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  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  */
  39
  40 #ifndef EXPORT_SYMTAB
  41 # define EXPORT_SYMTAB
  42 #endif
  43 #define DEBUG_SUBSYSTEM S_LMV
  44 #ifdef __KERNEL__
  45 #include <linux/slab.h>
  46 #include <linux/module.h>
  47 #include <linux/init.h>
  48 #include <linux/slab.h>
  49 #include <linux/pagemap.h>
  50 #include <linux/mm.h>
  51 #include <asm/div64.h>
  52 #include <linux/seq_file.h>
  53 #include <linux/namei.h>
  54 #else
  55 #include <liblustre.h>
  56 #endif
  57
  58 #include <lustre_log.h>
  59 #include <obd_support.h>
  60 #include <lustre_lib.h>
  61 #include <lustre_net.h>
  62 #include <obd_class.h>
  63 #include <lprocfs_status.h>
  64 #include <lustre_lite.h>
  65 #include <lustre_fid.h>
  66 #include "lmv_internal.h"
  67
  68 /* object cache. */
  69 cfs_mem_cache_t *lmv_object_cache;
  70 cfs_atomic_t lmv_object_count = CFS_ATOMIC_INIT(0);
  71
  72 static void lmv_activate_target(struct lmv_obd *lmv,
  73                                 struct lmv_tgt_desc *tgt,
  74                                 int activate)
  75 {
  76         if (tgt->ltd_active == activate)
  77                 return;
  78
  79         tgt->ltd_active = activate;
  80         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  81 }
  82
  83 /**
  84  * Error codes:
  85  *
  86  *  -EINVAL  : UUID can't be found in the LMV's target list
  87  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  88  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  89  */
  90 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  91                               int activate)
  92 {
  93         struct lmv_tgt_desc    *tgt;
  94         struct obd_device      *obd;
  95         int                     i;
  96         int                     rc = 0;
  97         ENTRY;
  98
  99         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
 100                lmv, uuid->uuid, activate);
 101
 102         cfs_spin_lock(&lmv->lmv_lock);
 103         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 104                 if (tgt->ltd_exp == NULL)
 105                         continue;
 106
 107                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n",
 108                        i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 109
 110                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 111                         break;
 112         }
 113
 114         if (i == lmv->desc.ld_tgt_count)
 115                 GOTO(out_lmv_lock, rc = -EINVAL);
 116
 117         obd = class_exp2obd(tgt->ltd_exp);
 118         if (obd == NULL)
 119                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 120
 121         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 122                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 123                obd->obd_type->typ_name, i);
 124         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 125
 126         if (tgt->ltd_active == activate) {
 127                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 128                        activate ? "" : "in");
 129                 GOTO(out_lmv_lock, rc);
 130         }
 131
 132         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 133                activate ? "" : "in");
 134         lmv_activate_target(lmv, tgt, activate);
 135         EXIT;
 136
 137  out_lmv_lock:
 138         cfs_spin_unlock(&lmv->lmv_lock);
 139         return rc;
 140 }
 141
 142 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
 143                             struct obd_connect_data *data)
 144 {
 145         struct lmv_tgt_desc    *tgt;
 146         int                     i;
 147         ENTRY;
 148
 149         LASSERT(data != NULL);
 150
 151         cfs_spin_lock(&lmv->lmv_lock);
 152         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 153                 if (tgt->ltd_exp == NULL)
 154                         continue;
 155
 156                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 157                         lmv->datas[tgt->ltd_idx] = *data;
 158                         break;
 159                 }
 160         }
 161         cfs_spin_unlock(&lmv->lmv_lock);
 162         RETURN(0);
 163 }
 164
 165 struct obd_uuid *lmv_get_uuid(struct obd_export *exp) {
 166         struct obd_device *obd = exp->exp_obd;
 167         struct lmv_obd *lmv = &obd->u.lmv;
 168         return obd_get_uuid(lmv->tgts[0].ltd_exp);
 169 }
 170
 171 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 172                       enum obd_notify_event ev, void *data)
 173 {
 174         struct obd_connect_data *conn_data;
 175         struct lmv_obd          *lmv = &obd->u.lmv;
 176         struct obd_uuid         *uuid;
 177         int                      rc = 0;
 178         ENTRY;
 179
 180         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 181                 CERROR("unexpected notification of %s %s!\n",
 182                        watched->obd_type->typ_name,
 183                        watched->obd_name);
 184                 RETURN(-EINVAL);
 185         }
 186
 187         uuid = &watched->u.cli.cl_target_uuid;
 188         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 189                 /*
 190                  * Set MDC as active before notifying the observer, so the
 191                  * observer can use the MDC normally.
 192                  */
 193                 rc = lmv_set_mdc_active(lmv, uuid,
 194                                         ev == OBD_NOTIFY_ACTIVE);
 195                 if (rc) {
 196                         CERROR("%sactivation of %s failed: %d\n",
 197                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 198                                uuid->uuid, rc);
 199                         RETURN(rc);
 200                 }
 201         } else if (ev == OBD_NOTIFY_OCD) {
 202                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 203
 204                 /*
 205                  * Set connect data to desired target, update exp_connect_flags.
 206                  */
 207                 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
 208                 if (rc) {
 209                         CERROR("can't set connect data to target %s, rc %d\n",
 210                                uuid->uuid, rc);
 211                         RETURN(rc);
 212                 }
 213
 214                 /*
 215                  * XXX: Make sure that ocd_connect_flags from all targets are
 216                  * the same. Otherwise one of MDTs runs wrong version or
 217                  * something like this.  --umka
 218                  */
 219                 obd->obd_self_export->exp_connect_flags =
 220                         conn_data->ocd_connect_flags;
 221         }
 222 #if 0
 223         else if (ev == OBD_NOTIFY_DISCON) {
 224                 /*
 225                  * For disconnect event, flush fld cache for failout MDS case.
 226                  */
 227                 fld_client_flush(&lmv->lmv_fld);
 228         }
 229 #endif
 230         /*
 231          * Pass the notification up the chain.
 232          */
 233         if (obd->obd_observer)
 234                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 235
 236         RETURN(rc);
 237 }
 238
 239 /**
 240  * This is fake connect function. Its purpose is to initialize lmv and say
 241  * caller that everything is okay. Real connection will be performed later.
 242  */
 243 static int lmv_connect(const struct lu_env *env,
 244                        struct obd_export **exp, struct obd_device *obd,
 245                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 246                        void *localdata)
 247 {
 248 #ifdef __KERNEL__
 249         struct proc_dir_entry *lmv_proc_dir;
 250 #endif
 251         struct lmv_obd        *lmv = &obd->u.lmv;
 252         struct lustre_handle  conn = { 0 };
 253         int                    rc = 0;
 254         ENTRY;
 255
 256         /*
 257          * We don't want to actually do the underlying connections more than
 258          * once, so keep track.
 259          */
 260         lmv->refcount++;
 261         if (lmv->refcount > 1) {
 262                 *exp = NULL;
 263                 RETURN(0);
 264         }
 265
 266         rc = class_connect(&conn, obd, cluuid);
 267         if (rc) {
 268                 CERROR("class_connection() returned %d\n", rc);
 269                 RETURN(rc);
 270         }
 271
 272         *exp = class_conn2export(&conn);
 273         class_export_get(*exp);
 274
 275         lmv->exp = *exp;
 276         lmv->connected = 0;
 277         lmv->cluuid = *cluuid;
 278
 279         if (data)
 280                 lmv->conn_data = *data;
 281
 282 #ifdef __KERNEL__
 283         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 284                                         NULL, NULL);
 285         if (IS_ERR(lmv_proc_dir)) {
 286                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 287                        obd->obd_type->typ_name, obd->obd_name);
 288                 lmv_proc_dir = NULL;
 289         }
 290 #endif
 291
 292         /*
 293          * All real clients should perform actual connection right away, because
 294          * it is possible, that LMV will not have opportunity to connect targets
 295          * and MDC stuff will be called directly, for instance while reading
 296          * ../mdc/../kbytesfree procfs file, etc.
 297          */
 298         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 299                 rc = lmv_check_connect(obd);
 300
 301 #ifdef __KERNEL__
 302         if (rc) {
 303                 if (lmv_proc_dir)
 304                         lprocfs_remove(&lmv_proc_dir);
 305         }
 306 #endif
 307
 308         RETURN(rc);
 309 }
 310
 311 static void lmv_set_timeouts(struct obd_device *obd)
 312 {
 313         struct lmv_tgt_desc   *tgts;
 314         struct lmv_obd        *lmv;
 315         int                    i;
 316
 317         lmv = &obd->u.lmv;
 318         if (lmv->server_timeout == 0)
 319                 return;
 320
 321         if (lmv->connected == 0)
 322                 return;
 323
 324         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 325                 if (tgts->ltd_exp == NULL)
 326                         continue;
 327
 328                 obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS),
 329                                    KEY_INTERMDS, 0, NULL, NULL);
 330         }
 331 }
 332
 333 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 334                             int def_easize, int cookiesize)
 335 {
 336         struct obd_device   *obd = exp->exp_obd;
 337         struct lmv_obd      *lmv = &obd->u.lmv;
 338         int                  i;
 339         int                  rc = 0;
 340         int                  change = 0;
 341         ENTRY;
 342
 343         if (lmv->max_easize < easize) {
 344                 lmv->max_easize = easize;
 345                 change = 1;
 346         }
 347         if (lmv->max_def_easize < def_easize) {
 348                 lmv->max_def_easize = def_easize;
 349                 change = 1;
 350         }
 351         if (lmv->max_cookiesize < cookiesize) {
 352                 lmv->max_cookiesize = cookiesize;
 353                 change = 1;
 354         }
 355         if (change == 0)
 356                 RETURN(0);
 357
 358         if (lmv->connected == 0)
 359                 RETURN(0);
 360
 361         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 362                 if (lmv->tgts[i].ltd_exp == NULL) {
 363                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 364                         continue;
 365                 }
 366
 367                 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
 368                                      cookiesize);
 369                 if (rc) {
 370                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 371                                "error %d.\n", i, rc);
 372                         break;
 373                 }
 374         }
 375         RETURN(rc);
 376 }
 377
 378 #define MAX_STRING_SIZE 128
 379
 380 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 381 {
 382 #ifdef __KERNEL__
 383         struct proc_dir_entry   *lmv_proc_dir;
 384 #endif
 385         struct lmv_obd          *lmv = &obd->u.lmv;
 386         struct obd_uuid         *cluuid = &lmv->cluuid;
 387         struct obd_connect_data *mdc_data = NULL;
 388         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 389         struct obd_device       *mdc_obd;
 390         struct obd_export       *mdc_exp;
 391         struct lu_fld_target     target;
 392         int                      rc;
 393         ENTRY;
 394
 395         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 396                                         &obd->obd_uuid);
 397         if (!mdc_obd) {
 398                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 399                 RETURN(-EINVAL);
 400         }
 401
 402         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 403                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 404                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 405                 cluuid->uuid);
 406
 407         if (!mdc_obd->obd_set_up) {
 408                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 409                 RETURN(-EINVAL);
 410         }
 411
 412         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 413                          &lmv->conn_data, NULL);
 414         if (rc) {
 415                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 416                 RETURN(rc);
 417         }
 418
 419         /*
 420          * Init fid sequence client for this mdc and add new fld target.
 421          */
 422         rc = obd_fid_init(mdc_exp);
 423         if (rc)
 424                 RETURN(rc);
 425
 426         target.ft_srv = NULL;
 427         target.ft_exp = mdc_exp;
 428         target.ft_idx = tgt->ltd_idx;
 429
 430         fld_client_add_target(&lmv->lmv_fld, &target);
 431
 432         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 433
 434         rc = obd_register_observer(mdc_obd, obd);
 435         if (rc) {
 436                 obd_disconnect(mdc_exp);
 437                 CERROR("target %s register_observer error %d\n",
 438                        tgt->ltd_uuid.uuid, rc);
 439                 RETURN(rc);
 440         }
 441
 442         if (obd->obd_observer) {
 443                 /*
 444                  * Tell the observer about the new target.
 445                  */
 446                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 447                                 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
 448                 if (rc) {
 449                         obd_disconnect(mdc_exp);
 450                         RETURN(rc);
 451                 }
 452         }
 453
 454         tgt->ltd_active = 1;
 455         tgt->ltd_exp = mdc_exp;
 456         lmv->desc.ld_active_tgt_count++;
 457
 458         /*
 459          * Copy connect data, it may be used later.
 460          */
 461         lmv->datas[tgt->ltd_idx] = *mdc_data;
 462
 463         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 464                         lmv->max_def_easize, lmv->max_cookiesize);
 465
 466         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 467                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 468                 cfs_atomic_read(&obd->obd_refcount));
 469
 470 #ifdef __KERNEL__
 471         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 472         if (lmv_proc_dir) {
 473                 struct proc_dir_entry *mdc_symlink;
 474
 475                 LASSERT(mdc_obd->obd_type != NULL);
 476                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 477                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 478                                                   lmv_proc_dir,
 479                                                   "../../../%s/%s",
 480                                                   mdc_obd->obd_type->typ_name,
 481                                                   mdc_obd->obd_name);
 482                 if (mdc_symlink == NULL) {
 483                         CERROR("Could not register LMV target "
 484                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 485                                obd->obd_type->typ_name, obd->obd_name,
 486                                mdc_obd->obd_name);
 487                         lprocfs_remove(&lmv_proc_dir);
 488                         lmv_proc_dir = NULL;
 489                 }
 490         }
 491 #endif
 492         RETURN(0);
 493 }
 494
 495 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 496 {
 497         struct lmv_obd      *lmv = &obd->u.lmv;
 498         struct lmv_tgt_desc *tgt;
 499         int                  rc = 0;
 500         ENTRY;
 501
 502         CDEBUG(D_CONFIG, "Target uuid: %s.\n", tgt_uuid->uuid);
 503
 504         lmv_init_lock(lmv);
 505
 506         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 507                 lmv_init_unlock(lmv);
 508                 CERROR("Can't add %s, LMV module compiled for %d MDCs. "
 509                        "That many MDCs already configured.\n",
 510                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 511                 RETURN(-EINVAL);
 512         }
 513         if (lmv->desc.ld_tgt_count == 0) {
 514                 struct obd_device *mdc_obd;
 515
 516                 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
 517                                                 &obd->obd_uuid);
 518                 if (!mdc_obd) {
 519                         lmv_init_unlock(lmv);
 520                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 521                         RETURN(-EINVAL);
 522                 }
 523
 524                 rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, NULL);
 525                 if (rc) {
 526                         lmv_init_unlock(lmv);
 527                         CERROR("lmv failed to setup llogging subsystems\n");
 528                 }
 529         }
 530         cfs_spin_lock(&lmv->lmv_lock);
 531         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 532         tgt->ltd_uuid = *tgt_uuid;
 533         cfs_spin_unlock(&lmv->lmv_lock);
 534
 535         if (lmv->connected) {
 536                 rc = lmv_connect_mdc(obd, tgt);
 537                 if (rc) {
 538                         cfs_spin_lock(&lmv->lmv_lock);
 539                         lmv->desc.ld_tgt_count--;
 540                         memset(tgt, 0, sizeof(*tgt));
 541                         cfs_spin_unlock(&lmv->lmv_lock);
 542                 } else {
 543                         int easize = sizeof(struct lmv_stripe_md) +
 544                                      lmv->desc.ld_tgt_count *
 545                                      sizeof(struct lu_fid);
 546                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 547                 }
 548         }
 549
 550         lmv_init_unlock(lmv);
 551         RETURN(rc);
 552 }
 553
 554 int lmv_check_connect(struct obd_device *obd)
 555 {
 556         struct lmv_obd       *lmv = &obd->u.lmv;
 557         struct lmv_tgt_desc  *tgt;
 558         int                   i;
 559         int                   rc;
 560         int                   easize;
 561         ENTRY;
 562
 563         if (lmv->connected)
 564                 RETURN(0);
 565
 566         lmv_init_lock(lmv);
 567         if (lmv->connected) {
 568                 lmv_init_unlock(lmv);
 569                 RETURN(0);
 570         }
 571
 572         if (lmv->desc.ld_tgt_count == 0) {
 573                 CERROR("%s: no targets configured.\n", obd->obd_name);
 574                 RETURN(-EINVAL);
 575         }
 576
 577         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 578                lmv->cluuid.uuid, obd->obd_name);
 579
 580         LASSERT(lmv->tgts != NULL);
 581
 582         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 583                 rc = lmv_connect_mdc(obd, tgt);
 584                 if (rc)
 585                         GOTO(out_disc, rc);
 586         }
 587
 588         lmv_set_timeouts(obd);
 589         class_export_put(lmv->exp);
 590         lmv->connected = 1;
 591         easize = lmv_get_easize(lmv);
 592         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 593         lmv_init_unlock(lmv);
 594         RETURN(0);
 595
 596  out_disc:
 597         while (i-- > 0) {
 598                 int rc2;
 599                 --tgt;
 600                 tgt->ltd_active = 0;
 601                 if (tgt->ltd_exp) {
 602                         --lmv->desc.ld_active_tgt_count;
 603                         rc2 = obd_disconnect(tgt->ltd_exp);
 604                         if (rc2) {
 605                                 CERROR("LMV target %s disconnect on "
 606                                        "MDC idx %d: error %d\n",
 607                                        tgt->ltd_uuid.uuid, i, rc2);
 608                         }
 609                 }
 610         }
 611         class_disconnect(lmv->exp);
 612         lmv_init_unlock(lmv);
 613         RETURN(rc);
 614 }
 615
 616 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 617 {
 618 #ifdef __KERNEL__
 619         struct proc_dir_entry  *lmv_proc_dir;
 620 #endif
 621         struct lmv_obd         *lmv = &obd->u.lmv;
 622         struct obd_device      *mdc_obd;
 623         int                     rc;
 624         ENTRY;
 625
 626         LASSERT(tgt != NULL);
 627         LASSERT(obd != NULL);
 628
 629         mdc_obd = class_exp2obd(tgt->ltd_exp);
 630
 631         if (mdc_obd) {
 632                 mdc_obd->obd_force = obd->obd_force;
 633                 mdc_obd->obd_fail = obd->obd_fail;
 634                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 635         }
 636
 637 #ifdef __KERNEL__
 638         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 639         if (lmv_proc_dir) {
 640                 struct proc_dir_entry *mdc_symlink;
 641
 642                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 643                 if (mdc_symlink) {
 644                         lprocfs_remove(&mdc_symlink);
 645                 } else {
 646                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 647                                obd->obd_type->typ_name, obd->obd_name,
 648                                mdc_obd->obd_name);
 649                 }
 650         }
 651 #endif
 652         rc = obd_fid_fini(tgt->ltd_exp);
 653         if (rc)
 654                 CERROR("Can't finanize fids factory\n");
 655
 656         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 657                tgt->ltd_exp->exp_obd->obd_name,
 658                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 659
 660         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 661         rc = obd_disconnect(tgt->ltd_exp);
 662         if (rc) {
 663                 if (tgt->ltd_active) {
 664                         CERROR("Target %s disconnect error %d\n",
 665                                tgt->ltd_uuid.uuid, rc);
 666                 }
 667         }
 668
 669         lmv_activate_target(lmv, tgt, 0);
 670         tgt->ltd_exp = NULL;
 671         RETURN(0);
 672 }
 673
 674 static int lmv_disconnect(struct obd_export *exp)
 675 {
 676         struct obd_device     *obd = class_exp2obd(exp);
 677 #ifdef __KERNEL__
 678         struct proc_dir_entry *lmv_proc_dir;
 679 #endif
 680         struct lmv_obd        *lmv = &obd->u.lmv;
 681         int                    rc;
 682         int                    i;
 683         ENTRY;
 684
 685         if (!lmv->tgts)
 686                 goto out_local;
 687
 688         /*
 689          * Only disconnect the underlying layers on the final disconnect.
 690          */
 691         lmv->refcount--;
 692         if (lmv->refcount != 0)
 693                 goto out_local;
 694
 695         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 696                 if (lmv->tgts[i].ltd_exp == NULL)
 697                         continue;
 698                 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
 699         }
 700
 701 #ifdef __KERNEL__
 702         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 703         if (lmv_proc_dir) {
 704                 lprocfs_remove(&lmv_proc_dir);
 705         } else {
 706                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 707                        obd->obd_type->typ_name, obd->obd_name);
 708         }
 709 #endif
 710
 711 out_local:
 712         /*
 713          * This is the case when no real connection is established by
 714          * lmv_check_connect().
 715          */
 716         if (!lmv->connected)
 717                 class_export_put(exp);
 718         rc = class_disconnect(exp);
 719         if (lmv->refcount == 0)
 720                 lmv->connected = 0;
 721         RETURN(rc);
 722 }
 723
 724 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 725                          int len, void *karg, void *uarg)
 726 {
 727         struct obd_device    *obddev = class_exp2obd(exp);
 728         struct lmv_obd       *lmv = &obddev->u.lmv;
 729         int                   i = 0;
 730         int                   rc = 0;
 731         int                   set = 0;
 732         int                   count = lmv->desc.ld_tgt_count;
 733         ENTRY;
 734
 735         if (count == 0)
 736                 RETURN(-ENOTTY);
 737
 738         switch (cmd) {
 739         case IOC_OBD_STATFS: {
 740                 struct obd_ioctl_data *data = karg;
 741                 struct obd_device *mdc_obd;
 742                 struct obd_statfs stat_buf = {0};
 743                 __u32 index;
 744
 745                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 746                 if ((index >= count))
 747                         RETURN(-ENODEV);
 748
 749                 if (!lmv->tgts[index].ltd_active)
 750                         RETURN(-ENODATA);
 751
 752                 mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp);
 753                 if (!mdc_obd)
 754                         RETURN(-EINVAL);
 755
 756                 /* copy UUID */
 757                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 758                                      min((int) data->ioc_plen2,
 759                                          (int) sizeof(struct obd_uuid))))
 760                         RETURN(-EFAULT);
 761
 762                 rc = obd_statfs(mdc_obd, &stat_buf,
 763                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 764                                 0);
 765                 if (rc)
 766                         RETURN(rc);
 767                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
 768                                      min((int) data->ioc_plen1,
 769                                          (int) sizeof(stat_buf))))
 770                         RETURN(-EFAULT);
 771                 break;
 772         }
 773         case OBD_IOC_QUOTACTL: {
 774                 struct if_quotactl *qctl = karg;
 775                 struct lmv_tgt_desc *tgt = NULL;
 776                 struct obd_quotactl *oqctl;
 777
 778                 if (qctl->qc_valid == QC_MDTIDX) {
 779                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 780                                 RETURN(-EINVAL);
 781
 782                         tgt = &lmv->tgts[qctl->qc_idx];
 783                         if (!tgt->ltd_exp)
 784                                 RETURN(-EINVAL);
 785                 } else if (qctl->qc_valid == QC_UUID) {
 786                         for (i = 0; i < count; i++) {
 787                                 tgt = &lmv->tgts[i];
 788                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 789                                                      &qctl->obd_uuid))
 790                                         continue;
 791
 792                                 if (tgt->ltd_exp == NULL)
 793                                         RETURN(-EINVAL);
 794
 795                                 break;
 796                         }
 797                 } else {
 798                         RETURN(-EINVAL);
 799                 }
 800
 801                 if (i >= count)
 802                         RETURN(-EAGAIN);
 803
 804                 LASSERT(tgt && tgt->ltd_exp);
 805                 OBD_ALLOC_PTR(oqctl);
 806                 if (!oqctl)
 807                         RETURN(-ENOMEM);
 808
 809                 QCTL_COPY(oqctl, qctl);
 810                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 811                 if (rc == 0) {
 812                         QCTL_COPY(qctl, oqctl);
 813                         qctl->qc_valid = QC_MDTIDX;
 814                         qctl->obd_uuid = tgt->ltd_uuid;
 815                 }
 816                 OBD_FREE_PTR(oqctl);
 817                 break;
 818         }
 819         case OBD_IOC_CHANGELOG_SEND:
 820         case OBD_IOC_CHANGELOG_CLEAR: {
 821                 struct ioc_changelog *icc = karg;
 822
 823                 if (icc->icc_mdtindex >= count)
 824                         RETURN(-ENODEV);
 825
 826                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex].ltd_exp,
 827                                    sizeof(*icc), icc, NULL);
 828                 break;
 829         }
 830         case LL_IOC_GET_CONNECT_FLAGS: {
 831                 rc = obd_iocontrol(cmd, lmv->tgts[0].ltd_exp, len, karg, uarg);
 832                 break;
 833         }
 834
 835         default : {
 836                 for (i = 0; i < count; i++) {
 837                         int err;
 838                         struct obd_device *mdc_obd;
 839
 840                         if (lmv->tgts[i].ltd_exp == NULL)
 841                                 continue;
 842                         /* ll_umount_begin() sets force flag but for lmv, not
 843                          * mdc. Let's pass it through */
 844                         mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 845                         mdc_obd->obd_force = obddev->obd_force;
 846                         err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len,
 847                                             karg, uarg);
 848                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
 849                                 RETURN(err);
 850                         } else if (err) {
 851                                 if (lmv->tgts[i].ltd_active) {
 852                                         CERROR("error: iocontrol MDC %s on MDT"
 853                                                "idx %d cmd %x: err = %d\n",
 854                                                 lmv->tgts[i].ltd_uuid.uuid,
 855                                                 i, cmd, err);
 856                                         if (!rc)
 857                                                 rc = err;
 858                                 }
 859                         } else
 860                                 set = 1;
 861                 }
 862                 if (!set && !rc)
 863                         rc = -EIO;
 864         }
 865         }
 866         RETURN(rc);
 867 }
 868
 869 static int lmv_all_chars_policy(int count, const char *name,
 870                                 int len)
 871 {
 872         unsigned int c = 0;
 873
 874         while (len > 0)
 875                 c += name[--len];
 876         c = c % count;
 877         return c;
 878 }
 879
 880 static int lmv_nid_policy(struct lmv_obd *lmv)
 881 {
 882         struct obd_import *imp;
 883         __u32              id;
 884
 885         /*
 886          * XXX: To get nid we assume that underlying obd device is mdc.
 887          */
 888         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 889         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 890         return id % lmv->desc.ld_tgt_count;
 891 }
 892
 893 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 894                           placement_policy_t placement)
 895 {
 896         switch (placement) {
 897         case PLACEMENT_CHAR_POLICY:
 898                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 899                                             op_data->op_name,
 900                                             op_data->op_namelen);
 901         case PLACEMENT_NID_POLICY:
 902                 return lmv_nid_policy(lmv);
 903
 904         default:
 905                 break;
 906         }
 907
 908         CERROR("Unsupported placement policy %x\n", placement);
 909         return -EINVAL;
 910 }
 911
 912 /**
 913  * This is _inode_ placement policy function (not name).
 914  */
 915 static int lmv_placement_policy(struct obd_device *obd,
 916                                 struct md_op_data *op_data,
 917                                 mdsno_t *mds)
 918 {
 919         struct lmv_obd          *lmv = &obd->u.lmv;
 920         struct lmv_object       *obj;
 921         int                      rc;
 922         ENTRY;
 923
 924         LASSERT(mds != NULL);
 925
 926         if (lmv->desc.ld_tgt_count == 1) {
 927                 *mds = 0;
 928                 RETURN(0);
 929         }
 930
 931         /*
 932          * Allocate new fid on target according to operation type and parent
 933          * home mds.
 934          */
 935         obj = lmv_object_find(obd, &op_data->op_fid1);
 936         if (obj != NULL || op_data->op_name == NULL ||
 937             op_data->op_opc != LUSTRE_OPC_MKDIR) {
 938                 /*
 939                  * Allocate fid for non-dir or for null name or for case parent
 940                  * dir is split.
 941                  */
 942                 if (obj) {
 943                         lmv_object_put(obj);
 944
 945                         /*
 946                          * If we have this flag turned on, and we see that
 947                          * parent dir is split, this means, that caller did not
 948                          * notice split yet. This is race and we would like to
 949                          * let caller know that.
 950                          */
 951                         if (op_data->op_bias & MDS_CHECK_SPLIT)
 952                                 RETURN(-ERESTART);
 953                 }
 954
 955                 /*
 956                  * Allocate new fid on same mds where parent fid is located and
 957                  * where operation will be sent. In case of split dir, ->op_fid1
 958                  * and ->op_mds here will contain fid and mds of slave directory
 959                  * object (assigned by caller).
 960                  */
 961                 *mds = op_data->op_mds;
 962                 rc = 0;
 963         } else {
 964                 /*
 965                  * Parent directory is not split and we want to create a
 966                  * directory in it. Let's calculate where to place it according
 967                  * to operation data @op_data.
 968                  */
 969                 *mds = lmv_choose_mds(lmv, op_data, lmv->lmv_placement);
 970                 rc = 0;
 971         }
 972
 973         if (rc) {
 974                 CERROR("Can't choose MDS, err = %d\n", rc);
 975         } else {
 976                 LASSERT(*mds < lmv->desc.ld_tgt_count);
 977         }
 978
 979         RETURN(rc);
 980 }
 981
 982 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
 983                     mdsno_t mds)
 984 {
 985         struct lmv_tgt_desc *tgt;
 986         int                  rc;
 987         ENTRY;
 988
 989         tgt = lmv_get_target(lmv, mds);
 990
 991         /*
 992          * New seq alloc and FLD setup should be atomic. Otherwise we may find
 993          * on server that seq in new allocated fid is not yet known.
 994          */
 995         cfs_down(&tgt->ltd_fid_sem);
 996
 997         if (!tgt->ltd_active)
 998                 GOTO(out, rc = -ENODEV);
 999
1000         /*
1001          * Asking underlaying tgt layer to allocate new fid.
1002          */
1003         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1004         if (rc > 0) {
1005                 LASSERT(fid_is_sane(fid));
1006                 rc = 0;
1007         }
1008
1009         EXIT;
1010 out:
1011         cfs_up(&tgt->ltd_fid_sem);
1012         return rc;
1013 }
1014
1015 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1016                   struct md_op_data *op_data)
1017 {
1018         struct obd_device     *obd = class_exp2obd(exp);
1019         struct lmv_obd        *lmv = &obd->u.lmv;
1020         mdsno_t                mds = 0;
1021         int                    rc;
1022         ENTRY;
1023
1024         LASSERT(op_data != NULL);
1025         LASSERT(fid != NULL);
1026
1027         rc = lmv_placement_policy(obd, op_data, &mds);
1028         if (rc) {
1029                 CERROR("Can't get target for allocating fid, "
1030                        "rc %d\n", rc);
1031                 RETURN(rc);
1032         }
1033
1034         rc = __lmv_fid_alloc(lmv, fid, mds);
1035         if (rc) {
1036                 CERROR("Can't alloc new fid, rc %d\n", rc);
1037                 RETURN(rc);
1038         }
1039
1040         RETURN(rc);
1041 }
1042
1043 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1044 {
1045         struct lmv_obd             *lmv = &obd->u.lmv;
1046         struct lprocfs_static_vars  lvars;
1047         struct lmv_desc            *desc;
1048         int                         rc;
1049         int                         i = 0;
1050         ENTRY;
1051
1052         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1053                 CERROR("LMV setup requires a descriptor\n");
1054                 RETURN(-EINVAL);
1055         }
1056
1057         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1058         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1059                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1060                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1061                 RETURN(-EINVAL);
1062         }
1063
1064         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
1065
1066         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
1067         if (lmv->tgts == NULL)
1068                 RETURN(-ENOMEM);
1069
1070         for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
1071                 cfs_sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
1072                 lmv->tgts[i].ltd_idx = i;
1073         }
1074
1075         lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
1076
1077         OBD_ALLOC(lmv->datas, lmv->datas_size);
1078         if (lmv->datas == NULL)
1079                 GOTO(out_free_tgts, rc = -ENOMEM);
1080
1081         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1082         lmv->desc.ld_tgt_count = 0;
1083         lmv->desc.ld_active_tgt_count = 0;
1084         lmv->max_cookiesize = 0;
1085         lmv->max_def_easize = 0;
1086         lmv->max_easize = 0;
1087         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1088
1089         cfs_spin_lock_init(&lmv->lmv_lock);
1090         cfs_sema_init(&lmv->init_sem, 1);
1091
1092         rc = lmv_object_setup(obd);
1093         if (rc) {
1094                 CERROR("Can't setup LMV object manager, error %d.\n", rc);
1095                 GOTO(out_free_datas, rc);
1096         }
1097
1098         lprocfs_lmv_init_vars(&lvars);
1099         lprocfs_obd_setup(obd, lvars.obd_vars);
1100 #ifdef LPROCFS
1101         {
1102                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status",
1103                                         0444, &lmv_proc_target_fops, obd);
1104                 if (rc)
1105                         CWARN("Error adding target_obd_stats file (%d)\n", rc);
1106        }
1107 #endif
1108         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1109                              LUSTRE_CLI_FLD_HASH_DHT);
1110         if (rc) {
1111                 CERROR("Can't init FLD, err %d\n", rc);
1112                 GOTO(out_free_datas, rc);
1113         }
1114
1115         RETURN(0);
1116
1117 out_free_datas:
1118         OBD_FREE(lmv->datas, lmv->datas_size);
1119         lmv->datas = NULL;
1120 out_free_tgts:
1121         OBD_FREE(lmv->tgts, lmv->tgts_size);
1122         lmv->tgts = NULL;
1123         return rc;
1124 }
1125
1126 static int lmv_cleanup(struct obd_device *obd)
1127 {
1128         struct lmv_obd   *lmv = &obd->u.lmv;
1129         ENTRY;
1130
1131         fld_client_fini(&lmv->lmv_fld);
1132         lmv_object_cleanup(obd);
1133         OBD_FREE(lmv->datas, lmv->datas_size);
1134         OBD_FREE(lmv->tgts, lmv->tgts_size);
1135
1136         RETURN(0);
1137 }
1138
1139 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1140 {
1141         struct lustre_cfg     *lcfg = buf;
1142         struct obd_uuid        tgt_uuid;
1143         int                    rc;
1144         ENTRY;
1145
1146         switch(lcfg->lcfg_command) {
1147         case LCFG_ADD_MDC:
1148                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1149                         GOTO(out, rc = -EINVAL);
1150
1151                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1152                 rc = lmv_add_target(obd, &tgt_uuid);
1153                 GOTO(out, rc);
1154         default: {
1155                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1156                 GOTO(out, rc = -EINVAL);
1157         }
1158         }
1159 out:
1160         RETURN(rc);
1161 }
1162
1163 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1164                       __u64 max_age, __u32 flags)
1165 {
1166         struct lmv_obd        *lmv = &obd->u.lmv;
1167         struct obd_statfs     *temp;
1168         int                    rc = 0;
1169         int                    i;
1170         ENTRY;
1171
1172         rc = lmv_check_connect(obd);
1173         if (rc)
1174                 RETURN(rc);
1175
1176         OBD_ALLOC(temp, sizeof(*temp));
1177         if (temp == NULL)
1178                 RETURN(-ENOMEM);
1179
1180         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1181                 if (lmv->tgts[i].ltd_exp == NULL)
1182                         continue;
1183
1184                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp,
1185                                 max_age, flags);
1186                 if (rc) {
1187                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1188                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1189                                rc);
1190                         GOTO(out_free_temp, rc);
1191                 }
1192                 if (i == 0) {
1193                         *osfs = *temp;
1194                 } else {
1195                         osfs->os_bavail += temp->os_bavail;
1196                         osfs->os_blocks += temp->os_blocks;
1197                         osfs->os_ffree += temp->os_ffree;
1198                         osfs->os_files += temp->os_files;
1199                 }
1200         }
1201
1202         EXIT;
1203 out_free_temp:
1204         OBD_FREE(temp, sizeof(*temp));
1205         return rc;
1206 }
1207
1208 static int lmv_getstatus(struct obd_export *exp,
1209                          struct lu_fid *fid,
1210                          struct obd_capa **pc)
1211 {
1212         struct obd_device    *obd = exp->exp_obd;
1213         struct lmv_obd       *lmv = &obd->u.lmv;
1214         int                   rc;
1215         ENTRY;
1216
1217         rc = lmv_check_connect(obd);
1218         if (rc)
1219                 RETURN(rc);
1220
1221         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1222         RETURN(rc);
1223 }
1224
1225 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1226                         struct obd_capa *oc, obd_valid valid, const char *name,
1227                         const char *input, int input_size, int output_size,
1228                         int flags, struct ptlrpc_request **request)
1229 {
1230         struct obd_device      *obd = exp->exp_obd;
1231         struct lmv_obd         *lmv = &obd->u.lmv;
1232         struct lmv_tgt_desc    *tgt;
1233         int                     rc;
1234         ENTRY;
1235
1236         rc = lmv_check_connect(obd);
1237         if (rc)
1238                 RETURN(rc);
1239
1240         tgt = lmv_find_target(lmv, fid);
1241         if (IS_ERR(tgt))
1242                 RETURN(PTR_ERR(tgt));
1243
1244         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1245                          input_size, output_size, flags, request);
1246
1247         RETURN(rc);
1248 }
1249
1250 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1251                         struct obd_capa *oc, obd_valid valid, const char *name,
1252                         const char *input, int input_size, int output_size,
1253                         int flags, __u32 suppgid,
1254                         struct ptlrpc_request **request)
1255 {
1256         struct obd_device      *obd = exp->exp_obd;
1257         struct lmv_obd         *lmv = &obd->u.lmv;
1258         struct lmv_tgt_desc    *tgt;
1259         int                     rc;
1260         ENTRY;
1261
1262         rc = lmv_check_connect(obd);
1263         if (rc)
1264                 RETURN(rc);
1265
1266         tgt = lmv_find_target(lmv, fid);
1267         if (IS_ERR(tgt))
1268                 RETURN(PTR_ERR(tgt));
1269
1270         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1271                          input_size, output_size, flags, suppgid,
1272                          request);
1273
1274         RETURN(rc);
1275 }
1276
1277 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1278                        struct ptlrpc_request **request)
1279 {
1280         struct obd_device       *obd = exp->exp_obd;
1281         struct lmv_obd          *lmv = &obd->u.lmv;
1282         struct lmv_tgt_desc     *tgt;
1283         struct lmv_object       *obj;
1284         int                      rc;
1285         int                      i;
1286         ENTRY;
1287
1288         rc = lmv_check_connect(obd);
1289         if (rc)
1290                 RETURN(rc);
1291
1292         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1293         if (IS_ERR(tgt))
1294                 RETURN(PTR_ERR(tgt));
1295
1296         if (op_data->op_valid & OBD_MD_MDTIDX) {
1297                 op_data->op_mds = tgt->ltd_idx;
1298                 RETURN(0);
1299         }
1300
1301         rc = md_getattr(tgt->ltd_exp, op_data, request);
1302         if (rc)
1303                 RETURN(rc);
1304
1305         obj = lmv_object_find_lock(obd, &op_data->op_fid1);
1306
1307         CDEBUG(D_INODE, "GETATTR for "DFID" %s\n", PFID(&op_data->op_fid1),
1308                obj ? "(split)" : "");
1309
1310         /*
1311          * If object is split, then we loop over all the slaves and gather size
1312          * attribute. In ideal world we would have to gather also mds field from
1313          * all slaves, as object is spread over the cluster and this is
1314          * definitely interesting information and it is not good to loss it,
1315          * but...
1316          */
1317         if (obj) {
1318                 struct mdt_body *body;
1319
1320                 if (*request == NULL) {
1321                         lmv_object_put(obj);
1322                         RETURN(rc);
1323                 }
1324
1325                 body = req_capsule_server_get(&(*request)->rq_pill,
1326                                               &RMF_MDT_BODY);
1327                 LASSERT(body != NULL);
1328
1329                 for (i = 0; i < obj->lo_objcount; i++) {
1330                         if (lmv->tgts[i].ltd_exp == NULL) {
1331                                 CWARN("%s: NULL export for %d\n",
1332                                       obd->obd_name, i);
1333                                 continue;
1334                         }
1335
1336                         /*
1337                          * Skip master object.
1338                          */
1339                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid))
1340                                 continue;
1341
1342                         body->size += obj->lo_stripes[i].ls_size;
1343                 }
1344
1345                 lmv_object_put_unlock(obj);
1346         }
1347
1348         RETURN(rc);
1349 }
1350
1351 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1352                              ldlm_iterator_t it, void *data)
1353 {
1354         struct obd_device   *obd = exp->exp_obd;
1355         struct lmv_obd      *lmv = &obd->u.lmv;
1356         int                  i;
1357         int                  rc;
1358         ENTRY;
1359
1360         rc = lmv_check_connect(obd);
1361         if (rc)
1362                 RETURN(rc);
1363
1364         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1365
1366         /*
1367          * With CMD every object can have two locks in different namespaces:
1368          * lookup lock in space of mds storing direntry and update/open lock in
1369          * space of mds storing inode.
1370          */
1371         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1372                 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1373
1374         RETURN(0);
1375 }
1376
1377 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1378                            ldlm_iterator_t it, void *data)
1379 {
1380         struct obd_device   *obd = exp->exp_obd;
1381         struct lmv_obd      *lmv = &obd->u.lmv;
1382         int                  i;
1383         int                  rc;
1384         ENTRY;
1385
1386         rc = lmv_check_connect(obd);
1387         if (rc)
1388                 RETURN(rc);
1389
1390         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1391
1392         /*
1393          * With CMD every object can have two locks in different namespaces:
1394          * lookup lock in space of mds storing direntry and update/open lock in
1395          * space of mds storing inode.
1396          */
1397         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1398                 rc = md_find_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1399                 if (rc)
1400                         RETURN(rc);
1401         }
1402
1403         RETURN(rc);
1404 }
1405
1406
1407 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1408                      struct md_open_data *mod, struct ptlrpc_request **request)
1409 {
1410         struct obd_device     *obd = exp->exp_obd;
1411         struct lmv_obd        *lmv = &obd->u.lmv;
1412         struct lmv_tgt_desc   *tgt;
1413         int                    rc;
1414         ENTRY;
1415
1416         rc = lmv_check_connect(obd);
1417         if (rc)
1418                 RETURN(rc);
1419
1420         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1421         if (IS_ERR(tgt))
1422                 RETURN(PTR_ERR(tgt));
1423
1424         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1425         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1426         RETURN(rc);
1427 }
1428
1429 /**
1430  * Called in the case MDS returns -ERESTART on create on open, what means that
1431  * directory is split and its LMV presentation object has to be updated.
1432  */
1433 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1434 {
1435         struct obd_device       *obd = exp->exp_obd;
1436         struct lmv_obd          *lmv = &obd->u.lmv;
1437         struct ptlrpc_request   *req = NULL;
1438         struct lmv_tgt_desc     *tgt;
1439         struct lmv_object       *obj;
1440         struct lustre_md         md;
1441         struct md_op_data       *op_data;
1442         int                      mealen;
1443         int                      rc;
1444         __u64                    valid;
1445         ENTRY;
1446
1447         md.mea = NULL;
1448         mealen = lmv_get_easize(lmv);
1449
1450         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1451
1452         tgt = lmv_find_target(lmv, fid);
1453         if (IS_ERR(tgt))
1454                 RETURN(PTR_ERR(tgt));
1455
1456         /*
1457          * Time to update mea of parent fid.
1458          */
1459
1460         OBD_ALLOC_PTR(op_data);
1461         if (op_data == NULL)
1462                 RETURN(-ENOMEM);
1463
1464         op_data->op_fid1 = *fid;
1465         op_data->op_mode = mealen;
1466         op_data->op_valid = valid;
1467
1468         rc = md_getattr(tgt->ltd_exp, op_data, &req);
1469         OBD_FREE_PTR(op_data);
1470         if (rc) {
1471                 CERROR("md_getattr() failed, error %d\n", rc);
1472                 GOTO(cleanup, rc);
1473         }
1474
1475         rc = md_get_lustre_md(tgt->ltd_exp, req, NULL, exp, &md);
1476         if (rc) {
1477                 CERROR("md_get_lustre_md() failed, error %d\n", rc);
1478                 GOTO(cleanup, rc);
1479         }
1480
1481         if (md.mea == NULL)
1482                 GOTO(cleanup, rc = -ENODATA);
1483
1484         obj = lmv_object_create(exp, fid, md.mea);
1485         if (IS_ERR(obj))
1486                 rc = PTR_ERR(obj);
1487         else
1488                 lmv_object_put(obj);
1489
1490         obd_free_memmd(exp, (void *)&md.mea);
1491         EXIT;
1492 cleanup:
1493         if (req)
1494                 ptlrpc_req_finished(req);
1495         return rc;
1496 }
1497
1498 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1499                const void *data, int datalen, int mode, __u32 uid,
1500                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1501                struct ptlrpc_request **request)
1502 {
1503         struct obd_device       *obd = exp->exp_obd;
1504         struct lmv_obd          *lmv = &obd->u.lmv;
1505         struct lmv_tgt_desc     *tgt;
1506         struct lmv_object       *obj;
1507         int                      rc;
1508         int                      loop = 0;
1509         int                      sidx;
1510         ENTRY;
1511
1512         rc = lmv_check_connect(obd);
1513         if (rc)
1514                 RETURN(rc);
1515
1516         if (!lmv->desc.ld_active_tgt_count)
1517                 RETURN(-EIO);
1518 repeat:
1519         ++loop;
1520         LASSERT(loop <= 2);
1521
1522         obj = lmv_object_find(obd, &op_data->op_fid1);
1523         if (obj) {
1524                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1525                                        op_data->op_name, op_data->op_namelen);
1526                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1527                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1528                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1529                 tgt = lmv_get_target(lmv, op_data->op_mds);
1530                 lmv_object_put(obj);
1531         } else {
1532                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1533                 op_data->op_bias |= MDS_CHECK_SPLIT;
1534                 op_data->op_mds = tgt->ltd_idx;
1535         }
1536
1537         if (IS_ERR(tgt))
1538                 RETURN(PTR_ERR(tgt));
1539
1540         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1541         if (rc == -ERESTART)
1542                 goto repeat;
1543         else if (rc)
1544                 RETURN(rc);
1545
1546         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1547                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1548                op_data->op_mds);
1549
1550         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1551         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1552                        cap_effective, rdev, request);
1553         if (rc == 0) {
1554                 if (*request == NULL)
1555                         RETURN(rc);
1556                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1557         } else if (rc == -ERESTART) {
1558                 LASSERT(*request != NULL);
1559                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1560                           "Got -ERESTART during create!\n");
1561                 ptlrpc_req_finished(*request);
1562                 *request = NULL;
1563
1564                 /*
1565                  * Directory got split. Time to update local object and repeat
1566                  * the request with proper MDS.
1567                  */
1568                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1569                 if (rc == 0) {
1570                         rc = lmv_allocate_slaves(obd, &op_data->op_fid1,
1571                                                  op_data, &op_data->op_fid2);
1572                         if (rc)
1573                                 RETURN(rc);
1574                         goto repeat;
1575                 }
1576         }
1577         RETURN(rc);
1578 }
1579
1580 static int lmv_done_writing(struct obd_export *exp,
1581                             struct md_op_data *op_data,
1582                             struct md_open_data *mod)
1583 {
1584         struct obd_device     *obd = exp->exp_obd;
1585         struct lmv_obd        *lmv = &obd->u.lmv;
1586         struct lmv_tgt_desc   *tgt;
1587         int                    rc;
1588         ENTRY;
1589
1590         rc = lmv_check_connect(obd);
1591         if (rc)
1592                 RETURN(rc);
1593
1594         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1595         if (IS_ERR(tgt))
1596                 RETURN(PTR_ERR(tgt));
1597
1598         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1599         RETURN(rc);
1600 }
1601
1602 static int
1603 lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1604                    struct lookup_intent *it, struct md_op_data *op_data,
1605                    struct lustre_handle *lockh, void *lmm, int lmmsize)
1606 {
1607         struct obd_device     *obd = exp->exp_obd;
1608         struct lmv_obd        *lmv = &obd->u.lmv;
1609         struct lmv_stripe_md  *mea = op_data->op_mea1;
1610         struct md_op_data     *op_data2;
1611         struct lmv_tgt_desc   *tgt;
1612         int                    i;
1613         int                    rc = 0;
1614         ENTRY;
1615
1616         OBD_ALLOC_PTR(op_data2);
1617         if (op_data2 == NULL)
1618                 RETURN(-ENOMEM);
1619
1620         LASSERT(mea != NULL);
1621         for (i = 0; i < mea->mea_count; i++) {
1622                 memset(op_data2, 0, sizeof(*op_data2));
1623                 op_data2->op_fid1 = mea->mea_ids[i];
1624                 op_data2->op_bias = 0;
1625
1626                 tgt = lmv_find_target(lmv, &op_data2->op_fid1);
1627                 if (IS_ERR(tgt))
1628                         GOTO(cleanup, rc = PTR_ERR(tgt));
1629
1630                 if (tgt->ltd_exp == NULL)
1631                         continue;
1632
1633                 rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data2,
1634                                 lockh + i, lmm, lmmsize, NULL, 0);
1635
1636                 CDEBUG(D_INODE, "Take lock on slave "DFID" -> %d/%d\n",
1637                        PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1638
1639                 if (rc)
1640                         GOTO(cleanup, rc);
1641
1642                 if (it->d.lustre.it_data) {
1643                         struct ptlrpc_request *req;
1644                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1645                         ptlrpc_req_finished(req);
1646                 }
1647
1648                 if (it->d.lustre.it_status)
1649                         GOTO(cleanup, rc = it->d.lustre.it_status);
1650         }
1651
1652         EXIT;
1653 cleanup:
1654         OBD_FREE_PTR(op_data2);
1655
1656         if (rc != 0) {
1657                 /*
1658                  * Drop all taken locks.
1659                  */
1660                 while (--i >= 0) {
1661                         if (lockh[i].cookie)
1662                                 ldlm_lock_decref(lockh + i, einfo->ei_mode);
1663                         lockh[i].cookie = 0;
1664                 }
1665         }
1666         return rc;
1667 }
1668
1669 static int
1670 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1671                    struct lookup_intent *it, struct md_op_data *op_data,
1672                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1673                    int extra_lock_flags)
1674 {
1675         struct ptlrpc_request      *req = it->d.lustre.it_data;
1676         struct obd_device          *obd = exp->exp_obd;
1677         struct lmv_obd             *lmv = &obd->u.lmv;
1678         struct lustre_handle        plock;
1679         struct lmv_tgt_desc        *tgt;
1680         struct md_op_data          *rdata;
1681         struct lu_fid               fid1;
1682         struct mdt_body            *body;
1683         int                         rc = 0;
1684         int                         pmode;
1685         ENTRY;
1686
1687         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1688         LASSERT(body != NULL);
1689
1690         if (!(body->valid & OBD_MD_MDS))
1691                 RETURN(0);
1692
1693         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1694                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1695
1696         /*
1697          * We got LOOKUP lock, but we really need attrs.
1698          */
1699         pmode = it->d.lustre.it_lock_mode;
1700         LASSERT(pmode != 0);
1701         memcpy(&plock, lockh, sizeof(plock));
1702         it->d.lustre.it_lock_mode = 0;
1703         it->d.lustre.it_data = NULL;
1704         fid1 = body->fid1;
1705
1706         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1707         ptlrpc_req_finished(req);
1708
1709         tgt = lmv_find_target(lmv, &fid1);
1710         if (IS_ERR(tgt))
1711                 GOTO(out, rc = PTR_ERR(tgt));
1712
1713         OBD_ALLOC_PTR(rdata);
1714         if (rdata == NULL)
1715                 GOTO(out, rc = -ENOMEM);
1716
1717         rdata->op_fid1 = fid1;
1718         rdata->op_bias = MDS_CROSS_REF;
1719
1720         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1721                         lmm, lmmsize, NULL, extra_lock_flags);
1722         OBD_FREE_PTR(rdata);
1723         EXIT;
1724 out:
1725         ldlm_lock_decref(&plock, pmode);
1726         return rc;
1727 }
1728
1729 static int
1730 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1731             struct lookup_intent *it, struct md_op_data *op_data,
1732             struct lustre_handle *lockh, void *lmm, int lmmsize,
1733             struct ptlrpc_request **req, int extra_lock_flags)
1734 {
1735         struct obd_device        *obd = exp->exp_obd;
1736         struct lmv_obd           *lmv = &obd->u.lmv;
1737         struct lmv_tgt_desc      *tgt;
1738         struct lmv_object        *obj;
1739         int                       sidx;
1740         int                       rc;
1741         ENTRY;
1742
1743         rc = lmv_check_connect(obd);
1744         if (rc)
1745                 RETURN(rc);
1746
1747         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1748                LL_IT2STR(it), PFID(&op_data->op_fid1));
1749
1750         if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) {
1751                 rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
1752                                         lockh, lmm, lmmsize);
1753                 RETURN(rc);
1754         }
1755
1756         obj = lmv_object_find(obd, &op_data->op_fid1);
1757         if (obj && op_data->op_namelen) {
1758                 sidx = raw_name2idx(obj->lo_hashtype,
1759                                        obj->lo_objcount,
1760                                        (char *)op_data->op_name,
1761                                        op_data->op_namelen);
1762                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
1763                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1764         } else {
1765                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1766         }
1767         if (obj)
1768                 lmv_object_put(obj);
1769
1770         if (IS_ERR(tgt))
1771                 RETURN(PTR_ERR(tgt));
1772
1773         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1774                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1775
1776         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1777                         lmm, lmmsize, req, extra_lock_flags);
1778
1779         if (rc == 0 && it && it->it_op == IT_OPEN) {
1780                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1781                                         lmm, lmmsize, extra_lock_flags);
1782         }
1783         RETURN(rc);
1784 }
1785
1786 static int
1787 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1788                  struct ptlrpc_request **request)
1789 {
1790         struct ptlrpc_request   *req = NULL;
1791         struct obd_device       *obd = exp->exp_obd;
1792         struct lmv_obd          *lmv = &obd->u.lmv;
1793         struct lu_fid            rid = op_data->op_fid1;
1794         struct lmv_tgt_desc     *tgt;
1795         struct mdt_body         *body;
1796         struct lmv_object       *obj;
1797         obd_valid                valid = op_data->op_valid;
1798         int                      rc;
1799         int                      loop = 0;
1800         int                      sidx;
1801         ENTRY;
1802
1803         rc = lmv_check_connect(obd);
1804         if (rc)
1805                 RETURN(rc);
1806
1807 repeat:
1808         ++loop;
1809         LASSERT(loop <= 2);
1810         obj = lmv_object_find(obd, &rid);
1811         if (obj) {
1812                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1813                                     op_data->op_name, op_data->op_namelen);
1814                 rid = obj->lo_stripes[sidx].ls_fid;
1815                 tgt = lmv_get_target(lmv, obj->lo_stripes[sidx].ls_mds);
1816                 op_data->op_mds = obj->lo_stripes[sidx].ls_mds;
1817                 valid &= ~OBD_MD_FLCKSPLIT;
1818                 lmv_object_put(obj);
1819         } else {
1820                 tgt = lmv_find_target(lmv, &rid);
1821                 valid |= OBD_MD_FLCKSPLIT;
1822                 op_data->op_mds = tgt->ltd_idx;
1823         }
1824         if (IS_ERR(tgt))
1825                 RETURN(PTR_ERR(tgt));
1826
1827         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" - "DFID" -> mds #%d\n",
1828                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1829                PFID(&rid), tgt->ltd_idx);
1830
1831         op_data->op_valid = valid;
1832         op_data->op_fid1 = rid;
1833         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1834         if (rc == 0) {
1835                 body = req_capsule_server_get(&(*request)->rq_pill,
1836                                               &RMF_MDT_BODY);
1837                 LASSERT(body != NULL);
1838
1839                 if (body->valid & OBD_MD_MDS) {
1840                         rid = body->fid1;
1841                         CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1842                                PFID(&rid));
1843
1844                         tgt = lmv_find_target(lmv, &rid);
1845                         if (IS_ERR(tgt)) {
1846                                 ptlrpc_req_finished(*request);
1847                                 RETURN(PTR_ERR(tgt));
1848                         }
1849
1850                         op_data->op_fid1 = rid;
1851                         op_data->op_valid |= OBD_MD_FLCROSSREF;
1852                         op_data->op_namelen = 0;
1853                         op_data->op_name = NULL;
1854                         rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1855                         ptlrpc_req_finished(*request);
1856                         *request = req;
1857                 }
1858         } else if (rc == -ERESTART) {
1859                 LASSERT(*request != NULL);
1860                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1861                           "Got -ERESTART during getattr!\n");
1862                 ptlrpc_req_finished(*request);
1863                 *request = NULL;
1864
1865                 /*
1866                  * Directory got split. Time to update local object and repeat
1867                  * the request with proper MDS.
1868                  */
1869                 rc = lmv_handle_split(exp, &rid);
1870                 if (rc == 0)
1871                         goto repeat;
1872         }
1873         RETURN(rc);
1874 }
1875
1876 #define md_op_data_fid(op_data, fl)                     \
1877         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1878          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1879          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1880          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1881          NULL)
1882
1883 static int lmv_early_cancel_slaves(struct obd_export *exp,
1884                                    struct md_op_data *op_data, int op_tgt,
1885                                    ldlm_mode_t mode, int bits, int flag)
1886 {
1887         struct obd_device      *obd = exp->exp_obd;
1888         struct lmv_obd         *lmv = &obd->u.lmv;
1889         ldlm_policy_data_t      policy = {{0}};
1890         struct lu_fid          *op_fid;
1891         struct lu_fid          *st_fid;
1892         struct lmv_tgt_desc    *tgt;
1893         struct lmv_object      *obj;
1894         int                     rc = 0;
1895         int                     i;
1896         ENTRY;
1897
1898         op_fid = md_op_data_fid(op_data, flag);
1899         if (!fid_is_sane(op_fid))
1900                 RETURN(0);
1901
1902         obj = lmv_object_find(obd, op_fid);
1903         if (obj == NULL)
1904                 RETURN(-EALREADY);
1905
1906         policy.l_inodebits.bits = bits;
1907         for (i = 0; i < obj->lo_objcount; i++) {
1908                 tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
1909                 st_fid = &obj->lo_stripes[i].ls_fid;
1910                 if (op_tgt != tgt->ltd_idx) {
1911                         CDEBUG(D_INODE, "EARLY_CANCEL slave "DFID" -> mds #%d\n",
1912                                PFID(st_fid), tgt->ltd_idx);
1913                         rc = md_cancel_unused(tgt->ltd_exp, st_fid, &policy,
1914                                               mode, LCF_ASYNC, NULL);
1915                         if (rc)
1916                                 GOTO(out_put_obj, rc);
1917                 } else {
1918                         CDEBUG(D_INODE,
1919                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1920                                op_tgt, PFID(st_fid));
1921                         /*
1922                          * Do not cancel locks for operation target, they will
1923                          * be handled later in underlaying layer when calling
1924                          * function we run on behalf of.
1925                          */
1926                         *op_fid = *st_fid;
1927                         op_data->op_flags |= flag;
1928                 }
1929         }
1930         EXIT;
1931 out_put_obj:
1932         lmv_object_put(obj);
1933         return rc;
1934 }
1935
1936 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1937                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1938 {
1939         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1940         struct obd_device      *obd = exp->exp_obd;
1941         struct lmv_obd         *lmv = &obd->u.lmv;
1942         struct lmv_tgt_desc    *tgt;
1943         ldlm_policy_data_t      policy = {{0}};
1944         struct lmv_object      *obj;
1945         int                     rc = 0;
1946         ENTRY;
1947
1948         if (!fid_is_sane(fid))
1949                 RETURN(0);
1950
1951         obj = lmv_object_find(obd, fid);
1952         if (obj) {
1953                 rc = lmv_early_cancel_slaves(exp, op_data, op_tgt, mode,
1954                                              bits, flag);
1955                 lmv_object_put(obj);
1956         } else {
1957                 tgt = lmv_find_target(lmv, fid);
1958                 if (IS_ERR(tgt))
1959                         RETURN(PTR_ERR(tgt));
1960
1961                 if (tgt->ltd_idx != op_tgt) {
1962                         CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1963                         policy.l_inodebits.bits = bits;
1964                         rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1965                                               mode, LCF_ASYNC, NULL);
1966                 } else {
1967                         CDEBUG(D_INODE,
1968                                "EARLY_CANCEL skip operation target %d on "DFID"\n",
1969                                op_tgt, PFID(fid));
1970                         op_data->op_flags |= flag;
1971                         rc = 0;
1972                 }
1973
1974         }
1975         RETURN(rc);
1976 }
1977
1978 /*
1979  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1980  * op_data->op_fid2
1981  */
1982 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1983                     struct ptlrpc_request **request)
1984 {
1985         struct obd_device       *obd = exp->exp_obd;
1986         struct lmv_obd          *lmv = &obd->u.lmv;
1987         struct lmv_tgt_desc     *tgt;
1988         struct lmv_object       *obj;
1989         int                      rc;
1990         int                      loop = 0;
1991         mdsno_t                  mds;
1992         int                      sidx;
1993         ENTRY;
1994
1995         rc = lmv_check_connect(obd);
1996         if (rc)
1997                 RETURN(rc);
1998
1999 repeat:
2000         ++loop;
2001         LASSERT(loop <= 2);
2002         LASSERT(op_data->op_namelen != 0);
2003
2004         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2005                PFID(&op_data->op_fid2), op_data->op_namelen,
2006                op_data->op_name, PFID(&op_data->op_fid1));
2007
2008         obj = lmv_object_find(obd, &op_data->op_fid2);
2009         if (obj) {
2010                 sidx = raw_name2idx(obj->lo_hashtype,
2011                                     obj->lo_objcount,
2012                                     op_data->op_name,
2013                                     op_data->op_namelen);
2014                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2015                 mds = obj->lo_stripes[sidx].ls_mds;
2016                 lmv_object_put(obj);
2017         } else {
2018                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
2019                 if (rc)
2020                         RETURN(rc);
2021         }
2022
2023         CDEBUG(D_INODE, "Forward to mds #%x ("DFID")\n",
2024                mds, PFID(&op_data->op_fid1));
2025
2026         op_data->op_fsuid = cfs_curproc_fsuid();
2027         op_data->op_fsgid = cfs_curproc_fsgid();
2028         op_data->op_cap = cfs_curproc_cap_pack();
2029         tgt = lmv_get_target(lmv, mds);
2030
2031         /*
2032          * Cancel UPDATE lock on child (fid1).
2033          */
2034         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2035         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2036                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2037         if (rc == 0)
2038                 rc = md_link(tgt->ltd_exp, op_data, request);
2039         if (rc == -ERESTART) {
2040                 LASSERT(*request != NULL);
2041                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2042                           "Got -ERESTART during link!\n");
2043                 ptlrpc_req_finished(*request);
2044                 *request = NULL;
2045
2046                 /*
2047                  * Directory got split. Time to update local object and repeat
2048                  * the request with proper MDS.
2049                  */
2050                 rc = lmv_handle_split(exp, &op_data->op_fid2);
2051                 if (rc == 0)
2052                         goto repeat;
2053         }
2054
2055         RETURN(rc);
2056 }
2057
2058 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2059                       const char *old, int oldlen, const char *new, int newlen,
2060                       struct ptlrpc_request **request)
2061 {
2062         struct obd_device       *obd = exp->exp_obd;
2063         struct lmv_obd          *lmv = &obd->u.lmv;
2064         struct lmv_tgt_desc     *src_tgt;
2065         struct lmv_tgt_desc     *tgt_tgt;
2066         int                      rc;
2067         int                      sidx;
2068         int                      loop = 0;
2069         struct lmv_object       *obj;
2070         mdsno_t                  mds1;
2071         mdsno_t                  mds2;
2072         ENTRY;
2073
2074         LASSERT(oldlen != 0);
2075
2076         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2077                oldlen, old, PFID(&op_data->op_fid1),
2078                newlen, new, PFID(&op_data->op_fid2));
2079
2080         rc = lmv_check_connect(obd);
2081         if (rc)
2082                 RETURN(rc);
2083
2084 repeat:
2085         ++loop;
2086         LASSERT(loop <= 2);
2087         obj = lmv_object_find(obd, &op_data->op_fid1);
2088         if (obj) {
2089                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2090                                     (char *)old, oldlen);
2091                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2092                 mds1 = obj->lo_stripes[sidx].ls_mds;
2093                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
2094                 lmv_object_put(obj);
2095         } else {
2096                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
2097                 if (rc)
2098                         RETURN(rc);
2099         }
2100
2101         obj = lmv_object_find(obd, &op_data->op_fid2);
2102         if (obj) {
2103                 /*
2104                  * Directory is already split, so we have to forward request to
2105                  * the right MDS.
2106                  */
2107                 sidx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
2108                                        (char *)new, newlen);
2109
2110                 mds2 = obj->lo_stripes[sidx].ls_mds;
2111                 op_data->op_fid2 = obj->lo_stripes[sidx].ls_fid;
2112                 CDEBUG(D_INODE, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
2113                 lmv_object_put(obj);
2114         } else {
2115                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
2116                 if (rc)
2117                         RETURN(rc);
2118         }
2119
2120         op_data->op_fsuid = cfs_curproc_fsuid();
2121         op_data->op_fsgid = cfs_curproc_fsgid();
2122         op_data->op_cap = cfs_curproc_cap_pack();
2123
2124         src_tgt = lmv_get_target(lmv, mds1);
2125         tgt_tgt = lmv_get_target(lmv, mds2);
2126
2127         /*
2128          * LOOKUP lock on src child (fid3) should also be cancelled for
2129          * src_tgt in mdc_rename.
2130          */
2131         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2132
2133         /*
2134          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2135          * own target.
2136          */
2137         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2138                               LCK_EX, MDS_INODELOCK_UPDATE,
2139                               MF_MDC_CANCEL_FID2);
2140
2141         /*
2142          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2143          */
2144         if (rc == 0) {
2145                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2146                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2147                                       MF_MDC_CANCEL_FID4);
2148         }
2149
2150         /*
2151          * Cancel all the locks on tgt child (fid4).
2152          */
2153         if (rc == 0)
2154                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2155                                       LCK_EX, MDS_INODELOCK_FULL,
2156                                       MF_MDC_CANCEL_FID4);
2157
2158         if (rc == 0)
2159                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2160                                new, newlen, request);
2161
2162         if (rc == -ERESTART) {
2163                 LASSERT(*request != NULL);
2164                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2165                           "Got -ERESTART during rename!\n");
2166                 ptlrpc_req_finished(*request);
2167                 *request = NULL;
2168
2169                 /*
2170                  * Directory got split. Time to update local object and repeat
2171                  * the request with proper MDS.
2172                  */
2173                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2174                 if (rc == 0)
2175                         goto repeat;
2176         }
2177         RETURN(rc);
2178 }
2179
2180 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2181                        void *ea, int ealen, void *ea2, int ea2len,
2182                        struct ptlrpc_request **request,
2183                        struct md_open_data **mod)
2184 {
2185         struct obd_device       *obd = exp->exp_obd;
2186         struct lmv_obd          *lmv = &obd->u.lmv;
2187         struct ptlrpc_request   *req;
2188         struct lmv_tgt_desc     *tgt;
2189         struct lmv_object       *obj;
2190         int                      rc = 0;
2191         int                      i;
2192         ENTRY;
2193
2194         rc = lmv_check_connect(obd);
2195         if (rc)
2196                 RETURN(rc);
2197
2198         obj = lmv_object_find(obd, &op_data->op_fid1);
2199
2200         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x%s\n",
2201                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2202                obj ? ", split" : "");
2203
2204         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2205         if (obj) {
2206                 for (i = 0; i < obj->lo_objcount; i++) {
2207                         op_data->op_fid1 = obj->lo_stripes[i].ls_fid;
2208
2209                         tgt = lmv_get_target(lmv, obj->lo_stripes[i].ls_mds);
2210                         if (IS_ERR(tgt)) {
2211                                 rc = PTR_ERR(tgt);
2212                                 break;
2213                         }
2214
2215                         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen,
2216                                         ea2, ea2len, &req, mod);
2217
2218                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_stripes[i].ls_fid)) {
2219                                 /*
2220                                  * This is master object and this request should
2221                                  * be returned back to llite.
2222                                  */
2223                                 *request = req;
2224                         } else {
2225                                 ptlrpc_req_finished(req);
2226                         }
2227
2228                         if (rc)
2229                                 break;
2230                 }
2231                 lmv_object_put(obj);
2232         } else {
2233                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2234                 if (IS_ERR(tgt))
2235                         RETURN(PTR_ERR(tgt));
2236
2237                 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2238                                 ea2len, request, mod);
2239         }
2240         RETURN(rc);
2241 }
2242
2243 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2244                     struct obd_capa *oc, struct ptlrpc_request **request)
2245 {
2246         struct obd_device         *obd = exp->exp_obd;
2247         struct lmv_obd            *lmv = &obd->u.lmv;
2248         struct lmv_tgt_desc       *tgt;
2249         int                        rc;
2250         ENTRY;
2251
2252         rc = lmv_check_connect(obd);
2253         if (rc)
2254                 RETURN(rc);
2255
2256         tgt = lmv_find_target(lmv, fid);
2257         if (IS_ERR(tgt))
2258                 RETURN(PTR_ERR(tgt));
2259
2260         rc = md_sync(tgt->ltd_exp, fid, oc, request);
2261         RETURN(rc);
2262 }
2263
2264 /**
2265  * Main purpose of LMV blocking ast is to remove split directory LMV
2266  * presentation object (struct lmv_object) attached to the lock being revoked.
2267  */
2268 int lmv_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2269                      void *data, int flag)
2270 {
2271         struct lustre_handle    lockh;
2272         struct lmv_object      *obj;
2273         int                     rc;
2274         ENTRY;
2275
2276         switch (flag) {
2277         case LDLM_CB_BLOCKING:
2278                 ldlm_lock2handle(lock, &lockh);
2279                 rc = ldlm_cli_cancel(&lockh);
2280                 if (rc < 0) {
2281                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2282                         RETURN(rc);
2283                 }
2284                 break;
2285         case LDLM_CB_CANCELING:
2286                 /*
2287                  * Time to drop cached attrs for split directory object
2288                  */
2289                 obj = lock->l_ast_data;
2290                 if (obj) {
2291                         CDEBUG(D_INODE, "Cancel %s on "LPU64"/"LPU64
2292                                ", master "DFID"\n",
2293                                lock->l_resource->lr_name.name[3] == 1 ?
2294                                "LOOKUP" : "UPDATE",
2295                                lock->l_resource->lr_name.name[0],
2296                                lock->l_resource->lr_name.name[1],
2297                                PFID(&obj->lo_fid));
2298                         lmv_object_put(obj);
2299                 }
2300                 break;
2301         default:
2302                 LBUG();
2303         }
2304         RETURN(0);
2305 }
2306
2307 static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj)
2308 {
2309         __u64         val;
2310
2311         val = le64_to_cpu(*hash);
2312         if (val < hash_adj)
2313                 val += MAX_HASH_SIZE;
2314         if (val != MDS_DIR_END_OFF)
2315                 *hash = cpu_to_le64(val - hash_adj);
2316 }
2317
2318 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2319 {
2320         __u64              id;
2321         struct obd_import *imp;
2322
2323         /*
2324          * XXX: to get nid we assume that underlying obd device is mdc.
2325          */
2326         imp  = class_exp2cliimp(exp);
2327         id   = imp->imp_connection->c_self + fid_flatten(fid);
2328
2329         CDEBUG(D_INODE, "Readpage node rank: "LPX64" "DFID" "LPX64" "LPX64"\n",
2330                imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2331
2332         return id ^ (id >> 32);
2333 }
2334
2335 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2336                         struct obd_capa *oc, __u64 offset64,
2337                         struct page **pages, unsigned npages,
2338                         struct ptlrpc_request **request)
2339 {
2340         struct obd_device       *obd = exp->exp_obd;
2341         struct lmv_obd          *lmv = &obd->u.lmv;
2342         struct lu_fid            rid = *fid;
2343         struct lmv_object       *obj;
2344         __u64                    offset;
2345         __u64                    hash_adj = 0;
2346         __u32                    rank = 0;
2347         __u64                    seg_size = 0;
2348         __u64                    tgt_tmp = 0;
2349         int                      tgt_idx = 0;
2350         int                      tgt0_idx = 0;
2351         int                      rc;
2352         int                      nr = 0;
2353         int                      i;
2354         /* number of pages read, in CFS_PAGE_SIZE */
2355         int                      nrdpgs;
2356         /* number of pages transferred in LU_PAGE_SIZE */
2357         int                      nlupgs;
2358         struct lmv_stripe       *los;
2359         struct lmv_tgt_desc     *tgt;
2360         struct lu_dirpage       *dp;
2361         struct lu_dirent        *ent;
2362         ENTRY;
2363
2364         offset = offset64;
2365
2366         rc = lmv_check_connect(obd);
2367         if (rc)
2368                 RETURN(rc);
2369
2370         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid));
2371
2372         /*
2373          * This case handle directory lookup in clustered metadata case (i.e.
2374          * split directory is located on multiple md servers.)
2375          * each server keeps directory entries for certain range of hashes.
2376          * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
2377          * first server will keep records with hashes [ 0 ... MAX_HASH / N  - 1],
2378          * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
2379          * so on....
2380          *      readdir can simply start reading entries from 0 - N server in
2381          * order but that will not scale well as all client will request dir in
2382          * to server in same order.
2383          * Following algorithm does optimization:
2384          * Instead of doing readdir in 1, 2, ...., N order, client with a
2385          * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
2386          * (every client has rank R)
2387          *      But ll_readdir() expect offset range [0 to MAX_HASH/N) but
2388          * since client ask dir from MDS{R} client has pages with offsets
2389          * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
2390          * on hash  values that we get.
2391          */
2392         obj = lmv_object_find_lock(obd, fid);
2393         if (obj) {
2394                 nr       = obj->lo_objcount;
2395                 LASSERT(nr > 0);
2396                 seg_size = MAX_HASH_SIZE;
2397                 do_div(seg_size, nr);
2398                 los      = obj->lo_stripes;
2399                 tgt      = lmv_get_target(lmv, los[0].ls_mds);
2400                 rank     = lmv_node_rank(tgt->ltd_exp, fid) % nr;
2401                 tgt_tmp  = offset;
2402                 do_div(tgt_tmp, seg_size);
2403                 tgt0_idx = do_div(tgt_tmp,  nr);
2404                 tgt_idx  = (tgt0_idx + rank) % nr;
2405
2406                 if (tgt_idx < tgt0_idx)
2407                         /*
2408                          * Wrap around.
2409                          *
2410                          * Last segment has unusual length due to division
2411                          * rounding.
2412                          */
2413                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
2414                 else
2415                         hash_adj = 0;
2416
2417                 hash_adj += rank * seg_size;
2418
2419                 CDEBUG(D_INODE, "Readpage hash adjustment: %x "LPX64" "
2420                        LPX64"/%x -> "LPX64"/%x\n", rank, hash_adj,
2421                        offset, tgt0_idx, offset + hash_adj, tgt_idx);
2422
2423                 offset = (offset + hash_adj) & MAX_HASH_SIZE;
2424                 rid = obj->lo_stripes[tgt_idx].ls_fid;
2425                 tgt = lmv_get_target(lmv, los[tgt_idx].ls_mds);
2426
2427                 CDEBUG(D_INODE, "Forward to "DFID" with offset %lu i %d\n",
2428                        PFID(&rid), (unsigned long)offset, tgt_idx);
2429         } else
2430                 tgt = lmv_find_target(lmv, &rid);
2431
2432         if (IS_ERR(tgt))
2433                 GOTO(cleanup, rc = PTR_ERR(tgt));
2434
2435         rc = md_readpage(tgt->ltd_exp, &rid, oc, offset, pages, npages,
2436                          request);
2437         if (rc)
2438                 GOTO(cleanup, rc);
2439
2440         nrdpgs = ((*request)->rq_bulk->bd_nob_transferred + CFS_PAGE_SIZE - 1)
2441                  >> CFS_PAGE_SHIFT;
2442         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2443         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2444         LASSERT(nrdpgs > 0 && nrdpgs <= npages);
2445
2446         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", nrdpgs, nlupgs, npages);
2447
2448         for (i = 0; i < nrdpgs; i++) {
2449 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2450                 struct lu_dirpage *first;
2451                 __u64 hash_end = 0;
2452                 __u32 flags = 0;
2453 #endif
2454                 struct lu_dirent *tmp = NULL;
2455
2456                 dp = cfs_kmap(pages[i]);
2457                 if (obj) {
2458                         lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2459                         lmv_hash_adjust(&dp->ldp_hash_end,   hash_adj);
2460                         LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64);
2461
2462                         if ((tgt0_idx != nr - 1) &&
2463                             (le64_to_cpu(dp->ldp_hash_end) == MDS_DIR_END_OFF))
2464                         {
2465                                 dp->ldp_hash_end = cpu_to_le32(seg_size *
2466                                                                (tgt0_idx + 1));
2467                                 CDEBUG(D_INODE,
2468                                        ""DFID" reset end "LPX64" tgt %d\n",
2469                                        PFID(&rid),
2470                                        (__u64)le64_to_cpu(dp->ldp_hash_end),
2471                                        tgt_idx);
2472                         }
2473                 }
2474
2475                 ent = lu_dirent_start(dp);
2476 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2477                 first = dp;
2478                 hash_end = dp->ldp_hash_end;
2479 repeat:
2480 #endif
2481                 nlupgs--;
2482                 for (tmp = ent; ent != NULL;
2483                      tmp = ent, ent = lu_dirent_next(ent)) {
2484                         if (obj)
2485                                 lmv_hash_adjust(&ent->lde_hash, hash_adj);
2486                 }
2487
2488 #if CFS_PAGE_SIZE > LU_PAGE_SIZE
2489                 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2490                 if (((unsigned long)dp & ~CFS_PAGE_MASK) && nlupgs > 0) {
2491                         ent = lu_dirent_start(dp);
2492
2493                         if (obj) {
2494                                 lmv_hash_adjust(&dp->ldp_hash_end, hash_adj);
2495                                 if ((tgt0_idx != nr - 1) &&
2496                                     (le64_to_cpu(dp->ldp_hash_end) ==
2497                                      MDS_DIR_END_OFF)) {
2498                                         hash_end = cpu_to_le32(seg_size *
2499                                                                (tgt0_idx + 1));
2500                                         CDEBUG(D_INODE,
2501                                             ""DFID" reset end "LPX64" tgt %d\n",
2502                                             PFID(&rid),
2503                                             (__u64)le64_to_cpu(hash_end),
2504                                             tgt_idx);
2505                                 }
2506                         }
2507                         hash_end = dp->ldp_hash_end;
2508                         flags = dp->ldp_flags;
2509
2510                         if (tmp) {
2511                                 /* enlarge the end entry lde_reclen from 0 to
2512                                  * first entry of next lu_dirpage, in this way
2513                                  * several lu_dirpages can be stored into one
2514                                  * client page on client. */
2515                                 tmp = ((void *)tmp) +
2516                                       le16_to_cpu(tmp->lde_reclen);
2517                                 tmp->lde_reclen =
2518                                         cpu_to_le16((char *)(dp->ldp_entries) -
2519                                                     (char *)tmp);
2520                                 goto repeat;
2521                         }
2522                 }
2523                 first->ldp_hash_end = hash_end;
2524                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2525                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2526 #endif
2527                 cfs_kunmap(pages[i]);
2528         }
2529         EXIT;
2530 cleanup:
2531         if (obj)
2532                 lmv_object_put_unlock(obj);
2533         return rc;
2534 }
2535
2536 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2537                       struct ptlrpc_request **request)
2538 {
2539         struct obd_device       *obd = exp->exp_obd;
2540         struct lmv_obd          *lmv = &obd->u.lmv;
2541         struct lmv_tgt_desc     *tgt = NULL;
2542         struct lmv_object       *obj;
2543         int                      rc;
2544         int                      sidx;
2545         int                      loop = 0;
2546         ENTRY;
2547
2548         rc = lmv_check_connect(obd);
2549         if (rc)
2550                 RETURN(rc);
2551
2552 repeat:
2553         ++loop;
2554         LASSERT(loop <= 2);
2555         LASSERT(op_data->op_namelen != 0);
2556
2557         obj = lmv_object_find(obd, &op_data->op_fid1);
2558         if (obj) {
2559                 sidx = raw_name2idx(obj->lo_hashtype,
2560                                     obj->lo_objcount,
2561                                     op_data->op_name,
2562                                     op_data->op_namelen);
2563                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
2564                 op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
2565                 tgt = lmv_get_target(lmv,
2566                                      obj->lo_stripes[sidx].ls_mds);
2567                 lmv_object_put(obj);
2568                 CDEBUG(D_INODE, "UNLINK '%*s' in "DFID" -> %u\n",
2569                        op_data->op_namelen, op_data->op_name,
2570                        PFID(&op_data->op_fid1), sidx);
2571         }
2572
2573         if (tgt == NULL) {
2574                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2575                 if (IS_ERR(tgt))
2576                         RETURN(PTR_ERR(tgt));
2577                 op_data->op_bias |= MDS_CHECK_SPLIT;
2578         }
2579
2580         op_data->op_fsuid = cfs_curproc_fsuid();
2581         op_data->op_fsgid = cfs_curproc_fsgid();
2582         op_data->op_cap = cfs_curproc_cap_pack();
2583
2584         /*
2585          * If child's fid is given, cancel unused locks for it if it is from
2586          * another export than parent.
2587          *
2588          * LOOKUP lock for child (fid3) should also be cancelled on parent
2589          * tgt_tgt in mdc_unlink().
2590          */
2591         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2592
2593         /*
2594          * Cancel FULL locks on child (fid3).
2595          */
2596         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2597                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2598
2599         if (rc == 0)
2600                 rc = md_unlink(tgt->ltd_exp, op_data, request);
2601
2602         if (rc == -ERESTART) {
2603                 LASSERT(*request != NULL);
2604                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2605                           "Got -ERESTART during unlink!\n");
2606                 ptlrpc_req_finished(*request);
2607                 *request = NULL;
2608
2609                 /*
2610                  * Directory got split. Time to update local object and repeat
2611                  * the request with proper MDS.
2612                  */
2613                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2614                 if (rc == 0)
2615                         goto repeat;
2616         }
2617         RETURN(rc);
2618 }
2619
2620 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2621 {
2622         struct lmv_obd *lmv = &obd->u.lmv;
2623         int rc = 0;
2624
2625         switch (stage) {
2626         case OBD_CLEANUP_EARLY:
2627                 /* XXX: here should be calling obd_precleanup() down to
2628                  * stack. */
2629                 break;
2630         case OBD_CLEANUP_EXPORTS:
2631                 fld_client_proc_fini(&lmv->lmv_fld);
2632                 lprocfs_obd_cleanup(obd);
2633                 rc = obd_llog_finish(obd, 0);
2634                 if (rc != 0)
2635                         CERROR("failed to cleanup llogging subsystems\n");
2636                 break;
2637         default:
2638                 break;
2639         }
2640         RETURN(rc);
2641 }
2642
2643 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2644                         void *key, __u32 *vallen, void *val,
2645                         struct lov_stripe_md *lsm)
2646 {
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
2659         lmv = &obd->u.lmv;
2660         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2661                 struct lmv_tgt_desc *tgts;
2662                 int i;
2663
2664                 rc = lmv_check_connect(obd);
2665                 if (rc)
2666                         RETURN(rc);
2667
2668                 LASSERT(*vallen == sizeof(__u32));
2669                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2670                      i++, tgts++) {
2671
2672                         /*
2673                          * All tgts should be connected when this gets called.
2674                          */
2675                         if (!tgts || !tgts->ltd_exp) {
2676                                 CERROR("target not setup?\n");
2677                                 continue;
2678                         }
2679
2680                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2681                                           vallen, val, NULL))
2682                                 RETURN(0);
2683                 }
2684                 RETURN(-EINVAL);
2685         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2686                 rc = lmv_check_connect(obd);
2687                 if (rc)
2688                         RETURN(rc);
2689
2690                 /*
2691                  * Forwarding this request to first MDS, it should know LOV
2692                  * desc.
2693                  */
2694                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2695                                   vallen, val, NULL);
2696                 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2697                         exp->exp_connect_flags =
2698                         ((struct obd_connect_data *)val)->ocd_connect_flags;
2699                 }
2700                 RETURN(rc);
2701         } else if (KEY_IS(KEY_TGT_COUNT)) {
2702                 *((int *)val) = lmv->desc.ld_tgt_count;
2703                 RETURN(0);
2704         }
2705
2706         CDEBUG(D_IOCTL, "Invalid key\n");
2707         RETURN(-EINVAL);
2708 }
2709
2710 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2711                        void *key, obd_count vallen, void *val,
2712                        struct ptlrpc_request_set *set)
2713 {
2714         struct lmv_tgt_desc    *tgt;
2715         struct obd_device      *obd;
2716         struct lmv_obd         *lmv;
2717         int rc = 0;
2718         ENTRY;
2719
2720         obd = class_exp2obd(exp);
2721         if (obd == NULL) {
2722                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2723                        exp->exp_handle.h_cookie);
2724                 RETURN(-EINVAL);
2725         }
2726         lmv = &obd->u.lmv;
2727
2728         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2729                 int i, err = 0;
2730
2731                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2732                         tgt = &lmv->tgts[i];
2733
2734                         if (!tgt->ltd_exp)
2735                                 continue;
2736
2737                         err = obd_set_info_async(tgt->ltd_exp,
2738                                                  keylen, key, vallen, val, set);
2739                         if (err && rc == 0)
2740                                 rc = err;
2741                 }
2742
2743                 RETURN(rc);
2744         }
2745
2746         RETURN(-EINVAL);
2747 }
2748
2749 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2750                struct lov_stripe_md *lsm)
2751 {
2752         struct obd_device         *obd = class_exp2obd(exp);
2753         struct lmv_obd            *lmv = &obd->u.lmv;
2754         struct lmv_stripe_md      *meap;
2755         struct lmv_stripe_md      *lsmp;
2756         int                        mea_size;
2757         int                        i;
2758         ENTRY;
2759
2760         mea_size = lmv_get_easize(lmv);
2761         if (!lmmp)
2762                 RETURN(mea_size);
2763
2764         if (*lmmp && !lsm) {
2765                 OBD_FREE_LARGE(*lmmp, mea_size);
2766                 *lmmp = NULL;
2767                 RETURN(0);
2768         }
2769
2770         if (*lmmp == NULL) {
2771                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2772                 if (*lmmp == NULL)
2773                         RETURN(-ENOMEM);
2774         }
2775
2776         if (!lsm)
2777                 RETURN(mea_size);
2778
2779         lsmp = (struct lmv_stripe_md *)lsm;
2780         meap = (struct lmv_stripe_md *)*lmmp;
2781
2782         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2783             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2784                 RETURN(-EINVAL);
2785
2786         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2787         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2788         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2789
2790         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2791                 meap->mea_ids[i] = meap->mea_ids[i];
2792                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2793         }
2794
2795         RETURN(mea_size);
2796 }
2797
2798 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2799                  struct lov_mds_md *lmm, int lmm_size)
2800 {
2801         struct obd_device          *obd = class_exp2obd(exp);
2802         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2803         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2804         struct lmv_obd             *lmv = &obd->u.lmv;
2805         int                         mea_size;
2806         int                         i;
2807         __u32                       magic;
2808         ENTRY;
2809
2810         mea_size = lmv_get_easize(lmv);
2811         if (lsmp == NULL)
2812                 return mea_size;
2813
2814         if (*lsmp != NULL && lmm == NULL) {
2815                 OBD_FREE_LARGE(*tmea, mea_size);
2816                 *lsmp = NULL;
2817                 RETURN(0);
2818         }
2819
2820         LASSERT(mea_size == lmm_size);
2821
2822         OBD_ALLOC_LARGE(*tmea, mea_size);
2823         if (*tmea == NULL)
2824                 RETURN(-ENOMEM);
2825
2826         if (!lmm)
2827                 RETURN(mea_size);
2828
2829         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2830             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2831             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2832         {
2833                 magic = le32_to_cpu(mea->mea_magic);
2834         } else {
2835                 /*
2836                  * Old mea is not handled here.
2837                  */
2838                 CERROR("Old not supportable EA is found\n");
2839                 LBUG();
2840         }
2841
2842         (*tmea)->mea_magic = magic;
2843         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2844         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2845
2846         for (i = 0; i < (*tmea)->mea_count; i++) {
2847                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2848                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2849         }
2850         RETURN(mea_size);
2851 }
2852
2853 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2854                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2855                              ldlm_cancel_flags_t flags, void *opaque)
2856 {
2857         struct obd_device       *obd = exp->exp_obd;
2858         struct lmv_obd          *lmv = &obd->u.lmv;
2859         int                      rc = 0;
2860         int                      err;
2861         int                      i;
2862         ENTRY;
2863
2864         LASSERT(fid != NULL);
2865
2866         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2867                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2868                         continue;
2869
2870                 err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
2871                                        policy, mode, flags, opaque);
2872                 if (!rc)
2873                         rc = err;
2874         }
2875         RETURN(rc);
2876 }
2877
2878 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2879                       __u32 *bits)
2880 {
2881         struct obd_device       *obd = exp->exp_obd;
2882         struct lmv_obd          *lmv = &obd->u.lmv;
2883         int                      rc;
2884         ENTRY;
2885
2886         rc =  md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data, bits);
2887         RETURN(rc);
2888 }
2889
2890 ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
2891                            const struct lu_fid *fid, ldlm_type_t type,
2892                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2893                            struct lustre_handle *lockh)
2894 {
2895         struct obd_device       *obd = exp->exp_obd;
2896         struct lmv_obd          *lmv = &obd->u.lmv;
2897         ldlm_mode_t              rc;
2898         int                      i;
2899         ENTRY;
2900
2901         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2902
2903         /*
2904          * With CMD every object can have two locks in different namespaces:
2905          * lookup lock in space of mds storing direntry and update/open lock in
2906          * space of mds storing inode. Thus we check all targets, not only that
2907          * one fid was created in.
2908          */
2909         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2910                 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2911                                    type, policy, mode, lockh);
2912                 if (rc)
2913                         RETURN(rc);
2914         }
2915
2916         RETURN(0);
2917 }
2918
2919 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2920                       struct obd_export *dt_exp, struct obd_export *md_exp,
2921                       struct lustre_md *md)
2922 {
2923         struct obd_device       *obd = exp->exp_obd;
2924         struct lmv_obd          *lmv = &obd->u.lmv;
2925         int                      rc;
2926         ENTRY;
2927         rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
2928         RETURN(rc);
2929 }
2930
2931 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2932 {
2933         struct obd_device       *obd = exp->exp_obd;
2934         struct lmv_obd          *lmv = &obd->u.lmv;
2935         ENTRY;
2936
2937         if (md->mea)
2938                 obd_free_memmd(exp, (void *)&md->mea);
2939         RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2940 }
2941
2942 int lmv_set_open_replay_data(struct obd_export *exp,
2943                              struct obd_client_handle *och,
2944                              struct ptlrpc_request *open_req)
2945 {
2946         struct obd_device       *obd = exp->exp_obd;
2947         struct lmv_obd          *lmv = &obd->u.lmv;
2948         struct lmv_tgt_desc     *tgt;
2949         ENTRY;
2950
2951         tgt = lmv_find_target(lmv, &och->och_fid);
2952         if (IS_ERR(tgt))
2953                 RETURN(PTR_ERR(tgt));
2954
2955         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2956 }
2957
2958 int lmv_clear_open_replay_data(struct obd_export *exp,
2959                                struct obd_client_handle *och)
2960 {
2961         struct obd_device       *obd = exp->exp_obd;
2962         struct lmv_obd          *lmv = &obd->u.lmv;
2963         struct lmv_tgt_desc     *tgt;
2964         ENTRY;
2965
2966         tgt = lmv_find_target(lmv, &och->och_fid);
2967         if (IS_ERR(tgt))
2968                 RETURN(PTR_ERR(tgt));
2969
2970         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2971 }
2972
2973 static int lmv_get_remote_perm(struct obd_export *exp,
2974                                const struct lu_fid *fid,
2975                                struct obd_capa *oc, __u32 suppgid,
2976                                struct ptlrpc_request **request)
2977 {
2978         struct obd_device       *obd = exp->exp_obd;
2979         struct lmv_obd          *lmv = &obd->u.lmv;
2980         struct lmv_tgt_desc     *tgt;
2981         int                      rc;
2982         ENTRY;
2983
2984         rc = lmv_check_connect(obd);
2985         if (rc)
2986                 RETURN(rc);
2987
2988         tgt = lmv_find_target(lmv, fid);
2989         if (IS_ERR(tgt))
2990                 RETURN(PTR_ERR(tgt));
2991
2992         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2993         RETURN(rc);
2994 }
2995
2996 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2997                           renew_capa_cb_t cb)
2998 {
2999         struct obd_device       *obd = exp->exp_obd;
3000         struct lmv_obd          *lmv = &obd->u.lmv;
3001         struct lmv_tgt_desc     *tgt;
3002         int                      rc;
3003         ENTRY;
3004
3005         rc = lmv_check_connect(obd);
3006         if (rc)
3007                 RETURN(rc);
3008
3009         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3010         if (IS_ERR(tgt))
3011                 RETURN(PTR_ERR(tgt));
3012
3013         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3014         RETURN(rc);
3015 }
3016
3017 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3018                     const struct req_msg_field *field, struct obd_capa **oc)
3019 {
3020         struct obd_device *obd = exp->exp_obd;
3021         struct lmv_obd *lmv = &obd->u.lmv;
3022         int rc;
3023
3024         ENTRY;
3025         rc = md_unpack_capa(lmv->tgts[0].ltd_exp, req, field, oc);
3026         RETURN(rc);
3027 }
3028
3029 int lmv_intent_getattr_async(struct obd_export *exp,
3030                              struct md_enqueue_info *minfo,
3031                              struct ldlm_enqueue_info *einfo)
3032 {
3033         struct md_op_data       *op_data = &minfo->mi_data;
3034         struct obd_device       *obd = exp->exp_obd;
3035         struct lmv_obd          *lmv = &obd->u.lmv;
3036         struct lmv_object       *obj;
3037         struct lmv_tgt_desc     *tgt;
3038         int                      rc;
3039         int                      sidx;
3040         ENTRY;
3041
3042         rc = lmv_check_connect(obd);
3043         if (rc)
3044                 RETURN(rc);
3045
3046         if (!fid_is_sane(&op_data->op_fid2)) {
3047                 obj = lmv_object_find(obd, &op_data->op_fid1);
3048                 if (obj && op_data->op_namelen) {
3049                         sidx = raw_name2idx(obj->lo_hashtype,
3050                                             obj->lo_objcount,
3051                                             (char *)op_data->op_name,
3052                                             op_data->op_namelen);
3053                         op_data->op_fid1 = obj->lo_stripes[sidx].ls_fid;
3054                         tgt = lmv_get_target(lmv,
3055                                              obj->lo_stripes[sidx].ls_mds);
3056                         CDEBUG(D_INODE,
3057                                "Choose slave dir ("DFID") -> mds #%d\n",
3058                                PFID(&op_data->op_fid1), tgt->ltd_idx);
3059                 } else {
3060                         tgt = lmv_find_target(lmv, &op_data->op_fid1);
3061                 }
3062                 if (obj)
3063                         lmv_object_put(obj);
3064         } else {
3065                 op_data->op_fid1 = op_data->op_fid2;
3066                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
3067                 op_data->op_bias = MDS_CROSS_REF;
3068                 /*
3069                  * Unfortunately, we have to lie to MDC/MDS to retrieve
3070                  * attributes llite needs.
3071                 */
3072                 if (minfo->mi_it.it_op & IT_LOOKUP)
3073                         minfo->mi_it.it_op = IT_GETATTR;
3074         }
3075
3076         if (IS_ERR(tgt))
3077                 RETURN(PTR_ERR(tgt));
3078
3079         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3080         RETURN(rc);
3081 }
3082
3083 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3084                         struct lu_fid *fid)
3085 {
3086         struct obd_device       *obd = exp->exp_obd;
3087         struct lmv_obd          *lmv = &obd->u.lmv;
3088         struct lmv_tgt_desc     *tgt;
3089         int                      rc;
3090         ENTRY;
3091
3092         rc = lmv_check_connect(obd);
3093         if (rc)
3094                 RETURN(rc);
3095
3096         tgt = lmv_find_target(lmv, fid);
3097         if (IS_ERR(tgt))
3098                 RETURN(PTR_ERR(tgt));
3099
3100         rc = md_revalidate_lock(tgt->ltd_exp, it, fid);
3101         RETURN(rc);
3102 }
3103
3104
3105 struct obd_ops lmv_obd_ops = {
3106         .o_owner                = THIS_MODULE,
3107         .o_setup                = lmv_setup,
3108         .o_cleanup              = lmv_cleanup,
3109         .o_precleanup           = lmv_precleanup,
3110         .o_process_config       = lmv_process_config,
3111         .o_connect              = lmv_connect,
3112         .o_disconnect           = lmv_disconnect,
3113         .o_statfs               = lmv_statfs,
3114         .o_get_info             = lmv_get_info,
3115         .o_set_info_async       = lmv_set_info_async,
3116         .o_packmd               = lmv_packmd,
3117         .o_unpackmd             = lmv_unpackmd,
3118         .o_notify               = lmv_notify,
3119         .o_get_uuid             = lmv_get_uuid,
3120         .o_iocontrol            = lmv_iocontrol,
3121 };
3122
3123 struct md_ops lmv_md_ops = {
3124         .m_getstatus            = lmv_getstatus,
3125         .m_change_cbdata        = lmv_change_cbdata,
3126         .m_find_cbdata          = lmv_find_cbdata,
3127         .m_close                = lmv_close,
3128         .m_create               = lmv_create,
3129         .m_done_writing         = lmv_done_writing,
3130         .m_enqueue              = lmv_enqueue,
3131         .m_getattr              = lmv_getattr,
3132         .m_getxattr             = lmv_getxattr,
3133         .m_getattr_name         = lmv_getattr_name,
3134         .m_intent_lock          = lmv_intent_lock,
3135         .m_link                 = lmv_link,
3136         .m_rename               = lmv_rename,
3137         .m_setattr              = lmv_setattr,
3138         .m_setxattr             = lmv_setxattr,
3139         .m_sync                 = lmv_sync,
3140         .m_readpage             = lmv_readpage,
3141         .m_unlink               = lmv_unlink,
3142         .m_init_ea_size         = lmv_init_ea_size,
3143         .m_cancel_unused        = lmv_cancel_unused,
3144         .m_set_lock_data        = lmv_set_lock_data,
3145         .m_lock_match           = lmv_lock_match,
3146         .m_get_lustre_md        = lmv_get_lustre_md,
3147         .m_free_lustre_md       = lmv_free_lustre_md,
3148         .m_set_open_replay_data = lmv_set_open_replay_data,
3149         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3150         .m_renew_capa           = lmv_renew_capa,
3151         .m_unpack_capa          = lmv_unpack_capa,
3152         .m_get_remote_perm      = lmv_get_remote_perm,
3153         .m_intent_getattr_async = lmv_intent_getattr_async,
3154         .m_revalidate_lock      = lmv_revalidate_lock
3155 };
3156
3157 static quota_interface_t *quota_interface;
3158 extern quota_interface_t lmv_quota_interface;
3159
3160 int __init lmv_init(void)
3161 {
3162         struct lprocfs_static_vars lvars;
3163         int                        rc;
3164
3165         lmv_object_cache = cfs_mem_cache_create("lmv_objects",
3166                                                 sizeof(struct lmv_object),
3167                                                 0, 0);
3168         if (!lmv_object_cache) {
3169                 CERROR("Error allocating lmv objects cache\n");
3170                 return -ENOMEM;
3171         }
3172
3173         lprocfs_lmv_init_vars(&lvars);
3174
3175         cfs_request_module("lquota");
3176         quota_interface = PORTAL_SYMBOL_GET(lmv_quota_interface);
3177         init_obd_quota_ops(quota_interface, &lmv_obd_ops);
3178
3179         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
3180                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
3181         if (rc) {
3182                 if (quota_interface)
3183                         PORTAL_SYMBOL_PUT(lmv_quota_interface);
3184                 cfs_mem_cache_destroy(lmv_object_cache);
3185         }
3186
3187         return rc;
3188 }
3189
3190 #ifdef __KERNEL__
3191 static void lmv_exit(void)
3192 {
3193         if (quota_interface)
3194                 PORTAL_SYMBOL_PUT(lmv_quota_interface);
3195
3196         class_unregister_type(LUSTRE_LMV_NAME);
3197
3198         LASSERTF(cfs_atomic_read(&lmv_object_count) == 0,
3199                  "Can't free lmv objects cache, %d object(s) busy\n",
3200                  cfs_atomic_read(&lmv_object_count));
3201         cfs_mem_cache_destroy(lmv_object_cache);
3202 }
3203
3204 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3205 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3206 MODULE_LICENSE("GPL");
3207
3208 module_init(lmv_init);
3209 module_exit(lmv_exit);
3210 #endif