lustre/lmv/lmv_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003, 2004, 2005, 2006 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/mm.h>
  33 #include <asm/div64.h>
  34 #include <linux/seq_file.h>
  35 #include <linux/namei.h>
  36 #else
  37 #include <liblustre.h>
  38 #endif
  39
  40 #include <linux/ext2_fs.h>
  41
  42 #include <lustre/lustre_idl.h>
  43 #include <lustre_log.h>
  44 #include <obd_support.h>
  45 #include <lustre_lib.h>
  46 #include <lustre_net.h>
  47 #include <obd_class.h>
  48 #include <lprocfs_status.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_fid.h>
  51 #include "lmv_internal.h"
  52
  53 /* not defined for liblustre building */
  54 #if !defined(ATOMIC_INIT)
  55 #define ATOMIC_INIT(val) { (val) }
  56 #endif
  57
  58 /* object cache. */
  59 cfs_mem_cache_t *obj_cache;
  60 atomic_t obj_cache_count = ATOMIC_INIT(0);
  61
  62 static void lmv_activate_target(struct lmv_obd *lmv,
  63                                 struct lmv_tgt_desc *tgt,
  64                                 int activate)
  65 {
  66         if (tgt->ltd_active == activate)
  67                 return;
  68
  69         tgt->ltd_active = activate;
  70         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  71 }
  72
  73 /* Error codes:
  74  *
  75  *  -EINVAL  : UUID can't be found in the LMV's target list
  76  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  77  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  78  */
  79 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  80                               int activate)
  81 {
  82         struct lmv_tgt_desc *tgt;
  83         struct obd_device *obd;
  84         int i, rc = 0;
  85         ENTRY;
  86
  87         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  88                lmv, uuid->uuid, activate);
  89
  90         spin_lock(&lmv->lmv_lock);
  91         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  92                 if (tgt->ltd_exp == NULL)
  93                         continue;
  94
  95                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  96                        i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  97
  98                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
  99                         break;
 100         }
 101
 102         if (i == lmv->desc.ld_tgt_count)
 103                 GOTO(out_lmv_lock, rc = -EINVAL);
 104
 105         obd = class_exp2obd(tgt->ltd_exp);
 106         if (obd == NULL)
 107                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 108
 109         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 110                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 111                obd->obd_type->typ_name, i);
 112         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 113
 114         if (tgt->ltd_active == activate) {
 115                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 116                        activate ? "" : "in");
 117                 GOTO(out_lmv_lock, rc);
 118         }
 119
 120         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 121                obd, activate ? "" : "in");
 122
 123         lmv_activate_target(lmv, tgt, activate);
 124
 125         EXIT;
 126
 127  out_lmv_lock:
 128         spin_unlock(&lmv->lmv_lock);
 129         return rc;
 130 }
 131
 132 static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
 133                             struct obd_connect_data *data)
 134 {
 135         struct lmv_tgt_desc *tgt;
 136         int i;
 137         ENTRY;
 138
 139         LASSERT(data != NULL);
 140
 141         spin_lock(&lmv->lmv_lock);
 142         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 143                 if (tgt->ltd_exp == NULL)
 144                         continue;
 145
 146                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
 147                         lmv->datas[tgt->ltd_idx] = *data;
 148                         break;
 149                 }
 150         }
 151         spin_unlock(&lmv->lmv_lock);
 152         RETURN(0);
 153 }
 154
 155 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 156                       enum obd_notify_event ev, void *data)
 157 {
 158         struct lmv_obd *lmv = &obd->u.lmv;
 159         struct obd_uuid *uuid;
 160         int rc = 0;
 161         ENTRY;
 162
 163         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 164                 CERROR("unexpected notification of %s %s!\n",
 165                        watched->obd_type->typ_name,
 166                        watched->obd_name);
 167                 RETURN(-EINVAL);
 168         }
 169
 170         uuid = &watched->u.cli.cl_target_uuid;
 171         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 172                 /*
 173                  * Set MDC as active before notifying the observer, so the
 174                  * observer can use the MDC normally.
 175                  */
 176                 rc = lmv_set_mdc_active(lmv, uuid,
 177                                         ev == OBD_NOTIFY_ACTIVE);
 178                 if (rc) {
 179                         CERROR("%sactivation of %s failed: %d\n",
 180                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 181                                uuid->uuid, rc);
 182                         RETURN(rc);
 183                 }
 184         } else if (ev == OBD_NOTIFY_OCD) {
 185                 struct obd_connect_data *conn_data =
 186                         &watched->u.cli.cl_import->imp_connect_data;
 187
 188                 /* Set connect data to desired target, update
 189                  * exp_connect_flags. */
 190                 rc = lmv_set_mdc_data(lmv, uuid, conn_data);
 191                 if (rc) {
 192                         CERROR("can't set connect data to target %s, rc %d\n",
 193                                uuid->uuid, rc);
 194                         RETURN(rc);
 195                 }
 196
 197                 /*
 198                  * XXX: Make sure that ocd_connect_flags from all targets are
 199                  * the same. Otherwise one of MDTs runs wrong version or
 200                  * something like this.  --umka
 201                  */
 202                 obd->obd_self_export->exp_connect_flags =
 203                         conn_data->ocd_connect_flags;
 204         }
 205 #if 0
 206         else if (ev == OBD_NOTIFY_DISCON) {
 207                 /* For disconnect event, flush fld cache for failout MDS case. */
 208                 fld_client_flush(&lmv->lmv_fld);
 209         }
 210 #endif
 211         /* Pass the notification up the chain. */
 212         if (obd->obd_observer)
 213                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 214
 215         RETURN(rc);
 216 }
 217
 218 /* this is fake connect function. Its purpose is to initialize lmv and say
 219  * caller that everything is okay. Real connection will be performed later. */
 220 static int lmv_connect(const struct lu_env *env,
 221                        struct lustre_handle *conn, struct obd_device *obd,
 222                        struct obd_uuid *cluuid, struct obd_connect_data *data)
 223 {
 224 #ifdef __KERNEL__
 225         struct proc_dir_entry *lmv_proc_dir;
 226 #endif
 227         struct lmv_obd *lmv = &obd->u.lmv;
 228         struct obd_export *exp;
 229         int rc = 0;
 230         ENTRY;
 231
 232         rc = class_connect(conn, obd, cluuid);
 233         if (rc) {
 234                 CERROR("class_connection() returned %d\n", rc);
 235                 RETURN(rc);
 236         }
 237
 238         exp = class_conn2export(conn);
 239
 240         /* we don't want to actually do the underlying connections more than
 241          * once, so keep track. */
 242         lmv->refcount++;
 243         if (lmv->refcount > 1) {
 244                 class_export_put(exp);
 245                 RETURN(0);
 246         }
 247
 248         lmv->exp = exp;
 249         lmv->connected = 0;
 250         lmv->cluuid = *cluuid;
 251
 252         if (data)
 253                 lmv->conn_data = *data;
 254
 255 #ifdef __KERNEL__
 256         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 257                                         NULL, NULL);
 258         if (IS_ERR(lmv_proc_dir)) {
 259                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 260                        obd->obd_type->typ_name, obd->obd_name);
 261                 lmv_proc_dir = NULL;
 262         }
 263 #endif
 264
 265         /* all real clients should perform actual connection right away, because
 266          * it is possible, that LMV will not have opportunity to connect targets
 267          * and MDC stuff will be called directly, for instance while reading
 268          * ../mdc/../kbytesfree procfs file, etc. */
 269         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 270                 rc = lmv_check_connect(obd);
 271
 272 #ifdef __KERNEL__
 273         if (rc) {
 274                 if (lmv_proc_dir)
 275                         lprocfs_remove(&lmv_proc_dir);
 276         }
 277 #endif
 278
 279         RETURN(rc);
 280 }
 281
 282 static void lmv_set_timeouts(struct obd_device *obd)
 283 {
 284         struct lmv_tgt_desc *tgts;
 285         struct lmv_obd *lmv;
 286         int i;
 287
 288         lmv = &obd->u.lmv;
 289         if (lmv->server_timeout == 0)
 290                 return;
 291
 292         if (lmv->connected == 0)
 293                 return;
 294
 295         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 296                 if (tgts->ltd_exp == NULL)
 297                         continue;
 298
 299                 obd_set_info_async(tgts->ltd_exp, strlen("inter_mds"),
 300                                    "inter_mds", 0, NULL, NULL);
 301         }
 302 }
 303
 304 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 305                             int def_easize, int cookiesize)
 306 {
 307         struct obd_device *obd = exp->exp_obd;
 308         struct lmv_obd *lmv = &obd->u.lmv;
 309         int i, rc = 0, change = 0;
 310         ENTRY;
 311
 312         if (lmv->max_easize < easize) {
 313                 lmv->max_easize = easize;
 314                 change = 1;
 315         }
 316         if (lmv->max_def_easize < def_easize) {
 317                 lmv->max_def_easize = def_easize;
 318                 change = 1;
 319         }
 320         if (lmv->max_cookiesize < cookiesize) {
 321                 lmv->max_cookiesize = cookiesize;
 322                 change = 1;
 323         }
 324         if (change == 0)
 325                 RETURN(0);
 326
 327         if (lmv->connected == 0)
 328                 RETURN(0);
 329
 330         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 331                 if (lmv->tgts[i].ltd_exp == NULL) {
 332                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 333                         continue;
 334                 }
 335
 336                 rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
 337                                      cookiesize);
 338                 if (rc) {
 339                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 340                                "error %d.\n", i, rc);
 341                         break;
 342                 }
 343         }
 344         RETURN(rc);
 345 }
 346
 347 #define MAX_STRING_SIZE 128
 348
 349 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 350 {
 351         struct lmv_obd *lmv = &obd->u.lmv;
 352         struct obd_uuid *cluuid = &lmv->cluuid;
 353         struct obd_connect_data *mdc_data = NULL;
 354         struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 355         struct lustre_handle conn = {0, };
 356         struct obd_device *mdc_obd;
 357         struct obd_export *mdc_exp;
 358         struct lu_fld_target target;
 359         int rc;
 360 #ifdef __KERNEL__
 361         struct proc_dir_entry *lmv_proc_dir;
 362 #endif
 363         ENTRY;
 364
 365         /* for MDS: don't connect to yourself */
 366         if (obd_uuid_equals(&tgt->ltd_uuid, cluuid)) {
 367                 CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
 368                 /* XXX - the old code didn't increment active tgt count.
 369                  *       should we ? */
 370                 RETURN(0);
 371         }
 372
 373         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 374                                         &obd->obd_uuid);
 375         if (!mdc_obd) {
 376                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 377                 RETURN(-EINVAL);
 378         }
 379
 380         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 381                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 382                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 383                 cluuid->uuid);
 384
 385         if (!mdc_obd->obd_set_up) {
 386                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 387                 RETURN(-EINVAL);
 388         }
 389
 390         rc = obd_connect(NULL, &conn, mdc_obd, &lmv_mdc_uuid,
 391                          &lmv->conn_data);
 392         if (rc) {
 393                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 394                 RETURN(rc);
 395         }
 396
 397         mdc_exp = class_conn2export(&conn);
 398
 399         /* Init fid sequence client for this mdc. */
 400         rc = obd_fid_init(mdc_exp);
 401         if (rc)
 402                 RETURN(rc);
 403
 404         /* Add new FLD target. */
 405         target.ft_srv = NULL;
 406         target.ft_exp = mdc_exp;
 407         target.ft_idx = tgt->ltd_idx;
 408
 409         fld_client_add_target(&lmv->lmv_fld, &target);
 410
 411         mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
 412
 413         rc = obd_register_observer(mdc_obd, obd);
 414         if (rc) {
 415                 obd_disconnect(mdc_exp);
 416                 CERROR("target %s register_observer error %d\n",
 417                        tgt->ltd_uuid.uuid, rc);
 418                 RETURN(rc);
 419         }
 420
 421         if (obd->obd_observer) {
 422                 /* tell the mds_lmv about the new target */
 423                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 424                                 OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
 425                 if (rc) {
 426                         obd_disconnect(mdc_exp);
 427                         RETURN(rc);
 428                 }
 429         }
 430
 431         tgt->ltd_active = 1;
 432         tgt->ltd_exp = mdc_exp;
 433         lmv->desc.ld_active_tgt_count++;
 434
 435         /* copy connect data, it may be used later */
 436         lmv->datas[tgt->ltd_idx] = *mdc_data;
 437
 438         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 439                         lmv->max_def_easize, lmv->max_cookiesize);
 440
 441         CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
 442                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 443                 atomic_read(&obd->obd_refcount));
 444
 445 #ifdef __KERNEL__
 446         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 447         if (lmv_proc_dir) {
 448                 struct proc_dir_entry *mdc_symlink;
 449                 char name[MAX_STRING_SIZE + 1];
 450
 451                 LASSERT(mdc_obd->obd_type != NULL);
 452                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 453                 name[MAX_STRING_SIZE] = '\0';
 454                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 455                          mdc_obd->obd_type->typ_name,
 456                          mdc_obd->obd_name);
 457                 mdc_symlink = proc_symlink(mdc_obd->obd_name,
 458                                            lmv_proc_dir, name);
 459                 if (mdc_symlink == NULL) {
 460                         CERROR("could not register LMV target "
 461                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 462                                obd->obd_type->typ_name, obd->obd_name,
 463                                mdc_obd->obd_name);
 464                         lprocfs_remove(&lmv_proc_dir);
 465                         lmv_proc_dir = NULL;
 466                 }
 467         }
 468 #endif
 469         RETURN(0);
 470 }
 471
 472 int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 473 {
 474         struct lmv_obd *lmv = &obd->u.lmv;
 475         struct lmv_tgt_desc *tgt;
 476         int rc = 0;
 477         ENTRY;
 478
 479         CDEBUG(D_CONFIG, "tgt_uuid: %s.\n", tgt_uuid->uuid);
 480
 481         lmv_init_lock(lmv);
 482
 483         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 484                 lmv_init_unlock(lmv);
 485                 CERROR("can't add %s, LMV module compiled for %d MDCs. "
 486                        "That many MDCs already configured.\n",
 487                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 488                 RETURN(-EINVAL);
 489         }
 490         if (lmv->desc.ld_tgt_count == 0) {
 491                 struct obd_device *mdc_obd;
 492
 493                 mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
 494                                                 &obd->obd_uuid);
 495                 if (!mdc_obd) {
 496                         lmv_init_unlock(lmv);
 497                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 498                         RETURN(-EINVAL);
 499                 }
 500
 501                 rc = obd_llog_init(obd, NULL, mdc_obd, 0, NULL, tgt_uuid);
 502                 if (rc) {
 503                         lmv_init_unlock(lmv);
 504                         CERROR("lmv failed to setup llogging subsystems\n");
 505                 }
 506         }
 507         spin_lock(&lmv->lmv_lock);
 508         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 509         tgt->ltd_uuid = *tgt_uuid;
 510         spin_unlock(&lmv->lmv_lock);
 511
 512         if (lmv->connected) {
 513                 rc = lmv_connect_mdc(obd, tgt);
 514                 if (rc) {
 515                         spin_lock(&lmv->lmv_lock);
 516                         lmv->desc.ld_tgt_count--;
 517                         memset(tgt, 0, sizeof(*tgt));
 518                         spin_unlock(&lmv->lmv_lock);
 519                 } else {
 520                         int easize = sizeof(struct lmv_stripe_md) +
 521                                      lmv->desc.ld_tgt_count *
 522                                      sizeof(struct lu_fid);
 523                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 524                 }
 525         }
 526
 527         lmv_init_unlock(lmv);
 528         RETURN(rc);
 529 }
 530
 531 /* performs a check if passed obd is connected. If no - connect it. */
 532 int lmv_check_connect(struct obd_device *obd)
 533 {
 534         struct lmv_obd *lmv = &obd->u.lmv;
 535         struct lmv_tgt_desc *tgt;
 536         int i, rc, easize;
 537         ENTRY;
 538
 539         if (lmv->connected)
 540                 RETURN(0);
 541
 542         lmv_init_lock(lmv);
 543         if (lmv->connected) {
 544                 lmv_init_unlock(lmv);
 545                 RETURN(0);
 546         }
 547
 548         if (lmv->desc.ld_tgt_count == 0) {
 549                 CERROR("%s: no targets configured.\n", obd->obd_name);
 550                 RETURN(-EINVAL);
 551         }
 552
 553         CDEBUG(D_CONFIG, "time to connect %s to %s\n",
 554                lmv->cluuid.uuid, obd->obd_name);
 555
 556         LASSERT(lmv->tgts != NULL);
 557
 558         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 559                 rc = lmv_connect_mdc(obd, tgt);
 560                 if (rc)
 561                         GOTO(out_disc, rc);
 562         }
 563
 564         lmv_set_timeouts(obd);
 565         class_export_put(lmv->exp);
 566         lmv->connected = 1;
 567         easize = lmv_get_easize(lmv);
 568         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 569         lmv_init_unlock(lmv);
 570         RETURN(0);
 571
 572  out_disc:
 573         while (i-- > 0) {
 574                 int rc2;
 575                 --tgt;
 576                 tgt->ltd_active = 0;
 577                 if (tgt->ltd_exp) {
 578                         --lmv->desc.ld_active_tgt_count;
 579                         rc2 = obd_disconnect(tgt->ltd_exp);
 580                         if (rc2) {
 581                                 CERROR("error: LMV target %s disconnect on "
 582                                        "MDC idx %d: error %d\n",
 583                                        tgt->ltd_uuid.uuid, i, rc2);
 584                         }
 585                 }
 586         }
 587         class_disconnect(lmv->exp);
 588         lmv_init_unlock(lmv);
 589         RETURN(rc);
 590 }
 591
 592 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 593 {
 594 #ifdef __KERNEL__
 595         struct proc_dir_entry *lmv_proc_dir;
 596 #endif
 597         struct lmv_obd *lmv = &obd->u.lmv;
 598         struct obd_device *mdc_obd;
 599         int rc;
 600         ENTRY;
 601
 602         LASSERT(tgt != NULL);
 603         LASSERT(obd != NULL);
 604
 605         mdc_obd = class_exp2obd(tgt->ltd_exp);
 606
 607         if (mdc_obd)
 608                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 609
 610 #ifdef __KERNEL__
 611         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 612         if (lmv_proc_dir) {
 613                 struct proc_dir_entry *mdc_symlink;
 614
 615                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 616                 if (mdc_symlink) {
 617                         lprocfs_remove(&mdc_symlink);
 618                 } else {
 619                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 620                                obd->obd_type->typ_name, obd->obd_name,
 621                                mdc_obd->obd_name);
 622                 }
 623         }
 624 #endif
 625         rc = obd_fid_fini(tgt->ltd_exp);
 626         if (rc)
 627                 CERROR("Can't finanize fids factory\n");
 628
 629         CDEBUG(D_OTHER, "Disconnected from %s(%s) successfully\n",
 630                tgt->ltd_exp->exp_obd->obd_name,
 631                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 632
 633         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 634         rc = obd_disconnect(tgt->ltd_exp);
 635         if (rc) {
 636                 if (tgt->ltd_active) {
 637                         CERROR("Target %s disconnect error %d\n",
 638                                tgt->ltd_uuid.uuid, rc);
 639                 }
 640         }
 641
 642         lmv_activate_target(lmv, tgt, 0);
 643         tgt->ltd_exp = NULL;
 644         RETURN(0);
 645 }
 646
 647 static int lmv_disconnect(struct obd_export *exp)
 648 {
 649         struct obd_device *obd = class_exp2obd(exp);
 650 #ifdef __KERNEL__
 651         struct proc_dir_entry *lmv_proc_dir;
 652 #endif
 653         struct lmv_obd *lmv = &obd->u.lmv;
 654         int rc, i;
 655         ENTRY;
 656
 657         if (!lmv->tgts)
 658                 goto out_local;
 659
 660         /* Only disconnect the underlying layers on the final disconnect. */
 661         lmv->refcount--;
 662         if (lmv->refcount != 0)
 663                 goto out_local;
 664
 665         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 666                 if (lmv->tgts[i].ltd_exp == NULL)
 667                         continue;
 668                 lmv_disconnect_mdc(obd, &lmv->tgts[i]);
 669         }
 670
 671 #ifdef __KERNEL__
 672         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 673         if (lmv_proc_dir) {
 674                 lprocfs_remove(&lmv_proc_dir);
 675         } else {
 676                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 677                        obd->obd_type->typ_name, obd->obd_name);
 678         }
 679 #endif
 680
 681 out_local:
 682         /*
 683          * This is the case when no real connection is established by
 684          * lmv_check_connect().
 685          */
 686         if (!lmv->connected)
 687                 class_export_put(exp);
 688         rc = class_disconnect(exp);
 689         if (lmv->refcount == 0)
 690                 lmv->connected = 0;
 691         RETURN(rc);
 692 }
 693
 694 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 695                          int len, void *karg, void *uarg)
 696 {
 697         struct obd_device *obddev = class_exp2obd(exp);
 698         struct lmv_obd *lmv = &obddev->u.lmv;
 699         int i, rc = 0, set = 0;
 700         ENTRY;
 701
 702         if (lmv->desc.ld_tgt_count == 0)
 703                 RETURN(-ENOTTY);
 704
 705         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 706                 int err;
 707
 708                 if (lmv->tgts[i].ltd_exp == NULL)
 709                         continue;
 710
 711                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 712                 if (err) {
 713                         if (lmv->tgts[i].ltd_active) {
 714                                 CERROR("error: iocontrol MDC %s on MDT"
 715                                        "idx %d: err = %d\n",
 716                                        lmv->tgts[i].ltd_uuid.uuid, i, err);
 717                                 if (!rc)
 718                                         rc = err;
 719                         }
 720                 } else
 721                         set = 1;
 722         }
 723         if (!set && !rc)
 724                 rc = -EIO;
 725
 726         RETURN(rc);
 727 }
 728
 729 enum MDS_POLICY {
 730      CHAR_TYPE,
 731      NID_TYPE
 732 };
 733
 734 static int lmv_all_chars_policy(int count, const char *name,
 735                                 int len)
 736 {
 737         unsigned int c = 0;
 738
 739         while (len > 0)
 740                 c += name[--len];
 741         c = c % count;
 742         return c;
 743 }
 744
 745 static int lmv_nid_policy(struct lmv_obd *lmv)
 746 {
 747         struct obd_import *imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
 748         __u32 id;
 749         /*
 750          * XXX Hack: to get nid we assume that underlying obd device is mdc.
 751          */
 752         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
 753         return id % lmv->desc.ld_tgt_count;
 754 }
 755
 756 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
 757                           int type)
 758 {
 759         switch (type) {
 760         case CHAR_TYPE:
 761                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
 762                                             op_data->op_name,
 763                                             op_data->op_namelen);
 764         case NID_TYPE:
 765                 return lmv_nid_policy(lmv);
 766
 767         default:
 768                 break;
 769         }
 770
 771         CERROR("unsupport type %d \n", type);
 772         return -EINVAL;
 773 }
 774
 775 /* This is _inode_ placement policy function (not name). */
 776 static int lmv_placement_policy(struct obd_device *obd,
 777                                 struct md_op_data *op_data,
 778                                 mdsno_t *mds)
 779 {
 780         struct lmv_obd *lmv = &obd->u.lmv;
 781         struct lmv_obj *obj;
 782         int rc;
 783         ENTRY;
 784
 785         LASSERT(mds != NULL);
 786
 787         /*
 788          * Allocate new fid on target according to operation type and parent
 789          * home mds.
 790          */
 791         obj = lmv_obj_grab(obd, &op_data->op_fid1);
 792         if (obj != NULL || op_data->op_name == NULL ||
 793             op_data->op_opc != LUSTRE_OPC_MKDIR) {
 794                 /*
 795                  * Allocate fid for non-dir or for null name or for case parent
 796                  * dir is split.
 797                  */
 798                 if (obj) {
 799                         lmv_obj_put(obj);
 800
 801                         /*
 802                          * If we have this flag turned on, and we see that
 803                          * parent dir is split, this means, that caller did not
 804                          * notice split yet. This is race and we would like to
 805                          * let caller know that.
 806                          */
 807                         if (op_data->op_bias & MDS_CHECK_SPLIT)
 808                                 RETURN(-ERESTART);
 809                 }
 810
 811                 /*
 812                  * Allocate new fid on same mds where parent fid is located and
 813                  * where operation will be sent. In case of split dir, ->op_fid1
 814                  * and ->op_mds here will contain fid and mds of slave directory
 815                  * object (assigned by caller).
 816                  */
 817                 *mds = op_data->op_mds;
 818                 rc = 0;
 819
 820 #if 0
 821                 /* XXX: This should be removed later wehn we sure it is not
 822                  * needed. */
 823                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, mds);
 824                 if (rc)
 825                         GOTO(out, rc);
 826 #endif
 827         } else {
 828                 /*
 829                  * Parent directory is not split and we want to create a
 830                  * directory in it. Let's calculate where to place it according
 831                  * to name.
 832                  */
 833                 *mds = lmv_choose_mds(lmv, op_data, NID_TYPE);
 834                 rc = 0;
 835         }
 836         EXIT;
 837 #if 0
 838 out:
 839 #endif
 840         if (rc) {
 841                 CERROR("Can't choose MDS, err = %d\n", rc);
 842         } else {
 843                 LASSERT(*mds < lmv->desc.ld_tgt_count);
 844         }
 845
 846         return rc;
 847 }
 848
 849 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
 850                     mdsno_t mds)
 851 {
 852         struct lmv_tgt_desc *tgt = &lmv->tgts[mds];
 853         int rc;
 854         ENTRY;
 855
 856         /* New seq alloc and FLD setup should be atomic. */
 857         down(&tgt->ltd_fid_sem);
 858
 859         /* Asking underlaying tgt layer to allocate new fid. */
 860         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
 861         if (rc > 0) {
 862                 LASSERT(fid_is_sane(fid));
 863
 864                 /* Client switches to new sequence, setup FLD. */
 865                 rc = fld_client_create(&lmv->lmv_fld, fid_seq(fid),
 866                                        mds, NULL);
 867                 if (rc) {
 868                         CERROR("Can't create fld entry, rc %d\n", rc);
 869                         /* Delete just allocated fid sequence */
 870                         obd_fid_delete(tgt->ltd_exp, NULL);
 871                 }
 872         }
 873         up(&tgt->ltd_fid_sem);
 874         RETURN(rc);
 875 }
 876
 877 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
 878                   struct md_op_data *op_data)
 879 {
 880         struct obd_device *obd = class_exp2obd(exp);
 881         struct lmv_obd *lmv = &obd->u.lmv;
 882         mdsno_t mds;
 883         int rc;
 884         ENTRY;
 885
 886         LASSERT(op_data != NULL);
 887         LASSERT(fid != NULL);
 888
 889         rc = lmv_placement_policy(obd, op_data, &mds);
 890         if (rc) {
 891                 CERROR("Can't get target for allocating fid, "
 892                        "rc %d\n", rc);
 893                 RETURN(rc);
 894         }
 895
 896         rc = __lmv_fid_alloc(lmv, fid, mds);
 897         if (rc) {
 898                 CERROR("Can't alloc new fid, rc %d\n", rc);
 899                 RETURN(rc);
 900         }
 901
 902         RETURN(rc);
 903 }
 904
 905 static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
 906 {
 907         ENTRY;
 908
 909         LASSERT(exp && fid);
 910         if (lmv_obj_delete(exp, fid)) {
 911                 CDEBUG(D_OTHER, "lmv object "DFID" is destroyed.\n",
 912                        PFID(fid));
 913         }
 914         RETURN(0);
 915 }
 916
 917 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 918 {
 919         struct lmv_obd *lmv = &obd->u.lmv;
 920         struct lprocfs_static_vars lvars;
 921         struct lmv_desc *desc;
 922         int rc, i = 0;
 923         ENTRY;
 924
 925         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 926                 CERROR("LMV setup requires a descriptor\n");
 927                 RETURN(-EINVAL);
 928         }
 929
 930         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
 931         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 932                 CERROR("descriptor size wrong: %d > %d\n",
 933                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 934                 RETURN(-EINVAL);
 935         }
 936
 937         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
 938
 939         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 940         if (lmv->tgts == NULL)
 941                 RETURN(-ENOMEM);
 942
 943         for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
 944                 sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
 945                 lmv->tgts[i].ltd_idx = i;
 946         }
 947
 948         lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
 949
 950         OBD_ALLOC(lmv->datas, lmv->datas_size);
 951         if (lmv->datas == NULL)
 952                 GOTO(out_free_tgts, rc = -ENOMEM);
 953
 954         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
 955         lmv->desc.ld_tgt_count = 0;
 956         lmv->desc.ld_active_tgt_count = 0;
 957         lmv->max_cookiesize = 0;
 958         lmv->max_def_easize = 0;
 959         lmv->max_easize = 0;
 960
 961         spin_lock_init(&lmv->lmv_lock);
 962         sema_init(&lmv->init_sem, 1);
 963
 964         rc = lmv_obj_setup(obd);
 965         if (rc) {
 966                 CERROR("Can't setup LMV object manager, "
 967                        "error %d.\n", rc);
 968                 GOTO(out_free_datas, rc);
 969         }
 970
 971         lprocfs_init_vars(lmv, &lvars);
 972         lprocfs_obd_setup(obd, lvars.obd_vars);
 973 #ifdef LPROCFS
 974         {
 975                 struct proc_dir_entry *entry;
 976
 977                 entry = create_proc_entry("target_obd_status", 0444,
 978                                           obd->obd_proc_entry);
 979                 if (entry != NULL) {
 980                         entry->proc_fops = &lmv_proc_target_fops;
 981                         entry->data = obd;
 982                 }
 983        }
 984 #endif
 985         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
 986                              LUSTRE_CLI_FLD_HASH_DHT);
 987         if (rc) {
 988                 CERROR("can't init FLD, err %d\n",
 989                        rc);
 990                 GOTO(out_free_datas, rc);
 991         }
 992
 993         RETURN(0);
 994
 995 out_free_datas:
 996         OBD_FREE(lmv->datas, lmv->datas_size);
 997         lmv->datas = NULL;
 998 out_free_tgts:
 999         OBD_FREE(lmv->tgts, lmv->tgts_size);
1000         lmv->tgts = NULL;
1001         return rc;
1002 }
1003
1004 static int lmv_cleanup(struct obd_device *obd)
1005 {
1006         struct lmv_obd *lmv = &obd->u.lmv;
1007         ENTRY;
1008
1009         fld_client_fini(&lmv->lmv_fld);
1010         lprocfs_obd_cleanup(obd);
1011         lmv_obj_cleanup(obd);
1012         OBD_FREE(lmv->datas, lmv->datas_size);
1013         OBD_FREE(lmv->tgts, lmv->tgts_size);
1014
1015         RETURN(0);
1016 }
1017
1018 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1019 {
1020         struct lustre_cfg *lcfg = buf;
1021         struct obd_uuid tgt_uuid;
1022         int rc;
1023         ENTRY;
1024
1025         switch(lcfg->lcfg_command) {
1026         case LCFG_ADD_MDC:
1027                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
1028                         GOTO(out, rc = -EINVAL);
1029
1030                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
1031                 rc = lmv_add_target(obd, &tgt_uuid);
1032                 GOTO(out, rc);
1033         default: {
1034                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1035                 GOTO(out, rc = -EINVAL);
1036         }
1037         }
1038 out:
1039         RETURN(rc);
1040 }
1041
1042 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1043                       __u64 max_age)
1044 {
1045         struct lmv_obd *lmv = &obd->u.lmv;
1046         struct obd_statfs *temp;
1047         int rc = 0, i;
1048         ENTRY;
1049
1050         rc = lmv_check_connect(obd);
1051         if (rc)
1052                 RETURN(rc);
1053
1054         OBD_ALLOC(temp, sizeof(*temp));
1055         if (temp == NULL)
1056                 RETURN(-ENOMEM);
1057
1058         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1059                 if (lmv->tgts[i].ltd_exp == NULL)
1060                         continue;
1061
1062                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
1063                 if (rc) {
1064                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1065                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
1066                                rc);
1067                         GOTO(out_free_temp, rc);
1068                 }
1069                 if (i == 0) {
1070                         *osfs = *temp;
1071                 } else {
1072                         osfs->os_bavail += temp->os_bavail;
1073                         osfs->os_blocks += temp->os_blocks;
1074                         osfs->os_ffree += temp->os_ffree;
1075                         osfs->os_files += temp->os_files;
1076                 }
1077         }
1078
1079         EXIT;
1080 out_free_temp:
1081         OBD_FREE(temp, sizeof(*temp));
1082         return rc;
1083 }
1084
1085 static int lmv_getstatus(struct obd_export *exp,
1086                          struct lu_fid *fid,
1087                          struct obd_capa **pc)
1088 {
1089         struct obd_device *obd = exp->exp_obd;
1090         struct lmv_obd *lmv = &obd->u.lmv;
1091         int rc;
1092         ENTRY;
1093
1094         rc = lmv_check_connect(obd);
1095         if (rc)
1096                 RETURN(rc);
1097
1098         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
1099
1100         RETURN(rc);
1101 }
1102
1103 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1104                         struct obd_capa *oc, obd_valid valid, const char *name,
1105                         const char *input, int input_size, int output_size,
1106                         int flags, struct ptlrpc_request **request)
1107 {
1108         struct obd_device *obd = exp->exp_obd;
1109         struct lmv_obd *lmv = &obd->u.lmv;
1110         struct obd_export *tgt_exp;
1111         int rc;
1112         ENTRY;
1113
1114         rc = lmv_check_connect(obd);
1115         if (rc)
1116                 RETURN(rc);
1117
1118         tgt_exp = lmv_find_export(lmv, fid);
1119         if (IS_ERR(tgt_exp))
1120                 RETURN(PTR_ERR(tgt_exp));
1121
1122         rc = md_getxattr(tgt_exp, fid, oc, valid, name, input, input_size,
1123                          output_size, flags, request);
1124
1125         RETURN(rc);
1126 }
1127
1128 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1129                         struct obd_capa *oc, obd_valid valid, const char *name,
1130                         const char *input, int input_size, int output_size,
1131                         int flags, struct ptlrpc_request **request)
1132 {
1133         struct obd_device *obd = exp->exp_obd;
1134         struct lmv_obd *lmv = &obd->u.lmv;
1135         struct obd_export *tgt_exp;
1136         int rc;
1137         ENTRY;
1138
1139         rc = lmv_check_connect(obd);
1140         if (rc)
1141                 RETURN(rc);
1142
1143         tgt_exp = lmv_find_export(lmv, fid);
1144         if (IS_ERR(tgt_exp))
1145                 RETURN(PTR_ERR(tgt_exp));
1146
1147         rc = md_setxattr(tgt_exp, fid, oc, valid, name,
1148                          input, input_size, output_size, flags, request);
1149
1150         RETURN(rc);
1151 }
1152
1153 static int lmv_getattr(struct obd_export *exp, const struct lu_fid *fid,
1154                        struct obd_capa *oc, obd_valid valid, int ea_size,
1155                        struct ptlrpc_request **request)
1156 {
1157         struct obd_device *obd = exp->exp_obd;
1158         struct lmv_obd *lmv = &obd->u.lmv;
1159         struct obd_export *tgt_exp;
1160         struct lmv_obj *obj;
1161         int rc, i;
1162         ENTRY;
1163
1164         rc = lmv_check_connect(obd);
1165         if (rc)
1166                 RETURN(rc);
1167
1168         tgt_exp = lmv_find_export(lmv, fid);
1169         if (IS_ERR(tgt_exp))
1170                 RETURN(PTR_ERR(tgt_exp));
1171
1172         rc = md_getattr(tgt_exp, fid, oc, valid, ea_size, request);
1173         if (rc)
1174                 RETURN(rc);
1175
1176         obj = lmv_obj_grab(obd, fid);
1177
1178         CDEBUG(D_OTHER, "GETATTR for "DFID" %s\n", PFID(fid),
1179                obj ? "(split)" : "");
1180
1181         /*
1182          * If object is split, then we loop over all the slaves and gather size
1183          * attribute. In ideal world we would have to gather also mds field from
1184          * all slaves, as object is spread over the cluster and this is
1185          * definitely interesting information and it is not good to loss it,
1186          * but...
1187          */
1188         if (obj) {
1189                 struct mdt_body *body;
1190
1191                 if (*request == NULL) {
1192                         lmv_obj_put(obj);
1193                         RETURN(rc);
1194                 }
1195
1196                 body = lustre_msg_buf((*request)->rq_repmsg, REPLY_REC_OFF,
1197                                       sizeof(*body));
1198                 LASSERT(body != NULL);
1199                 LASSERT_REPSWABBED((*request), REPLY_REC_OFF);
1200
1201                 lmv_obj_lock(obj);
1202
1203                 for (i = 0; i < obj->lo_objcount; i++) {
1204                         if (lmv->tgts[i].ltd_exp == NULL) {
1205                                 CWARN("%s: NULL export for %d\n",
1206                                       obd->obd_name, i);
1207                                 continue;
1208                         }
1209
1210                         /* skip master obj. */
1211                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid))
1212                                 continue;
1213
1214                         lmv_update_body(body, &obj->lo_inodes[i]);
1215                 }
1216
1217                 lmv_obj_unlock(obj);
1218                 lmv_obj_put(obj);
1219         }
1220
1221         RETURN(rc);
1222 }
1223
1224 static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1225                              ldlm_iterator_t it, void *data)
1226 {
1227         struct obd_device *obd = exp->exp_obd;
1228         struct lmv_obd *lmv = &obd->u.lmv;
1229         int i, rc;
1230         ENTRY;
1231
1232         rc = lmv_check_connect(obd);
1233         if (rc)
1234                 RETURN(rc);
1235
1236         CDEBUG(D_OTHER, "CBDATA for "DFID"\n", PFID(fid));
1237
1238         /*
1239          * With CMD every object can have two locks in different namespaces:
1240          * lookup lock in space of mds storing direntry and update/open lock in
1241          * space of mds storing inode.
1242          */
1243         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
1244                 md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
1245
1246         RETURN(0);
1247 }
1248
1249 static int lmv_close(struct obd_export *exp,
1250                      struct md_op_data *op_data,
1251                      struct obd_client_handle *och,
1252                      struct ptlrpc_request **request)
1253 {
1254         struct obd_device *obd = exp->exp_obd;
1255         struct lmv_obd *lmv = &obd->u.lmv;
1256         struct obd_export *tgt_exp;
1257         int rc;
1258         ENTRY;
1259
1260         rc = lmv_check_connect(obd);
1261         if (rc)
1262                 RETURN(rc);
1263
1264         tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1265         if (IS_ERR(tgt_exp))
1266                 RETURN(PTR_ERR(tgt_exp));
1267
1268         CDEBUG(D_OTHER, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1269         rc = md_close(tgt_exp, op_data, och, request);
1270         RETURN(rc);
1271 }
1272
1273 /*
1274  * Called in the case MDS returns -ERESTART on create on open, what means that
1275  * directory is split and its LMV presentation object has to be updated.
1276  */
1277 int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
1278 {
1279         struct obd_device *obd = exp->exp_obd;
1280         struct lmv_obd *lmv = &obd->u.lmv;
1281         struct ptlrpc_request *req = NULL;
1282         struct obd_export *tgt_exp;
1283         struct lmv_obj *obj;
1284         struct lustre_md md;
1285         int mealen, rc;
1286         __u64 valid;
1287         ENTRY;
1288
1289         md.mea = NULL;
1290         mealen = lmv_get_easize(lmv);
1291
1292         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1293
1294         tgt_exp = lmv_find_export(lmv, fid);
1295         if (IS_ERR(tgt_exp))
1296                 RETURN(PTR_ERR(tgt_exp));
1297
1298         /* time to update mea of parent fid */
1299         rc = md_getattr(tgt_exp, fid, NULL, valid, mealen, &req);
1300         if (rc) {
1301                 CERROR("md_getattr() failed, error %d\n", rc);
1302                 GOTO(cleanup, rc);
1303         }
1304
1305         rc = md_get_lustre_md(tgt_exp, req, 1, NULL, exp, &md);
1306         if (rc) {
1307                 CERROR("mdc_get_lustre_md() failed, error %d\n", rc);
1308                 GOTO(cleanup, rc);
1309         }
1310
1311         if (md.mea == NULL)
1312                 GOTO(cleanup, rc = -ENODATA);
1313
1314         obj = lmv_obj_create(exp, fid, md.mea);
1315         if (IS_ERR(obj))
1316                 rc = PTR_ERR(obj);
1317         else
1318                 lmv_obj_put(obj);
1319
1320         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
1321
1322         EXIT;
1323 cleanup:
1324         if (req)
1325                 ptlrpc_req_finished(req);
1326         return rc;
1327 }
1328
1329 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1330                const void *data, int datalen, int mode, __u32 uid,
1331                __u32 gid, __u32 cap_effective,  __u64 rdev,
1332                struct ptlrpc_request **request)
1333 {
1334         struct obd_device *obd = exp->exp_obd;
1335         struct lmv_obd *lmv = &obd->u.lmv;
1336         struct obd_export *tgt_exp;
1337         struct lmv_obj *obj;
1338         int rc, loop = 0;
1339         ENTRY;
1340
1341         rc = lmv_check_connect(obd);
1342         if (rc)
1343                 RETURN(rc);
1344
1345         if (!lmv->desc.ld_active_tgt_count)
1346                 RETURN(-EIO);
1347 repeat:
1348         ++loop;
1349         LASSERT(loop <= 2);
1350         obj = lmv_obj_grab(obd, &op_data->op_fid1);
1351         if (obj) {
1352                 int mea_idx;
1353
1354                 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1355                                        op_data->op_name, op_data->op_namelen);
1356                 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1357                 op_data->op_bias &= ~MDS_CHECK_SPLIT;
1358                 op_data->op_mds = obj->lo_inodes[mea_idx].li_mds;
1359                 tgt_exp = lmv_get_export(lmv, op_data->op_mds);
1360                 lmv_obj_put(obj);
1361         } else {
1362                 struct lmv_tgt_desc *tgt;
1363
1364                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1365                 op_data->op_bias |= MDS_CHECK_SPLIT;
1366                 op_data->op_mds = tgt->ltd_idx;
1367                 tgt_exp = tgt->ltd_exp;
1368         }
1369
1370         if (IS_ERR(tgt_exp))
1371                 RETURN(PTR_ERR(tgt_exp));
1372
1373         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1374         if (rc == -ERESTART)
1375                 goto repeat;
1376         else if (rc)
1377                 RETURN(rc);
1378
1379         CDEBUG(D_OTHER, "CREATE '%*s' on "DFID"\n", op_data->op_namelen,
1380                op_data->op_name, PFID(&op_data->op_fid1));
1381
1382         rc = md_create(tgt_exp, op_data, data, datalen, mode, uid, gid,
1383                        cap_effective, rdev, request);
1384         if (rc == 0) {
1385                 if (*request == NULL)
1386                         RETURN(rc);
1387                 CDEBUG(D_OTHER, "created - "DFID"\n", PFID(&op_data->op_fid1));
1388         } else if (rc == -ERESTART) {
1389                 LASSERT(*request != NULL);
1390                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1391                           "Got -ERESTART during create!\n");
1392                 ptlrpc_req_finished(*request);
1393                 *request = NULL;
1394
1395                 /*
1396                  * Directory got split. Time to update local object and repeat
1397                  * the request with proper MDS.
1398                  */
1399                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1400                 if (rc == 0) {
1401                         rc = lmv_alloc_slave_fids(obd, &op_data->op_fid1,
1402                                                   op_data, &op_data->op_fid2);
1403                         if (rc)
1404                                 RETURN(rc);
1405                         goto repeat;
1406                 }
1407         }
1408         RETURN(rc);
1409 }
1410
1411 static int lmv_done_writing(struct obd_export *exp,
1412                             struct md_op_data *op_data,
1413                             struct obd_client_handle *och)
1414 {
1415         struct obd_device *obd = exp->exp_obd;
1416         struct lmv_obd *lmv = &obd->u.lmv;
1417         struct obd_export *tgt_exp;
1418         int rc;
1419         ENTRY;
1420
1421         rc = lmv_check_connect(obd);
1422         if (rc)
1423                 RETURN(rc);
1424
1425         tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1426         if (IS_ERR(tgt_exp))
1427                 RETURN(PTR_ERR(tgt_exp));
1428
1429         rc = md_done_writing(tgt_exp, op_data, och);
1430         RETURN(rc);
1431 }
1432
1433 static int
1434 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1435                    struct lookup_intent *it, int lockmode,
1436                    struct md_op_data *op_data, struct lustre_handle *lockh,
1437                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1438                    ldlm_blocking_callback cb_blocking, void *cb_data)
1439 {
1440         struct obd_device *obd = exp->exp_obd;
1441         struct lmv_obd *lmv = &obd->u.lmv;
1442         struct lmv_stripe_md *mea = op_data->op_mea1;
1443         struct md_op_data *op_data2;
1444         struct obd_export *tgt_exp;
1445         int i, rc = 0;
1446         ENTRY;
1447
1448         OBD_ALLOC_PTR(op_data2);
1449         if (op_data2 == NULL)
1450                 RETURN(-ENOMEM);
1451
1452         LASSERT(mea != NULL);
1453         for (i = 0; i < mea->mea_count; i++) {
1454                 memset(op_data2, 0, sizeof(*op_data2));
1455                 op_data2->op_fid1 = mea->mea_ids[i];
1456                 op_data2->op_bias = 0;
1457
1458                 tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
1459                 if (IS_ERR(tgt_exp))
1460                         GOTO(cleanup, rc = PTR_ERR(tgt_exp));
1461
1462                 if (tgt_exp == NULL)
1463                         continue;
1464
1465                 rc = md_enqueue(tgt_exp, locktype, it, lockmode, op_data2,
1466                                 lockh + i, lmm, lmmsize, cb_compl, cb_blocking,
1467                                 cb_data, 0);
1468
1469                 CDEBUG(D_OTHER, "take lock on slave "DFID" -> %d/%d\n",
1470                        PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
1471
1472                 if (rc)
1473                         GOTO(cleanup, rc);
1474
1475                 if (it->d.lustre.it_data) {
1476                         struct ptlrpc_request *req;
1477                         req = (struct ptlrpc_request *)it->d.lustre.it_data;
1478                         ptlrpc_req_finished(req);
1479                 }
1480
1481                 if (it->d.lustre.it_status)
1482                         GOTO(cleanup, rc = it->d.lustre.it_status);
1483         }
1484
1485         EXIT;
1486 cleanup:
1487         OBD_FREE_PTR(op_data2);
1488
1489         if (rc != 0) {
1490                 /* drop all taken locks */
1491                 while (--i >= 0) {
1492                         if (lockh[i].cookie)
1493                                 ldlm_lock_decref(lockh + i, lockmode);
1494                         lockh[i].cookie = 0;
1495                 }
1496         }
1497         return rc;
1498 }
1499
1500 static int
1501 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1502                    struct lookup_intent *it, int lock_mode,
1503                    struct md_op_data *op_data, struct lustre_handle *lockh,
1504                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1505                    ldlm_blocking_callback cb_blocking, void *cb_data,
1506                    int extra_lock_flags)
1507 {
1508         struct ptlrpc_request *req = it->d.lustre.it_data;
1509         struct obd_device *obd = exp->exp_obd;
1510         struct lmv_obd *lmv = &obd->u.lmv;
1511         struct lustre_handle plock;
1512         struct obd_export *tgt_exp;
1513         struct md_op_data *rdata;
1514         struct lu_fid fid_copy;
1515         struct mdt_body *body;
1516         int rc = 0, pmode;
1517         ENTRY;
1518
1519         body = lustre_msg_buf(req->rq_repmsg,
1520                               DLM_REPLY_REC_OFF, sizeof(*body));
1521         LASSERT(body != NULL);
1522         LASSERT_REPSWABBED(req, DLM_REPLY_REC_OFF);
1523
1524         if (!(body->valid & OBD_MD_MDS))
1525                 RETURN(0);
1526
1527         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID" -> "DFID"\n",
1528                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1529
1530         /* We got LOOKUP lock, but we really need attrs */
1531         pmode = it->d.lustre.it_lock_mode;
1532         LASSERT(pmode != 0);
1533         memcpy(&plock, lockh, sizeof(plock));
1534         it->d.lustre.it_lock_mode = 0;
1535         it->d.lustre.it_data = NULL;
1536         fid_copy = body->fid1;
1537
1538         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1539         ptlrpc_req_finished(req);
1540
1541         tgt_exp = lmv_find_export(lmv, &fid_copy);
1542         if (IS_ERR(tgt_exp))
1543                 GOTO(out, rc = PTR_ERR(tgt_exp));
1544
1545         OBD_ALLOC_PTR(rdata);
1546         if (rdata == NULL)
1547                 GOTO(out, rc = -ENOMEM);
1548
1549         rdata->op_fid1 = fid_copy;
1550         rdata->op_bias = MDS_CROSS_REF;
1551
1552         rc = md_enqueue(tgt_exp, lock_type, it, lock_mode, rdata,
1553                         lockh, lmm, lmmsize, cb_compl, cb_blocking,
1554                         cb_data, extra_lock_flags);
1555         OBD_FREE_PTR(rdata);
1556         EXIT;
1557 out:
1558         ldlm_lock_decref(&plock, pmode);
1559         return rc;
1560 }
1561
1562 static int
1563 lmv_enqueue(struct obd_export *exp, int lock_type,
1564             struct lookup_intent *it, int lock_mode,
1565             struct md_op_data *op_data, struct lustre_handle *lockh,
1566             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1567             ldlm_blocking_callback cb_blocking, void *cb_data,
1568             int extra_lock_flags)
1569 {
1570         struct obd_device *obd = exp->exp_obd;
1571         struct lmv_obd *lmv = &obd->u.lmv;
1572         struct obd_export *tgt_exp = NULL;
1573         struct lmv_obj *obj;
1574         int rc;
1575         ENTRY;
1576
1577         rc = lmv_check_connect(obd);
1578         if (rc)
1579                 RETURN(rc);
1580
1581         if (op_data->op_mea1 && it->it_op == IT_UNLINK) {
1582                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1583                                         op_data, lockh, lmm, lmmsize,
1584                                         cb_compl, cb_blocking, cb_data);
1585                 RETURN(rc);
1586         }
1587
1588         if (op_data->op_namelen) {
1589                 obj = lmv_obj_grab(obd, &op_data->op_fid1);
1590                 if (obj) {
1591                         int mea_idx;
1592
1593                         /* directory is split. look for right mds for this
1594                          * name */
1595                         mea_idx = raw_name2idx(obj->lo_hashtype,
1596                                                obj->lo_objcount,
1597                                                (char *)op_data->op_name,
1598                                                op_data->op_namelen);
1599                         op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1600                         tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
1601                         lmv_obj_put(obj);
1602                 }
1603         }
1604
1605         if (tgt_exp == NULL)
1606                 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1607         if (IS_ERR(tgt_exp))
1608                 RETURN(PTR_ERR(tgt_exp));
1609
1610         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID"\n", LL_IT2STR(it),
1611                PFID(&op_data->op_fid1));
1612
1613         rc = md_enqueue(tgt_exp, lock_type, it, lock_mode, op_data, lockh,
1614                         lmm, lmmsize, cb_compl, cb_blocking, cb_data,
1615                         extra_lock_flags);
1616
1617         if (rc == 0 && it->it_op == IT_OPEN)
1618                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1619                                         op_data, lockh, lmm, lmmsize,
1620                                         cb_compl, cb_blocking, cb_data,
1621                                         extra_lock_flags);
1622         RETURN(rc);
1623 }
1624
1625 static int
1626 lmv_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
1627                  struct obd_capa *oc, const char *filename, int namelen,
1628                  obd_valid valid, int ea_size, struct ptlrpc_request **request)
1629 {
1630         struct obd_device *obd = exp->exp_obd;
1631         struct lmv_obd *lmv = &obd->u.lmv;
1632         struct lu_fid rid = *fid;
1633         struct obd_export *tgt_exp;
1634         struct mdt_body *body;
1635         struct lmv_obj *obj;
1636         int rc, loop = 0;
1637         ENTRY;
1638
1639         rc = lmv_check_connect(obd);
1640         if (rc)
1641                 RETURN(rc);
1642
1643 repeat:
1644         ++loop;
1645         LASSERT(loop <= 2);
1646         obj = lmv_obj_grab(obd, &rid);
1647         if (obj) {
1648                 int mea_idx;
1649
1650                 /* Directory is split. Look for right mds for this name */
1651                 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1652                                        filename, namelen - 1);
1653                 rid = obj->lo_inodes[mea_idx].li_fid;
1654                 tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
1655                 lmv_obj_put(obj);
1656                 valid &= ~OBD_MD_FLCKSPLIT;
1657         } else {
1658                 tgt_exp = lmv_find_export(lmv, &rid);
1659                 valid |= OBD_MD_FLCKSPLIT;
1660         }
1661         if (IS_ERR(tgt_exp))
1662                 RETURN(PTR_ERR(tgt_exp));
1663
1664         CDEBUG(D_OTHER, "getattr_name for %*s on "DFID" -> "DFID"\n",
1665                namelen, filename, PFID(fid), PFID(&rid));
1666
1667         rc = md_getattr_name(tgt_exp, &rid, oc, filename, namelen, valid,
1668                              ea_size, request);
1669         if (rc == 0) {
1670                 body = lustre_msg_buf((*request)->rq_repmsg,
1671                                       REQ_REC_OFF, sizeof(*body));
1672                 LASSERT(body != NULL);
1673                 LASSERT_REPSWABBED((*request), REQ_REC_OFF);
1674
1675                 if (body->valid & OBD_MD_MDS) {
1676                         struct ptlrpc_request *req = NULL;
1677
1678                         rid = body->fid1;
1679                         CDEBUG(D_OTHER, "request attrs for "DFID"\n",
1680                                PFID(&rid));
1681
1682                         tgt_exp = lmv_find_export(lmv, &rid);
1683                         if (IS_ERR(tgt_exp)) {
1684                                 ptlrpc_req_finished(*request);
1685                                 RETURN(PTR_ERR(tgt_exp));
1686                         }
1687
1688                         rc = md_getattr_name(tgt_exp, &rid, NULL, NULL, 1,
1689                                              valid | OBD_MD_FLCROSSREF,
1690                                              ea_size, &req);
1691                         ptlrpc_req_finished(*request);
1692                         *request = req;
1693                 }
1694         } else if (rc == -ERESTART) {
1695                 LASSERT(*request != NULL);
1696                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1697                           "Got -ERESTART during getattr!\n");
1698                 ptlrpc_req_finished(*request);
1699                 *request = NULL;
1700
1701                 /*
1702                  * Directory got split. Time to update local object and repeat
1703                  * the request with proper MDS.
1704                  */
1705                 rc = lmv_handle_split(exp, &rid);
1706                 if (rc == 0)
1707                         goto repeat;
1708         }
1709         RETURN(rc);
1710 }
1711
1712 /*
1713  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1714  * op_data->op_fid2
1715  */
1716 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1717                     struct ptlrpc_request **request)
1718 {
1719         struct obd_device *obd = exp->exp_obd;
1720         struct lmv_obd *lmv = &obd->u.lmv;
1721         struct lmv_obj *obj;
1722         int rc, loop = 0;
1723         mdsno_t mds;
1724         ENTRY;
1725
1726         rc = lmv_check_connect(obd);
1727         if (rc)
1728                 RETURN(rc);
1729
1730 repeat:
1731         ++loop;
1732         LASSERT(loop <= 2);
1733         if (op_data->op_namelen != 0) {
1734                 int mea_idx;
1735
1736                 /* Usual link request */
1737                 obj = lmv_obj_grab(obd, &op_data->op_fid2);
1738                 if (obj) {
1739                         mea_idx = raw_name2idx(obj->lo_hashtype,
1740                                                obj->lo_objcount,
1741                                                op_data->op_name,
1742                                                op_data->op_namelen);
1743                         op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1744                         mds = obj->lo_inodes[mea_idx].li_mds;
1745                         lmv_obj_put(obj);
1746                 } else {
1747                         rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
1748                         if (rc)
1749                                 RETURN(rc);
1750                 }
1751
1752                 CDEBUG(D_OTHER,"link "DFID":%*s to "DFID"\n",
1753                        PFID(&op_data->op_fid2), op_data->op_namelen,
1754                        op_data->op_name, PFID(&op_data->op_fid1));
1755         } else {
1756                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
1757                 if (rc)
1758                         RETURN(rc);
1759
1760                 /* request from MDS to acquire i_links for inode by fid1 */
1761                 CDEBUG(D_OTHER, "inc i_nlinks for "DFID"\n",
1762                        PFID(&op_data->op_fid1));
1763         }
1764
1765         CDEBUG(D_OTHER, "forward to MDS #"LPU64" ("DFID")\n",
1766                mds, PFID(&op_data->op_fid1));
1767
1768         op_data->op_fsuid = current->fsuid;
1769         op_data->op_fsgid = current->fsgid;
1770         op_data->op_cap   = current->cap_effective;
1771
1772         rc = md_link(lmv->tgts[mds].ltd_exp, op_data, request);
1773         if (rc == -ERESTART) {
1774                 LASSERT(*request != NULL);
1775                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1776                           "Got -ERESTART during link!\n");
1777                 ptlrpc_req_finished(*request);
1778                 *request = NULL;
1779
1780                 /*
1781                  * Directory got split. Time to update local object and repeat
1782                  * the request with proper MDS.
1783                  */
1784                 rc = lmv_handle_split(exp, &op_data->op_fid2);
1785                 if (rc == 0)
1786                         goto repeat;
1787         }
1788
1789         RETURN(rc);
1790 }
1791
1792 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1793                       const char *old, int oldlen, const char *new, int newlen,
1794                       struct ptlrpc_request **request)
1795 {
1796         struct obd_device *obd = exp->exp_obd;
1797         struct lmv_obd *lmv = &obd->u.lmv;
1798         int rc, mea_idx, loop = 0;
1799         struct lmv_obj *obj;
1800         mdsno_t mds;
1801         ENTRY;
1802
1803         CDEBUG(D_OTHER, "rename %*s in "DFID" to %*s in "DFID"\n",
1804                oldlen, old, PFID(&op_data->op_fid1),
1805                newlen, new, PFID(&op_data->op_fid2));
1806
1807         rc = lmv_check_connect(obd);
1808         if (rc)
1809                 RETURN(rc);
1810
1811         if (oldlen == 0) {
1812                 /*
1813                  * MDS with old dir entry is asking another MDS to create name
1814                  * there.
1815                  */
1816                 CDEBUG(D_OTHER,
1817                        "create %*s(%d/%d) in "DFID" pointing "
1818                        "to "DFID"\n", newlen, new, oldlen, newlen,
1819                        PFID(&op_data->op_fid2), PFID(&op_data->op_fid1));
1820
1821                 rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
1822                 if (rc)
1823                         RETURN(rc);
1824
1825                 /*
1826                  * Target directory can be split, sowe should forward request to
1827                  * the right MDS.
1828                  */
1829                 obj = lmv_obj_grab(obd, &op_data->op_fid2);
1830                 if (obj) {
1831                         mea_idx = raw_name2idx(obj->lo_hashtype,
1832                                                obj->lo_objcount,
1833                                                (char *)new, newlen);
1834                         op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1835                         CDEBUG(D_OTHER, "Parent obj "DFID"\n",
1836                                PFID(&op_data->op_fid2));
1837                         lmv_obj_put(obj);
1838                 }
1839                 goto request;
1840         }
1841
1842 repeat:
1843         ++loop;
1844         LASSERT(loop <= 2);
1845         obj = lmv_obj_grab(obd, &op_data->op_fid1);
1846         if (obj) {
1847                 /*
1848                  * directory is already split, so we have to forward request to
1849                  * the right MDS.
1850                  */
1851                 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1852                                        (char *)old, oldlen);
1853                 op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
1854                 mds = obj->lo_inodes[mea_idx].li_mds;
1855                 CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
1856                 lmv_obj_put(obj);
1857         } else {
1858                 rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
1859                 if (rc)
1860                         RETURN(rc);
1861         }
1862
1863         obj = lmv_obj_grab(obd, &op_data->op_fid2);
1864         if (obj) {
1865                 /*
1866                  * Directory is already split, so we have to forward request to
1867                  * the right MDS.
1868                  */
1869                 mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
1870                                        (char *)new, newlen);
1871
1872                 op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
1873                 CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
1874                 lmv_obj_put(obj);
1875         }
1876
1877 request:
1878         op_data->op_fsuid = current->fsuid;
1879         op_data->op_fsgid = current->fsgid;
1880         op_data->op_cap   = current->cap_effective;
1881
1882         rc = md_rename(lmv->tgts[mds].ltd_exp, op_data, old, oldlen,
1883                        new, newlen, request);
1884         if (rc == -ERESTART) {
1885                 LASSERT(*request != NULL);
1886                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
1887                           "Got -ERESTART during rename!\n");
1888                 ptlrpc_req_finished(*request);
1889                 *request = NULL;
1890
1891                 /*
1892                  * Directory got split. Time to update local object and repeat
1893                  * the request with proper MDS.
1894                  */
1895                 rc = lmv_handle_split(exp, &op_data->op_fid1);
1896                 if (rc == 0)
1897                         goto repeat;
1898         }
1899         RETURN(rc);
1900 }
1901
1902 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1903                        void *ea, int ealen, void *ea2, int ea2len,
1904                        struct ptlrpc_request **request)
1905 {
1906         struct obd_device *obd = exp->exp_obd;
1907         struct lmv_obd *lmv = &obd->u.lmv;
1908         struct ptlrpc_request *req;
1909         struct obd_export *tgt_exp;
1910         struct lmv_obj *obj;
1911         int rc = 0, i;
1912         ENTRY;
1913
1914         rc = lmv_check_connect(obd);
1915         if (rc)
1916                 RETURN(rc);
1917
1918         obj = lmv_obj_grab(obd, &op_data->op_fid1);
1919
1920         CDEBUG(D_OTHER, "SETATTR for "DFID", valid 0x%x%s\n",
1921                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
1922                obj ? ", split" : "");
1923
1924         if (obj) {
1925                 for (i = 0; i < obj->lo_objcount; i++) {
1926                         op_data->op_fid1 = obj->lo_inodes[i].li_fid;
1927
1928                         tgt_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
1929                         if (IS_ERR(tgt_exp)) {
1930                                 rc = PTR_ERR(tgt_exp);
1931                                 break;
1932                         }
1933
1934                         rc = md_setattr(tgt_exp, op_data, ea, ealen,
1935                                         ea2, ea2len, &req);
1936
1937                         if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid)) {
1938                                 /*
1939                                  * this is master object and this request should
1940                                  * be returned back to llite.
1941                                  */
1942                                 *request = req;
1943                         } else {
1944                                 ptlrpc_req_finished(req);
1945                         }
1946
1947                         if (rc)
1948                                 break;
1949                 }
1950                 lmv_obj_put(obj);
1951         } else {
1952                 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
1953                 if (IS_ERR(tgt_exp))
1954                         RETURN(PTR_ERR(tgt_exp));
1955
1956                 rc = md_setattr(tgt_exp, op_data, ea, ealen, ea2,
1957                                 ea2len, request);
1958         }
1959         RETURN(rc);
1960 }
1961
1962 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
1963                     struct obd_capa *oc, struct ptlrpc_request **request)
1964 {
1965         struct obd_device *obd = exp->exp_obd;
1966         struct lmv_obd *lmv = &obd->u.lmv;
1967         struct obd_export *tgt_exp;
1968         int rc;
1969         ENTRY;
1970
1971         rc = lmv_check_connect(obd);
1972         if (rc)
1973                 RETURN(rc);
1974
1975         tgt_exp = lmv_find_export(lmv, fid);
1976         if (IS_ERR(tgt_exp))
1977                 RETURN(PTR_ERR(tgt_exp));
1978
1979         rc = md_sync(tgt_exp, fid, oc, request);
1980         RETURN(rc);
1981 }
1982
1983 /* main purpose of LMV blocking ast is to remove split directory LMV
1984  * presentation object (struct lmv_obj) attached to the lock being revoked. */
1985 int lmv_blocking_ast(struct ldlm_lock *lock,
1986                      struct ldlm_lock_desc *desc,
1987                      void *data, int flag)
1988 {
1989         struct lustre_handle lockh;
1990         struct lmv_obj *obj;
1991         int rc;
1992         ENTRY;
1993
1994         switch (flag) {
1995         case LDLM_CB_BLOCKING:
1996                 ldlm_lock2handle(lock, &lockh);
1997                 rc = ldlm_cli_cancel(&lockh);
1998                 if (rc < 0) {
1999                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2000                         RETURN(rc);
2001                 }
2002                 break;
2003         case LDLM_CB_CANCELING:
2004                 /* time to drop cached attrs for dirobj */
2005                 obj = lock->l_ast_data;
2006                 if (obj) {
2007                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
2008                                ", master "DFID"\n",
2009                                lock->l_resource->lr_name.name[3] == 1 ?
2010                                "LOOKUP" : "UPDATE",
2011                                lock->l_resource->lr_name.name[0],
2012                                lock->l_resource->lr_name.name[1],
2013                                PFID(&obj->lo_fid));
2014                         lmv_obj_put(obj);
2015                 }
2016                 break;
2017         default:
2018                 LBUG();
2019         }
2020         RETURN(0);
2021 }
2022
2023 static void lmv_hash_adjust(__u32 *hash, __u32 hash_adj)
2024 {
2025         __u32 val;
2026
2027         val = le32_to_cpu(*hash);
2028         if (val < hash_adj)
2029                 val += MAX_HASH_SIZE;
2030         if (val != DIR_END_OFF)
2031                 *hash = cpu_to_le32(val - hash_adj);
2032 }
2033
2034 static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
2035 {
2036         __u64 id;
2037         struct obd_import *imp;
2038
2039         /*
2040          * XXX Hack: to get nid we assume that underlying obd device is mdc.
2041          */
2042         imp  = class_exp2cliimp(exp);
2043         id   = imp->imp_connection->c_self + fid_flatten(fid);
2044
2045         CDEBUG(D_INFO, "node rank: %llx "DFID" %llx %llx\n",
2046                imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
2047
2048         return id ^ (id >> 32);
2049 }
2050
2051 static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
2052                         struct obd_capa *oc, __u64 offset64, struct page *page,
2053                         struct ptlrpc_request **request)
2054 {
2055         struct obd_device *obd = exp->exp_obd;
2056         struct lmv_obd *lmv = &obd->u.lmv;
2057         struct obd_export *tgt_exp;
2058         struct lu_fid rid = *fid;
2059         struct lmv_obj *obj;
2060         __u32 offset0;
2061         __u32 offset;
2062         __u32 hash_adj = 0;
2063         __u32 rank = 0;
2064         __u32 seg_size = 0;
2065         int tgt = 0;
2066         int tgt0 = 0;
2067         int rc;
2068         int nr = 0;
2069         ENTRY;
2070
2071         offset0 = offset = offset64;
2072         /*
2073          * Check that offset is representable by 32bit number.
2074          */
2075         LASSERT((__u64)offset == offset64);
2076
2077         rc = lmv_check_connect(obd);
2078         if (rc)
2079                 RETURN(rc);
2080
2081         CDEBUG(D_INFO, "READPAGE at %x from "DFID"\n", offset, PFID(&rid));
2082
2083         obj = lmv_obj_grab(obd, fid);
2084         if (obj) {
2085                 struct lmv_inode *loi;
2086
2087                 lmv_obj_lock(obj);
2088
2089                 nr       = obj->lo_objcount;
2090                 LASSERT(nr > 0);
2091                 seg_size = MAX_HASH_SIZE / nr;
2092                 loi      = obj->lo_inodes;
2093                 rank     = lmv_node_rank(lmv_get_export(lmv, loi[0].li_mds),
2094                                          fid) % nr;
2095                 tgt0     = (offset / seg_size) % nr;
2096                 tgt      = (tgt0 + rank) % nr;
2097
2098                 if (tgt < tgt0)
2099                         /*
2100                          * Wrap around.
2101                          *
2102                          * Last segment has unusual length due to division
2103                          * rounding.
2104                          */
2105                         hash_adj = MAX_HASH_SIZE - seg_size * nr;
2106                 else
2107                         hash_adj = 0;
2108
2109                 hash_adj += rank * seg_size;
2110
2111                 CDEBUG(D_INFO, "hash_adj: %x %x %x/%x -> %x/%x\n",
2112                        rank, hash_adj, offset, tgt0, offset + hash_adj, tgt);
2113
2114                 offset = (offset + hash_adj) % MAX_HASH_SIZE;
2115                 rid = obj->lo_inodes[tgt].li_fid;
2116                 tgt_exp = lmv_get_export(lmv, loi[tgt].li_mds);
2117
2118                 CDEBUG(D_INFO, "forward to "DFID" with offset %lu i %d\n",
2119                        PFID(&rid), (unsigned long)offset, tgt);
2120         } else
2121                 tgt_exp = lmv_find_export(lmv, &rid);
2122
2123         if (IS_ERR(tgt_exp))
2124                 GOTO(cleanup, rc = PTR_ERR(tgt_exp));
2125
2126         rc = md_readpage(tgt_exp, &rid, oc, offset, page, request);
2127         if (rc)
2128                 GOTO(cleanup, rc);
2129         if (obj) {
2130                 struct lu_dirpage *dp;
2131                 struct lu_dirent  *ent;
2132
2133                 dp = cfs_kmap(page);
2134
2135                 lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
2136                 lmv_hash_adjust(&dp->ldp_hash_end,   hash_adj);
2137                 LASSERT(cpu_to_le32(dp->ldp_hash_start) <= offset0);
2138
2139                 for (ent = lu_dirent_start(dp); ent != NULL;
2140                      ent = lu_dirent_next(ent))
2141                         lmv_hash_adjust(&ent->lde_hash, hash_adj);
2142
2143                 if (tgt0 != nr - 1) {
2144                         __u32 end;
2145
2146                         end = le32_to_cpu(dp->ldp_hash_end);
2147                         if (end == DIR_END_OFF) {
2148                                 dp->ldp_hash_end = cpu_to_le32(seg_size *
2149                                                                (tgt0 + 1));
2150                                 CDEBUG(D_INFO, ""DFID" reset end %x tgt %d\n",
2151                                        PFID(&rid),
2152                                        le32_to_cpu(dp->ldp_hash_end), tgt);
2153                         }
2154                 }
2155                 cfs_kunmap(page);
2156         }
2157         /*
2158          * Here we could remove "." and ".." from all pages which at not from
2159          * master. But MDS has only "." and ".." for master dir.
2160          */
2161         EXIT;
2162 cleanup:
2163         if (obj) {
2164                 lmv_obj_unlock(obj);
2165                 lmv_obj_put(obj);
2166         }
2167         return rc;
2168 }
2169
2170 static int lmv_unlink_slaves(struct obd_export *exp,
2171                              struct md_op_data *op_data,
2172                              struct ptlrpc_request **req)
2173 {
2174         struct obd_device *obd = exp->exp_obd;
2175         struct lmv_obd *lmv = &obd->u.lmv;
2176         struct lmv_stripe_md *mea = op_data->op_mea1;
2177         struct md_op_data *op_data2;
2178         struct obd_export *tgt_exp;
2179         int i, rc = 0;
2180         ENTRY;
2181
2182         OBD_ALLOC_PTR(op_data2);
2183         if (op_data2 == NULL)
2184                 RETURN(-ENOMEM);
2185
2186         op_data2->op_mode = S_IFDIR;
2187         op_data2->op_fsuid = current->fsuid;
2188         op_data2->op_fsgid = current->fsgid;
2189         op_data2->op_bias = 0;
2190
2191         LASSERT(mea != NULL);
2192         for (i = 0; i < mea->mea_count; i++) {
2193                 memset(op_data2, 0, sizeof(*op_data2));
2194                 op_data2->op_fid1 = mea->mea_ids[i];
2195                 tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
2196                 if (IS_ERR(tgt_exp))
2197                         GOTO(out_free_op_data2, rc = PTR_ERR(tgt_exp));
2198
2199                 if (tgt_exp == NULL)
2200                         continue;
2201
2202                 rc = md_unlink(tgt_exp, op_data2, req);
2203
2204                 CDEBUG(D_OTHER, "unlink slave "DFID" -> %d\n",
2205                        PFID(&mea->mea_ids[i]), rc);
2206
2207                 if (*req) {
2208                         ptlrpc_req_finished(*req);
2209                         *req = NULL;
2210                 }
2211                 if (rc)
2212                         GOTO(out_free_op_data2, rc);
2213         }
2214
2215         EXIT;
2216 out_free_op_data2:
2217         OBD_FREE_PTR(op_data2);
2218         return rc;
2219 }
2220
2221 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2222                       struct ptlrpc_request **request)
2223 {
2224         struct obd_device *obd = exp->exp_obd;
2225         struct lmv_obd *lmv = &obd->u.lmv;
2226         struct obd_export *tgt_exp = NULL;
2227         int rc, loop = 0;
2228         ENTRY;
2229
2230         rc = lmv_check_connect(obd);
2231         if (rc)
2232                 RETURN(rc);
2233
2234         if (op_data->op_namelen == 0 && op_data->op_mea1 != NULL) {
2235                 /* mds asks to remove slave objects */
2236                 rc = lmv_unlink_slaves(exp, op_data, request);
2237                 RETURN(rc);
2238         }
2239
2240 repeat:
2241         ++loop;
2242         LASSERT(loop <= 2);
2243         if (op_data->op_namelen != 0) {
2244                 struct lmv_obj *obj;
2245                 int mea_idx;
2246
2247                 obj = lmv_obj_grab(obd, &op_data->op_fid1);
2248                 if (obj) {
2249                         mea_idx = raw_name2idx(obj->lo_hashtype,
2250                                                obj->lo_objcount,
2251                                                op_data->op_name,
2252                                                op_data->op_namelen);
2253                         op_data->op_bias &= ~MDS_CHECK_SPLIT;
2254                         op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
2255                         tgt_exp = lmv_get_export(lmv,
2256                                                  obj->lo_inodes[mea_idx].li_mds);
2257                         lmv_obj_put(obj);
2258                         CDEBUG(D_OTHER, "unlink '%*s' in "DFID" -> %u\n",
2259                                op_data->op_namelen, op_data->op_name,
2260                                PFID(&op_data->op_fid1), mea_idx);
2261                 }
2262         } else {
2263                 CDEBUG(D_OTHER, "drop i_nlink on "DFID"\n",
2264                        PFID(&op_data->op_fid1));
2265         }
2266         if (tgt_exp == NULL) {
2267                 tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
2268                 if (IS_ERR(tgt_exp))
2269                         RETURN(PTR_ERR(tgt_exp));
2270                 op_data->op_bias |= MDS_CHECK_SPLIT;
2271         }
2272
2273         op_data->op_fsuid = current->fsuid;
2274         op_data->op_fsgid = current->fsgid;
2275         op_data->op_cap   = current->cap_effective;
2276
2277         rc = md_unlink(tgt_exp, op_data, request);
2278         if (rc == -ERESTART) {
2279                 LASSERT(*request != NULL);
2280                 DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
2281                           "Got -ERESTART during unlink!\n");
2282                 ptlrpc_req_finished(*request);
2283                 *request = NULL;
2284
2285                 /*
2286                  * Directory got split. Time to update local object and repeat
2287                  * the request with proper MDS.
2288                  */
2289                 rc = lmv_handle_split(exp, &op_data->op_fid1);
2290                 if (rc == 0)
2291                         goto repeat;
2292         }
2293         RETURN(rc);
2294 }
2295
2296 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs* llogs,
2297                          struct obd_device *tgt, int count,
2298                          struct llog_catid *logid, struct obd_uuid *uuid)
2299 {
2300         struct llog_ctxt *ctxt;
2301         int rc;
2302         ENTRY;
2303
2304         rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
2305                         &llog_client_ops);
2306         if (rc == 0) {
2307                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
2308                 ctxt->loc_imp = tgt->u.cli.cl_import;
2309         }
2310
2311         RETURN(rc);
2312 }
2313
2314 static int lmv_llog_finish(struct obd_device *obd, int count)
2315 {
2316         int rc;
2317         ENTRY;
2318
2319         rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
2320         RETURN(rc);
2321 }
2322
2323 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2324 {
2325         int rc = 0;
2326
2327         switch (stage) {
2328         case OBD_CLEANUP_EARLY:
2329                 /* XXX: here should be calling obd_precleanup() down to
2330                  * stack. */
2331                 break;
2332         case OBD_CLEANUP_SELF_EXP:
2333                 rc = obd_llog_finish(obd, 0);
2334                 if (rc != 0)
2335                         CERROR("failed to cleanup llogging subsystems\n");
2336                 break;
2337         default:
2338                 break;
2339         }
2340         RETURN(rc);
2341 }
2342
2343 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
2344                         void *key, __u32 *vallen, void *val)
2345 {
2346         struct obd_device *obd;
2347         struct lmv_obd *lmv;
2348         int rc = 0;
2349         ENTRY;
2350
2351         obd = class_exp2obd(exp);
2352         if (obd == NULL) {
2353                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2354                        exp->exp_handle.h_cookie);
2355                 RETURN(-EINVAL);
2356         }
2357
2358         lmv = &obd->u.lmv;
2359         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2360                 struct lmv_tgt_desc *tgts;
2361                 int i;
2362
2363                 rc = lmv_check_connect(obd);
2364                 if (rc)
2365                         RETURN(rc);
2366
2367                 LASSERT(*vallen == sizeof(__u32));
2368                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2369                      i++, tgts++) {
2370
2371                         /* all tgts should be connected when this get called. */
2372                         if (!tgts || !tgts->ltd_exp) {
2373                                 CERROR("target not setup?\n");
2374                                 continue;
2375                         }
2376
2377                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2378                                           vallen, val))
2379                                 RETURN(0);
2380                 }
2381                 RETURN(-EINVAL);
2382         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2383                 rc = lmv_check_connect(obd);
2384                 if (rc)
2385                         RETURN(rc);
2386
2387                 /* forwarding this request to first MDS, it should know LOV
2388                  * desc. */
2389                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2390                                   vallen, val);
2391                 if (!rc && KEY_IS(KEY_CONN_DATA)) {
2392                         exp->exp_connect_flags =
2393                         ((struct obd_connect_data *)val)->ocd_connect_flags;
2394                 }
2395                 RETURN(rc);
2396         }
2397
2398         CDEBUG(D_IOCTL, "invalid key\n");
2399         RETURN(-EINVAL);
2400 }
2401
2402 int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
2403                        void *key, obd_count vallen, void *val,
2404                        struct ptlrpc_request_set *set)
2405 {
2406         struct lmv_tgt_desc    *tgt;
2407         struct obd_device      *obd;
2408         struct lmv_obd         *lmv;
2409         int rc = 0;
2410         ENTRY;
2411
2412         obd = class_exp2obd(exp);
2413         if (obd == NULL) {
2414                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2415                        exp->exp_handle.h_cookie);
2416                 RETURN(-EINVAL);
2417         }
2418         lmv = &obd->u.lmv;
2419
2420         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2421             KEY_IS(KEY_INIT_RECOV_BACKUP)) {
2422                 int i, err = 0;
2423
2424                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2425                         tgt = &lmv->tgts[i];
2426
2427                         if (!tgt->ltd_exp)
2428                                 continue;
2429
2430                         err = obd_set_info_async(tgt->ltd_exp,
2431                                                  keylen, key, vallen, val, set);
2432                         if (err && rc == 0)
2433                                 rc = err;
2434                 }
2435
2436                 RETURN(rc);
2437         }
2438
2439         RETURN(-EINVAL);
2440 }
2441
2442 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2443                struct lov_stripe_md *lsm)
2444 {
2445         struct obd_device *obd = class_exp2obd(exp);
2446         struct lmv_obd *lmv = &obd->u.lmv;
2447         struct lmv_stripe_md *meap, *lsmp;
2448         int mea_size, i;
2449         ENTRY;
2450
2451         mea_size = lmv_get_easize(lmv);
2452         if (!lmmp)
2453                 RETURN(mea_size);
2454
2455         if (*lmmp && !lsm) {
2456                 OBD_FREE(*lmmp, mea_size);
2457                 *lmmp = NULL;
2458                 RETURN(0);
2459         }
2460
2461         if (*lmmp == NULL) {
2462                 OBD_ALLOC(*lmmp, mea_size);
2463                 if (*lmmp == NULL)
2464                         RETURN(-ENOMEM);
2465         }
2466
2467         if (!lsm)
2468                 RETURN(mea_size);
2469
2470         lsmp = (struct lmv_stripe_md *)lsm;
2471         meap = (struct lmv_stripe_md *)*lmmp;
2472
2473         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2474             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2475                 RETURN(-EINVAL);
2476
2477         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2478         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2479         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2480
2481         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2482                 meap->mea_ids[i] = meap->mea_ids[i];
2483                 fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
2484         }
2485
2486         RETURN(mea_size);
2487 }
2488
2489 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2490                  struct lov_mds_md *lmm, int lmm_size)
2491 {
2492         struct obd_device *obd = class_exp2obd(exp);
2493         struct lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp;
2494         struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm;
2495         struct lmv_obd *lmv = &obd->u.lmv;
2496         int mea_size, i;
2497         __u32 magic;
2498         ENTRY;
2499
2500         mea_size = lmv_get_easize(lmv);
2501         if (lsmp == NULL)
2502                 return mea_size;
2503
2504         if (*lsmp != NULL && lmm == NULL) {
2505                 OBD_FREE(*tmea, mea_size);
2506                 *lsmp = NULL;
2507                 RETURN(0);
2508         }
2509
2510         LASSERT(mea_size == lmm_size);
2511
2512         OBD_ALLOC(*tmea, mea_size);
2513         if (*tmea == NULL)
2514                 RETURN(-ENOMEM);
2515
2516         if (!lmm)
2517                 RETURN(mea_size);
2518
2519         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2520             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2521             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2522         {
2523                 magic = le32_to_cpu(mea->mea_magic);
2524         } else {
2525                 /* old mea is not handled here */
2526                 LBUG();
2527         }
2528
2529         (*tmea)->mea_magic = magic;
2530         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2531         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2532
2533         for (i = 0; i < (*tmea)->mea_count; i++) {
2534                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2535                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2536         }
2537         RETURN(mea_size);
2538 }
2539
2540 static int lmv_cancel_unused(struct obd_export *exp,
2541                              const struct lu_fid *fid,
2542                              int flags, void *opaque)
2543 {
2544         struct obd_device *obd = exp->exp_obd;
2545         struct lmv_obd *lmv = &obd->u.lmv;
2546         int rc = 0, err, i;
2547         ENTRY;
2548
2549         LASSERT(fid != NULL);
2550
2551         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2552                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
2553                         continue;
2554
2555                 err = md_cancel_unused(lmv->tgts[i].ltd_exp,
2556                                        fid, flags, opaque);
2557                 if (!rc)
2558                         rc = err;
2559         }
2560         RETURN(rc);
2561 }
2562
2563 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
2564 {
2565         struct obd_device *obd = exp->exp_obd;
2566         struct lmv_obd *lmv = &obd->u.lmv;
2567
2568         ENTRY;
2569         RETURN(md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data));
2570 }
2571
2572 int lmv_lock_match(struct obd_export *exp, int flags,
2573                    const struct lu_fid *fid, ldlm_type_t type,
2574                    ldlm_policy_data_t *policy, ldlm_mode_t mode,
2575                    struct lustre_handle *lockh)
2576 {
2577         struct obd_device *obd = exp->exp_obd;
2578         struct lmv_obd *lmv = &obd->u.lmv;
2579         int i, rc = 0;
2580         ENTRY;
2581
2582         CDEBUG(D_OTHER, "lock match for "DFID"\n", PFID(fid));
2583
2584         /* with CMD every object can have two locks in different namespaces:
2585          * lookup lock in space of mds storing direntry and update/open lock in
2586          * space of mds storing inode. Thus we check all targets, not only that
2587          * one fid was created in. */
2588         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2589                 rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
2590                                    type, policy, mode, lockh);
2591                 if (rc)
2592                         RETURN(1);
2593         }
2594
2595         RETURN(rc);
2596 }
2597
2598 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2599                       int offset, struct obd_export *dt_exp,
2600                       struct obd_export *md_exp, struct lustre_md *md)
2601 {
2602         struct obd_device *obd = exp->exp_obd;
2603         struct lmv_obd *lmv = &obd->u.lmv;
2604         int rc;
2605
2606         ENTRY;
2607         rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, offset, dt_exp, md_exp,
2608                               md);
2609         RETURN(rc);
2610 }
2611
2612 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2613 {
2614         struct obd_device *obd = exp->exp_obd;
2615         struct lmv_obd *lmv = &obd->u.lmv;
2616
2617         ENTRY;
2618         if (md->mea)
2619                 obd_free_memmd(exp, (struct lov_stripe_md**)&md->mea);
2620         RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
2621 }
2622
2623 int lmv_set_open_replay_data(struct obd_export *exp,
2624                              struct obd_client_handle *och,
2625                              struct ptlrpc_request *open_req)
2626 {
2627         struct obd_device *obd = exp->exp_obd;
2628         struct lmv_obd *lmv = &obd->u.lmv;
2629         struct obd_export *tgt_exp;
2630
2631         ENTRY;
2632
2633         tgt_exp = lmv_find_export(lmv, &och->och_fid);
2634         if (IS_ERR(tgt_exp))
2635                 RETURN(PTR_ERR(tgt_exp));
2636
2637         RETURN(md_set_open_replay_data(tgt_exp, och, open_req));
2638 }
2639
2640 int lmv_clear_open_replay_data(struct obd_export *exp,
2641                                struct obd_client_handle *och)
2642 {
2643         struct obd_device *obd = exp->exp_obd;
2644         struct lmv_obd *lmv = &obd->u.lmv;
2645         struct obd_export *tgt_exp;
2646         ENTRY;
2647
2648         tgt_exp = lmv_find_export(lmv, &och->och_fid);
2649         if (IS_ERR(tgt_exp))
2650                 RETURN(PTR_ERR(tgt_exp));
2651
2652         RETURN(md_clear_open_replay_data(tgt_exp, och));
2653 }
2654
2655 static int lmv_get_remote_perm(struct obd_export *exp,
2656                                const struct lu_fid *fid,
2657                                struct obd_capa *oc,
2658                                struct ptlrpc_request **request)
2659 {
2660         struct obd_device *obd = exp->exp_obd;
2661         struct lmv_obd *lmv = &obd->u.lmv;
2662         struct obd_export *tgt_exp;
2663         int rc;
2664
2665         ENTRY;
2666
2667         rc = lmv_check_connect(obd);
2668         if (rc)
2669                 RETURN(rc);
2670
2671         tgt_exp = lmv_find_export(lmv, fid);
2672         if (IS_ERR(tgt_exp))
2673                 RETURN(PTR_ERR(tgt_exp));
2674
2675         rc = md_get_remote_perm(tgt_exp, fid, oc, request);
2676
2677         RETURN(rc);
2678 }
2679
2680 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2681                           renew_capa_cb_t cb)
2682 {
2683         struct obd_device *obd = exp->exp_obd;
2684         struct lmv_obd *lmv = &obd->u.lmv;
2685         struct obd_export *tgt_exp;
2686         int rc;
2687         ENTRY;
2688
2689         rc = lmv_check_connect(obd);
2690         if (rc)
2691                 RETURN(rc);
2692
2693         tgt_exp = lmv_find_export(lmv, &oc->c_capa.lc_fid);
2694         if (IS_ERR(tgt_exp))
2695                 RETURN(PTR_ERR(tgt_exp));
2696
2697         rc = md_renew_capa(tgt_exp, oc, cb);
2698         RETURN(rc);
2699 }
2700
2701 struct obd_ops lmv_obd_ops = {
2702         .o_owner                = THIS_MODULE,
2703         .o_setup                = lmv_setup,
2704         .o_cleanup              = lmv_cleanup,
2705         .o_precleanup           = lmv_precleanup,
2706         .o_process_config       = lmv_process_config,
2707         .o_connect              = lmv_connect,
2708         .o_disconnect           = lmv_disconnect,
2709         .o_statfs               = lmv_statfs,
2710         .o_llog_init            = lmv_llog_init,
2711         .o_llog_finish          = lmv_llog_finish,
2712         .o_get_info             = lmv_get_info,
2713         .o_set_info_async       = lmv_set_info_async,
2714         .o_packmd               = lmv_packmd,
2715         .o_unpackmd             = lmv_unpackmd,
2716         .o_notify               = lmv_notify,
2717         .o_iocontrol            = lmv_iocontrol,
2718         .o_fid_delete           = lmv_fid_delete
2719 };
2720
2721 struct md_ops lmv_md_ops = {
2722         .m_getstatus            = lmv_getstatus,
2723         .m_change_cbdata        = lmv_change_cbdata,
2724         .m_close                = lmv_close,
2725         .m_create               = lmv_create,
2726         .m_done_writing         = lmv_done_writing,
2727         .m_enqueue              = lmv_enqueue,
2728         .m_getattr              = lmv_getattr,
2729         .m_getxattr             = lmv_getxattr,
2730         .m_getattr_name         = lmv_getattr_name,
2731         .m_intent_lock          = lmv_intent_lock,
2732         .m_link                 = lmv_link,
2733         .m_rename               = lmv_rename,
2734         .m_setattr              = lmv_setattr,
2735         .m_setxattr             = lmv_setxattr,
2736         .m_sync                 = lmv_sync,
2737         .m_readpage             = lmv_readpage,
2738         .m_unlink               = lmv_unlink,
2739         .m_init_ea_size         = lmv_init_ea_size,
2740         .m_cancel_unused        = lmv_cancel_unused,
2741         .m_set_lock_data        = lmv_set_lock_data,
2742         .m_lock_match           = lmv_lock_match,
2743         .m_get_lustre_md        = lmv_get_lustre_md,
2744         .m_free_lustre_md       = lmv_free_lustre_md,
2745         .m_set_open_replay_data = lmv_set_open_replay_data,
2746         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2747         .m_get_remote_perm      = lmv_get_remote_perm,
2748         .m_renew_capa           = lmv_renew_capa
2749 };
2750
2751 int __init lmv_init(void)
2752 {
2753         struct lprocfs_static_vars lvars;
2754         int rc;
2755
2756         obj_cache = cfs_mem_cache_create("lmv_objects",
2757                                       sizeof(struct lmv_obj),
2758                                       0, 0);
2759         if (!obj_cache) {
2760                 CERROR("error allocating lmv objects cache\n");
2761                 return -ENOMEM;
2762         }
2763
2764         lprocfs_init_vars(lmv, &lvars);
2765         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2766                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2767         if (rc)
2768                 cfs_mem_cache_destroy(obj_cache);
2769
2770         return rc;
2771 }
2772
2773 #ifdef __KERNEL__
2774 static void lmv_exit(void)
2775 {
2776         int rc;
2777
2778         class_unregister_type(LUSTRE_LMV_NAME);
2779
2780         rc = cfs_mem_cache_destroy(obj_cache);
2781         LASSERTF(rc == 0,
2782                  "can't free lmv objects cache, %d object(s)"
2783                  "still in use\n", atomic_read(&obj_cache_count));
2784 }
2785
2786 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2787 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2788 MODULE_LICENSE("GPL");
2789
2790 module_init(lmv_init);
2791 module_exit(lmv_exit);
2792 #endif