lustre/lov/lov_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LOV
  45 #ifdef __KERNEL__
  46 #include <libcfs/libcfs.h>
  47 #else
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <obd_support.h>
  52 #include <lustre_lib.h>
  53 #include <lustre_net.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_mds.h>
  57 #include <lustre_debug.h>
  58 #include <obd_class.h>
  59 #include <obd_lov.h>
  60 #include <obd_ost.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_param.h>
  63 #include <cl_object.h>
  64 #include <lustre/ll_fiemap.h>
  65 #include <lustre_log.h>
  66
  67 #include "lov_internal.h"
  68
  69 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  70    Any function that expects lov_tgts to remain stationary must take a ref. */
  71 static void lov_getref(struct obd_device *obd)
  72 {
  73         struct lov_obd *lov = &obd->u.lov;
  74
  75         /* nobody gets through here until lov_putref is done */
  76         cfs_mutex_lock(&lov->lov_lock);
  77         cfs_atomic_inc(&lov->lov_refcount);
  78         cfs_mutex_unlock(&lov->lov_lock);
  79         return;
  80 }
  81
  82 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  83
  84 static void lov_putref(struct obd_device *obd)
  85 {
  86         struct lov_obd *lov = &obd->u.lov;
  87
  88         cfs_mutex_lock(&lov->lov_lock);
  89         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  90         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  91                 CFS_LIST_HEAD(kill);
  92                 int i;
  93                 struct lov_tgt_desc *tgt, *n;
  94                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  95                        lov->lov_death_row);
  96                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  97                         tgt = lov->lov_tgts[i];
  98
  99                         if (!tgt || !tgt->ltd_reap)
 100                                 continue;
 101                         cfs_list_add(&tgt->ltd_kill, &kill);
 102                         /* XXX - right now there is a dependency on ld_tgt_count
 103                          * being the maximum tgt index for computing the
 104                          * mds_max_easize. So we can't shrink it. */
 105                         lov_ost_pool_remove(&lov->lov_packed, i);
 106                         lov->lov_tgts[i] = NULL;
 107                         lov->lov_death_row--;
 108                 }
 109                 cfs_mutex_unlock(&lov->lov_lock);
 110
 111                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 112                         cfs_list_del(&tgt->ltd_kill);
 113                         /* Disconnect */
 114                         __lov_del_obd(obd, tgt);
 115                 }
 116         } else {
 117                 cfs_mutex_unlock(&lov->lov_lock);
 118         }
 119 }
 120
 121 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 122                               enum obd_notify_event ev);
 123 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 124                       enum obd_notify_event ev, void *data);
 125
 126
 127 #define MAX_STRING_SIZE 128
 128 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 129                     struct obd_connect_data *data)
 130 {
 131         struct lov_obd *lov = &obd->u.lov;
 132         struct obd_uuid *tgt_uuid;
 133         struct obd_device *tgt_obd;
 134         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 135         struct obd_import *imp;
 136 #ifdef __KERNEL__
 137         cfs_proc_dir_entry_t *lov_proc_dir;
 138 #endif
 139         int rc;
 140         ENTRY;
 141
 142         if (!lov->lov_tgts[index])
 143                 RETURN(-EINVAL);
 144
 145         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 146         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 147
 148         if (!tgt_obd->obd_set_up) {
 149                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 150                 RETURN(-EINVAL);
 151         }
 152
 153         /* override the sp_me from lov */
 154         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 155
 156         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 157                 data->ocd_index = index;
 158
 159         /*
 160          * Divine LOV knows that OBDs under it are OSCs.
 161          */
 162         imp = tgt_obd->u.cli.cl_import;
 163
 164         if (activate) {
 165                 tgt_obd->obd_no_recov = 0;
 166                 /* FIXME this is probably supposed to be
 167                    ptlrpc_set_import_active.  Horrible naming. */
 168                 ptlrpc_activate_import(imp);
 169         }
 170
 171         rc = obd_register_observer(tgt_obd, obd);
 172         if (rc) {
 173                 CERROR("Target %s register_observer error %d\n",
 174                        obd_uuid2str(tgt_uuid), rc);
 175                 RETURN(rc);
 176         }
 177
 178
 179         if (imp->imp_invalid) {
 180                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
 181                        "disabled\n", obd_uuid2str(tgt_uuid));
 182                 RETURN(0);
 183         }
 184
 185         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 186                          &lov_osc_uuid, data, NULL);
 187         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 188                 CERROR("Target %s connect error %d\n",
 189                        obd_uuid2str(tgt_uuid), rc);
 190                 RETURN(-ENODEV);
 191         }
 192
 193         lov->lov_tgts[index]->ltd_reap = 0;
 194
 195         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 196                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 197
 198 #ifdef __KERNEL__
 199         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 200         if (lov_proc_dir) {
 201                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 202                 cfs_proc_dir_entry_t *osc_symlink;
 203
 204                 LASSERT(osc_obd != NULL);
 205                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 206                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 207
 208                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
 209                                                   lov_proc_dir,
 210                                                   "../../../%s/%s",
 211                                                   osc_obd->obd_type->typ_name,
 212                                                   osc_obd->obd_name);
 213                 if (osc_symlink == NULL) {
 214                         CERROR("could not register LOV target "
 215                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
 216                                 obd->obd_type->typ_name, obd->obd_name,
 217                                 osc_obd->obd_name);
 218                         lprocfs_remove(&lov_proc_dir);
 219                 }
 220         }
 221 #endif
 222
 223         rc = qos_add_tgt(obd, index);
 224         if (rc)
 225                 CERROR("qos_add_tgt failed %d\n", rc);
 226
 227         RETURN(0);
 228 }
 229
 230 static int lov_connect(const struct lu_env *env,
 231                        struct obd_export **exp, struct obd_device *obd,
 232                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 233                        void *localdata)
 234 {
 235         struct lov_obd *lov = &obd->u.lov;
 236         struct lov_tgt_desc *tgt;
 237         struct lustre_handle conn;
 238         int i, rc;
 239         ENTRY;
 240
 241         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 242
 243         rc = class_connect(&conn, obd, cluuid);
 244         if (rc)
 245                 RETURN(rc);
 246
 247         *exp = class_conn2export(&conn);
 248
 249         /* Why should there ever be more than 1 connect? */
 250         lov->lov_connects++;
 251         LASSERT(lov->lov_connects == 1);
 252
 253         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 254         if (data)
 255                 lov->lov_ocd = *data;
 256
 257         obd_getref(obd);
 258         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 259                 tgt = lov->lov_tgts[i];
 260                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 261                         continue;
 262                 /* Flags will be lowest common denominator */
 263                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 264                 if (rc) {
 265                         CERROR("%s: lov connect tgt %d failed: %d\n",
 266                                obd->obd_name, i, rc);
 267                         continue;
 268                 }
 269                 /* connect to administrative disabled ost */
 270                 if (!lov->lov_tgts[i]->ltd_exp)
 271                         continue;
 272
 273                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 274                                 OBD_NOTIFY_CONNECT, (void *)&i);
 275                 if (rc) {
 276                         CERROR("%s error sending notify %d\n",
 277                                obd->obd_name, rc);
 278                 }
 279         }
 280         obd_putref(obd);
 281
 282         RETURN(0);
 283 }
 284
 285 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 286 {
 287         cfs_proc_dir_entry_t *lov_proc_dir;
 288         struct lov_obd *lov = &obd->u.lov;
 289         struct obd_device *osc_obd;
 290         int rc;
 291         ENTRY;
 292
 293         osc_obd = class_exp2obd(tgt->ltd_exp);
 294         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 295                obd->obd_name, osc_obd->obd_name);
 296
 297         if (tgt->ltd_active) {
 298                 tgt->ltd_active = 0;
 299                 lov->desc.ld_active_tgt_count--;
 300                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 301         }
 302
 303         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 304         if (lov_proc_dir) {
 305                 cfs_proc_dir_entry_t *osc_symlink;
 306
 307                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 308                 if (osc_symlink) {
 309                         lprocfs_remove(&osc_symlink);
 310                 } else {
 311                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 312                                obd->obd_type->typ_name, obd->obd_name,
 313                                osc_obd->obd_name);
 314                 }
 315         }
 316
 317         if (osc_obd) {
 318                 /* Pass it on to our clients.
 319                  * XXX This should be an argument to disconnect,
 320                  * XXX not a back-door flag on the OBD.  Ah well.
 321                  */
 322                 osc_obd->obd_force = obd->obd_force;
 323                 osc_obd->obd_fail = obd->obd_fail;
 324                 osc_obd->obd_no_recov = obd->obd_no_recov;
 325         }
 326
 327         obd_register_observer(osc_obd, NULL);
 328
 329         rc = obd_disconnect(tgt->ltd_exp);
 330         if (rc) {
 331                 CERROR("Target %s disconnect error %d\n",
 332                        tgt->ltd_uuid.uuid, rc);
 333                 rc = 0;
 334         }
 335
 336         qos_del_tgt(obd, tgt);
 337
 338         tgt->ltd_exp = NULL;
 339         RETURN(0);
 340 }
 341
 342 static int lov_disconnect(struct obd_export *exp)
 343 {
 344         struct obd_device *obd = class_exp2obd(exp);
 345         struct lov_obd *lov = &obd->u.lov;
 346         int i, rc;
 347         ENTRY;
 348
 349         if (!lov->lov_tgts)
 350                 goto out;
 351
 352         /* Only disconnect the underlying layers on the final disconnect. */
 353         lov->lov_connects--;
 354         if (lov->lov_connects != 0) {
 355                 /* why should there be more than 1 connect? */
 356                 CERROR("disconnect #%d\n", lov->lov_connects);
 357                 goto out;
 358         }
 359
 360         /* Let's hold another reference so lov_del_obd doesn't spin through
 361            putref every time */
 362         obd_getref(obd);
 363
 364         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 365                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 366                         /* Disconnection is the last we know about an obd */
 367                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 368                 }
 369         }
 370         obd_putref(obd);
 371
 372 out:
 373         rc = class_disconnect(exp); /* bz 9811 */
 374         RETURN(rc);
 375 }
 376
 377 /* Error codes:
 378  *
 379  *  -EINVAL  : UUID can't be found in the LOV's target list
 380  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 381  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 382  *  any >= 0 : is log target index
 383  */
 384 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 385                               enum obd_notify_event ev)
 386 {
 387         struct lov_obd *lov = &obd->u.lov;
 388         struct lov_tgt_desc *tgt;
 389         int index, activate, active;
 390         ENTRY;
 391
 392         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 393                lov, uuid->uuid, ev);
 394
 395         obd_getref(obd);
 396         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 397                 tgt = lov->lov_tgts[index];
 398                 if (!tgt || !tgt->ltd_exp)
 399                         continue;
 400
 401                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 402                        index, obd_uuid2str(&tgt->ltd_uuid),
 403                        tgt->ltd_exp->exp_handle.h_cookie);
 404                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 405                         break;
 406         }
 407
 408         if (index == lov->desc.ld_tgt_count)
 409                 GOTO(out, index = -EINVAL);
 410
 411         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 412                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 413
 414                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 415                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 416                                uuid->uuid, activate ? "" : "de");
 417                 } else {
 418                         lov->lov_tgts[index]->ltd_activate = activate;
 419                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 420                                activate ? "" : "de", obd_uuid2str(uuid));
 421                 }
 422
 423         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 424                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 425
 426                 if (lov->lov_tgts[index]->ltd_active == active) {
 427                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 428                                uuid->uuid, active ? "" : "in");
 429                         GOTO(out, index);
 430                 } else {
 431                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 432                                obd_uuid2str(uuid), active ? "" : "in");
 433                 }
 434
 435                 lov->lov_tgts[index]->ltd_active = active;
 436                 if (active) {
 437                         lov->desc.ld_active_tgt_count++;
 438                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 439                 } else {
 440                         lov->desc.ld_active_tgt_count--;
 441                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 442                 }
 443                 /* remove any old qos penalty */
 444                 lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 445         } else {
 446                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 447         }
 448
 449  out:
 450         obd_putref(obd);
 451         RETURN(index);
 452 }
 453
 454 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 455                       enum obd_notify_event ev, void *data)
 456 {
 457         int rc = 0;
 458         ENTRY;
 459
 460         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 461             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 462                 struct obd_uuid *uuid;
 463
 464                 LASSERT(watched);
 465
 466                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 467                         CERROR("unexpected notification of %s %s!\n",
 468                                watched->obd_type->typ_name,
 469                                watched->obd_name);
 470                         RETURN(-EINVAL);
 471                 }
 472                 uuid = &watched->u.cli.cl_target_uuid;
 473
 474                 /* Set OSC as active before notifying the observer, so the
 475                  * observer can use the OSC normally.
 476                  */
 477                 rc = lov_set_osc_active(obd, uuid, ev);
 478                 if (rc < 0) {
 479                         CERROR("event(%d) of %s failed: %d\n", ev,
 480                                obd_uuid2str(uuid), rc);
 481                         RETURN(rc);
 482                 }
 483                 /* active event should be pass lov target index as data */
 484                 data = &rc;
 485         }
 486
 487         /* Pass the notification up the chain. */
 488         if (watched) {
 489                 rc = obd_notify_observer(obd, watched, ev, data);
 490         } else {
 491                 /* NULL watched means all osc's in the lov (only for syncs) */
 492                 /* sync event should be send lov idx as data */
 493                 struct lov_obd *lov = &obd->u.lov;
 494                 int i, is_sync;
 495
 496                 data = &i;
 497                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 498                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 499
 500                 obd_getref(obd);
 501                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 502                         if (!lov->lov_tgts[i])
 503                                 continue;
 504
 505                         /* don't send sync event if target not
 506                          * connected/activated */
 507                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 508                                 continue;
 509
 510                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 511                                                  ev, data);
 512                         if (rc) {
 513                                 CERROR("%s: notify %s of %s failed %d\n",
 514                                        obd->obd_name,
 515                                        obd->obd_observer->obd_name,
 516                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 517                                        rc);
 518                         }
 519                 }
 520                 obd_putref(obd);
 521         }
 522
 523         RETURN(rc);
 524 }
 525
 526 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 527                           __u32 index, int gen, int active)
 528 {
 529         struct lov_obd *lov = &obd->u.lov;
 530         struct lov_tgt_desc *tgt;
 531         struct obd_device *tgt_obd;
 532         int rc;
 533         ENTRY;
 534
 535         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 536                uuidp->uuid, index, gen, active);
 537
 538         if (gen <= 0) {
 539                 CERROR("request to add OBD %s with invalid generation: %d\n",
 540                        uuidp->uuid, gen);
 541                 RETURN(-EINVAL);
 542         }
 543
 544         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 545                                         &obd->obd_uuid);
 546         if (tgt_obd == NULL)
 547                 RETURN(-EINVAL);
 548
 549         cfs_mutex_lock(&lov->lov_lock);
 550
 551         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 552                 tgt = lov->lov_tgts[index];
 553                 CERROR("UUID %s already assigned at LOV target index %d\n",
 554                        obd_uuid2str(&tgt->ltd_uuid), index);
 555                 cfs_mutex_unlock(&lov->lov_lock);
 556                 RETURN(-EEXIST);
 557         }
 558
 559         if (index >= lov->lov_tgt_size) {
 560                 /* We need to reallocate the lov target array. */
 561                 struct lov_tgt_desc **newtgts, **old = NULL;
 562                 __u32 newsize, oldsize = 0;
 563
 564                 newsize = max(lov->lov_tgt_size, (__u32)2);
 565                 while (newsize < index + 1)
 566                         newsize = newsize << 1;
 567                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 568                 if (newtgts == NULL) {
 569                         cfs_mutex_unlock(&lov->lov_lock);
 570                         RETURN(-ENOMEM);
 571                 }
 572
 573                 if (lov->lov_tgt_size) {
 574                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 575                                lov->lov_tgt_size);
 576                         old = lov->lov_tgts;
 577                         oldsize = lov->lov_tgt_size;
 578                 }
 579
 580                 lov->lov_tgts = newtgts;
 581                 lov->lov_tgt_size = newsize;
 582 #ifdef __KERNEL__
 583                 smp_rmb();
 584 #endif
 585                 if (old)
 586                         OBD_FREE(old, sizeof(*old) * oldsize);
 587
 588                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 589                        lov->lov_tgts, lov->lov_tgt_size);
 590         }
 591
 592         OBD_ALLOC_PTR(tgt);
 593         if (!tgt) {
 594                 cfs_mutex_unlock(&lov->lov_lock);
 595                 RETURN(-ENOMEM);
 596         }
 597
 598         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 599         if (rc) {
 600                 cfs_mutex_unlock(&lov->lov_lock);
 601                 OBD_FREE_PTR(tgt);
 602                 RETURN(rc);
 603         }
 604
 605         tgt->ltd_uuid = *uuidp;
 606         tgt->ltd_obd = tgt_obd;
 607         /* XXX - add a sanity check on the generation number. */
 608         tgt->ltd_gen = gen;
 609         tgt->ltd_index = index;
 610         tgt->ltd_activate = active;
 611         lov->lov_tgts[index] = tgt;
 612         if (index >= lov->desc.ld_tgt_count)
 613                 lov->desc.ld_tgt_count = index + 1;
 614
 615         cfs_mutex_unlock(&lov->lov_lock);
 616
 617         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 618                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 619
 620         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 621
 622         if (lov->lov_connects == 0) {
 623                 /* lov_connect hasn't been called yet. We'll do the
 624                    lov_connect_obd on this target when that fn first runs,
 625                    because we don't know the connect flags yet. */
 626                 RETURN(0);
 627         }
 628
 629         obd_getref(obd);
 630
 631         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 632         if (rc)
 633                 GOTO(out, rc);
 634
 635         /* connect to administrative disabled ost */
 636         if (!tgt->ltd_exp)
 637                 GOTO(out, rc = 0);
 638
 639         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 640                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 641                         (void *)&index);
 642
 643 out:
 644         if (rc) {
 645                 CERROR("add failed (%d), deleting %s\n", rc,
 646                        obd_uuid2str(&tgt->ltd_uuid));
 647                 lov_del_target(obd, index, 0, 0);
 648         }
 649         obd_putref(obd);
 650         RETURN(rc);
 651 }
 652
 653 /* Schedule a target for deletion */
 654 int lov_del_target(struct obd_device *obd, __u32 index,
 655                    struct obd_uuid *uuidp, int gen)
 656 {
 657         struct lov_obd *lov = &obd->u.lov;
 658         int count = lov->desc.ld_tgt_count;
 659         int rc = 0;
 660         ENTRY;
 661
 662         if (index >= count) {
 663                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 664                        index, count);
 665                 RETURN(-EINVAL);
 666         }
 667
 668         obd_getref(obd);
 669
 670         if (!lov->lov_tgts[index]) {
 671                 CERROR("LOV target at index %d is not setup.\n", index);
 672                 GOTO(out, rc = -EINVAL);
 673         }
 674
 675         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 676                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 677                        lov_uuid2str(lov, index), index,
 678                        obd_uuid2str(uuidp));
 679                 GOTO(out, rc = -EINVAL);
 680         }
 681
 682         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 683                lov_uuid2str(lov, index), index,
 684                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 685                lov->lov_tgts[index]->ltd_active);
 686
 687         lov->lov_tgts[index]->ltd_reap = 1;
 688         lov->lov_death_row++;
 689         /* we really delete it from obd_putref */
 690 out:
 691         obd_putref(obd);
 692
 693         RETURN(rc);
 694 }
 695
 696 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 697 {
 698         struct obd_device *osc_obd;
 699
 700         LASSERT(tgt);
 701         LASSERT(tgt->ltd_reap);
 702
 703         osc_obd = class_exp2obd(tgt->ltd_exp);
 704
 705         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 706                tgt->ltd_uuid.uuid,
 707                osc_obd ? osc_obd->obd_name : "<no obd>");
 708
 709         if (tgt->ltd_exp)
 710                 lov_disconnect_obd(obd, tgt);
 711
 712         OBD_FREE_PTR(tgt);
 713
 714         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 715            do it ourselves. And we can't do it from lov_cleanup,
 716            because we just lost our only reference to it. */
 717         if (osc_obd)
 718                 class_manual_cleanup(osc_obd);
 719 }
 720
 721 void lov_fix_desc_stripe_size(__u64 *val)
 722 {
 723         if (*val < PTLRPC_MAX_BRW_SIZE) {
 724                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 725                               PTLRPC_MAX_BRW_SIZE);
 726                 *val = PTLRPC_MAX_BRW_SIZE;
 727         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 728                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 729                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 730                               "multiple of %u)\n",
 731                               *val, LOV_MIN_STRIPE_SIZE);
 732         }
 733 }
 734
 735 void lov_fix_desc_stripe_count(__u32 *val)
 736 {
 737         if (*val == 0)
 738                 *val = 1;
 739 }
 740
 741 void lov_fix_desc_pattern(__u32 *val)
 742 {
 743         /* from lov_setstripe */
 744         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 745                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 746                 *val = 0;
 747         }
 748 }
 749
 750 void lov_fix_desc_qos_maxage(__u32 *val)
 751 {
 752         /* fix qos_maxage */
 753         if (*val == 0)
 754                 *val = QOS_DEFAULT_MAXAGE;
 755 }
 756
 757 void lov_fix_desc(struct lov_desc *desc)
 758 {
 759         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 760         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 761         lov_fix_desc_pattern(&desc->ld_pattern);
 762         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 763 }
 764
 765 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 766 {
 767         struct lprocfs_static_vars lvars = { 0 };
 768         struct lov_desc *desc;
 769         struct lov_obd *lov = &obd->u.lov;
 770         int rc;
 771         ENTRY;
 772
 773         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 774                 CERROR("LOV setup requires a descriptor\n");
 775                 RETURN(-EINVAL);
 776         }
 777
 778         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 779
 780         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 781                 CERROR("descriptor size wrong: %d > %d\n",
 782                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 783                 RETURN(-EINVAL);
 784         }
 785
 786         if (desc->ld_magic != LOV_DESC_MAGIC) {
 787                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 788                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 789                                    obd->obd_name, desc);
 790                             lustre_swab_lov_desc(desc);
 791                 } else {
 792                         CERROR("%s: Bad lov desc magic: %#x\n",
 793                                obd->obd_name, desc->ld_magic);
 794                         RETURN(-EINVAL);
 795                 }
 796         }
 797
 798         lov_fix_desc(desc);
 799
 800         desc->ld_active_tgt_count = 0;
 801         lov->desc = *desc;
 802         lov->lov_tgt_size = 0;
 803
 804         cfs_mutex_init(&lov->lov_lock);
 805         cfs_atomic_set(&lov->lov_refcount, 0);
 806         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 807         cfs_init_rwsem(&lov->lov_qos.lq_rw_sem);
 808         lov->lov_sp_me = LUSTRE_SP_CLI;
 809         lov->lov_qos.lq_dirty = 1;
 810         lov->lov_qos.lq_rr.lqr_dirty = 1;
 811         lov->lov_qos.lq_reset = 1;
 812         /* Default priority is toward free space balance */
 813         lov->lov_qos.lq_prio_free = 232;
 814         /* Default threshold for rr (roughly 17%) */
 815         lov->lov_qos.lq_threshold_rr = 43;
 816         /* Init statfs fields */
 817         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 818         if (NULL == lov->lov_qos.lq_statfs_data)
 819                 RETURN(-ENOMEM);
 820         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 821
 822         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 823                                                    HASH_POOLS_MAX_BITS,
 824                                                    HASH_POOLS_BKT_BITS, 0,
 825                                                    CFS_HASH_MIN_THETA,
 826                                                    CFS_HASH_MAX_THETA,
 827                                                    &pool_hash_operations,
 828                                                    CFS_HASH_DEFAULT);
 829         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 830         lov->lov_pool_count = 0;
 831         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 832         if (rc)
 833                 GOTO(out_free_statfs, rc);
 834         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 835         if (rc)
 836                 GOTO(out_free_lov_packed, rc);
 837
 838         lprocfs_lov_init_vars(&lvars);
 839         lprocfs_obd_setup(obd, lvars.obd_vars);
 840 #ifdef LPROCFS
 841         {
 842                 int rc;
 843
 844                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 845                                         0444, &lov_proc_target_fops, obd);
 846                 if (rc)
 847                         CWARN("Error adding the target_obd file\n");
 848         }
 849 #endif
 850         lov->lov_pool_proc_entry = lprocfs_register("pools",
 851                                                     obd->obd_proc_entry,
 852                                                     NULL, NULL);
 853
 854         RETURN(0);
 855
 856 out_free_lov_packed:
 857         lov_ost_pool_free(&lov->lov_packed);
 858 out_free_statfs:
 859         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 860         return rc;
 861 }
 862
 863 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 864 {
 865         int rc = 0;
 866         struct lov_obd *lov = &obd->u.lov;
 867
 868         ENTRY;
 869
 870         switch (stage) {
 871         case OBD_CLEANUP_EARLY: {
 872                 int i;
 873                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 874                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 875                                 continue;
 876                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 877                                        OBD_CLEANUP_EARLY);
 878                 }
 879                 break;
 880         }
 881         case OBD_CLEANUP_EXPORTS:
 882                 rc = obd_llog_finish(obd, 0);
 883                 if (rc != 0)
 884                         CERROR("failed to cleanup llogging subsystems\n");
 885                 break;
 886         }
 887         RETURN(rc);
 888 }
 889
 890 static int lov_cleanup(struct obd_device *obd)
 891 {
 892         struct lov_obd *lov = &obd->u.lov;
 893         cfs_list_t *pos, *tmp;
 894         struct pool_desc *pool;
 895         ENTRY;
 896
 897         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 898                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 899                 /* free pool structs */
 900                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 901                 /* In the function below, .hs_keycmp resolves to
 902                  * pool_hashkey_keycmp() */
 903                 /* coverity[overrun-buffer-val] */
 904                 lov_pool_del(obd, pool->pool_name);
 905         }
 906         cfs_hash_putref(lov->lov_pools_hash_body);
 907         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 908         lov_ost_pool_free(&lov->lov_packed);
 909
 910         lprocfs_obd_cleanup(obd);
 911         if (lov->lov_tgts) {
 912                 int i;
 913                 obd_getref(obd);
 914                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 915                         if (!lov->lov_tgts[i])
 916                                 continue;
 917
 918                         /* Inactive targets may never have connected */
 919                         if (lov->lov_tgts[i]->ltd_active ||
 920                             cfs_atomic_read(&lov->lov_refcount))
 921                             /* We should never get here - these
 922                                should have been removed in the
 923                              disconnect. */
 924                                 CERROR("lov tgt %d not cleaned!"
 925                                        " deathrow=%d, lovrc=%d\n",
 926                                        i, lov->lov_death_row,
 927                                        cfs_atomic_read(&lov->lov_refcount));
 928                         lov_del_target(obd, i, 0, 0);
 929                 }
 930                 obd_putref(obd);
 931                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 932                          lov->lov_tgt_size);
 933                 lov->lov_tgt_size = 0;
 934         }
 935         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
 936         RETURN(0);
 937 }
 938
 939 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 940                             __u32 *indexp, int *genp)
 941 {
 942         struct obd_uuid obd_uuid;
 943         int cmd;
 944         int rc = 0;
 945         ENTRY;
 946
 947         switch(cmd = lcfg->lcfg_command) {
 948         case LCFG_LOV_ADD_OBD:
 949         case LCFG_LOV_ADD_INA:
 950         case LCFG_LOV_DEL_OBD: {
 951                 __u32 index;
 952                 int gen;
 953                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 954                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 955                         GOTO(out, rc = -EINVAL);
 956
 957                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 958
 959                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 960                         GOTO(out, rc = -EINVAL);
 961                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 962                         GOTO(out, rc = -EINVAL);
 963                 index = *indexp;
 964                 gen = *genp;
 965                 if (cmd == LCFG_LOV_ADD_OBD)
 966                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 967                 else if (cmd == LCFG_LOV_ADD_INA)
 968                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 969                 else
 970                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 971                 GOTO(out, rc);
 972         }
 973         case LCFG_PARAM: {
 974                 struct lprocfs_static_vars lvars = { 0 };
 975                 struct lov_desc *desc = &(obd->u.lov.desc);
 976
 977                 if (!desc)
 978                         GOTO(out, rc = -EINVAL);
 979
 980                 lprocfs_lov_init_vars(&lvars);
 981
 982                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 983                                               lcfg, obd);
 984                 if (rc > 0)
 985                         rc = 0;
 986                 GOTO(out, rc);
 987         }
 988         case LCFG_POOL_NEW:
 989         case LCFG_POOL_ADD:
 990         case LCFG_POOL_DEL:
 991         case LCFG_POOL_REM:
 992                 GOTO(out, rc);
 993
 994         default: {
 995                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 996                 GOTO(out, rc = -EINVAL);
 997
 998         }
 999         }
1000 out:
1001         RETURN(rc);
1002 }
1003
1004 #ifndef log2
1005 #define log2(n) cfs_ffz(~(n))
1006 #endif
1007
1008 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1009                              struct lov_stripe_md **ea,
1010                              struct obd_trans_info *oti)
1011 {
1012         struct lov_obd *lov;
1013         struct obdo *tmp_oa;
1014         struct obd_uuid *ost_uuid = NULL;
1015         int rc = 0, i;
1016         ENTRY;
1017
1018         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1019                 src_oa->o_flags == OBD_FL_DELORPHAN);
1020
1021         lov = &export->exp_obd->u.lov;
1022
1023         OBDO_ALLOC(tmp_oa);
1024         if (tmp_oa == NULL)
1025                 RETURN(-ENOMEM);
1026
1027         if (oti->oti_ost_uuid) {
1028                 ost_uuid = oti->oti_ost_uuid;
1029                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1030                        ost_uuid->uuid);
1031         }
1032
1033         obd_getref(export->exp_obd);
1034         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1035                 struct lov_stripe_md obj_md;
1036                 struct lov_stripe_md *obj_mdp = &obj_md;
1037                 struct lov_tgt_desc *tgt;
1038                 int err;
1039
1040                 tgt = lov->lov_tgts[i];
1041                 if (!tgt)
1042                         continue;
1043
1044                 /* if called for a specific target, we don't
1045                    care if it is not active. */
1046                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1047                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1048                         continue;
1049                 }
1050
1051                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1052                         continue;
1053
1054                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1055                        obd_uuid2str(ost_uuid));
1056
1057                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1058
1059                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1060                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1061                 err = obd_create(NULL, lov->lov_tgts[i]->ltd_exp,
1062                                  tmp_oa, &obj_mdp, oti);
1063                 if (err) {
1064                         /* This export will be disabled until it is recovered,
1065                            and then orphan recovery will be completed. */
1066                         CERROR("error in orphan recovery on OST idx %d/%d: "
1067                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1068                         rc = err;
1069                 }
1070
1071                 if (ost_uuid)
1072                         break;
1073         }
1074         obd_putref(export->exp_obd);
1075
1076         OBDO_FREE(tmp_oa);
1077         RETURN(rc);
1078 }
1079
1080 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1081                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1082 {
1083         struct lov_stripe_md *obj_mdp, *lsm;
1084         struct lov_obd *lov = &exp->exp_obd->u.lov;
1085         unsigned ost_idx;
1086         int rc, i;
1087         ENTRY;
1088
1089         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1090                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1091
1092         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1093         if (obj_mdp == NULL)
1094                 RETURN(-ENOMEM);
1095
1096         ost_idx = src_oa->o_nlink;
1097         lsm = *ea;
1098         if (lsm == NULL)
1099                 GOTO(out, rc = -EINVAL);
1100         if (ost_idx >= lov->desc.ld_tgt_count ||
1101             !lov->lov_tgts[ost_idx])
1102                 GOTO(out, rc = -EINVAL);
1103
1104         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1105                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1106                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1107                                 GOTO(out, rc = -EINVAL);
1108                         break;
1109                 }
1110         }
1111         if (i == lsm->lsm_stripe_count)
1112                 GOTO(out, rc = -EINVAL);
1113
1114         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1115                         src_oa, &obj_mdp, oti);
1116 out:
1117         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1118         RETURN(rc);
1119 }
1120
1121 /* the LOV expects oa->o_id to be set to the LOV object id */
1122 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1123                       struct obdo *src_oa, struct lov_stripe_md **ea,
1124                       struct obd_trans_info *oti)
1125 {
1126         struct lov_obd *lov;
1127         struct obd_info oinfo;
1128         struct lov_request_set *set = NULL;
1129         struct lov_request *req;
1130         struct l_wait_info  lwi = { 0 };
1131         int rc = 0;
1132         ENTRY;
1133
1134         LASSERT(ea != NULL);
1135         if (exp == NULL)
1136                 RETURN(-EINVAL);
1137
1138         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1139             src_oa->o_flags == OBD_FL_DELORPHAN) {
1140                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1141                 RETURN(rc);
1142         }
1143
1144         lov = &exp->exp_obd->u.lov;
1145         if (!lov->desc.ld_active_tgt_count)
1146                 RETURN(-EIO);
1147
1148         obd_getref(exp->exp_obd);
1149         /* Recreate a specific object id at the given OST index */
1150         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1151             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1152                  rc = lov_recreate(exp, src_oa, ea, oti);
1153                  GOTO(out, rc);
1154         }
1155
1156         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1157          * later in alloc_qos(), we will wait for those rpcs to complete if
1158          * the osfs age is older than 2 * qos_maxage */
1159         qos_statfs_update(exp->exp_obd,
1160                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1161                                             OBD_STATFS_CACHE_SECONDS),
1162                           0);
1163
1164         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1165         if (rc)
1166                 GOTO(out, rc);
1167
1168         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
1169                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1170                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1171                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1172         }
1173
1174         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1175          * longer then this */
1176         l_wait_event(set->set_waitq, lov_set_finished(set, 1), &lwi);
1177
1178         /* we not have ptlrpc set for assign set->interpret and should
1179          * be call interpret function himself. calling from cb_create_update
1180          * not permited because lov_fini_create_set can sleep for long time,
1181          * but we must avoid sleeping in ptlrpcd interpret function. */
1182         rc = lov_fini_create_set(set, ea);
1183 out:
1184         obd_putref(exp->exp_obd);
1185         RETURN(rc);
1186 }
1187
1188 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1189 do {                                                                            \
1190         LASSERT((lsmp) != NULL);                                                \
1191         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1192                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1193                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1194 } while (0)
1195
1196 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1197                        struct obdo *oa, struct lov_stripe_md *lsm,
1198                        struct obd_trans_info *oti, struct obd_export *md_exp,
1199                        void *capa)
1200 {
1201         struct lov_request_set *set;
1202         struct obd_info oinfo;
1203         struct lov_request *req;
1204         cfs_list_t *pos;
1205         struct lov_obd *lov;
1206         int rc = 0, err = 0;
1207         ENTRY;
1208
1209         ASSERT_LSM_MAGIC(lsm);
1210
1211         if (!exp || !exp->exp_obd)
1212                 RETURN(-ENODEV);
1213
1214         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1215                 LASSERT(oti);
1216                 LASSERT(oti->oti_logcookies);
1217         }
1218
1219         lov = &exp->exp_obd->u.lov;
1220         obd_getref(exp->exp_obd);
1221         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1222         if (rc)
1223                 GOTO(out, rc);
1224
1225         cfs_list_for_each (pos, &set->set_list) {
1226                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1227
1228                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1229                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1230
1231                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1232                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1233                 err = lov_update_common_set(set, req, err);
1234                 if (err) {
1235                         CERROR("error: destroying objid "LPX64" subobj "
1236                                LPX64" on OST idx %d: rc = %d\n",
1237                                oa->o_id, req->rq_oi.oi_oa->o_id,
1238                                req->rq_idx, err);
1239                         if (!rc)
1240                                 rc = err;
1241                 }
1242         }
1243
1244         if (rc == 0) {
1245                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1246                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1247         }
1248         err = lov_fini_destroy_set(set);
1249 out:
1250         obd_putref(exp->exp_obd);
1251         RETURN(rc ? rc : err);
1252 }
1253
1254 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1255                        struct obd_info *oinfo)
1256 {
1257         struct lov_request_set *set;
1258         struct lov_request *req;
1259         cfs_list_t *pos;
1260         struct lov_obd *lov;
1261         int err = 0, rc = 0;
1262         ENTRY;
1263
1264         LASSERT(oinfo);
1265         ASSERT_LSM_MAGIC(oinfo->oi_md);
1266
1267         if (!exp || !exp->exp_obd)
1268                 RETURN(-ENODEV);
1269
1270         lov = &exp->exp_obd->u.lov;
1271
1272         rc = lov_prep_getattr_set(exp, oinfo, &set);
1273         if (rc)
1274                 RETURN(rc);
1275
1276         cfs_list_for_each (pos, &set->set_list) {
1277                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1278
1279                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1280                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1281                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1282
1283                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1284                                  &req->rq_oi);
1285                 err = lov_update_common_set(set, req, rc);
1286                 if (err) {
1287                         CERROR("error: getattr objid "LPX64" subobj "
1288                                LPX64" on OST idx %d: rc = %d\n",
1289                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1290                                req->rq_idx, err);
1291                         break;
1292                 }
1293         }
1294
1295         rc = lov_fini_getattr_set(set);
1296         if (err)
1297                 rc = err;
1298         RETURN(rc);
1299 }
1300
1301 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1302                                  void *data, int rc)
1303 {
1304         struct lov_request_set *lovset = (struct lov_request_set *)data;
1305         int err;
1306         ENTRY;
1307
1308         /* don't do attribute merge if this aysnc op failed */
1309         if (rc)
1310                 cfs_atomic_set(&lovset->set_completes, 0);
1311         err = lov_fini_getattr_set(lovset);
1312         RETURN(rc ? rc : err);
1313 }
1314
1315 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1316                               struct ptlrpc_request_set *rqset)
1317 {
1318         struct lov_request_set *lovset;
1319         struct lov_obd *lov;
1320         cfs_list_t *pos;
1321         struct lov_request *req;
1322         int rc = 0, err;
1323         ENTRY;
1324
1325         LASSERT(oinfo);
1326         ASSERT_LSM_MAGIC(oinfo->oi_md);
1327
1328         if (!exp || !exp->exp_obd)
1329                 RETURN(-ENODEV);
1330
1331         lov = &exp->exp_obd->u.lov;
1332
1333         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1334         if (rc)
1335                 RETURN(rc);
1336
1337         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1338                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1339                oinfo->oi_md->lsm_stripe_size);
1340
1341         cfs_list_for_each (pos, &lovset->set_list) {
1342                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1343
1344                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1345                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1346                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1347                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1348                                        &req->rq_oi, rqset);
1349                 if (rc) {
1350                         CERROR("error: getattr objid "LPX64" subobj "
1351                                LPX64" on OST idx %d: rc = %d\n",
1352                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1353                                req->rq_idx, rc);
1354                         GOTO(out, rc);
1355                 }
1356         }
1357
1358         if (!cfs_list_empty(&rqset->set_requests)) {
1359                 LASSERT(rc == 0);
1360                 LASSERT (rqset->set_interpret == NULL);
1361                 rqset->set_interpret = lov_getattr_interpret;
1362                 rqset->set_arg = (void *)lovset;
1363                 RETURN(rc);
1364         }
1365 out:
1366         if (rc)
1367                 cfs_atomic_set(&lovset->set_completes, 0);
1368         err = lov_fini_getattr_set(lovset);
1369         RETURN(rc ? rc : err);
1370 }
1371
1372 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1373                        struct obd_info *oinfo, struct obd_trans_info *oti)
1374 {
1375         struct lov_request_set *set;
1376         struct lov_obd *lov;
1377         cfs_list_t *pos;
1378         struct lov_request *req;
1379         int err = 0, rc = 0;
1380         ENTRY;
1381
1382         LASSERT(oinfo);
1383         ASSERT_LSM_MAGIC(oinfo->oi_md);
1384
1385         if (!exp || !exp->exp_obd)
1386                 RETURN(-ENODEV);
1387
1388         /* for now, we only expect the following updates here */
1389         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1390                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1391                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1392                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1393                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1394                                             OBD_MD_FLGID | OBD_MD_FLFID |
1395                                             OBD_MD_FLGENER)));
1396         lov = &exp->exp_obd->u.lov;
1397         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1398         if (rc)
1399                 RETURN(rc);
1400
1401         cfs_list_for_each (pos, &set->set_list) {
1402                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1403
1404                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1405                                  &req->rq_oi, NULL);
1406                 err = lov_update_setattr_set(set, req, rc);
1407                 if (err) {
1408                         CERROR("error: setattr objid "LPX64" subobj "
1409                                LPX64" on OST idx %d: rc = %d\n",
1410                                set->set_oi->oi_oa->o_id,
1411                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1412                         if (!rc)
1413                                 rc = err;
1414                 }
1415         }
1416         err = lov_fini_setattr_set(set);
1417         if (!rc)
1418                 rc = err;
1419         RETURN(rc);
1420 }
1421
1422 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1423                                  void *data, int rc)
1424 {
1425         struct lov_request_set *lovset = (struct lov_request_set *)data;
1426         int err;
1427         ENTRY;
1428
1429         if (rc)
1430                 cfs_atomic_set(&lovset->set_completes, 0);
1431         err = lov_fini_setattr_set(lovset);
1432         RETURN(rc ? rc : err);
1433 }
1434
1435 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1436    needed. Otherwise, a client is waiting for responses. */
1437 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1438                              struct obd_trans_info *oti,
1439                              struct ptlrpc_request_set *rqset)
1440 {
1441         struct lov_request_set *set;
1442         struct lov_request *req;
1443         cfs_list_t *pos;
1444         struct lov_obd *lov;
1445         int rc = 0;
1446         ENTRY;
1447
1448         LASSERT(oinfo);
1449         ASSERT_LSM_MAGIC(oinfo->oi_md);
1450         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1451                 LASSERT(oti);
1452                 LASSERT(oti->oti_logcookies);
1453         }
1454
1455         if (!exp || !exp->exp_obd)
1456                 RETURN(-ENODEV);
1457
1458         lov = &exp->exp_obd->u.lov;
1459         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1460         if (rc)
1461                 RETURN(rc);
1462
1463         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1464                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1465                oinfo->oi_md->lsm_stripe_size);
1466
1467         cfs_list_for_each (pos, &set->set_list) {
1468                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1469
1470                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1471                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1472
1473                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1474                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1475                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1476
1477                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1478                                        &req->rq_oi, oti, rqset);
1479                 if (rc) {
1480                         CERROR("error: setattr objid "LPX64" subobj "
1481                                LPX64" on OST idx %d: rc = %d\n",
1482                                set->set_oi->oi_oa->o_id,
1483                                req->rq_oi.oi_oa->o_id,
1484                                req->rq_idx, rc);
1485                         break;
1486                 }
1487         }
1488
1489         /* If we are not waiting for responses on async requests, return. */
1490         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1491                 int err;
1492                 if (rc)
1493                         cfs_atomic_set(&set->set_completes, 0);
1494                 err = lov_fini_setattr_set(set);
1495                 RETURN(rc ? rc : err);
1496         }
1497
1498         LASSERT(rqset->set_interpret == NULL);
1499         rqset->set_interpret = lov_setattr_interpret;
1500         rqset->set_arg = (void *)set;
1501
1502         RETURN(0);
1503 }
1504
1505 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1506                                void *data, int rc)
1507 {
1508         struct lov_request_set *lovset = (struct lov_request_set *)data;
1509         int err;
1510         ENTRY;
1511
1512         if (rc)
1513                 cfs_atomic_set(&lovset->set_completes, 0);
1514         err = lov_fini_punch_set(lovset);
1515         RETURN(rc ? rc : err);
1516 }
1517
1518 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1519  * we can send this 'punch' to just the authoritative node and the nodes
1520  * that the punch will affect. */
1521 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1522                      struct obd_info *oinfo, struct obd_trans_info *oti,
1523                      struct ptlrpc_request_set *rqset)
1524 {
1525         struct lov_request_set *set;
1526         struct lov_obd *lov;
1527         cfs_list_t *pos;
1528         struct lov_request *req;
1529         int rc = 0;
1530         ENTRY;
1531
1532         LASSERT(oinfo);
1533         ASSERT_LSM_MAGIC(oinfo->oi_md);
1534
1535         if (!exp || !exp->exp_obd)
1536                 RETURN(-ENODEV);
1537
1538         lov = &exp->exp_obd->u.lov;
1539         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1540         if (rc)
1541                 RETURN(rc);
1542
1543         cfs_list_for_each (pos, &set->set_list) {
1544                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1545
1546                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1547                                &req->rq_oi, NULL, rqset);
1548                 if (rc) {
1549                         CERROR("error: punch objid "LPX64" subobj "LPX64
1550                                " on OST idx %d: rc = %d\n",
1551                                set->set_oi->oi_oa->o_id,
1552                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1553                         break;
1554                 }
1555         }
1556
1557         if (rc || cfs_list_empty(&rqset->set_requests)) {
1558                 int err;
1559                 err = lov_fini_punch_set(set);
1560                 RETURN(rc ? rc : err);
1561         }
1562
1563         LASSERT(rqset->set_interpret == NULL);
1564         rqset->set_interpret = lov_punch_interpret;
1565         rqset->set_arg = (void *)set;
1566
1567         RETURN(0);
1568 }
1569
1570 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1571                               void *data, int rc)
1572 {
1573         struct lov_request_set *lovset = data;
1574         int err;
1575         ENTRY;
1576
1577         if (rc)
1578                 cfs_atomic_set(&lovset->set_completes, 0);
1579         err = lov_fini_sync_set(lovset);
1580         RETURN(rc ?: err);
1581 }
1582
1583 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1584                     struct obd_info *oinfo, obd_off start, obd_off end,
1585                     struct ptlrpc_request_set *rqset)
1586 {
1587         struct lov_request_set *set = NULL;
1588         struct lov_obd *lov;
1589         cfs_list_t *pos;
1590         struct lov_request *req;
1591         int rc = 0;
1592         ENTRY;
1593
1594         ASSERT_LSM_MAGIC(oinfo->oi_md);
1595         LASSERT(rqset != NULL);
1596
1597         if (!exp->exp_obd)
1598                 RETURN(-ENODEV);
1599
1600         lov = &exp->exp_obd->u.lov;
1601         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1602         if (rc)
1603                 RETURN(rc);
1604
1605         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1606                set->set_oi->oi_oa->o_id, start, end);
1607
1608         cfs_list_for_each (pos, &set->set_list) {
1609                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1610
1611                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1612                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1613                               req->rq_oi.oi_policy.l_extent.end, rqset);
1614                 if (rc) {
1615                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1616                                " on OST idx %d: rc = %d\n",
1617                                set->set_oi->oi_oa->o_id,
1618                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1619                         break;
1620                 }
1621         }
1622
1623         /* If we are not waiting for responses on async requests, return. */
1624         if (rc || cfs_list_empty(&rqset->set_requests)) {
1625                 int err = lov_fini_sync_set(set);
1626
1627                 RETURN(rc ?: err);
1628         }
1629
1630         LASSERT(rqset->set_interpret == NULL);
1631         rqset->set_interpret = lov_sync_interpret;
1632         rqset->set_arg = (void *)set;
1633
1634         RETURN(0);
1635 }
1636
1637 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1638                          obd_count oa_bufs, struct brw_page *pga)
1639 {
1640         struct obd_info oinfo = { { { 0 } } };
1641         int i, rc = 0;
1642
1643         oinfo.oi_oa = lov_oinfo->oi_oa;
1644
1645         /* The caller just wants to know if there's a chance that this
1646          * I/O can succeed */
1647         for (i = 0; i < oa_bufs; i++) {
1648                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1649                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1650                 obd_off start, end;
1651
1652                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1653                                            pga[i].off + pga[i].count - 1,
1654                                            &start, &end))
1655                         continue;
1656
1657                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1658                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1659                         return -EIO;
1660                 }
1661
1662                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1663                              1, &pga[i], NULL);
1664                 if (rc)
1665                         break;
1666         }
1667         return rc;
1668 }
1669
1670 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1671                    obd_count oa_bufs, struct brw_page *pga,
1672                    struct obd_trans_info *oti)
1673 {
1674         struct lov_request_set *set;
1675         struct lov_request *req;
1676         cfs_list_t *pos;
1677         struct lov_obd *lov = &exp->exp_obd->u.lov;
1678         int err, rc = 0;
1679         ENTRY;
1680
1681         ASSERT_LSM_MAGIC(oinfo->oi_md);
1682
1683         if (cmd == OBD_BRW_CHECK) {
1684                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1685                 RETURN(rc);
1686         }
1687
1688         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1689         if (rc)
1690                 RETURN(rc);
1691
1692         cfs_list_for_each (pos, &set->set_list) {
1693                 struct obd_export *sub_exp;
1694                 struct brw_page *sub_pga;
1695                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1696
1697                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1698                 sub_pga = set->set_pga + req->rq_pgaidx;
1699                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1700                              sub_pga, oti);
1701                 if (rc)
1702                         break;
1703                 lov_update_common_set(set, req, rc);
1704         }
1705
1706         err = lov_fini_brw_set(set);
1707         if (!rc)
1708                 rc = err;
1709         RETURN(rc);
1710 }
1711
1712 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1713                                  void *data, int rc)
1714 {
1715         struct lov_request_set *lovset = (struct lov_request_set *)data;
1716         ENTRY;
1717         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1718         RETURN(rc);
1719 }
1720
1721 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1722                        struct ldlm_enqueue_info *einfo,
1723                        struct ptlrpc_request_set *rqset)
1724 {
1725         ldlm_mode_t mode = einfo->ei_mode;
1726         struct lov_request_set *set;
1727         struct lov_request *req;
1728         cfs_list_t *pos;
1729         struct lov_obd *lov;
1730         ldlm_error_t rc;
1731         ENTRY;
1732
1733         LASSERT(oinfo);
1734         ASSERT_LSM_MAGIC(oinfo->oi_md);
1735         LASSERT(mode == (mode & -mode));
1736
1737         /* we should never be asked to replay a lock this way. */
1738         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1739
1740         if (!exp || !exp->exp_obd)
1741                 RETURN(-ENODEV);
1742
1743         lov = &exp->exp_obd->u.lov;
1744         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1745         if (rc)
1746                 RETURN(rc);
1747
1748         cfs_list_for_each (pos, &set->set_list) {
1749                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1750
1751                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1752                                  &req->rq_oi, einfo, rqset);
1753                 if (rc != ELDLM_OK)
1754                         GOTO(out, rc);
1755         }
1756
1757         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1758                 LASSERT(rc == 0);
1759                 LASSERT(rqset->set_interpret == NULL);
1760                 rqset->set_interpret = lov_enqueue_interpret;
1761                 rqset->set_arg = (void *)set;
1762                 RETURN(rc);
1763         }
1764 out:
1765         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1766         RETURN(rc);
1767 }
1768
1769 static int lov_change_cbdata(struct obd_export *exp,
1770                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1771                              void *data)
1772 {
1773         struct lov_obd *lov;
1774         int rc = 0, i;
1775         ENTRY;
1776
1777         ASSERT_LSM_MAGIC(lsm);
1778
1779         if (!exp || !exp->exp_obd)
1780                 RETURN(-ENODEV);
1781
1782         lov = &exp->exp_obd->u.lov;
1783         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1784                 struct lov_stripe_md submd;
1785                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1786
1787                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1788                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1789                         continue;
1790                 }
1791
1792                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1793                 submd.lsm_object_id = loi->loi_id;
1794                 submd.lsm_object_seq = loi->loi_seq;
1795                 submd.lsm_stripe_count = 0;
1796                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1797                                        &submd, it, data);
1798         }
1799         RETURN(rc);
1800 }
1801
1802 /* find any ldlm lock of the inode in lov
1803  * return 0    not find
1804  *        1    find one
1805  *      < 0    error */
1806 static int lov_find_cbdata(struct obd_export *exp,
1807                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1808                            void *data)
1809 {
1810         struct lov_obd *lov;
1811         int rc = 0, i;
1812         ENTRY;
1813
1814         ASSERT_LSM_MAGIC(lsm);
1815
1816         if (!exp || !exp->exp_obd)
1817                 RETURN(-ENODEV);
1818
1819         lov = &exp->exp_obd->u.lov;
1820         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1821                 struct lov_stripe_md submd;
1822                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1823
1824                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1825                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1826                         continue;
1827                 }
1828
1829                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1830                 submd.lsm_object_id = loi->loi_id;
1831                 submd.lsm_object_seq = loi->loi_seq;
1832                 submd.lsm_stripe_count = 0;
1833                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1834                                      &submd, it, data);
1835                 if (rc != 0)
1836                         RETURN(rc);
1837         }
1838         RETURN(rc);
1839 }
1840
1841 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1842                       __u32 mode, struct lustre_handle *lockh)
1843 {
1844         struct lov_request_set *set;
1845         struct obd_info oinfo;
1846         struct lov_request *req;
1847         cfs_list_t *pos;
1848         struct lov_obd *lov;
1849         struct lustre_handle *lov_lockhp;
1850         int err = 0, rc = 0;
1851         ENTRY;
1852
1853         ASSERT_LSM_MAGIC(lsm);
1854
1855         if (!exp || !exp->exp_obd)
1856                 RETURN(-ENODEV);
1857
1858         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1859         LASSERT(lockh);
1860         lov = &exp->exp_obd->u.lov;
1861         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1862         if (rc)
1863                 RETURN(rc);
1864
1865         cfs_list_for_each (pos, &set->set_list) {
1866                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1867                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1868
1869                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1870                                 req->rq_oi.oi_md, mode, lov_lockhp);
1871                 rc = lov_update_common_set(set, req, rc);
1872                 if (rc) {
1873                         CERROR("error: cancel objid "LPX64" subobj "
1874                                LPX64" on OST idx %d: rc = %d\n",
1875                                lsm->lsm_object_id,
1876                                req->rq_oi.oi_md->lsm_object_id,
1877                                req->rq_idx, rc);
1878                         err = rc;
1879                 }
1880
1881         }
1882         lov_fini_cancel_set(set);
1883         RETURN(err);
1884 }
1885
1886 static int lov_cancel_unused(struct obd_export *exp,
1887                              struct lov_stripe_md *lsm,
1888                              ldlm_cancel_flags_t flags, void *opaque)
1889 {
1890         struct lov_obd *lov;
1891         int rc = 0, i;
1892         ENTRY;
1893
1894         if (!exp || !exp->exp_obd)
1895                 RETURN(-ENODEV);
1896
1897         lov = &exp->exp_obd->u.lov;
1898         if (lsm == NULL) {
1899                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1900                         int err;
1901                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1902                                 continue;
1903
1904                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1905                                                 flags, opaque);
1906                         if (!rc)
1907                                 rc = err;
1908                 }
1909                 RETURN(rc);
1910         }
1911
1912         ASSERT_LSM_MAGIC(lsm);
1913
1914         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1915         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1916                 struct lov_stripe_md submd;
1917                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1918                 int err;
1919
1920                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1921                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1922                         continue;
1923                 }
1924
1925                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1926                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1927
1928                 submd.lsm_object_id = loi->loi_id;
1929                 submd.lsm_object_seq = loi->loi_seq;
1930                 submd.lsm_stripe_count = 0;
1931                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1932                                         &submd, flags, opaque);
1933                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1934                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1935                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1936                                loi->loi_id, loi->loi_ost_idx, err);
1937                         if (!rc)
1938                                 rc = err;
1939                 }
1940         }
1941         RETURN(rc);
1942 }
1943
1944 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1945 {
1946         struct lov_request_set *lovset = (struct lov_request_set *)data;
1947         int err;
1948         ENTRY;
1949
1950         if (rc)
1951                 cfs_atomic_set(&lovset->set_completes, 0);
1952
1953         err = lov_fini_statfs_set(lovset);
1954         RETURN(rc ? rc : err);
1955 }
1956
1957 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1958                             __u64 max_age, struct ptlrpc_request_set *rqset)
1959 {
1960         struct obd_device      *obd = class_exp2obd(exp);
1961         struct lov_request_set *set;
1962         struct lov_request *req;
1963         cfs_list_t *pos;
1964         struct lov_obd *lov;
1965         int rc = 0;
1966         ENTRY;
1967
1968         LASSERT(oinfo != NULL);
1969         LASSERT(oinfo->oi_osfs != NULL);
1970
1971         lov = &obd->u.lov;
1972         rc = lov_prep_statfs_set(obd, oinfo, &set);
1973         if (rc)
1974                 RETURN(rc);
1975
1976         cfs_list_for_each (pos, &set->set_list) {
1977                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1978                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1979                                       &req->rq_oi, max_age, rqset);
1980                 if (rc)
1981                         break;
1982         }
1983
1984         if (rc || cfs_list_empty(&rqset->set_requests)) {
1985                 int err;
1986                 if (rc)
1987                         cfs_atomic_set(&set->set_completes, 0);
1988                 err = lov_fini_statfs_set(set);
1989                 RETURN(rc ? rc : err);
1990         }
1991
1992         LASSERT(rqset->set_interpret == NULL);
1993         rqset->set_interpret = lov_statfs_interpret;
1994         rqset->set_arg = (void *)set;
1995         RETURN(0);
1996 }
1997
1998 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1999                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
2000 {
2001         struct ptlrpc_request_set *set = NULL;
2002         struct obd_info oinfo = { { { 0 } } };
2003         int rc = 0;
2004         ENTRY;
2005
2006
2007         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2008          * statfs requests */
2009         set = ptlrpc_prep_set();
2010         if (set == NULL)
2011                 RETURN(-ENOMEM);
2012
2013         oinfo.oi_osfs = osfs;
2014         oinfo.oi_flags = flags;
2015         rc = lov_statfs_async(exp, &oinfo, max_age, set);
2016         if (rc == 0)
2017                 rc = ptlrpc_set_wait(set);
2018         ptlrpc_set_destroy(set);
2019
2020         RETURN(rc);
2021 }
2022
2023 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2024                          void *karg, void *uarg)
2025 {
2026         struct obd_device *obddev = class_exp2obd(exp);
2027         struct lov_obd *lov = &obddev->u.lov;
2028         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
2029         struct obd_uuid *uuidp;
2030         ENTRY;
2031
2032         switch (cmd) {
2033         case IOC_OBD_STATFS: {
2034                 struct obd_ioctl_data *data = karg;
2035                 struct obd_device *osc_obd;
2036                 struct obd_statfs stat_buf = {0};
2037                 __u32 index;
2038                 __u32 flags;
2039
2040                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2041                 if ((index >= count))
2042                         RETURN(-ENODEV);
2043
2044                 if (!lov->lov_tgts[index])
2045                         /* Try again with the next index */
2046                         RETURN(-EAGAIN);
2047                 if (!lov->lov_tgts[index]->ltd_active)
2048                         RETURN(-ENODATA);
2049
2050                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2051                 if (!osc_obd)
2052                         RETURN(-EINVAL);
2053
2054                 /* copy UUID */
2055                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2056                                      min((int) data->ioc_plen2,
2057                                          (int) sizeof(struct obd_uuid))))
2058                         RETURN(-EFAULT);
2059
2060                 flags = uarg ? *(__u32*)uarg : 0;
2061                 /* got statfs data */
2062                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
2063                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2064                                 flags);
2065                 if (rc)
2066                         RETURN(rc);
2067                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
2068                                      min((int) data->ioc_plen1,
2069                                          (int) sizeof(stat_buf))))
2070                         RETURN(-EFAULT);
2071                 break;
2072         }
2073         case OBD_IOC_LOV_GET_CONFIG: {
2074                 struct obd_ioctl_data *data;
2075                 struct lov_desc *desc;
2076                 char *buf = NULL;
2077                 __u32 *genp;
2078
2079                 len = 0;
2080                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2081                         RETURN(-EINVAL);
2082
2083                 data = (struct obd_ioctl_data *)buf;
2084
2085                 if (sizeof(*desc) > data->ioc_inllen1) {
2086                         obd_ioctl_freedata(buf, len);
2087                         RETURN(-EINVAL);
2088                 }
2089
2090                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2091                         obd_ioctl_freedata(buf, len);
2092                         RETURN(-EINVAL);
2093                 }
2094
2095                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2096                         obd_ioctl_freedata(buf, len);
2097                         RETURN(-EINVAL);
2098                 }
2099
2100                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2101                 memcpy(desc, &(lov->desc), sizeof(*desc));
2102
2103                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2104                 genp = (__u32 *)data->ioc_inlbuf3;
2105                 /* the uuid will be empty for deleted OSTs */
2106                 for (i = 0; i < count; i++, uuidp++, genp++) {
2107                         if (!lov->lov_tgts[i])
2108                                 continue;
2109                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2110                         *genp = lov->lov_tgts[i]->ltd_gen;
2111                 }
2112
2113                 if (cfs_copy_to_user((void *)uarg, buf, len))
2114                         rc = -EFAULT;
2115                 obd_ioctl_freedata(buf, len);
2116                 break;
2117         }
2118         case LL_IOC_LOV_SETSTRIPE:
2119                 rc = lov_setstripe(exp, len, karg, uarg);
2120                 break;
2121         case LL_IOC_LOV_GETSTRIPE:
2122                 rc = lov_getstripe(exp, karg, uarg);
2123                 break;
2124         case LL_IOC_LOV_SETEA:
2125                 rc = lov_setea(exp, karg, uarg);
2126                 break;
2127         case OBD_IOC_QUOTACTL: {
2128                 struct if_quotactl *qctl = karg;
2129                 struct lov_tgt_desc *tgt = NULL;
2130                 struct obd_quotactl *oqctl;
2131
2132                 if (qctl->qc_valid == QC_OSTIDX) {
2133                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2134                                 RETURN(-EINVAL);
2135
2136                         tgt = lov->lov_tgts[qctl->qc_idx];
2137                         if (!tgt || !tgt->ltd_exp)
2138                                 RETURN(-EINVAL);
2139                 } else if (qctl->qc_valid == QC_UUID) {
2140                         for (i = 0; i < count; i++) {
2141                                 tgt = lov->lov_tgts[i];
2142                                 if (!tgt ||
2143                                     !obd_uuid_equals(&tgt->ltd_uuid,
2144                                                      &qctl->obd_uuid))
2145                                         continue;
2146
2147                                 if (tgt->ltd_exp == NULL)
2148                                         RETURN(-EINVAL);
2149
2150                                 break;
2151                         }
2152                 } else {
2153                         RETURN(-EINVAL);
2154                 }
2155
2156                 if (i >= count)
2157                         RETURN(-EAGAIN);
2158
2159                 LASSERT(tgt && tgt->ltd_exp);
2160                 OBD_ALLOC_PTR(oqctl);
2161                 if (!oqctl)
2162                         RETURN(-ENOMEM);
2163
2164                 QCTL_COPY(oqctl, qctl);
2165                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2166                 if (rc == 0) {
2167                         QCTL_COPY(qctl, oqctl);
2168                         qctl->qc_valid = QC_OSTIDX;
2169                         qctl->obd_uuid = tgt->ltd_uuid;
2170                 }
2171                 OBD_FREE_PTR(oqctl);
2172                 break;
2173         }
2174         default: {
2175                 int set = 0;
2176
2177                 if (count == 0)
2178                         RETURN(-ENOTTY);
2179
2180                 for (i = 0; i < count; i++) {
2181                         int err;
2182                         struct obd_device *osc_obd;
2183
2184                         /* OST was disconnected */
2185                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2186                                 continue;
2187
2188                         /* ll_umount_begin() sets force flag but for lov, not
2189                          * osc. Let's pass it through */
2190                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2191                         osc_obd->obd_force = obddev->obd_force;
2192                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2193                                             len, karg, uarg);
2194                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2195                                 RETURN(err);
2196                         } else if (err) {
2197                                 if (lov->lov_tgts[i]->ltd_active) {
2198                                         CDEBUG(err == -ENOTTY ?
2199                                                D_IOCTL : D_WARNING,
2200                                                "iocontrol OSC %s on OST "
2201                                                "idx %d cmd %x: err = %d\n",
2202                                                lov_uuid2str(lov, i),
2203                                                i, cmd, err);
2204                                         if (!rc)
2205                                                 rc = err;
2206                                 }
2207                         } else {
2208                                 set = 1;
2209                         }
2210                 }
2211                 if (!set && !rc)
2212                         rc = -EIO;
2213         }
2214         }
2215
2216         RETURN(rc);
2217 }
2218
2219 #define FIEMAP_BUFFER_SIZE 4096
2220
2221 /**
2222  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2223  * call. The local end offset and the device are sent in the first
2224  * fm_extent. This function calculates the stripe number from the index.
2225  * This function returns a stripe_no on which mapping is to be restarted.
2226  *
2227  * This function returns fm_end_offset which is the in-OST offset at which
2228  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2229  * will re-calculate proper offset in next stripe.
2230  * Note that the first extent is passed to lov_get_info via the value field.
2231  *
2232  * \param fiemap fiemap request header
2233  * \param lsm striping information for the file
2234  * \param fm_start logical start of mapping
2235  * \param fm_end logical end of mapping
2236  * \param start_stripe starting stripe will be returned in this
2237  */
2238 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2239                                    struct lov_stripe_md *lsm, obd_size fm_start,
2240                                    obd_size fm_end, int *start_stripe)
2241 {
2242         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2243         obd_off lun_start, lun_end;
2244         obd_size fm_end_offset;
2245         int stripe_no = -1, i;
2246
2247         if (fiemap->fm_extent_count == 0 ||
2248             fiemap->fm_extents[0].fe_logical == 0)
2249                 return 0;
2250
2251         /* Find out stripe_no from ost_index saved in the fe_device */
2252         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2253                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2254                                         fiemap->fm_extents[0].fe_device) {
2255                         stripe_no = i;
2256                         break;
2257                 }
2258         }
2259         if (stripe_no == -1)
2260                 return -EINVAL;
2261
2262         /* If we have finished mapping on previous device, shift logical
2263          * offset to start of next device */
2264         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2265                                    &lun_start, &lun_end)) != 0 &&
2266                                    local_end < lun_end) {
2267                 fm_end_offset = local_end;
2268                 *start_stripe = stripe_no;
2269         } else {
2270                 /* This is a special value to indicate that caller should
2271                  * calculate offset in next stripe. */
2272                 fm_end_offset = 0;
2273                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2274         }
2275
2276         return fm_end_offset;
2277 }
2278
2279 /**
2280  * We calculate on which OST the mapping will end. If the length of mapping
2281  * is greater than (stripe_size * stripe_count) then the last_stripe will
2282  * will be one just before start_stripe. Else we check if the mapping
2283  * intersects each OST and find last_stripe.
2284  * This function returns the last_stripe and also sets the stripe_count
2285  * over which the mapping is spread
2286  *
2287  * \param lsm striping information for the file
2288  * \param fm_start logical start of mapping
2289  * \param fm_end logical end of mapping
2290  * \param start_stripe starting stripe of the mapping
2291  * \param stripe_count the number of stripes across which to map is returned
2292  *
2293  * \retval last_stripe return the last stripe of the mapping
2294  */
2295 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2296                             obd_size fm_end, int start_stripe,
2297                             int *stripe_count)
2298 {
2299         int last_stripe;
2300         obd_off obd_start, obd_end;
2301         int i, j;
2302
2303         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2304                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2305                                                               start_stripe - 1);
2306                 *stripe_count = lsm->lsm_stripe_count;
2307         } else {
2308                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2309                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2310                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2311                                                    &obd_start, &obd_end)) == 0)
2312                                 break;
2313                 }
2314                 *stripe_count = j;
2315                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2316         }
2317
2318         return last_stripe;
2319 }
2320
2321 /**
2322  * Set fe_device and copy extents from local buffer into main return buffer.
2323  *
2324  * \param fiemap fiemap request header
2325  * \param lcl_fm_ext array of local fiemap extents to be copied
2326  * \param ost_index OST index to be written into the fm_device field for each
2327                     extent
2328  * \param ext_count number of extents to be copied
2329  * \param current_extent where to start copying in main extent array
2330  */
2331 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2332                                   struct ll_fiemap_extent *lcl_fm_ext,
2333                                   int ost_index, unsigned int ext_count,
2334                                   int current_extent)
2335 {
2336         char *to;
2337         int ext;
2338
2339         for (ext = 0; ext < ext_count; ext++) {
2340                 lcl_fm_ext[ext].fe_device = ost_index;
2341                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2342         }
2343
2344         /* Copy fm_extent's from fm_local to return buffer */
2345         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2346         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2347 }
2348
2349 /**
2350  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2351  * This also handles the restarting of FIEMAP calls in case mapping overflows
2352  * the available number of extents in single call.
2353  */
2354 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2355                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2356 {
2357         struct ll_fiemap_info_key *fm_key = key;
2358         struct ll_user_fiemap *fiemap = val;
2359         struct ll_user_fiemap *fm_local = NULL;
2360         struct ll_fiemap_extent *lcl_fm_ext;
2361         int count_local;
2362         unsigned int get_num_extents = 0;
2363         int ost_index = 0, actual_start_stripe, start_stripe;
2364         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2365         obd_size curr_loc;
2366         int current_extent = 0, rc = 0, i;
2367         int ost_eof = 0; /* EOF for object */
2368         int ost_done = 0; /* done with required mapping for this OST? */
2369         int last_stripe;
2370         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2371         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2372
2373         if (lsm == NULL)
2374                 GOTO(out, rc = 0);
2375
2376         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2377                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2378
2379         OBD_ALLOC_LARGE(fm_local, buffer_size);
2380         if (fm_local == NULL)
2381                 GOTO(out, rc = -ENOMEM);
2382         lcl_fm_ext = &fm_local->fm_extents[0];
2383
2384         count_local = fiemap_size_to_count(buffer_size);
2385
2386         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2387         fm_start = fiemap->fm_start;
2388         fm_length = fiemap->fm_length;
2389         /* Calculate start stripe, last stripe and length of mapping */
2390         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2391         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2392                                                 fm_start + fm_length - 1);
2393         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2394         if (fm_end > fm_key->oa.o_size)
2395                 fm_end = fm_key->oa.o_size;
2396
2397         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2398                                             actual_start_stripe, &stripe_count);
2399
2400         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2401                                                   fm_end, &start_stripe);
2402         if (fm_end_offset == -EINVAL)
2403                 return -EINVAL;
2404
2405         if (fiemap->fm_extent_count == 0) {
2406                 get_num_extents = 1;
2407                 count_local = 0;
2408         }
2409
2410         /* Check each stripe */
2411         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2412              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2413                 obd_size req_fm_len; /* Stores length of required mapping */
2414                 obd_size len_mapped_single_call;
2415                 obd_off lun_start, lun_end, obd_object_end;
2416                 unsigned int ext_count;
2417
2418                 cur_stripe_wrap = cur_stripe;
2419
2420                 /* Find out range of mapping on this stripe */
2421                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2422                                            &lun_start, &obd_object_end)) == 0)
2423                         continue;
2424
2425                 /* If this is a continuation FIEMAP call and we are on
2426                  * starting stripe then lun_start needs to be set to
2427                  * fm_end_offset */
2428                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2429                         lun_start = fm_end_offset;
2430
2431                 if (fm_length != ~0ULL) {
2432                         /* Handle fm_start + fm_length overflow */
2433                         if (fm_start + fm_length < fm_start)
2434                                 fm_length = ~0ULL - fm_start;
2435                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2436                                                      cur_stripe);
2437                 } else {
2438                         lun_end = ~0ULL;
2439                 }
2440
2441                 if (lun_start == lun_end)
2442                         continue;
2443
2444                 req_fm_len = obd_object_end - lun_start;
2445                 fm_local->fm_length = 0;
2446                 len_mapped_single_call = 0;
2447
2448                 /* If the output buffer is very large and the objects have many
2449                  * extents we may need to loop on a single OST repeatedly */
2450                 ost_eof = 0;
2451                 ost_done = 0;
2452                 do {
2453                         if (get_num_extents == 0) {
2454                                 /* Don't get too many extents. */
2455                                 if (current_extent + count_local >
2456                                     fiemap->fm_extent_count)
2457                                         count_local = fiemap->fm_extent_count -
2458                                                                  current_extent;
2459                         }
2460
2461                         lun_start += len_mapped_single_call;
2462                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2463                         req_fm_len = fm_local->fm_length;
2464                         fm_local->fm_extent_count = count_local;
2465                         fm_local->fm_mapped_extents = 0;
2466                         fm_local->fm_flags = fiemap->fm_flags;
2467
2468                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2469                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2470                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2471
2472                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2473                                 GOTO(out, rc = -EINVAL);
2474
2475                         /* If OST is inactive, return extent with UNKNOWN flag */
2476                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2477                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2478                                 fm_local->fm_mapped_extents = 1;
2479
2480                                 lcl_fm_ext[0].fe_logical = lun_start;
2481                                 lcl_fm_ext[0].fe_length = obd_object_end -
2482                                                                       lun_start;
2483                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2484
2485                                 goto inactive_tgt;
2486                         }
2487
2488                         fm_local->fm_start = lun_start;
2489                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2490                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2491                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2492                         rc = obd_get_info(NULL,
2493                                           lov->lov_tgts[ost_index]->ltd_exp,
2494                                           keylen, key, vallen, fm_local, lsm);
2495                         if (rc != 0)
2496                                 GOTO(out, rc);
2497
2498 inactive_tgt:
2499                         ext_count = fm_local->fm_mapped_extents;
2500                         if (ext_count == 0) {
2501                                 ost_done = 1;
2502                                 /* If last stripe has hole at the end,
2503                                  * then we need to return */
2504                                 if (cur_stripe_wrap == last_stripe) {
2505                                         fiemap->fm_mapped_extents = 0;
2506                                         goto finish;
2507                                 }
2508                                 break;
2509                         }
2510
2511                         /* If we just need num of extents then go to next device */
2512                         if (get_num_extents) {
2513                                 current_extent += ext_count;
2514                                 break;
2515                         }
2516
2517                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2518                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2519
2520                         /* Have we finished mapping on this device? */
2521                         if (req_fm_len <= len_mapped_single_call)
2522                                 ost_done = 1;
2523
2524                         /* Clear the EXTENT_LAST flag which can be present on
2525                          * last extent */
2526                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2527                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2528                                                             ~FIEMAP_EXTENT_LAST;
2529
2530                         curr_loc = lov_stripe_size(lsm,
2531                                            lcl_fm_ext[ext_count - 1].fe_logical+
2532                                            lcl_fm_ext[ext_count - 1].fe_length,
2533                                            cur_stripe);
2534                         if (curr_loc >= fm_key->oa.o_size)
2535                                 ost_eof = 1;
2536
2537                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2538                                                      ost_index, ext_count,
2539                                                      current_extent);
2540
2541                         current_extent += ext_count;
2542
2543                         /* Ran out of available extents? */
2544                         if (current_extent >= fiemap->fm_extent_count)
2545                                 goto finish;
2546                 } while (ost_done == 0 && ost_eof == 0);
2547
2548                 if (cur_stripe_wrap == last_stripe)
2549                         goto finish;
2550         }
2551
2552 finish:
2553         /* Indicate that we are returning device offsets unless file just has
2554          * single stripe */
2555         if (lsm->lsm_stripe_count > 1)
2556                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2557
2558         if (get_num_extents)
2559                 goto skip_last_device_calc;
2560
2561         /* Check if we have reached the last stripe and whether mapping for that
2562          * stripe is done. */
2563         if (cur_stripe_wrap == last_stripe) {
2564                 if (ost_done || ost_eof)
2565                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2566                                                              FIEMAP_EXTENT_LAST;
2567         }
2568
2569 skip_last_device_calc:
2570         fiemap->fm_mapped_extents = current_extent;
2571
2572 out:
2573         OBD_FREE_LARGE(fm_local, buffer_size);
2574         return rc;
2575 }
2576
2577 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2578                         __u32 keylen, void *key, __u32 *vallen, void *val,
2579                         struct lov_stripe_md *lsm)
2580 {
2581         struct obd_device *obddev = class_exp2obd(exp);
2582         struct lov_obd *lov = &obddev->u.lov;
2583         int i, rc;
2584         ENTRY;
2585
2586         if (!vallen || !val)
2587                 RETURN(-EFAULT);
2588
2589         obd_getref(obddev);
2590
2591         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2592                 struct {
2593                         char name[16];
2594                         struct ldlm_lock *lock;
2595                 } *data = key;
2596                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2597                 struct lov_oinfo *loi;
2598                 __u32 *stripe = val;
2599
2600                 if (*vallen < sizeof(*stripe))
2601                         GOTO(out, rc = -EFAULT);
2602                 *vallen = sizeof(*stripe);
2603
2604                 /* XXX This is another one of those bits that will need to
2605                  * change if we ever actually support nested LOVs.  It uses
2606                  * the lock's export to find out which stripe it is. */
2607                 /* XXX - it's assumed all the locks for deleted OSTs have
2608                  * been cancelled. Also, the export for deleted OSTs will
2609                  * be NULL and won't match the lock's export. */
2610                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2611                         loi = lsm->lsm_oinfo[i];
2612                         if (!lov->lov_tgts[loi->loi_ost_idx])
2613                                 continue;
2614                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2615                             data->lock->l_conn_export &&
2616                             osc_res_name_eq(loi->loi_id, loi->loi_seq, res_id)) {
2617                                 *stripe = i;
2618                                 GOTO(out, rc = 0);
2619                         }
2620                 }
2621                 LDLM_ERROR(data->lock, "lock on inode without such object");
2622                 dump_lsm(D_ERROR, lsm);
2623                 GOTO(out, rc = -ENXIO);
2624         } else if (KEY_IS(KEY_LAST_ID)) {
2625                 struct obd_id_info *info = val;
2626                 __u32 size = sizeof(obd_id);
2627                 struct lov_tgt_desc *tgt;
2628
2629                 LASSERT(*vallen == sizeof(struct obd_id_info));
2630                 tgt = lov->lov_tgts[info->idx];
2631
2632                 if (!tgt || !tgt->ltd_active)
2633                         GOTO(out, rc = -ESRCH);
2634
2635                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2636                                   &size, info->data, NULL);
2637                 GOTO(out, rc = 0);
2638         } else if (KEY_IS(KEY_LOVDESC)) {
2639                 struct lov_desc *desc_ret = val;
2640                 *desc_ret = lov->desc;
2641
2642                 GOTO(out, rc = 0);
2643         } else if (KEY_IS(KEY_FIEMAP)) {
2644                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2645                 GOTO(out, rc);
2646         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2647                 struct lov_tgt_desc *tgt;
2648                 __u64 ost_idx = *((__u64*)val);
2649
2650                 LASSERT(*vallen == sizeof(__u64));
2651                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2652                 tgt = lov->lov_tgts[ost_idx];
2653
2654                 if (!tgt || !tgt->ltd_exp)
2655                         GOTO(out, rc = -ESRCH);
2656
2657                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2658                 GOTO(out, rc = 0);
2659         } else if (KEY_IS(KEY_TGT_COUNT)) {
2660                 *((int *)val) = lov->desc.ld_tgt_count;
2661                 GOTO(out, rc = 0);
2662         }
2663
2664         rc = -EINVAL;
2665
2666 out:
2667         obd_putref(obddev);
2668         RETURN(rc);
2669 }
2670
2671 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2672                               obd_count keylen, void *key, obd_count vallen,
2673                               void *val, struct ptlrpc_request_set *set)
2674 {
2675         struct obd_device *obddev = class_exp2obd(exp);
2676         struct lov_obd *lov = &obddev->u.lov;
2677         obd_count count;
2678         int i, rc = 0, err;
2679         struct lov_tgt_desc *tgt;
2680         unsigned incr, check_uuid,
2681                  do_inactive, no_set;
2682         unsigned next_id = 0,  mds_con = 0, capa = 0;
2683         ENTRY;
2684
2685         incr = check_uuid = do_inactive = no_set = 0;
2686         if (set == NULL) {
2687                 no_set = 1;
2688                 set = ptlrpc_prep_set();
2689                 if (!set)
2690                         RETURN(-ENOMEM);
2691         }
2692
2693         obd_getref(obddev);
2694         count = lov->desc.ld_tgt_count;
2695
2696         if (KEY_IS(KEY_NEXT_ID)) {
2697                 count = vallen / sizeof(struct obd_id_info);
2698                 vallen = sizeof(obd_id);
2699                 incr = sizeof(struct obd_id_info);
2700                 do_inactive = 1;
2701                 next_id = 1;
2702         } else if (KEY_IS(KEY_CHECKSUM)) {
2703                 do_inactive = 1;
2704         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2705                 /* use defaults:  do_inactive = incr = 0; */
2706         } else if (KEY_IS(KEY_MDS_CONN)) {
2707                 mds_con = 1;
2708         } else if (KEY_IS(KEY_CAPA_KEY)) {
2709                 capa = 1;
2710         }
2711
2712         for (i = 0; i < count; i++, val = (char *)val + incr) {
2713                 if (next_id) {
2714                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2715                 } else {
2716                         tgt = lov->lov_tgts[i];
2717                 }
2718                 /* OST was disconnected */
2719                 if (!tgt || !tgt->ltd_exp)
2720                         continue;
2721
2722                 /* OST is inactive and we don't want inactive OSCs */
2723                 if (!tgt->ltd_active && !do_inactive)
2724                         continue;
2725
2726                 if (mds_con) {
2727                         struct mds_group_info *mgi;
2728
2729                         LASSERT(vallen == sizeof(*mgi));
2730                         mgi = (struct mds_group_info *)val;
2731
2732                         /* Only want a specific OSC */
2733                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2734                                                 &tgt->ltd_uuid))
2735                                 continue;
2736
2737                         err = obd_set_info_async(env, tgt->ltd_exp,
2738                                          keylen, key, sizeof(int),
2739                                          &mgi->group, set);
2740                 } else if (next_id) {
2741                         err = obd_set_info_async(env, tgt->ltd_exp,
2742                                          keylen, key, vallen,
2743                                          ((struct obd_id_info*)val)->data, set);
2744                 } else if (capa) {
2745                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2746
2747                         LASSERT(vallen == sizeof(*info));
2748
2749                          /* Only want a specific OSC */
2750                         if (info->uuid &&
2751                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2752                                 continue;
2753
2754                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2755                                                  key, sizeof(*info->capa),
2756                                                  info->capa, set);
2757                 } else {
2758                         /* Only want a specific OSC */
2759                         if (check_uuid &&
2760                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2761                                 continue;
2762
2763                         err = obd_set_info_async(env, tgt->ltd_exp,
2764                                          keylen, key, vallen, val, set);
2765                 }
2766
2767                 if (!rc)
2768                         rc = err;
2769         }
2770
2771         obd_putref(obddev);
2772         if (no_set) {
2773                 err = ptlrpc_set_wait(set);
2774                 if (!rc)
2775                         rc = err;
2776                 ptlrpc_set_destroy(set);
2777         }
2778         RETURN(rc);
2779 }
2780
2781 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2782                            int cmd, __u64 *offset)
2783 {
2784         __u32 ssize = lsm->lsm_stripe_size;
2785         __u64 start;
2786
2787         start = *offset;
2788         lov_do_div64(start, ssize);
2789         start = start * ssize;
2790
2791         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2792                            ", end "LPU64"\n", *offset, ssize, start,
2793                            start + ssize - 1);
2794         if (cmd == OBD_CALC_STRIPE_END) {
2795                 *offset = start + ssize - 1;
2796         } else if (cmd == OBD_CALC_STRIPE_START) {
2797                 *offset = start;
2798         } else {
2799                 LBUG();
2800         }
2801
2802         RETURN(0);
2803 }
2804
2805 void lov_stripe_lock(struct lov_stripe_md *md)
2806 {
2807         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2808         cfs_spin_lock(&md->lsm_lock);
2809         LASSERT(md->lsm_lock_owner == 0);
2810         md->lsm_lock_owner = cfs_curproc_pid();
2811 }
2812 EXPORT_SYMBOL(lov_stripe_lock);
2813
2814 void lov_stripe_unlock(struct lov_stripe_md *md)
2815 {
2816         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2817         md->lsm_lock_owner = 0;
2818         cfs_spin_unlock(&md->lsm_lock);
2819 }
2820 EXPORT_SYMBOL(lov_stripe_unlock);
2821
2822 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2823                         struct obd_quotactl *oqctl)
2824 {
2825         struct lov_obd      *lov = &obd->u.lov;
2826         struct lov_tgt_desc *tgt;
2827         __u64                curspace = 0;
2828         __u64                bhardlimit = 0;
2829         int                  i, rc = 0;
2830         ENTRY;
2831
2832         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2833             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2834             oqctl->qc_cmd != Q_GETOQUOTA &&
2835             oqctl->qc_cmd != Q_INITQUOTA &&
2836             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2837             oqctl->qc_cmd != Q_FINVALIDATE) {
2838                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2839                 RETURN(-EFAULT);
2840         }
2841
2842         /* for lov tgt */
2843         obd_getref(obd);
2844         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2845                 int err;
2846
2847                 tgt = lov->lov_tgts[i];
2848
2849                 if (!tgt)
2850                         continue;
2851
2852                 if (!tgt->ltd_active || tgt->ltd_reap) {
2853                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2854                             lov->lov_tgts[i]->ltd_activate) {
2855                                 rc = -EREMOTEIO;
2856                                 CERROR("ost %d is inactive\n", i);
2857                         } else {
2858                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2859                         }
2860                         continue;
2861                 }
2862
2863                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2864                 if (err) {
2865                         if (tgt->ltd_active && !rc)
2866                                 rc = err;
2867                         continue;
2868                 }
2869
2870                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2871                         curspace += oqctl->qc_dqblk.dqb_curspace;
2872                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2873                 }
2874         }
2875         obd_putref(obd);
2876
2877         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2878                 oqctl->qc_dqblk.dqb_curspace = curspace;
2879                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2880         }
2881         RETURN(rc);
2882 }
2883
2884 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2885                           struct obd_quotactl *oqctl)
2886 {
2887         struct lov_obd *lov = &obd->u.lov;
2888         int             i, rc = 0;
2889         ENTRY;
2890
2891         obd_getref(obd);
2892
2893         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2894                 if (!lov->lov_tgts[i])
2895                         continue;
2896
2897                 /* Skip quota check on the administratively disabled OSTs. */
2898                 if (!lov->lov_tgts[i]->ltd_activate) {
2899                         CWARN("lov idx %d was administratively disabled, "
2900                               "skip quotacheck on it.\n", i);
2901                         continue;
2902                 }
2903
2904                 if (!lov->lov_tgts[i]->ltd_active) {
2905                         CERROR("lov idx %d inactive\n", i);
2906                         rc = -EIO;
2907                         goto out;
2908                 }
2909         }
2910
2911         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2912                 int err;
2913
2914                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2915                         continue;
2916
2917                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2918                 if (err && !rc)
2919                         rc = err;
2920         }
2921
2922 out:
2923         obd_putref(obd);
2924
2925         RETURN(rc);
2926 }
2927
2928 struct obd_ops lov_obd_ops = {
2929         .o_owner               = THIS_MODULE,
2930         .o_setup               = lov_setup,
2931         .o_precleanup          = lov_precleanup,
2932         .o_cleanup             = lov_cleanup,
2933         //.o_process_config      = lov_process_config,
2934         .o_connect             = lov_connect,
2935         .o_disconnect          = lov_disconnect,
2936         .o_statfs              = lov_statfs,
2937         .o_statfs_async        = lov_statfs_async,
2938         .o_packmd              = lov_packmd,
2939         .o_unpackmd            = lov_unpackmd,
2940         .o_create              = lov_create,
2941         .o_destroy             = lov_destroy,
2942         .o_getattr             = lov_getattr,
2943         .o_getattr_async       = lov_getattr_async,
2944         .o_setattr             = lov_setattr,
2945         .o_setattr_async       = lov_setattr_async,
2946         .o_brw                 = lov_brw,
2947         .o_merge_lvb           = lov_merge_lvb,
2948         .o_adjust_kms          = lov_adjust_kms,
2949         .o_punch               = lov_punch,
2950         .o_sync                = lov_sync,
2951         .o_enqueue             = lov_enqueue,
2952         .o_change_cbdata       = lov_change_cbdata,
2953         .o_find_cbdata         = lov_find_cbdata,
2954         .o_cancel              = lov_cancel,
2955         .o_cancel_unused       = lov_cancel_unused,
2956         .o_iocontrol           = lov_iocontrol,
2957         .o_get_info            = lov_get_info,
2958         .o_set_info_async      = lov_set_info_async,
2959         .o_extent_calc         = lov_extent_calc,
2960         .o_llog_init           = lov_llog_init,
2961         .o_llog_finish         = lov_llog_finish,
2962         .o_notify              = lov_notify,
2963         .o_pool_new            = lov_pool_new,
2964         .o_pool_rem            = lov_pool_remove,
2965         .o_pool_add            = lov_pool_add,
2966         .o_pool_del            = lov_pool_del,
2967         .o_getref              = lov_getref,
2968         .o_putref              = lov_putref,
2969         .o_quotactl            = lov_quotactl,
2970         .o_quotacheck          = lov_quotacheck,
2971 };
2972
2973 cfs_mem_cache_t *lov_oinfo_slab;
2974
2975 extern struct lu_kmem_descr lov_caches[];
2976
2977 int __init lov_init(void)
2978 {
2979         struct lprocfs_static_vars lvars = { 0 };
2980         int rc, rc2;
2981         ENTRY;
2982
2983         /* print an address of _any_ initialized kernel symbol from this
2984          * module, to allow debugging with gdb that doesn't support data
2985          * symbols from modules.*/
2986         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2987
2988         rc = lu_kmem_init(lov_caches);
2989         if (rc)
2990                 return rc;
2991
2992         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2993                                               sizeof(struct lov_oinfo),
2994                                               0, CFS_SLAB_HWCACHE_ALIGN);
2995         if (lov_oinfo_slab == NULL) {
2996                 lu_kmem_fini(lov_caches);
2997                 return -ENOMEM;
2998         }
2999         lprocfs_lov_init_vars(&lvars);
3000
3001         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3002                                  LUSTRE_LOV_NAME, &lov_device_type);
3003
3004         if (rc) {
3005                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3006                 LASSERT(rc2 == 0);
3007                 lu_kmem_fini(lov_caches);
3008         }
3009
3010         RETURN(rc);
3011 }
3012
3013 #ifdef __KERNEL__
3014 static void /*__exit*/ lov_exit(void)
3015 {
3016         int rc;
3017
3018         class_unregister_type(LUSTRE_LOV_NAME);
3019         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3020         LASSERT(rc == 0);
3021
3022         lu_kmem_fini(lov_caches);
3023 }
3024
3025 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3026 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3027 MODULE_LICENSE("GPL");
3028
3029 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
3030 #endif