lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * 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 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <lustre_cache.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71
  72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  73    Any function that expects lov_tgts to remain stationary must take a ref. */
  74 static void lov_getref(struct obd_device *obd)
  75 {
  76         struct lov_obd *lov = &obd->u.lov;
  77
  78         /* nobody gets through here until lov_putref is done */
  79         mutex_down(&lov->lov_lock);
  80         atomic_inc(&lov->lov_refcount);
  81         mutex_up(&lov->lov_lock);
  82         return;
  83 }
  84
  85 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  86
  87 static void lov_putref(struct obd_device *obd)
  88 {
  89         struct lov_obd *lov = &obd->u.lov;
  90         CFS_LIST_HEAD(kill);
  91         struct lov_tgt_desc *tgt;
  92
  93         mutex_down(&lov->lov_lock);
  94         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  95         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  96                 int i;
  97                 struct lov_tgt_desc *n;
  98                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  99                        lov->lov_death_row);
 100                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 101                         tgt = lov->lov_tgts[i];
 102
 103                         if (!tgt || !tgt->ltd_reap)
 104                                 continue;
 105                         list_add(&tgt->ltd_kill, &kill);
 106                         /* XXX - right now there is a dependency on ld_tgt_count
 107                          * being the maximum tgt index for computing the
 108                          * mds_max_easize. So we can't shrink it. */
 109                         lov_ost_pool_remove(&lov->lov_packed, i);
 110                         lov->lov_tgts[i] = NULL;
 111                         lov->lov_death_row--;
 112                 }
 113                 mutex_up(&lov->lov_lock);
 114
 115                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 116                         list_del(&tgt->ltd_kill);
 117                         /* Disconnect */
 118                         __lov_del_obd(obd, tgt);
 119                 }
 120         } else {
 121                 mutex_up(&lov->lov_lock);
 122         }
 123 }
 124
 125 static int lov_obd_register_page_removal_cb(struct obd_device *obd,
 126                                         obd_page_removal_cb_t func,
 127                                         obd_pin_extent_cb pin_cb)
 128 {
 129         struct lov_obd *lov = &obd->u.lov;
 130
 131         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 132                 return -EBUSY;
 133
 134         if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
 135                 return -EBUSY;
 136
 137         lov->lov_page_removal_cb = func;
 138         lov->lov_page_pin_cb = pin_cb;
 139         return 0;
 140 }
 141
 142 static int lov_obd_unregister_page_removal_cb(struct obd_device *obd,
 143                                               obd_page_removal_cb_t func)
 144 {
 145         struct lov_obd *lov = &obd->u.lov;
 146         int i, rc = 0;
 147
 148         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 149                 return -EINVAL;
 150
 151         lov->lov_page_removal_cb = NULL;
 152         lov->lov_page_pin_cb = NULL;
 153
 154         obd_getref(obd);
 155         for (i = 0; i < lov->desc.ld_tgt_count; ++i) {
 156                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 157                         continue;
 158                 rc |= obd_unregister_page_removal_cb(
 159                                 lov->lov_tgts[i]->ltd_exp->exp_obd, func);
 160         }
 161         obd_putref(obd);
 162
 163         return rc;
 164 }
 165
 166 static int lov_obd_register_lock_cancel_cb(struct obd_device *obd,
 167                                            obd_lock_cancel_cb func)
 168 {
 169         struct lov_obd *lov = &obd->u.lov;
 170
 171         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 172                 return -EBUSY;
 173
 174         lov->lov_lock_cancel_cb = func;
 175
 176         return 0;
 177 }
 178
 179 static int lov_obd_unregister_lock_cancel_cb(struct obd_device *obd,
 180                                              obd_lock_cancel_cb func)
 181 {
 182         struct lov_obd *lov = &obd->u.lov;
 183
 184         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 185                 return -EINVAL;
 186
 187         lov->lov_lock_cancel_cb = NULL;
 188         return 0;
 189
 190 }
 191
 192 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 193                               int activate);
 194
 195 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 196                       enum obd_notify_event ev, void *data)
 197 {
 198         int rc = 0;
 199         ENTRY;
 200
 201         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 202                 LASSERT(watched);
 203
 204                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 205                         CERROR("unexpected notification of %s %s!\n",
 206                                watched->obd_type->typ_name,
 207                                watched->obd_name);
 208                         RETURN(-EINVAL);
 209                 }
 210
 211                 /* Set OSC as active before notifying the observer, so the
 212                  * observer can use the OSC normally.
 213                  */
 214                 rc = lov_set_osc_active(obd, &watched->u.cli.cl_target_uuid,
 215                                         ev == OBD_NOTIFY_ACTIVE);
 216                 if (rc < 0) {
 217                         CERROR("%sactivation of %s failed: %d\n",
 218                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 219                                obd_uuid2str(&watched->u.cli.cl_target_uuid),
 220                                rc);
 221                         RETURN(rc);
 222                 }
 223                 /* active event should be pass lov target index as data */
 224                 data = &rc;
 225         }
 226
 227         /* Pass the notification up the chain. */
 228         if (watched) {
 229                 rc = obd_notify_observer(obd, watched, ev, data);
 230         } else {
 231                 /* NULL watched means all osc's in the lov (only for syncs) */
 232                 /* sync event should be send lov idx as data */
 233                 struct lov_obd *lov = &obd->u.lov;
 234                 int i, is_sync;
 235
 236                 data = &i;
 237                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 238                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 239
 240                 obd_getref(obd);
 241                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 242
 243                         if (!lov->lov_tgts[i])
 244                                 continue;
 245                         /* don't send sync event if target not
 246                          * connected/activated */
 247                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 248                                 continue;
 249
 250                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 251                                                  ev, data);
 252                         if (rc) {
 253                                 CERROR("%s: notify %s of %s failed %d\n",
 254                                        obd->obd_name,
 255                                        obd->obd_observer->obd_name,
 256                                        lov->lov_tgts[i]->ltd_obd->obd_name, rc);
 257                         }
 258                 }
 259                 obd_putref(obd);
 260         }
 261
 262         RETURN(rc);
 263 }
 264
 265 #define MAX_STRING_SIZE 128
 266 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 267                            struct obd_connect_data *data)
 268 {
 269         struct lov_obd *lov = &obd->u.lov;
 270         struct obd_uuid tgt_uuid;
 271         struct obd_device *tgt_obd;
 272         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 273         struct lustre_handle conn = {0, };
 274         struct obd_import *imp;
 275 #ifdef __KERNEL__
 276         cfs_proc_dir_entry_t *lov_proc_dir;
 277 #endif
 278         int rc;
 279         ENTRY;
 280
 281         if (!lov->lov_tgts[index])
 282                 RETURN(-EINVAL);
 283
 284         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 285         if (!tgt_obd->obd_set_up) {
 286                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 287                 RETURN(-EINVAL);
 288         }
 289
 290         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 291                 data->ocd_index = index;
 292
 293         /*
 294          * Divine LOV knows that OBDs under it are OSCs.
 295          */
 296         imp = tgt_obd->u.cli.cl_import;
 297
 298         if (activate) {
 299                 tgt_obd->obd_no_recov = 0;
 300                 /* FIXME this is probably supposed to be
 301                    ptlrpc_set_import_active.  Horrible naming. */
 302                 ptlrpc_activate_import(imp);
 303         }
 304
 305         rc = obd_register_observer(tgt_obd, obd);
 306         if (rc) {
 307                 CERROR("Target %s register_observer error %d; "
 308                         "will not be able to reactivate\n",
 309                         obd_uuid2str(&tgt_uuid), rc);
 310                 RETURN(rc);
 311         }
 312
 313         if (imp->imp_invalid) {
 314                 CERROR("not connecting OSC %s; administratively "
 315                        "disabled\n", obd_uuid2str(&tgt_uuid));
 316                 RETURN(0);
 317         }
 318         if (lov->lov_lock_cancel_cb)
 319                 rc = obd_register_lock_cancel_cb(tgt_obd, lov->lov_lock_cancel_cb);
 320                 if (rc)
 321                         RETURN(rc);
 322
 323         if (lov->lov_page_removal_cb)
 324                 rc = obd_register_page_removal_cb(tgt_obd, lov->lov_page_removal_cb,
 325                                                   lov->lov_page_pin_cb);
 326                 if (rc)
 327                         GOTO(out_lock_cb, rc);
 328
 329         rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, data, &lov->lov_tgts[index]->ltd_exp);
 330         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 331                 CERROR("Target %s connect error %d\n",
 332                        obd_uuid2str(&tgt_uuid), rc);
 333                 GOTO(out_page_cb, rc);
 334         }
 335
 336 #ifdef __KERNEL__
 337         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 338         if (lov_proc_dir) {
 339                 struct obd_device *osc_obd = class_conn2obd(&conn);
 340                 cfs_proc_dir_entry_t *osc_symlink;
 341                 char name[MAX_STRING_SIZE];
 342
 343                 LASSERT(osc_obd != NULL);
 344                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 345                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 346                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 347                          osc_obd->obd_type->typ_name,
 348                          osc_obd->obd_name);
 349                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 350                                            name);
 351                 if (osc_symlink == NULL) {
 352                         CERROR("could not register LOV target "
 353                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 354                                obd->obd_type->typ_name, obd->obd_name,
 355                                osc_obd->obd_name);
 356                         lprocfs_remove(&lov_proc_dir);
 357                 }
 358         }
 359 #endif
 360         rc = qos_add_tgt(obd, index);
 361         if (rc)
 362                 CERROR("qos_add_tgt failed %d\n", rc);
 363
 364         RETURN(0);
 365
 366 out_page_cb:
 367         obd_unregister_page_removal_cb(obd, lov->lov_page_removal_cb);
 368 out_lock_cb:
 369         obd_unregister_lock_cancel_cb(obd, lov->lov_lock_cancel_cb);
 370
 371         RETURN(rc);
 372 }
 373
 374 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
 375                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 376                        void *localdata)
 377 {
 378         struct lov_obd *lov = &obd->u.lov;
 379         struct lov_tgt_desc *tgt;
 380         struct obd_export **exp = localdata;
 381         int i, rc;
 382         ENTRY;
 383
 384         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 385
 386         rc = class_connect(conn, obd, cluuid);
 387         if (rc)
 388                 RETURN(rc);
 389
 390         *exp = class_conn2export(conn);
 391
 392         /* Why should there ever be more than 1 connect? */
 393         lov->lov_connects++;
 394         LASSERT(lov->lov_connects == 1);
 395
 396         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 397         if (data)
 398                 lov->lov_ocd = *data;
 399
 400         obd_getref(obd);
 401         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 402                 tgt = lov->lov_tgts[i];
 403                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 404                         continue;
 405                 /* Flags will be lowest common denominator */
 406                 rc = lov_connect_obd(obd, i, lov->lov_tgts[i]->ltd_activate,
 407                                      &lov->lov_ocd);
 408                 if (rc) {
 409                         CERROR("%s: lov connect tgt %d failed: %d\n",
 410                                obd->obd_name, i, rc);
 411                         continue;
 412                 }
 413                 /* connect to administrative disabled ost */
 414                 if (!lov->lov_tgts[i]->ltd_exp)
 415                         continue;
 416
 417                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 418                                 OBD_NOTIFY_CONNECT, (void *)&i);
 419                 if (rc) {
 420                         CERROR("%s error sending notify %d\n",
 421                                obd->obd_name, rc);
 422                 }
 423         }
 424         obd_putref(obd);
 425
 426         RETURN(0);
 427 }
 428
 429 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 430 {
 431         cfs_proc_dir_entry_t *lov_proc_dir;
 432         struct lov_obd *lov = &obd->u.lov;
 433         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
 434         int rc;
 435
 436         ENTRY;
 437
 438         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 439                obd->obd_name, osc_obd->obd_name);
 440
 441         if (tgt->ltd_active) {
 442                 tgt->ltd_active = 0;
 443                 lov->desc.ld_active_tgt_count--;
 444                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 445         }
 446
 447         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 448         if (lov_proc_dir) {
 449                 cfs_proc_dir_entry_t *osc_symlink;
 450
 451                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 452                 if (osc_symlink) {
 453                         lprocfs_remove(&osc_symlink);
 454                 } else {
 455                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 456                                obd->obd_type->typ_name, obd->obd_name,
 457                                osc_obd->obd_name);
 458                 }
 459         }
 460
 461         if (obd->obd_force) {
 462                 /* Pass it on to our clients.
 463                  * XXX This should be an argument to disconnect,
 464                  * XXX not a back-door flag on the OBD.  Ah well.
 465                  */
 466                 if (osc_obd)
 467                         osc_obd->obd_force = 1;
 468         }
 469
 470         obd_register_observer(osc_obd, NULL);
 471
 472         obd_unregister_page_removal_cb(osc_obd, lov->lov_page_removal_cb);
 473         obd_unregister_lock_cancel_cb(osc_obd, lov->lov_lock_cancel_cb);
 474
 475         rc = obd_disconnect(tgt->ltd_exp);
 476         if (rc) {
 477                 CERROR("Target %s disconnect error %d\n",
 478                        tgt->ltd_uuid.uuid, rc);
 479                 rc = 0;
 480         }
 481
 482         qos_del_tgt(obd, tgt);
 483
 484         tgt->ltd_exp = NULL;
 485
 486         RETURN(0);
 487 }
 488
 489 static int lov_del_target(struct obd_device *obd, __u32 index,
 490                           struct obd_uuid *uuidp, int gen);
 491
 492 static int lov_disconnect(struct obd_export *exp)
 493 {
 494         struct obd_device *obd = class_exp2obd(exp);
 495         struct lov_obd *lov = &obd->u.lov;
 496         int i, rc;
 497         ENTRY;
 498
 499         if (!lov->lov_tgts)
 500                 goto out;
 501
 502         /* Only disconnect the underlying layers on the final disconnect. */
 503         lov->lov_connects--;
 504         if (lov->lov_connects != 0) {
 505                 /* why should there be more than 1 connect? */
 506                 CERROR("disconnect #%d\n", lov->lov_connects);
 507                 goto out;
 508         }
 509
 510         /* Let's hold another reference so lov_del_obd doesn't spin through
 511            putref every time */
 512         obd_getref(obd);
 513         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 514                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 515                         /* Disconnection is the last we know about an obd */
 516                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 517                 }
 518         }
 519         obd_putref(obd);
 520
 521 out:
 522         rc = class_disconnect(exp); /* bz 9811 */
 523         RETURN(rc);
 524 }
 525
 526 /* Error codes:
 527  *
 528  *  -EINVAL  : UUID can't be found in the LOV's target list
 529  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 530  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 531  *  - any above 0 is lov index
 532  */
 533 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 534                               int activate)
 535 {
 536         struct lov_obd *lov = &obd->u.lov;
 537         struct lov_tgt_desc *tgt;
 538         int i = 0;
 539         ENTRY;
 540
 541         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 542                lov, uuid->uuid, activate);
 543
 544         obd_getref(obd);
 545         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 546                 tgt = lov->lov_tgts[i];
 547                 if (!tgt || !tgt->ltd_exp)
 548                         continue;
 549
 550                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 551                        i, obd_uuid2str(&tgt->ltd_uuid),
 552                        tgt->ltd_exp->exp_handle.h_cookie);
 553                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 554                         break;
 555         }
 556
 557         if (i == lov->desc.ld_tgt_count)
 558                 GOTO(out, i = -EINVAL);
 559
 560         if (lov->lov_tgts[i]->ltd_active == activate) {
 561                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 562                        activate ? "" : "in");
 563                 GOTO(out, i);
 564         }
 565
 566         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 567                activate ? "" : "in");
 568
 569         lov->lov_tgts[i]->ltd_active = activate;
 570
 571         if (activate) {
 572                 lov->desc.ld_active_tgt_count++;
 573                 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
 574         } else {
 575                 lov->desc.ld_active_tgt_count--;
 576                 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
 577         }
 578         /* remove any old qos penalty */
 579         lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
 580
 581  out:
 582         obd_putref(obd);
 583         RETURN(i);
 584 }
 585
 586
 587 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 588                           __u32 index, int gen, int active)
 589 {
 590         struct obd_device *tgt_obd;
 591         struct lov_obd *lov = &obd->u.lov;
 592         struct lov_tgt_desc *tgt;
 593         int rc;
 594         ENTRY;
 595
 596         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 597                uuidp->uuid, index, gen, active);
 598
 599         if (gen <= 0) {
 600                 CERROR("request to add OBD %s with invalid generation: %d\n",
 601                        uuidp->uuid, gen);
 602                 RETURN(-EINVAL);
 603         }
 604
 605
 606         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 607                                         &obd->obd_uuid);
 608         if (!tgt_obd) {
 609                 CERROR("Target %s not attached\n", obd_uuid2str(uuidp));
 610                 RETURN(-EINVAL);
 611         }
 612
 613         mutex_down(&lov->lov_lock);
 614
 615         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 616                 tgt = lov->lov_tgts[index];
 617                 CERROR("UUID %s already assigned at LOV target index %d\n",
 618                        obd_uuid2str(&tgt->ltd_uuid), index);
 619                 GOTO(err_unlock, rc = -EEXIST);
 620         }
 621
 622         if (index >= lov->lov_tgt_size) {
 623                 /* We need to reallocate the lov target array. */
 624                 struct lov_tgt_desc **newtgts, **old = NULL;
 625                 __u32 newsize, oldsize = 0;
 626
 627                 newsize = max(lov->lov_tgt_size, (__u32)2);
 628                 while (newsize < index + 1)
 629                         newsize = newsize << 1;
 630                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 631                 if (newtgts == NULL)
 632                         GOTO(err_unlock, rc = -ENOMEM);
 633
 634                 if (lov->lov_tgt_size) {
 635                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 636                                lov->lov_tgt_size);
 637                         old = lov->lov_tgts;
 638                         oldsize = lov->lov_tgt_size;
 639                 }
 640
 641                 lov->lov_tgts = newtgts;
 642                 lov->lov_tgt_size = newsize;
 643 #ifdef __KERNEL__
 644                 smp_rmb();
 645 #endif
 646                 if (old)
 647                         OBD_FREE(old, sizeof(*old) * oldsize);
 648
 649                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 650                        lov->lov_tgts, lov->lov_tgt_size);
 651         }
 652
 653         OBD_ALLOC_PTR(tgt);
 654         if (!tgt)
 655                 GOTO(err_unlock, rc = -ENOMEM);
 656
 657         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 658         if (rc)
 659                 GOTO(err_free_tgt, rc = -EEXIST);
 660
 661         memset(tgt, 0, sizeof(*tgt));
 662         tgt->ltd_obd = tgt_obd;
 663         tgt->ltd_uuid = *uuidp;
 664         /* XXX - add a sanity check on the generation number. */
 665         tgt->ltd_gen = gen;
 666         tgt->ltd_index = index;
 667         tgt->ltd_activate = active;
 668
 669         lov->lov_tgts[index] = tgt;
 670         if (index >= lov->desc.ld_tgt_count)
 671                 lov->desc.ld_tgt_count = index + 1;
 672
 673         mutex_up(&lov->lov_lock);
 674
 675         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 676                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 677
 678         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 679
 680         if (lov->lov_connects == 0) {
 681                 /* lov_connect hasn't been called yet. We'll do the
 682                    lov_connect_obd on this target when that fn first runs,
 683                    because we don't know the connect flags yet. */
 684                 RETURN(0);
 685         }
 686
 687         obd_getref(obd);
 688
 689         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 690         if (rc)
 691                 GOTO(out, rc);
 692
 693         /* connect to administrative disabled ost */
 694         if (!tgt->ltd_exp)
 695                 GOTO(out, rc = 0);
 696
 697         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 698                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 699                         (void *)&index);
 700 out:
 701         if (rc) {
 702                 /* connect or notify failed - we can try connect later
 703                  * instead of complete delete target */
 704                 CERROR("connect or notify failed (%d) for %s\n", rc,
 705                        obd_uuid2str(&tgt->ltd_uuid));
 706         }
 707         obd_putref(obd);
 708         RETURN(rc);
 709
 710 err_free_tgt:
 711         OBD_FREE_PTR(tgt);
 712 err_unlock:
 713         mutex_up(&lov->lov_lock);
 714         RETURN(rc);
 715 }
 716
 717 /* Schedule a target for deletion */
 718 static int lov_del_target(struct obd_device *obd, __u32 index,
 719                           struct obd_uuid *uuidp, int gen)
 720 {
 721         struct lov_obd *lov = &obd->u.lov;
 722         int count = lov->desc.ld_tgt_count;
 723         int rc = 0;
 724         ENTRY;
 725
 726         if (index >= count) {
 727                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 728                        index, count);
 729                 RETURN(-EINVAL);
 730         }
 731
 732         obd_getref(obd);
 733
 734         if (!lov->lov_tgts[index]) {
 735                 CERROR("LOV target at index %d is not setup.\n", index);
 736                 GOTO(out, rc = -EINVAL);
 737         }
 738
 739         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 740                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 741                        lov_uuid2str(lov, index), index,
 742                        obd_uuid2str(uuidp));
 743                 GOTO(out, rc = -EINVAL);
 744         }
 745
 746         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 747                lov_uuid2str(lov, index), index,
 748                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 749                lov->lov_tgts[index]->ltd_active);
 750
 751         lov->lov_tgts[index]->ltd_reap = 1;
 752         lov->lov_death_row++;
 753         /* we really delete it from obd_putref */
 754 out:
 755         obd_putref(obd);
 756
 757         RETURN(rc);
 758 }
 759
 760 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 761 {
 762         struct obd_device *osc_obd;
 763
 764         LASSERT(tgt);
 765         LASSERT(tgt->ltd_reap);
 766
 767         osc_obd = class_exp2obd(tgt->ltd_exp);
 768
 769         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 770                tgt->ltd_uuid.uuid,
 771                osc_obd ? osc_obd->obd_name : "<no obd>");
 772
 773         if (tgt->ltd_exp)
 774                 lov_disconnect_obd(obd, tgt);
 775
 776         OBD_FREE_PTR(tgt);
 777
 778         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 779            do it ourselves. And we can't do it from lov_cleanup,
 780            because we just lost our only reference to it. */
 781         if (osc_obd) {
 782                 /* Use lov's force/fail flags. */
 783                 osc_obd->obd_force = obd->obd_force;
 784                 osc_obd->obd_fail = obd->obd_fail;
 785                 class_manual_cleanup(osc_obd);
 786         }
 787 }
 788
 789 void lov_fix_desc_stripe_size(__u64 *val)
 790 {
 791         if (*val < PTLRPC_MAX_BRW_SIZE) {
 792                 if (*val)
 793                         LCONSOLE_WARN("Increasing default stripe size from "
 794                                       LPU64" to %u\n",*val,PTLRPC_MAX_BRW_SIZE);
 795                 *val = PTLRPC_MAX_BRW_SIZE;
 796         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 797                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 798                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 799                               "multiple of %u)\n",
 800                               *val, LOV_MIN_STRIPE_SIZE);
 801         }
 802 }
 803
 804 void lov_fix_desc_stripe_count(__u32 *val)
 805 {
 806         if (*val == 0)
 807                 *val = 1;
 808 }
 809
 810 void lov_fix_desc_pattern(__u32 *val)
 811 {
 812         /* from lov_setstripe */
 813         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 814                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 815                 *val = 0;
 816         }
 817 }
 818
 819 void lov_fix_desc_qos_maxage(__u32 *val)
 820 {
 821         /* fix qos_maxage */
 822         if (*val == 0)
 823                 *val = QOS_DEFAULT_MAXAGE;
 824 }
 825
 826 void lov_fix_desc(struct lov_desc *desc)
 827 {
 828         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 829         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 830         lov_fix_desc_pattern(&desc->ld_pattern);
 831         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 832 }
 833
 834 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
 835 {
 836         struct lprocfs_static_vars lvars = { 0 };
 837         struct lustre_cfg *lcfg = buf;
 838         struct lov_desc *desc;
 839         struct lov_obd *lov = &obd->u.lov;
 840         int rc;
 841         ENTRY;
 842
 843         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 844                 CERROR("LOV setup requires a descriptor\n");
 845                 RETURN(-EINVAL);
 846         }
 847
 848         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 849
 850         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 851                 CERROR("descriptor size wrong: %d > %d\n",
 852                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 853                 RETURN(-EINVAL);
 854         }
 855
 856         if (desc->ld_magic != LOV_DESC_MAGIC) {
 857                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 858                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 859                                    obd->obd_name, desc);
 860                             lustre_swab_lov_desc(desc);
 861                 } else {
 862                         CERROR("%s: Bad lov desc magic: %#x\n",
 863                                obd->obd_name, desc->ld_magic);
 864                         RETURN(-EINVAL);
 865                 }
 866         }
 867
 868         lov_fix_desc(desc);
 869
 870         desc->ld_active_tgt_count = 0;
 871         lov->desc = *desc;
 872         lov->lov_tgt_size = 0;
 873
 874         sema_init(&lov->lov_lock, 1);
 875         atomic_set(&lov->lov_refcount, 0);
 876         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 877         init_rwsem(&lov->lov_qos.lq_rw_sem);
 878         lov->lov_qos.lq_dirty = 1;
 879         lov->lov_qos.lq_rr.lqr_dirty = 1;
 880         lov->lov_qos.lq_reset = 1;
 881         /* Default priority is toward free space balance */
 882         lov->lov_qos.lq_prio_free = 232;
 883         /* Default threshold for rr (roughly 17%) */
 884         lov->lov_qos.lq_threshold_rr = 43;
 885         /* Init statfs fields */
 886         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 887         if (NULL == lov->lov_qos.lq_statfs_data)
 888                 RETURN(-ENOMEM);
 889         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 890
 891         lov->lov_pools_hash_body = lustre_hash_init("POOLS",
 892                                                     HASH_POOLS_CUR_BITS,
 893                                                     HASH_POOLS_MAX_BITS,
 894                                                     &pool_hash_operations, 0);
 895         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 896         lov->lov_pool_count = 0;
 897         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 898         if (rc)
 899                 RETURN(rc);
 900         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 901         if (rc) {
 902                 lov_ost_pool_free(&lov->lov_packed);
 903                 RETURN(rc);
 904         }
 905
 906         lprocfs_lov_init_vars(&lvars);
 907         lprocfs_obd_setup(obd, lvars.obd_vars);
 908 #ifdef LPROCFS
 909         {
 910                 cfs_proc_dir_entry_t *entry;
 911
 912                 entry = create_proc_entry("target_obd", 0444,
 913                                           obd->obd_proc_entry);
 914                 if (entry != NULL) {
 915                         entry->proc_fops = &lov_proc_target_fops;
 916                         entry->data = obd;
 917                 }
 918         }
 919 #endif
 920         lov->lov_pool_proc_entry = lprocfs_register("pools",
 921                                                     obd->obd_proc_entry,
 922                                                     NULL, NULL);
 923
 924         RETURN(0);
 925 }
 926
 927 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 928 {
 929         int rc = 0;
 930         ENTRY;
 931
 932         switch (stage) {
 933         case OBD_CLEANUP_EARLY: {
 934                 struct lov_obd *lov = &obd->u.lov;
 935                 int i;
 936                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 937                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 938                                 continue;
 939                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 940                                        OBD_CLEANUP_EARLY);
 941                 }
 942                 break;
 943         }
 944         case OBD_CLEANUP_EXPORTS:
 945                 break;
 946         case OBD_CLEANUP_SELF_EXP:
 947                 rc = obd_llog_finish(obd, 0);
 948                 if (rc != 0)
 949                         CERROR("failed to cleanup llogging subsystems\n");
 950                 break;
 951         case OBD_CLEANUP_OBD:
 952                 break;
 953         }
 954         RETURN(rc);
 955 }
 956
 957 static int lov_cleanup(struct obd_device *obd)
 958 {
 959         struct lov_obd *lov = &obd->u.lov;
 960         struct list_head *pos, *tmp;
 961         struct pool_desc *pool;
 962
 963         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 964                 pool = list_entry(pos, struct pool_desc, pool_list);
 965                 /* free the pool structs */
 966                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 967                 lov_pool_del(obd, pool->pool_name);
 968         }
 969
 970         lustre_hash_exit(lov->lov_pools_hash_body);
 971
 972         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 973         lov_ost_pool_free(&lov->lov_packed);
 974
 975         if (lov->lov_tgts) {
 976                 int i;
 977                 obd_getref(obd);
 978                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 979                         if (lov->lov_tgts[i]) {
 980                                 /* Inactive targets may never have connected */
 981                                 if (lov->lov_tgts[i]->ltd_active ||
 982                                     atomic_read(&lov->lov_refcount))
 983                                         /* We should never get here - these
 984                                            should have been removed in the
 985                                            disconnect. */
 986                                         CERROR("lov tgt %d not cleaned!"
 987                                                " deathrow=%d, lovrc=%d\n",
 988                                                i, lov->lov_death_row,
 989                                                atomic_read(&lov->lov_refcount));
 990                                 lov_del_target(obd, i, 0, 0);
 991                         }
 992                 }
 993                 obd_putref(obd);
 994                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 995                          lov->lov_tgt_size);
 996                 lov->lov_tgt_size = 0;
 997         }
 998
 999         /* clear pools parent proc entry only after all pools is killed */
1000         lprocfs_obd_cleanup(obd);
1001
1002         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
1003         RETURN(0);
1004 }
1005
1006 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
1007 {
1008         struct lustre_cfg *lcfg = buf;
1009         struct obd_uuid obd_uuid;
1010         int cmd;
1011         int rc = 0;
1012         ENTRY;
1013
1014         switch(cmd = lcfg->lcfg_command) {
1015         case LCFG_LOV_ADD_OBD:
1016         case LCFG_LOV_ADD_INA:
1017         case LCFG_LOV_DEL_OBD: {
1018                 __u32 index;
1019                 int gen;
1020                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
1021                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1022                         GOTO(out, rc = -EINVAL);
1023
1024                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1025
1026                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1027                         GOTO(out, rc = -EINVAL);
1028                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1029                         GOTO(out, rc = -EINVAL);
1030                 if (cmd == LCFG_LOV_ADD_OBD)
1031                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
1032                 else if (cmd == LCFG_LOV_ADD_INA)
1033                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
1034                 else
1035                         rc = lov_del_target(obd, index, &obd_uuid, gen);
1036                 GOTO(out, rc);
1037         }
1038         case LCFG_PARAM: {
1039                 struct lprocfs_static_vars lvars = { 0 };
1040                 struct lov_desc *desc = &(obd->u.lov.desc);
1041
1042                 if (!desc)
1043                         GOTO(out, rc = -EINVAL);
1044
1045                 lprocfs_lov_init_vars(&lvars);
1046
1047                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1048                                               lcfg, obd);
1049                 GOTO(out, rc);
1050         }
1051         case LCFG_POOL_NEW:
1052         case LCFG_POOL_ADD:
1053         case LCFG_POOL_DEL:
1054         case LCFG_POOL_REM:
1055                 GOTO(out, rc);
1056
1057         default: {
1058                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1059                 GOTO(out, rc = -EINVAL);
1060
1061         }
1062         }
1063 out:
1064         RETURN(rc);
1065 }
1066
1067 #ifndef log2
1068 #define log2(n) ffz(~(n))
1069 #endif
1070
1071 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1072                              struct lov_stripe_md **ea,
1073                              struct obd_trans_info *oti)
1074 {
1075         struct lov_obd *lov;
1076         struct obdo *tmp_oa;
1077         struct obd_uuid *ost_uuid = NULL;
1078         int rc = 0, i;
1079         ENTRY;
1080
1081         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1082                 src_oa->o_flags == OBD_FL_DELORPHAN);
1083
1084         lov = &export->exp_obd->u.lov;
1085
1086         OBDO_ALLOC(tmp_oa);
1087         if (tmp_oa == NULL)
1088                 RETURN(-ENOMEM);
1089
1090         if (oti->oti_ost_uuid) {
1091                 ost_uuid = oti->oti_ost_uuid;
1092                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1093                        ost_uuid->uuid);
1094         }
1095
1096         obd_getref(export->exp_obd);
1097         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1098                 struct lov_stripe_md obj_md;
1099                 struct lov_stripe_md *obj_mdp = &obj_md;
1100                 struct lov_tgt_desc *tgt;
1101                 int err;
1102
1103                 tgt = lov->lov_tgts[i];
1104                 if (!tgt)
1105                         continue;
1106
1107                 /* if called for a specific target, we don't
1108                    care if it is not active. */
1109                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1110                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1111                         continue;
1112                 }
1113
1114                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1115                         continue;
1116
1117                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1118                        obd_uuid2str(ost_uuid));
1119
1120                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1121
1122                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1123                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1124                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1125                                  tmp_oa, &obj_mdp, oti);
1126                 if (err) {
1127                         /* This export will be disabled until it is recovered,
1128                            and then orphan recovery will be completed. */
1129                         CERROR("error in orphan recovery on OST idx %d/%d: "
1130                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1131                         rc = err;
1132                 }
1133
1134                 if (ost_uuid)
1135                         break;
1136         }
1137         obd_putref(export->exp_obd);
1138
1139         OBDO_FREE(tmp_oa);
1140         RETURN(rc);
1141 }
1142
1143 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1144                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1145 {
1146         struct lov_stripe_md *obj_mdp, *lsm;
1147         struct lov_obd *lov = &exp->exp_obd->u.lov;
1148         unsigned ost_idx;
1149         int rc, i;
1150         ENTRY;
1151
1152         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1153                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1154
1155         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1156         if (obj_mdp == NULL)
1157                 RETURN(-ENOMEM);
1158
1159         ost_idx = src_oa->o_nlink;
1160         lsm = *ea;
1161         if (lsm == NULL)
1162                 GOTO(out, rc = -EINVAL);
1163         if (ost_idx >= lov->desc.ld_tgt_count ||
1164             !lov->lov_tgts[ost_idx])
1165                 GOTO(out, rc = -EINVAL);
1166
1167         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1168                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1169                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1170                                 GOTO(out, rc = -EINVAL);
1171                         break;
1172                 }
1173         }
1174         if (i == lsm->lsm_stripe_count)
1175                 GOTO(out, rc = -EINVAL);
1176
1177         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1178 out:
1179         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1180         RETURN(rc);
1181 }
1182
1183 /* the LOV expects oa->o_id to be set to the LOV object id */
1184 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1185                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1186 {
1187         struct lov_obd *lov;
1188         struct obd_info *oinfo;
1189         struct lov_request_set *set = NULL;
1190         struct lov_request *req;
1191         struct l_wait_info  lwi = { 0 };
1192         int rc = 0;
1193         ENTRY;
1194
1195         LASSERT(ea != NULL);
1196         if (exp == NULL)
1197                 RETURN(-EINVAL);
1198
1199         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1200             src_oa->o_flags == OBD_FL_DELORPHAN) {
1201                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1202                 RETURN(rc);
1203         }
1204
1205         lov = &exp->exp_obd->u.lov;
1206         if (!lov->desc.ld_active_tgt_count)
1207                 RETURN(-EIO);
1208
1209         OBD_ALLOC_PTR(oinfo);
1210         if (NULL == oinfo)
1211                 RETURN(-ENOMEM);
1212
1213         obd_getref(exp->exp_obd);
1214         /* Recreate a specific object id at the given OST index */
1215         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1216             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1217                  rc = lov_recreate(exp, src_oa, ea, oti);
1218                  GOTO(out, rc);
1219         }
1220
1221         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1222          * later in alloc_qos(), we will wait for those rpcs to complete if
1223          * the osfs age is older than 2 * qos_maxage */
1224         qos_statfs_update(exp->exp_obd,
1225                           cfs_time_shift_64(-lov->desc.ld_qos_maxage +
1226                                             OBD_STATFS_CACHE_SECONDS),
1227                           0);
1228
1229         rc = lov_prep_create_set(exp, oinfo, ea, src_oa, oti, &set);
1230         if (rc)
1231                 GOTO(out, rc);
1232
1233         list_for_each_entry(req, &set->set_list, rq_link) {
1234                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1235                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1236                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1237         }
1238
1239         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1240          * longer then this */
1241         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1242
1243         /* we not have ptlrpc set for assign set->interpret and should
1244          * be call interpret function himself. calling from cb_create_update
1245          * not permited because lov_fini_create_set can sleep for long time,
1246          * but we must avoid sleeping in ptlrpcd interpret function. */
1247         rc = lov_fini_create_set(set, ea);
1248 out:
1249         obd_putref(exp->exp_obd);
1250         OBD_FREE_PTR(oinfo);
1251         RETURN(rc);
1252 }
1253
1254 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1255 do {                                                                            \
1256         LASSERT((lsmp) != NULL);                                                \
1257         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1258                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1259                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1260                  (lsmp), (lsmp)->lsm_magic);                                    \
1261 } while (0)
1262
1263 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1264                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1265                        struct obd_export *md_exp)
1266 {
1267         struct lov_request_set *set;
1268         struct obd_info *oinfo;
1269         struct lov_request *req;
1270         struct list_head *pos;
1271         struct lov_obd *lov;
1272         int rc = 0, err = 0;
1273         ENTRY;
1274
1275         ASSERT_LSM_MAGIC(lsm);
1276
1277         if (!exp || !exp->exp_obd)
1278                 RETURN(-ENODEV);
1279
1280         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1281                 LASSERT(oti);
1282                 LASSERT(oti->oti_logcookies);
1283         }
1284
1285         OBD_ALLOC_PTR(oinfo);
1286         if (NULL == oinfo)
1287                 RETURN(-ENOMEM);
1288
1289         lov = &exp->exp_obd->u.lov;
1290         obd_getref(exp->exp_obd);
1291         rc = lov_prep_destroy_set(exp, oinfo, oa, lsm, oti, &set);
1292         if (rc)
1293                 GOTO(out, rc);
1294
1295         list_for_each (pos, &set->set_list) {
1296                 req = list_entry(pos, struct lov_request, rq_link);
1297
1298                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1299                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1300
1301                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1302                                   req->rq_oi.oi_oa, NULL, oti, NULL);
1303                 err = lov_update_common_set(set, req, err);
1304                 if (err) {
1305                         CERROR("error: destroying objid "LPX64" subobj "
1306                                LPX64" on OST idx %d: rc = %d\n",
1307                                oa->o_id, req->rq_oi.oi_oa->o_id,
1308                                req->rq_idx, err);
1309                         if (!rc)
1310                                 rc = err;
1311                 }
1312         }
1313
1314         if (rc == 0) {
1315                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1316                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1317         }
1318         err = lov_fini_destroy_set(set);
1319 out:
1320         obd_putref(exp->exp_obd);
1321         OBD_FREE_PTR(oinfo);
1322         RETURN(rc ? rc : err);
1323 }
1324
1325 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1326 {
1327         struct lov_request_set *set;
1328         struct lov_request *req;
1329         struct list_head *pos;
1330         struct lov_obd *lov;
1331         int err = 0, rc = 0;
1332         ENTRY;
1333
1334         LASSERT(oinfo);
1335         ASSERT_LSM_MAGIC(oinfo->oi_md);
1336
1337         if (!exp || !exp->exp_obd)
1338                 RETURN(-ENODEV);
1339
1340         lov = &exp->exp_obd->u.lov;
1341
1342         rc = lov_prep_getattr_set(exp, oinfo, &set);
1343         if (rc)
1344                 RETURN(rc);
1345
1346         list_for_each (pos, &set->set_list) {
1347                 req = list_entry(pos, struct lov_request, rq_link);
1348
1349                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1350                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1351                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1352
1353                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1354                                  &req->rq_oi);
1355                 err = lov_update_common_set(set, req, rc);
1356                 if (err) {
1357                         CERROR("error: getattr objid "LPX64" subobj "
1358                                LPX64" on OST idx %d: rc = %d\n",
1359                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1360                                req->rq_idx, err);
1361                         break;
1362                 }
1363         }
1364
1365         rc = lov_fini_getattr_set(set);
1366         if (err)
1367                 rc = err;
1368         RETURN(rc);
1369 }
1370
1371 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1372                                  void *data, int rc)
1373 {
1374         struct lov_request_set *lovset = (struct lov_request_set *)data;
1375         int err;
1376         ENTRY;
1377
1378         /* don't do attribute merge if this aysnc op failed */
1379         if (rc)
1380                 lovset->set_completes = 0;
1381         err = lov_fini_getattr_set(lovset);
1382         RETURN(rc ? rc : err);
1383 }
1384
1385 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1386                               struct ptlrpc_request_set *rqset)
1387 {
1388         struct lov_request_set *lovset;
1389         struct lov_obd *lov;
1390         struct list_head *pos;
1391         struct lov_request *req;
1392         int rc = 0, err;
1393         ENTRY;
1394
1395         LASSERT(oinfo);
1396         ASSERT_LSM_MAGIC(oinfo->oi_md);
1397
1398         if (!exp || !exp->exp_obd)
1399                 RETURN(-ENODEV);
1400
1401         lov = &exp->exp_obd->u.lov;
1402
1403         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1404         if (rc)
1405                 RETURN(rc);
1406
1407         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1408                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1409                oinfo->oi_md->lsm_stripe_size);
1410
1411         list_for_each (pos, &lovset->set_list) {
1412                 req = list_entry(pos, struct lov_request, rq_link);
1413
1414                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1415                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1416                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1417                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1418                                        &req->rq_oi, rqset);
1419                 if (rc) {
1420                         CERROR("error: getattr objid "LPX64" subobj "
1421                                LPX64" on OST idx %d: rc = %d\n",
1422                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1423                                req->rq_idx, rc);
1424                         GOTO(out, rc);
1425                 }
1426         }
1427
1428         if (!list_empty(&rqset->set_requests)) {
1429                 LASSERT(rc == 0);
1430                 LASSERT (rqset->set_interpret == NULL);
1431                 rqset->set_interpret = lov_getattr_interpret;
1432                 rqset->set_arg = (void *)lovset;
1433                 RETURN(rc);
1434         }
1435 out:
1436         if (rc)
1437                 lovset->set_completes = 0;
1438         err = lov_fini_getattr_set(lovset);
1439         RETURN(rc ? rc : err);
1440 }
1441
1442 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1443                        struct obd_trans_info *oti)
1444 {
1445         struct lov_request_set *set;
1446         struct lov_obd *lov;
1447         struct list_head *pos;
1448         struct lov_request *req;
1449         int err = 0, rc = 0;
1450         ENTRY;
1451
1452         LASSERT(oinfo);
1453         ASSERT_LSM_MAGIC(oinfo->oi_md);
1454
1455         if (!exp || !exp->exp_obd)
1456                 RETURN(-ENODEV);
1457
1458         /* for now, we only expect the following updates here */
1459         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1460                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1461                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1462                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1463                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1464                                             OBD_MD_FLGID | OBD_MD_FLFID |
1465                                             OBD_MD_FLGENER)));
1466         lov = &exp->exp_obd->u.lov;
1467         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1468         if (rc)
1469                 RETURN(rc);
1470
1471         list_for_each (pos, &set->set_list) {
1472                 req = list_entry(pos, struct lov_request, rq_link);
1473
1474                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1475                                  &req->rq_oi, NULL);
1476                 err = lov_update_setattr_set(set, req, rc);
1477                 if (err) {
1478                         CERROR("error: setattr objid "LPX64" subobj "
1479                                LPX64" on OST idx %d: rc = %d\n",
1480                                set->set_oi->oi_oa->o_id,
1481                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1482                         if (!rc)
1483                                 rc = err;
1484                 }
1485         }
1486         err = lov_fini_setattr_set(set);
1487         if (!rc)
1488                 rc = err;
1489         RETURN(rc);
1490 }
1491
1492 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1493                                  void *data, int rc)
1494 {
1495         struct lov_request_set *lovset = (struct lov_request_set *)data;
1496         int err;
1497         ENTRY;
1498
1499         if (rc)
1500                 lovset->set_completes = 0;
1501         err = lov_fini_setattr_set(lovset);
1502         RETURN(rc ? rc : err);
1503 }
1504
1505 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1506    needed. Otherwise, a client is waiting for responses. */
1507 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1508                              struct obd_trans_info *oti,
1509                              struct ptlrpc_request_set *rqset)
1510 {
1511         struct lov_request_set *set;
1512         struct lov_request *req;
1513         struct list_head *pos;
1514         struct lov_obd *lov;
1515         int rc = 0;
1516         ENTRY;
1517
1518         LASSERT(oinfo);
1519         ASSERT_LSM_MAGIC(oinfo->oi_md);
1520         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1521                 LASSERT(oti);
1522                 LASSERT(oti->oti_logcookies);
1523         }
1524
1525         if (!exp || !exp->exp_obd)
1526                 RETURN(-ENODEV);
1527
1528         lov = &exp->exp_obd->u.lov;
1529         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1530         if (rc)
1531                 RETURN(rc);
1532
1533         CDEBUG(D_INFO, "objid "LPX64"@"LPX64": %ux%u byte stripes\n",
1534                oinfo->oi_md->lsm_object_id,
1535                oinfo->oi_md->lsm_object_gr,
1536                oinfo->oi_md->lsm_stripe_count,
1537                oinfo->oi_md->lsm_stripe_size);
1538
1539         list_for_each (pos, &set->set_list) {
1540                 req = list_entry(pos, struct lov_request, rq_link);
1541
1542                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1543                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1544
1545                 CDEBUG(D_INFO, "objid "LPX64"@"LPX64"[%d] has subobj "LPX64
1546                        " at idx %u\n", oinfo->oi_oa->o_id, oinfo->oi_oa->o_gr,
1547                        req->rq_stripe, req->rq_oi.oi_oa->o_id, req->rq_idx);
1548
1549                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1550                                        &req->rq_oi, oti, rqset);
1551                 if (rc) {
1552                         CERROR("error: setattr objid "LPX64" subobj "
1553                                LPX64" on OST idx %d: rc = %d\n",
1554                                set->set_oi->oi_oa->o_id,
1555                                req->rq_oi.oi_oa->o_id,
1556                                req->rq_idx, rc);
1557                         break;
1558                 }
1559         }
1560
1561         /* If we are not waiting for responses on async requests, return. */
1562         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1563                 int err;
1564                 if (rc)
1565                         set->set_completes = 0;
1566                 err = lov_fini_setattr_set(set);
1567                 RETURN(rc ? rc : err);
1568         }
1569
1570         LASSERT(rqset->set_interpret == NULL);
1571         rqset->set_interpret = lov_setattr_interpret;
1572         rqset->set_arg = (void *)set;
1573
1574         RETURN(0);
1575 }
1576
1577 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1578                                void *data, int rc)
1579 {
1580         struct lov_request_set *lovset = (struct lov_request_set *)data;
1581         int err;
1582         ENTRY;
1583
1584         if (rc)
1585                 lovset->set_completes = 0;
1586         err = lov_fini_punch_set(lovset);
1587         RETURN(rc ? rc : err);
1588 }
1589
1590 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1591  * we can send this 'punch' to just the authoritative node and the nodes
1592  * that the punch will affect. */
1593 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1594                      struct obd_trans_info *oti,
1595                      struct ptlrpc_request_set *rqset)
1596 {
1597         struct lov_request_set *set;
1598         struct lov_obd *lov;
1599         struct list_head *pos;
1600         struct lov_request *req;
1601         int rc = 0;
1602         ENTRY;
1603
1604         LASSERT(oinfo);
1605         ASSERT_LSM_MAGIC(oinfo->oi_md);
1606
1607         if (!exp || !exp->exp_obd)
1608                 RETURN(-ENODEV);
1609
1610         lov = &exp->exp_obd->u.lov;
1611         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1612         if (rc)
1613                 RETURN(rc);
1614
1615         list_for_each (pos, &set->set_list) {
1616                 req = list_entry(pos, struct lov_request, rq_link);
1617
1618                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1619                                &req->rq_oi, NULL, rqset);
1620                 if (rc) {
1621                         CERROR("error: punch objid "LPX64" subobj "LPX64
1622                                " on OST idx %d: rc = %d\n",
1623                                set->set_oi->oi_oa->o_id,
1624                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1625                         break;
1626                 }
1627         }
1628
1629         if (rc || list_empty(&rqset->set_requests)) {
1630                 int err;
1631                 err = lov_fini_punch_set(set);
1632                 RETURN(rc ? rc : err);
1633         }
1634
1635         LASSERT(rqset->set_interpret == NULL);
1636         rqset->set_interpret = lov_punch_interpret;
1637         rqset->set_arg = (void *)set;
1638
1639         RETURN(0);
1640 }
1641
1642 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1643                               void *data, int rc)
1644 {
1645         struct lov_request_set *lovset = (struct lov_request_set *)data;
1646         int err;
1647         ENTRY;
1648
1649         if (rc)
1650                 lovset->set_completes = 0;
1651         err = lov_fini_sync_set(lovset);
1652         RETURN(rc ? rc : err);
1653 }
1654
1655 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1656                     obd_off start, obd_off end,
1657                     struct ptlrpc_request_set *rqset)
1658 {
1659         struct lov_request_set *set = NULL;
1660         struct lov_obd *lov;
1661         struct list_head *pos;
1662         struct lov_request *req;
1663         int    rc = 0;
1664         ENTRY;
1665
1666         ASSERT_LSM_MAGIC(oinfo->oi_md);
1667         LASSERT(rqset != NULL);
1668
1669         if (!exp->exp_obd)
1670                 RETURN(-ENODEV);
1671
1672         lov = &exp->exp_obd->u.lov;
1673         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1674         if (rc)
1675                 RETURN(rc);
1676
1677         list_for_each (pos, &set->set_list) {
1678                 req = list_entry(pos, struct lov_request, rq_link);
1679
1680                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1681                               req->rq_oi.oi_policy.l_extent.start,
1682                               req->rq_oi.oi_policy.l_extent.end, rqset);
1683                 if (rc) {
1684                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1685                                " on OST idx %d: rc = %d\n",
1686                                set->set_oi->oi_oa->o_id,
1687                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1688                         break;
1689                 }
1690         }
1691
1692         /* If we are not waiting for responses on async requests, return. */
1693         if (rc || list_empty(&rqset->set_requests)) {
1694                 int err = lov_fini_sync_set(set);
1695                 RETURN(rc ? rc : err);
1696         }
1697
1698         LASSERT(rqset->set_interpret == NULL);
1699         rqset->set_interpret = lov_sync_interpret;
1700         rqset->set_arg = (void *)set;
1701
1702         RETURN(0);
1703 }
1704
1705 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1706                          obd_count oa_bufs, struct brw_page *pga)
1707 {
1708         struct obd_info oinfo = { { { 0 } } };
1709         int i, rc = 0;
1710
1711         oinfo.oi_oa = lov_oinfo->oi_oa;
1712
1713         /* The caller just wants to know if there's a chance that this
1714          * I/O can succeed */
1715         for (i = 0; i < oa_bufs; i++) {
1716                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1717                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1718                 obd_off start, end;
1719
1720                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1721                                            pga[i].off + pga[i].count - 1,
1722                                            &start, &end))
1723                         continue;
1724
1725                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1726                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1727                         return -EIO;
1728                 }
1729
1730                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1731                              1, &pga[i], NULL);
1732                 if (rc)
1733                         break;
1734         }
1735         return rc;
1736 }
1737
1738 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1739                    obd_count oa_bufs, struct brw_page *pga,
1740                    struct obd_trans_info *oti)
1741 {
1742         struct lov_request_set *set;
1743         struct lov_request *req;
1744         struct list_head *pos;
1745         struct lov_obd *lov = &exp->exp_obd->u.lov;
1746         int err, rc = 0;
1747         ENTRY;
1748
1749         ASSERT_LSM_MAGIC(oinfo->oi_md);
1750
1751         if (cmd == OBD_BRW_CHECK) {
1752                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1753                 RETURN(rc);
1754         }
1755
1756         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1757         if (rc)
1758                 RETURN(rc);
1759
1760         list_for_each (pos, &set->set_list) {
1761                 struct obd_export *sub_exp;
1762                 struct brw_page *sub_pga;
1763                 req = list_entry(pos, struct lov_request, rq_link);
1764
1765                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1766                 sub_pga = set->set_pga + req->rq_pgaidx;
1767                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1768                              sub_pga, oti);
1769                 if (rc)
1770                         break;
1771                 lov_update_common_set(set, req, rc);
1772         }
1773
1774         err = lov_fini_brw_set(set);
1775         if (!rc)
1776                 rc = err;
1777         RETURN(rc);
1778 }
1779
1780 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1781                              int rc)
1782 {
1783         struct lov_request_set *lovset = (struct lov_request_set *)data;
1784         ENTRY;
1785
1786         if (rc) {
1787                 lovset->set_completes = 0;
1788                 lov_fini_brw_set(lovset);
1789         } else {
1790                 rc = lov_fini_brw_set(lovset);
1791         }
1792
1793         RETURN(rc);
1794 }
1795
1796 static int lov_brw_async(int cmd, struct obd_export *exp,
1797                          struct obd_info *oinfo, obd_count oa_bufs,
1798                          struct brw_page *pga, struct obd_trans_info *oti,
1799                          struct ptlrpc_request_set *set, int pshift)
1800 {
1801         struct lov_request_set *lovset;
1802         struct lov_request *req;
1803         struct list_head *pos;
1804         struct lov_obd *lov = &exp->exp_obd->u.lov;
1805         int rc = 0;
1806         ENTRY;
1807
1808         LASSERT(oinfo);
1809         ASSERT_LSM_MAGIC(oinfo->oi_md);
1810
1811         if (cmd == OBD_BRW_CHECK) {
1812                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1813                 RETURN(rc);
1814         }
1815
1816         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1817         if (rc)
1818                 RETURN(rc);
1819
1820         list_for_each (pos, &lovset->set_list) {
1821                 struct obd_export *sub_exp;
1822                 struct brw_page *sub_pga;
1823                 req = list_entry(pos, struct lov_request, rq_link);
1824
1825                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1826                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1827                 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1828                                    sub_pga, oti, set, pshift);
1829                 if (rc)
1830                         GOTO(out, rc);
1831                 lov_update_common_set(lovset, req, rc);
1832         }
1833         LASSERT(rc == 0);
1834         LASSERT(set->set_interpret == NULL);
1835         LASSERT(set->set_arg == NULL);
1836         rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1837         if (rc)
1838                 GOTO(out, rc);
1839
1840         RETURN(rc);
1841 out:
1842         lov_fini_brw_set(lovset);
1843         RETURN(rc);
1844 }
1845
1846 static int lov_ap_make_ready(void *data, int cmd)
1847 {
1848         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1849
1850         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1851 }
1852
1853 static int lov_ap_refresh_count(void *data, int cmd)
1854 {
1855         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1856
1857         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1858                                                      cmd);
1859 }
1860
1861 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1862 {
1863         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1864
1865         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1866         /* XXX woah, shouldn't we be altering more here?  size? */
1867         oa->o_id = lap->lap_loi_id;
1868         oa->o_stripe_idx = lap->lap_stripe;
1869 }
1870
1871 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1872                                obd_valid valid)
1873 {
1874         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1875
1876         lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1877 }
1878
1879 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1880 {
1881         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1882
1883         /* in a raid1 regime this would down a count of many ios
1884          * in flight, onl calling the caller_ops completion when all
1885          * the raid1 ios are complete */
1886         rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1887         return rc;
1888 }
1889
1890 static struct obd_async_page_ops lov_async_page_ops = {
1891         .ap_make_ready =        lov_ap_make_ready,
1892         .ap_refresh_count =     lov_ap_refresh_count,
1893         .ap_fill_obdo =         lov_ap_fill_obdo,
1894         .ap_update_obdo =       lov_ap_update_obdo,
1895         .ap_completion =        lov_ap_completion,
1896 };
1897
1898 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1899                         struct lov_oinfo *loi, cfs_page_t *page,
1900                         obd_off offset, struct obd_async_page_ops *ops,
1901                         void *data, void **res, int flags,
1902                         struct lustre_handle *lockh)
1903 {
1904         struct lov_obd *lov = &exp->exp_obd->u.lov;
1905         struct lov_async_page *lap;
1906         struct lov_lock_handles *lov_lockh = NULL;
1907         int rc = 0;
1908         ENTRY;
1909
1910         if (!page) {
1911                 int i = 0;
1912                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1913                    Only because of this layering limitation will a client
1914                    mount with no osts fail */
1915                 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1916                        !lov->lov_tgts[i]->ltd_exp) {
1917                         i++;
1918                         if (i >= lov->desc.ld_tgt_count)
1919                                 RETURN(-ENOMEDIUM);
1920                 }
1921                 rc = size_round(sizeof(*lap)) +
1922                         obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1923                                             NULL, NULL, 0, NULL, NULL, NULL, 0,
1924                                             NULL);
1925                 RETURN(rc);
1926         }
1927         ASSERT_LSM_MAGIC(lsm);
1928         LASSERT(loi == NULL);
1929
1930         lap = *res;
1931         lap->lap_magic = LOV_AP_MAGIC;
1932         lap->lap_caller_ops = ops;
1933         lap->lap_caller_data = data;
1934
1935         /* for now only raid 0 which passes through */
1936         lap->lap_stripe = lov_stripe_number(lsm, offset);
1937         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1938         loi = lsm->lsm_oinfo[lap->lap_stripe];
1939
1940         /* so the callback doesn't need the lsm */
1941         lap->lap_loi_id = loi->loi_id;
1942
1943         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1944
1945         if (lockh && lustre_handle_is_used(lockh) && !(flags & OBD_FAST_LOCK)) {
1946                 lov_lockh = lov_handle2llh(lockh);
1947                 if (lov_lockh) {
1948                         lockh = lov_lockh->llh_handles + lap->lap_stripe;
1949                 }
1950         }
1951
1952         rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1953                                  lsm, loi, page, lap->lap_sub_offset,
1954                                  &lov_async_page_ops, lap,
1955                                  &lap->lap_sub_cookie, flags, lockh);
1956         if (lov_lockh)
1957                 lov_llh_put(lov_lockh);
1958         if (rc)
1959                 RETURN(rc);
1960         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1961                lap->lap_sub_cookie, offset);
1962         RETURN(0);
1963 }
1964
1965 static int lov_queue_async_io(struct obd_export *exp,
1966                               struct lov_stripe_md *lsm,
1967                               struct lov_oinfo *loi, void *cookie,
1968                               int cmd, obd_off off, int count,
1969                               obd_flag brw_flags, obd_flag async_flags)
1970 {
1971         struct lov_obd *lov = &exp->exp_obd->u.lov;
1972         struct lov_async_page *lap;
1973         int rc;
1974
1975         LASSERT(loi == NULL);
1976
1977         ASSERT_LSM_MAGIC(lsm);
1978
1979         lap = LAP_FROM_COOKIE(cookie);
1980
1981         loi = lsm->lsm_oinfo[lap->lap_stripe];
1982
1983         rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1984                                 loi, lap->lap_sub_cookie, cmd, off, count,
1985                                 brw_flags, async_flags);
1986         RETURN(rc);
1987 }
1988
1989 static int lov_set_async_flags(struct obd_export *exp,
1990                                struct lov_stripe_md *lsm,
1991                                struct lov_oinfo *loi, void *cookie,
1992                                obd_flag async_flags)
1993 {
1994         struct lov_obd *lov = &exp->exp_obd->u.lov;
1995         struct lov_async_page *lap;
1996         int rc;
1997
1998         LASSERT(loi == NULL);
1999
2000         ASSERT_LSM_MAGIC(lsm);
2001
2002         lap = LAP_FROM_COOKIE(cookie);
2003
2004         loi = lsm->lsm_oinfo[lap->lap_stripe];
2005
2006         rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2007                                  lsm, loi, lap->lap_sub_cookie, async_flags);
2008         RETURN(rc);
2009 }
2010
2011 static int lov_queue_group_io(struct obd_export *exp,
2012                               struct lov_stripe_md *lsm,
2013                               struct lov_oinfo *loi,
2014                               struct obd_io_group *oig, void *cookie,
2015                               int cmd, obd_off off, int count,
2016                               obd_flag brw_flags, obd_flag async_flags)
2017 {
2018         struct lov_obd *lov = &exp->exp_obd->u.lov;
2019         struct lov_async_page *lap;
2020         int rc;
2021
2022         LASSERT(loi == NULL);
2023
2024         ASSERT_LSM_MAGIC(lsm);
2025
2026         lap = LAP_FROM_COOKIE(cookie);
2027
2028         loi = lsm->lsm_oinfo[lap->lap_stripe];
2029
2030         rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
2031                                 loi, oig, lap->lap_sub_cookie, cmd, off, count,
2032                                 brw_flags, async_flags);
2033         RETURN(rc);
2034 }
2035
2036 /* this isn't exactly optimal.  we may have queued sync io in oscs on
2037  * all stripes, but we don't record that fact at queue time.  so we
2038  * trigger sync io on all stripes. */
2039 static int lov_trigger_group_io(struct obd_export *exp,
2040                                 struct lov_stripe_md *lsm,
2041                                 struct lov_oinfo *loi,
2042                                 struct obd_io_group *oig)
2043 {
2044         struct lov_obd *lov = &exp->exp_obd->u.lov;
2045         int rc = 0, i, err;
2046
2047         LASSERT(loi == NULL);
2048
2049         ASSERT_LSM_MAGIC(lsm);
2050
2051         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2052                 loi = lsm->lsm_oinfo[i];
2053                 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2054                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2055                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2056                         continue;
2057                 }
2058
2059                 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2060                                            lsm, loi, oig);
2061                 if (rc == 0 && err != 0)
2062                         rc = err;
2063         };
2064         RETURN(rc);
2065 }
2066
2067 static int lov_teardown_async_page(struct obd_export *exp,
2068                                    struct lov_stripe_md *lsm,
2069                                    struct lov_oinfo *loi, void *cookie)
2070 {
2071         struct lov_obd *lov = &exp->exp_obd->u.lov;
2072         struct lov_async_page *lap;
2073         int rc;
2074
2075         LASSERT(loi == NULL);
2076
2077         ASSERT_LSM_MAGIC(lsm);
2078
2079         lap = LAP_FROM_COOKIE(cookie);
2080
2081         loi = lsm->lsm_oinfo[lap->lap_stripe];
2082
2083         rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2084                                      lsm, loi, lap->lap_sub_cookie);
2085         if (rc) {
2086                 CERROR("unable to teardown sub cookie %p: %d\n",
2087                        lap->lap_sub_cookie, rc);
2088                 RETURN(rc);
2089         }
2090         RETURN(rc);
2091 }
2092
2093 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2094                                  void *data, int rc)
2095 {
2096         struct lov_request_set *lovset = (struct lov_request_set *)data;
2097         ENTRY;
2098         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2099         RETURN(rc);
2100 }
2101
2102 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2103                        struct ldlm_enqueue_info *einfo,
2104                        struct ptlrpc_request_set *rqset)
2105 {
2106         ldlm_mode_t mode = einfo->ei_mode;
2107         struct lov_request_set *set;
2108         struct lov_request *req;
2109         struct list_head *pos;
2110         struct lov_obd *lov;
2111         ldlm_error_t rc;
2112         ENTRY;
2113
2114         LASSERT(oinfo);
2115         ASSERT_LSM_MAGIC(oinfo->oi_md);
2116         LASSERT(mode == (mode & -mode));
2117
2118         /* we should never be asked to replay a lock this way. */
2119         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2120
2121         if (!exp || !exp->exp_obd)
2122                 RETURN(-ENODEV);
2123
2124         lov = &exp->exp_obd->u.lov;
2125         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2126         if (rc)
2127                 RETURN(rc);
2128
2129         list_for_each (pos, &set->set_list) {
2130                 req = list_entry(pos, struct lov_request, rq_link);
2131
2132                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2133                                  &req->rq_oi, einfo, rqset);
2134                 if (rc != ELDLM_OK)
2135                         GOTO(out, rc);
2136         }
2137
2138         if (rqset && !list_empty(&rqset->set_requests)) {
2139                 LASSERT(rc == 0);
2140                 LASSERT(rqset->set_interpret == NULL);
2141                 rqset->set_interpret = lov_enqueue_interpret;
2142                 rqset->set_arg = (void *)set;
2143                 RETURN(rc);
2144         }
2145 out:
2146         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2147         RETURN(rc);
2148 }
2149
2150 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2151                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2152                      int *flags, void *data, struct lustre_handle *lockh,
2153                      int *n_matches)
2154 {
2155         struct lov_request_set *set;
2156         struct obd_info oinfo;
2157         struct lov_request *req;
2158         struct list_head *pos;
2159         struct lov_obd *lov;
2160         struct lustre_handle *lov_lockhp;
2161         int lov_flags, rc = 0;
2162         ENTRY;
2163
2164         ASSERT_LSM_MAGIC(lsm);
2165         LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2166
2167         if (!exp || !exp->exp_obd)
2168                 RETURN(-ENODEV);
2169
2170         lov = &exp->exp_obd->u.lov;
2171         rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2172         if (rc)
2173                 RETURN(rc);
2174
2175         list_for_each (pos, &set->set_list) {
2176                 ldlm_policy_data_t sub_policy;
2177                 req = list_entry(pos, struct lov_request, rq_link);
2178                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2179                 LASSERT(lov_lockhp);
2180
2181                 lov_flags = *flags;
2182                 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2183
2184                 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2185                                req->rq_oi.oi_md, type, &sub_policy,
2186                                mode, &lov_flags, data, lov_lockhp,
2187                                n_matches);
2188                 rc = lov_update_match_set(set, req, rc);
2189                 if (rc <= 0)
2190                         break;
2191         }
2192         lov_fini_match_set(set, mode, *flags);
2193         RETURN(rc);
2194 }
2195
2196 static int lov_change_cbdata(struct obd_export *exp,
2197                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
2198                              void *data)
2199 {
2200         struct lov_obd *lov;
2201         struct lov_oinfo *loi;
2202         int rc = 0, i;
2203         ENTRY;
2204
2205         ASSERT_LSM_MAGIC(lsm);
2206
2207         if (!exp || !exp->exp_obd)
2208                 RETURN(-ENODEV);
2209
2210         lov = &exp->exp_obd->u.lov;
2211         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2212                 struct lov_stripe_md submd;
2213
2214                 loi = lsm->lsm_oinfo[i];
2215                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2216                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2217                         continue;
2218                 }
2219                 submd.lsm_object_id = loi->loi_id;
2220                 submd.lsm_object_gr = loi->loi_gr;
2221                 submd.lsm_stripe_count = 0;
2222                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2223                                        &submd, it, data);
2224         }
2225         RETURN(rc);
2226 }
2227
2228 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2229                       __u32 mode, struct lustre_handle *lockh, int flags,
2230                       obd_off end)
2231 {
2232         struct lov_request_set *set;
2233         struct obd_info oinfo;
2234         struct lov_request *req;
2235         struct list_head *pos;
2236         struct lov_obd *lov;
2237         struct lustre_handle *lov_lockhp;
2238         ldlm_mode_t this_mode;
2239         int err = 0, rc = 0;
2240         ENTRY;
2241
2242         ASSERT_LSM_MAGIC(lsm);
2243
2244         if (!exp || !exp->exp_obd)
2245                 RETURN(-ENODEV);
2246
2247         LASSERT(lockh);
2248         lov = &exp->exp_obd->u.lov;
2249         if (flags & OBD_FAST_LOCK) {
2250                 int stripe = lov_stripe_number(lsm, end);
2251                 RETURN(obd_cancel(lov->lov_tgts[lsm->lsm_oinfo[stripe]->
2252                                   loi_ost_idx]->ltd_exp, NULL, mode, lockh,
2253                                   flags, end));
2254         }
2255
2256         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2257         if (rc)
2258                 RETURN(rc);
2259
2260         list_for_each (pos, &set->set_list) {
2261                 req = list_entry(pos, struct lov_request, rq_link);
2262                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2263
2264                 /* If this lock was used for a write or truncate, the object
2265                  * will have been recreated by the OST, cancel the lock
2266                  * (setting LCK_GROUP incidentally causes immediate cancel). */
2267                 if (OST_LVB_IS_ERR(lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_blocks) &&
2268                     (mode == LCK_PW || mode == LCK_CW))
2269                         this_mode = LCK_GROUP;
2270                 else
2271                         this_mode = mode;
2272
2273                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2274                                 req->rq_oi.oi_md, this_mode, lov_lockhp, flags,
2275                                 end);
2276                 rc = lov_update_common_set(set, req, rc);
2277                 if (rc) {
2278                         CERROR("error: cancel objid "LPX64" subobj "
2279                                LPX64" on OST idx %d: rc = %d\n",
2280                                lsm->lsm_object_id,
2281                                req->rq_oi.oi_md->lsm_object_id,
2282                                req->rq_idx, rc);
2283                         err = rc;
2284                 }
2285
2286         }
2287         lov_fini_cancel_set(set);
2288         RETURN(err);
2289 }
2290
2291 static int lov_cancel_unused(struct obd_export *exp,
2292                              struct lov_stripe_md *lsm, int flags, void *opaque)
2293 {
2294         struct lov_obd *lov;
2295         struct lov_oinfo *loi;
2296         int rc = 0, i;
2297         ENTRY;
2298
2299         if (!exp || !exp->exp_obd)
2300                 RETURN(-ENODEV);
2301
2302         lov = &exp->exp_obd->u.lov;
2303         if (lsm == NULL) {
2304                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2305                         int err;
2306                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2307                                 continue;
2308
2309                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2310                                                 flags, opaque);
2311                         if (!rc)
2312                                 rc = err;
2313                 }
2314                 RETURN(rc);
2315         }
2316
2317         ASSERT_LSM_MAGIC(lsm);
2318
2319         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2320                 struct lov_stripe_md submd;
2321                 int err;
2322
2323                 loi = lsm->lsm_oinfo[i];
2324                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2325                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2326                         continue;
2327                 }
2328
2329                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2330                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2331
2332                 submd.lsm_object_id = loi->loi_id;
2333                 submd.lsm_stripe_count = 0;
2334                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2335                                         &submd, flags, opaque);
2336                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2337                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2338                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2339                                loi->loi_id, loi->loi_ost_idx, err);
2340                         if (!rc)
2341                                 rc = err;
2342                 }
2343         }
2344         RETURN(rc);
2345 }
2346
2347 static int lov_join_lru(struct obd_export *exp,
2348                         struct lov_stripe_md *lsm, int join)
2349 {
2350         struct lov_obd *lov;
2351         struct lov_oinfo *loi;
2352         int i, count = 0;
2353         ENTRY;
2354
2355         ASSERT_LSM_MAGIC(lsm);
2356         if (!exp || !exp->exp_obd)
2357                 RETURN(-ENODEV);
2358
2359         lov = &exp->exp_obd->u.lov;
2360         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2361                 struct lov_stripe_md submd;
2362                 int rc = 0;
2363
2364                 loi = lsm->lsm_oinfo[i];
2365                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2366                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2367                         continue;
2368                 }
2369
2370                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2371                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2372
2373                 submd.lsm_object_id = loi->loi_id;
2374                 submd.lsm_stripe_count = 0;
2375                 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2376                                   &submd, join);
2377                 if (rc < 0) {
2378                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2379                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2380                                loi->loi_id, loi->loi_ost_idx, rc);
2381                         return rc;
2382                 } else {
2383                         count += rc;
2384                 }
2385         }
2386         RETURN(count);
2387 }
2388
2389 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
2390 {
2391         struct lov_request_set *lovset = (struct lov_request_set *)data;
2392         int err;
2393         ENTRY;
2394
2395         if (rc)
2396                 lovset->set_completes = 0;
2397
2398         err = lov_fini_statfs_set(lovset);
2399         RETURN(rc ? rc : err);
2400 }
2401
2402 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2403                             __u64 max_age, struct ptlrpc_request_set *rqset)
2404 {
2405         struct lov_request_set *set;
2406         struct lov_request *req;
2407         struct list_head *pos;
2408         struct lov_obd *lov;
2409         int rc = 0;
2410         ENTRY;
2411
2412         LASSERT(oinfo != NULL);
2413         LASSERT(oinfo->oi_osfs != NULL);
2414
2415         lov = &obd->u.lov;
2416         rc = lov_prep_statfs_set(obd, oinfo, &set);
2417         if (rc)
2418                 RETURN(rc);
2419
2420         list_for_each (pos, &set->set_list) {
2421                 struct obd_device *osc_obd;
2422
2423                 req = list_entry(pos, struct lov_request, rq_link);
2424
2425                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2426                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2427                 if (rc)
2428                         break;
2429         }
2430
2431         if (rc || list_empty(&rqset->set_requests)) {
2432                 int err;
2433                 if (rc)
2434                         set->set_completes = 0;
2435                 err = lov_fini_statfs_set(set);
2436                 RETURN(rc ? rc : err);
2437         }
2438
2439         LASSERT(rqset->set_interpret == NULL);
2440         rqset->set_interpret = lov_statfs_interpret;
2441         rqset->set_arg = (void *)set;
2442         RETURN(0);
2443 }
2444
2445 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2446                       __u64 max_age, __u32 flags)
2447 {
2448         struct ptlrpc_request_set *set = NULL;
2449         struct obd_info oinfo = { { { 0 } } };
2450         int rc = 0;
2451         ENTRY;
2452
2453         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2454          * statfs requests */
2455         set = ptlrpc_prep_set();
2456         if (set == NULL)
2457                 RETURN(-ENOMEM);
2458
2459         oinfo.oi_osfs = osfs;
2460         oinfo.oi_flags = flags;
2461         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2462         if (rc == 0)
2463                 rc = ptlrpc_set_wait(set);
2464         ptlrpc_set_destroy(set);
2465
2466         RETURN(rc);
2467 }
2468
2469 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2470                          void *karg, void *uarg)
2471 {
2472         struct obd_device *obddev = class_exp2obd(exp);
2473         struct lov_obd *lov = &obddev->u.lov;
2474         int i, rc, count = lov->desc.ld_tgt_count;
2475         struct obd_uuid *uuidp;
2476         ENTRY;
2477
2478         switch (cmd) {
2479         case OBD_IOC_LOV_GET_CONFIG: {
2480                 struct obd_ioctl_data *data;
2481                 struct lov_desc *desc;
2482                 char *buf = NULL;
2483                 __u32 *genp;
2484
2485                 len = 0;
2486                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2487                         RETURN(-EINVAL);
2488
2489                 data = (struct obd_ioctl_data *)buf;
2490
2491                 if (sizeof(*desc) > data->ioc_inllen1) {
2492                         obd_ioctl_freedata(buf, len);
2493                         RETURN(-EINVAL);
2494                 }
2495
2496                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2497                         obd_ioctl_freedata(buf, len);
2498                         RETURN(-EINVAL);
2499                 }
2500
2501                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2502                         obd_ioctl_freedata(buf, len);
2503                         RETURN(-EINVAL);
2504                 }
2505
2506                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2507                 memcpy(desc, &(lov->desc), sizeof(*desc));
2508
2509                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2510                 genp = (__u32 *)data->ioc_inlbuf3;
2511                 /* the uuid will be empty for deleted OSTs */
2512                 for (i = 0; i < count; i++, uuidp++, genp++) {
2513                         if (!lov->lov_tgts[i])
2514                                 continue;
2515                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2516                         *genp = lov->lov_tgts[i]->ltd_gen;
2517                 }
2518
2519                 rc = copy_to_user((void *)uarg, buf, len);
2520                 if (rc)
2521                         rc = -EFAULT;
2522                 obd_ioctl_freedata(buf, len);
2523                 break;
2524         }
2525         case LL_IOC_LOV_SETSTRIPE:
2526                 rc = lov_setstripe(exp, karg, uarg);
2527                 break;
2528         case LL_IOC_LOV_GETSTRIPE:
2529                 rc = lov_getstripe(exp, karg, uarg);
2530                 break;
2531         case LL_IOC_LOV_SETEA:
2532                 rc = lov_setea(exp, karg, uarg);
2533                 break;
2534         default: {
2535                 int set = 0;
2536
2537                 if (count == 0)
2538                         RETURN(-ENOTTY);
2539
2540                 rc = 0;
2541                 for (i = 0; i < count; i++) {
2542                         int err;
2543
2544                         /* OST was disconnected */
2545                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2546                                 continue;
2547
2548                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2549                                             len, karg, uarg);
2550                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2551                                 RETURN(err);
2552                         } else if (err) {
2553                                 if (lov->lov_tgts[i]->ltd_active) {
2554                                         CDEBUG(err == -ENOTTY ?
2555                                                D_IOCTL : D_WARNING,
2556                                                "iocontrol OSC %s on OST "
2557                                                "idx %d cmd %x: err = %d\n",
2558                                                lov_uuid2str(lov, i),
2559                                                i, cmd, err);
2560                                         if (!rc)
2561                                                 rc = err;
2562                                 }
2563                         } else {
2564                                 set = 1;
2565                         }
2566                 }
2567                 if (!set && !rc)
2568                         rc = -EIO;
2569         }
2570         }
2571
2572         RETURN(rc);
2573 }
2574
2575 #define FIEMAP_BUFFER_SIZE 4096
2576
2577 /* Non-zero fe_logical indicates that this is a continuation FIEMAP
2578  * call. The local end offset and the device are sent in the first
2579  * fm_extent. This function calculates the stripe number from the index.
2580  * This function returns a stripe_no on which mapping is to be restarted.
2581  *
2582  * This function returns fm_end_offset which is the in-OST offset at which
2583  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2584  * will re-calculate proper offset in next stripe.
2585  * Note that the first extent is passed to lov_get_info via the value field */
2586 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2587                                    struct lov_stripe_md *lsm, obd_size fm_start,
2588                                    obd_size fm_end, int *start_stripe)
2589 {
2590         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2591         obd_off lun_start, lun_end;
2592         obd_size fm_end_offset;
2593         int stripe_no = -1, i;
2594
2595         if (fiemap->fm_extent_count == 0 ||
2596             fiemap->fm_extents[0].fe_logical == 0)
2597                 return 0;
2598
2599         /* Find out stripe_no from ost_index saved in the fe_device */
2600         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2601                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2602                                         fiemap->fm_extents[0].fe_device) {
2603                         stripe_no = i;
2604                         break;
2605                 }
2606         }
2607
2608         /* If we have finished mapping on previous device, shift logical
2609          * offset to start of next device */
2610         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2611                                    &lun_start, &lun_end)) != 0 &&
2612                                    local_end < lun_end) {
2613                 fm_end_offset = local_end;
2614                 *start_stripe = stripe_no;
2615         } else {
2616                 /* This is a special value to indicate that caller should
2617                  * calculate offset in next stripe. */
2618                 fm_end_offset = 0;
2619                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2620         }
2621
2622         return fm_end_offset;
2623 }
2624
2625 /* We calculate on which OST the mapping will end. If the length of mapping
2626  * is greater than (stripe_size * stripe_count) then the last_stripe will
2627  * will be one just before start_stripe. Else we check if the mapping
2628  * intersects each OST and find last_stripe.
2629  * This function returns the last_stripe and also sets the stripe_count
2630  * over which the mapping is spread */
2631 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2632                             obd_size fm_end, int start_stripe,
2633                             int *stripe_count)
2634 {
2635         int last_stripe;
2636         obd_off obd_start, obd_end;
2637         int i, j;
2638
2639         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2640                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2641                                                               start_stripe - 1);
2642                 *stripe_count = lsm->lsm_stripe_count;
2643         } else {
2644                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2645                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2646                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2647                                                    &obd_start, &obd_end)) == 0)
2648                                 break;
2649                 }
2650                 *stripe_count = j;
2651                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2652         }
2653
2654         return last_stripe;
2655 }
2656
2657 /* Set fe_device and copy extents from local buffer into main return buffer */
2658 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2659                                   struct ll_fiemap_extent *lcl_fm_ext,
2660                                   int ost_index, unsigned int ext_count,
2661                                   int current_extent)
2662 {
2663         char *to;
2664         int ext;
2665
2666         for (ext = 0; ext < ext_count; ext++) {
2667                 lcl_fm_ext[ext].fe_device = ost_index;
2668                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2669         }
2670
2671         /* Copy fm_extent's from fm_local to return buffer */
2672         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2673         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2674 }
2675
2676 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2677                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2678 {
2679         struct ll_fiemap_info_key *fm_key = key;
2680         struct ll_user_fiemap *fiemap = val;
2681         struct ll_user_fiemap *fm_local = NULL;
2682         struct ll_fiemap_extent *lcl_fm_ext;
2683         int count_local;
2684         unsigned int get_num_extents = 0;
2685         int ost_index = 0, actual_start_stripe, start_stripe;
2686         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2687         obd_size curr_loc;
2688         int current_extent = 0, rc = 0, i;
2689         int ost_eof = 0; /* EOF for object */
2690         int ost_done = 0; /* done with required mapping for this OST? */
2691         int last_stripe;
2692         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2693         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2694
2695         if (lsm == NULL)
2696                 GOTO(out, rc = 0);
2697
2698         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2699                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2700
2701         OBD_ALLOC(fm_local, buffer_size);
2702         if (fm_local == NULL)
2703                 GOTO(out, rc = -ENOMEM);
2704         lcl_fm_ext = &fm_local->fm_extents[0];
2705
2706         count_local = fiemap_size_to_count(buffer_size);
2707
2708         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2709         fm_start = fiemap->fm_start;
2710         fm_length = fiemap->fm_length;
2711         /* Calculate start stripe, last stripe and length of mapping */
2712         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2713         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2714                                                 fm_start + fm_length - 1);
2715         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2716         if (fm_end > fm_key->oa.o_size)
2717                 fm_end = fm_key->oa.o_size;
2718
2719         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2720                                               actual_start_stripe, &stripe_count);
2721
2722         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2723                                                   &start_stripe);
2724
2725         if (fiemap->fm_extent_count == 0) {
2726                 get_num_extents = 1;
2727                 count_local = 0;
2728         }
2729
2730         /* Check each stripe */
2731         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2732              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2733                 obd_size req_fm_len; /* Stores length of required mapping */
2734                 obd_size len_mapped_single_call;
2735                 obd_off lun_start, lun_end, obd_object_end;
2736                 unsigned int ext_count;
2737
2738                 cur_stripe_wrap = cur_stripe;
2739
2740                 /* Find out range of mapping on this stripe */
2741                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2742                                            &lun_start, &obd_object_end)) == 0)
2743                         continue;
2744
2745                 /* If this is a continuation FIEMAP call and we are on
2746                  * starting stripe then lun_start needs to be set to
2747                  * fm_end_offset */
2748                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2749                         lun_start = fm_end_offset;
2750
2751                 if (fm_length != ~0ULL) {
2752                         /* Handle fm_start + fm_length overflow */
2753                         if (fm_start + fm_length < fm_start)
2754                                 fm_length = ~0ULL - fm_start;
2755                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2756                                                      cur_stripe);
2757                 } else {
2758                         lun_end = ~0ULL;
2759                 }
2760
2761                 if (lun_start == lun_end)
2762                         continue;
2763
2764                 req_fm_len = obd_object_end - lun_start;
2765                 fm_local->fm_length = 0;
2766                 len_mapped_single_call = 0;
2767
2768                 /* If the output buffer is very large and the objects have many
2769                  * extents we may need to loop on a single OST repeatedly */
2770                 ost_eof = 0;
2771                 ost_done = 0;
2772                 do {
2773                         if (get_num_extents == 0) {
2774                                 /* Don't get too many extents. */
2775                                 if (current_extent + count_local >
2776                                     fiemap->fm_extent_count)
2777                                         count_local = fiemap->fm_extent_count -
2778                                                                  current_extent;
2779                         }
2780
2781                         lun_start += len_mapped_single_call;
2782                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2783                         req_fm_len = fm_local->fm_length;
2784                         fm_local->fm_extent_count = count_local;
2785                         fm_local->fm_mapped_extents = 0;
2786                         fm_local->fm_flags = fiemap->fm_flags;
2787
2788                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2789                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2790
2791                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2792                                 GOTO(out, rc = -EINVAL);
2793
2794                         /* If OST is inactive, return extent with UNKNOWN flag */
2795                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2796                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2797                                 fm_local->fm_mapped_extents = 1;
2798
2799                                 lcl_fm_ext[0].fe_logical = lun_start;
2800                                 lcl_fm_ext[0].fe_length = obd_object_end -
2801                                                                       lun_start;
2802                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2803
2804                                 goto inactive_tgt;
2805                         }
2806
2807                         fm_local->fm_start = lun_start;
2808                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2809                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2810                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2811                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2812                                           keylen, key, vallen, fm_local, lsm);
2813                         if (rc != 0)
2814                                 GOTO(out, rc);
2815
2816 inactive_tgt:
2817                         ext_count = fm_local->fm_mapped_extents;
2818                         if (ext_count == 0) {
2819                                 ost_done = 1;
2820                                 /* If last stripe has hole at the end,
2821                                  * then we need to return */
2822                                 if (cur_stripe_wrap == last_stripe) {
2823                                         fiemap->fm_mapped_extents = 0;
2824                                         goto finish;
2825                                 }
2826                                 break;
2827                         }
2828
2829                         /* If we just need num of extents then go to next device */
2830                         if (get_num_extents) {
2831                                 current_extent += ext_count;
2832                                 break;
2833                         }
2834
2835                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2836                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2837
2838                         /* Have we finished mapping on this device? */
2839                         if (req_fm_len <= len_mapped_single_call)
2840                                 ost_done = 1;
2841
2842                         /* Clear the EXTENT_LAST flag which can be present on
2843                          * last extent */
2844                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2845                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2846                                                             ~FIEMAP_EXTENT_LAST;
2847
2848                         curr_loc = lov_stripe_size(lsm,
2849                                            lcl_fm_ext[ext_count - 1].fe_logical+
2850                                            lcl_fm_ext[ext_count - 1].fe_length,
2851                                            cur_stripe);
2852                         if (curr_loc >= fm_key->oa.o_size)
2853                                 ost_eof = 1;
2854
2855                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2856                                                      ost_index, ext_count,
2857                                                      current_extent);
2858
2859                         current_extent += ext_count;
2860
2861                         /* Ran out of available extents? */
2862                         if (current_extent >= fiemap->fm_extent_count)
2863                                 goto finish;
2864                 } while (ost_done == 0 && ost_eof == 0);
2865
2866                 if (cur_stripe_wrap == last_stripe)
2867                         goto finish;
2868         }
2869
2870 finish:
2871         /* Indicate that we are returning device offsets unless file just has
2872          * single stripe */
2873         if (lsm->lsm_stripe_count > 1)
2874                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2875
2876         if (get_num_extents)
2877                 goto skip_last_device_calc;
2878
2879         /* Check if we have reached the last stripe and whether mapping for that
2880          * stripe is done. */
2881         if (cur_stripe_wrap == last_stripe) {
2882                 if (ost_done || ost_eof)
2883                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2884                                                              FIEMAP_EXTENT_LAST;
2885         }
2886
2887 skip_last_device_calc:
2888         fiemap->fm_mapped_extents = current_extent;
2889
2890 out:
2891         OBD_FREE(fm_local, buffer_size);
2892         return rc;
2893 }
2894
2895 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2896                         void *key, __u32 *vallen, void *val,
2897                         struct lov_stripe_md *lsm)
2898 {
2899         struct obd_device *obddev = class_exp2obd(exp);
2900         struct lov_obd *lov = &obddev->u.lov;
2901         int i, rc;
2902         ENTRY;
2903
2904         if (!vallen || !val)
2905                 RETURN(-EFAULT);
2906
2907         obd_getref(obddev);
2908
2909         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2910                 struct {
2911                         char name[16];
2912                         struct ldlm_lock *lock;
2913                 } *data = key;
2914                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2915                 struct lov_oinfo *loi;
2916                 __u32 *stripe = val;
2917
2918                 if (*vallen < sizeof(*stripe))
2919                         GOTO(out, rc = -EFAULT);
2920                 *vallen = sizeof(*stripe);
2921
2922                 /* XXX This is another one of those bits that will need to
2923                  * change if we ever actually support nested LOVs.  It uses
2924                  * the lock's export to find out which stripe it is. */
2925                 /* XXX - it's assumed all the locks for deleted OSTs have
2926                  * been cancelled. Also, the export for deleted OSTs will
2927                  * be NULL and won't match the lock's export. */
2928                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2929                         loi = lsm->lsm_oinfo[i];
2930                         if (!lov->lov_tgts[loi->loi_ost_idx])
2931                                 continue;
2932                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2933                             data->lock->l_conn_export &&
2934                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2935                                 *stripe = i;
2936                                 GOTO(out, rc = 0);
2937                         }
2938                 }
2939                 LDLM_ERROR(data->lock, "lock on inode without such object");
2940                 dump_lsm(D_ERROR, lsm);
2941                 GOTO(out, rc = -ENXIO);
2942         } else if (KEY_IS(KEY_LAST_ID)) {
2943                 struct obd_id_info *info = val;
2944                 __u32 size = sizeof(obd_id);
2945                 struct lov_tgt_desc *tgt;
2946
2947                 LASSERT(*vallen == sizeof(struct obd_id_info));
2948                 tgt = lov->lov_tgts[info->idx];
2949
2950                 if (!tgt || !tgt->ltd_active)
2951                         GOTO(out, rc = -ESRCH);
2952
2953                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2954                 GOTO(out, rc = 0);
2955         } else if (KEY_IS(KEY_LOVDESC)) {
2956                 struct lov_desc *desc_ret = val;
2957                 *desc_ret = lov->desc;
2958
2959                 GOTO(out, rc = 0);
2960         } else if (KEY_IS(KEY_FIEMAP)) {
2961                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2962                 GOTO(out, rc);
2963         } else if (KEY_IS(KEY_OFF_RPCSIZE)) {
2964                 __u64 *offset = val;
2965                 struct lov_tgt_desc *tgt;
2966                 struct lov_oinfo *loi;
2967                 int stripe;
2968
2969                 LASSERT(*vallen == sizeof(__u64));
2970                 stripe = lov_stripe_number(lsm, *offset);
2971                 loi = lsm->lsm_oinfo[stripe];
2972                 tgt = lov->lov_tgts[loi->loi_ost_idx];
2973                 if (!tgt || !tgt->ltd_active)
2974                         GOTO(out, rc = -ESRCH);
2975                 rc = obd_get_info(tgt->ltd_exp, keylen, key, vallen, val, NULL);
2976                 GOTO(out, rc);
2977         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2978                 struct lov_tgt_desc *tgt;
2979                 __u64 ost_idx = *((__u64*)val);
2980
2981                 LASSERT(*vallen == sizeof(__u64));
2982                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2983                 tgt = lov->lov_tgts[ost_idx];
2984
2985                 if (!tgt || !tgt->ltd_exp)
2986                         GOTO(out, rc = -ESRCH);
2987
2988                 *((__u64*)val) = tgt->ltd_exp->exp_connect_flags;
2989                 GOTO(out, rc = 0);
2990         }
2991
2992         rc = -EINVAL;
2993 out:
2994         obd_putref(obddev);
2995         RETURN(rc);
2996 }
2997
2998 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2999                               void *key, obd_count vallen, void *val,
3000                               struct ptlrpc_request_set *set)
3001 {
3002         struct obd_device *obddev = class_exp2obd(exp);
3003         struct lov_obd *lov = &obddev->u.lov;
3004         obd_count count;
3005         int i, rc = 0, err, incr = 0, check_uuid = 0, do_inactive = 0;
3006         int no_set = !set;
3007         unsigned next_id = 0;
3008         struct lov_tgt_desc *tgt;
3009         void *data;
3010         ENTRY;
3011
3012         if (no_set) {
3013                 set = ptlrpc_prep_set();
3014                 if (!set)
3015                         RETURN(-ENOMEM);
3016         }
3017
3018         obd_getref(obddev);
3019         count = lov->desc.ld_tgt_count;
3020
3021         if (KEY_IS(KEY_NEXT_ID)) {
3022                 count = vallen / sizeof(struct obd_id_info);
3023                 vallen = sizeof(obd_id);
3024                 incr = sizeof(struct obd_id_info);
3025                 do_inactive = 1;
3026                 next_id = 1;
3027         } else if (KEY_IS(KEY_CHECKSUM)) {
3028                 do_inactive = 1;
3029         } else if (KEY_IS(KEY_MDS_CONN)) {
3030                 check_uuid = val ? 1 : 0;
3031         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3032                 /* use defaults:
3033                 do_inactive = incr = 0;
3034                  */
3035         }
3036
3037         for (i = 0; i < count; i++, val = (char *)val + incr) {
3038                 if (next_id) {
3039                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3040                         data = ((struct obd_id_info*)val)->data;
3041                 } else {
3042                         tgt = lov->lov_tgts[i];
3043                         data = val;
3044                 }
3045                 /* OST was disconnected */
3046                 if (!tgt || !tgt->ltd_exp)
3047                         continue;
3048
3049                 /* OST is inactive and we don't want inactive OSCs */
3050                 if (!tgt->ltd_active && !do_inactive)
3051                         continue;
3052
3053                 /* Only want a specific OSC */
3054                 if (check_uuid &&
3055                     !obd_uuid_equals(val, &tgt->ltd_uuid))
3056                         continue;
3057
3058                 err = obd_set_info_async(tgt->ltd_exp,
3059                                          keylen, key, vallen, data, set);
3060                 if (!rc)
3061                         rc = err;
3062         }
3063         obd_putref(obddev);
3064         if (no_set) {
3065                 err = ptlrpc_set_wait(set);
3066                 if (!rc)
3067                         rc = err;
3068                 ptlrpc_set_destroy(set);
3069         }
3070         RETURN(rc);
3071 }
3072
3073 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3074                        struct lov_stripe_md *lsm)
3075 {
3076         int rc;
3077         ENTRY;
3078
3079         if (!lsm)
3080                 RETURN(0);
3081         LASSERT(md_exp);
3082         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3083         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3084
3085         RETURN(rc);
3086 }
3087
3088 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3089 {
3090         struct lov_oinfo *loi;
3091         int i, rc = 0;
3092         ENTRY;
3093
3094         for (i = 0; i < lsm->lsm_stripe_count; i++) {
3095                 loi = lsm->lsm_oinfo[i];
3096                 if (loi->loi_ar.ar_rc && !rc)
3097                         rc = loi->loi_ar.ar_rc;
3098                 loi->loi_ar.ar_rc = 0;
3099         }
3100         RETURN(rc);
3101 }
3102 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3103
3104
3105 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3106                            int cmd, __u64 *offset)
3107 {
3108         __u64 start;
3109         __u32 ssize  = lsm->lsm_stripe_size;
3110
3111         if (cmd & OBD_CALC_STRIPE_RPC_ALIGN)
3112                 ssize = ssize > PTLRPC_MAX_BRW_SIZE ?
3113                         PTLRPC_MAX_BRW_SIZE : ssize;
3114
3115         start = *offset;
3116         do_div(start, ssize);
3117         start = start * ssize;
3118
3119         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3120                ", end "LPU64"\n", *offset, ssize, start, start + ssize - 1);
3121         if (cmd & OBD_CALC_STRIPE_END)
3122                 *offset = start + ssize - 1;
3123         else if (cmd & OBD_CALC_STRIPE_START)
3124                 *offset = start;
3125         else
3126                 LBUG();
3127
3128         RETURN(0);
3129 }
3130
3131 #if 0
3132 struct lov_multi_wait {
3133         struct ldlm_lock *lock;
3134         wait_queue_t      wait;
3135         int               completed;
3136         int               generation;
3137 };
3138
3139 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3140                       struct lustre_handle *lockh)
3141 {
3142         struct lov_lock_handles *lov_lockh = NULL;
3143         struct lustre_handle *lov_lockhp;
3144         struct lov_obd *lov;
3145         struct lov_oinfo *loi;
3146         struct lov_multi_wait *queues;
3147         int rc = 0, i;
3148         ENTRY;
3149
3150         ASSERT_LSM_MAGIC(lsm);
3151
3152         if (!exp || !exp->exp_obd)
3153                 RETURN(-ENODEV);
3154
3155         LASSERT(lockh != NULL);
3156         if (lsm->lsm_stripe_count > 1) {
3157                 lov_lockh = lov_handle2llh(lockh);
3158                 if (lov_lockh == NULL) {
3159                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3160                         RETURN(-EINVAL);
3161                 }
3162
3163                 lov_lockhp = lov_lockh->llh_handles;
3164         } else {
3165                 lov_lockhp = lockh;
3166         }
3167
3168         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3169         if (queues == NULL)
3170                 GOTO(out, rc = -ENOMEM);
3171
3172         lov = &exp->exp_obd->u.lov;
3173         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3174              i++, loi++, lov_lockhp++) {
3175                 struct ldlm_lock *lock;
3176                 struct obd_device *obd;
3177
3178                 lock = ldlm_handle2lock(lov_lockhp);
3179                 if (lock == NULL) {
3180                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3181                                loi->loi_ost_idx, loi->loi_id);
3182                         queues[i].completed = 1;
3183                         continue;
3184                 }
3185
3186                 queues[i].lock = lock;
3187                 init_waitqueue_entry(&(queues[i].wait), current);
3188                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3189
3190                 obd = class_exp2obd(lock->l_conn_export);
3191                 if (obd != NULL)
3192                         imp = obd->u.cli.cl_import;
3193                 if (imp != NULL) {
3194                         spin_lock(&imp->imp_lock);
3195                         queues[i].generation = imp->imp_generation;
3196                         spin_unlock(&imp->imp_lock);
3197                 }
3198         }
3199
3200         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3201                                interrupted_completion_wait, &lwd);
3202         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3203
3204         for (i = 0; i < lsm->lsm_stripe_count; i++)
3205                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3206
3207         if (rc == -EINTR || rc == -ETIMEDOUT) {
3208
3209
3210         }
3211
3212  out:
3213         if (lov_lockh != NULL)
3214                 lov_llh_put(lov_lockh);
3215         RETURN(rc);
3216 }
3217 #endif
3218
3219 void lov_stripe_lock(struct lov_stripe_md *md)
3220 {
3221         LASSERT(md->lsm_lock_owner != cfs_current());
3222         spin_lock(&md->lsm_lock);
3223         LASSERT(md->lsm_lock_owner == NULL);
3224         md->lsm_lock_owner = cfs_current();
3225 }
3226 EXPORT_SYMBOL(lov_stripe_lock);
3227
3228 void lov_stripe_unlock(struct lov_stripe_md *md)
3229 {
3230         LASSERT(md->lsm_lock_owner == cfs_current());
3231         md->lsm_lock_owner = NULL;
3232         spin_unlock(&md->lsm_lock);
3233 }
3234 EXPORT_SYMBOL(lov_stripe_unlock);
3235
3236 static int lov_get_lock(struct obd_export *exp, struct lov_stripe_md *lsm,
3237                         void **res, int rw, obd_off start, obd_off end,
3238                         struct lustre_handle *lockh, int flags)
3239 {
3240         struct lov_async_page *l = *res;
3241         obd_off stripe_start, stripe_end = start;
3242         struct lov_lock_handles *lov_lockh = NULL;
3243         int rc;
3244
3245         ENTRY;
3246
3247         if (lockh && lustre_handle_is_used(lockh) &&
3248             !(flags & OBD_FAST_LOCK)) {
3249                 lov_lockh = lov_handle2llh(lockh);
3250                 if (lov_lockh == NULL) {
3251                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3252                         RETURN(-EINVAL);
3253                 }
3254                 lockh = lov_lockh->llh_handles + l->lap_stripe;
3255         }
3256
3257         /* ensure we don't cross stripe boundaries */
3258         lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3259         if (stripe_end < end)
3260                 GOTO(out, rc = 0);
3261
3262         /* map the region limits to the object limits */
3263         lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3264         lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3265
3266         rc = obd_get_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3267                             lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3268                             ltd_exp, NULL, &l->lap_sub_cookie,
3269                             rw, stripe_start, stripe_end, lockh, flags);
3270 out:
3271         if (lov_lockh != NULL)
3272                 lov_llh_put(lov_lockh);
3273         RETURN(rc);
3274 }
3275
3276 struct obd_ops lov_obd_ops = {
3277         .o_owner               = THIS_MODULE,
3278         .o_setup               = lov_setup,
3279         .o_precleanup          = lov_precleanup,
3280         .o_cleanup             = lov_cleanup,
3281         .o_process_config      = lov_process_config,
3282         .o_connect             = lov_connect,
3283         .o_disconnect          = lov_disconnect,
3284         .o_statfs              = lov_statfs,
3285         .o_statfs_async        = lov_statfs_async,
3286         .o_packmd              = lov_packmd,
3287         .o_unpackmd            = lov_unpackmd,
3288         .o_checkmd             = lov_checkmd,
3289         .o_create              = lov_create,
3290         .o_destroy             = lov_destroy,
3291         .o_getattr             = lov_getattr,
3292         .o_getattr_async       = lov_getattr_async,
3293         .o_setattr             = lov_setattr,
3294         .o_setattr_async       = lov_setattr_async,
3295         .o_brw                 = lov_brw,
3296         .o_brw_async           = lov_brw_async,
3297         .o_prep_async_page     = lov_prep_async_page,
3298         .o_get_lock            = lov_get_lock,
3299         .o_queue_async_io      = lov_queue_async_io,
3300         .o_set_async_flags     = lov_set_async_flags,
3301         .o_queue_group_io      = lov_queue_group_io,
3302         .o_trigger_group_io    = lov_trigger_group_io,
3303         .o_teardown_async_page = lov_teardown_async_page,
3304         .o_merge_lvb           = lov_merge_lvb,
3305         .o_update_lvb          = lov_update_lvb,
3306         .o_adjust_kms          = lov_adjust_kms,
3307         .o_punch               = lov_punch,
3308         .o_sync                = lov_sync,
3309         .o_enqueue             = lov_enqueue,
3310         .o_match               = lov_match,
3311         .o_change_cbdata       = lov_change_cbdata,
3312         .o_cancel              = lov_cancel,
3313         .o_cancel_unused       = lov_cancel_unused,
3314         .o_join_lru            = lov_join_lru,
3315         .o_iocontrol           = lov_iocontrol,
3316         .o_get_info            = lov_get_info,
3317         .o_set_info_async      = lov_set_info_async,
3318         .o_extent_calc         = lov_extent_calc,
3319         .o_llog_init           = lov_llog_init,
3320         .o_llog_finish         = lov_llog_finish,
3321         .o_notify              = lov_notify,
3322         .o_register_page_removal_cb = lov_obd_register_page_removal_cb,
3323         .o_unregister_page_removal_cb = lov_obd_unregister_page_removal_cb,
3324         .o_register_lock_cancel_cb = lov_obd_register_lock_cancel_cb,
3325         .o_unregister_lock_cancel_cb = lov_obd_unregister_lock_cancel_cb,
3326         .o_pool_new            = lov_pool_new,
3327         .o_pool_rem            = lov_pool_remove,
3328         .o_pool_add            = lov_pool_add,
3329         .o_pool_del            = lov_pool_del,
3330         .o_getref              = lov_getref,
3331         .o_putref              = lov_putref,
3332 };
3333
3334 static quota_interface_t *quota_interface;
3335 extern quota_interface_t lov_quota_interface;
3336
3337 cfs_mem_cache_t *lov_oinfo_slab;
3338
3339 int __init lov_init(void)
3340 {
3341         struct lprocfs_static_vars lvars = { 0 };
3342         int rc, rc2;
3343         ENTRY;
3344
3345         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3346                                               sizeof(struct lov_oinfo),
3347                                               0, SLAB_HWCACHE_ALIGN);
3348         if (lov_oinfo_slab == NULL)
3349                 return -ENOMEM;
3350         lprocfs_lov_init_vars(&lvars);
3351
3352         request_module("lquota");
3353         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3354         init_obd_quota_ops(quota_interface, &lov_obd_ops);
3355
3356         rc = class_register_type(&lov_obd_ops, lvars.module_vars,
3357                                  LUSTRE_LOV_NAME);
3358         if (rc) {
3359                 if (quota_interface)
3360                         PORTAL_SYMBOL_PUT(lov_quota_interface);
3361                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3362                 LASSERT(rc2 == 0);
3363         }
3364
3365         RETURN(rc);
3366 }
3367
3368 #ifdef __KERNEL__
3369 static void /*__exit*/ lov_exit(void)
3370 {
3371         int rc;
3372
3373         if (quota_interface)
3374                 PORTAL_SYMBOL_PUT(lov_quota_interface);
3375
3376         class_unregister_type(LUSTRE_LOV_NAME);
3377         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3378         LASSERT(rc == 0);
3379 }
3380
3381 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3382 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3383 MODULE_LICENSE("GPL");
3384
3385 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
3386 #endif