os_blocks;
1206                         osfs->os_ffree += temp->os_ffree;
1207                         osfs->os_files += temp->os_files;
1208                 }
1209         }
1210
1211         EXIT;
1212 out_free_temp:
1213         OBD_FREE(temp, sizeof(*temp));
1214         return rc;
1215 }
1216
1217 static int lmv_getstatus(struct obd_export *exp,
1218                          struct lu_fid *fid,
1219                          struct obd_capa **pc)
1220 {
1221         struct obd_device    *obd = exp->exp_obd;
1222         struct lmv_obd       *lmv = &obd->u.lmv;
1223         int                   rc;
1224         ENTRY;
1225
1226         rc = lmv_check_connect(obd);
1227         if (rc)
1228                 RETURN(rc);
1229
1230         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1231         RETURN(rc);
1232 }
1233
1234 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1235                         struct obd_capa *oc, obd_valid valid, const char *name,
1236                         const char *input, int input_size, int output_size,
1237                         int flags, struct ptlrpc_request **request)
1238 {
1239         struct obd_device      *obd = exp->exp_obd;
1240         struct lmv_obd         *lmv = &obd->u.lmv;
1241         struct lmv_tgt_desc    *tgt;
1242         int                     rc;
1243         ENTRY;
1244
1245         rc = lmv_check_connect(obd);
1246         if (rc)
1247                 RETURN(rc);
1248
1249         tgt = lmv_find_target(lmv, fid);
1250         if (IS_ERR(tgt))
1251                 RETURN(PTR_ERR(tgt));
1252
1253         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1254                          input_size, output_size, flags, request);
1255
1256         RETURN(rc);
1257 }
1258
1259 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1260                         struct obd_capa *oc, obd_valid valid, const char *name,
1261                         const char *input, int input_size, int output_size,
1262                         int flags, __u32 suppgid,
1263                         struct ptlrpc_request **request)
1264 {
1265         struct obd_device      *obd = exp->exp_obd;
1266         struct lmv_obd         *lmv = &obd->u.lmv;
1267         struct lmv_tgt_desc    *tgt;
1268         int                     rc;
1269         ENTRY;
1270
1271         rc = lmv_check_connect(obd);
1272         if (rc)
1273                 RETURN(rc);
1274
1275         tgt = lmv_find_target(lmv, fid);
1276         if (IS_ERR(tgt))
1277                 RETURN(PTR_ERR(tgt));
1278
1279         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1280                          input_size, output_size, flags, suppgid,
1281                          request);
1282
1283         RETURN(rc);
1284 }
1285
1286 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1287                        struct ptlrpc_request **request)
1288 {
1289         struct obd_device       *obd = exp->exp_obd;
1290         struct lmv_obd          *lmv = &obd->u.lmv;
1291         struct lmv_tgt_desc     *tgt;
1292         struct lmv_object       *obj;
1293         int                      rc;
1294         int                      i;
1295         ENTRY;
1296
1297         rc = lmv_check_connect(obd);
1298         if (rc)
1299                 RETURN(rc);
1300
1301         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1302         if (IS_ERR(tgt))
1303                 RETURN(PTR_ERR(tgt));
1304
1305         if (op_data->op_valid & OBD_MD_MDTIDX) {
1306                 op_data->op_mds = tgt->ltd_idx;
1307                 RETURN(0);
1308         }
1309
1310         rc = md_getattr(tgt->ltd_exp, op_data, request);
1311         if (rc)
1312                 RETURN(rc);
1313
1314         obj = lmv_object_find_lock(obd, &op_data->op_fid1);
1315
1316         CDEBUG(D_INODE, "GETATTR for "DFID" %s\n", PFID(&op_data->op_fid1),
1317                obj ? "(split)" : "");
1318
1319         /*
1320          * If object is split, then we loop over all the slaves and gather size
1321          * attribute. In ideal world we would have to gather also mds field from
1322          * all slaves, as object is spread over the cluster and this is
1323          * definitely interesting information and it is not good to loss it,
1324          * but...
1325          */
1326         if (obj) {
1327                 struct mdt_body *body;
1328
1329                 if (*request == NULL) {
1330                         lmv_object_put(obj);
1331                         RETURN(rc);
1332                 }
1333
1334                 body = req_capsule_server_get(&(*request)->rq_pill,
1335                                               &RMF_MDT_BODY);
1336                 LASSERT(body != NULL);
1337
1338                 for (i = 0; i < obj->lo_objcount; i++) {
1339                         if (lmv->tgts[i].ltd_exp == NULL) {
1340                                 CWARN("%s: NULL export for %d\n",
1341                                       obd->obd_name, i);
1342                                 continue;
1343                         }
1344
1345                         /*
1346                          * Skip master object.
1347                          */
1348                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid))
1349                                 continue;
1350
1351                         body->size += obj->lo_stripes[i].ls_size;
1352                 }
1353
1354                 lmv_object_put_unlock(obj);
1355         }
1356
1357         RETURN(rc);
1358 }
1359
1360 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1361                              ldlm_iterator_t it, void *data)
1362 {
1363         struct obd_device   *obd = exp->exp_obd;
1364         struct lmv_obd      *lmv = &obd->u.lmv;
1365         int                  i;
1366         int                  rc;
1367         ENTRY;
1368
1369         rc = lmv_check_connect(obd);
1370         if (rc)
1371                 RETURN(rc);
1372
1373         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1374
1375         /*
1376          * With CMD every object can have two locks in different namespaces:
1377          * lookup lock in space of mds storing direntry and update/open lock in
1378          * space of mds storing inode.
1379          */
1380         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1381                 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1382
1383         RETURN(0);
1384 }
1385
1386 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1387                            ldlm_iterator_t it, void *data)
1388 {
1389         struct obd_device   *obd = exp->exp_obd;
1390         struct lmv_obd      *lmv = &obd->u.lmv;
1391         int                  i;
1392         int                  rc;
1393         ENTRY;
1394
1395         rc = lmv_check_connect(obd);
1396         if (rc)
1397                 RETURN(rc);
1398
1399         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1400
1401         /*
1402          * With CMD every object can have two locks in different namespaces:
1403          * lookup lock in space of mds storing direntry and update/open lock in
1404          * space of mds storing inode.
1405          */
1406         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1407                 rc = md_find_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1408                 if (rc)
1409                         RETURN(rc);
1410         }
1411
1412         RETURN(rc);
1413 }
1414
1415
1416 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1417                      struct md_open_data *mod, struct ptlrpc_request **request)
1418 {
1419         struct obd_device     *obd = exp->exp_obd;
1420         struct lmv_obd        *lmv = &obd->u.lmv;
1421         struct lmv_tgt_desc   *tgt;
1422         int                    rc;
1423         ENTRY;
1424
1425         rc = lmv_check_connect(obd);
1426         if (rc)
1427                 RETURN(rc);
1428
1429         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1430         if (IS_ERR(tgt))
1431                 RETURN(PTR_ERR(tgt));
1432
1433         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1434         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1435         RETURN(rc);
1436 }
1437
1438 /**
1439  * Called in the case MDS returns -ERESTART on create on open, what means that
1440  * directory is split and its LMV presentation object has to be updated.
1441  */
1442 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1443 {
1444         struct obd_device       *obd = exp->exp_obd;
1445         struct lmv_obd          *lmv = &obd->u.lmv;
1446         struct ptlrpc_request   *req = NULL;
1447         struct lmv_tgt_desc     *tgt;
1448         struct lmv_object       *obj;
1449         struct lustre_md         md;
1450         struct md_op_data       *op_data;
1451         int                      mealen;
1452         int                      rc;
1453         __u64                    valid;
1454         ENTRY;
1455
1456         md.mea = NULL;
1457         mealen = lmv_get_easize(lmv);
1458
1459         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1460
1461         tgt = lmv_find_target(lmv, fid);
1462         if (IS_ERR(tgt))
1463                 RETURN(PTR_ERR(tgt));
1464
1465         /*
1466          * Time to update mea of parent fid.
1467          */
1468
1469         OBD_ALLOC_PTR(op_data);
1470         if (op_data == NULL)
1471                 RETURN(-ENOMEM);
1472
1473         op_data->op_fid1 = *fid;
1474         op_data->op_mode = mealen;
1475         op_data->op_valid = valid;
1476
1477         rc = md_getattr(tgt->ltd_exp, op_data, &req);
1478         OBD_FREE_PTR(op_data);
1479         if (rc) {
1480                 CERROR("md_getattr() failed, error %d\n", rc);
1481                 GOTO(cleanup, rc);
1482         }
1483
1484         rc = md_get_lustre_md(tgt->ltd_exp, req, NULL, exp, &md);
1485         if (rc) {
1486                 CERROR("md_get_lustre_md() failed, error %d\n", rc);
1487                 GOTO(cleanup, rc);
1488         }
1489
1490         if (md.mea == NULL)
1491                 GOTO(cleanup, rc = -ENODATA);
1492
1493         obj = lmv_object_create(exp, fid, md.mea);
1494         if (IS_ERR(obj))
1495                 rc = PTR_ERR(obj);
1496         else
1497                 lmv_object_put(obj);
1498
1499         obd_free_memmd(exp, (void *)&md.mea);
1500         EXIT;
1501 cleanup:
1502         if (req)
1503                 ptlrpc_req_finished(req);
1504         return rc;
1505 }
1506
1507 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1508                const void *data, int datalen, int mode, __u32 uid,
1509                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1510                struct ptlrpc_request **request)
1511 {
1512         struct obd_device       *obd = exp->exp_obd;
1513         struct lmv_obd          *lmv = &obd->u.lmv;
1514         struct lmv_tgt_desc     *tgt;
1515         struct lmv_object       *obj;
1516         int                      rc;
1517         int                      loop = 0;
1518         int                      sidx;
1519         ENTRY;
1520
1521         rc = lmv_check_connect(obd);
1522         if (rc)
1523                 RETURN(rc);
1524
1525         if (!lmv->desc.ld_active_tgt_count)
1526                 RETURN(-EIO);
1527 repeat:
1528         ++loop;
1529         LASSERT(loop <= 2);
1530
1531         obj = lmv_object_find(obd, &op_data->op_fid1);
1532         if (obj) {
1533                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1534                                        op_data->op_name, op_data->op_namelen);
1535                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1536                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1537                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1538                 tgt = lmv_get_target(lmv, op_data->op_mds);
1539                 lmv_object_put(obj);
1540         } else {
1541                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1542                 op_data->op_bias |= MDS_CHECK_SPLIT;
1543                 op_data->op_mds = tgt->ltd_idx;
1544         }
1545
1546         if (IS_ERR(tgt))
1547                 RETURN(PTR_ERR(tgt));
1548
1549         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1550         if (rc == -ERESTART)
1551                 goto repeat;
1552         else if (rc)
1553                 RETURN(rc);
1554
1555         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1556                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1557                op_data->op_mds);
1558
1559         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1560         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1561                        cap_effective, rdev, request);
1562         if (rc == 0) {
1563                 if (*request == NULL)
1564                         RETURN(rc);
1565                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1566         } else if (rc == -ERESTART) {
1567                 LASSERT(*request != NULL);
1568                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1569                           "Got -ERESTART during create!\n");
1570                 ptlrpc_req_finished(*request);
1571                 *request = NULL;
1572
1573                 /*
1574                  * Directory got split. Time to update local object and repeat
1575                  * the request with proper MDS.
1576                  */
1577                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1578                 if (rc == 0) {
1579                         rc = lmv_allocate_slaves(obd, &op_data->op_fid1,
1580                                                  op_data, &op_data->op_fid2);
1581                         if (rc)
1582                                 RETURN(rc);
1583                         goto repeat;
1584                 }
1585         }
1586         RETURN(rc);
1587 }
1588
1589 static int lmv_done_writing(struct obd_export *exp,
1590                             struct md_op_data *op_data,
1591                             struct md_open_data *mod)
1592 {
1593         struct obd_device     *obd = exp->exp_obd;
1594         struct lmv_obd        *lmv = &obd->u.lmv;
1595         struct lmv_tgt_desc   *tgt;
1596         int                    rc;
1597         ENTRY;
1598
1599         rc = lmv_check_connect(obd);
1600         if (rc)
1601                 RETURN(rc);
1602
1603         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1604         if (IS_ERR(tgt))
1605                 RETURN(PTR_ERR(tgt));
1606
1607         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1608         RETURN(rc);
1609 }
1610
1611 static int
1612 lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1613                    struct lookup_intent *it, struct md_op_data *op_data,
1614                    struct lustre_handle *lockh, void *lmm, int lmmsize)
1615 {
1616         struct obd_device     *obd = exp->exp_obd;
1617         struct lmv_obd        *lmv = &obd->u.lmv;
1618         struct lmv_stripe_md  *mea = op_data->op_mea1;
1619         struct md_op_data     *op_data2;
1620         struct lmv_tgt_desc   *tgt;
1621         int                    i;
1622         int                    rc = 0;
1623         ENTRY;
1624
1625         OBD_ALLOC_PTR(op_data2);
1626         if (op_data2 == NULL)
1627                 RETURN(-ENOMEM);
1628
1629         LASSERT(mea != NULL);
1630         for (i = 0; i < mea->mea_count; i++) {
1631                 memset(op_data2, 0, sizeof(*op_data2));
1632                 op_data2->op_fid1 = mea->mea_ids[i];
1633                 op_data2->op_bias = 0;
1634
1635                 tgt = lmv_find_target(lmv, &op_data2->op_fid1);
1636                 if (IS_ERR(tgt))
1637                         GOTO(cleanup, rc = PTR_ERR(tgt));
1638
1639                 if (tgt->ltd_exp == NULL)
1640                         continue;
1641
1642                 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data2,
1643                                 lockh + i, lmm, lmmsize, NULL, 0);
1644
1645                 CDEBUG(D_INODE, "Take lock on slave "DFID" -> %d/%d\n",
1646                        PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1647
1648                 if (rc)
1649                         GOTO(cleanup, rc);
1650
1651                 if (it->d.lustre.it_data) {
1652                         struct ptlrpc_request *req;
1653                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1654                         ptlrpc_req_finished(req);
1655                 }
1656
1657                 if (it->d.lustre.it_status)
1658                         GOTO(cleanup, rc = it->d.lustre.it_status);
1659         }
1660
1661         EXIT;
1662 cleanup:
1663         OBD_FREE_PTR(op_data2);
1664
1665         if (rc != 0) {
1666                 /*
1667                  * Drop all taken locks.
1668                  */
1669                 while (--i >= 0) {
1670                         if (lockh[i].cookie)
1671                                 ldlm_lock_decref(lockh + i, einfo->ei_mode);
1672                         lockh[i].cookie = 0;
1673                 }
1674         }
1675         return rc;
1676 }
1677
1678 static int
1679 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1680                    struct lookup_intent *it, struct md_op_data *op_data,
1681                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1682                    int extra_lock_flags)
1683 {
1684         struct ptlrpc_request      *req = it->d.lustre.it_data;
1685         struct obd_device          *obd = exp->exp_obd;
1686         struct lmv_obd             *lmv = &obd->u.lmv;
1687         struct lustre_handle        plock;
1688         struct lmv_tgt_desc        *tgt;
1689         struct md_op_data          *rdata;
1690         struct lu_fid               fid1;
1691         struct mdt_body            *body;
1692         int                         rc = 0;
1693         int                         pmode;
1694         ENTRY;
1695
1696         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1697         LASSERT(body != NULL);
1698
1699         if (!(body->valid & OBD_MD_MDS))
1700                 RETURN(0);
1701
1702         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1703                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1704
1705         /*
1706          * We got LOOKUP lock, but we really need attrs.
1707          */
1708         pmode = it->d.lustre.it_lock_mode;
1709         LASSERT(pmode != 0);
1710         memcpy(&plock, lockh, sizeof(plock));
1711         it->d.lustre.it_lock_mode = 0;
1712         it->d.lustre.it_data = NULL;
1713         fid1 = body->fid1;
1714
1715         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1716         ptlrpc_req_finished(req);
1717
1718         tgt = lmv_find_target(lmv, &fid1);
1719         if (IS_ERR(tgt))
1720                 GOTO(out, rc = PTR_ERR(tgt));
1721
1722         OBD_ALLOC_PTR(rdata);
1723         if (rdata == NULL)
1724                 GOTO(out, rc = -ENOMEM);
1725
1726         rdata->op_fid1 = fid1;
1727         rdata->op_bias = MDS_CROSS_REF;
1728
1729         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1730                         lmm, lmmsize, NULL, extra_lock_flags);
1731         OBD_FREE_PTR(rdata);
1732         EXIT;
1733 out:
1734         ldlm_lock_decref(&plock, pmode);
1735         return rc;
1736 }
1737
1738 static int
1739 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1740             struct lookup_intent *it, struct md_op_data *op_data,
1741             struct lustre_handle *lockh, void *lmm, int lmmsize,
1742             struct ptlrpc_request **req, int extra_lock_flags)
1743 {
1744         struct obd_device        *obd = exp->exp_obd;
1745         struct lmv_obd           *lmv = &obd->u.lmv;
1746         struct lmv_tgt_desc      *tgt;
1747         struct lmv_object        *obj;
1748         int                       sidx;
1749         int                       rc;
1750         ENTRY;
1751
1752         rc = lmv_check_connect(obd);
1753         if (rc)
1754                 RETURN(rc);
1755
1756         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1757                LL_IT2STR(it), PFID(&op_data->op_fid1));
1758
1759         if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) {
1760                 rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
1761                                         lockh, lmm, lmmsize);
1762                 RETURN(rc);
1763         }
1764
1765         obj = lmv_object_find(obd, &op_data->op_fid1);
1766         if (obj && op_data->op_namelen) {
1767                 sidx = raw_name2idx(obj->lo_hashtype,
1768                                        obj->lo_objcount,
1769                                        (char *)op_data->op_name,
1770                                        op_data->op_namelen);
1771                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1772                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1773         } else {
1774                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1775         }
1776         if (obj)
1777                 lmv_object_put(obj);
1778
1779         if (IS_ERR(tgt))
1780                 RETURN(PTR_ERR(tgt));
1781
1782         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1783                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1784
1785         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1786                         lmm, lmmsize, req, extra_lock_flags);
1787
1788         if (rc == 0 && it && it->it_op == IT_OPEN) {
1789                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1790                                         lmm, lmmsize, extra_lock_flags);
1791         }
1792         RETURN(rc);
1793 }
1794
1795 static int
1796 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1797                  struct ptlrpc_request **request)
1798 {
1799         struct ptlrpc_request   *req = NULL;
1800         struct obd_device       *obd = exp->exp_obd;
1801         struct lmv_obd          *lmv = &obd->u.lmv;
1802         struct lu_fid            rid = op_data->op_fid1;
1803         struct lmv_tgt_desc     *tgt;
1804         struct mdt_body         *body;
1805         struct lmv_object       *obj;
1806         obd_valid                valid = op_data->op_valid;
1807         int                      rc;
1808         int                      loop = 0;
1809         int                      sidx;
1810         ENTRY;
1811
1812         rc = lmv_check_connect(obd);
1813         if (rc)
1814                 RETURN(rc);
1815
1816 repeat:
1817         ++loop;
1818         LASSERT(loop <= 2);
1819         obj = lmv_object_find(obd, &rid);
1820         if (obj) {
1821                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1822                                     op_data->op_name, op_data->op_namelen);
1823                 rid = obj->lo_stripes[sidx].ls_fid;
1824                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1825                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1826                 valid &= ~OBD_MD_FLCKSPLIT;
1827                 lmv_object_put(obj);
1828         } else {
1829                 tgt = lmv_find_target(lmv, &rid);
1830                 valid |= OBD_MD_FLCKSPLIT;
1831                 op_data->op_mds = tgt->ltd_idx;
1832         }
1833         if (IS_ERR(tgt))
1834                 RETURN(PTR_ERR(tgt));
1835
1836         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" - "DFID" -> mds #%d\n",
1837                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1838                PFID(&rid), tgt->ltd_idx);
1839
1840         op_data->op_valid = valid;
1841         op_data->op_fid1 = rid;
1842         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1843         if (rc == 0) {
1844                 body = req_capsule_server_get(&(*request)->rq_pill,
1845                                               &RMF_MDT_BODY);
1846                 LASSERT(body != NULL);
1847
1848                 if (body->valid & OBD_MD_MDS) {
1849                         rid = body->fid1;
1850                         CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1851                                PFID(&rid));
1852
1853                         tgt = lmv_find_target(lmv, &rid);
1854                         if (IS_ERR(tgt)) {
1855                                 ptlrpc_req_finished(*request);
1856                                 RETURN(PTR_ERR(tgt));
1857                         }
1858
1859                         op_data->op_fid1 = rid;
1860                         op_data->op_valid |= OBD_MD_FLCROSSREF;
1861                         op_data->op_namelen = 0;
1862                         op_data->op_name = NULL;
1863                         rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1864                         ptlrpc_req_finished(*request);
1865                         *request = req;
1866                 }
1867         } else if (rc == -ERESTART) {
1868                 LASSERT(*request != NULL);
1869                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1870                           "Got -ERESTART during getattr!\n");
1871                 ptlrpc_req_finished(*request);
1872                 *request = NULL;
1873
1874                 /*
1875                  * Directory got split. Time to update local object and repeat
1876                  * the request with proper MDS.
1877                  */
1878                 rc = lmv_handle_split(exp, &rid);
1879                 if (rc == 0)
1880                         goto repeat;
1881         }
1882         RETURN(rc);
1883 }
1884
1885 #define md_op_data_fid(op_data, fl)                     \
1886         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1887          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1888          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1889          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1890          NULL)
1891
1892 static int lmv_early_cancel_slaves(struct obd_export *exp,
1893                                    struct md_op_data *op_data, int op_tgt,
1894                                    ldlm_mode_t mode, int bits, int flag)
1895 {
1896         struct obd_device      *obd = exp->exp_obd;
1897         struct lmv_obd         *lmv = &obd->u.lmv;
1898         ldlm_policy_data_t      policy = {{0}};
1899         struct lu_fid          *op_fid;
1900         struct lu_fid          *st_fid;
1901         struct lmv_tgt_desc    *tgt;
1902         struct lmv_object      *obj;
1903         int                     rc = 0;
1904         int                     i;
1905         ENTRY;
1906
1907         op_fid = md_op_data_fid(op_data, flag);
1908         if (!fid_is_sane(op_fid))
1909                 RETURN(0);
1910
1911         obj = lmv_object_find(obd, op_fid);
1912         if (obj == NULL)
1913                 RETURN(-EALREADY);
1914
1915         policy.l_inodebits.bits = bits;
1916         for (i = 0; i < obj->lo_objcount; i++) {
1917                 tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
1918                 st_fid = &obj->lo_stripes[i].ls_fid;
1919                 if (op_tgt != tgt->ltd_idx) {
1920                         CDEBUG(D_INODE, "EARLY_CANCEL slave "DFID" -> mds #%d\n",
1921                                PFID(st_fid), tgt->ltd_idx);
1922                         rc = md_cancel_unused(tgt->ltd_exp, st_fid, &policy,
1923                                               mode, LCF_ASYNC, NULL);
1924                         if (rc)
1925                                 GOTO(out_put_obj, rc);
1926                 } else {
1927                         CDEBUG(D_INODE,
1928                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1929                                op_tgt, PFID(st_fid));
1930                         /*
1931                          * Do not cancel locks for operation target, they will
1932                          * be handled later in underlaying layer when calling
1933                          * function we run on behalf of.
1934                          */
1935                         *op_fid = *st_fid;
1936                         op_data->op_flags |= flag;
1937                 }
1938         }
1939         EXIT;
1940 out_put_obj:
1941         lmv_object_put(obj);
1942         return rc;
1943 }
1944
1945 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1946                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1947 {
1948         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1949         struct obd_device      *obd = exp->exp_obd;
1950         struct lmv_obd         *lmv = &obd->u.lmv;
1951         struct lmv_tgt_desc    *tgt;
1952         ldlm_policy_data_t      policy = {{0}};
1953         struct lmv_object      *obj;
1954         int                     rc = 0;
1955         ENTRY;
1956
1957         if (!fid_is_sane(fid))
1958                 RETURN(0);
1959
1960         obj = lmv_object_find(obd, fid);
1961         if (obj) {
1962                 rc = lmv_early_cancel_slaves(exp, op_data, op_tgt, mode,
1963                                              bits, flag);
1964                 lmv_object_put(obj);
1965         } else {
1966                 tgt = lmv_find_target(lmv, fid);
1967                 if (IS_ERR(tgt))
1968                         RETURN(PTR_ERR(tgt));
1969
1970                 if (tgt->ltd_idx != op_tgt) {
1971                         CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1972                         policy.l_inodebits.bits = bits;
1973                         rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1974                                               mode, LCF_ASYNC, NULL);
1975                 } else {
1976                         CDEBUG(D_INODE,
1977                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1978                                op_tgt, PFID(fid));
1979                         op_data->op_flags |= flag;
1980                         rc = 0;
1981                 }
1982
1983         }
1984         RETURN(rc);
1985 }
1986
1987 /*
1988  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1989  * op_data->op_fid2
1990  */
1991 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1992                     struct ptlrpc_request **request)
1993 {
1994         struct obd_device       *obd = exp->exp_obd;
1995         struct lmv_obd          *lmv = &obd->u.lmv;
1996         struct lmv_tgt_desc     *tgt;
1997         struct lmv_object       *obj;
1998         int                      rc;
1999         int                      loop = 0;
2000         mdsno_t                  mds;
2001         int                      sidx;
2002         ENTRY;
2003
2004         rc = lmv_check_connect(obd);
2005         if (rc)
2006                 RETURN(rc);
2007
2008 repeat:
2009         ++loop;
2010         LASSERT(loop <= 2);
2011         LASSERT(op_data->op_namelen != 0);
2012
2013         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2014                PFID(&op_data->op_fid2), op_data->op_namelen,
2015                op_data->op_name, PFID(&op_data->op_fid1));
2016
2017         obj = lmv_object_find(obd, &op_data->op_fid2);
2018         if (obj) {
2019                 sidx = raw_name2idx(obj->lo_hashtype,
2020                                     obj->lo_objcount,
2021                                     op_data->op_name,
2022                                     op_data->op_namelen);
2023                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2024                 mds = obj->lo_stripes[sidx].ls_mds;
2025                 lmv_object_put(obj);
2026         } else {
2027                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
2028                 if (rc)
2029                         RETURN(rc);
2030         }
2031
2032         CDEBUG(D_INODE, "Forward to mds #%x ("DFID")\n",
2033                mds, PFID(&op_data->op_fid1));
2034
2035         op_data->op_fsuid = cfs_curproc_fsuid();
2036         op_data->op_fsgid = cfs_curproc_fsgid();
2037         op_data->op_cap = cfs_curproc_cap_pack();
2038         tgt = lmv_get_target(lmv, mds);
2039
2040         /*
2041          * Cancel UPDATE lock on child (fid1).
2042          */
2043         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2044         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2045                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2046         if (rc == 0)
2047                 rc = md_link(tgt->ltd_exp, op_data, request);
2048         if (rc == -ERESTART) {
2049                 LASSERT(*request != NULL);
2050                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2051                           "Got -ERESTART during link!\n");
2052                 ptlrpc_req_finished(*request);
2053                 *request = NULL;
2054
2055                 /*
2056                  * Directory got split. Time to update local object and repeat
2057                  * the request with proper MDS.
2058                  */
2059                 rc = lmv_handle_split(exp, &op_data->op_fid2);
2060                 if (rc == 0)
2061                         goto repeat;
2062         }
2063
2064         RETURN(rc);
2065 }
2066
2067 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2068                       const char *old, int oldlen, const char *new, int newlen,
2069                       struct ptlrpc_request **request)
2070 {
2071         struct obd_device       *obd = exp->exp_obd;
2072         struct lmv_obd          *lmv = &obd->u.lmv;
2073         struct lmv_tgt_desc     *src_tgt;
2074         struct lmv_tgt_desc     *tgt_tgt;
2075         int                      rc;
2076         int                      sidx;
2077         int                      loop = 0;
2078         struct lmv_object       *obj;
2079         mdsno_t                  mds1;
2080         mdsno_t                  mds2;
2081         ENTRY;
2082
2083         LASSERT(oldlen != 0);
2084
2085         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2086                oldlen, old, PFID(&op_data->op_fid1),
2087                newlen, new, PFID(&op_data->op_fid2));
2088
2089         rc = lmv_check_connect(obd);
2090         if (rc)
2091                 RETURN(rc);
2092
2093 repeat:
2094         ++loop;
2095         LASSERT(loop <= 2);
2096         obj = lmv_object_find(obd, &op_data->op_fid1);
2097         if (obj) {
2098                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2099                                     (char *)old, oldlen);
2100                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2101                 mds1 = obj->lo_stripes[sidx].ls_mds;
2102                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
2103                 lmv_object_put(obj);
2104         } else {
2105                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
2106                 if (rc)
2107                         RETURN(rc);
2108         }
2109
2110         obj = lmv_object_find(obd, &op_data->op_fid2);
2111         if (obj) {
2112                 /*
2113                  * Directory is already split, so we have to forward request to
2114                  * the right MDS.
2115                  */
2116                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2117                                        (char *)new, newlen);
2118
2119                 mds2 = obj->lo_stripes[sidx].ls_mds;
2120                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2121                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
2122                 lmv_object_put(obj);
2123         } else {
2124                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
2125                 if (rc)
2126                         RETURN(rc);
2127         }
2128
2129         op_data->op_fsuid = cfs_curproc_fsuid();
2130         op_data->op_fsgid = cfs_curproc_fsgid();
2131         op_data->op_cap = cfs_curproc_cap_pack();
2132
2133         src_tgt = lmv_get_target(lmv, mds1);
2134         tgt_tgt = lmv_get_target(lmv, mds2);
2135
2136         /*
2137          * LOOKUP lock on src child (fid3) should also be cancelled for
2138          * src_tgt in mdc_rename.
2139          */
2140         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2141
2142         /*
2143          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2144          * own target.
2145          */
2146         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2147                               LCK_EX, MDS_INODELOCK_UPDATE,
2148                               MF_MDC_CANCEL_FID2);
2149
2150         /*
2151          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2152          */
2153         if (rc == 0) {
2154                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2155                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2156                                       MF_MDC_CANCEL_FID4);
2157         }
2158
2159         /*
2160          * Cancel all the locks on tgt child (fid4).
2161          */
2162         if (rc == 0)
2163                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2164                                       LCK_EX, MDS_INODELOCK_FULL,
2165                                       MF_MDC_CANCEL_FID4);
2166
2167         if (rc == 0)
2168                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2169                                new, newlen, request);
2170
2171         if (rc == -ERESTART) {
2172                 LASSERT(*request != NULL);
2173                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2174                           "Got -ERESTART during rename!\n");
2175                 ptlrpc_req_finished(*request);
2176                 *request = NULL;
2177
2178                 /*
2179                  * Directory got split. Time to update local object and repeat
2180                  * the request with proper MDS.
2181                  */
2182                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2183                 if (rc == 0)
2184                         goto repeat;
2185         }
2186         RETURN(rc);
2187 }
2188
2189 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2190                        void *ea, int ealen, void *ea2, int ea2len,
2191                        struct ptlrpc_request **request,
2192                        struct md_open_data **mod)
2193 {
2194         struct obd_device       *obd = exp->exp_obd;
2195         struct lmv_obd          *lmv = &obd->u.lmv;
2196         struct ptlrpc_request   *req;
2197         struct lmv_tgt_desc     *tgt;
2198         struct lmv_object       *obj;
2199         int                      rc = 0;
2200         int                      i;
2201         ENTRY;
2202
2203         rc = lmv_check_connect(obd);
2204         if (rc)
2205                 RETURN(rc);
2206
2207         obj = lmv_object_find(obd, &op_data->op_fid1);
2208
2209         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x%s\n",
2210                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2211                obj ? ", split" : "");
2212
2213         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2214         if (obj) {
2215                 for (i = 0; i < obj->lo_objcount; i++) {
2216                         op_data->op_fid1 = obj->lo_stripes[i].ls_fid;
2217
2218                         tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
2219                         if (IS_ERR(tgt)) {
2220                                 rc = PTR_ERR(tgt);
2221                                 break;
2222                         }
2223
2224                         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen,
2225                                         ea2, ea2len, &req, mod);
2226
2227                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) {
2228                                 /*
2229                                  * This is master object and this request should
2230                                  * be returned back to llite.
2231                                  */
2232                                 *request = req;
2233                         } else {
2234                                 ptlrpc_req_finished(req);
2235                         }
2236
2237                         if (rc)
2238                                 break;
2239                 }
2240                 lmv_object_put(obj);
2241         } else {
2242                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2243                 if (IS_ERR(tgt))
2244                         RETURN(PTR_ERR(tgt));
2245
2246                 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2247                                 ea2len, request, mod);
2248         }
2249         RETURN(rc);
2250 }
2251
2252 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2253                     struct obd_capa *oc, struct ptlrpc_request **request)
2254 {
2255         struct obd_device         *obd = exp->exp_obd;
2256         struct lmv_obd            *lmv = &obd->u.lmv;
2257         struct lmv_tgt_desc       *tgt;
2258         int                        rc;
2259         ENTRY;
2260
2261         rc = lmv_check_connect(obd);
2262         if (rc)
2263                 RETURN(rc);
2264
2265         tgt = lmv_find_target(lmv, fid);
2266         if (IS_ERR(tgt))
2267                 RETURN(PTR_ERR(tgt));
2268
2269         rc = md_sync(tgt->ltd_exp, fid, oc, request);
2270         RETURN(rc);
2271 }
2272
2273 /**
2274  * Main purpose of LMV blocking ast is to remove split directory LMV
2275  * presentation object (struct lmv_object) attached to the lock being revoked.
2276  */
2277 int lmv_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2278                      void *data, int flag)
2279 {
2280         struct lustre_handle    lockh;
2281         struct lmv_object      *obj;
2282         int                     rc;
2283         ENTRY;
2284
2285         switch (flag) {
2286         case LDLM_CB_BLOCKING:
2287                 ldlm_lock2handle(lock, &lockh);
2288                 rc = ldlm_cli_cancel(&lockh);
2289                 if (rc < 0) {
2290                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2291                         RETURN(rc);
2292                 }
2293                 break;
2294         case LDLM_CB_CANCELING:
2295                 /*
2296                  * Time to drop cached attrs for split directory object
2297                  */
2298                 obj = lock->l_ast_data;
2299                 if (obj) {
2300                         CDEBUG(D_INODE, "Cancel %s on "LPU64"/"LPU64
2301                                ", master "DFID"\n",
2302                                lock->l_resource->lr_name.name[3] == 1 ?
2303                                "LOOKUP" : "UPDATE",
2304                                lock->l_resource->lr_name.name[0],
2305                                lock->l_resource->lr_name.name[1],
2306                                PFID(&obj->lo_fid));
2307                         lmv_object_put(obj);
2308                 }
2309                 break;
2310         default:
2311                 LBUG();
2312         }
2313         RETURN(0);
2314 }
2315
2316 static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj)
2317 {
2318         __u64         val;
2319
2320         val = le64_to_cpu(*hash);
2321         if (val < hash_adj)
2322                 val += MAX_HASH_SIZE;
2323         if (val != MDS_DIR_END_OFF)
2324                 *hash = cpu_to_le64(val - hash_adj);
2325 }
2326
2327 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2328 {
2329         __u64              id;
2330         struct obd_import *imp;
2331
2332         /*
2333          * XXX: to get nid we assume that underlying obd device is mdc.
2334          */
2335         imp  = class_exp2cliimp(exp);
2336         id   = imp->imp_connection->c_self + fid_flatten(fid);
2337
2338         CDEBUG(D_INODE, "Readpage node rank: "LPX64" "DFID" "LPX64" "LPX64"\n",
2339                imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2340
2341         return id ^ (id >> 32);
2342 }
2343
2344 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2345                         struct obd_capa *oc, __u64 offset64, struct page *page,
2346                         struct ptlrpc_request **request)
2347 {
2348         struct obd_device       *obd = exp->exp_obd;
2349         struct lmv_obd          *lmv = &obd->u.lmv;
2350         struct lu_fid            rid = *fid;
2351         struct lmv_object       *obj;
2352         __u64                    offset;
2353         __u64                    hash_adj = 0;
2354         __u32                    rank = 0;
2355         __u64                    seg_size = 0;
2356         __u64                    tgt_tmp = 0;
2357         int                      tgt_idx = 0;
2358         int                      tgt0_idx = 0;
2359         int                      rc;
2360         int                      nr = 0;
2361         struct lmv_stripe       *los;
2362         struct lmv_tgt_desc     *tgt;
2363         struct lu_dirpage       *dp;
2364         struct lu_dirent        *ent;
2365         ENTRY;
2366
2367         offset = offset64;
2368
2369         rc = lmv_check_connect(obd);
2370         if (rc)
2371                 RETURN(rc);
2372
2373         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid));
2374
2375         /*
2376          * This case handle directory lookup in clustered metadata case (i.e.
2377          * split directory is located on multiple md servers.)
2378          * each server keeps directory entries for certain range of hashes.
2379          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
2380          * first server will keep records with hashes [ 0 ... MAX_HASH / N  - 1],
2381          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
2382          * so on....
2383          *      readdir can simply start reading entries from 0 - N server in
2384          * order but that will not scale well as all client will request dir in
2385          * to server in same order.
2386          * Following algorithm does optimization:
2387          * Instead of doing readdir in 1, 2, ...., N order, client with a
2388          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
2389          * (every client has rank R)
2390          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
2391          * since client ask dir from MDS{R} client has pages with offsets
2392          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
2393          * on hash  values that we get.
2394          */
2395         obj = lmv_object_find_lock(obd, fid);
2396         if (obj) {
2397                 nr       = obj->lo_objcount;
2398                 LASSERT(nr > 0);
2399                 seg_size = MAX_HASH_SIZE;
2400                 do_div(seg_size, nr);
2401                 los      = obj->lo_stripes;
2402                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
2403                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
2404                 tgt_tmp  = offset;
2405                 do_div(tgt_tmp, seg_size);
2406                 tgt0_idx = do_div(tgt_tmp,  nr);
2407                 tgt_idx  = (tgt0_idx + rank) % nr;
2408
2409                 if (tgt_idx < tgt0_idx)
2410                         /*
2411                          * Wrap around.
2412                          *
2413                          * Last segment has unusual length due to division
2414                          * rounding.
2415                          */
2416                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
2417                 else
2418                         hash_adj = 0;
2419
2420                 hash_adj += rank * seg_size;
2421
2422                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
2423                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
2424                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
2425
2426                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
2427                 rid = obj->lo_stripes[tgt_idx].ls_fid;
2428                 tgt = lmv_get_target(lmv, los[tgt_idx].ls_mds);
2429
2430                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
2431                        PFID(&rid), (unsigned long)offset, tgt_idx);
2432         } else
2433                 tgt = lmv_find_target(lmv, &rid);
2434
2435         if (IS_ERR(tgt))
2436                 GOTO(cleanup, rc = PTR_ERR(tgt));
2437
2438         rc = md_readpage(tgt->ltd_exp, &rid, oc, offset, page, request);
2439         if (rc)
2440                 GOTO(cleanup, rc);
2441         if (obj) {
2442                 dp = cfs_kmap(page);
2443
2444                 lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2445                 lmv_hash_adjust(&dp->ldp_hash_end,   hash_adj);
2446                 LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64);
2447
2448                 for (ent = lu_dirent_start(dp); ent != NULL;
2449                      ent = lu_dirent_next(ent))
2450                         lmv_hash_adjust(&ent->lde_hash, hash_adj);
2451
2452                 if (tgt0_idx != nr - 1) {
2453                         __u64 end;
2454
2455                         end = le64_to_cpu(dp->ldp_hash_end);
2456                         if (end == MDS_DIR_END_OFF) {
2457                                 dp->ldp_hash_end = cpu_to_le32(seg_size *
2458                                                                (tgt0_idx + 1));
2459                                 CDEBUG(D_INODE,
2460                                        ""DFID" reset end "LPX64" tgt %d\n",
2461                                        PFID(&rid),
2462                                        (__u64)le64_to_cpu(dp->ldp_hash_end), tgt_idx);
2463                         }
2464                 }
2465                 cfs_kunmap(page);
2466         }
2467         EXIT;
2468 cleanup:
2469         if (obj)
2470                 lmv_object_put_unlock(obj);
2471         return rc;
2472 }
2473
2474 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2475                       struct ptlrpc_request **request)
2476 {
2477         struct obd_device       *obd = exp->exp_obd;
2478         struct lmv_obd          *lmv = &obd->u.lmv;
2479         struct lmv_tgt_desc     *tgt = NULL;
2480         struct lmv_object       *obj;
2481         int                      rc;
2482         int                      sidx;
2483         int                      loop = 0;
2484         ENTRY;
2485
2486         rc = lmv_check_connect(obd);
2487         if (rc)
2488                 RETURN(rc);
2489
2490 repeat:
2491         ++loop;
2492         LASSERT(loop <= 2);
2493         LASSERT(op_data->op_namelen != 0);
2494
2495         obj = lmv_object_find(obd, &op_data->op_fid1);
2496         if (obj) {
2497                 sidx = raw_name2idx(obj->lo_hashtype,
2498                                     obj->lo_objcount,
2499                                     op_data->op_name,
2500                                     op_data->op_namelen);
2501                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
2502                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2503                 tgt = lmv_get_target(lmv,
2504                                      obj->lo_stripes[sidx].ls_mds);
2505                 lmv_object_put(obj);
2506                 CDEBUG(D_INODE, "UNLINK '%*s' in "DFID" -> %u\n",
2507                        op_data->op_namelen, op_data->op_name,
2508                        PFID(&op_data->op_fid1), sidx);
2509         }
2510
2511         if (tgt == NULL) {
2512                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2513                 if (IS_ERR(tgt))
2514                         RETURN(PTR_ERR(tgt));
2515                 op_data->op_bias |= MDS_CHECK_SPLIT;
2516         }
2517
2518         op_data->op_fsuid = cfs_curproc_fsuid();
2519         op_data->op_fsgid = cfs_curproc_fsgid();
2520         op_data->op_cap = cfs_curproc_cap_pack();
2521
2522         /*
2523          * If child's fid is given, cancel unused locks for it if it is from
2524          * another export than parent.
2525          *
2526          * LOOKUP lock for child (fid3) should also be cancelled on parent
2527          * tgt_tgt in mdc_unlink().
2528          */
2529         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2530
2531         /*
2532          * Cancel FULL locks on child (fid3).
2533          */
2534         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2535                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2536
2537         if (rc == 0)
2538                 rc = md_unlink(tgt->ltd_exp, op_data, request);
2539
2540         if (rc == -ERESTART) {
2541                 LASSERT(*request != NULL);
2542                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2543                           "Got -ERESTART during unlink!\n");
2544                 ptlrpc_req_finished(*request);
2545                 *request = NULL;
2546
2547                 /*
2548                  * Directory got split. Time to update local object and repeat
2549                  * the request with proper MDS.
2550                  */
2551                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2552                 if (rc == 0)
2553                         goto repeat;
2554         }
2555         RETURN(rc);
2556 }
2557
2558 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2559 {
2560         int        rc = 0;
2561
2562         switch (stage) {
2563         case OBD_CLEANUP_EARLY:
2564                 /* XXX: here should be calling obd_precleanup() down to
2565                  * stack. */
2566                 break;
2567         case OBD_CLEANUP_EXPORTS:
2568                 rc = obd_llog_finish(obd, 0);
2569                 if (rc != 0)
2570                         CERROR("failed to cleanup llogging subsystems\n");
2571                 break;
2572         default:
2573                 break;
2574         }
2575         RETURN(rc);
2576 }
2577
2578 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2579                         void *key, __u32 *vallen, void *val,
2580                         struct lov_stripe_md *lsm)
2581 {
2582         struct obd_device       *obd;
2583         struct lmv_obd          *lmv;
2584         int                      rc = 0;
2585         ENTRY;
2586
2587         obd = class_exp2obd(exp);
2588         if (obd == NULL) {
2589                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2590                        exp->exp_handle.h_cookie);
2591                 RETURN(-EINVAL);
2592         }
2593
2594         lmv = &obd->u.lmv;
2595         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2596                 struct lmv_tgt_desc *tgts;
2597                 int i;
2598
2599                 rc = lmv_check_connect(obd);
2600                 if (rc)
2601                         RETURN(rc);
2602
2603                 LASSERT(*vallen == sizeof(__u32));
2604                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2605                      i++, tgts++) {
2606
2607                         /*
2608                          * All tgts should be connected when this gets called.
2609                          */
2610                         if (!tgts || !tgts->ltd_exp) {
2611                                 CERROR("target not setup?\n");
2612                                 continue;
2613                         }
2614
2615                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2616                                           vallen, val, NULL))
2617                                 RETURN(0);
2618                 }
2619                 RETURN(-EINVAL);
2620         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2621                 rc = lmv_check_connect(obd);
2622                 if (rc)
2623                         RETURN(rc);
2624
2625                 /*
2626                  * Forwarding this request to first MDS, it should know LOV
2627                  * desc.
2628                  */
2629                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2630                                   vallen, val, NULL);
2631                 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2632                         exp->exp_connect_flags =
2633                         ((struct obd_connect_data *)val)->ocd_connect_flags;
2634                 }
2635                 RETURN(rc);
2636         }
2637
2638         CDEBUG(D_IOCTL, "Invalid key\n");
2639         RETURN(-EINVAL);
2640 }
2641
2642 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2643                        void *key, obd_count vallen, void *val,
2644                        struct ptlrpc_request_set *set)
2645 {
2646         struct lmv_tgt_desc    *tgt;
2647         struct obd_device      *obd;
2648         struct lmv_obd         *lmv;
2649         int rc = 0;
2650         ENTRY;
2651
2652         obd = class_exp2obd(exp);
2653         if (obd == NULL) {
2654                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2655                        exp->exp_handle.h_cookie);
2656                 RETURN(-EINVAL);
2657         }
2658         lmv = &obd->u.lmv;
2659
2660         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2661                 int i, err = 0;
2662
2663                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2664                         tgt = &lmv->tgts[i];
2665
2666                         if (!tgt->ltd_exp)
2667                                 continue;
2668
2669                         err = obd_set_info_async(tgt->ltd_exp,
2670                                                  keylen, key, vallen, val, set);
2671                         if (err && rc == 0)
2672                                 rc = err;
2673                 }
2674
2675                 RETURN(rc);
2676         }
2677
2678         RETURN(-EINVAL);
2679 }
2680
2681 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2682                struct lov_stripe_md *lsm)
2683 {
2684         struct obd_device         *obd = class_exp2obd(exp);
2685         struct lmv_obd            *lmv = &obd->u.lmv;
2686         struct lmv_stripe_md      *meap;
2687         struct lmv_stripe_md      *lsmp;
2688         int                        mea_size;
2689         int                        i;
2690         ENTRY;
2691
2692         mea_size = lmv_get_easize(lmv);
2693         if (!lmmp)
2694                 RETURN(mea_size);
2695
2696         if (*lmmp && !lsm) {
2697                 OBD_FREE_LARGE(*lmmp, mea_size);
2698                 *lmmp = NULL;
2699                 RETURN(0);
2700         }
2701
2702         if (*lmmp == NULL) {
2703                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2704                 if (*lmmp == NULL)
2705                         RETURN(-ENOMEM);
2706         }
2707
2708         if (!lsm)
2709                 RETURN(mea_size);
2710
2711         lsmp = (struct lmv_stripe_md *)lsm;
2712         meap = (struct lmv_stripe_md *)*lmmp;
2713
2714         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2715             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2716                 RETURN(-EINVAL);
2717
2718         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2719         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2720         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2721
2722         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2723                 meap->mea_ids[i] = meap->mea_ids[i];
2724                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2725         }
2726
2727         RETURN(mea_size);
2728 }
2729
2730 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2731                  struct lov_mds_md *lmm, int lmm_size)
2732 {
2733         struct obd_device          *obd = class_exp2obd(exp);
2734         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2735         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2736         struct lmv_obd             *lmv = &obd->u.lmv;
2737         int                         mea_size;
2738         int                         i;
2739         __u32                       magic;
2740         ENTRY;
2741
2742         mea_size = lmv_get_easize(lmv);
2743         if (lsmp == NULL)
2744                 return mea_size;
2745
2746         if (*lsmp != NULL && lmm == NULL) {
2747                 OBD_FREE_LARGE(*tmea, mea_size);
2748                 *lsmp = NULL;
2749                 RETURN(0);
2750         }
2751
2752         LASSERT(mea_size == lmm_size);
2753
2754         OBD_ALLOC_LARGE(*tmea, mea_size);
2755         if (*tmea == NULL)
2756                 RETURN(-ENOMEM);
2757
2758         if (!lmm)
2759                 RETURN(mea_size);
2760
2761         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2762             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2763             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2764         {
2765                 magic = le32_to_cpu(mea->mea_magic);
2766         } else {
2767                 /*
2768                  * Old mea is not handled here.
2769                  */
2770                 CERROR("Old not supportable EA is found\n");
2771                 LBUG();
2772         }
2773
2774         (*tmea)->mea_magic = magic;
2775         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2776         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2777
2778         for (i = 0; i < (*tmea)->mea_count; i++) {
2779                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2780                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2781         }
2782         RETURN(mea_size);
2783 }
2784
2785 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2786                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2787                              ldlm_cancel_flags_t flags, void *opaque)
2788 {
2789         struct obd_device       *obd = exp->exp_obd;
2790         struct lmv_obd          *lmv = &obd->u.lmv;
2791         int                      rc = 0;
2792         int                      err;
2793         int                      i;
2794         ENTRY;
2795
2796         LASSERT(fid != NULL);
2797
2798         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2799                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2800                         continue;
2801
2802                 err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
2803                                        policy, mode, flags, opaque);
2804                 if (!rc)
2805                         rc = err;
2806         }
2807         RETURN(rc);
2808 }
2809
2810 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2811                       __u32 *bits)
2812 {
2813         struct obd_device       *obd = exp->exp_obd;
2814         struct lmv_obd          *lmv = &obd->u.lmv;
2815         int                      rc;
2816         ENTRY;
2817
2818         rc =  md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data, bits);
2819         RETURN(rc);
2820 }
2821
2822 ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
2823                            const struct lu_fid *fid, ldlm_type_t type,
2824                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2825                            struct lustre_handle *lockh)
2826 {
2827         struct obd_device       *obd = exp->exp_obd;
2828         struct lmv_obd          *lmv = &obd->u.lmv;
2829         ldlm_mode_t              rc;
2830         int                      i;
2831         ENTRY;
2832
2833         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2834
2835         /*
2836          * With CMD every object can have two locks in different namespaces:
2837          * lookup lock in space of mds storing direntry and update/open lock in
2838          * space of mds storing inode. Thus we check all targets, not only that
2839          * one fid was created in.
2840          */
2841         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2842                 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2843                                    type, policy, mode, lockh);
2844                 if (rc)
2845                         RETURN(rc);
2846         }
2847
2848         RETURN(0);
2849 }
2850
2851 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2852                       struct obd_export *dt_exp, struct obd_export *md_exp,
2853                       struct lustre_md *md)
2854 {
2855         struct obd_device       *obd = exp->exp_obd;
2856         struct lmv_obd          *lmv = &obd->u.lmv;
2857         int                      rc;
2858         ENTRY;
2859         rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
2860         RETURN(rc);
2861 }
2862
2863 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2864 {
2865         struct obd_device       *obd = exp->exp_obd;
2866         struct lmv_obd          *lmv = &obd->u.lmv;
2867         ENTRY;
2868
2869         if (md->mea)
2870                 obd_free_memmd(exp, (void *)&md->mea);
2871         RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2872 }
2873
2874 int lmv_set_open_replay_data(struct obd_export *exp,
2875                              struct obd_client_handle *och,
2876                              struct ptlrpc_request *open_req)
2877 {
2878         struct obd_device       *obd = exp->exp_obd;
2879         struct lmv_obd          *lmv = &obd->u.lmv;
2880         struct lmv_tgt_desc     *tgt;
2881         ENTRY;
2882
2883         tgt = lmv_find_target(lmv, &och->och_fid);
2884         if (IS_ERR(tgt))
2885                 RETURN(PTR_ERR(tgt));
2886
2887         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2888 }
2889
2890 int lmv_clear_open_replay_data(struct obd_export *exp,
2891                                struct obd_client_handle *och)
2892 {
2893         struct obd_device       *obd = exp->exp_obd;
2894         struct lmv_obd          *lmv = &obd->u.lmv;
2895         struct lmv_tgt_desc     *tgt;
2896         ENTRY;
2897
2898         tgt = lmv_find_target(lmv, &och->och_fid);
2899         if (IS_ERR(tgt))
2900                 RETURN(PTR_ERR(tgt));
2901
2902         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2903 }
2904
2905 static int lmv_get_remote_perm(struct obd_export *exp,
2906                                const struct lu_fid *fid,
2907                                struct obd_capa *oc, __u32 suppgid,
2908                                struct ptlrpc_request **request)
2909 {
2910         struct obd_device       *obd = exp->exp_obd;
2911         struct lmv_obd          *lmv = &obd->u.lmv;
2912         struct lmv_tgt_desc     *tgt;
2913         int                      rc;
2914         ENTRY;
2915
2916         rc = lmv_check_connect(obd);
2917         if (rc)
2918                 RETURN(rc);
2919
2920         tgt = lmv_find_target(lmv, fid);
2921         if (IS_ERR(tgt))
2922                 RETURN(PTR_ERR(tgt));
2923
2924         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2925         RETURN(rc);
2926 }
2927
2928 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2929                           renew_capa_cb_t cb)
2930 {
2931         struct obd_device       *obd = exp->exp_obd;
2932         struct lmv_obd          *lmv = &obd->u.lmv;
2933         struct lmv_tgt_desc     *tgt;
2934         int                      rc;
2935         ENTRY;
2936
2937         rc = lmv_check_connect(obd);
2938         if (rc)
2939                 RETURN(rc);
2940
2941         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2942         if (IS_ERR(tgt))
2943                 RETURN(PTR_ERR(tgt));
2944
2945         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2946         RETURN(rc);
2947 }
2948
2949 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2950                     const struct req_msg_field *field, struct obd_capa **oc)
2951 {
2952         struct obd_device *obd = exp->exp_obd;
2953         struct lmv_obd *lmv = &obd->u.lmv;
2954         int rc;
2955
2956         ENTRY;
2957         rc = md_unpack_capa(lmv->tgts[0].ltd_exp, req, field, oc);
2958         RETURN(rc);
2959 }
2960
2961 int lmv_intent_getattr_async(struct obd_export *exp,
2962                              struct md_enqueue_info *minfo,
2963                              struct ldlm_enqueue_info *einfo)
2964 {
2965         struct md_op_data       *op_data = &minfo->mi_data;
2966         struct obd_device       *obd = exp->exp_obd;
2967         struct lmv_obd          *lmv = &obd->u.lmv;
2968         struct lmv_object       *obj;
2969         struct lmv_tgt_desc     *tgt;
2970         int                      rc;
2971         int                      sidx;
2972         ENTRY;
2973
2974         rc = lmv_check_connect(obd);
2975         if (rc)
2976                 RETURN(rc);
2977
2978         if (!fid_is_sane(&op_data->op_fid2)) {
2979                 obj = lmv_object_find(obd, &op_data->op_fid1);
2980                 if (obj && op_data->op_namelen) {
2981                         sidx = raw_name2idx(obj->lo_hashtype,
2982                                             obj->lo_objcount,
2983                                             (char *)op_data->op_name,
2984                                             op_data->op_namelen);
2985                         op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2986                         tgt = lmv_get_target(lmv,
2987                                              obj->lo_stripes[sidx].ls_mds);
2988                         CDEBUG(D_INODE,
2989                                "Choose slave dir ("DFID") -> mds #%d\n",
2990                                PFID(&op_data->op_fid1), tgt->ltd_idx);
2991                 } else {
2992                         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2993                 }
2994                 if (obj)
2995                         lmv_object_put(obj);
2996         } else {
2997                 op_data->op_fid1 = op_data->op_fid2;
2998                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2999                 op_data->op_bias = MDS_CROSS_REF;
3000                 /*
3001                  * Unfortunately, we have to lie to MDC/MDS to retrieve
3002                  * attributes llite needs.
3003                 */
3004                 if (minfo->mi_it.it_op & IT_LOOKUP)
3005                         minfo->mi_it.it_op = IT_GETATTR;
3006         }
3007
3008         if (IS_ERR(tgt))
3009                 RETURN(PTR_ERR(tgt));
3010
3011         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3012         RETURN(rc);
3013 }
3014
3015 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3016                         struct lu_fid *fid)
3017 {
3018         struct obd_device       *obd = exp->exp_obd;
3019         struct lmv_obd          *lmv = &obd->u.lmv;
3020         struct lmv_tgt_desc     *tgt;
3021         int                      rc;
3022         ENTRY;
3023
3024         rc = lmv_check_connect(obd);
3025         if (rc)
3026                 RETURN(rc);
3027
3028         tgt = lmv_find_target(lmv, fid);
3029         if (IS_ERR(tgt))
3030                 RETURN(PTR_ERR(tgt));
3031
3032         rc = md_revalidate_lock(tgt->ltd_exp, it, fid);
3033         RETURN(rc);
3034 }
3035
3036
3037 struct obd_ops lmv_obd_ops = {
3038         .o_owner                = THIS_MODULE,
3039         .o_setup                = lmv_setup,
3040         .o_cleanup              = lmv_cleanup,
3041         .o_precleanup           = lmv_precleanup,
3042         .o_process_config       = lmv_process_config,
3043         .o_connect              = lmv_connect,
3044         .o_disconnect           = lmv_disconnect,
3045         .o_statfs               = lmv_statfs,
3046         .o_get_info             = lmv_get_info,
3047         .o_set_info_async       = lmv_set_info_async,
3048         .o_packmd               = lmv_packmd,
3049         .o_unpackmd             = lmv_unpackmd,
3050         .o_notify               = lmv_notify,
3051         .o_get_uuid             = lmv_get_uuid,
3052         .o_iocontrol            = lmv_iocontrol,
3053         .o_fid_delete           = lmv_fid_delete
3054 };
3055
3056 struct md_ops lmv_md_ops = {
3057         .m_getstatus            = lmv_getstatus,
3058         .m_change_cbdata        = lmv_change_cbdata,
3059         .m_find_cbdata          = lmv_find_cbdata,
3060         .m_close                = lmv_close,
3061         .m_create               = lmv_create,
3062         .m_done_writing         = lmv_done_writing,
3063         .m_enqueue              = lmv_enqueue,
3064         .m_getattr              = lmv_getattr,
3065         .m_getxattr             = lmv_getxattr,
3066         .m_getattr_name         = lmv_getattr_name,
3067         .m_intent_lock          = lmv_intent_lock,
3068         .m_link                 = lmv_link,
3069         .m_rename               = lmv_rename,
3070         .m_setattr              = lmv_setattr,
3071         .m_setxattr             = lmv_setxattr,
3072         .m_sync                 = lmv_sync,
3073         .m_readpage             = lmv_readpage,
3074         .m_unlink               = lmv_unlink,
3075         .m_init_ea_size         = lmv_init_ea_size,
3076         .m_cancel_unused        = lmv_cancel_unused,
3077         .m_set_lock_data        = lmv_set_lock_data,
3078         .m_lock_match           = lmv_lock_match,
3079         .m_get_lustre_md        = lmv_get_lustre_md,
3080         .m_free_lustre_md       = lmv_free_lustre_md,
3081         .m_set_open_replay_data = lmv_set_open_replay_data,
3082         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3083         .m_renew_capa           = lmv_renew_capa,
3084         .m_unpack_capa          = lmv_unpack_capa,
3085         .m_get_remote_perm      = lmv_get_remote_perm,
3086         .m_intent_getattr_async = lmv_intent_getattr_async,
3087         .m_revalidate_lock      = lmv_revalidate_lock
3088 };
3089
3090 static quota_interface_t *quota_interface;
3091 extern quota_interface_t lmv_quota_interface;
3092
3093 int __init lmv_init(void)
3094 {
3095         struct lprocfs_static_vars lvars;
3096         int                        rc;
3097
3098         lmv_object_cache = cfs_mem_cache_create("lmv_objects",
3099                                                 sizeof(struct lmv_object),
3100                                                 0, 0);
3101         if (!lmv_object_cache) {
3102                 CERROR("Error allocating lmv objects cache\n");
3103                 return -ENOMEM;
3104         }
3105
3106         lprocfs_lmv_init_vars(&lvars);
3107
3108         cfs_request_module("lquota");
3109         quota_interface = PORTAL_SYMBOL_GET(lmv_quota_interface);
3110         init_obd_quota_ops(quota_interface, &lmv_obd_ops);
3111
3112         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
3113                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
3114         if (rc) {
3115                 if (quota_interface)
3116                         PORTAL_SYMBOL_PUT(lmv_quota_interface);
3117                 cfs_mem_cache_destroy(lmv_object_cache);
3118         }
3119
3120         return rc;
3121 }
3122
3123 #ifdef __KERNEL__
3124 static void lmv_exit(void)
3125 {
3126         if (quota_interface)
3127                 PORTAL_SYMBOL_PUT(lmv_quota_interface);
3128
3129         class_unregister_type(LUSTRE_LMV_NAME);
3130
3131         LASSERTF(cfs_atomic_read(&lmv_object_count) == 0,
3132                  "Can't free lmv objects cache, %d object(s) busy\n",
3133                  cfs_atomic_read(&lmv_object_count));
3134         cfs_mem_cache_destroy(lmv_object_cache);
3135 }
3136
3137 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3138 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3139 MODULE_LICENSE("GPL");
3140
3141 module_init(lmv_init);
3142 module_exit(lmv_exit);
3143 #endif