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 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, __u32 index);
  86
  87 void lov_putref(struct obd_device *obd)
  88 {
  89         struct lov_obd *lov = &obd->u.lov;
  90         mutex_down(&lov->lov_lock);
  91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  92         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  93                 int i;
  94                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  95                        lov->lov_death_row);
  96                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  97                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
  98                                 continue;
  99                         /* Disconnect and delete from list */
 100                         __lov_del_obd(obd, i);
 101                         lov->lov_death_row--;
 102                 }
 103         }
 104         mutex_up(&lov->lov_lock);
 105 }
 106
 107 static int lov_register_page_removal_cb(struct obd_export *exp,
 108                                         obd_page_removal_cb_t func,
 109                                         obd_pin_extent_cb pin_cb)
 110 {
 111         struct lov_obd *lov = &exp->exp_obd->u.lov;
 112         int i, rc = 0;
 113
 114         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 115                 return -EBUSY;
 116
 117         if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
 118                 return -EBUSY;
 119
 120         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 121                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 122                         continue;
 123                 rc |= obd_register_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
 124                                                    func, pin_cb);
 125         }
 126
 127         lov->lov_page_removal_cb = func;
 128         lov->lov_page_pin_cb = pin_cb;
 129
 130         return rc;
 131 }
 132
 133 static int lov_unregister_page_removal_cb(struct obd_export *exp,
 134                                         obd_page_removal_cb_t func)
 135 {
 136         struct lov_obd *lov = &exp->exp_obd->u.lov;
 137         int i, rc = 0;
 138
 139         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 140                 return -EINVAL;
 141
 142         lov->lov_page_removal_cb = NULL;
 143         lov->lov_page_pin_cb = NULL;
 144
 145         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 146                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 147                         continue;
 148                 rc |= obd_unregister_page_removal_cb(lov->lov_tgts[i]->ltd_exp,
 149                                                      func);
 150         }
 151
 152         return rc;
 153 }
 154
 155 static int lov_register_lock_cancel_cb(struct obd_export *exp,
 156                                          obd_lock_cancel_cb func)
 157 {
 158         struct lov_obd *lov = &exp->exp_obd->u.lov;
 159         int i, rc = 0;
 160
 161         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 162                 return -EBUSY;
 163
 164         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 165                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 166                         continue;
 167                 rc |= obd_register_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
 168                                                   func);
 169         }
 170
 171         lov->lov_lock_cancel_cb = func;
 172
 173         return rc;
 174 }
 175
 176 static int lov_unregister_lock_cancel_cb(struct obd_export *exp,
 177                                          obd_lock_cancel_cb func)
 178 {
 179         struct lov_obd *lov = &exp->exp_obd->u.lov;
 180         int i, rc = 0;
 181
 182         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 183                 return -EINVAL;
 184
 185         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 186                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 187                         continue;
 188                 rc |= obd_unregister_lock_cancel_cb(lov->lov_tgts[i]->ltd_exp,
 189                                                     func);
 190         }
 191         lov->lov_lock_cancel_cb = NULL;
 192         return rc;
 193 }
 194
 195 #define MAX_STRING_SIZE 128
 196 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 197                            struct obd_connect_data *data)
 198 {
 199         struct lov_obd *lov = &obd->u.lov;
 200         struct obd_uuid tgt_uuid;
 201         struct obd_device *tgt_obd;
 202         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 203         struct lustre_handle conn = {0, };
 204         struct obd_import *imp;
 205
 206 #ifdef __KERNEL__
 207         cfs_proc_dir_entry_t *lov_proc_dir;
 208 #endif
 209         int rc;
 210         ENTRY;
 211
 212         if (!lov->lov_tgts[index])
 213                 RETURN(-EINVAL);
 214
 215         tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
 216
 217         tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
 218                                         &obd->obd_uuid);
 219
 220         if (!tgt_obd) {
 221                 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
 222                 RETURN(-EINVAL);
 223         }
 224         if (!tgt_obd->obd_set_up) {
 225                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 226                 RETURN(-EINVAL);
 227         }
 228
 229         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 230                 data->ocd_index = index;
 231
 232         /*
 233          * Divine LOV knows that OBDs under it are OSCs.
 234          */
 235         imp = tgt_obd->u.cli.cl_import;
 236
 237         if (activate) {
 238                 tgt_obd->obd_no_recov = 0;
 239                 /* FIXME this is probably supposed to be
 240                    ptlrpc_set_import_active.  Horrible naming. */
 241                 ptlrpc_activate_import(imp);
 242         }
 243
 244         if (imp->imp_invalid) {
 245                 CERROR("not connecting OSC %s; administratively "
 246                        "disabled\n", obd_uuid2str(&tgt_uuid));
 247                 rc = obd_register_observer(tgt_obd, obd);
 248                 if (rc) {
 249                         CERROR("Target %s register_observer error %d; "
 250                                "will not be able to reactivate\n",
 251                                obd_uuid2str(&tgt_uuid), rc);
 252                 }
 253                 RETURN(0);
 254         }
 255
 256         rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data, NULL);
 257         if (rc) {
 258                 CERROR("Target %s connect error %d\n",
 259                        obd_uuid2str(&tgt_uuid), rc);
 260                 RETURN(rc);
 261         }
 262         lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
 263         if (!lov->lov_tgts[index]->ltd_exp) {
 264                 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
 265                 RETURN(-ENODEV);
 266         }
 267
 268         rc = obd_register_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
 269                                           lov->lov_page_removal_cb,
 270                                           lov->lov_page_pin_cb);
 271         if (rc) {
 272                 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 273                 lov->lov_tgts[index]->ltd_exp = NULL;
 274                 RETURN(rc);
 275         }
 276
 277         rc = obd_register_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
 278                                          lov->lov_lock_cancel_cb);
 279         if (rc) {
 280                 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
 281                                                lov->lov_page_removal_cb);
 282                 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 283                 lov->lov_tgts[index]->ltd_exp = NULL;
 284                 RETURN(rc);
 285         }
 286
 287         rc = obd_register_observer(tgt_obd, obd);
 288         if (rc) {
 289                 CERROR("Target %s register_observer error %d\n",
 290                        obd_uuid2str(&tgt_uuid), rc);
 291                 obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
 292                                               lov->lov_lock_cancel_cb);
 293                 obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
 294                                                lov->lov_page_removal_cb);
 295                 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 296                 lov->lov_tgts[index]->ltd_exp = NULL;
 297                 RETURN(rc);
 298         }
 299
 300         lov->lov_tgts[index]->ltd_reap = 0;
 301         if (activate) {
 302                 lov->lov_tgts[index]->ltd_active = 1;
 303                 lov->desc.ld_active_tgt_count++;
 304                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 305         }
 306         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 307                obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 308
 309 #ifdef __KERNEL__
 310         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 311         if (lov_proc_dir) {
 312                 struct obd_device *osc_obd = class_conn2obd(&conn);
 313                 cfs_proc_dir_entry_t *osc_symlink;
 314                 char name[MAX_STRING_SIZE];
 315
 316                 LASSERT(osc_obd != NULL);
 317                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 318                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 319                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 320                          osc_obd->obd_type->typ_name,
 321                          osc_obd->obd_name);
 322                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name, lov_proc_dir,
 323                                                   name);
 324                 if (osc_symlink == NULL) {
 325                         CERROR("could not register LOV target "
 326                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 327                                obd->obd_type->typ_name, obd->obd_name,
 328                                osc_obd->obd_name);
 329                         lprocfs_remove(&lov_proc_dir);
 330                 }
 331         }
 332 #endif
 333
 334         rc = qos_add_tgt(obd, index);
 335         if (rc)
 336                 CERROR("qos_add_tgt failed %d\n", rc);
 337
 338         RETURN(0);
 339 }
 340
 341 static int lov_connect(const struct lu_env *env,
 342                        struct lustre_handle *conn, struct obd_device *obd,
 343                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 344                        void *localdata)
 345 {
 346         struct lov_obd *lov = &obd->u.lov;
 347         struct lov_tgt_desc *tgt;
 348         int i, rc;
 349         ENTRY;
 350
 351         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 352
 353         rc = class_connect(conn, obd, cluuid);
 354         if (rc)
 355                 RETURN(rc);
 356
 357         /* Why should there ever be more than 1 connect? */
 358         lov->lov_connects++;
 359         LASSERT(lov->lov_connects == 1);
 360
 361         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 362         if (data)
 363                 lov->lov_ocd = *data;
 364
 365         lov_getref(obd);
 366         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 367                 tgt = lov->lov_tgts[i];
 368                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 369                         continue;
 370                 /* Flags will be lowest common denominator */
 371                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 372                 if (rc) {
 373                         CERROR("%s: lov connect tgt %d failed: %d\n",
 374                                obd->obd_name, i, rc);
 375                         continue;
 376                 }
 377         }
 378         lov_putref(obd);
 379
 380         RETURN(0);
 381 }
 382
 383 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
 384 {
 385         cfs_proc_dir_entry_t *lov_proc_dir;
 386         struct lov_obd *lov = &obd->u.lov;
 387         struct obd_device *osc_obd;
 388         int rc;
 389
 390         ENTRY;
 391
 392         if (lov->lov_tgts[index] == NULL)
 393                 RETURN(-EINVAL);
 394
 395         osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
 396         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 397                obd->obd_name, osc_obd->obd_name);
 398
 399         obd_unregister_lock_cancel_cb(lov->lov_tgts[index]->ltd_exp,
 400                                       lov->lov_lock_cancel_cb);
 401         obd_unregister_page_removal_cb(lov->lov_tgts[index]->ltd_exp,
 402                                        lov->lov_page_removal_cb);
 403
 404         if (lov->lov_tgts[index]->ltd_active) {
 405                 lov->lov_tgts[index]->ltd_active = 0;
 406                 lov->desc.ld_active_tgt_count--;
 407                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 408         }
 409
 410         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 411         if (lov_proc_dir) {
 412                 cfs_proc_dir_entry_t *osc_symlink;
 413
 414                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 415                 if (osc_symlink) {
 416                         lprocfs_remove(&osc_symlink);
 417                 } else {
 418                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 419                                obd->obd_type->typ_name, obd->obd_name,
 420                                osc_obd->obd_name);
 421                 }
 422         }
 423
 424         if (osc_obd) {
 425                 /* Pass it on to our clients.
 426                  * XXX This should be an argument to disconnect,
 427                  * XXX not a back-door flag on the OBD.  Ah well.
 428                  */
 429                 osc_obd->obd_force = obd->obd_force;
 430                 osc_obd->obd_fail = obd->obd_fail;
 431                 osc_obd->obd_no_recov = obd->obd_no_recov;
 432         }
 433
 434         obd_register_observer(osc_obd, NULL);
 435
 436         rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 437         if (rc) {
 438                 CERROR("Target %s disconnect error %d\n",
 439                        lov_uuid2str(lov, index), rc);
 440                 rc = 0;
 441         }
 442
 443         qos_del_tgt(obd, index);
 444
 445         lov->lov_tgts[index]->ltd_exp = NULL;
 446         RETURN(0);
 447 }
 448
 449 static int lov_del_target(struct obd_device *obd, __u32 index,
 450                           struct obd_uuid *uuidp, int gen);
 451
 452 static int lov_disconnect(struct obd_export *exp)
 453 {
 454         struct obd_device *obd = class_exp2obd(exp);
 455         struct lov_obd *lov = &obd->u.lov;
 456         int i, rc;
 457         ENTRY;
 458
 459         if (!lov->lov_tgts)
 460                 goto out;
 461
 462         /* Only disconnect the underlying layers on the final disconnect. */
 463         lov->lov_connects--;
 464         if (lov->lov_connects != 0) {
 465                 /* why should there be more than 1 connect? */
 466                 CERROR("disconnect #%d\n", lov->lov_connects);
 467                 goto out;
 468         }
 469
 470         /* Let's hold another reference so lov_del_obd doesn't spin through
 471            putref every time */
 472         lov_getref(obd);
 473         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 474                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 475                         /* Disconnection is the last we know about an obd */
 476                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 477                 }
 478         }
 479         lov_putref(obd);
 480
 481 out:
 482         rc = class_disconnect(exp); /* bz 9811 */
 483         RETURN(rc);
 484 }
 485
 486 /* Error codes:
 487  *
 488  *  -EINVAL  : UUID can't be found in the LOV's target list
 489  * - any other is lov index
 490  */
 491 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 492                               int activate)
 493 {
 494         struct lov_obd *lov = &obd->u.lov;
 495         struct lov_tgt_desc *tgt;
 496         int index;
 497         ENTRY;
 498
 499         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 500                lov, uuid->uuid, activate);
 501
 502         lov_getref(obd);
 503         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 504                 tgt = lov->lov_tgts[index];
 505                 if (!tgt || !tgt->ltd_exp)
 506                         continue;
 507
 508                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 509                        index, obd_uuid2str(&tgt->ltd_uuid),
 510                        tgt->ltd_exp->exp_handle.h_cookie);
 511                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 512                         break;
 513         }
 514
 515         if (index == lov->desc.ld_tgt_count)
 516                 GOTO(out, index = -EINVAL);
 517
 518         if (lov->lov_tgts[index]->ltd_active == activate) {
 519                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 520                        activate ? "" : "in");
 521                 GOTO(out, index);
 522         }
 523
 524         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 525                activate ? "" : "in");
 526
 527         lov->lov_tgts[index]->ltd_active = activate;
 528
 529         if (activate) {
 530                 lov->desc.ld_active_tgt_count++;
 531                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 532         } else {
 533                 lov->desc.ld_active_tgt_count--;
 534                 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 535         }
 536         /* remove any old qos penalty */
 537         lov->lov_tgts[index]->ltd_qos.ltq_penalty = 0;
 538
 539  out:
 540         lov_putref(obd);
 541         RETURN(index);
 542 }
 543
 544 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 545                       enum obd_notify_event ev, void *data)
 546 {
 547         int rc = 0;
 548         ENTRY;
 549
 550         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 551                 struct obd_uuid *uuid;
 552
 553                 LASSERT(watched);
 554
 555                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 556                         CERROR("unexpected notification of %s %s!\n",
 557                                watched->obd_type->typ_name,
 558                                watched->obd_name);
 559                         RETURN(-EINVAL);
 560                 }
 561                 uuid = &watched->u.cli.cl_target_uuid;
 562
 563                 /* Set OSC as active before notifying the observer, so the
 564                  * observer can use the OSC normally.
 565                  */
 566                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 567                 if (rc < 0) {
 568                         CERROR("%sactivation of %s failed: %d\n",
 569                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 570                                obd_uuid2str(uuid), rc);
 571                         RETURN(rc);
 572                 }
 573                 data = &rc;
 574         }
 575
 576         /* Pass the notification up the chain. */
 577         if (watched) {
 578                 rc = obd_notify_observer(obd, watched, ev, data);
 579         } else {
 580                 /* NULL watched means all osc's in the lov (only for syncs) */
 581                 /* sync event should be send lov idx as data */
 582                 struct lov_obd *lov = &obd->u.lov;
 583                 struct obd_device *tgt_obd;
 584                 int i;
 585                 lov_getref(obd);
 586                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 587                         if (!lov->lov_tgts[i])
 588                                 continue;
 589
 590                         if ((ev == OBD_NOTIFY_SYNC) ||
 591                             (ev == OBD_NOTIFY_SYNC_NONBLOCK))
 592                                 data = &i;
 593
 594                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
 595                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 596                         if (rc) {
 597                                 CERROR("%s: notify %s of %s failed %d\n",
 598                                        obd->obd_name,
 599                                        obd->obd_observer->obd_name,
 600                                        tgt_obd->obd_name, rc);
 601                                 break;
 602                         }
 603                 }
 604                 lov_putref(obd);
 605         }
 606
 607         RETURN(rc);
 608 }
 609
 610 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 611                           __u32 index, int gen, int active)
 612 {
 613         struct lov_obd *lov = &obd->u.lov;
 614         struct lov_tgt_desc *tgt;
 615         int rc;
 616         ENTRY;
 617
 618         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 619                uuidp->uuid, index, gen, active);
 620
 621         if (gen <= 0) {
 622                 CERROR("request to add OBD %s with invalid generation: %d\n",
 623                        uuidp->uuid, gen);
 624                 RETURN(-EINVAL);
 625         }
 626
 627         mutex_down(&lov->lov_lock);
 628
 629         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 630                 tgt = lov->lov_tgts[index];
 631                 CERROR("UUID %s already assigned at LOV target index %d\n",
 632                        obd_uuid2str(&tgt->ltd_uuid), index);
 633                 mutex_up(&lov->lov_lock);
 634                 RETURN(-EEXIST);
 635         }
 636
 637         if (index >= lov->lov_tgt_size) {
 638                 /* We need to reallocate the lov target array. */
 639                 struct lov_tgt_desc **newtgts, **old = NULL;
 640                 __u32 newsize, oldsize = 0;
 641
 642                 newsize = max(lov->lov_tgt_size, (__u32)2);
 643                 while (newsize < index + 1)
 644                         newsize = newsize << 1;
 645                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 646                 if (newtgts == NULL) {
 647                         mutex_up(&lov->lov_lock);
 648                         RETURN(-ENOMEM);
 649                 }
 650
 651                 if (lov->lov_tgt_size) {
 652                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 653                                lov->lov_tgt_size);
 654                         old = lov->lov_tgts;
 655                         oldsize = lov->lov_tgt_size;
 656                 }
 657
 658                 lov->lov_tgts = newtgts;
 659                 lov->lov_tgt_size = newsize;
 660 #ifdef __KERNEL__
 661                 smp_rmb();
 662 #endif
 663                 if (old)
 664                         OBD_FREE(old, sizeof(*old) * oldsize);
 665
 666                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 667                        lov->lov_tgts, lov->lov_tgt_size);
 668         }
 669
 670         OBD_ALLOC_PTR(tgt);
 671         if (!tgt) {
 672                 mutex_up(&lov->lov_lock);
 673                 RETURN(-ENOMEM);
 674         }
 675
 676         memset(tgt, 0, sizeof(*tgt));
 677         tgt->ltd_uuid = *uuidp;
 678         /* XXX - add a sanity check on the generation number. */
 679         tgt->ltd_gen = gen;
 680         tgt->ltd_index = index;
 681         tgt->ltd_activate = active;
 682         lov->lov_tgts[index] = tgt;
 683         if (index >= lov->desc.ld_tgt_count)
 684                 lov->desc.ld_tgt_count = index + 1;
 685
 686         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 687         if (rc)
 688                 RETURN(rc);
 689
 690         mutex_up(&lov->lov_lock);
 691
 692         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 693                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 694
 695         if (lov->lov_connects == 0) {
 696                 /* lov_connect hasn't been called yet. We'll do the
 697                    lov_connect_obd on this target when that fn first runs,
 698                    because we don't know the connect flags yet. */
 699                 RETURN(0);
 700         }
 701
 702         lov_getref(obd);
 703
 704         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 705         if (rc)
 706                 GOTO(out, rc);
 707
 708         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 709                         active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
 710                         (void *)&index);
 711
 712 out:
 713         if (rc) {
 714                 CERROR("add failed (%d), deleting %s\n", rc,
 715                        obd_uuid2str(&tgt->ltd_uuid));
 716                 lov_del_target(obd, index, 0, 0);
 717         }
 718         lov_putref(obd);
 719         RETURN(rc);
 720 }
 721
 722 /* Schedule a target for deletion */
 723 static int lov_del_target(struct obd_device *obd, __u32 index,
 724                           struct obd_uuid *uuidp, int gen)
 725 {
 726         struct lov_obd *lov = &obd->u.lov;
 727         int count = lov->desc.ld_tgt_count;
 728         int rc = 0;
 729         ENTRY;
 730
 731         if (index >= count) {
 732                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 733                        index, count);
 734                 RETURN(-EINVAL);
 735         }
 736
 737         lov_getref(obd);
 738
 739         if (!lov->lov_tgts[index]) {
 740                 CERROR("LOV target at index %d is not setup.\n", index);
 741                 GOTO(out, rc = -EINVAL);
 742         }
 743
 744         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 745                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 746                        lov_uuid2str(lov, index), index,
 747                        obd_uuid2str(uuidp));
 748                 GOTO(out, rc = -EINVAL);
 749         }
 750
 751         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 752                lov_uuid2str(lov, index), index,
 753                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 754                lov->lov_tgts[index]->ltd_active);
 755
 756         lov->lov_tgts[index]->ltd_reap = 1;
 757         lov->lov_death_row++;
 758         /* we really delete it from lov_putref */
 759 out:
 760         lov_putref(obd);
 761
 762         RETURN(rc);
 763 }
 764
 765 /* We are holding lov_lock */
 766 static void __lov_del_obd(struct obd_device *obd, __u32 index)
 767 {
 768         struct lov_obd *lov = &obd->u.lov;
 769         struct obd_device *osc_obd;
 770         struct lov_tgt_desc *tgt = lov->lov_tgts[index];
 771
 772         LASSERT(tgt);
 773         LASSERT(tgt->ltd_reap);
 774
 775         osc_obd = class_exp2obd(tgt->ltd_exp);
 776
 777         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 778                lov_uuid2str(lov, index),
 779                osc_obd ? osc_obd->obd_name : "<no obd>");
 780
 781         if (tgt->ltd_exp)
 782                 lov_disconnect_obd(obd, index);
 783
 784         /* XXX - right now there is a dependency on ld_tgt_count being the
 785          * maximum tgt index for computing the mds_max_easize. So we can't
 786          * shrink it. */
 787
 788         lov_ost_pool_remove(&lov->lov_packed, index);
 789         lov->lov_tgts[index] = NULL;
 790         OBD_FREE_PTR(tgt);
 791
 792         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 793            do it ourselves. And we can't do it from lov_cleanup,
 794            because we just lost our only reference to it. */
 795         if (osc_obd)
 796                 class_manual_cleanup(osc_obd);
 797 }
 798
 799 void lov_fix_desc_stripe_size(__u64 *val)
 800 {
 801         if (*val < PTLRPC_MAX_BRW_SIZE) {
 802                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 803                               PTLRPC_MAX_BRW_SIZE);
 804                 *val = PTLRPC_MAX_BRW_SIZE;
 805         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 806                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 807                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 808                               "multiple of %u)\n",
 809                               *val, LOV_MIN_STRIPE_SIZE);
 810         }
 811 }
 812
 813 void lov_fix_desc_stripe_count(__u32 *val)
 814 {
 815         if (*val == 0)
 816                 *val = 1;
 817 }
 818
 819 void lov_fix_desc_pattern(__u32 *val)
 820 {
 821         /* from lov_setstripe */
 822         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 823                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 824                 *val = 0;
 825         }
 826 }
 827
 828 void lov_fix_desc_qos_maxage(__u32 *val)
 829 {
 830         /* fix qos_maxage */
 831         if (*val == 0)
 832                 *val = QOS_DEFAULT_MAXAGE;
 833 }
 834
 835 void lov_fix_desc(struct lov_desc *desc)
 836 {
 837         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 838         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 839         lov_fix_desc_pattern(&desc->ld_pattern);
 840         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 841 }
 842
 843 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 844 {
 845         struct lprocfs_static_vars lvars = { 0 };
 846         struct lov_desc *desc;
 847         struct lov_obd *lov = &obd->u.lov;
 848         int count;
 849         int rc;
 850         ENTRY;
 851
 852         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 853                 CERROR("LOV setup requires a descriptor\n");
 854                 RETURN(-EINVAL);
 855         }
 856
 857         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 858
 859         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 860                 CERROR("descriptor size wrong: %d > %d\n",
 861                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 862                 RETURN(-EINVAL);
 863         }
 864
 865         if (desc->ld_magic != LOV_DESC_MAGIC) {
 866                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 867                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 868                                    obd->obd_name, desc);
 869                             lustre_swab_lov_desc(desc);
 870                 } else {
 871                         CERROR("%s: Bad lov desc magic: %#x\n",
 872                                obd->obd_name, desc->ld_magic);
 873                         RETURN(-EINVAL);
 874                 }
 875         }
 876
 877         lov_fix_desc(desc);
 878
 879         /* Because of 64-bit divide/mod operations only work with a 32-bit
 880          * divisor in a 32-bit kernel, we cannot support a stripe width
 881          * of 4GB or larger on 32-bit CPUs. */
 882         count = desc->ld_default_stripe_count;
 883         if ((count > 0 ? count : desc->ld_tgt_count) *
 884             desc->ld_default_stripe_size > 0xffffffff) {
 885                 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
 886                        desc->ld_default_stripe_size, count);
 887                 RETURN(-EINVAL);
 888         }
 889
 890         desc->ld_active_tgt_count = 0;
 891         lov->desc = *desc;
 892         lov->lov_tgt_size = 0;
 893
 894         sema_init(&lov->lov_lock, 1);
 895         atomic_set(&lov->lov_refcount, 0);
 896         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 897         init_rwsem(&lov->lov_qos.lq_rw_sem);
 898         lov->lov_qos.lq_dirty = 1;
 899         lov->lov_qos.lq_rr.lqr_dirty = 1;
 900         lov->lov_qos.lq_reset = 1;
 901         /* Default priority is toward free space balance */
 902         lov->lov_qos.lq_prio_free = 232;
 903
 904         lov->lov_pools_hash_body = lustre_hash_init("POOLS", 128, 128,
 905                                                     &pool_hash_operations, 0);
 906         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 907         lov->lov_pool_count = 0;
 908         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 909         if (rc)
 910                 RETURN(rc);
 911         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 912         if (rc) {
 913                 lov_ost_pool_free(&lov->lov_packed);
 914                 RETURN(rc);
 915         }
 916
 917         lprocfs_lov_init_vars(&lvars);
 918         lprocfs_obd_setup(obd, lvars.obd_vars);
 919 #ifdef LPROCFS
 920         {
 921                 int rc;
 922
 923                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 924                                         0444, &lov_proc_target_fops, obd);
 925                 if (rc)
 926                         CWARN("Error adding the target_obd file\n");
 927         }
 928 #endif
 929         lov->lov_pool_proc_entry = lprocfs_register("pools",
 930                                                     obd->obd_proc_entry,
 931                                                     NULL, NULL);
 932
 933         RETURN(0);
 934 }
 935
 936 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 937 {
 938         int rc = 0;
 939         ENTRY;
 940
 941         switch (stage) {
 942         case OBD_CLEANUP_EARLY: {
 943                 struct lov_obd *lov = &obd->u.lov;
 944                 int i;
 945                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 946                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 947                                 continue;
 948                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 949                                        OBD_CLEANUP_EARLY);
 950                 }
 951                 break;
 952         }
 953         case OBD_CLEANUP_EXPORTS:
 954                 rc = obd_llog_finish(obd, 0);
 955                 if (rc != 0)
 956                         CERROR("failed to cleanup llogging subsystems\n");
 957                 break;
 958         }
 959         RETURN(rc);
 960 }
 961
 962 static int lov_cleanup(struct obd_device *obd)
 963 {
 964         struct lov_obd *lov = &obd->u.lov;
 965         struct list_head *pos, *tmp;
 966         struct pool_desc *pool;
 967
 968         lprocfs_obd_cleanup(obd);
 969
 970         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 971                 pool = list_entry(pos, struct pool_desc, pool_list);
 972                 list_del(&pool->pool_list);
 973                 lov_ost_pool_free(&(pool->pool_rr.lqr_pool));
 974                 lov_ost_pool_free(&(pool->pool_obds));
 975                 OBD_FREE_PTR(pool);
 976         }
 977         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 978         lov_ost_pool_free(&lov->lov_packed);
 979         lustre_hash_exit(lov->lov_pools_hash_body);
 980
 981         if (lov->lov_tgts) {
 982                 int i;
 983                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 984                         if (!lov->lov_tgts[i])
 985                                 continue;
 986
 987                         /* Inactive targets may never have connected */
 988                         if (lov->lov_tgts[i]->ltd_active ||
 989                             atomic_read(&lov->lov_refcount))
 990                             /* We should never get here - these
 991                                should have been removed in the
 992                              disconnect. */
 993                                 CERROR("lov tgt %d not cleaned!"
 994                                        " deathrow=%d, lovrc=%d\n",
 995                                        i, lov->lov_death_row,
 996                                        atomic_read(&lov->lov_refcount));
 997                         lov_del_target(obd, i, 0, 0);
 998                 }
 999                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
1000                          lov->lov_tgt_size);
1001                 lov->lov_tgt_size = 0;
1002         }
1003
1004         RETURN(0);
1005 }
1006
1007 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
1008 {
1009         struct lustre_cfg *lcfg = buf;
1010         struct obd_uuid obd_uuid;
1011         int cmd;
1012         int rc = 0;
1013         ENTRY;
1014
1015         switch(cmd = lcfg->lcfg_command) {
1016         case LCFG_LOV_ADD_OBD:
1017         case LCFG_LOV_ADD_INA:
1018         case LCFG_LOV_DEL_OBD: {
1019                 __u32 index;
1020                 int gen;
1021                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
1022                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1023                         GOTO(out, rc = -EINVAL);
1024
1025                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1026
1027                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1028                         GOTO(out, rc = -EINVAL);
1029                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1030                         GOTO(out, rc = -EINVAL);
1031                 if (cmd == LCFG_LOV_ADD_OBD)
1032                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
1033                 else if (cmd == LCFG_LOV_ADD_INA)
1034                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
1035                 else
1036                         rc = lov_del_target(obd, index, &obd_uuid, gen);
1037                 GOTO(out, rc);
1038         }
1039         case LCFG_PARAM: {
1040                 struct lprocfs_static_vars lvars = { 0 };
1041                 struct lov_desc *desc = &(obd->u.lov.desc);
1042
1043                 if (!desc)
1044                         GOTO(out, rc = -EINVAL);
1045
1046                 lprocfs_lov_init_vars(&lvars);
1047
1048                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1049                                               lcfg, obd);
1050                 GOTO(out, rc);
1051         }
1052         case LCFG_POOL_NEW:
1053         case LCFG_POOL_ADD:
1054         case LCFG_POOL_DEL:
1055         case LCFG_POOL_REM:
1056                 GOTO(out, rc);
1057
1058         default: {
1059                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1060                 GOTO(out, rc = -EINVAL);
1061
1062         }
1063         }
1064 out:
1065         RETURN(rc);
1066 }
1067
1068 #ifndef log2
1069 #define log2(n) ffz(~(n))
1070 #endif
1071
1072 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1073                              struct lov_stripe_md **ea,
1074                              struct obd_trans_info *oti)
1075 {
1076         struct lov_obd *lov;
1077         struct obdo *tmp_oa;
1078         struct obd_uuid *ost_uuid = NULL;
1079         int rc = 0, i;
1080         ENTRY;
1081
1082         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1083                 src_oa->o_flags == OBD_FL_DELORPHAN);
1084
1085         lov = &export->exp_obd->u.lov;
1086
1087         OBDO_ALLOC(tmp_oa);
1088         if (tmp_oa == NULL)
1089                 RETURN(-ENOMEM);
1090
1091         if (oti->oti_ost_uuid) {
1092                 ost_uuid = oti->oti_ost_uuid;
1093                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1094                        ost_uuid->uuid);
1095         }
1096
1097         lov_getref(export->exp_obd);
1098         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1099                 struct lov_stripe_md obj_md;
1100                 struct lov_stripe_md *obj_mdp = &obj_md;
1101                 struct lov_tgt_desc *tgt;
1102                 int err;
1103
1104                 tgt = lov->lov_tgts[i];
1105                 if (!tgt)
1106                         continue;
1107
1108                 /* if called for a specific target, we don't
1109                    care if it is not active. */
1110                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1111                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1112                         continue;
1113                 }
1114
1115                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1116                         continue;
1117
1118                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1119                        obd_uuid2str(ost_uuid));
1120
1121                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1122
1123                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1124                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1125                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1126                                  tmp_oa, &obj_mdp, oti);
1127                 if (err) {
1128                         /* This export will be disabled until it is recovered,
1129                            and then orphan recovery will be completed. */
1130                         CERROR("error in orphan recovery on OST idx %d/%d: "
1131                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1132                         rc = err;
1133                 }
1134
1135                 if (ost_uuid)
1136                         break;
1137         }
1138         lov_putref(export->exp_obd);
1139
1140         OBDO_FREE(tmp_oa);
1141         RETURN(rc);
1142 }
1143
1144 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1145                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1146 {
1147         struct lov_stripe_md *obj_mdp, *lsm;
1148         struct lov_obd *lov = &exp->exp_obd->u.lov;
1149         unsigned ost_idx;
1150         int rc, i;
1151         ENTRY;
1152
1153         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1154                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1155
1156         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1157         if (obj_mdp == NULL)
1158                 RETURN(-ENOMEM);
1159
1160         ost_idx = src_oa->o_nlink;
1161         lsm = *ea;
1162         if (lsm == NULL)
1163                 GOTO(out, rc = -EINVAL);
1164         if (ost_idx >= lov->desc.ld_tgt_count ||
1165             !lov->lov_tgts[ost_idx])
1166                 GOTO(out, rc = -EINVAL);
1167
1168         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1169                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1170                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1171                                 GOTO(out, rc = -EINVAL);
1172                         break;
1173                 }
1174         }
1175         if (i == lsm->lsm_stripe_count)
1176                 GOTO(out, rc = -EINVAL);
1177
1178         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1179 out:
1180         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1181         RETURN(rc);
1182 }
1183
1184 /* the LOV expects oa->o_id to be set to the LOV object id */
1185 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1186                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1187 {
1188         struct lov_obd *lov;
1189         struct obd_info oinfo;
1190         struct lov_request_set *set = NULL;
1191         struct lov_request *req;
1192         struct obd_statfs osfs;
1193         __u64 maxage;
1194         int rc = 0;
1195         ENTRY;
1196
1197         LASSERT(ea != NULL);
1198         if (exp == NULL)
1199                 RETURN(-EINVAL);
1200
1201         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1202             src_oa->o_flags == OBD_FL_DELORPHAN) {
1203                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1204                 RETURN(rc);
1205         }
1206
1207         lov = &exp->exp_obd->u.lov;
1208         if (!lov->desc.ld_active_tgt_count)
1209                 RETURN(-EIO);
1210
1211         /* Recreate a specific object id at the given OST index */
1212         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1213             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1214                  rc = lov_recreate(exp, src_oa, ea, oti);
1215                  RETURN(rc);
1216         }
1217
1218         maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1219         obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1220
1221         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1222         if (rc)
1223                 RETURN(rc);
1224
1225         list_for_each_entry(req, &set->set_list, rq_link) {
1226                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1227                 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1228                                 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1229                 lov_update_create_set(set, req, rc);
1230         }
1231         rc = lov_fini_create_set(set, ea);
1232         RETURN(rc);
1233 }
1234
1235 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1236 do {                                                                            \
1237         LASSERT((lsmp) != NULL);                                                \
1238         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1239                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1240                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1241                  (lsmp), (lsmp)->lsm_magic);                                    \
1242 } while (0)
1243
1244 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1245                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1246                        struct obd_export *md_exp)
1247 {
1248         struct lov_request_set *set;
1249         struct obd_info oinfo;
1250         struct lov_request *req;
1251         struct list_head *pos;
1252         struct lov_obd *lov;
1253         int rc = 0, err;
1254         ENTRY;
1255
1256         ASSERT_LSM_MAGIC(lsm);
1257
1258         if (!exp || !exp->exp_obd)
1259                 RETURN(-ENODEV);
1260
1261         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1262                 LASSERT(oti);
1263                 LASSERT(oti->oti_logcookies);
1264         }
1265
1266         lov = &exp->exp_obd->u.lov;
1267         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1268         if (rc)
1269                 RETURN(rc);
1270
1271         list_for_each (pos, &set->set_list) {
1272                 int err;
1273                 req = list_entry(pos, struct lov_request, rq_link);
1274
1275                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1276                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1277
1278                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1279                                   req->rq_oi.oi_oa, NULL, oti, NULL);
1280                 err = lov_update_common_set(set, req, err);
1281                 if (err) {
1282                         CERROR("error: destroying objid "LPX64" subobj "
1283                                LPX64" on OST idx %d: rc = %d\n",
1284                                oa->o_id, req->rq_oi.oi_oa->o_id,
1285                                req->rq_idx, err);
1286                         if (!rc)
1287                                 rc = err;
1288                 }
1289         }
1290
1291         if (rc == 0) {
1292                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1293                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1294         }
1295         err = lov_fini_destroy_set(set);
1296         RETURN(rc ? rc : err);
1297 }
1298
1299 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1300 {
1301         struct lov_request_set *set;
1302         struct lov_request *req;
1303         struct list_head *pos;
1304         struct lov_obd *lov;
1305         int err = 0, rc = 0;
1306         ENTRY;
1307
1308         LASSERT(oinfo);
1309         ASSERT_LSM_MAGIC(oinfo->oi_md);
1310
1311         if (!exp || !exp->exp_obd)
1312                 RETURN(-ENODEV);
1313
1314         lov = &exp->exp_obd->u.lov;
1315
1316         rc = lov_prep_getattr_set(exp, oinfo, &set);
1317         if (rc)
1318                 RETURN(rc);
1319
1320         list_for_each (pos, &set->set_list) {
1321                 req = list_entry(pos, struct lov_request, rq_link);
1322
1323                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1324                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1325                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1326
1327                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1328                                  &req->rq_oi);
1329                 err = lov_update_common_set(set, req, rc);
1330                 if (err) {
1331                         CERROR("error: getattr objid "LPX64" subobj "
1332                                LPX64" on OST idx %d: rc = %d\n",
1333                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1334                                req->rq_idx, err);
1335                         break;
1336                 }
1337         }
1338
1339         rc = lov_fini_getattr_set(set);
1340         if (err)
1341                 rc = err;
1342         RETURN(rc);
1343 }
1344
1345 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1346                                  void *data, int rc)
1347 {
1348         struct lov_request_set *lovset = (struct lov_request_set *)data;
1349         int err;
1350         ENTRY;
1351
1352         /* don't do attribute merge if this aysnc op failed */
1353         if (rc)
1354                 lovset->set_completes = 0;
1355         err = lov_fini_getattr_set(lovset);
1356         RETURN(rc ? rc : err);
1357 }
1358
1359 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1360                               struct ptlrpc_request_set *rqset)
1361 {
1362         struct lov_request_set *lovset;
1363         struct lov_obd *lov;
1364         struct list_head *pos;
1365         struct lov_request *req;
1366         int rc = 0, err;
1367         ENTRY;
1368
1369         LASSERT(oinfo);
1370         ASSERT_LSM_MAGIC(oinfo->oi_md);
1371
1372         if (!exp || !exp->exp_obd)
1373                 RETURN(-ENODEV);
1374
1375         lov = &exp->exp_obd->u.lov;
1376
1377         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1378         if (rc)
1379                 RETURN(rc);
1380
1381         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1382                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1383                oinfo->oi_md->lsm_stripe_size);
1384
1385         list_for_each (pos, &lovset->set_list) {
1386                 req = list_entry(pos, struct lov_request, rq_link);
1387
1388                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1389                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1390                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1391                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1392                                        &req->rq_oi, rqset);
1393                 if (rc) {
1394                         CERROR("error: getattr objid "LPX64" subobj "
1395                                LPX64" on OST idx %d: rc = %d\n",
1396                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1397                                req->rq_idx, rc);
1398                         GOTO(out, rc);
1399                 }
1400         }
1401
1402         if (!list_empty(&rqset->set_requests)) {
1403                 LASSERT(rc == 0);
1404                 LASSERT (rqset->set_interpret == NULL);
1405                 rqset->set_interpret = lov_getattr_interpret;
1406                 rqset->set_arg = (void *)lovset;
1407                 RETURN(rc);
1408         }
1409 out:
1410         if (rc)
1411                 lovset->set_completes = 0;
1412         err = lov_fini_getattr_set(lovset);
1413         RETURN(rc ? rc : err);
1414 }
1415
1416 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1417                        struct obd_trans_info *oti)
1418 {
1419         struct lov_request_set *set;
1420         struct lov_obd *lov;
1421         struct list_head *pos;
1422         struct lov_request *req;
1423         int err = 0, rc = 0;
1424         ENTRY;
1425
1426         LASSERT(oinfo);
1427         ASSERT_LSM_MAGIC(oinfo->oi_md);
1428
1429         if (!exp || !exp->exp_obd)
1430                 RETURN(-ENODEV);
1431
1432         /* for now, we only expect the following updates here */
1433         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1434                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1435                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1436                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1437                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1438                                             OBD_MD_FLGID | OBD_MD_FLFID |
1439                                             OBD_MD_FLGENER)));
1440         lov = &exp->exp_obd->u.lov;
1441         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1442         if (rc)
1443                 RETURN(rc);
1444
1445         list_for_each (pos, &set->set_list) {
1446                 req = list_entry(pos, struct lov_request, rq_link);
1447
1448                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1449                                  &req->rq_oi, NULL);
1450                 err = lov_update_setattr_set(set, req, rc);
1451                 if (err) {
1452                         CERROR("error: setattr objid "LPX64" subobj "
1453                                LPX64" on OST idx %d: rc = %d\n",
1454                                set->set_oi->oi_oa->o_id,
1455                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1456                         if (!rc)
1457                                 rc = err;
1458                 }
1459         }
1460         err = lov_fini_setattr_set(set);
1461         if (!rc)
1462                 rc = err;
1463         RETURN(rc);
1464 }
1465
1466 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1467                                  void *data, int rc)
1468 {
1469         struct lov_request_set *lovset = (struct lov_request_set *)data;
1470         int err;
1471         ENTRY;
1472
1473         if (rc)
1474                 lovset->set_completes = 0;
1475         err = lov_fini_setattr_set(lovset);
1476         RETURN(rc ? rc : err);
1477 }
1478
1479 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1480    needed. Otherwise, a client is waiting for responses. */
1481 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1482                              struct obd_trans_info *oti,
1483                              struct ptlrpc_request_set *rqset)
1484 {
1485         struct lov_request_set *set;
1486         struct lov_request *req;
1487         struct list_head *pos;
1488         struct lov_obd *lov;
1489         int rc = 0;
1490         ENTRY;
1491
1492         LASSERT(oinfo);
1493         ASSERT_LSM_MAGIC(oinfo->oi_md);
1494         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1495                 LASSERT(oti);
1496                 LASSERT(oti->oti_logcookies);
1497         }
1498
1499         if (!exp || !exp->exp_obd)
1500                 RETURN(-ENODEV);
1501
1502         lov = &exp->exp_obd->u.lov;
1503         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1504         if (rc)
1505                 RETURN(rc);
1506
1507         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1508                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1509                oinfo->oi_md->lsm_stripe_size);
1510
1511         list_for_each (pos, &set->set_list) {
1512                 req = list_entry(pos, struct lov_request, rq_link);
1513
1514                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1515                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1516
1517                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1518                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1519                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1520
1521                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1522                                        &req->rq_oi, oti, rqset);
1523                 if (rc) {
1524                         CERROR("error: setattr objid "LPX64" subobj "
1525                                LPX64" on OST idx %d: rc = %d\n",
1526                                set->set_oi->oi_oa->o_id,
1527                                req->rq_oi.oi_oa->o_id,
1528                                req->rq_idx, rc);
1529                         break;
1530                 }
1531         }
1532
1533         /* If we are not waiting for responses on async requests, return. */
1534         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1535                 int err;
1536                 if (rc)
1537                         set->set_completes = 0;
1538                 err = lov_fini_setattr_set(set);
1539                 RETURN(rc ? rc : err);
1540         }
1541
1542         LASSERT(rqset->set_interpret == NULL);
1543         rqset->set_interpret = lov_setattr_interpret;
1544         rqset->set_arg = (void *)set;
1545
1546         RETURN(0);
1547 }
1548
1549 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1550                                void *data, int rc)
1551 {
1552         struct lov_request_set *lovset = (struct lov_request_set *)data;
1553         int err;
1554         ENTRY;
1555
1556         if (rc)
1557                 lovset->set_completes = 0;
1558         err = lov_fini_punch_set(lovset);
1559         RETURN(rc ? rc : err);
1560 }
1561
1562 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1563  * we can send this 'punch' to just the authoritative node and the nodes
1564  * that the punch will affect. */
1565 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1566                      struct obd_trans_info *oti,
1567                      struct ptlrpc_request_set *rqset)
1568 {
1569         struct lov_request_set *set;
1570         struct lov_obd *lov;
1571         struct list_head *pos;
1572         struct lov_request *req;
1573         int rc = 0;
1574         ENTRY;
1575
1576         LASSERT(oinfo);
1577         ASSERT_LSM_MAGIC(oinfo->oi_md);
1578
1579         if (!exp || !exp->exp_obd)
1580                 RETURN(-ENODEV);
1581
1582         lov = &exp->exp_obd->u.lov;
1583         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1584         if (rc)
1585                 RETURN(rc);
1586
1587         list_for_each (pos, &set->set_list) {
1588                 req = list_entry(pos, struct lov_request, rq_link);
1589
1590                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1591                                &req->rq_oi, NULL, rqset);
1592                 if (rc) {
1593                         CERROR("error: punch objid "LPX64" subobj "LPX64
1594                                " on OST idx %d: rc = %d\n",
1595                                set->set_oi->oi_oa->o_id,
1596                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1597                         break;
1598                 }
1599         }
1600
1601         if (rc || list_empty(&rqset->set_requests)) {
1602                 int err;
1603                 err = lov_fini_punch_set(set);
1604                 RETURN(rc ? rc : err);
1605         }
1606
1607         LASSERT(rqset->set_interpret == NULL);
1608         rqset->set_interpret = lov_punch_interpret;
1609         rqset->set_arg = (void *)set;
1610
1611         RETURN(0);
1612 }
1613
1614 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1615                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1616                     void *capa)
1617 {
1618         struct lov_request_set *set;
1619         struct obd_info oinfo;
1620         struct lov_obd *lov;
1621         struct list_head *pos;
1622         struct lov_request *req;
1623         int err = 0, rc = 0;
1624         ENTRY;
1625
1626         ASSERT_LSM_MAGIC(lsm);
1627
1628         if (!exp->exp_obd)
1629                 RETURN(-ENODEV);
1630
1631         lov = &exp->exp_obd->u.lov;
1632         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1633         if (rc)
1634                 RETURN(rc);
1635
1636         list_for_each (pos, &set->set_list) {
1637                 req = list_entry(pos, struct lov_request, rq_link);
1638
1639                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1640                               req->rq_oi.oi_oa, NULL,
1641                               req->rq_oi.oi_policy.l_extent.start,
1642                               req->rq_oi.oi_policy.l_extent.end, capa);
1643                 err = lov_update_common_set(set, req, rc);
1644                 if (err) {
1645                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1646                                " on OST idx %d: rc = %d\n",
1647                                set->set_oi->oi_oa->o_id,
1648                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1649                         if (!rc)
1650                                 rc = err;
1651                 }
1652         }
1653         err = lov_fini_sync_set(set);
1654         if (!rc)
1655                 rc = err;
1656         RETURN(rc);
1657 }
1658
1659 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1660                          obd_count oa_bufs, struct brw_page *pga)
1661 {
1662         struct obd_info oinfo = { { { 0 } } };
1663         int i, rc = 0;
1664
1665         oinfo.oi_oa = lov_oinfo->oi_oa;
1666
1667         /* The caller just wants to know if there's a chance that this
1668          * I/O can succeed */
1669         for (i = 0; i < oa_bufs; i++) {
1670                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1671                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1672                 obd_off start, end;
1673
1674                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1675                                            pga[i].off + pga[i].count,
1676                                            &start, &end))
1677                         continue;
1678
1679                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1680                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1681                         return -EIO;
1682                 }
1683
1684                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1685                              1, &pga[i], NULL);
1686                 if (rc)
1687                         break;
1688         }
1689         return rc;
1690 }
1691
1692 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1693                    obd_count oa_bufs, struct brw_page *pga,
1694                    struct obd_trans_info *oti)
1695 {
1696         struct lov_request_set *set;
1697         struct lov_request *req;
1698         struct list_head *pos;
1699         struct lov_obd *lov = &exp->exp_obd->u.lov;
1700         int err, rc = 0;
1701         ENTRY;
1702
1703         ASSERT_LSM_MAGIC(oinfo->oi_md);
1704
1705         if (cmd == OBD_BRW_CHECK) {
1706                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1707                 RETURN(rc);
1708         }
1709
1710         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1711         if (rc)
1712                 RETURN(rc);
1713
1714         list_for_each (pos, &set->set_list) {
1715                 struct obd_export *sub_exp;
1716                 struct brw_page *sub_pga;
1717                 req = list_entry(pos, struct lov_request, rq_link);
1718
1719                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1720                 sub_pga = set->set_pga + req->rq_pgaidx;
1721                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1722                              sub_pga, oti);
1723                 if (rc)
1724                         break;
1725                 lov_update_common_set(set, req, rc);
1726         }
1727
1728         err = lov_fini_brw_set(set);
1729         if (!rc)
1730                 rc = err;
1731         RETURN(rc);
1732 }
1733
1734 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1735                              int rc)
1736 {
1737         struct lov_request_set *lovset = (struct lov_request_set *)data;
1738         ENTRY;
1739
1740         if (rc) {
1741                 lovset->set_completes = 0;
1742                 lov_fini_brw_set(lovset);
1743         } else {
1744                 rc = lov_fini_brw_set(lovset);
1745         }
1746
1747         RETURN(rc);
1748 }
1749
1750 static int lov_brw_async(int cmd, struct obd_export *exp,
1751                          struct obd_info *oinfo, obd_count oa_bufs,
1752                          struct brw_page *pga, struct obd_trans_info *oti,
1753                          struct ptlrpc_request_set *set)
1754 {
1755         struct lov_request_set *lovset;
1756         struct lov_request *req;
1757         struct list_head *pos;
1758         struct lov_obd *lov = &exp->exp_obd->u.lov;
1759         int rc = 0;
1760         ENTRY;
1761
1762         LASSERT(oinfo);
1763         ASSERT_LSM_MAGIC(oinfo->oi_md);
1764
1765         if (cmd == OBD_BRW_CHECK) {
1766                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1767                 RETURN(rc);
1768         }
1769
1770         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1771         if (rc)
1772                 RETURN(rc);
1773
1774         list_for_each (pos, &lovset->set_list) {
1775                 struct obd_export *sub_exp;
1776                 struct brw_page *sub_pga;
1777                 req = list_entry(pos, struct lov_request, rq_link);
1778
1779                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1780                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1781                 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1782                                    sub_pga, oti, set);
1783                 if (rc)
1784                         GOTO(out, rc);
1785                 lov_update_common_set(lovset, req, rc);
1786         }
1787         LASSERT(rc == 0);
1788         LASSERT(set->set_interpret == NULL);
1789         LASSERT(set->set_arg == NULL);
1790         rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1791         if (rc)
1792                 GOTO(out, rc);
1793
1794         RETURN(rc);
1795 out:
1796         lov_fini_brw_set(lovset);
1797         RETURN(rc);
1798 }
1799
1800 static int lov_ap_make_ready(void *data, int cmd)
1801 {
1802         struct lov_async_page *lap = lap_from_cookie(data);
1803
1804         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1805 }
1806
1807 static int lov_ap_refresh_count(void *data, int cmd)
1808 {
1809         struct lov_async_page *lap = lap_from_cookie(data);
1810
1811         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1812                                                      cmd);
1813 }
1814
1815 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1816 {
1817         struct lov_async_page *lap = lap_from_cookie(data);
1818
1819         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1820         /* XXX woah, shouldn't we be altering more here?  size? */
1821         oa->o_id = lap->lap_loi_id;
1822         oa->o_gr = lap->lap_loi_gr;
1823         oa->o_valid |= OBD_MD_FLGROUP;
1824         oa->o_stripe_idx = lap->lap_stripe;
1825 }
1826
1827 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1828                                obd_valid valid)
1829 {
1830         struct lov_async_page *lap = lap_from_cookie(data);
1831
1832         lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1833 }
1834
1835 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1836 {
1837         struct lov_async_page *lap = lap_from_cookie(data);
1838
1839         /* in a raid1 regime this would down a count of many ios
1840          * in flight, onl calling the caller_ops completion when all
1841          * the raid1 ios are complete */
1842         rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1843         return rc;
1844 }
1845
1846 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1847 {
1848         struct lov_async_page *lap = lap_from_cookie(data);
1849         return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1850 }
1851
1852 static struct obd_async_page_ops lov_async_page_ops = {
1853         .ap_make_ready =        lov_ap_make_ready,
1854         .ap_refresh_count =     lov_ap_refresh_count,
1855         .ap_fill_obdo =         lov_ap_fill_obdo,
1856         .ap_update_obdo =       lov_ap_update_obdo,
1857         .ap_completion =        lov_ap_completion,
1858         .ap_lookup_capa =       lov_ap_lookup_capa,
1859 };
1860
1861 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1862                            struct lov_oinfo *loi, cfs_page_t *page,
1863                            obd_off offset, struct obd_async_page_ops *ops,
1864                            void *data, void **res, int nocache,
1865                            struct lustre_handle *lockh)
1866 {
1867         struct lov_obd *lov = &exp->exp_obd->u.lov;
1868         struct lov_async_page *lap;
1869         struct lov_lock_handles *lov_lockh = NULL;
1870         int rc = 0;
1871         ENTRY;
1872
1873         if (!page) {
1874                 int i = 0;
1875                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1876                    Only because of this layering limitation will a client
1877                    mount with no osts fail */
1878                 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1879                        !lov->lov_tgts[i]->ltd_exp) {
1880                         i++;
1881                         if (i >= lov->desc.ld_tgt_count)
1882                                 RETURN(-ENOMEDIUM);
1883                 }
1884                 rc = size_round(sizeof(*lap)) +
1885                         obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1886                                             NULL, NULL, 0, NULL, NULL, NULL, 0,
1887                                             NULL);
1888                 RETURN(rc);
1889         }
1890         ASSERT_LSM_MAGIC(lsm);
1891         LASSERT(loi == NULL);
1892
1893         lap = *res;
1894         lap->lap_magic = LOV_AP_MAGIC;
1895         lap->lap_caller_ops = ops;
1896         lap->lap_caller_data = data;
1897
1898         /* for now only raid 0 which passes through */
1899         lap->lap_stripe = lov_stripe_number(lsm, offset);
1900         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1901         loi = lsm->lsm_oinfo[lap->lap_stripe];
1902
1903         /* so the callback doesn't need the lsm */
1904         lap->lap_loi_id = loi->loi_id;
1905         lap->lap_loi_gr = lsm->lsm_object_gr;
1906         LASSERT(lsm->lsm_object_gr > 0);
1907
1908         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1909
1910         if (lockh) {
1911                 lov_lockh = lov_handle2llh(lockh);
1912                 if (lov_lockh) {
1913                         lockh = lov_lockh->llh_handles + lap->lap_stripe;
1914                 }
1915         }
1916
1917         rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1918                                  lsm, loi, page, lap->lap_sub_offset,
1919                                  &lov_async_page_ops, lap,
1920                                  &lap->lap_sub_cookie, nocache, lockh);
1921         if (lov_lockh)
1922                 lov_llh_put(lov_lockh);
1923         if (rc)
1924                 RETURN(rc);
1925         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1926                lap->lap_sub_cookie, offset);
1927         RETURN(0);
1928 }
1929
1930 static int lov_queue_async_io(struct obd_export *exp,
1931                               struct lov_stripe_md *lsm,
1932                               struct lov_oinfo *loi, void *cookie,
1933                               int cmd, obd_off off, int count,
1934                               obd_flag brw_flags, obd_flag async_flags)
1935 {
1936         struct lov_obd *lov = &exp->exp_obd->u.lov;
1937         struct lov_async_page *lap;
1938         int rc;
1939
1940         LASSERT(loi == NULL);
1941
1942         ASSERT_LSM_MAGIC(lsm);
1943
1944         lap = lap_from_cookie(cookie);
1945
1946         loi = lsm->lsm_oinfo[lap->lap_stripe];
1947
1948         rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1949                                 loi, lap->lap_sub_cookie, cmd, off, count,
1950                                 brw_flags, async_flags);
1951         RETURN(rc);
1952 }
1953
1954 static int lov_set_async_flags(struct obd_export *exp,
1955                                struct lov_stripe_md *lsm,
1956                                struct lov_oinfo *loi, void *cookie,
1957                                obd_flag async_flags)
1958 {
1959         struct lov_obd *lov = &exp->exp_obd->u.lov;
1960         struct lov_async_page *lap;
1961         int rc;
1962
1963         LASSERT(loi == NULL);
1964
1965         ASSERT_LSM_MAGIC(lsm);
1966
1967         lap = lap_from_cookie(cookie);
1968
1969         loi = lsm->lsm_oinfo[lap->lap_stripe];
1970
1971         rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1972                                  lsm, loi, lap->lap_sub_cookie, async_flags);
1973         RETURN(rc);
1974 }
1975
1976 static int lov_queue_group_io(struct obd_export *exp,
1977                               struct lov_stripe_md *lsm,
1978                               struct lov_oinfo *loi,
1979                               struct obd_io_group *oig, void *cookie,
1980                               int cmd, obd_off off, int count,
1981                               obd_flag brw_flags, obd_flag async_flags)
1982 {
1983         struct lov_obd *lov = &exp->exp_obd->u.lov;
1984         struct lov_async_page *lap;
1985         int rc;
1986
1987         LASSERT(loi == NULL);
1988
1989         ASSERT_LSM_MAGIC(lsm);
1990
1991         lap = lap_from_cookie(cookie);
1992
1993         loi = lsm->lsm_oinfo[lap->lap_stripe];
1994
1995         rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1996                                 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1997                                 brw_flags, async_flags);
1998         RETURN(rc);
1999 }
2000
2001 /* this isn't exactly optimal.  we may have queued sync io in oscs on
2002  * all stripes, but we don't record that fact at queue time.  so we
2003  * trigger sync io on all stripes. */
2004 static int lov_trigger_group_io(struct obd_export *exp,
2005                                 struct lov_stripe_md *lsm,
2006                                 struct lov_oinfo *loi,
2007                                 struct obd_io_group *oig)
2008 {
2009         struct lov_obd *lov = &exp->exp_obd->u.lov;
2010         int rc = 0, i, err;
2011
2012         LASSERT(loi == NULL);
2013
2014         ASSERT_LSM_MAGIC(lsm);
2015
2016         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2017                 loi = lsm->lsm_oinfo[i];
2018                 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2019                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2020                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2021                         continue;
2022                 }
2023
2024                 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2025                                            lsm, loi, oig);
2026                 if (rc == 0 && err != 0)
2027                         rc = err;
2028         };
2029         RETURN(rc);
2030 }
2031
2032 static int lov_teardown_async_page(struct obd_export *exp,
2033                                    struct lov_stripe_md *lsm,
2034                                    struct lov_oinfo *loi, void *cookie)
2035 {
2036         struct lov_obd *lov = &exp->exp_obd->u.lov;
2037         struct lov_async_page *lap;
2038         int rc;
2039
2040         LASSERT(loi == NULL);
2041
2042         ASSERT_LSM_MAGIC(lsm);
2043
2044         lap = lap_from_cookie(cookie);
2045
2046         loi = lsm->lsm_oinfo[lap->lap_stripe];
2047
2048         rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2049                                      lsm, loi, lap->lap_sub_cookie);
2050         if (rc) {
2051                 CERROR("unable to teardown sub cookie %p: %d\n",
2052                        lap->lap_sub_cookie, rc);
2053                 RETURN(rc);
2054         }
2055         RETURN(rc);
2056 }
2057
2058 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2059                                  void *data, int rc)
2060 {
2061         struct lov_request_set *lovset = (struct lov_request_set *)data;
2062         ENTRY;
2063         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2064         RETURN(rc);
2065 }
2066
2067 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2068                        struct ldlm_enqueue_info *einfo,
2069                        struct ptlrpc_request_set *rqset)
2070 {
2071         ldlm_mode_t mode = einfo->ei_mode;
2072         struct lov_request_set *set;
2073         struct lov_request *req;
2074         struct list_head *pos;
2075         struct lov_obd *lov;
2076         ldlm_error_t rc;
2077         ENTRY;
2078
2079         LASSERT(oinfo);
2080         ASSERT_LSM_MAGIC(oinfo->oi_md);
2081         LASSERT(mode == (mode & -mode));
2082
2083         /* we should never be asked to replay a lock this way. */
2084         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2085
2086         if (!exp || !exp->exp_obd)
2087                 RETURN(-ENODEV);
2088
2089         lov = &exp->exp_obd->u.lov;
2090         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2091         if (rc)
2092                 RETURN(rc);
2093
2094         list_for_each (pos, &set->set_list) {
2095                 req = list_entry(pos, struct lov_request, rq_link);
2096
2097                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2098                                  &req->rq_oi, einfo, rqset);
2099                 if (rc != ELDLM_OK)
2100                         GOTO(out, rc);
2101         }
2102
2103         if (rqset && !list_empty(&rqset->set_requests)) {
2104                 LASSERT(rc == 0);
2105                 LASSERT(rqset->set_interpret == NULL);
2106                 rqset->set_interpret = lov_enqueue_interpret;
2107                 rqset->set_arg = (void *)set;
2108                 RETURN(rc);
2109         }
2110 out:
2111         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2112         RETURN(rc);
2113 }
2114
2115 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2116                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2117                      int *flags, void *data, struct lustre_handle *lockh)
2118 {
2119         struct lov_request_set *set;
2120         struct obd_info oinfo;
2121         struct lov_request *req;
2122         struct list_head *pos;
2123         struct lov_obd *lov = &exp->exp_obd->u.lov;
2124         struct lustre_handle *lov_lockhp;
2125         int lov_flags, rc = 0;
2126         ENTRY;
2127
2128         ASSERT_LSM_MAGIC(lsm);
2129         LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2130
2131         if (!exp || !exp->exp_obd)
2132                 RETURN(-ENODEV);
2133
2134         lov = &exp->exp_obd->u.lov;
2135         rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2136         if (rc)
2137                 RETURN(rc);
2138
2139         list_for_each (pos, &set->set_list) {
2140                 ldlm_policy_data_t sub_policy;
2141                 req = list_entry(pos, struct lov_request, rq_link);
2142                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2143                 LASSERT(lov_lockhp);
2144
2145                 lov_flags = *flags;
2146                 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2147
2148                 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2149                                req->rq_oi.oi_md, type, &sub_policy,
2150                                mode, &lov_flags, data, lov_lockhp);
2151                 rc = lov_update_match_set(set, req, rc);
2152                 if (rc <= 0)
2153                         break;
2154         }
2155         lov_fini_match_set(set, mode, *flags);
2156         RETURN(rc);
2157 }
2158
2159 static int lov_change_cbdata(struct obd_export *exp,
2160                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
2161                              void *data)
2162 {
2163         struct lov_obd *lov;
2164         int rc = 0, i;
2165         ENTRY;
2166
2167         ASSERT_LSM_MAGIC(lsm);
2168
2169         if (!exp || !exp->exp_obd)
2170                 RETURN(-ENODEV);
2171
2172         LASSERT(lsm->lsm_object_gr > 0);
2173
2174         lov = &exp->exp_obd->u.lov;
2175         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2176                 struct lov_stripe_md submd;
2177                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2178
2179                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2180                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2181                         continue;
2182                 }
2183
2184                 submd.lsm_object_id = loi->loi_id;
2185                 submd.lsm_object_gr = lsm->lsm_object_gr;
2186                 submd.lsm_stripe_count = 0;
2187                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2188                                        &submd, it, data);
2189         }
2190         RETURN(rc);
2191 }
2192
2193 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2194                       __u32 mode, struct lustre_handle *lockh)
2195 {
2196         struct lov_request_set *set;
2197         struct obd_info oinfo;
2198         struct lov_request *req;
2199         struct list_head *pos;
2200         struct lov_obd *lov = &exp->exp_obd->u.lov;
2201         struct lustre_handle *lov_lockhp;
2202         int err = 0, rc = 0;
2203         ENTRY;
2204
2205         ASSERT_LSM_MAGIC(lsm);
2206
2207         if (!exp || !exp->exp_obd)
2208                 RETURN(-ENODEV);
2209
2210         LASSERT(lsm->lsm_object_gr > 0);
2211         LASSERT(lockh);
2212         lov = &exp->exp_obd->u.lov;
2213         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2214         if (rc)
2215                 RETURN(rc);
2216
2217         list_for_each (pos, &set->set_list) {
2218                 req = list_entry(pos, struct lov_request, rq_link);
2219                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2220
2221                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2222                                 req->rq_oi.oi_md, mode, lov_lockhp);
2223                 rc = lov_update_common_set(set, req, rc);
2224                 if (rc) {
2225                         CERROR("error: cancel objid "LPX64" subobj "
2226                                LPX64" on OST idx %d: rc = %d\n",
2227                                lsm->lsm_object_id,
2228                                req->rq_oi.oi_md->lsm_object_id,
2229                                req->rq_idx, rc);
2230                         err = rc;
2231                 }
2232
2233         }
2234         lov_fini_cancel_set(set);
2235         RETURN(err);
2236 }
2237
2238 static int lov_cancel_unused(struct obd_export *exp,
2239                              struct lov_stripe_md *lsm,
2240                              int flags, void *opaque)
2241 {
2242         struct lov_obd *lov;
2243         int rc = 0, i;
2244         ENTRY;
2245
2246         if (!exp || !exp->exp_obd)
2247                 RETURN(-ENODEV);
2248
2249         lov = &exp->exp_obd->u.lov;
2250         if (lsm == NULL) {
2251                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2252                         int err;
2253                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2254                                 continue;
2255
2256                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2257                                                 flags, opaque);
2258                         if (!rc)
2259                                 rc = err;
2260                 }
2261                 RETURN(rc);
2262         }
2263
2264         ASSERT_LSM_MAGIC(lsm);
2265
2266         LASSERT(lsm->lsm_object_gr > 0);
2267         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2268                 struct lov_stripe_md submd;
2269                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2270                 int err;
2271
2272                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2273                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2274                         continue;
2275                 }
2276
2277                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2278                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2279
2280                 submd.lsm_object_id = loi->loi_id;
2281                 submd.lsm_object_gr = lsm->lsm_object_gr;
2282                 submd.lsm_stripe_count = 0;
2283                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2284                                         &submd, flags, opaque);
2285                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2286                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2287                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2288                                loi->loi_id, loi->loi_ost_idx, err);
2289                         if (!rc)
2290                                 rc = err;
2291                 }
2292         }
2293         RETURN(rc);
2294 }
2295
2296 static int lov_join_lru(struct obd_export *exp,
2297                         struct lov_stripe_md *lsm, int join)
2298 {
2299         struct lov_obd *lov;
2300         int i, count = 0;
2301         ENTRY;
2302
2303         ASSERT_LSM_MAGIC(lsm);
2304         if (!exp || !exp->exp_obd)
2305                 RETURN(-ENODEV);
2306
2307         lov = &exp->exp_obd->u.lov;
2308         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2309                 struct lov_stripe_md submd;
2310                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2311                 int rc = 0;
2312
2313                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2314                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2315                         continue;
2316                 }
2317
2318                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2319                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2320
2321                 submd.lsm_object_id = loi->loi_id;
2322                 submd.lsm_object_gr = lsm->lsm_object_gr;
2323                 submd.lsm_stripe_count = 0;
2324                 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2325                                   &submd, join);
2326                 if (rc < 0) {
2327                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2328                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2329                                loi->loi_id, loi->loi_ost_idx, rc);
2330                         return rc;
2331                 } else {
2332                         count += rc;
2333                 }
2334         }
2335         RETURN(count);
2336 }
2337
2338 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2339                                 void *data, int rc)
2340 {
2341         struct lov_request_set *lovset = (struct lov_request_set *)data;
2342         int err;
2343         ENTRY;
2344
2345         if (rc)
2346                 lovset->set_completes = 0;
2347
2348         err = lov_fini_statfs_set(lovset);
2349         RETURN(rc ? rc : err);
2350 }
2351
2352 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2353                             __u64 max_age, struct ptlrpc_request_set *rqset)
2354 {
2355         struct lov_request_set *set;
2356         struct lov_request *req;
2357         struct list_head *pos;
2358         struct lov_obd *lov;
2359         int rc = 0;
2360         ENTRY;
2361
2362         LASSERT(oinfo != NULL);
2363         LASSERT(oinfo->oi_osfs != NULL);
2364
2365         lov = &obd->u.lov;
2366         rc = lov_prep_statfs_set(obd, oinfo, &set);
2367         if (rc)
2368                 RETURN(rc);
2369
2370         list_for_each (pos, &set->set_list) {
2371                 struct obd_device *osc_obd;
2372
2373                 req = list_entry(pos, struct lov_request, rq_link);
2374
2375                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2376                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2377                 if (rc)
2378                         break;
2379         }
2380
2381         if (rc || list_empty(&rqset->set_requests)) {
2382                 int err;
2383                 if (rc)
2384                         set->set_completes = 0;
2385                 err = lov_fini_statfs_set(set);
2386                 RETURN(rc ? rc : err);
2387         }
2388
2389         LASSERT(rqset->set_interpret == NULL);
2390         rqset->set_interpret = lov_statfs_interpret;
2391         rqset->set_arg = (void *)set;
2392         RETURN(0);
2393 }
2394
2395 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2396                       __u64 max_age, __u32 flags)
2397 {
2398         struct ptlrpc_request_set *set = NULL;
2399         struct obd_info oinfo = { { { 0 } } };
2400         int rc = 0;
2401         ENTRY;
2402
2403
2404         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2405          * statfs requests */
2406         set = ptlrpc_prep_set();
2407         if (set == NULL)
2408                 RETURN(-ENOMEM);
2409
2410         oinfo.oi_osfs = osfs;
2411         oinfo.oi_flags = flags;
2412         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2413         if (rc == 0)
2414                 rc = ptlrpc_set_wait(set);
2415         ptlrpc_set_destroy(set);
2416
2417         RETURN(rc);
2418 }
2419
2420 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2421                          void *karg, void *uarg)
2422 {
2423         struct obd_device *obddev = class_exp2obd(exp);
2424         struct lov_obd *lov = &obddev->u.lov;
2425         int i, rc, count = lov->desc.ld_tgt_count;
2426         struct obd_uuid *uuidp;
2427         ENTRY;
2428
2429         switch (cmd) {
2430         case IOC_OBD_STATFS: {
2431                 struct obd_ioctl_data *data = karg;
2432                 struct obd_device *osc_obd;
2433                 struct obd_statfs stat_buf = {0};
2434                 __u32 index;
2435
2436                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2437                 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
2438
2439                 if ((index >= count))
2440                         RETURN(-ENODEV);
2441
2442                 if (!lov->lov_tgts[index])
2443                         /* Try again with the next index */
2444                         RETURN(-EAGAIN);
2445                 if (!lov->lov_tgts[index]->ltd_active)
2446                         RETURN(-ENODATA);
2447
2448                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2449                 if (!osc_obd)
2450                         RETURN(-EINVAL);
2451
2452                 /* got statfs data */
2453                 rc = obd_statfs(osc_obd, &stat_buf,
2454                                 cfs_time_current_64() - HZ, 0);
2455                 if (rc)
2456                         RETURN(rc);
2457                 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2458                         RETURN(rc);
2459                 /* copy UUID */
2460                 rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2461                                   data->ioc_plen2);
2462                 break;
2463         }
2464         case OBD_IOC_LOV_GET_CONFIG: {
2465                 struct obd_ioctl_data *data;
2466                 struct lov_desc *desc;
2467                 char *buf = NULL;
2468                 __u32 *genp;
2469
2470                 len = 0;
2471                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2472                         RETURN(-EINVAL);
2473
2474                 data = (struct obd_ioctl_data *)buf;
2475
2476                 if (sizeof(*desc) > data->ioc_inllen1) {
2477                         obd_ioctl_freedata(buf, len);
2478                         RETURN(-EINVAL);
2479                 }
2480
2481                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2482                         obd_ioctl_freedata(buf, len);
2483                         RETURN(-EINVAL);
2484                 }
2485
2486                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2487                         obd_ioctl_freedata(buf, len);
2488                         RETURN(-EINVAL);
2489                 }
2490
2491                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2492                 memcpy(desc, &(lov->desc), sizeof(*desc));
2493
2494                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2495                 genp = (__u32 *)data->ioc_inlbuf3;
2496                 /* the uuid will be empty for deleted OSTs */
2497                 for (i = 0; i < count; i++, uuidp++, genp++) {
2498                         if (!lov->lov_tgts[i])
2499                                 continue;
2500                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2501                         *genp = lov->lov_tgts[i]->ltd_gen;
2502                 }
2503
2504                 rc = copy_to_user((void *)uarg, buf, len);
2505                 if (rc)
2506                         rc = -EFAULT;
2507                 obd_ioctl_freedata(buf, len);
2508                 break;
2509         }
2510         case LL_IOC_LOV_SETSTRIPE:
2511                 rc = lov_setstripe(exp, karg, uarg);
2512                 break;
2513         case LL_IOC_LOV_GETSTRIPE:
2514                 rc = lov_getstripe(exp, karg, uarg);
2515                 break;
2516         case LL_IOC_LOV_SETEA:
2517                 rc = lov_setea(exp, karg, uarg);
2518                 break;
2519         default: {
2520                 int set = 0;
2521
2522                 if (count == 0)
2523                         RETURN(-ENOTTY);
2524
2525                 rc = 0;
2526                 for (i = 0; i < count; i++) {
2527                         int err;
2528
2529                         /* OST was disconnected */
2530                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2531                                 continue;
2532
2533                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2534                                             len, karg, uarg);
2535                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2536                                 RETURN(err);
2537                         } else if (err) {
2538                                 if (lov->lov_tgts[i]->ltd_active) {
2539                                         CDEBUG(err == -ENOTTY ?
2540                                                D_IOCTL : D_WARNING,
2541                                                "iocontrol OSC %s on OST "
2542                                                "idx %d cmd %x: err = %d\n",
2543                                                lov_uuid2str(lov, i),
2544                                                i, cmd, err);
2545                                         if (!rc)
2546                                                 rc = err;
2547                                 }
2548                         } else {
2549                                 set = 1;
2550                         }
2551                 }
2552                 if (!set && !rc)
2553                         rc = -EIO;
2554         }
2555         }
2556
2557         RETURN(rc);
2558 }
2559
2560 #define FIEMAP_BUFFER_SIZE 4096
2561
2562 /**
2563  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2564  * call. The local end offset and the device are sent in the first
2565  * fm_extent. This function calculates the stripe number from the index.
2566  * This function returns a stripe_no on which mapping is to be restarted.
2567  *
2568  * This function returns fm_end_offset which is the in-OST offset at which
2569  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2570  * will re-calculate proper offset in next stripe.
2571  * Note that the first extent is passed to lov_get_info via the value field.
2572  *
2573  * \param fiemap fiemap request header
2574  * \param lsm striping information for the file
2575  * \param fm_start logical start of mapping
2576  * \param fm_end logical end of mapping
2577  * \param start_stripe starting stripe will be returned in this
2578  */
2579 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2580                                    struct lov_stripe_md *lsm, obd_size fm_start,
2581                                    obd_size fm_end, int *start_stripe)
2582 {
2583         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2584         obd_off lun_start, lun_end;
2585         obd_size fm_end_offset;
2586         int stripe_no = -1, i;
2587
2588         if (fiemap->fm_extent_count == 0 ||
2589             fiemap->fm_extents[0].fe_logical == 0)
2590                 return 0;
2591
2592         /* Find out stripe_no from ost_index saved in the fe_device */
2593         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2594                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2595                                         fiemap->fm_extents[0].fe_device) {
2596                         stripe_no = i;
2597                         break;
2598                 }
2599         }
2600
2601         /* If we have finished mapping on previous device, shift logical
2602          * offset to start of next device */
2603         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2604                                    &lun_start, &lun_end)) != 0 &&
2605                                    local_end < lun_end) {
2606                 fm_end_offset = local_end;
2607                 *start_stripe = stripe_no;
2608         } else {
2609                 /* This is a special value to indicate that caller should
2610                  * calculate offset in next stripe. */
2611                 fm_end_offset = 0;
2612                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2613         }
2614
2615         return fm_end_offset;
2616 }
2617
2618 /**
2619  * We calculate on which OST the mapping will end. If the length of mapping
2620  * is greater than (stripe_size * stripe_count) then the last_stripe will
2621  * will be one just before start_stripe. Else we check if the mapping
2622  * intersects each OST and find last_stripe.
2623  * This function returns the last_stripe and also sets the stripe_count
2624  * over which the mapping is spread
2625  *
2626  * \param lsm striping information for the file
2627  * \param fm_start logical start of mapping
2628  * \param fm_end logical end of mapping
2629  * \param start_stripe starting stripe of the mapping
2630  * \param stripe_count the number of stripes across which to map is returned
2631  *
2632  * \retval last_stripe return the last stripe of the mapping
2633  */
2634 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2635                             obd_size fm_end, int start_stripe,
2636                             int *stripe_count)
2637 {
2638         int last_stripe;
2639         obd_off obd_start, obd_end;
2640         int i, j;
2641
2642         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2643                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2644                                                               start_stripe - 1);
2645                 *stripe_count = lsm->lsm_stripe_count;
2646         } else {
2647                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2648                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2649                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2650                                                    &obd_start, &obd_end)) == 0)
2651                                 break;
2652                 }
2653                 *stripe_count = j;
2654                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2655         }
2656
2657         return last_stripe;
2658 }
2659
2660 /**
2661  * Set fe_device and copy extents from local buffer into main return buffer.
2662  *
2663  * \param fiemap fiemap request header
2664  * \param lcl_fm_ext array of local fiemap extents to be copied
2665  * \param ost_index OST index to be written into the fm_device field for each
2666                     extent
2667  * \param ext_count number of extents to be copied
2668  * \param current_extent where to start copying in main extent array
2669  */
2670 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2671                                   struct ll_fiemap_extent *lcl_fm_ext,
2672                                   int ost_index, unsigned int ext_count,
2673                                   int current_extent)
2674 {
2675         char *to;
2676         int ext;
2677
2678         for (ext = 0; ext < ext_count; ext++) {
2679                 lcl_fm_ext[ext].fe_device = ost_index;
2680                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2681         }
2682
2683         /* Copy fm_extent's from fm_local to return buffer */
2684         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2685         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2686 }
2687
2688 /**
2689  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2690  * This also handles the restarting of FIEMAP calls in case mapping overflows
2691  * the available number of extents in single call.
2692  */
2693 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2694                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2695 {
2696         struct ll_fiemap_info_key *fm_key = key;
2697         struct ll_user_fiemap *fiemap = val;
2698         struct ll_user_fiemap *fm_local = NULL;
2699         struct ll_fiemap_extent *lcl_fm_ext;
2700         int count_local;
2701         unsigned int get_num_extents = 0;
2702         int ost_index = 0, actual_start_stripe, start_stripe;
2703         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2704         obd_size curr_loc;
2705         int current_extent = 0, rc = 0, i;
2706         int ost_eof = 0; /* EOF for object */
2707         int ost_done = 0; /* done with required mapping for this OST? */
2708         int last_stripe;
2709         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2710         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2711
2712         if (lsm == NULL)
2713                 GOTO(out, rc = 0);
2714
2715         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2716                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2717
2718         OBD_ALLOC(fm_local, buffer_size);
2719         if (fm_local == NULL)
2720                 GOTO(out, rc = -ENOMEM);
2721         lcl_fm_ext = &fm_local->fm_extents[0];
2722
2723         count_local = fiemap_size_to_count(buffer_size);
2724
2725         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2726         fm_start = fiemap->fm_start;
2727         fm_length = fiemap->fm_length;
2728         /* Calculate start stripe, last stripe and length of mapping */
2729         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2730         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2731                                                 fm_start + fm_length - 1);
2732         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2733         if (fm_end > fm_key->oa.o_size)
2734                 fm_end = fm_key->oa.o_size;
2735
2736         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2737                                             actual_start_stripe, &stripe_count);
2738
2739         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2740                                                   &start_stripe);
2741
2742         if (fiemap->fm_extent_count == 0) {
2743                 get_num_extents = 1;
2744                 count_local = 0;
2745         }
2746
2747         /* Check each stripe */
2748         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2749              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2750                 obd_size req_fm_len; /* Stores length of required mapping */
2751                 obd_size len_mapped_single_call;
2752                 obd_off lun_start, lun_end, obd_object_end;
2753                 unsigned int ext_count;
2754
2755                 cur_stripe_wrap = cur_stripe;
2756
2757                 /* Find out range of mapping on this stripe */
2758                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2759                                            &lun_start, &obd_object_end)) == 0)
2760                         continue;
2761
2762                 /* If this is a continuation FIEMAP call and we are on
2763                  * starting stripe then lun_start needs to be set to
2764                  * fm_end_offset */
2765                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2766                         lun_start = fm_end_offset;
2767
2768                 if (fm_length != ~0ULL) {
2769                         /* Handle fm_start + fm_length overflow */
2770                         if (fm_start + fm_length < fm_start)
2771                                 fm_length = ~0ULL - fm_start;
2772                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2773                                                      cur_stripe);
2774                 } else {
2775                         lun_end = ~0ULL;
2776                 }
2777
2778                 if (lun_start == lun_end)
2779                         continue;
2780
2781                 req_fm_len = obd_object_end - lun_start;
2782                 fm_local->fm_length = 0;
2783                 len_mapped_single_call = 0;
2784
2785                 /* If the output buffer is very large and the objects have many
2786                  * extents we may need to loop on a single OST repeatedly */
2787                 ost_eof = 0;
2788                 ost_done = 0;
2789                 do {
2790                         if (get_num_extents == 0) {
2791                                 /* Don't get too many extents. */
2792                                 if (current_extent + count_local >
2793                                     fiemap->fm_extent_count)
2794                                         count_local = fiemap->fm_extent_count -
2795                                                                  current_extent;
2796                         }
2797
2798                         lun_start += len_mapped_single_call;
2799                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2800                         req_fm_len = fm_local->fm_length;
2801                         fm_local->fm_extent_count = count_local;
2802                         fm_local->fm_mapped_extents = 0;
2803                         fm_local->fm_flags = fiemap->fm_flags;
2804
2805                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2806                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2807
2808                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2809                                 GOTO(out, rc = -EINVAL);
2810
2811                         /* If OST is inactive, return extent with UNKNOWN flag */
2812                         if (lov && !lov->lov_tgts[ost_index]->ltd_active) {
2813                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2814                                 fm_local->fm_mapped_extents = 1;
2815
2816                                 lcl_fm_ext[0].fe_logical = lun_start;
2817                                 lcl_fm_ext[0].fe_length = obd_object_end -
2818                                                                       lun_start;
2819                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2820
2821                                 goto inactive_tgt;
2822                         }
2823
2824                         fm_local->fm_start = lun_start;
2825                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2826                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2827                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2828                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2829                                           keylen, key, vallen, fm_local, lsm);
2830                         if (rc != 0)
2831                                 GOTO(out, rc);
2832
2833 inactive_tgt:
2834                         ext_count = fm_local->fm_mapped_extents;
2835                         if (ext_count == 0) {
2836                                 ost_done = 1;
2837                                 /* If last stripe has hole at the end,
2838                                  * then we need to return */
2839                                 if (cur_stripe_wrap == last_stripe) {
2840                                         fiemap->fm_mapped_extents = 0;
2841                                         goto finish;
2842                                 }
2843                                 break;
2844                         }
2845
2846                         /* If we just need num of extents then go to next device */
2847                         if (get_num_extents) {
2848                                 current_extent += ext_count;
2849                                 break;
2850                         }
2851
2852                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2853                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2854
2855                         /* Have we finished mapping on this device? */
2856                         if (req_fm_len <= len_mapped_single_call)
2857                                 ost_done = 1;
2858
2859                         /* Clear the EXTENT_LAST flag which can be present on
2860                          * last extent */
2861                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2862                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2863                                                             ~FIEMAP_EXTENT_LAST;
2864
2865                         curr_loc = lov_stripe_size(lsm,
2866                                            lcl_fm_ext[ext_count - 1].fe_logical+
2867                                            lcl_fm_ext[ext_count - 1].fe_length,
2868                                            cur_stripe);
2869                         if (curr_loc >= fm_key->oa.o_size)
2870                                 ost_eof = 1;
2871
2872                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2873                                                      ost_index, ext_count,
2874                                                      current_extent);
2875
2876                         current_extent += ext_count;
2877
2878                         /* Ran out of available extents? */
2879                         if (current_extent >= fiemap->fm_extent_count)
2880                                 goto finish;
2881                 } while (ost_done == 0 && ost_eof == 0);
2882
2883                 if (cur_stripe_wrap == last_stripe)
2884                         goto finish;
2885         }
2886
2887 finish:
2888         /* Indicate that we are returning device offsets unless file just has
2889          * single stripe */
2890         if (lsm->lsm_stripe_count > 1)
2891                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2892
2893         if (get_num_extents)
2894                 goto skip_last_device_calc;
2895
2896         /* Check if we have reached the last stripe and whether mapping for that
2897          * stripe is done. */
2898         if (cur_stripe_wrap == last_stripe) {
2899                 if (ost_done || ost_eof)
2900                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2901                                                              FIEMAP_EXTENT_LAST;
2902         }
2903
2904 skip_last_device_calc:
2905         fiemap->fm_mapped_extents = current_extent;
2906
2907 out:
2908         OBD_FREE(fm_local, buffer_size);
2909         return rc;
2910 }
2911
2912 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2913                         void *key, __u32 *vallen, void *val,
2914                         struct lov_stripe_md *lsm)
2915 {
2916         struct obd_device *obddev = class_exp2obd(exp);
2917         struct lov_obd *lov = &obddev->u.lov;
2918         int i, rc;
2919         ENTRY;
2920
2921         if (!vallen || !val)
2922                 RETURN(-EFAULT);
2923
2924         lov_getref(obddev);
2925
2926         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2927                 struct {
2928                         char name[16];
2929                         struct ldlm_lock *lock;
2930                 } *data = key;
2931                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2932                 struct lov_oinfo *loi;
2933                 __u32 *stripe = val;
2934
2935                 if (*vallen < sizeof(*stripe))
2936                         GOTO(out, rc = -EFAULT);
2937                 *vallen = sizeof(*stripe);
2938
2939                 /* XXX This is another one of those bits that will need to
2940                  * change if we ever actually support nested LOVs.  It uses
2941                  * the lock's export to find out which stripe it is. */
2942                 /* XXX - it's assumed all the locks for deleted OSTs have
2943                  * been cancelled. Also, the export for deleted OSTs will
2944                  * be NULL and won't match the lock's export. */
2945                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2946                         loi = lsm->lsm_oinfo[i];
2947                         if (!lov->lov_tgts[loi->loi_ost_idx])
2948                                 continue;
2949                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2950                             data->lock->l_conn_export &&
2951                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2952                                 *stripe = i;
2953                                 GOTO(out, rc = 0);
2954                         }
2955                 }
2956                 LDLM_ERROR(data->lock, "lock on inode without such object");
2957                 dump_lsm(D_ERROR, lsm);
2958                 GOTO(out, rc = -ENXIO);
2959         } else if (KEY_IS(KEY_LAST_ID)) {
2960                 struct obd_id_info *info = val;
2961                 __u32 size = sizeof(obd_id);
2962                 struct lov_tgt_desc *tgt;
2963
2964                 LASSERT(*vallen == sizeof(struct obd_id_info));
2965                 tgt = lov->lov_tgts[info->idx];
2966
2967                 if (!tgt || !tgt->ltd_active)
2968                         GOTO(out, rc = -ESRCH);
2969
2970                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2971                 GOTO(out, rc = 0);
2972         } else if (KEY_IS(KEY_LOVDESC)) {
2973                 struct lov_desc *desc_ret = val;
2974                 *desc_ret = lov->desc;
2975
2976                 GOTO(out, rc = 0);
2977         } else if (KEY_IS(KEY_FIEMAP)) {
2978                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2979                 GOTO(out, rc);
2980         }
2981
2982         rc = -EINVAL;
2983
2984 out:
2985         lov_putref(obddev);
2986         RETURN(rc);
2987 }
2988
2989 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2990                               void *key, obd_count vallen, void *val,
2991                               struct ptlrpc_request_set *set)
2992 {
2993         struct obd_device *obddev = class_exp2obd(exp);
2994         struct lov_obd *lov = &obddev->u.lov;
2995         obd_count count;
2996         int i, rc = 0, err;
2997         struct lov_tgt_desc *tgt;
2998         unsigned incr, check_uuid,
2999                  do_inactive, no_set;
3000         unsigned next_id = 0,  mds_con = 0;
3001         ENTRY;
3002
3003         incr = check_uuid = do_inactive = no_set = 0;
3004         if (set == NULL) {
3005                 no_set = 1;
3006                 set = ptlrpc_prep_set();
3007                 if (!set)
3008                         RETURN(-ENOMEM);
3009         }
3010
3011         lov_getref(obddev);
3012         count = lov->desc.ld_tgt_count;
3013
3014         if (KEY_IS(KEY_NEXT_ID)) {
3015                 count = vallen / sizeof(struct obd_id_info);
3016                 vallen = sizeof(obd_id);
3017                 incr = sizeof(struct obd_id_info);
3018                 do_inactive = 1;
3019                 next_id = 1;
3020         } else if (KEY_IS(KEY_CHECKSUM)) {
3021                 do_inactive = 1;
3022         } else if (KEY_IS(KEY_UNLINKED)) {
3023                 check_uuid = val ? 1 : 0;
3024         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3025                 /* use defaults:  do_inactive = incr = 0; */
3026         } else if (KEY_IS(KEY_MDS_CONN)) {
3027                 mds_con = 1;
3028         }
3029
3030         for (i = 0; i < count; i++, val = (char *)val + incr) {
3031                 if (next_id) {
3032                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3033                 } else {
3034                         tgt = lov->lov_tgts[i];
3035                 }
3036                 /* OST was disconnected */
3037                 if (!tgt || !tgt->ltd_exp)
3038                         continue;
3039
3040                 /* OST is inactive and we don't want inactive OSCs */
3041                 if (!tgt->ltd_active && !do_inactive)
3042                         continue;
3043
3044                 if (mds_con) {
3045                         struct mds_group_info *mgi;
3046
3047                         LASSERT(vallen == sizeof(*mgi));
3048                         mgi = (struct mds_group_info *)val;
3049
3050                         /* Only want a specific OSC */
3051                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
3052                                                 &tgt->ltd_uuid))
3053                                 continue;
3054
3055                         err = obd_set_info_async(tgt->ltd_exp,
3056                                          keylen, key, sizeof(int),
3057                                          &mgi->group, set);
3058                 } else if (next_id) {
3059                         err = obd_set_info_async(tgt->ltd_exp,
3060                                          keylen, key, vallen,
3061                                          ((struct obd_id_info*)val)->data, set);
3062                 } else  {
3063                         /* Only want a specific OSC */
3064                         if (check_uuid &&
3065                             !obd_uuid_equals(val, &tgt->ltd_uuid))
3066                                 continue;
3067
3068                         err = obd_set_info_async(tgt->ltd_exp,
3069                                          keylen, key, vallen, val, set);
3070                 }
3071
3072                 if (!rc)
3073                         rc = err;
3074         }
3075
3076         lov_putref(obddev);
3077         if (no_set) {
3078                 err = ptlrpc_set_wait(set);
3079                 if (!rc)
3080                         rc = err;
3081                 ptlrpc_set_destroy(set);
3082         }
3083         RETURN(rc);
3084 }
3085
3086 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3087                        struct lov_stripe_md *lsm)
3088 {
3089         int rc;
3090         ENTRY;
3091
3092         if (!lsm)
3093                 RETURN(0);
3094         LASSERT(md_exp);
3095         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3096         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3097
3098         RETURN(rc);
3099 }
3100
3101 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3102 {
3103         int i, rc = 0;
3104         ENTRY;
3105
3106         for (i = 0; i < lsm->lsm_stripe_count; i++) {
3107                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
3108                 if (loi->loi_ar.ar_rc && !rc)
3109                         rc = loi->loi_ar.ar_rc;
3110                 loi->loi_ar.ar_rc = 0;
3111         }
3112         RETURN(rc);
3113 }
3114 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3115
3116
3117 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3118                            int cmd, __u64 *offset)
3119 {
3120         __u32 ssize = lsm->lsm_stripe_size;
3121         __u64 start;
3122
3123         start = *offset;
3124         do_div(start, ssize);
3125         start = start * ssize;
3126
3127         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3128                            ", end "LPU64"\n", *offset, ssize, start,
3129                            start + ssize - 1);
3130         if (cmd == OBD_CALC_STRIPE_END) {
3131                 *offset = start + ssize - 1;
3132         } else if (cmd == OBD_CALC_STRIPE_START) {
3133                 *offset = start;
3134         } else {
3135                 LBUG();
3136         }
3137
3138         RETURN(0);
3139 }
3140
3141
3142 #if 0
3143 struct lov_multi_wait {
3144         struct ldlm_lock *lock;
3145         wait_queue_t      wait;
3146         int               completed;
3147         int               generation;
3148 };
3149
3150 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3151                       struct lustre_handle *lockh)
3152 {
3153         struct lov_lock_handles *lov_lockh = NULL;
3154         struct lustre_handle *lov_lockhp;
3155         struct lov_obd *lov;
3156         struct lov_oinfo *loi;
3157         struct lov_multi_wait *queues;
3158         int rc = 0, i;
3159         ENTRY;
3160
3161         ASSERT_LSM_MAGIC(lsm);
3162
3163         if (!exp || !exp->exp_obd)
3164                 RETURN(-ENODEV);
3165
3166         LASSERT(lockh != NULL);
3167         if (lsm->lsm_stripe_count > 1) {
3168                 lov_lockh = lov_handle2llh(lockh);
3169                 if (lov_lockh == NULL) {
3170                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3171                         RETURN(-EINVAL);
3172                 }
3173
3174                 lov_lockhp = lov_lockh->llh_handles;
3175         } else {
3176                 lov_lockhp = lockh;
3177         }
3178
3179         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3180         if (queues == NULL)
3181                 GOTO(out, rc = -ENOMEM);
3182
3183         lov = &exp->exp_obd->u.lov;
3184         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3185              i++, loi++, lov_lockhp++) {
3186                 struct ldlm_lock *lock;
3187                 struct obd_device *obd;
3188
3189                 lock = ldlm_handle2lock(lov_lockhp);
3190                 if (lock == NULL) {
3191                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3192                                loi->loi_ost_idx, loi->loi_id);
3193                         queues[i].completed = 1;
3194                         continue;
3195                 }
3196
3197                 queues[i].lock = lock;
3198                 init_waitqueue_entry(&(queues[i].wait), current);
3199                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3200
3201                 obd = class_exp2obd(lock->l_conn_export);
3202                 if (obd != NULL)
3203                         imp = obd->u.cli.cl_import;
3204                 if (imp != NULL) {
3205                         spin_lock(&imp->imp_lock);
3206                         queues[i].generation = imp->imp_generation;
3207                         spin_unlock(&imp->imp_lock);
3208                 }
3209         }
3210
3211         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3212                                interrupted_completion_wait, &lwd);
3213         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3214
3215         for (i = 0; i < lsm->lsm_stripe_count; i++)
3216                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3217
3218         if (rc == -EINTR || rc == -ETIMEDOUT) {
3219
3220
3221         }
3222
3223  out:
3224         if (lov_lockh != NULL)
3225                 lov_llh_put(lov_lockh);
3226         RETURN(rc);
3227 }
3228 #endif
3229
3230 void lov_stripe_lock(struct lov_stripe_md *md)
3231 {
3232         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
3233         spin_lock(&md->lsm_lock);
3234         LASSERT(md->lsm_lock_owner == 0);
3235         md->lsm_lock_owner = cfs_curproc_pid();
3236 }
3237 EXPORT_SYMBOL(lov_stripe_lock);
3238
3239 void lov_stripe_unlock(struct lov_stripe_md *md)
3240 {
3241         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
3242         md->lsm_lock_owner = 0;
3243         spin_unlock(&md->lsm_lock);
3244 }
3245 EXPORT_SYMBOL(lov_stripe_unlock);
3246
3247 /**
3248  * Checks if requested extent lock is compatible with a lock under the page.
3249  *
3250  * Checks if the lock under \a page is compatible with a read or write lock
3251  * (specified by \a rw) for an extent [\a start , \a end].
3252  *
3253  * \param exp lov export
3254  * \param lsm striping information for the file
3255  * \param res lov_async_page placeholder
3256  * \param rw OBD_BRW_READ if requested for reading,
3257  *           OBD_BRW_WRITE if requested for writing
3258  * \param start start of the requested extent
3259  * \param end end of the requested extent
3260  * \param cookie transparent parameter for passing locking context
3261  *
3262  * \post result == 1, *cookie == context, appropriate lock is referenced or
3263  * \post result == 0
3264  *
3265  * \retval 1 owned lock is reused for the request
3266  * \retval 0 no lock reused for the request
3267  *
3268  * \see lov_release_short_lock
3269  */
3270 static int lov_reget_short_lock(struct obd_export *exp,
3271                                 struct lov_stripe_md *lsm,
3272                                 void **res, int rw,
3273                                 obd_off start, obd_off end,
3274                                 void **cookie)
3275 {
3276         struct lov_async_page *l = *res;
3277         obd_off stripe_start, stripe_end = start;
3278
3279         ENTRY;
3280
3281         /* ensure we don't cross stripe boundaries */
3282         lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3283         if (stripe_end <= end)
3284                 RETURN(0);
3285
3286         /* map the region limits to the object limits */
3287         lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3288         lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3289
3290         RETURN(obd_reget_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3291                                     lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3292                                     ltd_exp, NULL, &l->lap_sub_cookie,
3293                                     rw, stripe_start, stripe_end, cookie));
3294 }
3295
3296 /**
3297  * Releases a reference to a lock taken in a "fast" way.
3298  *
3299  * Releases a read or a write (specified by \a rw) lock
3300  * referenced by \a cookie.
3301  *
3302  * \param exp lov export
3303  * \param lsm striping information for the file
3304  * \param end end of the locked extent
3305  * \param rw OBD_BRW_READ if requested for reading,
3306  *           OBD_BRW_WRITE if requested for writing
3307  * \param cookie transparent parameter for passing locking context
3308  *
3309  * \post appropriate lock is dereferenced
3310  *
3311  * \see lov_reget_short_lock
3312  */
3313 static int lov_release_short_lock(struct obd_export *exp,
3314                                   struct lov_stripe_md *lsm, obd_off end,
3315                                   void *cookie, int rw)
3316 {
3317         int stripe;
3318
3319         ENTRY;
3320
3321         stripe = lov_stripe_number(lsm, end);
3322
3323         RETURN(obd_release_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3324                                       lsm_oinfo[stripe]->loi_ost_idx]->
3325                                       ltd_exp, NULL, end, cookie, rw));
3326 }
3327
3328 struct obd_ops lov_obd_ops = {
3329         .o_owner               = THIS_MODULE,
3330         .o_setup               = lov_setup,
3331         .o_precleanup          = lov_precleanup,
3332         .o_cleanup             = lov_cleanup,
3333         .o_process_config      = lov_process_config,
3334         .o_connect             = lov_connect,
3335         .o_disconnect          = lov_disconnect,
3336         .o_statfs              = lov_statfs,
3337         .o_statfs_async        = lov_statfs_async,
3338         .o_packmd              = lov_packmd,
3339         .o_unpackmd            = lov_unpackmd,
3340         .o_checkmd             = lov_checkmd,
3341         .o_create              = lov_create,
3342         .o_destroy             = lov_destroy,
3343         .o_getattr             = lov_getattr,
3344         .o_getattr_async       = lov_getattr_async,
3345         .o_setattr             = lov_setattr,
3346         .o_setattr_async       = lov_setattr_async,
3347         .o_brw                 = lov_brw,
3348         .o_brw_async           = lov_brw_async,
3349         .o_prep_async_page     = lov_prep_async_page,
3350         .o_reget_short_lock    = lov_reget_short_lock,
3351         .o_release_short_lock  = lov_release_short_lock,
3352         .o_queue_async_io      = lov_queue_async_io,
3353         .o_set_async_flags     = lov_set_async_flags,
3354         .o_queue_group_io      = lov_queue_group_io,
3355         .o_trigger_group_io    = lov_trigger_group_io,
3356         .o_teardown_async_page = lov_teardown_async_page,
3357         .o_merge_lvb           = lov_merge_lvb,
3358         .o_adjust_kms          = lov_adjust_kms,
3359         .o_punch               = lov_punch,
3360         .o_sync                = lov_sync,
3361         .o_enqueue             = lov_enqueue,
3362         .o_match               = lov_match,
3363         .o_change_cbdata       = lov_change_cbdata,
3364         .o_cancel              = lov_cancel,
3365         .o_cancel_unused       = lov_cancel_unused,
3366         .o_join_lru            = lov_join_lru,
3367         .o_iocontrol           = lov_iocontrol,
3368         .o_get_info            = lov_get_info,
3369         .o_set_info_async      = lov_set_info_async,
3370         .o_extent_calc         = lov_extent_calc,
3371         .o_llog_init           = lov_llog_init,
3372         .o_llog_finish         = lov_llog_finish,
3373         .o_notify              = lov_notify,
3374         .o_register_page_removal_cb = lov_register_page_removal_cb,
3375         .o_unregister_page_removal_cb = lov_unregister_page_removal_cb,
3376         .o_register_lock_cancel_cb = lov_register_lock_cancel_cb,
3377         .o_unregister_lock_cancel_cb = lov_unregister_lock_cancel_cb,
3378         .o_pool_new            = lov_pool_new,
3379         .o_pool_rem            = lov_pool_remove,
3380         .o_pool_add            = lov_pool_add,
3381         .o_pool_del            = lov_pool_del,
3382 };
3383
3384 static quota_interface_t *quota_interface;
3385 extern quota_interface_t lov_quota_interface;
3386
3387 cfs_mem_cache_t *lov_oinfo_slab;
3388
3389 int __init lov_init(void)
3390 {
3391         struct lprocfs_static_vars lvars = { 0 };
3392         int rc, rc2;
3393         ENTRY;
3394
3395         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3396                                               sizeof(struct lov_oinfo),
3397                                               0, SLAB_HWCACHE_ALIGN);
3398         if (lov_oinfo_slab == NULL)
3399                 return -ENOMEM;
3400         lprocfs_lov_init_vars(&lvars);
3401
3402         request_module("lquota");
3403         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3404         init_obd_quota_ops(quota_interface, &lov_obd_ops);
3405
3406         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3407                                  LUSTRE_LOV_NAME, NULL);
3408         if (rc) {
3409                 if (quota_interface)
3410                         PORTAL_SYMBOL_PUT(lov_quota_interface);
3411                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3412                 LASSERT(rc2 == 0);
3413         }
3414
3415         RETURN(rc);
3416 }
3417
3418 #ifdef __KERNEL__
3419 static void /*__exit*/ lov_exit(void)
3420 {
3421         int rc;
3422
3423         if (quota_interface)
3424                 PORTAL_SYMBOL_PUT(lov_quota_interface);
3425
3426         class_unregister_type(LUSTRE_LOV_NAME);
3427         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3428         LASSERT(rc == 0);
3429 }
3430
3431 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3432 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3433 MODULE_LICENSE("GPL");
3434
3435 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
3436 #endif