lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #ifndef EXPORT_SYMTAB
  45 # define EXPORT_SYMTAB
  46 #endif
  47 #define DEBUG_SUBSYSTEM S_LOV
  48 #ifdef __KERNEL__
  49 #include <libcfs/libcfs.h>
  50 #else
  51 #include <liblustre.h>
  52 #endif
  53
  54 #include <obd_support.h>
  55 #include <lustre_lib.h>
  56 #include <lustre_net.h>
  57 #include <lustre/lustre_idl.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_mds.h>
  60 #include <lustre_debug.h>
  61 #include <obd_class.h>
  62 #include <obd_lov.h>
  63 #include <obd_ost.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_param.h>
  66 #include <lustre_cache.h>
  67 #include <lustre/ll_fiemap.h>
  68
  69 #include "lov_internal.h"
  70
  71
  72 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  73    Any function that expects lov_tgts to remain stationary must take a ref. */
  74 static void lov_getref(struct obd_device *obd)
  75 {
  76         struct lov_obd *lov = &obd->u.lov;
  77
  78         /* nobody gets through here until lov_putref is done */
  79         mutex_down(&lov->lov_lock);
  80         atomic_inc(&lov->lov_refcount);
  81         mutex_up(&lov->lov_lock);
  82         return;
  83 }
  84
  85 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  86
  87 static void lov_putref(struct obd_device *obd)
  88 {
  89         struct lov_obd *lov = &obd->u.lov;
  90         CFS_LIST_HEAD(kill);
  91         struct lov_tgt_desc *tgt;
  92
  93         mutex_down(&lov->lov_lock);
  94         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  95         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  96                 int i;
  97                 struct lov_tgt_desc *n;
  98                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  99                        lov->lov_death_row);
 100                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 101                         tgt = lov->lov_tgts[i];
 102
 103                         if (!tgt || !tgt->ltd_reap)
 104                                 continue;
 105                         list_add(&tgt->ltd_kill, &kill);
 106                         /* XXX - right now there is a dependency on ld_tgt_count
 107                          * being the maximum tgt index for computing the
 108                          * mds_max_easize. So we can't shrink it. */
 109                         lov_ost_pool_remove(&lov->lov_packed, i);
 110                         lov->lov_tgts[i] = NULL;
 111                         lov->lov_death_row--;
 112                 }
 113                 mutex_up(&lov->lov_lock);
 114
 115                 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 116                         list_del(&tgt->ltd_kill);
 117                         /* Disconnect */
 118                         __lov_del_obd(obd, tgt);
 119                 }
 120         } else {
 121                 mutex_up(&lov->lov_lock);
 122         }
 123 }
 124
 125 static int lov_obd_register_page_removal_cb(struct obd_device *obd,
 126                                         obd_page_removal_cb_t func,
 127                                         obd_pin_extent_cb pin_cb)
 128 {
 129         struct lov_obd *lov = &obd->u.lov;
 130
 131         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 132                 return -EBUSY;
 133
 134         if (lov->lov_page_pin_cb && lov->lov_page_pin_cb != pin_cb)
 135                 return -EBUSY;
 136
 137         lov->lov_page_removal_cb = func;
 138         lov->lov_page_pin_cb = pin_cb;
 139         return 0;
 140 }
 141
 142 static int lov_obd_unregister_page_removal_cb(struct obd_device *obd,
 143                                               obd_page_removal_cb_t func)
 144 {
 145         struct lov_obd *lov = &obd->u.lov;
 146         int i, rc = 0;
 147
 148         if (lov->lov_page_removal_cb && lov->lov_page_removal_cb != func)
 149                 return -EINVAL;
 150
 151         lov->lov_page_removal_cb = NULL;
 152         lov->lov_page_pin_cb = NULL;
 153
 154         obd_getref(obd);
 155         for (i = 0; i < lov->desc.ld_tgt_count; ++i) {
 156                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
 157                         continue;
 158                 rc |= obd_unregister_page_removal_cb(
 159                                 lov->lov_tgts[i]->ltd_exp->exp_obd, func);
 160         }
 161         obd_putref(obd);
 162
 163         return rc;
 164 }
 165
 166 static int lov_obd_register_lock_cancel_cb(struct obd_device *obd,
 167                                            obd_lock_cancel_cb func)
 168 {
 169         struct lov_obd *lov = &obd->u.lov;
 170
 171         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 172                 return -EBUSY;
 173
 174         lov->lov_lock_cancel_cb = func;
 175
 176         return 0;
 177 }
 178
 179 static int lov_obd_unregister_lock_cancel_cb(struct obd_device *obd,
 180                                              obd_lock_cancel_cb func)
 181 {
 182         struct lov_obd *lov = &obd->u.lov;
 183
 184         if (lov->lov_lock_cancel_cb && lov->lov_lock_cancel_cb != func)
 185                 return -EINVAL;
 186
 187         lov->lov_lock_cancel_cb = NULL;
 188         return 0;
 189
 190 }
 191
 192 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 193                               int activate);
 194
 195 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 196                       enum obd_notify_event ev, void *data)
 197 {
 198         int rc = 0;
 199         ENTRY;
 200
 201         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 202                 LASSERT(watched);
 203
 204                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 205                         CERROR("unexpected notification of %s %s!\n",
 206                                watched->obd_type->typ_name,
 207                                watched->obd_name);
 208                         RETURN(-EINVAL);
 209                 }
 210
 211                 /* Set OSC as active before notifying the observer, so the
 212                  * observer can use the OSC normally.
 213                  */
 214                 rc = lov_set_osc_active(obd, &watched->u.cli.cl_target_uuid,
 215                                         ev == OBD_NOTIFY_ACTIVE);
 216                 if (rc < 0) {
 217                         CERROR("%sactivation of %s failed: %d\n",
 218                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 219                                obd_uuid2str(&watched->u.cli.cl_target_uuid),
 220                                rc);
 221                         RETURN(rc);
 222                 }
 223                 /* active event should be pass lov target index as data */
 224                 data = &rc;
 225         }
 226
 227         /* Pass the notification up the chain. */
 228         if (watched) {
 229                 rc = obd_notify_observer(obd, watched, ev, data);
 230         } else {
 231                 /* NULL watched means all osc's in the lov (only for syncs) */
 232                 /* sync event should be send lov idx as data */
 233                 struct lov_obd *lov = &obd->u.lov;
 234                 int i, is_sync;
 235
 236                 data = &i;
 237                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 238                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 239
 240                 obd_getref(obd);
 241                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 242
 243                         if (!lov->lov_tgts[i])
 244                                 continue;
 245                         /* don't send sync event if target not
 246                          * connected/activated */
 247                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 248                                 continue;
 249
 250                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 251                                                  ev, data);
 252                         if (rc) {
 253                                 CERROR("%s: notify %s of %s failed %d\n",
 254                                        obd->obd_name,
 255                                        obd->obd_observer->obd_name,
 256                                        lov->lov_tgts[i]->ltd_obd->obd_name, rc);
 257                         }
 258                 }
 259                 obd_putref(obd);
 260         }
 261
 262         RETURN(rc);
 263 }
 264
 265 #define MAX_STRING_SIZE 128
 266 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 267                            struct obd_connect_data *data)
 268 {
 269         struct lov_obd *lov = &obd->u.lov;
 270         struct obd_uuid tgt_uuid;
 271         struct obd_device *tgt_obd;
 272         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 273         struct lustre_handle conn = {0, };
 274         struct obd_import *imp;
 275 #ifdef __KERNEL__
 276         cfs_proc_dir_entry_t *lov_proc_dir;
 277 #endif
 278         int rc;
 279         ENTRY;
 280
 281         if (!lov->lov_tgts[index])
 282                 RETURN(-EINVAL);
 283
 284         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 285         if (!tgt_obd->obd_set_up) {
 286                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 287                 RETURN(-EINVAL);
 288         }
 289
 290         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 291                 data->ocd_index = index;
 292
 293         /*
 294          * Divine LOV knows that OBDs under it are OSCs.
 295          */
 296         imp = tgt_obd->u.cli.cl_import;
 297
 298         if (activate) {
 299                 tgt_obd->obd_no_recov = 0;
 300                 /* FIXME this is probably supposed to be
 301                    ptlrpc_set_import_active.  Horrible naming. */
 302                 ptlrpc_activate_import(imp);
 303         }
 304
 305         rc = obd_register_observer(tgt_obd, obd);
 306         if (rc) {
 307                 CERROR("Target %s register_observer error %d; "
 308                         "will not be able to reactivate\n",
 309                         obd_uuid2str(&tgt_uuid), rc);
 310                 RETURN(rc);
 311         }
 312
 313         if (imp->imp_invalid) {
 314                 CERROR("not connecting OSC %s; administratively "
 315                        "disabled\n", obd_uuid2str(&tgt_uuid));
 316                 RETURN(0);
 317         }
 318         if (lov->lov_lock_cancel_cb)
 319                 rc = obd_register_lock_cancel_cb(tgt_obd, lov->lov_lock_cancel_cb);
 320                 if (rc)
 321                         RETURN(rc);
 322
 323         if (lov->lov_page_removal_cb)
 324                 rc = obd_register_page_removal_cb(tgt_obd, lov->lov_page_removal_cb,
 325                                                   lov->lov_page_pin_cb);
 326                 if (rc)
 327                         GOTO(out_lock_cb, rc);
 328
 329         rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, data, &lov->lov_tgts[index]->ltd_exp);
 330         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 331                 CERROR("Target %s connect error %d\n",
 332                        obd_uuid2str(&tgt_uuid), rc);
 333                 GOTO(out_page_cb, rc);
 334         }
 335
 336 #ifdef __KERNEL__
 337         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 338         if (lov_proc_dir) {
 339                 struct obd_device *osc_obd = class_conn2obd(&conn);
 340                 cfs_proc_dir_entry_t *osc_symlink;
 341                 char name[MAX_STRING_SIZE];
 342
 343                 LASSERT(osc_obd != NULL);
 344                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 345                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 346                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 347                          osc_obd->obd_type->typ_name,
 348                          osc_obd->obd_name);
 349                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 350                                            name);
 351                 if (osc_symlink == NULL) {
 352                         CERROR("could not register LOV target "
 353                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 354                                obd->obd_type->typ_name, obd->obd_name,
 355                                osc_obd->obd_name);
 356                         lprocfs_remove(&lov_proc_dir);
 357                 }
 358         }
 359 #endif
 360         rc = qos_add_tgt(obd, index);
 361         if (rc)
 362                 CERROR("qos_add_tgt failed %d\n", rc);
 363
 364         RETURN(0);
 365
 366 out_page_cb:
 367         obd_unregister_page_removal_cb(obd, lov->lov_page_removal_cb);
 368 out_lock_cb:
 369         obd_unregister_lock_cancel_cb(obd, lov->lov_lock_cancel_cb);
 370
 371         RETURN(rc);
 372 }
 373
 374 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
 375                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 376                        void *localdata)
 377 {
 378         struct lov_obd *lov = &obd->u.lov;
 379         struct lov_tgt_desc *tgt;
 380         struct obd_export **exp = localdata;
 381         int i, rc;
 382         ENTRY;
 383
 384         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 385
 386         rc = class_connect(conn, obd, cluuid);
 387         if (rc)
 388                 RETURN(rc);
 389
 390         *exp = class_conn2export(conn);
 391
 392         /* Why should there ever be more than 1 connect? */
 393         lov->lov_connects++;
 394         LASSERT(lov->lov_connects == 1);
 395
 396         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 397         if (data)
 398                 lov->lov_ocd = *data;
 399
 400         obd_getref(obd);
 401         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 402                 tgt = lov->lov_tgts[i];
 403                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 404                         continue;
 405                 /* Flags will be lowest common denominator */
 406                 rc = lov_connect_obd(obd, i, lov->lov_tgts[i]->ltd_activate,
 407                                      &lov->lov_ocd);
 408                 if (rc) {
 409                         CERROR("%s: lov connect tgt %d failed: %d\n",
 410                                obd->obd_name, i, rc);
 411                         continue;
 412                 }
 413                 /* connect to administrative disabled ost */
 414                 if (!lov->lov_tgts[i]->ltd_exp)
 415                         continue;
 416
 417                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 418                                 OBD_NOTIFY_CONNECT, (void *)&i);
 419                 if (rc) {
 420                         CERROR("%s error sending notify %d\n",
 421                                obd->obd_name, rc);
 422                 }
 423         }
 424         obd_putref(obd);
 425
 426         RETURN(0);
 427 }
 428
 429 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 430 {
 431         cfs_proc_dir_entry_t *lov_proc_dir;
 432         struct lov_obd *lov = &obd->u.lov;
 433         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
 434         int rc;
 435
 436         ENTRY;
 437
 438         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 439                obd->obd_name, osc_obd->obd_name);
 440
 441         if (tgt->ltd_active) {
 442                 tgt->ltd_active = 0;
 443                 lov->desc.ld_active_tgt_count--;
 444                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 445         }
 446
 447         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 448         if (lov_proc_dir) {
 449                 cfs_proc_dir_entry_t *osc_symlink;
 450
 451                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 452                 if (osc_symlink) {
 453                         lprocfs_remove(&osc_symlink);
 454                 } else {
 455                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 456                                obd->obd_type->typ_name, obd->obd_name,
 457                                osc_obd->obd_name);
 458                 }
 459         }
 460
 461         if (obd->obd_force) {
 462                 /* Pass it on to our clients.
 463                  * XXX This should be an argument to disconnect,
 464                  * XXX not a back-door flag on the OBD.  Ah well.
 465                  */
 466                 if (osc_obd)
 467                         osc_obd->obd_force = 1;
 468         }
 469
 470         obd_register_observer(osc_obd, NULL);
 471
 472         obd_unregister_page_removal_cb(osc_obd, lov->lov_page_removal_cb);
 473         obd_unregister_lock_cancel_cb(osc_obd, lov->lov_lock_cancel_cb);
 474
 475         rc = obd_disconnect(tgt->ltd_exp);
 476         if (rc) {
 477                 CERROR("Target %s disconnect error %d\n",
 478                        tgt->ltd_uuid.uuid, rc);
 479                 rc = 0;
 480         }
 481
 482         qos_del_tgt(obd, tgt);
 483
 484         tgt->ltd_exp = NULL;
 485
 486         RETURN(0);
 487 }
 488
 489 static int lov_del_target(struct obd_device *obd, __u32 index,
 490                           struct obd_uuid *uuidp, int gen);
 491
 492 static int lov_disconnect(struct obd_export *exp)
 493 {
 494         struct obd_device *obd = class_exp2obd(exp);
 495         struct lov_obd *lov = &obd->u.lov;
 496         int i, rc;
 497         ENTRY;
 498
 499         if (!lov->lov_tgts)
 500                 goto out;
 501
 502         /* Only disconnect the underlying layers on the final disconnect. */
 503         lov->lov_connects--;
 504         if (lov->lov_connects != 0) {
 505                 /* why should there be more than 1 connect? */
 506                 CERROR("disconnect #%d\n", lov->lov_connects);
 507                 goto out;
 508         }
 509
 510         /* Let's hold another reference so lov_del_obd doesn't spin through
 511            putref every time */
 512         obd_getref(obd);
 513         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 514                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 515                         /* Disconnection is the last we know about an obd */
 516                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 517                 }
 518         }
 519         obd_putref(obd);
 520
 521 out:
 522         rc = class_disconnect(exp); /* bz 9811 */
 523         RETURN(rc);
 524 }
 525
 526 /* Error codes:
 527  *
 528  *  -EINVAL  : UUID can't be found in the LOV's target list
 529  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 530  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 531  *  - any above 0 is lov index
 532  */
 533 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 534                               int activate)
 535 {
 536         struct lov_obd *lov = &obd->u.lov;
 537         struct lov_tgt_desc *tgt;
 538         int i = 0;
 539         ENTRY;
 540
 541         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 542                lov, uuid->uuid, activate);
 543
 544         obd_getref(obd);
 545         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 546                 tgt = lov->lov_tgts[i];
 547                 if (!tgt || !tgt->ltd_exp)
 548                         continue;
 549
 550                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 551                        i, obd_uuid2str(&tgt->ltd_uuid),
 552                        tgt->ltd_exp->exp_handle.h_cookie);
 553                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 554                         break;
 555         }
 556
 557         if (i == lov->desc.ld_tgt_count)
 558                 GOTO(out, i = -EINVAL);
 559
 560         if (lov->lov_tgts[i]->ltd_active == activate) {
 561                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 562                        activate ? "" : "in");
 563                 GOTO(out, i);
 564         }
 565
 566         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 567                activate ? "" : "in");
 568
 569         lov->lov_tgts[i]->ltd_active = activate;
 570
 571         if (activate) {
 572                 lov->desc.ld_active_tgt_count++;
 573                 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 0;
 574         } else {
 575                 lov->desc.ld_active_tgt_count--;
 576                 lov->lov_tgts[i]->ltd_exp->exp_obd->obd_inactive = 1;
 577         }
 578         /* remove any old qos penalty */
 579         lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
 580
 581  out:
 582         obd_putref(obd);
 583         RETURN(i);
 584 }
 585
 586
 587 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 588                           __u32 index, int gen, int active)
 589 {
 590         struct obd_device *tgt_obd;
 591         struct lov_obd *lov = &obd->u.lov;
 592         struct lov_tgt_desc *tgt;
 593         int rc;
 594         ENTRY;
 595
 596         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 597                uuidp->uuid, index, gen, active);
 598
 599         if (gen <= 0) {
 600                 CERROR("request to add OBD %s with invalid generation: %d\n",
 601                        uuidp->uuid, gen);
 602                 RETURN(-EINVAL);
 603         }
 604
 605
 606         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 607                                         &obd->obd_uuid);
 608         if (!tgt_obd) {
 609                 CERROR("Target %s not attached\n", obd_uuid2str(uuidp));
 610                 RETURN(-EINVAL);
 611         }
 612
 613         mutex_down(&lov->lov_lock);
 614
 615         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 616                 tgt = lov->lov_tgts[index];
 617                 CERROR("UUID %s already assigned at LOV target index %d\n",
 618                        obd_uuid2str(&tgt->ltd_uuid), index);
 619                 GOTO(err_unlock, rc = -EEXIST);
 620         }
 621
 622         if (index >= lov->lov_tgt_size) {
 623                 /* We need to reallocate the lov target array. */
 624                 struct lov_tgt_desc **newtgts, **old = NULL;
 625                 __u32 newsize, oldsize = 0;
 626
 627                 newsize = max(lov->lov_tgt_size, (__u32)2);
 628                 while (newsize < index + 1)
 629                         newsize = newsize << 1;
 630                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 631                 if (newtgts == NULL)
 632                         GOTO(err_unlock, rc = -ENOMEM);
 633
 634                 if (lov->lov_tgt_size) {
 635                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 636                                lov->lov_tgt_size);
 637                         old = lov->lov_tgts;
 638                         oldsize = lov->lov_tgt_size;
 639                 }
 640
 641                 lov->lov_tgts = newtgts;
 642                 lov->lov_tgt_size = newsize;
 643 #ifdef __KERNEL__
 644                 smp_rmb();
 645 #endif
 646                 if (old)
 647                         OBD_FREE(old, sizeof(*old) * oldsize);
 648
 649                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 650                        lov->lov_tgts, lov->lov_tgt_size);
 651         }
 652
 653         OBD_ALLOC_PTR(tgt);
 654         if (!tgt)
 655                 GOTO(err_unlock, rc = -ENOMEM);
 656
 657         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 658         if (rc)
 659                 GOTO(err_free_tgt, rc = -EEXIST);
 660
 661         memset(tgt, 0, sizeof(*tgt));
 662         tgt->ltd_obd = tgt_obd;
 663         tgt->ltd_uuid = *uuidp;
 664         /* XXX - add a sanity check on the generation number. */
 665         tgt->ltd_gen = gen;
 666         tgt->ltd_index = index;
 667         tgt->ltd_activate = active;
 668
 669         lov->lov_tgts[index] = tgt;
 670         if (index >= lov->desc.ld_tgt_count)
 671                 lov->desc.ld_tgt_count = index + 1;
 672
 673         mutex_up(&lov->lov_lock);
 674
 675         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 676                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 677
 678         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 679
 680         if (lov->lov_connects == 0) {
 681                 /* lov_connect hasn't been called yet. We'll do the
 682                    lov_connect_obd on this target when that fn first runs,
 683                    because we don't know the connect flags yet. */
 684                 RETURN(0);
 685         }
 686
 687         obd_getref(obd);
 688
 689         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 690         if (rc)
 691                 GOTO(out, rc);
 692
 693         /* connect to administrative disabled ost */
 694         if (!tgt->ltd_exp)
 695                 GOTO(out, rc = 0);
 696
 697         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 698                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 699                         (void *)&index);
 700 out:
 701         if (rc) {
 702                 /* connect or notify failed - we can try connect later
 703                  * instead of complete delete target */
 704                 CERROR("connect or notify failed (%d) for %s\n", rc,
 705                        obd_uuid2str(&tgt->ltd_uuid));
 706         }
 707         obd_putref(obd);
 708         RETURN(rc);
 709
 710 err_free_tgt:
 711         OBD_FREE_PTR(tgt);
 712 err_unlock:
 713         mutex_up(&lov->lov_lock);
 714         RETURN(rc);
 715 }
 716
 717 /* Schedule a target for deletion */
 718 static int lov_del_target(struct obd_device *obd, __u32 index,
 719                           struct obd_uuid *uuidp, int gen)
 720 {
 721         struct lov_obd *lov = &obd->u.lov;
 722         int count = lov->desc.ld_tgt_count;
 723         int rc = 0;
 724         ENTRY;
 725
 726         if (index >= count) {
 727                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 728                        index, count);
 729                 RETURN(-EINVAL);
 730         }
 731
 732         obd_getref(obd);
 733
 734         if (!lov->lov_tgts[index]) {
 735                 CERROR("LOV target at index %d is not setup.\n", index);
 736                 GOTO(out, rc = -EINVAL);
 737         }
 738
 739         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 740                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 741                        lov_uuid2str(lov, index), index,
 742                        obd_uuid2str(uuidp));
 743                 GOTO(out, rc = -EINVAL);
 744         }
 745
 746         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 747                lov_uuid2str(lov, index), index,
 748                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 749                lov->lov_tgts[index]->ltd_active);
 750
 751         lov->lov_tgts[index]->ltd_reap = 1;
 752         lov->lov_death_row++;
 753         /* we really delete it from obd_putref */
 754 out:
 755         obd_putref(obd);
 756
 757         RETURN(rc);
 758 }
 759
 760 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 761 {
 762         struct obd_device *osc_obd;
 763
 764         LASSERT(tgt);
 765         LASSERT(tgt->ltd_reap);
 766
 767         osc_obd = class_exp2obd(tgt->ltd_exp);
 768
 769         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 770                tgt->ltd_uuid.uuid,
 771                osc_obd ? osc_obd->obd_name : "<no obd>");
 772
 773         if (tgt->ltd_exp)
 774                 lov_disconnect_obd(obd, tgt);
 775
 776         OBD_FREE_PTR(tgt);
 777
 778         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 779            do it ourselves. And we can't do it from lov_cleanup,
 780            because we just lost our only reference to it. */
 781         if (osc_obd) {
 782                 /* Use lov's force/fail flags. */
 783                 osc_obd->obd_force = obd->obd_force;
 784                 osc_obd->obd_fail = obd->obd_fail;
 785                 class_manual_cleanup(osc_obd);
 786         }
 787 }
 788
 789 void lov_fix_desc_stripe_size(__u64 *val)
 790 {
 791         if (*val < PTLRPC_MAX_BRW_SIZE) {
 792                 if (*val)
 793                         LCONSOLE_WARN("Increasing default stripe size from "
 794                                       LPU64" to %u\n",*val,PTLRPC_MAX_BRW_SIZE);
 795                 *val = PTLRPC_MAX_BRW_SIZE;
 796         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 797                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 798                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 799                               "multiple of %u)\n",
 800                               *val, LOV_MIN_STRIPE_SIZE);
 801         }
 802 }
 803
 804 void lov_fix_desc_stripe_count(__u32 *val)
 805 {
 806         if (*val == 0)
 807                 *val = 1;
 808 }
 809
 810 void lov_fix_desc_pattern(__u32 *val)
 811 {
 812         /* from lov_setstripe */
 813         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 814                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 815                 *val = 0;
 816         }
 817 }
 818
 819 void lov_fix_desc_qos_maxage(__u32 *val)
 820 {
 821         /* fix qos_maxage */
 822         if (*val == 0)
 823                 *val = QOS_DEFAULT_MAXAGE;
 824 }
 825
 826 void lov_fix_desc(struct lov_desc *desc)
 827 {
 828         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 829         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 830         lov_fix_desc_pattern(&desc->ld_pattern);
 831         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 832 }
 833
 834 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
 835 {
 836         struct lprocfs_static_vars lvars = { 0 };
 837         struct lustre_cfg *lcfg = buf;
 838         struct lov_desc *desc;
 839         struct lov_obd *lov = &obd->u.lov;
 840         int rc;
 841         ENTRY;
 842
 843         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 844                 CERROR("LOV setup requires a descriptor\n");
 845                 RETURN(-EINVAL);
 846         }
 847
 848         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 849
 850         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 851                 CERROR("descriptor size wrong: %d > %d\n",
 852                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 853                 RETURN(-EINVAL);
 854         }
 855
 856         if (desc->ld_magic != LOV_DESC_MAGIC) {
 857                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 858                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 859                                    obd->obd_name, desc);
 860                             lustre_swab_lov_desc(desc);
 861                 } else {
 862                         CERROR("%s: Bad lov desc magic: %#x\n",
 863                                obd->obd_name, desc->ld_magic);
 864                         RETURN(-EINVAL);
 865                 }
 866         }
 867
 868         lov_fix_desc(desc);
 869
 870         desc->ld_active_tgt_count = 0;
 871         lov->desc = *desc;
 872         lov->lov_tgt_size = 0;
 873
 874         sema_init(&lov->lov_lock, 1);
 875         atomic_set(&lov->lov_refcount, 0);
 876         CFS_INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 877         init_rwsem(&lov->lov_qos.lq_rw_sem);
 878         lov->lov_qos.lq_dirty = 1;
 879         lov->lov_qos.lq_rr.lqr_dirty = 1;
 880         lov->lov_qos.lq_reset = 1;
 881         /* Default priority is toward free space balance */
 882         lov->lov_qos.lq_prio_free = 232;
 883         /* Default threshold for rr (roughly 17%) */
 884         lov->lov_qos.lq_threshold_rr = 43;
 885         /* Init statfs fields */
 886         OBD_ALLOC_PTR(lov->lov_qos.lq_statfs_data);
 887         if (NULL == lov->lov_qos.lq_statfs_data)
 888                 RETURN(-ENOMEM);
 889         cfs_waitq_init(&lov->lov_qos.lq_statfs_waitq);
 890
 891         lov->lov_pools_hash_body = lustre_hash_init("POOLS",
 892                                                     HASH_POOLS_CUR_BITS,
 893                                                     HASH_POOLS_MAX_BITS,
 894                                                     &pool_hash_operations, 0);
 895         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 896         lov->lov_pool_count = 0;
 897         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 898         if (rc)
 899                 RETURN(rc);
 900         rc = lov_ost_pool_init(&lov->lov_qos.lq_rr.lqr_pool, 0);
 901         if (rc) {
 902                 lov_ost_pool_free(&lov->lov_packed);
 903                 RETURN(rc);
 904         }
 905
 906         lprocfs_lov_init_vars(&lvars);
 907         lprocfs_obd_setup(obd, lvars.obd_vars);
 908 #ifdef LPROCFS
 909         {
 910                 cfs_proc_dir_entry_t *entry;
 911
 912                 entry = create_proc_entry("target_obd", 0444,
 913                                           obd->obd_proc_entry);
 914                 if (entry != NULL) {
 915                         entry->proc_fops = &lov_proc_target_fops;
 916                         entry->data = obd;
 917                 }
 918         }
 919 #endif
 920         lov->lov_pool_proc_entry = lprocfs_register("pools",
 921                                                     obd->obd_proc_entry,
 922                                                     NULL, NULL);
 923
 924         RETURN(0);
 925 }
 926
 927 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 928 {
 929         int rc = 0;
 930         ENTRY;
 931
 932         switch (stage) {
 933         case OBD_CLEANUP_EARLY: {
 934                 struct lov_obd *lov = &obd->u.lov;
 935                 int i;
 936                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 937                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 938                                 continue;
 939                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 940                                        OBD_CLEANUP_EARLY);
 941                 }
 942                 break;
 943         }
 944         case OBD_CLEANUP_EXPORTS:
 945                 break;
 946         case OBD_CLEANUP_SELF_EXP:
 947                 rc = obd_llog_finish(obd, 0);
 948                 if (rc != 0)
 949                         CERROR("failed to cleanup llogging subsystems\n");
 950                 break;
 951         case OBD_CLEANUP_OBD:
 952                 break;
 953         }
 954         RETURN(rc);
 955 }
 956
 957 static int lov_cleanup(struct obd_device *obd)
 958 {
 959         struct lov_obd *lov = &obd->u.lov;
 960         struct list_head *pos, *tmp;
 961         struct pool_desc *pool;
 962
 963         list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 964                 pool = list_entry(pos, struct pool_desc, pool_list);
 965                 /* free the pool structs */
 966                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 967                 lov_pool_del(obd, pool->pool_name);
 968         }
 969
 970         lustre_hash_exit(lov->lov_pools_hash_body);
 971
 972         lov_ost_pool_free(&(lov->lov_qos.lq_rr.lqr_pool));
 973         lov_ost_pool_free(&lov->lov_packed);
 974
 975         if (lov->lov_tgts) {
 976                 int i;
 977                 obd_getref(obd);
 978                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 979                         if (lov->lov_tgts[i]) {
 980                                 /* Inactive targets may never have connected */
 981                                 if (lov->lov_tgts[i]->ltd_active ||
 982                                     atomic_read(&lov->lov_refcount))
 983                                         /* We should never get here - these
 984                                            should have been removed in the
 985                                            disconnect. */
 986                                         CERROR("lov tgt %d not cleaned!"
 987                                                " deathrow=%d, lovrc=%d\n",
 988                                                i, lov->lov_death_row,
 989                                                atomic_read(&lov->lov_refcount));
 990                                 lov_del_target(obd, i, 0, 0);
 991                         }
 992                 }
 993                 obd_putref(obd);
 994                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 995                          lov->lov_tgt_size);
 996                 lov->lov_tgt_size = 0;
 997         }
 998
 999         /* clear pools parent proc entry only after all pools is killed */
1000         lprocfs_obd_cleanup(obd);
1001
1002         OBD_FREE_PTR(lov->lov_qos.lq_statfs_data);
1003         RETURN(0);
1004 }
1005
1006 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
1007 {
1008         struct lustre_cfg *lcfg = buf;
1009         struct obd_uuid obd_uuid;
1010         int cmd;
1011         int rc = 0;
1012         ENTRY;
1013
1014         switch(cmd = lcfg->lcfg_command) {
1015         case LCFG_LOV_ADD_OBD:
1016         case LCFG_LOV_ADD_INA:
1017         case LCFG_LOV_DEL_OBD: {
1018                 __u32 index;
1019                 int gen;
1020                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
1021                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1022                         GOTO(out, rc = -EINVAL);
1023
1024                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1025
1026                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1027                         GOTO(out, rc = -EINVAL);
1028                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1029                         GOTO(out, rc = -EINVAL);
1030                 if (cmd == LCFG_LOV_ADD_OBD)
1031                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
1032                 else if (cmd == LCFG_LOV_ADD_INA)
1033                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
1034                 else
1035                         rc = lov_del_target(obd, index, &obd_uuid, gen);
1036                 GOTO(out, rc);
1037         }
1038         case LCFG_PARAM: {
1039                 struct lprocfs_static_vars lvars = { 0 };
1040                 struct lov_desc *desc = &(obd->u.lov.desc);
1041
1042                 if (!desc)
1043                         GOTO(out, rc = -EINVAL);
1044
1045                 lprocfs_lov_init_vars(&lvars);
1046
1047                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
1048                                               lcfg, obd);
1049                 GOTO(out, rc);
1050         }
1051         case LCFG_POOL_NEW:
1052         case LCFG_POOL_ADD:
1053         case LCFG_POOL_DEL:
1054         case LCFG_POOL_REM:
1055                 GOTO(out, rc);
1056
1057         default: {
1058                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1059                 GOTO(out, rc = -EINVAL);
1060
1061         }
1062         }
1063 out:
1064         RETURN(rc);
1065 }
1066
1067 #ifndef log2
1068 #define log2(n) ffz(~(n))
1069 #endif
1070
1071 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
1072                              struct lov_stripe_md **ea,
1073                              struct obd_trans_info *oti)
1074 {
1075         struct lov_obd *lov;
1076         struct obdo *tmp_oa;
1077         struct obd_uuid *ost_uuid = NULL;
1078         int rc = 0, i;
1079         ENTRY;
1080
1081         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1082                 src_oa->o_flags == OBD_FL_DELORPHAN);
1083
1084         lov = &export->exp_obd->u.lov;
1085
1086         OBDO_ALLOC(tmp_oa);
1087         if (tmp_oa == NULL)
1088                 RETURN(-ENOMEM);
1089
1090         if (oti->oti_ost_uuid) {
1091                 ost_uuid = oti->oti_ost_uuid;
1092                 CDEBUG(D_HA, "clearing orphans only for %s\n",
1093                        ost_uuid->uuid);
1094         }
1095
1096         obd_getref(export->exp_obd);
1097         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1098                 struct lov_stripe_md obj_md;
1099                 struct lov_stripe_md *obj_mdp = &obj_md;
1100                 struct lov_tgt_desc *tgt;
1101                 int err;
1102
1103                 tgt = lov->lov_tgts[i];
1104                 if (!tgt)
1105                         continue;
1106
1107                 /* if called for a specific target, we don't
1108                    care if it is not active. */
1109                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
1110                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1111                         continue;
1112                 }
1113
1114                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
1115                         continue;
1116
1117                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
1118                        obd_uuid2str(ost_uuid));
1119
1120                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1121
1122                 LASSERT(lov->lov_tgts[i]->ltd_exp);
1123                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1124                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
1125                                  tmp_oa, &obj_mdp, oti);
1126                 if (err) {
1127                         /* This export will be disabled until it is recovered,
1128                            and then orphan recovery will be completed. */
1129                         CERROR("error in orphan recovery on OST idx %d/%d: "
1130                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
1131                         rc = err;
1132                 }
1133
1134                 if (ost_uuid)
1135                         break;
1136         }
1137         obd_putref(export->exp_obd);
1138
1139         OBDO_FREE(tmp_oa);
1140         RETURN(rc);
1141 }
1142
1143 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1144                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
1145 {
1146         struct lov_stripe_md *obj_mdp, *lsm;
1147         struct lov_obd *lov = &exp->exp_obd->u.lov;
1148         unsigned ost_idx;
1149         int rc, i;
1150         ENTRY;
1151
1152         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1153                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1154
1155         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1156         if (obj_mdp == NULL)
1157                 RETURN(-ENOMEM);
1158
1159         ost_idx = src_oa->o_nlink;
1160         lsm = *ea;
1161         if (lsm == NULL)
1162                 GOTO(out, rc = -EINVAL);
1163         if (ost_idx >= lov->desc.ld_tgt_count ||
1164             !lov->lov_tgts[ost_idx])
1165                 GOTO(out, rc = -EINVAL);
1166
1167         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1168                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1169                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1170                                 GOTO(out, rc = -EINVAL);
1171                         break;
1172                 }
1173         }
1174         if (i == lsm->lsm_stripe_count)
1175                 GOTO(out, rc = -EINVAL);
1176
1177         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
1178 out:
1179         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1180         RETURN(rc);
1181 }
1182
1183 /* the LOV expects oa->o_id to be set to the LOV object id */
1184 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
1185                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
1186 {
1187         struct lov_obd *lov;
1188         struct obd_info *oinfo;
1189         struct lov_request_set *set = NULL;
1190         struct lov_request *req;
1191         struct l_wait_info  lwi = { 0 };
1192         int rc = 0;
1193         ENTRY;
1194
1195         LASSERT(ea != NULL);
1196         if (exp == NULL)
1197                 RETURN(-EINVAL);
1198
1199         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1200             src_oa->o_flags == OBD_FL_DELORPHAN) {
1201                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
1202                 RETURN(rc);
1203         }
1204
1205         lov = &exp->exp_obd->u.lov;
1206         if (!lov->desc.ld_active_tgt_count)
1207                 RETURN(-EIO);
1208
1209         OBD_ALLOC_PTR(oinfo);
1210         if (NULL == oinfo)
1211                 RETURN(-ENOMEM);
1212
1213         obd_getref(exp->exp_obd);
1214         /* Recreate a specific object id at the given OST index */
1215         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1216             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1217                  rc = lov_recreate(exp, src_oa, ea, oti);
1218                  GOTO(out, rc);
1219         }
1220
1221         /* issue statfs rpcs if the osfs data is older than qos_maxage - 1s,
1222          * later in alloc_qos(), we will wait for those rpcs to complete if
1223          * the osfs age is older than 2 * qos_maxage */
1224         qos_statfs_update(exp->exp_obd,
1225                           cfs_time_shift_64(-lov->desc.ld_qos_maxage) + HZ, 0);
1226
1227         rc = lov_prep_create_set(exp, oinfo, ea, src_oa, oti, &set);
1228         if (rc)
1229                 GOTO(out, rc);
1230
1231         list_for_each_entry(req, &set->set_list, rq_link) {
1232                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1233                 rc = obd_create_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1234                                       &req->rq_oi, &req->rq_oi.oi_md, oti);
1235         }
1236
1237         /* osc_create have timeout equ obd_timeout/2 so waiting don't be
1238          * longer then this */
1239         l_wait_event(set->set_waitq, lov_finished_set(set), &lwi);
1240
1241         /* we not have ptlrpc set for assign set->interpret and should
1242          * be call interpret function himself. calling from cb_create_update
1243          * not permited because lov_fini_create_set can sleep for long time,
1244          * but we must avoid sleeping in ptlrpcd interpret function. */
1245         rc = lov_fini_create_set(set, ea);
1246 out:
1247         obd_putref(exp->exp_obd);
1248         OBD_FREE_PTR(oinfo);
1249         RETURN(rc);
1250 }
1251
1252 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1253 do {                                                                            \
1254         LASSERT((lsmp) != NULL);                                                \
1255         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1256                  (lsmp)->lsm_magic == LOV_MAGIC_V3 ||                           \
1257                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1258                  (lsmp), (lsmp)->lsm_magic);                                    \
1259 } while (0)
1260
1261 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1262                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1263                        struct obd_export *md_exp)
1264 {
1265         struct lov_request_set *set;
1266         struct obd_info *oinfo;
1267         struct lov_request *req;
1268         struct list_head *pos;
1269         struct lov_obd *lov;
1270         int rc = 0, err = 0;
1271         ENTRY;
1272
1273         ASSERT_LSM_MAGIC(lsm);
1274
1275         if (!exp || !exp->exp_obd)
1276                 RETURN(-ENODEV);
1277
1278         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1279                 LASSERT(oti);
1280                 LASSERT(oti->oti_logcookies);
1281         }
1282
1283         OBD_ALLOC_PTR(oinfo);
1284         if (NULL == oinfo)
1285                 RETURN(-ENOMEM);
1286
1287         lov = &exp->exp_obd->u.lov;
1288         obd_getref(exp->exp_obd);
1289         rc = lov_prep_destroy_set(exp, oinfo, oa, lsm, oti, &set);
1290         if (rc)
1291                 GOTO(out, rc);
1292
1293         list_for_each (pos, &set->set_list) {
1294                 req = list_entry(pos, struct lov_request, rq_link);
1295
1296                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1297                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1298
1299                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1300                                   req->rq_oi.oi_oa, NULL, oti, NULL);
1301                 err = lov_update_common_set(set, req, err);
1302                 if (err) {
1303                         CERROR("error: destroying objid "LPX64" subobj "
1304                                LPX64" on OST idx %d: rc = %d\n",
1305                                oa->o_id, req->rq_oi.oi_oa->o_id,
1306                                req->rq_idx, err);
1307                         if (!rc)
1308                                 rc = err;
1309                 }
1310         }
1311
1312         if (rc == 0) {
1313                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1314                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1315         }
1316         err = lov_fini_destroy_set(set);
1317 out:
1318         obd_putref(exp->exp_obd);
1319         OBD_FREE_PTR(oinfo);
1320         RETURN(rc ? rc : err);
1321 }
1322
1323 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1324 {
1325         struct lov_request_set *set;
1326         struct lov_request *req;
1327         struct list_head *pos;
1328         struct lov_obd *lov;
1329         int err = 0, rc = 0;
1330         ENTRY;
1331
1332         LASSERT(oinfo);
1333         ASSERT_LSM_MAGIC(oinfo->oi_md);
1334
1335         if (!exp || !exp->exp_obd)
1336                 RETURN(-ENODEV);
1337
1338         lov = &exp->exp_obd->u.lov;
1339
1340         rc = lov_prep_getattr_set(exp, oinfo, &set);
1341         if (rc)
1342                 RETURN(rc);
1343
1344         list_for_each (pos, &set->set_list) {
1345                 req = list_entry(pos, struct lov_request, rq_link);
1346
1347                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1348                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1349                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1350
1351                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1352                                  &req->rq_oi);
1353                 err = lov_update_common_set(set, req, rc);
1354                 if (err) {
1355                         CERROR("error: getattr objid "LPX64" subobj "
1356                                LPX64" on OST idx %d: rc = %d\n",
1357                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1358                                req->rq_idx, err);
1359                         break;
1360                 }
1361         }
1362
1363         rc = lov_fini_getattr_set(set);
1364         if (err)
1365                 rc = err;
1366         RETURN(rc);
1367 }
1368
1369 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1370                                  void *data, int rc)
1371 {
1372         struct lov_request_set *lovset = (struct lov_request_set *)data;
1373         int err;
1374         ENTRY;
1375
1376         /* don't do attribute merge if this aysnc op failed */
1377         if (rc)
1378                 lovset->set_completes = 0;
1379         err = lov_fini_getattr_set(lovset);
1380         RETURN(rc ? rc : err);
1381 }
1382
1383 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1384                               struct ptlrpc_request_set *rqset)
1385 {
1386         struct lov_request_set *lovset;
1387         struct lov_obd *lov;
1388         struct list_head *pos;
1389         struct lov_request *req;
1390         int rc = 0, err;
1391         ENTRY;
1392
1393         LASSERT(oinfo);
1394         ASSERT_LSM_MAGIC(oinfo->oi_md);
1395
1396         if (!exp || !exp->exp_obd)
1397                 RETURN(-ENODEV);
1398
1399         lov = &exp->exp_obd->u.lov;
1400
1401         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1402         if (rc)
1403                 RETURN(rc);
1404
1405         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1406                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1407                oinfo->oi_md->lsm_stripe_size);
1408
1409         list_for_each (pos, &lovset->set_list) {
1410                 req = list_entry(pos, struct lov_request, rq_link);
1411
1412                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1413                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1414                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1415                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1416                                        &req->rq_oi, rqset);
1417                 if (rc) {
1418                         CERROR("error: getattr objid "LPX64" subobj "
1419                                LPX64" on OST idx %d: rc = %d\n",
1420                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1421                                req->rq_idx, rc);
1422                         GOTO(out, rc);
1423                 }
1424         }
1425
1426         if (!list_empty(&rqset->set_requests)) {
1427                 LASSERT(rc == 0);
1428                 LASSERT (rqset->set_interpret == NULL);
1429                 rqset->set_interpret = lov_getattr_interpret;
1430                 rqset->set_arg = (void *)lovset;
1431                 RETURN(rc);
1432         }
1433 out:
1434         if (rc)
1435                 lovset->set_completes = 0;
1436         err = lov_fini_getattr_set(lovset);
1437         RETURN(rc ? rc : err);
1438 }
1439
1440 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1441                        struct obd_trans_info *oti)
1442 {
1443         struct lov_request_set *set;
1444         struct lov_obd *lov;
1445         struct list_head *pos;
1446         struct lov_request *req;
1447         int err = 0, rc = 0;
1448         ENTRY;
1449
1450         LASSERT(oinfo);
1451         ASSERT_LSM_MAGIC(oinfo->oi_md);
1452
1453         if (!exp || !exp->exp_obd)
1454                 RETURN(-ENODEV);
1455
1456         /* for now, we only expect the following updates here */
1457         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1458                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1459                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1460                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1461                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1462                                             OBD_MD_FLGID | OBD_MD_FLFID |
1463                                             OBD_MD_FLGENER)));
1464         lov = &exp->exp_obd->u.lov;
1465         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1466         if (rc)
1467                 RETURN(rc);
1468
1469         list_for_each (pos, &set->set_list) {
1470                 req = list_entry(pos, struct lov_request, rq_link);
1471
1472                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1473                                  &req->rq_oi, NULL);
1474                 err = lov_update_setattr_set(set, req, rc);
1475                 if (err) {
1476                         CERROR("error: setattr objid "LPX64" subobj "
1477                                LPX64" on OST idx %d: rc = %d\n",
1478                                set->set_oi->oi_oa->o_id,
1479                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1480                         if (!rc)
1481                                 rc = err;
1482                 }
1483         }
1484         err = lov_fini_setattr_set(set);
1485         if (!rc)
1486                 rc = err;
1487         RETURN(rc);
1488 }
1489
1490 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1491                                  void *data, int rc)
1492 {
1493         struct lov_request_set *lovset = (struct lov_request_set *)data;
1494         int err;
1495         ENTRY;
1496
1497         if (rc)
1498                 lovset->set_completes = 0;
1499         err = lov_fini_setattr_set(lovset);
1500         RETURN(rc ? rc : err);
1501 }
1502
1503 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1504    needed. Otherwise, a client is waiting for responses. */
1505 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1506                              struct obd_trans_info *oti,
1507                              struct ptlrpc_request_set *rqset)
1508 {
1509         struct lov_request_set *set;
1510         struct lov_request *req;
1511         struct list_head *pos;
1512         struct lov_obd *lov;
1513         int rc = 0;
1514         ENTRY;
1515
1516         LASSERT(oinfo);
1517         ASSERT_LSM_MAGIC(oinfo->oi_md);
1518         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1519                 LASSERT(oti);
1520                 LASSERT(oti->oti_logcookies);
1521         }
1522
1523         if (!exp || !exp->exp_obd)
1524                 RETURN(-ENODEV);
1525
1526         lov = &exp->exp_obd->u.lov;
1527         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1528         if (rc)
1529                 RETURN(rc);
1530
1531         CDEBUG(D_INFO, "objid "LPX64"@"LPX64": %ux%u byte stripes\n",
1532                oinfo->oi_md->lsm_object_id,
1533                oinfo->oi_md->lsm_object_gr,
1534                oinfo->oi_md->lsm_stripe_count,
1535                oinfo->oi_md->lsm_stripe_size);
1536
1537         list_for_each (pos, &set->set_list) {
1538                 req = list_entry(pos, struct lov_request, rq_link);
1539
1540                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1541                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1542
1543                 CDEBUG(D_INFO, "objid "LPX64"@"LPX64"[%d] has subobj "LPX64
1544                        " at idx %u\n", oinfo->oi_oa->o_id, oinfo->oi_oa->o_gr,
1545                        req->rq_stripe, req->rq_oi.oi_oa->o_id, req->rq_idx);
1546
1547                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1548                                        &req->rq_oi, oti, rqset);
1549                 if (rc) {
1550                         CERROR("error: setattr objid "LPX64" subobj "
1551                                LPX64" on OST idx %d: rc = %d\n",
1552                                set->set_oi->oi_oa->o_id,
1553                                req->rq_oi.oi_oa->o_id,
1554                                req->rq_idx, rc);
1555                         break;
1556                 }
1557         }
1558
1559         /* If we are not waiting for responses on async requests, return. */
1560         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1561                 int err;
1562                 if (rc)
1563                         set->set_completes = 0;
1564                 err = lov_fini_setattr_set(set);
1565                 RETURN(rc ? rc : err);
1566         }
1567
1568         LASSERT(rqset->set_interpret == NULL);
1569         rqset->set_interpret = lov_setattr_interpret;
1570         rqset->set_arg = (void *)set;
1571
1572         RETURN(0);
1573 }
1574
1575 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1576                                void *data, int rc)
1577 {
1578         struct lov_request_set *lovset = (struct lov_request_set *)data;
1579         int err;
1580         ENTRY;
1581
1582         if (rc)
1583                 lovset->set_completes = 0;
1584         err = lov_fini_punch_set(lovset);
1585         RETURN(rc ? rc : err);
1586 }
1587
1588 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1589  * we can send this 'punch' to just the authoritative node and the nodes
1590  * that the punch will affect. */
1591 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1592                      struct obd_trans_info *oti,
1593                      struct ptlrpc_request_set *rqset)
1594 {
1595         struct lov_request_set *set;
1596         struct lov_obd *lov;
1597         struct list_head *pos;
1598         struct lov_request *req;
1599         int rc = 0;
1600         ENTRY;
1601
1602         LASSERT(oinfo);
1603         ASSERT_LSM_MAGIC(oinfo->oi_md);
1604
1605         if (!exp || !exp->exp_obd)
1606                 RETURN(-ENODEV);
1607
1608         lov = &exp->exp_obd->u.lov;
1609         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1610         if (rc)
1611                 RETURN(rc);
1612
1613         list_for_each (pos, &set->set_list) {
1614                 req = list_entry(pos, struct lov_request, rq_link);
1615
1616                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1617                                &req->rq_oi, NULL, rqset);
1618                 if (rc) {
1619                         CERROR("error: punch objid "LPX64" subobj "LPX64
1620                                " on OST idx %d: rc = %d\n",
1621                                set->set_oi->oi_oa->o_id,
1622                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1623                         break;
1624                 }
1625         }
1626
1627         if (rc || list_empty(&rqset->set_requests)) {
1628                 int err;
1629                 err = lov_fini_punch_set(set);
1630                 RETURN(rc ? rc : err);
1631         }
1632
1633         LASSERT(rqset->set_interpret == NULL);
1634         rqset->set_interpret = lov_punch_interpret;
1635         rqset->set_arg = (void *)set;
1636
1637         RETURN(0);
1638 }
1639
1640 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1641                               void *data, int rc)
1642 {
1643         struct lov_request_set *lovset = (struct lov_request_set *)data;
1644         int err;
1645         ENTRY;
1646
1647         if (rc)
1648                 lovset->set_completes = 0;
1649         err = lov_fini_sync_set(lovset);
1650         RETURN(rc ? rc : err);
1651 }
1652
1653 static int lov_sync(struct obd_export *exp, struct obd_info *oinfo,
1654                     obd_off start, obd_off end,
1655                     struct ptlrpc_request_set *rqset)
1656 {
1657         struct lov_request_set *set = NULL;
1658         struct lov_obd *lov;
1659         struct list_head *pos;
1660         struct lov_request *req;
1661         int    rc = 0;
1662         ENTRY;
1663
1664         ASSERT_LSM_MAGIC(oinfo->oi_md);
1665         LASSERT(rqset != NULL);
1666
1667         if (!exp->exp_obd)
1668                 RETURN(-ENODEV);
1669
1670         lov = &exp->exp_obd->u.lov;
1671         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1672         if (rc)
1673                 RETURN(rc);
1674
1675         list_for_each (pos, &set->set_list) {
1676                 req = list_entry(pos, struct lov_request, rq_link);
1677
1678                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp, &req->rq_oi,
1679                               req->rq_oi.oi_policy.l_extent.start,
1680                               req->rq_oi.oi_policy.l_extent.end, rqset);
1681                 if (rc) {
1682                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1683                                " on OST idx %d: rc = %d\n",
1684                                set->set_oi->oi_oa->o_id,
1685                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1686                         break;
1687                 }
1688         }
1689
1690         /* If we are not waiting for responses on async requests, return. */
1691         if (rc || list_empty(&rqset->set_requests)) {
1692                 int err = lov_fini_sync_set(set);
1693                 RETURN(rc ? rc : err);
1694         }
1695
1696         LASSERT(rqset->set_interpret == NULL);
1697         rqset->set_interpret = lov_sync_interpret;
1698         rqset->set_arg = (void *)set;
1699
1700         RETURN(0);
1701 }
1702
1703 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1704                          obd_count oa_bufs, struct brw_page *pga)
1705 {
1706         struct obd_info oinfo = { { { 0 } } };
1707         int i, rc = 0;
1708
1709         oinfo.oi_oa = lov_oinfo->oi_oa;
1710
1711         /* The caller just wants to know if there's a chance that this
1712          * I/O can succeed */
1713         for (i = 0; i < oa_bufs; i++) {
1714                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1715                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1716                 obd_off start, end;
1717
1718                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1719                                            pga[i].off + pga[i].count - 1,
1720                                            &start, &end))
1721                         continue;
1722
1723                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1724                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1725                         return -EIO;
1726                 }
1727
1728                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1729                              1, &pga[i], NULL);
1730                 if (rc)
1731                         break;
1732         }
1733         return rc;
1734 }
1735
1736 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1737                    obd_count oa_bufs, struct brw_page *pga,
1738                    struct obd_trans_info *oti)
1739 {
1740         struct lov_request_set *set;
1741         struct lov_request *req;
1742         struct list_head *pos;
1743         struct lov_obd *lov = &exp->exp_obd->u.lov;
1744         int err, rc = 0;
1745         ENTRY;
1746
1747         ASSERT_LSM_MAGIC(oinfo->oi_md);
1748
1749         if (cmd == OBD_BRW_CHECK) {
1750                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1751                 RETURN(rc);
1752         }
1753
1754         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1755         if (rc)
1756                 RETURN(rc);
1757
1758         list_for_each (pos, &set->set_list) {
1759                 struct obd_export *sub_exp;
1760                 struct brw_page *sub_pga;
1761                 req = list_entry(pos, struct lov_request, rq_link);
1762
1763                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1764                 sub_pga = set->set_pga + req->rq_pgaidx;
1765                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1766                              sub_pga, oti);
1767                 if (rc)
1768                         break;
1769                 lov_update_common_set(set, req, rc);
1770         }
1771
1772         err = lov_fini_brw_set(set);
1773         if (!rc)
1774                 rc = err;
1775         RETURN(rc);
1776 }
1777
1778 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1779                              int rc)
1780 {
1781         struct lov_request_set *lovset = (struct lov_request_set *)data;
1782         ENTRY;
1783
1784         if (rc) {
1785                 lovset->set_completes = 0;
1786                 lov_fini_brw_set(lovset);
1787         } else {
1788                 rc = lov_fini_brw_set(lovset);
1789         }
1790
1791         RETURN(rc);
1792 }
1793
1794 static int lov_brw_async(int cmd, struct obd_export *exp,
1795                          struct obd_info *oinfo, obd_count oa_bufs,
1796                          struct brw_page *pga, struct obd_trans_info *oti,
1797                          struct ptlrpc_request_set *set, int pshift)
1798 {
1799         struct lov_request_set *lovset;
1800         struct lov_request *req;
1801         struct list_head *pos;
1802         struct lov_obd *lov = &exp->exp_obd->u.lov;
1803         int rc = 0;
1804         ENTRY;
1805
1806         LASSERT(oinfo);
1807         ASSERT_LSM_MAGIC(oinfo->oi_md);
1808
1809         if (cmd == OBD_BRW_CHECK) {
1810                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1811                 RETURN(rc);
1812         }
1813
1814         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1815         if (rc)
1816                 RETURN(rc);
1817
1818         list_for_each (pos, &lovset->set_list) {
1819                 struct obd_export *sub_exp;
1820                 struct brw_page *sub_pga;
1821                 req = list_entry(pos, struct lov_request, rq_link);
1822
1823                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1824                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1825                 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1826                                    sub_pga, oti, set, pshift);
1827                 if (rc)
1828                         GOTO(out, rc);
1829                 lov_update_common_set(lovset, req, rc);
1830         }
1831         LASSERT(rc == 0);
1832         LASSERT(set->set_interpret == NULL);
1833         LASSERT(set->set_arg == NULL);
1834         rc = ptlrpc_set_add_cb(set, lov_brw_interpret, lovset);
1835         if (rc)
1836                 GOTO(out, rc);
1837
1838         RETURN(rc);
1839 out:
1840         lov_fini_brw_set(lovset);
1841         RETURN(rc);
1842 }
1843
1844 static int lov_ap_make_ready(void *data, int cmd)
1845 {
1846         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1847
1848         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1849 }
1850
1851 static int lov_ap_refresh_count(void *data, int cmd)
1852 {
1853         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1854
1855         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1856                                                      cmd);
1857 }
1858
1859 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1860 {
1861         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1862
1863         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1864         /* XXX woah, shouldn't we be altering more here?  size? */
1865         oa->o_id = lap->lap_loi_id;
1866         oa->o_stripe_idx = lap->lap_stripe;
1867 }
1868
1869 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1870                                obd_valid valid)
1871 {
1872         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1873
1874         lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1875 }
1876
1877 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1878 {
1879         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1880
1881         /* in a raid1 regime this would down a count of many ios
1882          * in flight, onl calling the caller_ops completion when all
1883          * the raid1 ios are complete */
1884         rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1885         return rc;
1886 }
1887
1888 static struct obd_async_page_ops lov_async_page_ops = {
1889         .ap_make_ready =        lov_ap_make_ready,
1890         .ap_refresh_count =     lov_ap_refresh_count,
1891         .ap_fill_obdo =         lov_ap_fill_obdo,
1892         .ap_update_obdo =       lov_ap_update_obdo,
1893         .ap_completion =        lov_ap_completion,
1894 };
1895
1896 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1897                         struct lov_oinfo *loi, cfs_page_t *page,
1898                         obd_off offset, struct obd_async_page_ops *ops,
1899                         void *data, void **res, int flags,
1900                         struct lustre_handle *lockh)
1901 {
1902         struct lov_obd *lov = &exp->exp_obd->u.lov;
1903         struct lov_async_page *lap;
1904         struct lov_lock_handles *lov_lockh = NULL;
1905         int rc = 0;
1906         ENTRY;
1907
1908         if (!page) {
1909                 int i = 0;
1910                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1911                    Only because of this layering limitation will a client
1912                    mount with no osts fail */
1913                 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1914                        !lov->lov_tgts[i]->ltd_exp) {
1915                         i++;
1916                         if (i >= lov->desc.ld_tgt_count)
1917                                 RETURN(-ENOMEDIUM);
1918                 }
1919                 rc = size_round(sizeof(*lap)) +
1920                         obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1921                                             NULL, NULL, 0, NULL, NULL, NULL, 0,
1922                                             NULL);
1923                 RETURN(rc);
1924         }
1925         ASSERT_LSM_MAGIC(lsm);
1926         LASSERT(loi == NULL);
1927
1928         lap = *res;
1929         lap->lap_magic = LOV_AP_MAGIC;
1930         lap->lap_caller_ops = ops;
1931         lap->lap_caller_data = data;
1932
1933         /* for now only raid 0 which passes through */
1934         lap->lap_stripe = lov_stripe_number(lsm, offset);
1935         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1936         loi = lsm->lsm_oinfo[lap->lap_stripe];
1937
1938         /* so the callback doesn't need the lsm */
1939         lap->lap_loi_id = loi->loi_id;
1940
1941         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1942
1943         if (lockh && !(flags & OBD_FAST_LOCK)) {
1944                 lov_lockh = lov_handle2llh(lockh);
1945                 if (lov_lockh) {
1946                         lockh = lov_lockh->llh_handles + lap->lap_stripe;
1947                 }
1948         }
1949
1950         rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1951                                  lsm, loi, page, lap->lap_sub_offset,
1952                                  &lov_async_page_ops, lap,
1953                                  &lap->lap_sub_cookie, flags, lockh);
1954         if (lov_lockh)
1955                 lov_llh_put(lov_lockh);
1956         if (rc)
1957                 RETURN(rc);
1958         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1959                lap->lap_sub_cookie, offset);
1960         RETURN(0);
1961 }
1962
1963 static int lov_queue_async_io(struct obd_export *exp,
1964                               struct lov_stripe_md *lsm,
1965                               struct lov_oinfo *loi, void *cookie,
1966                               int cmd, obd_off off, int count,
1967                               obd_flag brw_flags, obd_flag async_flags)
1968 {
1969         struct lov_obd *lov = &exp->exp_obd->u.lov;
1970         struct lov_async_page *lap;
1971         int rc;
1972
1973         LASSERT(loi == NULL);
1974
1975         ASSERT_LSM_MAGIC(lsm);
1976
1977         lap = LAP_FROM_COOKIE(cookie);
1978
1979         loi = lsm->lsm_oinfo[lap->lap_stripe];
1980
1981         rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1982                                 loi, lap->lap_sub_cookie, cmd, off, count,
1983                                 brw_flags, async_flags);
1984         RETURN(rc);
1985 }
1986
1987 static int lov_set_async_flags(struct obd_export *exp,
1988                                struct lov_stripe_md *lsm,
1989                                struct lov_oinfo *loi, void *cookie,
1990                                obd_flag async_flags)
1991 {
1992         struct lov_obd *lov = &exp->exp_obd->u.lov;
1993         struct lov_async_page *lap;
1994         int rc;
1995
1996         LASSERT(loi == NULL);
1997
1998         ASSERT_LSM_MAGIC(lsm);
1999
2000         lap = LAP_FROM_COOKIE(cookie);
2001
2002         loi = lsm->lsm_oinfo[lap->lap_stripe];
2003
2004         rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2005                                  lsm, loi, lap->lap_sub_cookie, async_flags);
2006         RETURN(rc);
2007 }
2008
2009 static int lov_queue_group_io(struct obd_export *exp,
2010                               struct lov_stripe_md *lsm,
2011                               struct lov_oinfo *loi,
2012                               struct obd_io_group *oig, void *cookie,
2013                               int cmd, obd_off off, int count,
2014                               obd_flag brw_flags, obd_flag async_flags)
2015 {
2016         struct lov_obd *lov = &exp->exp_obd->u.lov;
2017         struct lov_async_page *lap;
2018         int rc;
2019
2020         LASSERT(loi == NULL);
2021
2022         ASSERT_LSM_MAGIC(lsm);
2023
2024         lap = LAP_FROM_COOKIE(cookie);
2025
2026         loi = lsm->lsm_oinfo[lap->lap_stripe];
2027
2028         rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
2029                                 loi, oig, lap->lap_sub_cookie, cmd, off, count,
2030                                 brw_flags, async_flags);
2031         RETURN(rc);
2032 }
2033
2034 /* this isn't exactly optimal.  we may have queued sync io in oscs on
2035  * all stripes, but we don't record that fact at queue time.  so we
2036  * trigger sync io on all stripes. */
2037 static int lov_trigger_group_io(struct obd_export *exp,
2038                                 struct lov_stripe_md *lsm,
2039                                 struct lov_oinfo *loi,
2040                                 struct obd_io_group *oig)
2041 {
2042         struct lov_obd *lov = &exp->exp_obd->u.lov;
2043         int rc = 0, i, err;
2044
2045         LASSERT(loi == NULL);
2046
2047         ASSERT_LSM_MAGIC(lsm);
2048
2049         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2050                 loi = lsm->lsm_oinfo[i];
2051                 if (!lov->lov_tgts[loi->loi_ost_idx] ||
2052                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2053                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2054                         continue;
2055                 }
2056
2057                 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2058                                            lsm, loi, oig);
2059                 if (rc == 0 && err != 0)
2060                         rc = err;
2061         };
2062         RETURN(rc);
2063 }
2064
2065 static int lov_teardown_async_page(struct obd_export *exp,
2066                                    struct lov_stripe_md *lsm,
2067                                    struct lov_oinfo *loi, void *cookie)
2068 {
2069         struct lov_obd *lov = &exp->exp_obd->u.lov;
2070         struct lov_async_page *lap;
2071         int rc;
2072
2073         LASSERT(loi == NULL);
2074
2075         ASSERT_LSM_MAGIC(lsm);
2076
2077         lap = LAP_FROM_COOKIE(cookie);
2078
2079         loi = lsm->lsm_oinfo[lap->lap_stripe];
2080
2081         rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2082                                      lsm, loi, lap->lap_sub_cookie);
2083         if (rc) {
2084                 CERROR("unable to teardown sub cookie %p: %d\n",
2085                        lap->lap_sub_cookie, rc);
2086                 RETURN(rc);
2087         }
2088         RETURN(rc);
2089 }
2090
2091 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
2092                                  void *data, int rc)
2093 {
2094         struct lov_request_set *lovset = (struct lov_request_set *)data;
2095         ENTRY;
2096         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
2097         RETURN(rc);
2098 }
2099
2100 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
2101                        struct ldlm_enqueue_info *einfo,
2102                        struct ptlrpc_request_set *rqset)
2103 {
2104         ldlm_mode_t mode = einfo->ei_mode;
2105         struct lov_request_set *set;
2106         struct lov_request *req;
2107         struct list_head *pos;
2108         struct lov_obd *lov;
2109         ldlm_error_t rc;
2110         ENTRY;
2111
2112         LASSERT(oinfo);
2113         ASSERT_LSM_MAGIC(oinfo->oi_md);
2114         LASSERT(mode == (mode & -mode));
2115
2116         /* we should never be asked to replay a lock this way. */
2117         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
2118
2119         if (!exp || !exp->exp_obd)
2120                 RETURN(-ENODEV);
2121
2122         lov = &exp->exp_obd->u.lov;
2123         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
2124         if (rc)
2125                 RETURN(rc);
2126
2127         list_for_each (pos, &set->set_list) {
2128                 req = list_entry(pos, struct lov_request, rq_link);
2129
2130                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
2131                                  &req->rq_oi, einfo, rqset);
2132                 if (rc != ELDLM_OK)
2133                         GOTO(out, rc);
2134         }
2135
2136         if (rqset && !list_empty(&rqset->set_requests)) {
2137                 LASSERT(rc == 0);
2138                 LASSERT(rqset->set_interpret == NULL);
2139                 rqset->set_interpret = lov_enqueue_interpret;
2140                 rqset->set_arg = (void *)set;
2141                 RETURN(rc);
2142         }
2143 out:
2144         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
2145         RETURN(rc);
2146 }
2147
2148 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
2149                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
2150                      int *flags, void *data, struct lustre_handle *lockh,
2151                      int *n_matches)
2152 {
2153         struct lov_request_set *set;
2154         struct obd_info oinfo;
2155         struct lov_request *req;
2156         struct list_head *pos;
2157         struct lov_obd *lov;
2158         struct lustre_handle *lov_lockhp;
2159         int lov_flags, rc = 0;
2160         ENTRY;
2161
2162         ASSERT_LSM_MAGIC(lsm);
2163         LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
2164
2165         if (!exp || !exp->exp_obd)
2166                 RETURN(-ENODEV);
2167
2168         lov = &exp->exp_obd->u.lov;
2169         rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
2170         if (rc)
2171                 RETURN(rc);
2172
2173         list_for_each (pos, &set->set_list) {
2174                 ldlm_policy_data_t sub_policy;
2175                 req = list_entry(pos, struct lov_request, rq_link);
2176                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2177                 LASSERT(lov_lockhp);
2178
2179                 lov_flags = *flags;
2180                 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
2181
2182                 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
2183                                req->rq_oi.oi_md, type, &sub_policy,
2184                                mode, &lov_flags, data, lov_lockhp,
2185                                n_matches);
2186                 rc = lov_update_match_set(set, req, rc);
2187                 if (rc <= 0)
2188                         break;
2189         }
2190         lov_fini_match_set(set, mode, *flags);
2191         RETURN(rc);
2192 }
2193
2194 static int lov_change_cbdata(struct obd_export *exp,
2195                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
2196                              void *data)
2197 {
2198         struct lov_obd *lov;
2199         struct lov_oinfo *loi;
2200         int rc = 0, i;
2201         ENTRY;
2202
2203         ASSERT_LSM_MAGIC(lsm);
2204
2205         if (!exp || !exp->exp_obd)
2206                 RETURN(-ENODEV);
2207
2208         lov = &exp->exp_obd->u.lov;
2209         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2210                 struct lov_stripe_md submd;
2211
2212                 loi = lsm->lsm_oinfo[i];
2213                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2214                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
2215                         continue;
2216                 }
2217                 submd.lsm_object_id = loi->loi_id;
2218                 submd.lsm_object_gr = loi->loi_gr;
2219                 submd.lsm_stripe_count = 0;
2220                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2221                                        &submd, it, data);
2222         }
2223         RETURN(rc);
2224 }
2225
2226 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
2227                       __u32 mode, struct lustre_handle *lockh, int flags,
2228                       obd_off end)
2229 {
2230         struct lov_request_set *set;
2231         struct obd_info oinfo;
2232         struct lov_request *req;
2233         struct list_head *pos;
2234         struct lov_obd *lov;
2235         struct lustre_handle *lov_lockhp;
2236         ldlm_mode_t this_mode;
2237         int err = 0, rc = 0;
2238         ENTRY;
2239
2240         ASSERT_LSM_MAGIC(lsm);
2241
2242         if (!exp || !exp->exp_obd)
2243                 RETURN(-ENODEV);
2244
2245         LASSERT(lockh);
2246         lov = &exp->exp_obd->u.lov;
2247         if (flags & OBD_FAST_LOCK) {
2248                 int stripe = lov_stripe_number(lsm, end);
2249                 RETURN(obd_cancel(lov->lov_tgts[lsm->lsm_oinfo[stripe]->
2250                                   loi_ost_idx]->ltd_exp, NULL, mode, lockh,
2251                                   flags, end));
2252         }
2253
2254         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
2255         if (rc)
2256                 RETURN(rc);
2257
2258         list_for_each (pos, &set->set_list) {
2259                 req = list_entry(pos, struct lov_request, rq_link);
2260                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
2261
2262                 /* If this lock was used for a write or truncate, the object
2263                  * will have been recreated by the OST, cancel the lock
2264                  * (setting LCK_GROUP incidentally causes immediate cancel). */
2265                 if (OST_LVB_IS_ERR(lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_blocks) &&
2266                     (mode == LCK_PW || mode == LCK_CW))
2267                         this_mode = LCK_GROUP;
2268                 else
2269                         this_mode = mode;
2270
2271                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2272                                 req->rq_oi.oi_md, this_mode, lov_lockhp, flags,
2273                                 end);
2274                 rc = lov_update_common_set(set, req, rc);
2275                 if (rc) {
2276                         CERROR("error: cancel objid "LPX64" subobj "
2277                                LPX64" on OST idx %d: rc = %d\n",
2278                                lsm->lsm_object_id,
2279                                req->rq_oi.oi_md->lsm_object_id,
2280                                req->rq_idx, rc);
2281                         err = rc;
2282                 }
2283
2284         }
2285         lov_fini_cancel_set(set);
2286         RETURN(err);
2287 }
2288
2289 static int lov_cancel_unused(struct obd_export *exp,
2290                              struct lov_stripe_md *lsm, int flags, void *opaque)
2291 {
2292         struct lov_obd *lov;
2293         struct lov_oinfo *loi;
2294         int rc = 0, i;
2295         ENTRY;
2296
2297         if (!exp || !exp->exp_obd)
2298                 RETURN(-ENODEV);
2299
2300         lov = &exp->exp_obd->u.lov;
2301         if (lsm == NULL) {
2302                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2303                         int err;
2304                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2305                                 continue;
2306
2307                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2308                                                 flags, opaque);
2309                         if (!rc)
2310                                 rc = err;
2311                 }
2312                 RETURN(rc);
2313         }
2314
2315         ASSERT_LSM_MAGIC(lsm);
2316
2317         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2318                 struct lov_stripe_md submd;
2319                 int err;
2320
2321                 loi = lsm->lsm_oinfo[i];
2322                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2323                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2324                         continue;
2325                 }
2326
2327                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2328                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2329
2330                 submd.lsm_object_id = loi->loi_id;
2331                 submd.lsm_stripe_count = 0;
2332                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2333                                         &submd, flags, opaque);
2334                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2335                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2336                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2337                                loi->loi_id, loi->loi_ost_idx, err);
2338                         if (!rc)
2339                                 rc = err;
2340                 }
2341         }
2342         RETURN(rc);
2343 }
2344
2345 static int lov_join_lru(struct obd_export *exp,
2346                         struct lov_stripe_md *lsm, int join)
2347 {
2348         struct lov_obd *lov;
2349         struct lov_oinfo *loi;
2350         int i, count = 0;
2351         ENTRY;
2352
2353         ASSERT_LSM_MAGIC(lsm);
2354         if (!exp || !exp->exp_obd)
2355                 RETURN(-ENODEV);
2356
2357         lov = &exp->exp_obd->u.lov;
2358         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2359                 struct lov_stripe_md submd;
2360                 int rc = 0;
2361
2362                 loi = lsm->lsm_oinfo[i];
2363                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2364                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2365                         continue;
2366                 }
2367
2368                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2369                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2370
2371                 submd.lsm_object_id = loi->loi_id;
2372                 submd.lsm_stripe_count = 0;
2373                 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2374                                   &submd, join);
2375                 if (rc < 0) {
2376                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2377                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2378                                loi->loi_id, loi->loi_ost_idx, rc);
2379                         return rc;
2380                 } else {
2381                         count += rc;
2382                 }
2383         }
2384         RETURN(count);
2385 }
2386
2387 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
2388 {
2389         struct lov_request_set *lovset = (struct lov_request_set *)data;
2390         int err;
2391         ENTRY;
2392
2393         if (rc)
2394                 lovset->set_completes = 0;
2395
2396         err = lov_fini_statfs_set(lovset);
2397         RETURN(rc ? rc : err);
2398 }
2399
2400 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2401                             __u64 max_age, struct ptlrpc_request_set *rqset)
2402 {
2403         struct lov_request_set *set;
2404         struct lov_request *req;
2405         struct list_head *pos;
2406         struct lov_obd *lov;
2407         int rc = 0;
2408         ENTRY;
2409
2410         LASSERT(oinfo != NULL);
2411         LASSERT(oinfo->oi_osfs != NULL);
2412
2413         lov = &obd->u.lov;
2414         rc = lov_prep_statfs_set(obd, oinfo, &set);
2415         if (rc)
2416                 RETURN(rc);
2417
2418         list_for_each (pos, &set->set_list) {
2419                 struct obd_device *osc_obd;
2420
2421                 req = list_entry(pos, struct lov_request, rq_link);
2422
2423                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2424                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2425                 if (rc)
2426                         break;
2427         }
2428
2429         if (rc || list_empty(&rqset->set_requests)) {
2430                 int err;
2431                 if (rc)
2432                         set->set_completes = 0;
2433                 err = lov_fini_statfs_set(set);
2434                 RETURN(rc ? rc : err);
2435         }
2436
2437         LASSERT(rqset->set_interpret == NULL);
2438         rqset->set_interpret = lov_statfs_interpret;
2439         rqset->set_arg = (void *)set;
2440         RETURN(0);
2441 }
2442
2443 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2444                       __u64 max_age, __u32 flags)
2445 {
2446         struct ptlrpc_request_set *set = NULL;
2447         struct obd_info oinfo = { { { 0 } } };
2448         int rc = 0;
2449         ENTRY;
2450
2451         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
2452          * statfs requests */
2453         set = ptlrpc_prep_set();
2454         if (set == NULL)
2455                 RETURN(-ENOMEM);
2456
2457         oinfo.oi_osfs = osfs;
2458         oinfo.oi_flags = flags;
2459         rc = lov_statfs_async(obd, &oinfo, max_age, set);
2460         if (rc == 0)
2461                 rc = ptlrpc_set_wait(set);
2462         ptlrpc_set_destroy(set);
2463
2464         RETURN(rc);
2465 }
2466
2467 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2468                          void *karg, void *uarg)
2469 {
2470         struct obd_device *obddev = class_exp2obd(exp);
2471         struct lov_obd *lov = &obddev->u.lov;
2472         int i, rc, count = lov->desc.ld_tgt_count;
2473         struct obd_uuid *uuidp;
2474         ENTRY;
2475
2476         switch (cmd) {
2477         case OBD_IOC_LOV_GET_CONFIG: {
2478                 struct obd_ioctl_data *data;
2479                 struct lov_desc *desc;
2480                 char *buf = NULL;
2481                 __u32 *genp;
2482
2483                 len = 0;
2484                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2485                         RETURN(-EINVAL);
2486
2487                 data = (struct obd_ioctl_data *)buf;
2488
2489                 if (sizeof(*desc) > data->ioc_inllen1) {
2490                         obd_ioctl_freedata(buf, len);
2491                         RETURN(-EINVAL);
2492                 }
2493
2494                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2495                         obd_ioctl_freedata(buf, len);
2496                         RETURN(-EINVAL);
2497                 }
2498
2499                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2500                         obd_ioctl_freedata(buf, len);
2501                         RETURN(-EINVAL);
2502                 }
2503
2504                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2505                 memcpy(desc, &(lov->desc), sizeof(*desc));
2506
2507                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2508                 genp = (__u32 *)data->ioc_inlbuf3;
2509                 /* the uuid will be empty for deleted OSTs */
2510                 for (i = 0; i < count; i++, uuidp++, genp++) {
2511                         if (!lov->lov_tgts[i])
2512                                 continue;
2513                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2514                         *genp = lov->lov_tgts[i]->ltd_gen;
2515                 }
2516
2517                 rc = copy_to_user((void *)uarg, buf, len);
2518                 if (rc)
2519                         rc = -EFAULT;
2520                 obd_ioctl_freedata(buf, len);
2521                 break;
2522         }
2523         case LL_IOC_LOV_SETSTRIPE:
2524                 rc = lov_setstripe(exp, karg, uarg);
2525                 break;
2526         case LL_IOC_LOV_GETSTRIPE:
2527                 rc = lov_getstripe(exp, karg, uarg);
2528                 break;
2529         case LL_IOC_LOV_SETEA:
2530                 rc = lov_setea(exp, karg, uarg);
2531                 break;
2532         default: {
2533                 int set = 0;
2534
2535                 if (count == 0)
2536                         RETURN(-ENOTTY);
2537
2538                 rc = 0;
2539                 for (i = 0; i < count; i++) {
2540                         int err;
2541
2542                         /* OST was disconnected */
2543                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2544                                 continue;
2545
2546                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2547                                             len, karg, uarg);
2548                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2549                                 RETURN(err);
2550                         } else if (err) {
2551                                 if (lov->lov_tgts[i]->ltd_active) {
2552                                         CDEBUG(err == -ENOTTY ?
2553                                                D_IOCTL : D_WARNING,
2554                                                "iocontrol OSC %s on OST "
2555                                                "idx %d cmd %x: err = %d\n",
2556                                                lov_uuid2str(lov, i),
2557                                                i, cmd, err);
2558                                         if (!rc)
2559                                                 rc = err;
2560                                 }
2561                         } else {
2562                                 set = 1;
2563                         }
2564                 }
2565                 if (!set && !rc)
2566                         rc = -EIO;
2567         }
2568         }
2569
2570         RETURN(rc);
2571 }
2572
2573 #define FIEMAP_BUFFER_SIZE 4096
2574
2575 /* Non-zero fe_logical indicates that this is a continuation FIEMAP
2576  * call. The local end offset and the device are sent in the first
2577  * fm_extent. This function calculates the stripe number from the index.
2578  * This function returns a stripe_no on which mapping is to be restarted.
2579  *
2580  * This function returns fm_end_offset which is the in-OST offset at which
2581  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2582  * will re-calculate proper offset in next stripe.
2583  * Note that the first extent is passed to lov_get_info via the value field */
2584 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2585                                    struct lov_stripe_md *lsm, obd_size fm_start,
2586                                    obd_size fm_end, int *start_stripe)
2587 {
2588         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2589         obd_off lun_start, lun_end;
2590         obd_size fm_end_offset;
2591         int stripe_no = -1, i;
2592
2593         if (fiemap->fm_extent_count == 0 ||
2594             fiemap->fm_extents[0].fe_logical == 0)
2595                 return 0;
2596
2597         /* Find out stripe_no from ost_index saved in the fe_device */
2598         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2599                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2600                                         fiemap->fm_extents[0].fe_device) {
2601                         stripe_no = i;
2602                         break;
2603                 }
2604         }
2605
2606         /* If we have finished mapping on previous device, shift logical
2607          * offset to start of next device */
2608         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2609                                    &lun_start, &lun_end)) != 0 &&
2610                                    local_end < lun_end) {
2611                 fm_end_offset = local_end;
2612                 *start_stripe = stripe_no;
2613         } else {
2614                 /* This is a special value to indicate that caller should
2615                  * calculate offset in next stripe. */
2616                 fm_end_offset = 0;
2617                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2618         }
2619
2620         return fm_end_offset;
2621 }
2622
2623 /* We calculate on which OST the mapping will end. If the length of mapping
2624  * is greater than (stripe_size * stripe_count) then the last_stripe will
2625  * will be one just before start_stripe. Else we check if the mapping
2626  * intersects each OST and find last_stripe.
2627  * This function returns the last_stripe and also sets the stripe_count
2628  * over which the mapping is spread */
2629 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2630                             obd_size fm_end, int start_stripe,
2631                             int *stripe_count)
2632 {
2633         int last_stripe;
2634         obd_off obd_start, obd_end;
2635         int i, j;
2636
2637         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2638                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2639                                                               start_stripe - 1);
2640                 *stripe_count = lsm->lsm_stripe_count;
2641         } else {
2642                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2643                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2644                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2645                                                    &obd_start, &obd_end)) == 0)
2646                                 break;
2647                 }
2648                 *stripe_count = j;
2649                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2650         }
2651
2652         return last_stripe;
2653 }
2654
2655 /* Set fe_device and copy extents from local buffer into main return buffer */
2656 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2657                                   struct ll_fiemap_extent *lcl_fm_ext,
2658                                   int ost_index, unsigned int ext_count,
2659                                   int current_extent)
2660 {
2661         char *to;
2662         int ext;
2663
2664         for (ext = 0; ext < ext_count; ext++) {
2665                 lcl_fm_ext[ext].fe_device = ost_index;
2666                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2667         }
2668
2669         /* Copy fm_extent's from fm_local to return buffer */
2670         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2671         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2672 }
2673
2674 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2675                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2676 {
2677         struct ll_fiemap_info_key *fm_key = key;
2678         struct ll_user_fiemap *fiemap = val;
2679         struct ll_user_fiemap *fm_local = NULL;
2680         struct ll_fiemap_extent *lcl_fm_ext;
2681         int count_local;
2682         unsigned int get_num_extents = 0;
2683         int ost_index = 0, actual_start_stripe, start_stripe;
2684         obd_size fm_start, fm_end, fm_length, fm_end_offset = 0;
2685         obd_size curr_loc;
2686         int current_extent = 0, rc = 0, i;
2687         int ost_eof = 0; /* EOF for object */
2688         int ost_done = 0; /* done with required mapping for this OST? */
2689         int last_stripe;
2690         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2691         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2692
2693         if (lsm == NULL)
2694                 GOTO(out, rc = 0);
2695
2696         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2697                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2698
2699         OBD_ALLOC(fm_local, buffer_size);
2700         if (fm_local == NULL)
2701                 GOTO(out, rc = -ENOMEM);
2702         lcl_fm_ext = &fm_local->fm_extents[0];
2703
2704         count_local = fiemap_size_to_count(buffer_size);
2705
2706         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2707         fm_start = fiemap->fm_start;
2708         fm_length = fiemap->fm_length;
2709         /* Calculate start stripe, last stripe and length of mapping */
2710         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2711         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2712                                                 fm_start + fm_length - 1);
2713         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2714         if (fm_end > fm_key->oa.o_size)
2715                 fm_end = fm_key->oa.o_size;
2716
2717         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2718                                               actual_start_stripe, &stripe_count);
2719
2720         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start, fm_end,
2721                                                   &start_stripe);
2722
2723         if (fiemap->fm_extent_count == 0) {
2724                 get_num_extents = 1;
2725                 count_local = 0;
2726         }
2727
2728         /* Check each stripe */
2729         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2730              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2731                 obd_size req_fm_len; /* Stores length of required mapping */
2732                 obd_size len_mapped_single_call;
2733                 obd_off lun_start, lun_end, obd_object_end;
2734                 unsigned int ext_count;
2735
2736                 cur_stripe_wrap = cur_stripe;
2737
2738                 /* Find out range of mapping on this stripe */
2739                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2740                                            &lun_start, &obd_object_end)) == 0)
2741                         continue;
2742
2743                 /* If this is a continuation FIEMAP call and we are on
2744                  * starting stripe then lun_start needs to be set to
2745                  * fm_end_offset */
2746                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2747                         lun_start = fm_end_offset;
2748
2749                 if (fm_length != ~0ULL) {
2750                         /* Handle fm_start + fm_length overflow */
2751                         if (fm_start + fm_length < fm_start)
2752                                 fm_length = ~0ULL - fm_start;
2753                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2754                                                      cur_stripe);
2755                 } else {
2756                         lun_end = ~0ULL;
2757                 }
2758
2759                 if (lun_start == lun_end)
2760                         continue;
2761
2762                 req_fm_len = obd_object_end - lun_start;
2763                 fm_local->fm_length = 0;
2764                 len_mapped_single_call = 0;
2765
2766                 /* If the output buffer is very large and the objects have many
2767                  * extents we may need to loop on a single OST repeatedly */
2768                 ost_eof = 0;
2769                 ost_done = 0;
2770                 do {
2771                         if (get_num_extents == 0) {
2772                                 /* Don't get too many extents. */
2773                                 if (current_extent + count_local >
2774                                     fiemap->fm_extent_count)
2775                                         count_local = fiemap->fm_extent_count -
2776                                                                  current_extent;
2777                         }
2778
2779                         lun_start += len_mapped_single_call;
2780                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2781                         req_fm_len = fm_local->fm_length;
2782                         fm_local->fm_extent_count = count_local;
2783                         fm_local->fm_mapped_extents = 0;
2784                         fm_local->fm_flags = fiemap->fm_flags;
2785
2786                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2787                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2788
2789                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2790                                 GOTO(out, rc = -EINVAL);
2791
2792                         /* If OST is inactive, return extent with UNKNOWN flag */
2793                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2794                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2795                                 fm_local->fm_mapped_extents = 1;
2796
2797                                 lcl_fm_ext[0].fe_logical = lun_start;
2798                                 lcl_fm_ext[0].fe_length = obd_object_end -
2799                                                                       lun_start;
2800                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2801
2802                                 goto inactive_tgt;
2803                         }
2804
2805                         fm_local->fm_start = lun_start;
2806                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2807                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2808                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2809                         rc = obd_get_info(lov->lov_tgts[ost_index]->ltd_exp,
2810                                           keylen, key, vallen, fm_local, lsm);
2811                         if (rc != 0)
2812                                 GOTO(out, rc);
2813
2814 inactive_tgt:
2815                         ext_count = fm_local->fm_mapped_extents;
2816                         if (ext_count == 0) {
2817                                 ost_done = 1;
2818                                 /* If last stripe has hole at the end,
2819                                  * then we need to return */
2820                                 if (cur_stripe_wrap == last_stripe) {
2821                                         fiemap->fm_mapped_extents = 0;
2822                                         goto finish;
2823                                 }
2824                                 break;
2825                         }
2826
2827                         /* If we just need num of extents then go to next device */
2828                         if (get_num_extents) {
2829                                 current_extent += ext_count;
2830                                 break;
2831                         }
2832
2833                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2834                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2835
2836                         /* Have we finished mapping on this device? */
2837                         if (req_fm_len <= len_mapped_single_call)
2838                                 ost_done = 1;
2839
2840                         /* Clear the EXTENT_LAST flag which can be present on
2841                          * last extent */
2842                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2843                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2844                                                             ~FIEMAP_EXTENT_LAST;
2845
2846                         curr_loc = lov_stripe_size(lsm,
2847                                            lcl_fm_ext[ext_count - 1].fe_logical+
2848                                            lcl_fm_ext[ext_count - 1].fe_length,
2849                                            cur_stripe);
2850                         if (curr_loc >= fm_key->oa.o_size)
2851                                 ost_eof = 1;
2852
2853                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2854                                                      ost_index, ext_count,
2855                                                      current_extent);
2856
2857                         current_extent += ext_count;
2858
2859                         /* Ran out of available extents? */
2860                         if (current_extent >= fiemap->fm_extent_count)
2861                                 goto finish;
2862                 } while (ost_done == 0 && ost_eof == 0);
2863
2864                 if (cur_stripe_wrap == last_stripe)
2865                         goto finish;
2866         }
2867
2868 finish:
2869         /* Indicate that we are returning device offsets unless file just has
2870          * single stripe */
2871         if (lsm->lsm_stripe_count > 1)
2872                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2873
2874         if (get_num_extents)
2875                 goto skip_last_device_calc;
2876
2877         /* Check if we have reached the last stripe and whether mapping for that
2878          * stripe is done. */
2879         if (cur_stripe_wrap == last_stripe) {
2880                 if (ost_done || ost_eof)
2881                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2882                                                              FIEMAP_EXTENT_LAST;
2883         }
2884
2885 skip_last_device_calc:
2886         fiemap->fm_mapped_extents = current_extent;
2887
2888 out:
2889         OBD_FREE(fm_local, buffer_size);
2890         return rc;
2891 }
2892
2893 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2894                         void *key, __u32 *vallen, void *val,
2895                         struct lov_stripe_md *lsm)
2896 {
2897         struct obd_device *obddev = class_exp2obd(exp);
2898         struct lov_obd *lov = &obddev->u.lov;
2899         int i, rc;
2900         ENTRY;
2901
2902         if (!vallen || !val)
2903                 RETURN(-EFAULT);
2904
2905         obd_getref(obddev);
2906
2907         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2908                 struct {
2909                         char name[16];
2910                         struct ldlm_lock *lock;
2911                 } *data = key;
2912                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2913                 struct lov_oinfo *loi;
2914                 __u32 *stripe = val;
2915
2916                 if (*vallen < sizeof(*stripe))
2917                         GOTO(out, rc = -EFAULT);
2918                 *vallen = sizeof(*stripe);
2919
2920                 /* XXX This is another one of those bits that will need to
2921                  * change if we ever actually support nested LOVs.  It uses
2922                  * the lock's export to find out which stripe it is. */
2923                 /* XXX - it's assumed all the locks for deleted OSTs have
2924                  * been cancelled. Also, the export for deleted OSTs will
2925                  * be NULL and won't match the lock's export. */
2926                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2927                         loi = lsm->lsm_oinfo[i];
2928                         if (!lov->lov_tgts[loi->loi_ost_idx])
2929                                 continue;
2930                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2931                             data->lock->l_conn_export &&
2932                             osc_res_name_eq(loi->loi_id, loi->loi_gr, res_id)) {
2933                                 *stripe = i;
2934                                 GOTO(out, rc = 0);
2935                         }
2936                 }
2937                 LDLM_ERROR(data->lock, "lock on inode without such object");
2938                 dump_lsm(D_ERROR, lsm);
2939                 GOTO(out, rc = -ENXIO);
2940         } else if (KEY_IS(KEY_LAST_ID)) {
2941                 struct obd_id_info *info = val;
2942                 __u32 size = sizeof(obd_id);
2943                 struct lov_tgt_desc *tgt;
2944
2945                 LASSERT(*vallen == sizeof(struct obd_id_info));
2946                 tgt = lov->lov_tgts[info->idx];
2947
2948                 if (!tgt || !tgt->ltd_active)
2949                         GOTO(out, rc = -ESRCH);
2950
2951                 rc = obd_get_info(tgt->ltd_exp, keylen, key, &size, info->data, NULL);
2952                 GOTO(out, rc = 0);
2953         } else if (KEY_IS(KEY_LOVDESC)) {
2954                 struct lov_desc *desc_ret = val;
2955                 *desc_ret = lov->desc;
2956
2957                 GOTO(out, rc = 0);
2958         } else if (KEY_IS(KEY_FIEMAP)) {
2959                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2960                 GOTO(out, rc);
2961         } else if (KEY_IS(KEY_OFF_RPCSIZE)) {
2962                 __u64 *offset = val;
2963                 struct lov_tgt_desc *tgt;
2964                 struct lov_oinfo *loi;
2965                 int stripe;
2966
2967                 LASSERT(*vallen == sizeof(__u64));
2968                 stripe = lov_stripe_number(lsm, *offset);
2969                 loi = lsm->lsm_oinfo[stripe];
2970                 tgt = lov->lov_tgts[loi->loi_ost_idx];
2971                 if (!tgt || !tgt->ltd_active)
2972                         GOTO(out, rc = -ESRCH);
2973                 rc = obd_get_info(tgt->ltd_exp, keylen, key, vallen, val, NULL);
2974                 GOTO(out, rc);
2975         }
2976
2977         rc = -EINVAL;
2978 out:
2979         obd_putref(obddev);
2980         RETURN(rc);
2981 }
2982
2983 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2984                               void *key, obd_count vallen, void *val,
2985                               struct ptlrpc_request_set *set)
2986 {
2987         struct obd_device *obddev = class_exp2obd(exp);
2988         struct lov_obd *lov = &obddev->u.lov;
2989         obd_count count;
2990         int i, rc = 0, err, incr = 0, check_uuid = 0, do_inactive = 0;
2991         int no_set = !set;
2992         unsigned next_id = 0;
2993         struct lov_tgt_desc *tgt;
2994         void *data;
2995         ENTRY;
2996
2997         if (no_set) {
2998                 set = ptlrpc_prep_set();
2999                 if (!set)
3000                         RETURN(-ENOMEM);
3001         }
3002
3003         obd_getref(obddev);
3004         count = lov->desc.ld_tgt_count;
3005
3006         if (KEY_IS(KEY_NEXT_ID)) {
3007                 count = vallen / sizeof(struct obd_id_info);
3008                 vallen = sizeof(obd_id);
3009                 incr = sizeof(struct obd_id_info);
3010                 do_inactive = 1;
3011                 next_id = 1;
3012         } else if (KEY_IS(KEY_CHECKSUM)) {
3013                 do_inactive = 1;
3014         } else if (KEY_IS(KEY_MDS_CONN) || KEY_IS(KEY_UNLINKED)) {
3015                 check_uuid = val ? 1 : 0;
3016         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
3017                 /* use defaults:
3018                 do_inactive = incr = 0;
3019                  */
3020         }
3021
3022         for (i = 0; i < count; i++, val = (char *)val + incr) {
3023                 if (next_id) {
3024                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
3025                         data = ((struct obd_id_info*)val)->data;
3026                 } else {
3027                         tgt = lov->lov_tgts[i];
3028                         data = val;
3029                 }
3030                 /* OST was disconnected */
3031                 if (!tgt || !tgt->ltd_exp)
3032                         continue;
3033
3034                 /* OST is inactive and we don't want inactive OSCs */
3035                 if (!tgt->ltd_active && !do_inactive)
3036                         continue;
3037
3038                 /* Only want a specific OSC */
3039                 if (check_uuid &&
3040                     !obd_uuid_equals(val, &tgt->ltd_uuid))
3041                         continue;
3042
3043                 err = obd_set_info_async(tgt->ltd_exp,
3044                                          keylen, key, vallen, data, set);
3045                 if (!rc)
3046                         rc = err;
3047         }
3048         obd_putref(obddev);
3049         if (no_set) {
3050                 err = ptlrpc_set_wait(set);
3051                 if (!rc)
3052                         rc = err;
3053                 ptlrpc_set_destroy(set);
3054         }
3055         RETURN(rc);
3056 }
3057
3058 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
3059                        struct lov_stripe_md *lsm)
3060 {
3061         int rc;
3062         ENTRY;
3063
3064         if (!lsm)
3065                 RETURN(0);
3066         LASSERT(md_exp);
3067         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
3068         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
3069
3070         RETURN(rc);
3071 }
3072
3073 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
3074 {
3075         struct lov_oinfo *loi;
3076         int i, rc = 0;
3077         ENTRY;
3078
3079         for (i = 0; i < lsm->lsm_stripe_count; i++) {
3080                 loi = lsm->lsm_oinfo[i];
3081                 if (loi->loi_ar.ar_rc && !rc)
3082                         rc = loi->loi_ar.ar_rc;
3083                 loi->loi_ar.ar_rc = 0;
3084         }
3085         RETURN(rc);
3086 }
3087 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
3088
3089
3090 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
3091                            int cmd, __u64 *offset)
3092 {
3093         __u64 start;
3094         __u32 ssize  = lsm->lsm_stripe_size;
3095
3096         if (cmd & OBD_CALC_STRIPE_RPC_ALIGN)
3097                 ssize = ssize > PTLRPC_MAX_BRW_SIZE ?
3098                         PTLRPC_MAX_BRW_SIZE : ssize;
3099
3100         start = *offset;
3101         do_div(start, ssize);
3102         start = start * ssize;
3103
3104         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
3105                ", end "LPU64"\n", *offset, ssize, start, start + ssize - 1);
3106         if (cmd & OBD_CALC_STRIPE_END)
3107                 *offset = start + ssize - 1;
3108         else if (cmd & OBD_CALC_STRIPE_START)
3109                 *offset = start;
3110         else
3111                 LBUG();
3112
3113         RETURN(0);
3114 }
3115
3116 #if 0
3117 struct lov_multi_wait {
3118         struct ldlm_lock *lock;
3119         wait_queue_t      wait;
3120         int               completed;
3121         int               generation;
3122 };
3123
3124 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
3125                       struct lustre_handle *lockh)
3126 {
3127         struct lov_lock_handles *lov_lockh = NULL;
3128         struct lustre_handle *lov_lockhp;
3129         struct lov_obd *lov;
3130         struct lov_oinfo *loi;
3131         struct lov_multi_wait *queues;
3132         int rc = 0, i;
3133         ENTRY;
3134
3135         ASSERT_LSM_MAGIC(lsm);
3136
3137         if (!exp || !exp->exp_obd)
3138                 RETURN(-ENODEV);
3139
3140         LASSERT(lockh != NULL);
3141         if (lsm->lsm_stripe_count > 1) {
3142                 lov_lockh = lov_handle2llh(lockh);
3143                 if (lov_lockh == NULL) {
3144                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3145                         RETURN(-EINVAL);
3146                 }
3147
3148                 lov_lockhp = lov_lockh->llh_handles;
3149         } else {
3150                 lov_lockhp = lockh;
3151         }
3152
3153         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
3154         if (queues == NULL)
3155                 GOTO(out, rc = -ENOMEM);
3156
3157         lov = &exp->exp_obd->u.lov;
3158         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
3159              i++, loi++, lov_lockhp++) {
3160                 struct ldlm_lock *lock;
3161                 struct obd_device *obd;
3162
3163                 lock = ldlm_handle2lock(lov_lockhp);
3164                 if (lock == NULL) {
3165                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
3166                                loi->loi_ost_idx, loi->loi_id);
3167                         queues[i].completed = 1;
3168                         continue;
3169                 }
3170
3171                 queues[i].lock = lock;
3172                 init_waitqueue_entry(&(queues[i].wait), current);
3173                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
3174
3175                 obd = class_exp2obd(lock->l_conn_export);
3176                 if (obd != NULL)
3177                         imp = obd->u.cli.cl_import;
3178                 if (imp != NULL) {
3179                         spin_lock(&imp->imp_lock);
3180                         queues[i].generation = imp->imp_generation;
3181                         spin_unlock(&imp->imp_lock);
3182                 }
3183         }
3184
3185         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
3186                                interrupted_completion_wait, &lwd);
3187         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
3188
3189         for (i = 0; i < lsm->lsm_stripe_count; i++)
3190                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
3191
3192         if (rc == -EINTR || rc == -ETIMEDOUT) {
3193
3194
3195         }
3196
3197  out:
3198         if (lov_lockh != NULL)
3199                 lov_llh_put(lov_lockh);
3200         RETURN(rc);
3201 }
3202 #endif
3203
3204 void lov_stripe_lock(struct lov_stripe_md *md)
3205 {
3206         LASSERT(md->lsm_lock_owner != cfs_current());
3207         spin_lock(&md->lsm_lock);
3208         LASSERT(md->lsm_lock_owner == NULL);
3209         md->lsm_lock_owner = cfs_current();
3210 }
3211 EXPORT_SYMBOL(lov_stripe_lock);
3212
3213 void lov_stripe_unlock(struct lov_stripe_md *md)
3214 {
3215         LASSERT(md->lsm_lock_owner == cfs_current());
3216         md->lsm_lock_owner = NULL;
3217         spin_unlock(&md->lsm_lock);
3218 }
3219 EXPORT_SYMBOL(lov_stripe_unlock);
3220
3221 static int lov_get_lock(struct obd_export *exp, struct lov_stripe_md *lsm,
3222                         void **res, int rw, obd_off start, obd_off end,
3223                         struct lustre_handle *lockh, int flags)
3224 {
3225         struct lov_async_page *l = *res;
3226         obd_off stripe_start, stripe_end = start;
3227         struct lov_lock_handles *lov_lockh = NULL;
3228         int rc;
3229
3230         ENTRY;
3231
3232         if (lockh && lustre_handle_is_used(lockh) &&
3233             !(flags & OBD_FAST_LOCK)) {
3234                 lov_lockh = lov_handle2llh(lockh);
3235                 if (lov_lockh == NULL) {
3236                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
3237                         RETURN(-EINVAL);
3238                 }
3239                 lockh = lov_lockh->llh_handles + l->lap_stripe;
3240         }
3241
3242         /* ensure we don't cross stripe boundaries */
3243         lov_extent_calc(exp, lsm, OBD_CALC_STRIPE_END, &stripe_end);
3244         if (stripe_end <= end)
3245                 GOTO(out, rc = 0);
3246
3247         /* map the region limits to the object limits */
3248         lov_stripe_offset(lsm, start, l->lap_stripe, &stripe_start);
3249         lov_stripe_offset(lsm, end, l->lap_stripe, &stripe_end);
3250
3251         rc = obd_get_lock(exp->exp_obd->u.lov.lov_tgts[lsm->
3252                             lsm_oinfo[l->lap_stripe]->loi_ost_idx]->
3253                             ltd_exp, NULL, &l->lap_sub_cookie,
3254                             rw, stripe_start, stripe_end, lockh, flags);
3255 out:
3256         if (lov_lockh != NULL)
3257                 lov_llh_put(lov_lockh);
3258         RETURN(rc);
3259 }
3260
3261 struct obd_ops lov_obd_ops = {
3262         .o_owner               = THIS_MODULE,
3263         .o_setup               = lov_setup,
3264         .o_precleanup          = lov_precleanup,
3265         .o_cleanup             = lov_cleanup,
3266         .o_process_config      = lov_process_config,
3267         .o_connect             = lov_connect,
3268         .o_disconnect          = lov_disconnect,
3269         .o_statfs              = lov_statfs,
3270         .o_statfs_async        = lov_statfs_async,
3271         .o_packmd              = lov_packmd,
3272         .o_unpackmd            = lov_unpackmd,
3273         .o_checkmd             = lov_checkmd,
3274         .o_create              = lov_create,
3275         .o_destroy             = lov_destroy,
3276         .o_getattr             = lov_getattr,
3277         .o_getattr_async       = lov_getattr_async,
3278         .o_setattr             = lov_setattr,
3279         .o_setattr_async       = lov_setattr_async,
3280         .o_brw                 = lov_brw,
3281         .o_brw_async           = lov_brw_async,
3282         .o_prep_async_page     = lov_prep_async_page,
3283         .o_get_lock            = lov_get_lock,
3284         .o_queue_async_io      = lov_queue_async_io,
3285         .o_set_async_flags     = lov_set_async_flags,
3286         .o_queue_group_io      = lov_queue_group_io,
3287         .o_trigger_group_io    = lov_trigger_group_io,
3288         .o_teardown_async_page = lov_teardown_async_page,
3289         .o_merge_lvb           = lov_merge_lvb,
3290         .o_update_lvb          = lov_update_lvb,
3291         .o_adjust_kms          = lov_adjust_kms,
3292         .o_punch               = lov_punch,
3293         .o_sync                = lov_sync,
3294         .o_enqueue             = lov_enqueue,
3295         .o_match               = lov_match,
3296         .o_change_cbdata       = lov_change_cbdata,
3297         .o_cancel              = lov_cancel,
3298         .o_cancel_unused       = lov_cancel_unused,
3299         .o_join_lru            = lov_join_lru,
3300         .o_iocontrol           = lov_iocontrol,
3301         .o_get_info            = lov_get_info,
3302         .o_set_info_async      = lov_set_info_async,
3303         .o_extent_calc         = lov_extent_calc,
3304         .o_llog_init           = lov_llog_init,
3305         .o_llog_finish         = lov_llog_finish,
3306         .o_notify              = lov_notify,
3307         .o_register_page_removal_cb = lov_obd_register_page_removal_cb,
3308         .o_unregister_page_removal_cb = lov_obd_unregister_page_removal_cb,
3309         .o_register_lock_cancel_cb = lov_obd_register_lock_cancel_cb,
3310         .o_unregister_lock_cancel_cb = lov_obd_unregister_lock_cancel_cb,
3311         .o_pool_new            = lov_pool_new,
3312         .o_pool_rem            = lov_pool_remove,
3313         .o_pool_add            = lov_pool_add,
3314         .o_pool_del            = lov_pool_del,
3315         .o_getref              = lov_getref,
3316         .o_putref              = lov_putref,
3317 };
3318
3319 static quota_interface_t *quota_interface;
3320 extern quota_interface_t lov_quota_interface;
3321
3322 cfs_mem_cache_t *lov_oinfo_slab;
3323
3324 int __init lov_init(void)
3325 {
3326         struct lprocfs_static_vars lvars = { 0 };
3327         int rc, rc2;
3328         ENTRY;
3329
3330         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
3331                                               sizeof(struct lov_oinfo),
3332                                               0, SLAB_HWCACHE_ALIGN);
3333         if (lov_oinfo_slab == NULL)
3334                 return -ENOMEM;
3335         lprocfs_lov_init_vars(&lvars);
3336
3337         request_module("lquota");
3338         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
3339         init_obd_quota_ops(quota_interface, &lov_obd_ops);
3340
3341         rc = class_register_type(&lov_obd_ops, lvars.module_vars,
3342                                  LUSTRE_LOV_NAME);
3343         if (rc) {
3344                 if (quota_interface)
3345                         PORTAL_SYMBOL_PUT(lov_quota_interface);
3346                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
3347                 LASSERT(rc2 == 0);
3348         }
3349
3350         RETURN(rc);
3351 }
3352
3353 #ifdef __KERNEL__
3354 static void /*__exit*/ lov_exit(void)
3355 {
3356         int rc;
3357
3358         if (quota_interface)
3359                 PORTAL_SYMBOL_PUT(lov_quota_interface);
3360
3361         class_unregister_type(LUSTRE_LOV_NAME);
3362         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
3363         LASSERT(rc == 0);
3364 }
3365
3366 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3367 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
3368 MODULE_LICENSE("GPL");
3369
3370 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
3371 #endif