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
1133                 if (ost_uuid)
1134                         break;
1135         }
1136         lov_putref(export->exp_obd);
1137
1138         OBDO_FREE(tmp_oa);
1139         RETURN(rc);
1140 }
1141
1142 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1143                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1144 {
1145         struct lov_stripe_md *obj_mdp, *lsm;
1146         struct lov_obd *lov = &exp->exp_obd->u.lov;
1147         unsigned ost_idx;
1148         int rc, i;
1149         ENTRY;
1150
1151         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1152                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1153
1154         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1155         if (obj_mdp == NULL)
1156                 RETURN(-ENOMEM);
1157
1158         ost_idx = src_oa->o_nlink;
1159         lsm = *ea;
1160         if (lsm == NULL)
1161                 GOTO(out, rc = -EINVAL);
1162         if (ost_idx >= lov->desc.ld_tgt_count ||
1163             !lov->lov_tgts[ost_idx])
1164                 GOTO(out, rc = -EINVAL);
1165
1166         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1167                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1168                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1169                                 GOTO(out, rc = -EINVAL);
1170                         break;
1171                 }
1172         }
1173         if (i == lsm->lsm_stripe_count)
1174                 GOTO(out, rc = -EINVAL);
1175
1176         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1177 out:
1178         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1179         RETURN(rc);
1180 }
1181
1182 /* the LOV expects oa->o_id to be set to the LOV object id */
1183 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1184                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1185 {
1186         struct lov_obd *lov;
1187         struct obd_info oinfo;
1188         struct lov_request_set *set = NULL;
1189         struct lov_request *req;
1190         struct obd_statfs osfs;
1191         __u64 maxage;
1192         int rc = 0;
1193         ENTRY;
1194
1195         LASSERT(ea != NULL);
1196         if (exp == NULL)
1197                 RETURN(-EINVAL);
1198
1199         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1200             src_oa->o_flags == OBD_FL_DELORPHAN) {
1201                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1202                 RETURN(rc);
1203         }
1204
1205         lov = &exp->exp_obd->u.lov;
1206         if (!lov->desc.ld_active_tgt_count)
1207                 RETURN(-EIO);
1208
1209         /* Recreate a specific object id at the given OST index */
1210         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1211             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1212                  rc = lov_recreate(exp, src_oa, ea, oti);
1213                  RETURN(rc);
1214         }
1215
1216         maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1217         obd_statfs_rqset(exp->exp_obd, &osfs, maxage, OBD_STATFS_NODELAY);
1218
1219         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1220         if (rc)
1221                 RETURN(rc);
1222
1223         list_for_each_entry(req, &set->set_list, rq_link) {
1224                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1225                 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1226                                 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1227                 lov_update_create_set(set, req, rc);
1228         }
1229         rc = lov_fini_create_set(set, ea);
1230         RETURN(rc);
1231 }
1232
1233 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1234 do {                                                                            \
1235         LASSERT((lsmp) != NULL);                                                \
1236         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1237                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1238                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1239                  (lsmp), (lsmp)->lsm_magic);                                    \
1240 } while (0)
1241
1242 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1243                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1244                        struct obd_export *md_exp)
1245 {
1246         struct lov_request_set *set;
1247         struct obd_info oinfo;
1248         struct lov_request *req;
1249         struct list_head *pos;
1250         struct lov_obd *lov;
1251         int rc = 0, err;
1252         ENTRY;
1253
1254         ASSERT_LSM_MAGIC(lsm);
1255
1256         if (!exp || !exp->exp_obd)
1257                 RETURN(-ENODEV);
1258
1259         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1260                 LASSERT(oti);
1261                 LASSERT(oti->oti_logcookies);
1262         }
1263
1264         lov = &exp->exp_obd->u.lov;
1265         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1266         if (rc)
1267                 RETURN(rc);
1268
1269         list_for_each (pos, &set->set_list) {
1270                 int err;
1271                 req = list_entry(pos, struct lov_request, rq_link);
1272
1273                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1274                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1275
1276                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1277                                   req->rq_oi.oi_oa, NULL, oti, NULL);
1278                 err = lov_update_common_set(set, req, err);
1279                 if (err) {
1280                         CERROR("error: destroying objid "LPX64" subobj "
1281                                LPX64" on OST idx %d: rc = %d\n",
1282                                oa->o_id, req->rq_oi.oi_oa->o_id,
1283                                req->rq_idx, err);
1284                         if (!rc)
1285                                 rc = err;
1286                 }
1287         }
1288
1289         if (rc == 0) {
1290                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1291                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1292         }
1293         err = lov_fini_destroy_set(set);
1294         RETURN(rc ? rc : err);
1295 }
1296
1297 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1298 {
1299         struct lov_request_set *set;
1300         struct lov_request *req;
1301         struct list_head *pos;
1302         struct lov_obd *lov;
1303         int err = 0, rc = 0;
1304         ENTRY;
1305
1306         LASSERT(oinfo);
1307         ASSERT_LSM_MAGIC(oinfo->oi_md);
1308
1309         if (!exp || !exp->exp_obd)
1310                 RETURN(-ENODEV);
1311
1312         lov = &exp->exp_obd->u.lov;
1313
1314         rc = lov_prep_getattr_set(exp, oinfo, &set);
1315         if (rc)
1316                 RETURN(rc);
1317
1318         list_for_each (pos, &set->set_list) {
1319                 req = list_entry(pos, struct lov_request, rq_link);
1320
1321                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1322                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1323                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1324
1325                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1326                                  &req->rq_oi);
1327                 err = lov_update_common_set(set, req, rc);
1328                 if (err) {
1329                         CERROR("error: getattr objid "LPX64" subobj "
1330                                LPX64" on OST idx %d: rc = %d\n",
1331                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1332                                req->rq_idx, err);
1333                         break;
1334                 }
1335         }
1336
1337         rc = lov_fini_getattr_set(set);
1338         if (err)
1339                 rc = err;
1340         RETURN(rc);
1341 }
1342
1343 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1344                                  void *data, int rc)
1345 {
1346         struct lov_request_set *lovset = (struct lov_request_set *)data;
1347         int err;
1348         ENTRY;
1349
1350         /* don't do attribute merge if this aysnc op failed */
1351         if (rc)
1352                 lovset->set_completes = 0;
1353         err = lov_fini_getattr_set(lovset);
1354         RETURN(rc ? rc : err);
1355 }
1356
1357 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1358                               struct ptlrpc_request_set *rqset)
1359 {
1360         struct lov_request_set *lovset;
1361         struct lov_obd *lov;
1362         struct list_head *pos;
1363         struct lov_request *req;
1364         int rc = 0, err;
1365         ENTRY;
1366
1367         LASSERT(oinfo);
1368         ASSERT_LSM_MAGIC(oinfo->oi_md);
1369
1370         if (!exp || !exp->exp_obd)
1371                 RETURN(-ENODEV);
1372
1373         lov = &exp->exp_obd->u.lov;
1374
1375         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1376         if (rc)
1377                 RETURN(rc);
1378
1379         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1380                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1381                oinfo->oi_md->lsm_stripe_size);
1382
1383         list_for_each (pos, &lovset->set_list) {
1384                 req = list_entry(pos, struct lov_request, rq_link);
1385
1386                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1387                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1388                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1389                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1390                                        &req->rq_oi, rqset);
1391                 if (rc) {
1392                         CERROR("error: getattr objid "LPX64" subobj "
1393                                LPX64" on OST idx %d: rc = %d\n",
1394                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1395                                req->rq_idx, rc);
1396                         GOTO(out, rc);
1397                 }
1398         }
1399
1400         if (!list_empty(&rqset->set_requests)) {
1401                 LASSERT(rc == 0);
1402                 LASSERT (rqset->set_interpret == NULL);
1403                 rqset->set_interpret = lov_getattr_interpret;
1404                 rqset->set_arg = (void *)lovset;
1405                 RETURN(rc);
1406         }
1407 out:
1408         if (rc)
1409                 lovset->set_completes = 0;
1410         err = lov_fini_getattr_set(lovset);
1411         RETURN(rc ? rc : err);
1412 }
1413
1414 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1415                        struct obd_trans_info *oti)
1416 {
1417         struct lov_request_set *set;
1418         struct lov_obd *lov;
1419         struct list_head *pos;
1420         struct lov_request *req;
1421         int err = 0, rc = 0;
1422         ENTRY;
1423
1424         LASSERT(oinfo);
1425         ASSERT_LSM_MAGIC(oinfo->oi_md);
1426
1427         if (!exp || !exp->exp_obd)
1428                 RETURN(-ENODEV);
1429
1430         /* for now, we only expect the following updates here */
1431         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1432                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1433                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1434                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1435                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1436                                             OBD_MD_FLGID | OBD_MD_FLFID |
1437                                             OBD_MD_FLGENER)));
1438         lov = &exp->exp_obd->u.lov;
1439         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1440         if (rc)
1441                 RETURN(rc);
1442
1443         list_for_each (pos, &set->set_list) {
1444                 req = list_entry(pos, struct lov_request, rq_link);
1445
1446                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1447                                  &req->rq_oi, NULL);
1448                 err = lov_update_setattr_set(set, req, rc);
1449                 if (err) {
1450                         CERROR("error: setattr objid "LPX64" subobj "
1451                                LPX64" on OST idx %d: rc = %d\n",
1452                                set->set_oi->oi_oa->o_id,
1453                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1454                         if (!rc)
1455                                 rc = err;
1456                 }
1457         }
1458         err = lov_fini_setattr_set(set);
1459         if (!rc)
1460                 rc = err;
1461         RETURN(rc);
1462 }
1463
1464 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1465                                  void *data, int rc)
1466 {
1467         struct lov_request_set *lovset = (struct lov_request_set *)data;
1468         int err;
1469         ENTRY;
1470
1471         if (rc)
1472                 lovset->set_completes = 0;
1473         err = lov_fini_setattr_set(lovset);
1474         RETURN(rc ? rc : err);
1475 }
1476
1477 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1478    needed. Otherwise, a client is waiting for responses. */
1479 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1480                              struct obd_trans_info *oti,
1481                              struct ptlrpc_request_set *rqset)
1482 {
1483         struct lov_request_set *set;
1484         struct lov_request *req;
1485         struct list_head *pos;
1486         struct lov_obd *lov;
1487         int rc = 0;
1488         ENTRY;
1489
1490         LASSERT(oinfo);
1491         ASSERT_LSM_MAGIC(oinfo->oi_md);
1492         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1493                 LASSERT(oti);
1494                 LASSERT(oti->oti_logcookies);
1495         }
1496
1497         if (!exp || !exp->exp_obd)
1498                 RETURN(-ENODEV);
1499
1500         lov = &exp->exp_obd->u.lov;
1501         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1502         if (rc)
1503                 RETURN(rc);
1504
1505         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1506                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1507                oinfo->oi_md->lsm_stripe_size);
1508
1509         list_for_each (pos, &set->set_list) {
1510                 req = list_entry(pos, struct lov_request, rq_link);
1511
1512                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1513                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1514
1515                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1516                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1517                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1518
1519                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1520                                        &req->rq_oi, oti, rqset);
1521                 if (rc) {
1522                         CERROR("error: setattr objid "LPX64" subobj "
1523                                LPX64" on OST idx %d: rc = %d\n",
1524                                set->set_oi->oi_oa->o_id,
1525                                req->rq_oi.oi_oa->o_id,
1526                                req->rq_idx, rc);
1527                         break;
1528                 }
1529         }
1530
1531         /* If we are not waiting for responses on async requests, return. */
1532         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1533                 int err;
1534                 if (rc)
1535                         set->set_completes = 0;
1536                 err = lov_fini_setattr_set(set);
1537                 RETURN(rc ? rc : err);
1538         }
1539
1540         LASSERT(rqset->set_interpret == NULL);
1541         rqset->set_interpret = lov_setattr_interpret;
1542         rqset->set_arg = (void *)set;
1543
1544         RETURN(0);
1545 }
1546
1547 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1548                                void *data, int rc)
1549 {
1550         struct lov_request_set *lovset = (struct lov_request_set *)data;
1551         int err;
1552         ENTRY;
1553
1554         if (rc)
1555                 lovset->set_completes = 0;
1556         err = lov_fini_punch_set(lovset);
1557         RETURN(rc ? rc : err);
1558 }
1559
1560 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1561  * we can send this 'punch' to just the authoritative node and the nodes
1562  * that the punch will affect. */
1563 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1564                      struct obd_trans_info *oti,
1565                      struct ptlrpc_request_set *rqset)
1566 {
1567         struct lov_request_set *set;
1568         struct lov_obd *lov;
1569         struct list_head *pos;
1570         struct lov_request *req;
1571         int rc = 0;
1572         ENTRY;
1573
1574         LASSERT(oinfo);
1575         ASSERT_LSM_MAGIC(oinfo->oi_md);
1576
1577         if (!exp || !exp->exp_obd)
1578                 RETURN(-ENODEV);
1579
1580         lov = &exp->exp_obd->u.lov;
1581         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1582         if (rc)
1583                 RETURN(rc);
1584
1585         list_for_each (pos, &set->set_list) {
1586                 req = list_entry(pos, struct lov_request, rq_link);
1587
1588                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1589                                &req->rq_oi, NULL, rqset);
1590                 if (rc) {
1591                         CERROR("error: punch objid "LPX64" subobj "LPX64
1592                                " on OST idx %d: rc = %d\n",
1593                                set->set_oi->oi_oa->o_id,
1594                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1595                         break;
1596                 }
1597         }
1598
1599         if (rc || list_empty(&rqset->set_requests)) {
1600                 int err;
1601                 err = lov_fini_punch_set(set);
1602                 RETURN(rc ? rc : err);
1603         }
1604
1605         LASSERT(rqset->set_interpret == NULL);
1606         rqset->set_interpret = lov_punch_interpret;
1607         rqset->set_arg = (void *)set;
1608
1609         RETURN(0);
1610 }
1611
1612 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1613                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1614                     void *capa)
1615 {
1616         struct lov_request_set *set;
1617         struct obd_info oinfo;
1618         struct lov_obd *lov;
1619         struct list_head *pos;
1620         struct lov_request *req;
1621         int err = 0, rc = 0;
1622         ENTRY;
1623
1624         ASSERT_LSM_MAGIC(lsm);
1625
1626         if (!exp->exp_obd)
1627                 RETURN(-ENODEV);
1628
1629         lov = &exp->exp_obd->u.lov;
1630         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1631         if (rc)
1632                 RETURN(rc);
1633
1634         list_for_each (pos, &set->set_list) {
1635                 req = list_entry(pos, struct lov_request, rq_link);
1636
1637                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1638                               req->rq_oi.oi_oa, NULL,
1639                               req->rq_oi.oi_policy.l_extent.start,
1640                               req->rq_oi.oi_policy.l_extent.end, capa);
1641                 err = lov_update_common_set(set, req, rc);
1642                 if (err) {
1643                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1644                                " on OST idx %d: rc = %d\n",
1645                                set->set_oi->oi_oa->o_id,
1646                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1647                         if (!rc)
1648                                 rc = err;
1649                 }
1650         }
1651         err = lov_fini_sync_set(set);
1652         if (!rc)
1653                 rc = err;
1654         RETURN(rc);
1655 }
1656
1657 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1658                          obd_count oa_bufs, struct brw_page *pga)
1659 {
1660         struct obd_info oinfo = { { { 0 } } };
1661         int i, rc = 0;
1662
1663         oinfo.oi_oa = lov_oinfo->oi_oa;
1664
1665         /* The caller just wants to know if there's a chance that this
1666          * I/O can succeed */
1667         for (i = 0; i < oa_bufs; i++) {
1668                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1669                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1670                 obd_off start, end;
1671
1672                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1673                                            pga[i].off + pga[i].count,
1674                                            &start, &end))
1675                         continue;
1676
1677                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1678                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1679                         return -EIO;
1680                 }
1681
1682                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1683                              1, &pga[i], NULL);
1684                 if (rc)
1685                         break;
1686         }
1687         return rc;
1688 }
1689
1690 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1691                    obd_count oa_bufs, struct brw_page *pga,
1692                    struct obd_trans_info *oti)
1693 {
1694         struct lov_request_set *set;
1695         struct lov_request *req;
1696         struct list_head *pos;
1697         struct lov_obd *lov = &exp->exp_obd->u.lov;
1698         int err, rc = 0;
1699         ENTRY;
1700
1701         ASSERT_LSM_MAGIC(oinfo->oi_md);
1702
1703         if (cmd == OBD_BRW_CHECK) {
1704                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1705                 RETURN(rc);
1706         }
1707
1708         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1709         if (rc)
1710                 RETURN(rc);
1711
1712         list_for_each (pos, &set->set_list) {
1713                 struct obd_export *sub_exp;
1714                 struct brw_page *sub_pga;
1715                 req = list_entry(pos, struct lov_request, rq_link);
1716
1717                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1718                 sub_pga = set->set_pga + req->rq_pgaidx;
1719                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1720                              sub_pga, oti);
1721                 if (rc)
1722                         break;
1723                 lov_update_common_set(set, req, rc);
1724         }
1725
1726         err = lov_fini_brw_set(set);
1727         if (!rc)
1728                 rc = err;
1729         RETURN(rc);
1730 }
1731
1732 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1733                              int rc)
1734 {
1735         struct lov_request_set *lovset = (struct lov_request_set *)data;
1736         ENTRY;
1737
1738         if (rc) {
1739                 lovset->set_completes = 0;
1740                 lov_fini_brw_set(lovset);
1741         } else {
1742                 rc = lov_fini_brw_set(lovset);
1743         }
1744
1745         RETURN(rc);
1746 }
1747
1748 static int lov_brw_async(int cmd, struct obd_export *exp,
1749                          struct obd_info *oinfo, obd_count oa_bufs,
1750                          struct brw_page *pga, struct obd_trans_info *oti,
1751                          struct ptlrpc_request_set *set)
1752 {
1753         struct lov_request_set *lovset;
1754         struct lov_request *req;
1755         struct list_head *pos;
1756         struct lov_obd *lov = &exp->exp_obd->u.lov;
1757         int rc = 0;
1758         ENTRY;
1759
1760         LASSERT(oinfo);
1761         ASSERT_LSM_MAGIC(oinfo->oi_md);
1762
1763         if (cmd == OBD_BRW_CHECK) {
1764                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1765                 RETURN(rc);
1766         }
1767
1768         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1769         if (rc)
1770                 RETURN(rc);
1771
1772         list_for_each (pos, &lovset->set_list) {
1773                 struct obd_export *sub_exp;
1774                 struct brw_page *sub_pga;
1775                 req = list_entry(pos, struct lov_request, rq_link);
1776
1777                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1778                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1779                 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1780                                    sub_pga, oti, set);
1781                 if (rc)
1782                         GOTO(out, rc);
1783                 lov_update_common_set(lovset, req, rc);
1784         }
1785         LASSERT(rc == 0);
1786         LASSERT(set->set_interpret == NULL);
1787         LASSERT(set->set_arg == NULL);
1788         rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1789         if (rc)
1790                 GOTO(out, rc);
1791
1792         RETURN(rc);
1793 out:
1794         lov_fini_brw_set(lovset);
1795         RETURN(rc);
1796 }
1797
1798 static int lov_ap_make_ready(void *data, int cmd)
1799 {
1800         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1801
1802         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1803 }
1804
1805 static int lov_ap_refresh_count(void *data, int cmd)
1806 {
1807         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1808
1809         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1810                                                      cmd);
1811 }
1812
1813 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1814 {
1815         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1816
1817         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1818         /* XXX woah, shouldn't we be altering more here?  size? */
1819         oa->o_id = lap->lap_loi_id;
1820         oa->o_gr = lap->lap_loi_gr;
1821         oa->o_valid |= OBD_MD_FLGROUP;
1822         oa->o_stripe_idx = lap->lap_stripe;
1823 }
1824
1825 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1826                                obd_valid valid)
1827 {
1828         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1829
1830         lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1831 }
1832
1833 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1834 {
1835         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1836
1837         /* in a raid1 regime this would down a count of many ios
1838          * in flight, onl calling the caller_ops completion when all
1839          * the raid1 ios are complete */
1840         rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1841         return rc;
1842 }
1843
1844 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1845 {
1846         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1847         return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1848 }
1849
1850 static struct obd_async_page_ops lov_async_page_ops = {
1851         .ap_make_ready =        lov_ap_make_ready,
1852         .ap_refresh_count =     lov_ap_refresh_count,
1853         .ap_fill_obdo =         lov_ap_fill_obdo,
1854         .ap_update_obdo =       lov_ap_update_obdo,
1855         .ap_completion =        lov_ap_completion,
1856         .ap_lookup_capa =       lov_ap_lookup_capa,
1857 };
1858
1859 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1860                            struct lov_oinfo *loi, cfs_page_t *page,
1861                            obd_off offset, struct obd_async_page_ops *ops,
1862                            void *data, void **res, int nocache,
1863                            struct lustre_handle *lockh)
1864 {
1865         struct lov_obd *lov = &exp->exp_obd->u.lov;
1866         struct lov_async_page *lap;
1867         struct lov_lock_handles *lov_lockh = NULL;
1868         int rc = 0;
1869         ENTRY;
1870
1871         if (!page) {
1872                 int i = 0;
1873                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1874                    Only because of this layering limitation will a client
1875                    mount with no osts fail */
1876                 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1877                        !lov->lov_tgts[i]->ltd_exp) {
1878                         i++;
1879                         if (i >= lov->desc.ld_tgt_count)
1880                                 RETURN(-ENOMEDIUM);
1881                 }
1882                 rc = size_round(sizeof(*lap)) +
1883                         obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1884                                             NULL, NULL, 0, NULL, NULL, NULL, 0,
1885                                             NULL);
1886                 RETURN(rc);
1887         }
1888         ASSERT_LSM_MAGIC(lsm);
1889         LASSERT(loi == NULL);
1890
1891         lap = *res;
1892         lap->lap_magic = LOV_AP_MAGIC;
1893         lap->lap_caller_ops = ops;
1894         lap->lap_caller_data = data;
1895
1896         /* for now only raid 0 which passes through */
1897         lap->lap_stripe = lov_stripe_number(lsm, offset);
1898         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1899         loi = lsm->lsm_oinfo[lap->lap_stripe];
1900
1901         /* so the callback doesn't need the lsm */
1902         lap->lap_loi_id = loi->loi_id;
1903         lap->lap_loi_gr = lsm->lsm_object_gr;
1904         LASSERT(lsm->lsm_object_gr > 0);
1905
1906         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1907
1908         if (lockh) {
1909                 lov_lockh = lov_handle2llh(lockh);
1910                 if (lov_lockh) {
1911                         lockh = lov_lockh->llh_handles + lap->lap_stripe;
1912                 }
1913         }
1914
1915         rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1916                                  lsm, loi, page, lap->lap_sub_offset,
1917                                  &lov_async_page_ops, lap,
1918                                  &lap->lap_sub_cookie, nocache, lockh);
1919         if (lov_lockh)
1920                 lov_llh_put(lov_lockh);
1921         if (rc)
1922                 RETURN(rc);
1923         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1924                lap->lap_sub_cookie, offset);
1925         RETURN(0);
1926 }
1927
1928 static int lov_queue_async_io(struct obd_export *exp,
1929                               struct lov_stripe_md *lsm,
1930                               struct lov_oinfo *loi, void *cookie,
1931                               int cmd, obd_off off, int count,
1932                               obd_flag brw_flags, obd_flag async_flags)
1933 {
1934         struct lov_obd *lov = &exp->exp_obd->u.lov;
1935         struct lov_async_page *lap;
1936         int rc;
1937
1938         LASSERT(loi == NULL);
1939
1940         ASSERT_LSM_MAGIC(lsm);
1941
1942         lap = LAP_FROM_COOKIE(cookie);
1943
1944         loi = lsm->lsm_oinfo[lap->lap_stripe];
1945
1946         rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1947                                 loi, lap->lap_sub_cookie, cmd, off, count,
1948                                 brw_flags, async_flags);
1949         RETURN(rc);
1950 }
1951
1952 static int lov_set_async_flags(struct obd_export *exp,
1953                                struct lov_stripe_md *lsm,
1954                                struct lov_oinfo *loi, void *cookie,
1955                                obd_flag async_flags)
1956 {
1957         struct lov_obd *lov = &exp->exp_obd->u.lov;
1958         struct lov_async_page *lap;
1959         int rc;
1960
1961         LASSERT(loi == NULL);
1962
1963         ASSERT_LSM_MAGIC(lsm);
1964
1965         lap = LAP_FROM_COOKIE(cookie);
1966
1967         loi = lsm->lsm_oinfo[lap->lap_stripe];
1968
1969         rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1970                                  lsm, loi, lap->lap_sub_cookie, async_flags);
1971         RETURN(rc);
1972 }
1973
1974 static int lov_queue_group_io(struct obd_export *exp,
1975                               struct lov_stripe_md *lsm,
1976                               struct lov_oinfo *loi,
1977                               struct obd_io_group *oig, void *cookie,
1978                               int cmd, obd_off off, int count,
1979                               obd_flag brw_flags, obd_flag async_flags)
1980 {
1981         struct lov_obd *lov = &exp->exp_obd->u.lov;
1982         struct lov_async_page *lap;
1983         int rc;
1984
1985         LASSERT(loi == NULL);
1986
1987         ASSERT_LSM_MAGIC(lsm);
1988
1989         lap = LAP_FROM_COOKIE(cookie);
1990
1991         loi = lsm->lsm_oinfo[lap->lap_stripe];
1992
1993         rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1994                                 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1995                                 brw_flags, async_flags);
1996         RETURN(rc);
1997 }
1998
1999 /* this isn't exactly optimal.  we may have queued sync io in oscs on
2000  * all stripes, but we don't record that fact at queue time.  so we
2001  * trigger sync io on all stripes. */
2002 static int lov_trigger_group_io(struct obd_export *exp,
2003                                 struct lov_stripe_md *lsm,
2004                                 struct lov_oinfo *loi,
2005                                 struct obd_io_group *oig)
2006 {
2007         struct lov_obd *lov = &exp->exp_obd->u.lov;
2008         int rc = 0, i, err;
2009
2010         LASSERT(loi == NULL);
2011
2012         ASSERT_LSM_MAGIC(lsm);
2013
2014         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2015                 loi = lsm->lsm_oinfo[i];
2016                 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2017                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2018                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2019                         continue;
2020                 }
2021
2022                 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2023                                            lsm, loi, oig);
2024                 if (rc == 0 && err != 0)
2025                         rc = err;
2026         };
2027         RETURN(rc);
2028 }
2029
2030 static int lov_teardown_async_page(struct obd_export *exp,
2031                                    struct lov_stripe_md *lsm,
2032                                    struct lov_oinfo *loi, void *cookie)
2033 {
2034         struct lov_obd *lov = &exp->exp_obd->u.lov;
2035         struct lov_async_page *lap;
2036         int rc;
2037
2038         LASSERT(loi == NULL);
2039
2040         ASSERT_LSM_MAGIC(lsm);
2041
2042         lap = LAP_FROM_COOKIE(cookie);
2043
2044         loi = lsm->lsm_oinfo[lap->lap_stripe];
2045
2046         rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2047                                      lsm, loi, lap->lap_sub_cookie);
2048         if (rc) {
2049                 CERROR("unable to teardown sub cookie %p: %d\n",
2050                        lap->lap_sub_cookie, rc);
2051                 RETURN(rc);
2052         }
2053         RETURN(rc);
2054 }
2055
2056 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2057                                  void *data, int rc)
2058 {
2059         struct lov_request_set *lovset = (struct lov_request_set *)data;
2060         ENTRY;
2061         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2062         RETURN(rc);
2063 }
2064
2065 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2066                        struct ldlm_enqueue_info *einfo,
2067                        struct ptlrpc_request_set *rqset)
2068 {
2069         ldlm_mode_t mode = einfo->ei_mode;
2070         struct lov_request_set *set;
2071         struct lov_request *req;
2072         struct list_head *pos;
2073         struct lov_obd *lov;
2074         ldlm_error_t rc;
2075         ENTRY;
2076
2077         LASSERT(oinfo);
2078         ASSERT_LSM_MAGIC(oinfo->oi_md);
2079         LASSERT(mode == (mode & -mode));
2080
2081         /* we should never be asked to replay a lock this way. */
2082         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2083
2084         if (!exp || !exp->exp_obd)
2085                 RETURN(-ENODEV);
2086
2087         lov = &exp->exp_obd->u.lov;
2088         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2089         if (rc)
2090                 RETURN(rc);
2091
2092         list_for_each (pos, &set->set_list) {
2093                 req = list_entry(pos, struct lov_request, rq_link);
2094
2095                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2096                                  &req->rq_oi, einfo, rqset);
2097                 if (rc != ELDLM_OK)
2098                         GOTO(out, rc);
2099         }
2100
2101         if (rqset && !list_empty(&rqset->set_requests)) {
2102                 LASSERT(rc == 0);
2103                 LASSERT(rqset->set_interpret == NULL);
2104                 rqset->set_interpret = lov_enqueue_interpret;
2105                 rqset->set_arg = (void *)set;
2106                 RETURN(rc);
2107         }
2108 out:
2109         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2110         RETURN(rc);
2111 }
2112
2113 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2114                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2115                      int *flags, void *data, struct lustre_handle *lockh)
2116 {
2117         struct lov_request_set *set;
2118         struct obd_info oinfo;
2119         struct lov_request *req;
2120         struct list_head *pos;
2121         struct lov_obd *lov = &exp->exp_obd->u.lov;
2122         struct lustre_handle *lov_lockhp;
2123         int lov_flags, rc = 0;
2124         ENTRY;
2125
2126         ASSERT_LSM_MAGIC(lsm);
2127         LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2128
2129         if (!exp || !exp->exp_obd)
2130                 RETURN(-ENODEV);
2131
2132         lov = &exp->exp_obd->u.lov;
2133         rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2134         if (rc)
2135                 RETURN(rc);
2136
2137         list_for_each (pos, &set->set_list) {
2138                 ldlm_policy_data_t sub_policy;
2139                 req = list_entry(pos, struct lov_request, rq_link);
2140                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2141                 LASSERT(lov_lockhp);
2142
2143                 lov_flags = *flags;
2144                 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2145
2146                 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2147                                req->rq_oi.oi_md, type, &sub_policy,
2148                                mode, &lov_flags, data, lov_lockhp);
2149                 rc = lov_update_match_set(set, req, rc);
2150                 if (rc <= 0)
2151                         break;
2152         }
2153         lov_fini_match_set(set, mode, *flags);
2154         RETURN(rc);
2155 }
2156
2157 static int lov_change_cbdata(struct obd_export *exp,
2158                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
2159                              void *data)
2160 {
2161         struct lov_obd *lov;
2162         int rc = 0, i;
2163         ENTRY;
2164
2165         ASSERT_LSM_MAGIC(lsm);
2166
2167         if (!exp || !exp->exp_obd)
2168                 RETURN(-ENODEV);
2169
2170         LASSERT(lsm->lsm_object_gr > 0);
2171
2172         lov = &exp->exp_obd->u.lov;
2173         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2174                 struct lov_stripe_md submd;
2175                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2176
2177                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2178                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2179                         continue;
2180                 }
2181
2182                 submd.lsm_object_id = loi->loi_id;
2183                 submd.lsm_object_gr = lsm->lsm_object_gr;
2184                 submd.lsm_stripe_count = 0;
2185                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2186                                        &submd, it, data);
2187         }
2188         RETURN(rc);
2189 }
2190
2191 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2192                       __u32 mode, struct lustre_handle *lockh)
2193 {
2194         struct lov_request_set *set;
2195         struct obd_info oinfo;
2196         struct lov_request *req;
2197         struct list_head *pos;
2198         struct lov_obd *lov = &exp->exp_obd->u.lov;
2199         struct lustre_handle *lov_lockhp;
2200         int err = 0, rc = 0;
2201         ENTRY;
2202
2203         ASSERT_LSM_MAGIC(lsm);
2204
2205         if (!exp || !exp->exp_obd)
2206                 RETURN(-ENODEV);
2207
2208         LASSERT(lsm->lsm_object_gr > 0);
2209         LASSERT(lockh);
2210         lov = &exp->exp_obd->u.lov;
2211         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2212         if (rc)
2213                 RETURN(rc);
2214
2215         list_for_each (pos, &set->set_list) {
2216                 req = list_entry(pos, struct lov_request, rq_link);
2217                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2218
2219                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2220                                 req->rq_oi.oi_md, mode, lov_lockhp);
2221                 rc = lov_update_common_set(set, req, rc);
2222                 if (rc) {
2223                         CERROR("error: cancel objid "LPX64" subobj "
2224                                LPX64" on OST idx %d: rc = %d\n",
2225                                lsm->lsm_object_id,
2226                                req->rq_oi.oi_md->lsm_object_id,
2227                                req->rq_idx, rc);
2228                         err = rc;
2229                 }
2230
2231         }
2232         lov_fini_cancel_set(set);
2233         RETURN(err);
2234 }
2235
2236 static int lov_cancel_unused(struct obd_export *exp,
2237                              struct lov_stripe_md *lsm,
2238                              int flags, void *opaque)
2239 {
2240         struct lov_obd *lov;
2241         int rc = 0, i;
2242         ENTRY;
2243
2244         if (!exp || !exp->exp_obd)
2245                 RETURN(-ENODEV);
2246
2247         lov = &exp->exp_obd->u.lov;
2248         if (lsm == NULL) {
2249                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2250                         int err;
2251                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2252                                 continue;
2253
2254                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2255                                                 flags, opaque);
2256                         if (!rc)
2257                                 rc = err;
2258                 }
2259                 RETURN(rc);
2260         }
2261
2262         ASSERT_LSM_MAGIC(lsm);
2263
2264         LASSERT(lsm->lsm_object_gr > 0);
2265         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2266                 struct lov_stripe_md submd;
2267                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2268                 int err;
2269
2270                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2271                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2272                         continue;
2273                 }
2274
2275                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2276                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2277
2278                 submd.lsm_object_id = loi->loi_id;
2279                 submd.lsm_object_gr = lsm->lsm_object_gr;
2280                 submd.lsm_stripe_count = 0;
2281                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2282                                         &submd, flags, opaque);
2283                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2284                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2285                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2286                                loi->loi_id, loi->loi_ost_idx, err);
2287                         if (!rc)
2288                                 rc = err;
2289                 }
2290         }
2291         RETURN(rc);
2292 }
2293
2294 static int lov_join_lru(struct obd_export *exp,
2295                         struct lov_stripe_md *lsm, int join)
2296 {
2297         struct lov_obd *lov;
2298         int i, count = 0;
2299         ENTRY;
2300
2301         ASSERT_LSM_MAGIC(lsm);
2302         if (!exp || !exp->exp_obd)
2303                 RETURN(-ENODEV);
2304
2305         lov = &exp->exp_obd->u.lov;
2306         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2307                 struct lov_stripe_md submd;
2308                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2309                 int rc = 0;
2310
2311                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2312                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2313                         continue;
2314                 }
2315
2316                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2317                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2318
2319                 submd.lsm_object_id = loi->loi_id;
2320                 submd.lsm_object_gr = lsm->lsm_object_gr;
2321                 submd.lsm_stripe_count = 0;
2322                 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2323                                   &submd, join);
2324                 if (rc < 0) {
2325                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2326                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2327                                loi->loi_id, loi->loi_ost_idx, rc);
2328                         return rc;
2329                 } else {
2330                         count += rc;
2331                 }
2332         }
2333         RETURN(count);
2334 }
2335
2336 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2337                                 void *data, int rc)
2338 {
2339         struct lov_request_set *lovset = (struct lov_request_set *)data;
2340         int err;
2341         ENTRY;
2342
2343         if (rc)
2344                 lovset->set_completes = 0;
2345
2346         err = lov_fini_statfs_set(lovset);
2347         RETURN(rc ? rc : err);
2348 }
2349
2350 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2351                             __u64 max_age, struct ptlrpc_request_set *rqset)
2352 {
2353         struct lov_request_set *set;
2354         struct lov_request *req;
2355         struct list_head *pos;
2356         struct lov_obd *lov;
2357         int rc = 0;
2358         ENTRY;
2359
2360         LASSERT(oinfo != NULL);
2361         LASSERT(oinfo->oi_osfs != NULL);
2362
2363         lov = &obd->u.lov;
2364         rc = lov_prep_statfs_set(obd, oinfo, &set);
2365         if (rc)
2366                 RETURN(rc);
2367
2368         list_for_each (pos, &set->set_list) {
2369                 struct obd_device *osc_obd;
2370
2371                 req = list_entry(pos, struct lov_request, rq_link);
2372
2373                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2374                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2375                 if (rc)
2376                         break;
2377         }
2378
2379         if (rc || list_empty(&rqset->set_requests)) {
2380                 int err;
2381                 if (rc)
2382                         set->set_completes = 0;
2383                 err = lov_fini_statfs_set(set);
2384                 RETURN(rc ? rc : err);
2385         }
2386
2387         LASSERT(rqset->set_interpret == NULL);
2388         rqset->set_interpret = lov_statfs_interpret;
2389         rqset->set_arg = (void *)set;
2390         RETURN(0);
2391 }
2392
2393 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2394                       __u64 max_age, __u32 flags)
2395 {
2396         struct ptlrpc_request_set *set = NULL;
2397         struct obd_info oinfo = { { { 0 } } };
2398         int rc = 0;
2399         ENTRY;
2400
2401
2402         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2403          * statfs requests */
2404         set = ptlrpc_prep_set();
2405         if (set == NULL)
2406                 RETURN(-ENOMEM);
2407
2408         oinfo.oi_osfs = osfs;
2409         oinfo.oi_flags = flags;
2410         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2411         if (rc == 0)
2412                 rc = ptlrpc_set_wait(set);
2413         ptlrpc_set_destroy(set);
2414
2415         RETURN(rc);
2416 }
2417
2418 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2419                          void *karg, void *uarg)
2420 {
2421         struct obd_device *obddev = class_exp2obd(exp);
2422         struct lov_obd *lov = &obddev->u.lov;
2423         int i, rc, count = lov->desc.ld_tgt_count;
2424         struct obd_uuid *uuidp;
2425         ENTRY;
2426
2427         switch (cmd) {
2428         case IOC_OBD_STATFS: {
2429                 struct obd_ioctl_data *data = karg;
2430                 struct obd_device *osc_obd;
2431                 struct obd_statfs stat_buf = {0};
2432                 __u32 index;
2433
2434                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2435                 LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
2436
2437                 if ((index >= count))
2438                         RETURN(-ENODEV);
2439
2440                 if (!lov->lov_tgts[index])
2441                         /* Try again with the next index */
2442                         RETURN(-EAGAIN);
2443                 if (!lov->lov_tgts[index]->ltd_active)
2444                         RETURN(-ENODATA);
2445
2446                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2447                 if (!osc_obd)
2448                         RETURN(-EINVAL);
2449
2450                 /* got statfs data */
2451                 rc = obd_statfs(osc_obd, &stat_buf,
2452                                 cfs_time_current_64() - HZ, 0);
2453                 if (rc)
2454                         RETURN(rc);
2455                 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2456                         RETURN(rc);
2457                 /* copy UUID */
2458                 rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
2459                                   data->ioc_plen2);
2460                 break;
2461         }
2462         case OBD_IOC_LOV_GET_CONFIG: {
2463                 struct obd_ioctl_data *data;
2464                 struct lov_desc *desc;
2465                 char *buf = NULL;
2466                 __u32 *genp;
2467
2468                 len = 0;
2469                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2470                         RETURN(-EINVAL);
2471
2472                 data = (struct obd_ioctl_data *)buf;
2473
2474                 if (sizeof(*desc) > data->ioc_inllen1) {
2475                         obd_ioctl_freedata(buf, len);
2476                         RETURN(-EINVAL);
2477                 }
2478
2479                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2480                         obd_ioctl_freedata(buf, len);
2481                         RETURN(-EINVAL);
2482                 }
2483
2484                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2485                         obd_ioctl_freedata(buf, len);
2486                         RETURN(-EINVAL);
2487                 }
2488
2489                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2490                 memcpy(desc, &(lov->desc), sizeof(*desc));
2491
2492                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2493                 genp = (__u32 *)data->ioc_inlbuf3;
2494                 /* the uuid will be empty for deleted OSTs */
2495                 for (i = 0; i < count; i++, uuidp++, genp++) {
2496                         if (!lov->lov_tgts[i])
2497                                 continue;
2498                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2499                         *genp = lov->lov_tgts[i]->ltd_gen;
2500                 }
2501
2502                 rc = copy_to_user((void *)uarg, buf, len);
2503                 if (rc)
2504                         rc = -EFAULT;
2505                 obd_ioctl_freedata(buf, len);
2506                 break;
2507         }
2508         case LL_IOC_LOV_SETSTRIPE:
2509                 rc = lov_setstripe(exp, karg, uarg);
2510                 break;
2511         case LL_IOC_LOV_GETSTRIPE:
2512                 rc = lov_getstripe(exp, karg, uarg);
2513                 break;
2514         case LL_IOC_LOV_SETEA:
2515                 rc = lov_setea(exp, karg, uarg);
2516                 break;
2517         default: {
2518                 int set = 0;
2519
2520                 if (count == 0)
2521                         RETURN(-ENOTTY);
2522
2523                 rc = 0;
2524                 for (i = 0; i < count; i++) {
2525                         int err;
2526
2527                         /* OST was disconnected */
2528                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2529                                 continue;
2530
2531                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2532                                             len, karg, uarg);
2533                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2534                                 RETURN(err);
2535                         } else if (err) {
2536                                 if (lov->lov_tgts[i]->ltd_active) {
2537                                         CDEBUG(err == -ENOTTY ?
2538                                                D_IOCTL : D_WARNING,
2539                                                "iocontrol OSC %s on OST "
2540                                                "idx %d cmd %x: err = %d\n",
2541                                                lov_uuid2str(lov, i),
2542                                                i, cmd, err);
2543                                         if (!rc)
2544                                                 rc = err;
2545                                 }
2546                         } else {
2547                                 set = 1;
2548                         }
2549                 }
2550                 if (!set && !rc)
2551                         rc = -EIO;
2552         }
2553         }
2554
2555         RETURN(rc);
2556 }
2557
2558 #define FIEMAP_BUFFER_SIZE 4096
2559
2560 /**
2561  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2562  * call. The local end offset and the device are sent in the first
2563  * fm_extent. This function calculates the stripe number from the index.
2564  * This function returns a stripe_no on which mapping is to be restarted.
2565  *
2566  * This function returns fm_end_offset which is the in-OST offset at which
2567  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2568  * will re-calculate proper offset in next stripe.
2569  * Note that the first extent is passed to lov_get_info via the value field.
2570  *
2571  * \param fiemap fiemap request header
2572  * \param lsm striping information for the file
2573  * \param fm_start logical start of mapping
2574  * \param fm_end logical end of mapping
2575  * \param start_stripe starting stripe will be returned in this
2576  */
2577 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2578                                    struct lov_stripe_md *lsm, obd_size fm_start,
2579                                    obd_size fm_end, int *start_stripe)
2580 {
2581         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2582         obd_off lun_start, lun_end;
2583         obd_size fm_end_offset;
2584         int stripe_no = -1, i;
2585
2586         if (fiemap->fm_extent_count == 0 ||
2587             fiemap->fm_extents[0].fe_logical == 0)
2588                 return 0;
2589
2590         /* Find out stripe_no from ost_index saved in the fe_device */
2591         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2592                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2593                                         fiemap->fm_extents[0].fe_device) {
2594                         stripe_no = i;
2595                         break;
2596                 }
2597         }
2598
2599         /* If we have finished mapping on previous device, shift logical
2600          * offset to start of next device */
2601         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2602                                    &lun_start, &lun_end)) != 0 &&
2603                                    local_end < lun_end) {
2604                 fm_end_offset = local_end;
2605                 *start_stripe = stripe_no;
2606         } else {
2607                 /* This is a special value to indicate that caller should
2608                  * calculate offset in next stripe. */
2609                 fm_end_offset = 0;
2610                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2611         }
2612
2613         return fm_end_offset;
2614 }
2615
2616 /**
2617  * We calculate on which OST the mapping will end. If the length of mapping
2618  * is greater than (stripe_size * stripe_count) then the last_stripe will
2619  * will be one just before start_stripe. Else we check if the mapping
2620  * intersects each OST and find last_stripe.
2621  * This function returns the last_stripe and also sets the stripe_count
2622  * over which the mapping is spread
2623  *
2624  * \param lsm striping information for the file
2625  * \param fm_start logical start of mapping
2626  * \param fm_end logical end of mapping
2627  * \param start_stripe starting stripe of the mapping
2628  * \param stripe_count the number of stripes across which to map is returned
2629  *
2630  * \retval last_stripe return the last stripe of the mapping
2631  */
2632 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2633                             obd_size fm_end, int start_stripe,
2634                             int *stripe_count)
2635 {
2636         int last_stripe;
2637         obd_off obd_start, obd_end;
2638         int i, j;
2639
2640         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2641                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2642                                                               start_stripe - 1);
2643                 *stripe_count = lsm->lsm_stripe_count;
2644         } else {
2645                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2646                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2647                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2648                                                    &obd_start, &obd_end)) == 0)
2649                                 break;
2650                 }
2651                 *stripe_count = j;
2652                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2653         }
2654
2655         return last_stripe;
2656 }
2657
2658 /**
2659  * Set fe_device and copy extents from local buffer into main return buffer.
2660  *
2661  * \param fiemap fiemap request header
2662  * \param lcl_fm_ext array of local fiemap extents to be copied
2663  * \param ost_index OST index to be written into the fm_device field for each
2664                     extent
2665  * \param ext_count number of extents to be copied
2666  * \param current_extent where to start copying in main extent array
2667  */
2668 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2669                                   struct ll_fiemap_extent *lcl_fm_ext,
2670                                   int ost_index, unsigned int ext_count,
2671                                   int current_extent)
2672 {
2673         char *to;
2674         int ext;
2675
2676         for (ext = 0; ext < ext_count; ext++) {
2677                 lcl_fm_ext[ext].fe_device = ost_index;
2678                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2679         }
2680
2681         /* Copy fm_extent's from fm_local to return buffer */
2682         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2683         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2684 }
2685
2686 /**
2687  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2688  * This also handles the restarting of FIEMAP calls in case mapping overflows
2689  * the available number of extents in single call.
2690  */
2691 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2692                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2693 {
2694         struct ll_fiemap_info_key *fm_key = key;
2695         struct ll_user_fiemap *fiemap = val;
2696         struct ll_user_fiemap *fm_local = NULL;
2697         struct ll_fiemap_extent *lcl_fm_ext;
2698         int count_local;
2699         unsigned int get_num_extents = 0;
2700         int ost_index = 0, actual_start_stripe, start_stripe;
2701         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2702         obd_size curr_loc;
2703         int current_extent = 0, rc = 0, i;
2704         int ost_eof = 0; /* EOF for object */
2705         int ost_done = 0; /* done with required mapping for this OST? */
2706         int last_stripe;
2707         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2708         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2709
2710         if (lsm == NULL)
2711                 GOTO(out, rc = 0);
2712
2713         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2714                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2715
2716         OBD_ALLOC(fm_local, buffer_size);
2717         if (fm_local == NULL)
2718                 GOTO(out, rc = -ENOMEM);
2719         lcl_fm_ext = &fm_local->fm_extents[0];
2720
2721         count_local = fiemap_size_to_count(buffer_size);
2722
2723         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2724         fm_start = fiemap->fm_start;
2725         fm_length = fiemap->fm_length;
2726         /* Calculate start stripe, last stripe and length of mapping */
2727         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2728         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2729                                                 fm_start + fm_length - 1);
2730         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2731         if (fm_end > fm_key->oa.o_size)
2732                 fm_end = fm_key->oa.o_size;
2733
2734         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2735                                             actual_start_stripe, &stripe_count);
2736
2737         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2738                                                   &start_stripe);
2739
2740         if (fiemap->fm_extent_count == 0) {
2741                 get_num_extents = 1;
2742                 count_local = 0;
2743         }
2744
2745         /* Check each stripe */
2746         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2747              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2748                 obd_size req_fm_len; /* Stores length of required mapping */
2749                 obd_size len_mapped_single_call;
2750                 obd_off lun_start, lun_end, obd_object_end;
2751                 unsigned int ext_count;
2752
2753                 cur_stripe_wrap = cur_stripe;
2754
2755                 /* Find out range of mapping on this stripe */
2756                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2757                                            &lun_start, &obd_object_end)) == 0)
2758                         continue;
2759
2760                 /* If this is a continuation FIEMAP call and we are on
2761                  * starting stripe then lun_start needs to be set to
2762                  * fm_end_offset */
2763                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2764                         lun_start = fm_end_offset;
2765
2766                 if (fm_length != ~0ULL) {
2767                         /* Handle fm_start + fm_length overflow */
2768                         if (fm_start + fm_length < fm_start)
2769                                 fm_length = ~0ULL - fm_start;
2770                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2771                                                      cur_stripe);
2772                 } else {
2773                         lun_end = ~0ULL;
2774                 }
2775
2776                 if (lun_start == lun_end)
2777                         continue;
2778
2779                 req_fm_len = obd_object_end - lun_start;
2780                 fm_local->fm_length = 0;
2781                 len_mapped_single_call = 0;
2782
2783                 /* If the output buffer is very large and the objects have many
2784                  * extents we may need to loop on a single OST repeatedly */
2785                 ost_eof = 0;
2786                 ost_done = 0;
2787                 do {
2788                         if (get_num_extents == 0) {
2789                                 /* Don't get too many extents. */
2790                                 if (current_extent + count_local >
2791                                     fiemap->fm_extent_count)
2792                                         count_local = fiemap->fm_extent_count -
2793                                                                  current_extent;
2794                         }
2795
2796                         lun_start += len_mapped_single_call;
2797                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2798                         req_fm_len = fm_local->fm_length;
2799                         fm_local->fm_extent_count = count_local;
2800                         fm_local->fm_mapped_extents = 0;
2801                         fm_local->fm_flags = fiemap->fm_flags;
2802
2803                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2804                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2805
2806                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2807                                 GOTO(out, rc = -EINVAL);
2808
2809                         /* If OST is inactive, return extent with UNKNOWN flag */
2810                         if (lov && !lov->lov_tgts[ost_index]->ltd_active) {
2811                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2812                                 fm_local->fm_mapped_extents = 1;
2813
2814                                 lcl_fm_ext[0].fe_logical = lun_start;
2815                                 lcl_fm_ext[0].fe_length = obd_object_end -
2816                                                                       lun_start;
2817                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2818
2819                                 goto inactive_tgt;
2820                         }
2821
2822                         fm_local->fm_start = lun_start;
2823                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2824                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2825                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2826                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2827                                           keylen, key, vallen, fm_local, lsm);
2828                         if (rc != 0)
2829                                 GOTO(out, rc);
2830
2831 inactive_tgt:
2832                         ext_count = fm_local->fm_mapped_extents;
2833                         if (ext_count == 0) {
2834                                 ost_done = 1;
2835                                 /* If last stripe has hole at the end,
2836                                  * then we need to return */
2837                                 if (cur_stripe_wrap == last_stripe) {
2838                                         fiemap->fm_mapped_extents = 0;
2839                                         goto finish;
2840                                 }
2841                                 break;
2842                         }
2843
2844                         /* If we just need num of extents then go to next device */
2845                         if (get_num_extents) {
2846                                 current_extent += ext_count;
2847                                 break;
2848                         }
2849
2850                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2851                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2852
2853                         /* Have we finished mapping on this device? */
2854                         if (req_fm_len <= len_mapped_single_call)
2855                                 ost_done = 1;
2856
2857                         /* Clear the EXTENT_LAST flag which can be present on
2858                          * last extent */
2859                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2860                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2861                                                             ~FIEMAP_EXTENT_LAST;
2862
2863                         curr_loc = lov_stripe_size(lsm,
2864                                            lcl_fm_ext[ext_count - 1].fe_logical+
2865                                            lcl_fm_ext[ext_count - 1].fe_length,
2866                                            cur_stripe);
2867                         if (curr_loc >= fm_key->oa.o_size)
2868                                 ost_eof = 1;
2869
2870                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2871                                                      ost_index, ext_count,
2872                                                      current_extent);
2873
2874                         current_extent += ext_count;
2875
2876                         /* Ran out of available extents? */
2877                         if (current_extent >= fiemap->fm_extent_count)
2878                                 goto finish;
2879                 } while (ost_done == 0 && ost_eof == 0);
2880
2881                 if (cur_stripe_wrap == last_stripe)
2882                         goto finish;
2883         }
2884
2885 finish:
2886         /* Indicate that we are returning device offsets unless file just has
2887          * single stripe */
2888         if (lsm->lsm_stripe_count > 1)
2889                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2890
2891         if (get_num_extents)
2892                 goto skip_last_device_calc;
2893
2894         /* Check if we have reached the last stripe and whether mapping for that
2895          * stripe is done. */
2896         if (cur_stripe_wrap == last_stripe) {
2897                 if (ost_done || ost_eof)
2898                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2899                                                              FIEMAP_EXTENT_LAST;
2900         }
2901
2902 skip_last_device_calc:
2903         fiemap->fm_mapped_extents = current_extent;
2904
2905 out:
2906         OBD_FREE(fm_local, buffer_size);
2907         return rc;
2908 }
2909
2910 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2911                         void *key, __u32 *vallen, void *val,
2912                         struct lov_stripe_md *lsm)
2913 {
2914         struct obd_device *obddev = class_exp2obd(exp);
2915         struct lov_obd *lov = &obddev->u.lov;
2916         int i, rc;
2917         ENTRY;
2918
2919         if (!vallen || !val)
2920                 RETURN(-EFAULT);
2921
2922         lov_getref(obddev);
2923
2924         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2925                 struct {
2926                         char name[16];
2927                         struct ldlm_lock *lock;
2928                 } *data = key;
2929                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2930                 struct lov_oinfo *loi;
2931                 __u32 *stripe = val;
2932
2933                 if (*vallen < sizeof(*stripe))
2934                         GOTO(out, rc = -EFAULT);
2935                 *vallen = sizeof(*stripe);
2936
2937                 /* XXX This is another one of those bits that will need to
2938                  * change if we ever actually support nested LOVs.  It uses
2939                  * the lock's export to find out which stripe it is. */
2940                 /* XXX - it's assumed all the locks for deleted OSTs have
2941                  * been cancelled. Also, the export for deleted OSTs will
2942                  * be NULL and won't match the lock's export. */
2943                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2944                         loi = lsm->lsm_oinfo[i];
2945                         if (!lov->lov_tgts[loi->loi_ost_idx])
2946                                 continue;
2947                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2948                             data->lock->l_conn_export &&
2949                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2950                                 *stripe = i;
2951                                 GOTO(out, rc = 0);
2952                         }
2953                 }
2954                 LDLM_ERROR(data->lock, "lock on inode without such object");
2955                 dump_lsm(D_ERROR, lsm);
2956                 GOTO(out, rc = -ENXIO);
2957         } else if (KEY_IS(KEY_LAST_ID)) {
2958                 struct obd_id_info *info = val;
2959                 __u32 size = sizeof(obd_id);
2960                 struct lov_tgt_desc *tgt;
2961
2962                 LASSERT(*vallen == sizeof(struct obd_id_info));
2963                 tgt = lov->lov_tgts[info->idx];
2964
2965                 if (!tgt || !tgt->ltd_active)
2966                         GOTO(out, rc = -ESRCH);
2967
2968                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2969                 GOTO(out, rc = 0);
2970         } else if (KEY_IS(KEY_LOVDESC)) {
2971                 struct lov_desc *desc_ret = val;
2972                 *desc_ret = lov->desc;
2973
2974                 GOTO(out, rc = 0);
2975         } else if (KEY_IS(KEY_FIEMAP)) {
2976                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2977                 GOTO(out, rc);
2978         }
2979
2980         rc = -EINVAL;
2981
2982 out:
2983         lov_putref(obddev);
2984         RETURN(rc);
2985 }
2986
2987 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2988                               void *key, obd_count vallen, void *val,
2989                               struct ptlrpc_request_set *set)
2990 {
2991         struct obd_device *obddev = class_exp2obd(exp);
2992         struct lov_obd *lov = &obddev->u.lov;
2993         obd_count count;
2994         int i, rc = 0, err;
2995         struct lov_tgt_desc *tgt;
2996         unsigned incr, check_uuid,
2997                  do_inactive, no_set;
2998         unsigned next_id = 0,  mds_con = 0;
2999         ENTRY;
3000
3001         incr = check_uuid = do_inactive = no_set = 0;
3002         if (set == NULL) {
3003                 no_set = 1;
3004                 set = ptlrpc_prep_set();
3005                 if (!set)
3006                         RETURN(-ENOMEM);
3007         }
3008
3009         lov_getref(obddev);
3010         count = lov->desc.ld_tgt_count;
3011
3012         if (KEY_IS(KEY_NEXT_ID)) {
3013                 count = vallen / sizeof(struct obd_id_info);
3014                 vallen = sizeof(obd_id);
3015                 incr = sizeof(struct obd_id_info);
3016                 do_inactive = 1;
3017                 next_id = 1;
3018         } else if (KEY_IS(KEY_CHECKSUM)) {
3019                 do_inactive = 1;
3020         } else if (KEY_IS(KEY_UNLINKED)) {
3021                 check_uuid = val ? 1 : 0;
3022         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3023                 /* use defaults:  do_inactive = incr = 0; */
3024         } else if (KEY_IS(KEY_MDS_CONN)) {
3025                 mds_con = 1;
3026         }
3027
3028         for (i = 0; i < count; i++, val = (char *)val + incr) {
3029                 if (next_id) {
3030                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3031                 } else {
3032                         tgt = lov->lov_tgts[i];
3033                 }
3034                 /* OST was disconnected */
3035                 if (!tgt || !tgt->ltd_exp)
3036                         continue;
3037
3038                 /* OST is inactive and we don't want inactive OSCs */
3039                 if (!tgt->ltd_active && !do_inactive)
3040                         continue;
3041
3042                 if (mds_con) {
3043                         struct mds_group_info *mgi;
3044
3045                         LASSERT(vallen == sizeof(*mgi));
3046                         mgi = (struct mds_group_info *)val;
3047
3048                         /* Only want a specific OSC */
3049                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
3050                                                 &tgt->ltd_uuid))
3051                                 continue;
3052
3053                         err = obd_set_info_async(tgt->ltd_exp,
3054                                          keylen, key, sizeof(int),
3055                                          &mgi->group, set);
3056                 } else if (next_id) {
3057                         err = obd_set_info_async(tgt->ltd_exp,
3058                                          keylen, key, vallen,
3059                                          ((struct obd_id_info*)val)->data, set);
3060                 } else  {
3061                         /* Only want a specific OSC */
3062                         if (check_uuid &&
3063                             !obd_uuid_equals(val, &tgt->ltd_uuid))
3064                                 continue;
3065
3066                         err = obd_set_info_async(tgt->ltd_exp,
3067                                          keylen, key, vallen, val, set);
3068                 }
3069
3070                 if (!rc)
3071                         rc = err;
3072         }
3073
3074         lov_putref(obddev);
3075         if (no_set) {
3076                 err = ptlrpc_set_wait(set);
3077                 if (!rc)
3078                         rc = err;
3079                 ptlrpc_set_destroy(set);
3080         }
3081         RETURN(rc);
3082 }
3083
3084 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3085                        struct lov_stripe_md *lsm)
3086 {
3087         int rc;
3088         ENTRY;
3089
3090         if (!lsm)
3091                 RETURN(0);
3092         LASSERT(md_exp);
3093         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3094         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3095
3096         RETURN(rc);
3097 }
3098
3099 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3100 {
3101         int i, rc = 0;
3102         ENTRY;
3103
3104         for (i = 0; i < lsm->lsm_stripe_count; i++) {
3105                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
3106                 if (loi->loi_ar.ar_rc && !rc)
3107                         rc = loi->loi_ar.ar_rc;
3108                 loi->loi_ar.ar_rc = 0;
3109         }
3110         RETURN(rc);
3111 }
3112 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3113
3114
3115 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3116                            int cmd, __u64 *offset)
3117 {
3118         __u32 ssize = lsm->lsm_stripe_size;
3119         __u64 start;
3120
3121         start = *offset;
3122         do_div(start, ssize);
3123         start = start * ssize;
3124
3125         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3126                            ", end "LPU64"\n", *offset, ssize, start,
3127                            start + ssize - 1);
3128         if (cmd == OBD_CALC_STRIPE_END) {
3129                 *offset = start + ssize - 1;
3130         } else if (cmd == OBD_CALC_STRIPE_START) {
3131                 *offset = start;
3132         } else {
3133                 LBUG();
3134         }
3135
3136         RETURN(0);
3137 }
3138
3139
3140 #if 0
3141 struct lov_multi_wait {
3142         struct ldlm_lock *lock;
3143         wait_queue_t      wait;
3144         int               completed;
3145         int               generation;
3146 };
3147
3148 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3149                       struct lustre_handle *lockh)
3150 {
3151         struct lov_lock_handles *lov_lockh = NULL;
3152         struct lustre_handle *lov_lockhp;
3153         struct lov_obd *lov;
3154         struct lov_oinfo *loi;
3155         struct lov_multi_wait *queues;
3156         int rc = 0, i;
3157         ENTRY;
3158
3159         ASSERT_LSM_MAGIC(lsm);
3160
3161         if (!exp || !exp->exp_obd)
3162                 RETURN(-ENODEV);
3163
3164         LASSERT(lockh != NULL);
3165         if (lsm->lsm_stripe_count > 1) {
3166                 lov_lockh = lov_handle2llh(lockh);
3167                 if (lov_lockh == NULL) {
3168                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3169                         RETURN(-EINVAL);
3170                 }
3171
3172                 lov_lockhp = lov_lockh->llh_handles;
3173         } else {
3174                 lov_lockhp = lockh;
3175         }
3176
3177         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3178         if (queues == NULL)
3179                 GOTO(out, rc = -ENOMEM);
3180
3181         lov = &exp->exp_obd->u.lov;
3182         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3183              i++, loi++, lov_lockhp++) {
3184                 struct ldlm_lock *lock;
3185                 struct obd_device *obd;
3186
3187                 lock = ldlm_handle2lock(lov_lockhp);
3188                 if (lock == NULL) {
3189                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3190                                loi->loi_ost_idx, loi->loi_id);
3191                         queues[i].completed = 1;
3192                         continue;
3193                 }
3194
3195                 queues[i].lock = lock;
3196                 init_waitqueue_entry(&(queues[i].wait), current);
3197                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3198
3199                 obd = class_exp2obd(lock->l_conn_export);
3200                 if (obd != NULL)
3201                         imp = obd->u.cli.cl_import;
3202                 if (imp != NULL) {
3203                         spin_lock(&imp->imp_lock);
3204                         queues[i].generation = imp->imp_generation;
3205                         spin_unlock(&imp->imp_lock);
3206                 }
3207         }
3208
3209         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3210                                interrupted_completion_wait, &lwd);
3211         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3212
3213         for (i = 0; i < lsm->lsm_stripe_count; i++)
3214                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3215
3216         if (rc == -EINTR || rc == -ETIMEDOUT) {
3217
3218
3219         }
3220
3221  out:
3222         if (lov_lockh != NULL)
3223                 lov_llh_put(lov_lockh);
3224         RETURN(rc);
3225 }
3226 #endif
3227
3228 void lov_stripe_lock(struct lov_stripe_md *md)
3229 {
3230         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
3231         spin_lock(&md->lsm_lock);
3232         LASSERT(md->lsm_lock_owner == 0);
3233         md->lsm_lock_owner = cfs_curproc_pid();
3234 }
3235 EXPORT_SYMBOL(lov_stripe_lock);
3236
3237 void lov_stripe_unlock(struct lov_stripe_md *md)
3238 {
3239         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
3240         md->lsm_lock_owner = 0;
3241         spin_unlock(&md->lsm_lock);
3242 }
3243 EXPORT_SYMBOL(lov_stripe_unlock);
3244
3245 /**
3246  * Checks if requested extent lock is compatible with a lock under the page.
3247  *
3248  * Checks if the lock under \a page is compatible with a read or write lock
3249  * (specified by \a rw) for an extent [\a start , \a end].
3250  *
3251  * \param exp lov export
3252  * \param lsm striping information for the file
3253  * \param res lov_async_page placeholder
3254  * \param rw OBD_BRW_READ if requested for reading,
3255  *           OBD_BRW_WRITE if requested for writing
3256  * \param start start of the requested extent
3257  * \param end end of the requested extent
3258  * \param cookie transparent parameter for passing locking context
3259  *
3260  * \post result == 1, *cookie == context, appropriate lock is referenced or
3261  * \post result == 0
3262  *
3263  * \retval 1 owned lock is reused for the request
3264  * \retval 0 no lock reused for the request
3265  *
3266  * \see lov_release_short_lock
3267  */
3268 static int lov_reget_short_lock(struct obd_export *exp,
3269                                 struct lov_stripe_md *lsm,
3270                                 void **res, int rw,
3271                                 obd_off start, obd_off end,
3272                                 void **cookie)
3273 {
3274         struct lov_async_page *l = *res;
3275         obd_off stripe_start, stripe_end = start;
3276
3277         ENTRY;
3278
3279         /* ensure we don't cross stripe boundaries */
3280         lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3281         if (stripe_end <= end)
3282                 RETURN(0);
3283
3284         /* map the region limits to the object limits */
3285         lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3286         lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3287
3288         RETURN(obd_reget_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3289                                     lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3290                                     ltd_exp, NULL, &l->lap_sub_cookie,
3291                                     rw, stripe_start, stripe_end, cookie));
3292 }
3293
3294 /**
3295  * Releases a reference to a lock taken in a "fast" way.
3296  *
3297  * Releases a read or a write (specified by \a rw) lock
3298  * referenced by \a cookie.
3299  *
3300  * \param exp lov export
3301  * \param lsm striping information for the file
3302  * \param end end of the locked extent
3303  * \param rw OBD_BRW_READ if requested for reading,
3304  *           OBD_BRW_WRITE if requested for writing
3305  * \param cookie transparent parameter for passing locking context
3306  *
3307  * \post appropriate lock is dereferenced
3308  *
3309  * \see lov_reget_short_lock
3310  */
3311 static int lov_release_short_lock(struct obd_export *exp,
3312                                   struct lov_stripe_md *lsm, obd_off end,
3313                                   void *cookie, int rw)
3314 {
3315         int stripe;
3316
3317         ENTRY;
3318
3319         stripe = lov_stripe_number(lsm, end);
3320
3321         RETURN(obd_release_short_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3322                                       lsm_oinfo[stripe]->loi_ost_idx]->
3323                                       ltd_exp, NULL, end, cookie, rw));
3324 }
3325
3326 struct obd_ops lov_obd_ops = {
3327         .o_owner               = THIS_MODULE,
3328         .o_setup               = lov_setup,
3329         .o_precleanup          = lov_precleanup,
3330         .o_cleanup             = lov_cleanup,
3331         .o_process_config      = lov_process_config,
3332         .o_connect             = lov_connect,
3333         .o_disconnect          = lov_disconnect,
3334         .o_statfs              = lov_statfs,
3335         .o_statfs_async        = lov_statfs_async,
3336         .o_packmd              = lov_packmd,
3337         .o_unpackmd            = lov_unpackmd,
3338         .o_checkmd             = lov_checkmd,
3339         .o_create              = lov_create,
3340         .o_destroy             = lov_destroy,
3341         .o_getattr             = lov_getattr,
3342         .o_getattr_async       = lov_getattr_async,
3343         .o_setattr             = lov_setattr,
3344         .o_setattr_async       = lov_setattr_async,
3345         .o_brw                 = lov_brw,
3346         .o_brw_async           = lov_brw_async,
3347         .o_prep_async_page     = lov_prep_async_page,
3348         .o_reget_short_lock    = lov_reget_short_lock,
3349         .o_release_short_lock  = lov_release_short_lock,
3350         .o_queue_async_io      = lov_queue_async_io,
3351         .o_set_async_flags     = lov_set_async_flags,
3352         .o_queue_group_io      = lov_queue_group_io,
3353         .o_trigger_group_io    = lov_trigger_group_io,
3354         .o_teardown_async_page = lov_teardown_async_page,
3355         .o_merge_lvb           = lov_merge_lvb,
3356         .o_adjust_kms          = lov_adjust_kms,
3357         .o_punch               = lov_punch,
3358         .o_sync                = lov_sync,
3359         .o_enqueue             = lov_enqueue,
3360         .o_match               = lov_match,
3361         .o_change_cbdata       = lov_change_cbdata,
3362         .o_cancel              = lov_cancel,
3363         .o_cancel_unused       = lov_cancel_unused,
3364         .o_join_lru            = lov_join_lru,
3365         .o_iocontrol           = lov_iocontrol,
3366         .o_get_info            = lov_get_info,
3367         .o_set_info_async      = lov_set_info_async,
3368         .o_extent_calc         = lov_extent_calc,
3369         .o_llog_init           = lov_llog_init,
3370         .o_llog_finish         = lov_llog_finish,
3371         .o_notify              = lov_notify,
3372         .o_register_page_removal_cb = lov_register_page_removal_cb,
3373         .o_unregister_page_removal_cb = lov_unregister_page_removal_cb,
3374         .o_register_lock_cancel_cb = lov_register_lock_cancel_cb,
3375         .o_unregister_lock_cancel_cb = lov_unregister_lock_cancel_cb,
3376         .o_pool_new            = lov_pool_new,
3377         .o_pool_rem            = lov_pool_remove,
3378         .o_pool_add            = lov_pool_add,
3379         .o_pool_del            = lov_pool_del,
3380 };
3381
3382 static quota_interface_t *quota_interface;
3383 extern quota_interface_t lov_quota_interface;
3384
3385 cfs_mem_cache_t *lov_oinfo_slab;
3386
3387 int __init lov_init(void)
3388 {
3389         struct lprocfs_static_vars lvars = { 0 };
3390         int rc, rc2;
3391         ENTRY;
3392
3393         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3394                                               sizeof(struct lov_oinfo),
3395                                               0, SLAB_HWCACHE_ALIGN);
3396         if (lov_oinfo_slab == NULL)
3397                 return -ENOMEM;
3398         lprocfs_lov_init_vars(&lvars);
3399
3400         request_module("lquota");
3401         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3402         init_obd_quota_ops(quota_interface, &lov_obd_ops);
3403
3404         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
3405                                  LUSTRE_LOV_NAME, NULL);
3406         if (rc) {
3407                 if (quota_interface)
3408                         PORTAL_SYMBOL_PUT(lov_quota_interface);
3409                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3410                 LASSERT(rc2 == 0);
3411         }
3412
3413         RETURN(rc);
3414 }
3415
3416 #ifdef __KERNEL__
3417 static void /*__exit*/ lov_exit(void)
3418 {
3419         int rc;
3420
3421         if (quota_interface)
3422                 PORTAL_SYMBOL_PUT(lov_quota_interface);
3423
3424         class_unregister_type(LUSTRE_LOV_NAME);
3425         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3426         LASSERT(rc == 0);
3427 }
3428
3429 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3430 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3431 MODULE_LICENSE("GPL");
3432
3433 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
3434 #endif